linux-bk/net/core/dev.c History
<<
>>
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, <bir7@leland.Stanford.Edu>
  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 <asm/system.h>
  77#include <asm/bitops.h>
  78#include <linux/config.h>
  79#include <linux/cpu.h>
  80#include <linux/types.h>
  81#include <linux/kernel.h>
  82#include <linux/sched.h>
  83#include <linux/string.h>
  84#include <linux/mm.h>
  85#include <linux/socket.h>
  86#include <linux/sockios.h>
  87#include <linux/errno.h>
  88#include <linux/interrupt.h>
  89#include <linux/if_ether.h>
  90#include <linux/netdevice.h>
  91#include <linux/etherdevice.h>
  92#include <linux/notifier.h>
  93#include <linux/skbuff.h>
  94#include <net/sock.h>
  95#include <linux/rtnetlink.h>
  96#include <linux/proc_fs.h>
  97#include <linux/seq_file.h>
  98#include <linux/stat.h>
  99#include <linux/if_bridge.h>
 100#include <linux/divert.h>
 101#include <net/dst.h>
 102#include <net/pkt_sched.h>
 103#include <net/checksum.h>
 104#include <linux/highmem.h>
 105#include <linux/init.h>
 106#include <linux/kmod.h>
 107#include <linux/module.h>
 108#include <linux/kallsyms.h>
 109#include <linux/netpoll.h>
 110#include <linux/rcupdate.h>
 111#ifdef CONFIG_NET_RADIO
 112#include <linux/wireless.h>             /* Note : will define WIRELESS_EXT */
 113#include <net/iw_handler.h>
 114#endif  /* CONFIG_NET_RADIO */
 115#include <asm/current.h>
 116
 117/* This define, if set, will randomly drop a packet when congestion
 118 * is more than moderate.  It helps fairness in the multi-interface
 119 * case when one of them is a hog, but it kills performance for the
 120 * single interface case so it is off now by default.
 121 */
 122#undef RAND_LIE
 123
 124/* Setting this will sample the queue lengths and thus congestion
 125 * via a timer instead of as each packet is received.
 126 */
 127#undef OFFLINE_SAMPLE
 128
 129/*
 130 *      The list of packet types we will receive (as opposed to discard)
 131 *      and the routines to invoke.
 132 *
 133 *      Why 16. Because with 16 the only overlap we get on a hash of the
 134 *      low nibble of the protocol value is RARP/SNAP/X.25.
 135 *
 136 *      NOTE:  That is no longer true with the addition of VLAN tags.  Not
 137 *             sure which should go first, but I bet it won't make much
 138 *             difference if we are running VLANs.  The good news is that
 139 *             this protocol won't be in the list unless compiled in, so
 140 *             the average user (w/out VLANs) will not be adversly affected.
 141 *             --BLG
 142 *
 143 *              0800    IP
 144 *              8100    802.1Q VLAN
 145 *              0001    802.3
 146 *              0002    AX.25
 147 *              0004    802.2
 148 *              8035    RARP
 149 *              0005    SNAP
 150 *              0805    X.25
 151 *              0806    ARP
 152 *              8137    IPX
 153 *              0009    Localtalk
 154 *              86DD    IPv6
 155 */
 156
 157static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
 158static struct list_head ptype_base[16]; /* 16 way hashed list */
 159static struct list_head ptype_all;              /* Taps */
 160
 161#ifdef OFFLINE_SAMPLE
 162static void sample_queue(unsigned long dummy);
 163static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
 164#endif
 165
 166/*
 167 * The @dev_base list is protected by @dev_base_lock and the rtln
 168 * semaphore.
 169 *
 170 * Pure readers hold dev_base_lock for reading.
 171 *
 172 * Writers must hold the rtnl semaphore while they loop through the
 173 * dev_base list, and hold dev_base_lock for writing when they do the
 174 * actual updates.  This allows pure readers to access the list even
 175 * while a writer is preparing to update it.
 176 *
 177 * To put it another way, dev_base_lock is held for writing only to
 178 * protect against pure readers; the rtnl semaphore provides the
 179 * protection against other writers.
 180 *
 181 * See, for example usages, register_netdevice() and
 182 * unregister_netdevice(), which must be called with the rtnl
 183 * semaphore held.
 184 */
 185struct net_device *dev_base;
 186struct net_device **dev_tail = &dev_base;
 187rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
 188
 189EXPORT_SYMBOL(dev_base);
 190EXPORT_SYMBOL(dev_base_lock);
 191
 192#define NETDEV_HASHBITS 8
 193static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
 194static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
 195
 196static inline struct hlist_head *dev_name_hash(const char *name)
 197{
 198        unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
 199        return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
 200}
 201
 202static inline struct hlist_head *dev_index_hash(int ifindex)
 203{
 204        return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
 205}
 206
 207/*
 208 *      Our notifier list
 209 */
 210
 211static struct notifier_block *netdev_chain;
 212
 213/*
 214 *      Device drivers call our routines to queue packets here. We empty the
 215 *      queue in the local softnet handler.
 216 */
 217DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
 218
 219#ifdef CONFIG_SYSFS
 220extern int netdev_sysfs_init(void);
 221extern int netdev_register_sysfs(struct net_device *);
 222extern void netdev_unregister_sysfs(struct net_device *);
 223#else
 224#define netdev_sysfs_init()             (0)
 225#define netdev_register_sysfs(dev)      (0)
 226#define netdev_unregister_sysfs(dev)    do { } while(0)
 227#endif
 228
 229
 230/*******************************************************************************
 231
 232                Protocol management and registration routines
 233
 234*******************************************************************************/
 235
 236/*
 237 *      For efficiency
 238 */
 239
 240int netdev_nit;
 241
 242/*
 243 *      Add a protocol ID to the list. Now that the input handler is
 244 *      smarter we can dispense with all the messy stuff that used to be
 245 *      here.
 246 *
 247 *      BEWARE!!! Protocol handlers, mangling input packets,
 248 *      MUST BE last in hash buckets and checking protocol handlers
 249 *      MUST start from promiscuous ptype_all chain in net_bh.
 250 *      It is true now, do not change it.
 251 *      Explanation follows: if protocol handler, mangling packet, will
 252 *      be the first on list, it is not able to sense, that packet
 253 *      is cloned and should be copied-on-write, so that it will
 254 *      change it and subsequent readers will get broken packet.
 255 *                                                      --ANK (980803)
 256 */
 257
 258/**
 259 *      dev_add_pack - add packet handler
 260 *      @pt: packet type declaration
 261 *
 262 *      Add a protocol handler to the networking stack. The passed &packet_type
 263 *      is linked into kernel lists and may not be freed until it has been
 264 *      removed from the kernel lists.
 265 *
 266 *      This call does not sleep therefore it can not 
 267 *      guarantee all CPU's that are in middle of receiving packets
 268 *      will see the new packet type (until the next received packet).
 269 */
 270
 271void dev_add_pack(struct packet_type *pt)
 272{
 273        int hash;
 274
 275        spin_lock_bh(&ptype_lock);
 276        if (pt->type == htons(ETH_P_ALL)) {
 277                netdev_nit++;
 278                list_add_rcu(&pt->list, &ptype_all);
 279        } else {
 280                hash = ntohs(pt->type) & 15;
 281                list_add_rcu(&pt->list, &ptype_base[hash]);
 282        }
 283        spin_unlock_bh(&ptype_lock);
 284}
 285
 286extern void linkwatch_run_queue(void);
 287
 288
 289
 290/**
 291 *      __dev_remove_pack        - remove packet handler
 292 *      @pt: packet type declaration
 293 *
 294 *      Remove a protocol handler that was previously added to the kernel
 295 *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 296 *      from the kernel lists and can be freed or reused once this function
 297 *      returns. 
 298 *
 299 *      The packet type might still be in use by receivers
 300 *      and must not be freed until after all the CPU's have gone
 301 *      through a quiescent state.
 302 */
 303void __dev_remove_pack(struct packet_type *pt)
 304{
 305        struct list_head *head;
 306        struct packet_type *pt1;
 307
 308        spin_lock_bh(&ptype_lock);
 309
 310        if (pt->type == htons(ETH_P_ALL)) {
 311                netdev_nit--;
 312                head = &ptype_all;
 313        } else
 314                head = &ptype_base[ntohs(pt->type) & 15];
 315
 316        list_for_each_entry(pt1, head, list) {
 317                if (pt == pt1) {
 318                        list_del_rcu(&pt->list);
 319                        goto out;
 320                }
 321        }
 322
 323        printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
 324out:
 325        spin_unlock_bh(&ptype_lock);
 326}
 327/**
 328 *      dev_remove_pack  - remove packet handler
 329 *      @pt: packet type declaration
 330 *
 331 *      Remove a protocol handler that was previously added to the kernel
 332 *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 333 *      from the kernel lists and can be freed or reused once this function
 334 *      returns.
 335 *
 336 *      This call sleeps to guarantee that no CPU is looking at the packet
 337 *      type after return.
 338 */
 339void dev_remove_pack(struct packet_type *pt)
 340{
 341        __dev_remove_pack(pt);
 342        
 343        synchronize_net();
 344}
 345
 346/******************************************************************************
 347
 348                      Device Boot-time Settings Routines
 349
 350*******************************************************************************/
 351
 352/* Boot time configuration table */
 353static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
 354
 355/**
 356 *      netdev_boot_setup_add   - add new setup entry
 357 *      @name: name of the device
 358 *      @map: configured settings for the device
 359 *
 360 *      Adds new setup entry to the dev_boot_setup list.  The function
 361 *      returns 0 on error and 1 on success.  This is a generic routine to
 362 *      all netdevices.
 363 */
 364int netdev_boot_setup_add(char *name, struct ifmap *map)
 365{
 366        struct netdev_boot_setup *s;
 367        int i;
 368
 369        s = dev_boot_setup;
 370        for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 371                if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
 372                        memset(s[i].name, 0, sizeof(s[i].name));
 373                        strcpy(s[i].name, name);
 374                        memcpy(&s[i].map, map, sizeof(s[i].map));
 375                        break;
 376                }
 377        }
 378
 379        return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
 380}
 381
 382/**
 383 *      netdev_boot_setup_check - check boot time settings
 384 *      @dev: the netdevice
 385 *
 386 *      Check boot time settings for the device.
 387 *      The found settings are set for the device to be used
 388 *      later in the device probing.
 389 *      Returns 0 if no settings found, 1 if they are.
 390 */
 391int netdev_boot_setup_check(struct net_device *dev)
 392{
 393        struct netdev_boot_setup *s = dev_boot_setup;
 394        int i;
 395
 396        for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 397                if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
 398                    !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
 399                        dev->irq        = s[i].map.irq;
 400                        dev->base_addr  = s[i].map.base_addr;
 401                        dev->mem_start  = s[i].map.mem_start;
 402                        dev->mem_end    = s[i].map.mem_end;
 403                        return 1;
 404                }
 405        }
 406        return 0;
 407}
 408
 409
 410/**
 411 *      netdev_boot_base        - get address from boot time settings
 412 *      @prefix: prefix for network device
 413 *      @unit: id for network device
 414 *
 415 *      Check boot time settings for the base address of device.
 416 *      The found settings are set for the device to be used
 417 *      later in the device probing.
 418 *      Returns 0 if no settings found.
 419 */
 420unsigned long netdev_boot_base(const char *prefix, int unit)
 421{
 422        const struct netdev_boot_setup *s = dev_boot_setup;
 423        char name[IFNAMSIZ];
 424        int i;
 425
 426        sprintf(name, "%s%d", prefix, unit);
 427
 428        /*
 429         * If device already registered then return base of 1
 430         * to indicate not to probe for this interface
 431         */
 432        if (__dev_get_by_name(name))
 433                return 1;
 434
 435        for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
 436                if (!strcmp(name, s[i].name))
 437                        return s[i].map.base_addr;
 438        return 0;
 439}
 440
 441/*
 442 * Saves at boot time configured settings for any netdevice.
 443 */
 444int __init netdev_boot_setup(char *str)
 445{
 446        int ints[5];
 447        struct ifmap map;
 448
 449        str = get_options(str, ARRAY_SIZE(ints), ints);
 450        if (!str || !*str)
 451                return 0;
 452
 453        /* Save settings */
 454        memset(&map, 0, sizeof(map));
 455        if (ints[0] > 0)
 456                map.irq = ints[1];
 457        if (ints[0] > 1)
 458                map.base_addr = ints[2];
 459        if (ints[0] > 2)
 460                map.mem_start = ints[3];
 461        if (ints[0] > 3)
 462                map.mem_end = ints[4];
 463
 464        /* Add new entry to the list */
 465        return netdev_boot_setup_add(str, &map);
 466}
 467
 468__setup("netdev=", netdev_boot_setup);
 469
 470/*******************************************************************************
 471
 472                            Device Interface Subroutines
 473
 474*******************************************************************************/
 475
 476/**
 477 *      __dev_get_by_name       - find a device by its name
 478 *      @name: name to find
 479 *
 480 *      Find an interface by name. Must be called under RTNL semaphore
 481 *      or @dev_base_lock. If the name is found a pointer to the device
 482 *      is returned. If the name is not found then %NULL is returned. The
 483 *      reference counters are not incremented so the caller must be
 484 *      careful with locks.
 485 */
 486
 487struct net_device *__dev_get_by_name(const char *name)
 488{
 489        struct hlist_node *p;
 490
 491        hlist_for_each(p, dev_name_hash(name)) {
 492                struct net_device *dev
 493                        = hlist_entry(p, struct net_device, name_hlist);
 494                if (!strncmp(dev->name, name, IFNAMSIZ))
 495                        return dev;
 496        }
 497        return NULL;
 498}
 499
 500/**
 501 *      dev_get_by_name         - find a device by its name
 502 *      @name: name to find
 503 *
 504 *      Find an interface by name. This can be called from any
 505 *      context and does its own locking. The returned handle has
 506 *      the usage count incremented and the caller must use dev_put() to
 507 *      release it when it is no longer needed. %NULL is returned if no
 508 *      matching device is found.
 509 */
 510
 511struct net_device *dev_get_by_name(const char *name)
 512{
 513        struct net_device *dev;
 514
 515        read_lock(&dev_base_lock);
 516        dev = __dev_get_by_name(name);
 517        if (dev)
 518                dev_hold(dev);
 519        read_unlock(&dev_base_lock);
 520        return dev;
 521}
 522
 523/*
 524   Return value is changed to int to prevent illegal usage in future.
 525   It is still legal to use to check for device existence.
 526
 527   User should understand, that the result returned by this function
 528   is meaningless, if it was not issued under rtnl semaphore.
 529 */
 530
 531/**
 532 *      dev_get -       test if a device exists
 533 *      @name:  name to test for
 534 *
 535 *      Test if a name exists. Returns true if the name is found. In order
 536 *      to be sure the name is not allocated or removed during the test the
 537 *      caller must hold the rtnl semaphore.
 538 *
 539 *      This function exists only for back compatibility with older
 540 *      drivers.
 541 */
 542int __dev_get(const char *name)
 543{
 544        struct net_device *dev;
 545
 546        read_lock(&dev_base_lock);
 547        dev = __dev_get_by_name(name);
 548        read_unlock(&dev_base_lock);
 549        return dev != NULL;
 550}
 551
 552/**
 553 *      __dev_get_by_index - find a device by its ifindex
 554 *      @ifindex: index of device
 555 *
 556 *      Search for an interface by index. Returns %NULL if the device
 557 *      is not found or a pointer to the device. The device has not
 558 *      had its reference counter increased so the caller must be careful
 559 *      about locking. The caller must hold either the RTNL semaphore
 560 *      or @dev_base_lock.
 561 */
 562
 563struct net_device *__dev_get_by_index(int ifindex)
 564{
 565        struct hlist_node *p;
 566
 567        hlist_for_each(p, dev_index_hash(ifindex)) {
 568                struct net_device *dev
 569                        = hlist_entry(p, struct net_device, index_hlist);
 570                if (dev->ifindex == ifindex)
 571                        return dev;
 572        }
 573        return NULL;
 574}
 575
 576
 577/**
 578 *      dev_get_by_index - find a device by its ifindex
 579 *      @ifindex: index of device
 580 *
 581 *      Search for an interface by index. Returns NULL if the device
 582 *      is not found or a pointer to the device. The device returned has
 583 *      had a reference added and the pointer is safe until the user calls
 584 *      dev_put to indicate they have finished with it.
 585 */
 586
 587struct net_device *dev_get_by_index(int ifindex)
 588{
 589        struct net_device *dev;
 590
 591        read_lock(&dev_base_lock);
 592        dev = __dev_get_by_index(ifindex);
 593        if (dev)
 594                dev_hold(dev);
 595        read_unlock(&dev_base_lock);
 596        return dev;
 597}
 598
 599/**
 600 *      dev_getbyhwaddr - find a device by its hardware address
 601 *      @type: media type of device
 602 *      @ha: hardware address
 603 *
 604 *      Search for an interface by MAC address. Returns NULL if the device
 605 *      is not found or a pointer to the device. The caller must hold the
 606 *      rtnl semaphore. The returned device has not had its ref count increased
 607 *      and the caller must therefore be careful about locking
 608 *
 609 *      BUGS:
 610 *      If the API was consistent this would be __dev_get_by_hwaddr
 611 */
 612
 613struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
 614{
 615        struct net_device *dev;
 616
 617        ASSERT_RTNL();
 618
 619        for (dev = dev_base; dev; dev = dev->next)
 620                if (dev->type == type &&
 621                    !memcmp(dev->dev_addr, ha, dev->addr_len))
 622                        break;
 623        return dev;
 624}
 625
 626struct net_device *__dev_getfirstbyhwtype(unsigned short type)
 627{
 628        struct net_device *dev;
 629
 630        for (dev = dev_base; dev; dev = dev->next)
 631                if (dev->type == type)
 632                        break;
 633        return dev;
 634}
 635
 636EXPORT_SYMBOL(__dev_getfirstbyhwtype);
 637
 638struct net_device *dev_getfirstbyhwtype(unsigned short type)
 639{
 640        struct net_device *dev;
 641
 642        rtnl_lock();
 643        dev = __dev_getfirstbyhwtype(type);
 644        if (dev)
 645                dev_hold(dev);
 646        rtnl_unlock();
 647        return dev;
 648}
 649
 650EXPORT_SYMBOL(dev_getfirstbyhwtype);
 651
 652/**
 653 *      dev_get_by_flags - find any device with given flags
 654 *      @if_flags: IFF_* values
 655 *      @mask: bitmask of bits in if_flags to check
 656 *
 657 *      Search for any interface with the given flags. Returns NULL if a device
 658 *      is not found or a pointer to the device. The device returned has 
 659 *      had a reference added and the pointer is safe until the user calls
 660 *      dev_put to indicate they have finished with it.
 661 */
 662
 663struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
 664{
 665        struct net_device *dev;
 666
 667        read_lock(&dev_base_lock);
 668        dev = __dev_get_by_flags(if_flags, mask);
 669        if (dev)
 670                dev_hold(dev);
 671        read_unlock(&dev_base_lock);
 672        return dev;
 673}
 674
 675/**
 676 *      __dev_get_by_flags - find any device with given flags
 677 *      @if_flags: IFF_* values
 678 *      @mask: bitmask of bits in if_flags to check
 679 *
 680 *      Search for any interface with the given flags. Returns NULL if a device
 681 *      is not found or a pointer to the device. The caller must hold either
 682 *      the RTNL semaphore or @dev_base_lock.
 683 */
 684
 685struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
 686{
 687        struct net_device *dev;
 688
 689        for (dev = dev_base; dev != NULL; dev = dev->next) {
 690                if (((dev->flags ^ if_flags) & mask) == 0)
 691                        return dev;
 692        }
 693        return NULL;
 694}
 695
 696/**
 697 *      dev_valid_name - check if name is okay for network device
 698 *      @name: name string
 699 *
 700 *      Network device names need to be valid file names to
 701 *      to allow sysfs to work
 702 */
 703int dev_valid_name(const char *name)
 704{
 705        return !(*name == '\0' 
 706                 || !strcmp(name, ".")
 707                 || !strcmp(name, "..")
 708                 || strchr(name, '/'));
 709}
 710
 711/**
 712 *      dev_alloc_name - allocate a name for a device
 713 *      @dev: device
 714 *      @name: name format string
 715 *
 716 *      Passed a format string - eg "lt%d" it will try and find a suitable
 717 *      id. Not efficient for many devices, not called a lot. The caller
 718 *      must hold the dev_base or rtnl lock while allocating the name and
 719 *      adding the device in order to avoid duplicates. Returns the number
 720 *      of the unit assigned or a negative errno code.
 721 */
 722
 723int dev_alloc_name(struct net_device *dev, const char *name)
 724{
 725        int i = 0;
 726        char buf[IFNAMSIZ];
 727        const char *p;
 728        const int max_netdevices = 8*PAGE_SIZE;
 729        long *inuse;
 730        struct net_device *d;
 731
 732        p = strnchr(name, IFNAMSIZ-1, '%');
 733        if (p) {
 734                /*
 735                 * Verify the string as this thing may have come from
 736                 * the user.  There must be either one "%d" and no other "%"
 737                 * characters.
 738                 */
 739                if (p[1] != 'd' || strchr(p + 2, '%'))
 740                        return -EINVAL;
 741
 742                /* Use one page as a bit array of possible slots */
 743                inuse = (long *) get_zeroed_page(GFP_ATOMIC);
 744                if (!inuse)
 745                        return -ENOMEM;
 746
 747                for (d = dev_base; d; d = d->next) {
 748                        if (!sscanf(d->name, name, &i))
 749                                continue;
 750                        if (i < 0 || i >= max_netdevices)
 751                                continue;
 752
 753                        /*  avoid cases where sscanf is not exact inverse of printf */
 754                        snprintf(buf, sizeof(buf), name, i);
 755                        if (!strncmp(buf, d->name, IFNAMSIZ))
 756                                set_bit(i, inuse);
 757                }
 758
 759                i = find_first_zero_bit(inuse, max_netdevices);
 760                free_page((unsigned long) inuse);
 761        }
 762
 763        snprintf(buf, sizeof(buf), name, i);
 764        if (!__dev_get_by_name(buf)) {
 765                strlcpy(dev->name, buf, IFNAMSIZ);
 766                return i;
 767        }
 768
 769        /* It is possible to run out of possible slots
 770         * when the name is long and there isn't enough space left
 771         * for the digits, or if all bits are used.
 772         */
 773        return -ENFILE;
 774}
 775
 776
 777/**
 778 *      dev_change_name - change name of a device
 779 *      @dev: device
 780 *      @newname: name (or format string) must be at least IFNAMSIZ
 781 *
 782 *      Change name of a device, can pass format strings "eth%d".
 783 *      for wildcarding.
 784 */
 785int dev_change_name(struct net_device *dev, char *newname)
 786{
 787        int err = 0;
 788
 789        ASSERT_RTNL();
 790
 791        if (dev->flags & IFF_UP)
 792                return -EBUSY;
 793
 794        if (!dev_valid_name(newname))
 795                return -EINVAL;
 796
 797        if (strchr(newname, '%')) {
 798                err = dev_alloc_name(dev, newname);
 799                if (err < 0)
 800                        return err;
 801                strcpy(newname, dev->name);
 802        }
 803        else if (__dev_get_by_name(newname))
 804                return -EEXIST;
 805        else
 806                strlcpy(dev->name, newname, IFNAMSIZ);
 807
 808        err = class_device_rename(&dev->class_dev, dev->name);
 809        if (!err) {
 810                hlist_del(&dev->name_hlist);
 811                hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
 812                notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
 813        }
 814
 815        return err;
 816}
 817
 818/**
 819 *      netdev_state_change - device changes state
 820 *      @dev: device to cause notification
 821 *
 822 *      Called to indicate a device has changed state. This function calls
 823 *      the notifier chains for netdev_chain and sends a NEWLINK message
 824 *      to the routing socket.
 825 */
 826void netdev_state_change(struct net_device *dev)
 827{
 828        if (dev->flags & IFF_UP) {
 829                notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
 830                rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
 831        }
 832}
 833
 834/**
 835 *      dev_load        - load a network module
 836 *      @name: name of interface
 837 *
 838 *      If a network interface is not present and the process has suitable
 839 *      privileges this function loads the module. If module loading is not
 840 *      available in this kernel then it becomes a nop.
 841 */
 842
 843void dev_load(const char *name)
 844{
 845        struct net_device *dev;  
 846
 847        read_lock(&dev_base_lock);
 848        dev = __dev_get_by_name(name);
 849        read_unlock(&dev_base_lock);
 850
 851        if (!dev && capable(CAP_SYS_MODULE))
 852                request_module("%s", name);
 853}
 854
 855static int default_rebuild_header(struct sk_buff *skb)
 856{
 857        printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
 858               skb->dev ? skb->dev->name : "NULL!!!");
 859        kfree_skb(skb);
 860        return 1;
 861}
 862
 863
 864/*
 865 * Some old buggy device drivers change get_stats after registering
 866 * the device.  Try and trap them here.
 867 * This can be elimnated when all devices are known fixed.
 868 */
 869static inline int get_stats_changed(struct net_device *dev)
 870{
 871        int changed = dev->last_stats != dev->get_stats;
 872        dev->last_stats = dev->get_stats;
 873        return changed;
 874}
 875
 876/**
 877 *      dev_open        - prepare an interface for use.
 878 *      @dev:   device to open
 879 *
 880 *      Takes a device from down to up state. The device's private open
 881 *      function is invoked and then the multicast lists are loaded. Finally
 882 *      the device is moved into the up state and a %NETDEV_UP message is
 883 *      sent to the netdev notifier chain.
 884 *
 885 *      Calling this function on an active interface is a nop. On a failure
 886 *      a negative errno code is returned.
 887 */
 888int dev_open(struct net_device *dev)
 889{
 890        int ret = 0;
 891
 892        /*
 893         *      Is it already up?
 894         */
 895
 896        if (dev->flags & IFF_UP)
 897                return 0;
 898
 899        /*
 900         *       Check for broken device drivers.
 901         */
 902        if (get_stats_changed(dev) && net_ratelimit()) {
 903                printk(KERN_ERR "%s: driver changed get_stats after register\n",
 904                       dev->name);
 905        }
 906
 907        /*
 908         *      Is it even present?
 909         */
 910        if (!netif_device_present(dev))
 911                return -ENODEV;
 912
 913        /*
 914         *      Call device private open method
 915         */
 916        set_bit(__LINK_STATE_START, &dev->state);
 917        if (dev->open) {
 918                ret = dev->open(dev);
 919                if (ret)
 920                        clear_bit(__LINK_STATE_START, &dev->state);
 921        }
 922
 923        /*
 924         *      Check for more broken device drivers.
 925         */
 926        if (get_stats_changed(dev) && net_ratelimit()) {
 927                printk(KERN_ERR "%s: driver changed get_stats in open\n",
 928                       dev->name);
 929        }
 930
 931        /*
 932         *      If it went open OK then:
 933         */
 934
 935        if (!ret) {
 936                /*
 937                 *      Set the flags.
 938                 */
 939                dev->flags |= IFF_UP;
 940
 941                /*
 942                 *      Initialize multicasting status
 943                 */
 944                dev_mc_upload(dev);
 945
 946                /*
 947                 *      Wakeup transmit queue engine
 948                 */
 949                dev_activate(dev);
 950
 951                /*
 952                 *      ... and announce new interface.
 953                 */
 954                notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
 955        }
 956        return ret;
 957}
 958
 959/**
 960 *      dev_close - shutdown an interface.
 961 *      @dev: device to shutdown
 962 *
 963 *      This function moves an active device into down state. A
 964 *      %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
 965 *      is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
 966 *      chain.
 967 */
 968int dev_close(struct net_device *dev)
 969{
 970        if (!(dev->flags & IFF_UP))
 971                return 0;
 972
 973        /*
 974         *      Tell people we are going down, so that they can
 975         *      prepare to death, when device is still operating.
 976         */
 977        notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
 978
 979        dev_deactivate(dev);
 980
 981        clear_bit(__LINK_STATE_START, &dev->state);
 982
 983        /* Synchronize to scheduled poll. We cannot touch poll list,
 984         * it can be even on different cpu. So just clear netif_running(),
 985         * and wait when poll really will happen. Actually, the best place
 986         * for this is inside dev->stop() after device stopped its irq
 987         * engine, but this requires more changes in devices. */
 988
 989        smp_mb__after_clear_bit(); /* Commit netif_running(). */
 990        while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
 991                /* No hurry. */
 992                current->state = TASK_INTERRUPTIBLE;
 993                schedule_timeout(1);
 994        }
 995
 996        /*
 997         *      Call the device specific close. This cannot fail.
 998         *      Only if device is UP
 999         *
1000         *      We allow it to be called even after a DETACH hot-plug
1001         *      event.
1002         */
1003        if (dev->stop)
1004                dev->stop(dev);
1005
1006        /*
1007         *      Device is now down.
1008         */
1009
1010        dev->flags &= ~IFF_UP;
1011
1012        /*
1013         * Tell people we are down
1014         */
1015        notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1016
1017        return 0;
1018}
1019
1020
1021/*
1022 *      Device change register/unregister. These are not inline or static
1023 *      as we export them to the world.
1024 */
1025
1026/**
1027 *      register_netdevice_notifier - register a network notifier block
1028 *      @nb: notifier
1029 *
1030 *      Register a notifier to be called when network device events occur.
1031 *      The notifier passed is linked into the kernel structures and must
1032 *      not be reused until it has been unregistered. A negative errno code
1033 *      is returned on a failure.
1034 *
1035 *      When registered all registration and up events are replayed
1036 *      to the new notifier to allow device to have a race free 
1037 *      view of the network device list.
1038 */
1039
1040int register_netdevice_notifier(struct notifier_block *nb)
1041{
1042        struct net_device *dev;
1043        int err;
1044
1045        rtnl_lock();
1046        err = notifier_chain_register(&netdev_chain, nb);
1047        if (!err) {
1048                for (dev = dev_base; dev; dev = dev->next) {
1049                        nb->notifier_call(nb, NETDEV_REGISTER, dev);
1050
1051                        if (dev->flags & IFF_UP) 
1052                                nb->notifier_call(nb, NETDEV_UP, dev);
1053                }
1054        }
1055        rtnl_unlock();
1056        return err;
1057}
1058
1059/**
1060 *      unregister_netdevice_notifier - unregister a network notifier block
1061 *      @nb: notifier
1062 *
1063 *      Unregister a notifier previously registered by
1064 *      register_netdevice_notifier(). The notifier is unlinked into the
1065 *      kernel structures and may then be reused. A negative errno code
1066 *      is returned on a failure.
1067 */
1068
1069int unregister_netdevice_notifier(struct notifier_block *nb)
1070{
1071        return notifier_chain_unregister(&netdev_chain, nb);
1072}
1073
1074/**
1075 *      call_netdevice_notifiers - call all network notifier blocks
1076 *      @val: value passed unmodified to notifier function
1077 *      @v:   pointer passed unmodified to notifier function
1078 *
1079 *      Call all network notifier blocks.  Parameters and return value
1080 *      are as for notifier_call_chain().
1081 */
1082
1083int call_netdevice_notifiers(unsigned long val, void *v)
1084{
1085        return notifier_call_chain(&netdev_chain, val, v);
1086}
1087
1088/*
1089 *      Support routine. Sends outgoing frames to any network
1090 *      taps currently in use.
1091 */
1092
1093void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1094{
1095        struct packet_type *ptype;
1096        net_timestamp(&skb->stamp);
1097
1098        rcu_read_lock();
1099        list_for_each_entry_rcu(ptype, &ptype_all, list) {
1100                /* Never send packets back to the socket
1101                 * they originated from - MvS (miquels@drinkel.ow.org)
1102                 */
1103                if ((ptype->dev == dev || !ptype->dev) &&
1104                    (ptype->af_packet_priv == NULL ||
1105                     (struct sock *)ptype->af_packet_priv != skb->sk)) {
1106                        struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1107                        if (!skb2)
1108                                break;
1109
1110                        /* skb->nh should be correctly
1111                           set by sender, so that the second statement is
1112                           just protection against buggy protocols.
1113                         */
1114                        skb2->mac.raw = skb2->data;
1115
1116                        if (skb2->nh.raw < skb2->data ||
1117                            skb2->nh.raw > skb2->tail) {
1118                                if (net_ratelimit())
1119                                        printk(KERN_CRIT "protocol %04x is "
1120                                               "buggy, dev %s\n",
1121                                               skb2->protocol, dev->name);
1122                                skb2->nh.raw = skb2->data;
1123                        }
1124
1125                        skb2->h.raw = skb2->nh.raw;
1126                        skb2->pkt_type = PACKET_OUTGOING;
1127                        ptype->func(skb2, skb->dev, ptype);
1128                }
1129        }
1130        rcu_read_unlock();
1131}
1132
1133/*
1134 * Invalidate hardware checksum when packet is to be mangled, and
1135 * complete checksum manually on outgoing path.
1136 */
1137int skb_checksum_help(struct sk_buff **pskb, int inward)
1138{
1139        unsigned int csum;
1140        int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1141
1142        if (inward) {
1143                (*pskb)->ip_summed = CHECKSUM_NONE;
1144                goto out;
1145        }
1146
1147        if (skb_shared(*pskb)  || skb_cloned(*pskb)) {
1148                struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1149                if (!newskb) {
1150                        ret = -ENOMEM;
1151                        goto out;
1152                }
1153                if ((*pskb)->sk)
1154                        skb_set_owner_w(newskb, (*pskb)->sk);
1155                kfree_skb(*pskb);
1156                *pskb = newskb;
1157        }
1158
1159        if (offset > (int)(*pskb)->len)
1160                BUG();
1161        csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1162
1163        offset = (*pskb)->tail - (*pskb)->h.raw;
1164        if (offset <= 0)
1165                BUG();
1166        if ((*pskb)->csum + 2 > offset)
1167                BUG();
1168
1169        *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1170        (*pskb)->ip_summed = CHECKSUM_NONE;
1171out:    
1172        return ret;
1173}
1174
1175#ifdef CONFIG_HIGHMEM
1176/* Actually, we should eliminate this check as soon as we know, that:
1177 * 1. IOMMU is present and allows to map all the memory.
1178 * 2. No high memory really exists on this machine.
1179 */
1180
1181static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1182{
1183        int i;
1184
1185        if (dev->features & NETIF_F_HIGHDMA)
1186                return 0;
1187
1188        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1189                if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1190                        return 1;
1191
1192        return 0;
1193}
1194#else
1195#define illegal_highdma(dev, skb)       (0)
1196#endif
1197
1198extern void skb_release_data(struct sk_buff *);
1199
1200/* Keep head the same: replace data */
1201int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1202{
1203        unsigned int size;
1204        u8 *data;
1205        long offset;
1206        struct skb_shared_info *ninfo;
1207        int headerlen = skb->data - skb->head;
1208        int expand = (skb->tail + skb->data_len) - skb->end;
1209
1210        if (skb_shared(skb))
1211                BUG();
1212
1213        if (expand <= 0)
1214                expand = 0;
1215
1216        size = skb->end - skb->head + expand;
1217        size = SKB_DATA_ALIGN(size);
1218        data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1219        if (!data)
1220                return -ENOMEM;
1221
1222        /* Copy entire thing */
1223        if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1224                BUG();
1225
1226        /* Set up shinfo */
1227        ninfo = (struct skb_shared_info*)(data + size);
1228        atomic_set(&ninfo->dataref, 1);
1229        ninfo->tso_size = skb_shinfo(skb)->tso_size;
1230        ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1231        ninfo->nr_frags = 0;
1232        ninfo->frag_list = NULL;
1233
1234        /* Offset between the two in bytes */
1235        offset = data - skb->head;
1236
1237        /* Free old data. */
1238        skb_release_data(skb);
1239
1240        skb->head = data;
1241        skb->end  = data + size;
1242
1243        /* Set up new pointers */
1244        skb->h.raw   += offset;
1245        skb->nh.raw  += offset;
1246        skb->mac.raw += offset;
1247        skb->tail    += offset;
1248        skb->data    += offset;
1249
1250        /* We are no longer a clone, even if we were. */
1251        skb->cloned    = 0;
1252
1253        skb->tail     += skb->data_len;
1254        skb->data_len  = 0;
1255        return 0;
1256}
1257
1258#define HARD_TX_LOCK_BH(dev, cpu) {                     \
1259        if ((dev->features & NETIF_F_LLTX) == 0) {      \
1260                spin_lock_bh(&dev->xmit_lock);          \
1261                dev->xmit_lock_owner = cpu;             \
1262        }                                               \
1263}
1264
1265#define HARD_TX_UNLOCK_BH(dev) {                        \
1266        if ((dev->features & NETIF_F_LLTX) == 0) {      \
1267                dev->xmit_lock_owner = -1;              \
1268                spin_unlock_bh(&dev->xmit_lock);        \
1269        }                                               \
1270}
1271
1272static inline void qdisc_run(struct net_device *dev)
1273{
1274        while (!netif_queue_stopped(dev) &&
1275               qdisc_restart(dev)<0)
1276                /* NOTHING */;
1277}
1278
1279/**
1280 *      dev_queue_xmit - transmit a buffer
1281 *      @skb: buffer to transmit
1282 *
1283 *      Queue a buffer for transmission to a network device. The caller must
1284 *      have set the device and priority and built the buffer before calling
1285 *      this function. The function can be called from an interrupt.
1286 *
1287 *      A negative errno code is returned on a failure. A success does not
1288 *      guarantee the frame will be transmitted as it may be dropped due
1289 *      to congestion or traffic shaping.
1290 */
1291
1292int dev_queue_xmit(struct sk_buff *skb)
1293{
1294        struct net_device *dev = skb->dev;
1295        struct Qdisc *q;
1296        int rc = -ENOMEM;
1297
1298        if (skb_shinfo(skb)->frag_list &&
1299            !(dev->features & NETIF_F_FRAGLIST) &&
1300            __skb_linearize(skb, GFP_ATOMIC))
1301                goto out_kfree_skb;
1302
1303        /* Fragmented skb is linearized if device does not support SG,
1304         * or if at least one of fragments is in highmem and device
1305         * does not support DMA from it.
1306         */
1307        if (skb_shinfo(skb)->nr_frags &&
1308            (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1309            __skb_linearize(skb, GFP_ATOMIC))
1310                goto out_kfree_skb;
1311
1312        /* If packet is not checksummed and device does not support
1313         * checksumming for this protocol, complete checksumming here.
1314         */
1315        if (skb->ip_summed == CHECKSUM_HW &&
1316            (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1317             (!(dev->features & NETIF_F_IP_CSUM) ||
1318              skb->protocol != htons(ETH_P_IP))))
1319                if (skb_checksum_help(&skb, 0))
1320                        goto out_kfree_skb;
1321
1322        rcu_read_lock();
1323        /* Updates of qdisc are serialized by queue_lock. 
1324         * The struct Qdisc which is pointed to by qdisc is now a 
1325         * rcu structure - it may be accessed without acquiring 
1326         * a lock (but the structure may be stale.) The freeing of the
1327         * qdisc will be deferred until it's known that there are no 
1328         * more references to it.
1329         * 
1330         * If the qdisc has an enqueue function, we still need to 
1331         * hold the queue_lock before calling it, since queue_lock
1332         * also serializes access to the device queue.
1333         */
1334
1335        q = dev->qdisc;
1336        smp_read_barrier_depends();
1337#ifdef CONFIG_NET_CLS_ACT
1338        skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1339#endif
1340        if (q->enqueue) {
1341                /* Grab device queue */
1342                spin_lock_bh(&dev->queue_lock);
1343
1344                rc = q->enqueue(skb, q);
1345
1346                qdisc_run(dev);
1347
1348                spin_unlock_bh(&dev->queue_lock);
1349                rcu_read_unlock();
1350                rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1351                goto out;
1352        }
1353        rcu_read_unlock();
1354
1355        /* The device has no queue. Common case for software devices:
1356           loopback, all the sorts of tunnels...
1357
1358           Really, it is unlikely that xmit_lock protection is necessary here.
1359           (f.e. loopback and IP tunnels are clean ignoring statistics
1360           counters.)
1361           However, it is possible, that they rely on protection
1362           made by us here.
1363
1364           Check this and shot the lock. It is not prone from deadlocks.
1365           Either shot noqueue qdisc, it is even simpler 8)
1366         */
1367        if (dev->flags & IFF_UP) {
1368                int cpu = get_cpu();
1369
1370                if (dev->xmit_lock_owner != cpu) {
1371
1372                        HARD_TX_LOCK_BH(dev, cpu);
1373                        put_cpu();
1374
1375                        if (!netif_queue_stopped(dev)) {
1376                                if (netdev_nit)
1377                                        dev_queue_xmit_nit(skb, dev);
1378
1379                                rc = 0;
1380                                if (!dev->hard_start_xmit(skb, dev)) {
1381                                        HARD_TX_UNLOCK_BH(dev);
1382                                        goto out;
1383                                }
1384                        }
1385                        HARD_TX_UNLOCK_BH(dev);
1386                        if (net_ratelimit())
1387                                printk(KERN_CRIT "Virtual device %s asks to "
1388                                       "queue packet!\n", dev->name);
1389                        goto out_enetdown;
1390                } else {
1391                        put_cpu();
1392                        /* Recursion is detected! It is possible,
1393                         * unfortunately */
1394                        if (net_ratelimit())
1395                                printk(KERN_CRIT "Dead loop on virtual device "
1396                                       "%s, fix it urgently!\n", dev->name);
1397                }
1398        }
1399out_enetdown:
1400        rc = -ENETDOWN;
1401out_kfree_skb:
1402        kfree_skb(skb);
1403out:
1404        return rc;
1405}
1406
1407
1408/*=======================================================================
1409                        Receiver routines
1410  =======================================================================*/
1411
1412int netdev_max_backlog = 300;
1413int weight_p = 64;            /* old backlog weight */
1414/* These numbers are selected based on intuition and some
1415 * experimentatiom, if you have more scientific way of doing this
1416 * please go ahead and fix things.
1417 */
1418int no_cong_thresh = 10;
1419int no_cong = 20;
1420int lo_cong = 100;
1421int mod_cong = 290;
1422
1423DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1424
1425
1426#ifdef CONFIG_NET_HW_FLOWCONTROL
1427atomic_t netdev_dropping = ATOMIC_INIT(0);
1428static unsigned long netdev_fc_mask = 1;
1429unsigned long netdev_fc_xoff;
1430spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1431
1432static struct
1433{
1434        void (*stimul)(struct net_device *);
1435        struct net_device *dev;
1436} netdev_fc_slots[BITS_PER_LONG];
1437
1438int netdev_register_fc(struct net_device *dev,
1439                       void (*stimul)(struct net_device *dev))
1440{
1441        int bit = 0;
1442        unsigned long flags;
1443
1444        spin_lock_irqsave(&netdev_fc_lock, flags);
1445        if (netdev_fc_mask != ~0UL) {
1446                bit = ffz(netdev_fc_mask);
1447                netdev_fc_slots[bit].stimul = stimul;
1448                netdev_fc_slots[bit].dev = dev;
1449                set_bit(bit, &netdev_fc_mask);
1450                clear_bit(bit, &netdev_fc_xoff);
1451        }
1452        spin_unlock_irqrestore(&netdev_fc_lock, flags);
1453        return bit;
1454}
1455
1456void netdev_unregister_fc(int bit)
1457{
1458        unsigned long flags;
1459
1460        spin_lock_irqsave(&netdev_fc_lock, flags);
1461        if (bit > 0) {
1462                netdev_fc_slots[bit].stimul = NULL;
1463                netdev_fc_slots[bit].dev = NULL;
1464                clear_bit(bit, &netdev_fc_mask);
1465                clear_bit(bit, &netdev_fc_xoff);
1466        }
1467        spin_unlock_irqrestore(&netdev_fc_lock, flags);
1468}
1469
1470static void netdev_wakeup(void)
1471{
1472        unsigned long xoff;
1473
1474        spin_lock(&netdev_fc_lock);
1475        xoff = netdev_fc_xoff;
1476        netdev_fc_xoff = 0;
1477        while (xoff) {
1478                int i = ffz(~xoff);
1479                xoff &= ~(1 << i);
1480                netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1481        }
1482        spin_unlock(&netdev_fc_lock);
1483}
1484#endif
1485
1486static void get_sample_stats(int cpu)
1487{
1488#ifdef RAND_LIE
1489        unsigned long rd;
1490        int rq;
1491#endif
1492        struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1493        int blog = sd->input_pkt_queue.qlen;
1494        int avg_blog = sd->avg_blog;
1495
1496        avg_blog = (avg_blog >> 1) + (blog >> 1);
1497
1498        if (avg_blog > mod_cong) {
1499                /* Above moderate congestion levels. */
1500                sd->cng_level = NET_RX_CN_HIGH;
1501#ifdef RAND_LIE
1502                rd = net_random();
1503                rq = rd % netdev_max_backlog;
1504                if (rq < avg_blog) /* unlucky bastard */
1505                        sd->cng_level = NET_RX_DROP;
1506#endif
1507        } else if (avg_blog > lo_cong) {
1508                sd->cng_level = NET_RX_CN_MOD;
1509#ifdef RAND_LIE
1510                rd = net_random();
1511                rq = rd % netdev_max_backlog;
1512                        if (rq < avg_blog) /* unlucky bastard */
1513                                sd->cng_level = NET_RX_CN_HIGH;
1514#endif
1515        } else if (avg_blog > no_cong)
1516                sd->cng_level = NET_RX_CN_LOW;
1517        else  /* no congestion */
1518                sd->cng_level = NET_RX_SUCCESS;
1519
1520        sd->avg_blog = avg_blog;
1521}
1522
1523#ifdef OFFLINE_SAMPLE
1524static void sample_queue(unsigned long dummy)
1525{
1526/* 10 ms 0r 1ms -- i don't care -- JHS */
1527        int next_tick = 1;
1528        int cpu = smp_processor_id();
1529
1530        get_sample_stats(cpu);
1531        next_tick += jiffies;
1532        mod_timer(&samp_timer, next_tick);
1533}
1534#endif
1535
1536
1537/**
1538 *      netif_rx        -       post buffer to the network code
1539 *      @skb: buffer to post
1540 *
1541 *      This function receives a packet from a device driver and queues it for
1542 *      the upper (protocol) levels to process.  It always succeeds. The buffer
1543 *      may be dropped during processing for congestion control or by the
1544 *      protocol layers.
1545 *
1546 *      return values:
1547 *      NET_RX_SUCCESS  (no congestion)
1548 *      NET_RX_CN_LOW   (low congestion)
1549 *      NET_RX_CN_MOD   (moderate congestion)
1550 *      NET_RX_CN_HIGH  (high congestion)
1551 *      NET_RX_DROP     (packet was dropped)
1552 *
1553 */
1554
1555int netif_rx(struct sk_buff *skb)
1556{
1557        int this_cpu;
1558        struct softnet_data *queue;
1559        unsigned long flags;
1560
1561#ifdef CONFIG_NETPOLL_RX
1562        if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1563                kfree_skb(skb);
1564                return NET_RX_DROP;
1565        }
1566#endif
1567        
1568        if (!skb->stamp.tv_sec)
1569                net_timestamp(&skb->stamp);
1570
1571        /*
1572         * The code is rearranged so that the path is the most
1573         * short when CPU is congested, but is still operating.
1574         */
1575        local_irq_save(flags);
1576        this_cpu = smp_processor_id();
1577        queue = &__get_cpu_var(softnet_data);
1578
1579        __get_cpu_var(netdev_rx_stat).total++;
1580        if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1581                if (queue->input_pkt_queue.qlen) {
1582                        if (queue->throttle)
1583                                goto drop;
1584
1585enqueue:
1586                        dev_hold(skb->dev);
1587                        __skb_queue_tail(&queue->input_pkt_queue, skb);
1588#ifndef OFFLINE_SAMPLE
1589                        get_sample_stats(this_cpu);
1590#endif
1591                        local_irq_restore(flags);
1592                        return queue->cng_level;
1593                }
1594
1595                if (queue->throttle) {
1596                        queue->throttle = 0;
1597#ifdef CONFIG_NET_HW_FLOWCONTROL
1598                        if (atomic_dec_and_test(&netdev_dropping))
1599                                netdev_wakeup();
1600#endif
1601                }
1602
1603                netif_rx_schedule(&queue->backlog_dev);
1604                goto enqueue;
1605        }
1606
1607        if (!queue->throttle) {
1608                queue->throttle = 1;
1609                __get_cpu_var(netdev_rx_stat).throttled++;
1610#ifdef CONFIG_NET_HW_FLOWCONTROL
1611                atomic_inc(&netdev_dropping);
1612#endif
1613        }
1614
1615drop:
1616        __get_cpu_var(netdev_rx_stat).dropped++;
1617        local_irq_restore(flags);
1618
1619        kfree_skb(skb);
1620        return NET_RX_DROP;
1621}
1622
1623static __inline__ void skb_bond(struct sk_buff *skb)
1624{
1625        struct net_device *dev = skb->dev;
1626
1627        if (dev->master) {
1628                skb->real_dev = skb->dev;
1629                skb->dev = dev->master;
1630        }
1631}
1632
1633static void net_tx_action(struct softirq_action *h)
1634{
1635        struct softnet_data *sd = &__get_cpu_var(softnet_data);
1636
1637        if (sd->completion_queue) {
1638                struct sk_buff *clist;
1639
1640                local_irq_disable();
1641                clist = sd->completion_queue;
1642                sd->completion_queue = NULL;
1643                local_irq_enable();
1644
1645                while (clist) {
1646                        struct sk_buff *skb = clist;
1647                        clist = clist->next;
1648
1649                        BUG_TRAP(!atomic_read(&skb->users));
1650                        __kfree_skb(skb);
1651                }
1652        }
1653
1654        if (sd->output_queue) {
1655                struct net_device *head;
1656
1657                local_irq_disable();
1658                head = sd->output_queue;
1659                sd->output_queue = NULL;
1660                local_irq_enable();
1661
1662                while (head) {
1663                        struct net_device *dev = head;
1664                        head = head->next_sched;
1665
1666                        smp_mb__before_clear_bit();
1667                        clear_bit(__LINK_STATE_SCHED, &dev->state);
1668
1669                        if (spin_trylock(&dev->queue_lock)) {
1670                                qdisc_run(dev);
1671                                spin_unlock(&dev->queue_lock);
1672                        } else {
1673                                netif_schedule(dev);
1674                        }
1675                }
1676        }
1677}
1678
1679static __inline__ int deliver_skb(struct sk_buff *skb,
1680                                  struct packet_type *pt_prev, int last)
1681{
1682        atomic_inc(&skb->users);
1683        return pt_prev->func(skb, skb->dev, pt_prev);
1684}
1685
1686
1687#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1688int (*br_handle_frame_hook)(struct sk_buff *skb);
1689
1690static __inline__ int handle_bridge(struct sk_buff *skb,
1691                                     struct packet_type *pt_prev)
1692{
1693        int ret = NET_RX_DROP;
1694        if (pt_prev)
1695                ret = deliver_skb(skb, pt_prev, 0);
1696
1697        return ret;
1698}
1699
1700#endif
1701
1702static inline int __handle_bridge(struct sk_buff *skb,
1703                        struct packet_type **pt_prev, int *ret)
1704{
1705#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1706        if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1707                *ret = handle_bridge(skb, *pt_prev);
1708                if (br_handle_frame_hook(skb) == 0)
1709                        return 1;
1710
1711                *pt_prev = NULL;
1712        }
1713#endif
1714        return 0;
1715}
1716
1717
1718#ifdef CONFIG_NET_CLS_ACT
1719/* TODO: Maybe we should just force sch_ingress to be compiled in
1720 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1721 * a compare and 2 stores extra right now if we dont have it on
1722 * but have CONFIG_NET_CLS_ACT
1723 * NOTE: This doesnt stop any functionality; if you dont have 
1724 * the ingress scheduler, you just cant add policies on ingress.
1725 *
1726 */
1727int ing_filter(struct sk_buff *skb) 
1728{
1729        struct Qdisc *q;
1730        struct net_device *dev = skb->dev;
1731        int result = TC_ACT_OK;
1732        
1733        if (dev->qdisc_ingress) {
1734                __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1735                if (MAX_RED_LOOP < ttl++) {
1736                        printk("Redir loop detected Dropping packet (%s->%s)\n",
1737                                skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1738                        return TC_ACT_SHOT;
1739                }
1740
1741                skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1742
1743                skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1744                if (NULL == skb->input_dev) {
1745                        skb->input_dev = skb->dev;
1746                        printk("ing_filter:  fixed  %s out %s\n",skb->input_dev->name,skb->dev->name);
1747                }
1748                spin_lock(&dev->ingress_lock);
1749                if ((q = dev->qdisc_ingress) != NULL)
1750                        result = q->enqueue(skb, q);
1751                spin_unlock(&dev->ingress_lock);
1752
1753        }
1754
1755        return result;
1756}
1757#endif
1758
1759int netif_receive_skb(struct sk_buff *skb)
1760{
1761        struct packet_type *ptype, *pt_prev;
1762        int ret = NET_RX_DROP;
1763        unsigned short type;
1764
1765#ifdef CONFIG_NETPOLL_RX
1766        if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1767                kfree_skb(skb);
1768                return NET_RX_DROP;
1769        }
1770#endif
1771
1772        if (!skb->stamp.tv_sec)
1773                net_timestamp(&skb->stamp);
1774
1775        skb_bond(skb);
1776
1777        __get_cpu_var(netdev_rx_stat).total++;
1778
1779        skb->h.raw = skb->nh.raw = skb->data;
1780        skb->mac_len = skb->nh.raw - skb->mac.raw;
1781
1782        pt_prev = NULL;
1783#ifdef CONFIG_NET_CLS_ACT
1784        if (skb->tc_verd & TC_NCLS) {
1785                skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1786                rcu_read_lock();
1787                goto ncls;
1788        }
1789 #endif
1790
1791        rcu_read_lock();
1792        list_for_each_entry_rcu(ptype, &ptype_all, list) {
1793                if (!ptype->dev || ptype->dev == skb->dev) {
1794                        if (pt_prev) 
1795                                ret = deliver_skb(skb, pt_prev, 0);
1796                        pt_prev = ptype;
1797                }
1798        }
1799
1800#ifdef CONFIG_NET_CLS_ACT
1801        if (pt_prev) {
1802                atomic_inc(&skb->users);
1803                ret = pt_prev->func(skb, skb->dev, pt_prev);
1804                pt_prev = NULL; /* noone else should process this after*/
1805        } else {
1806                skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1807        }
1808
1809        ret = ing_filter(skb);
1810
1811        if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1812                kfree_skb(skb);
1813                goto out;
1814        }
1815
1816        skb->tc_verd = 0;
1817ncls:
1818#endif
1819
1820        handle_diverter(skb);
1821
1822        if (__handle_bridge(skb, &pt_prev, &ret))
1823                goto out;
1824
1825        type = skb->protocol;
1826        list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1827                if (ptype->type == type &&
1828                    (!ptype->dev || ptype->dev == skb->dev)) {
1829                        if (pt_prev) 
1830                                ret = deliver_skb(skb, pt_prev, 0);
1831                        pt_prev = ptype;
1832                }
1833        }
1834
1835        if (pt_prev) {
1836                ret = pt_prev->func(skb, skb->dev, pt_prev);
1837        } else {
1838                kfree_skb(skb);
1839                /* Jamal, now you will not able to escape explaining
1840                 * me how you were going to use this. :-)
1841                 */
1842                ret = NET_RX_DROP;
1843        }
1844
1845out:
1846        rcu_read_unlock();
1847        return ret;
1848}
1849
1850static int process_backlog(struct net_device *backlog_dev, int *budget)
1851{
1852        int work = 0;
1853        int quota = min(backlog_dev->quota, *budget);
1854        struct softnet_data *queue = &__get_cpu_var(softnet_data);
1855        unsigned long start_time = jiffies;
1856
1857        for (;;) {
1858                struct sk_buff *skb;
1859                struct net_device *dev;
1860
1861                local_irq_disable();
1862                skb = __skb_dequeue(&queue->input_pkt_queue);
1863                if (!skb)
1864                        goto job_done;
1865                local_irq_enable();
1866
1867                dev = skb->dev;
1868
1869                netif_receive_skb(skb);
1870
1871                dev_put(dev);
1872
1873                work++;
1874
1875                if (work >= quota || jiffies - start_time > 1)
1876                        break;
1877
1878#ifdef CONFIG_NET_HW_FLOWCONTROL
1879                if (queue->throttle &&
1880                    queue->input_pkt_queue.qlen < no_cong_thresh ) {
1881                        queue->throttle = 0;
1882                        if (atomic_dec_and_test(&netdev_dropping)) {
1883                                netdev_wakeup();
1884                                break;
1885                        }
1886                }
1887#endif
1888        }
1889
1890        backlog_dev->quota -= work;
1891        *budget -= work;
1892        return -1;
1893
1894job_done:
1895        backlog_dev->quota -= work;
1896        *budget -= work;
1897
1898        list_del(&backlog_dev->poll_list);
1899        smp_mb__before_clear_bit();
1900        netif_poll_enable(backlog_dev);
1901
1902        if (queue->throttle) {
1903                queue->throttle = 0;
1904#ifdef CONFIG_NET_HW_FLOWCONTROL
1905                if (atomic_dec_and_test(&netdev_dropping))
1906                        netdev_wakeup();
1907#endif
1908        }
1909        local_irq_enable();
1910        return 0;
1911}
1912
1913static void net_rx_action(struct softirq_action *h)
1914{
1915        struct softnet_data *queue = &__get_cpu_var(softnet_data);
1916        unsigned long start_time = jiffies;
1917        int budget = netdev_max_backlog;
1918
1919        
1920        local_irq_disable();
1921
1922        while (!list_empty(&queue->poll_list)) {
1923                struct net_device *dev;
1924
1925                if (budget <= 0 || jiffies - start_time > 1)
1926                        goto softnet_break;
1927
1928                local_irq_enable();
1929
1930                dev = list_entry(queue->poll_list.next,
1931                                 struct net_device, poll_list);
1932
1933                if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1934                        local_irq_disable();
1935                        list_del(&dev->poll_list);
1936                        list_add_tail(&dev->poll_list, &queue->poll_list);
1937                        if (dev->quota < 0)
1938                                dev->quota += dev->weight;
1939                        else
1940                                dev->quota = dev->weight;
1941                } else {
1942                        dev_put(dev);
1943                        local_irq_disable();
1944                }
1945        }
1946out:
1947        local_irq_enable();
1948        return;
1949
1950softnet_break:
1951        __get_cpu_var(netdev_rx_stat).time_squeeze++;
1952        __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1953        goto out;
1954}
1955
1956static gifconf_func_t * gifconf_list [NPROTO];
1957
1958/**
1959 *      register_gifconf        -       register a SIOCGIF handler
1960 *      @family: Address family
1961 *      @gifconf: Function handler
1962 *
1963 *      Register protocol dependent address dumping routines. The handler
1964 *      that is passed must not be freed or reused until it has been replaced
1965 *      by another handler.
1966 */
1967int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1968{
1969        if (family >= NPROTO)
1970                return -EINVAL;
1971        gifconf_list[family] = gifconf;
1972        return 0;
1973}
1974
1975
1976/*
1977 *      Map an interface index to its name (SIOCGIFNAME)
1978 */
1979
1980/*
1981 *      We need this ioctl for efficient implementation of the
1982 *      if_indextoname() function required by the IPv6 API.  Without
1983 *      it, we would have to search all the interfaces to find a
1984 *      match.  --pb
1985 */
1986
1987static int dev_ifname(struct ifreq __user *arg)
1988{
1989        struct net_device *dev;
1990        struct ifreq ifr;
1991
1992        /*
1993         *      Fetch the caller's info block.
1994         */
1995
1996        if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1997                return -EFAULT;
1998
1999        read_lock(&dev_base_lock);
2000        dev = __dev_get_by_index(ifr.ifr_ifindex);
2001        if (!dev) {
2002                read_unlock(&dev_base_lock);
2003                return -ENODEV;
2004        }
2005
2006        strcpy(ifr.ifr_name, dev->name);
2007        read_unlock(&dev_base_lock);
2008
2009        if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2010                return -EFAULT;
2011        return 0;
2012}
2013
2014/*
2015 *      Perform a SIOCGIFCONF call. This structure will change
2016 *      size eventually, and there is nothing I can do about it.
2017 *      Thus we will need a 'compatibility mode'.
2018 */
2019
2020static int dev_ifconf(char __user *arg)
2021{
2022        struct ifconf ifc;
2023        struct net_device *dev;
2024        char __user *pos;
2025        int len;
2026        int total;
2027        int i;
2028
2029        /*
2030         *      Fetch the caller's info block.
2031         */
2032
2033        if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2034                return -EFAULT;
2035
2036        pos = ifc.ifc_buf;
2037        len = ifc.ifc_len;
2038
2039        /*
2040         *      Loop over the interfaces, and write an info block for each.
2041         */
2042
2043        total = 0;
2044        for (dev = dev_base; dev; dev = dev->next) {
2045                for (i = 0; i < NPROTO; i++) {
2046                        if (gifconf_list[i]) {
2047                                int done;
2048                                if (!pos)
2049                                        done = gifconf_list[i](dev, NULL, 0);
2050                                else
2051                                        done = gifconf_list[i](dev, pos + total,
2052                                                               len - total);
2053                                if (done < 0)
2054                                        return -EFAULT;
2055                                total += done;
2056                        }
2057                }
2058        }
2059
2060        /*
2061         *      All done.  Write the updated control block back to the caller.
2062         */
2063        ifc.ifc_len = total;
2064
2065        /*
2066         *      Both BSD and Solaris return 0 here, so we do too.
2067         */
2068        return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2069}
2070
2071#ifdef CONFIG_PROC_FS
2072/*
2073 *      This is invoked by the /proc filesystem handler to display a device
2074 *      in detail.
2075 */
2076static __inline__ struct net_device *dev_get_idx(loff_t pos)
2077{
2078        struct net_device *dev;
2079        loff_t i;
2080
2081        for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2082
2083        return i == pos ? dev : NULL;
2084}
2085
2086void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2087{
2088        read_lock(&dev_base_lock);
2089        return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2090}
2091
2092void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2093{
2094        ++*pos;
2095        return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2096}
2097
2098void dev_seq_stop(struct seq_file *seq, void *v)
2099{
2100        read_unlock(&dev_base_lock);
2101}
2102
2103static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2104{
2105        if (dev->get_stats) {
2106                struct net_device_stats *stats = dev->get_stats(dev);
2107
2108                seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2109                                "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2110                           dev->name, stats->rx_bytes, stats->rx_packets,
2111                           stats->rx_errors,
2112                           stats->rx_dropped + stats->rx_missed_errors,
2113                           stats->rx_fifo_errors,
2114                           stats->rx_length_errors + stats->rx_over_errors +
2115                             stats->rx_crc_errors + stats->rx_frame_errors,
2116                           stats->rx_compressed, stats->multicast,
2117                           stats->tx_bytes, stats->tx_packets,
2118                           stats->tx_errors, stats->tx_dropped,
2119                           stats->tx_fifo_errors, stats->collisions,
2120                           stats->tx_carrier_errors +
2121                             stats->tx_aborted_errors +
2122                             stats->tx_window_errors +
2123                             stats->tx_heartbeat_errors,
2124                           stats->tx_compressed);
2125        } else
2126                seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2127}
2128
2129/*
2130 *      Called from the PROCfs module. This now uses the new arbitrary sized
2131 *      /proc/net interface to create /proc/net/dev
2132 */
2133static int dev_seq_show(struct seq_file *seq, void *v)
2134{
2135        if (v == SEQ_START_TOKEN)
2136                seq_puts(seq, "Inter-|   Receive                            "
2137                              "                    |  Transmit\n"
2138                              " face |bytes    packets errs drop fifo frame "
2139                              "compressed multicast|bytes    packets errs "
2140                              "drop fifo colls carrier compressed\n");
2141        else
2142                dev_seq_printf_stats(seq, v);
2143        return 0;
2144}
2145
2146static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2147{
2148        struct netif_rx_stats *rc = NULL;
2149
2150        while (*pos < NR_CPUS)
2151                if (cpu_online(*pos)) {
2152                        rc = &per_cpu(netdev_rx_stat, *pos);
2153                        break;
2154                } else
2155                        ++*pos;
2156        return rc;
2157}
2158
2159static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2160{
2161        return softnet_get_online(pos);
2162}
2163
2164static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2165{
2166        ++*pos;
2167        return softnet_get_online(pos);
2168}
2169
2170static void softnet_seq_stop(struct seq_file *seq, void *v)
2171{
2172}
2173
2174static int softnet_seq_show(struct seq_file *seq, void *v)
2175{
2176        struct netif_rx_stats *s = v;
2177
2178        seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2179                   s->total, s->dropped, s->time_squeeze, s->throttled,
2180                   s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2181                   s->fastroute_deferred_out,
2182#if 0
2183                   s->fastroute_latency_reduction
2184#else
2185                   s->cpu_collision
2186#endif
2187                  );
2188        return 0;
2189}
2190
2191static struct seq_operations dev_seq_ops = {
2192        .start = dev_seq_start,
2193        .next  = dev_seq_next,
2194        .stop  = dev_seq_stop,
2195        .show  = dev_seq_show,
2196};
2197
2198static int dev_seq_open(struct inode *inode, struct file *file)
2199{
2200        return seq_open(file, &dev_seq_ops);
2201}
2202
2203static struct file_operations dev_seq_fops = {
2204        .owner   = THIS_MODULE,
2205        .open    = dev_seq_open,
2206        .read    = seq_read,
2207        .llseek  = seq_lseek,
2208        .release = seq_release,
2209};
2210
2211static struct seq_operations softnet_seq_ops = {
2212        .start = softnet_seq_start,
2213        .next  = softnet_seq_next,
2214        .stop  = softnet_seq_stop,
2215        .show  = softnet_seq_show,
2216};
2217
2218static int softnet_seq_open(struct inode *inode, struct file *file)
2219{
2220        return seq_open(file, &softnet_seq_ops);
2221}
2222
2223static struct file_operations softnet_seq_fops = {
2224        .owner   = THIS_MODULE,
2225        .open    = softnet_seq_open,
2226        .read    = seq_read,
2227        .llseek  = seq_lseek,
2228        .release = seq_release,
2229};
2230
2231#ifdef WIRELESS_EXT
2232extern int wireless_proc_init(void);
2233#else
2234#define wireless_proc_init() 0
2235#endif
2236
2237static int __init dev_proc_init(void)
2238{
2239        int rc = -ENOMEM;
2240
2241        if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2242                goto out;
2243        if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2244                goto out_dev;
2245        if (wireless_proc_init())
2246                goto out_softnet;
2247        rc = 0;
2248out:
2249        return rc;
2250out_softnet:
2251        proc_net_remove("softnet_stat");
2252out_dev:
2253        proc_net_remove("dev");
2254        goto out;
2255}
2256#else
2257#define dev_proc_init() 0
2258#endif  /* CONFIG_PROC_FS */
2259
2260
2261/**
2262 *      netdev_set_master       -       set up master/slave pair
2263 *      @slave: slave device
2264 *      @master: new master device
2265 *
2266 *      Changes the master device of the slave. Pass %NULL to break the
2267 *      bonding. The caller must hold the RTNL semaphore. On a failure
2268 *      a negative errno code is returned. On success the reference counts
2269 *      are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2270 *      function returns zero.
2271 */
2272int netdev_set_master(struct net_device *slave, struct net_device *master)
2273{
2274        struct net_device *old = slave->master;
2275
2276        ASSERT_RTNL();
2277
2278        if (master) {
2279                if (old)
2280                        return -EBUSY;
2281                dev_hold(master);
2282        }
2283
2284        slave->master = master;
2285        
2286        synchronize_net();
2287
2288        if (old)
2289                dev_put(old);
2290
2291        if (master)
2292                slave->flags |= IFF_SLAVE;
2293        else
2294                slave->flags &= ~IFF_SLAVE;
2295
2296        rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2297        return 0;
2298}
2299
2300/**
2301 *      dev_set_promiscuity     - update promiscuity count on a device
2302 *      @dev: device
2303 *      @inc: modifier
2304 *
2305 *      Add or remove promsicuity from a device. While the count in the device
2306 *      remains above zero the interface remains promiscuous. Once it hits zero
2307 *      the device reverts back to normal filtering operation. A negative inc
2308 *      value is used to drop promiscuity on the device.
2309 */
2310void dev_set_promiscuity(struct net_device *dev, int inc)
2311{
2312        unsigned short old_flags = dev->flags;
2313
2314        dev->flags |= IFF_PROMISC;
2315        if ((dev->promiscuity += inc) == 0)
2316                dev->flags &= ~IFF_PROMISC;
2317        if (dev->flags ^ old_flags) {
2318                dev_mc_upload(dev);
2319                printk(KERN_INFO "device %s %s promiscuous mode\n",
2320                       dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2321                                                               "left");
2322        }
2323}
2324
2325/**
2326 *      dev_set_allmulti        - update allmulti count on a device
2327 *      @dev: device
2328 *      @inc: modifier
2329 *
2330 *      Add or remove reception of all multicast frames to a device. While the
2331 *      count in the device remains above zero the interface remains listening
2332 *      to all interfaces. Once it hits zero the device reverts back to normal
2333 *      filtering operation. A negative @inc value is used to drop the counter
2334 *      when releasing a resource needing all multicasts.
2335 */
2336
2337void dev_set_allmulti(struct net_device *dev, int inc)
2338{
2339        unsigned short old_flags = dev->flags;
2340
2341        dev->flags |= IFF_ALLMULTI;
2342        if ((dev->allmulti += inc) == 0)
2343                dev->flags &= ~IFF_ALLMULTI;
2344        if (dev->flags ^ old_flags)
2345                dev_mc_upload(dev);
2346}
2347
2348unsigned dev_get_flags(const struct net_device *dev)
2349{
2350        unsigned flags;
2351
2352        flags = (dev->flags & ~(IFF_PROMISC |
2353                                IFF_ALLMULTI |
2354                                IFF_RUNNING)) | 
2355                (dev->gflags & (IFF_PROMISC |
2356                                IFF_ALLMULTI));
2357
2358        if (netif_running(dev) && netif_carrier_ok(dev))
2359                flags |= IFF_RUNNING;
2360
2361        return flags;
2362}
2363
2364int dev_change_flags(struct net_device *dev, unsigned flags)
2365{
2366        int ret;
2367        int old_flags = dev->flags;
2368
2369        /*
2370         *      Set the flags on our device.
2371         */
2372
2373        dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2374                               IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2375                               IFF_AUTOMEDIA)) |
2376                     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2377                                    IFF_ALLMULTI));
2378
2379        /*
2380         *      Load in the correct multicast list now the flags have changed.
2381         */
2382
2383        dev_mc_upload(dev);
2384
2385        /*
2386         *      Have we downed the interface. We handle IFF_UP ourselves
2387         *      according to user attempts to set it, rather than blindly
2388         *      setting it.
2389         */
2390
2391        ret = 0;
2392        if ((old_flags ^ flags) & IFF_UP) {     /* Bit is different  ? */
2393                ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2394
2395                if (!ret)
2396                        dev_mc_upload(dev);
2397        }
2398
2399        if (dev->flags & IFF_UP &&
2400            ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2401                                          IFF_VOLATILE)))
2402                notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2403
2404        if ((flags ^ dev->gflags) & IFF_PROMISC) {
2405                int inc = (flags & IFF_PROMISC) ? +1 : -1;
2406                dev->gflags ^= IFF_PROMISC;
2407                dev_set_promiscuity(dev, inc);
2408        }
2409
2410        /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2411           is important. Some (broken) drivers set IFF_PROMISC, when
2412           IFF_ALLMULTI is requested not asking us and not reporting.
2413         */
2414        if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2415                int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2416                dev->gflags ^= IFF_ALLMULTI;
2417                dev_set_allmulti(dev, inc);
2418        }
2419
2420        if (old_flags ^ dev->flags)
2421                rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2422
2423        return ret;
2424}
2425
2426int dev_set_mtu(struct net_device *dev, int new_mtu)
2427{
2428        int err;
2429
2430        if (new_mtu == dev->mtu)
2431                return 0;
2432
2433        /*      MTU must be positive.    */
2434        if (new_mtu < 0)
2435                return -EINVAL;
2436
2437        if (!netif_device_present(dev))
2438                return -ENODEV;
2439
2440        err = 0;
2441        if (dev->change_mtu)
2442                err = dev->change_mtu(dev, new_mtu);
2443        else
2444                dev->mtu = new_mtu;
2445        if (!err && dev->flags & IFF_UP)
2446                notifier_call_chain(&netdev_chain,
2447                                    NETDEV_CHANGEMTU, dev);
2448        return err;
2449}
2450
2451
2452/*
2453 *      Perform the SIOCxIFxxx calls.
2454 */
2455static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2456{
2457        int err;
2458        struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2459
2460        if (!dev)
2461                return -ENODEV;
2462
2463        switch (cmd) {
2464                case SIOCGIFFLAGS:      /* Get interface flags */
2465                        ifr->ifr_flags = dev_get_flags(dev);
2466                        return 0;
2467
2468                case SIOCSIFFLAGS:      /* Set interface flags */
2469                        return dev_change_flags(dev, ifr->ifr_flags);
2470
2471                case SIOCGIFMETRIC:     /* Get the metric on the interface
2472                                           (currently unused) */
2473                        ifr->ifr_metric = 0;
2474                        return 0;
2475
2476                case SIOCSIFMETRIC:     /* Set the metric on the interface
2477                                           (currently unused) */
2478                        return -EOPNOTSUPP;
2479
2480                case SIOCGIFMTU:        /* Get the MTU of a device */
2481                        ifr->ifr_mtu = dev->mtu;
2482                        return 0;
2483
2484                case SIOCSIFMTU:        /* Set the MTU of a device */
2485                        return dev_set_mtu(dev, ifr->ifr_mtu);
2486
2487                case SIOCGIFHWADDR:
2488                        memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2489                               min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2490                        ifr->ifr_hwaddr.sa_family = dev->type;
2491                        return 0;
2492
2493                case SIOCSIFHWADDR:
2494                        if (!dev->set_mac_address)
2495                                return -EOPNOTSUPP;
2496                        if (ifr->ifr_hwaddr.sa_family != dev->type)
2497                                return -EINVAL;
2498                        if (!netif_device_present(dev))
2499                                return -ENODEV;
2500                        err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2501                        if (!err)
2502                                notifier_call_chain(&netdev_chain,
2503                                                    NETDEV_CHANGEADDR, dev);
2504                        return err;
2505
2506                case SIOCSIFHWBROADCAST:
2507                        if (ifr->ifr_hwaddr.sa_family != dev->type)
2508                                return -EINVAL;
2509                        memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2510                               min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2511                        notifier_call_chain(&netdev_chain,
2512                                            NETDEV_CHANGEADDR, dev);
2513                        return 0;
2514
2515                case SIOCGIFMAP:
2516                        ifr->ifr_map.mem_start = dev->mem_start;
2517                        ifr->ifr_map.mem_end   = dev->mem_end;
2518                        ifr->ifr_map.base_addr = dev->base_addr;
2519                        ifr->ifr_map.irq       = dev->irq;
2520                        ifr->ifr_map.dma       = dev->dma;
2521                        ifr->ifr_map.port      = dev->if_port;
2522                        return 0;
2523
2524                case SIOCSIFMAP:
2525                        if (dev->set_config) {
2526                                if (!netif_device_present(dev))
2527                                        return -ENODEV;
2528                                return dev->set_config(dev, &ifr->ifr_map);
2529                        }
2530                        return -EOPNOTSUPP;
2531
2532                case SIOCADDMULTI:
2533                        if (!dev->set_multicast_list ||
2534                            ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2535                                return -EINVAL;
2536                        if (!netif_device_present(dev))
2537                                return -ENODEV;
2538                        return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2539                                          dev->addr_len, 1);
2540
2541                case SIOCDELMULTI:
2542                        if (!dev->set_multicast_list ||
2543                            ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2544                                return -EINVAL;
2545                        if (!netif_device_present(dev))
2546                                return -ENODEV;
2547                        return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2548                                             dev->addr_len, 1);
2549
2550                case SIOCGIFINDEX:
2551                        ifr->ifr_ifindex = dev->ifindex;
2552                        return 0;
2553
2554                case SIOCGIFTXQLEN:
2555                        ifr->ifr_qlen = dev->tx_queue_len;
2556                        return 0;
2557
2558                case SIOCSIFTXQLEN:
2559                        if (ifr->ifr_qlen < 0)
2560                                return -EINVAL;
2561                        dev->tx_queue_len = ifr->ifr_qlen;
2562                        return 0;
2563
2564                case SIOCSIFNAME:
2565                        ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2566                        return dev_change_name(dev, ifr->ifr_newname);
2567
2568                /*
2569                 *      Unknown or private ioctl
2570                 */
2571
2572                default:
2573                        if ((cmd >= SIOCDEVPRIVATE &&
2574                            cmd <= SIOCDEVPRIVATE + 15) ||
2575                            cmd == SIOCBONDENSLAVE ||
2576                            cmd == SIOCBONDRELEASE ||
2577                            cmd == SIOCBONDSETHWADDR ||
2578                            cmd == SIOCBONDSLAVEINFOQUERY ||
2579                            cmd == SIOCBONDINFOQUERY ||
2580                            cmd == SIOCBONDCHANGEACTIVE ||
2581                            cmd == SIOCGMIIPHY ||
2582                            cmd == SIOCGMIIREG ||
2583                            cmd == SIOCSMIIREG ||
2584                            cmd == SIOCBRADDIF ||
2585                            cmd == SIOCBRDELIF ||
2586                            cmd == SIOCWANDEV) {
2587                                err = -EOPNOTSUPP;
2588                                if (dev->do_ioctl) {
2589                                        if (netif_device_present(dev))
2590                                                err = dev->do_ioctl(dev, ifr,
2591                                                                    cmd);
2592                                        else
2593                                                err = -ENODEV;
2594                                }
2595                        } else
2596                                err = -EINVAL;
2597
2598        }
2599        return err;
2600}
2601
2602/*
2603 *      This function handles all "interface"-type I/O control requests. The actual
2604 *      'doing' part of this is dev_ifsioc above.
2605 */
2606
2607/**
2608 *      dev_ioctl       -       network device ioctl
2609 *      @cmd: command to issue
2610 *      @arg: pointer to a struct ifreq in user space
2611 *
2612 *      Issue ioctl functions to devices. This is normally called by the
2613 *      user space syscall interfaces but can sometimes be useful for
2614 *      other purposes. The return value is the return from the syscall if
2615 *      positive or a negative errno code on error.
2616 */
2617
2618int dev_ioctl(unsigned int cmd, void __user *arg)
2619{
2620        struct ifreq ifr;
2621        int ret;
2622        char *colon;
2623
2624        /* One special case: SIOCGIFCONF takes ifconf argument
2625           and requires shared lock, because it sleeps writing
2626           to user space.
2627         */
2628
2629        if (cmd == SIOCGIFCONF) {
2630                rtnl_shlock();
2631                ret = dev_ifconf((char __user *) arg);
2632                rtnl_shunlock();
2633                return ret;
2634        }
2635        if (cmd == SIOCGIFNAME)
2636                return dev_ifname((struct ifreq __user *)arg);
2637
2638        if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2639                return -EFAULT;
2640
2641        ifr.ifr_name[IFNAMSIZ-1] = 0;
2642
2643        colon = strchr(ifr.ifr_name, ':');
2644        if (colon)
2645                *colon = 0;
2646
2647        /*
2648         *      See which interface the caller is talking about.
2649         */
2650
2651        switch (cmd) {
2652                /*
2653                 *      These ioctl calls:
2654                 *      - can be done by all.
2655                 *      - atomic and do not require locking.
2656                 *      - return a value
2657                 */
2658                case SIOCGIFFLAGS:
2659                case SIOCGIFMETRIC:
2660                case SIOCGIFMTU:
2661                case SIOCGIFHWADDR:
2662                case SIOCGIFSLAVE:
2663                case SIOCGIFMAP:
2664                case SIOCGIFINDEX:
2665                case SIOCGIFTXQLEN:
2666                        dev_load(ifr.ifr_name);
2667                        read_lock(&dev_base_lock);
2668                        ret = dev_ifsioc(&ifr, cmd);
2669                        read_unlock(&dev_base_lock);
2670                        if (!ret) {
2671                                if (colon)
2672                                        *colon = ':';
2673                                if (copy_to_user(arg, &ifr,
2674                                                 sizeof(struct ifreq)))
2675                                        ret = -EFAULT;
2676                        }
2677                        return ret;
2678
2679                case SIOCETHTOOL:
2680                        dev_load(ifr.ifr_name);
2681                        rtnl_lock();
2682                        ret = dev_ethtool(&ifr);
2683                        rtnl_unlock();
2684                        if (!ret) {
2685                                if (colon)
2686                                        *colon = ':';
2687                                if (copy_to_user(arg, &ifr,
2688                                                 sizeof(struct ifreq)))
2689                                        ret = -EFAULT;
2690                        }
2691                        return ret;
2692
2693                /*
2694                 *      These ioctl calls:
2695                 *      - require superuser power.
2696                 *      - require strict serialization.
2697                 *      - return a value
2698                 */
2699                case SIOCGMIIPHY:
2700                case SIOCGMIIREG:
2701                case SIOCSIFNAME:
2702                        if (!capable(CAP_NET_ADMIN))
2703                                return -EPERM;
2704                        dev_load(ifr.ifr_name);
2705                        rtnl_lock();
2706                        ret = dev_ifsioc(&ifr, cmd);
2707                        rtnl_unlock();
2708                        if (!ret) {
2709                                if (colon)
2710                                        *colon = ':';
2711                                if (copy_to_user(arg, &ifr,
2712                                                 sizeof(struct ifreq)))
2713                                        ret = -EFAULT;
2714                        }
2715                        return ret;
2716
2717                /*
2718                 *      These ioctl calls:
2719                 *      - require superuser power.
2720                 *      - require strict serialization.
2721                 *      - do not return a value
2722                 */
2723                case SIOCSIFFLAGS:
2724                case SIOCSIFMETRIC:
2725                case SIOCSIFMTU:
2726                case SIOCSIFMAP:
2727                case SIOCSIFHWADDR:
2728                case SIOCSIFSLAVE:
2729                case SIOCADDMULTI:
2730                case SIOCDELMULTI:
2731                case SIOCSIFHWBROADCAST:
2732                case SIOCSIFTXQLEN:
2733                case SIOCSMIIREG:
2734                case SIOCBONDENSLAVE:
2735                case SIOCBONDRELEASE:
2736                case SIOCBONDSETHWADDR:
2737                case SIOCBONDSLAVEINFOQUERY:
2738                case SIOCBONDINFOQUERY:
2739                case SIOCBONDCHANGEACTIVE:
2740                case SIOCBRADDIF:
2741                case SIOCBRDELIF:
2742                        if (!capable(CAP_NET_ADMIN))
2743                                return -EPERM;
2744                        dev_load(ifr.ifr_name);
2745                        rtnl_lock();
2746                        ret = dev_ifsioc(&ifr, cmd);
2747                        rtnl_unlock();
2748                        return ret;
2749
2750                case SIOCGIFMEM:
2751                        /* Get the per device memory space. We can add this but
2752                         * currently do not support it */
2753                case SIOCSIFMEM:
2754                        /* Set the per device memory buffer space.
2755                         * Not applicable in our case */
2756                case SIOCSIFLINK:
2757                        return -EINVAL;
2758
2759                /*
2760                 *      Unknown or private ioctl.
2761                 */
2762                default:
2763                        if (cmd == SIOCWANDEV ||
2764                            (cmd >= SIOCDEVPRIVATE &&
2765                             cmd <= SIOCDEVPRIVATE + 15)) {
2766                                dev_load(ifr.ifr_name);
2767                                rtnl_lock();
2768                                ret = dev_ifsioc(&ifr, cmd);
2769                                rtnl_unlock();
2770                                if (!ret && copy_to_user(arg, &ifr,
2771                                                         sizeof(struct ifreq)))
2772                                        ret = -EFAULT;
2773                                return ret;
2774                        }
2775#ifdef WIRELESS_EXT
2776                        /* Take care of Wireless Extensions */
2777                        if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2778                                /* If command is `set a parameter', or
2779                                 * `get the encoding parameters', check if
2780                                 * the user has the right to do it */
2781                                if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2782                                        if (!capable(CAP_NET_ADMIN))
2783                                                return -EPERM;
2784                                }
2785                                dev_load(ifr.ifr_name);
2786                                rtnl_lock();
2787                                /* Follow me in net/core/wireless.c */
2788                                ret = wireless_process_ioctl(&ifr, cmd);
2789                                rtnl_unlock();
2790                                if (!ret && IW_IS_GET(cmd) &&
2791                                    copy_to_user(arg, &ifr,
2792                                                 sizeof(struct ifreq)))
2793                                        ret = -EFAULT;
2794                                return ret;
2795                        }
2796#endif  /* WIRELESS_EXT */
2797                        return -EINVAL;
2798        }
2799}
2800
2801
2802/**
2803 *      dev_new_index   -       allocate an ifindex
2804 *
2805 *      Returns a suitable unique value for a new device interface
2806 *      number.  The caller must hold the rtnl semaphore or the
2807 *      dev_base_lock to be sure it remains unique.
2808 */
2809int dev_new_index(void)
2810{
2811        static int ifindex;
2812        for (;;) {
2813                if (++ifindex <= 0)
2814                        ifindex = 1;
2815                if (!__dev_get_by_index(ifindex))
2816                        return ifindex;
2817        }
2818}
2819
2820static int dev_boot_phase = 1;
2821
2822/* Delayed registration/unregisteration */
2823static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2824static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2825
2826static inline void net_set_todo(struct net_device *dev)
2827{
2828        spin_lock(&net_todo_list_lock);
2829        list_add_tail(&dev->todo_list, &net_todo_list);
2830        spin_unlock(&net_todo_list_lock);
2831}
2832
2833/**
2834 *      register_netdevice      - register a network device
2835 *      @dev: device to register
2836 *
2837 *      Take a completed network device structure and add it to the kernel
2838 *      interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2839 *      chain. 0 is returned on success. A negative errno code is returned
2840 *      on a failure to set up the device, or if the name is a duplicate.
2841 *
2842 *      Callers must hold the rtnl semaphore.  See the comment at the
2843 *      end of Space.c for details about the locking.  You may want
2844 *      register_netdev() instead of this.
2845 *
2846 *      BUGS:
2847 *      The locking appears insufficient to guarantee two parallel registers
2848 *      will not get the same name.
2849 */
2850
2851int register_netdevice(struct net_device *dev)
2852{
2853        struct hlist_head *head;
2854        struct hlist_node *p;
2855        int ret;
2856
2857        BUG_ON(dev_boot_phase);
2858        ASSERT_RTNL();
2859
2860        /* When net_device's are persistent, this will be fatal. */
2861        BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2862
2863        spin_lock_init(&dev->queue_lock);
2864        spin_lock_init(&dev->xmit_lock);
2865        dev->xmit_lock_owner = -1;
2866#ifdef CONFIG_NET_CLS_ACT
2867        spin_lock_init(&dev->ingress_lock);
2868#endif
2869
2870        ret = alloc_divert_blk(dev);
2871        if (ret)
2872                goto out;
2873
2874        dev->iflink = -1;
2875
2876        /* Init, if this function is available */
2877        if (dev->init) {
2878                ret = dev->init(dev);
2879                if (ret) {
2880                        if (ret > 0)
2881                                ret = -EIO;
2882                        goto out_err;
2883                }
2884        }
2885 
2886        if (!dev_valid_name(dev->name)) {
2887                ret = -EINVAL;
2888                goto out_err;
2889        }
2890
2891        dev->ifindex = dev_new_index();
2892        if (dev->iflink == -1)
2893                dev->iflink = dev->ifindex;
2894
2895        /* Check for existence of name */
2896        head = dev_name_hash(dev->name);
2897        hlist_for_each(p, head) {
2898                struct net_device *d
2899                        = hlist_entry(p, struct net_device, name_hlist);
2900                if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2901                        ret = -EEXIST;
2902                        goto out_err;
2903                }
2904        }
2905
2906        /* Fix illegal SG+CSUM combinations. */
2907        if ((dev->features & NETIF_F_SG) &&
2908            !(dev->features & (NETIF_F_IP_CSUM |
2909                               NETIF_F_NO_CSUM |
2910                               NETIF_F_HW_CSUM))) {
2911                printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2912                       dev->name);
2913                dev->features &= ~NETIF_F_SG;
2914        }
2915
2916        /*
2917         *      nil rebuild_header routine,
2918         *      that should be never called and used as just bug trap.
2919         */
2920
2921        if (!dev->rebuild_header)
2922                dev->rebuild_header = default_rebuild_header;
2923
2924        /*
2925         *      Default initial state at registry is that the
2926         *      device is present.
2927         */
2928
2929        set_bit(__LINK_STATE_PRESENT, &dev->state);
2930
2931        dev->next = NULL;
2932        dev_init_scheduler(dev);
2933        write_lock_bh(&dev_base_lock);
2934        *dev_tail = dev;
2935        dev_tail = &dev->next;
2936        hlist_add_head(&dev->name_hlist, head);
2937        hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2938        dev_hold(dev);
2939        dev->reg_state = NETREG_REGISTERING;
2940        write_unlock_bh(&dev_base_lock);
2941
2942        /* Notify protocols, that a new device appeared. */
2943        notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2944
2945        /* Finish registration after unlock */
2946        net_set_todo(dev);
2947        ret = 0;
2948
2949out:
2950        return ret;
2951out_err:
2952        free_divert_blk(dev);
2953        goto out;
2954}
2955
2956/*
2957 * netdev_wait_allrefs - wait until all references are gone.
2958 *
2959 * This is called when unregistering network devices.
2960 *
2961 * Any protocol or device that holds a reference should register
2962 * for netdevice notification, and cleanup and put back the
2963 * reference if they receive an UNREGISTER event.
2964 * We can get stuck here if buggy protocols don't correctly
2965 * call dev_put. 
2966 */
2967static void netdev_wait_allrefs(struct net_device *dev)
2968{
2969        unsigned long rebroadcast_time, warning_time;
2970
2971        rebroadcast_time = warning_time = jiffies;
2972        while (atomic_read(&dev->refcnt) != 0) {
2973                if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2974                        rtnl_shlock();
2975
2976                        /* Rebroadcast unregister notification */
2977                        notifier_call_chain(&netdev_chain,
2978                                            NETDEV_UNREGISTER, dev);
2979
2980                        if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2981                                     &dev->state)) {
2982                                /* We must not have linkwatch events
2983                                 * pending on unregister. If this
2984                                 * happens, we simply run the queue
2985                                 * unscheduled, resulting in a noop
2986                                 * for this device.
2987                                 */
2988                                linkwatch_run_queue();
2989                        }
2990
2991                        rtnl_shunlock();
2992
2993                        rebroadcast_time = jiffies;
2994                }
2995
2996                current->state = TASK_INTERRUPTIBLE;
2997                schedule_timeout(HZ / 4);
2998
2999                if (time_after(jiffies, warning_time + 10 * HZ)) {
3000                        printk(KERN_EMERG "unregister_netdevice: "
3001                               "waiting for %s to become free. Usage "
3002                               "count = %d\n",
3003                               dev->name, atomic_read(&dev->refcnt));
3004                        warning_time = jiffies;
3005                }
3006        }
3007}
3008
3009/* The sequence is:
3010 *
3011 *      rtnl_lock();
3012 *      ...
3013 *      register_netdevice(x1);
3014 *      register_netdevice(x2);
3015 *      ...
3016 *      unregister_netdevice(y1);
3017 *      unregister_netdevice(y2);
3018 *      ...
3019 *      rtnl_unlock();
3020 *      free_netdev(y1);
3021 *      free_netdev(y2);
3022 *
3023 * We are invoked by rtnl_unlock() after it drops the semaphore.
3024 * This allows us to deal with problems:
3025 * 1) We can create/delete sysfs objects which invoke hotplug
3026 *    without deadlocking with linkwatch via keventd.
3027 * 2) Since we run with the RTNL semaphore not held, we can sleep
3028 *    safely in order to wait for the netdev refcnt to drop to zero.
3029 */
3030static DECLARE_MUTEX(net_todo_run_mutex);
3031void netdev_run_todo(void)
3032{
3033        struct list_head list = LIST_HEAD_INIT(list);
3034        int err;
3035
3036
3037        /* Need to guard against multiple cpu's getting out of order. */
3038        down(&net_todo_run_mutex);
3039
3040        /* Not safe to do outside the semaphore.  We must not return
3041         * until all unregister events invoked by the local processor
3042         * have been completed (either by this todo run, or one on
3043         * another cpu).
3044         */
3045        if (list_empty(&net_todo_list))
3046                goto out;
3047
3048        /* Snapshot list, allow later requests */
3049        spin_lock(&net_todo_list_lock);
3050        list_splice_init(&net_todo_list, &list);
3051        spin_unlock(&net_todo_list_lock);
3052                
3053        while (!list_empty(&list)) {
3054                struct net_device *dev
3055                        = list_entry(list.next, struct net_device, todo_list);
3056                list_del(&dev->todo_list);
3057
3058                switch(dev->reg_state) {
3059                case NETREG_REGISTERING:
3060                        err = netdev_register_sysfs(dev);
3061                        if (err)
3062                                printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3063                                       dev->name, err);
3064                        dev->reg_state = NETREG_REGISTERED;
3065                        break;
3066
3067                case NETREG_UNREGISTERING:
3068                        netdev_unregister_sysfs(dev);
3069                        dev->reg_state = NETREG_UNREGISTERED;
3070
3071                        netdev_wait_allrefs(dev);
3072
3073                        /* paranoia */
3074                        BUG_ON(atomic_read(&dev->refcnt));
3075                        BUG_TRAP(!dev->ip_ptr);
3076                        BUG_TRAP(!dev->ip6_ptr);
3077                        BUG_TRAP(!dev->dn_ptr);
3078
3079
3080                        /* It must be the very last action, 
3081                         * after this 'dev' may point to freed up memory.
3082                         */
3083                        if (dev->destructor)
3084                                dev->destructor(dev);
3085                        break;
3086
3087                default:
3088                        printk(KERN_ERR "network todo '%s' but state %d\n",
3089                               dev->name, dev->reg_state);
3090                        break;
3091                }
3092        }
3093
3094out:
3095        up(&net_todo_run_mutex);
3096}
3097
3098/**
3099 *      free_netdev - free network device
3100 *      @dev: device
3101 *
3102 *      This function does the last stage of destroying an allocated device 
3103 *      interface. The reference to the device object is released.  
3104 *      If this is the last reference then it will be freed.
3105 */
3106void free_netdev(struct net_device *dev)
3107{
3108#ifdef CONFIG_SYSFS
3109        /*  Compatiablity with error handling in drivers */
3110        if (dev->reg_state == NETREG_UNINITIALIZED) {
3111                kfree((char *)dev - dev->padded);
3112                return;
3113        }
3114
3115        BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3116        dev->reg_state = NETREG_RELEASED;
3117
3118        /* will free via class release */
3119        class_device_put(&dev->class_dev);
3120#else
3121        kfree((char *)dev - dev->padded);
3122#endif
3123}
3124 
3125/* Synchronize with packet receive processing. */
3126void synchronize_net(void) 
3127{
3128        might_sleep();
3129        synchronize_kernel();
3130}
3131
3132/**
3133 *      unregister_netdevice - remove device from the kernel
3134 *      @dev: device
3135 *
3136 *      This function shuts down a device interface and removes it
3137 *      from the kernel tables. On success 0 is returned, on a failure
3138 *      a negative errno code is returned.
3139 *
3140 *      Callers must hold the rtnl semaphore.  See the comment at the
3141 *      end of Space.c for details about the locking.  You may want
3142 *      unregister_netdev() instead of this.
3143 */
3144
3145int unregister_netdevice(struct net_device *dev)
3146{
3147        struct net_device *d, **dp;
3148
3149        BUG_ON(dev_boot_phase);
3150        ASSERT_RTNL();
3151
3152        /* Some devices call without registering for initialization unwind. */
3153        if (dev->reg_state == NETREG_UNINITIALIZED) {
3154                printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3155                                  "was registered\n", dev->name, dev);
3156                return -ENODEV;
3157        }
3158
3159        BUG_ON(dev->reg_state != NETREG_REGISTERED);
3160
3161        /* If device is running, close it first. */
3162        if (dev->flags & IFF_UP)
3163                dev_close(dev);
3164
3165        /* And unlink it from device chain. */
3166        for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3167                if (d == dev) {
3168                        write_lock_bh(&dev_base_lock);
3169                        hlist_del(&dev->name_hlist);
3170                        hlist_del(&dev->index_hlist);
3171                        if (dev_tail == &dev->next)
3172                                dev_tail = dp;
3173                        *dp = d->next;
3174                        write_unlock_bh(&dev_base_lock);
3175                        break;
3176                }
3177        }
3178        if (!d) {
3179                printk(KERN_ERR "unregister net_device: '%s' not found\n",
3180                       dev->name);
3181                return -ENODEV;
3182        }
3183
3184        dev->reg_state = NETREG_UNREGISTERING;
3185
3186        synchronize_net();
3187
3188        /* Shutdown queueing discipline. */
3189        dev_shutdown(dev);
3190
3191        
3192        /* Notify protocols, that we are about to destroy
3193           this device. They should clean all the things.
3194        */
3195        notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3196        
3197        /*
3198         *      Flush the multicast chain
3199         */
3200        dev_mc_discard(dev);
3201
3202        if (dev->uninit)
3203                dev->uninit(dev);
3204
3205        /* Notifier chain MUST detach us from master device. */
3206        BUG_TRAP(!dev->master);
3207
3208        free_divert_blk(dev);
3209
3210        /* Finish processing unregister after unlock */
3211        net_set_todo(dev);
3212
3213        synchronize_net();
3214
3215        dev_put(dev);
3216        return 0;
3217}
3218
3219#ifdef CONFIG_HOTPLUG_CPU
3220static int dev_cpu_callback(struct notifier_block *nfb,
3221                            unsigned long action,
3222                            void *ocpu)
3223{
3224        struct sk_buff **list_skb;
3225        struct net_device **list_net;
3226        struct sk_buff *skb;
3227        unsigned int cpu, oldcpu = (unsigned long)ocpu;
3228        struct softnet_data *sd, *oldsd;
3229
3230        if (action != CPU_DEAD)
3231                return NOTIFY_OK;
3232
3233        local_irq_disable();
3234        cpu = smp_processor_id();
3235        sd = &per_cpu(softnet_data, cpu);
3236        oldsd = &per_cpu(softnet_data, oldcpu);
3237
3238        /* Find end of our completion_queue. */
3239        list_skb = &sd->completion_queue;
3240        while (*list_skb)
3241                list_skb = &(*list_skb)->next;
3242        /* Append completion queue from offline CPU. */
3243        *list_skb = oldsd->completion_queue;
3244        oldsd->completion_queue = NULL;
3245
3246        /* Find end of our output_queue. */
3247        list_net = &sd->output_queue;
3248        while (*list_net)
3249                list_net = &(*list_net)->next_sched;
3250        /* Append output queue from offline CPU. */
3251        *list_net = oldsd->output_queue;
3252        oldsd->output_queue = NULL;
3253
3254        raise_softirq_irqoff(NET_TX_SOFTIRQ);
3255        local_irq_enable();
3256
3257        /* Process offline CPU's input_pkt_queue */
3258        while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3259                netif_rx(skb);
3260
3261        return NOTIFY_OK;
3262}
3263#endif /* CONFIG_HOTPLUG_CPU */
3264
3265
3266/*
3267 *      Initialize the DEV module. At boot time this walks the device list and
3268 *      unhooks any devices that fail to initialise (normally hardware not
3269 *      present) and leaves us with a valid list of present and active devices.
3270 *
3271 */
3272
3273/*
3274 *       This is called single threaded during boot, so no need
3275 *       to take the rtnl semaphore.
3276 */
3277static int __init net_dev_init(void)
3278{
3279        int i, rc = -ENOMEM;
3280
3281        BUG_ON(!dev_boot_phase);
3282
3283        if (dev_proc_init())
3284                goto out;
3285
3286        if (netdev_sysfs_init())
3287                goto out;
3288
3289        INIT_LIST_HEAD(&ptype_all);
3290        for (i = 0; i < 16; i++) 
3291                INIT_LIST_HEAD(&ptype_base[i]);
3292
3293        for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3294                INIT_HLIST_HEAD(&dev_name_head[i]);
3295
3296        for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3297                INIT_HLIST_HEAD(&dev_index_head[i]);
3298
3299        /*
3300         *      Initialise the packet receive queues.
3301         */
3302
3303        for (i = 0; i < NR_CPUS; i++) {
3304                struct softnet_data *queue;
3305
3306                queue = &per_cpu(softnet_data, i);
3307                skb_queue_head_init(&queue->input_pkt_queue);
3308                queue->throttle = 0;
3309                queue->cng_level = 0;
3310                queue->avg_blog = 10; /* arbitrary non-zero */
3311                queue->completion_queue = NULL;
3312                INIT_LIST_HEAD(&queue->poll_list);
3313                set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3314                queue->backlog_dev.weight = weight_p;
3315                queue->backlog_dev.poll = process_backlog;
3316                atomic_set(&queue->backlog_dev.refcnt, 1);
3317        }
3318
3319#ifdef OFFLINE_SAMPLE
3320        samp_timer.expires = jiffies + (10 * HZ);
3321        add_timer(&samp_timer);
3322#endif
3323
3324        dev_boot_phase = 0;
3325
3326        open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3327        open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3328
3329        hotcpu_notifier(dev_cpu_callback, 0);
3330        dst_init();
3331        dev_mcast_init();
3332        rc = 0;
3333out:
3334        return rc;
3335}
3336
3337subsys_initcall(net_dev_init);
3338
3339EXPORT_SYMBOL(__dev_get);
3340EXPORT_SYMBOL(__dev_get_by_flags);
3341EXPORT_SYMBOL(__dev_get_by_index);
3342EXPORT_SYMBOL(__dev_get_by_name);
3343EXPORT_SYMBOL(__dev_remove_pack);
3344EXPORT_SYMBOL(__skb_linearize);
3345EXPORT_SYMBOL(call_netdevice_notifiers);
3346EXPORT_SYMBOL(dev_add_pack);
3347EXPORT_SYMBOL(dev_alloc_name);
3348EXPORT_SYMBOL(dev_close);
3349EXPORT_SYMBOL(dev_get_by_flags);
3350EXPORT_SYMBOL(dev_get_by_index);
3351EXPORT_SYMBOL(dev_get_by_name);
3352EXPORT_SYMBOL(dev_getbyhwaddr);
3353EXPORT_SYMBOL(dev_ioctl);
3354EXPORT_SYMBOL(dev_new_index);
3355EXPORT_SYMBOL(dev_open);
3356EXPORT_SYMBOL(dev_queue_xmit);
3357EXPORT_SYMBOL(dev_queue_xmit_nit);
3358EXPORT_SYMBOL(dev_remove_pack);
3359EXPORT_SYMBOL(dev_set_allmulti);
3360EXPORT_SYMBOL(dev_set_promiscuity);
3361EXPORT_SYMBOL(dev_change_flags);
3362EXPORT_SYMBOL(dev_set_mtu);
3363EXPORT_SYMBOL(free_netdev);
3364EXPORT_SYMBOL(netdev_boot_setup_check);
3365EXPORT_SYMBOL(netdev_set_master);
3366EXPORT_SYMBOL(netdev_state_change);
3367EXPORT_SYMBOL(netif_receive_skb);
3368EXPORT_SYMBOL(netif_rx);
3369EXPORT_SYMBOL(register_gifconf);
3370EXPORT_SYMBOL(register_netdevice);
3371EXPORT_SYMBOL(register_netdevice_notifier);
3372EXPORT_SYMBOL(skb_checksum_help);
3373EXPORT_SYMBOL(synchronize_net);
3374EXPORT_SYMBOL(unregister_netdevice);
3375EXPORT_SYMBOL(unregister_netdevice_notifier);
3376
3377#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3378EXPORT_SYMBOL(br_handle_frame_hook);
3379#endif
3380
3381#ifdef CONFIG_KMOD
3382EXPORT_SYMBOL(dev_load);
3383#endif
3384#ifdef CONFIG_NET_HW_FLOWCONTROL
3385EXPORT_SYMBOL(netdev_dropping);
3386EXPORT_SYMBOL(netdev_fc_xoff);
3387EXPORT_SYMBOL(netdev_register_fc);
3388EXPORT_SYMBOL(netdev_unregister_fc);
3389#endif
3390
3391#ifdef CONFIG_NET_CLS_ACT
3392EXPORT_SYMBOL(ing_filter);
3393#endif
3394
3395
3396EXPORT_PER_CPU_SYMBOL(softnet_data);
3397
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.