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