at v2.6.17 84 kB view raw
1/* 2 * NET3 Protocol independent device support routines. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Derived from the non IP parts of dev.c 1.0.19 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Mark Evans, <evansmp@uhura.aston.ac.uk> 13 * 14 * Additional Authors: 15 * Florian la Roche <rzsfl@rz.uni-sb.de> 16 * Alan Cox <gw4pts@gw4pts.ampr.org> 17 * David Hinds <dahinds@users.sourceforge.net> 18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 19 * Adam Sulmicki <adam@cfar.umd.edu> 20 * Pekka Riikonen <priikone@poesidon.pspt.fi> 21 * 22 * Changes: 23 * D.J. Barrow : Fixed bug where dev->refcnt gets set 24 * to 2 if register_netdev gets called 25 * before net_dev_init & also removed a 26 * few lines of code in the process. 27 * Alan Cox : device private ioctl copies fields back. 28 * Alan Cox : Transmit queue code does relevant 29 * stunts to keep the queue safe. 30 * Alan Cox : Fixed double lock. 31 * Alan Cox : Fixed promisc NULL pointer trap 32 * ???????? : Support the full private ioctl range 33 * Alan Cox : Moved ioctl permission check into 34 * drivers 35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 36 * Alan Cox : 100 backlog just doesn't cut it when 37 * you start doing multicast video 8) 38 * Alan Cox : Rewrote net_bh and list manager. 39 * Alan Cox : Fix ETH_P_ALL echoback lengths. 40 * Alan Cox : Took out transmit every packet pass 41 * Saved a few bytes in the ioctl handler 42 * Alan Cox : Network driver sets packet type before 43 * calling netif_rx. Saves a function 44 * call a packet. 45 * Alan Cox : Hashed net_bh() 46 * Richard Kooijman: Timestamp fixes. 47 * Alan Cox : Wrong field in SIOCGIFDSTADDR 48 * Alan Cox : Device lock protection. 49 * Alan Cox : Fixed nasty side effect of device close 50 * changes. 51 * Rudi Cilibrasi : Pass the right thing to 52 * set_mac_address() 53 * Dave Miller : 32bit quantity for the device lock to 54 * make it work out on a Sparc. 55 * Bjorn Ekwall : Added KERNELD hack. 56 * Alan Cox : Cleaned up the backlog initialise. 57 * Craig Metz : SIOCGIFCONF fix if space for under 58 * 1 device. 59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 60 * is no device open function. 61 * Andi Kleen : Fix error reporting for SIOCGIFCONF 62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 63 * Cyrus Durgin : Cleaned for KMOD 64 * Adam Sulmicki : Bug Fix : Network Device Unload 65 * A network device unload needs to purge 66 * the backlog queue. 67 * Paul Rusty Russell : SIOCSIFNAME 68 * Pekka Riikonen : Netdev boot-time settings code 69 * Andrew Morton : Make unregister_netdevice wait 70 * indefinitely on dev->refcnt 71 * J Hadi Salim : - Backlog queue sampling 72 * - netif_rx() feedback 73 */ 74 75#include <asm/uaccess.h> 76#include <asm/system.h> 77#include <linux/bitops.h> 78#include <linux/capability.h> 79#include <linux/config.h> 80#include <linux/cpu.h> 81#include <linux/types.h> 82#include <linux/kernel.h> 83#include <linux/sched.h> 84#include <linux/mutex.h> 85#include <linux/string.h> 86#include <linux/mm.h> 87#include <linux/socket.h> 88#include <linux/sockios.h> 89#include <linux/errno.h> 90#include <linux/interrupt.h> 91#include <linux/if_ether.h> 92#include <linux/netdevice.h> 93#include <linux/etherdevice.h> 94#include <linux/notifier.h> 95#include <linux/skbuff.h> 96#include <net/sock.h> 97#include <linux/rtnetlink.h> 98#include <linux/proc_fs.h> 99#include <linux/seq_file.h> 100#include <linux/stat.h> 101#include <linux/if_bridge.h> 102#include <linux/divert.h> 103#include <net/dst.h> 104#include <net/pkt_sched.h> 105#include <net/checksum.h> 106#include <linux/highmem.h> 107#include <linux/init.h> 108#include <linux/kmod.h> 109#include <linux/module.h> 110#include <linux/kallsyms.h> 111#include <linux/netpoll.h> 112#include <linux/rcupdate.h> 113#include <linux/delay.h> 114#include <linux/wireless.h> 115#include <net/iw_handler.h> 116#include <asm/current.h> 117#include <linux/audit.h> 118 119/* 120 * The list of packet types we will receive (as opposed to discard) 121 * and the routines to invoke. 122 * 123 * Why 16. Because with 16 the only overlap we get on a hash of the 124 * low nibble of the protocol value is RARP/SNAP/X.25. 125 * 126 * NOTE: That is no longer true with the addition of VLAN tags. Not 127 * sure which should go first, but I bet it won't make much 128 * difference if we are running VLANs. The good news is that 129 * this protocol won't be in the list unless compiled in, so 130 * the average user (w/out VLANs) will not be adversely affected. 131 * --BLG 132 * 133 * 0800 IP 134 * 8100 802.1Q VLAN 135 * 0001 802.3 136 * 0002 AX.25 137 * 0004 802.2 138 * 8035 RARP 139 * 0005 SNAP 140 * 0805 X.25 141 * 0806 ARP 142 * 8137 IPX 143 * 0009 Localtalk 144 * 86DD IPv6 145 */ 146 147static DEFINE_SPINLOCK(ptype_lock); 148static struct list_head ptype_base[16]; /* 16 way hashed list */ 149static struct list_head ptype_all; /* Taps */ 150 151/* 152 * The @dev_base list is protected by @dev_base_lock and the rtnl 153 * semaphore. 154 * 155 * Pure readers hold dev_base_lock for reading. 156 * 157 * Writers must hold the rtnl semaphore while they loop through the 158 * dev_base list, and hold dev_base_lock for writing when they do the 159 * actual updates. This allows pure readers to access the list even 160 * while a writer is preparing to update it. 161 * 162 * To put it another way, dev_base_lock is held for writing only to 163 * protect against pure readers; the rtnl semaphore provides the 164 * protection against other writers. 165 * 166 * See, for example usages, register_netdevice() and 167 * unregister_netdevice(), which must be called with the rtnl 168 * semaphore held. 169 */ 170struct net_device *dev_base; 171static struct net_device **dev_tail = &dev_base; 172DEFINE_RWLOCK(dev_base_lock); 173 174EXPORT_SYMBOL(dev_base); 175EXPORT_SYMBOL(dev_base_lock); 176 177#define NETDEV_HASHBITS 8 178static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS]; 179static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS]; 180 181static inline struct hlist_head *dev_name_hash(const char *name) 182{ 183 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); 184 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)]; 185} 186 187static inline struct hlist_head *dev_index_hash(int ifindex) 188{ 189 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)]; 190} 191 192/* 193 * Our notifier list 194 */ 195 196static RAW_NOTIFIER_HEAD(netdev_chain); 197 198/* 199 * Device drivers call our routines to queue packets here. We empty the 200 * queue in the local softnet handler. 201 */ 202DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL }; 203 204#ifdef CONFIG_SYSFS 205extern int netdev_sysfs_init(void); 206extern int netdev_register_sysfs(struct net_device *); 207extern void netdev_unregister_sysfs(struct net_device *); 208#else 209#define netdev_sysfs_init() (0) 210#define netdev_register_sysfs(dev) (0) 211#define netdev_unregister_sysfs(dev) do { } while(0) 212#endif 213 214 215/******************************************************************************* 216 217 Protocol management and registration routines 218 219*******************************************************************************/ 220 221/* 222 * For efficiency 223 */ 224 225int netdev_nit; 226 227/* 228 * Add a protocol ID to the list. Now that the input handler is 229 * smarter we can dispense with all the messy stuff that used to be 230 * here. 231 * 232 * BEWARE!!! Protocol handlers, mangling input packets, 233 * MUST BE last in hash buckets and checking protocol handlers 234 * MUST start from promiscuous ptype_all chain in net_bh. 235 * It is true now, do not change it. 236 * Explanation follows: if protocol handler, mangling packet, will 237 * be the first on list, it is not able to sense, that packet 238 * is cloned and should be copied-on-write, so that it will 239 * change it and subsequent readers will get broken packet. 240 * --ANK (980803) 241 */ 242 243/** 244 * dev_add_pack - add packet handler 245 * @pt: packet type declaration 246 * 247 * Add a protocol handler to the networking stack. The passed &packet_type 248 * is linked into kernel lists and may not be freed until it has been 249 * removed from the kernel lists. 250 * 251 * This call does not sleep therefore it can not 252 * guarantee all CPU's that are in middle of receiving packets 253 * will see the new packet type (until the next received packet). 254 */ 255 256void dev_add_pack(struct packet_type *pt) 257{ 258 int hash; 259 260 spin_lock_bh(&ptype_lock); 261 if (pt->type == htons(ETH_P_ALL)) { 262 netdev_nit++; 263 list_add_rcu(&pt->list, &ptype_all); 264 } else { 265 hash = ntohs(pt->type) & 15; 266 list_add_rcu(&pt->list, &ptype_base[hash]); 267 } 268 spin_unlock_bh(&ptype_lock); 269} 270 271/** 272 * __dev_remove_pack - remove packet handler 273 * @pt: packet type declaration 274 * 275 * Remove a protocol handler that was previously added to the kernel 276 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 277 * from the kernel lists and can be freed or reused once this function 278 * returns. 279 * 280 * The packet type might still be in use by receivers 281 * and must not be freed until after all the CPU's have gone 282 * through a quiescent state. 283 */ 284void __dev_remove_pack(struct packet_type *pt) 285{ 286 struct list_head *head; 287 struct packet_type *pt1; 288 289 spin_lock_bh(&ptype_lock); 290 291 if (pt->type == htons(ETH_P_ALL)) { 292 netdev_nit--; 293 head = &ptype_all; 294 } else 295 head = &ptype_base[ntohs(pt->type) & 15]; 296 297 list_for_each_entry(pt1, head, list) { 298 if (pt == pt1) { 299 list_del_rcu(&pt->list); 300 goto out; 301 } 302 } 303 304 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); 305out: 306 spin_unlock_bh(&ptype_lock); 307} 308/** 309 * dev_remove_pack - remove packet handler 310 * @pt: packet type declaration 311 * 312 * Remove a protocol handler that was previously added to the kernel 313 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 314 * from the kernel lists and can be freed or reused once this function 315 * returns. 316 * 317 * This call sleeps to guarantee that no CPU is looking at the packet 318 * type after return. 319 */ 320void dev_remove_pack(struct packet_type *pt) 321{ 322 __dev_remove_pack(pt); 323 324 synchronize_net(); 325} 326 327/****************************************************************************** 328 329 Device Boot-time Settings Routines 330 331*******************************************************************************/ 332 333/* Boot time configuration table */ 334static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 335 336/** 337 * netdev_boot_setup_add - add new setup entry 338 * @name: name of the device 339 * @map: configured settings for the device 340 * 341 * Adds new setup entry to the dev_boot_setup list. The function 342 * returns 0 on error and 1 on success. This is a generic routine to 343 * all netdevices. 344 */ 345static int netdev_boot_setup_add(char *name, struct ifmap *map) 346{ 347 struct netdev_boot_setup *s; 348 int i; 349 350 s = dev_boot_setup; 351 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 352 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 353 memset(s[i].name, 0, sizeof(s[i].name)); 354 strcpy(s[i].name, name); 355 memcpy(&s[i].map, map, sizeof(s[i].map)); 356 break; 357 } 358 } 359 360 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 361} 362 363/** 364 * netdev_boot_setup_check - check boot time settings 365 * @dev: the netdevice 366 * 367 * Check boot time settings for the device. 368 * The found settings are set for the device to be used 369 * later in the device probing. 370 * Returns 0 if no settings found, 1 if they are. 371 */ 372int netdev_boot_setup_check(struct net_device *dev) 373{ 374 struct netdev_boot_setup *s = dev_boot_setup; 375 int i; 376 377 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 378 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 379 !strncmp(dev->name, s[i].name, strlen(s[i].name))) { 380 dev->irq = s[i].map.irq; 381 dev->base_addr = s[i].map.base_addr; 382 dev->mem_start = s[i].map.mem_start; 383 dev->mem_end = s[i].map.mem_end; 384 return 1; 385 } 386 } 387 return 0; 388} 389 390 391/** 392 * netdev_boot_base - get address from boot time settings 393 * @prefix: prefix for network device 394 * @unit: id for network device 395 * 396 * Check boot time settings for the base address of device. 397 * The found settings are set for the device to be used 398 * later in the device probing. 399 * Returns 0 if no settings found. 400 */ 401unsigned long netdev_boot_base(const char *prefix, int unit) 402{ 403 const struct netdev_boot_setup *s = dev_boot_setup; 404 char name[IFNAMSIZ]; 405 int i; 406 407 sprintf(name, "%s%d", prefix, unit); 408 409 /* 410 * If device already registered then return base of 1 411 * to indicate not to probe for this interface 412 */ 413 if (__dev_get_by_name(name)) 414 return 1; 415 416 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 417 if (!strcmp(name, s[i].name)) 418 return s[i].map.base_addr; 419 return 0; 420} 421 422/* 423 * Saves at boot time configured settings for any netdevice. 424 */ 425int __init netdev_boot_setup(char *str) 426{ 427 int ints[5]; 428 struct ifmap map; 429 430 str = get_options(str, ARRAY_SIZE(ints), ints); 431 if (!str || !*str) 432 return 0; 433 434 /* Save settings */ 435 memset(&map, 0, sizeof(map)); 436 if (ints[0] > 0) 437 map.irq = ints[1]; 438 if (ints[0] > 1) 439 map.base_addr = ints[2]; 440 if (ints[0] > 2) 441 map.mem_start = ints[3]; 442 if (ints[0] > 3) 443 map.mem_end = ints[4]; 444 445 /* Add new entry to the list */ 446 return netdev_boot_setup_add(str, &map); 447} 448 449__setup("netdev=", netdev_boot_setup); 450 451/******************************************************************************* 452 453 Device Interface Subroutines 454 455*******************************************************************************/ 456 457/** 458 * __dev_get_by_name - find a device by its name 459 * @name: name to find 460 * 461 * Find an interface by name. Must be called under RTNL semaphore 462 * or @dev_base_lock. If the name is found a pointer to the device 463 * is returned. If the name is not found then %NULL is returned. The 464 * reference counters are not incremented so the caller must be 465 * careful with locks. 466 */ 467 468struct net_device *__dev_get_by_name(const char *name) 469{ 470 struct hlist_node *p; 471 472 hlist_for_each(p, dev_name_hash(name)) { 473 struct net_device *dev 474 = hlist_entry(p, struct net_device, name_hlist); 475 if (!strncmp(dev->name, name, IFNAMSIZ)) 476 return dev; 477 } 478 return NULL; 479} 480 481/** 482 * dev_get_by_name - find a device by its name 483 * @name: name to find 484 * 485 * Find an interface by name. This can be called from any 486 * context and does its own locking. The returned handle has 487 * the usage count incremented and the caller must use dev_put() to 488 * release it when it is no longer needed. %NULL is returned if no 489 * matching device is found. 490 */ 491 492struct net_device *dev_get_by_name(const char *name) 493{ 494 struct net_device *dev; 495 496 read_lock(&dev_base_lock); 497 dev = __dev_get_by_name(name); 498 if (dev) 499 dev_hold(dev); 500 read_unlock(&dev_base_lock); 501 return dev; 502} 503 504/** 505 * __dev_get_by_index - find a device by its ifindex 506 * @ifindex: index of device 507 * 508 * Search for an interface by index. Returns %NULL if the device 509 * is not found or a pointer to the device. The device has not 510 * had its reference counter increased so the caller must be careful 511 * about locking. The caller must hold either the RTNL semaphore 512 * or @dev_base_lock. 513 */ 514 515struct net_device *__dev_get_by_index(int ifindex) 516{ 517 struct hlist_node *p; 518 519 hlist_for_each(p, dev_index_hash(ifindex)) { 520 struct net_device *dev 521 = hlist_entry(p, struct net_device, index_hlist); 522 if (dev->ifindex == ifindex) 523 return dev; 524 } 525 return NULL; 526} 527 528 529/** 530 * dev_get_by_index - find a device by its ifindex 531 * @ifindex: index of device 532 * 533 * Search for an interface by index. Returns NULL if the device 534 * is not found or a pointer to the device. The device returned has 535 * had a reference added and the pointer is safe until the user calls 536 * dev_put to indicate they have finished with it. 537 */ 538 539struct net_device *dev_get_by_index(int ifindex) 540{ 541 struct net_device *dev; 542 543 read_lock(&dev_base_lock); 544 dev = __dev_get_by_index(ifindex); 545 if (dev) 546 dev_hold(dev); 547 read_unlock(&dev_base_lock); 548 return dev; 549} 550 551/** 552 * dev_getbyhwaddr - find a device by its hardware address 553 * @type: media type of device 554 * @ha: hardware address 555 * 556 * Search for an interface by MAC address. Returns NULL if the device 557 * is not found or a pointer to the device. The caller must hold the 558 * rtnl semaphore. The returned device has not had its ref count increased 559 * and the caller must therefore be careful about locking 560 * 561 * BUGS: 562 * If the API was consistent this would be __dev_get_by_hwaddr 563 */ 564 565struct net_device *dev_getbyhwaddr(unsigned short type, char *ha) 566{ 567 struct net_device *dev; 568 569 ASSERT_RTNL(); 570 571 for (dev = dev_base; dev; dev = dev->next) 572 if (dev->type == type && 573 !memcmp(dev->dev_addr, ha, dev->addr_len)) 574 break; 575 return dev; 576} 577 578EXPORT_SYMBOL(dev_getbyhwaddr); 579 580struct net_device *dev_getfirstbyhwtype(unsigned short type) 581{ 582 struct net_device *dev; 583 584 rtnl_lock(); 585 for (dev = dev_base; dev; dev = dev->next) { 586 if (dev->type == type) { 587 dev_hold(dev); 588 break; 589 } 590 } 591 rtnl_unlock(); 592 return dev; 593} 594 595EXPORT_SYMBOL(dev_getfirstbyhwtype); 596 597/** 598 * dev_get_by_flags - find any device with given flags 599 * @if_flags: IFF_* values 600 * @mask: bitmask of bits in if_flags to check 601 * 602 * Search for any interface with the given flags. Returns NULL if a device 603 * is not found or a pointer to the device. The device returned has 604 * had a reference added and the pointer is safe until the user calls 605 * dev_put to indicate they have finished with it. 606 */ 607 608struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask) 609{ 610 struct net_device *dev; 611 612 read_lock(&dev_base_lock); 613 for (dev = dev_base; dev != NULL; dev = dev->next) { 614 if (((dev->flags ^ if_flags) & mask) == 0) { 615 dev_hold(dev); 616 break; 617 } 618 } 619 read_unlock(&dev_base_lock); 620 return dev; 621} 622 623/** 624 * dev_valid_name - check if name is okay for network device 625 * @name: name string 626 * 627 * Network device names need to be valid file names to 628 * to allow sysfs to work 629 */ 630int dev_valid_name(const char *name) 631{ 632 return !(*name == '\0' 633 || !strcmp(name, ".") 634 || !strcmp(name, "..") 635 || strchr(name, '/')); 636} 637 638/** 639 * dev_alloc_name - allocate a name for a device 640 * @dev: device 641 * @name: name format string 642 * 643 * Passed a format string - eg "lt%d" it will try and find a suitable 644 * id. It scans list of devices to build up a free map, then chooses 645 * the first empty slot. The caller must hold the dev_base or rtnl lock 646 * while allocating the name and adding the device in order to avoid 647 * duplicates. 648 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 649 * Returns the number of the unit assigned or a negative errno code. 650 */ 651 652int dev_alloc_name(struct net_device *dev, const char *name) 653{ 654 int i = 0; 655 char buf[IFNAMSIZ]; 656 const char *p; 657 const int max_netdevices = 8*PAGE_SIZE; 658 long *inuse; 659 struct net_device *d; 660 661 p = strnchr(name, IFNAMSIZ-1, '%'); 662 if (p) { 663 /* 664 * Verify the string as this thing may have come from 665 * the user. There must be either one "%d" and no other "%" 666 * characters. 667 */ 668 if (p[1] != 'd' || strchr(p + 2, '%')) 669 return -EINVAL; 670 671 /* Use one page as a bit array of possible slots */ 672 inuse = (long *) get_zeroed_page(GFP_ATOMIC); 673 if (!inuse) 674 return -ENOMEM; 675 676 for (d = dev_base; d; d = d->next) { 677 if (!sscanf(d->name, name, &i)) 678 continue; 679 if (i < 0 || i >= max_netdevices) 680 continue; 681 682 /* avoid cases where sscanf is not exact inverse of printf */ 683 snprintf(buf, sizeof(buf), name, i); 684 if (!strncmp(buf, d->name, IFNAMSIZ)) 685 set_bit(i, inuse); 686 } 687 688 i = find_first_zero_bit(inuse, max_netdevices); 689 free_page((unsigned long) inuse); 690 } 691 692 snprintf(buf, sizeof(buf), name, i); 693 if (!__dev_get_by_name(buf)) { 694 strlcpy(dev->name, buf, IFNAMSIZ); 695 return i; 696 } 697 698 /* It is possible to run out of possible slots 699 * when the name is long and there isn't enough space left 700 * for the digits, or if all bits are used. 701 */ 702 return -ENFILE; 703} 704 705 706/** 707 * dev_change_name - change name of a device 708 * @dev: device 709 * @newname: name (or format string) must be at least IFNAMSIZ 710 * 711 * Change name of a device, can pass format strings "eth%d". 712 * for wildcarding. 713 */ 714int dev_change_name(struct net_device *dev, char *newname) 715{ 716 int err = 0; 717 718 ASSERT_RTNL(); 719 720 if (dev->flags & IFF_UP) 721 return -EBUSY; 722 723 if (!dev_valid_name(newname)) 724 return -EINVAL; 725 726 if (strchr(newname, '%')) { 727 err = dev_alloc_name(dev, newname); 728 if (err < 0) 729 return err; 730 strcpy(newname, dev->name); 731 } 732 else if (__dev_get_by_name(newname)) 733 return -EEXIST; 734 else 735 strlcpy(dev->name, newname, IFNAMSIZ); 736 737 err = class_device_rename(&dev->class_dev, dev->name); 738 if (!err) { 739 hlist_del(&dev->name_hlist); 740 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name)); 741 raw_notifier_call_chain(&netdev_chain, 742 NETDEV_CHANGENAME, dev); 743 } 744 745 return err; 746} 747 748/** 749 * netdev_features_change - device changes features 750 * @dev: device to cause notification 751 * 752 * Called to indicate a device has changed features. 753 */ 754void netdev_features_change(struct net_device *dev) 755{ 756 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev); 757} 758EXPORT_SYMBOL(netdev_features_change); 759 760/** 761 * netdev_state_change - device changes state 762 * @dev: device to cause notification 763 * 764 * Called to indicate a device has changed state. This function calls 765 * the notifier chains for netdev_chain and sends a NEWLINK message 766 * to the routing socket. 767 */ 768void netdev_state_change(struct net_device *dev) 769{ 770 if (dev->flags & IFF_UP) { 771 raw_notifier_call_chain(&netdev_chain, 772 NETDEV_CHANGE, dev); 773 rtmsg_ifinfo(RTM_NEWLINK, dev, 0); 774 } 775} 776 777/** 778 * dev_load - load a network module 779 * @name: name of interface 780 * 781 * If a network interface is not present and the process has suitable 782 * privileges this function loads the module. If module loading is not 783 * available in this kernel then it becomes a nop. 784 */ 785 786void dev_load(const char *name) 787{ 788 struct net_device *dev; 789 790 read_lock(&dev_base_lock); 791 dev = __dev_get_by_name(name); 792 read_unlock(&dev_base_lock); 793 794 if (!dev && capable(CAP_SYS_MODULE)) 795 request_module("%s", name); 796} 797 798static int default_rebuild_header(struct sk_buff *skb) 799{ 800 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n", 801 skb->dev ? skb->dev->name : "NULL!!!"); 802 kfree_skb(skb); 803 return 1; 804} 805 806 807/** 808 * dev_open - prepare an interface for use. 809 * @dev: device to open 810 * 811 * Takes a device from down to up state. The device's private open 812 * function is invoked and then the multicast lists are loaded. Finally 813 * the device is moved into the up state and a %NETDEV_UP message is 814 * sent to the netdev notifier chain. 815 * 816 * Calling this function on an active interface is a nop. On a failure 817 * a negative errno code is returned. 818 */ 819int dev_open(struct net_device *dev) 820{ 821 int ret = 0; 822 823 /* 824 * Is it already up? 825 */ 826 827 if (dev->flags & IFF_UP) 828 return 0; 829 830 /* 831 * Is it even present? 832 */ 833 if (!netif_device_present(dev)) 834 return -ENODEV; 835 836 /* 837 * Call device private open method 838 */ 839 set_bit(__LINK_STATE_START, &dev->state); 840 if (dev->open) { 841 ret = dev->open(dev); 842 if (ret) 843 clear_bit(__LINK_STATE_START, &dev->state); 844 } 845 846 /* 847 * If it went open OK then: 848 */ 849 850 if (!ret) { 851 /* 852 * Set the flags. 853 */ 854 dev->flags |= IFF_UP; 855 856 /* 857 * Initialize multicasting status 858 */ 859 dev_mc_upload(dev); 860 861 /* 862 * Wakeup transmit queue engine 863 */ 864 dev_activate(dev); 865 866 /* 867 * ... and announce new interface. 868 */ 869 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev); 870 } 871 return ret; 872} 873 874/** 875 * dev_close - shutdown an interface. 876 * @dev: device to shutdown 877 * 878 * This function moves an active device into down state. A 879 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 880 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 881 * chain. 882 */ 883int dev_close(struct net_device *dev) 884{ 885 if (!(dev->flags & IFF_UP)) 886 return 0; 887 888 /* 889 * Tell people we are going down, so that they can 890 * prepare to death, when device is still operating. 891 */ 892 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev); 893 894 dev_deactivate(dev); 895 896 clear_bit(__LINK_STATE_START, &dev->state); 897 898 /* Synchronize to scheduled poll. We cannot touch poll list, 899 * it can be even on different cpu. So just clear netif_running(), 900 * and wait when poll really will happen. Actually, the best place 901 * for this is inside dev->stop() after device stopped its irq 902 * engine, but this requires more changes in devices. */ 903 904 smp_mb__after_clear_bit(); /* Commit netif_running(). */ 905 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) { 906 /* No hurry. */ 907 msleep(1); 908 } 909 910 /* 911 * Call the device specific close. This cannot fail. 912 * Only if device is UP 913 * 914 * We allow it to be called even after a DETACH hot-plug 915 * event. 916 */ 917 if (dev->stop) 918 dev->stop(dev); 919 920 /* 921 * Device is now down. 922 */ 923 924 dev->flags &= ~IFF_UP; 925 926 /* 927 * Tell people we are down 928 */ 929 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev); 930 931 return 0; 932} 933 934 935/* 936 * Device change register/unregister. These are not inline or static 937 * as we export them to the world. 938 */ 939 940/** 941 * register_netdevice_notifier - register a network notifier block 942 * @nb: notifier 943 * 944 * Register a notifier to be called when network device events occur. 945 * The notifier passed is linked into the kernel structures and must 946 * not be reused until it has been unregistered. A negative errno code 947 * is returned on a failure. 948 * 949 * When registered all registration and up events are replayed 950 * to the new notifier to allow device to have a race free 951 * view of the network device list. 952 */ 953 954int register_netdevice_notifier(struct notifier_block *nb) 955{ 956 struct net_device *dev; 957 int err; 958 959 rtnl_lock(); 960 err = raw_notifier_chain_register(&netdev_chain, nb); 961 if (!err) { 962 for (dev = dev_base; dev; dev = dev->next) { 963 nb->notifier_call(nb, NETDEV_REGISTER, dev); 964 965 if (dev->flags & IFF_UP) 966 nb->notifier_call(nb, NETDEV_UP, dev); 967 } 968 } 969 rtnl_unlock(); 970 return err; 971} 972 973/** 974 * unregister_netdevice_notifier - unregister a network notifier block 975 * @nb: notifier 976 * 977 * Unregister a notifier previously registered by 978 * register_netdevice_notifier(). The notifier is unlinked into the 979 * kernel structures and may then be reused. A negative errno code 980 * is returned on a failure. 981 */ 982 983int unregister_netdevice_notifier(struct notifier_block *nb) 984{ 985 int err; 986 987 rtnl_lock(); 988 err = raw_notifier_chain_unregister(&netdev_chain, nb); 989 rtnl_unlock(); 990 return err; 991} 992 993/** 994 * call_netdevice_notifiers - call all network notifier blocks 995 * @val: value passed unmodified to notifier function 996 * @v: pointer passed unmodified to notifier function 997 * 998 * Call all network notifier blocks. Parameters and return value 999 * are as for raw_notifier_call_chain(). 1000 */ 1001 1002int call_netdevice_notifiers(unsigned long val, void *v) 1003{ 1004 return raw_notifier_call_chain(&netdev_chain, val, v); 1005} 1006 1007/* When > 0 there are consumers of rx skb time stamps */ 1008static atomic_t netstamp_needed = ATOMIC_INIT(0); 1009 1010void net_enable_timestamp(void) 1011{ 1012 atomic_inc(&netstamp_needed); 1013} 1014 1015void net_disable_timestamp(void) 1016{ 1017 atomic_dec(&netstamp_needed); 1018} 1019 1020void __net_timestamp(struct sk_buff *skb) 1021{ 1022 struct timeval tv; 1023 1024 do_gettimeofday(&tv); 1025 skb_set_timestamp(skb, &tv); 1026} 1027EXPORT_SYMBOL(__net_timestamp); 1028 1029static inline void net_timestamp(struct sk_buff *skb) 1030{ 1031 if (atomic_read(&netstamp_needed)) 1032 __net_timestamp(skb); 1033 else { 1034 skb->tstamp.off_sec = 0; 1035 skb->tstamp.off_usec = 0; 1036 } 1037} 1038 1039/* 1040 * Support routine. Sends outgoing frames to any network 1041 * taps currently in use. 1042 */ 1043 1044void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1045{ 1046 struct packet_type *ptype; 1047 1048 net_timestamp(skb); 1049 1050 rcu_read_lock(); 1051 list_for_each_entry_rcu(ptype, &ptype_all, list) { 1052 /* Never send packets back to the socket 1053 * they originated from - MvS (miquels@drinkel.ow.org) 1054 */ 1055 if ((ptype->dev == dev || !ptype->dev) && 1056 (ptype->af_packet_priv == NULL || 1057 (struct sock *)ptype->af_packet_priv != skb->sk)) { 1058 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); 1059 if (!skb2) 1060 break; 1061 1062 /* skb->nh should be correctly 1063 set by sender, so that the second statement is 1064 just protection against buggy protocols. 1065 */ 1066 skb2->mac.raw = skb2->data; 1067 1068 if (skb2->nh.raw < skb2->data || 1069 skb2->nh.raw > skb2->tail) { 1070 if (net_ratelimit()) 1071 printk(KERN_CRIT "protocol %04x is " 1072 "buggy, dev %s\n", 1073 skb2->protocol, dev->name); 1074 skb2->nh.raw = skb2->data; 1075 } 1076 1077 skb2->h.raw = skb2->nh.raw; 1078 skb2->pkt_type = PACKET_OUTGOING; 1079 ptype->func(skb2, skb->dev, ptype, skb->dev); 1080 } 1081 } 1082 rcu_read_unlock(); 1083} 1084 1085 1086void __netif_schedule(struct net_device *dev) 1087{ 1088 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { 1089 unsigned long flags; 1090 struct softnet_data *sd; 1091 1092 local_irq_save(flags); 1093 sd = &__get_cpu_var(softnet_data); 1094 dev->next_sched = sd->output_queue; 1095 sd->output_queue = dev; 1096 raise_softirq_irqoff(NET_TX_SOFTIRQ); 1097 local_irq_restore(flags); 1098 } 1099} 1100EXPORT_SYMBOL(__netif_schedule); 1101 1102void __netif_rx_schedule(struct net_device *dev) 1103{ 1104 unsigned long flags; 1105 1106 local_irq_save(flags); 1107 dev_hold(dev); 1108 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list); 1109 if (dev->quota < 0) 1110 dev->quota += dev->weight; 1111 else 1112 dev->quota = dev->weight; 1113 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 1114 local_irq_restore(flags); 1115} 1116EXPORT_SYMBOL(__netif_rx_schedule); 1117 1118void dev_kfree_skb_any(struct sk_buff *skb) 1119{ 1120 if (in_irq() || irqs_disabled()) 1121 dev_kfree_skb_irq(skb); 1122 else 1123 dev_kfree_skb(skb); 1124} 1125EXPORT_SYMBOL(dev_kfree_skb_any); 1126 1127 1128/* Hot-plugging. */ 1129void netif_device_detach(struct net_device *dev) 1130{ 1131 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 1132 netif_running(dev)) { 1133 netif_stop_queue(dev); 1134 } 1135} 1136EXPORT_SYMBOL(netif_device_detach); 1137 1138void netif_device_attach(struct net_device *dev) 1139{ 1140 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 1141 netif_running(dev)) { 1142 netif_wake_queue(dev); 1143 __netdev_watchdog_up(dev); 1144 } 1145} 1146EXPORT_SYMBOL(netif_device_attach); 1147 1148 1149/* 1150 * Invalidate hardware checksum when packet is to be mangled, and 1151 * complete checksum manually on outgoing path. 1152 */ 1153int skb_checksum_help(struct sk_buff *skb, int inward) 1154{ 1155 unsigned int csum; 1156 int ret = 0, offset = skb->h.raw - skb->data; 1157 1158 if (inward) { 1159 skb->ip_summed = CHECKSUM_NONE; 1160 goto out; 1161 } 1162 1163 if (skb_cloned(skb)) { 1164 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 1165 if (ret) 1166 goto out; 1167 } 1168 1169 BUG_ON(offset > (int)skb->len); 1170 csum = skb_checksum(skb, offset, skb->len-offset, 0); 1171 1172 offset = skb->tail - skb->h.raw; 1173 BUG_ON(offset <= 0); 1174 BUG_ON(skb->csum + 2 > offset); 1175 1176 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum); 1177 skb->ip_summed = CHECKSUM_NONE; 1178out: 1179 return ret; 1180} 1181 1182/* Take action when hardware reception checksum errors are detected. */ 1183#ifdef CONFIG_BUG 1184void netdev_rx_csum_fault(struct net_device *dev) 1185{ 1186 if (net_ratelimit()) { 1187 printk(KERN_ERR "%s: hw csum failure.\n", 1188 dev ? dev->name : "<unknown>"); 1189 dump_stack(); 1190 } 1191} 1192EXPORT_SYMBOL(netdev_rx_csum_fault); 1193#endif 1194 1195#ifdef CONFIG_HIGHMEM 1196/* Actually, we should eliminate this check as soon as we know, that: 1197 * 1. IOMMU is present and allows to map all the memory. 1198 * 2. No high memory really exists on this machine. 1199 */ 1200 1201static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 1202{ 1203 int i; 1204 1205 if (dev->features & NETIF_F_HIGHDMA) 1206 return 0; 1207 1208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1209 if (PageHighMem(skb_shinfo(skb)->frags[i].page)) 1210 return 1; 1211 1212 return 0; 1213} 1214#else 1215#define illegal_highdma(dev, skb) (0) 1216#endif 1217 1218/* Keep head the same: replace data */ 1219int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask) 1220{ 1221 unsigned int size; 1222 u8 *data; 1223 long offset; 1224 struct skb_shared_info *ninfo; 1225 int headerlen = skb->data - skb->head; 1226 int expand = (skb->tail + skb->data_len) - skb->end; 1227 1228 if (skb_shared(skb)) 1229 BUG(); 1230 1231 if (expand <= 0) 1232 expand = 0; 1233 1234 size = skb->end - skb->head + expand; 1235 size = SKB_DATA_ALIGN(size); 1236 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 1237 if (!data) 1238 return -ENOMEM; 1239 1240 /* Copy entire thing */ 1241 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len)) 1242 BUG(); 1243 1244 /* Set up shinfo */ 1245 ninfo = (struct skb_shared_info*)(data + size); 1246 atomic_set(&ninfo->dataref, 1); 1247 ninfo->tso_size = skb_shinfo(skb)->tso_size; 1248 ninfo->tso_segs = skb_shinfo(skb)->tso_segs; 1249 ninfo->nr_frags = 0; 1250 ninfo->frag_list = NULL; 1251 1252 /* Offset between the two in bytes */ 1253 offset = data - skb->head; 1254 1255 /* Free old data. */ 1256 skb_release_data(skb); 1257 1258 skb->head = data; 1259 skb->end = data + size; 1260 1261 /* Set up new pointers */ 1262 skb->h.raw += offset; 1263 skb->nh.raw += offset; 1264 skb->mac.raw += offset; 1265 skb->tail += offset; 1266 skb->data += offset; 1267 1268 /* We are no longer a clone, even if we were. */ 1269 skb->cloned = 0; 1270 1271 skb->tail += skb->data_len; 1272 skb->data_len = 0; 1273 return 0; 1274} 1275 1276#define HARD_TX_LOCK(dev, cpu) { \ 1277 if ((dev->features & NETIF_F_LLTX) == 0) { \ 1278 spin_lock(&dev->xmit_lock); \ 1279 dev->xmit_lock_owner = cpu; \ 1280 } \ 1281} 1282 1283#define HARD_TX_UNLOCK(dev) { \ 1284 if ((dev->features & NETIF_F_LLTX) == 0) { \ 1285 dev->xmit_lock_owner = -1; \ 1286 spin_unlock(&dev->xmit_lock); \ 1287 } \ 1288} 1289 1290/** 1291 * dev_queue_xmit - transmit a buffer 1292 * @skb: buffer to transmit 1293 * 1294 * Queue a buffer for transmission to a network device. The caller must 1295 * have set the device and priority and built the buffer before calling 1296 * this function. The function can be called from an interrupt. 1297 * 1298 * A negative errno code is returned on a failure. A success does not 1299 * guarantee the frame will be transmitted as it may be dropped due 1300 * to congestion or traffic shaping. 1301 * 1302 * ----------------------------------------------------------------------------------- 1303 * I notice this method can also return errors from the queue disciplines, 1304 * including NET_XMIT_DROP, which is a positive value. So, errors can also 1305 * be positive. 1306 * 1307 * Regardless of the return value, the skb is consumed, so it is currently 1308 * difficult to retry a send to this method. (You can bump the ref count 1309 * before sending to hold a reference for retry if you are careful.) 1310 * 1311 * When calling this method, interrupts MUST be enabled. This is because 1312 * the BH enable code must have IRQs enabled so that it will not deadlock. 1313 * --BLG 1314 */ 1315 1316int dev_queue_xmit(struct sk_buff *skb) 1317{ 1318 struct net_device *dev = skb->dev; 1319 struct Qdisc *q; 1320 int rc = -ENOMEM; 1321 1322 if (skb_shinfo(skb)->frag_list && 1323 !(dev->features & NETIF_F_FRAGLIST) && 1324 __skb_linearize(skb, GFP_ATOMIC)) 1325 goto out_kfree_skb; 1326 1327 /* Fragmented skb is linearized if device does not support SG, 1328 * or if at least one of fragments is in highmem and device 1329 * does not support DMA from it. 1330 */ 1331 if (skb_shinfo(skb)->nr_frags && 1332 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && 1333 __skb_linearize(skb, GFP_ATOMIC)) 1334 goto out_kfree_skb; 1335 1336 /* If packet is not checksummed and device does not support 1337 * checksumming for this protocol, complete checksumming here. 1338 */ 1339 if (skb->ip_summed == CHECKSUM_HW && 1340 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) && 1341 (!(dev->features & NETIF_F_IP_CSUM) || 1342 skb->protocol != htons(ETH_P_IP)))) 1343 if (skb_checksum_help(skb, 0)) 1344 goto out_kfree_skb; 1345 1346 spin_lock_prefetch(&dev->queue_lock); 1347 1348 /* Disable soft irqs for various locks below. Also 1349 * stops preemption for RCU. 1350 */ 1351 local_bh_disable(); 1352 1353 /* Updates of qdisc are serialized by queue_lock. 1354 * The struct Qdisc which is pointed to by qdisc is now a 1355 * rcu structure - it may be accessed without acquiring 1356 * a lock (but the structure may be stale.) The freeing of the 1357 * qdisc will be deferred until it's known that there are no 1358 * more references to it. 1359 * 1360 * If the qdisc has an enqueue function, we still need to 1361 * hold the queue_lock before calling it, since queue_lock 1362 * also serializes access to the device queue. 1363 */ 1364 1365 q = rcu_dereference(dev->qdisc); 1366#ifdef CONFIG_NET_CLS_ACT 1367 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); 1368#endif 1369 if (q->enqueue) { 1370 /* Grab device queue */ 1371 spin_lock(&dev->queue_lock); 1372 1373 rc = q->enqueue(skb, q); 1374 1375 qdisc_run(dev); 1376 1377 spin_unlock(&dev->queue_lock); 1378 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc; 1379 goto out; 1380 } 1381 1382 /* The device has no queue. Common case for software devices: 1383 loopback, all the sorts of tunnels... 1384 1385 Really, it is unlikely that xmit_lock protection is necessary here. 1386 (f.e. loopback and IP tunnels are clean ignoring statistics 1387 counters.) 1388 However, it is possible, that they rely on protection 1389 made by us here. 1390 1391 Check this and shot the lock. It is not prone from deadlocks. 1392 Either shot noqueue qdisc, it is even simpler 8) 1393 */ 1394 if (dev->flags & IFF_UP) { 1395 int cpu = smp_processor_id(); /* ok because BHs are off */ 1396 1397 if (dev->xmit_lock_owner != cpu) { 1398 1399 HARD_TX_LOCK(dev, cpu); 1400 1401 if (!netif_queue_stopped(dev)) { 1402 if (netdev_nit) 1403 dev_queue_xmit_nit(skb, dev); 1404 1405 rc = 0; 1406 if (!dev->hard_start_xmit(skb, dev)) { 1407 HARD_TX_UNLOCK(dev); 1408 goto out; 1409 } 1410 } 1411 HARD_TX_UNLOCK(dev); 1412 if (net_ratelimit()) 1413 printk(KERN_CRIT "Virtual device %s asks to " 1414 "queue packet!\n", dev->name); 1415 } else { 1416 /* Recursion is detected! It is possible, 1417 * unfortunately */ 1418 if (net_ratelimit()) 1419 printk(KERN_CRIT "Dead loop on virtual device " 1420 "%s, fix it urgently!\n", dev->name); 1421 } 1422 } 1423 1424 rc = -ENETDOWN; 1425 local_bh_enable(); 1426 1427out_kfree_skb: 1428 kfree_skb(skb); 1429 return rc; 1430out: 1431 local_bh_enable(); 1432 return rc; 1433} 1434 1435 1436/*======================================================================= 1437 Receiver routines 1438 =======================================================================*/ 1439 1440int netdev_max_backlog = 1000; 1441int netdev_budget = 300; 1442int weight_p = 64; /* old backlog weight */ 1443 1444DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; 1445 1446 1447/** 1448 * netif_rx - post buffer to the network code 1449 * @skb: buffer to post 1450 * 1451 * This function receives a packet from a device driver and queues it for 1452 * the upper (protocol) levels to process. It always succeeds. The buffer 1453 * may be dropped during processing for congestion control or by the 1454 * protocol layers. 1455 * 1456 * return values: 1457 * NET_RX_SUCCESS (no congestion) 1458 * NET_RX_CN_LOW (low congestion) 1459 * NET_RX_CN_MOD (moderate congestion) 1460 * NET_RX_CN_HIGH (high congestion) 1461 * NET_RX_DROP (packet was dropped) 1462 * 1463 */ 1464 1465int netif_rx(struct sk_buff *skb) 1466{ 1467 struct softnet_data *queue; 1468 unsigned long flags; 1469 1470 /* if netpoll wants it, pretend we never saw it */ 1471 if (netpoll_rx(skb)) 1472 return NET_RX_DROP; 1473 1474 if (!skb->tstamp.off_sec) 1475 net_timestamp(skb); 1476 1477 /* 1478 * The code is rearranged so that the path is the most 1479 * short when CPU is congested, but is still operating. 1480 */ 1481 local_irq_save(flags); 1482 queue = &__get_cpu_var(softnet_data); 1483 1484 __get_cpu_var(netdev_rx_stat).total++; 1485 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { 1486 if (queue->input_pkt_queue.qlen) { 1487enqueue: 1488 dev_hold(skb->dev); 1489 __skb_queue_tail(&queue->input_pkt_queue, skb); 1490 local_irq_restore(flags); 1491 return NET_RX_SUCCESS; 1492 } 1493 1494 netif_rx_schedule(&queue->backlog_dev); 1495 goto enqueue; 1496 } 1497 1498 __get_cpu_var(netdev_rx_stat).dropped++; 1499 local_irq_restore(flags); 1500 1501 kfree_skb(skb); 1502 return NET_RX_DROP; 1503} 1504 1505int netif_rx_ni(struct sk_buff *skb) 1506{ 1507 int err; 1508 1509 preempt_disable(); 1510 err = netif_rx(skb); 1511 if (local_softirq_pending()) 1512 do_softirq(); 1513 preempt_enable(); 1514 1515 return err; 1516} 1517 1518EXPORT_SYMBOL(netif_rx_ni); 1519 1520static inline struct net_device *skb_bond(struct sk_buff *skb) 1521{ 1522 struct net_device *dev = skb->dev; 1523 1524 if (dev->master) { 1525 /* 1526 * On bonding slaves other than the currently active 1527 * slave, suppress duplicates except for 802.3ad 1528 * ETH_P_SLOW and alb non-mcast/bcast. 1529 */ 1530 if (dev->priv_flags & IFF_SLAVE_INACTIVE) { 1531 if (dev->master->priv_flags & IFF_MASTER_ALB) { 1532 if (skb->pkt_type != PACKET_BROADCAST && 1533 skb->pkt_type != PACKET_MULTICAST) 1534 goto keep; 1535 } 1536 1537 if (dev->master->priv_flags & IFF_MASTER_8023AD && 1538 skb->protocol == __constant_htons(ETH_P_SLOW)) 1539 goto keep; 1540 1541 kfree_skb(skb); 1542 return NULL; 1543 } 1544keep: 1545 skb->dev = dev->master; 1546 } 1547 1548 return dev; 1549} 1550 1551static void net_tx_action(struct softirq_action *h) 1552{ 1553 struct softnet_data *sd = &__get_cpu_var(softnet_data); 1554 1555 if (sd->completion_queue) { 1556 struct sk_buff *clist; 1557 1558 local_irq_disable(); 1559 clist = sd->completion_queue; 1560 sd->completion_queue = NULL; 1561 local_irq_enable(); 1562 1563 while (clist) { 1564 struct sk_buff *skb = clist; 1565 clist = clist->next; 1566 1567 BUG_TRAP(!atomic_read(&skb->users)); 1568 __kfree_skb(skb); 1569 } 1570 } 1571 1572 if (sd->output_queue) { 1573 struct net_device *head; 1574 1575 local_irq_disable(); 1576 head = sd->output_queue; 1577 sd->output_queue = NULL; 1578 local_irq_enable(); 1579 1580 while (head) { 1581 struct net_device *dev = head; 1582 head = head->next_sched; 1583 1584 smp_mb__before_clear_bit(); 1585 clear_bit(__LINK_STATE_SCHED, &dev->state); 1586 1587 if (spin_trylock(&dev->queue_lock)) { 1588 qdisc_run(dev); 1589 spin_unlock(&dev->queue_lock); 1590 } else { 1591 netif_schedule(dev); 1592 } 1593 } 1594 } 1595} 1596 1597static __inline__ int deliver_skb(struct sk_buff *skb, 1598 struct packet_type *pt_prev, 1599 struct net_device *orig_dev) 1600{ 1601 atomic_inc(&skb->users); 1602 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1603} 1604 1605#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) 1606int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb); 1607struct net_bridge; 1608struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, 1609 unsigned char *addr); 1610void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent); 1611 1612static __inline__ int handle_bridge(struct sk_buff **pskb, 1613 struct packet_type **pt_prev, int *ret, 1614 struct net_device *orig_dev) 1615{ 1616 struct net_bridge_port *port; 1617 1618 if ((*pskb)->pkt_type == PACKET_LOOPBACK || 1619 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL) 1620 return 0; 1621 1622 if (*pt_prev) { 1623 *ret = deliver_skb(*pskb, *pt_prev, orig_dev); 1624 *pt_prev = NULL; 1625 } 1626 1627 return br_handle_frame_hook(port, pskb); 1628} 1629#else 1630#define handle_bridge(skb, pt_prev, ret, orig_dev) (0) 1631#endif 1632 1633#ifdef CONFIG_NET_CLS_ACT 1634/* TODO: Maybe we should just force sch_ingress to be compiled in 1635 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions 1636 * a compare and 2 stores extra right now if we dont have it on 1637 * but have CONFIG_NET_CLS_ACT 1638 * NOTE: This doesnt stop any functionality; if you dont have 1639 * the ingress scheduler, you just cant add policies on ingress. 1640 * 1641 */ 1642static int ing_filter(struct sk_buff *skb) 1643{ 1644 struct Qdisc *q; 1645 struct net_device *dev = skb->dev; 1646 int result = TC_ACT_OK; 1647 1648 if (dev->qdisc_ingress) { 1649 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd); 1650 if (MAX_RED_LOOP < ttl++) { 1651 printk("Redir loop detected Dropping packet (%s->%s)\n", 1652 skb->input_dev->name, skb->dev->name); 1653 return TC_ACT_SHOT; 1654 } 1655 1656 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl); 1657 1658 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS); 1659 1660 spin_lock(&dev->ingress_lock); 1661 if ((q = dev->qdisc_ingress) != NULL) 1662 result = q->enqueue(skb, q); 1663 spin_unlock(&dev->ingress_lock); 1664 1665 } 1666 1667 return result; 1668} 1669#endif 1670 1671int netif_receive_skb(struct sk_buff *skb) 1672{ 1673 struct packet_type *ptype, *pt_prev; 1674 struct net_device *orig_dev; 1675 int ret = NET_RX_DROP; 1676 unsigned short type; 1677 1678 /* if we've gotten here through NAPI, check netpoll */ 1679 if (skb->dev->poll && netpoll_rx(skb)) 1680 return NET_RX_DROP; 1681 1682 if (!skb->tstamp.off_sec) 1683 net_timestamp(skb); 1684 1685 if (!skb->input_dev) 1686 skb->input_dev = skb->dev; 1687 1688 orig_dev = skb_bond(skb); 1689 1690 if (!orig_dev) 1691 return NET_RX_DROP; 1692 1693 __get_cpu_var(netdev_rx_stat).total++; 1694 1695 skb->h.raw = skb->nh.raw = skb->data; 1696 skb->mac_len = skb->nh.raw - skb->mac.raw; 1697 1698 pt_prev = NULL; 1699 1700 rcu_read_lock(); 1701 1702#ifdef CONFIG_NET_CLS_ACT 1703 if (skb->tc_verd & TC_NCLS) { 1704 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); 1705 goto ncls; 1706 } 1707#endif 1708 1709 list_for_each_entry_rcu(ptype, &ptype_all, list) { 1710 if (!ptype->dev || ptype->dev == skb->dev) { 1711 if (pt_prev) 1712 ret = deliver_skb(skb, pt_prev, orig_dev); 1713 pt_prev = ptype; 1714 } 1715 } 1716 1717#ifdef CONFIG_NET_CLS_ACT 1718 if (pt_prev) { 1719 ret = deliver_skb(skb, pt_prev, orig_dev); 1720 pt_prev = NULL; /* noone else should process this after*/ 1721 } else { 1722 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); 1723 } 1724 1725 ret = ing_filter(skb); 1726 1727 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) { 1728 kfree_skb(skb); 1729 goto out; 1730 } 1731 1732 skb->tc_verd = 0; 1733ncls: 1734#endif 1735 1736 handle_diverter(skb); 1737 1738 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev)) 1739 goto out; 1740 1741 type = skb->protocol; 1742 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) { 1743 if (ptype->type == type && 1744 (!ptype->dev || ptype->dev == skb->dev)) { 1745 if (pt_prev) 1746 ret = deliver_skb(skb, pt_prev, orig_dev); 1747 pt_prev = ptype; 1748 } 1749 } 1750 1751 if (pt_prev) { 1752 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1753 } else { 1754 kfree_skb(skb); 1755 /* Jamal, now you will not able to escape explaining 1756 * me how you were going to use this. :-) 1757 */ 1758 ret = NET_RX_DROP; 1759 } 1760 1761out: 1762 rcu_read_unlock(); 1763 return ret; 1764} 1765 1766static int process_backlog(struct net_device *backlog_dev, int *budget) 1767{ 1768 int work = 0; 1769 int quota = min(backlog_dev->quota, *budget); 1770 struct softnet_data *queue = &__get_cpu_var(softnet_data); 1771 unsigned long start_time = jiffies; 1772 1773 backlog_dev->weight = weight_p; 1774 for (;;) { 1775 struct sk_buff *skb; 1776 struct net_device *dev; 1777 1778 local_irq_disable(); 1779 skb = __skb_dequeue(&queue->input_pkt_queue); 1780 if (!skb) 1781 goto job_done; 1782 local_irq_enable(); 1783 1784 dev = skb->dev; 1785 1786 netif_receive_skb(skb); 1787 1788 dev_put(dev); 1789 1790 work++; 1791 1792 if (work >= quota || jiffies - start_time > 1) 1793 break; 1794 1795 } 1796 1797 backlog_dev->quota -= work; 1798 *budget -= work; 1799 return -1; 1800 1801job_done: 1802 backlog_dev->quota -= work; 1803 *budget -= work; 1804 1805 list_del(&backlog_dev->poll_list); 1806 smp_mb__before_clear_bit(); 1807 netif_poll_enable(backlog_dev); 1808 1809 local_irq_enable(); 1810 return 0; 1811} 1812 1813static void net_rx_action(struct softirq_action *h) 1814{ 1815 struct softnet_data *queue = &__get_cpu_var(softnet_data); 1816 unsigned long start_time = jiffies; 1817 int budget = netdev_budget; 1818 void *have; 1819 1820 local_irq_disable(); 1821 1822 while (!list_empty(&queue->poll_list)) { 1823 struct net_device *dev; 1824 1825 if (budget <= 0 || jiffies - start_time > 1) 1826 goto softnet_break; 1827 1828 local_irq_enable(); 1829 1830 dev = list_entry(queue->poll_list.next, 1831 struct net_device, poll_list); 1832 have = netpoll_poll_lock(dev); 1833 1834 if (dev->quota <= 0 || dev->poll(dev, &budget)) { 1835 netpoll_poll_unlock(have); 1836 local_irq_disable(); 1837 list_move_tail(&dev->poll_list, &queue->poll_list); 1838 if (dev->quota < 0) 1839 dev->quota += dev->weight; 1840 else 1841 dev->quota = dev->weight; 1842 } else { 1843 netpoll_poll_unlock(have); 1844 dev_put(dev); 1845 local_irq_disable(); 1846 } 1847 } 1848out: 1849 local_irq_enable(); 1850 return; 1851 1852softnet_break: 1853 __get_cpu_var(netdev_rx_stat).time_squeeze++; 1854 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 1855 goto out; 1856} 1857 1858static gifconf_func_t * gifconf_list [NPROTO]; 1859 1860/** 1861 * register_gifconf - register a SIOCGIF handler 1862 * @family: Address family 1863 * @gifconf: Function handler 1864 * 1865 * Register protocol dependent address dumping routines. The handler 1866 * that is passed must not be freed or reused until it has been replaced 1867 * by another handler. 1868 */ 1869int register_gifconf(unsigned int family, gifconf_func_t * gifconf) 1870{ 1871 if (family >= NPROTO) 1872 return -EINVAL; 1873 gifconf_list[family] = gifconf; 1874 return 0; 1875} 1876 1877 1878/* 1879 * Map an interface index to its name (SIOCGIFNAME) 1880 */ 1881 1882/* 1883 * We need this ioctl for efficient implementation of the 1884 * if_indextoname() function required by the IPv6 API. Without 1885 * it, we would have to search all the interfaces to find a 1886 * match. --pb 1887 */ 1888 1889static int dev_ifname(struct ifreq __user *arg) 1890{ 1891 struct net_device *dev; 1892 struct ifreq ifr; 1893 1894 /* 1895 * Fetch the caller's info block. 1896 */ 1897 1898 if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) 1899 return -EFAULT; 1900 1901 read_lock(&dev_base_lock); 1902 dev = __dev_get_by_index(ifr.ifr_ifindex); 1903 if (!dev) { 1904 read_unlock(&dev_base_lock); 1905 return -ENODEV; 1906 } 1907 1908 strcpy(ifr.ifr_name, dev->name); 1909 read_unlock(&dev_base_lock); 1910 1911 if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) 1912 return -EFAULT; 1913 return 0; 1914} 1915 1916/* 1917 * Perform a SIOCGIFCONF call. This structure will change 1918 * size eventually, and there is nothing I can do about it. 1919 * Thus we will need a 'compatibility mode'. 1920 */ 1921 1922static int dev_ifconf(char __user *arg) 1923{ 1924 struct ifconf ifc; 1925 struct net_device *dev; 1926 char __user *pos; 1927 int len; 1928 int total; 1929 int i; 1930 1931 /* 1932 * Fetch the caller's info block. 1933 */ 1934 1935 if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) 1936 return -EFAULT; 1937 1938 pos = ifc.ifc_buf; 1939 len = ifc.ifc_len; 1940 1941 /* 1942 * Loop over the interfaces, and write an info block for each. 1943 */ 1944 1945 total = 0; 1946 for (dev = dev_base; dev; dev = dev->next) { 1947 for (i = 0; i < NPROTO; i++) { 1948 if (gifconf_list[i]) { 1949 int done; 1950 if (!pos) 1951 done = gifconf_list[i](dev, NULL, 0); 1952 else 1953 done = gifconf_list[i](dev, pos + total, 1954 len - total); 1955 if (done < 0) 1956 return -EFAULT; 1957 total += done; 1958 } 1959 } 1960 } 1961 1962 /* 1963 * All done. Write the updated control block back to the caller. 1964 */ 1965 ifc.ifc_len = total; 1966 1967 /* 1968 * Both BSD and Solaris return 0 here, so we do too. 1969 */ 1970 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; 1971} 1972 1973#ifdef CONFIG_PROC_FS 1974/* 1975 * This is invoked by the /proc filesystem handler to display a device 1976 * in detail. 1977 */ 1978static __inline__ struct net_device *dev_get_idx(loff_t pos) 1979{ 1980 struct net_device *dev; 1981 loff_t i; 1982 1983 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next); 1984 1985 return i == pos ? dev : NULL; 1986} 1987 1988void *dev_seq_start(struct seq_file *seq, loff_t *pos) 1989{ 1990 read_lock(&dev_base_lock); 1991 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN; 1992} 1993 1994void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1995{ 1996 ++*pos; 1997 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next; 1998} 1999 2000void dev_seq_stop(struct seq_file *seq, void *v) 2001{ 2002 read_unlock(&dev_base_lock); 2003} 2004 2005static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) 2006{ 2007 if (dev->get_stats) { 2008 struct net_device_stats *stats = dev->get_stats(dev); 2009 2010 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " 2011 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", 2012 dev->name, stats->rx_bytes, stats->rx_packets, 2013 stats->rx_errors, 2014 stats->rx_dropped + stats->rx_missed_errors, 2015 stats->rx_fifo_errors, 2016 stats->rx_length_errors + stats->rx_over_errors + 2017 stats->rx_crc_errors + stats->rx_frame_errors, 2018 stats->rx_compressed, stats->multicast, 2019 stats->tx_bytes, stats->tx_packets, 2020 stats->tx_errors, stats->tx_dropped, 2021 stats->tx_fifo_errors, stats->collisions, 2022 stats->tx_carrier_errors + 2023 stats->tx_aborted_errors + 2024 stats->tx_window_errors + 2025 stats->tx_heartbeat_errors, 2026 stats->tx_compressed); 2027 } else 2028 seq_printf(seq, "%6s: No statistics available.\n", dev->name); 2029} 2030 2031/* 2032 * Called from the PROCfs module. This now uses the new arbitrary sized 2033 * /proc/net interface to create /proc/net/dev 2034 */ 2035static int dev_seq_show(struct seq_file *seq, void *v) 2036{ 2037 if (v == SEQ_START_TOKEN) 2038 seq_puts(seq, "Inter-| Receive " 2039 " | Transmit\n" 2040 " face |bytes packets errs drop fifo frame " 2041 "compressed multicast|bytes packets errs " 2042 "drop fifo colls carrier compressed\n"); 2043 else 2044 dev_seq_printf_stats(seq, v); 2045 return 0; 2046} 2047 2048static struct netif_rx_stats *softnet_get_online(loff_t *pos) 2049{ 2050 struct netif_rx_stats *rc = NULL; 2051 2052 while (*pos < NR_CPUS) 2053 if (cpu_online(*pos)) { 2054 rc = &per_cpu(netdev_rx_stat, *pos); 2055 break; 2056 } else 2057 ++*pos; 2058 return rc; 2059} 2060 2061static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) 2062{ 2063 return softnet_get_online(pos); 2064} 2065 2066static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2067{ 2068 ++*pos; 2069 return softnet_get_online(pos); 2070} 2071 2072static void softnet_seq_stop(struct seq_file *seq, void *v) 2073{ 2074} 2075 2076static int softnet_seq_show(struct seq_file *seq, void *v) 2077{ 2078 struct netif_rx_stats *s = v; 2079 2080 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", 2081 s->total, s->dropped, s->time_squeeze, 0, 2082 0, 0, 0, 0, /* was fastroute */ 2083 s->cpu_collision ); 2084 return 0; 2085} 2086 2087static struct seq_operations dev_seq_ops = { 2088 .start = dev_seq_start, 2089 .next = dev_seq_next, 2090 .stop = dev_seq_stop, 2091 .show = dev_seq_show, 2092}; 2093 2094static int dev_seq_open(struct inode *inode, struct file *file) 2095{ 2096 return seq_open(file, &dev_seq_ops); 2097} 2098 2099static struct file_operations dev_seq_fops = { 2100 .owner = THIS_MODULE, 2101 .open = dev_seq_open, 2102 .read = seq_read, 2103 .llseek = seq_lseek, 2104 .release = seq_release, 2105}; 2106 2107static struct seq_operations softnet_seq_ops = { 2108 .start = softnet_seq_start, 2109 .next = softnet_seq_next, 2110 .stop = softnet_seq_stop, 2111 .show = softnet_seq_show, 2112}; 2113 2114static int softnet_seq_open(struct inode *inode, struct file *file) 2115{ 2116 return seq_open(file, &softnet_seq_ops); 2117} 2118 2119static struct file_operations softnet_seq_fops = { 2120 .owner = THIS_MODULE, 2121 .open = softnet_seq_open, 2122 .read = seq_read, 2123 .llseek = seq_lseek, 2124 .release = seq_release, 2125}; 2126 2127#ifdef CONFIG_WIRELESS_EXT 2128extern int wireless_proc_init(void); 2129#else 2130#define wireless_proc_init() 0 2131#endif 2132 2133static int __init dev_proc_init(void) 2134{ 2135 int rc = -ENOMEM; 2136 2137 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops)) 2138 goto out; 2139 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops)) 2140 goto out_dev; 2141 if (wireless_proc_init()) 2142 goto out_softnet; 2143 rc = 0; 2144out: 2145 return rc; 2146out_softnet: 2147 proc_net_remove("softnet_stat"); 2148out_dev: 2149 proc_net_remove("dev"); 2150 goto out; 2151} 2152#else 2153#define dev_proc_init() 0 2154#endif /* CONFIG_PROC_FS */ 2155 2156 2157/** 2158 * netdev_set_master - set up master/slave pair 2159 * @slave: slave device 2160 * @master: new master device 2161 * 2162 * Changes the master device of the slave. Pass %NULL to break the 2163 * bonding. The caller must hold the RTNL semaphore. On a failure 2164 * a negative errno code is returned. On success the reference counts 2165 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the 2166 * function returns zero. 2167 */ 2168int netdev_set_master(struct net_device *slave, struct net_device *master) 2169{ 2170 struct net_device *old = slave->master; 2171 2172 ASSERT_RTNL(); 2173 2174 if (master) { 2175 if (old) 2176 return -EBUSY; 2177 dev_hold(master); 2178 } 2179 2180 slave->master = master; 2181 2182 synchronize_net(); 2183 2184 if (old) 2185 dev_put(old); 2186 2187 if (master) 2188 slave->flags |= IFF_SLAVE; 2189 else 2190 slave->flags &= ~IFF_SLAVE; 2191 2192 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); 2193 return 0; 2194} 2195 2196/** 2197 * dev_set_promiscuity - update promiscuity count on a device 2198 * @dev: device 2199 * @inc: modifier 2200 * 2201 * Add or remove promiscuity from a device. While the count in the device 2202 * remains above zero the interface remains promiscuous. Once it hits zero 2203 * the device reverts back to normal filtering operation. A negative inc 2204 * value is used to drop promiscuity on the device. 2205 */ 2206void dev_set_promiscuity(struct net_device *dev, int inc) 2207{ 2208 unsigned short old_flags = dev->flags; 2209 2210 if ((dev->promiscuity += inc) == 0) 2211 dev->flags &= ~IFF_PROMISC; 2212 else 2213 dev->flags |= IFF_PROMISC; 2214 if (dev->flags != old_flags) { 2215 dev_mc_upload(dev); 2216 printk(KERN_INFO "device %s %s promiscuous mode\n", 2217 dev->name, (dev->flags & IFF_PROMISC) ? "entered" : 2218 "left"); 2219 audit_log(current->audit_context, GFP_ATOMIC, 2220 AUDIT_ANOM_PROMISCUOUS, 2221 "dev=%s prom=%d old_prom=%d auid=%u", 2222 dev->name, (dev->flags & IFF_PROMISC), 2223 (old_flags & IFF_PROMISC), 2224 audit_get_loginuid(current->audit_context)); 2225 } 2226} 2227 2228/** 2229 * dev_set_allmulti - update allmulti count on a device 2230 * @dev: device 2231 * @inc: modifier 2232 * 2233 * Add or remove reception of all multicast frames to a device. While the 2234 * count in the device remains above zero the interface remains listening 2235 * to all interfaces. Once it hits zero the device reverts back to normal 2236 * filtering operation. A negative @inc value is used to drop the counter 2237 * when releasing a resource needing all multicasts. 2238 */ 2239 2240void dev_set_allmulti(struct net_device *dev, int inc) 2241{ 2242 unsigned short old_flags = dev->flags; 2243 2244 dev->flags |= IFF_ALLMULTI; 2245 if ((dev->allmulti += inc) == 0) 2246 dev->flags &= ~IFF_ALLMULTI; 2247 if (dev->flags ^ old_flags) 2248 dev_mc_upload(dev); 2249} 2250 2251unsigned dev_get_flags(const struct net_device *dev) 2252{ 2253 unsigned flags; 2254 2255 flags = (dev->flags & ~(IFF_PROMISC | 2256 IFF_ALLMULTI | 2257 IFF_RUNNING | 2258 IFF_LOWER_UP | 2259 IFF_DORMANT)) | 2260 (dev->gflags & (IFF_PROMISC | 2261 IFF_ALLMULTI)); 2262 2263 if (netif_running(dev)) { 2264 if (netif_oper_up(dev)) 2265 flags |= IFF_RUNNING; 2266 if (netif_carrier_ok(dev)) 2267 flags |= IFF_LOWER_UP; 2268 if (netif_dormant(dev)) 2269 flags |= IFF_DORMANT; 2270 } 2271 2272 return flags; 2273} 2274 2275int dev_change_flags(struct net_device *dev, unsigned flags) 2276{ 2277 int ret; 2278 int old_flags = dev->flags; 2279 2280 /* 2281 * Set the flags on our device. 2282 */ 2283 2284 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 2285 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 2286 IFF_AUTOMEDIA)) | 2287 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 2288 IFF_ALLMULTI)); 2289 2290 /* 2291 * Load in the correct multicast list now the flags have changed. 2292 */ 2293 2294 dev_mc_upload(dev); 2295 2296 /* 2297 * Have we downed the interface. We handle IFF_UP ourselves 2298 * according to user attempts to set it, rather than blindly 2299 * setting it. 2300 */ 2301 2302 ret = 0; 2303 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ 2304 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); 2305 2306 if (!ret) 2307 dev_mc_upload(dev); 2308 } 2309 2310 if (dev->flags & IFF_UP && 2311 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | 2312 IFF_VOLATILE))) 2313 raw_notifier_call_chain(&netdev_chain, 2314 NETDEV_CHANGE, dev); 2315 2316 if ((flags ^ dev->gflags) & IFF_PROMISC) { 2317 int inc = (flags & IFF_PROMISC) ? +1 : -1; 2318 dev->gflags ^= IFF_PROMISC; 2319 dev_set_promiscuity(dev, inc); 2320 } 2321 2322 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 2323 is important. Some (broken) drivers set IFF_PROMISC, when 2324 IFF_ALLMULTI is requested not asking us and not reporting. 2325 */ 2326 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 2327 int inc = (flags & IFF_ALLMULTI) ? +1 : -1; 2328 dev->gflags ^= IFF_ALLMULTI; 2329 dev_set_allmulti(dev, inc); 2330 } 2331 2332 if (old_flags ^ dev->flags) 2333 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags); 2334 2335 return ret; 2336} 2337 2338int dev_set_mtu(struct net_device *dev, int new_mtu) 2339{ 2340 int err; 2341 2342 if (new_mtu == dev->mtu) 2343 return 0; 2344 2345 /* MTU must be positive. */ 2346 if (new_mtu < 0) 2347 return -EINVAL; 2348 2349 if (!netif_device_present(dev)) 2350 return -ENODEV; 2351 2352 err = 0; 2353 if (dev->change_mtu) 2354 err = dev->change_mtu(dev, new_mtu); 2355 else 2356 dev->mtu = new_mtu; 2357 if (!err && dev->flags & IFF_UP) 2358 raw_notifier_call_chain(&netdev_chain, 2359 NETDEV_CHANGEMTU, dev); 2360 return err; 2361} 2362 2363int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 2364{ 2365 int err; 2366 2367 if (!dev->set_mac_address) 2368 return -EOPNOTSUPP; 2369 if (sa->sa_family != dev->type) 2370 return -EINVAL; 2371 if (!netif_device_present(dev)) 2372 return -ENODEV; 2373 err = dev->set_mac_address(dev, sa); 2374 if (!err) 2375 raw_notifier_call_chain(&netdev_chain, 2376 NETDEV_CHANGEADDR, dev); 2377 return err; 2378} 2379 2380/* 2381 * Perform the SIOCxIFxxx calls. 2382 */ 2383static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd) 2384{ 2385 int err; 2386 struct net_device *dev = __dev_get_by_name(ifr->ifr_name); 2387 2388 if (!dev) 2389 return -ENODEV; 2390 2391 switch (cmd) { 2392 case SIOCGIFFLAGS: /* Get interface flags */ 2393 ifr->ifr_flags = dev_get_flags(dev); 2394 return 0; 2395 2396 case SIOCSIFFLAGS: /* Set interface flags */ 2397 return dev_change_flags(dev, ifr->ifr_flags); 2398 2399 case SIOCGIFMETRIC: /* Get the metric on the interface 2400 (currently unused) */ 2401 ifr->ifr_metric = 0; 2402 return 0; 2403 2404 case SIOCSIFMETRIC: /* Set the metric on the interface 2405 (currently unused) */ 2406 return -EOPNOTSUPP; 2407 2408 case SIOCGIFMTU: /* Get the MTU of a device */ 2409 ifr->ifr_mtu = dev->mtu; 2410 return 0; 2411 2412 case SIOCSIFMTU: /* Set the MTU of a device */ 2413 return dev_set_mtu(dev, ifr->ifr_mtu); 2414 2415 case SIOCGIFHWADDR: 2416 if (!dev->addr_len) 2417 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); 2418 else 2419 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, 2420 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); 2421 ifr->ifr_hwaddr.sa_family = dev->type; 2422 return 0; 2423 2424 case SIOCSIFHWADDR: 2425 return dev_set_mac_address(dev, &ifr->ifr_hwaddr); 2426 2427 case SIOCSIFHWBROADCAST: 2428 if (ifr->ifr_hwaddr.sa_family != dev->type) 2429 return -EINVAL; 2430 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, 2431 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); 2432 raw_notifier_call_chain(&netdev_chain, 2433 NETDEV_CHANGEADDR, dev); 2434 return 0; 2435 2436 case SIOCGIFMAP: 2437 ifr->ifr_map.mem_start = dev->mem_start; 2438 ifr->ifr_map.mem_end = dev->mem_end; 2439 ifr->ifr_map.base_addr = dev->base_addr; 2440 ifr->ifr_map.irq = dev->irq; 2441 ifr->ifr_map.dma = dev->dma; 2442 ifr->ifr_map.port = dev->if_port; 2443 return 0; 2444 2445 case SIOCSIFMAP: 2446 if (dev->set_config) { 2447 if (!netif_device_present(dev)) 2448 return -ENODEV; 2449 return dev->set_config(dev, &ifr->ifr_map); 2450 } 2451 return -EOPNOTSUPP; 2452 2453 case SIOCADDMULTI: 2454 if (!dev->set_multicast_list || 2455 ifr->ifr_hwaddr.sa_family != AF_UNSPEC) 2456 return -EINVAL; 2457 if (!netif_device_present(dev)) 2458 return -ENODEV; 2459 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, 2460 dev->addr_len, 1); 2461 2462 case SIOCDELMULTI: 2463 if (!dev->set_multicast_list || 2464 ifr->ifr_hwaddr.sa_family != AF_UNSPEC) 2465 return -EINVAL; 2466 if (!netif_device_present(dev)) 2467 return -ENODEV; 2468 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, 2469 dev->addr_len, 1); 2470 2471 case SIOCGIFINDEX: 2472 ifr->ifr_ifindex = dev->ifindex; 2473 return 0; 2474 2475 case SIOCGIFTXQLEN: 2476 ifr->ifr_qlen = dev->tx_queue_len; 2477 return 0; 2478 2479 case SIOCSIFTXQLEN: 2480 if (ifr->ifr_qlen < 0) 2481 return -EINVAL; 2482 dev->tx_queue_len = ifr->ifr_qlen; 2483 return 0; 2484 2485 case SIOCSIFNAME: 2486 ifr->ifr_newname[IFNAMSIZ-1] = '\0'; 2487 return dev_change_name(dev, ifr->ifr_newname); 2488 2489 /* 2490 * Unknown or private ioctl 2491 */ 2492 2493 default: 2494 if ((cmd >= SIOCDEVPRIVATE && 2495 cmd <= SIOCDEVPRIVATE + 15) || 2496 cmd == SIOCBONDENSLAVE || 2497 cmd == SIOCBONDRELEASE || 2498 cmd == SIOCBONDSETHWADDR || 2499 cmd == SIOCBONDSLAVEINFOQUERY || 2500 cmd == SIOCBONDINFOQUERY || 2501 cmd == SIOCBONDCHANGEACTIVE || 2502 cmd == SIOCGMIIPHY || 2503 cmd == SIOCGMIIREG || 2504 cmd == SIOCSMIIREG || 2505 cmd == SIOCBRADDIF || 2506 cmd == SIOCBRDELIF || 2507 cmd == SIOCWANDEV) { 2508 err = -EOPNOTSUPP; 2509 if (dev->do_ioctl) { 2510 if (netif_device_present(dev)) 2511 err = dev->do_ioctl(dev, ifr, 2512 cmd); 2513 else 2514 err = -ENODEV; 2515 } 2516 } else 2517 err = -EINVAL; 2518 2519 } 2520 return err; 2521} 2522 2523/* 2524 * This function handles all "interface"-type I/O control requests. The actual 2525 * 'doing' part of this is dev_ifsioc above. 2526 */ 2527 2528/** 2529 * dev_ioctl - network device ioctl 2530 * @cmd: command to issue 2531 * @arg: pointer to a struct ifreq in user space 2532 * 2533 * Issue ioctl functions to devices. This is normally called by the 2534 * user space syscall interfaces but can sometimes be useful for 2535 * other purposes. The return value is the return from the syscall if 2536 * positive or a negative errno code on error. 2537 */ 2538 2539int dev_ioctl(unsigned int cmd, void __user *arg) 2540{ 2541 struct ifreq ifr; 2542 int ret; 2543 char *colon; 2544 2545 /* One special case: SIOCGIFCONF takes ifconf argument 2546 and requires shared lock, because it sleeps writing 2547 to user space. 2548 */ 2549 2550 if (cmd == SIOCGIFCONF) { 2551 rtnl_lock(); 2552 ret = dev_ifconf((char __user *) arg); 2553 rtnl_unlock(); 2554 return ret; 2555 } 2556 if (cmd == SIOCGIFNAME) 2557 return dev_ifname((struct ifreq __user *)arg); 2558 2559 if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) 2560 return -EFAULT; 2561 2562 ifr.ifr_name[IFNAMSIZ-1] = 0; 2563 2564 colon = strchr(ifr.ifr_name, ':'); 2565 if (colon) 2566 *colon = 0; 2567 2568 /* 2569 * See which interface the caller is talking about. 2570 */ 2571 2572 switch (cmd) { 2573 /* 2574 * These ioctl calls: 2575 * - can be done by all. 2576 * - atomic and do not require locking. 2577 * - return a value 2578 */ 2579 case SIOCGIFFLAGS: 2580 case SIOCGIFMETRIC: 2581 case SIOCGIFMTU: 2582 case SIOCGIFHWADDR: 2583 case SIOCGIFSLAVE: 2584 case SIOCGIFMAP: 2585 case SIOCGIFINDEX: 2586 case SIOCGIFTXQLEN: 2587 dev_load(ifr.ifr_name); 2588 read_lock(&dev_base_lock); 2589 ret = dev_ifsioc(&ifr, cmd); 2590 read_unlock(&dev_base_lock); 2591 if (!ret) { 2592 if (colon) 2593 *colon = ':'; 2594 if (copy_to_user(arg, &ifr, 2595 sizeof(struct ifreq))) 2596 ret = -EFAULT; 2597 } 2598 return ret; 2599 2600 case SIOCETHTOOL: 2601 dev_load(ifr.ifr_name); 2602 rtnl_lock(); 2603 ret = dev_ethtool(&ifr); 2604 rtnl_unlock(); 2605 if (!ret) { 2606 if (colon) 2607 *colon = ':'; 2608 if (copy_to_user(arg, &ifr, 2609 sizeof(struct ifreq))) 2610 ret = -EFAULT; 2611 } 2612 return ret; 2613 2614 /* 2615 * These ioctl calls: 2616 * - require superuser power. 2617 * - require strict serialization. 2618 * - return a value 2619 */ 2620 case SIOCGMIIPHY: 2621 case SIOCGMIIREG: 2622 case SIOCSIFNAME: 2623 if (!capable(CAP_NET_ADMIN)) 2624 return -EPERM; 2625 dev_load(ifr.ifr_name); 2626 rtnl_lock(); 2627 ret = dev_ifsioc(&ifr, cmd); 2628 rtnl_unlock(); 2629 if (!ret) { 2630 if (colon) 2631 *colon = ':'; 2632 if (copy_to_user(arg, &ifr, 2633 sizeof(struct ifreq))) 2634 ret = -EFAULT; 2635 } 2636 return ret; 2637 2638 /* 2639 * These ioctl calls: 2640 * - require superuser power. 2641 * - require strict serialization. 2642 * - do not return a value 2643 */ 2644 case SIOCSIFFLAGS: 2645 case SIOCSIFMETRIC: 2646 case SIOCSIFMTU: 2647 case SIOCSIFMAP: 2648 case SIOCSIFHWADDR: 2649 case SIOCSIFSLAVE: 2650 case SIOCADDMULTI: 2651 case SIOCDELMULTI: 2652 case SIOCSIFHWBROADCAST: 2653 case SIOCSIFTXQLEN: 2654 case SIOCSMIIREG: 2655 case SIOCBONDENSLAVE: 2656 case SIOCBONDRELEASE: 2657 case SIOCBONDSETHWADDR: 2658 case SIOCBONDCHANGEACTIVE: 2659 case SIOCBRADDIF: 2660 case SIOCBRDELIF: 2661 if (!capable(CAP_NET_ADMIN)) 2662 return -EPERM; 2663 /* fall through */ 2664 case SIOCBONDSLAVEINFOQUERY: 2665 case SIOCBONDINFOQUERY: 2666 dev_load(ifr.ifr_name); 2667 rtnl_lock(); 2668 ret = dev_ifsioc(&ifr, cmd); 2669 rtnl_unlock(); 2670 return ret; 2671 2672 case SIOCGIFMEM: 2673 /* Get the per device memory space. We can add this but 2674 * currently do not support it */ 2675 case SIOCSIFMEM: 2676 /* Set the per device memory buffer space. 2677 * Not applicable in our case */ 2678 case SIOCSIFLINK: 2679 return -EINVAL; 2680 2681 /* 2682 * Unknown or private ioctl. 2683 */ 2684 default: 2685 if (cmd == SIOCWANDEV || 2686 (cmd >= SIOCDEVPRIVATE && 2687 cmd <= SIOCDEVPRIVATE + 15)) { 2688 dev_load(ifr.ifr_name); 2689 rtnl_lock(); 2690 ret = dev_ifsioc(&ifr, cmd); 2691 rtnl_unlock(); 2692 if (!ret && copy_to_user(arg, &ifr, 2693 sizeof(struct ifreq))) 2694 ret = -EFAULT; 2695 return ret; 2696 } 2697#ifdef CONFIG_WIRELESS_EXT 2698 /* Take care of Wireless Extensions */ 2699 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) { 2700 /* If command is `set a parameter', or 2701 * `get the encoding parameters', check if 2702 * the user has the right to do it */ 2703 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE 2704 || cmd == SIOCGIWENCODEEXT) { 2705 if (!capable(CAP_NET_ADMIN)) 2706 return -EPERM; 2707 } 2708 dev_load(ifr.ifr_name); 2709 rtnl_lock(); 2710 /* Follow me in net/core/wireless.c */ 2711 ret = wireless_process_ioctl(&ifr, cmd); 2712 rtnl_unlock(); 2713 if (IW_IS_GET(cmd) && 2714 copy_to_user(arg, &ifr, 2715 sizeof(struct ifreq))) 2716 ret = -EFAULT; 2717 return ret; 2718 } 2719#endif /* CONFIG_WIRELESS_EXT */ 2720 return -EINVAL; 2721 } 2722} 2723 2724 2725/** 2726 * dev_new_index - allocate an ifindex 2727 * 2728 * Returns a suitable unique value for a new device interface 2729 * number. The caller must hold the rtnl semaphore or the 2730 * dev_base_lock to be sure it remains unique. 2731 */ 2732static int dev_new_index(void) 2733{ 2734 static int ifindex; 2735 for (;;) { 2736 if (++ifindex <= 0) 2737 ifindex = 1; 2738 if (!__dev_get_by_index(ifindex)) 2739 return ifindex; 2740 } 2741} 2742 2743static int dev_boot_phase = 1; 2744 2745/* Delayed registration/unregisteration */ 2746static DEFINE_SPINLOCK(net_todo_list_lock); 2747static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list); 2748 2749static inline void net_set_todo(struct net_device *dev) 2750{ 2751 spin_lock(&net_todo_list_lock); 2752 list_add_tail(&dev->todo_list, &net_todo_list); 2753 spin_unlock(&net_todo_list_lock); 2754} 2755 2756/** 2757 * register_netdevice - register a network device 2758 * @dev: device to register 2759 * 2760 * Take a completed network device structure and add it to the kernel 2761 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 2762 * chain. 0 is returned on success. A negative errno code is returned 2763 * on a failure to set up the device, or if the name is a duplicate. 2764 * 2765 * Callers must hold the rtnl semaphore. You may want 2766 * register_netdev() instead of this. 2767 * 2768 * BUGS: 2769 * The locking appears insufficient to guarantee two parallel registers 2770 * will not get the same name. 2771 */ 2772 2773int register_netdevice(struct net_device *dev) 2774{ 2775 struct hlist_head *head; 2776 struct hlist_node *p; 2777 int ret; 2778 2779 BUG_ON(dev_boot_phase); 2780 ASSERT_RTNL(); 2781 2782 might_sleep(); 2783 2784 /* When net_device's are persistent, this will be fatal. */ 2785 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 2786 2787 spin_lock_init(&dev->queue_lock); 2788 spin_lock_init(&dev->xmit_lock); 2789 dev->xmit_lock_owner = -1; 2790#ifdef CONFIG_NET_CLS_ACT 2791 spin_lock_init(&dev->ingress_lock); 2792#endif 2793 2794 ret = alloc_divert_blk(dev); 2795 if (ret) 2796 goto out; 2797 2798 dev->iflink = -1; 2799 2800 /* Init, if this function is available */ 2801 if (dev->init) { 2802 ret = dev->init(dev); 2803 if (ret) { 2804 if (ret > 0) 2805 ret = -EIO; 2806 goto out_err; 2807 } 2808 } 2809 2810 if (!dev_valid_name(dev->name)) { 2811 ret = -EINVAL; 2812 goto out_err; 2813 } 2814 2815 dev->ifindex = dev_new_index(); 2816 if (dev->iflink == -1) 2817 dev->iflink = dev->ifindex; 2818 2819 /* Check for existence of name */ 2820 head = dev_name_hash(dev->name); 2821 hlist_for_each(p, head) { 2822 struct net_device *d 2823 = hlist_entry(p, struct net_device, name_hlist); 2824 if (!strncmp(d->name, dev->name, IFNAMSIZ)) { 2825 ret = -EEXIST; 2826 goto out_err; 2827 } 2828 } 2829 2830 /* Fix illegal SG+CSUM combinations. */ 2831 if ((dev->features & NETIF_F_SG) && 2832 !(dev->features & (NETIF_F_IP_CSUM | 2833 NETIF_F_NO_CSUM | 2834 NETIF_F_HW_CSUM))) { 2835 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n", 2836 dev->name); 2837 dev->features &= ~NETIF_F_SG; 2838 } 2839 2840 /* TSO requires that SG is present as well. */ 2841 if ((dev->features & NETIF_F_TSO) && 2842 !(dev->features & NETIF_F_SG)) { 2843 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n", 2844 dev->name); 2845 dev->features &= ~NETIF_F_TSO; 2846 } 2847 if (dev->features & NETIF_F_UFO) { 2848 if (!(dev->features & NETIF_F_HW_CSUM)) { 2849 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " 2850 "NETIF_F_HW_CSUM feature.\n", 2851 dev->name); 2852 dev->features &= ~NETIF_F_UFO; 2853 } 2854 if (!(dev->features & NETIF_F_SG)) { 2855 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " 2856 "NETIF_F_SG feature.\n", 2857 dev->name); 2858 dev->features &= ~NETIF_F_UFO; 2859 } 2860 } 2861 2862 /* 2863 * nil rebuild_header routine, 2864 * that should be never called and used as just bug trap. 2865 */ 2866 2867 if (!dev->rebuild_header) 2868 dev->rebuild_header = default_rebuild_header; 2869 2870 ret = netdev_register_sysfs(dev); 2871 if (ret) 2872 goto out_err; 2873 dev->reg_state = NETREG_REGISTERED; 2874 2875 /* 2876 * Default initial state at registry is that the 2877 * device is present. 2878 */ 2879 2880 set_bit(__LINK_STATE_PRESENT, &dev->state); 2881 2882 dev->next = NULL; 2883 dev_init_scheduler(dev); 2884 write_lock_bh(&dev_base_lock); 2885 *dev_tail = dev; 2886 dev_tail = &dev->next; 2887 hlist_add_head(&dev->name_hlist, head); 2888 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex)); 2889 dev_hold(dev); 2890 write_unlock_bh(&dev_base_lock); 2891 2892 /* Notify protocols, that a new device appeared. */ 2893 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev); 2894 2895 ret = 0; 2896 2897out: 2898 return ret; 2899out_err: 2900 free_divert_blk(dev); 2901 goto out; 2902} 2903 2904/** 2905 * register_netdev - register a network device 2906 * @dev: device to register 2907 * 2908 * Take a completed network device structure and add it to the kernel 2909 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 2910 * chain. 0 is returned on success. A negative errno code is returned 2911 * on a failure to set up the device, or if the name is a duplicate. 2912 * 2913 * This is a wrapper around register_netdev that takes the rtnl semaphore 2914 * and expands the device name if you passed a format string to 2915 * alloc_netdev. 2916 */ 2917int register_netdev(struct net_device *dev) 2918{ 2919 int err; 2920 2921 rtnl_lock(); 2922 2923 /* 2924 * If the name is a format string the caller wants us to do a 2925 * name allocation. 2926 */ 2927 if (strchr(dev->name, '%')) { 2928 err = dev_alloc_name(dev, dev->name); 2929 if (err < 0) 2930 goto out; 2931 } 2932 2933 /* 2934 * Back compatibility hook. Kill this one in 2.5 2935 */ 2936 if (dev->name[0] == 0 || dev->name[0] == ' ') { 2937 err = dev_alloc_name(dev, "eth%d"); 2938 if (err < 0) 2939 goto out; 2940 } 2941 2942 err = register_netdevice(dev); 2943out: 2944 rtnl_unlock(); 2945 return err; 2946} 2947EXPORT_SYMBOL(register_netdev); 2948 2949/* 2950 * netdev_wait_allrefs - wait until all references are gone. 2951 * 2952 * This is called when unregistering network devices. 2953 * 2954 * Any protocol or device that holds a reference should register 2955 * for netdevice notification, and cleanup and put back the 2956 * reference if they receive an UNREGISTER event. 2957 * We can get stuck here if buggy protocols don't correctly 2958 * call dev_put. 2959 */ 2960static void netdev_wait_allrefs(struct net_device *dev) 2961{ 2962 unsigned long rebroadcast_time, warning_time; 2963 2964 rebroadcast_time = warning_time = jiffies; 2965 while (atomic_read(&dev->refcnt) != 0) { 2966 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 2967 rtnl_lock(); 2968 2969 /* Rebroadcast unregister notification */ 2970 raw_notifier_call_chain(&netdev_chain, 2971 NETDEV_UNREGISTER, dev); 2972 2973 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 2974 &dev->state)) { 2975 /* We must not have linkwatch events 2976 * pending on unregister. If this 2977 * happens, we simply run the queue 2978 * unscheduled, resulting in a noop 2979 * for this device. 2980 */ 2981 linkwatch_run_queue(); 2982 } 2983 2984 __rtnl_unlock(); 2985 2986 rebroadcast_time = jiffies; 2987 } 2988 2989 msleep(250); 2990 2991 if (time_after(jiffies, warning_time + 10 * HZ)) { 2992 printk(KERN_EMERG "unregister_netdevice: " 2993 "waiting for %s to become free. Usage " 2994 "count = %d\n", 2995 dev->name, atomic_read(&dev->refcnt)); 2996 warning_time = jiffies; 2997 } 2998 } 2999} 3000 3001/* The sequence is: 3002 * 3003 * rtnl_lock(); 3004 * ... 3005 * register_netdevice(x1); 3006 * register_netdevice(x2); 3007 * ... 3008 * unregister_netdevice(y1); 3009 * unregister_netdevice(y2); 3010 * ... 3011 * rtnl_unlock(); 3012 * free_netdev(y1); 3013 * free_netdev(y2); 3014 * 3015 * We are invoked by rtnl_unlock() after it drops the semaphore. 3016 * This allows us to deal with problems: 3017 * 1) We can delete sysfs objects which invoke hotplug 3018 * without deadlocking with linkwatch via keventd. 3019 * 2) Since we run with the RTNL semaphore not held, we can sleep 3020 * safely in order to wait for the netdev refcnt to drop to zero. 3021 */ 3022static DEFINE_MUTEX(net_todo_run_mutex); 3023void netdev_run_todo(void) 3024{ 3025 struct list_head list = LIST_HEAD_INIT(list); 3026 3027 /* Need to guard against multiple cpu's getting out of order. */ 3028 mutex_lock(&net_todo_run_mutex); 3029 3030 /* Not safe to do outside the semaphore. We must not return 3031 * until all unregister events invoked by the local processor 3032 * have been completed (either by this todo run, or one on 3033 * another cpu). 3034 */ 3035 if (list_empty(&net_todo_list)) 3036 goto out; 3037 3038 /* Snapshot list, allow later requests */ 3039 spin_lock(&net_todo_list_lock); 3040 list_splice_init(&net_todo_list, &list); 3041 spin_unlock(&net_todo_list_lock); 3042 3043 while (!list_empty(&list)) { 3044 struct net_device *dev 3045 = list_entry(list.next, struct net_device, todo_list); 3046 list_del(&dev->todo_list); 3047 3048 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 3049 printk(KERN_ERR "network todo '%s' but state %d\n", 3050 dev->name, dev->reg_state); 3051 dump_stack(); 3052 continue; 3053 } 3054 3055 netdev_unregister_sysfs(dev); 3056 dev->reg_state = NETREG_UNREGISTERED; 3057 3058 netdev_wait_allrefs(dev); 3059 3060 /* paranoia */ 3061 BUG_ON(atomic_read(&dev->refcnt)); 3062 BUG_TRAP(!dev->ip_ptr); 3063 BUG_TRAP(!dev->ip6_ptr); 3064 BUG_TRAP(!dev->dn_ptr); 3065 3066 /* It must be the very last action, 3067 * after this 'dev' may point to freed up memory. 3068 */ 3069 if (dev->destructor) 3070 dev->destructor(dev); 3071 } 3072 3073out: 3074 mutex_unlock(&net_todo_run_mutex); 3075} 3076 3077/** 3078 * alloc_netdev - allocate network device 3079 * @sizeof_priv: size of private data to allocate space for 3080 * @name: device name format string 3081 * @setup: callback to initialize device 3082 * 3083 * Allocates a struct net_device with private data area for driver use 3084 * and performs basic initialization. 3085 */ 3086struct net_device *alloc_netdev(int sizeof_priv, const char *name, 3087 void (*setup)(struct net_device *)) 3088{ 3089 void *p; 3090 struct net_device *dev; 3091 int alloc_size; 3092 3093 /* ensure 32-byte alignment of both the device and private area */ 3094 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; 3095 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST; 3096 3097 p = kzalloc(alloc_size, GFP_KERNEL); 3098 if (!p) { 3099 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n"); 3100 return NULL; 3101 } 3102 3103 dev = (struct net_device *) 3104 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); 3105 dev->padded = (char *)dev - (char *)p; 3106 3107 if (sizeof_priv) 3108 dev->priv = netdev_priv(dev); 3109 3110 setup(dev); 3111 strcpy(dev->name, name); 3112 return dev; 3113} 3114EXPORT_SYMBOL(alloc_netdev); 3115 3116/** 3117 * free_netdev - free network device 3118 * @dev: device 3119 * 3120 * This function does the last stage of destroying an allocated device 3121 * interface. The reference to the device object is released. 3122 * If this is the last reference then it will be freed. 3123 */ 3124void free_netdev(struct net_device *dev) 3125{ 3126#ifdef CONFIG_SYSFS 3127 /* Compatibility with error handling in drivers */ 3128 if (dev->reg_state == NETREG_UNINITIALIZED) { 3129 kfree((char *)dev - dev->padded); 3130 return; 3131 } 3132 3133 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 3134 dev->reg_state = NETREG_RELEASED; 3135 3136 /* will free via class release */ 3137 class_device_put(&dev->class_dev); 3138#else 3139 kfree((char *)dev - dev->padded); 3140#endif 3141} 3142 3143/* Synchronize with packet receive processing. */ 3144void synchronize_net(void) 3145{ 3146 might_sleep(); 3147 synchronize_rcu(); 3148} 3149 3150/** 3151 * unregister_netdevice - remove device from the kernel 3152 * @dev: device 3153 * 3154 * This function shuts down a device interface and removes it 3155 * from the kernel tables. On success 0 is returned, on a failure 3156 * a negative errno code is returned. 3157 * 3158 * Callers must hold the rtnl semaphore. You may want 3159 * unregister_netdev() instead of this. 3160 */ 3161 3162int unregister_netdevice(struct net_device *dev) 3163{ 3164 struct net_device *d, **dp; 3165 3166 BUG_ON(dev_boot_phase); 3167 ASSERT_RTNL(); 3168 3169 /* Some devices call without registering for initialization unwind. */ 3170 if (dev->reg_state == NETREG_UNINITIALIZED) { 3171 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " 3172 "was registered\n", dev->name, dev); 3173 return -ENODEV; 3174 } 3175 3176 BUG_ON(dev->reg_state != NETREG_REGISTERED); 3177 3178 /* If device is running, close it first. */ 3179 if (dev->flags & IFF_UP) 3180 dev_close(dev); 3181 3182 /* And unlink it from device chain. */ 3183 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) { 3184 if (d == dev) { 3185 write_lock_bh(&dev_base_lock); 3186 hlist_del(&dev->name_hlist); 3187 hlist_del(&dev->index_hlist); 3188 if (dev_tail == &dev->next) 3189 dev_tail = dp; 3190 *dp = d->next; 3191 write_unlock_bh(&dev_base_lock); 3192 break; 3193 } 3194 } 3195 if (!d) { 3196 printk(KERN_ERR "unregister net_device: '%s' not found\n", 3197 dev->name); 3198 return -ENODEV; 3199 } 3200 3201 dev->reg_state = NETREG_UNREGISTERING; 3202 3203 synchronize_net(); 3204 3205 /* Shutdown queueing discipline. */ 3206 dev_shutdown(dev); 3207 3208 3209 /* Notify protocols, that we are about to destroy 3210 this device. They should clean all the things. 3211 */ 3212 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev); 3213 3214 /* 3215 * Flush the multicast chain 3216 */ 3217 dev_mc_discard(dev); 3218 3219 if (dev->uninit) 3220 dev->uninit(dev); 3221 3222 /* Notifier chain MUST detach us from master device. */ 3223 BUG_TRAP(!dev->master); 3224 3225 free_divert_blk(dev); 3226 3227 /* Finish processing unregister after unlock */ 3228 net_set_todo(dev); 3229 3230 synchronize_net(); 3231 3232 dev_put(dev); 3233 return 0; 3234} 3235 3236/** 3237 * unregister_netdev - remove device from the kernel 3238 * @dev: device 3239 * 3240 * This function shuts down a device interface and removes it 3241 * from the kernel tables. On success 0 is returned, on a failure 3242 * a negative errno code is returned. 3243 * 3244 * This is just a wrapper for unregister_netdevice that takes 3245 * the rtnl semaphore. In general you want to use this and not 3246 * unregister_netdevice. 3247 */ 3248void unregister_netdev(struct net_device *dev) 3249{ 3250 rtnl_lock(); 3251 unregister_netdevice(dev); 3252 rtnl_unlock(); 3253} 3254 3255EXPORT_SYMBOL(unregister_netdev); 3256 3257#ifdef CONFIG_HOTPLUG_CPU 3258static int dev_cpu_callback(struct notifier_block *nfb, 3259 unsigned long action, 3260 void *ocpu) 3261{ 3262 struct sk_buff **list_skb; 3263 struct net_device **list_net; 3264 struct sk_buff *skb; 3265 unsigned int cpu, oldcpu = (unsigned long)ocpu; 3266 struct softnet_data *sd, *oldsd; 3267 3268 if (action != CPU_DEAD) 3269 return NOTIFY_OK; 3270 3271 local_irq_disable(); 3272 cpu = smp_processor_id(); 3273 sd = &per_cpu(softnet_data, cpu); 3274 oldsd = &per_cpu(softnet_data, oldcpu); 3275 3276 /* Find end of our completion_queue. */ 3277 list_skb = &sd->completion_queue; 3278 while (*list_skb) 3279 list_skb = &(*list_skb)->next; 3280 /* Append completion queue from offline CPU. */ 3281 *list_skb = oldsd->completion_queue; 3282 oldsd->completion_queue = NULL; 3283 3284 /* Find end of our output_queue. */ 3285 list_net = &sd->output_queue; 3286 while (*list_net) 3287 list_net = &(*list_net)->next_sched; 3288 /* Append output queue from offline CPU. */ 3289 *list_net = oldsd->output_queue; 3290 oldsd->output_queue = NULL; 3291 3292 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3293 local_irq_enable(); 3294 3295 /* Process offline CPU's input_pkt_queue */ 3296 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) 3297 netif_rx(skb); 3298 3299 return NOTIFY_OK; 3300} 3301#endif /* CONFIG_HOTPLUG_CPU */ 3302 3303 3304/* 3305 * Initialize the DEV module. At boot time this walks the device list and 3306 * unhooks any devices that fail to initialise (normally hardware not 3307 * present) and leaves us with a valid list of present and active devices. 3308 * 3309 */ 3310 3311/* 3312 * This is called single threaded during boot, so no need 3313 * to take the rtnl semaphore. 3314 */ 3315static int __init net_dev_init(void) 3316{ 3317 int i, rc = -ENOMEM; 3318 3319 BUG_ON(!dev_boot_phase); 3320 3321 net_random_init(); 3322 3323 if (dev_proc_init()) 3324 goto out; 3325 3326 if (netdev_sysfs_init()) 3327 goto out; 3328 3329 INIT_LIST_HEAD(&ptype_all); 3330 for (i = 0; i < 16; i++) 3331 INIT_LIST_HEAD(&ptype_base[i]); 3332 3333 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++) 3334 INIT_HLIST_HEAD(&dev_name_head[i]); 3335 3336 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++) 3337 INIT_HLIST_HEAD(&dev_index_head[i]); 3338 3339 /* 3340 * Initialise the packet receive queues. 3341 */ 3342 3343 for_each_possible_cpu(i) { 3344 struct softnet_data *queue; 3345 3346 queue = &per_cpu(softnet_data, i); 3347 skb_queue_head_init(&queue->input_pkt_queue); 3348 queue->completion_queue = NULL; 3349 INIT_LIST_HEAD(&queue->poll_list); 3350 set_bit(__LINK_STATE_START, &queue->backlog_dev.state); 3351 queue->backlog_dev.weight = weight_p; 3352 queue->backlog_dev.poll = process_backlog; 3353 atomic_set(&queue->backlog_dev.refcnt, 1); 3354 } 3355 3356 dev_boot_phase = 0; 3357 3358 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL); 3359 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL); 3360 3361 hotcpu_notifier(dev_cpu_callback, 0); 3362 dst_init(); 3363 dev_mcast_init(); 3364 rc = 0; 3365out: 3366 return rc; 3367} 3368 3369subsys_initcall(net_dev_init); 3370 3371EXPORT_SYMBOL(__dev_get_by_index); 3372EXPORT_SYMBOL(__dev_get_by_name); 3373EXPORT_SYMBOL(__dev_remove_pack); 3374EXPORT_SYMBOL(__skb_linearize); 3375EXPORT_SYMBOL(dev_valid_name); 3376EXPORT_SYMBOL(dev_add_pack); 3377EXPORT_SYMBOL(dev_alloc_name); 3378EXPORT_SYMBOL(dev_close); 3379EXPORT_SYMBOL(dev_get_by_flags); 3380EXPORT_SYMBOL(dev_get_by_index); 3381EXPORT_SYMBOL(dev_get_by_name); 3382EXPORT_SYMBOL(dev_open); 3383EXPORT_SYMBOL(dev_queue_xmit); 3384EXPORT_SYMBOL(dev_remove_pack); 3385EXPORT_SYMBOL(dev_set_allmulti); 3386EXPORT_SYMBOL(dev_set_promiscuity); 3387EXPORT_SYMBOL(dev_change_flags); 3388EXPORT_SYMBOL(dev_set_mtu); 3389EXPORT_SYMBOL(dev_set_mac_address); 3390EXPORT_SYMBOL(free_netdev); 3391EXPORT_SYMBOL(netdev_boot_setup_check); 3392EXPORT_SYMBOL(netdev_set_master); 3393EXPORT_SYMBOL(netdev_state_change); 3394EXPORT_SYMBOL(netif_receive_skb); 3395EXPORT_SYMBOL(netif_rx); 3396EXPORT_SYMBOL(register_gifconf); 3397EXPORT_SYMBOL(register_netdevice); 3398EXPORT_SYMBOL(register_netdevice_notifier); 3399EXPORT_SYMBOL(skb_checksum_help); 3400EXPORT_SYMBOL(synchronize_net); 3401EXPORT_SYMBOL(unregister_netdevice); 3402EXPORT_SYMBOL(unregister_netdevice_notifier); 3403EXPORT_SYMBOL(net_enable_timestamp); 3404EXPORT_SYMBOL(net_disable_timestamp); 3405EXPORT_SYMBOL(dev_get_flags); 3406 3407#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) 3408EXPORT_SYMBOL(br_handle_frame_hook); 3409EXPORT_SYMBOL(br_fdb_get_hook); 3410EXPORT_SYMBOL(br_fdb_put_hook); 3411#endif 3412 3413#ifdef CONFIG_KMOD 3414EXPORT_SYMBOL(dev_load); 3415#endif 3416 3417EXPORT_PER_CPU_SYMBOL(softnet_data);