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