linux/net/core/dev.c
<<
>>
Prefs
   1/*
   2 *      NET3    Protocol independent device support routines.
   3 *
   4 *              This program is free software; you can redistribute it and/or
   5 *              modify it under the terms of the GNU General Public License
   6 *              as published by the Free Software Foundation; either version
   7 *              2 of the License, or (at your option) any later version.
   8 *
   9 *      Derived from the non IP parts of dev.c 1.0.19
  10 *              Authors:        Ross Biro
  11 *                              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12 *                              Mark Evans, <evansmp@uhura.aston.ac.uk>
  13 *
  14 *      Additional Authors:
  15 *              Florian la Roche <rzsfl@rz.uni-sb.de>
  16 *              Alan Cox <gw4pts@gw4pts.ampr.org>
  17 *              David Hinds <dahinds@users.sourceforge.net>
  18 *              Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
  19 *              Adam Sulmicki <adam@cfar.umd.edu>
  20 *              Pekka Riikonen <priikone@poesidon.pspt.fi>
  21 *
  22 *      Changes:
  23 *              D.J. Barrow     :       Fixed bug where dev->refcnt gets set
  24 *                                      to 2 if register_netdev gets called
  25 *                                      before net_dev_init & also removed a
  26 *                                      few lines of code in the process.
  27 *              Alan Cox        :       device private ioctl copies fields back.
  28 *              Alan Cox        :       Transmit queue code does relevant
  29 *                                      stunts to keep the queue safe.
  30 *              Alan Cox        :       Fixed double lock.
  31 *              Alan Cox        :       Fixed promisc NULL pointer trap
  32 *              ????????        :       Support the full private ioctl range
  33 *              Alan Cox        :       Moved ioctl permission check into
  34 *                                      drivers
  35 *              Tim Kordas      :       SIOCADDMULTI/SIOCDELMULTI
  36 *              Alan Cox        :       100 backlog just doesn't cut it when
  37 *                                      you start doing multicast video 8)
  38 *              Alan Cox        :       Rewrote net_bh and list manager.
  39 *              Alan Cox        :       Fix ETH_P_ALL echoback lengths.
  40 *              Alan Cox        :       Took out transmit every packet pass
  41 *                                      Saved a few bytes in the ioctl handler
  42 *              Alan Cox        :       Network driver sets packet type before
  43 *                                      calling netif_rx. Saves a function
  44 *                                      call a packet.
  45 *              Alan Cox        :       Hashed net_bh()
  46 *              Richard Kooijman:       Timestamp fixes.
  47 *              Alan Cox        :       Wrong field in SIOCGIFDSTADDR
  48 *              Alan Cox        :       Device lock protection.
  49 *              Alan Cox        :       Fixed nasty side effect of device close
  50 *                                      changes.
  51 *              Rudi Cilibrasi  :       Pass the right thing to
  52 *                                      set_mac_address()
  53 *              Dave Miller     :       32bit quantity for the device lock to
  54 *                                      make it work out on a Sparc.
  55 *              Bjorn Ekwall    :       Added KERNELD hack.
  56 *              Alan Cox        :       Cleaned up the backlog initialise.
  57 *              Craig Metz      :       SIOCGIFCONF fix if space for under
  58 *                                      1 device.
  59 *          Thomas Bogendoerfer :       Return ENODEV for dev_open, if there
  60 *                                      is no device open function.
  61 *              Andi Kleen      :       Fix error reporting for SIOCGIFCONF
  62 *          Michael Chastain    :       Fix signed/unsigned for SIOCGIFCONF
  63 *              Cyrus Durgin    :       Cleaned for KMOD
  64 *              Adam Sulmicki   :       Bug Fix : Network Device Unload
  65 *                                      A network device unload needs to purge
  66 *                                      the backlog queue.
  67 *      Paul Rusty Russell      :       SIOCSIFNAME
  68 *              Pekka Riikonen  :       Netdev boot-time settings code
  69 *              Andrew Morton   :       Make unregister_netdevice wait
  70 *                                      indefinitely on dev->refcnt
  71 *              J Hadi Salim    :       - Backlog queue sampling
  72 *                                      - netif_rx() feedback
  73 */
  74
  75#include <asm/uaccess.h>
  76#include <linux/bitops.h>
  77#include <linux/capability.h>
  78#include <linux/cpu.h>
  79#include <linux/types.h>
  80#include <linux/kernel.h>
  81#include <linux/hash.h>
  82#include <linux/slab.h>
  83#include <linux/sched.h>
  84#include <linux/mutex.h>
  85#include <linux/string.h>
  86#include <linux/mm.h>
  87#include <linux/socket.h>
  88#include <linux/sockios.h>
  89#include <linux/errno.h>
  90#include <linux/interrupt.h>
  91#include <linux/if_ether.h>
  92#include <linux/netdevice.h>
  93#include <linux/etherdevice.h>
  94#include <linux/ethtool.h>
  95#include <linux/notifier.h>
  96#include <linux/skbuff.h>
  97#include <net/net_namespace.h>
  98#include <net/sock.h>
  99#include <linux/rtnetlink.h>
 100#include <linux/stat.h>
 101#include <net/dst.h>
 102#include <net/pkt_sched.h>
 103#include <net/checksum.h>
 104#include <net/xfrm.h>
 105#include <linux/highmem.h>
 106#include <linux/init.h>
 107#include <linux/module.h>
 108#include <linux/netpoll.h>
 109#include <linux/rcupdate.h>
 110#include <linux/delay.h>
 111#include <net/iw_handler.h>
 112#include <asm/current.h>
 113#include <linux/audit.h>
 114#include <linux/dmaengine.h>
 115#include <linux/err.h>
 116#include <linux/ctype.h>
 117#include <linux/if_arp.h>
 118#include <linux/if_vlan.h>
 119#include <linux/ip.h>
 120#include <net/ip.h>
 121#include <linux/ipv6.h>
 122#include <linux/in.h>
 123#include <linux/jhash.h>
 124#include <linux/random.h>
 125#include <trace/events/napi.h>
 126#include <trace/events/net.h>
 127#include <trace/events/skb.h>
 128#include <linux/pci.h>
 129#include <linux/inetdevice.h>
 130#include <linux/cpu_rmap.h>
 131#include <linux/static_key.h>
 132#include <linux/hashtable.h>
 133#include <linux/vmalloc.h>
 134#include <linux/if_macvlan.h>
 135
 136#include "net-sysfs.h"
 137
 138/* Instead of increasing this, you should create a hash table. */
 139#define MAX_GRO_SKBS 8
 140
 141/* This should be increased if a protocol with a bigger head is added. */
 142#define GRO_MAX_HEAD (MAX_HEADER + 128)
 143
 144static DEFINE_SPINLOCK(ptype_lock);
 145static DEFINE_SPINLOCK(offload_lock);
 146struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
 147struct list_head ptype_all __read_mostly;       /* Taps */
 148static struct list_head offload_base __read_mostly;
 149
 150static int netif_rx_internal(struct sk_buff *skb);
 151
 152/*
 153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
 154 * semaphore.
 155 *
 156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
 157 *
 158 * Writers must hold the rtnl semaphore while they loop through the
 159 * dev_base_head list, and hold dev_base_lock for writing when they do the
 160 * actual updates.  This allows pure readers to access the list even
 161 * while a writer is preparing to update it.
 162 *
 163 * To put it another way, dev_base_lock is held for writing only to
 164 * protect against pure readers; the rtnl semaphore provides the
 165 * protection against other writers.
 166 *
 167 * See, for example usages, register_netdevice() and
 168 * unregister_netdevice(), which must be called with the rtnl
 169 * semaphore held.
 170 */
 171DEFINE_RWLOCK(dev_base_lock);
 172EXPORT_SYMBOL(dev_base_lock);
 173
 174/* protects napi_hash addition/deletion and napi_gen_id */
 175static DEFINE_SPINLOCK(napi_hash_lock);
 176
 177static unsigned int napi_gen_id;
 178static DEFINE_HASHTABLE(napi_hash, 8);
 179
 180static seqcount_t devnet_rename_seq;
 181
 182static inline void dev_base_seq_inc(struct net *net)
 183{
 184        while (++net->dev_base_seq == 0);
 185}
 186
 187static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
 188{
 189        unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
 190
 191        return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
 192}
 193
 194static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
 195{
 196        return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
 197}
 198
 199static inline void rps_lock(struct softnet_data *sd)
 200{
 201#ifdef CONFIG_RPS
 202        spin_lock(&sd->input_pkt_queue.lock);
 203#endif
 204}
 205
 206static inline void rps_unlock(struct softnet_data *sd)
 207{
 208#ifdef CONFIG_RPS
 209        spin_unlock(&sd->input_pkt_queue.lock);
 210#endif
 211}
 212
 213/* Device list insertion */
 214static void list_netdevice(struct net_device *dev)
 215{
 216        struct net *net = dev_net(dev);
 217
 218        ASSERT_RTNL();
 219
 220        write_lock_bh(&dev_base_lock);
 221        list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
 222        hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
 223        hlist_add_head_rcu(&dev->index_hlist,
 224                           dev_index_hash(net, dev->ifindex));
 225        write_unlock_bh(&dev_base_lock);
 226
 227        dev_base_seq_inc(net);
 228}
 229
 230/* Device list removal
 231 * caller must respect a RCU grace period before freeing/reusing dev
 232 */
 233static void unlist_netdevice(struct net_device *dev)
 234{
 235        ASSERT_RTNL();
 236
 237        /* Unlink dev from the device chain */
 238        write_lock_bh(&dev_base_lock);
 239        list_del_rcu(&dev->dev_list);
 240        hlist_del_rcu(&dev->name_hlist);
 241        hlist_del_rcu(&dev->index_hlist);
 242        write_unlock_bh(&dev_base_lock);
 243
 244        dev_base_seq_inc(dev_net(dev));
 245}
 246
 247/*
 248 *      Our notifier list
 249 */
 250
 251static RAW_NOTIFIER_HEAD(netdev_chain);
 252
 253/*
 254 *      Device drivers call our routines to queue packets here. We empty the
 255 *      queue in the local softnet handler.
 256 */
 257
 258DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 259EXPORT_PER_CPU_SYMBOL(softnet_data);
 260
 261#ifdef CONFIG_LOCKDEP
 262/*
 263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
 264 * according to dev->type
 265 */
 266static const unsigned short netdev_lock_type[] =
 267        {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
 268         ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
 269         ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
 270         ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
 271         ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
 272         ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
 273         ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
 274         ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
 275         ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
 276         ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
 277         ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
 278         ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
 279         ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
 280         ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
 281         ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
 282
 283static const char *const netdev_lock_name[] =
 284        {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
 285         "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
 286         "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
 287         "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
 288         "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
 289         "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
 290         "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
 291         "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
 292         "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
 293         "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
 294         "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
 295         "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
 296         "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
 297         "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
 298         "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
 299
 300static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
 301static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
 302
 303static inline unsigned short netdev_lock_pos(unsigned short dev_type)
 304{
 305        int i;
 306
 307        for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
 308                if (netdev_lock_type[i] == dev_type)
 309                        return i;
 310        /* the last key is used by default */
 311        return ARRAY_SIZE(netdev_lock_type) - 1;
 312}
 313
 314static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 315                                                 unsigned short dev_type)
 316{
 317        int i;
 318
 319        i = netdev_lock_pos(dev_type);
 320        lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
 321                                   netdev_lock_name[i]);
 322}
 323
 324static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 325{
 326        int i;
 327
 328        i = netdev_lock_pos(dev->type);
 329        lockdep_set_class_and_name(&dev->addr_list_lock,
 330                                   &netdev_addr_lock_key[i],
 331                                   netdev_lock_name[i]);
 332}
 333#else
 334static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 335                                                 unsigned short dev_type)
 336{
 337}
 338static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 339{
 340}
 341#endif
 342
 343/*******************************************************************************
 344
 345                Protocol management and registration routines
 346
 347*******************************************************************************/
 348
 349/*
 350 *      Add a protocol ID to the list. Now that the input handler is
 351 *      smarter we can dispense with all the messy stuff that used to be
 352 *      here.
 353 *
 354 *      BEWARE!!! Protocol handlers, mangling input packets,
 355 *      MUST BE last in hash buckets and checking protocol handlers
 356 *      MUST start from promiscuous ptype_all chain in net_bh.
 357 *      It is true now, do not change it.
 358 *      Explanation follows: if protocol handler, mangling packet, will
 359 *      be the first on list, it is not able to sense, that packet
 360 *      is cloned and should be copied-on-write, so that it will
 361 *      change it and subsequent readers will get broken packet.
 362 *                                                      --ANK (980803)
 363 */
 364
 365static inline struct list_head *ptype_head(const struct packet_type *pt)
 366{
 367        if (pt->type == htons(ETH_P_ALL))
 368                return &ptype_all;
 369        else
 370                return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
 371}
 372
 373/**
 374 *      dev_add_pack - add packet handler
 375 *      @pt: packet type declaration
 376 *
 377 *      Add a protocol handler to the networking stack. The passed &packet_type
 378 *      is linked into kernel lists and may not be freed until it has been
 379 *      removed from the kernel lists.
 380 *
 381 *      This call does not sleep therefore it can not
 382 *      guarantee all CPU's that are in middle of receiving packets
 383 *      will see the new packet type (until the next received packet).
 384 */
 385
 386void dev_add_pack(struct packet_type *pt)
 387{
 388        struct list_head *head = ptype_head(pt);
 389
 390        spin_lock(&ptype_lock);
 391        list_add_rcu(&pt->list, head);
 392        spin_unlock(&ptype_lock);
 393}
 394EXPORT_SYMBOL(dev_add_pack);
 395
 396/**
 397 *      __dev_remove_pack        - remove packet handler
 398 *      @pt: packet type declaration
 399 *
 400 *      Remove a protocol handler that was previously added to the kernel
 401 *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 402 *      from the kernel lists and can be freed or reused once this function
 403 *      returns.
 404 *
 405 *      The packet type might still be in use by receivers
 406 *      and must not be freed until after all the CPU's have gone
 407 *      through a quiescent state.
 408 */
 409void __dev_remove_pack(struct packet_type *pt)
 410{
 411        struct list_head *head = ptype_head(pt);
 412        struct packet_type *pt1;
 413
 414        spin_lock(&ptype_lock);
 415
 416        list_for_each_entry(pt1, head, list) {
 417                if (pt == pt1) {
 418                        list_del_rcu(&pt->list);
 419                        goto out;
 420                }
 421        }
 422
 423        pr_warn("dev_remove_pack: %p not found\n", pt);
 424out:
 425        spin_unlock(&ptype_lock);
 426}
 427EXPORT_SYMBOL(__dev_remove_pack);
 428
 429/**
 430 *      dev_remove_pack  - remove packet handler
 431 *      @pt: packet type declaration
 432 *
 433 *      Remove a protocol handler that was previously added to the kernel
 434 *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 435 *      from the kernel lists and can be freed or reused once this function
 436 *      returns.
 437 *
 438 *      This call sleeps to guarantee that no CPU is looking at the packet
 439 *      type after return.
 440 */
 441void dev_remove_pack(struct packet_type *pt)
 442{
 443        __dev_remove_pack(pt);
 444
 445        synchronize_net();
 446}
 447EXPORT_SYMBOL(dev_remove_pack);
 448
 449
 450/**
 451 *      dev_add_offload - register offload handlers
 452 *      @po: protocol offload declaration
 453 *
 454 *      Add protocol offload handlers to the networking stack. The passed
 455 *      &proto_offload is linked into kernel lists and may not be freed until
 456 *      it has been removed from the kernel lists.
 457 *
 458 *      This call does not sleep therefore it can not
 459 *      guarantee all CPU's that are in middle of receiving packets
 460 *      will see the new offload handlers (until the next received packet).
 461 */
 462void dev_add_offload(struct packet_offload *po)
 463{
 464        struct list_head *head = &offload_base;
 465
 466        spin_lock(&offload_lock);
 467        list_add_rcu(&po->list, head);
 468        spin_unlock(&offload_lock);
 469}
 470EXPORT_SYMBOL(dev_add_offload);
 471
 472/**
 473 *      __dev_remove_offload     - remove offload handler
 474 *      @po: packet offload declaration
 475 *
 476 *      Remove a protocol offload handler that was previously added to the
 477 *      kernel offload handlers by dev_add_offload(). The passed &offload_type
 478 *      is removed from the kernel lists and can be freed or reused once this
 479 *      function returns.
 480 *
 481 *      The packet type might still be in use by receivers
 482 *      and must not be freed until after all the CPU's have gone
 483 *      through a quiescent state.
 484 */
 485static void __dev_remove_offload(struct packet_offload *po)
 486{
 487        struct list_head *head = &offload_base;
 488        struct packet_offload *po1;
 489
 490        spin_lock(&offload_lock);
 491
 492        list_for_each_entry(po1, head, list) {
 493                if (po == po1) {
 494                        list_del_rcu(&po->list);
 495                        goto out;
 496                }
 497        }
 498
 499        pr_warn("dev_remove_offload: %p not found\n", po);
 500out:
 501        spin_unlock(&offload_lock);
 502}
 503
 504/**
 505 *      dev_remove_offload       - remove packet offload handler
 506 *      @po: packet offload declaration
 507 *
 508 *      Remove a packet offload handler that was previously added to the kernel
 509 *      offload handlers by dev_add_offload(). The passed &offload_type is
 510 *      removed from the kernel lists and can be freed or reused once this
 511 *      function returns.
 512 *
 513 *      This call sleeps to guarantee that no CPU is looking at the packet
 514 *      type after return.
 515 */
 516void dev_remove_offload(struct packet_offload *po)
 517{
 518        __dev_remove_offload(po);
 519
 520        synchronize_net();
 521}
 522EXPORT_SYMBOL(dev_remove_offload);
 523
 524/******************************************************************************
 525
 526                      Device Boot-time Settings Routines
 527
 528*******************************************************************************/
 529
 530/* Boot time configuration table */
 531static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
 532
 533/**
 534 *      netdev_boot_setup_add   - add new setup entry
 535 *      @name: name of the device
 536 *      @map: configured settings for the device
 537 *
 538 *      Adds new setup entry to the dev_boot_setup list.  The function
 539 *      returns 0 on error and 1 on success.  This is a generic routine to
 540 *      all netdevices.
 541 */
 542static int netdev_boot_setup_add(char *name, struct ifmap *map)
 543{
 544        struct netdev_boot_setup *s;
 545        int i;
 546
 547        s = dev_boot_setup;
 548        for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 549                if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
 550                        memset(s[i].name, 0, sizeof(s[i].name));
 551                        strlcpy(s[i].name, name, IFNAMSIZ);
 552                        memcpy(&s[i].map, map, sizeof(s[i].map));
 553                        break;
 554                }
 555        }
 556
 557        return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
 558}
 559
 560/**
 561 *      netdev_boot_setup_check - check boot time settings
 562 *      @dev: the netdevice
 563 *
 564 *      Check boot time settings for the device.
 565 *      The found settings are set for the device to be used
 566 *      later in the device probing.
 567 *      Returns 0 if no settings found, 1 if they are.
 568 */
 569int netdev_boot_setup_check(struct net_device *dev)
 570{
 571        struct netdev_boot_setup *s = dev_boot_setup;
 572        int i;
 573
 574        for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 575                if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
 576                    !strcmp(dev->name, s[i].name)) {
 577                        dev->irq        = s[i].map.irq;
 578                        dev->base_addr  = s[i].map.base_addr;
 579                        dev->mem_start  = s[i].map.mem_start;
 580                        dev->mem_end    = s[i].map.mem_end;
 581                        return 1;
 582                }
 583        }
 584        return 0;
 585}
 586EXPORT_SYMBOL(netdev_boot_setup_check);
 587
 588
 589/**
 590 *      netdev_boot_base        - get address from boot time settings
 591 *      @prefix: prefix for network device
 592 *      @unit: id for network device
 593 *
 594 *      Check boot time settings for the base address of device.
 595 *      The found settings are set for the device to be used
 596 *      later in the device probing.
 597 *      Returns 0 if no settings found.
 598 */
 599unsigned long netdev_boot_base(const char *prefix, int unit)
 600{
 601        const struct netdev_boot_setup *s = dev_boot_setup;
 602        char name[IFNAMSIZ];
 603        int i;
 604
 605        sprintf(name, "%s%d", prefix, unit);
 606
 607        /*
 608         * If device already registered then return base of 1
 609         * to indicate not to probe for this interface
 610         */
 611        if (__dev_get_by_name(&init_net, name))
 612                return 1;
 613
 614        for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
 615                if (!strcmp(name, s[i].name))
 616                        return s[i].map.base_addr;
 617        return 0;
 618}
 619
 620/*
 621 * Saves at boot time configured settings for any netdevice.
 622 */
 623int __init netdev_boot_setup(char *str)
 624{
 625        int ints[5];
 626        struct ifmap map;
 627
 628        str = get_options(str, ARRAY_SIZE(ints), ints);
 629        if (!str || !*str)
 630                return 0;
 631
 632        /* Save settings */
 633        memset(&map, 0, sizeof(map));
 634        if (ints[0] > 0)
 635                map.irq = ints[1];
 636        if (ints[0] > 1)
 637                map.base_addr = ints[2];
 638        if (ints[0] > 2)
 639                map.mem_start = ints[3];
 640        if (ints[0] > 3)
 641                map.mem_end = ints[4];
 642
 643        /* Add new entry to the list */
 644        return netdev_boot_setup_add(str, &map);
 645}
 646
 647__setup("netdev=", netdev_boot_setup);
 648
 649/*******************************************************************************
 650
 651                            Device Interface Subroutines
 652
 653*******************************************************************************/
 654
 655/**
 656 *      __dev_get_by_name       - find a device by its name
 657 *      @net: the applicable net namespace
 658 *      @name: name to find
 659 *
 660 *      Find an interface by name. Must be called under RTNL semaphore
 661 *      or @dev_base_lock. If the name is found a pointer to the device
 662 *      is returned. If the name is not found then %NULL is returned. The
 663 *      reference counters are not incremented so the caller must be
 664 *      careful with locks.
 665 */
 666
 667struct net_device *__dev_get_by_name(struct net *net, const char *name)
 668{
 669        struct net_device *dev;
 670        struct hlist_head *head = dev_name_hash(net, name);
 671
 672        hlist_for_each_entry(dev, head, name_hlist)
 673                if (!strncmp(dev->name, name, IFNAMSIZ))
 674                        return dev;
 675
 676        return NULL;
 677}
 678EXPORT_SYMBOL(__dev_get_by_name);
 679
 680/**
 681 *      dev_get_by_name_rcu     - find a device by its name
 682 *      @net: the applicable net namespace
 683 *      @name: name to find
 684 *
 685 *      Find an interface by name.
 686 *      If the name is found a pointer to the device is returned.
 687 *      If the name is not found then %NULL is returned.
 688 *      The reference counters are not incremented so the caller must be
 689 *      careful with locks. The caller must hold RCU lock.
 690 */
 691
 692struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
 693{
 694        struct net_device *dev;
 695        struct hlist_head *head = dev_name_hash(net, name);
 696
 697        hlist_for_each_entry_rcu(dev, head, name_hlist)
 698                if (!strncmp(dev->name, name, IFNAMSIZ))
 699                        return dev;
 700
 701        return NULL;
 702}
 703EXPORT_SYMBOL(dev_get_by_name_rcu);
 704
 705/**
 706 *      dev_get_by_name         - find a device by its name
 707 *      @net: the applicable net namespace
 708 *      @name: name to find
 709 *
 710 *      Find an interface by name. This can be called from any
 711 *      context and does its own locking. The returned handle has
 712 *      the usage count incremented and the caller must use dev_put() to
 713 *      release it when it is no longer needed. %NULL is returned if no
 714 *      matching device is found.
 715 */
 716
 717struct net_device *dev_get_by_name(struct net *net, const char *name)
 718{
 719        struct net_device *dev;
 720
 721        rcu_read_lock();
 722        dev = dev_get_by_name_rcu(net, name);
 723        if (dev)
 724                dev_hold(dev);
 725        rcu_read_unlock();
 726        return dev;
 727}
 728EXPORT_SYMBOL(dev_get_by_name);
 729
 730/**
 731 *      __dev_get_by_index - find a device by its ifindex
 732 *      @net: the applicable net namespace
 733 *      @ifindex: index of device
 734 *
 735 *      Search for an interface by index. Returns %NULL if the device
 736 *      is not found or a pointer to the device. The device has not
 737 *      had its reference counter increased so the caller must be careful
 738 *      about locking. The caller must hold either the RTNL semaphore
 739 *      or @dev_base_lock.
 740 */
 741
 742struct net_device *__dev_get_by_index(struct net *net, int ifindex)
 743{
 744        struct net_device *dev;
 745        struct hlist_head *head = dev_index_hash(net, ifindex);
 746
 747        hlist_for_each_entry(dev, head, index_hlist)
 748                if (dev->ifindex == ifindex)
 749                        return dev;
 750
 751        return NULL;
 752}
 753EXPORT_SYMBOL(__dev_get_by_index);
 754
 755/**
 756 *      dev_get_by_index_rcu - find a device by its ifindex
 757 *      @net: the applicable net namespace
 758 *      @ifindex: index of device
 759 *
 760 *      Search for an interface by index. Returns %NULL if the device
 761 *      is not found or a pointer to the device. The device has not
 762 *      had its reference counter increased so the caller must be careful
 763 *      about locking. The caller must hold RCU lock.
 764 */
 765
 766struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
 767{
 768        struct net_device *dev;
 769        struct hlist_head *head = dev_index_hash(net, ifindex);
 770
 771        hlist_for_each_entry_rcu(dev, head, index_hlist)
 772                if (dev->ifindex == ifindex)
 773                        return dev;
 774
 775        return NULL;
 776}
 777EXPORT_SYMBOL(dev_get_by_index_rcu);
 778
 779
 780/**
 781 *      dev_get_by_index - find a device by its ifindex
 782 *      @net: the applicable net namespace
 783 *      @ifindex: index of device
 784 *
 785 *      Search for an interface by index. Returns NULL if the device
 786 *      is not found or a pointer to the device. The device returned has
 787 *      had a reference added and the pointer is safe until the user calls
 788 *      dev_put to indicate they have finished with it.
 789 */
 790
 791struct net_device *dev_get_by_index(struct net *net, int ifindex)
 792{
 793        struct net_device *dev;
 794
 795        rcu_read_lock();
 796        dev = dev_get_by_index_rcu(net, ifindex);
 797        if (dev)
 798                dev_hold(dev);
 799        rcu_read_unlock();
 800        return dev;
 801}
 802EXPORT_SYMBOL(dev_get_by_index);
 803
 804/**
 805 *      netdev_get_name - get a netdevice name, knowing its ifindex.
 806 *      @net: network namespace
 807 *      @name: a pointer to the buffer where the name will be stored.
 808 *      @ifindex: the ifindex of the interface to get the name from.
 809 *
 810 *      The use of raw_seqcount_begin() and cond_resched() before
 811 *      retrying is required as we want to give the writers a chance
 812 *      to complete when CONFIG_PREEMPT is not set.
 813 */
 814int netdev_get_name(struct net *net, char *name, int ifindex)
 815{
 816        struct net_device *dev;
 817        unsigned int seq;
 818
 819retry:
 820        seq = raw_seqcount_begin(&devnet_rename_seq);
 821        rcu_read_lock();
 822        dev = dev_get_by_index_rcu(net, ifindex);
 823        if (!dev) {
 824                rcu_read_unlock();
 825                return -ENODEV;
 826        }
 827
 828        strcpy(name, dev->name);
 829        rcu_read_unlock();
 830        if (read_seqcount_retry(&devnet_rename_seq, seq)) {
 831                cond_resched();
 832                goto retry;
 833        }
 834
 835        return 0;
 836}
 837
 838/**
 839 *      dev_getbyhwaddr_rcu - find a device by its hardware address
 840 *      @net: the applicable net namespace
 841 *      @type: media type of device
 842 *      @ha: hardware address
 843 *
 844 *      Search for an interface by MAC address. Returns NULL if the device
 845 *      is not found or a pointer to the device.
 846 *      The caller must hold RCU or RTNL.
 847 *      The returned device has not had its ref count increased
 848 *      and the caller must therefore be careful about locking
 849 *
 850 */
 851
 852struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
 853                                       const char *ha)
 854{
 855        struct net_device *dev;
 856
 857        for_each_netdev_rcu(net, dev)
 858                if (dev->type == type &&
 859                    !memcmp(dev->dev_addr, ha, dev->addr_len))
 860                        return dev;
 861
 862        return NULL;
 863}
 864EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
 865
 866struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
 867{
 868        struct net_device *dev;
 869
 870        ASSERT_RTNL();
 871        for_each_netdev(net, dev)
 872                if (dev->type == type)
 873                        return dev;
 874
 875        return NULL;
 876}
 877EXPORT_SYMBOL(__dev_getfirstbyhwtype);
 878
 879struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
 880{
 881        struct net_device *dev, *ret = NULL;
 882
 883        rcu_read_lock();
 884        for_each_netdev_rcu(net, dev)
 885                if (dev->type == type) {
 886                        dev_hold(dev);
 887                        ret = dev;
 888                        break;
 889                }
 890        rcu_read_unlock();
 891        return ret;
 892}
 893EXPORT_SYMBOL(dev_getfirstbyhwtype);
 894
 895/**
 896 *      dev_get_by_flags_rcu - find any device with given flags
 897 *      @net: the applicable net namespace
 898 *      @if_flags: IFF_* values
 899 *      @mask: bitmask of bits in if_flags to check
 900 *
 901 *      Search for any interface with the given flags. Returns NULL if a device
 902 *      is not found or a pointer to the device. Must be called inside
 903 *      rcu_read_lock(), and result refcount is unchanged.
 904 */
 905
 906struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
 907                                    unsigned short mask)
 908{
 909        struct net_device *dev, *ret;
 910
 911        ret = NULL;
 912        for_each_netdev_rcu(net, dev) {
 913                if (((dev->flags ^ if_flags) & mask) == 0) {
 914                        ret = dev;
 915                        break;
 916                }
 917        }
 918        return ret;
 919}
 920EXPORT_SYMBOL(dev_get_by_flags_rcu);
 921
 922/**
 923 *      dev_valid_name - check if name is okay for network device
 924 *      @name: name string
 925 *
 926 *      Network device names need to be valid file names to
 927 *      to allow sysfs to work.  We also disallow any kind of
 928 *      whitespace.
 929 */
 930bool dev_valid_name(const char *name)
 931{
 932        if (*name == '\0')
 933                return false;
 934        if (strlen(name) >= IFNAMSIZ)
 935                return false;
 936        if (!strcmp(name, ".") || !strcmp(name, ".."))
 937                return false;
 938
 939        while (*name) {
 940                if (*name == '/' || isspace(*name))
 941                        return false;
 942                name++;
 943        }
 944        return true;
 945}
 946EXPORT_SYMBOL(dev_valid_name);
 947
 948/**
 949 *      __dev_alloc_name - allocate a name for a device
 950 *      @net: network namespace to allocate the device name in
 951 *      @name: name format string
 952 *      @buf:  scratch buffer and result name string
 953 *
 954 *      Passed a format string - eg "lt%d" it will try and find a suitable
 955 *      id. It scans list of devices to build up a free map, then chooses
 956 *      the first empty slot. The caller must hold the dev_base or rtnl lock
 957 *      while allocating the name and adding the device in order to avoid
 958 *      duplicates.
 959 *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
 960 *      Returns the number of the unit assigned or a negative errno code.
 961 */
 962
 963static int __dev_alloc_name(struct net *net, const char *name, char *buf)
 964{
 965        int i = 0;
 966        const char *p;
 967        const int max_netdevices = 8*PAGE_SIZE;
 968        unsigned long *inuse;
 969        struct net_device *d;
 970
 971        p = strnchr(name, IFNAMSIZ-1, '%');
 972        if (p) {
 973                /*
 974                 * Verify the string as this thing may have come from
 975                 * the user.  There must be either one "%d" and no other "%"
 976                 * characters.
 977                 */
 978                if (p[1] != 'd' || strchr(p + 2, '%'))
 979                        return -EINVAL;
 980
 981                /* Use one page as a bit array of possible slots */
 982                inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
 983                if (!inuse)
 984                        return -ENOMEM;
 985
 986                for_each_netdev(net, d) {
 987                        if (!sscanf(d->name, name, &i))
 988                                continue;
 989                        if (i < 0 || i >= max_netdevices)
 990                                continue;
 991
 992                        /*  avoid cases where sscanf is not exact inverse of printf */
 993                        snprintf(buf, IFNAMSIZ, name, i);
 994                        if (!strncmp(buf, d->name, IFNAMSIZ))
 995                                set_bit(i, inuse);
 996                }
 997
 998                i = find_first_zero_bit(inuse, max_netdevices);
 999                free_page((unsigned long) inuse);
1000        }
1001
1002        if (buf != name)
1003                snprintf(buf, IFNAMSIZ, name, i);
1004        if (!__dev_get_by_name(net, buf))
1005                return i;
1006
1007        /* It is possible to run out of possible slots
1008         * when the name is long and there isn't enough space left
1009         * for the digits, or if all bits are used.
1010         */
1011        return -ENFILE;
1012}
1013
1014/**
1015 *      dev_alloc_name - allocate a name for a device
1016 *      @dev: device
1017 *      @name: name format string
1018 *
1019 *      Passed a format string - eg "lt%d" it will try and find a suitable
1020 *      id. It scans list of devices to build up a free map, then chooses
1021 *      the first empty slot. The caller must hold the dev_base or rtnl lock
1022 *      while allocating the name and adding the device in order to avoid
1023 *      duplicates.
1024 *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 *      Returns the number of the unit assigned or a negative errno code.
1026 */
1027
1028int dev_alloc_name(struct net_device *dev, const char *name)
1029{
1030        char buf[IFNAMSIZ];
1031        struct net *net;
1032        int ret;
1033
1034        BUG_ON(!dev_net(dev));
1035        net = dev_net(dev);
1036        ret = __dev_alloc_name(net, name, buf);
1037        if (ret >= 0)
1038                strlcpy(dev->name, buf, IFNAMSIZ);
1039        return ret;
1040}
1041EXPORT_SYMBOL(dev_alloc_name);
1042
1043static int dev_alloc_name_ns(struct net *net,
1044                             struct net_device *dev,
1045                             const char *name)
1046{
1047        char buf[IFNAMSIZ];
1048        int ret;
1049
1050        ret = __dev_alloc_name(net, name, buf);
1051        if (ret >= 0)
1052                strlcpy(dev->name, buf, IFNAMSIZ);
1053        return ret;
1054}
1055
1056static int dev_get_valid_name(struct net *net,
1057                              struct net_device *dev,
1058                              const char *name)
1059{
1060        BUG_ON(!net);
1061
1062        if (!dev_valid_name(name))
1063                return -EINVAL;
1064
1065        if (strchr(name, '%'))
1066                return dev_alloc_name_ns(net, dev, name);
1067        else if (__dev_get_by_name(net, name))
1068                return -EEXIST;
1069        else if (dev->name != name)
1070                strlcpy(dev->name, name, IFNAMSIZ);
1071
1072        return 0;
1073}
1074
1075/**
1076 *      dev_change_name - change name of a device
1077 *      @dev: device
1078 *      @newname: name (or format string) must be at least IFNAMSIZ
1079 *
1080 *      Change name of a device, can pass format strings "eth%d".
1081 *      for wildcarding.
1082 */
1083int dev_change_name(struct net_device *dev, const char *newname)
1084{
1085        char oldname[IFNAMSIZ];
1086        int err = 0;
1087        int ret;
1088        struct net *net;
1089
1090        ASSERT_RTNL();
1091        BUG_ON(!dev_net(dev));
1092
1093        net = dev_net(dev);
1094        if (dev->flags & IFF_UP)
1095                return -EBUSY;
1096
1097        write_seqcount_begin(&devnet_rename_seq);
1098
1099        if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1100                write_seqcount_end(&devnet_rename_seq);
1101                return 0;
1102        }
1103
1104        memcpy(oldname, dev->name, IFNAMSIZ);
1105
1106        err = dev_get_valid_name(net, dev, newname);
1107        if (err < 0) {
1108                write_seqcount_end(&devnet_rename_seq);
1109                return err;
1110        }
1111
1112rollback:
1113        ret = device_rename(&dev->dev, dev->name);
1114        if (ret) {
1115                memcpy(dev->name, oldname, IFNAMSIZ);
1116                write_seqcount_end(&devnet_rename_seq);
1117                return ret;
1118        }
1119
1120        write_seqcount_end(&devnet_rename_seq);
1121
1122        netdev_adjacent_rename_links(dev, oldname);
1123
1124        write_lock_bh(&dev_base_lock);
1125        hlist_del_rcu(&dev->name_hlist);
1126        write_unlock_bh(&dev_base_lock);
1127
1128        synchronize_rcu();
1129
1130        write_lock_bh(&dev_base_lock);
1131        hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1132        write_unlock_bh(&dev_base_lock);
1133
1134        ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1135        ret = notifier_to_errno(ret);
1136
1137        if (ret) {
1138                /* err >= 0 after dev_alloc_name() or stores the first errno */
1139                if (err >= 0) {
1140                        err = ret;
1141                        write_seqcount_begin(&devnet_rename_seq);
1142                        memcpy(dev->name, oldname, IFNAMSIZ);
1143                        memcpy(oldname, newname, IFNAMSIZ);
1144                        goto rollback;
1145                } else {
1146                        pr_err("%s: name change rollback failed: %d\n",
1147                               dev->name, ret);
1148                }
1149        }
1150
1151        return err;
1152}
1153
1154/**
1155 *      dev_set_alias - change ifalias of a device
1156 *      @dev: device
1157 *      @alias: name up to IFALIASZ
1158 *      @len: limit of bytes to copy from info
1159 *
1160 *      Set ifalias for a device,
1161 */
1162int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1163{
1164        char *new_ifalias;
1165
1166        ASSERT_RTNL();
1167
1168        if (len >= IFALIASZ)
1169                return -EINVAL;
1170
1171        if (!len) {
1172                kfree(dev->ifalias);
1173                dev->ifalias = NULL;
1174                return 0;
1175        }
1176
1177        new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1178        if (!new_ifalias)
1179                return -ENOMEM;
1180        dev->ifalias = new_ifalias;
1181
1182        strlcpy(dev->ifalias, alias, len+1);
1183        return len;
1184}
1185
1186
1187/**
1188 *      netdev_features_change - device changes features
1189 *      @dev: device to cause notification
1190 *
1191 *      Called to indicate a device has changed features.
1192 */
1193void netdev_features_change(struct net_device *dev)
1194{
1195        call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1196}
1197EXPORT_SYMBOL(netdev_features_change);
1198
1199/**
1200 *      netdev_state_change - device changes state
1201 *      @dev: device to cause notification
1202 *
1203 *      Called to indicate a device has changed state. This function calls
1204 *      the notifier chains for netdev_chain and sends a NEWLINK message
1205 *      to the routing socket.
1206 */
1207void netdev_state_change(struct net_device *dev)
1208{
1209        if (dev->flags & IFF_UP) {
1210                call_netdevice_notifiers(NETDEV_CHANGE, dev);
1211                rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1212        }
1213}
1214EXPORT_SYMBOL(netdev_state_change);
1215
1216/**
1217 *      netdev_notify_peers - notify network peers about existence of @dev
1218 *      @dev: network device
1219 *
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1224 * migration.
1225 */
1226void netdev_notify_peers(struct net_device *dev)
1227{
1228        rtnl_lock();
1229        call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1230        rtnl_unlock();
1231}
1232EXPORT_SYMBOL(netdev_notify_peers);
1233
1234static int __dev_open(struct net_device *dev)
1235{
1236        const struct net_device_ops *ops = dev->netdev_ops;
1237        int ret;
1238
1239        ASSERT_RTNL();
1240
1241        if (!netif_device_present(dev))
1242                return -ENODEV;
1243
1244        /* Block netpoll from trying to do any rx path servicing.
1245         * If we don't do this there is a chance ndo_poll_controller
1246         * or ndo_poll may be running while we open the device
1247         */
1248        netpoll_rx_disable(dev);
1249
1250        ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1251        ret = notifier_to_errno(ret);
1252        if (ret)
1253                return ret;
1254
1255        set_bit(__LINK_STATE_START, &dev->state);
1256
1257        if (ops->ndo_validate_addr)
1258                ret = ops->ndo_validate_addr(dev);
1259
1260        if (!ret && ops->ndo_open)
1261                ret = ops->ndo_open(dev);
1262
1263        netpoll_rx_enable(dev);
1264
1265        if (ret)
1266                clear_bit(__LINK_STATE_START, &dev->state);
1267        else {
1268                dev->flags |= IFF_UP;
1269                net_dmaengine_get();
1270                dev_set_rx_mode(dev);
1271                dev_activate(dev);
1272                add_device_randomness(dev->dev_addr, dev->addr_len);
1273        }
1274
1275        return ret;
1276}
1277
1278/**
1279 *      dev_open        - prepare an interface for use.
1280 *      @dev:   device to open
1281 *
1282 *      Takes a device from down to up state. The device's private open
1283 *      function is invoked and then the multicast lists are loaded. Finally
1284 *      the device is moved into the up state and a %NETDEV_UP message is
1285 *      sent to the netdev notifier chain.
1286 *
1287 *      Calling this function on an active interface is a nop. On a failure
1288 *      a negative errno code is returned.
1289 */
1290int dev_open(struct net_device *dev)
1291{
1292        int ret;
1293
1294        if (dev->flags & IFF_UP)
1295                return 0;
1296
1297        ret = __dev_open(dev);
1298        if (ret < 0)
1299                return ret;
1300
1301        rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1302        call_netdevice_notifiers(NETDEV_UP, dev);
1303
1304        return ret;
1305}
1306EXPORT_SYMBOL(dev_open);
1307
1308static int __dev_close_many(struct list_head *head)
1309{
1310        struct net_device *dev;
1311
1312        ASSERT_RTNL();
1313        might_sleep();
1314
1315        list_for_each_entry(dev, head, close_list) {
1316                call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1317
1318                clear_bit(__LINK_STATE_START, &dev->state);
1319
1320                /* Synchronize to scheduled poll. We cannot touch poll list, it
1321                 * can be even on different cpu. So just clear netif_running().
1322                 *
1323                 * dev->stop() will invoke napi_disable() on all of it's
1324                 * napi_struct instances on this device.
1325                 */
1326                smp_mb__after_clear_bit(); /* Commit netif_running(). */
1327        }
1328
1329        dev_deactivate_many(head);
1330
1331        list_for_each_entry(dev, head, close_list) {
1332                const struct net_device_ops *ops = dev->netdev_ops;
1333
1334                /*
1335                 *      Call the device specific close. This cannot fail.
1336                 *      Only if device is UP
1337                 *
1338                 *      We allow it to be called even after a DETACH hot-plug
1339                 *      event.
1340                 */
1341                if (ops->ndo_stop)
1342                        ops->ndo_stop(dev);
1343
1344                dev->flags &= ~IFF_UP;
1345                net_dmaengine_put();
1346        }
1347
1348        return 0;
1349}
1350
1351static int __dev_close(struct net_device *dev)
1352{
1353        int retval;
1354        LIST_HEAD(single);
1355
1356        /* Temporarily disable netpoll until the interface is down */
1357        netpoll_rx_disable(dev);
1358
1359        list_add(&dev->close_list, &single);
1360        retval = __dev_close_many(&single);
1361        list_del(&single);
1362
1363        netpoll_rx_enable(dev);
1364        return retval;
1365}
1366
1367static int dev_close_many(struct list_head *head)
1368{
1369        struct net_device *dev, *tmp;
1370
1371        /* Remove the devices that don't need to be closed */
1372        list_for_each_entry_safe(dev, tmp, head, close_list)
1373                if (!(dev->flags & IFF_UP))
1374                        list_del_init(&dev->close_list);
1375
1376        __dev_close_many(head);
1377
1378        list_for_each_entry_safe(dev, tmp, head, close_list) {
1379                rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1380                call_netdevice_notifiers(NETDEV_DOWN, dev);
1381                list_del_init(&dev->close_list);
1382        }
1383
1384        return 0;
1385}
1386
1387/**
1388 *      dev_close - shutdown an interface.
1389 *      @dev: device to shutdown
1390 *
1391 *      This function moves an active device into down state. A
1392 *      %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 *      is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1394 *      chain.
1395 */
1396int dev_close(struct net_device *dev)
1397{
1398        if (dev->flags & IFF_UP) {
1399                LIST_HEAD(single);
1400
1401                /* Block netpoll rx while the interface is going down */
1402                netpoll_rx_disable(dev);
1403
1404                list_add(&dev->close_list, &single);
1405                dev_close_many(&single);
1406                list_del(&single);
1407
1408                netpoll_rx_enable(dev);
1409        }
1410        return 0;
1411}
1412EXPORT_SYMBOL(dev_close);
1413
1414
1415/**
1416 *      dev_disable_lro - disable Large Receive Offload on a device
1417 *      @dev: device
1418 *
1419 *      Disable Large Receive Offload (LRO) on a net device.  Must be
1420 *      called under RTNL.  This is needed if received packets may be
1421 *      forwarded to another interface.
1422 */
1423void dev_disable_lro(struct net_device *dev)
1424{
1425        /*
1426         * If we're trying to disable lro on a vlan device
1427         * use the underlying physical device instead
1428         */
1429        if (is_vlan_dev(dev))
1430                dev = vlan_dev_real_dev(dev);
1431
1432        /* the same for macvlan devices */
1433        if (netif_is_macvlan(dev))
1434                dev = macvlan_dev_real_dev(dev);
1435
1436        dev->wanted_features &= ~NETIF_F_LRO;
1437        netdev_update_features(dev);
1438
1439        if (unlikely(dev->features & NETIF_F_LRO))
1440                netdev_WARN(dev, "failed to disable LRO!\n");
1441}
1442EXPORT_SYMBOL(dev_disable_lro);
1443
1444static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1445                                   struct net_device *dev)
1446{
1447        struct netdev_notifier_info info;
1448
1449        netdev_notifier_info_init(&info, dev);
1450        return nb->notifier_call(nb, val, &info);
1451}
1452
1453static int dev_boot_phase = 1;
1454
1455/**
1456 *      register_netdevice_notifier - register a network notifier block
1457 *      @nb: notifier
1458 *
1459 *      Register a notifier to be called when network device events occur.
1460 *      The notifier passed is linked into the kernel structures and must
1461 *      not be reused until it has been unregistered. A negative errno code
1462 *      is returned on a failure.
1463 *
1464 *      When registered all registration and up events are replayed
1465 *      to the new notifier to allow device to have a race free
1466 *      view of the network device list.
1467 */
1468
1469int register_netdevice_notifier(struct notifier_block *nb)
1470{
1471        struct net_device *dev;
1472        struct net_device *last;
1473        struct net *net;
1474        int err;
1475
1476        rtnl_lock();
1477        err = raw_notifier_chain_register(&netdev_chain, nb);
1478        if (err)
1479                goto unlock;
1480        if (dev_boot_phase)
1481                goto unlock;
1482        for_each_net(net) {
1483                for_each_netdev(net, dev) {
1484                        err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1485                        err = notifier_to_errno(err);
1486                        if (err)
1487                                goto rollback;
1488
1489                        if (!(dev->flags & IFF_UP))
1490                                continue;
1491
1492                        call_netdevice_notifier(nb, NETDEV_UP, dev);
1493                }
1494        }
1495
1496unlock:
1497        rtnl_unlock();
1498        return err;
1499
1500rollback:
1501        last = dev;
1502        for_each_net(net) {
1503                for_each_netdev(net, dev) {
1504                        if (dev == last)
1505                                goto outroll;
1506
1507                        if (dev->flags & IFF_UP) {
1508                                call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1509                                                        dev);
1510                                call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1511                        }
1512                        call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1513                }
1514        }
1515
1516outroll:
1517        raw_notifier_chain_unregister(&netdev_chain, nb);
1518        goto unlock;
1519}
1520EXPORT_SYMBOL(register_netdevice_notifier);
1521
1522/**
1523 *      unregister_netdevice_notifier - unregister a network notifier block
1524 *      @nb: notifier
1525 *
1526 *      Unregister a notifier previously registered by
1527 *      register_netdevice_notifier(). The notifier is unlinked into the
1528 *      kernel structures and may then be reused. A negative errno code
1529 *      is returned on a failure.
1530 *
1531 *      After unregistering unregister and down device events are synthesized
1532 *      for all devices on the device list to the removed notifier to remove
1533 *      the need for special case cleanup code.
1534 */
1535
1536int unregister_netdevice_notifier(struct notifier_block *nb)
1537{
1538        struct net_device *dev;
1539        struct net *net;
1540        int err;
1541
1542        rtnl_lock();
1543        err = raw_notifier_chain_unregister(&netdev_chain, nb);
1544        if (err)
1545                goto unlock;
1546
1547        for_each_net(net) {
1548                for_each_netdev(net, dev) {
1549                        if (dev->flags & IFF_UP) {
1550                                call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1551                                                        dev);
1552                                call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1553                        }
1554                        call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1555                }
1556        }
1557unlock:
1558        rtnl_unlock();
1559        return err;
1560}
1561EXPORT_SYMBOL(unregister_netdevice_notifier);
1562
1563/**
1564 *      call_netdevice_notifiers_info - call all network notifier blocks
1565 *      @val: value passed unmodified to notifier function
1566 *      @dev: net_device pointer passed unmodified to notifier function
1567 *      @info: notifier information data
1568 *
1569 *      Call all network notifier blocks.  Parameters and return value
1570 *      are as for raw_notifier_call_chain().
1571 */
1572
1573static int call_netdevice_notifiers_info(unsigned long val,
1574                                         struct net_device *dev,
1575                                         struct netdev_notifier_info *info)
1576{
1577        ASSERT_RTNL();
1578        netdev_notifier_info_init(info, dev);
1579        return raw_notifier_call_chain(&netdev_chain, val, info);
1580}
1581
1582/**
1583 *      call_netdevice_notifiers - call all network notifier blocks
1584 *      @val: value passed unmodified to notifier function
1585 *      @dev: net_device pointer passed unmodified to notifier function
1586 *
1587 *      Call all network notifier blocks.  Parameters and return value
1588 *      are as for raw_notifier_call_chain().
1589 */
1590
1591int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1592{
1593        struct netdev_notifier_info info;
1594
1595        return call_netdevice_notifiers_info(val, dev, &info);
1596}
1597EXPORT_SYMBOL(call_netdevice_notifiers);
1598
1599static struct static_key netstamp_needed __read_mostly;
1600#ifdef HAVE_JUMP_LABEL
1601/* We are not allowed to call static_key_slow_dec() from irq context
1602 * If net_disable_timestamp() is called from irq context, defer the
1603 * static_key_slow_dec() calls.
1604 */
1605static atomic_t netstamp_needed_deferred;
1606#endif
1607
1608void net_enable_timestamp(void)
1609{
1610#ifdef HAVE_JUMP_LABEL
1611        int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1612
1613        if (deferred) {
1614                while (--deferred)
1615                        static_key_slow_dec(&netstamp_needed);
1616                return;
1617        }
1618#endif
1619        static_key_slow_inc(&netstamp_needed);
1620}
1621EXPORT_SYMBOL(net_enable_timestamp);
1622
1623void net_disable_timestamp(void)
1624{
1625#ifdef HAVE_JUMP_LABEL
1626        if (in_interrupt()) {
1627                atomic_inc(&netstamp_needed_deferred);
1628                return;
1629        }
1630#endif
1631        static_key_slow_dec(&netstamp_needed);
1632}
1633EXPORT_SYMBOL(net_disable_timestamp);
1634
1635static inline void net_timestamp_set(struct sk_buff *skb)
1636{
1637        skb->tstamp.tv64 = 0;
1638        if (static_key_false(&netstamp_needed))
1639                __net_timestamp(skb);
1640}
1641
1642#define net_timestamp_check(COND, SKB)                  \
1643        if (static_key_false(&netstamp_needed)) {               \
1644                if ((COND) && !(SKB)->tstamp.tv64)      \
1645                        __net_timestamp(SKB);           \
1646        }                                               \
1647
1648static inline bool is_skb_forwardable(struct net_device *dev,
1649                                      struct sk_buff *skb)
1650{
1651        unsigned int len;
1652
1653        if (!(dev->flags & IFF_UP))
1654                return false;
1655
1656        len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1657        if (skb->len <= len)
1658                return true;
1659
1660        /* if TSO is enabled, we don't care about the length as the packet
1661         * could be forwarded without being segmented before
1662         */
1663        if (skb_is_gso(skb))
1664                return true;
1665
1666        return false;
1667}
1668
1669/**
1670 * dev_forward_skb - loopback an skb to another netif
1671 *
1672 * @dev: destination network device
1673 * @skb: buffer to forward
1674 *
1675 * return values:
1676 *      NET_RX_SUCCESS  (no congestion)
1677 *      NET_RX_DROP     (packet was dropped, but freed)
1678 *
1679 * dev_forward_skb can be used for injecting an skb from the
1680 * start_xmit function of one device into the receive queue
1681 * of another device.
1682 *
1683 * The receiving device may be in another namespace, so
1684 * we have to clear all information in the skb that could
1685 * impact namespace isolation.
1686 */
1687int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1688{
1689        if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1690                if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1691                        atomic_long_inc(&dev->rx_dropped);
1692                        kfree_skb(skb);
1693                        return NET_RX_DROP;
1694                }
1695        }
1696
1697        if (unlikely(!is_skb_forwardable(dev, skb))) {
1698                atomic_long_inc(&dev->rx_dropped);
1699                kfree_skb(skb);
1700                return NET_RX_DROP;
1701        }
1702
1703        skb_scrub_packet(skb, true);
1704        skb->protocol = eth_type_trans(skb, dev);
1705
1706        return netif_rx_internal(skb);
1707}
1708EXPORT_SYMBOL_GPL(dev_forward_skb);
1709
1710static inline int deliver_skb(struct sk_buff *skb,
1711                              struct packet_type *pt_prev,
1712                              struct net_device *orig_dev)
1713{
1714        if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1715                return -ENOMEM;
1716        atomic_inc(&skb->users);
1717        return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1718}
1719
1720static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1721{
1722        if (!ptype->af_packet_priv || !skb->sk)
1723                return false;
1724
1725        if (ptype->id_match)
1726                return ptype->id_match(ptype, skb->sk);
1727        else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1728                return true;
1729
1730        return false;
1731}
1732
1733/*
1734 *      Support routine. Sends outgoing frames to any network
1735 *      taps currently in use.
1736 */
1737
1738static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1739{
1740        struct packet_type *ptype;
1741        struct sk_buff *skb2 = NULL;
1742        struct packet_type *pt_prev = NULL;
1743
1744        rcu_read_lock();
1745        list_for_each_entry_rcu(ptype, &ptype_all, list) {
1746                /* Never send packets back to the socket
1747                 * they originated from - MvS (miquels@drinkel.ow.org)
1748                 */
1749                if ((ptype->dev == dev || !ptype->dev) &&
1750                    (!skb_loop_sk(ptype, skb))) {
1751                        if (pt_prev) {
1752                                deliver_skb(skb2, pt_prev, skb->dev);
1753                                pt_prev = ptype;
1754                                continue;
1755                        }
1756
1757                        skb2 = skb_clone(skb, GFP_ATOMIC);
1758                        if (!skb2)
1759                                break;
1760
1761                        net_timestamp_set(skb2);
1762
1763                        /* skb->nh should be correctly
1764                           set by sender, so that the second statement is
1765                           just protection against buggy protocols.
1766                         */
1767                        skb_reset_mac_header(skb2);
1768
1769                        if (skb_network_header(skb2) < skb2->data ||
1770                            skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1771                                net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1772                                                     ntohs(skb2->protocol),
1773                                                     dev->name);
1774                                skb_reset_network_header(skb2);
1775                        }
1776
1777                        skb2->transport_header = skb2->network_header;
1778                        skb2->pkt_type = PACKET_OUTGOING;
1779                        pt_prev = ptype;
1780                }
1781        }
1782        if (pt_prev)
1783                pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1784        rcu_read_unlock();
1785}
1786
1787/**
1788 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1789 * @dev: Network device
1790 * @txq: number of queues available
1791 *
1792 * If real_num_tx_queues is changed the tc mappings may no longer be
1793 * valid. To resolve this verify the tc mapping remains valid and if
1794 * not NULL the mapping. With no priorities mapping to this
1795 * offset/count pair it will no longer be used. In the worst case TC0
1796 * is invalid nothing can be done so disable priority mappings. If is
1797 * expected that drivers will fix this mapping if they can before
1798 * calling netif_set_real_num_tx_queues.
1799 */
1800static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1801{
1802        int i;
1803        struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1804
1805        /* If TC0 is invalidated disable TC mapping */
1806        if (tc->offset + tc->count > txq) {
1807                pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1808                dev->num_tc = 0;
1809                return;
1810        }
1811
1812        /* Invalidated prio to tc mappings set to TC0 */
1813        for (i = 1; i < TC_BITMASK + 1; i++) {
1814                int q = netdev_get_prio_tc_map(dev, i);
1815
1816                tc = &dev->tc_to_txq[q];
1817                if (tc->offset + tc->count > txq) {
1818                        pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1819                                i, q);
1820                        netdev_set_prio_tc_map(dev, i, 0);
1821                }
1822        }
1823}
1824
1825#ifdef CONFIG_XPS
1826static DEFINE_MUTEX(xps_map_mutex);
1827#define xmap_dereference(P)             \
1828        rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1829
1830static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1831                                        int cpu, u16 index)
1832{
1833        struct xps_map *map = NULL;
1834        int pos;
1835
1836        if (dev_maps)
1837                map = xmap_dereference(dev_maps->cpu_map[cpu]);
1838
1839        for (pos = 0; map && pos < map->len; pos++) {
1840                if (map->queues[pos] == index) {
1841                        if (map->len > 1) {
1842                                map->queues[pos] = map->queues[--map->len];
1843                        } else {
1844                                RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1845                                kfree_rcu(map, rcu);
1846                                map = NULL;
1847                        }
1848                        break;
1849                }
1850        }
1851
1852        return map;
1853}
1854
1855static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1856{
1857        struct xps_dev_maps *dev_maps;
1858        int cpu, i;
1859        bool active = false;
1860
1861        mutex_lock(&xps_map_mutex);
1862        dev_maps = xmap_dereference(dev->xps_maps);
1863
1864        if (!dev_maps)
1865                goto out_no_maps;
1866
1867        for_each_possible_cpu(cpu) {
1868                for (i = index; i < dev->num_tx_queues; i++) {
1869                        if (!remove_xps_queue(dev_maps, cpu, i))
1870                                break;
1871                }
1872                if (i == dev->num_tx_queues)
1873                        active = true;
1874        }
1875
1876        if (!active) {
1877                RCU_INIT_POINTER(dev->xps_maps, NULL);
1878                kfree_rcu(dev_maps, rcu);
1879        }
1880
1881        for (i = index; i < dev->num_tx_queues; i++)
1882                netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1883                                             NUMA_NO_NODE);
1884
1885out_no_maps:
1886        mutex_unlock(&xps_map_mutex);
1887}
1888
1889static struct xps_map *expand_xps_map(struct xps_map *map,
1890                                      int cpu, u16 index)
1891{
1892        struct xps_map *new_map;
1893        int alloc_len = XPS_MIN_MAP_ALLOC;
1894        int i, pos;
1895
1896        for (pos = 0; map && pos < map->len; pos++) {
1897                if (map->queues[pos] != index)
1898                        continue;
1899                return map;
1900        }
1901
1902        /* Need to add queue to this CPU's existing map */
1903        if (map) {
1904                if (pos < map->alloc_len)
1905                        return map;
1906
1907                alloc_len = map->alloc_len * 2;
1908        }
1909
1910        /* Need to allocate new map to store queue on this CPU's map */
1911        new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1912                               cpu_to_node(cpu));
1913        if (!new_map)
1914                return NULL;
1915
1916        for (i = 0; i < pos; i++)
1917                new_map->queues[i] = map->queues[i];
1918        new_map->alloc_len = alloc_len;
1919        new_map->len = pos;
1920
1921        return new_map;
1922}
1923
1924int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1925                        u16 index)
1926{
1927        struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1928        struct xps_map *map, *new_map;
1929        int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1930        int cpu, numa_node_id = -2;
1931        bool active = false;
1932
1933        mutex_lock(&xps_map_mutex);
1934
1935        dev_maps = xmap_dereference(dev->xps_maps);
1936
1937        /* allocate memory for queue storage */
1938        for_each_online_cpu(cpu) {
1939                if (!cpumask_test_cpu(cpu, mask))
1940                        continue;
1941
1942                if (!new_dev_maps)
1943                        new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1944                if (!new_dev_maps) {
1945                        mutex_unlock(&xps_map_mutex);
1946                        return -ENOMEM;
1947                }
1948
1949                map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1950                                 NULL;
1951
1952                map = expand_xps_map(map, cpu, index);
1953                if (!map)
1954                        goto error;
1955
1956                RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1957        }
1958
1959        if (!new_dev_maps)
1960                goto out_no_new_maps;
1961
1962        for_each_possible_cpu(cpu) {
1963                if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1964                        /* add queue to CPU maps */
1965                        int pos = 0;
1966
1967                        map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1968                        while ((pos < map->len) && (map->queues[pos] != index))
1969                                pos++;
1970
1971                        if (pos == map->len)
1972                                map->queues[map->len++] = index;
1973#ifdef CONFIG_NUMA
1974                        if (numa_node_id == -2)
1975                                numa_node_id = cpu_to_node(cpu);
1976                        else if (numa_node_id != cpu_to_node(cpu))
1977                                numa_node_id = -1;
1978#endif
1979                } else if (dev_maps) {
1980                        /* fill in the new device map from the old device map */
1981                        map = xmap_dereference(dev_maps->cpu_map[cpu]);
1982                        RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1983                }
1984
1985        }
1986
1987        rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1988
1989        /* Cleanup old maps */
1990        if (dev_maps) {
1991                for_each_possible_cpu(cpu) {
1992                        new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1993                        map = xmap_dereference(dev_maps->cpu_map[cpu]);
1994                        if (map && map != new_map)
1995                                kfree_rcu(map, rcu);
1996                }
1997
1998                kfree_rcu(dev_maps, rcu);
1999        }
2000
2001        dev_maps = new_dev_maps;
2002        active = true;
2003
2004out_no_new_maps:
2005        /* update Tx queue numa node */
2006        netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2007                                     (numa_node_id >= 0) ? numa_node_id :
2008                                     NUMA_NO_NODE);
2009
2010        if (!dev_maps)
2011                goto out_no_maps;
2012
2013        /* removes queue from unused CPUs */
2014        for_each_possible_cpu(cpu) {
2015                if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2016                        continue;
2017
2018                if (remove_xps_queue(dev_maps, cpu, index))
2019                        active = true;
2020        }
2021
2022        /* free map if not active */
2023        if (!active) {
2024                RCU_INIT_POINTER(dev->xps_maps, NULL);
2025                kfree_rcu(dev_maps, rcu);
2026        }
2027
2028out_no_maps:
2029        mutex_unlock(&xps_map_mutex);
2030
2031        return 0;
2032error:
2033        /* remove any maps that we added */
2034        for_each_possible_cpu(cpu) {
2035                new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2036                map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2037                                 NULL;
2038                if (new_map && new_map != map)
2039                        kfree(new_map);
2040        }
2041
2042        mutex_unlock(&xps_map_mutex);
2043
2044        kfree(new_dev_maps);
2045        return -ENOMEM;
2046}
2047EXPORT_SYMBOL(netif_set_xps_queue);
2048
2049#endif
2050/*
2051 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2052 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2053 */
2054int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2055{
2056        int rc;
2057
2058        if (txq < 1 || txq > dev->num_tx_queues)
2059                return -EINVAL;
2060
2061        if (dev->reg_state == NETREG_REGISTERED ||
2062            dev->reg_state == NETREG_UNREGISTERING) {
2063                ASSERT_RTNL();
2064
2065                rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2066                                                  txq);
2067                if (rc)
2068                        return rc;
2069
2070                if (dev->num_tc)
2071                        netif_setup_tc(dev, txq);
2072
2073                if (txq < dev->real_num_tx_queues) {
2074                        qdisc_reset_all_tx_gt(dev, txq);
2075#ifdef CONFIG_XPS
2076                        netif_reset_xps_queues_gt(dev, txq);
2077#endif
2078                }
2079        }
2080
2081        dev->real_num_tx_queues = txq;
2082        return 0;
2083}
2084EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2085
2086#ifdef CONFIG_SYSFS
2087/**
2088 *      netif_set_real_num_rx_queues - set actual number of RX queues used
2089 *      @dev: Network device
2090 *      @rxq: Actual number of RX queues
2091 *
2092 *      This must be called either with the rtnl_lock held or before
2093 *      registration of the net device.  Returns 0 on success, or a
2094 *      negative error code.  If called before registration, it always
2095 *      succeeds.
2096 */
2097int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2098{
2099        int rc;
2100
2101        if (rxq < 1 || rxq > dev->num_rx_queues)
2102                return -EINVAL;
2103
2104        if (dev->reg_state == NETREG_REGISTERED) {
2105                ASSERT_RTNL();
2106
2107                rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2108                                                  rxq);
2109                if (rc)
2110                        return rc;
2111        }
2112
2113        dev->real_num_rx_queues = rxq;
2114        return 0;
2115}
2116EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2117#endif
2118
2119/**
2120 * netif_get_num_default_rss_queues - default number of RSS queues
2121 *
2122 * This routine should set an upper limit on the number of RSS queues
2123 * used by default by multiqueue devices.
2124 */
2125int netif_get_num_default_rss_queues(void)
2126{
2127        return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2128}
2129EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2130
2131static inline void __netif_reschedule(struct Qdisc *q)
2132{
2133        struct softnet_data *sd;
2134        unsigned long flags;
2135
2136        local_irq_save(flags);
2137        sd = &__get_cpu_var(softnet_data);
2138        q->next_sched = NULL;
2139        *sd->output_queue_tailp = q;
2140        sd->output_queue_tailp = &q->next_sched;
2141        raise_softirq_irqoff(NET_TX_SOFTIRQ);
2142        local_irq_restore(flags);
2143}
2144
2145void __netif_schedule(struct Qdisc *q)
2146{
2147        if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2148                __netif_reschedule(q);
2149}
2150EXPORT_SYMBOL(__netif_schedule);
2151
2152struct dev_kfree_skb_cb {
2153        enum skb_free_reason reason;
2154};
2155
2156static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2157{
2158        return (struct dev_kfree_skb_cb *)skb->cb;
2159}
2160
2161void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2162{
2163        unsigned long flags;
2164
2165        if (likely(atomic_read(&skb->users) == 1)) {
2166                smp_rmb();
2167                atomic_set(&skb->users, 0);
2168        } else if (likely(!atomic_dec_and_test(&skb->users))) {
2169                return;
2170        }
2171        get_kfree_skb_cb(skb)->reason = reason;
2172        local_irq_save(flags);
2173        skb->next = __this_cpu_read(softnet_data.completion_queue);
2174        __this_cpu_write(softnet_data.completion_queue, skb);
2175        raise_softirq_irqoff(NET_TX_SOFTIRQ);
2176        local_irq_restore(flags);
2177}
2178EXPORT_SYMBOL(__dev_kfree_skb_irq);
2179
2180void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2181{
2182        if (in_irq() || irqs_disabled())
2183                __dev_kfree_skb_irq(skb, reason);
2184        else
2185                dev_kfree_skb(skb);
2186}
2187EXPORT_SYMBOL(__dev_kfree_skb_any);
2188
2189
2190/**
2191 * netif_device_detach - mark device as removed
2192 * @dev: network device
2193 *
2194 * Mark device as removed from system and therefore no longer available.
2195 */
2196void netif_device_detach(struct net_device *dev)
2197{
2198        if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2199            netif_running(dev)) {
2200                netif_tx_stop_all_queues(dev);
2201        }
2202}
2203EXPORT_SYMBOL(netif_device_detach);
2204
2205/**
2206 * netif_device_attach - mark device as attached
2207 * @dev: network device
2208 *
2209 * Mark device as attached from system and restart if needed.
2210 */
2211void netif_device_attach(struct net_device *dev)
2212{
2213        if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2214            netif_running(dev)) {
2215                netif_tx_wake_all_queues(dev);
2216                __netdev_watchdog_up(dev);
2217        }
2218}
2219EXPORT_SYMBOL(netif_device_attach);
2220
2221static void skb_warn_bad_offload(const struct sk_buff *skb)
2222{
2223        static const netdev_features_t null_features = 0;
2224        struct net_device *dev = skb->dev;
2225        const char *driver = "";
2226
2227        if (!net_ratelimit())
2228                return;
2229
2230        if (dev && dev->dev.parent)
2231                driver = dev_driver_string(dev->dev.parent);
2232
2233        WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2234             "gso_type=%d ip_summed=%d\n",
2235             driver, dev ? &dev->features : &null_features,
2236             skb->sk ? &skb->sk->sk_route_caps : &null_features,
2237             skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2238             skb_shinfo(skb)->gso_type, skb->ip_summed);
2239}
2240
2241/*
2242 * Invalidate hardware checksum when packet is to be mangled, and
2243 * complete checksum manually on outgoing path.
2244 */
2245int skb_checksum_help(struct sk_buff *skb)
2246{
2247        __wsum csum;
2248        int ret = 0, offset;
2249
2250        if (skb->ip_summed == CHECKSUM_COMPLETE)
2251                goto out_set_summed;
2252
2253        if (unlikely(skb_shinfo(skb)->gso_size)) {
2254                skb_warn_bad_offload(skb);
2255                return -EINVAL;
2256        }
2257
2258        /* Before computing a checksum, we should make sure no frag could
2259         * be modified by an external entity : checksum could be wrong.
2260         */
2261        if (skb_has_shared_frag(skb)) {
2262                ret = __skb_linearize(skb);
2263                if (ret)
2264                        goto out;
2265        }
2266
2267        offset = skb_checksum_start_offset(skb);
2268        BUG_ON(offset >= skb_headlen(skb));
2269        csum = skb_checksum(skb, offset, skb->len - offset, 0);
2270
2271        offset += skb->csum_offset;
2272        BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2273
2274        if (skb_cloned(skb) &&
2275            !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2276                ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2277                if (ret)
2278                        goto out;
2279        }
2280
2281        *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2282out_set_summed:
2283        skb->ip_summed = CHECKSUM_NONE;
2284out:
2285        return ret;
2286}
2287EXPORT_SYMBOL(skb_checksum_help);
2288
2289__be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2290{
2291        __be16 type = skb->protocol;
2292        int vlan_depth = ETH_HLEN;
2293
2294        /* Tunnel gso handlers can set protocol to ethernet. */
2295        if (type == htons(ETH_P_TEB)) {
2296                struct ethhdr *eth;
2297
2298                if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2299                        return 0;
2300
2301                eth = (struct ethhdr *)skb_mac_header(skb);
2302                type = eth->h_proto;
2303        }
2304
2305        while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2306                struct vlan_hdr *vh;
2307
2308                if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2309                        return 0;
2310
2311                vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2312                type = vh->h_vlan_encapsulated_proto;
2313                vlan_depth += VLAN_HLEN;
2314        }
2315
2316        *depth = vlan_depth;
2317
2318        return type;
2319}
2320
2321/**
2322 *      skb_mac_gso_segment - mac layer segmentation handler.
2323 *      @skb: buffer to segment
2324 *      @features: features for the output path (see dev->features)
2325 */
2326struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2327                                    netdev_features_t features)
2328{
2329        struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2330        struct packet_offload *ptype;
2331        int vlan_depth = skb->mac_len;
2332        __be16 type = skb_network_protocol(skb, &vlan_depth);
2333
2334        if (unlikely(!type))
2335                return ERR_PTR(-EINVAL);
2336
2337        __skb_pull(skb, vlan_depth);
2338
2339        rcu_read_lock();
2340        list_for_each_entry_rcu(ptype, &offload_base, list) {
2341                if (ptype->type == type && ptype->callbacks.gso_segment) {
2342                        if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2343                                int err;
2344
2345                                err = ptype->callbacks.gso_send_check(skb);
2346                                segs = ERR_PTR(err);
2347                                if (err || skb_gso_ok(skb, features))
2348                                        break;
2349                                __skb_push(skb, (skb->data -
2350                                                 skb_network_header(skb)));
2351                        }
2352                        segs = ptype->callbacks.gso_segment(skb, features);
2353                        break;
2354                }
2355        }
2356        rcu_read_unlock();
2357
2358        __skb_push(skb, skb->data - skb_mac_header(skb));
2359
2360        return segs;
2361}
2362EXPORT_SYMBOL(skb_mac_gso_segment);
2363
2364
2365/* openvswitch calls this on rx path, so we need a different check.
2366 */
2367static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2368{
2369        if (tx_path)
2370                return skb->ip_summed != CHECKSUM_PARTIAL;
2371        else
2372                return skb->ip_summed == CHECKSUM_NONE;
2373}
2374
2375/**
2376 *      __skb_gso_segment - Perform segmentation on skb.
2377 *      @skb: buffer to segment
2378 *      @features: features for the output path (see dev->features)
2379 *      @tx_path: whether it is called in TX path
2380 *
2381 *      This function segments the given skb and returns a list of segments.
2382 *
2383 *      It may return NULL if the skb requires no segmentation.  This is
2384 *      only possible when GSO is used for verifying header integrity.
2385 */
2386struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2387                                  netdev_features_t features, bool tx_path)
2388{
2389        if (unlikely(skb_needs_check(skb, tx_path))) {
2390                int err;
2391
2392                skb_warn_bad_offload(skb);
2393
2394                if (skb_header_cloned(skb) &&
2395                    (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2396                        return ERR_PTR(err);
2397        }
2398
2399        SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2400        SKB_GSO_CB(skb)->encap_level = 0;
2401
2402        skb_reset_mac_header(skb);
2403        skb_reset_mac_len(skb);
2404
2405        return skb_mac_gso_segment(skb, features);
2406}
2407EXPORT_SYMBOL(__skb_gso_segment);
2408
2409/* Take action when hardware reception checksum errors are detected. */
2410#ifdef CONFIG_BUG
2411void netdev_rx_csum_fault(struct net_device *dev)
2412{
2413        if (net_ratelimit()) {
2414                pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2415                dump_stack();
2416        }
2417}
2418EXPORT_SYMBOL(netdev_rx_csum_fault);
2419#endif
2420
2421/* Actually, we should eliminate this check as soon as we know, that:
2422 * 1. IOMMU is present and allows to map all the memory.
2423 * 2. No high memory really exists on this machine.
2424 */
2425
2426static int illegal_highdma(const struct net_device *dev, struct sk_buff *skb)
2427{
2428#ifdef CONFIG_HIGHMEM
2429        int i;
2430        if (!(dev->features & NETIF_F_HIGHDMA)) {
2431                for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2432                        skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2433                        if (PageHighMem(skb_frag_page(frag)))
2434                                return 1;
2435                }
2436        }
2437
2438        if (PCI_DMA_BUS_IS_PHYS) {
2439                struct device *pdev = dev->dev.parent;
2440
2441                if (!pdev)
2442                        return 0;
2443                for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2444                        skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2445                        dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2446                        if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2447                                return 1;
2448                }
2449        }
2450#endif
2451        return 0;
2452}
2453
2454struct dev_gso_cb {
2455        void (*destructor)(struct sk_buff *skb);
2456};
2457
2458#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2459
2460static void dev_gso_skb_destructor(struct sk_buff *skb)
2461{
2462        struct dev_gso_cb *cb;
2463
2464        kfree_skb_list(skb->next);
2465        skb->next = NULL;
2466
2467        cb = DEV_GSO_CB(skb);
2468        if (cb->destructor)
2469                cb->destructor(skb);
2470}
2471
2472/**
2473 *      dev_gso_segment - Perform emulated hardware segmentation on skb.
2474 *      @skb: buffer to segment
2475 *      @features: device features as applicable to this skb
2476 *
2477 *      This function segments the given skb and stores the list of segments
2478 *      in skb->next.
2479 */
2480static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2481{
2482        struct sk_buff *segs;
2483
2484        segs = skb_gso_segment(skb, features);
2485
2486        /* Verifying header integrity only. */
2487        if (!segs)
2488                return 0;
2489
2490        if (IS_ERR(segs))
2491                return PTR_ERR(segs);
2492
2493        skb->next = segs;
2494        DEV_GSO_CB(skb)->destructor = skb->destructor;
2495        skb->destructor = dev_gso_skb_destructor;
2496
2497        return 0;
2498}
2499
2500static netdev_features_t harmonize_features(struct sk_buff *skb,
2501                                            const struct net_device *dev,
2502                                            netdev_features_t features)
2503{
2504        int tmp;
2505
2506        if (skb->ip_summed != CHECKSUM_NONE &&
2507            !can_checksum_protocol(features, skb_network_protocol(skb, &tmp))) {
2508                features &= ~NETIF_F_ALL_CSUM;
2509        } else if (illegal_highdma(dev, skb)) {
2510                features &= ~NETIF_F_SG;
2511        }
2512
2513        return features;
2514}
2515
2516netdev_features_t netif_skb_dev_features(struct sk_buff *skb,
2517                                         const struct net_device *dev)
2518{
2519        __be16 protocol = skb->protocol;
2520        netdev_features_t features = dev->features;
2521
2522        if (skb_shinfo(skb)->gso_segs > dev->gso_max_segs)
2523                features &= ~NETIF_F_GSO_MASK;
2524
2525        if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2526                struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2527                protocol = veh->h_vlan_encapsulated_proto;
2528        } else if (!vlan_tx_tag_present(skb)) {
2529                return harmonize_features(skb, dev, features);
2530        }
2531
2532        features &= (dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2533                                               NETIF_F_HW_VLAN_STAG_TX);
2534
2535        if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2536                features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2537                                NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2538                                NETIF_F_HW_VLAN_STAG_TX;
2539
2540        return harmonize_features(skb, dev, features);
2541}
2542EXPORT_SYMBOL(netif_skb_dev_features);
2543
2544int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2545                        struct netdev_queue *txq)
2546{
2547        const struct net_device_ops *ops = dev->netdev_ops;
2548        int rc = NETDEV_TX_OK;
2549        unsigned int skb_len;
2550
2551        if (likely(!skb->next)) {
2552                netdev_features_t features;
2553
2554                /*
2555                 * If device doesn't need skb->dst, release it right now while
2556                 * its hot in this cpu cache
2557                 */
2558                if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2559                        skb_dst_drop(skb);
2560
2561                features = netif_skb_features(skb);
2562
2563                if (vlan_tx_tag_present(skb) &&
2564                    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2565                        skb = __vlan_put_tag(skb, skb->vlan_proto,
2566                                             vlan_tx_tag_get(skb));
2567                        if (unlikely(!skb))
2568                                goto out;
2569
2570                        skb->vlan_tci = 0;
2571                }
2572
2573                /* If encapsulation offload request, verify we are testing
2574                 * hardware encapsulation features instead of standard
2575                 * features for the netdev
2576                 */
2577                if (skb->encapsulation)
2578                        features &= dev->hw_enc_features;
2579
2580                if (netif_needs_gso(skb, features)) {
2581                        if (unlikely(dev_gso_segment(skb, features)))
2582                                goto out_kfree_skb;
2583                        if (skb->next)
2584                                goto gso;
2585                } else {
2586                        if (skb_needs_linearize(skb, features) &&
2587                            __skb_linearize(skb))
2588                                goto out_kfree_skb;
2589
2590                        /* If packet is not checksummed and device does not
2591                         * support checksumming for this protocol, complete
2592                         * checksumming here.
2593                         */
2594                        if (skb->ip_summed == CHECKSUM_PARTIAL) {
2595                                if (skb->encapsulation)
2596                                        skb_set_inner_transport_header(skb,
2597                                                skb_checksum_start_offset(skb));
2598                                else
2599                                        skb_set_transport_header(skb,
2600                                                skb_checksum_start_offset(skb));
2601                                if (!(features & NETIF_F_ALL_CSUM) &&
2602                                     skb_checksum_help(skb))
2603                                        goto out_kfree_skb;
2604                        }
2605                }
2606
2607                if (!list_empty(&ptype_all))
2608                        dev_queue_xmit_nit(skb, dev);
2609
2610                skb_len = skb->len;
2611                trace_net_dev_start_xmit(skb, dev);
2612                rc = ops->ndo_start_xmit(skb, dev);
2613                trace_net_dev_xmit(skb, rc, dev, skb_len);
2614                if (rc == NETDEV_TX_OK)
2615                        txq_trans_update(txq);
2616                return rc;
2617        }
2618
2619gso:
2620        do {
2621                struct sk_buff *nskb = skb->next;
2622
2623                skb->next = nskb->next;
2624                nskb->next = NULL;
2625
2626                if (!list_empty(&ptype_all))
2627                        dev_queue_xmit_nit(nskb, dev);
2628
2629                skb_len = nskb->len;
2630                trace_net_dev_start_xmit(nskb, dev);
2631                rc = ops->ndo_start_xmit(nskb, dev);
2632                trace_net_dev_xmit(nskb, rc, dev, skb_len);
2633                if (unlikely(rc != NETDEV_TX_OK)) {
2634                        if (rc & ~NETDEV_TX_MASK)
2635                                goto out_kfree_gso_skb;
2636                        nskb->next = skb->next;
2637                        skb->next = nskb;
2638                        return rc;
2639                }
2640                txq_trans_update(txq);
2641                if (unlikely(netif_xmit_stopped(txq) && skb->next))
2642                        return NETDEV_TX_BUSY;
2643        } while (skb->next);
2644
2645out_kfree_gso_skb:
2646        if (likely(skb->next == NULL)) {
2647                skb->destructor = DEV_GSO_CB(skb)->destructor;
2648                consume_skb(skb);
2649                return rc;
2650        }
2651out_kfree_skb:
2652        kfree_skb(skb);
2653out:
2654        return rc;
2655}
2656EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2657
2658static void qdisc_pkt_len_init(struct sk_buff *skb)
2659{
2660        const struct skb_shared_info *shinfo = skb_shinfo(skb);
2661
2662        qdisc_skb_cb(skb)->pkt_len = skb->len;
2663
2664        /* To get more precise estimation of bytes sent on wire,
2665         * we add to pkt_len the headers size of all segments
2666         */
2667        if (shinfo->gso_size)  {
2668                unsigned int hdr_len;
2669                u16 gso_segs = shinfo->gso_segs;
2670
2671                /* mac layer + network layer */
2672                hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2673
2674                /* + transport layer */
2675                if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2676                        hdr_len += tcp_hdrlen(skb);
2677                else
2678                        hdr_len += sizeof(struct udphdr);
2679
2680                if (shinfo->gso_type & SKB_GSO_DODGY)
2681                        gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2682                                                shinfo->gso_size);
2683
2684                qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2685        }
2686}
2687
2688static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2689                                 struct net_device *dev,
2690                                 struct netdev_queue *txq)
2691{
2692        spinlock_t *root_lock = qdisc_lock(q);
2693        bool contended;
2694        int rc;
2695
2696        qdisc_pkt_len_init(skb);
2697        qdisc_calculate_pkt_len(skb, q);
2698        /*
2699         * Heuristic to force contended enqueues to serialize on a
2700         * separate lock before trying to get qdisc main lock.
2701         * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2702         * and dequeue packets faster.
2703         */
2704        contended = qdisc_is_running(q);
2705        if (unlikely(contended))
2706                spin_lock(&q->busylock);
2707
2708        spin_lock(root_lock);
2709        if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2710                kfree_skb(skb);
2711                rc = NET_XMIT_DROP;
2712        } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2713                   qdisc_run_begin(q)) {
2714                /*
2715                 * This is a work-conserving queue; there are no old skbs
2716                 * waiting to be sent out; and the qdisc is not running -
2717                 * xmit the skb directly.
2718                 */
2719                if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2720                        skb_dst_force(skb);
2721
2722                qdisc_bstats_update(q, skb);
2723
2724                if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2725                        if (unlikely(contended)) {
2726                                spin_unlock(&q->busylock);
2727                                contended = false;
2728                        }
2729                        __qdisc_run(q);
2730                } else
2731                        qdisc_run_end(q);
2732
2733                rc = NET_XMIT_SUCCESS;
2734        } else {
2735                skb_dst_force(skb);
2736                rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2737                if (qdisc_run_begin(q)) {
2738                        if (unlikely(contended)) {
2739                                spin_unlock(&q->busylock);
2740                                contended = false;
2741                        }
2742                        __qdisc_run(q);
2743                }
2744        }
2745        spin_unlock(root_lock);
2746        if (unlikely(contended))
2747                spin_unlock(&q->busylock);
2748        return rc;
2749}
2750
2751#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2752static void skb_update_prio(struct sk_buff *skb)
2753{
2754        struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2755
2756        if (!skb->priority && skb->sk && map) {
2757                unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2758
2759                if (prioidx < map->priomap_len)
2760                        skb->priority = map->priomap[prioidx];
2761        }
2762}
2763#else
2764#define skb_update_prio(skb)
2765#endif
2766
2767static DEFINE_PER_CPU(int, xmit_recursion);
2768#define RECURSION_LIMIT 10
2769
2770/**
2771 *      dev_loopback_xmit - loop back @skb
2772 *      @skb: buffer to transmit
2773 */
2774int dev_loopback_xmit(struct sk_buff *skb)
2775{
2776        skb_reset_mac_header(skb);
2777        __skb_pull(skb, skb_network_offset(skb));
2778        skb->pkt_type = PACKET_LOOPBACK;
2779        skb->ip_summed = CHECKSUM_UNNECESSARY;
2780        WARN_ON(!skb_dst(skb));
2781        skb_dst_force(skb);
2782        netif_rx_ni(skb);
2783        return 0;
2784}
2785EXPORT_SYMBOL(dev_loopback_xmit);
2786
2787/**
2788 *      __dev_queue_xmit - transmit a buffer
2789 *      @skb: buffer to transmit
2790 *      @accel_priv: private data used for L2 forwarding offload
2791 *
2792 *      Queue a buffer for transmission to a network device. The caller must
2793 *      have set the device and priority and built the buffer before calling
2794 *      this function. The function can be called from an interrupt.
2795 *
2796 *      A negative errno code is returned on a failure. A success does not
2797 *      guarantee the frame will be transmitted as it may be dropped due
2798 *      to congestion or traffic shaping.
2799 *
2800 * -----------------------------------------------------------------------------------
2801 *      I notice this method can also return errors from the queue disciplines,
2802 *      including NET_XMIT_DROP, which is a positive value.  So, errors can also
2803 *      be positive.
2804 *
2805 *      Regardless of the return value, the skb is consumed, so it is currently
2806 *      difficult to retry a send to this method.  (You can bump the ref count
2807 *      before sending to hold a reference for retry if you are careful.)
2808 *
2809 *      When calling this method, interrupts MUST be enabled.  This is because
2810 *      the BH enable code must have IRQs enabled so that it will not deadlock.
2811 *          --BLG
2812 */
2813static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2814{
2815        struct net_device *dev = skb->dev;
2816        struct netdev_queue *txq;
2817        struct Qdisc *q;
2818        int rc = -ENOMEM;
2819
2820        skb_reset_mac_header(skb);
2821
2822        /* Disable soft irqs for various locks below. Also
2823         * stops preemption for RCU.
2824         */
2825        rcu_read_lock_bh();
2826
2827        skb_update_prio(skb);
2828
2829        txq = netdev_pick_tx(dev, skb, accel_priv);
2830        q = rcu_dereference_bh(txq->qdisc);
2831
2832#ifdef CONFIG_NET_CLS_ACT
2833        skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2834#endif
2835        trace_net_dev_queue(skb);
2836        if (q->enqueue) {
2837                rc = __dev_xmit_skb(skb, q, dev, txq);
2838                goto out;
2839        }
2840
2841        /* The device has no queue. Common case for software devices:
2842           loopback, all the sorts of tunnels...
2843
2844           Really, it is unlikely that netif_tx_lock protection is necessary
2845           here.  (f.e. loopback and IP tunnels are clean ignoring statistics
2846           counters.)
2847           However, it is possible, that they rely on protection
2848           made by us here.
2849
2850           Check this and shot the lock. It is not prone from deadlocks.
2851           Either shot noqueue qdisc, it is even simpler 8)
2852         */
2853        if (dev->flags & IFF_UP) {
2854                int cpu = smp_processor_id(); /* ok because BHs are off */
2855
2856                if (txq->xmit_lock_owner != cpu) {
2857
2858                        if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2859                                goto recursion_alert;
2860
2861                        HARD_TX_LOCK(dev, txq, cpu);
2862
2863                        if (!netif_xmit_stopped(txq)) {
2864                                __this_cpu_inc(xmit_recursion);
2865                                rc = dev_hard_start_xmit(skb, dev, txq);
2866                                __this_cpu_dec(xmit_recursion);
2867                                if (dev_xmit_complete(rc)) {
2868                                        HARD_TX_UNLOCK(dev, txq);
2869                                        goto out;
2870                                }
2871                        }
2872                        HARD_TX_UNLOCK(dev, txq);
2873                        net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2874                                             dev->name);
2875                } else {
2876                        /* Recursion is detected! It is possible,
2877                         * unfortunately
2878                         */
2879recursion_alert:
2880                        net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2881                                             dev->name);
2882                }
2883        }
2884
2885        rc = -ENETDOWN;
2886        rcu_read_unlock_bh();
2887
2888        kfree_skb(skb);
2889        return rc;
2890out:
2891        rcu_read_unlock_bh();
2892        return rc;
2893}
2894
2895int dev_queue_xmit(struct sk_buff *skb)
2896{
2897        return __dev_queue_xmit(skb, NULL);
2898}
2899EXPORT_SYMBOL(dev_queue_xmit);
2900
2901int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2902{
2903        return __dev_queue_xmit(skb, accel_priv);
2904}
2905EXPORT_SYMBOL(dev_queue_xmit_accel);
2906
2907
2908/*=======================================================================
2909                        Receiver routines
2910  =======================================================================*/
2911
2912int netdev_max_backlog __read_mostly = 1000;
2913EXPORT_SYMBOL(netdev_max_backlog);
2914
2915int netdev_tstamp_prequeue __read_mostly = 1;
2916int netdev_budget __read_mostly = 300;
2917int weight_p __read_mostly = 64;            /* old backlog weight */
2918
2919/* Called with irq disabled */
2920static inline void ____napi_schedule(struct softnet_data *sd,
2921                                     struct napi_struct *napi)
2922{
2923        list_add_tail(&napi->poll_list, &sd->poll_list);
2924        __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2925}
2926
2927#ifdef CONFIG_RPS
2928
2929/* One global table that all flow-based protocols share. */
2930struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2931EXPORT_SYMBOL(rps_sock_flow_table);
2932
2933struct static_key rps_needed __read_mostly;
2934
2935static struct rps_dev_flow *
2936set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2937            struct rps_dev_flow *rflow, u16 next_cpu)
2938{
2939        if (next_cpu != RPS_NO_CPU) {
2940#ifdef CONFIG_RFS_ACCEL
2941                struct netdev_rx_queue *rxqueue;
2942                struct rps_dev_flow_table *flow_table;
2943                struct rps_dev_flow *old_rflow;
2944                u32 flow_id;
2945                u16 rxq_index;
2946                int rc;
2947
2948                /* Should we steer this flow to a different hardware queue? */
2949                if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2950                    !(dev->features & NETIF_F_NTUPLE))
2951                        goto out;
2952                rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2953                if (rxq_index == skb_get_rx_queue(skb))
2954                        goto out;
2955
2956                rxqueue = dev->_rx + rxq_index;
2957                flow_table = rcu_dereference(rxqueue->rps_flow_table);
2958                if (!flow_table)
2959                        goto out;
2960                flow_id = skb->rxhash & flow_table->mask;
2961                rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2962                                                        rxq_index, flow_id);
2963                if (rc < 0)
2964                        goto out;
2965                old_rflow = rflow;
2966                rflow = &flow_table->flows[flow_id];
2967                rflow->filter = rc;
2968                if (old_rflow->filter == rflow->filter)
2969                        old_rflow->filter = RPS_NO_FILTER;
2970        out:
2971#endif
2972                rflow->last_qtail =
2973                        per_cpu(softnet_data, next_cpu).input_queue_head;
2974        }
2975
2976        rflow->cpu = next_cpu;
2977        return rflow;
2978}
2979
2980/*
2981 * get_rps_cpu is called from netif_receive_skb and returns the target
2982 * CPU from the RPS map of the receiving queue for a given skb.
2983 * rcu_read_lock must be held on entry.
2984 */
2985static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2986                       struct rps_dev_flow **rflowp)
2987{
2988        struct netdev_rx_queue *rxqueue;
2989        struct rps_map *map;
2990        struct rps_dev_flow_table *flow_table;
2991        struct rps_sock_flow_table *sock_flow_table;
2992        int cpu = -1;
2993        u16 tcpu;
2994
2995        if (skb_rx_queue_recorded(skb)) {
2996                u16 index = skb_get_rx_queue(skb);
2997                if (unlikely(index >= dev->real_num_rx_queues)) {
2998                        WARN_ONCE(dev->real_num_rx_queues > 1,
2999                                  "%s received packet on queue %u, but number "
3000                                  "of RX queues is %u\n",
3001                                  dev->name, index, dev->real_num_rx_queues);
3002                        goto done;
3003                }
3004                rxqueue = dev->_rx + index;
3005        } else
3006                rxqueue = dev->_rx;
3007
3008        map = rcu_dereference(rxqueue->rps_map);
3009        if (map) {
3010                if (map->len == 1 &&
3011                    !rcu_access_pointer(rxqueue->rps_flow_table)) {
3012                        tcpu = map->cpus[0];
3013                        if (cpu_online(tcpu))
3014                                cpu = tcpu;
3015                        goto done;
3016                }
3017        } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3018                goto done;
3019        }
3020
3021        skb_reset_network_header(skb);
3022        if (!skb_get_hash(skb))
3023                goto done;
3024
3025        flow_table = rcu_dereference(rxqueue->rps_flow_table);
3026        sock_flow_table = rcu_dereference(rps_sock_flow_table);
3027        if (flow_table && sock_flow_table) {
3028                u16 next_cpu;
3029                struct rps_dev_flow *rflow;
3030
3031                rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3032                tcpu = rflow->cpu;
3033
3034                next_cpu = sock_flow_table->ents[skb->rxhash &
3035                    sock_flow_table->mask];
3036
3037                /*
3038                 * If the desired CPU (where last recvmsg was done) is
3039                 * different from current CPU (one in the rx-queue flow
3040                 * table entry), switch if one of the following holds:
3041                 *   - Current CPU is unset (equal to RPS_NO_CPU).
3042                 *   - Current CPU is offline.
3043                 *   - The current CPU's queue tail has advanced beyond the
3044                 *     last packet that was enqueued using this table entry.
3045                 *     This guarantees that all previous packets for the flow
3046                 *     have been dequeued, thus preserving in order delivery.
3047                 */
3048                if (unlikely(tcpu != next_cpu) &&
3049                    (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3050                     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3051                      rflow->last_qtail)) >= 0)) {
3052                        tcpu = next_cpu;
3053                        rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3054                }
3055
3056                if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3057                        *rflowp = rflow;
3058                        cpu = tcpu;
3059                        goto done;
3060                }
3061        }
3062
3063        if (map) {
3064                tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3065
3066                if (cpu_online(tcpu)) {
3067                        cpu = tcpu;
3068                        goto done;
3069                }
3070        }
3071
3072done:
3073        return cpu;
3074}
3075
3076#ifdef CONFIG_RFS_ACCEL
3077
3078/**
3079 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3080 * @dev: Device on which the filter was set
3081 * @rxq_index: RX queue index
3082 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3083 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3084 *
3085 * Drivers that implement ndo_rx_flow_steer() should periodically call
3086 * this function for each installed filter and remove the filters for
3087 * which it returns %true.
3088 */
3089bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3090                         u32 flow_id, u16 filter_id)
3091{
3092        struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3093        struct rps_dev_flow_table *flow_table;
3094        struct rps_dev_flow *rflow;
3095        bool expire = true;
3096        int cpu;
3097
3098        rcu_read_lock();
3099        flow_table = rcu_dereference(rxqueue->rps_flow_table);
3100        if (flow_table && flow_id <= flow_table->mask) {
3101                rflow = &flow_table->flows[flow_id];
3102                cpu = ACCESS_ONCE(rflow->cpu);
3103                if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3104                    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3105                           rflow->last_qtail) <
3106                     (int)(10 * flow_table->mask)))
3107                        expire = false;
3108        }
3109        rcu_read_unlock();
3110        return expire;
3111}
3112EXPORT_SYMBOL(rps_may_expire_flow);
3113
3114#endif /* CONFIG_RFS_ACCEL */
3115
3116/* Called from hardirq (IPI) context */
3117static void rps_trigger_softirq(void *data)
3118{
3119        struct softnet_data *sd = data;
3120
3121        ____napi_schedule(sd, &sd->backlog);
3122        sd->received_rps++;
3123}
3124
3125#endif /* CONFIG_RPS */
3126
3127/*
3128 * Check if this softnet_data structure is another cpu one
3129 * If yes, queue it to our IPI list and return 1
3130 * If no, return 0
3131 */
3132static int rps_ipi_queued(struct softnet_data *sd)
3133{
3134#ifdef CONFIG_RPS
3135        struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3136
3137        if (sd != mysd) {
3138                sd->rps_ipi_next = mysd->rps_ipi_list;
3139                mysd->rps_ipi_list = sd;
3140
3141                __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3142                return 1;
3143        }
3144#endif /* CONFIG_RPS */
3145        return 0;
3146}
3147
3148#ifdef CONFIG_NET_FLOW_LIMIT
3149int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3150#endif
3151
3152static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3153{
3154#ifdef CONFIG_NET_FLOW_LIMIT
3155        struct sd_flow_limit *fl;
3156        struct softnet_data *sd;
3157        unsigned int old_flow, new_flow;
3158
3159        if (qlen < (netdev_max_backlog >> 1))
3160                return false;
3161
3162        sd = &__get_cpu_var(softnet_data);
3163
3164        rcu_read_lock();
3165        fl = rcu_dereference(sd->flow_limit);
3166        if (fl) {
3167                new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3168                old_flow = fl->history[fl->history_head];
3169                fl->history[fl->history_head] = new_flow;
3170
3171                fl->history_head++;
3172                fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3173
3174                if (likely(fl->buckets[old_flow]))
3175                        fl->buckets[old_flow]--;
3176
3177                if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3178                        fl->count++;
3179                        rcu_read_unlock();
3180                        return true;
3181                }
3182        }
3183        rcu_read_unlock();
3184#endif
3185        return false;
3186}
3187
3188/*
3189 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3190 * queue (may be a remote CPU queue).
3191 */
3192static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3193                              unsigned int *qtail)
3194{
3195        struct softnet_data *sd;
3196        unsigned long flags;
3197        unsigned int qlen;
3198
3199        sd = &per_cpu(softnet_data, cpu);
3200
3201        local_irq_save(flags);
3202
3203        rps_lock(sd);
3204        qlen = skb_queue_len(&sd->input_pkt_queue);
3205        if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3206                if (skb_queue_len(&sd->input_pkt_queue)) {
3207enqueue:
3208                        __skb_queue_tail(&sd->input_pkt_queue, skb);
3209                        input_queue_tail_incr_save(sd, qtail);
3210                        rps_unlock(sd);
3211                        local_irq_restore(flags);
3212                        return NET_RX_SUCCESS;
3213                }
3214
3215                /* Schedule NAPI for backlog device
3216                 * We can use non atomic operation since we own the queue lock
3217                 */
3218                if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3219                        if (!rps_ipi_queued(sd))
3220                                ____napi_schedule(sd, &sd->backlog);
3221                }
3222                goto enqueue;
3223        }
3224
3225        sd->dropped++;
3226        rps_unlock(sd);
3227
3228        local_irq_restore(flags);
3229
3230        atomic_long_inc(&skb->dev->rx_dropped);
3231        kfree_skb(skb);
3232        return NET_RX_DROP;
3233}
3234
3235static int netif_rx_internal(struct sk_buff *skb)
3236{
3237        int ret;
3238
3239        /* if netpoll wants it, pretend we never saw it */
3240        if (netpoll_rx(skb))
3241                return NET_RX_DROP;
3242
3243        net_timestamp_check(netdev_tstamp_prequeue, skb);
3244
3245        trace_netif_rx(skb);
3246#ifdef CONFIG_RPS
3247        if (static_key_false(&rps_needed)) {
3248                struct rps_dev_flow voidflow, *rflow = &voidflow;
3249                int cpu;
3250
3251                preempt_disable();
3252                rcu_read_lock();
3253
3254                cpu = get_rps_cpu(skb->dev, skb, &rflow);
3255                if (cpu < 0)
3256                        cpu = smp_processor_id();
3257
3258                ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3259
3260                rcu_read_unlock();
3261                preempt_enable();
3262        } else
3263#endif
3264        {
3265                unsigned int qtail;
3266                ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3267                put_cpu();
3268        }
3269        return ret;
3270}
3271
3272/**
3273 *      netif_rx        -       post buffer to the network code
3274 *      @skb: buffer to post
3275 *
3276 *      This function receives a packet from a device driver and queues it for
3277 *      the upper (protocol) levels to process.  It always succeeds. The buffer
3278 *      may be dropped during processing for congestion control or by the
3279 *      protocol layers.
3280 *
3281 *      return values:
3282 *      NET_RX_SUCCESS  (no congestion)
3283 *      NET_RX_DROP     (packet was dropped)
3284 *
3285 */
3286
3287int netif_rx(struct sk_buff *skb)
3288{
3289        trace_netif_rx_entry(skb);
3290
3291        return netif_rx_internal(skb);
3292}
3293EXPORT_SYMBOL(netif_rx);
3294
3295int netif_rx_ni(struct sk_buff *skb)
3296{
3297        int err;
3298
3299        trace_netif_rx_ni_entry(skb);
3300
3301        preempt_disable();
3302        err = netif_rx_internal(skb);
3303        if (local_softirq_pending())
3304                do_softirq();
3305        preempt_enable();
3306
3307        return err;
3308}
3309EXPORT_SYMBOL(netif_rx_ni);
3310
3311static void net_tx_action(struct softirq_action *h)
3312{
3313        struct softnet_data *sd = &__get_cpu_var(softnet_data);
3314
3315        if (sd->completion_queue) {
3316                struct sk_buff *clist;
3317
3318                local_irq_disable();
3319                clist = sd->completion_queue;
3320                sd->completion_queue = NULL;
3321                local_irq_enable();
3322
3323                while (clist) {
3324                        struct sk_buff *skb = clist;
3325                        clist = clist->next;
3326
3327                        WARN_ON(atomic_read(&skb->users));
3328                        if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3329                                trace_consume_skb(skb);
3330                        else
3331                                trace_kfree_skb(skb, net_tx_action);
3332                        __kfree_skb(skb);
3333                }
3334        }
3335
3336        if (sd->output_queue) {
3337                struct Qdisc *head;
3338
3339                local_irq_disable();
3340                head = sd->output_queue;
3341                sd->output_queue = NULL;
3342                sd->output_queue_tailp = &sd->output_queue;
3343                local_irq_enable();
3344
3345                while (head) {
3346                        struct Qdisc *q = head;
3347                        spinlock_t *root_lock;
3348
3349                        head = head->next_sched;
3350
3351                        root_lock = qdisc_lock(q);
3352                        if (spin_trylock(root_lock)) {
3353                                smp_mb__before_clear_bit();
3354                                clear_bit(__QDISC_STATE_SCHED,
3355                                          &q->state);
3356                                qdisc_run(q);
3357                                spin_unlock(root_lock);
3358                        } else {
3359                                if (!test_bit(__QDISC_STATE_DEACTIVATED,
3360                                              &q->state)) {
3361                                        __netif_reschedule(q);
3362                                } else {
3363                                        smp_mb__before_clear_bit();
3364                                        clear_bit(__QDISC_STATE_SCHED,
3365                                                  &q->state);
3366                                }
3367                        }
3368                }
3369        }
3370}
3371
3372#if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3373    (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3374/* This hook is defined here for ATM LANE */
3375int (*br_fdb_test_addr_hook)(struct net_device *dev,
3376                             unsigned char *addr) __read_mostly;
3377EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3378#endif
3379
3380#ifdef CONFIG_NET_CLS_ACT
3381/* TODO: Maybe we should just force sch_ingress to be compiled in
3382 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3383 * a compare and 2 stores extra right now if we dont have it on
3384 * but have CONFIG_NET_CLS_ACT
3385 * NOTE: This doesn't stop any functionality; if you dont have
3386 * the ingress scheduler, you just can't add policies on ingress.
3387 *
3388 */
3389static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3390{
3391        struct net_device *dev = skb->dev;
3392        u32 ttl = G_TC_RTTL(skb->tc_verd);
3393        int result = TC_ACT_OK;
3394        struct Qdisc *q;
3395
3396        if (unlikely(MAX_RED_LOOP < ttl++)) {
3397                net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3398                                     skb->skb_iif, dev->ifindex);
3399                return TC_ACT_SHOT;
3400        }
3401
3402        skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3403        skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3404
3405        q = rxq->qdisc;
3406        if (q != &noop_qdisc) {
3407                spin_lock(qdisc_lock(q));
3408                if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3409                        result = qdisc_enqueue_root(skb, q);
3410                spin_unlock(qdisc_lock(q));
3411        }
3412
3413        return result;
3414}
3415
3416static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3417                                         struct packet_type **pt_prev,
3418                                         int *ret, struct net_device *orig_dev)
3419{
3420        struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3421
3422        if (!rxq || rxq->qdisc == &noop_qdisc)
3423                goto out;
3424
3425        if (*pt_prev) {
3426                *ret = deliver_skb(skb, *pt_prev, orig_dev);
3427                *pt_prev = NULL;
3428        }
3429
3430        switch (ing_filter(skb, rxq)) {
3431        case TC_ACT_SHOT:
3432        case TC_ACT_STOLEN:
3433                kfree_skb(skb);
3434                return NULL;
3435        }
3436
3437out:
3438        skb->tc_verd = 0;
3439        return skb;
3440}
3441#endif
3442
3443/**
3444 *      netdev_rx_handler_register - register receive handler
3445 *      @dev: device to register a handler for
3446 *      @rx_handler: receive handler to register
3447 *      @rx_handler_data: data pointer that is used by rx handler
3448 *
3449 *      Register a receive hander for a device. This handler will then be
3450 *      called from __netif_receive_skb. A negative errno code is returned
3451 *      on a failure.
3452 *
3453 *      The caller must hold the rtnl_mutex.
3454 *
3455 *      For a general description of rx_handler, see enum rx_handler_result.
3456 */
3457int netdev_rx_handler_register(struct net_device *dev,
3458                               rx_handler_func_t *rx_handler,
3459                               void *rx_handler_data)
3460{
3461        ASSERT_RTNL();
3462
3463        if (dev->rx_handler)
3464                return -EBUSY;
3465
3466        /* Note: rx_handler_data must be set before rx_handler */
3467        rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3468        rcu_assign_pointer(dev->rx_handler, rx_handler);
3469
3470        return 0;
3471}
3472EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3473
3474/**
3475 *      netdev_rx_handler_unregister - unregister receive handler
3476 *      @dev: device to unregister a handler from
3477 *
3478 *      Unregister a receive handler from a device.
3479 *
3480 *      The caller must hold the rtnl_mutex.
3481 */
3482void netdev_rx_handler_unregister(struct net_device *dev)
3483{
3484
3485        ASSERT_RTNL();
3486        RCU_INIT_POINTER(dev->rx_handler, NULL);
3487        /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3488         * section has a guarantee to see a non NULL rx_handler_data
3489         * as well.
3490         */
3491        synchronize_net();
3492        RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3493}
3494EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3495
3496/*
3497 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3498 * the special handling of PFMEMALLOC skbs.
3499 */
3500static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3501{
3502        switch (skb->protocol) {
3503        case __constant_htons(ETH_P_ARP):
3504        case __constant_htons(ETH_P_IP):
3505        case __constant_htons(ETH_P_IPV6):
3506        case __constant_htons(ETH_P_8021Q):
3507        case __constant_htons(ETH_P_8021AD):
3508                return true;
3509        default:
3510                return false;
3511        }
3512}
3513
3514static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3515{
3516        struct packet_type *ptype, *pt_prev;
3517        rx_handler_func_t *rx_handler;
3518        struct net_device *orig_dev;
3519        struct net_device *null_or_dev;
3520        bool deliver_exact = false;
3521        int ret = NET_RX_DROP;
3522        __be16 type;
3523
3524        net_timestamp_check(!netdev_tstamp_prequeue, skb);
3525
3526        trace_netif_receive_skb(skb);
3527
3528        /* if we've gotten here through NAPI, check netpoll */
3529        if (netpoll_receive_skb(skb))
3530                goto out;
3531
3532        orig_dev = skb->dev;
3533
3534        skb_reset_network_header(skb);
3535        if (!skb_transport_header_was_set(skb))
3536                skb_reset_transport_header(skb);
3537        skb_reset_mac_len(skb);
3538
3539        pt_prev = NULL;
3540
3541        rcu_read_lock();
3542
3543another_round:
3544        skb->skb_iif = skb->dev->ifindex;
3545
3546        __this_cpu_inc(softnet_data.processed);
3547
3548        if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3549            skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3550                skb = vlan_untag(skb);
3551                if (unlikely(!skb))
3552                        goto unlock;
3553        }
3554
3555#ifdef CONFIG_NET_CLS_ACT
3556        if (skb->tc_verd & TC_NCLS) {
3557                skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3558                goto ncls;
3559        }
3560#endif
3561
3562        if (pfmemalloc)
3563                goto skip_taps;
3564
3565        list_for_each_entry_rcu(ptype, &ptype_all, list) {
3566                if (!ptype->dev || ptype->dev == skb->dev) {
3567                        if (pt_prev)
3568                                ret = deliver_skb(skb, pt_prev, orig_dev);
3569                        pt_prev = ptype;
3570                }
3571        }
3572
3573skip_taps:
3574#ifdef CONFIG_NET_CLS_ACT
3575        skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3576        if (!skb)
3577                goto unlock;
3578ncls:
3579#endif
3580
3581        if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3582                goto drop;
3583
3584        if (vlan_tx_tag_present(skb)) {
3585                if (pt_prev) {
3586                        ret = deliver_skb(skb, pt_prev, orig_dev);
3587                        pt_prev = NULL;
3588                }
3589                if (vlan_do_receive(&skb))
3590                        goto another_round;
3591                else if (unlikely(!skb))
3592                        goto unlock;
3593        }
3594
3595        rx_handler = rcu_dereference(skb->dev->rx_handler);
3596        if (rx_handler) {
3597                if (pt_prev) {
3598                        ret = deliver_skb(skb, pt_prev, orig_dev);
3599                        pt_prev = NULL;
3600                }
3601                switch (rx_handler(&skb)) {
3602                case RX_HANDLER_CONSUMED:
3603                        ret = NET_RX_SUCCESS;
3604                        goto unlock;
3605                case RX_HANDLER_ANOTHER:
3606                        goto another_round;
3607                case RX_HANDLER_EXACT:
3608                        deliver_exact = true;
3609                case RX_HANDLER_PASS:
3610                        break;
3611                default:
3612                        BUG();
3613                }
3614        }
3615
3616        if (unlikely(vlan_tx_tag_present(skb))) {
3617                if (vlan_tx_tag_get_id(skb))
3618                        skb->pkt_type = PACKET_OTHERHOST;
3619                /* Note: we might in the future use prio bits
3620                 * and set skb->priority like in vlan_do_receive()
3621                 * For the time being, just ignore Priority Code Point
3622                 */
3623                skb->vlan_tci = 0;
3624        }
3625
3626        /* deliver only exact match when indicated */
3627        null_or_dev = deliver_exact ? skb->dev : NULL;
3628
3629        type = skb->protocol;
3630        list_for_each_entry_rcu(ptype,
3631                        &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3632                if (ptype->type == type &&
3633                    (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3634                     ptype->dev == orig_dev)) {
3635                        if (pt_prev)
3636                                ret = deliver_skb(skb, pt_prev, orig_dev);
3637                        pt_prev = ptype;
3638                }
3639        }
3640
3641        if (pt_prev) {
3642                if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3643                        goto drop;
3644                else
3645                        ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3646        } else {
3647drop:
3648                atomic_long_inc(&skb->dev->rx_dropped);
3649                kfree_skb(skb);
3650                /* Jamal, now you will not able to escape explaining
3651                 * me how you were going to use this. :-)
3652                 */
3653                ret = NET_RX_DROP;
3654        }
3655
3656unlock:
3657        rcu_read_unlock();
3658out:
3659        return ret;
3660}
3661
3662static int __netif_receive_skb(struct sk_buff *skb)
3663{
3664        int ret;
3665
3666        if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3667                unsigned long pflags = current->flags;
3668
3669                /*
3670                 * PFMEMALLOC skbs are special, they should
3671                 * - be delivered to SOCK_MEMALLOC sockets only
3672                 * - stay away from userspace
3673                 * - have bounded memory usage
3674                 *
3675                 * Use PF_MEMALLOC as this saves us from propagating the allocation
3676                 * context down to all allocation sites.
3677                 */
3678                current->flags |= PF_MEMALLOC;
3679                ret = __netif_receive_skb_core(skb, true);
3680                tsk_restore_flags(current, pflags, PF_MEMALLOC);
3681        } else
3682                ret = __netif_receive_skb_core(skb, false);
3683
3684        return ret;
3685}
3686
3687static int netif_receive_skb_internal(struct sk_buff *skb)
3688{
3689        net_timestamp_check(netdev_tstamp_prequeue, skb);
3690
3691        if (skb_defer_rx_timestamp(skb))
3692                return NET_RX_SUCCESS;
3693
3694#ifdef CONFIG_RPS
3695        if (static_key_false(&rps_needed)) {
3696                struct rps_dev_flow voidflow, *rflow = &voidflow;
3697                int cpu, ret;
3698
3699                rcu_read_lock();
3700
3701                cpu = get_rps_cpu(skb->dev, skb, &rflow);
3702
3703                if (cpu >= 0) {
3704                        ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3705                        rcu_read_unlock();
3706                        return ret;
3707                }
3708                rcu_read_unlock();
3709        }
3710#endif
3711        return __netif_receive_skb(skb);
3712}
3713
3714/**
3715 *      netif_receive_skb - process receive buffer from network
3716 *      @skb: buffer to process
3717 *
3718 *      netif_receive_skb() is the main receive data processing function.
3719 *      It always succeeds. The buffer may be dropped during processing
3720 *      for congestion control or by the protocol layers.
3721 *
3722 *      This function may only be called from softirq context and interrupts
3723 *      should be enabled.
3724 *
3725 *      Return values (usually ignored):
3726 *      NET_RX_SUCCESS: no congestion
3727 *      NET_RX_DROP: packet was dropped
3728 */
3729int netif_receive_skb(struct sk_buff *skb)
3730{
3731        trace_netif_receive_skb_entry(skb);
3732
3733        return netif_receive_skb_internal(skb);
3734}
3735EXPORT_SYMBOL(netif_receive_skb);
3736
3737/* Network device is going away, flush any packets still pending
3738 * Called with irqs disabled.
3739 */
3740static void flush_backlog(void *arg)
3741{
3742        struct net_device *dev = arg;
3743        struct softnet_data *sd = &__get_cpu_var(softnet_data);
3744        struct sk_buff *skb, *tmp;
3745
3746        rps_lock(sd);
3747        skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3748                if (skb->dev == dev) {
3749                        __skb_unlink(skb, &sd->input_pkt_queue);
3750                        kfree_skb(skb);
3751                        input_queue_head_incr(sd);
3752                }
3753        }
3754        rps_unlock(sd);
3755
3756        skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3757                if (skb->dev == dev) {
3758                        __skb_unlink(skb, &sd->process_queue);
3759                        kfree_skb(skb);
3760                        input_queue_head_incr(sd);
3761                }
3762        }
3763}
3764
3765static int napi_gro_complete(struct sk_buff *skb)
3766{
3767        struct packet_offload *ptype;
3768        __be16 type = skb->protocol;
3769        struct list_head *head = &offload_base;
3770        int err = -ENOENT;
3771
3772        BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3773
3774        if (NAPI_GRO_CB(skb)->count == 1) {
3775                skb_shinfo(skb)->gso_size = 0;
3776                goto out;
3777        }
3778
3779        rcu_read_lock();
3780        list_for_each_entry_rcu(ptype, head, list) {
3781                if (ptype->type != type || !ptype->callbacks.gro_complete)
3782                        continue;
3783
3784                err = ptype->callbacks.gro_complete(skb, 0);
3785                break;
3786        }
3787        rcu_read_unlock();
3788
3789        if (err) {
3790                WARN_ON(&ptype->list == head);
3791                kfree_skb(skb);
3792                return NET_RX_SUCCESS;
3793        }
3794
3795out:
3796        return netif_receive_skb_internal(skb);
3797}
3798
3799/* napi->gro_list contains packets ordered by age.
3800 * youngest packets at the head of it.
3801 * Complete skbs in reverse order to reduce latencies.
3802 */
3803void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3804{
3805        struct sk_buff *skb, *prev = NULL;
3806
3807        /* scan list and build reverse chain */
3808        for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3809                skb->prev = prev;
3810                prev = skb;
3811        }
3812
3813        for (skb = prev; skb; skb = prev) {
3814                skb->next = NULL;
3815
3816                if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3817                        return;
3818
3819                prev = skb->prev;
3820                napi_gro_complete(skb);
3821                napi->gro_count--;
3822        }
3823
3824        napi->gro_list = NULL;
3825}
3826EXPORT_SYMBOL(napi_gro_flush);
3827
3828static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3829{
3830        struct sk_buff *p;
3831        unsigned int maclen = skb->dev->hard_header_len;
3832        u32 hash = skb_get_hash_raw(skb);
3833
3834        for (p = napi->gro_list; p; p = p->next) {
3835                unsigned long diffs;
3836
3837                NAPI_GRO_CB(p)->flush = 0;
3838
3839                if (hash != skb_get_hash_raw(p)) {
3840                        NAPI_GRO_CB(p)->same_flow = 0;
3841                        continue;
3842                }
3843
3844                diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3845                diffs |= p->vlan_tci ^ skb->vlan_tci;
3846                if (maclen == ETH_HLEN)
3847                        diffs |= compare_ether_header(skb_mac_header(p),
3848                                                      skb_gro_mac_header(skb));
3849                else if (!diffs)
3850                        diffs = memcmp(skb_mac_header(p),
3851                                       skb_gro_mac_header(skb),
3852                                       maclen);
3853                NAPI_GRO_CB(p)->same_flow = !diffs;
3854        }
3855}
3856
3857static void skb_gro_reset_offset(struct sk_buff *skb)
3858{
3859        const struct skb_shared_info *pinfo = skb_shinfo(skb);
3860        const skb_frag_t *frag0 = &pinfo->frags[0];
3861
3862        NAPI_GRO_CB(skb)->data_offset = 0;
3863        NAPI_GRO_CB(skb)->frag0 = NULL;
3864        NAPI_GRO_CB(skb)->frag0_len = 0;
3865
3866        if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3867            pinfo->nr_frags &&
3868            !PageHighMem(skb_frag_page(frag0))) {
3869                NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3870                NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3871        }
3872}
3873
3874static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3875{
3876        struct sk_buff **pp = NULL;
3877        struct packet_offload *ptype;
3878        __be16 type = skb->protocol;
3879        struct list_head *head = &offload_base;
3880        int same_flow;
3881        enum gro_result ret;
3882
3883        if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3884                goto normal;
3885
3886        if (skb_is_gso(skb) || skb_has_frag_list(skb))
3887                goto normal;
3888
3889        skb_gro_reset_offset(skb);
3890        gro_list_prepare(napi, skb);
3891        NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3892
3893        rcu_read_lock();
3894        list_for_each_entry_rcu(ptype, head, list) {
3895                if (ptype->type != type || !ptype->callbacks.gro_receive)
3896                        continue;
3897
3898                skb_set_network_header(skb, skb_gro_offset(skb));
3899                skb_reset_mac_len(skb);
3900                NAPI_GRO_CB(skb)->same_flow = 0;
3901                NAPI_GRO_CB(skb)->flush = 0;
3902                NAPI_GRO_CB(skb)->free = 0;
3903                NAPI_GRO_CB(skb)->udp_mark = 0;
3904
3905                pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3906                break;
3907        }
3908        rcu_read_unlock();
3909
3910        if (&ptype->list == head)
3911                goto normal;
3912
3913        same_flow = NAPI_GRO_CB(skb)->same_flow;
3914        ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3915
3916        if (pp) {
3917                struct sk_buff *nskb = *pp;
3918
3919                *pp = nskb->next;
3920                nskb->next = NULL;
3921                napi_gro_complete(nskb);
3922                napi->gro_count--;
3923        }
3924
3925        if (same_flow)
3926                goto ok;
3927
3928        if (NAPI_GRO_CB(skb)->flush)
3929                goto normal;
3930
3931        if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3932                struct sk_buff *nskb = napi->gro_list;
3933
3934                /* locate the end of the list to select the 'oldest' flow */
3935                while (nskb->next) {
3936                        pp = &nskb->next;
3937                        nskb = *pp;
3938                }
3939                *pp = NULL;
3940                nskb->next = NULL;
3941                napi_gro_complete(nskb);
3942        } else {
3943                napi->gro_count++;
3944        }
3945        NAPI_GRO_CB(skb)->count = 1;
3946        NAPI_GRO_CB(skb)->age = jiffies;
3947        skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3948        skb->next = napi->gro_list;
3949        napi->gro_list = skb;
3950        ret = GRO_HELD;
3951
3952pull:
3953        if (skb_headlen(skb) < skb_gro_offset(skb)) {
3954                int grow = skb_gro_offset(skb) - skb_headlen(skb);
3955
3956                BUG_ON(skb->end - skb->tail < grow);
3957
3958                memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3959
3960                skb->tail += grow;
3961                skb->data_len -= grow;
3962
3963                skb_shinfo(skb)->frags[0].page_offset += grow;
3964                skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3965
3966                if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3967                        skb_frag_unref(skb, 0);
3968                        memmove(skb_shinfo(skb)->frags,
3969                                skb_shinfo(skb)->frags + 1,
3970                                --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3971                }
3972        }
3973
3974ok:
3975        return ret;
3976
3977normal:
3978        ret = GRO_NORMAL;
3979        goto pull;
3980}
3981
3982struct packet_offload *gro_find_receive_by_type(__be16 type)
3983{
3984        struct list_head *offload_head = &offload_base;
3985        struct packet_offload *ptype;
3986
3987        list_for_each_entry_rcu(ptype, offload_head, list) {
3988                if (ptype->type != type || !ptype->callbacks.gro_receive)
3989                        continue;
3990                return ptype;
3991        }
3992        return NULL;
3993}
3994EXPORT_SYMBOL(gro_find_receive_by_type);
3995
3996struct packet_offload *gro_find_complete_by_type(__be16 type)
3997{
3998        struct list_head *offload_head = &offload_base;
3999        struct packet_offload *ptype;
4000
4001        list_for_each_entry_rcu(ptype, offload_head, list) {
4002                if (ptype->type != type || !ptype->callbacks.gro_complete)
4003                        continue;
4004                return ptype;
4005        }
4006        return NULL;
4007}
4008EXPORT_SYMBOL(gro_find_complete_by_type);
4009
4010static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4011{
4012        switch (ret) {
4013        case GRO_NORMAL:
4014                if (netif_receive_skb_internal(skb))
4015                        ret = GRO_DROP;
4016                break;
4017
4018        case GRO_DROP:
4019                kfree_skb(skb);
4020                break;
4021
4022        case GRO_MERGED_FREE:
4023                if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4024                        kmem_cache_free(skbuff_head_cache, skb);
4025                else
4026                        __kfree_skb(skb);
4027                break;
4028
4029        case GRO_HELD:
4030        case GRO_MERGED:
4031                break;
4032        }
4033
4034        return ret;
4035}
4036
4037gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4038{
4039        trace_napi_gro_receive_entry(skb);
4040
4041        return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4042}
4043EXPORT_SYMBOL(napi_gro_receive);
4044
4045static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4046{
4047        __skb_pull(skb, skb_headlen(skb));
4048        /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4049        skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4050        skb->vlan_tci = 0;
4051        skb->dev = napi->dev;
4052        skb->skb_iif = 0;
4053
4054        napi->skb = skb;
4055}
4056
4057struct sk_buff *napi_get_frags(struct napi_struct *napi)
4058{
4059        struct sk_buff *skb = napi->skb;
4060
4061        if (!skb) {
4062                skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4063                napi->skb = skb;
4064        }
4065        return skb;
4066}
4067EXPORT_SYMBOL(napi_get_frags);
4068
4069static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4070                               gro_result_t ret)
4071{
4072        switch (ret) {
4073        case GRO_NORMAL:
4074                if (netif_receive_skb_internal(skb))
4075                        ret = GRO_DROP;
4076                break;
4077
4078        case GRO_DROP:
4079        case GRO_MERGED_FREE:
4080                napi_reuse_skb(napi, skb);
4081                break;
4082
4083        case GRO_HELD:
4084        case GRO_MERGED:
4085                break;
4086        }
4087
4088        return ret;
4089}
4090
4091static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4092{
4093        struct sk_buff *skb = napi->skb;
4094
4095        napi->skb = NULL;
4096
4097        if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4098                napi_reuse_skb(napi, skb);
4099                return NULL;
4100        }
4101        skb->protocol = eth_type_trans(skb, skb->dev);
4102
4103        return skb;
4104}
4105
4106gro_result_t napi_gro_frags(struct napi_struct *napi)
4107{
4108        struct sk_buff *skb = napi_frags_skb(napi);
4109
4110        if (!skb)
4111                return GRO_DROP;
4112
4113        trace_napi_gro_frags_entry(skb);
4114
4115        return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4116}
4117EXPORT_SYMBOL(napi_gro_frags);
4118
4119/*
4120 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4121 * Note: called with local irq disabled, but exits with local irq enabled.
4122 */
4123static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4124{
4125#ifdef CONFIG_RPS
4126        struct softnet_data *remsd = sd->rps_ipi_list;
4127
4128        if (remsd) {
4129                sd->rps_ipi_list = NULL;
4130
4131                local_irq_enable();
4132
4133                /* Send pending IPI's to kick RPS processing on remote cpus. */
4134                while (remsd) {
4135                        struct softnet_data *next = remsd->rps_ipi_next;
4136
4137                        if (cpu_online(remsd->cpu))
4138                                __smp_call_function_single(remsd->cpu,
4139                                                           &remsd->csd, 0);
4140                        remsd = next;
4141                }
4142        } else
4143#endif
4144                local_irq_enable();
4145}
4146
4147static int process_backlog(struct napi_struct *napi, int quota)
4148{
4149        int work = 0;
4150        struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4151
4152#ifdef CONFIG_RPS
4153        /* Check if we have pending ipi, its better to send them now,
4154         * not waiting net_rx_action() end.
4155         */
4156        if (sd->rps_ipi_list) {
4157                local_irq_disable();
4158                net_rps_action_and_irq_enable(sd);
4159        }
4160#endif
4161        napi->weight = weight_p;
4162        local_irq_disable();
4163        while (work < quota) {
4164                struct sk_buff *skb;
4165                unsigned int qlen;
4166
4167                while ((skb = __skb_dequeue(&sd->process_queue))) {
4168                        local_irq_enable();
4169                        __netif_receive_skb(skb);
4170                        local_irq_disable();
4171                        input_queue_head_incr(sd);
4172                        if (++work >= quota) {
4173                                local_irq_enable();
4174                                return work;
4175                        }
4176                }
4177
4178                rps_lock(sd);
4179                qlen = skb_queue_len(&sd->input_pkt_queue);
4180                if (qlen)
4181                        skb_queue_splice_tail_init(&sd->input_pkt_queue,
4182                                                   &sd->process_queue);
4183
4184                if (qlen < quota - work) {
4185                        /*
4186                         * Inline a custom version of __napi_complete().
4187                         * only current cpu owns and manipulates this napi,
4188                         * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4189                         * we can use a plain write instead of clear_bit(),
4190                         * and we dont need an smp_mb() memory barrier.
4191                         */
4192                        list_del(&napi->poll_list);
4193                        napi->state = 0;
4194
4195                        quota = work + qlen;
4196                }
4197                rps_unlock(sd);
4198        }
4199        local_irq_enable();
4200
4201        return work;
4202}
4203
4204/**
4205 * __napi_schedule - schedule for receive
4206 * @n: entry to schedule
4207 *
4208 * The entry's receive function will be scheduled to run
4209 */
4210void __napi_schedule(struct napi_struct *n)
4211{
4212        unsigned long flags;
4213
4214        local_irq_save(flags);
4215        ____napi_schedule(&__get_cpu_var(softnet_data), n);
4216        local_irq_restore(flags);
4217}
4218EXPORT_SYMBOL(__napi_schedule);
4219
4220void __napi_complete(struct napi_struct *n)
4221{
4222        BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4223        BUG_ON(n->gro_list);
4224
4225        list_del(&n->poll_list);
4226        smp_mb__before_clear_bit();
4227        clear_bit(NAPI_STATE_SCHED, &n->state);
4228}
4229EXPORT_SYMBOL(__napi_complete);
4230
4231void napi_complete(struct napi_struct *n)
4232{
4233        unsigned long flags;
4234
4235        /*
4236         * don't let napi dequeue from the cpu poll list
4237         * just in case its running on a different cpu
4238         */
4239        if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4240                return;
4241
4242        napi_gro_flush(n, false);
4243        local_irq_save(flags);
4244        __napi_complete(n);
4245        local_irq_restore(flags);
4246}
4247EXPORT_SYMBOL(napi_complete);
4248
4249/* must be called under rcu_read_lock(), as we dont take a reference */
4250struct napi_struct *napi_by_id(unsigned int napi_id)
4251{
4252        unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4253        struct napi_struct *napi;
4254
4255        hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4256                if (napi->napi_id == napi_id)
4257                        return napi;
4258
4259        return NULL;
4260}
4261EXPORT_SYMBOL_GPL(napi_by_id);
4262
4263void napi_hash_add(struct napi_struct *napi)
4264{
4265        if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4266
4267                spin_lock(&napi_hash_lock);
4268
4269                /* 0 is not a valid id, we also skip an id that is taken
4270                 * we expect both events to be extremely rare
4271                 */
4272                napi->napi_id = 0;
4273                while (!napi->napi_id) {
4274                        napi->napi_id = ++napi_gen_id;
4275                        if (napi_by_id(napi->napi_id))
4276                                napi->napi_id = 0;
4277                }
4278
4279                hlist_add_head_rcu(&napi->napi_hash_node,
4280                        &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4281
4282                spin_unlock(&napi_hash_lock);
4283        }
4284}
4285EXPORT_SYMBOL_GPL(napi_hash_add);
4286
4287/* Warning : caller is responsible to make sure rcu grace period
4288 * is respected before freeing memory containing @napi
4289 */
4290void napi_hash_del(struct napi_struct *napi)
4291{
4292        spin_lock(&napi_hash_lock);
4293
4294        if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4295                hlist_del_rcu(&napi->napi_hash_node);
4296
4297        spin_unlock(&napi_hash_lock);
4298}
4299EXPORT_SYMBOL_GPL(napi_hash_del);
4300
4301void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4302                    int (*poll)(struct napi_struct *, int), int weight)
4303{
4304        INIT_LIST_HEAD(&napi->poll_list);
4305        napi->gro_count = 0;
4306        napi->gro_list = NULL;
4307        napi->skb = NULL;
4308        napi->poll = poll;
4309        if (weight > NAPI_POLL_WEIGHT)
4310                pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4311                            weight, dev->name);
4312        napi->weight = weight;
4313        list_add(&napi->dev_list, &dev->napi_list);
4314        napi->dev = dev;
4315#ifdef CONFIG_NETPOLL
4316        spin_lock_init(&napi->poll_lock);
4317        napi->poll_owner = -1;
4318#endif
4319        set_bit(NAPI_STATE_SCHED, &napi->state);
4320}
4321EXPORT_SYMBOL(netif_napi_add);
4322
4323void netif_napi_del(struct napi_struct *napi)
4324{
4325        list_del_init(&napi->dev_list);
4326        napi_free_frags(napi);
4327
4328        kfree_skb_list(napi->gro_list);
4329        napi->gro_list = NULL;
4330        napi->gro_count = 0;
4331}
4332EXPORT_SYMBOL(netif_napi_del);
4333
4334static void net_rx_action(struct softirq_action *h)
4335{
4336        struct softnet_data *sd = &__get_cpu_var(softnet_data);
4337        unsigned long time_limit = jiffies + 2;
4338        int budget = netdev_budget;
4339        void *have;
4340
4341        local_irq_disable();
4342
4343        while (!list_empty(&sd->poll_list)) {
4344                struct napi_struct *n;
4345                int work, weight;
4346
4347                /* If softirq window is exhuasted then punt.
4348                 * Allow this to run for 2 jiffies since which will allow
4349                 * an average latency of 1.5/HZ.
4350                 */
4351                if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4352                        goto softnet_break;
4353
4354                local_irq_enable();
4355
4356                /* Even though interrupts have been re-enabled, this
4357                 * access is safe because interrupts can only add new
4358                 * entries to the tail of this list, and only ->poll()
4359                 * calls can remove this head entry from the list.
4360                 */
4361                n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4362
4363                have = netpoll_poll_lock(n);
4364
4365                weight = n->weight;
4366
4367                /* This NAPI_STATE_SCHED test is for avoiding a race
4368                 * with netpoll's poll_napi().  Only the entity which
4369                 * obtains the lock and sees NAPI_STATE_SCHED set will
4370                 * actually make the ->poll() call.  Therefore we avoid
4371                 * accidentally calling ->poll() when NAPI is not scheduled.
4372                 */
4373                work = 0;
4374                if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4375                        work = n->poll(n, weight);
4376                        trace_napi_poll(n);
4377                }
4378
4379                WARN_ON_ONCE(work > weight);
4380
4381                budget -= work;
4382
4383                local_irq_disable();
4384
4385                /* Drivers must not modify the NAPI state if they
4386                 * consume the entire weight.  In such cases this code
4387                 * still "owns" the NAPI instance and therefore can
4388                 * move the instance around on the list at-will.
4389                 */
4390                if (unlikely(work == weight)) {
4391                        if (unlikely(napi_disable_pending(n))) {
4392                                local_irq_enable();
4393                                napi_complete(n);
4394                                local_irq_disable();
4395                        } else {
4396                                if (n->gro_list) {
4397                                        /* flush too old packets
4398                                         * If HZ < 1000, flush all packets.
4399                                         */
4400                                        local_irq_enable();
4401                                        napi_gro_flush(n, HZ >= 1000);
4402                                        local_irq_disable();
4403                                }
4404                                list_move_tail(&n->poll_list, &sd->poll_list);
4405                        }
4406                }
4407
4408                netpoll_poll_unlock(have);
4409        }
4410out:
4411        net_rps_action_and_irq_enable(sd);
4412
4413#ifdef CONFIG_NET_DMA
4414        /*
4415         * There may not be any more sk_buffs coming right now, so push
4416         * any pending DMA copies to hardware
4417         */
4418        dma_issue_pending_all();
4419#endif
4420
4421        return;
4422
4423softnet_break:
4424        sd->time_squeeze++;
4425        __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4426        goto out;
4427}
4428
4429struct netdev_adjacent {
4430        struct net_device *dev;
4431
4432        /* upper master flag, there can only be one master device per list */
4433        bool master;
4434
4435        /* counter for the number of times this device was added to us */
4436        u16 ref_nr;
4437
4438        /* private field for the users */
4439        void *private;
4440
4441        struct list_head list;
4442        struct rcu_head rcu;
4443};
4444
4445static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4446                                                 struct net_device *adj_dev,
4447                                                 struct list_head *adj_list)
4448{
4449        struct netdev_adjacent *adj;
4450
4451        list_for_each_entry(adj, adj_list, list) {
4452                if (adj->dev == adj_dev)
4453                        return adj;
4454        }
4455        return NULL;
4456}
4457
4458/**
4459 * netdev_has_upper_dev - Check if device is linked to an upper device
4460 * @dev: device
4461 * @upper_dev: upper device to check
4462 *
4463 * Find out if a device is linked to specified upper device and return true
4464 * in case it is. Note that this checks only immediate upper device,
4465 * not through a complete stack of devices. The caller must hold the RTNL lock.
4466 */
4467bool netdev_has_upper_dev(struct net_device *dev,
4468                          struct net_device *upper_dev)
4469{
4470        ASSERT_RTNL();
4471
4472        return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4473}
4474EXPORT_SYMBOL(netdev_has_upper_dev);
4475
4476/**
4477 * netdev_has_any_upper_dev - Check if device is linked to some device
4478 * @dev: device
4479 *
4480 * Find out if a device is linked to an upper device and return true in case
4481 * it is. The caller must hold the RTNL lock.
4482 */
4483static bool netdev_has_any_upper_dev(struct net_device *dev)
4484{
4485        ASSERT_RTNL();
4486
4487        return !list_empty(&dev->all_adj_list.upper);
4488}
4489
4490/**
4491 * netdev_master_upper_dev_get - Get master upper device
4492 * @dev: device
4493 *
4494 * Find a master upper device and return pointer to it or NULL in case
4495 * it's not there. The caller must hold the RTNL lock.
4496 */
4497struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4498{
4499        struct netdev_adjacent *upper;
4500
4501        ASSERT_RTNL();
4502
4503        if (list_empty(&dev->adj_list.upper))
4504                return NULL;
4505
4506        upper = list_first_entry(&dev->adj_list.upper,
4507                                 struct netdev_adjacent, list);
4508        if (likely(upper->master))
4509                return upper->dev;
4510        return NULL;
4511}
4512EXPORT_SYMBOL(netdev_master_upper_dev_get);
4513
4514void *netdev_adjacent_get_private(struct list_head *adj_list)
4515{
4516        struct netdev_adjacent *adj;
4517
4518        adj = list_entry(adj_list, struct netdev_adjacent, list);
4519
4520        return adj->private;
4521}
4522EXPORT_SYMBOL(netdev_adjacent_get_private);
4523
4524/**
4525 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4526 * @dev: device
4527 * @iter: list_head ** of the current position
4528 *
4529 * Gets the next device from the dev's upper list, starting from iter
4530 * position. The caller must hold RCU read lock.
4531 */
4532struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4533                                                     struct list_head **iter)
4534{
4535        struct netdev_adjacent *upper;
4536
4537        WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4538
4539        upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4540
4541        if (&upper->list == &dev->all_adj_list.upper)
4542                return NULL;
4543
4544        *iter = &upper->list;
4545
4546        return upper->dev;
4547}
4548EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4549
4550/**
4551 * netdev_lower_get_next_private - Get the next ->private from the
4552 *                                 lower neighbour list
4553 * @dev: device
4554 * @iter: list_head ** of the current position
4555 *
4556 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4557 * list, starting from iter position. The caller must hold either hold the
4558 * RTNL lock or its own locking that guarantees that the neighbour lower
4559 * list will remain unchainged.
4560 */
4561void *netdev_lower_get_next_private(struct net_device *dev,
4562                                    struct list_head **iter)
4563{
4564        struct netdev_adjacent *lower;
4565
4566        lower = list_entry(*iter, struct netdev_adjacent, list);
4567
4568        if (&lower->list == &dev->adj_list.lower)
4569                return NULL;
4570
4571        if (iter)
4572                *iter = lower->list.next;
4573
4574        return lower->private;
4575}
4576EXPORT_SYMBOL(netdev_lower_get_next_private);
4577
4578/**
4579 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4580 *                                     lower neighbour list, RCU
4581 *                                     variant
4582 * @dev: device
4583 * @iter: list_head ** of the current position
4584 *
4585 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4586 * list, starting from iter position. The caller must hold RCU read lock.
4587 */
4588void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4589                                        struct list_head **iter)
4590{
4591        struct netdev_adjacent *lower;
4592
4593        WARN_ON_ONCE(!rcu_read_lock_held());
4594
4595        lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4596
4597        if (&lower->list == &dev->adj_list.lower)
4598                return NULL;
4599
4600        if (iter)
4601                *iter = &lower->list;
4602
4603        return lower->private;
4604}
4605EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4606
4607/**
4608 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4609 *                                     lower neighbour list, RCU
4610 *                                     variant
4611 * @dev: device
4612 *
4613 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4614 * list. The caller must hold RCU read lock.
4615 */
4616void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4617{
4618        struct netdev_adjacent *lower;
4619
4620        lower = list_first_or_null_rcu(&dev->adj_list.lower,
4621                        struct netdev_adjacent, list);
4622        if (lower)
4623                return lower->private;
4624        return NULL;
4625}
4626EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4627
4628/**
4629 * netdev_master_upper_dev_get_rcu - Get master upper device
4630 * @dev: device
4631 *
4632 * Find a master upper device and return pointer to it or NULL in case
4633 * it's not there. The caller must hold the RCU read lock.
4634 */
4635struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4636{
4637        struct netdev_adjacent *upper;
4638
4639        upper = list_first_or_null_rcu(&dev->adj_list.upper,
4640                                       struct netdev_adjacent, list);
4641        if (upper && likely(upper->master))
4642                return upper->dev;
4643        return NULL;
4644}
4645EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4646
4647static int netdev_adjacent_sysfs_add(struct net_device *dev,
4648                              struct net_device *adj_dev,
4649                              struct list_head *dev_list)
4650{
4651        char linkname[IFNAMSIZ+7];
4652        sprintf(linkname, dev_list == &dev->adj_list.upper ?
4653                "upper_%s" : "lower_%s", adj_dev->name);
4654        return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4655                                 linkname);
4656}
4657static void netdev_adjacent_sysfs_del(struct net_device *dev,
4658                               char *name,
4659                               struct list_head *dev_list)
4660{
4661        char linkname[IFNAMSIZ+7];
4662        sprintf(linkname, dev_list == &dev->adj_list.upper ?
4663                "upper_%s" : "lower_%s", name);
4664        sysfs_remove_link(&(dev->dev.kobj), linkname);
4665}
4666
4667#define netdev_adjacent_is_neigh_list(dev, dev_list) \
4668                (dev_list == &dev->adj_list.upper || \
4669                 dev_list == &dev->adj_list.lower)
4670
4671static int __netdev_adjacent_dev_insert(struct net_device *dev,
4672                                        struct net_device *adj_dev,
4673                                        struct list_head *dev_list,
4674                                        void *private, bool master)
4675{
4676        struct netdev_adjacent *adj;
4677        int ret;
4678
4679        adj = __netdev_find_adj(dev, adj_dev, dev_list);
4680
4681        if (adj) {
4682                adj->ref_nr++;
4683                return 0;
4684        }
4685
4686        adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4687        if (!adj)
4688                return -ENOMEM;
4689
4690        adj->dev = adj_dev;
4691        adj->master = master;
4692        adj->ref_nr = 1;
4693        adj->private = private;
4694        dev_hold(adj_dev);
4695
4696        pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4697                 adj_dev->name, dev->name, adj_dev->name);
4698
4699        if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4700                ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4701                if (ret)
4702                        goto free_adj;
4703        }
4704
4705        /* Ensure that master link is always the first item in list. */
4706        if (master) {
4707                ret = sysfs_create_link(&(dev->dev.kobj),
4708                                        &(adj_dev->dev.kobj), "master");
4709                if (ret)
4710                        goto remove_symlinks;
4711
4712                list_add_rcu(&adj->list, dev_list);
4713        } else {
4714                list_add_tail_rcu(&adj->list, dev_list);
4715        }
4716
4717        return 0;
4718
4719remove_symlinks:
4720        if (netdev_adjacent_is_neigh_list(dev, dev_list))
4721                netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4722free_adj:
4723        kfree(adj);
4724        dev_put(adj_dev);
4725
4726        return ret;
4727}
4728
4729static void __netdev_adjacent_dev_remove(struct net_device *dev,
4730                                         struct net_device *adj_dev,
4731                                         struct list_head *dev_list)
4732{
4733        struct netdev_adjacent *adj;
4734
4735        adj = __netdev_find_adj(dev, adj_dev, dev_list);
4736
4737        if (!adj) {
4738                pr_err("tried to remove device %s from %s\n",
4739                       dev->name, adj_dev->name);
4740                BUG();
4741        }
4742
4743        if (adj->ref_nr > 1) {
4744                pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4745                         adj->ref_nr-1);
4746                adj->ref_nr--;
4747                return;
4748        }
4749
4750        if (adj->master)
4751                sysfs_remove_link(&(dev->dev.kobj), "master");
4752
4753        if (netdev_adjacent_is_neigh_list(dev, dev_list))
4754                netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4755
4756        list_del_rcu(&adj->list);
4757        pr_debug("dev_put for %s, because link removed from %s to %s\n",
4758                 adj_dev->name, dev->name, adj_dev->name);
4759        dev_put(adj_dev);
4760        kfree_rcu(adj, rcu);
4761}
4762
4763static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4764                                            struct net_device *upper_dev,
4765                                            struct list_head *up_list,
4766                                            struct list_head *down_list,
4767                                            void *private, bool master)
4768{
4769        int ret;
4770
4771        ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4772                                           master);
4773        if (ret)
4774                return ret;
4775
4776        ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4777                                           false);
4778        if (ret) {
4779                __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4780                return ret;
4781        }
4782
4783        return 0;
4784}
4785
4786static int __netdev_adjacent_dev_link(struct net_device *dev,
4787                                      struct net_device *upper_dev)
4788{
4789        return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4790                                                &dev->all_adj_list.upper,
4791                                                &upper_dev->all_adj_list.lower,
4792                                                NULL, false);
4793}
4794
4795static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4796                                               struct net_device *upper_dev,
4797                                               struct list_head *up_list,
4798                                               struct list_head *down_list)
4799{
4800        __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4801        __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4802}
4803
4804static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4805                                         struct net_device *upper_dev)
4806{
4807        __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4808                                           &dev->all_adj_list.upper,
4809                                           &upper_dev->all_adj_list.lower);
4810}
4811
4812static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4813                                                struct net_device *upper_dev,
4814                                                void *private, bool master)
4815{
4816        int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4817
4818        if (ret)
4819                return ret;
4820
4821        ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4822                                               &dev->adj_list.upper,
4823                                               &upper_dev->adj_list.lower,
4824                                               private, master);
4825        if (ret) {
4826                __netdev_adjacent_dev_unlink(dev, upper_dev);
4827                return ret;
4828        }
4829
4830        return 0;
4831}
4832
4833static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4834                                                   struct net_device *upper_dev)
4835{
4836        __netdev_adjacent_dev_unlink(dev, upper_dev);
4837        __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4838                                           &dev->adj_list.upper,
4839                                           &upper_dev->adj_list.lower);
4840}
4841
4842static int __netdev_upper_dev_link(struct net_device *dev,
4843                                   struct net_device *upper_dev, bool master,
4844                                   void *private)
4845{
4846        struct netdev_adjacent *i, *j, *to_i, *to_j;
4847        int ret = 0;
4848
4849        ASSERT_RTNL();
4850
4851        if (dev == upper_dev)
4852                return -EBUSY;
4853
4854        /* To prevent loops, check if dev is not upper device to upper_dev. */
4855        if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4856                return -EBUSY;
4857
4858        if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4859                return -EEXIST;
4860
4861        if (master && netdev_master_upper_dev_get(dev))
4862                return -EBUSY;
4863
4864        ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4865                                                   master);
4866        if (ret)
4867                return ret;
4868
4869        /* Now that we linked these devs, make all the upper_dev's
4870         * all_adj_list.upper visible to every dev's all_adj_list.lower an
4871         * versa, and don't forget the devices itself. All of these
4872         * links are non-neighbours.
4873         */
4874        list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4875                list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4876                        pr_debug("Interlinking %s with %s, non-neighbour\n",
4877                                 i->dev->name, j->dev->name);
4878                        ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4879                        if (ret)
4880                                goto rollback_mesh;
4881                }
4882        }
4883
4884        /* add dev to every upper_dev's upper device */
4885        list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4886                pr_debug("linking %s's upper device %s with %s\n",
4887                         upper_dev->name, i->dev->name, dev->name);
4888                ret = __netdev_adjacent_dev_link(dev, i->dev);
4889                if (ret)
4890                        goto rollback_upper_mesh;
4891        }
4892
4893        /* add upper_dev to every dev's lower device */
4894        list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4895                pr_debug("linking %s's lower device %s with %s\n", dev->name,
4896                         i->dev->name, upper_dev->name);
4897                ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4898                if (ret)
4899                        goto rollback_lower_mesh;
4900        }
4901
4902        call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4903        return 0;
4904
4905rollback_lower_mesh:
4906        to_i = i;
4907        list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4908                if (i == to_i)
4909                        break;
4910                __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4911        }
4912
4913        i = NULL;
4914
4915rollback_upper_mesh:
4916        to_i = i;
4917        list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4918                if (i == to_i)
4919                        break;
4920                __netdev_adjacent_dev_unlink(dev, i->dev);
4921        }
4922
4923        i = j = NULL;
4924
4925rollback_mesh:
4926        to_i = i;
4927        to_j = j;
4928        list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4929                list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4930                        if (i == to_i && j == to_j)
4931                                break;
4932                        __netdev_adjacent_dev_unlink(i->dev, j->dev);
4933                }
4934                if (i == to_i)
4935                        break;
4936        }
4937
4938        __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4939
4940        return ret;
4941}
4942
4943/**
4944 * netdev_upper_dev_link - Add a link to the upper device
4945 * @dev: device
4946 * @upper_dev: new upper device
4947 *
4948 * Adds a link to device which is upper to this one. The caller must hold
4949 * the RTNL lock. On a failure a negative errno code is returned.
4950 * On success the reference counts are adjusted and the function
4951 * returns zero.
4952 */
4953int netdev_upper_dev_link(struct net_device *dev,
4954                          struct net_device *upper_dev)
4955{
4956        return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4957}
4958EXPORT_SYMBOL(netdev_upper_dev_link);
4959
4960/**
4961 * netdev_master_upper_dev_link - Add a master link to the upper device
4962 * @dev: device
4963 * @upper_dev: new upper device
4964 *
4965 * Adds a link to device which is upper to this one. In this case, only
4966 * one master upper device can be linked, although other non-master devices
4967 * might be linked as well. The caller must hold the RTNL lock.
4968 * On a failure a negative errno code is returned. On success the reference
4969 * counts are adjusted and the function returns zero.
4970 */
4971int netdev_master_upper_dev_link(struct net_device *dev,
4972                                 struct net_device *upper_dev)
4973{
4974        return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4975}
4976EXPORT_SYMBOL(netdev_master_upper_dev_link);
4977
4978int netdev_master_upper_dev_link_private(struct net_device *dev,
4979                                         struct net_device *upper_dev,
4980                                         void *private)
4981{
4982        return __netdev_upper_dev_link(dev, upper_dev, true, private);
4983}
4984EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4985
4986/**
4987 * netdev_upper_dev_unlink - Removes a link to upper device
4988 * @dev: device
4989 * @upper_dev: new upper device
4990 *
4991 * Removes a link to device which is upper to this one. The caller must hold
4992 * the RTNL lock.
4993 */
4994void netdev_upper_dev_unlink(struct net_device *dev,
4995                             struct net_device *upper_dev)
4996{
4997        struct netdev_adjacent *i, *j;
4998        ASSERT_RTNL();
4999
5000        __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5001
5002        /* Here is the tricky part. We must remove all dev's lower
5003         * devices from all upper_dev's upper devices and vice
5004         * versa, to maintain the graph relationship.
5005         */
5006        list_for_each_entry(i, &dev->all_adj_list.lower, list)
5007                list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5008                        __netdev_adjacent_dev_unlink(i->dev, j->dev);
5009
5010        /* remove also the devices itself from lower/upper device
5011         * list
5012         */
5013        list_for_each_entry(i, &dev->all_adj_list.lower, list)
5014                __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5015
5016        list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5017                __netdev_adjacent_dev_unlink(dev, i->dev);
5018
5019        call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5020}
5021EXPORT_SYMBOL(netdev_upper_dev_unlink);
5022
5023void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5024{
5025        struct netdev_adjacent *iter;
5026
5027        list_for_each_entry(iter, &dev->adj_list.upper, list) {
5028                netdev_adjacent_sysfs_del(iter->dev, oldname,
5029                                          &iter->dev->adj_list.lower);
5030                netdev_adjacent_sysfs_add(iter->dev, dev,
5031                                          &iter->dev->adj_list.lower);
5032        }
5033
5034        list_for_each_entry(iter, &dev->adj_list.lower, list) {
5035                netdev_adjacent_sysfs_del(iter->dev, oldname,
5036                                          &iter->dev->adj_list.upper);
5037                netdev_adjacent_sysfs_add(iter->dev, dev,
5038                                          &iter->dev->adj_list.upper);
5039        }
5040}
5041
5042void *netdev_lower_dev_get_private(struct net_device *dev,
5043                                   struct net_device *lower_dev)
5044{
5045        struct netdev_adjacent *lower;
5046
5047        if (!lower_dev)
5048                return NULL;
5049        lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5050        if (!lower)
5051                return NULL;
5052
5053        return lower->private;
5054}
5055EXPORT_SYMBOL(netdev_lower_dev_get_private);
5056
5057static void dev_change_rx_flags(struct net_device *dev, int flags)
5058{
5059        const struct net_device_ops *ops = dev->netdev_ops;
5060
5061        if (ops->ndo_change_rx_flags)
5062                ops->ndo_change_rx_flags(dev, flags);
5063}
5064
5065static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5066{
5067        unsigned int old_flags = dev->flags;
5068        kuid_t uid;
5069        kgid_t gid;
5070
5071        ASSERT_RTNL();
5072
5073        dev->flags |= IFF_PROMISC;
5074        dev->promiscuity += inc;
5075        if (dev->promiscuity == 0) {
5076                /*
5077                 * Avoid overflow.
5078                 * If inc causes overflow, untouch promisc and return error.
5079                 */
5080                if (inc < 0)
5081                        dev->flags &= ~IFF_PROMISC;
5082                else {
5083                        dev->promiscuity -= inc;
5084                        pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5085                                dev->name);
5086                        return -EOVERFLOW;
5087                }
5088        }
5089        if (dev->flags != old_flags) {
5090                pr_info("device %s %s promiscuous mode\n",
5091                        dev->name,
5092                        dev->flags & IFF_PROMISC ? "entered" : "left");
5093                if (audit_enabled) {
5094                        current_uid_gid(&uid, &gid);
5095                        audit_log(current->audit_context, GFP_ATOMIC,
5096                                AUDIT_ANOM_PROMISCUOUS,
5097                                "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5098                                dev->name, (dev->flags & IFF_PROMISC),
5099                                (old_flags & IFF_PROMISC),
5100                                from_kuid(&init_user_ns, audit_get_loginuid(current)),
5101                                from_kuid(&init_user_ns, uid),
5102                                from_kgid(&init_user_ns, gid),
5103                                audit_get_sessionid(current));
5104                }
5105
5106                dev_change_rx_flags(dev, IFF_PROMISC);
5107        }
5108        if (notify)
5109                __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5110        return 0;
5111}
5112
5113/**
5114 *      dev_set_promiscuity     - update promiscuity count on a device
5115 *      @dev: device
5116 *      @inc: modifier
5117 *
5118 *      Add or remove promiscuity from a device. While the count in the device
5119 *      remains above zero the interface remains promiscuous. Once it hits zero
5120 *      the device reverts back to normal filtering operation. A negative inc
5121 *      value is used to drop promiscuity on the device.
5122 *      Return 0 if successful or a negative errno code on error.
5123 */
5124int dev_set_promiscuity(struct net_device *dev, int inc)
5125{
5126        unsigned int old_flags = dev->flags;
5127        int err;
5128
5129        err = __dev_set_promiscuity(dev, inc, true);
5130        if (err < 0)
5131                return err;
5132        if (dev->flags != old_flags)
5133                dev_set_rx_mode(dev);
5134        return err;
5135}
5136EXPORT_SYMBOL(dev_set_promiscuity);
5137
5138static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5139{
5140        unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5141
5142        ASSERT_RTNL();
5143
5144        dev->flags |= IFF_ALLMULTI;
5145        dev->allmulti += inc;
5146        if (dev->allmulti == 0) {
5147                /*
5148                 * Avoid overflow.
5149                 * If inc causes overflow, untouch allmulti and return error.
5150                 */
5151                if (inc < 0)
5152                        dev->flags &= ~IFF_ALLMULTI;
5153                else {
5154                        dev->allmulti -= inc;
5155                        pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5156                                dev->name);
5157                        return -EOVERFLOW;
5158                }
5159        }
5160        if (dev->flags ^ old_flags) {
5161                dev_change_rx_flags(dev, IFF_ALLMULTI);
5162                dev_set_rx_mode(dev);
5163                if (notify)
5164                        __dev_notify_flags(dev, old_flags,
5165                                           dev->gflags ^ old_gflags);
5166        }
5167        return 0;
5168}
5169
5170/**
5171 *      dev_set_allmulti        - update allmulti count on a device
5172 *      @dev: device
5173 *      @inc: modifier
5174 *
5175 *      Add or remove reception of all multicast frames to a device. While the
5176 *      count in the device remains above zero the interface remains listening
5177 *      to all interfaces. Once it hits zero the device reverts back to normal
5178 *      filtering operation. A negative @inc value is used to drop the counter
5179 *      when releasing a resource needing all multicasts.
5180 *      Return 0 if successful or a negative errno code on error.
5181 */
5182
5183int dev_set_allmulti(struct net_device *dev, int inc)
5184{
5185        return __dev_set_allmulti(dev, inc, true);
5186}
5187EXPORT_SYMBOL(dev_set_allmulti);
5188
5189/*
5190 *      Upload unicast and multicast address lists to device and
5191 *      configure RX filtering. When the device doesn't support unicast
5192 *      filtering it is put in promiscuous mode while unicast addresses
5193 *      are present.
5194 */
5195void __dev_set_rx_mode(struct net_device *dev)
5196{
5197        const struct net_device_ops *ops = dev->netdev_ops;
5198
5199        /* dev_open will call this function so the list will stay sane. */
5200        if (!(dev->flags&IFF_UP))
5201                return;
5202
5203        if (!netif_device_present(dev))
5204                return;
5205
5206        if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5207                /* Unicast addresses changes may only happen under the rtnl,
5208                 * therefore calling __dev_set_promiscuity here is safe.
5209                 */
5210                if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5211                        __dev_set_promiscuity(dev, 1, false);
5212                        dev->uc_promisc = true;
5213                } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5214                        __dev_set_promiscuity(dev, -1, false);
5215                        dev->uc_promisc = false;
5216                }
5217        }
5218
5219        if (ops->ndo_set_rx_mode)
5220                ops->ndo_set_rx_mode(dev);
5221}
5222
5223void dev_set_rx_mode(struct net_device *dev)
5224{
5225        netif_addr_lock_bh(dev);
5226        __dev_set_rx_mode(dev);
5227        netif_addr_unlock_bh(dev);
5228}
5229
5230/**
5231 *      dev_get_flags - get flags reported to userspace
5232 *      @dev: device
5233 *
5234 *      Get the combination of flag bits exported through APIs to userspace.
5235 */
5236unsigned int dev_get_flags(const struct net_device *dev)
5237{
5238        unsigned int flags;
5239
5240        flags = (dev->flags & ~(IFF_PROMISC |
5241                                IFF_ALLMULTI |
5242                                IFF_RUNNING |
5243                                IFF_LOWER_UP |
5244                                IFF_DORMANT)) |
5245                (dev->gflags & (IFF_PROMISC |
5246                                IFF_ALLMULTI));
5247
5248        if (netif_running(dev)) {
5249                if (netif_oper_up(dev))
5250                        flags |= IFF_RUNNING;
5251                if (netif_carrier_ok(dev))
5252                        flags |= IFF_LOWER_UP;
5253                if (netif_dormant(dev))
5254                        flags |= IFF_DORMANT;
5255        }
5256
5257        return flags;
5258}
5259EXPORT_SYMBOL(dev_get_flags);
5260
5261int __dev_change_flags(struct net_device *dev, unsigned int flags)
5262{
5263        unsigned int old_flags = dev->flags;
5264        int ret;
5265
5266        ASSERT_RTNL();
5267
5268        /*
5269         *      Set the flags on our device.
5270         */
5271
5272        dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5273                               IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5274                               IFF_AUTOMEDIA)) |
5275                     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5276                                    IFF_ALLMULTI));
5277
5278        /*
5279         *      Load in the correct multicast list now the flags have changed.
5280         */
5281
5282        if ((old_flags ^ flags) & IFF_MULTICAST)
5283                dev_change_rx_flags(dev, IFF_MULTICAST);
5284
5285        dev_set_rx_mode(dev);
5286
5287        /*
5288         *      Have we downed the interface. We handle IFF_UP ourselves
5289         *      according to user attempts to set it, rather than blindly
5290         *      setting it.
5291         */
5292
5293        ret = 0;
5294        if ((old_flags ^ flags) & IFF_UP) {     /* Bit is different  ? */
5295                ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5296
5297                if (!ret)
5298                        dev_set_rx_mode(dev);
5299        }
5300
5301        if ((flags ^ dev->gflags) & IFF_PROMISC) {
5302                int inc = (flags & IFF_PROMISC) ? 1 : -1;
5303                unsigned int old_flags = dev->flags;
5304
5305                dev->gflags ^= IFF_PROMISC;
5306
5307                if (__dev_set_promiscuity(dev, inc, false) >= 0)
5308                        if (dev->flags != old_flags)
5309                                dev_set_rx_mode(dev);
5310        }
5311
5312        /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5313           is important. Some (broken) drivers set IFF_PROMISC, when
5314           IFF_ALLMULTI is requested not asking us and not reporting.
5315         */
5316        if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5317                int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5318
5319                dev->gflags ^= IFF_ALLMULTI;
5320                __dev_set_allmulti(dev, inc, false);
5321        }
5322
5323        return ret;
5324}
5325
5326void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5327                        unsigned int gchanges)
5328{
5329        unsigned int changes = dev->flags ^ old_flags;
5330
5331        if (gchanges)
5332                rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5333
5334        if (changes & IFF_UP) {
5335                if (dev->flags & IFF_UP)
5336                        call_netdevice_notifiers(NETDEV_UP, dev);
5337                else
5338                        call_netdevice_notifiers(NETDEV_DOWN, dev);
5339        }
5340
5341        if (dev->flags & IFF_UP &&
5342            (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5343                struct netdev_notifier_change_info change_info;
5344
5345                change_info.flags_changed = changes;
5346                call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5347                                              &change_info.info);
5348        }
5349}
5350
5351/**
5352 *      dev_change_flags - change device settings
5353 *      @dev: device
5354 *      @flags: device state flags
5355 *
5356 *      Change settings on device based state flags. The flags are
5357 *      in the userspace exported format.
5358 */
5359int dev_change_flags(struct net_device *dev, unsigned int flags)
5360{
5361        int ret;
5362        unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5363
5364        ret = __dev_change_flags(dev, flags);
5365        if (ret < 0)
5366                return ret;
5367
5368        changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5369        __dev_notify_flags(dev, old_flags, changes);
5370        return ret;
5371}
5372EXPORT_SYMBOL(dev_change_flags);
5373
5374static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5375{
5376        const struct net_device_ops *ops = dev->netdev_ops;
5377
5378        if (ops->ndo_change_mtu)
5379                return ops->ndo_change_mtu(dev, new_mtu);
5380
5381        dev->mtu = new_mtu;
5382        return 0;
5383}
5384
5385/**
5386 *      dev_set_mtu - Change maximum transfer unit
5387 *      @dev: device
5388 *      @new_mtu: new transfer unit
5389 *
5390 *      Change the maximum transfer size of the network device.
5391 */
5392int dev_set_mtu(struct net_device *dev, int new_mtu)
5393{
5394        int err, orig_mtu;
5395
5396        if (new_mtu == dev->mtu)
5397                return 0;
5398
5399        /*      MTU must be positive.    */
5400        if (new_mtu < 0)
5401                return -EINVAL;
5402
5403        if (!netif_device_present(dev))
5404                return -ENODEV;
5405
5406        err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5407        err = notifier_to_errno(err);
5408        if (err)
5409                return err;
5410
5411        orig_mtu = dev->mtu;
5412        err = __dev_set_mtu(dev, new_mtu);
5413
5414        if (!err) {
5415                err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5416                err = notifier_to_errno(err);
5417                if (err) {
5418                        /* setting mtu back and notifying everyone again,
5419                         * so that they have a chance to revert changes.
5420                         */
5421                        __dev_set_mtu(dev, orig_mtu);
5422                        call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5423                }
5424        }
5425        return err;
5426}
5427EXPORT_SYMBOL(dev_set_mtu);
5428
5429/**
5430 *      dev_set_group - Change group this device belongs to
5431 *      @dev: device
5432 *      @new_group: group this device should belong to
5433 */
5434void dev_set_group(struct net_device *dev, int new_group)
5435{
5436        dev->group = new_group;
5437}
5438EXPORT_SYMBOL(dev_set_group);
5439
5440/**
5441 *      dev_set_mac_address - Change Media Access Control Address
5442 *      @dev: device
5443 *      @sa: new address
5444 *
5445 *      Change the hardware (MAC) address of the device
5446 */
5447int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5448{
5449        const struct net_device_ops *ops = dev->netdev_ops;
5450        int err;
5451
5452        if (!ops->ndo_set_mac_address)
5453                return -EOPNOTSUPP;
5454        if (sa->sa_family != dev->type)
5455                return -EINVAL;
5456        if (!netif_device_present(dev))
5457                return -ENODEV;
5458        err = ops->ndo_set_mac_address(dev, sa);
5459        if (err)
5460                return err;
5461        dev->addr_assign_type = NET_ADDR_SET;
5462        call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5463        add_device_randomness(dev->dev_addr, dev->addr_len);
5464        return 0;
5465}
5466EXPORT_SYMBOL(dev_set_mac_address);
5467
5468/**
5469 *      dev_change_carrier - Change device carrier
5470 *      @dev: device
5471 *      @new_carrier: new value
5472 *
5473 *      Change device carrier
5474 */
5475int dev_change_carrier(struct net_device *dev, bool new_carrier)
5476{
5477        const struct net_device_ops *ops = dev->netdev_ops;
5478
5479        if (!ops->ndo_change_carrier)
5480                return -EOPNOTSUPP;
5481        if (!netif_device_present(dev))
5482                return -ENODEV;
5483        return ops->ndo_change_carrier(dev, new_carrier);
5484}
5485EXPORT_SYMBOL(dev_change_carrier);
5486
5487/**
5488 *      dev_get_phys_port_id - Get device physical port ID
5489 *      @dev: device
5490 *      @ppid: port ID
5491 *
5492 *      Get device physical port ID
5493 */
5494int dev_get_phys_port_id(struct net_device *dev,
5495                         struct netdev_phys_port_id *ppid)
5496{
5497        const struct net_device_ops *ops = dev->netdev_ops;
5498
5499        if (!ops->ndo_get_phys_port_id)
5500                return -EOPNOTSUPP;
5501        return ops->ndo_get_phys_port_id(dev, ppid);
5502}
5503EXPORT_SYMBOL(dev_get_phys_port_id);
5504
5505/**
5506 *      dev_new_index   -       allocate an ifindex
5507 *      @net: the applicable net namespace
5508 *
5509 *      Returns a suitable unique value for a new device interface
5510 *      number.  The caller must hold the rtnl semaphore or the
5511 *      dev_base_lock to be sure it remains unique.
5512 */
5513static int dev_new_index(struct net *net)
5514{
5515        int ifindex = net->ifindex;
5516        for (;;) {
5517                if (++ifindex <= 0)
5518                        ifindex = 1;
5519                if (!__dev_get_by_index(net, ifindex))
5520                        return net->ifindex = ifindex;
5521        }
5522}
5523
5524/* Delayed registration/unregisteration */
5525static LIST_HEAD(net_todo_list);
5526static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5527
5528static void net_set_todo(struct net_device *dev)
5529{
5530        list_add_tail(&dev->todo_list, &net_todo_list);
5531        dev_net(dev)->dev_unreg_count++;
5532}
5533
5534static void rollback_registered_many(struct list_head *head)
5535{
5536        struct net_device *dev, *tmp;
5537        LIST_HEAD(close_head);
5538
5539        BUG_ON(dev_boot_phase);
5540        ASSERT_RTNL();
5541
5542        list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5543                /* Some devices call without registering
5544                 * for initialization unwind. Remove those
5545                 * devices and proceed with the remaining.
5546                 */
5547                if (dev->reg_state == NETREG_UNINITIALIZED) {
5548                        pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5549                                 dev->name, dev);
5550
5551                        WARN_ON(1);
5552                        list_del(&dev->unreg_list);
5553                        continue;
5554                }
5555                dev->dismantle = true;
5556                BUG_ON(dev->reg_state != NETREG_REGISTERED);
5557        }
5558
5559        /* If device is running, close it first. */
5560        list_for_each_entry(dev, head, unreg_list)
5561                list_add_tail(&dev->close_list, &close_head);
5562        dev_close_many(&close_head);
5563
5564        list_for_each_entry(dev, head, unreg_list) {
5565                /* And unlink it from device chain. */
5566                unlist_netdevice(dev);
5567
5568                dev->reg_state = NETREG_UNREGISTERING;
5569        }
5570
5571        synchronize_net();
5572
5573        list_for_each_entry(dev, head, unreg_list) {
5574                /* Shutdown queueing discipline. */
5575                dev_shutdown(dev);
5576
5577
5578                /* Notify protocols, that we are about to destroy
5579                   this device. They should clean all the things.
5580                */
5581                call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5582
5583                if (!dev->rtnl_link_ops ||
5584                    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5585                        rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5586
5587                /*
5588                 *      Flush the unicast and multicast chains
5589                 */
5590                dev_uc_flush(dev);
5591                dev_mc_flush(dev);
5592
5593                if (dev->netdev_ops->ndo_uninit)
5594                        dev->netdev_ops->ndo_uninit(dev);
5595
5596                /* Notifier chain MUST detach us all upper devices. */
5597                WARN_ON(netdev_has_any_upper_dev(dev));
5598
5599                /* Remove entries from kobject tree */
5600                netdev_unregister_kobject(dev);
5601#ifdef CONFIG_XPS
5602                /* Remove XPS queueing entries */
5603                netif_reset_xps_queues_gt(dev, 0);
5604#endif
5605        }
5606
5607        synchronize_net();
5608
5609        list_for_each_entry(dev, head, unreg_list)
5610                dev_put(dev);
5611}
5612
5613static void rollback_registered(struct net_device *dev)
5614{
5615        LIST_HEAD(single);
5616
5617        list_add(&dev->unreg_list, &single);
5618        rollback_registered_many(&single);
5619        list_del(&single);
5620}
5621
5622static netdev_features_t netdev_fix_features(struct net_device *dev,
5623        netdev_features_t features)
5624{
5625        /* Fix illegal checksum combinations */
5626        if ((features & NETIF_F_HW_CSUM) &&
5627            (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5628                netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5629                features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5630        }
5631
5632        /* TSO requires that SG is present as well. */
5633        if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5634                netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5635                features &= ~NETIF_F_ALL_TSO;
5636        }
5637
5638        if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5639                                        !(features & NETIF_F_IP_CSUM)) {
5640                netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5641                features &= ~NETIF_F_TSO;
5642                features &= ~NETIF_F_TSO_ECN;
5643        }
5644
5645        if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5646                                         !(features & NETIF_F_IPV6_CSUM)) {
5647                netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5648                features &= ~NETIF_F_TSO6;
5649        }
5650
5651        /* TSO ECN requires that TSO is present as well. */
5652        if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5653                features &= ~NETIF_F_TSO_ECN;
5654
5655        /* Software GSO depends on SG. */
5656        if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5657                netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5658                features &= ~NETIF_F_GSO;
5659        }
5660
5661        /* UFO needs SG and checksumming */
5662        if (features & NETIF_F_UFO) {
5663                /* maybe split UFO into V4 and V6? */
5664                if (!((features & NETIF_F_GEN_CSUM) ||
5665                    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5666                            == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5667                        netdev_dbg(dev,
5668                                "Dropping NETIF_F_UFO since no checksum offload features.\n");
5669                        features &= ~NETIF_F_UFO;
5670                }
5671
5672                if (!(features & NETIF_F_SG)) {
5673                        netdev_dbg(dev,
5674                                "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5675                        features &= ~NETIF_F_UFO;
5676                }
5677        }
5678
5679        return features;
5680}
5681
5682int __netdev_update_features(struct net_device *dev)
5683{
5684        netdev_features_t features;
5685        int err = 0;
5686
5687        ASSERT_RTNL();
5688
5689        features = netdev_get_wanted_features(dev);
5690
5691        if (dev->netdev_ops->ndo_fix_features)
5692                features = dev->netdev_ops->ndo_fix_features(dev, features);
5693
5694        /* driver might be less strict about feature dependencies */
5695        features = netdev_fix_features(dev, features);
5696
5697        if (dev->features == features)
5698                return 0;
5699
5700        netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5701                &dev->features, &features);
5702
5703        if (dev->netdev_ops->ndo_set_features)
5704                err = dev->netdev_ops->ndo_set_features(dev, features);
5705
5706        if (unlikely(err < 0)) {
5707                netdev_err(dev,
5708                        "set_features() failed (%d); wanted %pNF, left %pNF\n",
5709                        err, &features, &dev->features);
5710                return -1;
5711        }
5712
5713        if (!err)
5714                dev->features = features;
5715
5716        return 1;
5717}
5718
5719/**
5720 *      netdev_update_features - recalculate device features
5721 *      @dev: the device to check
5722 *
5723 *      Recalculate dev->features set and send notifications if it
5724 *      has changed. Should be called after driver or hardware dependent
5725 *      conditions might have changed that influence the features.
5726 */
5727void netdev_update_features(struct net_device *dev)
5728{
5729        if (__netdev_update_features(dev))
5730                netdev_features_change(dev);
5731}
5732EXPORT_SYMBOL(netdev_update_features);
5733
5734/**
5735 *      netdev_change_features - recalculate device features
5736 *      @dev: the device to check
5737 *
5738 *      Recalculate dev->features set and send notifications even
5739 *      if they have not changed. Should be called instead of
5740 *      netdev_update_features() if also dev->vlan_features might
5741 *      have changed to allow the changes to be propagated to stacked
5742 *      VLAN devices.
5743 */
5744void netdev_change_features(struct net_device *dev)
5745{
5746        __netdev_update_features(dev);
5747        netdev_features_change(dev);
5748}
5749EXPORT_SYMBOL(netdev_change_features);
5750
5751/**
5752 *      netif_stacked_transfer_operstate -      transfer operstate
5753 *      @rootdev: the root or lower level device to transfer state from
5754 *      @dev: the device to transfer operstate to
5755 *
5756 *      Transfer operational state from root to device. This is normally
5757 *      called when a stacking relationship exists between the root
5758 *      device and the device(a leaf device).
5759 */
5760void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5761                                        struct net_device *dev)
5762{
5763        if (rootdev->operstate == IF_OPER_DORMANT)
5764                netif_dormant_on(dev);
5765        else
5766                netif_dormant_off(dev);
5767
5768        if (netif_carrier_ok(rootdev)) {
5769                if (!netif_carrier_ok(dev))
5770                        netif_carrier_on(dev);
5771        } else {
5772                if (netif_carrier_ok(dev))
5773                        netif_carrier_off(dev);
5774        }
5775}
5776EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5777
5778#ifdef CONFIG_SYSFS
5779static int netif_alloc_rx_queues(struct net_device *dev)
5780{
5781        unsigned int i, count = dev->num_rx_queues;
5782        struct netdev_rx_queue *rx;
5783
5784        BUG_ON(count < 1);
5785
5786        rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5787        if (!rx)
5788                return -ENOMEM;
5789
5790        dev->_rx = rx;
5791
5792        for (i = 0; i < count; i++)
5793                rx[i].dev = dev;