1 /* 2 * NET3 Protocol independent device support routines. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Derived from the non IP parts of dev.c 1.0.19 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Mark Evans, <evansmp@uhura.aston.ac.uk> 13 * 14 * Additional Authors: 15 * Florian la Roche <rzsfl@rz.uni-sb.de> 16 * Alan Cox <gw4pts@gw4pts.ampr.org> 17 * David Hinds <dahinds@users.sourceforge.net> 18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 19 * Adam Sulmicki <adam@cfar.umd.edu> 20 * Pekka Riikonen <priikone@poesidon.pspt.fi> 21 * 22 * Changes: 23 * D.J. Barrow : Fixed bug where dev->refcnt gets set 24 * to 2 if register_netdev gets called 25 * before net_dev_init & also removed a 26 * few lines of code in the process. 27 * Alan Cox : device private ioctl copies fields back. 28 * Alan Cox : Transmit queue code does relevant 29 * stunts to keep the queue safe. 30 * Alan Cox : Fixed double lock. 31 * Alan Cox : Fixed promisc NULL pointer trap 32 * ???????? : Support the full private ioctl range 33 * Alan Cox : Moved ioctl permission check into 34 * drivers 35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 36 * Alan Cox : 100 backlog just doesn't cut it when 37 * you start doing multicast video 8) 38 * Alan Cox : Rewrote net_bh and list manager. 39 * Alan Cox : Fix ETH_P_ALL echoback lengths. 40 * Alan Cox : Took out transmit every packet pass 41 * Saved a few bytes in the ioctl handler 42 * Alan Cox : Network driver sets packet type before 43 * calling netif_rx. Saves a function 44 * call a packet. 45 * Alan Cox : Hashed net_bh() 46 * Richard Kooijman: Timestamp fixes. 47 * Alan Cox : Wrong field in SIOCGIFDSTADDR 48 * Alan Cox : Device lock protection. 49 * Alan Cox : Fixed nasty side effect of device close 50 * changes. 51 * Rudi Cilibrasi : Pass the right thing to 52 * set_mac_address() 53 * Dave Miller : 32bit quantity for the device lock to 54 * make it work out on a Sparc. 55 * Bjorn Ekwall : Added KERNELD hack. 56 * Alan Cox : Cleaned up the backlog initialise. 57 * Craig Metz : SIOCGIFCONF fix if space for under 58 * 1 device. 59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 60 * is no device open function. 61 * Andi Kleen : Fix error reporting for SIOCGIFCONF 62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 63 * Cyrus Durgin : Cleaned for KMOD 64 * Adam Sulmicki : Bug Fix : Network Device Unload 65 * A network device unload needs to purge 66 * the backlog queue. 67 * Paul Rusty Russell : SIOCSIFNAME 68 * Pekka Riikonen : Netdev boot-time settings code 69 * Andrew Morton : Make unregister_netdevice wait 70 * indefinitely on dev->refcnt 71 * J Hadi Salim : - Backlog queue sampling 72 * - netif_rx() feedback 73 */ 74 75 #include <asm/uaccess.h> 76 #include <asm/system.h> 77 #include <linux/bitops.h> 78 #include <linux/capability.h> 79 #include <linux/cpu.h> 80 #include <linux/types.h> 81 #include <linux/kernel.h> 82 #include <linux/sched.h> 83 #include <linux/mutex.h> 84 #include <linux/string.h> 85 #include <linux/mm.h> 86 #include <linux/socket.h> 87 #include <linux/sockios.h> 88 #include <linux/errno.h> 89 #include <linux/interrupt.h> 90 #include <linux/if_ether.h> 91 #include <linux/netdevice.h> 92 #include <linux/etherdevice.h> 93 #include <linux/ethtool.h> 94 #include <linux/notifier.h> 95 #include <linux/skbuff.h> 96 #include <net/net_namespace.h> 97 #include <net/sock.h> 98 #include <linux/rtnetlink.h> 99 #include <linux/proc_fs.h> 100 #include <linux/seq_file.h> 101 #include <linux/stat.h> 102 #include <linux/if_bridge.h> 103 #include <linux/if_macvlan.h> 104 #include <net/dst.h> 105 #include <net/pkt_sched.h> 106 #include <net/checksum.h> 107 #include <linux/highmem.h> 108 #include <linux/init.h> 109 #include <linux/kmod.h> 110 #include <linux/module.h> 111 #include <linux/netpoll.h> 112 #include <linux/rcupdate.h> 113 #include <linux/delay.h> 114 #include <net/wext.h> 115 #include <net/iw_handler.h> 116 #include <asm/current.h> 117 #include <linux/audit.h> 118 #include <linux/dmaengine.h> 119 #include <linux/err.h> 120 #include <linux/ctype.h> 121 #include <linux/if_arp.h> 122 #include <linux/if_vlan.h> 123 #include <linux/ip.h> 124 #include <net/ip.h> 125 #include <linux/ipv6.h> 126 #include <linux/in.h> 127 #include <linux/jhash.h> 128 #include <linux/random.h> 129 130 #include "net-sysfs.h" 131 132 /* Instead of increasing this, you should create a hash table. */ 133 #define MAX_GRO_SKBS 8 134 135 /* This should be increased if a protocol with a bigger head is added. */ 136 #define GRO_MAX_HEAD (MAX_HEADER + 128) 137 138 /* 139 * The list of packet types we will receive (as opposed to discard) 140 * and the routines to invoke. 141 * 142 * Why 16. Because with 16 the only overlap we get on a hash of the 143 * low nibble of the protocol value is RARP/SNAP/X.25. 144 * 145 * NOTE: That is no longer true with the addition of VLAN tags. Not 146 * sure which should go first, but I bet it won't make much 147 * difference if we are running VLANs. The good news is that 148 * this protocol won't be in the list unless compiled in, so 149 * the average user (w/out VLANs) will not be adversely affected. 150 * --BLG 151 * 152 * 0800 IP 153 * 8100 802.1Q VLAN 154 * 0001 802.3 155 * 0002 AX.25 156 * 0004 802.2 157 * 8035 RARP 158 * 0005 SNAP 159 * 0805 X.25 160 * 0806 ARP 161 * 8137 IPX 162 * 0009 Localtalk 163 * 86DD IPv6 164 */ 165 166 #define PTYPE_HASH_SIZE (16) 167 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) 168 169 static DEFINE_SPINLOCK(ptype_lock); 170 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 171 static struct list_head ptype_all __read_mostly; /* Taps */ 172 173 /* 174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 175 * semaphore. 176 * 177 * Pure readers hold dev_base_lock for reading. 178 * 179 * Writers must hold the rtnl semaphore while they loop through the 180 * dev_base_head list, and hold dev_base_lock for writing when they do the 181 * actual updates. This allows pure readers to access the list even 182 * while a writer is preparing to update it. 183 * 184 * To put it another way, dev_base_lock is held for writing only to 185 * protect against pure readers; the rtnl semaphore provides the 186 * protection against other writers. 187 * 188 * See, for example usages, register_netdevice() and 189 * unregister_netdevice(), which must be called with the rtnl 190 * semaphore held. 191 */ 192 DEFINE_RWLOCK(dev_base_lock); 193 194 EXPORT_SYMBOL(dev_base_lock); 195 196 #define NETDEV_HASHBITS 8 197 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) 198 199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 200 { 201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); 202 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)]; 203 } 204 205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 206 { 207 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)]; 208 } 209 210 /* Device list insertion */ 211 static int list_netdevice(struct net_device *dev) 212 { 213 struct net *net = dev_net(dev); 214 215 ASSERT_RTNL(); 216 217 write_lock_bh(&dev_base_lock); 218 list_add_tail(&dev->dev_list, &net->dev_base_head); 219 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); 220 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex)); 221 write_unlock_bh(&dev_base_lock); 222 return 0; 223 } 224 225 /* Device list removal */ 226 static void unlist_netdevice(struct net_device *dev) 227 { 228 ASSERT_RTNL(); 229 230 /* Unlink dev from the device chain */ 231 write_lock_bh(&dev_base_lock); 232 list_del(&dev->dev_list); 233 hlist_del(&dev->name_hlist); 234 hlist_del(&dev->index_hlist); 235 write_unlock_bh(&dev_base_lock); 236 } 237 238 /* 239 * Our notifier list 240 */ 241 242 static RAW_NOTIFIER_HEAD(netdev_chain); 243 244 /* 245 * Device drivers call our routines to queue packets here. We empty the 246 * queue in the local softnet handler. 247 */ 248 249 DEFINE_PER_CPU(struct softnet_data, softnet_data); 250 251 #ifdef CONFIG_LOCKDEP 252 /* 253 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 254 * according to dev->type 255 */ 256 static const unsigned short netdev_lock_type[] = 257 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 258 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 259 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 260 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 261 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 262 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 263 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 264 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 265 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 266 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 267 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 268 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 269 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, 270 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, 271 ARPHRD_PHONET_PIPE, ARPHRD_VOID, ARPHRD_NONE}; 272 273 static const char *netdev_lock_name[] = 274 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 275 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 276 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 277 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 278 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 279 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 280 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 281 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 282 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 283 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 284 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 285 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 286 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", 287 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", 288 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"}; 289 290 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 291 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 292 293 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 294 { 295 int i; 296 297 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 298 if (netdev_lock_type[i] == dev_type) 299 return i; 300 /* the last key is used by default */ 301 return ARRAY_SIZE(netdev_lock_type) - 1; 302 } 303 304 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 305 unsigned short dev_type) 306 { 307 int i; 308 309 i = netdev_lock_pos(dev_type); 310 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 311 netdev_lock_name[i]); 312 } 313 314 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 315 { 316 int i; 317 318 i = netdev_lock_pos(dev->type); 319 lockdep_set_class_and_name(&dev->addr_list_lock, 320 &netdev_addr_lock_key[i], 321 netdev_lock_name[i]); 322 } 323 #else 324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 325 unsigned short dev_type) 326 { 327 } 328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 329 { 330 } 331 #endif 332 333 /******************************************************************************* 334 335 Protocol management and registration routines 336 337 *******************************************************************************/ 338 339 /* 340 * Add a protocol ID to the list. Now that the input handler is 341 * smarter we can dispense with all the messy stuff that used to be 342 * here. 343 * 344 * BEWARE!!! Protocol handlers, mangling input packets, 345 * MUST BE last in hash buckets and checking protocol handlers 346 * MUST start from promiscuous ptype_all chain in net_bh. 347 * It is true now, do not change it. 348 * Explanation follows: if protocol handler, mangling packet, will 349 * be the first on list, it is not able to sense, that packet 350 * is cloned and should be copied-on-write, so that it will 351 * change it and subsequent readers will get broken packet. 352 * --ANK (980803) 353 */ 354 355 /** 356 * dev_add_pack - add packet handler 357 * @pt: packet type declaration 358 * 359 * Add a protocol handler to the networking stack. The passed &packet_type 360 * is linked into kernel lists and may not be freed until it has been 361 * removed from the kernel lists. 362 * 363 * This call does not sleep therefore it can not 364 * guarantee all CPU's that are in middle of receiving packets 365 * will see the new packet type (until the next received packet). 366 */ 367 368 void dev_add_pack(struct packet_type *pt) 369 { 370 int hash; 371 372 spin_lock_bh(&ptype_lock); 373 if (pt->type == htons(ETH_P_ALL)) 374 list_add_rcu(&pt->list, &ptype_all); 375 else { 376 hash = ntohs(pt->type) & PTYPE_HASH_MASK; 377 list_add_rcu(&pt->list, &ptype_base[hash]); 378 } 379 spin_unlock_bh(&ptype_lock); 380 } 381 382 /** 383 * __dev_remove_pack - remove packet handler 384 * @pt: packet type declaration 385 * 386 * Remove a protocol handler that was previously added to the kernel 387 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 388 * from the kernel lists and can be freed or reused once this function 389 * returns. 390 * 391 * The packet type might still be in use by receivers 392 * and must not be freed until after all the CPU's have gone 393 * through a quiescent state. 394 */ 395 void __dev_remove_pack(struct packet_type *pt) 396 { 397 struct list_head *head; 398 struct packet_type *pt1; 399 400 spin_lock_bh(&ptype_lock); 401 402 if (pt->type == htons(ETH_P_ALL)) 403 head = &ptype_all; 404 else 405 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 406 407 list_for_each_entry(pt1, head, list) { 408 if (pt == pt1) { 409 list_del_rcu(&pt->list); 410 goto out; 411 } 412 } 413 414 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); 415 out: 416 spin_unlock_bh(&ptype_lock); 417 } 418 /** 419 * dev_remove_pack - remove packet handler 420 * @pt: packet type declaration 421 * 422 * Remove a protocol handler that was previously added to the kernel 423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 424 * from the kernel lists and can be freed or reused once this function 425 * returns. 426 * 427 * This call sleeps to guarantee that no CPU is looking at the packet 428 * type after return. 429 */ 430 void dev_remove_pack(struct packet_type *pt) 431 { 432 __dev_remove_pack(pt); 433 434 synchronize_net(); 435 } 436 437 /****************************************************************************** 438 439 Device Boot-time Settings Routines 440 441 *******************************************************************************/ 442 443 /* Boot time configuration table */ 444 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 445 446 /** 447 * netdev_boot_setup_add - add new setup entry 448 * @name: name of the device 449 * @map: configured settings for the device 450 * 451 * Adds new setup entry to the dev_boot_setup list. The function 452 * returns 0 on error and 1 on success. This is a generic routine to 453 * all netdevices. 454 */ 455 static int netdev_boot_setup_add(char *name, struct ifmap *map) 456 { 457 struct netdev_boot_setup *s; 458 int i; 459 460 s = dev_boot_setup; 461 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 462 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 463 memset(s[i].name, 0, sizeof(s[i].name)); 464 strlcpy(s[i].name, name, IFNAMSIZ); 465 memcpy(&s[i].map, map, sizeof(s[i].map)); 466 break; 467 } 468 } 469 470 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 471 } 472 473 /** 474 * netdev_boot_setup_check - check boot time settings 475 * @dev: the netdevice 476 * 477 * Check boot time settings for the device. 478 * The found settings are set for the device to be used 479 * later in the device probing. 480 * Returns 0 if no settings found, 1 if they are. 481 */ 482 int netdev_boot_setup_check(struct net_device *dev) 483 { 484 struct netdev_boot_setup *s = dev_boot_setup; 485 int i; 486 487 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 488 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 489 !strcmp(dev->name, s[i].name)) { 490 dev->irq = s[i].map.irq; 491 dev->base_addr = s[i].map.base_addr; 492 dev->mem_start = s[i].map.mem_start; 493 dev->mem_end = s[i].map.mem_end; 494 return 1; 495 } 496 } 497 return 0; 498 } 499 500 501 /** 502 * netdev_boot_base - get address from boot time settings 503 * @prefix: prefix for network device 504 * @unit: id for network device 505 * 506 * Check boot time settings for the base address of device. 507 * The found settings are set for the device to be used 508 * later in the device probing. 509 * Returns 0 if no settings found. 510 */ 511 unsigned long netdev_boot_base(const char *prefix, int unit) 512 { 513 const struct netdev_boot_setup *s = dev_boot_setup; 514 char name[IFNAMSIZ]; 515 int i; 516 517 sprintf(name, "%s%d", prefix, unit); 518 519 /* 520 * If device already registered then return base of 1 521 * to indicate not to probe for this interface 522 */ 523 if (__dev_get_by_name(&init_net, name)) 524 return 1; 525 526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 527 if (!strcmp(name, s[i].name)) 528 return s[i].map.base_addr; 529 return 0; 530 } 531 532 /* 533 * Saves at boot time configured settings for any netdevice. 534 */ 535 int __init netdev_boot_setup(char *str) 536 { 537 int ints[5]; 538 struct ifmap map; 539 540 str = get_options(str, ARRAY_SIZE(ints), ints); 541 if (!str || !*str) 542 return 0; 543 544 /* Save settings */ 545 memset(&map, 0, sizeof(map)); 546 if (ints[0] > 0) 547 map.irq = ints[1]; 548 if (ints[0] > 1) 549 map.base_addr = ints[2]; 550 if (ints[0] > 2) 551 map.mem_start = ints[3]; 552 if (ints[0] > 3) 553 map.mem_end = ints[4]; 554 555 /* Add new entry to the list */ 556 return netdev_boot_setup_add(str, &map); 557 } 558 559 __setup("netdev=", netdev_boot_setup); 560 561 /******************************************************************************* 562 563 Device Interface Subroutines 564 565 *******************************************************************************/ 566 567 /** 568 * __dev_get_by_name - find a device by its name 569 * @net: the applicable net namespace 570 * @name: name to find 571 * 572 * Find an interface by name. Must be called under RTNL semaphore 573 * or @dev_base_lock. If the name is found a pointer to the device 574 * is returned. If the name is not found then %NULL is returned. The 575 * reference counters are not incremented so the caller must be 576 * careful with locks. 577 */ 578 579 struct net_device *__dev_get_by_name(struct net *net, const char *name) 580 { 581 struct hlist_node *p; 582 583 hlist_for_each(p, dev_name_hash(net, name)) { 584 struct net_device *dev 585 = hlist_entry(p, struct net_device, name_hlist); 586 if (!strncmp(dev->name, name, IFNAMSIZ)) 587 return dev; 588 } 589 return NULL; 590 } 591 592 /** 593 * dev_get_by_name - find a device by its name 594 * @net: the applicable net namespace 595 * @name: name to find 596 * 597 * Find an interface by name. This can be called from any 598 * context and does its own locking. The returned handle has 599 * the usage count incremented and the caller must use dev_put() to 600 * release it when it is no longer needed. %NULL is returned if no 601 * matching device is found. 602 */ 603 604 struct net_device *dev_get_by_name(struct net *net, const char *name) 605 { 606 struct net_device *dev; 607 608 read_lock(&dev_base_lock); 609 dev = __dev_get_by_name(net, name); 610 if (dev) 611 dev_hold(dev); 612 read_unlock(&dev_base_lock); 613 return dev; 614 } 615 616 /** 617 * __dev_get_by_index - find a device by its ifindex 618 * @net: the applicable net namespace 619 * @ifindex: index of device 620 * 621 * Search for an interface by index. Returns %NULL if the device 622 * is not found or a pointer to the device. The device has not 623 * had its reference counter increased so the caller must be careful 624 * about locking. The caller must hold either the RTNL semaphore 625 * or @dev_base_lock. 626 */ 627 628 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 629 { 630 struct hlist_node *p; 631 632 hlist_for_each(p, dev_index_hash(net, ifindex)) { 633 struct net_device *dev 634 = hlist_entry(p, struct net_device, index_hlist); 635 if (dev->ifindex == ifindex) 636 return dev; 637 } 638 return NULL; 639 } 640 641 642 /** 643 * dev_get_by_index - find a device by its ifindex 644 * @net: the applicable net namespace 645 * @ifindex: index of device 646 * 647 * Search for an interface by index. Returns NULL if the device 648 * is not found or a pointer to the device. The device returned has 649 * had a reference added and the pointer is safe until the user calls 650 * dev_put to indicate they have finished with it. 651 */ 652 653 struct net_device *dev_get_by_index(struct net *net, int ifindex) 654 { 655 struct net_device *dev; 656 657 read_lock(&dev_base_lock); 658 dev = __dev_get_by_index(net, ifindex); 659 if (dev) 660 dev_hold(dev); 661 read_unlock(&dev_base_lock); 662 return dev; 663 } 664 665 /** 666 * dev_getbyhwaddr - find a device by its hardware address 667 * @net: the applicable net namespace 668 * @type: media type of device 669 * @ha: hardware address 670 * 671 * Search for an interface by MAC address. Returns NULL if the device 672 * is not found or a pointer to the device. The caller must hold the 673 * rtnl semaphore. The returned device has not had its ref count increased 674 * and the caller must therefore be careful about locking 675 * 676 * BUGS: 677 * If the API was consistent this would be __dev_get_by_hwaddr 678 */ 679 680 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) 681 { 682 struct net_device *dev; 683 684 ASSERT_RTNL(); 685 686 for_each_netdev(net, dev) 687 if (dev->type == type && 688 !memcmp(dev->dev_addr, ha, dev->addr_len)) 689 return dev; 690 691 return NULL; 692 } 693 694 EXPORT_SYMBOL(dev_getbyhwaddr); 695 696 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 697 { 698 struct net_device *dev; 699 700 ASSERT_RTNL(); 701 for_each_netdev(net, dev) 702 if (dev->type == type) 703 return dev; 704 705 return NULL; 706 } 707 708 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 709 710 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 711 { 712 struct net_device *dev; 713 714 rtnl_lock(); 715 dev = __dev_getfirstbyhwtype(net, type); 716 if (dev) 717 dev_hold(dev); 718 rtnl_unlock(); 719 return dev; 720 } 721 722 EXPORT_SYMBOL(dev_getfirstbyhwtype); 723 724 /** 725 * dev_get_by_flags - find any device with given flags 726 * @net: the applicable net namespace 727 * @if_flags: IFF_* values 728 * @mask: bitmask of bits in if_flags to check 729 * 730 * Search for any interface with the given flags. Returns NULL if a device 731 * is not found or a pointer to the device. The device returned has 732 * had a reference added and the pointer is safe until the user calls 733 * dev_put to indicate they have finished with it. 734 */ 735 736 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask) 737 { 738 struct net_device *dev, *ret; 739 740 ret = NULL; 741 read_lock(&dev_base_lock); 742 for_each_netdev(net, dev) { 743 if (((dev->flags ^ if_flags) & mask) == 0) { 744 dev_hold(dev); 745 ret = dev; 746 break; 747 } 748 } 749 read_unlock(&dev_base_lock); 750 return ret; 751 } 752 753 /** 754 * dev_valid_name - check if name is okay for network device 755 * @name: name string 756 * 757 * Network device names need to be valid file names to 758 * to allow sysfs to work. We also disallow any kind of 759 * whitespace. 760 */ 761 int dev_valid_name(const char *name) 762 { 763 if (*name == '\0') 764 return 0; 765 if (strlen(name) >= IFNAMSIZ) 766 return 0; 767 if (!strcmp(name, ".") || !strcmp(name, "..")) 768 return 0; 769 770 while (*name) { 771 if (*name == '/' || isspace(*name)) 772 return 0; 773 name++; 774 } 775 return 1; 776 } 777 778 /** 779 * __dev_alloc_name - allocate a name for a device 780 * @net: network namespace to allocate the device name in 781 * @name: name format string 782 * @buf: scratch buffer and result name string 783 * 784 * Passed a format string - eg "lt%d" it will try and find a suitable 785 * id. It scans list of devices to build up a free map, then chooses 786 * the first empty slot. The caller must hold the dev_base or rtnl lock 787 * while allocating the name and adding the device in order to avoid 788 * duplicates. 789 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 790 * Returns the number of the unit assigned or a negative errno code. 791 */ 792 793 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 794 { 795 int i = 0; 796 const char *p; 797 const int max_netdevices = 8*PAGE_SIZE; 798 unsigned long *inuse; 799 struct net_device *d; 800 801 p = strnchr(name, IFNAMSIZ-1, '%'); 802 if (p) { 803 /* 804 * Verify the string as this thing may have come from 805 * the user. There must be either one "%d" and no other "%" 806 * characters. 807 */ 808 if (p[1] != 'd' || strchr(p + 2, '%')) 809 return -EINVAL; 810 811 /* Use one page as a bit array of possible slots */ 812 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 813 if (!inuse) 814 return -ENOMEM; 815 816 for_each_netdev(net, d) { 817 if (!sscanf(d->name, name, &i)) 818 continue; 819 if (i < 0 || i >= max_netdevices) 820 continue; 821 822 /* avoid cases where sscanf is not exact inverse of printf */ 823 snprintf(buf, IFNAMSIZ, name, i); 824 if (!strncmp(buf, d->name, IFNAMSIZ)) 825 set_bit(i, inuse); 826 } 827 828 i = find_first_zero_bit(inuse, max_netdevices); 829 free_page((unsigned long) inuse); 830 } 831 832 snprintf(buf, IFNAMSIZ, name, i); 833 if (!__dev_get_by_name(net, buf)) 834 return i; 835 836 /* It is possible to run out of possible slots 837 * when the name is long and there isn't enough space left 838 * for the digits, or if all bits are used. 839 */ 840 return -ENFILE; 841 } 842 843 /** 844 * dev_alloc_name - allocate a name for a device 845 * @dev: device 846 * @name: name format string 847 * 848 * Passed a format string - eg "lt%d" it will try and find a suitable 849 * id. It scans list of devices to build up a free map, then chooses 850 * the first empty slot. The caller must hold the dev_base or rtnl lock 851 * while allocating the name and adding the device in order to avoid 852 * duplicates. 853 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 854 * Returns the number of the unit assigned or a negative errno code. 855 */ 856 857 int dev_alloc_name(struct net_device *dev, const char *name) 858 { 859 char buf[IFNAMSIZ]; 860 struct net *net; 861 int ret; 862 863 BUG_ON(!dev_net(dev)); 864 net = dev_net(dev); 865 ret = __dev_alloc_name(net, name, buf); 866 if (ret >= 0) 867 strlcpy(dev->name, buf, IFNAMSIZ); 868 return ret; 869 } 870 871 872 /** 873 * dev_change_name - change name of a device 874 * @dev: device 875 * @newname: name (or format string) must be at least IFNAMSIZ 876 * 877 * Change name of a device, can pass format strings "eth%d". 878 * for wildcarding. 879 */ 880 int dev_change_name(struct net_device *dev, const char *newname) 881 { 882 char oldname[IFNAMSIZ]; 883 int err = 0; 884 int ret; 885 struct net *net; 886 887 ASSERT_RTNL(); 888 BUG_ON(!dev_net(dev)); 889 890 net = dev_net(dev); 891 if (dev->flags & IFF_UP) 892 return -EBUSY; 893 894 if (!dev_valid_name(newname)) 895 return -EINVAL; 896 897 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) 898 return 0; 899 900 memcpy(oldname, dev->name, IFNAMSIZ); 901 902 if (strchr(newname, '%')) { 903 err = dev_alloc_name(dev, newname); 904 if (err < 0) 905 return err; 906 } 907 else if (__dev_get_by_name(net, newname)) 908 return -EEXIST; 909 else 910 strlcpy(dev->name, newname, IFNAMSIZ); 911 912 rollback: 913 /* For now only devices in the initial network namespace 914 * are in sysfs. 915 */ 916 if (net == &init_net) { 917 ret = device_rename(&dev->dev, dev->name); 918 if (ret) { 919 memcpy(dev->name, oldname, IFNAMSIZ); 920 return ret; 921 } 922 } 923 924 write_lock_bh(&dev_base_lock); 925 hlist_del(&dev->name_hlist); 926 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); 927 write_unlock_bh(&dev_base_lock); 928 929 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 930 ret = notifier_to_errno(ret); 931 932 if (ret) { 933 if (err) { 934 printk(KERN_ERR 935 "%s: name change rollback failed: %d.\n", 936 dev->name, ret); 937 } else { 938 err = ret; 939 memcpy(dev->name, oldname, IFNAMSIZ); 940 goto rollback; 941 } 942 } 943 944 return err; 945 } 946 947 /** 948 * dev_set_alias - change ifalias of a device 949 * @dev: device 950 * @alias: name up to IFALIASZ 951 * @len: limit of bytes to copy from info 952 * 953 * Set ifalias for a device, 954 */ 955 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 956 { 957 ASSERT_RTNL(); 958 959 if (len >= IFALIASZ) 960 return -EINVAL; 961 962 if (!len) { 963 if (dev->ifalias) { 964 kfree(dev->ifalias); 965 dev->ifalias = NULL; 966 } 967 return 0; 968 } 969 970 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL); 971 if (!dev->ifalias) 972 return -ENOMEM; 973 974 strlcpy(dev->ifalias, alias, len+1); 975 return len; 976 } 977 978 979 /** 980 * netdev_features_change - device changes features 981 * @dev: device to cause notification 982 * 983 * Called to indicate a device has changed features. 984 */ 985 void netdev_features_change(struct net_device *dev) 986 { 987 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 988 } 989 EXPORT_SYMBOL(netdev_features_change); 990 991 /** 992 * netdev_state_change - device changes state 993 * @dev: device to cause notification 994 * 995 * Called to indicate a device has changed state. This function calls 996 * the notifier chains for netdev_chain and sends a NEWLINK message 997 * to the routing socket. 998 */ 999 void netdev_state_change(struct net_device *dev) 1000 { 1001 if (dev->flags & IFF_UP) { 1002 call_netdevice_notifiers(NETDEV_CHANGE, dev); 1003 rtmsg_ifinfo(RTM_NEWLINK, dev, 0); 1004 } 1005 } 1006 1007 void netdev_bonding_change(struct net_device *dev) 1008 { 1009 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev); 1010 } 1011 EXPORT_SYMBOL(netdev_bonding_change); 1012 1013 /** 1014 * dev_load - load a network module 1015 * @net: the applicable net namespace 1016 * @name: name of interface 1017 * 1018 * If a network interface is not present and the process has suitable 1019 * privileges this function loads the module. If module loading is not 1020 * available in this kernel then it becomes a nop. 1021 */ 1022 1023 void dev_load(struct net *net, const char *name) 1024 { 1025 struct net_device *dev; 1026 1027 read_lock(&dev_base_lock); 1028 dev = __dev_get_by_name(net, name); 1029 read_unlock(&dev_base_lock); 1030 1031 if (!dev && capable(CAP_SYS_MODULE)) 1032 request_module("%s", name); 1033 } 1034 1035 /** 1036 * dev_open - prepare an interface for use. 1037 * @dev: device to open 1038 * 1039 * Takes a device from down to up state. The device's private open 1040 * function is invoked and then the multicast lists are loaded. Finally 1041 * the device is moved into the up state and a %NETDEV_UP message is 1042 * sent to the netdev notifier chain. 1043 * 1044 * Calling this function on an active interface is a nop. On a failure 1045 * a negative errno code is returned. 1046 */ 1047 int dev_open(struct net_device *dev) 1048 { 1049 const struct net_device_ops *ops = dev->netdev_ops; 1050 int ret = 0; 1051 1052 ASSERT_RTNL(); 1053 1054 /* 1055 * Is it already up? 1056 */ 1057 1058 if (dev->flags & IFF_UP) 1059 return 0; 1060 1061 /* 1062 * Is it even present? 1063 */ 1064 if (!netif_device_present(dev)) 1065 return -ENODEV; 1066 1067 /* 1068 * Call device private open method 1069 */ 1070 set_bit(__LINK_STATE_START, &dev->state); 1071 1072 if (ops->ndo_validate_addr) 1073 ret = ops->ndo_validate_addr(dev); 1074 1075 if (!ret && ops->ndo_open) 1076 ret = ops->ndo_open(dev); 1077 1078 /* 1079 * If it went open OK then: 1080 */ 1081 1082 if (ret) 1083 clear_bit(__LINK_STATE_START, &dev->state); 1084 else { 1085 /* 1086 * Set the flags. 1087 */ 1088 dev->flags |= IFF_UP; 1089 1090 /* 1091 * Enable NET_DMA 1092 */ 1093 net_dmaengine_get(); 1094 1095 /* 1096 * Initialize multicasting status 1097 */ 1098 dev_set_rx_mode(dev); 1099 1100 /* 1101 * Wakeup transmit queue engine 1102 */ 1103 dev_activate(dev); 1104 1105 /* 1106 * ... and announce new interface. 1107 */ 1108 call_netdevice_notifiers(NETDEV_UP, dev); 1109 } 1110 1111 return ret; 1112 } 1113 1114 /** 1115 * dev_close - shutdown an interface. 1116 * @dev: device to shutdown 1117 * 1118 * This function moves an active device into down state. A 1119 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1120 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1121 * chain. 1122 */ 1123 int dev_close(struct net_device *dev) 1124 { 1125 const struct net_device_ops *ops = dev->netdev_ops; 1126 ASSERT_RTNL(); 1127 1128 might_sleep(); 1129 1130 if (!(dev->flags & IFF_UP)) 1131 return 0; 1132 1133 /* 1134 * Tell people we are going down, so that they can 1135 * prepare to death, when device is still operating. 1136 */ 1137 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1138 1139 clear_bit(__LINK_STATE_START, &dev->state); 1140 1141 /* Synchronize to scheduled poll. We cannot touch poll list, 1142 * it can be even on different cpu. So just clear netif_running(). 1143 * 1144 * dev->stop() will invoke napi_disable() on all of it's 1145 * napi_struct instances on this device. 1146 */ 1147 smp_mb__after_clear_bit(); /* Commit netif_running(). */ 1148 1149 dev_deactivate(dev); 1150 1151 /* 1152 * Call the device specific close. This cannot fail. 1153 * Only if device is UP 1154 * 1155 * We allow it to be called even after a DETACH hot-plug 1156 * event. 1157 */ 1158 if (ops->ndo_stop) 1159 ops->ndo_stop(dev); 1160 1161 /* 1162 * Device is now down. 1163 */ 1164 1165 dev->flags &= ~IFF_UP; 1166 1167 /* 1168 * Tell people we are down 1169 */ 1170 call_netdevice_notifiers(NETDEV_DOWN, dev); 1171 1172 /* 1173 * Shutdown NET_DMA 1174 */ 1175 net_dmaengine_put(); 1176 1177 return 0; 1178 } 1179 1180 1181 /** 1182 * dev_disable_lro - disable Large Receive Offload on a device 1183 * @dev: device 1184 * 1185 * Disable Large Receive Offload (LRO) on a net device. Must be 1186 * called under RTNL. This is needed if received packets may be 1187 * forwarded to another interface. 1188 */ 1189 void dev_disable_lro(struct net_device *dev) 1190 { 1191 if (dev->ethtool_ops && dev->ethtool_ops->get_flags && 1192 dev->ethtool_ops->set_flags) { 1193 u32 flags = dev->ethtool_ops->get_flags(dev); 1194 if (flags & ETH_FLAG_LRO) { 1195 flags &= ~ETH_FLAG_LRO; 1196 dev->ethtool_ops->set_flags(dev, flags); 1197 } 1198 } 1199 WARN_ON(dev->features & NETIF_F_LRO); 1200 } 1201 EXPORT_SYMBOL(dev_disable_lro); 1202 1203 1204 static int dev_boot_phase = 1; 1205 1206 /* 1207 * Device change register/unregister. These are not inline or static 1208 * as we export them to the world. 1209 */ 1210 1211 /** 1212 * register_netdevice_notifier - register a network notifier block 1213 * @nb: notifier 1214 * 1215 * Register a notifier to be called when network device events occur. 1216 * The notifier passed is linked into the kernel structures and must 1217 * not be reused until it has been unregistered. A negative errno code 1218 * is returned on a failure. 1219 * 1220 * When registered all registration and up events are replayed 1221 * to the new notifier to allow device to have a race free 1222 * view of the network device list. 1223 */ 1224 1225 int register_netdevice_notifier(struct notifier_block *nb) 1226 { 1227 struct net_device *dev; 1228 struct net_device *last; 1229 struct net *net; 1230 int err; 1231 1232 rtnl_lock(); 1233 err = raw_notifier_chain_register(&netdev_chain, nb); 1234 if (err) 1235 goto unlock; 1236 if (dev_boot_phase) 1237 goto unlock; 1238 for_each_net(net) { 1239 for_each_netdev(net, dev) { 1240 err = nb->notifier_call(nb, NETDEV_REGISTER, dev); 1241 err = notifier_to_errno(err); 1242 if (err) 1243 goto rollback; 1244 1245 if (!(dev->flags & IFF_UP)) 1246 continue; 1247 1248 nb->notifier_call(nb, NETDEV_UP, dev); 1249 } 1250 } 1251 1252 unlock: 1253 rtnl_unlock(); 1254 return err; 1255 1256 rollback: 1257 last = dev; 1258 for_each_net(net) { 1259 for_each_netdev(net, dev) { 1260 if (dev == last) 1261 break; 1262 1263 if (dev->flags & IFF_UP) { 1264 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); 1265 nb->notifier_call(nb, NETDEV_DOWN, dev); 1266 } 1267 nb->notifier_call(nb, NETDEV_UNREGISTER, dev); 1268 } 1269 } 1270 1271 raw_notifier_chain_unregister(&netdev_chain, nb); 1272 goto unlock; 1273 } 1274 1275 /** 1276 * unregister_netdevice_notifier - unregister a network notifier block 1277 * @nb: notifier 1278 * 1279 * Unregister a notifier previously registered by 1280 * register_netdevice_notifier(). The notifier is unlinked into the 1281 * kernel structures and may then be reused. A negative errno code 1282 * is returned on a failure. 1283 */ 1284 1285 int unregister_netdevice_notifier(struct notifier_block *nb) 1286 { 1287 int err; 1288 1289 rtnl_lock(); 1290 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1291 rtnl_unlock(); 1292 return err; 1293 } 1294 1295 /** 1296 * call_netdevice_notifiers - call all network notifier blocks 1297 * @val: value passed unmodified to notifier function 1298 * @dev: net_device pointer passed unmodified to notifier function 1299 * 1300 * Call all network notifier blocks. Parameters and return value 1301 * are as for raw_notifier_call_chain(). 1302 */ 1303 1304 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1305 { 1306 return raw_notifier_call_chain(&netdev_chain, val, dev); 1307 } 1308 1309 /* When > 0 there are consumers of rx skb time stamps */ 1310 static atomic_t netstamp_needed = ATOMIC_INIT(0); 1311 1312 void net_enable_timestamp(void) 1313 { 1314 atomic_inc(&netstamp_needed); 1315 } 1316 1317 void net_disable_timestamp(void) 1318 { 1319 atomic_dec(&netstamp_needed); 1320 } 1321 1322 static inline void net_timestamp(struct sk_buff *skb) 1323 { 1324 if (atomic_read(&netstamp_needed)) 1325 __net_timestamp(skb); 1326 else 1327 skb->tstamp.tv64 = 0; 1328 } 1329 1330 /* 1331 * Support routine. Sends outgoing frames to any network 1332 * taps currently in use. 1333 */ 1334 1335 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1336 { 1337 struct packet_type *ptype; 1338 1339 net_timestamp(skb); 1340 1341 rcu_read_lock(); 1342 list_for_each_entry_rcu(ptype, &ptype_all, list) { 1343 /* Never send packets back to the socket 1344 * they originated from - MvS (miquels@drinkel.ow.org) 1345 */ 1346 if ((ptype->dev == dev || !ptype->dev) && 1347 (ptype->af_packet_priv == NULL || 1348 (struct sock *)ptype->af_packet_priv != skb->sk)) { 1349 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); 1350 if (!skb2) 1351 break; 1352 1353 /* skb->nh should be correctly 1354 set by sender, so that the second statement is 1355 just protection against buggy protocols. 1356 */ 1357 skb_reset_mac_header(skb2); 1358 1359 if (skb_network_header(skb2) < skb2->data || 1360 skb2->network_header > skb2->tail) { 1361 if (net_ratelimit()) 1362 printk(KERN_CRIT "protocol %04x is " 1363 "buggy, dev %s\n", 1364 skb2->protocol, dev->name); 1365 skb_reset_network_header(skb2); 1366 } 1367 1368 skb2->transport_header = skb2->network_header; 1369 skb2->pkt_type = PACKET_OUTGOING; 1370 ptype->func(skb2, skb->dev, ptype, skb->dev); 1371 } 1372 } 1373 rcu_read_unlock(); 1374 } 1375 1376 1377 static inline void __netif_reschedule(struct Qdisc *q) 1378 { 1379 struct softnet_data *sd; 1380 unsigned long flags; 1381 1382 local_irq_save(flags); 1383 sd = &__get_cpu_var(softnet_data); 1384 q->next_sched = sd->output_queue; 1385 sd->output_queue = q; 1386 raise_softirq_irqoff(NET_TX_SOFTIRQ); 1387 local_irq_restore(flags); 1388 } 1389 1390 void __netif_schedule(struct Qdisc *q) 1391 { 1392 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 1393 __netif_reschedule(q); 1394 } 1395 EXPORT_SYMBOL(__netif_schedule); 1396 1397 void dev_kfree_skb_irq(struct sk_buff *skb) 1398 { 1399 if (atomic_dec_and_test(&skb->users)) { 1400 struct softnet_data *sd; 1401 unsigned long flags; 1402 1403 local_irq_save(flags); 1404 sd = &__get_cpu_var(softnet_data); 1405 skb->next = sd->completion_queue; 1406 sd->completion_queue = skb; 1407 raise_softirq_irqoff(NET_TX_SOFTIRQ); 1408 local_irq_restore(flags); 1409 } 1410 } 1411 EXPORT_SYMBOL(dev_kfree_skb_irq); 1412 1413 void dev_kfree_skb_any(struct sk_buff *skb) 1414 { 1415 if (in_irq() || irqs_disabled()) 1416 dev_kfree_skb_irq(skb); 1417 else 1418 dev_kfree_skb(skb); 1419 } 1420 EXPORT_SYMBOL(dev_kfree_skb_any); 1421 1422 1423 /** 1424 * netif_device_detach - mark device as removed 1425 * @dev: network device 1426 * 1427 * Mark device as removed from system and therefore no longer available. 1428 */ 1429 void netif_device_detach(struct net_device *dev) 1430 { 1431 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 1432 netif_running(dev)) { 1433 netif_stop_queue(dev); 1434 } 1435 } 1436 EXPORT_SYMBOL(netif_device_detach); 1437 1438 /** 1439 * netif_device_attach - mark device as attached 1440 * @dev: network device 1441 * 1442 * Mark device as attached from system and restart if needed. 1443 */ 1444 void netif_device_attach(struct net_device *dev) 1445 { 1446 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 1447 netif_running(dev)) { 1448 netif_wake_queue(dev); 1449 __netdev_watchdog_up(dev); 1450 } 1451 } 1452 EXPORT_SYMBOL(netif_device_attach); 1453 1454 static bool can_checksum_protocol(unsigned long features, __be16 protocol) 1455 { 1456 return ((features & NETIF_F_GEN_CSUM) || 1457 ((features & NETIF_F_IP_CSUM) && 1458 protocol == htons(ETH_P_IP)) || 1459 ((features & NETIF_F_IPV6_CSUM) && 1460 protocol == htons(ETH_P_IPV6)) || 1461 ((features & NETIF_F_FCOE_CRC) && 1462 protocol == htons(ETH_P_FCOE))); 1463 } 1464 1465 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) 1466 { 1467 if (can_checksum_protocol(dev->features, skb->protocol)) 1468 return true; 1469 1470 if (skb->protocol == htons(ETH_P_8021Q)) { 1471 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; 1472 if (can_checksum_protocol(dev->features & dev->vlan_features, 1473 veh->h_vlan_encapsulated_proto)) 1474 return true; 1475 } 1476 1477 return false; 1478 } 1479 1480 /* 1481 * Invalidate hardware checksum when packet is to be mangled, and 1482 * complete checksum manually on outgoing path. 1483 */ 1484 int skb_checksum_help(struct sk_buff *skb) 1485 { 1486 __wsum csum; 1487 int ret = 0, offset; 1488 1489 if (skb->ip_summed == CHECKSUM_COMPLETE) 1490 goto out_set_summed; 1491 1492 if (unlikely(skb_shinfo(skb)->gso_size)) { 1493 /* Let GSO fix up the checksum. */ 1494 goto out_set_summed; 1495 } 1496 1497 offset = skb->csum_start - skb_headroom(skb); 1498 BUG_ON(offset >= skb_headlen(skb)); 1499 csum = skb_checksum(skb, offset, skb->len - offset, 0); 1500 1501 offset += skb->csum_offset; 1502 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 1503 1504 if (skb_cloned(skb) && 1505 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 1506 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 1507 if (ret) 1508 goto out; 1509 } 1510 1511 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 1512 out_set_summed: 1513 skb->ip_summed = CHECKSUM_NONE; 1514 out: 1515 return ret; 1516 } 1517 1518 /** 1519 * skb_gso_segment - Perform segmentation on skb. 1520 * @skb: buffer to segment 1521 * @features: features for the output path (see dev->features) 1522 * 1523 * This function segments the given skb and returns a list of segments. 1524 * 1525 * It may return NULL if the skb requires no segmentation. This is 1526 * only possible when GSO is used for verifying header integrity. 1527 */ 1528 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) 1529 { 1530 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 1531 struct packet_type *ptype; 1532 __be16 type = skb->protocol; 1533 int err; 1534 1535 skb_reset_mac_header(skb); 1536 skb->mac_len = skb->network_header - skb->mac_header; 1537 __skb_pull(skb, skb->mac_len); 1538 1539 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { 1540 struct net_device *dev = skb->dev; 1541 struct ethtool_drvinfo info = {}; 1542 1543 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) 1544 dev->ethtool_ops->get_drvinfo(dev, &info); 1545 1546 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " 1547 "ip_summed=%d", 1548 info.driver, dev ? dev->features : 0L, 1549 skb->sk ? skb->sk->sk_route_caps : 0L, 1550 skb->len, skb->data_len, skb->ip_summed); 1551 1552 if (skb_header_cloned(skb) && 1553 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) 1554 return ERR_PTR(err); 1555 } 1556 1557 rcu_read_lock(); 1558 list_for_each_entry_rcu(ptype, 1559 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { 1560 if (ptype->type == type && !ptype->dev && ptype->gso_segment) { 1561 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { 1562 err = ptype->gso_send_check(skb); 1563 segs = ERR_PTR(err); 1564 if (err || skb_gso_ok(skb, features)) 1565 break; 1566 __skb_push(skb, (skb->data - 1567 skb_network_header(skb))); 1568 } 1569 segs = ptype->gso_segment(skb, features); 1570 break; 1571 } 1572 } 1573 rcu_read_unlock(); 1574 1575 __skb_push(skb, skb->data - skb_mac_header(skb)); 1576 1577 return segs; 1578 } 1579 1580 EXPORT_SYMBOL(skb_gso_segment); 1581 1582 /* Take action when hardware reception checksum errors are detected. */ 1583 #ifdef CONFIG_BUG 1584 void netdev_rx_csum_fault(struct net_device *dev) 1585 { 1586 if (net_ratelimit()) { 1587 printk(KERN_ERR "%s: hw csum failure.\n", 1588 dev ? dev->name : "<unknown>"); 1589 dump_stack(); 1590 } 1591 } 1592 EXPORT_SYMBOL(netdev_rx_csum_fault); 1593 #endif 1594 1595 /* Actually, we should eliminate this check as soon as we know, that: 1596 * 1. IOMMU is present and allows to map all the memory. 1597 * 2. No high memory really exists on this machine. 1598 */ 1599 1600 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 1601 { 1602 #ifdef CONFIG_HIGHMEM 1603 int i; 1604 1605 if (dev->features & NETIF_F_HIGHDMA) 1606 return 0; 1607 1608 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1609 if (PageHighMem(skb_shinfo(skb)->frags[i].page)) 1610 return 1; 1611 1612 #endif 1613 return 0; 1614 } 1615 1616 struct dev_gso_cb { 1617 void (*destructor)(struct sk_buff *skb); 1618 }; 1619 1620 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) 1621 1622 static void dev_gso_skb_destructor(struct sk_buff *skb) 1623 { 1624 struct dev_gso_cb *cb; 1625 1626 do { 1627 struct sk_buff *nskb = skb->next; 1628 1629 skb->next = nskb->next; 1630 nskb->next = NULL; 1631 kfree_skb(nskb); 1632 } while (skb->next); 1633 1634 cb = DEV_GSO_CB(skb); 1635 if (cb->destructor) 1636 cb->destructor(skb); 1637 } 1638 1639 /** 1640 * dev_gso_segment - Perform emulated hardware segmentation on skb. 1641 * @skb: buffer to segment 1642 * 1643 * This function segments the given skb and stores the list of segments 1644 * in skb->next. 1645 */ 1646 static int dev_gso_segment(struct sk_buff *skb) 1647 { 1648 struct net_device *dev = skb->dev; 1649 struct sk_buff *segs; 1650 int features = dev->features & ~(illegal_highdma(dev, skb) ? 1651 NETIF_F_SG : 0); 1652 1653 segs = skb_gso_segment(skb, features); 1654 1655 /* Verifying header integrity only. */ 1656 if (!segs) 1657 return 0; 1658 1659 if (IS_ERR(segs)) 1660 return PTR_ERR(segs); 1661 1662 skb->next = segs; 1663 DEV_GSO_CB(skb)->destructor = skb->destructor; 1664 skb->destructor = dev_gso_skb_destructor; 1665 1666 return 0; 1667 } 1668 1669 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, 1670 struct netdev_queue *txq) 1671 { 1672 const struct net_device_ops *ops = dev->netdev_ops; 1673 int rc; 1674 1675 if (likely(!skb->next)) { 1676 if (!list_empty(&ptype_all)) 1677 dev_queue_xmit_nit(skb, dev); 1678 1679 if (netif_needs_gso(dev, skb)) { 1680 if (unlikely(dev_gso_segment(skb))) 1681 goto out_kfree_skb; 1682 if (skb->next) 1683 goto gso; 1684 } 1685 1686 rc = ops->ndo_start_xmit(skb, dev); 1687 /* 1688 * TODO: if skb_orphan() was called by 1689 * dev->hard_start_xmit() (for example, the unmodified 1690 * igb driver does that; bnx2 doesn't), then 1691 * skb_tx_software_timestamp() will be unable to send 1692 * back the time stamp. 1693 * 1694 * How can this be prevented? Always create another 1695 * reference to the socket before calling 1696 * dev->hard_start_xmit()? Prevent that skb_orphan() 1697 * does anything in dev->hard_start_xmit() by clearing 1698 * the skb destructor before the call and restoring it 1699 * afterwards, then doing the skb_orphan() ourselves? 1700 */ 1701 return rc; 1702 } 1703 1704 gso: 1705 do { 1706 struct sk_buff *nskb = skb->next; 1707 1708 skb->next = nskb->next; 1709 nskb->next = NULL; 1710 rc = ops->ndo_start_xmit(nskb, dev); 1711 if (unlikely(rc)) { 1712 nskb->next = skb->next; 1713 skb->next = nskb; 1714 return rc; 1715 } 1716 if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) 1717 return NETDEV_TX_BUSY; 1718 } while (skb->next); 1719 1720 skb->destructor = DEV_GSO_CB(skb)->destructor; 1721 1722 out_kfree_skb: 1723 kfree_skb(skb); 1724 return 0; 1725 } 1726 1727 static u32 skb_tx_hashrnd; 1728 1729 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) 1730 { 1731 u32 hash; 1732 1733 if (skb_rx_queue_recorded(skb)) { 1734 hash = skb_get_rx_queue(skb); 1735 } else if (skb->sk && skb->sk->sk_hash) { 1736 hash = skb->sk->sk_hash; 1737 } else 1738 hash = skb->protocol; 1739 1740 hash = jhash_1word(hash, skb_tx_hashrnd); 1741 1742 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); 1743 } 1744 EXPORT_SYMBOL(skb_tx_hash); 1745 1746 static struct netdev_queue *dev_pick_tx(struct net_device *dev, 1747 struct sk_buff *skb) 1748 { 1749 const struct net_device_ops *ops = dev->netdev_ops; 1750 u16 queue_index = 0; 1751 1752 if (ops->ndo_select_queue) 1753 queue_index = ops->ndo_select_queue(dev, skb); 1754 else if (dev->real_num_tx_queues > 1) 1755 queue_index = skb_tx_hash(dev, skb); 1756 1757 skb_set_queue_mapping(skb, queue_index); 1758 return netdev_get_tx_queue(dev, queue_index); 1759 } 1760 1761 /** 1762 * dev_queue_xmit - transmit a buffer 1763 * @skb: buffer to transmit 1764 * 1765 * Queue a buffer for transmission to a network device. The caller must 1766 * have set the device and priority and built the buffer before calling 1767 * this function. The function can be called from an interrupt. 1768 * 1769 * A negative errno code is returned on a failure. A success does not 1770 * guarantee the frame will be transmitted as it may be dropped due 1771 * to congestion or traffic shaping. 1772 * 1773 * ----------------------------------------------------------------------------------- 1774 * I notice this method can also return errors from the queue disciplines, 1775 * including NET_XMIT_DROP, which is a positive value. So, errors can also 1776 * be positive. 1777 * 1778 * Regardless of the return value, the skb is consumed, so it is currently 1779 * difficult to retry a send to this method. (You can bump the ref count 1780 * before sending to hold a reference for retry if you are careful.) 1781 * 1782 * When calling this method, interrupts MUST be enabled. This is because 1783 * the BH enable code must have IRQs enabled so that it will not deadlock. 1784 * --BLG 1785 */ 1786 int dev_queue_xmit(struct sk_buff *skb) 1787 { 1788 struct net_device *dev = skb->dev; 1789 struct netdev_queue *txq; 1790 struct Qdisc *q; 1791 int rc = -ENOMEM; 1792 1793 /* GSO will handle the following emulations directly. */ 1794 if (netif_needs_gso(dev, skb)) 1795 goto gso; 1796 1797 if (skb_shinfo(skb)->frag_list && 1798 !(dev->features & NETIF_F_FRAGLIST) && 1799 __skb_linearize(skb)) 1800 goto out_kfree_skb; 1801 1802 /* Fragmented skb is linearized if device does not support SG, 1803 * or if at least one of fragments is in highmem and device 1804 * does not support DMA from it. 1805 */ 1806 if (skb_shinfo(skb)->nr_frags && 1807 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && 1808 __skb_linearize(skb)) 1809 goto out_kfree_skb; 1810 1811 /* If packet is not checksummed and device does not support 1812 * checksumming for this protocol, complete checksumming here. 1813 */ 1814 if (skb->ip_summed == CHECKSUM_PARTIAL) { 1815 skb_set_transport_header(skb, skb->csum_start - 1816 skb_headroom(skb)); 1817 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) 1818 goto out_kfree_skb; 1819 } 1820 1821 gso: 1822 /* Disable soft irqs for various locks below. Also 1823 * stops preemption for RCU. 1824 */ 1825 rcu_read_lock_bh(); 1826 1827 txq = dev_pick_tx(dev, skb); 1828 q = rcu_dereference(txq->qdisc); 1829 1830 #ifdef CONFIG_NET_CLS_ACT 1831 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); 1832 #endif 1833 if (q->enqueue) { 1834 spinlock_t *root_lock = qdisc_lock(q); 1835 1836 spin_lock(root_lock); 1837 1838 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 1839 kfree_skb(skb); 1840 rc = NET_XMIT_DROP; 1841 } else { 1842 rc = qdisc_enqueue_root(skb, q); 1843 qdisc_run(q); 1844 } 1845 spin_unlock(root_lock); 1846 1847 goto out; 1848 } 1849 1850 /* The device has no queue. Common case for software devices: 1851 loopback, all the sorts of tunnels... 1852 1853 Really, it is unlikely that netif_tx_lock protection is necessary 1854 here. (f.e. loopback and IP tunnels are clean ignoring statistics 1855 counters.) 1856 However, it is possible, that they rely on protection 1857 made by us here. 1858 1859 Check this and shot the lock. It is not prone from deadlocks. 1860 Either shot noqueue qdisc, it is even simpler 8) 1861 */ 1862 if (dev->flags & IFF_UP) { 1863 int cpu = smp_processor_id(); /* ok because BHs are off */ 1864 1865 if (txq->xmit_lock_owner != cpu) { 1866 1867 HARD_TX_LOCK(dev, txq, cpu); 1868 1869 if (!netif_tx_queue_stopped(txq)) { 1870 rc = 0; 1871 if (!dev_hard_start_xmit(skb, dev, txq)) { 1872 HARD_TX_UNLOCK(dev, txq); 1873 goto out; 1874 } 1875 } 1876 HARD_TX_UNLOCK(dev, txq); 1877 if (net_ratelimit()) 1878 printk(KERN_CRIT "Virtual device %s asks to " 1879 "queue packet!\n", dev->name); 1880 } else { 1881 /* Recursion is detected! It is possible, 1882 * unfortunately */ 1883 if (net_ratelimit()) 1884 printk(KERN_CRIT "Dead loop on virtual device " 1885 "%s, fix it urgently!\n", dev->name); 1886 } 1887 } 1888 1889 rc = -ENETDOWN; 1890 rcu_read_unlock_bh(); 1891 1892 out_kfree_skb: 1893 kfree_skb(skb); 1894 return rc; 1895 out: 1896 rcu_read_unlock_bh(); 1897 return rc; 1898 } 1899 1900 1901 /*======================================================================= 1902 Receiver routines 1903 =======================================================================*/ 1904 1905 int netdev_max_backlog __read_mostly = 1000; 1906 int netdev_budget __read_mostly = 300; 1907 int weight_p __read_mostly = 64; /* old backlog weight */ 1908 1909 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; 1910 1911 1912 /** 1913 * netif_rx - post buffer to the network code 1914 * @skb: buffer to post 1915 * 1916 * This function receives a packet from a device driver and queues it for 1917 * the upper (protocol) levels to process. It always succeeds. The buffer 1918 * may be dropped during processing for congestion control or by the 1919 * protocol layers. 1920 * 1921 * return values: 1922 * NET_RX_SUCCESS (no congestion) 1923 * NET_RX_DROP (packet was dropped) 1924 * 1925 */ 1926 1927 int netif_rx(struct sk_buff *skb) 1928 { 1929 struct softnet_data *queue; 1930 unsigned long flags; 1931 1932 /* if netpoll wants it, pretend we never saw it */ 1933 if (netpoll_rx(skb)) 1934 return NET_RX_DROP; 1935 1936 if (!skb->tstamp.tv64) 1937 net_timestamp(skb); 1938 1939 /* 1940 * The code is rearranged so that the path is the most 1941 * short when CPU is congested, but is still operating. 1942 */ 1943 local_irq_save(flags); 1944 queue = &__get_cpu_var(softnet_data); 1945 1946 __get_cpu_var(netdev_rx_stat).total++; 1947 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { 1948 if (queue->input_pkt_queue.qlen) { 1949 enqueue: 1950 __skb_queue_tail(&queue->input_pkt_queue, skb); 1951 local_irq_restore(flags); 1952 return NET_RX_SUCCESS; 1953 } 1954 1955 napi_schedule(&queue->backlog); 1956 goto enqueue; 1957 } 1958 1959 __get_cpu_var(netdev_rx_stat).dropped++; 1960 local_irq_restore(flags); 1961 1962 kfree_skb(skb); 1963 return NET_RX_DROP; 1964 } 1965 1966 int netif_rx_ni(struct sk_buff *skb) 1967 { 1968 int err; 1969 1970 preempt_disable(); 1971 err = netif_rx(skb); 1972 if (local_softirq_pending()) 1973 do_softirq(); 1974 preempt_enable(); 1975 1976 return err; 1977 } 1978 1979 EXPORT_SYMBOL(netif_rx_ni); 1980 1981 static void net_tx_action(struct softirq_action *h) 1982 { 1983 struct softnet_data *sd = &__get_cpu_var(softnet_data); 1984 1985 if (sd->completion_queue) { 1986 struct sk_buff *clist; 1987 1988 local_irq_disable(); 1989 clist = sd->completion_queue; 1990 sd->completion_queue = NULL; 1991 local_irq_enable(); 1992 1993 while (clist) { 1994 struct sk_buff *skb = clist; 1995 clist = clist->next; 1996 1997 WARN_ON(atomic_read(&skb->users)); 1998 __kfree_skb(skb); 1999 } 2000 } 2001 2002 if (sd->output_queue) { 2003 struct Qdisc *head; 2004 2005 local_irq_disable(); 2006 head = sd->output_queue; 2007 sd->output_queue = NULL; 2008 local_irq_enable(); 2009 2010 while (head) { 2011 struct Qdisc *q = head; 2012 spinlock_t *root_lock; 2013 2014 head = head->next_sched; 2015 2016 root_lock = qdisc_lock(q); 2017 if (spin_trylock(root_lock)) { 2018 smp_mb__before_clear_bit(); 2019 clear_bit(__QDISC_STATE_SCHED, 2020 &q->state); 2021 qdisc_run(q); 2022 spin_unlock(root_lock); 2023 } else { 2024 if (!test_bit(__QDISC_STATE_DEACTIVATED, 2025 &q->state)) { 2026 __netif_reschedule(q); 2027 } else { 2028 smp_mb__before_clear_bit(); 2029 clear_bit(__QDISC_STATE_SCHED, 2030 &q->state); 2031 } 2032 } 2033 } 2034 } 2035 } 2036 2037 static inline int deliver_skb(struct sk_buff *skb, 2038 struct packet_type *pt_prev, 2039 struct net_device *orig_dev) 2040 { 2041 atomic_inc(&skb->users); 2042 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2043 } 2044 2045 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) 2046 /* These hooks defined here for ATM */ 2047 struct net_bridge; 2048 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, 2049 unsigned char *addr); 2050 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly; 2051 2052 /* 2053 * If bridge module is loaded call bridging hook. 2054 * returns NULL if packet was consumed. 2055 */ 2056 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, 2057 struct sk_buff *skb) __read_mostly; 2058 static inline struct sk_buff *handle_bridge(struct sk_buff *skb, 2059 struct packet_type **pt_prev, int *ret, 2060 struct net_device *orig_dev) 2061 { 2062 struct net_bridge_port *port; 2063 2064 if (skb->pkt_type == PACKET_LOOPBACK || 2065 (port = rcu_dereference(skb->dev->br_port)) == NULL) 2066 return skb; 2067 2068 if (*pt_prev) { 2069 *ret = deliver_skb(skb, *pt_prev, orig_dev); 2070 *pt_prev = NULL; 2071 } 2072 2073 return br_handle_frame_hook(port, skb); 2074 } 2075 #else 2076 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) 2077 #endif 2078 2079 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) 2080 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; 2081 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); 2082 2083 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, 2084 struct packet_type **pt_prev, 2085 int *ret, 2086 struct net_device *orig_dev) 2087 { 2088 if (skb->dev->macvlan_port == NULL) 2089 return skb; 2090 2091 if (*pt_prev) { 2092 *ret = deliver_skb(skb, *pt_prev, orig_dev); 2093 *pt_prev = NULL; 2094 } 2095 return macvlan_handle_frame_hook(skb); 2096 } 2097 #else 2098 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) 2099 #endif 2100 2101 #ifdef CONFIG_NET_CLS_ACT 2102 /* TODO: Maybe we should just force sch_ingress to be compiled in 2103 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions 2104 * a compare and 2 stores extra right now if we dont have it on 2105 * but have CONFIG_NET_CLS_ACT 2106 * NOTE: This doesnt stop any functionality; if you dont have 2107 * the ingress scheduler, you just cant add policies on ingress. 2108 * 2109 */ 2110 static int ing_filter(struct sk_buff *skb) 2111 { 2112 struct net_device *dev = skb->dev; 2113 u32 ttl = G_TC_RTTL(skb->tc_verd); 2114 struct netdev_queue *rxq; 2115 int result = TC_ACT_OK; 2116 struct Qdisc *q; 2117 2118 if (MAX_RED_LOOP < ttl++) { 2119 printk(KERN_WARNING 2120 "Redir loop detected Dropping packet (%d->%d)\n", 2121 skb->iif, dev->ifindex); 2122 return TC_ACT_SHOT; 2123 } 2124 2125 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); 2126 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); 2127 2128 rxq = &dev->rx_queue; 2129 2130 q = rxq->qdisc; 2131 if (q != &noop_qdisc) { 2132 spin_lock(qdisc_lock(q)); 2133 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) 2134 result = qdisc_enqueue_root(skb, q); 2135 spin_unlock(qdisc_lock(q)); 2136 } 2137 2138 return result; 2139 } 2140 2141 static inline struct sk_buff *handle_ing(struct sk_buff *skb, 2142 struct packet_type **pt_prev, 2143 int *ret, struct net_device *orig_dev) 2144 { 2145 if (skb->dev->rx_queue.qdisc == &noop_qdisc) 2146 goto out; 2147 2148 if (*pt_prev) { 2149 *ret = deliver_skb(skb, *pt_prev, orig_dev); 2150 *pt_prev = NULL; 2151 } else { 2152 /* Huh? Why does turning on AF_PACKET affect this? */ 2153 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); 2154 } 2155 2156 switch (ing_filter(skb)) { 2157 case TC_ACT_SHOT: 2158 case TC_ACT_STOLEN: 2159 kfree_skb(skb); 2160 return NULL; 2161 } 2162 2163 out: 2164 skb->tc_verd = 0; 2165 return skb; 2166 } 2167 #endif 2168 2169 /* 2170 * netif_nit_deliver - deliver received packets to network taps 2171 * @skb: buffer 2172 * 2173 * This function is used to deliver incoming packets to network 2174 * taps. It should be used when the normal netif_receive_skb path 2175 * is bypassed, for example because of VLAN acceleration. 2176 */ 2177 void netif_nit_deliver(struct sk_buff *skb) 2178 { 2179 struct packet_type *ptype; 2180 2181 if (list_empty(&ptype_all)) 2182 return; 2183 2184 skb_reset_network_header(skb); 2185 skb_reset_transport_header(skb); 2186 skb->mac_len = skb->network_header - skb->mac_header; 2187 2188 rcu_read_lock(); 2189 list_for_each_entry_rcu(ptype, &ptype_all, list) { 2190 if (!ptype->dev || ptype->dev == skb->dev) 2191 deliver_skb(skb, ptype, skb->dev); 2192 } 2193 rcu_read_unlock(); 2194 } 2195 2196 /** 2197 * netif_receive_skb - process receive buffer from network 2198 * @skb: buffer to process 2199 * 2200 * netif_receive_skb() is the main receive data processing function. 2201 * It always succeeds. The buffer may be dropped during processing 2202 * for congestion control or by the protocol layers. 2203 * 2204 * This function may only be called from softirq context and interrupts 2205 * should be enabled. 2206 * 2207 * Return values (usually ignored): 2208 * NET_RX_SUCCESS: no congestion 2209 * NET_RX_DROP: packet was dropped 2210 */ 2211 int netif_receive_skb(struct sk_buff *skb) 2212 { 2213 struct packet_type *ptype, *pt_prev; 2214 struct net_device *orig_dev; 2215 struct net_device *null_or_orig; 2216 int ret = NET_RX_DROP; 2217 __be16 type; 2218 2219 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb)) 2220 return NET_RX_SUCCESS; 2221 2222 /* if we've gotten here through NAPI, check netpoll */ 2223 if (netpoll_receive_skb(skb)) 2224 return NET_RX_DROP; 2225 2226 if (!skb->tstamp.tv64) 2227 net_timestamp(skb); 2228 2229 if (!skb->iif) 2230 skb->iif = skb->dev->ifindex; 2231 2232 null_or_orig = NULL; 2233 orig_dev = skb->dev; 2234 if (orig_dev->master) { 2235 if (skb_bond_should_drop(skb)) 2236 null_or_orig = orig_dev; /* deliver only exact match */ 2237 else 2238 skb->dev = orig_dev->master; 2239 } 2240 2241 __get_cpu_var(netdev_rx_stat).total++; 2242 2243 skb_reset_network_header(skb); 2244 skb_reset_transport_header(skb); 2245 skb->mac_len = skb->network_header - skb->mac_header; 2246 2247 pt_prev = NULL; 2248 2249 rcu_read_lock(); 2250 2251 #ifdef CONFIG_NET_CLS_ACT 2252 if (skb->tc_verd & TC_NCLS) { 2253 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); 2254 goto ncls; 2255 } 2256 #endif 2257 2258 list_for_each_entry_rcu(ptype, &ptype_all, list) { 2259 if (ptype->dev == null_or_orig || ptype->dev == skb->dev || 2260 ptype->dev == orig_dev) { 2261 if (pt_prev) 2262 ret = deliver_skb(skb, pt_prev, orig_dev); 2263 pt_prev = ptype; 2264 } 2265 } 2266 2267 #ifdef CONFIG_NET_CLS_ACT 2268 skb = handle_ing(skb, &pt_prev, &ret, orig_dev); 2269 if (!skb) 2270 goto out; 2271 ncls: 2272 #endif 2273 2274 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); 2275 if (!skb) 2276 goto out; 2277 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); 2278 if (!skb) 2279 goto out; 2280 2281 skb_orphan(skb); 2282 2283 type = skb->protocol; 2284 list_for_each_entry_rcu(ptype, 2285 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { 2286 if (ptype->type == type && 2287 (ptype->dev == null_or_orig || ptype->dev == skb->dev || 2288 ptype->dev == orig_dev)) { 2289 if (pt_prev) 2290 ret = deliver_skb(skb, pt_prev, orig_dev); 2291 pt_prev = ptype; 2292 } 2293 } 2294 2295 if (pt_prev) { 2296 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2297 } else { 2298 kfree_skb(skb); 2299 /* Jamal, now you will not able to escape explaining 2300 * me how you were going to use this. :-) 2301 */ 2302 ret = NET_RX_DROP; 2303 } 2304 2305 out: 2306 rcu_read_unlock(); 2307 return ret; 2308 } 2309 2310 /* Network device is going away, flush any packets still pending */ 2311 static void flush_backlog(void *arg) 2312 { 2313 struct net_device *dev = arg; 2314 struct softnet_data *queue = &__get_cpu_var(softnet_data); 2315 struct sk_buff *skb, *tmp; 2316 2317 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) 2318 if (skb->dev == dev) { 2319 __skb_unlink(skb, &queue->input_pkt_queue); 2320 kfree_skb(skb); 2321 } 2322 } 2323 2324 static int napi_gro_complete(struct sk_buff *skb) 2325 { 2326 struct packet_type *ptype; 2327 __be16 type = skb->protocol; 2328 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; 2329 int err = -ENOENT; 2330 2331 if (NAPI_GRO_CB(skb)->count == 1) 2332 goto out; 2333 2334 rcu_read_lock(); 2335 list_for_each_entry_rcu(ptype, head, list) { 2336 if (ptype->type != type || ptype->dev || !ptype->gro_complete) 2337 continue; 2338 2339 err = ptype->gro_complete(skb); 2340 break; 2341 } 2342 rcu_read_unlock(); 2343 2344 if (err) { 2345 WARN_ON(&ptype->list == head); 2346 kfree_skb(skb); 2347 return NET_RX_SUCCESS; 2348 } 2349 2350 out: 2351 skb_shinfo(skb)->gso_size = 0; 2352 return netif_receive_skb(skb); 2353 } 2354 2355 void napi_gro_flush(struct napi_struct *napi) 2356 { 2357 struct sk_buff *skb, *next; 2358 2359 for (skb = napi->gro_list; skb; skb = next) { 2360 next = skb->next; 2361 skb->next = NULL; 2362 napi_gro_complete(skb); 2363 } 2364 2365 napi->gro_count = 0; 2366 napi->gro_list = NULL; 2367 } 2368 EXPORT_SYMBOL(napi_gro_flush); 2369 2370 void *skb_gro_header(struct sk_buff *skb, unsigned int hlen) 2371 { 2372 unsigned int offset = skb_gro_offset(skb); 2373 2374 hlen += offset; 2375 if (hlen <= skb_headlen(skb)) 2376 return skb->data + offset; 2377 2378 if (unlikely(!skb_shinfo(skb)->nr_frags || 2379 skb_shinfo(skb)->frags[0].size <= 2380 hlen - skb_headlen(skb) || 2381 PageHighMem(skb_shinfo(skb)->frags[0].page))) 2382 return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL; 2383 2384 return page_address(skb_shinfo(skb)->frags[0].page) + 2385 skb_shinfo(skb)->frags[0].page_offset + 2386 offset - skb_headlen(skb); 2387 } 2388 EXPORT_SYMBOL(skb_gro_header); 2389 2390 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 2391 { 2392 struct sk_buff **pp = NULL; 2393 struct packet_type *ptype; 2394 __be16 type = skb->protocol; 2395 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; 2396 int same_flow; 2397 int mac_len; 2398 int ret; 2399 2400 if (!(skb->dev->features & NETIF_F_GRO)) 2401 goto normal; 2402 2403 if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list) 2404 goto normal; 2405 2406 rcu_read_lock(); 2407 list_for_each_entry_rcu(ptype, head, list) { 2408 if (ptype->type != type || ptype->dev || !ptype->gro_receive) 2409 continue; 2410 2411 skb_set_network_header(skb, skb_gro_offset(skb)); 2412 mac_len = skb->network_header - skb->mac_header; 2413 skb->mac_len = mac_len; 2414 NAPI_GRO_CB(skb)->same_flow = 0; 2415 NAPI_GRO_CB(skb)->flush = 0; 2416 NAPI_GRO_CB(skb)->free = 0; 2417 2418 pp = ptype->gro_receive(&napi->gro_list, skb); 2419 break; 2420 } 2421 rcu_read_unlock(); 2422 2423 if (&ptype->list == head) 2424 goto normal; 2425 2426 same_flow = NAPI_GRO_CB(skb)->same_flow; 2427 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 2428 2429 if (pp) { 2430 struct sk_buff *nskb = *pp; 2431 2432 *pp = nskb->next; 2433 nskb->next = NULL; 2434 napi_gro_complete(nskb); 2435 napi->gro_count--; 2436 } 2437 2438 if (same_flow) 2439 goto ok; 2440 2441 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) 2442 goto normal; 2443 2444 napi->gro_count++; 2445 NAPI_GRO_CB(skb)->count = 1; 2446 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 2447 skb->next = napi->gro_list; 2448 napi->gro_list = skb; 2449 ret = GRO_HELD; 2450 2451 pull: 2452 if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) { 2453 if (napi->gro_list == skb) 2454 napi->gro_list = skb->next; 2455 ret = GRO_DROP; 2456 } 2457 2458 ok: 2459 return ret; 2460 2461 normal: 2462 ret = GRO_NORMAL; 2463 goto pull; 2464 } 2465 EXPORT_SYMBOL(dev_gro_receive); 2466 2467 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 2468 { 2469 struct sk_buff *p; 2470 2471 if (netpoll_rx_on(skb)) 2472 return GRO_NORMAL; 2473 2474 for (p = napi->gro_list; p; p = p->next) { 2475 NAPI_GRO_CB(p)->same_flow = !compare_ether_header( 2476 skb_mac_header(p), skb_gro_mac_header(skb)); 2477 NAPI_GRO_CB(p)->flush = 0; 2478 } 2479 2480 return dev_gro_receive(napi, skb); 2481 } 2482 2483 int napi_skb_finish(int ret, struct sk_buff *skb) 2484 { 2485 int err = NET_RX_SUCCESS; 2486 2487 switch (ret) { 2488 case GRO_NORMAL: 2489 return netif_receive_skb(skb); 2490 2491 case GRO_DROP: 2492 err = NET_RX_DROP; 2493 /* fall through */ 2494 2495 case GRO_MERGED_FREE: 2496 kfree_skb(skb); 2497 break; 2498 } 2499 2500 return err; 2501 } 2502 EXPORT_SYMBOL(napi_skb_finish); 2503 2504 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 2505 { 2506 skb_gro_reset_offset(skb); 2507 2508 return napi_skb_finish(__napi_gro_receive(napi, skb), skb); 2509 } 2510 EXPORT_SYMBOL(napi_gro_receive); 2511 2512 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 2513 { 2514 __skb_pull(skb, skb_headlen(skb)); 2515 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); 2516 2517 napi->skb = skb; 2518 } 2519 EXPORT_SYMBOL(napi_reuse_skb); 2520 2521 struct sk_buff *napi_fraginfo_skb(struct napi_struct *napi, 2522 struct napi_gro_fraginfo *info) 2523 { 2524 struct net_device *dev = napi->dev; 2525 struct sk_buff *skb = napi->skb; 2526 struct ethhdr *eth; 2527 skb_frag_t *frag; 2528 int i; 2529 2530 napi->skb = NULL; 2531 2532 if (!skb) { 2533 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN); 2534 if (!skb) 2535 goto out; 2536 2537 skb_reserve(skb, NET_IP_ALIGN); 2538 } 2539 2540 BUG_ON(info->nr_frags > MAX_SKB_FRAGS); 2541 frag = &info->frags[info->nr_frags - 1]; 2542 2543 for (i = skb_shinfo(skb)->nr_frags; i < info->nr_frags; i++) { 2544 skb_fill_page_desc(skb, i, frag->page, frag->page_offset, 2545 frag->size); 2546 frag++; 2547 } 2548 skb_shinfo(skb)->nr_frags = info->nr_frags; 2549 2550 skb->data_len = info->len; 2551 skb->len += info->len; 2552 skb->truesize += info->len; 2553 2554 skb_reset_mac_header(skb); 2555 skb_gro_reset_offset(skb); 2556 2557 eth = skb_gro_header(skb, sizeof(*eth)); 2558 if (!eth) { 2559 napi_reuse_skb(napi, skb); 2560 skb = NULL; 2561 goto out; 2562 } 2563 2564 skb_gro_pull(skb, sizeof(*eth)); 2565 2566 /* 2567 * This works because the only protocols we care about don't require 2568 * special handling. We'll fix it up properly at the end. 2569 */ 2570 skb->protocol = eth->h_proto; 2571 2572 skb->ip_summed = info->ip_summed; 2573 skb->csum = info->csum; 2574 2575 out: 2576 return skb; 2577 } 2578 EXPORT_SYMBOL(napi_fraginfo_skb); 2579 2580 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret) 2581 { 2582 int err = NET_RX_SUCCESS; 2583 2584 switch (ret) { 2585 case GRO_NORMAL: 2586 case GRO_HELD: 2587 skb->protocol = eth_type_trans(skb, napi->dev); 2588 2589 if (ret == GRO_NORMAL) 2590 return netif_receive_skb(skb); 2591 2592 skb_gro_pull(skb, -ETH_HLEN); 2593 break; 2594 2595 case GRO_DROP: 2596 err = NET_RX_DROP; 2597 /* fall through */ 2598 2599 case GRO_MERGED_FREE: 2600 napi_reuse_skb(napi, skb); 2601 break; 2602 } 2603 2604 return err; 2605 } 2606 EXPORT_SYMBOL(napi_frags_finish); 2607 2608 int napi_gro_frags(struct napi_struct *napi, struct napi_gro_fraginfo *info) 2609 { 2610 struct sk_buff *skb = napi_fraginfo_skb(napi, info); 2611 2612 if (!skb) 2613 return NET_RX_DROP; 2614 2615 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); 2616 } 2617 EXPORT_SYMBOL(napi_gro_frags); 2618 2619 static int process_backlog(struct napi_struct *napi, int quota) 2620 { 2621 int work = 0; 2622 struct softnet_data *queue = &__get_cpu_var(softnet_data); 2623 unsigned long start_time = jiffies; 2624 2625 napi->weight = weight_p; 2626 do { 2627 struct sk_buff *skb; 2628 2629 local_irq_disable(); 2630 skb = __skb_dequeue(&queue->input_pkt_queue); 2631 if (!skb) { 2632 __napi_complete(napi); 2633 local_irq_enable(); 2634 break; 2635 } 2636 local_irq_enable(); 2637 2638 netif_receive_skb(skb); 2639 } while (++work < quota && jiffies == start_time); 2640 2641 return work; 2642 } 2643 2644 /** 2645 * __napi_schedule - schedule for receive 2646 * @n: entry to schedule 2647 * 2648 * The entry's receive function will be scheduled to run 2649 */ 2650 void __napi_schedule(struct napi_struct *n) 2651 { 2652 unsigned long flags; 2653 2654 local_irq_save(flags); 2655 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); 2656 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 2657 local_irq_restore(flags); 2658 } 2659 EXPORT_SYMBOL(__napi_schedule); 2660 2661 void __napi_complete(struct napi_struct *n) 2662 { 2663 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 2664 BUG_ON(n->gro_list); 2665 2666 list_del(&n->poll_list); 2667 smp_mb__before_clear_bit(); 2668 clear_bit(NAPI_STATE_SCHED, &n->state); 2669 } 2670 EXPORT_SYMBOL(__napi_complete); 2671 2672 void napi_complete(struct napi_struct *n) 2673 { 2674 unsigned long flags; 2675 2676 /* 2677 * don't let napi dequeue from the cpu poll list 2678 * just in case its running on a different cpu 2679 */ 2680 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) 2681 return; 2682 2683 napi_gro_flush(n); 2684 local_irq_save(flags); 2685 __napi_complete(n); 2686 local_irq_restore(flags); 2687 } 2688 EXPORT_SYMBOL(napi_complete); 2689 2690 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 2691 int (*poll)(struct napi_struct *, int), int weight) 2692 { 2693 INIT_LIST_HEAD(&napi->poll_list); 2694 napi->gro_count = 0; 2695 napi->gro_list = NULL; 2696 napi->skb = NULL; 2697 napi->poll = poll; 2698 napi->weight = weight; 2699 list_add(&napi->dev_list, &dev->napi_list); 2700 napi->dev = dev; 2701 #ifdef CONFIG_NETPOLL 2702 spin_lock_init(&napi->poll_lock); 2703 napi->poll_owner = -1; 2704 #endif 2705 set_bit(NAPI_STATE_SCHED, &napi->state); 2706 } 2707 EXPORT_SYMBOL(netif_napi_add); 2708 2709 void netif_napi_del(struct napi_struct *napi) 2710 { 2711 struct sk_buff *skb, *next; 2712 2713 list_del_init(&napi->dev_list); 2714 kfree_skb(napi->skb); 2715 2716 for (skb = napi->gro_list; skb; skb = next) { 2717 next = skb->next; 2718 skb->next = NULL; 2719 kfree_skb(skb); 2720 } 2721 2722 napi->gro_list = NULL; 2723 napi->gro_count = 0; 2724 } 2725 EXPORT_SYMBOL(netif_napi_del); 2726 2727 2728 static void net_rx_action(struct softirq_action *h) 2729 { 2730 struct list_head *list = &__get_cpu_var(softnet_data).poll_list; 2731 unsigned long time_limit = jiffies + 2; 2732 int budget = netdev_budget; 2733 void *have; 2734 2735 local_irq_disable(); 2736 2737 while (!list_empty(list)) { 2738 struct napi_struct *n; 2739 int work, weight; 2740 2741 /* If softirq window is exhuasted then punt. 2742 * Allow this to run for 2 jiffies since which will allow 2743 * an average latency of 1.5/HZ. 2744 */ 2745 if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) 2746 goto softnet_break; 2747 2748 local_irq_enable(); 2749 2750 /* Even though interrupts have been re-enabled, this 2751 * access is safe because interrupts can only add new 2752 * entries to the tail of this list, and only ->poll() 2753 * calls can remove this head entry from the list. 2754 */ 2755 n = list_entry(list->next, struct napi_struct, poll_list); 2756 2757 have = netpoll_poll_lock(n); 2758 2759 weight = n->weight; 2760 2761 /* This NAPI_STATE_SCHED test is for avoiding a race 2762 * with netpoll's poll_napi(). Only the entity which 2763 * obtains the lock and sees NAPI_STATE_SCHED set will 2764 * actually make the ->poll() call. Therefore we avoid 2765 * accidently calling ->poll() when NAPI is not scheduled. 2766 */ 2767 work = 0; 2768 if (test_bit(NAPI_STATE_SCHED, &n->state)) 2769 work = n->poll(n, weight); 2770 2771 WARN_ON_ONCE(work > weight); 2772 2773 budget -= work; 2774 2775 local_irq_disable(); 2776 2777 /* Drivers must not modify the NAPI state if they 2778 * consume the entire weight. In such cases this code 2779 * still "owns" the NAPI instance and therefore can 2780 * move the instance around on the list at-will. 2781 */ 2782 if (unlikely(work == weight)) { 2783 if (unlikely(napi_disable_pending(n))) 2784 __napi_complete(n); 2785 else 2786 list_move_tail(&n->poll_list, list); 2787 } 2788 2789 netpoll_poll_unlock(have); 2790 } 2791 out: 2792 local_irq_enable(); 2793 2794 #ifdef CONFIG_NET_DMA 2795 /* 2796 * There may not be any more sk_buffs coming right now, so push 2797 * any pending DMA copies to hardware 2798 */ 2799 dma_issue_pending_all(); 2800 #endif 2801 2802 return; 2803 2804 softnet_break: 2805 __get_cpu_var(netdev_rx_stat).time_squeeze++; 2806 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 2807 goto out; 2808 } 2809 2810 static gifconf_func_t * gifconf_list [NPROTO]; 2811 2812 /** 2813 * register_gifconf - register a SIOCGIF handler 2814 * @family: Address family 2815 * @gifconf: Function handler 2816 * 2817 * Register protocol dependent address dumping routines. The handler 2818 * that is passed must not be freed or reused until it has been replaced 2819 * by another handler. 2820 */ 2821 int register_gifconf(unsigned int family, gifconf_func_t * gifconf) 2822 { 2823 if (family >= NPROTO) 2824 return -EINVAL; 2825 gifconf_list[family] = gifconf; 2826 return 0; 2827 } 2828 2829 2830 /* 2831 * Map an interface index to its name (SIOCGIFNAME) 2832 */ 2833 2834 /* 2835 * We need this ioctl for efficient implementation of the 2836 * if_indextoname() function required by the IPv6 API. Without 2837 * it, we would have to search all the interfaces to find a 2838 * match. --pb 2839 */ 2840 2841 static int dev_ifname(struct net *net, struct ifreq __user *arg) 2842 { 2843 struct net_device *dev; 2844 struct ifreq ifr; 2845 2846 /* 2847 * Fetch the caller's info block. 2848 */ 2849 2850 if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) 2851 return -EFAULT; 2852 2853 read_lock(&dev_base_lock); 2854 dev = __dev_get_by_index(net, ifr.ifr_ifindex); 2855 if (!dev) { 2856 read_unlock(&dev_base_lock); 2857 return -ENODEV; 2858 } 2859 2860 strcpy(ifr.ifr_name, dev->name); 2861 read_unlock(&dev_base_lock); 2862 2863 if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) 2864 return -EFAULT; 2865 return 0; 2866 } 2867 2868 /* 2869 * Perform a SIOCGIFCONF call. This structure will change 2870 * size eventually, and there is nothing I can do about it. 2871 * Thus we will need a 'compatibility mode'. 2872 */ 2873 2874 static int dev_ifconf(struct net *net, char __user *arg) 2875 { 2876 struct ifconf ifc; 2877 struct net_device *dev; 2878 char __user *pos; 2879 int len; 2880 int total; 2881 int i; 2882 2883 /* 2884 * Fetch the caller's info block. 2885 */ 2886 2887 if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) 2888 return -EFAULT; 2889 2890 pos = ifc.ifc_buf; 2891 len = ifc.ifc_len; 2892 2893 /* 2894 * Loop over the interfaces, and write an info block for each. 2895 */ 2896 2897 total = 0; 2898 for_each_netdev(net, dev) { 2899 for (i = 0; i < NPROTO; i++) { 2900 if (gifconf_list[i]) { 2901 int done; 2902 if (!pos) 2903 done = gifconf_list[i](dev, NULL, 0); 2904 else 2905 done = gifconf_list[i](dev, pos + total, 2906 len - total); 2907 if (done < 0) 2908 return -EFAULT; 2909 total += done; 2910 } 2911 } 2912 } 2913 2914 /* 2915 * All done. Write the updated control block back to the caller. 2916 */ 2917 ifc.ifc_len = total; 2918 2919 /* 2920 * Both BSD and Solaris return 0 here, so we do too. 2921 */ 2922 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; 2923 } 2924 2925 #ifdef CONFIG_PROC_FS 2926 /* 2927 * This is invoked by the /proc filesystem handler to display a device 2928 * in detail. 2929 */ 2930 void *dev_seq_start(struct seq_file *seq, loff_t *pos) 2931 __acquires(dev_base_lock) 2932 { 2933 struct net *net = seq_file_net(seq); 2934 loff_t off; 2935 struct net_device *dev; 2936 2937 read_lock(&dev_base_lock); 2938 if (!*pos) 2939 return SEQ_START_TOKEN; 2940 2941 off = 1; 2942 for_each_netdev(net, dev) 2943 if (off++ == *pos) 2944 return dev; 2945 2946 return NULL; 2947 } 2948 2949 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2950 { 2951 struct net *net = seq_file_net(seq); 2952 ++*pos; 2953 return v == SEQ_START_TOKEN ? 2954 first_net_device(net) : next_net_device((struct net_device *)v); 2955 } 2956 2957 void dev_seq_stop(struct seq_file *seq, void *v) 2958 __releases(dev_base_lock) 2959 { 2960 read_unlock(&dev_base_lock); 2961 } 2962 2963 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) 2964 { 2965 const struct net_device_stats *stats = dev_get_stats(dev); 2966 2967 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " 2968 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", 2969 dev->name, stats->rx_bytes, stats->rx_packets, 2970 stats->rx_errors, 2971 stats->rx_dropped + stats->rx_missed_errors, 2972 stats->rx_fifo_errors, 2973 stats->rx_length_errors + stats->rx_over_errors + 2974 stats->rx_crc_errors + stats->rx_frame_errors, 2975 stats->rx_compressed, stats->multicast, 2976 stats->tx_bytes, stats->tx_packets, 2977 stats->tx_errors, stats->tx_dropped, 2978 stats->tx_fifo_errors, stats->collisions, 2979 stats->tx_carrier_errors + 2980 stats->tx_aborted_errors + 2981 stats->tx_window_errors + 2982 stats->tx_heartbeat_errors, 2983 stats->tx_compressed); 2984 } 2985 2986 /* 2987 * Called from the PROCfs module. This now uses the new arbitrary sized 2988 * /proc/net interface to create /proc/net/dev 2989 */ 2990 static int dev_seq_show(struct seq_file *seq, void *v) 2991 { 2992 if (v == SEQ_START_TOKEN) 2993 seq_puts(seq, "Inter-| Receive " 2994 " | Transmit\n" 2995 " face |bytes packets errs drop fifo frame " 2996 "compressed multicast|bytes packets errs " 2997 "drop fifo colls carrier compressed\n"); 2998 else 2999 dev_seq_printf_stats(seq, v); 3000 return 0; 3001 } 3002 3003 static struct netif_rx_stats *softnet_get_online(loff_t *pos) 3004 { 3005 struct netif_rx_stats *rc = NULL; 3006 3007 while (*pos < nr_cpu_ids) 3008 if (cpu_online(*pos)) { 3009 rc = &per_cpu(netdev_rx_stat, *pos); 3010 break; 3011 } else 3012 ++*pos; 3013 return rc; 3014 } 3015 3016 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) 3017 { 3018 return softnet_get_online(pos); 3019 } 3020 3021 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3022 { 3023 ++*pos; 3024 return softnet_get_online(pos); 3025 } 3026 3027 static void softnet_seq_stop(struct seq_file *seq, void *v) 3028 { 3029 } 3030 3031 static int softnet_seq_show(struct seq_file *seq, void *v) 3032 { 3033 struct netif_rx_stats *s = v; 3034 3035 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", 3036 s->total, s->dropped, s->time_squeeze, 0, 3037 0, 0, 0, 0, /* was fastroute */ 3038 s->cpu_collision ); 3039 return 0; 3040 } 3041 3042 static const struct seq_operations dev_seq_ops = { 3043 .start = dev_seq_start, 3044 .next = dev_seq_next, 3045 .stop = dev_seq_stop, 3046 .show = dev_seq_show, 3047 }; 3048 3049 static int dev_seq_open(struct inode *inode, struct file *file) 3050 { 3051 return seq_open_net(inode, file, &dev_seq_ops, 3052 sizeof(struct seq_net_private)); 3053 } 3054 3055 static const struct file_operations dev_seq_fops = { 3056 .owner = THIS_MODULE, 3057 .open = dev_seq_open, 3058 .read = seq_read, 3059 .llseek = seq_lseek, 3060 .release = seq_release_net, 3061 }; 3062 3063 static const struct seq_operations softnet_seq_ops = { 3064 .start = softnet_seq_start, 3065 .next = softnet_seq_next, 3066 .stop = softnet_seq_stop, 3067 .show = softnet_seq_show, 3068 }; 3069 3070 static int softnet_seq_open(struct inode *inode, struct file *file) 3071 { 3072 return seq_open(file, &softnet_seq_ops); 3073 } 3074 3075 static const struct file_operations softnet_seq_fops = { 3076 .owner = THIS_MODULE, 3077 .open = softnet_seq_open, 3078 .read = seq_read, 3079 .llseek = seq_lseek, 3080 .release = seq_release, 3081 }; 3082 3083 static void *ptype_get_idx(loff_t pos) 3084 { 3085 struct packet_type *pt = NULL; 3086 loff_t i = 0; 3087 int t; 3088 3089 list_for_each_entry_rcu(pt, &ptype_all, list) { 3090 if (i == pos) 3091 return pt; 3092 ++i; 3093 } 3094 3095 for (t = 0; t < PTYPE_HASH_SIZE; t++) { 3096 list_for_each_entry_rcu(pt, &ptype_base[t], list) { 3097 if (i == pos) 3098 return pt; 3099 ++i; 3100 } 3101 } 3102 return NULL; 3103 } 3104 3105 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) 3106 __acquires(RCU) 3107 { 3108 rcu_read_lock(); 3109 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; 3110 } 3111 3112 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3113 { 3114 struct packet_type *pt; 3115 struct list_head *nxt; 3116 int hash; 3117 3118 ++*pos; 3119 if (v == SEQ_START_TOKEN) 3120 return ptype_get_idx(0); 3121 3122 pt = v; 3123 nxt = pt->list.next; 3124 if (pt->type == htons(ETH_P_ALL)) { 3125 if (nxt != &ptype_all) 3126 goto found; 3127 hash = 0; 3128 nxt = ptype_base[0].next; 3129 } else 3130 hash = ntohs(pt->type) & PTYPE_HASH_MASK; 3131 3132 while (nxt == &ptype_base[hash]) { 3133 if (++hash >= PTYPE_HASH_SIZE) 3134 return NULL; 3135 nxt = ptype_base[hash].next; 3136 } 3137 found: 3138 return list_entry(nxt, struct packet_type, list); 3139 } 3140 3141 static void ptype_seq_stop(struct seq_file *seq, void *v) 3142 __releases(RCU) 3143 { 3144 rcu_read_unlock(); 3145 } 3146 3147 static int ptype_seq_show(struct seq_file *seq, void *v) 3148 { 3149 struct packet_type *pt = v; 3150 3151 if (v == SEQ_START_TOKEN) 3152 seq_puts(seq, "Type Device Function\n"); 3153 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { 3154 if (pt->type == htons(ETH_P_ALL)) 3155 seq_puts(seq, "ALL "); 3156 else 3157 seq_printf(seq, "%04x", ntohs(pt->type)); 3158 3159 seq_printf(seq, " %-8s %pF\n", 3160 pt->dev ? pt->dev->name : "", pt->func); 3161 } 3162 3163 return 0; 3164 } 3165 3166 static const struct seq_operations ptype_seq_ops = { 3167 .start = ptype_seq_start, 3168 .next = ptype_seq_next, 3169 .stop = ptype_seq_stop, 3170 .show = ptype_seq_show, 3171 }; 3172 3173 static int ptype_seq_open(struct inode *inode, struct file *file) 3174 { 3175 return seq_open_net(inode, file, &ptype_seq_ops, 3176 sizeof(struct seq_net_private)); 3177 } 3178 3179 static const struct file_operations ptype_seq_fops = { 3180 .owner = THIS_MODULE, 3181 .open = ptype_seq_open, 3182 .read = seq_read, 3183 .llseek = seq_lseek, 3184 .release = seq_release_net, 3185 }; 3186 3187 3188 static int __net_init dev_proc_net_init(struct net *net) 3189 { 3190 int rc = -ENOMEM; 3191 3192 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) 3193 goto out; 3194 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) 3195 goto out_dev; 3196 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) 3197 goto out_softnet; 3198 3199 if (wext_proc_init(net)) 3200 goto out_ptype; 3201 rc = 0; 3202 out: 3203 return rc; 3204 out_ptype: 3205 proc_net_remove(net, "ptype"); 3206 out_softnet: 3207 proc_net_remove(net, "softnet_stat"); 3208 out_dev: 3209 proc_net_remove(net, "dev"); 3210 goto out; 3211 } 3212 3213 static void __net_exit dev_proc_net_exit(struct net *net) 3214 { 3215 wext_proc_exit(net); 3216 3217 proc_net_remove(net, "ptype"); 3218 proc_net_remove(net, "softnet_stat"); 3219 proc_net_remove(net, "dev"); 3220 } 3221 3222 static struct pernet_operations __net_initdata dev_proc_ops = { 3223 .init = dev_proc_net_init, 3224 .exit = dev_proc_net_exit, 3225 }; 3226 3227 static int __init dev_proc_init(void) 3228 { 3229 return register_pernet_subsys(&dev_proc_ops); 3230 } 3231 #else 3232 #define dev_proc_init() 0 3233 #endif /* CONFIG_PROC_FS */ 3234 3235 3236 /** 3237 * netdev_set_master - set up master/slave pair 3238 * @slave: slave device 3239 * @master: new master device 3240 * 3241 * Changes the master device of the slave. Pass %NULL to break the 3242 * bonding. The caller must hold the RTNL semaphore. On a failure 3243 * a negative errno code is returned. On success the reference counts 3244 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the 3245 * function returns zero. 3246 */ 3247 int netdev_set_master(struct net_device *slave, struct net_device *master) 3248 { 3249 struct net_device *old = slave->master; 3250 3251 ASSERT_RTNL(); 3252 3253 if (master) { 3254 if (old) 3255 return -EBUSY; 3256 dev_hold(master); 3257 } 3258 3259 slave->master = master; 3260 3261 synchronize_net(); 3262 3263 if (old) 3264 dev_put(old); 3265 3266 if (master) 3267 slave->flags |= IFF_SLAVE; 3268 else 3269 slave->flags &= ~IFF_SLAVE; 3270 3271 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); 3272 return 0; 3273 } 3274 3275 static void dev_change_rx_flags(struct net_device *dev, int flags) 3276 { 3277 const struct net_device_ops *ops = dev->netdev_ops; 3278 3279 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) 3280 ops->ndo_change_rx_flags(dev, flags); 3281 } 3282 3283 static int __dev_set_promiscuity(struct net_device *dev, int inc) 3284 { 3285 unsigned short old_flags = dev->flags; 3286 uid_t uid; 3287 gid_t gid; 3288 3289 ASSERT_RTNL(); 3290 3291 dev->flags |= IFF_PROMISC; 3292 dev->promiscuity += inc; 3293 if (dev->promiscuity == 0) { 3294 /* 3295 * Avoid overflow. 3296 * If inc causes overflow, untouch promisc and return error. 3297 */ 3298 if (inc < 0) 3299 dev->flags &= ~IFF_PROMISC; 3300 else { 3301 dev->promiscuity -= inc; 3302 printk(KERN_WARNING "%s: promiscuity touches roof, " 3303 "set promiscuity failed, promiscuity feature " 3304 "of device might be broken.\n", dev->name); 3305 return -EOVERFLOW; 3306 } 3307 } 3308 if (dev->flags != old_flags) { 3309 printk(KERN_INFO "device %s %s promiscuous mode\n", 3310 dev->name, (dev->flags & IFF_PROMISC) ? "entered" : 3311 "left"); 3312 if (audit_enabled) { 3313 current_uid_gid(&uid, &gid); 3314 audit_log(current->audit_context, GFP_ATOMIC, 3315 AUDIT_ANOM_PROMISCUOUS, 3316 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 3317 dev->name, (dev->flags & IFF_PROMISC), 3318 (old_flags & IFF_PROMISC), 3319 audit_get_loginuid(current), 3320 uid, gid, 3321 audit_get_sessionid(current)); 3322 } 3323 3324 dev_change_rx_flags(dev, IFF_PROMISC); 3325 } 3326 return 0; 3327 } 3328 3329 /** 3330 * dev_set_promiscuity - update promiscuity count on a device 3331 * @dev: device 3332 * @inc: modifier 3333 * 3334 * Add or remove promiscuity from a device. While the count in the device 3335 * remains above zero the interface remains promiscuous. Once it hits zero 3336 * the device reverts back to normal filtering operation. A negative inc 3337 * value is used to drop promiscuity on the device. 3338 * Return 0 if successful or a negative errno code on error. 3339 */ 3340 int dev_set_promiscuity(struct net_device *dev, int inc) 3341 { 3342 unsigned short old_flags = dev->flags; 3343 int err; 3344 3345 err = __dev_set_promiscuity(dev, inc); 3346 if (err < 0) 3347 return err; 3348 if (dev->flags != old_flags) 3349 dev_set_rx_mode(dev); 3350 return err; 3351 } 3352 3353 /** 3354 * dev_set_allmulti - update allmulti count on a device 3355 * @dev: device 3356 * @inc: modifier 3357 * 3358 * Add or remove reception of all multicast frames to a device. While the 3359 * count in the device remains above zero the interface remains listening 3360 * to all interfaces. Once it hits zero the device reverts back to normal 3361 * filtering operation. A negative @inc value is used to drop the counter 3362 * when releasing a resource needing all multicasts. 3363 * Return 0 if successful or a negative errno code on error. 3364 */ 3365 3366 int dev_set_allmulti(struct net_device *dev, int inc) 3367 { 3368 unsigned short old_flags = dev->flags; 3369 3370 ASSERT_RTNL(); 3371 3372 dev->flags |= IFF_ALLMULTI; 3373 dev->allmulti += inc; 3374 if (dev->allmulti == 0) { 3375 /* 3376 * Avoid overflow. 3377 * If inc causes overflow, untouch allmulti and return error. 3378 */ 3379 if (inc < 0) 3380 dev->flags &= ~IFF_ALLMULTI; 3381 else { 3382 dev->allmulti -= inc; 3383 printk(KERN_WARNING "%s: allmulti touches roof, " 3384 "set allmulti failed, allmulti feature of " 3385 "device might be broken.\n", dev->name); 3386 return -EOVERFLOW; 3387 } 3388 } 3389 if (dev->flags ^ old_flags) { 3390 dev_change_rx_flags(dev, IFF_ALLMULTI); 3391 dev_set_rx_mode(dev); 3392 } 3393 return 0; 3394 } 3395 3396 /* 3397 * Upload unicast and multicast address lists to device and 3398 * configure RX filtering. When the device doesn't support unicast 3399 * filtering it is put in promiscuous mode while unicast addresses 3400 * are present. 3401 */ 3402 void __dev_set_rx_mode(struct net_device *dev) 3403 { 3404 const struct net_device_ops *ops = dev->netdev_ops; 3405 3406 /* dev_open will call this function so the list will stay sane. */ 3407 if (!(dev->flags&IFF_UP)) 3408 return; 3409 3410 if (!netif_device_present(dev)) 3411 return; 3412 3413 if (ops->ndo_set_rx_mode) 3414 ops->ndo_set_rx_mode(dev); 3415 else { 3416 /* Unicast addresses changes may only happen under the rtnl, 3417 * therefore calling __dev_set_promiscuity here is safe. 3418 */ 3419 if (dev->uc_count > 0 && !dev->uc_promisc) { 3420 __dev_set_promiscuity(dev, 1); 3421 dev->uc_promisc = 1; 3422 } else if (dev->uc_count == 0 && dev->uc_promisc) { 3423 __dev_set_promiscuity(dev, -1); 3424 dev->uc_promisc = 0; 3425 } 3426 3427 if (ops->ndo_set_multicast_list) 3428 ops->ndo_set_multicast_list(dev); 3429 } 3430 } 3431 3432 void dev_set_rx_mode(struct net_device *dev) 3433 { 3434 netif_addr_lock_bh(dev); 3435 __dev_set_rx_mode(dev); 3436 netif_addr_unlock_bh(dev); 3437 } 3438 3439 int __dev_addr_delete(struct dev_addr_list **list, int *count, 3440 void *addr, int alen, int glbl) 3441 { 3442 struct dev_addr_list *da; 3443 3444 for (; (da = *list) != NULL; list = &da->next) { 3445 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && 3446 alen == da->da_addrlen) { 3447 if (glbl) { 3448 int old_glbl = da->da_gusers; 3449 da->da_gusers = 0; 3450 if (old_glbl == 0) 3451 break; 3452 } 3453 if (--da->da_users) 3454 return 0; 3455 3456 *list = da->next; 3457 kfree(da); 3458 (*count)--; 3459 return 0; 3460 } 3461 } 3462 return -ENOENT; 3463 } 3464 3465 int __dev_addr_add(struct dev_addr_list **list, int *count, 3466 void *addr, int alen, int glbl) 3467 { 3468 struct dev_addr_list *da; 3469 3470 for (da = *list; da != NULL; da = da->next) { 3471 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && 3472 da->da_addrlen == alen) { 3473 if (glbl) { 3474 int old_glbl = da->da_gusers; 3475 da->da_gusers = 1; 3476 if (old_glbl) 3477 return 0; 3478 } 3479 da->da_users++; 3480 return 0; 3481 } 3482 } 3483 3484 da = kzalloc(sizeof(*da), GFP_ATOMIC); 3485 if (da == NULL) 3486 return -ENOMEM; 3487 memcpy(da->da_addr, addr, alen); 3488 da->da_addrlen = alen; 3489 da->da_users = 1; 3490 da->da_gusers = glbl ? 1 : 0; 3491 da->next = *list; 3492 *list = da; 3493 (*count)++; 3494 return 0; 3495 } 3496 3497 /** 3498 * dev_unicast_delete - Release secondary unicast address. 3499 * @dev: device 3500 * @addr: address to delete 3501 * @alen: length of @addr 3502 * 3503 * Release reference to a secondary unicast address and remove it 3504 * from the device if the reference count drops to zero. 3505 * 3506 * The caller must hold the rtnl_mutex. 3507 */ 3508 int dev_unicast_delete(struct net_device *dev, void *addr, int alen) 3509 { 3510 int err; 3511 3512 ASSERT_RTNL(); 3513 3514 netif_addr_lock_bh(dev); 3515 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0); 3516 if (!err) 3517 __dev_set_rx_mode(dev); 3518 netif_addr_unlock_bh(dev); 3519 return err; 3520 } 3521 EXPORT_SYMBOL(dev_unicast_delete); 3522 3523 /** 3524 * dev_unicast_add - add a secondary unicast address 3525 * @dev: device 3526 * @addr: address to add 3527 * @alen: length of @addr 3528 * 3529 * Add a secondary unicast address to the device or increase 3530 * the reference count if it already exists. 3531 * 3532 * The caller must hold the rtnl_mutex. 3533 */ 3534 int dev_unicast_add(struct net_device *dev, void *addr, int alen) 3535 { 3536 int err; 3537 3538 ASSERT_RTNL(); 3539 3540 netif_addr_lock_bh(dev); 3541 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0); 3542 if (!err) 3543 __dev_set_rx_mode(dev); 3544 netif_addr_unlock_bh(dev); 3545 return err; 3546 } 3547 EXPORT_SYMBOL(dev_unicast_add); 3548 3549 int __dev_addr_sync(struct dev_addr_list **to, int *to_count, 3550 struct dev_addr_list **from, int *from_count) 3551 { 3552 struct dev_addr_list *da, *next; 3553 int err = 0; 3554 3555 da = *from; 3556 while (da != NULL) { 3557 next = da->next; 3558 if (!da->da_synced) { 3559 err = __dev_addr_add(to, to_count, 3560 da->da_addr, da->da_addrlen, 0); 3561 if (err < 0) 3562 break; 3563 da->da_synced = 1; 3564 da->da_users++; 3565 } else if (da->da_users == 1) { 3566 __dev_addr_delete(to, to_count, 3567 da->da_addr, da->da_addrlen, 0); 3568 __dev_addr_delete(from, from_count, 3569 da->da_addr, da->da_addrlen, 0); 3570 } 3571 da = next; 3572 } 3573 return err; 3574 } 3575 3576 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, 3577 struct dev_addr_list **from, int *from_count) 3578 { 3579 struct dev_addr_list *da, *next; 3580 3581 da = *from; 3582 while (da != NULL) { 3583 next = da->next; 3584 if (da->da_synced) { 3585 __dev_addr_delete(to, to_count, 3586 da->da_addr, da->da_addrlen, 0); 3587 da->da_synced = 0; 3588 __dev_addr_delete(from, from_count, 3589 da->da_addr, da->da_addrlen, 0); 3590 } 3591 da = next; 3592 } 3593 } 3594 3595 /** 3596 * dev_unicast_sync - Synchronize device's unicast list to another device 3597 * @to: destination device 3598 * @from: source device 3599 * 3600 * Add newly added addresses to the destination device and release 3601 * addresses that have no users left. The source device must be 3602 * locked by netif_tx_lock_bh. 3603 * 3604 * This function is intended to be called from the dev->set_rx_mode 3605 * function of layered software devices. 3606 */ 3607 int dev_unicast_sync(struct net_device *to, struct net_device *from) 3608 { 3609 int err = 0; 3610 3611 netif_addr_lock_bh(to); 3612 err = __dev_addr_sync(&to->uc_list, &to->uc_count, 3613 &from->uc_list, &from->uc_count); 3614 if (!err) 3615 __dev_set_rx_mode(to); 3616 netif_addr_unlock_bh(to); 3617 return err; 3618 } 3619 EXPORT_SYMBOL(dev_unicast_sync); 3620 3621 /** 3622 * dev_unicast_unsync - Remove synchronized addresses from the destination device 3623 * @to: destination device 3624 * @from: source device 3625 * 3626 * Remove all addresses that were added to the destination device by 3627 * dev_unicast_sync(). This function is intended to be called from the 3628 * dev->stop function of layered software devices. 3629 */ 3630 void dev_unicast_unsync(struct net_device *to, struct net_device *from) 3631 { 3632 netif_addr_lock_bh(from); 3633 netif_addr_lock(to); 3634 3635 __dev_addr_unsync(&to->uc_list, &to->uc_count, 3636 &from->uc_list, &from->uc_count); 3637 __dev_set_rx_mode(to); 3638 3639 netif_addr_unlock(to); 3640 netif_addr_unlock_bh(from); 3641 } 3642 EXPORT_SYMBOL(dev_unicast_unsync); 3643 3644 static void __dev_addr_discard(struct dev_addr_list **list) 3645 { 3646 struct dev_addr_list *tmp; 3647 3648 while (*list != NULL) { 3649 tmp = *list; 3650 *list = tmp->next; 3651 if (tmp->da_users > tmp->da_gusers) 3652 printk("__dev_addr_discard: address leakage! " 3653 "da_users=%d\n", tmp->da_users); 3654 kfree(tmp); 3655 } 3656 } 3657 3658 static void dev_addr_discard(struct net_device *dev) 3659 { 3660 netif_addr_lock_bh(dev); 3661 3662 __dev_addr_discard(&dev->uc_list); 3663 dev->uc_count = 0; 3664 3665 __dev_addr_discard(&dev->mc_list); 3666 dev->mc_count = 0; 3667 3668 netif_addr_unlock_bh(dev); 3669 } 3670 3671 /** 3672 * dev_get_flags - get flags reported to userspace 3673 * @dev: device 3674 * 3675 * Get the combination of flag bits exported through APIs to userspace. 3676 */ 3677 unsigned dev_get_flags(const struct net_device *dev) 3678 { 3679 unsigned flags; 3680 3681 flags = (dev->flags & ~(IFF_PROMISC | 3682 IFF_ALLMULTI | 3683 IFF_RUNNING | 3684 IFF_LOWER_UP | 3685 IFF_DORMANT)) | 3686 (dev->gflags & (IFF_PROMISC | 3687 IFF_ALLMULTI)); 3688 3689 if (netif_running(dev)) { 3690 if (netif_oper_up(dev)) 3691 flags |= IFF_RUNNING; 3692 if (netif_carrier_ok(dev)) 3693 flags |= IFF_LOWER_UP; 3694 if (netif_dormant(dev)) 3695 flags |= IFF_DORMANT; 3696 } 3697 3698 return flags; 3699 } 3700 3701 /** 3702 * dev_change_flags - change device settings 3703 * @dev: device 3704 * @flags: device state flags 3705 * 3706 * Change settings on device based state flags. The flags are 3707 * in the userspace exported format. 3708 */ 3709 int dev_change_flags(struct net_device *dev, unsigned flags) 3710 { 3711 int ret, changes; 3712 int old_flags = dev->flags; 3713 3714 ASSERT_RTNL(); 3715 3716 /* 3717 * Set the flags on our device. 3718 */ 3719 3720 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 3721 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 3722 IFF_AUTOMEDIA)) | 3723 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 3724 IFF_ALLMULTI)); 3725 3726 /* 3727 * Load in the correct multicast list now the flags have changed. 3728 */ 3729 3730 if ((old_flags ^ flags) & IFF_MULTICAST) 3731 dev_change_rx_flags(dev, IFF_MULTICAST); 3732 3733 dev_set_rx_mode(dev); 3734 3735 /* 3736 * Have we downed the interface. We handle IFF_UP ourselves 3737 * according to user attempts to set it, rather than blindly 3738 * setting it. 3739 */ 3740 3741 ret = 0; 3742 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ 3743 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); 3744 3745 if (!ret) 3746 dev_set_rx_mode(dev); 3747 } 3748 3749 if (dev->flags & IFF_UP && 3750 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | 3751 IFF_VOLATILE))) 3752 call_netdevice_notifiers(NETDEV_CHANGE, dev); 3753 3754 if ((flags ^ dev->gflags) & IFF_PROMISC) { 3755 int inc = (flags & IFF_PROMISC) ? +1 : -1; 3756 dev->gflags ^= IFF_PROMISC; 3757 dev_set_promiscuity(dev, inc); 3758 } 3759 3760 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 3761 is important. Some (broken) drivers set IFF_PROMISC, when 3762 IFF_ALLMULTI is requested not asking us and not reporting. 3763 */ 3764 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 3765 int inc = (flags & IFF_ALLMULTI) ? +1 : -1; 3766 dev->gflags ^= IFF_ALLMULTI; 3767 dev_set_allmulti(dev, inc); 3768 } 3769 3770 /* Exclude state transition flags, already notified */ 3771 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); 3772 if (changes) 3773 rtmsg_ifinfo(RTM_NEWLINK, dev, changes); 3774 3775 return ret; 3776 } 3777 3778 /** 3779 * dev_set_mtu - Change maximum transfer unit 3780 * @dev: device 3781 * @new_mtu: new transfer unit 3782 * 3783 * Change the maximum transfer size of the network device. 3784 */ 3785 int dev_set_mtu(struct net_device *dev, int new_mtu) 3786 { 3787 const struct net_device_ops *ops = dev->netdev_ops; 3788 int err; 3789 3790 if (new_mtu == dev->mtu) 3791 return 0; 3792 3793 /* MTU must be positive. */ 3794 if (new_mtu < 0) 3795 return -EINVAL; 3796 3797 if (!netif_device_present(dev)) 3798 return -ENODEV; 3799 3800 err = 0; 3801 if (ops->ndo_change_mtu) 3802 err = ops->ndo_change_mtu(dev, new_mtu); 3803 else 3804 dev->mtu = new_mtu; 3805 3806 if (!err && dev->flags & IFF_UP) 3807 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 3808 return err; 3809 } 3810 3811 /** 3812 * dev_set_mac_address - Change Media Access Control Address 3813 * @dev: device 3814 * @sa: new address 3815 * 3816 * Change the hardware (MAC) address of the device 3817 */ 3818 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 3819 { 3820 const struct net_device_ops *ops = dev->netdev_ops; 3821 int err; 3822 3823 if (!ops->ndo_set_mac_address) 3824 return -EOPNOTSUPP; 3825 if (sa->sa_family != dev->type) 3826 return -EINVAL; 3827 if (!netif_device_present(dev)) 3828 return -ENODEV; 3829 err = ops->ndo_set_mac_address(dev, sa); 3830 if (!err) 3831 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 3832 return err; 3833 } 3834 3835 /* 3836 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock) 3837 */ 3838 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) 3839 { 3840 int err; 3841 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); 3842 3843 if (!dev) 3844 return -ENODEV; 3845 3846 switch (cmd) { 3847 case SIOCGIFFLAGS: /* Get interface flags */ 3848 ifr->ifr_flags = dev_get_flags(dev); 3849 return 0; 3850 3851 case SIOCGIFMETRIC: /* Get the metric on the interface 3852 (currently unused) */ 3853 ifr->ifr_metric = 0; 3854 return 0; 3855 3856 case SIOCGIFMTU: /* Get the MTU of a device */ 3857 ifr->ifr_mtu = dev->mtu; 3858 return 0; 3859 3860 case SIOCGIFHWADDR: 3861 if (!dev->addr_len) 3862 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); 3863 else 3864 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, 3865 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); 3866 ifr->ifr_hwaddr.sa_family = dev->type; 3867 return 0; 3868 3869 case SIOCGIFSLAVE: 3870 err = -EINVAL; 3871 break; 3872 3873 case SIOCGIFMAP: 3874 ifr->ifr_map.mem_start = dev->mem_start; 3875 ifr->ifr_map.mem_end = dev->mem_end; 3876 ifr->ifr_map.base_addr = dev->base_addr; 3877 ifr->ifr_map.irq = dev->irq; 3878 ifr->ifr_map.dma = dev->dma; 3879 ifr->ifr_map.port = dev->if_port; 3880 return 0; 3881 3882 case SIOCGIFINDEX: 3883 ifr->ifr_ifindex = dev->ifindex; 3884 return 0; 3885 3886 case SIOCGIFTXQLEN: 3887 ifr->ifr_qlen = dev->tx_queue_len; 3888 return 0; 3889 3890 default: 3891 /* dev_ioctl() should ensure this case 3892 * is never reached 3893 */ 3894 WARN_ON(1); 3895 err = -EINVAL; 3896 break; 3897 3898 } 3899 return err; 3900 } 3901 3902 /* 3903 * Perform the SIOCxIFxxx calls, inside rtnl_lock() 3904 */ 3905 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) 3906 { 3907 int err; 3908 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); 3909 const struct net_device_ops *ops; 3910 3911 if (!dev) 3912 return -ENODEV; 3913 3914 ops = dev->netdev_ops; 3915 3916 switch (cmd) { 3917 case SIOCSIFFLAGS: /* Set interface flags */ 3918 return dev_change_flags(dev, ifr->ifr_flags); 3919 3920 case SIOCSIFMETRIC: /* Set the metric on the interface 3921 (currently unused) */ 3922 return -EOPNOTSUPP; 3923 3924 case SIOCSIFMTU: /* Set the MTU of a device */ 3925 return dev_set_mtu(dev, ifr->ifr_mtu); 3926 3927 case SIOCSIFHWADDR: 3928 return dev_set_mac_address(dev, &ifr->ifr_hwaddr); 3929 3930 case SIOCSIFHWBROADCAST: 3931 if (ifr->ifr_hwaddr.sa_family != dev->type) 3932 return -EINVAL; 3933 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, 3934 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); 3935 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 3936 return 0; 3937 3938 case SIOCSIFMAP: 3939 if (ops->ndo_set_config) { 3940 if (!netif_device_present(dev)) 3941 return -ENODEV; 3942 return ops->ndo_set_config(dev, &ifr->ifr_map); 3943 } 3944 return -EOPNOTSUPP; 3945 3946 case SIOCADDMULTI: 3947 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || 3948 ifr->ifr_hwaddr.sa_family != AF_UNSPEC) 3949 return -EINVAL; 3950 if (!netif_device_present(dev)) 3951 return -ENODEV; 3952 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, 3953 dev->addr_len, 1); 3954 3955 case SIOCDELMULTI: 3956 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || 3957 ifr->ifr_hwaddr.sa_family != AF_UNSPEC) 3958 return -EINVAL; 3959 if (!netif_device_present(dev)) 3960 return -ENODEV; 3961 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, 3962 dev->addr_len, 1); 3963 3964 case SIOCSIFTXQLEN: 3965 if (ifr->ifr_qlen < 0) 3966 return -EINVAL; 3967 dev->tx_queue_len = ifr->ifr_qlen; 3968 return 0; 3969 3970 case SIOCSIFNAME: 3971 ifr->ifr_newname[IFNAMSIZ-1] = '\0'; 3972 return dev_change_name(dev, ifr->ifr_newname); 3973 3974 /* 3975 * Unknown or private ioctl 3976 */ 3977 3978 default: 3979 if ((cmd >= SIOCDEVPRIVATE && 3980 cmd <= SIOCDEVPRIVATE + 15) || 3981 cmd == SIOCBONDENSLAVE || 3982 cmd == SIOCBONDRELEASE || 3983 cmd == SIOCBONDSETHWADDR || 3984 cmd == SIOCBONDSLAVEINFOQUERY || 3985 cmd == SIOCBONDINFOQUERY || 3986 cmd == SIOCBONDCHANGEACTIVE || 3987 cmd == SIOCGMIIPHY || 3988 cmd == SIOCGMIIREG || 3989 cmd == SIOCSMIIREG || 3990 cmd == SIOCBRADDIF || 3991 cmd == SIOCBRDELIF || 3992 cmd == SIOCSHWTSTAMP || 3993 cmd == SIOCWANDEV) { 3994 err = -EOPNOTSUPP; 3995 if (ops->ndo_do_ioctl) { 3996 if (netif_device_present(dev)) 3997 err = ops->ndo_do_ioctl(dev, ifr, cmd); 3998 else 3999 err = -ENODEV; 4000 } 4001 } else 4002 err = -EINVAL; 4003 4004 } 4005 return err; 4006 } 4007 4008 /* 4009 * This function handles all "interface"-type I/O control requests. The actual 4010 * 'doing' part of this is dev_ifsioc above. 4011 */ 4012 4013 /** 4014 * dev_ioctl - network device ioctl 4015 * @net: the applicable net namespace 4016 * @cmd: command to issue 4017 * @arg: pointer to a struct ifreq in user space 4018 * 4019 * Issue ioctl functions to devices. This is normally called by the 4020 * user space syscall interfaces but can sometimes be useful for 4021 * other purposes. The return value is the return from the syscall if 4022 * positive or a negative errno code on error. 4023 */ 4024 4025 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) 4026 { 4027 struct ifreq ifr; 4028 int ret; 4029 char *colon; 4030 4031 /* One special case: SIOCGIFCONF takes ifconf argument 4032 and requires shared lock, because it sleeps writing 4033 to user space. 4034 */ 4035 4036 if (cmd == SIOCGIFCONF) { 4037 rtnl_lock(); 4038 ret = dev_ifconf(net, (char __user *) arg); 4039 rtnl_unlock(); 4040 return ret; 4041 } 4042 if (cmd == SIOCGIFNAME) 4043 return dev_ifname(net, (struct ifreq __user *)arg); 4044 4045 if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) 4046 return -EFAULT; 4047 4048 ifr.ifr_name[IFNAMSIZ-1] = 0; 4049 4050 colon = strchr(ifr.ifr_name, ':'); 4051 if (colon) 4052 *colon = 0; 4053 4054 /* 4055 * See which interface the caller is talking about. 4056 */ 4057 4058 switch (cmd) { 4059 /* 4060 * These ioctl calls: 4061 * - can be done by all. 4062 * - atomic and do not require locking. 4063 * - return a value 4064 */ 4065 case SIOCGIFFLAGS: 4066 case SIOCGIFMETRIC: 4067 case SIOCGIFMTU: 4068 case SIOCGIFHWADDR: 4069 case SIOCGIFSLAVE: 4070 case SIOCGIFMAP: 4071 case SIOCGIFINDEX: 4072 case SIOCGIFTXQLEN: 4073 dev_load(net, ifr.ifr_name); 4074 read_lock(&dev_base_lock); 4075 ret = dev_ifsioc_locked(net, &ifr, cmd); 4076 read_unlock(&dev_base_lock); 4077 if (!ret) { 4078 if (colon) 4079 *colon = ':'; 4080 if (copy_to_user(arg, &ifr, 4081 sizeof(struct ifreq))) 4082 ret = -EFAULT; 4083 } 4084 return ret; 4085 4086 case SIOCETHTOOL: 4087 dev_load(net, ifr.ifr_name); 4088 rtnl_lock(); 4089 ret = dev_ethtool(net, &ifr); 4090 rtnl_unlock(); 4091 if (!ret) { 4092 if (colon) 4093 *colon = ':'; 4094 if (copy_to_user(arg, &ifr, 4095 sizeof(struct ifreq))) 4096 ret = -EFAULT; 4097 } 4098 return ret; 4099 4100 /* 4101 * These ioctl calls: 4102 * - require superuser power. 4103 * - require strict serialization. 4104 * - return a value 4105 */ 4106 case SIOCGMIIPHY: 4107 case SIOCGMIIREG: 4108 case SIOCSIFNAME: 4109 if (!capable(CAP_NET_ADMIN)) 4110 return -EPERM; 4111 dev_load(net, ifr.ifr_name); 4112 rtnl_lock(); 4113 ret = dev_ifsioc(net, &ifr, cmd); 4114 rtnl_unlock(); 4115 if (!ret) { 4116 if (colon) 4117 *colon = ':'; 4118 if (copy_to_user(arg, &ifr, 4119 sizeof(struct ifreq))) 4120 ret = -EFAULT; 4121 } 4122 return ret; 4123 4124 /* 4125 * These ioctl calls: 4126 * - require superuser power. 4127 * - require strict serialization. 4128 * - do not return a value 4129 */ 4130 case SIOCSIFFLAGS: 4131 case SIOCSIFMETRIC: 4132 case SIOCSIFMTU: 4133 case SIOCSIFMAP: 4134 case SIOCSIFHWADDR: 4135 case SIOCSIFSLAVE: 4136 case SIOCADDMULTI: 4137 case SIOCDELMULTI: 4138 case SIOCSIFHWBROADCAST: 4139 case SIOCSIFTXQLEN: 4140 case SIOCSMIIREG: 4141 case SIOCBONDENSLAVE: 4142 case SIOCBONDRELEASE: 4143 case SIOCBONDSETHWADDR: 4144 case SIOCBONDCHANGEACTIVE: 4145 case SIOCBRADDIF: 4146 case SIOCBRDELIF: 4147 case SIOCSHWTSTAMP: 4148 if (!capable(CAP_NET_ADMIN)) 4149 return -EPERM; 4150 /* fall through */ 4151 case SIOCBONDSLAVEINFOQUERY: 4152 case SIOCBONDINFOQUERY: 4153 dev_load(net, ifr.ifr_name); 4154 rtnl_lock(); 4155 ret = dev_ifsioc(net, &ifr, cmd); 4156 rtnl_unlock(); 4157 return ret; 4158 4159 case SIOCGIFMEM: 4160 /* Get the per device memory space. We can add this but 4161 * currently do not support it */ 4162 case SIOCSIFMEM: 4163 /* Set the per device memory buffer space. 4164 * Not applicable in our case */ 4165 case SIOCSIFLINK: 4166 return -EINVAL; 4167 4168 /* 4169 * Unknown or private ioctl. 4170 */ 4171 default: 4172 if (cmd == SIOCWANDEV || 4173 (cmd >= SIOCDEVPRIVATE && 4174 cmd <= SIOCDEVPRIVATE + 15)) { 4175 dev_load(net, ifr.ifr_name); 4176 rtnl_lock(); 4177 ret = dev_ifsioc(net, &ifr, cmd); 4178 rtnl_unlock(); 4179 if (!ret && copy_to_user(arg, &ifr, 4180 sizeof(struct ifreq))) 4181 ret = -EFAULT; 4182 return ret; 4183 } 4184 /* Take care of Wireless Extensions */ 4185 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) 4186 return wext_handle_ioctl(net, &ifr, cmd, arg); 4187 return -EINVAL; 4188 } 4189 } 4190 4191 4192 /** 4193 * dev_new_index - allocate an ifindex 4194 * @net: the applicable net namespace 4195 * 4196 * Returns a suitable unique value for a new device interface 4197 * number. The caller must hold the rtnl semaphore or the 4198 * dev_base_lock to be sure it remains unique. 4199 */ 4200 static int dev_new_index(struct net *net) 4201 { 4202 static int ifindex; 4203 for (;;) { 4204 if (++ifindex <= 0) 4205 ifindex = 1; 4206 if (!__dev_get_by_index(net, ifindex)) 4207 return ifindex; 4208 } 4209 } 4210 4211 /* Delayed registration/unregisteration */ 4212 static LIST_HEAD(net_todo_list); 4213 4214 static void net_set_todo(struct net_device *dev) 4215 { 4216 list_add_tail(&dev->todo_list, &net_todo_list); 4217 } 4218 4219 static void rollback_registered(struct net_device *dev) 4220 { 4221 BUG_ON(dev_boot_phase); 4222 ASSERT_RTNL(); 4223 4224 /* Some devices call without registering for initialization unwind. */ 4225 if (dev->reg_state == NETREG_UNINITIALIZED) { 4226 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " 4227 "was registered\n", dev->name, dev); 4228 4229 WARN_ON(1); 4230 return; 4231 } 4232 4233 BUG_ON(dev->reg_state != NETREG_REGISTERED); 4234 4235 /* If device is running, close it first. */ 4236 dev_close(dev); 4237 4238 /* And unlink it from device chain. */ 4239 unlist_netdevice(dev); 4240 4241 dev->reg_state = NETREG_UNREGISTERING; 4242 4243 synchronize_net(); 4244 4245 /* Shutdown queueing discipline. */ 4246 dev_shutdown(dev); 4247 4248 4249 /* Notify protocols, that we are about to destroy 4250 this device. They should clean all the things. 4251 */ 4252 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 4253 4254 /* 4255 * Flush the unicast and multicast chains 4256 */ 4257 dev_addr_discard(dev); 4258 4259 if (dev->netdev_ops->ndo_uninit) 4260 dev->netdev_ops->ndo_uninit(dev); 4261 4262 /* Notifier chain MUST detach us from master device. */ 4263 WARN_ON(dev->master); 4264 4265 /* Remove entries from kobject tree */ 4266 netdev_unregister_kobject(dev); 4267 4268 synchronize_net(); 4269 4270 dev_put(dev); 4271 } 4272 4273 static void __netdev_init_queue_locks_one(struct net_device *dev, 4274 struct netdev_queue *dev_queue, 4275 void *_unused) 4276 { 4277 spin_lock_init(&dev_queue->_xmit_lock); 4278 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); 4279 dev_queue->xmit_lock_owner = -1; 4280 } 4281 4282 static void netdev_init_queue_locks(struct net_device *dev) 4283 { 4284 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); 4285 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); 4286 } 4287 4288 unsigned long netdev_fix_features(unsigned long features, const char *name) 4289 { 4290 /* Fix illegal SG+CSUM combinations. */ 4291 if ((features & NETIF_F_SG) && 4292 !(features & NETIF_F_ALL_CSUM)) { 4293 if (name) 4294 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " 4295 "checksum feature.\n", name); 4296 features &= ~NETIF_F_SG; 4297 } 4298 4299 /* TSO requires that SG is present as well. */ 4300 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { 4301 if (name) 4302 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " 4303 "SG feature.\n", name); 4304 features &= ~NETIF_F_TSO; 4305 } 4306 4307 if (features & NETIF_F_UFO) { 4308 if (!(features & NETIF_F_GEN_CSUM)) { 4309 if (name) 4310 printk(KERN_ERR "%s: Dropping NETIF_F_UFO " 4311 "since no NETIF_F_HW_CSUM feature.\n", 4312 name); 4313 features &= ~NETIF_F_UFO; 4314 } 4315 4316 if (!(features & NETIF_F_SG)) { 4317 if (name) 4318 printk(KERN_ERR "%s: Dropping NETIF_F_UFO " 4319 "since no NETIF_F_SG feature.\n", name); 4320 features &= ~NETIF_F_UFO; 4321 } 4322 } 4323 4324 return features; 4325 } 4326 EXPORT_SYMBOL(netdev_fix_features); 4327 4328 /* Some devices need to (re-)set their netdev_ops inside 4329 * ->init() or similar. If that happens, we have to setup 4330 * the compat pointers again. 4331 */ 4332 void netdev_resync_ops(struct net_device *dev) 4333 { 4334 #ifdef CONFIG_COMPAT_NET_DEV_OPS 4335 const struct net_device_ops *ops = dev->netdev_ops; 4336 4337 dev->init = ops->ndo_init; 4338 dev->uninit = ops->ndo_uninit; 4339 dev->open = ops->ndo_open; 4340 dev->change_rx_flags = ops->ndo_change_rx_flags; 4341 dev->set_rx_mode = ops->ndo_set_rx_mode; 4342 dev->set_multicast_list = ops->ndo_set_multicast_list; 4343 dev->set_mac_address = ops->ndo_set_mac_address; 4344 dev->validate_addr = ops->ndo_validate_addr; 4345 dev->do_ioctl = ops->ndo_do_ioctl; 4346 dev->set_config = ops->ndo_set_config; 4347 dev->change_mtu = ops->ndo_change_mtu; 4348 dev->neigh_setup = ops->ndo_neigh_setup; 4349 dev->tx_timeout = ops->ndo_tx_timeout; 4350 dev->get_stats = ops->ndo_get_stats; 4351 dev->vlan_rx_register = ops->ndo_vlan_rx_register; 4352 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid; 4353 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid; 4354 #ifdef CONFIG_NET_POLL_CONTROLLER 4355 dev->poll_controller = ops->ndo_poll_controller; 4356 #endif 4357 #endif 4358 } 4359 EXPORT_SYMBOL(netdev_resync_ops); 4360 4361 /** 4362 * register_netdevice - register a network device 4363 * @dev: device to register 4364 * 4365 * Take a completed network device structure and add it to the kernel 4366 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 4367 * chain. 0 is returned on success. A negative errno code is returned 4368 * on a failure to set up the device, or if the name is a duplicate. 4369 * 4370 * Callers must hold the rtnl semaphore. You may want 4371 * register_netdev() instead of this. 4372 * 4373 * BUGS: 4374 * The locking appears insufficient to guarantee two parallel registers 4375 * will not get the same name. 4376 */ 4377 4378 int register_netdevice(struct net_device *dev) 4379 { 4380 struct hlist_head *head; 4381 struct hlist_node *p; 4382 int ret; 4383 struct net *net = dev_net(dev); 4384 4385 BUG_ON(dev_boot_phase); 4386 ASSERT_RTNL(); 4387 4388 might_sleep(); 4389 4390 /* When net_device's are persistent, this will be fatal. */ 4391 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 4392 BUG_ON(!net); 4393 4394 spin_lock_init(&dev->addr_list_lock); 4395 netdev_set_addr_lockdep_class(dev); 4396 netdev_init_queue_locks(dev); 4397 4398 dev->iflink = -1; 4399 4400 #ifdef CONFIG_COMPAT_NET_DEV_OPS 4401 /* Netdevice_ops API compatiability support. 4402 * This is temporary until all network devices are converted. 4403 */ 4404 if (dev->netdev_ops) { 4405 netdev_resync_ops(dev); 4406 } else { 4407 char drivername[64]; 4408 pr_info("%s (%s): not using net_device_ops yet\n", 4409 dev->name, netdev_drivername(dev, drivername, 64)); 4410 4411 /* This works only because net_device_ops and the 4412 compatiablity structure are the same. */ 4413 dev->netdev_ops = (void *) &(dev->init); 4414 } 4415 #endif 4416 4417 /* Init, if this function is available */ 4418 if (dev->netdev_ops->ndo_init) { 4419 ret = dev->netdev_ops->ndo_init(dev); 4420 if (ret) { 4421 if (ret > 0) 4422 ret = -EIO; 4423 goto out; 4424 } 4425 } 4426 4427 if (!dev_valid_name(dev->name)) { 4428 ret = -EINVAL; 4429 goto err_uninit; 4430 } 4431 4432 dev->ifindex = dev_new_index(net); 4433 if (dev->iflink == -1) 4434 dev->iflink = dev->ifindex; 4435 4436 /* Check for existence of name */ 4437 head = dev_name_hash(net, dev->name); 4438 hlist_for_each(p, head) { 4439 struct net_device *d 4440 = hlist_entry(p, struct net_device, name_hlist); 4441 if (!strncmp(d->name, dev->name, IFNAMSIZ)) { 4442 ret = -EEXIST; 4443 goto err_uninit; 4444 } 4445 } 4446 4447 /* Fix illegal checksum combinations */ 4448 if ((dev->features & NETIF_F_HW_CSUM) && 4449 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 4450 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", 4451 dev->name); 4452 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 4453 } 4454 4455 if ((dev->features & NETIF_F_NO_CSUM) && 4456 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 4457 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", 4458 dev->name); 4459 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); 4460 } 4461 4462 dev->features = netdev_fix_features(dev->features, dev->name); 4463 4464 /* Enable software GSO if SG is supported. */ 4465 if (dev->features & NETIF_F_SG) 4466 dev->features |= NETIF_F_GSO; 4467 4468 netdev_initialize_kobject(dev); 4469 ret = netdev_register_kobject(dev); 4470 if (ret) 4471 goto err_uninit; 4472 dev->reg_state = NETREG_REGISTERED; 4473 4474 /* 4475 * Default initial state at registry is that the 4476 * device is present. 4477 */ 4478 4479 set_bit(__LINK_STATE_PRESENT, &dev->state); 4480 4481 dev_init_scheduler(dev); 4482 dev_hold(dev); 4483 list_netdevice(dev); 4484 4485 /* Notify protocols, that a new device appeared. */ 4486 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 4487 ret = notifier_to_errno(ret); 4488 if (ret) { 4489 rollback_registered(dev); 4490 dev->reg_state = NETREG_UNREGISTERED; 4491 } 4492 4493 out: 4494 return ret; 4495 4496 err_uninit: 4497 if (dev->netdev_ops->ndo_uninit) 4498 dev->netdev_ops->ndo_uninit(dev); 4499 goto out; 4500 } 4501 4502 /** 4503 * init_dummy_netdev - init a dummy network device for NAPI 4504 * @dev: device to init 4505 * 4506 * This takes a network device structure and initialize the minimum 4507 * amount of fields so it can be used to schedule NAPI polls without 4508 * registering a full blown interface. This is to be used by drivers 4509 * that need to tie several hardware interfaces to a single NAPI 4510 * poll scheduler due to HW limitations. 4511 */ 4512 int init_dummy_netdev(struct net_device *dev) 4513 { 4514 /* Clear everything. Note we don't initialize spinlocks 4515 * are they aren't supposed to be taken by any of the 4516 * NAPI code and this dummy netdev is supposed to be 4517 * only ever used for NAPI polls 4518 */ 4519 memset(dev, 0, sizeof(struct net_device)); 4520 4521 /* make sure we BUG if trying to hit standard 4522 * register/unregister code path 4523 */ 4524 dev->reg_state = NETREG_DUMMY; 4525 4526 /* initialize the ref count */ 4527 atomic_set(&dev->refcnt, 1); 4528 4529 /* NAPI wants this */ 4530 INIT_LIST_HEAD(&dev->napi_list); 4531 4532 /* a dummy interface is started by default */ 4533 set_bit(__LINK_STATE_PRESENT, &dev->state); 4534 set_bit(__LINK_STATE_START, &dev->state); 4535 4536 return 0; 4537 } 4538 EXPORT_SYMBOL_GPL(init_dummy_netdev); 4539 4540 4541 /** 4542 * register_netdev - register a network device 4543 * @dev: device to register 4544 * 4545 * Take a completed network device structure and add it to the kernel 4546 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 4547 * chain. 0 is returned on success. A negative errno code is returned 4548 * on a failure to set up the device, or if the name is a duplicate. 4549 * 4550 * This is a wrapper around register_netdevice that takes the rtnl semaphore 4551 * and expands the device name if you passed a format string to 4552 * alloc_netdev. 4553 */ 4554 int register_netdev(struct net_device *dev) 4555 { 4556 int err; 4557 4558 rtnl_lock(); 4559 4560 /* 4561 * If the name is a format string the caller wants us to do a 4562 * name allocation. 4563 */ 4564 if (strchr(dev->name, '%')) { 4565 err = dev_alloc_name(dev, dev->name); 4566 if (err < 0) 4567 goto out; 4568 } 4569 4570 err = register_netdevice(dev); 4571 out: 4572 rtnl_unlock(); 4573 return err; 4574 } 4575 EXPORT_SYMBOL(register_netdev); 4576 4577 /* 4578 * netdev_wait_allrefs - wait until all references are gone. 4579 * 4580 * This is called when unregistering network devices. 4581 * 4582 * Any protocol or device that holds a reference should register 4583 * for netdevice notification, and cleanup and put back the 4584 * reference if they receive an UNREGISTER event. 4585 * We can get stuck here if buggy protocols don't correctly 4586 * call dev_put. 4587 */ 4588 static void netdev_wait_allrefs(struct net_device *dev) 4589 { 4590 unsigned long rebroadcast_time, warning_time; 4591 4592 rebroadcast_time = warning_time = jiffies; 4593 while (atomic_read(&dev->refcnt) != 0) { 4594 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 4595 rtnl_lock(); 4596 4597 /* Rebroadcast unregister notification */ 4598 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 4599 4600 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 4601 &dev->state)) { 4602 /* We must not have linkwatch events 4603 * pending on unregister. If this 4604 * happens, we simply run the queue 4605 * unscheduled, resulting in a noop 4606 * for this device. 4607 */ 4608 linkwatch_run_queue(); 4609 } 4610 4611 __rtnl_unlock(); 4612 4613 rebroadcast_time = jiffies; 4614 } 4615 4616 msleep(250); 4617 4618 if (time_after(jiffies, warning_time + 10 * HZ)) { 4619 printk(KERN_EMERG "unregister_netdevice: " 4620 "waiting for %s to become free. Usage " 4621 "count = %d\n", 4622 dev->name, atomic_read(&dev->refcnt)); 4623 warning_time = jiffies; 4624 } 4625 } 4626 } 4627 4628 /* The sequence is: 4629 * 4630 * rtnl_lock(); 4631 * ... 4632 * register_netdevice(x1); 4633 * register_netdevice(x2); 4634 * ... 4635 * unregister_netdevice(y1); 4636 * unregister_netdevice(y2); 4637 * ... 4638 * rtnl_unlock(); 4639 * free_netdev(y1); 4640 * free_netdev(y2); 4641 * 4642 * We are invoked by rtnl_unlock(). 4643 * This allows us to deal with problems: 4644 * 1) We can delete sysfs objects which invoke hotplug 4645 * without deadlocking with linkwatch via keventd. 4646 * 2) Since we run with the RTNL semaphore not held, we can sleep 4647 * safely in order to wait for the netdev refcnt to drop to zero. 4648 * 4649 * We must not return until all unregister events added during 4650 * the interval the lock was held have been completed. 4651 */ 4652 void netdev_run_todo(void) 4653 { 4654 struct list_head list; 4655 4656 /* Snapshot list, allow later requests */ 4657 list_replace_init(&net_todo_list, &list); 4658 4659 __rtnl_unlock(); 4660 4661 while (!list_empty(&list)) { 4662 struct net_device *dev 4663 = list_entry(list.next, struct net_device, todo_list); 4664 list_del(&dev->todo_list); 4665 4666 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 4667 printk(KERN_ERR "network todo '%s' but state %d\n", 4668 dev->name, dev->reg_state); 4669 dump_stack(); 4670 continue; 4671 } 4672 4673 dev->reg_state = NETREG_UNREGISTERED; 4674 4675 on_each_cpu(flush_backlog, dev, 1); 4676 4677 netdev_wait_allrefs(dev); 4678 4679 /* paranoia */ 4680 BUG_ON(atomic_read(&dev->refcnt)); 4681 WARN_ON(dev->ip_ptr); 4682 WARN_ON(dev->ip6_ptr); 4683 WARN_ON(dev->dn_ptr); 4684 4685 if (dev->destructor) 4686 dev->destructor(dev); 4687 4688 /* Free network device */ 4689 kobject_put(&dev->dev.kobj); 4690 } 4691 } 4692 4693 /** 4694 * dev_get_stats - get network device statistics 4695 * @dev: device to get statistics from 4696 * 4697 * Get network statistics from device. The device driver may provide 4698 * its own method by setting dev->netdev_ops->get_stats; otherwise 4699 * the internal statistics structure is used. 4700 */ 4701 const struct net_device_stats *dev_get_stats(struct net_device *dev) 4702 { 4703 const struct net_device_ops *ops = dev->netdev_ops; 4704 4705 if (ops->ndo_get_stats) 4706 return ops->ndo_get_stats(dev); 4707 else 4708 return &dev->stats; 4709 } 4710 EXPORT_SYMBOL(dev_get_stats); 4711 4712 static void netdev_init_one_queue(struct net_device *dev, 4713 struct netdev_queue *queue, 4714 void *_unused) 4715 { 4716 queue->dev = dev; 4717 } 4718 4719 static void netdev_init_queues(struct net_device *dev) 4720 { 4721 netdev_init_one_queue(dev, &dev->rx_queue, NULL); 4722 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 4723 spin_lock_init(&dev->tx_global_lock); 4724 } 4725 4726 /** 4727 * alloc_netdev_mq - allocate network device 4728 * @sizeof_priv: size of private data to allocate space for 4729 * @name: device name format string 4730 * @setup: callback to initialize device 4731 * @queue_count: the number of subqueues to allocate 4732 * 4733 * Allocates a struct net_device with private data area for driver use 4734 * and performs basic initialization. Also allocates subquue structs 4735 * for each queue on the device at the end of the netdevice. 4736 */ 4737 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, 4738 void (*setup)(struct net_device *), unsigned int queue_count) 4739 { 4740 struct netdev_queue *tx; 4741 struct net_device *dev; 4742 size_t alloc_size; 4743 void *p; 4744 4745 BUG_ON(strlen(name) >= sizeof(dev->name)); 4746 4747 alloc_size = sizeof(struct net_device); 4748 if (sizeof_priv) { 4749 /* ensure 32-byte alignment of private area */ 4750 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; 4751 alloc_size += sizeof_priv; 4752 } 4753 /* ensure 32-byte alignment of whole construct */ 4754 alloc_size += NETDEV_ALIGN_CONST; 4755 4756 p = kzalloc(alloc_size, GFP_KERNEL); 4757 if (!p) { 4758 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); 4759 return NULL; 4760 } 4761 4762 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); 4763 if (!tx) { 4764 printk(KERN_ERR "alloc_netdev: Unable to allocate " 4765 "tx qdiscs.\n"); 4766 kfree(p); 4767 return NULL; 4768 } 4769 4770 dev = (struct net_device *) 4771 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); 4772 dev->padded = (char *)dev - (char *)p; 4773 dev_net_set(dev, &init_net); 4774 4775 dev->_tx = tx; 4776 dev->num_tx_queues = queue_count; 4777 dev->real_num_tx_queues = queue_count; 4778 4779 dev->gso_max_size = GSO_MAX_SIZE; 4780 4781 netdev_init_queues(dev); 4782 4783 INIT_LIST_HEAD(&dev->napi_list); 4784 setup(dev); 4785 strcpy(dev->name, name); 4786 return dev; 4787 } 4788 EXPORT_SYMBOL(alloc_netdev_mq); 4789 4790 /** 4791 * free_netdev - free network device 4792 * @dev: device 4793 * 4794 * This function does the last stage of destroying an allocated device 4795 * interface. The reference to the device object is released. 4796 * If this is the last reference then it will be freed. 4797 */ 4798 void free_netdev(struct net_device *dev) 4799 { 4800 struct napi_struct *p, *n; 4801 4802 release_net(dev_net(dev)); 4803 4804 kfree(dev->_tx); 4805 4806 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 4807 netif_napi_del(p); 4808 4809 /* Compatibility with error handling in drivers */ 4810 if (dev->reg_state == NETREG_UNINITIALIZED) { 4811 kfree((char *)dev - dev->padded); 4812 return; 4813 } 4814 4815 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 4816 dev->reg_state = NETREG_RELEASED; 4817 4818 /* will free via device release */ 4819 put_device(&dev->dev); 4820 } 4821 4822 /** 4823 * synchronize_net - Synchronize with packet receive processing 4824 * 4825 * Wait for packets currently being received to be done. 4826 * Does not block later packets from starting. 4827 */ 4828 void synchronize_net(void) 4829 { 4830 might_sleep(); 4831 synchronize_rcu(); 4832 } 4833 4834 /** 4835 * unregister_netdevice - remove device from the kernel 4836 * @dev: device 4837 * 4838 * This function shuts down a device interface and removes it 4839 * from the kernel tables. 4840 * 4841 * Callers must hold the rtnl semaphore. You may want 4842 * unregister_netdev() instead of this. 4843 */ 4844 4845 void unregister_netdevice(struct net_device *dev) 4846 { 4847 ASSERT_RTNL(); 4848 4849 rollback_registered(dev); 4850 /* Finish processing unregister after unlock */ 4851 net_set_todo(dev); 4852 } 4853 4854 /** 4855 * unregister_netdev - remove device from the kernel 4856 * @dev: device 4857 * 4858 * This function shuts down a device interface and removes it 4859 * from the kernel tables. 4860 * 4861 * This is just a wrapper for unregister_netdevice that takes 4862 * the rtnl semaphore. In general you want to use this and not 4863 * unregister_netdevice. 4864 */ 4865 void unregister_netdev(struct net_device *dev) 4866 { 4867 rtnl_lock(); 4868 unregister_netdevice(dev); 4869 rtnl_unlock(); 4870 } 4871 4872 EXPORT_SYMBOL(unregister_netdev); 4873 4874 /** 4875 * dev_change_net_namespace - move device to different nethost namespace 4876 * @dev: device 4877 * @net: network namespace 4878 * @pat: If not NULL name pattern to try if the current device name 4879 * is already taken in the destination network namespace. 4880 * 4881 * This function shuts down a device interface and moves it 4882 * to a new network namespace. On success 0 is returned, on 4883 * a failure a netagive errno code is returned. 4884 * 4885 * Callers must hold the rtnl semaphore. 4886 */ 4887 4888 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 4889 { 4890 char buf[IFNAMSIZ]; 4891 const char *destname; 4892 int err; 4893 4894 ASSERT_RTNL(); 4895 4896 /* Don't allow namespace local devices to be moved. */ 4897 err = -EINVAL; 4898 if (dev->features & NETIF_F_NETNS_LOCAL) 4899 goto out; 4900 4901 #ifdef CONFIG_SYSFS 4902 /* Don't allow real devices to be moved when sysfs 4903 * is enabled. 4904 */ 4905 err = -EINVAL; 4906 if (dev->dev.parent) 4907 goto out; 4908 #endif 4909 4910 /* Ensure the device has been registrered */ 4911 err = -EINVAL; 4912 if (dev->reg_state != NETREG_REGISTERED) 4913 goto out; 4914 4915 /* Get out if there is nothing todo */ 4916 err = 0; 4917 if (net_eq(dev_net(dev), net)) 4918 goto out; 4919 4920 /* Pick the destination device name, and ensure 4921 * we can use it in the destination network namespace. 4922 */ 4923 err = -EEXIST; 4924 destname = dev->name; 4925 if (__dev_get_by_name(net, destname)) { 4926 /* We get here if we can't use the current device name */ 4927 if (!pat) 4928 goto out; 4929 if (!dev_valid_name(pat)) 4930 goto out; 4931 if (strchr(pat, '%')) { 4932 if (__dev_alloc_name(net, pat, buf) < 0) 4933 goto out; 4934 destname = buf; 4935 } else 4936 destname = pat; 4937 if (__dev_get_by_name(net, destname)) 4938 goto out; 4939 } 4940 4941 /* 4942 * And now a mini version of register_netdevice unregister_netdevice. 4943 */ 4944 4945 /* If device is running close it first. */ 4946 dev_close(dev); 4947 4948 /* And unlink it from device chain */ 4949 err = -ENODEV; 4950 unlist_netdevice(dev); 4951 4952 synchronize_net(); 4953 4954 /* Shutdown queueing discipline. */ 4955 dev_shutdown(dev); 4956 4957 /* Notify protocols, that we are about to destroy 4958 this device. They should clean all the things. 4959 */ 4960 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 4961 4962 /* 4963 * Flush the unicast and multicast chains 4964 */ 4965 dev_addr_discard(dev); 4966 4967 netdev_unregister_kobject(dev); 4968 4969 /* Actually switch the network namespace */ 4970 dev_net_set(dev, net); 4971 4972 /* Assign the new device name */ 4973 if (destname != dev->name) 4974 strcpy(dev->name, destname); 4975 4976 /* If there is an ifindex conflict assign a new one */ 4977 if (__dev_get_by_index(net, dev->ifindex)) { 4978 int iflink = (dev->iflink == dev->ifindex); 4979 dev->ifindex = dev_new_index(net); 4980 if (iflink) 4981 dev->iflink = dev->ifindex; 4982 } 4983 4984 /* Fixup kobjects */ 4985 err = netdev_register_kobject(dev); 4986 WARN_ON(err); 4987 4988 /* Add the device back in the hashes */ 4989 list_netdevice(dev); 4990 4991 /* Notify protocols, that a new device appeared. */ 4992 call_netdevice_notifiers(NETDEV_REGISTER, dev); 4993 4994 synchronize_net(); 4995 err = 0; 4996 out: 4997 return err; 4998 } 4999 5000 static int dev_cpu_callback(struct notifier_block *nfb, 5001 unsigned long action, 5002 void *ocpu) 5003 { 5004 struct sk_buff **list_skb; 5005 struct Qdisc **list_net; 5006 struct sk_buff *skb; 5007 unsigned int cpu, oldcpu = (unsigned long)ocpu; 5008 struct softnet_data *sd, *oldsd; 5009 5010 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) 5011 return NOTIFY_OK; 5012 5013 local_irq_disable(); 5014 cpu = smp_processor_id(); 5015 sd = &per_cpu(softnet_data, cpu); 5016 oldsd = &per_cpu(softnet_data, oldcpu); 5017 5018 /* Find end of our completion_queue. */ 5019 list_skb = &sd->completion_queue; 5020 while (*list_skb) 5021 list_skb = &(*list_skb)->next; 5022 /* Append completion queue from offline CPU. */ 5023 *list_skb = oldsd->completion_queue; 5024 oldsd->completion_queue = NULL; 5025 5026 /* Find end of our output_queue. */ 5027 list_net = &sd->output_queue; 5028 while (*list_net) 5029 list_net = &(*list_net)->next_sched; 5030 /* Append output queue from offline CPU. */ 5031 *list_net = oldsd->output_queue; 5032 oldsd->output_queue = NULL; 5033 5034 raise_softirq_irqoff(NET_TX_SOFTIRQ); 5035 local_irq_enable(); 5036 5037 /* Process offline CPU's input_pkt_queue */ 5038 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) 5039 netif_rx(skb); 5040 5041 return NOTIFY_OK; 5042 } 5043 5044 5045 /** 5046 * netdev_increment_features - increment feature set by one 5047 * @all: current feature set 5048 * @one: new feature set 5049 * @mask: mask feature set 5050 * 5051 * Computes a new feature set after adding a device with feature set 5052 * @one to the master device with current feature set @all. Will not 5053 * enable anything that is off in @mask. Returns the new feature set. 5054 */ 5055 unsigned long netdev_increment_features(unsigned long all, unsigned long one, 5056 unsigned long mask) 5057 { 5058 /* If device needs checksumming, downgrade to it. */ 5059 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) 5060 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); 5061 else if (mask & NETIF_F_ALL_CSUM) { 5062 /* If one device supports v4/v6 checksumming, set for all. */ 5063 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && 5064 !(all & NETIF_F_GEN_CSUM)) { 5065 all &= ~NETIF_F_ALL_CSUM; 5066 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 5067 } 5068 5069 /* If one device supports hw checksumming, set for all. */ 5070 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { 5071 all &= ~NETIF_F_ALL_CSUM; 5072 all |= NETIF_F_HW_CSUM; 5073 } 5074 } 5075 5076 one |= NETIF_F_ALL_CSUM; 5077 5078 one |= all & NETIF_F_ONE_FOR_ALL; 5079 all &= one | NETIF_F_LLTX | NETIF_F_GSO; 5080 all |= one & mask & NETIF_F_ONE_FOR_ALL; 5081 5082 return all; 5083 } 5084 EXPORT_SYMBOL(netdev_increment_features); 5085 5086 static struct hlist_head *netdev_create_hash(void) 5087 { 5088 int i; 5089 struct hlist_head *hash; 5090 5091 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); 5092 if (hash != NULL) 5093 for (i = 0; i < NETDEV_HASHENTRIES; i++) 5094 INIT_HLIST_HEAD(&hash[i]); 5095 5096 return hash; 5097 } 5098 5099 /* Initialize per network namespace state */ 5100 static int __net_init netdev_init(struct net *net) 5101 { 5102 INIT_LIST_HEAD(&net->dev_base_head); 5103 5104 net->dev_name_head = netdev_create_hash(); 5105 if (net->dev_name_head == NULL) 5106 goto err_name; 5107 5108 net->dev_index_head = netdev_create_hash(); 5109 if (net->dev_index_head == NULL) 5110 goto err_idx; 5111 5112 return 0; 5113 5114 err_idx: 5115 kfree(net->dev_name_head); 5116 err_name: 5117 return -ENOMEM; 5118 } 5119 5120 /** 5121 * netdev_drivername - network driver for the device 5122 * @dev: network device 5123 * @buffer: buffer for resulting name 5124 * @len: size of buffer 5125 * 5126 * Determine network driver for device. 5127 */ 5128 char *netdev_drivername(const struct net_device *dev, char *buffer, int len) 5129 { 5130 const struct device_driver *driver; 5131 const struct device *parent; 5132 5133 if (len <= 0 || !buffer) 5134 return buffer; 5135 buffer[0] = 0; 5136 5137 parent = dev->dev.parent; 5138 5139 if (!parent) 5140 return buffer; 5141 5142 driver = parent->driver; 5143 if (driver && driver->name) 5144 strlcpy(buffer, driver->name, len); 5145 return buffer; 5146 } 5147 5148 static void __net_exit netdev_exit(struct net *net) 5149 { 5150 kfree(net->dev_name_head); 5151 kfree(net->dev_index_head); 5152 } 5153 5154 static struct pernet_operations __net_initdata netdev_net_ops = { 5155 .init = netdev_init, 5156 .exit = netdev_exit, 5157 }; 5158 5159 static void __net_exit default_device_exit(struct net *net) 5160 { 5161 struct net_device *dev; 5162 /* 5163 * Push all migratable of the network devices back to the 5164 * initial network namespace 5165 */ 5166 rtnl_lock(); 5167 restart: 5168 for_each_netdev(net, dev) { 5169 int err; 5170 char fb_name[IFNAMSIZ]; 5171 5172 /* Ignore unmoveable devices (i.e. loopback) */ 5173 if (dev->features & NETIF_F_NETNS_LOCAL) 5174 continue; 5175 5176 /* Delete virtual devices */ 5177 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) { 5178 dev->rtnl_link_ops->dellink(dev); 5179 goto restart; 5180 } 5181 5182 /* Push remaing network devices to init_net */ 5183 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 5184 err = dev_change_net_namespace(dev, &init_net, fb_name); 5185 if (err) { 5186 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", 5187 __func__, dev->name, err); 5188 BUG(); 5189 } 5190 goto restart; 5191 } 5192 rtnl_unlock(); 5193 } 5194 5195 static struct pernet_operations __net_initdata default_device_ops = { 5196 .exit = default_device_exit, 5197 }; 5198 5199 /* 5200 * Initialize the DEV module. At boot time this walks the device list and 5201 * unhooks any devices that fail to initialise (normally hardware not 5202 * present) and leaves us with a valid list of present and active devices. 5203 * 5204 */ 5205 5206 /* 5207 * This is called single threaded during boot, so no need 5208 * to take the rtnl semaphore. 5209 */ 5210 static int __init net_dev_init(void) 5211 { 5212 int i, rc = -ENOMEM; 5213 5214 BUG_ON(!dev_boot_phase); 5215 5216 if (dev_proc_init()) 5217 goto out; 5218 5219 if (netdev_kobject_init()) 5220 goto out; 5221 5222 INIT_LIST_HEAD(&ptype_all); 5223 for (i = 0; i < PTYPE_HASH_SIZE; i++) 5224 INIT_LIST_HEAD(&ptype_base[i]); 5225 5226 if (register_pernet_subsys(&netdev_net_ops)) 5227 goto out; 5228 5229 /* 5230 * Initialise the packet receive queues. 5231 */ 5232 5233 for_each_possible_cpu(i) { 5234 struct softnet_data *queue; 5235 5236 queue = &per_cpu(softnet_data, i); 5237 skb_queue_head_init(&queue->input_pkt_queue); 5238 queue->completion_queue = NULL; 5239 INIT_LIST_HEAD(&queue->poll_list); 5240 5241 queue->backlog.poll = process_backlog; 5242 queue->backlog.weight = weight_p; 5243 queue->backlog.gro_list = NULL; 5244 queue->backlog.gro_count = 0; 5245 } 5246 5247 dev_boot_phase = 0; 5248 5249 /* The loopback device is special if any other network devices 5250 * is present in a network namespace the loopback device must 5251 * be present. Since we now dynamically allocate and free the 5252 * loopback device ensure this invariant is maintained by 5253 * keeping the loopback device as the first device on the 5254 * list of network devices. Ensuring the loopback devices 5255 * is the first device that appears and the last network device 5256 * that disappears. 5257 */ 5258 if (register_pernet_device(&loopback_net_ops)) 5259 goto out; 5260 5261 if (register_pernet_device(&default_device_ops)) 5262 goto out; 5263 5264 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 5265 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 5266 5267 hotcpu_notifier(dev_cpu_callback, 0); 5268 dst_init(); 5269 dev_mcast_init(); 5270 rc = 0; 5271 out: 5272 return rc; 5273 } 5274 5275 subsys_initcall(net_dev_init); 5276 5277 static int __init initialize_hashrnd(void) 5278 { 5279 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd)); 5280 return 0; 5281 } 5282 5283 late_initcall_sync(initialize_hashrnd); 5284 5285 EXPORT_SYMBOL(__dev_get_by_index); 5286 EXPORT_SYMBOL(__dev_get_by_name); 5287 EXPORT_SYMBOL(__dev_remove_pack); 5288 EXPORT_SYMBOL(dev_valid_name); 5289 EXPORT_SYMBOL(dev_add_pack); 5290 EXPORT_SYMBOL(dev_alloc_name); 5291 EXPORT_SYMBOL(dev_close); 5292 EXPORT_SYMBOL(dev_get_by_flags); 5293 EXPORT_SYMBOL(dev_get_by_index); 5294 EXPORT_SYMBOL(dev_get_by_name); 5295 EXPORT_SYMBOL(dev_open); 5296 EXPORT_SYMBOL(dev_queue_xmit); 5297 EXPORT_SYMBOL(dev_remove_pack); 5298 EXPORT_SYMBOL(dev_set_allmulti); 5299 EXPORT_SYMBOL(dev_set_promiscuity); 5300 EXPORT_SYMBOL(dev_change_flags); 5301 EXPORT_SYMBOL(dev_set_mtu); 5302 EXPORT_SYMBOL(dev_set_mac_address); 5303 EXPORT_SYMBOL(free_netdev); 5304 EXPORT_SYMBOL(netdev_boot_setup_check); 5305 EXPORT_SYMBOL(netdev_set_master); 5306 EXPORT_SYMBOL(netdev_state_change); 5307 EXPORT_SYMBOL(netif_receive_skb); 5308 EXPORT_SYMBOL(netif_rx); 5309 EXPORT_SYMBOL(register_gifconf); 5310 EXPORT_SYMBOL(register_netdevice); 5311 EXPORT_SYMBOL(register_netdevice_notifier); 5312 EXPORT_SYMBOL(skb_checksum_help); 5313 EXPORT_SYMBOL(synchronize_net); 5314 EXPORT_SYMBOL(unregister_netdevice); 5315 EXPORT_SYMBOL(unregister_netdevice_notifier); 5316 EXPORT_SYMBOL(net_enable_timestamp); 5317 EXPORT_SYMBOL(net_disable_timestamp); 5318 EXPORT_SYMBOL(dev_get_flags); 5319 5320 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) 5321 EXPORT_SYMBOL(br_handle_frame_hook); 5322 EXPORT_SYMBOL(br_fdb_get_hook); 5323 EXPORT_SYMBOL(br_fdb_put_hook); 5324 #endif 5325 5326 EXPORT_SYMBOL(dev_load); 5327 5328 EXPORT_PER_CPU_SYMBOL(softnet_data); 5329