at v2.6.12-rc6 3384 lines 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/config.h> 79#include <linux/cpu.h> 80#include <linux/types.h> 81#include <linux/kernel.h> 82#include <linux/sched.h> 83#include <linux/string.h> 84#include <linux/mm.h> 85#include <linux/socket.h> 86#include <linux/sockios.h> 87#include <linux/errno.h> 88#include <linux/interrupt.h> 89#include <linux/if_ether.h> 90#include <linux/netdevice.h> 91#include <linux/etherdevice.h> 92#include <linux/notifier.h> 93#include <linux/skbuff.h> 94#include <net/sock.h> 95#include <linux/rtnetlink.h> 96#include <linux/proc_fs.h> 97#include <linux/seq_file.h> 98#include <linux/stat.h> 99#include <linux/if_bridge.h> 100#include <linux/divert.h> 101#include <net/dst.h> 102#include <net/pkt_sched.h> 103#include <net/checksum.h> 104#include <linux/highmem.h> 105#include <linux/init.h> 106#include <linux/kmod.h> 107#include <linux/module.h> 108#include <linux/kallsyms.h> 109#include <linux/netpoll.h> 110#include <linux/rcupdate.h> 111#include <linux/delay.h> 112#ifdef CONFIG_NET_RADIO 113#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */ 114#include <net/iw_handler.h> 115#endif /* CONFIG_NET_RADIO */ 116#include <asm/current.h> 117 118/* This define, if set, will randomly drop a packet when congestion 119 * is more than moderate. It helps fairness in the multi-interface 120 * case when one of them is a hog, but it kills performance for the 121 * single interface case so it is off now by default. 122 */ 123#undef RAND_LIE 124 125/* Setting this will sample the queue lengths and thus congestion 126 * via a timer instead of as each packet is received. 127 */ 128#undef OFFLINE_SAMPLE 129 130/* 131 * The list of packet types we will receive (as opposed to discard) 132 * and the routines to invoke. 133 * 134 * Why 16. Because with 16 the only overlap we get on a hash of the 135 * low nibble of the protocol value is RARP/SNAP/X.25. 136 * 137 * NOTE: That is no longer true with the addition of VLAN tags. Not 138 * sure which should go first, but I bet it won't make much 139 * difference if we are running VLANs. The good news is that 140 * this protocol won't be in the list unless compiled in, so 141 * the average user (w/out VLANs) will not be adversly affected. 142 * --BLG 143 * 144 * 0800 IP 145 * 8100 802.1Q VLAN 146 * 0001 802.3 147 * 0002 AX.25 148 * 0004 802.2 149 * 8035 RARP 150 * 0005 SNAP 151 * 0805 X.25 152 * 0806 ARP 153 * 8137 IPX 154 * 0009 Localtalk 155 * 86DD IPv6 156 */ 157 158static DEFINE_SPINLOCK(ptype_lock); 159static struct list_head ptype_base[16]; /* 16 way hashed list */ 160static struct list_head ptype_all; /* Taps */ 161 162#ifdef OFFLINE_SAMPLE 163static void sample_queue(unsigned long dummy); 164static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0); 165#endif 166 167/* 168 * The @dev_base list is protected by @dev_base_lock and the rtln 169 * semaphore. 170 * 171 * Pure readers hold dev_base_lock for reading. 172 * 173 * Writers must hold the rtnl semaphore while they loop through the 174 * dev_base list, and hold dev_base_lock for writing when they do the 175 * actual updates. This allows pure readers to access the list even 176 * while a writer is preparing to update it. 177 * 178 * To put it another way, dev_base_lock is held for writing only to 179 * protect against pure readers; the rtnl semaphore provides the 180 * protection against other writers. 181 * 182 * See, for example usages, register_netdevice() and 183 * unregister_netdevice(), which must be called with the rtnl 184 * semaphore held. 185 */ 186struct net_device *dev_base; 187static struct net_device **dev_tail = &dev_base; 188DEFINE_RWLOCK(dev_base_lock); 189 190EXPORT_SYMBOL(dev_base); 191EXPORT_SYMBOL(dev_base_lock); 192 193#define NETDEV_HASHBITS 8 194static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS]; 195static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS]; 196 197static inline struct hlist_head *dev_name_hash(const char *name) 198{ 199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); 200 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)]; 201} 202 203static inline struct hlist_head *dev_index_hash(int ifindex) 204{ 205 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)]; 206} 207 208/* 209 * Our notifier list 210 */ 211 212static struct notifier_block *netdev_chain; 213 214/* 215 * Device drivers call our routines to queue packets here. We empty the 216 * queue in the local softnet handler. 217 */ 218DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, }; 219 220#ifdef CONFIG_SYSFS 221extern int netdev_sysfs_init(void); 222extern int netdev_register_sysfs(struct net_device *); 223extern void netdev_unregister_sysfs(struct net_device *); 224#else 225#define netdev_sysfs_init() (0) 226#define netdev_register_sysfs(dev) (0) 227#define netdev_unregister_sysfs(dev) do { } while(0) 228#endif 229 230 231/******************************************************************************* 232 233 Protocol management and registration routines 234 235*******************************************************************************/ 236 237/* 238 * For efficiency 239 */ 240 241int netdev_nit; 242 243/* 244 * Add a protocol ID to the list. Now that the input handler is 245 * smarter we can dispense with all the messy stuff that used to be 246 * here. 247 * 248 * BEWARE!!! Protocol handlers, mangling input packets, 249 * MUST BE last in hash buckets and checking protocol handlers 250 * MUST start from promiscuous ptype_all chain in net_bh. 251 * It is true now, do not change it. 252 * Explanation follows: if protocol handler, mangling packet, will 253 * be the first on list, it is not able to sense, that packet 254 * is cloned and should be copied-on-write, so that it will 255 * change it and subsequent readers will get broken packet. 256 * --ANK (980803) 257 */ 258 259/** 260 * dev_add_pack - add packet handler 261 * @pt: packet type declaration 262 * 263 * Add a protocol handler to the networking stack. The passed &packet_type 264 * is linked into kernel lists and may not be freed until it has been 265 * removed from the kernel lists. 266 * 267 * This call does not sleep therefore it can not 268 * guarantee all CPU's that are in middle of receiving packets 269 * will see the new packet type (until the next received packet). 270 */ 271 272void dev_add_pack(struct packet_type *pt) 273{ 274 int hash; 275 276 spin_lock_bh(&ptype_lock); 277 if (pt->type == htons(ETH_P_ALL)) { 278 netdev_nit++; 279 list_add_rcu(&pt->list, &ptype_all); 280 } else { 281 hash = ntohs(pt->type) & 15; 282 list_add_rcu(&pt->list, &ptype_base[hash]); 283 } 284 spin_unlock_bh(&ptype_lock); 285} 286 287extern void linkwatch_run_queue(void); 288 289 290 291/** 292 * __dev_remove_pack - remove packet handler 293 * @pt: packet type declaration 294 * 295 * Remove a protocol handler that was previously added to the kernel 296 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 297 * from the kernel lists and can be freed or reused once this function 298 * returns. 299 * 300 * The packet type might still be in use by receivers 301 * and must not be freed until after all the CPU's have gone 302 * through a quiescent state. 303 */ 304void __dev_remove_pack(struct packet_type *pt) 305{ 306 struct list_head *head; 307 struct packet_type *pt1; 308 309 spin_lock_bh(&ptype_lock); 310 311 if (pt->type == htons(ETH_P_ALL)) { 312 netdev_nit--; 313 head = &ptype_all; 314 } else 315 head = &ptype_base[ntohs(pt->type) & 15]; 316 317 list_for_each_entry(pt1, head, list) { 318 if (pt == pt1) { 319 list_del_rcu(&pt->list); 320 goto out; 321 } 322 } 323 324 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); 325out: 326 spin_unlock_bh(&ptype_lock); 327} 328/** 329 * dev_remove_pack - remove packet handler 330 * @pt: packet type declaration 331 * 332 * Remove a protocol handler that was previously added to the kernel 333 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 334 * from the kernel lists and can be freed or reused once this function 335 * returns. 336 * 337 * This call sleeps to guarantee that no CPU is looking at the packet 338 * type after return. 339 */ 340void dev_remove_pack(struct packet_type *pt) 341{ 342 __dev_remove_pack(pt); 343 344 synchronize_net(); 345} 346 347/****************************************************************************** 348 349 Device Boot-time Settings Routines 350 351*******************************************************************************/ 352 353/* Boot time configuration table */ 354static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 355 356/** 357 * netdev_boot_setup_add - add new setup entry 358 * @name: name of the device 359 * @map: configured settings for the device 360 * 361 * Adds new setup entry to the dev_boot_setup list. The function 362 * returns 0 on error and 1 on success. This is a generic routine to 363 * all netdevices. 364 */ 365static int netdev_boot_setup_add(char *name, struct ifmap *map) 366{ 367 struct netdev_boot_setup *s; 368 int i; 369 370 s = dev_boot_setup; 371 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 372 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 373 memset(s[i].name, 0, sizeof(s[i].name)); 374 strcpy(s[i].name, name); 375 memcpy(&s[i].map, map, sizeof(s[i].map)); 376 break; 377 } 378 } 379 380 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 381} 382 383/** 384 * netdev_boot_setup_check - check boot time settings 385 * @dev: the netdevice 386 * 387 * Check boot time settings for the device. 388 * The found settings are set for the device to be used 389 * later in the device probing. 390 * Returns 0 if no settings found, 1 if they are. 391 */ 392int netdev_boot_setup_check(struct net_device *dev) 393{ 394 struct netdev_boot_setup *s = dev_boot_setup; 395 int i; 396 397 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 398 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 399 !strncmp(dev->name, s[i].name, strlen(s[i].name))) { 400 dev->irq = s[i].map.irq; 401 dev->base_addr = s[i].map.base_addr; 402 dev->mem_start = s[i].map.mem_start; 403 dev->mem_end = s[i].map.mem_end; 404 return 1; 405 } 406 } 407 return 0; 408} 409 410 411/** 412 * netdev_boot_base - get address from boot time settings 413 * @prefix: prefix for network device 414 * @unit: id for network device 415 * 416 * Check boot time settings for the base address of device. 417 * The found settings are set for the device to be used 418 * later in the device probing. 419 * Returns 0 if no settings found. 420 */ 421unsigned long netdev_boot_base(const char *prefix, int unit) 422{ 423 const struct netdev_boot_setup *s = dev_boot_setup; 424 char name[IFNAMSIZ]; 425 int i; 426 427 sprintf(name, "%s%d", prefix, unit); 428 429 /* 430 * If device already registered then return base of 1 431 * to indicate not to probe for this interface 432 */ 433 if (__dev_get_by_name(name)) 434 return 1; 435 436 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 437 if (!strcmp(name, s[i].name)) 438 return s[i].map.base_addr; 439 return 0; 440} 441 442/* 443 * Saves at boot time configured settings for any netdevice. 444 */ 445int __init netdev_boot_setup(char *str) 446{ 447 int ints[5]; 448 struct ifmap map; 449 450 str = get_options(str, ARRAY_SIZE(ints), ints); 451 if (!str || !*str) 452 return 0; 453 454 /* Save settings */ 455 memset(&map, 0, sizeof(map)); 456 if (ints[0] > 0) 457 map.irq = ints[1]; 458 if (ints[0] > 1) 459 map.base_addr = ints[2]; 460 if (ints[0] > 2) 461 map.mem_start = ints[3]; 462 if (ints[0] > 3) 463 map.mem_end = ints[4]; 464 465 /* Add new entry to the list */ 466 return netdev_boot_setup_add(str, &map); 467} 468 469__setup("netdev=", netdev_boot_setup); 470 471/******************************************************************************* 472 473 Device Interface Subroutines 474 475*******************************************************************************/ 476 477/** 478 * __dev_get_by_name - find a device by its name 479 * @name: name to find 480 * 481 * Find an interface by name. Must be called under RTNL semaphore 482 * or @dev_base_lock. If the name is found a pointer to the device 483 * is returned. If the name is not found then %NULL is returned. The 484 * reference counters are not incremented so the caller must be 485 * careful with locks. 486 */ 487 488struct net_device *__dev_get_by_name(const char *name) 489{ 490 struct hlist_node *p; 491 492 hlist_for_each(p, dev_name_hash(name)) { 493 struct net_device *dev 494 = hlist_entry(p, struct net_device, name_hlist); 495 if (!strncmp(dev->name, name, IFNAMSIZ)) 496 return dev; 497 } 498 return NULL; 499} 500 501/** 502 * dev_get_by_name - find a device by its name 503 * @name: name to find 504 * 505 * Find an interface by name. This can be called from any 506 * context and does its own locking. The returned handle has 507 * the usage count incremented and the caller must use dev_put() to 508 * release it when it is no longer needed. %NULL is returned if no 509 * matching device is found. 510 */ 511 512struct net_device *dev_get_by_name(const char *name) 513{ 514 struct net_device *dev; 515 516 read_lock(&dev_base_lock); 517 dev = __dev_get_by_name(name); 518 if (dev) 519 dev_hold(dev); 520 read_unlock(&dev_base_lock); 521 return dev; 522} 523 524/** 525 * __dev_get_by_index - find a device by its ifindex 526 * @ifindex: index of device 527 * 528 * Search for an interface by index. Returns %NULL if the device 529 * is not found or a pointer to the device. The device has not 530 * had its reference counter increased so the caller must be careful 531 * about locking. The caller must hold either the RTNL semaphore 532 * or @dev_base_lock. 533 */ 534 535struct net_device *__dev_get_by_index(int ifindex) 536{ 537 struct hlist_node *p; 538 539 hlist_for_each(p, dev_index_hash(ifindex)) { 540 struct net_device *dev 541 = hlist_entry(p, struct net_device, index_hlist); 542 if (dev->ifindex == ifindex) 543 return dev; 544 } 545 return NULL; 546} 547 548 549/** 550 * dev_get_by_index - find a device by its ifindex 551 * @ifindex: index of device 552 * 553 * Search for an interface by index. Returns NULL if the device 554 * is not found or a pointer to the device. The device returned has 555 * had a reference added and the pointer is safe until the user calls 556 * dev_put to indicate they have finished with it. 557 */ 558 559struct net_device *dev_get_by_index(int ifindex) 560{ 561 struct net_device *dev; 562 563 read_lock(&dev_base_lock); 564 dev = __dev_get_by_index(ifindex); 565 if (dev) 566 dev_hold(dev); 567 read_unlock(&dev_base_lock); 568 return dev; 569} 570 571/** 572 * dev_getbyhwaddr - find a device by its hardware address 573 * @type: media type of device 574 * @ha: hardware address 575 * 576 * Search for an interface by MAC address. Returns NULL if the device 577 * is not found or a pointer to the device. The caller must hold the 578 * rtnl semaphore. The returned device has not had its ref count increased 579 * and the caller must therefore be careful about locking 580 * 581 * BUGS: 582 * If the API was consistent this would be __dev_get_by_hwaddr 583 */ 584 585struct net_device *dev_getbyhwaddr(unsigned short type, char *ha) 586{ 587 struct net_device *dev; 588 589 ASSERT_RTNL(); 590 591 for (dev = dev_base; dev; dev = dev->next) 592 if (dev->type == type && 593 !memcmp(dev->dev_addr, ha, dev->addr_len)) 594 break; 595 return dev; 596} 597 598struct net_device *dev_getfirstbyhwtype(unsigned short type) 599{ 600 struct net_device *dev; 601 602 rtnl_lock(); 603 for (dev = dev_base; dev; dev = dev->next) { 604 if (dev->type == type) { 605 dev_hold(dev); 606 break; 607 } 608 } 609 rtnl_unlock(); 610 return dev; 611} 612 613EXPORT_SYMBOL(dev_getfirstbyhwtype); 614 615/** 616 * dev_get_by_flags - find any device with given flags 617 * @if_flags: IFF_* values 618 * @mask: bitmask of bits in if_flags to check 619 * 620 * Search for any interface with the given flags. Returns NULL if a device 621 * is not found or a pointer to the device. The device returned has 622 * had a reference added and the pointer is safe until the user calls 623 * dev_put to indicate they have finished with it. 624 */ 625 626struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask) 627{ 628 struct net_device *dev; 629 630 read_lock(&dev_base_lock); 631 for (dev = dev_base; dev != NULL; dev = dev->next) { 632 if (((dev->flags ^ if_flags) & mask) == 0) { 633 dev_hold(dev); 634 break; 635 } 636 } 637 read_unlock(&dev_base_lock); 638 return dev; 639} 640 641/** 642 * dev_valid_name - check if name is okay for network device 643 * @name: name string 644 * 645 * Network device names need to be valid file names to 646 * to allow sysfs to work 647 */ 648static int dev_valid_name(const char *name) 649{ 650 return !(*name == '\0' 651 || !strcmp(name, ".") 652 || !strcmp(name, "..") 653 || strchr(name, '/')); 654} 655 656/** 657 * dev_alloc_name - allocate a name for a device 658 * @dev: device 659 * @name: name format string 660 * 661 * Passed a format string - eg "lt%d" it will try and find a suitable 662 * id. Not efficient for many devices, not called a lot. The caller 663 * must hold the dev_base or rtnl lock while allocating the name and 664 * adding the device in order to avoid duplicates. Returns the number 665 * of the unit assigned or a negative errno code. 666 */ 667 668int dev_alloc_name(struct net_device *dev, const char *name) 669{ 670 int i = 0; 671 char buf[IFNAMSIZ]; 672 const char *p; 673 const int max_netdevices = 8*PAGE_SIZE; 674 long *inuse; 675 struct net_device *d; 676 677 p = strnchr(name, IFNAMSIZ-1, '%'); 678 if (p) { 679 /* 680 * Verify the string as this thing may have come from 681 * the user. There must be either one "%d" and no other "%" 682 * characters. 683 */ 684 if (p[1] != 'd' || strchr(p + 2, '%')) 685 return -EINVAL; 686 687 /* Use one page as a bit array of possible slots */ 688 inuse = (long *) get_zeroed_page(GFP_ATOMIC); 689 if (!inuse) 690 return -ENOMEM; 691 692 for (d = dev_base; d; d = d->next) { 693 if (!sscanf(d->name, name, &i)) 694 continue; 695 if (i < 0 || i >= max_netdevices) 696 continue; 697 698 /* avoid cases where sscanf is not exact inverse of printf */ 699 snprintf(buf, sizeof(buf), name, i); 700 if (!strncmp(buf, d->name, IFNAMSIZ)) 701 set_bit(i, inuse); 702 } 703 704 i = find_first_zero_bit(inuse, max_netdevices); 705 free_page((unsigned long) inuse); 706 } 707 708 snprintf(buf, sizeof(buf), name, i); 709 if (!__dev_get_by_name(buf)) { 710 strlcpy(dev->name, buf, IFNAMSIZ); 711 return i; 712 } 713 714 /* It is possible to run out of possible slots 715 * when the name is long and there isn't enough space left 716 * for the digits, or if all bits are used. 717 */ 718 return -ENFILE; 719} 720 721 722/** 723 * dev_change_name - change name of a device 724 * @dev: device 725 * @newname: name (or format string) must be at least IFNAMSIZ 726 * 727 * Change name of a device, can pass format strings "eth%d". 728 * for wildcarding. 729 */ 730int dev_change_name(struct net_device *dev, char *newname) 731{ 732 int err = 0; 733 734 ASSERT_RTNL(); 735 736 if (dev->flags & IFF_UP) 737 return -EBUSY; 738 739 if (!dev_valid_name(newname)) 740 return -EINVAL; 741 742 if (strchr(newname, '%')) { 743 err = dev_alloc_name(dev, newname); 744 if (err < 0) 745 return err; 746 strcpy(newname, dev->name); 747 } 748 else if (__dev_get_by_name(newname)) 749 return -EEXIST; 750 else 751 strlcpy(dev->name, newname, IFNAMSIZ); 752 753 err = class_device_rename(&dev->class_dev, dev->name); 754 if (!err) { 755 hlist_del(&dev->name_hlist); 756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name)); 757 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev); 758 } 759 760 return err; 761} 762 763/** 764 * netdev_features_change - device changes fatures 765 * @dev: device to cause notification 766 * 767 * Called to indicate a device has changed features. 768 */ 769void netdev_features_change(struct net_device *dev) 770{ 771 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev); 772} 773EXPORT_SYMBOL(netdev_features_change); 774 775/** 776 * netdev_state_change - device changes state 777 * @dev: device to cause notification 778 * 779 * Called to indicate a device has changed state. This function calls 780 * the notifier chains for netdev_chain and sends a NEWLINK message 781 * to the routing socket. 782 */ 783void netdev_state_change(struct net_device *dev) 784{ 785 if (dev->flags & IFF_UP) { 786 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev); 787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0); 788 } 789} 790 791/** 792 * dev_load - load a network module 793 * @name: name of interface 794 * 795 * If a network interface is not present and the process has suitable 796 * privileges this function loads the module. If module loading is not 797 * available in this kernel then it becomes a nop. 798 */ 799 800void dev_load(const char *name) 801{ 802 struct net_device *dev; 803 804 read_lock(&dev_base_lock); 805 dev = __dev_get_by_name(name); 806 read_unlock(&dev_base_lock); 807 808 if (!dev && capable(CAP_SYS_MODULE)) 809 request_module("%s", name); 810} 811 812static int default_rebuild_header(struct sk_buff *skb) 813{ 814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n", 815 skb->dev ? skb->dev->name : "NULL!!!"); 816 kfree_skb(skb); 817 return 1; 818} 819 820 821/** 822 * dev_open - prepare an interface for use. 823 * @dev: device to open 824 * 825 * Takes a device from down to up state. The device's private open 826 * function is invoked and then the multicast lists are loaded. Finally 827 * the device is moved into the up state and a %NETDEV_UP message is 828 * sent to the netdev notifier chain. 829 * 830 * Calling this function on an active interface is a nop. On a failure 831 * a negative errno code is returned. 832 */ 833int dev_open(struct net_device *dev) 834{ 835 int ret = 0; 836 837 /* 838 * Is it already up? 839 */ 840 841 if (dev->flags & IFF_UP) 842 return 0; 843 844 /* 845 * Is it even present? 846 */ 847 if (!netif_device_present(dev)) 848 return -ENODEV; 849 850 /* 851 * Call device private open method 852 */ 853 set_bit(__LINK_STATE_START, &dev->state); 854 if (dev->open) { 855 ret = dev->open(dev); 856 if (ret) 857 clear_bit(__LINK_STATE_START, &dev->state); 858 } 859 860 /* 861 * If it went open OK then: 862 */ 863 864 if (!ret) { 865 /* 866 * Set the flags. 867 */ 868 dev->flags |= IFF_UP; 869 870 /* 871 * Initialize multicasting status 872 */ 873 dev_mc_upload(dev); 874 875 /* 876 * Wakeup transmit queue engine 877 */ 878 dev_activate(dev); 879 880 /* 881 * ... and announce new interface. 882 */ 883 notifier_call_chain(&netdev_chain, NETDEV_UP, dev); 884 } 885 return ret; 886} 887 888/** 889 * dev_close - shutdown an interface. 890 * @dev: device to shutdown 891 * 892 * This function moves an active device into down state. A 893 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 894 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 895 * chain. 896 */ 897int dev_close(struct net_device *dev) 898{ 899 if (!(dev->flags & IFF_UP)) 900 return 0; 901 902 /* 903 * Tell people we are going down, so that they can 904 * prepare to death, when device is still operating. 905 */ 906 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev); 907 908 dev_deactivate(dev); 909 910 clear_bit(__LINK_STATE_START, &dev->state); 911 912 /* Synchronize to scheduled poll. We cannot touch poll list, 913 * it can be even on different cpu. So just clear netif_running(), 914 * and wait when poll really will happen. Actually, the best place 915 * for this is inside dev->stop() after device stopped its irq 916 * engine, but this requires more changes in devices. */ 917 918 smp_mb__after_clear_bit(); /* Commit netif_running(). */ 919 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) { 920 /* No hurry. */ 921 current->state = TASK_INTERRUPTIBLE; 922 schedule_timeout(1); 923 } 924 925 /* 926 * Call the device specific close. This cannot fail. 927 * Only if device is UP 928 * 929 * We allow it to be called even after a DETACH hot-plug 930 * event. 931 */ 932 if (dev->stop) 933 dev->stop(dev); 934 935 /* 936 * Device is now down. 937 */ 938 939 dev->flags &= ~IFF_UP; 940 941 /* 942 * Tell people we are down 943 */ 944 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev); 945 946 return 0; 947} 948 949 950/* 951 * Device change register/unregister. These are not inline or static 952 * as we export them to the world. 953 */ 954 955/** 956 * register_netdevice_notifier - register a network notifier block 957 * @nb: notifier 958 * 959 * Register a notifier to be called when network device events occur. 960 * The notifier passed is linked into the kernel structures and must 961 * not be reused until it has been unregistered. A negative errno code 962 * is returned on a failure. 963 * 964 * When registered all registration and up events are replayed 965 * to the new notifier to allow device to have a race free 966 * view of the network device list. 967 */ 968 969int register_netdevice_notifier(struct notifier_block *nb) 970{ 971 struct net_device *dev; 972 int err; 973 974 rtnl_lock(); 975 err = notifier_chain_register(&netdev_chain, nb); 976 if (!err) { 977 for (dev = dev_base; dev; dev = dev->next) { 978 nb->notifier_call(nb, NETDEV_REGISTER, dev); 979 980 if (dev->flags & IFF_UP) 981 nb->notifier_call(nb, NETDEV_UP, dev); 982 } 983 } 984 rtnl_unlock(); 985 return err; 986} 987 988/** 989 * unregister_netdevice_notifier - unregister a network notifier block 990 * @nb: notifier 991 * 992 * Unregister a notifier previously registered by 993 * register_netdevice_notifier(). The notifier is unlinked into the 994 * kernel structures and may then be reused. A negative errno code 995 * is returned on a failure. 996 */ 997 998int unregister_netdevice_notifier(struct notifier_block *nb) 999{ 1000 return notifier_chain_unregister(&netdev_chain, nb); 1001} 1002 1003/** 1004 * call_netdevice_notifiers - call all network notifier blocks 1005 * @val: value passed unmodified to notifier function 1006 * @v: pointer passed unmodified to notifier function 1007 * 1008 * Call all network notifier blocks. Parameters and return value 1009 * are as for notifier_call_chain(). 1010 */ 1011 1012int call_netdevice_notifiers(unsigned long val, void *v) 1013{ 1014 return notifier_call_chain(&netdev_chain, val, v); 1015} 1016 1017/* When > 0 there are consumers of rx skb time stamps */ 1018static atomic_t netstamp_needed = ATOMIC_INIT(0); 1019 1020void net_enable_timestamp(void) 1021{ 1022 atomic_inc(&netstamp_needed); 1023} 1024 1025void net_disable_timestamp(void) 1026{ 1027 atomic_dec(&netstamp_needed); 1028} 1029 1030static inline void net_timestamp(struct timeval *stamp) 1031{ 1032 if (atomic_read(&netstamp_needed)) 1033 do_gettimeofday(stamp); 1034 else { 1035 stamp->tv_sec = 0; 1036 stamp->tv_usec = 0; 1037 } 1038} 1039 1040/* 1041 * Support routine. Sends outgoing frames to any network 1042 * taps currently in use. 1043 */ 1044 1045void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1046{ 1047 struct packet_type *ptype; 1048 net_timestamp(&skb->stamp); 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); 1080 } 1081 } 1082 rcu_read_unlock(); 1083} 1084 1085/* 1086 * Invalidate hardware checksum when packet is to be mangled, and 1087 * complete checksum manually on outgoing path. 1088 */ 1089int skb_checksum_help(struct sk_buff *skb, int inward) 1090{ 1091 unsigned int csum; 1092 int ret = 0, offset = skb->h.raw - skb->data; 1093 1094 if (inward) { 1095 skb->ip_summed = CHECKSUM_NONE; 1096 goto out; 1097 } 1098 1099 if (skb_cloned(skb)) { 1100 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 1101 if (ret) 1102 goto out; 1103 } 1104 1105 if (offset > (int)skb->len) 1106 BUG(); 1107 csum = skb_checksum(skb, offset, skb->len-offset, 0); 1108 1109 offset = skb->tail - skb->h.raw; 1110 if (offset <= 0) 1111 BUG(); 1112 if (skb->csum + 2 > offset) 1113 BUG(); 1114 1115 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum); 1116 skb->ip_summed = CHECKSUM_NONE; 1117out: 1118 return ret; 1119} 1120 1121#ifdef CONFIG_HIGHMEM 1122/* Actually, we should eliminate this check as soon as we know, that: 1123 * 1. IOMMU is present and allows to map all the memory. 1124 * 2. No high memory really exists on this machine. 1125 */ 1126 1127static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 1128{ 1129 int i; 1130 1131 if (dev->features & NETIF_F_HIGHDMA) 1132 return 0; 1133 1134 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1135 if (PageHighMem(skb_shinfo(skb)->frags[i].page)) 1136 return 1; 1137 1138 return 0; 1139} 1140#else 1141#define illegal_highdma(dev, skb) (0) 1142#endif 1143 1144extern void skb_release_data(struct sk_buff *); 1145 1146/* Keep head the same: replace data */ 1147int __skb_linearize(struct sk_buff *skb, int gfp_mask) 1148{ 1149 unsigned int size; 1150 u8 *data; 1151 long offset; 1152 struct skb_shared_info *ninfo; 1153 int headerlen = skb->data - skb->head; 1154 int expand = (skb->tail + skb->data_len) - skb->end; 1155 1156 if (skb_shared(skb)) 1157 BUG(); 1158 1159 if (expand <= 0) 1160 expand = 0; 1161 1162 size = skb->end - skb->head + expand; 1163 size = SKB_DATA_ALIGN(size); 1164 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 1165 if (!data) 1166 return -ENOMEM; 1167 1168 /* Copy entire thing */ 1169 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len)) 1170 BUG(); 1171 1172 /* Set up shinfo */ 1173 ninfo = (struct skb_shared_info*)(data + size); 1174 atomic_set(&ninfo->dataref, 1); 1175 ninfo->tso_size = skb_shinfo(skb)->tso_size; 1176 ninfo->tso_segs = skb_shinfo(skb)->tso_segs; 1177 ninfo->nr_frags = 0; 1178 ninfo->frag_list = NULL; 1179 1180 /* Offset between the two in bytes */ 1181 offset = data - skb->head; 1182 1183 /* Free old data. */ 1184 skb_release_data(skb); 1185 1186 skb->head = data; 1187 skb->end = data + size; 1188 1189 /* Set up new pointers */ 1190 skb->h.raw += offset; 1191 skb->nh.raw += offset; 1192 skb->mac.raw += offset; 1193 skb->tail += offset; 1194 skb->data += offset; 1195 1196 /* We are no longer a clone, even if we were. */ 1197 skb->cloned = 0; 1198 1199 skb->tail += skb->data_len; 1200 skb->data_len = 0; 1201 return 0; 1202} 1203 1204#define HARD_TX_LOCK(dev, cpu) { \ 1205 if ((dev->features & NETIF_F_LLTX) == 0) { \ 1206 spin_lock(&dev->xmit_lock); \ 1207 dev->xmit_lock_owner = cpu; \ 1208 } \ 1209} 1210 1211#define HARD_TX_UNLOCK(dev) { \ 1212 if ((dev->features & NETIF_F_LLTX) == 0) { \ 1213 dev->xmit_lock_owner = -1; \ 1214 spin_unlock(&dev->xmit_lock); \ 1215 } \ 1216} 1217 1218/** 1219 * dev_queue_xmit - transmit a buffer 1220 * @skb: buffer to transmit 1221 * 1222 * Queue a buffer for transmission to a network device. The caller must 1223 * have set the device and priority and built the buffer before calling 1224 * this function. The function can be called from an interrupt. 1225 * 1226 * A negative errno code is returned on a failure. A success does not 1227 * guarantee the frame will be transmitted as it may be dropped due 1228 * to congestion or traffic shaping. 1229 * 1230 * ----------------------------------------------------------------------------------- 1231 * I notice this method can also return errors from the queue disciplines, 1232 * including NET_XMIT_DROP, which is a positive value. So, errors can also 1233 * be positive. 1234 * 1235 * Regardless of the return value, the skb is consumed, so it is currently 1236 * difficult to retry a send to this method. (You can bump the ref count 1237 * before sending to hold a reference for retry if you are careful.) 1238 * 1239 * When calling this method, interrupts MUST be enabled. This is because 1240 * the BH enable code must have IRQs enabled so that it will not deadlock. 1241 * --BLG 1242 */ 1243 1244int dev_queue_xmit(struct sk_buff *skb) 1245{ 1246 struct net_device *dev = skb->dev; 1247 struct Qdisc *q; 1248 int rc = -ENOMEM; 1249 1250 if (skb_shinfo(skb)->frag_list && 1251 !(dev->features & NETIF_F_FRAGLIST) && 1252 __skb_linearize(skb, GFP_ATOMIC)) 1253 goto out_kfree_skb; 1254 1255 /* Fragmented skb is linearized if device does not support SG, 1256 * or if at least one of fragments is in highmem and device 1257 * does not support DMA from it. 1258 */ 1259 if (skb_shinfo(skb)->nr_frags && 1260 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && 1261 __skb_linearize(skb, GFP_ATOMIC)) 1262 goto out_kfree_skb; 1263 1264 /* If packet is not checksummed and device does not support 1265 * checksumming for this protocol, complete checksumming here. 1266 */ 1267 if (skb->ip_summed == CHECKSUM_HW && 1268 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) && 1269 (!(dev->features & NETIF_F_IP_CSUM) || 1270 skb->protocol != htons(ETH_P_IP)))) 1271 if (skb_checksum_help(skb, 0)) 1272 goto out_kfree_skb; 1273 1274 /* Disable soft irqs for various locks below. Also 1275 * stops preemption for RCU. 1276 */ 1277 local_bh_disable(); 1278 1279 /* Updates of qdisc are serialized by queue_lock. 1280 * The struct Qdisc which is pointed to by qdisc is now a 1281 * rcu structure - it may be accessed without acquiring 1282 * a lock (but the structure may be stale.) The freeing of the 1283 * qdisc will be deferred until it's known that there are no 1284 * more references to it. 1285 * 1286 * If the qdisc has an enqueue function, we still need to 1287 * hold the queue_lock before calling it, since queue_lock 1288 * also serializes access to the device queue. 1289 */ 1290 1291 q = rcu_dereference(dev->qdisc); 1292#ifdef CONFIG_NET_CLS_ACT 1293 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); 1294#endif 1295 if (q->enqueue) { 1296 /* Grab device queue */ 1297 spin_lock(&dev->queue_lock); 1298 1299 rc = q->enqueue(skb, q); 1300 1301 qdisc_run(dev); 1302 1303 spin_unlock(&dev->queue_lock); 1304 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc; 1305 goto out; 1306 } 1307 1308 /* The device has no queue. Common case for software devices: 1309 loopback, all the sorts of tunnels... 1310 1311 Really, it is unlikely that xmit_lock protection is necessary here. 1312 (f.e. loopback and IP tunnels are clean ignoring statistics 1313 counters.) 1314 However, it is possible, that they rely on protection 1315 made by us here. 1316 1317 Check this and shot the lock. It is not prone from deadlocks. 1318 Either shot noqueue qdisc, it is even simpler 8) 1319 */ 1320 if (dev->flags & IFF_UP) { 1321 int cpu = smp_processor_id(); /* ok because BHs are off */ 1322 1323 if (dev->xmit_lock_owner != cpu) { 1324 1325 HARD_TX_LOCK(dev, cpu); 1326 1327 if (!netif_queue_stopped(dev)) { 1328 if (netdev_nit) 1329 dev_queue_xmit_nit(skb, dev); 1330 1331 rc = 0; 1332 if (!dev->hard_start_xmit(skb, dev)) { 1333 HARD_TX_UNLOCK(dev); 1334 goto out; 1335 } 1336 } 1337 HARD_TX_UNLOCK(dev); 1338 if (net_ratelimit()) 1339 printk(KERN_CRIT "Virtual device %s asks to " 1340 "queue packet!\n", dev->name); 1341 } else { 1342 /* Recursion is detected! It is possible, 1343 * unfortunately */ 1344 if (net_ratelimit()) 1345 printk(KERN_CRIT "Dead loop on virtual device " 1346 "%s, fix it urgently!\n", dev->name); 1347 } 1348 } 1349 1350 rc = -ENETDOWN; 1351 local_bh_enable(); 1352 1353out_kfree_skb: 1354 kfree_skb(skb); 1355 return rc; 1356out: 1357 local_bh_enable(); 1358 return rc; 1359} 1360 1361 1362/*======================================================================= 1363 Receiver routines 1364 =======================================================================*/ 1365 1366int netdev_max_backlog = 300; 1367int weight_p = 64; /* old backlog weight */ 1368/* These numbers are selected based on intuition and some 1369 * experimentatiom, if you have more scientific way of doing this 1370 * please go ahead and fix things. 1371 */ 1372int no_cong_thresh = 10; 1373int no_cong = 20; 1374int lo_cong = 100; 1375int mod_cong = 290; 1376 1377DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; 1378 1379 1380static void get_sample_stats(int cpu) 1381{ 1382#ifdef RAND_LIE 1383 unsigned long rd; 1384 int rq; 1385#endif 1386 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 1387 int blog = sd->input_pkt_queue.qlen; 1388 int avg_blog = sd->avg_blog; 1389 1390 avg_blog = (avg_blog >> 1) + (blog >> 1); 1391 1392 if (avg_blog > mod_cong) { 1393 /* Above moderate congestion levels. */ 1394 sd->cng_level = NET_RX_CN_HIGH; 1395#ifdef RAND_LIE 1396 rd = net_random(); 1397 rq = rd % netdev_max_backlog; 1398 if (rq < avg_blog) /* unlucky bastard */ 1399 sd->cng_level = NET_RX_DROP; 1400#endif 1401 } else if (avg_blog > lo_cong) { 1402 sd->cng_level = NET_RX_CN_MOD; 1403#ifdef RAND_LIE 1404 rd = net_random(); 1405 rq = rd % netdev_max_backlog; 1406 if (rq < avg_blog) /* unlucky bastard */ 1407 sd->cng_level = NET_RX_CN_HIGH; 1408#endif 1409 } else if (avg_blog > no_cong) 1410 sd->cng_level = NET_RX_CN_LOW; 1411 else /* no congestion */ 1412 sd->cng_level = NET_RX_SUCCESS; 1413 1414 sd->avg_blog = avg_blog; 1415} 1416 1417#ifdef OFFLINE_SAMPLE 1418static void sample_queue(unsigned long dummy) 1419{ 1420/* 10 ms 0r 1ms -- i don't care -- JHS */ 1421 int next_tick = 1; 1422 int cpu = smp_processor_id(); 1423 1424 get_sample_stats(cpu); 1425 next_tick += jiffies; 1426 mod_timer(&samp_timer, next_tick); 1427} 1428#endif 1429 1430 1431/** 1432 * netif_rx - post buffer to the network code 1433 * @skb: buffer to post 1434 * 1435 * This function receives a packet from a device driver and queues it for 1436 * the upper (protocol) levels to process. It always succeeds. The buffer 1437 * may be dropped during processing for congestion control or by the 1438 * protocol layers. 1439 * 1440 * return values: 1441 * NET_RX_SUCCESS (no congestion) 1442 * NET_RX_CN_LOW (low congestion) 1443 * NET_RX_CN_MOD (moderate congestion) 1444 * NET_RX_CN_HIGH (high congestion) 1445 * NET_RX_DROP (packet was dropped) 1446 * 1447 */ 1448 1449int netif_rx(struct sk_buff *skb) 1450{ 1451 int this_cpu; 1452 struct softnet_data *queue; 1453 unsigned long flags; 1454 1455 /* if netpoll wants it, pretend we never saw it */ 1456 if (netpoll_rx(skb)) 1457 return NET_RX_DROP; 1458 1459 if (!skb->stamp.tv_sec) 1460 net_timestamp(&skb->stamp); 1461 1462 /* 1463 * The code is rearranged so that the path is the most 1464 * short when CPU is congested, but is still operating. 1465 */ 1466 local_irq_save(flags); 1467 this_cpu = smp_processor_id(); 1468 queue = &__get_cpu_var(softnet_data); 1469 1470 __get_cpu_var(netdev_rx_stat).total++; 1471 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { 1472 if (queue->input_pkt_queue.qlen) { 1473 if (queue->throttle) 1474 goto drop; 1475 1476enqueue: 1477 dev_hold(skb->dev); 1478 __skb_queue_tail(&queue->input_pkt_queue, skb); 1479#ifndef OFFLINE_SAMPLE 1480 get_sample_stats(this_cpu); 1481#endif 1482 local_irq_restore(flags); 1483 return queue->cng_level; 1484 } 1485 1486 if (queue->throttle) 1487 queue->throttle = 0; 1488 1489 netif_rx_schedule(&queue->backlog_dev); 1490 goto enqueue; 1491 } 1492 1493 if (!queue->throttle) { 1494 queue->throttle = 1; 1495 __get_cpu_var(netdev_rx_stat).throttled++; 1496 } 1497 1498drop: 1499 __get_cpu_var(netdev_rx_stat).dropped++; 1500 local_irq_restore(flags); 1501 1502 kfree_skb(skb); 1503 return NET_RX_DROP; 1504} 1505 1506int netif_rx_ni(struct sk_buff *skb) 1507{ 1508 int err; 1509 1510 preempt_disable(); 1511 err = netif_rx(skb); 1512 if (local_softirq_pending()) 1513 do_softirq(); 1514 preempt_enable(); 1515 1516 return err; 1517} 1518 1519EXPORT_SYMBOL(netif_rx_ni); 1520 1521static __inline__ void skb_bond(struct sk_buff *skb) 1522{ 1523 struct net_device *dev = skb->dev; 1524 1525 if (dev->master) { 1526 skb->real_dev = skb->dev; 1527 skb->dev = dev->master; 1528 } 1529} 1530 1531static void net_tx_action(struct softirq_action *h) 1532{ 1533 struct softnet_data *sd = &__get_cpu_var(softnet_data); 1534 1535 if (sd->completion_queue) { 1536 struct sk_buff *clist; 1537 1538 local_irq_disable(); 1539 clist = sd->completion_queue; 1540 sd->completion_queue = NULL; 1541 local_irq_enable(); 1542 1543 while (clist) { 1544 struct sk_buff *skb = clist; 1545 clist = clist->next; 1546 1547 BUG_TRAP(!atomic_read(&skb->users)); 1548 __kfree_skb(skb); 1549 } 1550 } 1551 1552 if (sd->output_queue) { 1553 struct net_device *head; 1554 1555 local_irq_disable(); 1556 head = sd->output_queue; 1557 sd->output_queue = NULL; 1558 local_irq_enable(); 1559 1560 while (head) { 1561 struct net_device *dev = head; 1562 head = head->next_sched; 1563 1564 smp_mb__before_clear_bit(); 1565 clear_bit(__LINK_STATE_SCHED, &dev->state); 1566 1567 if (spin_trylock(&dev->queue_lock)) { 1568 qdisc_run(dev); 1569 spin_unlock(&dev->queue_lock); 1570 } else { 1571 netif_schedule(dev); 1572 } 1573 } 1574 } 1575} 1576 1577static __inline__ int deliver_skb(struct sk_buff *skb, 1578 struct packet_type *pt_prev) 1579{ 1580 atomic_inc(&skb->users); 1581 return pt_prev->func(skb, skb->dev, pt_prev); 1582} 1583 1584#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) 1585int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb); 1586struct net_bridge; 1587struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, 1588 unsigned char *addr); 1589void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent); 1590 1591static __inline__ int handle_bridge(struct sk_buff **pskb, 1592 struct packet_type **pt_prev, int *ret) 1593{ 1594 struct net_bridge_port *port; 1595 1596 if ((*pskb)->pkt_type == PACKET_LOOPBACK || 1597 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL) 1598 return 0; 1599 1600 if (*pt_prev) { 1601 *ret = deliver_skb(*pskb, *pt_prev); 1602 *pt_prev = NULL; 1603 } 1604 1605 return br_handle_frame_hook(port, pskb); 1606} 1607#else 1608#define handle_bridge(skb, pt_prev, ret) (0) 1609#endif 1610 1611#ifdef CONFIG_NET_CLS_ACT 1612/* TODO: Maybe we should just force sch_ingress to be compiled in 1613 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions 1614 * a compare and 2 stores extra right now if we dont have it on 1615 * but have CONFIG_NET_CLS_ACT 1616 * NOTE: This doesnt stop any functionality; if you dont have 1617 * the ingress scheduler, you just cant add policies on ingress. 1618 * 1619 */ 1620static int ing_filter(struct sk_buff *skb) 1621{ 1622 struct Qdisc *q; 1623 struct net_device *dev = skb->dev; 1624 int result = TC_ACT_OK; 1625 1626 if (dev->qdisc_ingress) { 1627 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd); 1628 if (MAX_RED_LOOP < ttl++) { 1629 printk("Redir loop detected Dropping packet (%s->%s)\n", 1630 skb->input_dev?skb->input_dev->name:"??",skb->dev->name); 1631 return TC_ACT_SHOT; 1632 } 1633 1634 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl); 1635 1636 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS); 1637 if (NULL == skb->input_dev) { 1638 skb->input_dev = skb->dev; 1639 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name); 1640 } 1641 spin_lock(&dev->ingress_lock); 1642 if ((q = dev->qdisc_ingress) != NULL) 1643 result = q->enqueue(skb, q); 1644 spin_unlock(&dev->ingress_lock); 1645 1646 } 1647 1648 return result; 1649} 1650#endif 1651 1652int netif_receive_skb(struct sk_buff *skb) 1653{ 1654 struct packet_type *ptype, *pt_prev; 1655 int ret = NET_RX_DROP; 1656 unsigned short type; 1657 1658 /* if we've gotten here through NAPI, check netpoll */ 1659 if (skb->dev->poll && netpoll_rx(skb)) 1660 return NET_RX_DROP; 1661 1662 if (!skb->stamp.tv_sec) 1663 net_timestamp(&skb->stamp); 1664 1665 skb_bond(skb); 1666 1667 __get_cpu_var(netdev_rx_stat).total++; 1668 1669 skb->h.raw = skb->nh.raw = skb->data; 1670 skb->mac_len = skb->nh.raw - skb->mac.raw; 1671 1672 pt_prev = NULL; 1673 1674 rcu_read_lock(); 1675 1676#ifdef CONFIG_NET_CLS_ACT 1677 if (skb->tc_verd & TC_NCLS) { 1678 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); 1679 goto ncls; 1680 } 1681#endif 1682 1683 list_for_each_entry_rcu(ptype, &ptype_all, list) { 1684 if (!ptype->dev || ptype->dev == skb->dev) { 1685 if (pt_prev) 1686 ret = deliver_skb(skb, pt_prev); 1687 pt_prev = ptype; 1688 } 1689 } 1690 1691#ifdef CONFIG_NET_CLS_ACT 1692 if (pt_prev) { 1693 ret = deliver_skb(skb, pt_prev); 1694 pt_prev = NULL; /* noone else should process this after*/ 1695 } else { 1696 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); 1697 } 1698 1699 ret = ing_filter(skb); 1700 1701 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) { 1702 kfree_skb(skb); 1703 goto out; 1704 } 1705 1706 skb->tc_verd = 0; 1707ncls: 1708#endif 1709 1710 handle_diverter(skb); 1711 1712 if (handle_bridge(&skb, &pt_prev, &ret)) 1713 goto out; 1714 1715 type = skb->protocol; 1716 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) { 1717 if (ptype->type == type && 1718 (!ptype->dev || ptype->dev == skb->dev)) { 1719 if (pt_prev) 1720 ret = deliver_skb(skb, pt_prev); 1721 pt_prev = ptype; 1722 } 1723 } 1724 1725 if (pt_prev) { 1726 ret = pt_prev->func(skb, skb->dev, pt_prev); 1727 } else { 1728 kfree_skb(skb); 1729 /* Jamal, now you will not able to escape explaining 1730 * me how you were going to use this. :-) 1731 */ 1732 ret = NET_RX_DROP; 1733 } 1734 1735out: 1736 rcu_read_unlock(); 1737 return ret; 1738} 1739 1740static int process_backlog(struct net_device *backlog_dev, int *budget) 1741{ 1742 int work = 0; 1743 int quota = min(backlog_dev->quota, *budget); 1744 struct softnet_data *queue = &__get_cpu_var(softnet_data); 1745 unsigned long start_time = jiffies; 1746 1747 for (;;) { 1748 struct sk_buff *skb; 1749 struct net_device *dev; 1750 1751 local_irq_disable(); 1752 skb = __skb_dequeue(&queue->input_pkt_queue); 1753 if (!skb) 1754 goto job_done; 1755 local_irq_enable(); 1756 1757 dev = skb->dev; 1758 1759 netif_receive_skb(skb); 1760 1761 dev_put(dev); 1762 1763 work++; 1764 1765 if (work >= quota || jiffies - start_time > 1) 1766 break; 1767 1768 } 1769 1770 backlog_dev->quota -= work; 1771 *budget -= work; 1772 return -1; 1773 1774job_done: 1775 backlog_dev->quota -= work; 1776 *budget -= work; 1777 1778 list_del(&backlog_dev->poll_list); 1779 smp_mb__before_clear_bit(); 1780 netif_poll_enable(backlog_dev); 1781 1782 if (queue->throttle) 1783 queue->throttle = 0; 1784 local_irq_enable(); 1785 return 0; 1786} 1787 1788static void net_rx_action(struct softirq_action *h) 1789{ 1790 struct softnet_data *queue = &__get_cpu_var(softnet_data); 1791 unsigned long start_time = jiffies; 1792 int budget = netdev_max_backlog; 1793 1794 1795 local_irq_disable(); 1796 1797 while (!list_empty(&queue->poll_list)) { 1798 struct net_device *dev; 1799 1800 if (budget <= 0 || jiffies - start_time > 1) 1801 goto softnet_break; 1802 1803 local_irq_enable(); 1804 1805 dev = list_entry(queue->poll_list.next, 1806 struct net_device, poll_list); 1807 netpoll_poll_lock(dev); 1808 1809 if (dev->quota <= 0 || dev->poll(dev, &budget)) { 1810 netpoll_poll_unlock(dev); 1811 local_irq_disable(); 1812 list_del(&dev->poll_list); 1813 list_add_tail(&dev->poll_list, &queue->poll_list); 1814 if (dev->quota < 0) 1815 dev->quota += dev->weight; 1816 else 1817 dev->quota = dev->weight; 1818 } else { 1819 netpoll_poll_unlock(dev); 1820 dev_put(dev); 1821 local_irq_disable(); 1822 } 1823 } 1824out: 1825 local_irq_enable(); 1826 return; 1827 1828softnet_break: 1829 __get_cpu_var(netdev_rx_stat).time_squeeze++; 1830 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 1831 goto out; 1832} 1833 1834static gifconf_func_t * gifconf_list [NPROTO]; 1835 1836/** 1837 * register_gifconf - register a SIOCGIF handler 1838 * @family: Address family 1839 * @gifconf: Function handler 1840 * 1841 * Register protocol dependent address dumping routines. The handler 1842 * that is passed must not be freed or reused until it has been replaced 1843 * by another handler. 1844 */ 1845int register_gifconf(unsigned int family, gifconf_func_t * gifconf) 1846{ 1847 if (family >= NPROTO) 1848 return -EINVAL; 1849 gifconf_list[family] = gifconf; 1850 return 0; 1851} 1852 1853 1854/* 1855 * Map an interface index to its name (SIOCGIFNAME) 1856 */ 1857 1858/* 1859 * We need this ioctl for efficient implementation of the 1860 * if_indextoname() function required by the IPv6 API. Without 1861 * it, we would have to search all the interfaces to find a 1862 * match. --pb 1863 */ 1864 1865static int dev_ifname(struct ifreq __user *arg) 1866{ 1867 struct net_device *dev; 1868 struct ifreq ifr; 1869 1870 /* 1871 * Fetch the caller's info block. 1872 */ 1873 1874 if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) 1875 return -EFAULT; 1876 1877 read_lock(&dev_base_lock); 1878 dev = __dev_get_by_index(ifr.ifr_ifindex); 1879 if (!dev) { 1880 read_unlock(&dev_base_lock); 1881 return -ENODEV; 1882 } 1883 1884 strcpy(ifr.ifr_name, dev->name); 1885 read_unlock(&dev_base_lock); 1886 1887 if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) 1888 return -EFAULT; 1889 return 0; 1890} 1891 1892/* 1893 * Perform a SIOCGIFCONF call. This structure will change 1894 * size eventually, and there is nothing I can do about it. 1895 * Thus we will need a 'compatibility mode'. 1896 */ 1897 1898static int dev_ifconf(char __user *arg) 1899{ 1900 struct ifconf ifc; 1901 struct net_device *dev; 1902 char __user *pos; 1903 int len; 1904 int total; 1905 int i; 1906 1907 /* 1908 * Fetch the caller's info block. 1909 */ 1910 1911 if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) 1912 return -EFAULT; 1913 1914 pos = ifc.ifc_buf; 1915 len = ifc.ifc_len; 1916 1917 /* 1918 * Loop over the interfaces, and write an info block for each. 1919 */ 1920 1921 total = 0; 1922 for (dev = dev_base; dev; dev = dev->next) { 1923 for (i = 0; i < NPROTO; i++) { 1924 if (gifconf_list[i]) { 1925 int done; 1926 if (!pos) 1927 done = gifconf_list[i](dev, NULL, 0); 1928 else 1929 done = gifconf_list[i](dev, pos + total, 1930 len - total); 1931 if (done < 0) 1932 return -EFAULT; 1933 total += done; 1934 } 1935 } 1936 } 1937 1938 /* 1939 * All done. Write the updated control block back to the caller. 1940 */ 1941 ifc.ifc_len = total; 1942 1943 /* 1944 * Both BSD and Solaris return 0 here, so we do too. 1945 */ 1946 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; 1947} 1948 1949#ifdef CONFIG_PROC_FS 1950/* 1951 * This is invoked by the /proc filesystem handler to display a device 1952 * in detail. 1953 */ 1954static __inline__ struct net_device *dev_get_idx(loff_t pos) 1955{ 1956 struct net_device *dev; 1957 loff_t i; 1958 1959 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next); 1960 1961 return i == pos ? dev : NULL; 1962} 1963 1964void *dev_seq_start(struct seq_file *seq, loff_t *pos) 1965{ 1966 read_lock(&dev_base_lock); 1967 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN; 1968} 1969 1970void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1971{ 1972 ++*pos; 1973 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next; 1974} 1975 1976void dev_seq_stop(struct seq_file *seq, void *v) 1977{ 1978 read_unlock(&dev_base_lock); 1979} 1980 1981static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) 1982{ 1983 if (dev->get_stats) { 1984 struct net_device_stats *stats = dev->get_stats(dev); 1985 1986 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " 1987 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", 1988 dev->name, stats->rx_bytes, stats->rx_packets, 1989 stats->rx_errors, 1990 stats->rx_dropped + stats->rx_missed_errors, 1991 stats->rx_fifo_errors, 1992 stats->rx_length_errors + stats->rx_over_errors + 1993 stats->rx_crc_errors + stats->rx_frame_errors, 1994 stats->rx_compressed, stats->multicast, 1995 stats->tx_bytes, stats->tx_packets, 1996 stats->tx_errors, stats->tx_dropped, 1997 stats->tx_fifo_errors, stats->collisions, 1998 stats->tx_carrier_errors + 1999 stats->tx_aborted_errors + 2000 stats->tx_window_errors + 2001 stats->tx_heartbeat_errors, 2002 stats->tx_compressed); 2003 } else 2004 seq_printf(seq, "%6s: No statistics available.\n", dev->name); 2005} 2006 2007/* 2008 * Called from the PROCfs module. This now uses the new arbitrary sized 2009 * /proc/net interface to create /proc/net/dev 2010 */ 2011static int dev_seq_show(struct seq_file *seq, void *v) 2012{ 2013 if (v == SEQ_START_TOKEN) 2014 seq_puts(seq, "Inter-| Receive " 2015 " | Transmit\n" 2016 " face |bytes packets errs drop fifo frame " 2017 "compressed multicast|bytes packets errs " 2018 "drop fifo colls carrier compressed\n"); 2019 else 2020 dev_seq_printf_stats(seq, v); 2021 return 0; 2022} 2023 2024static struct netif_rx_stats *softnet_get_online(loff_t *pos) 2025{ 2026 struct netif_rx_stats *rc = NULL; 2027 2028 while (*pos < NR_CPUS) 2029 if (cpu_online(*pos)) { 2030 rc = &per_cpu(netdev_rx_stat, *pos); 2031 break; 2032 } else 2033 ++*pos; 2034 return rc; 2035} 2036 2037static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) 2038{ 2039 return softnet_get_online(pos); 2040} 2041 2042static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2043{ 2044 ++*pos; 2045 return softnet_get_online(pos); 2046} 2047 2048static void softnet_seq_stop(struct seq_file *seq, void *v) 2049{ 2050} 2051 2052static int softnet_seq_show(struct seq_file *seq, void *v) 2053{ 2054 struct netif_rx_stats *s = v; 2055 2056 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", 2057 s->total, s->dropped, s->time_squeeze, s->throttled, 2058 s->fastroute_hit, s->fastroute_success, s->fastroute_defer, 2059 s->fastroute_deferred_out, 2060#if 0 2061 s->fastroute_latency_reduction 2062#else 2063 s->cpu_collision 2064#endif 2065 ); 2066 return 0; 2067} 2068 2069static struct seq_operations dev_seq_ops = { 2070 .start = dev_seq_start, 2071 .next = dev_seq_next, 2072 .stop = dev_seq_stop, 2073 .show = dev_seq_show, 2074}; 2075 2076static int dev_seq_open(struct inode *inode, struct file *file) 2077{ 2078 return seq_open(file, &dev_seq_ops); 2079} 2080 2081static struct file_operations dev_seq_fops = { 2082 .owner = THIS_MODULE, 2083 .open = dev_seq_open, 2084 .read = seq_read, 2085 .llseek = seq_lseek, 2086 .release = seq_release, 2087}; 2088 2089static struct seq_operations softnet_seq_ops = { 2090 .start = softnet_seq_start, 2091 .next = softnet_seq_next, 2092 .stop = softnet_seq_stop, 2093 .show = softnet_seq_show, 2094}; 2095 2096static int softnet_seq_open(struct inode *inode, struct file *file) 2097{ 2098 return seq_open(file, &softnet_seq_ops); 2099} 2100 2101static struct file_operations softnet_seq_fops = { 2102 .owner = THIS_MODULE, 2103 .open = softnet_seq_open, 2104 .read = seq_read, 2105 .llseek = seq_lseek, 2106 .release = seq_release, 2107}; 2108 2109#ifdef WIRELESS_EXT 2110extern int wireless_proc_init(void); 2111#else 2112#define wireless_proc_init() 0 2113#endif 2114 2115static int __init dev_proc_init(void) 2116{ 2117 int rc = -ENOMEM; 2118 2119 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops)) 2120 goto out; 2121 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops)) 2122 goto out_dev; 2123 if (wireless_proc_init()) 2124 goto out_softnet; 2125 rc = 0; 2126out: 2127 return rc; 2128out_softnet: 2129 proc_net_remove("softnet_stat"); 2130out_dev: 2131 proc_net_remove("dev"); 2132 goto out; 2133} 2134#else 2135#define dev_proc_init() 0 2136#endif /* CONFIG_PROC_FS */ 2137 2138 2139/** 2140 * netdev_set_master - set up master/slave pair 2141 * @slave: slave device 2142 * @master: new master device 2143 * 2144 * Changes the master device of the slave. Pass %NULL to break the 2145 * bonding. The caller must hold the RTNL semaphore. On a failure 2146 * a negative errno code is returned. On success the reference counts 2147 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the 2148 * function returns zero. 2149 */ 2150int netdev_set_master(struct net_device *slave, struct net_device *master) 2151{ 2152 struct net_device *old = slave->master; 2153 2154 ASSERT_RTNL(); 2155 2156 if (master) { 2157 if (old) 2158 return -EBUSY; 2159 dev_hold(master); 2160 } 2161 2162 slave->master = master; 2163 2164 synchronize_net(); 2165 2166 if (old) 2167 dev_put(old); 2168 2169 if (master) 2170 slave->flags |= IFF_SLAVE; 2171 else 2172 slave->flags &= ~IFF_SLAVE; 2173 2174 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); 2175 return 0; 2176} 2177 2178/** 2179 * dev_set_promiscuity - update promiscuity count on a device 2180 * @dev: device 2181 * @inc: modifier 2182 * 2183 * Add or remove promsicuity from a device. While the count in the device 2184 * remains above zero the interface remains promiscuous. Once it hits zero 2185 * the device reverts back to normal filtering operation. A negative inc 2186 * value is used to drop promiscuity on the device. 2187 */ 2188void dev_set_promiscuity(struct net_device *dev, int inc) 2189{ 2190 unsigned short old_flags = dev->flags; 2191 2192 dev->flags |= IFF_PROMISC; 2193 if ((dev->promiscuity += inc) == 0) 2194 dev->flags &= ~IFF_PROMISC; 2195 if (dev->flags ^ old_flags) { 2196 dev_mc_upload(dev); 2197 printk(KERN_INFO "device %s %s promiscuous mode\n", 2198 dev->name, (dev->flags & IFF_PROMISC) ? "entered" : 2199 "left"); 2200 } 2201} 2202 2203/** 2204 * dev_set_allmulti - update allmulti count on a device 2205 * @dev: device 2206 * @inc: modifier 2207 * 2208 * Add or remove reception of all multicast frames to a device. While the 2209 * count in the device remains above zero the interface remains listening 2210 * to all interfaces. Once it hits zero the device reverts back to normal 2211 * filtering operation. A negative @inc value is used to drop the counter 2212 * when releasing a resource needing all multicasts. 2213 */ 2214 2215void dev_set_allmulti(struct net_device *dev, int inc) 2216{ 2217 unsigned short old_flags = dev->flags; 2218 2219 dev->flags |= IFF_ALLMULTI; 2220 if ((dev->allmulti += inc) == 0) 2221 dev->flags &= ~IFF_ALLMULTI; 2222 if (dev->flags ^ old_flags) 2223 dev_mc_upload(dev); 2224} 2225 2226unsigned dev_get_flags(const struct net_device *dev) 2227{ 2228 unsigned flags; 2229 2230 flags = (dev->flags & ~(IFF_PROMISC | 2231 IFF_ALLMULTI | 2232 IFF_RUNNING)) | 2233 (dev->gflags & (IFF_PROMISC | 2234 IFF_ALLMULTI)); 2235 2236 if (netif_running(dev) && netif_carrier_ok(dev)) 2237 flags |= IFF_RUNNING; 2238 2239 return flags; 2240} 2241 2242int dev_change_flags(struct net_device *dev, unsigned flags) 2243{ 2244 int ret; 2245 int old_flags = dev->flags; 2246 2247 /* 2248 * Set the flags on our device. 2249 */ 2250 2251 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 2252 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 2253 IFF_AUTOMEDIA)) | 2254 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 2255 IFF_ALLMULTI)); 2256 2257 /* 2258 * Load in the correct multicast list now the flags have changed. 2259 */ 2260 2261 dev_mc_upload(dev); 2262 2263 /* 2264 * Have we downed the interface. We handle IFF_UP ourselves 2265 * according to user attempts to set it, rather than blindly 2266 * setting it. 2267 */ 2268 2269 ret = 0; 2270 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ 2271 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); 2272 2273 if (!ret) 2274 dev_mc_upload(dev); 2275 } 2276 2277 if (dev->flags & IFF_UP && 2278 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | 2279 IFF_VOLATILE))) 2280 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev); 2281 2282 if ((flags ^ dev->gflags) & IFF_PROMISC) { 2283 int inc = (flags & IFF_PROMISC) ? +1 : -1; 2284 dev->gflags ^= IFF_PROMISC; 2285 dev_set_promiscuity(dev, inc); 2286 } 2287 2288 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 2289 is important. Some (broken) drivers set IFF_PROMISC, when 2290 IFF_ALLMULTI is requested not asking us and not reporting. 2291 */ 2292 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 2293 int inc = (flags & IFF_ALLMULTI) ? +1 : -1; 2294 dev->gflags ^= IFF_ALLMULTI; 2295 dev_set_allmulti(dev, inc); 2296 } 2297 2298 if (old_flags ^ dev->flags) 2299 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags); 2300 2301 return ret; 2302} 2303 2304int dev_set_mtu(struct net_device *dev, int new_mtu) 2305{ 2306 int err; 2307 2308 if (new_mtu == dev->mtu) 2309 return 0; 2310 2311 /* MTU must be positive. */ 2312 if (new_mtu < 0) 2313 return -EINVAL; 2314 2315 if (!netif_device_present(dev)) 2316 return -ENODEV; 2317 2318 err = 0; 2319 if (dev->change_mtu) 2320 err = dev->change_mtu(dev, new_mtu); 2321 else 2322 dev->mtu = new_mtu; 2323 if (!err && dev->flags & IFF_UP) 2324 notifier_call_chain(&netdev_chain, 2325 NETDEV_CHANGEMTU, dev); 2326 return err; 2327} 2328 2329int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 2330{ 2331 int err; 2332 2333 if (!dev->set_mac_address) 2334 return -EOPNOTSUPP; 2335 if (sa->sa_family != dev->type) 2336 return -EINVAL; 2337 if (!netif_device_present(dev)) 2338 return -ENODEV; 2339 err = dev->set_mac_address(dev, sa); 2340 if (!err) 2341 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev); 2342 return err; 2343} 2344 2345/* 2346 * Perform the SIOCxIFxxx calls. 2347 */ 2348static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd) 2349{ 2350 int err; 2351 struct net_device *dev = __dev_get_by_name(ifr->ifr_name); 2352 2353 if (!dev) 2354 return -ENODEV; 2355 2356 switch (cmd) { 2357 case SIOCGIFFLAGS: /* Get interface flags */ 2358 ifr->ifr_flags = dev_get_flags(dev); 2359 return 0; 2360 2361 case SIOCSIFFLAGS: /* Set interface flags */ 2362 return dev_change_flags(dev, ifr->ifr_flags); 2363 2364 case SIOCGIFMETRIC: /* Get the metric on the interface 2365 (currently unused) */ 2366 ifr->ifr_metric = 0; 2367 return 0; 2368 2369 case SIOCSIFMETRIC: /* Set the metric on the interface 2370 (currently unused) */ 2371 return -EOPNOTSUPP; 2372 2373 case SIOCGIFMTU: /* Get the MTU of a device */ 2374 ifr->ifr_mtu = dev->mtu; 2375 return 0; 2376 2377 case SIOCSIFMTU: /* Set the MTU of a device */ 2378 return dev_set_mtu(dev, ifr->ifr_mtu); 2379 2380 case SIOCGIFHWADDR: 2381 if (!dev->addr_len) 2382 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); 2383 else 2384 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, 2385 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); 2386 ifr->ifr_hwaddr.sa_family = dev->type; 2387 return 0; 2388 2389 case SIOCSIFHWADDR: 2390 return dev_set_mac_address(dev, &ifr->ifr_hwaddr); 2391 2392 case SIOCSIFHWBROADCAST: 2393 if (ifr->ifr_hwaddr.sa_family != dev->type) 2394 return -EINVAL; 2395 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, 2396 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); 2397 notifier_call_chain(&netdev_chain, 2398 NETDEV_CHANGEADDR, dev); 2399 return 0; 2400 2401 case SIOCGIFMAP: 2402 ifr->ifr_map.mem_start = dev->mem_start; 2403 ifr->ifr_map.mem_end = dev->mem_end; 2404 ifr->ifr_map.base_addr = dev->base_addr; 2405 ifr->ifr_map.irq = dev->irq; 2406 ifr->ifr_map.dma = dev->dma; 2407 ifr->ifr_map.port = dev->if_port; 2408 return 0; 2409 2410 case SIOCSIFMAP: 2411 if (dev->set_config) { 2412 if (!netif_device_present(dev)) 2413 return -ENODEV; 2414 return dev->set_config(dev, &ifr->ifr_map); 2415 } 2416 return -EOPNOTSUPP; 2417 2418 case SIOCADDMULTI: 2419 if (!dev->set_multicast_list || 2420 ifr->ifr_hwaddr.sa_family != AF_UNSPEC) 2421 return -EINVAL; 2422 if (!netif_device_present(dev)) 2423 return -ENODEV; 2424 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, 2425 dev->addr_len, 1); 2426 2427 case SIOCDELMULTI: 2428 if (!dev->set_multicast_list || 2429 ifr->ifr_hwaddr.sa_family != AF_UNSPEC) 2430 return -EINVAL; 2431 if (!netif_device_present(dev)) 2432 return -ENODEV; 2433 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, 2434 dev->addr_len, 1); 2435 2436 case SIOCGIFINDEX: 2437 ifr->ifr_ifindex = dev->ifindex; 2438 return 0; 2439 2440 case SIOCGIFTXQLEN: 2441 ifr->ifr_qlen = dev->tx_queue_len; 2442 return 0; 2443 2444 case SIOCSIFTXQLEN: 2445 if (ifr->ifr_qlen < 0) 2446 return -EINVAL; 2447 dev->tx_queue_len = ifr->ifr_qlen; 2448 return 0; 2449 2450 case SIOCSIFNAME: 2451 ifr->ifr_newname[IFNAMSIZ-1] = '\0'; 2452 return dev_change_name(dev, ifr->ifr_newname); 2453 2454 /* 2455 * Unknown or private ioctl 2456 */ 2457 2458 default: 2459 if ((cmd >= SIOCDEVPRIVATE && 2460 cmd <= SIOCDEVPRIVATE + 15) || 2461 cmd == SIOCBONDENSLAVE || 2462 cmd == SIOCBONDRELEASE || 2463 cmd == SIOCBONDSETHWADDR || 2464 cmd == SIOCBONDSLAVEINFOQUERY || 2465 cmd == SIOCBONDINFOQUERY || 2466 cmd == SIOCBONDCHANGEACTIVE || 2467 cmd == SIOCGMIIPHY || 2468 cmd == SIOCGMIIREG || 2469 cmd == SIOCSMIIREG || 2470 cmd == SIOCBRADDIF || 2471 cmd == SIOCBRDELIF || 2472 cmd == SIOCWANDEV) { 2473 err = -EOPNOTSUPP; 2474 if (dev->do_ioctl) { 2475 if (netif_device_present(dev)) 2476 err = dev->do_ioctl(dev, ifr, 2477 cmd); 2478 else 2479 err = -ENODEV; 2480 } 2481 } else 2482 err = -EINVAL; 2483 2484 } 2485 return err; 2486} 2487 2488/* 2489 * This function handles all "interface"-type I/O control requests. The actual 2490 * 'doing' part of this is dev_ifsioc above. 2491 */ 2492 2493/** 2494 * dev_ioctl - network device ioctl 2495 * @cmd: command to issue 2496 * @arg: pointer to a struct ifreq in user space 2497 * 2498 * Issue ioctl functions to devices. This is normally called by the 2499 * user space syscall interfaces but can sometimes be useful for 2500 * other purposes. The return value is the return from the syscall if 2501 * positive or a negative errno code on error. 2502 */ 2503 2504int dev_ioctl(unsigned int cmd, void __user *arg) 2505{ 2506 struct ifreq ifr; 2507 int ret; 2508 char *colon; 2509 2510 /* One special case: SIOCGIFCONF takes ifconf argument 2511 and requires shared lock, because it sleeps writing 2512 to user space. 2513 */ 2514 2515 if (cmd == SIOCGIFCONF) { 2516 rtnl_shlock(); 2517 ret = dev_ifconf((char __user *) arg); 2518 rtnl_shunlock(); 2519 return ret; 2520 } 2521 if (cmd == SIOCGIFNAME) 2522 return dev_ifname((struct ifreq __user *)arg); 2523 2524 if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) 2525 return -EFAULT; 2526 2527 ifr.ifr_name[IFNAMSIZ-1] = 0; 2528 2529 colon = strchr(ifr.ifr_name, ':'); 2530 if (colon) 2531 *colon = 0; 2532 2533 /* 2534 * See which interface the caller is talking about. 2535 */ 2536 2537 switch (cmd) { 2538 /* 2539 * These ioctl calls: 2540 * - can be done by all. 2541 * - atomic and do not require locking. 2542 * - return a value 2543 */ 2544 case SIOCGIFFLAGS: 2545 case SIOCGIFMETRIC: 2546 case SIOCGIFMTU: 2547 case SIOCGIFHWADDR: 2548 case SIOCGIFSLAVE: 2549 case SIOCGIFMAP: 2550 case SIOCGIFINDEX: 2551 case SIOCGIFTXQLEN: 2552 dev_load(ifr.ifr_name); 2553 read_lock(&dev_base_lock); 2554 ret = dev_ifsioc(&ifr, cmd); 2555 read_unlock(&dev_base_lock); 2556 if (!ret) { 2557 if (colon) 2558 *colon = ':'; 2559 if (copy_to_user(arg, &ifr, 2560 sizeof(struct ifreq))) 2561 ret = -EFAULT; 2562 } 2563 return ret; 2564 2565 case SIOCETHTOOL: 2566 dev_load(ifr.ifr_name); 2567 rtnl_lock(); 2568 ret = dev_ethtool(&ifr); 2569 rtnl_unlock(); 2570 if (!ret) { 2571 if (colon) 2572 *colon = ':'; 2573 if (copy_to_user(arg, &ifr, 2574 sizeof(struct ifreq))) 2575 ret = -EFAULT; 2576 } 2577 return ret; 2578 2579 /* 2580 * These ioctl calls: 2581 * - require superuser power. 2582 * - require strict serialization. 2583 * - return a value 2584 */ 2585 case SIOCGMIIPHY: 2586 case SIOCGMIIREG: 2587 case SIOCSIFNAME: 2588 if (!capable(CAP_NET_ADMIN)) 2589 return -EPERM; 2590 dev_load(ifr.ifr_name); 2591 rtnl_lock(); 2592 ret = dev_ifsioc(&ifr, cmd); 2593 rtnl_unlock(); 2594 if (!ret) { 2595 if (colon) 2596 *colon = ':'; 2597 if (copy_to_user(arg, &ifr, 2598 sizeof(struct ifreq))) 2599 ret = -EFAULT; 2600 } 2601 return ret; 2602 2603 /* 2604 * These ioctl calls: 2605 * - require superuser power. 2606 * - require strict serialization. 2607 * - do not return a value 2608 */ 2609 case SIOCSIFFLAGS: 2610 case SIOCSIFMETRIC: 2611 case SIOCSIFMTU: 2612 case SIOCSIFMAP: 2613 case SIOCSIFHWADDR: 2614 case SIOCSIFSLAVE: 2615 case SIOCADDMULTI: 2616 case SIOCDELMULTI: 2617 case SIOCSIFHWBROADCAST: 2618 case SIOCSIFTXQLEN: 2619 case SIOCSMIIREG: 2620 case SIOCBONDENSLAVE: 2621 case SIOCBONDRELEASE: 2622 case SIOCBONDSETHWADDR: 2623 case SIOCBONDSLAVEINFOQUERY: 2624 case SIOCBONDINFOQUERY: 2625 case SIOCBONDCHANGEACTIVE: 2626 case SIOCBRADDIF: 2627 case SIOCBRDELIF: 2628 if (!capable(CAP_NET_ADMIN)) 2629 return -EPERM; 2630 dev_load(ifr.ifr_name); 2631 rtnl_lock(); 2632 ret = dev_ifsioc(&ifr, cmd); 2633 rtnl_unlock(); 2634 return ret; 2635 2636 case SIOCGIFMEM: 2637 /* Get the per device memory space. We can add this but 2638 * currently do not support it */ 2639 case SIOCSIFMEM: 2640 /* Set the per device memory buffer space. 2641 * Not applicable in our case */ 2642 case SIOCSIFLINK: 2643 return -EINVAL; 2644 2645 /* 2646 * Unknown or private ioctl. 2647 */ 2648 default: 2649 if (cmd == SIOCWANDEV || 2650 (cmd >= SIOCDEVPRIVATE && 2651 cmd <= SIOCDEVPRIVATE + 15)) { 2652 dev_load(ifr.ifr_name); 2653 rtnl_lock(); 2654 ret = dev_ifsioc(&ifr, cmd); 2655 rtnl_unlock(); 2656 if (!ret && copy_to_user(arg, &ifr, 2657 sizeof(struct ifreq))) 2658 ret = -EFAULT; 2659 return ret; 2660 } 2661#ifdef WIRELESS_EXT 2662 /* Take care of Wireless Extensions */ 2663 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) { 2664 /* If command is `set a parameter', or 2665 * `get the encoding parameters', check if 2666 * the user has the right to do it */ 2667 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) { 2668 if (!capable(CAP_NET_ADMIN)) 2669 return -EPERM; 2670 } 2671 dev_load(ifr.ifr_name); 2672 rtnl_lock(); 2673 /* Follow me in net/core/wireless.c */ 2674 ret = wireless_process_ioctl(&ifr, cmd); 2675 rtnl_unlock(); 2676 if (IW_IS_GET(cmd) && 2677 copy_to_user(arg, &ifr, 2678 sizeof(struct ifreq))) 2679 ret = -EFAULT; 2680 return ret; 2681 } 2682#endif /* WIRELESS_EXT */ 2683 return -EINVAL; 2684 } 2685} 2686 2687 2688/** 2689 * dev_new_index - allocate an ifindex 2690 * 2691 * Returns a suitable unique value for a new device interface 2692 * number. The caller must hold the rtnl semaphore or the 2693 * dev_base_lock to be sure it remains unique. 2694 */ 2695static int dev_new_index(void) 2696{ 2697 static int ifindex; 2698 for (;;) { 2699 if (++ifindex <= 0) 2700 ifindex = 1; 2701 if (!__dev_get_by_index(ifindex)) 2702 return ifindex; 2703 } 2704} 2705 2706static int dev_boot_phase = 1; 2707 2708/* Delayed registration/unregisteration */ 2709static DEFINE_SPINLOCK(net_todo_list_lock); 2710static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list); 2711 2712static inline void net_set_todo(struct net_device *dev) 2713{ 2714 spin_lock(&net_todo_list_lock); 2715 list_add_tail(&dev->todo_list, &net_todo_list); 2716 spin_unlock(&net_todo_list_lock); 2717} 2718 2719/** 2720 * register_netdevice - register a network device 2721 * @dev: device to register 2722 * 2723 * Take a completed network device structure and add it to the kernel 2724 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 2725 * chain. 0 is returned on success. A negative errno code is returned 2726 * on a failure to set up the device, or if the name is a duplicate. 2727 * 2728 * Callers must hold the rtnl semaphore. You may want 2729 * register_netdev() instead of this. 2730 * 2731 * BUGS: 2732 * The locking appears insufficient to guarantee two parallel registers 2733 * will not get the same name. 2734 */ 2735 2736int register_netdevice(struct net_device *dev) 2737{ 2738 struct hlist_head *head; 2739 struct hlist_node *p; 2740 int ret; 2741 2742 BUG_ON(dev_boot_phase); 2743 ASSERT_RTNL(); 2744 2745 /* When net_device's are persistent, this will be fatal. */ 2746 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 2747 2748 spin_lock_init(&dev->queue_lock); 2749 spin_lock_init(&dev->xmit_lock); 2750 dev->xmit_lock_owner = -1; 2751#ifdef CONFIG_NET_CLS_ACT 2752 spin_lock_init(&dev->ingress_lock); 2753#endif 2754 2755 ret = alloc_divert_blk(dev); 2756 if (ret) 2757 goto out; 2758 2759 dev->iflink = -1; 2760 2761 /* Init, if this function is available */ 2762 if (dev->init) { 2763 ret = dev->init(dev); 2764 if (ret) { 2765 if (ret > 0) 2766 ret = -EIO; 2767 goto out_err; 2768 } 2769 } 2770 2771 if (!dev_valid_name(dev->name)) { 2772 ret = -EINVAL; 2773 goto out_err; 2774 } 2775 2776 dev->ifindex = dev_new_index(); 2777 if (dev->iflink == -1) 2778 dev->iflink = dev->ifindex; 2779 2780 /* Check for existence of name */ 2781 head = dev_name_hash(dev->name); 2782 hlist_for_each(p, head) { 2783 struct net_device *d 2784 = hlist_entry(p, struct net_device, name_hlist); 2785 if (!strncmp(d->name, dev->name, IFNAMSIZ)) { 2786 ret = -EEXIST; 2787 goto out_err; 2788 } 2789 } 2790 2791 /* Fix illegal SG+CSUM combinations. */ 2792 if ((dev->features & NETIF_F_SG) && 2793 !(dev->features & (NETIF_F_IP_CSUM | 2794 NETIF_F_NO_CSUM | 2795 NETIF_F_HW_CSUM))) { 2796 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n", 2797 dev->name); 2798 dev->features &= ~NETIF_F_SG; 2799 } 2800 2801 /* TSO requires that SG is present as well. */ 2802 if ((dev->features & NETIF_F_TSO) && 2803 !(dev->features & NETIF_F_SG)) { 2804 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n", 2805 dev->name); 2806 dev->features &= ~NETIF_F_TSO; 2807 } 2808 2809 /* 2810 * nil rebuild_header routine, 2811 * that should be never called and used as just bug trap. 2812 */ 2813 2814 if (!dev->rebuild_header) 2815 dev->rebuild_header = default_rebuild_header; 2816 2817 /* 2818 * Default initial state at registry is that the 2819 * device is present. 2820 */ 2821 2822 set_bit(__LINK_STATE_PRESENT, &dev->state); 2823 2824 dev->next = NULL; 2825 dev_init_scheduler(dev); 2826 write_lock_bh(&dev_base_lock); 2827 *dev_tail = dev; 2828 dev_tail = &dev->next; 2829 hlist_add_head(&dev->name_hlist, head); 2830 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex)); 2831 dev_hold(dev); 2832 dev->reg_state = NETREG_REGISTERING; 2833 write_unlock_bh(&dev_base_lock); 2834 2835 /* Notify protocols, that a new device appeared. */ 2836 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev); 2837 2838 /* Finish registration after unlock */ 2839 net_set_todo(dev); 2840 ret = 0; 2841 2842out: 2843 return ret; 2844out_err: 2845 free_divert_blk(dev); 2846 goto out; 2847} 2848 2849/** 2850 * register_netdev - register a network device 2851 * @dev: device to register 2852 * 2853 * Take a completed network device structure and add it to the kernel 2854 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 2855 * chain. 0 is returned on success. A negative errno code is returned 2856 * on a failure to set up the device, or if the name is a duplicate. 2857 * 2858 * This is a wrapper around register_netdev that takes the rtnl semaphore 2859 * and expands the device name if you passed a format string to 2860 * alloc_netdev. 2861 */ 2862int register_netdev(struct net_device *dev) 2863{ 2864 int err; 2865 2866 rtnl_lock(); 2867 2868 /* 2869 * If the name is a format string the caller wants us to do a 2870 * name allocation. 2871 */ 2872 if (strchr(dev->name, '%')) { 2873 err = dev_alloc_name(dev, dev->name); 2874 if (err < 0) 2875 goto out; 2876 } 2877 2878 /* 2879 * Back compatibility hook. Kill this one in 2.5 2880 */ 2881 if (dev->name[0] == 0 || dev->name[0] == ' ') { 2882 err = dev_alloc_name(dev, "eth%d"); 2883 if (err < 0) 2884 goto out; 2885 } 2886 2887 err = register_netdevice(dev); 2888out: 2889 rtnl_unlock(); 2890 return err; 2891} 2892EXPORT_SYMBOL(register_netdev); 2893 2894/* 2895 * netdev_wait_allrefs - wait until all references are gone. 2896 * 2897 * This is called when unregistering network devices. 2898 * 2899 * Any protocol or device that holds a reference should register 2900 * for netdevice notification, and cleanup and put back the 2901 * reference if they receive an UNREGISTER event. 2902 * We can get stuck here if buggy protocols don't correctly 2903 * call dev_put. 2904 */ 2905static void netdev_wait_allrefs(struct net_device *dev) 2906{ 2907 unsigned long rebroadcast_time, warning_time; 2908 2909 rebroadcast_time = warning_time = jiffies; 2910 while (atomic_read(&dev->refcnt) != 0) { 2911 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 2912 rtnl_shlock(); 2913 2914 /* Rebroadcast unregister notification */ 2915 notifier_call_chain(&netdev_chain, 2916 NETDEV_UNREGISTER, dev); 2917 2918 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 2919 &dev->state)) { 2920 /* We must not have linkwatch events 2921 * pending on unregister. If this 2922 * happens, we simply run the queue 2923 * unscheduled, resulting in a noop 2924 * for this device. 2925 */ 2926 linkwatch_run_queue(); 2927 } 2928 2929 rtnl_shunlock(); 2930 2931 rebroadcast_time = jiffies; 2932 } 2933 2934 msleep(250); 2935 2936 if (time_after(jiffies, warning_time + 10 * HZ)) { 2937 printk(KERN_EMERG "unregister_netdevice: " 2938 "waiting for %s to become free. Usage " 2939 "count = %d\n", 2940 dev->name, atomic_read(&dev->refcnt)); 2941 warning_time = jiffies; 2942 } 2943 } 2944} 2945 2946/* The sequence is: 2947 * 2948 * rtnl_lock(); 2949 * ... 2950 * register_netdevice(x1); 2951 * register_netdevice(x2); 2952 * ... 2953 * unregister_netdevice(y1); 2954 * unregister_netdevice(y2); 2955 * ... 2956 * rtnl_unlock(); 2957 * free_netdev(y1); 2958 * free_netdev(y2); 2959 * 2960 * We are invoked by rtnl_unlock() after it drops the semaphore. 2961 * This allows us to deal with problems: 2962 * 1) We can create/delete sysfs objects which invoke hotplug 2963 * without deadlocking with linkwatch via keventd. 2964 * 2) Since we run with the RTNL semaphore not held, we can sleep 2965 * safely in order to wait for the netdev refcnt to drop to zero. 2966 */ 2967static DECLARE_MUTEX(net_todo_run_mutex); 2968void netdev_run_todo(void) 2969{ 2970 struct list_head list = LIST_HEAD_INIT(list); 2971 int err; 2972 2973 2974 /* Need to guard against multiple cpu's getting out of order. */ 2975 down(&net_todo_run_mutex); 2976 2977 /* Not safe to do outside the semaphore. We must not return 2978 * until all unregister events invoked by the local processor 2979 * have been completed (either by this todo run, or one on 2980 * another cpu). 2981 */ 2982 if (list_empty(&net_todo_list)) 2983 goto out; 2984 2985 /* Snapshot list, allow later requests */ 2986 spin_lock(&net_todo_list_lock); 2987 list_splice_init(&net_todo_list, &list); 2988 spin_unlock(&net_todo_list_lock); 2989 2990 while (!list_empty(&list)) { 2991 struct net_device *dev 2992 = list_entry(list.next, struct net_device, todo_list); 2993 list_del(&dev->todo_list); 2994 2995 switch(dev->reg_state) { 2996 case NETREG_REGISTERING: 2997 err = netdev_register_sysfs(dev); 2998 if (err) 2999 printk(KERN_ERR "%s: failed sysfs registration (%d)\n", 3000 dev->name, err); 3001 dev->reg_state = NETREG_REGISTERED; 3002 break; 3003 3004 case NETREG_UNREGISTERING: 3005 netdev_unregister_sysfs(dev); 3006 dev->reg_state = NETREG_UNREGISTERED; 3007 3008 netdev_wait_allrefs(dev); 3009 3010 /* paranoia */ 3011 BUG_ON(atomic_read(&dev->refcnt)); 3012 BUG_TRAP(!dev->ip_ptr); 3013 BUG_TRAP(!dev->ip6_ptr); 3014 BUG_TRAP(!dev->dn_ptr); 3015 3016 3017 /* It must be the very last action, 3018 * after this 'dev' may point to freed up memory. 3019 */ 3020 if (dev->destructor) 3021 dev->destructor(dev); 3022 break; 3023 3024 default: 3025 printk(KERN_ERR "network todo '%s' but state %d\n", 3026 dev->name, dev->reg_state); 3027 break; 3028 } 3029 } 3030 3031out: 3032 up(&net_todo_run_mutex); 3033} 3034 3035/** 3036 * alloc_netdev - allocate network device 3037 * @sizeof_priv: size of private data to allocate space for 3038 * @name: device name format string 3039 * @setup: callback to initialize device 3040 * 3041 * Allocates a struct net_device with private data area for driver use 3042 * and performs basic initialization. 3043 */ 3044struct net_device *alloc_netdev(int sizeof_priv, const char *name, 3045 void (*setup)(struct net_device *)) 3046{ 3047 void *p; 3048 struct net_device *dev; 3049 int alloc_size; 3050 3051 /* ensure 32-byte alignment of both the device and private area */ 3052 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; 3053 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST; 3054 3055 p = kmalloc(alloc_size, GFP_KERNEL); 3056 if (!p) { 3057 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n"); 3058 return NULL; 3059 } 3060 memset(p, 0, alloc_size); 3061 3062 dev = (struct net_device *) 3063 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); 3064 dev->padded = (char *)dev - (char *)p; 3065 3066 if (sizeof_priv) 3067 dev->priv = netdev_priv(dev); 3068 3069 setup(dev); 3070 strcpy(dev->name, name); 3071 return dev; 3072} 3073EXPORT_SYMBOL(alloc_netdev); 3074 3075/** 3076 * free_netdev - free network device 3077 * @dev: device 3078 * 3079 * This function does the last stage of destroying an allocated device 3080 * interface. The reference to the device object is released. 3081 * If this is the last reference then it will be freed. 3082 */ 3083void free_netdev(struct net_device *dev) 3084{ 3085#ifdef CONFIG_SYSFS 3086 /* Compatiablity with error handling in drivers */ 3087 if (dev->reg_state == NETREG_UNINITIALIZED) { 3088 kfree((char *)dev - dev->padded); 3089 return; 3090 } 3091 3092 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 3093 dev->reg_state = NETREG_RELEASED; 3094 3095 /* will free via class release */ 3096 class_device_put(&dev->class_dev); 3097#else 3098 kfree((char *)dev - dev->padded); 3099#endif 3100} 3101 3102/* Synchronize with packet receive processing. */ 3103void synchronize_net(void) 3104{ 3105 might_sleep(); 3106 synchronize_rcu(); 3107} 3108 3109/** 3110 * unregister_netdevice - remove device from the kernel 3111 * @dev: device 3112 * 3113 * This function shuts down a device interface and removes it 3114 * from the kernel tables. On success 0 is returned, on a failure 3115 * a negative errno code is returned. 3116 * 3117 * Callers must hold the rtnl semaphore. You may want 3118 * unregister_netdev() instead of this. 3119 */ 3120 3121int unregister_netdevice(struct net_device *dev) 3122{ 3123 struct net_device *d, **dp; 3124 3125 BUG_ON(dev_boot_phase); 3126 ASSERT_RTNL(); 3127 3128 /* Some devices call without registering for initialization unwind. */ 3129 if (dev->reg_state == NETREG_UNINITIALIZED) { 3130 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " 3131 "was registered\n", dev->name, dev); 3132 return -ENODEV; 3133 } 3134 3135 BUG_ON(dev->reg_state != NETREG_REGISTERED); 3136 3137 /* If device is running, close it first. */ 3138 if (dev->flags & IFF_UP) 3139 dev_close(dev); 3140 3141 /* And unlink it from device chain. */ 3142 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) { 3143 if (d == dev) { 3144 write_lock_bh(&dev_base_lock); 3145 hlist_del(&dev->name_hlist); 3146 hlist_del(&dev->index_hlist); 3147 if (dev_tail == &dev->next) 3148 dev_tail = dp; 3149 *dp = d->next; 3150 write_unlock_bh(&dev_base_lock); 3151 break; 3152 } 3153 } 3154 if (!d) { 3155 printk(KERN_ERR "unregister net_device: '%s' not found\n", 3156 dev->name); 3157 return -ENODEV; 3158 } 3159 3160 dev->reg_state = NETREG_UNREGISTERING; 3161 3162 synchronize_net(); 3163 3164 /* Shutdown queueing discipline. */ 3165 dev_shutdown(dev); 3166 3167 3168 /* Notify protocols, that we are about to destroy 3169 this device. They should clean all the things. 3170 */ 3171 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev); 3172 3173 /* 3174 * Flush the multicast chain 3175 */ 3176 dev_mc_discard(dev); 3177 3178 if (dev->uninit) 3179 dev->uninit(dev); 3180 3181 /* Notifier chain MUST detach us from master device. */ 3182 BUG_TRAP(!dev->master); 3183 3184 free_divert_blk(dev); 3185 3186 /* Finish processing unregister after unlock */ 3187 net_set_todo(dev); 3188 3189 synchronize_net(); 3190 3191 dev_put(dev); 3192 return 0; 3193} 3194 3195/** 3196 * unregister_netdev - remove device from the kernel 3197 * @dev: device 3198 * 3199 * This function shuts down a device interface and removes it 3200 * from the kernel tables. On success 0 is returned, on a failure 3201 * a negative errno code is returned. 3202 * 3203 * This is just a wrapper for unregister_netdevice that takes 3204 * the rtnl semaphore. In general you want to use this and not 3205 * unregister_netdevice. 3206 */ 3207void unregister_netdev(struct net_device *dev) 3208{ 3209 rtnl_lock(); 3210 unregister_netdevice(dev); 3211 rtnl_unlock(); 3212} 3213 3214EXPORT_SYMBOL(unregister_netdev); 3215 3216#ifdef CONFIG_HOTPLUG_CPU 3217static int dev_cpu_callback(struct notifier_block *nfb, 3218 unsigned long action, 3219 void *ocpu) 3220{ 3221 struct sk_buff **list_skb; 3222 struct net_device **list_net; 3223 struct sk_buff *skb; 3224 unsigned int cpu, oldcpu = (unsigned long)ocpu; 3225 struct softnet_data *sd, *oldsd; 3226 3227 if (action != CPU_DEAD) 3228 return NOTIFY_OK; 3229 3230 local_irq_disable(); 3231 cpu = smp_processor_id(); 3232 sd = &per_cpu(softnet_data, cpu); 3233 oldsd = &per_cpu(softnet_data, oldcpu); 3234 3235 /* Find end of our completion_queue. */ 3236 list_skb = &sd->completion_queue; 3237 while (*list_skb) 3238 list_skb = &(*list_skb)->next; 3239 /* Append completion queue from offline CPU. */ 3240 *list_skb = oldsd->completion_queue; 3241 oldsd->completion_queue = NULL; 3242 3243 /* Find end of our output_queue. */ 3244 list_net = &sd->output_queue; 3245 while (*list_net) 3246 list_net = &(*list_net)->next_sched; 3247 /* Append output queue from offline CPU. */ 3248 *list_net = oldsd->output_queue; 3249 oldsd->output_queue = NULL; 3250 3251 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3252 local_irq_enable(); 3253 3254 /* Process offline CPU's input_pkt_queue */ 3255 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) 3256 netif_rx(skb); 3257 3258 return NOTIFY_OK; 3259} 3260#endif /* CONFIG_HOTPLUG_CPU */ 3261 3262 3263/* 3264 * Initialize the DEV module. At boot time this walks the device list and 3265 * unhooks any devices that fail to initialise (normally hardware not 3266 * present) and leaves us with a valid list of present and active devices. 3267 * 3268 */ 3269 3270/* 3271 * This is called single threaded during boot, so no need 3272 * to take the rtnl semaphore. 3273 */ 3274static int __init net_dev_init(void) 3275{ 3276 int i, rc = -ENOMEM; 3277 3278 BUG_ON(!dev_boot_phase); 3279 3280 net_random_init(); 3281 3282 if (dev_proc_init()) 3283 goto out; 3284 3285 if (netdev_sysfs_init()) 3286 goto out; 3287 3288 INIT_LIST_HEAD(&ptype_all); 3289 for (i = 0; i < 16; i++) 3290 INIT_LIST_HEAD(&ptype_base[i]); 3291 3292 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++) 3293 INIT_HLIST_HEAD(&dev_name_head[i]); 3294 3295 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++) 3296 INIT_HLIST_HEAD(&dev_index_head[i]); 3297 3298 /* 3299 * Initialise the packet receive queues. 3300 */ 3301 3302 for (i = 0; i < NR_CPUS; i++) { 3303 struct softnet_data *queue; 3304 3305 queue = &per_cpu(softnet_data, i); 3306 skb_queue_head_init(&queue->input_pkt_queue); 3307 queue->throttle = 0; 3308 queue->cng_level = 0; 3309 queue->avg_blog = 10; /* arbitrary non-zero */ 3310 queue->completion_queue = NULL; 3311 INIT_LIST_HEAD(&queue->poll_list); 3312 set_bit(__LINK_STATE_START, &queue->backlog_dev.state); 3313 queue->backlog_dev.weight = weight_p; 3314 queue->backlog_dev.poll = process_backlog; 3315 atomic_set(&queue->backlog_dev.refcnt, 1); 3316 } 3317 3318#ifdef OFFLINE_SAMPLE 3319 samp_timer.expires = jiffies + (10 * HZ); 3320 add_timer(&samp_timer); 3321#endif 3322 3323 dev_boot_phase = 0; 3324 3325 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL); 3326 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL); 3327 3328 hotcpu_notifier(dev_cpu_callback, 0); 3329 dst_init(); 3330 dev_mcast_init(); 3331 rc = 0; 3332out: 3333 return rc; 3334} 3335 3336subsys_initcall(net_dev_init); 3337 3338EXPORT_SYMBOL(__dev_get_by_index); 3339EXPORT_SYMBOL(__dev_get_by_name); 3340EXPORT_SYMBOL(__dev_remove_pack); 3341EXPORT_SYMBOL(__skb_linearize); 3342EXPORT_SYMBOL(dev_add_pack); 3343EXPORT_SYMBOL(dev_alloc_name); 3344EXPORT_SYMBOL(dev_close); 3345EXPORT_SYMBOL(dev_get_by_flags); 3346EXPORT_SYMBOL(dev_get_by_index); 3347EXPORT_SYMBOL(dev_get_by_name); 3348EXPORT_SYMBOL(dev_ioctl); 3349EXPORT_SYMBOL(dev_open); 3350EXPORT_SYMBOL(dev_queue_xmit); 3351EXPORT_SYMBOL(dev_remove_pack); 3352EXPORT_SYMBOL(dev_set_allmulti); 3353EXPORT_SYMBOL(dev_set_promiscuity); 3354EXPORT_SYMBOL(dev_change_flags); 3355EXPORT_SYMBOL(dev_set_mtu); 3356EXPORT_SYMBOL(dev_set_mac_address); 3357EXPORT_SYMBOL(free_netdev); 3358EXPORT_SYMBOL(netdev_boot_setup_check); 3359EXPORT_SYMBOL(netdev_set_master); 3360EXPORT_SYMBOL(netdev_state_change); 3361EXPORT_SYMBOL(netif_receive_skb); 3362EXPORT_SYMBOL(netif_rx); 3363EXPORT_SYMBOL(register_gifconf); 3364EXPORT_SYMBOL(register_netdevice); 3365EXPORT_SYMBOL(register_netdevice_notifier); 3366EXPORT_SYMBOL(skb_checksum_help); 3367EXPORT_SYMBOL(synchronize_net); 3368EXPORT_SYMBOL(unregister_netdevice); 3369EXPORT_SYMBOL(unregister_netdevice_notifier); 3370EXPORT_SYMBOL(net_enable_timestamp); 3371EXPORT_SYMBOL(net_disable_timestamp); 3372EXPORT_SYMBOL(dev_get_flags); 3373 3374#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) 3375EXPORT_SYMBOL(br_handle_frame_hook); 3376EXPORT_SYMBOL(br_fdb_get_hook); 3377EXPORT_SYMBOL(br_fdb_put_hook); 3378#endif 3379 3380#ifdef CONFIG_KMOD 3381EXPORT_SYMBOL(dev_load); 3382#endif 3383 3384EXPORT_PER_CPU_SYMBOL(softnet_data);