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 <linux/bitops.h> 77 #include <linux/capability.h> 78 #include <linux/cpu.h> 79 #include <linux/types.h> 80 #include <linux/kernel.h> 81 #include <linux/hash.h> 82 #include <linux/slab.h> 83 #include <linux/sched.h> 84 #include <linux/mutex.h> 85 #include <linux/string.h> 86 #include <linux/mm.h> 87 #include <linux/socket.h> 88 #include <linux/sockios.h> 89 #include <linux/errno.h> 90 #include <linux/interrupt.h> 91 #include <linux/if_ether.h> 92 #include <linux/netdevice.h> 93 #include <linux/etherdevice.h> 94 #include <linux/ethtool.h> 95 #include <linux/notifier.h> 96 #include <linux/skbuff.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <linux/rtnetlink.h> 100 #include <linux/stat.h> 101 #include <net/dst.h> 102 #include <net/pkt_sched.h> 103 #include <net/checksum.h> 104 #include <net/xfrm.h> 105 #include <linux/highmem.h> 106 #include <linux/init.h> 107 #include <linux/module.h> 108 #include <linux/netpoll.h> 109 #include <linux/rcupdate.h> 110 #include <linux/delay.h> 111 #include <net/iw_handler.h> 112 #include <asm/current.h> 113 #include <linux/audit.h> 114 #include <linux/dmaengine.h> 115 #include <linux/err.h> 116 #include <linux/ctype.h> 117 #include <linux/if_arp.h> 118 #include <linux/if_vlan.h> 119 #include <linux/ip.h> 120 #include <net/ip.h> 121 #include <linux/ipv6.h> 122 #include <linux/in.h> 123 #include <linux/jhash.h> 124 #include <linux/random.h> 125 #include <trace/events/napi.h> 126 #include <trace/events/net.h> 127 #include <trace/events/skb.h> 128 #include <linux/pci.h> 129 #include <linux/inetdevice.h> 130 #include <linux/cpu_rmap.h> 131 #include <linux/static_key.h> 132 #include <linux/hashtable.h> 133 #include <linux/vmalloc.h> 134 #include <linux/if_macvlan.h> 135 #include <linux/errqueue.h> 136 137 #include "net-sysfs.h" 138 139 /* Instead of increasing this, you should create a hash table. */ 140 #define MAX_GRO_SKBS 8 141 142 /* This should be increased if a protocol with a bigger head is added. */ 143 #define GRO_MAX_HEAD (MAX_HEADER + 128) 144 145 static DEFINE_SPINLOCK(ptype_lock); 146 static DEFINE_SPINLOCK(offload_lock); 147 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 148 struct list_head ptype_all __read_mostly; /* Taps */ 149 static struct list_head offload_base __read_mostly; 150 151 static int netif_rx_internal(struct sk_buff *skb); 152 static int call_netdevice_notifiers_info(unsigned long val, 153 struct net_device *dev, 154 struct netdev_notifier_info *info); 155 156 /* 157 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 158 * semaphore. 159 * 160 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 161 * 162 * Writers must hold the rtnl semaphore while they loop through the 163 * dev_base_head list, and hold dev_base_lock for writing when they do the 164 * actual updates. This allows pure readers to access the list even 165 * while a writer is preparing to update it. 166 * 167 * To put it another way, dev_base_lock is held for writing only to 168 * protect against pure readers; the rtnl semaphore provides the 169 * protection against other writers. 170 * 171 * See, for example usages, register_netdevice() and 172 * unregister_netdevice(), which must be called with the rtnl 173 * semaphore held. 174 */ 175 DEFINE_RWLOCK(dev_base_lock); 176 EXPORT_SYMBOL(dev_base_lock); 177 178 /* protects napi_hash addition/deletion and napi_gen_id */ 179 static DEFINE_SPINLOCK(napi_hash_lock); 180 181 static unsigned int napi_gen_id; 182 static DEFINE_HASHTABLE(napi_hash, 8); 183 184 static seqcount_t devnet_rename_seq; 185 186 static inline void dev_base_seq_inc(struct net *net) 187 { 188 while (++net->dev_base_seq == 0); 189 } 190 191 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 192 { 193 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); 194 195 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 196 } 197 198 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 199 { 200 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 201 } 202 203 static inline void rps_lock(struct softnet_data *sd) 204 { 205 #ifdef CONFIG_RPS 206 spin_lock(&sd->input_pkt_queue.lock); 207 #endif 208 } 209 210 static inline void rps_unlock(struct softnet_data *sd) 211 { 212 #ifdef CONFIG_RPS 213 spin_unlock(&sd->input_pkt_queue.lock); 214 #endif 215 } 216 217 /* Device list insertion */ 218 static void list_netdevice(struct net_device *dev) 219 { 220 struct net *net = dev_net(dev); 221 222 ASSERT_RTNL(); 223 224 write_lock_bh(&dev_base_lock); 225 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 226 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 227 hlist_add_head_rcu(&dev->index_hlist, 228 dev_index_hash(net, dev->ifindex)); 229 write_unlock_bh(&dev_base_lock); 230 231 dev_base_seq_inc(net); 232 } 233 234 /* Device list removal 235 * caller must respect a RCU grace period before freeing/reusing dev 236 */ 237 static void unlist_netdevice(struct net_device *dev) 238 { 239 ASSERT_RTNL(); 240 241 /* Unlink dev from the device chain */ 242 write_lock_bh(&dev_base_lock); 243 list_del_rcu(&dev->dev_list); 244 hlist_del_rcu(&dev->name_hlist); 245 hlist_del_rcu(&dev->index_hlist); 246 write_unlock_bh(&dev_base_lock); 247 248 dev_base_seq_inc(dev_net(dev)); 249 } 250 251 /* 252 * Our notifier list 253 */ 254 255 static RAW_NOTIFIER_HEAD(netdev_chain); 256 257 /* 258 * Device drivers call our routines to queue packets here. We empty the 259 * queue in the local softnet handler. 260 */ 261 262 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 263 EXPORT_PER_CPU_SYMBOL(softnet_data); 264 265 #ifdef CONFIG_LOCKDEP 266 /* 267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 268 * according to dev->type 269 */ 270 static const unsigned short netdev_lock_type[] = 271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 283 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 284 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 285 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 286 287 static const char *const netdev_lock_name[] = 288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 300 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 301 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 302 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 303 304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 306 307 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 308 { 309 int i; 310 311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 312 if (netdev_lock_type[i] == dev_type) 313 return i; 314 /* the last key is used by default */ 315 return ARRAY_SIZE(netdev_lock_type) - 1; 316 } 317 318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 319 unsigned short dev_type) 320 { 321 int i; 322 323 i = netdev_lock_pos(dev_type); 324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 325 netdev_lock_name[i]); 326 } 327 328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 329 { 330 int i; 331 332 i = netdev_lock_pos(dev->type); 333 lockdep_set_class_and_name(&dev->addr_list_lock, 334 &netdev_addr_lock_key[i], 335 netdev_lock_name[i]); 336 } 337 #else 338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 339 unsigned short dev_type) 340 { 341 } 342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 343 { 344 } 345 #endif 346 347 /******************************************************************************* 348 349 Protocol management and registration routines 350 351 *******************************************************************************/ 352 353 /* 354 * Add a protocol ID to the list. Now that the input handler is 355 * smarter we can dispense with all the messy stuff that used to be 356 * here. 357 * 358 * BEWARE!!! Protocol handlers, mangling input packets, 359 * MUST BE last in hash buckets and checking protocol handlers 360 * MUST start from promiscuous ptype_all chain in net_bh. 361 * It is true now, do not change it. 362 * Explanation follows: if protocol handler, mangling packet, will 363 * be the first on list, it is not able to sense, that packet 364 * is cloned and should be copied-on-write, so that it will 365 * change it and subsequent readers will get broken packet. 366 * --ANK (980803) 367 */ 368 369 static inline struct list_head *ptype_head(const struct packet_type *pt) 370 { 371 if (pt->type == htons(ETH_P_ALL)) 372 return &ptype_all; 373 else 374 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 375 } 376 377 /** 378 * dev_add_pack - add packet handler 379 * @pt: packet type declaration 380 * 381 * Add a protocol handler to the networking stack. The passed &packet_type 382 * is linked into kernel lists and may not be freed until it has been 383 * removed from the kernel lists. 384 * 385 * This call does not sleep therefore it can not 386 * guarantee all CPU's that are in middle of receiving packets 387 * will see the new packet type (until the next received packet). 388 */ 389 390 void dev_add_pack(struct packet_type *pt) 391 { 392 struct list_head *head = ptype_head(pt); 393 394 spin_lock(&ptype_lock); 395 list_add_rcu(&pt->list, head); 396 spin_unlock(&ptype_lock); 397 } 398 EXPORT_SYMBOL(dev_add_pack); 399 400 /** 401 * __dev_remove_pack - remove packet handler 402 * @pt: packet type declaration 403 * 404 * Remove a protocol handler that was previously added to the kernel 405 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 406 * from the kernel lists and can be freed or reused once this function 407 * returns. 408 * 409 * The packet type might still be in use by receivers 410 * and must not be freed until after all the CPU's have gone 411 * through a quiescent state. 412 */ 413 void __dev_remove_pack(struct packet_type *pt) 414 { 415 struct list_head *head = ptype_head(pt); 416 struct packet_type *pt1; 417 418 spin_lock(&ptype_lock); 419 420 list_for_each_entry(pt1, head, list) { 421 if (pt == pt1) { 422 list_del_rcu(&pt->list); 423 goto out; 424 } 425 } 426 427 pr_warn("dev_remove_pack: %p not found\n", pt); 428 out: 429 spin_unlock(&ptype_lock); 430 } 431 EXPORT_SYMBOL(__dev_remove_pack); 432 433 /** 434 * dev_remove_pack - remove packet handler 435 * @pt: packet type declaration 436 * 437 * Remove a protocol handler that was previously added to the kernel 438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 439 * from the kernel lists and can be freed or reused once this function 440 * returns. 441 * 442 * This call sleeps to guarantee that no CPU is looking at the packet 443 * type after return. 444 */ 445 void dev_remove_pack(struct packet_type *pt) 446 { 447 __dev_remove_pack(pt); 448 449 synchronize_net(); 450 } 451 EXPORT_SYMBOL(dev_remove_pack); 452 453 454 /** 455 * dev_add_offload - register offload handlers 456 * @po: protocol offload declaration 457 * 458 * Add protocol offload handlers to the networking stack. The passed 459 * &proto_offload is linked into kernel lists and may not be freed until 460 * it has been removed from the kernel lists. 461 * 462 * This call does not sleep therefore it can not 463 * guarantee all CPU's that are in middle of receiving packets 464 * will see the new offload handlers (until the next received packet). 465 */ 466 void dev_add_offload(struct packet_offload *po) 467 { 468 struct list_head *head = &offload_base; 469 470 spin_lock(&offload_lock); 471 list_add_rcu(&po->list, head); 472 spin_unlock(&offload_lock); 473 } 474 EXPORT_SYMBOL(dev_add_offload); 475 476 /** 477 * __dev_remove_offload - remove offload handler 478 * @po: packet offload declaration 479 * 480 * Remove a protocol offload handler that was previously added to the 481 * kernel offload handlers by dev_add_offload(). The passed &offload_type 482 * is removed from the kernel lists and can be freed or reused once this 483 * function returns. 484 * 485 * The packet type might still be in use by receivers 486 * and must not be freed until after all the CPU's have gone 487 * through a quiescent state. 488 */ 489 static void __dev_remove_offload(struct packet_offload *po) 490 { 491 struct list_head *head = &offload_base; 492 struct packet_offload *po1; 493 494 spin_lock(&offload_lock); 495 496 list_for_each_entry(po1, head, list) { 497 if (po == po1) { 498 list_del_rcu(&po->list); 499 goto out; 500 } 501 } 502 503 pr_warn("dev_remove_offload: %p not found\n", po); 504 out: 505 spin_unlock(&offload_lock); 506 } 507 508 /** 509 * dev_remove_offload - remove packet offload handler 510 * @po: packet offload declaration 511 * 512 * Remove a packet offload handler that was previously added to the kernel 513 * offload handlers by dev_add_offload(). The passed &offload_type is 514 * removed from the kernel lists and can be freed or reused once this 515 * function returns. 516 * 517 * This call sleeps to guarantee that no CPU is looking at the packet 518 * type after return. 519 */ 520 void dev_remove_offload(struct packet_offload *po) 521 { 522 __dev_remove_offload(po); 523 524 synchronize_net(); 525 } 526 EXPORT_SYMBOL(dev_remove_offload); 527 528 /****************************************************************************** 529 530 Device Boot-time Settings Routines 531 532 *******************************************************************************/ 533 534 /* Boot time configuration table */ 535 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 536 537 /** 538 * netdev_boot_setup_add - add new setup entry 539 * @name: name of the device 540 * @map: configured settings for the device 541 * 542 * Adds new setup entry to the dev_boot_setup list. The function 543 * returns 0 on error and 1 on success. This is a generic routine to 544 * all netdevices. 545 */ 546 static int netdev_boot_setup_add(char *name, struct ifmap *map) 547 { 548 struct netdev_boot_setup *s; 549 int i; 550 551 s = dev_boot_setup; 552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 553 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 554 memset(s[i].name, 0, sizeof(s[i].name)); 555 strlcpy(s[i].name, name, IFNAMSIZ); 556 memcpy(&s[i].map, map, sizeof(s[i].map)); 557 break; 558 } 559 } 560 561 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 562 } 563 564 /** 565 * netdev_boot_setup_check - check boot time settings 566 * @dev: the netdevice 567 * 568 * Check boot time settings for the device. 569 * The found settings are set for the device to be used 570 * later in the device probing. 571 * Returns 0 if no settings found, 1 if they are. 572 */ 573 int netdev_boot_setup_check(struct net_device *dev) 574 { 575 struct netdev_boot_setup *s = dev_boot_setup; 576 int i; 577 578 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 579 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 580 !strcmp(dev->name, s[i].name)) { 581 dev->irq = s[i].map.irq; 582 dev->base_addr = s[i].map.base_addr; 583 dev->mem_start = s[i].map.mem_start; 584 dev->mem_end = s[i].map.mem_end; 585 return 1; 586 } 587 } 588 return 0; 589 } 590 EXPORT_SYMBOL(netdev_boot_setup_check); 591 592 593 /** 594 * netdev_boot_base - get address from boot time settings 595 * @prefix: prefix for network device 596 * @unit: id for network device 597 * 598 * Check boot time settings for the base address of device. 599 * The found settings are set for the device to be used 600 * later in the device probing. 601 * Returns 0 if no settings found. 602 */ 603 unsigned long netdev_boot_base(const char *prefix, int unit) 604 { 605 const struct netdev_boot_setup *s = dev_boot_setup; 606 char name[IFNAMSIZ]; 607 int i; 608 609 sprintf(name, "%s%d", prefix, unit); 610 611 /* 612 * If device already registered then return base of 1 613 * to indicate not to probe for this interface 614 */ 615 if (__dev_get_by_name(&init_net, name)) 616 return 1; 617 618 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 619 if (!strcmp(name, s[i].name)) 620 return s[i].map.base_addr; 621 return 0; 622 } 623 624 /* 625 * Saves at boot time configured settings for any netdevice. 626 */ 627 int __init netdev_boot_setup(char *str) 628 { 629 int ints[5]; 630 struct ifmap map; 631 632 str = get_options(str, ARRAY_SIZE(ints), ints); 633 if (!str || !*str) 634 return 0; 635 636 /* Save settings */ 637 memset(&map, 0, sizeof(map)); 638 if (ints[0] > 0) 639 map.irq = ints[1]; 640 if (ints[0] > 1) 641 map.base_addr = ints[2]; 642 if (ints[0] > 2) 643 map.mem_start = ints[3]; 644 if (ints[0] > 3) 645 map.mem_end = ints[4]; 646 647 /* Add new entry to the list */ 648 return netdev_boot_setup_add(str, &map); 649 } 650 651 __setup("netdev=", netdev_boot_setup); 652 653 /******************************************************************************* 654 655 Device Interface Subroutines 656 657 *******************************************************************************/ 658 659 /** 660 * __dev_get_by_name - find a device by its name 661 * @net: the applicable net namespace 662 * @name: name to find 663 * 664 * Find an interface by name. Must be called under RTNL semaphore 665 * or @dev_base_lock. If the name is found a pointer to the device 666 * is returned. If the name is not found then %NULL is returned. The 667 * reference counters are not incremented so the caller must be 668 * careful with locks. 669 */ 670 671 struct net_device *__dev_get_by_name(struct net *net, const char *name) 672 { 673 struct net_device *dev; 674 struct hlist_head *head = dev_name_hash(net, name); 675 676 hlist_for_each_entry(dev, head, name_hlist) 677 if (!strncmp(dev->name, name, IFNAMSIZ)) 678 return dev; 679 680 return NULL; 681 } 682 EXPORT_SYMBOL(__dev_get_by_name); 683 684 /** 685 * dev_get_by_name_rcu - find a device by its name 686 * @net: the applicable net namespace 687 * @name: name to find 688 * 689 * Find an interface by name. 690 * If the name is found a pointer to the device is returned. 691 * If the name is not found then %NULL is returned. 692 * The reference counters are not incremented so the caller must be 693 * careful with locks. The caller must hold RCU lock. 694 */ 695 696 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 697 { 698 struct net_device *dev; 699 struct hlist_head *head = dev_name_hash(net, name); 700 701 hlist_for_each_entry_rcu(dev, head, name_hlist) 702 if (!strncmp(dev->name, name, IFNAMSIZ)) 703 return dev; 704 705 return NULL; 706 } 707 EXPORT_SYMBOL(dev_get_by_name_rcu); 708 709 /** 710 * dev_get_by_name - find a device by its name 711 * @net: the applicable net namespace 712 * @name: name to find 713 * 714 * Find an interface by name. This can be called from any 715 * context and does its own locking. The returned handle has 716 * the usage count incremented and the caller must use dev_put() to 717 * release it when it is no longer needed. %NULL is returned if no 718 * matching device is found. 719 */ 720 721 struct net_device *dev_get_by_name(struct net *net, const char *name) 722 { 723 struct net_device *dev; 724 725 rcu_read_lock(); 726 dev = dev_get_by_name_rcu(net, name); 727 if (dev) 728 dev_hold(dev); 729 rcu_read_unlock(); 730 return dev; 731 } 732 EXPORT_SYMBOL(dev_get_by_name); 733 734 /** 735 * __dev_get_by_index - find a device by its ifindex 736 * @net: the applicable net namespace 737 * @ifindex: index of device 738 * 739 * Search for an interface by index. Returns %NULL if the device 740 * is not found or a pointer to the device. The device has not 741 * had its reference counter increased so the caller must be careful 742 * about locking. The caller must hold either the RTNL semaphore 743 * or @dev_base_lock. 744 */ 745 746 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 747 { 748 struct net_device *dev; 749 struct hlist_head *head = dev_index_hash(net, ifindex); 750 751 hlist_for_each_entry(dev, head, index_hlist) 752 if (dev->ifindex == ifindex) 753 return dev; 754 755 return NULL; 756 } 757 EXPORT_SYMBOL(__dev_get_by_index); 758 759 /** 760 * dev_get_by_index_rcu - find a device by its ifindex 761 * @net: the applicable net namespace 762 * @ifindex: index of device 763 * 764 * Search for an interface by index. Returns %NULL if the device 765 * is not found or a pointer to the device. The device has not 766 * had its reference counter increased so the caller must be careful 767 * about locking. The caller must hold RCU lock. 768 */ 769 770 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 771 { 772 struct net_device *dev; 773 struct hlist_head *head = dev_index_hash(net, ifindex); 774 775 hlist_for_each_entry_rcu(dev, head, index_hlist) 776 if (dev->ifindex == ifindex) 777 return dev; 778 779 return NULL; 780 } 781 EXPORT_SYMBOL(dev_get_by_index_rcu); 782 783 784 /** 785 * dev_get_by_index - find a device by its ifindex 786 * @net: the applicable net namespace 787 * @ifindex: index of device 788 * 789 * Search for an interface by index. Returns NULL if the device 790 * is not found or a pointer to the device. The device returned has 791 * had a reference added and the pointer is safe until the user calls 792 * dev_put to indicate they have finished with it. 793 */ 794 795 struct net_device *dev_get_by_index(struct net *net, int ifindex) 796 { 797 struct net_device *dev; 798 799 rcu_read_lock(); 800 dev = dev_get_by_index_rcu(net, ifindex); 801 if (dev) 802 dev_hold(dev); 803 rcu_read_unlock(); 804 return dev; 805 } 806 EXPORT_SYMBOL(dev_get_by_index); 807 808 /** 809 * netdev_get_name - get a netdevice name, knowing its ifindex. 810 * @net: network namespace 811 * @name: a pointer to the buffer where the name will be stored. 812 * @ifindex: the ifindex of the interface to get the name from. 813 * 814 * The use of raw_seqcount_begin() and cond_resched() before 815 * retrying is required as we want to give the writers a chance 816 * to complete when CONFIG_PREEMPT is not set. 817 */ 818 int netdev_get_name(struct net *net, char *name, int ifindex) 819 { 820 struct net_device *dev; 821 unsigned int seq; 822 823 retry: 824 seq = raw_seqcount_begin(&devnet_rename_seq); 825 rcu_read_lock(); 826 dev = dev_get_by_index_rcu(net, ifindex); 827 if (!dev) { 828 rcu_read_unlock(); 829 return -ENODEV; 830 } 831 832 strcpy(name, dev->name); 833 rcu_read_unlock(); 834 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 835 cond_resched(); 836 goto retry; 837 } 838 839 return 0; 840 } 841 842 /** 843 * dev_getbyhwaddr_rcu - find a device by its hardware address 844 * @net: the applicable net namespace 845 * @type: media type of device 846 * @ha: hardware address 847 * 848 * Search for an interface by MAC address. Returns NULL if the device 849 * is not found or a pointer to the device. 850 * The caller must hold RCU or RTNL. 851 * The returned device has not had its ref count increased 852 * and the caller must therefore be careful about locking 853 * 854 */ 855 856 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 857 const char *ha) 858 { 859 struct net_device *dev; 860 861 for_each_netdev_rcu(net, dev) 862 if (dev->type == type && 863 !memcmp(dev->dev_addr, ha, dev->addr_len)) 864 return dev; 865 866 return NULL; 867 } 868 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 869 870 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 871 { 872 struct net_device *dev; 873 874 ASSERT_RTNL(); 875 for_each_netdev(net, dev) 876 if (dev->type == type) 877 return dev; 878 879 return NULL; 880 } 881 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 882 883 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 884 { 885 struct net_device *dev, *ret = NULL; 886 887 rcu_read_lock(); 888 for_each_netdev_rcu(net, dev) 889 if (dev->type == type) { 890 dev_hold(dev); 891 ret = dev; 892 break; 893 } 894 rcu_read_unlock(); 895 return ret; 896 } 897 EXPORT_SYMBOL(dev_getfirstbyhwtype); 898 899 /** 900 * __dev_get_by_flags - find any device with given flags 901 * @net: the applicable net namespace 902 * @if_flags: IFF_* values 903 * @mask: bitmask of bits in if_flags to check 904 * 905 * Search for any interface with the given flags. Returns NULL if a device 906 * is not found or a pointer to the device. Must be called inside 907 * rtnl_lock(), and result refcount is unchanged. 908 */ 909 910 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 911 unsigned short mask) 912 { 913 struct net_device *dev, *ret; 914 915 ASSERT_RTNL(); 916 917 ret = NULL; 918 for_each_netdev(net, dev) { 919 if (((dev->flags ^ if_flags) & mask) == 0) { 920 ret = dev; 921 break; 922 } 923 } 924 return ret; 925 } 926 EXPORT_SYMBOL(__dev_get_by_flags); 927 928 /** 929 * dev_valid_name - check if name is okay for network device 930 * @name: name string 931 * 932 * Network device names need to be valid file names to 933 * to allow sysfs to work. We also disallow any kind of 934 * whitespace. 935 */ 936 bool dev_valid_name(const char *name) 937 { 938 if (*name == '\0') 939 return false; 940 if (strlen(name) >= IFNAMSIZ) 941 return false; 942 if (!strcmp(name, ".") || !strcmp(name, "..")) 943 return false; 944 945 while (*name) { 946 if (*name == '/' || isspace(*name)) 947 return false; 948 name++; 949 } 950 return true; 951 } 952 EXPORT_SYMBOL(dev_valid_name); 953 954 /** 955 * __dev_alloc_name - allocate a name for a device 956 * @net: network namespace to allocate the device name in 957 * @name: name format string 958 * @buf: scratch buffer and result name string 959 * 960 * Passed a format string - eg "lt%d" it will try and find a suitable 961 * id. It scans list of devices to build up a free map, then chooses 962 * the first empty slot. The caller must hold the dev_base or rtnl lock 963 * while allocating the name and adding the device in order to avoid 964 * duplicates. 965 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 966 * Returns the number of the unit assigned or a negative errno code. 967 */ 968 969 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 970 { 971 int i = 0; 972 const char *p; 973 const int max_netdevices = 8*PAGE_SIZE; 974 unsigned long *inuse; 975 struct net_device *d; 976 977 p = strnchr(name, IFNAMSIZ-1, '%'); 978 if (p) { 979 /* 980 * Verify the string as this thing may have come from 981 * the user. There must be either one "%d" and no other "%" 982 * characters. 983 */ 984 if (p[1] != 'd' || strchr(p + 2, '%')) 985 return -EINVAL; 986 987 /* Use one page as a bit array of possible slots */ 988 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 989 if (!inuse) 990 return -ENOMEM; 991 992 for_each_netdev(net, d) { 993 if (!sscanf(d->name, name, &i)) 994 continue; 995 if (i < 0 || i >= max_netdevices) 996 continue; 997 998 /* avoid cases where sscanf is not exact inverse of printf */ 999 snprintf(buf, IFNAMSIZ, name, i); 1000 if (!strncmp(buf, d->name, IFNAMSIZ)) 1001 set_bit(i, inuse); 1002 } 1003 1004 i = find_first_zero_bit(inuse, max_netdevices); 1005 free_page((unsigned long) inuse); 1006 } 1007 1008 if (buf != name) 1009 snprintf(buf, IFNAMSIZ, name, i); 1010 if (!__dev_get_by_name(net, buf)) 1011 return i; 1012 1013 /* It is possible to run out of possible slots 1014 * when the name is long and there isn't enough space left 1015 * for the digits, or if all bits are used. 1016 */ 1017 return -ENFILE; 1018 } 1019 1020 /** 1021 * dev_alloc_name - allocate a name for a device 1022 * @dev: device 1023 * @name: name format string 1024 * 1025 * Passed a format string - eg "lt%d" it will try and find a suitable 1026 * id. It scans list of devices to build up a free map, then chooses 1027 * the first empty slot. The caller must hold the dev_base or rtnl lock 1028 * while allocating the name and adding the device in order to avoid 1029 * duplicates. 1030 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1031 * Returns the number of the unit assigned or a negative errno code. 1032 */ 1033 1034 int dev_alloc_name(struct net_device *dev, const char *name) 1035 { 1036 char buf[IFNAMSIZ]; 1037 struct net *net; 1038 int ret; 1039 1040 BUG_ON(!dev_net(dev)); 1041 net = dev_net(dev); 1042 ret = __dev_alloc_name(net, name, buf); 1043 if (ret >= 0) 1044 strlcpy(dev->name, buf, IFNAMSIZ); 1045 return ret; 1046 } 1047 EXPORT_SYMBOL(dev_alloc_name); 1048 1049 static int dev_alloc_name_ns(struct net *net, 1050 struct net_device *dev, 1051 const char *name) 1052 { 1053 char buf[IFNAMSIZ]; 1054 int ret; 1055 1056 ret = __dev_alloc_name(net, name, buf); 1057 if (ret >= 0) 1058 strlcpy(dev->name, buf, IFNAMSIZ); 1059 return ret; 1060 } 1061 1062 static int dev_get_valid_name(struct net *net, 1063 struct net_device *dev, 1064 const char *name) 1065 { 1066 BUG_ON(!net); 1067 1068 if (!dev_valid_name(name)) 1069 return -EINVAL; 1070 1071 if (strchr(name, '%')) 1072 return dev_alloc_name_ns(net, dev, name); 1073 else if (__dev_get_by_name(net, name)) 1074 return -EEXIST; 1075 else if (dev->name != name) 1076 strlcpy(dev->name, name, IFNAMSIZ); 1077 1078 return 0; 1079 } 1080 1081 /** 1082 * dev_change_name - change name of a device 1083 * @dev: device 1084 * @newname: name (or format string) must be at least IFNAMSIZ 1085 * 1086 * Change name of a device, can pass format strings "eth%d". 1087 * for wildcarding. 1088 */ 1089 int dev_change_name(struct net_device *dev, const char *newname) 1090 { 1091 unsigned char old_assign_type; 1092 char oldname[IFNAMSIZ]; 1093 int err = 0; 1094 int ret; 1095 struct net *net; 1096 1097 ASSERT_RTNL(); 1098 BUG_ON(!dev_net(dev)); 1099 1100 net = dev_net(dev); 1101 if (dev->flags & IFF_UP) 1102 return -EBUSY; 1103 1104 write_seqcount_begin(&devnet_rename_seq); 1105 1106 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1107 write_seqcount_end(&devnet_rename_seq); 1108 return 0; 1109 } 1110 1111 memcpy(oldname, dev->name, IFNAMSIZ); 1112 1113 err = dev_get_valid_name(net, dev, newname); 1114 if (err < 0) { 1115 write_seqcount_end(&devnet_rename_seq); 1116 return err; 1117 } 1118 1119 if (oldname[0] && !strchr(oldname, '%')) 1120 netdev_info(dev, "renamed from %s\n", oldname); 1121 1122 old_assign_type = dev->name_assign_type; 1123 dev->name_assign_type = NET_NAME_RENAMED; 1124 1125 rollback: 1126 ret = device_rename(&dev->dev, dev->name); 1127 if (ret) { 1128 memcpy(dev->name, oldname, IFNAMSIZ); 1129 dev->name_assign_type = old_assign_type; 1130 write_seqcount_end(&devnet_rename_seq); 1131 return ret; 1132 } 1133 1134 write_seqcount_end(&devnet_rename_seq); 1135 1136 netdev_adjacent_rename_links(dev, oldname); 1137 1138 write_lock_bh(&dev_base_lock); 1139 hlist_del_rcu(&dev->name_hlist); 1140 write_unlock_bh(&dev_base_lock); 1141 1142 synchronize_rcu(); 1143 1144 write_lock_bh(&dev_base_lock); 1145 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1146 write_unlock_bh(&dev_base_lock); 1147 1148 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1149 ret = notifier_to_errno(ret); 1150 1151 if (ret) { 1152 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1153 if (err >= 0) { 1154 err = ret; 1155 write_seqcount_begin(&devnet_rename_seq); 1156 memcpy(dev->name, oldname, IFNAMSIZ); 1157 memcpy(oldname, newname, IFNAMSIZ); 1158 dev->name_assign_type = old_assign_type; 1159 old_assign_type = NET_NAME_RENAMED; 1160 goto rollback; 1161 } else { 1162 pr_err("%s: name change rollback failed: %d\n", 1163 dev->name, ret); 1164 } 1165 } 1166 1167 return err; 1168 } 1169 1170 /** 1171 * dev_set_alias - change ifalias of a device 1172 * @dev: device 1173 * @alias: name up to IFALIASZ 1174 * @len: limit of bytes to copy from info 1175 * 1176 * Set ifalias for a device, 1177 */ 1178 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1179 { 1180 char *new_ifalias; 1181 1182 ASSERT_RTNL(); 1183 1184 if (len >= IFALIASZ) 1185 return -EINVAL; 1186 1187 if (!len) { 1188 kfree(dev->ifalias); 1189 dev->ifalias = NULL; 1190 return 0; 1191 } 1192 1193 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); 1194 if (!new_ifalias) 1195 return -ENOMEM; 1196 dev->ifalias = new_ifalias; 1197 1198 strlcpy(dev->ifalias, alias, len+1); 1199 return len; 1200 } 1201 1202 1203 /** 1204 * netdev_features_change - device changes features 1205 * @dev: device to cause notification 1206 * 1207 * Called to indicate a device has changed features. 1208 */ 1209 void netdev_features_change(struct net_device *dev) 1210 { 1211 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1212 } 1213 EXPORT_SYMBOL(netdev_features_change); 1214 1215 /** 1216 * netdev_state_change - device changes state 1217 * @dev: device to cause notification 1218 * 1219 * Called to indicate a device has changed state. This function calls 1220 * the notifier chains for netdev_chain and sends a NEWLINK message 1221 * to the routing socket. 1222 */ 1223 void netdev_state_change(struct net_device *dev) 1224 { 1225 if (dev->flags & IFF_UP) { 1226 struct netdev_notifier_change_info change_info; 1227 1228 change_info.flags_changed = 0; 1229 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 1230 &change_info.info); 1231 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1232 } 1233 } 1234 EXPORT_SYMBOL(netdev_state_change); 1235 1236 /** 1237 * netdev_notify_peers - notify network peers about existence of @dev 1238 * @dev: network device 1239 * 1240 * Generate traffic such that interested network peers are aware of 1241 * @dev, such as by generating a gratuitous ARP. This may be used when 1242 * a device wants to inform the rest of the network about some sort of 1243 * reconfiguration such as a failover event or virtual machine 1244 * migration. 1245 */ 1246 void netdev_notify_peers(struct net_device *dev) 1247 { 1248 rtnl_lock(); 1249 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1250 rtnl_unlock(); 1251 } 1252 EXPORT_SYMBOL(netdev_notify_peers); 1253 1254 static int __dev_open(struct net_device *dev) 1255 { 1256 const struct net_device_ops *ops = dev->netdev_ops; 1257 int ret; 1258 1259 ASSERT_RTNL(); 1260 1261 if (!netif_device_present(dev)) 1262 return -ENODEV; 1263 1264 /* Block netpoll from trying to do any rx path servicing. 1265 * If we don't do this there is a chance ndo_poll_controller 1266 * or ndo_poll may be running while we open the device 1267 */ 1268 netpoll_poll_disable(dev); 1269 1270 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); 1271 ret = notifier_to_errno(ret); 1272 if (ret) 1273 return ret; 1274 1275 set_bit(__LINK_STATE_START, &dev->state); 1276 1277 if (ops->ndo_validate_addr) 1278 ret = ops->ndo_validate_addr(dev); 1279 1280 if (!ret && ops->ndo_open) 1281 ret = ops->ndo_open(dev); 1282 1283 netpoll_poll_enable(dev); 1284 1285 if (ret) 1286 clear_bit(__LINK_STATE_START, &dev->state); 1287 else { 1288 dev->flags |= IFF_UP; 1289 dev_set_rx_mode(dev); 1290 dev_activate(dev); 1291 add_device_randomness(dev->dev_addr, dev->addr_len); 1292 } 1293 1294 return ret; 1295 } 1296 1297 /** 1298 * dev_open - prepare an interface for use. 1299 * @dev: device to open 1300 * 1301 * Takes a device from down to up state. The device's private open 1302 * function is invoked and then the multicast lists are loaded. Finally 1303 * the device is moved into the up state and a %NETDEV_UP message is 1304 * sent to the netdev notifier chain. 1305 * 1306 * Calling this function on an active interface is a nop. On a failure 1307 * a negative errno code is returned. 1308 */ 1309 int dev_open(struct net_device *dev) 1310 { 1311 int ret; 1312 1313 if (dev->flags & IFF_UP) 1314 return 0; 1315 1316 ret = __dev_open(dev); 1317 if (ret < 0) 1318 return ret; 1319 1320 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1321 call_netdevice_notifiers(NETDEV_UP, dev); 1322 1323 return ret; 1324 } 1325 EXPORT_SYMBOL(dev_open); 1326 1327 static int __dev_close_many(struct list_head *head) 1328 { 1329 struct net_device *dev; 1330 1331 ASSERT_RTNL(); 1332 might_sleep(); 1333 1334 list_for_each_entry(dev, head, close_list) { 1335 /* Temporarily disable netpoll until the interface is down */ 1336 netpoll_poll_disable(dev); 1337 1338 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1339 1340 clear_bit(__LINK_STATE_START, &dev->state); 1341 1342 /* Synchronize to scheduled poll. We cannot touch poll list, it 1343 * can be even on different cpu. So just clear netif_running(). 1344 * 1345 * dev->stop() will invoke napi_disable() on all of it's 1346 * napi_struct instances on this device. 1347 */ 1348 smp_mb__after_atomic(); /* Commit netif_running(). */ 1349 } 1350 1351 dev_deactivate_many(head); 1352 1353 list_for_each_entry(dev, head, close_list) { 1354 const struct net_device_ops *ops = dev->netdev_ops; 1355 1356 /* 1357 * Call the device specific close. This cannot fail. 1358 * Only if device is UP 1359 * 1360 * We allow it to be called even after a DETACH hot-plug 1361 * event. 1362 */ 1363 if (ops->ndo_stop) 1364 ops->ndo_stop(dev); 1365 1366 dev->flags &= ~IFF_UP; 1367 netpoll_poll_enable(dev); 1368 } 1369 1370 return 0; 1371 } 1372 1373 static int __dev_close(struct net_device *dev) 1374 { 1375 int retval; 1376 LIST_HEAD(single); 1377 1378 list_add(&dev->close_list, &single); 1379 retval = __dev_close_many(&single); 1380 list_del(&single); 1381 1382 return retval; 1383 } 1384 1385 static int dev_close_many(struct list_head *head) 1386 { 1387 struct net_device *dev, *tmp; 1388 1389 /* Remove the devices that don't need to be closed */ 1390 list_for_each_entry_safe(dev, tmp, head, close_list) 1391 if (!(dev->flags & IFF_UP)) 1392 list_del_init(&dev->close_list); 1393 1394 __dev_close_many(head); 1395 1396 list_for_each_entry_safe(dev, tmp, head, close_list) { 1397 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1398 call_netdevice_notifiers(NETDEV_DOWN, dev); 1399 list_del_init(&dev->close_list); 1400 } 1401 1402 return 0; 1403 } 1404 1405 /** 1406 * dev_close - shutdown an interface. 1407 * @dev: device to shutdown 1408 * 1409 * This function moves an active device into down state. A 1410 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1411 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1412 * chain. 1413 */ 1414 int dev_close(struct net_device *dev) 1415 { 1416 if (dev->flags & IFF_UP) { 1417 LIST_HEAD(single); 1418 1419 list_add(&dev->close_list, &single); 1420 dev_close_many(&single); 1421 list_del(&single); 1422 } 1423 return 0; 1424 } 1425 EXPORT_SYMBOL(dev_close); 1426 1427 1428 /** 1429 * dev_disable_lro - disable Large Receive Offload on a device 1430 * @dev: device 1431 * 1432 * Disable Large Receive Offload (LRO) on a net device. Must be 1433 * called under RTNL. This is needed if received packets may be 1434 * forwarded to another interface. 1435 */ 1436 void dev_disable_lro(struct net_device *dev) 1437 { 1438 /* 1439 * If we're trying to disable lro on a vlan device 1440 * use the underlying physical device instead 1441 */ 1442 if (is_vlan_dev(dev)) 1443 dev = vlan_dev_real_dev(dev); 1444 1445 /* the same for macvlan devices */ 1446 if (netif_is_macvlan(dev)) 1447 dev = macvlan_dev_real_dev(dev); 1448 1449 dev->wanted_features &= ~NETIF_F_LRO; 1450 netdev_update_features(dev); 1451 1452 if (unlikely(dev->features & NETIF_F_LRO)) 1453 netdev_WARN(dev, "failed to disable LRO!\n"); 1454 } 1455 EXPORT_SYMBOL(dev_disable_lro); 1456 1457 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1458 struct net_device *dev) 1459 { 1460 struct netdev_notifier_info info; 1461 1462 netdev_notifier_info_init(&info, dev); 1463 return nb->notifier_call(nb, val, &info); 1464 } 1465 1466 static int dev_boot_phase = 1; 1467 1468 /** 1469 * register_netdevice_notifier - register a network notifier block 1470 * @nb: notifier 1471 * 1472 * Register a notifier to be called when network device events occur. 1473 * The notifier passed is linked into the kernel structures and must 1474 * not be reused until it has been unregistered. A negative errno code 1475 * is returned on a failure. 1476 * 1477 * When registered all registration and up events are replayed 1478 * to the new notifier to allow device to have a race free 1479 * view of the network device list. 1480 */ 1481 1482 int register_netdevice_notifier(struct notifier_block *nb) 1483 { 1484 struct net_device *dev; 1485 struct net_device *last; 1486 struct net *net; 1487 int err; 1488 1489 rtnl_lock(); 1490 err = raw_notifier_chain_register(&netdev_chain, nb); 1491 if (err) 1492 goto unlock; 1493 if (dev_boot_phase) 1494 goto unlock; 1495 for_each_net(net) { 1496 for_each_netdev(net, dev) { 1497 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1498 err = notifier_to_errno(err); 1499 if (err) 1500 goto rollback; 1501 1502 if (!(dev->flags & IFF_UP)) 1503 continue; 1504 1505 call_netdevice_notifier(nb, NETDEV_UP, dev); 1506 } 1507 } 1508 1509 unlock: 1510 rtnl_unlock(); 1511 return err; 1512 1513 rollback: 1514 last = dev; 1515 for_each_net(net) { 1516 for_each_netdev(net, dev) { 1517 if (dev == last) 1518 goto outroll; 1519 1520 if (dev->flags & IFF_UP) { 1521 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1522 dev); 1523 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1524 } 1525 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1526 } 1527 } 1528 1529 outroll: 1530 raw_notifier_chain_unregister(&netdev_chain, nb); 1531 goto unlock; 1532 } 1533 EXPORT_SYMBOL(register_netdevice_notifier); 1534 1535 /** 1536 * unregister_netdevice_notifier - unregister a network notifier block 1537 * @nb: notifier 1538 * 1539 * Unregister a notifier previously registered by 1540 * register_netdevice_notifier(). The notifier is unlinked into the 1541 * kernel structures and may then be reused. A negative errno code 1542 * is returned on a failure. 1543 * 1544 * After unregistering unregister and down device events are synthesized 1545 * for all devices on the device list to the removed notifier to remove 1546 * the need for special case cleanup code. 1547 */ 1548 1549 int unregister_netdevice_notifier(struct notifier_block *nb) 1550 { 1551 struct net_device *dev; 1552 struct net *net; 1553 int err; 1554 1555 rtnl_lock(); 1556 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1557 if (err) 1558 goto unlock; 1559 1560 for_each_net(net) { 1561 for_each_netdev(net, dev) { 1562 if (dev->flags & IFF_UP) { 1563 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1564 dev); 1565 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1566 } 1567 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1568 } 1569 } 1570 unlock: 1571 rtnl_unlock(); 1572 return err; 1573 } 1574 EXPORT_SYMBOL(unregister_netdevice_notifier); 1575 1576 /** 1577 * call_netdevice_notifiers_info - call all network notifier blocks 1578 * @val: value passed unmodified to notifier function 1579 * @dev: net_device pointer passed unmodified to notifier function 1580 * @info: notifier information data 1581 * 1582 * Call all network notifier blocks. Parameters and return value 1583 * are as for raw_notifier_call_chain(). 1584 */ 1585 1586 static int call_netdevice_notifiers_info(unsigned long val, 1587 struct net_device *dev, 1588 struct netdev_notifier_info *info) 1589 { 1590 ASSERT_RTNL(); 1591 netdev_notifier_info_init(info, dev); 1592 return raw_notifier_call_chain(&netdev_chain, val, info); 1593 } 1594 1595 /** 1596 * call_netdevice_notifiers - call all network notifier blocks 1597 * @val: value passed unmodified to notifier function 1598 * @dev: net_device pointer passed unmodified to notifier function 1599 * 1600 * Call all network notifier blocks. Parameters and return value 1601 * are as for raw_notifier_call_chain(). 1602 */ 1603 1604 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1605 { 1606 struct netdev_notifier_info info; 1607 1608 return call_netdevice_notifiers_info(val, dev, &info); 1609 } 1610 EXPORT_SYMBOL(call_netdevice_notifiers); 1611 1612 static struct static_key netstamp_needed __read_mostly; 1613 #ifdef HAVE_JUMP_LABEL 1614 /* We are not allowed to call static_key_slow_dec() from irq context 1615 * If net_disable_timestamp() is called from irq context, defer the 1616 * static_key_slow_dec() calls. 1617 */ 1618 static atomic_t netstamp_needed_deferred; 1619 #endif 1620 1621 void net_enable_timestamp(void) 1622 { 1623 #ifdef HAVE_JUMP_LABEL 1624 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1625 1626 if (deferred) { 1627 while (--deferred) 1628 static_key_slow_dec(&netstamp_needed); 1629 return; 1630 } 1631 #endif 1632 static_key_slow_inc(&netstamp_needed); 1633 } 1634 EXPORT_SYMBOL(net_enable_timestamp); 1635 1636 void net_disable_timestamp(void) 1637 { 1638 #ifdef HAVE_JUMP_LABEL 1639 if (in_interrupt()) { 1640 atomic_inc(&netstamp_needed_deferred); 1641 return; 1642 } 1643 #endif 1644 static_key_slow_dec(&netstamp_needed); 1645 } 1646 EXPORT_SYMBOL(net_disable_timestamp); 1647 1648 static inline void net_timestamp_set(struct sk_buff *skb) 1649 { 1650 skb->tstamp.tv64 = 0; 1651 if (static_key_false(&netstamp_needed)) 1652 __net_timestamp(skb); 1653 } 1654 1655 #define net_timestamp_check(COND, SKB) \ 1656 if (static_key_false(&netstamp_needed)) { \ 1657 if ((COND) && !(SKB)->tstamp.tv64) \ 1658 __net_timestamp(SKB); \ 1659 } \ 1660 1661 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) 1662 { 1663 unsigned int len; 1664 1665 if (!(dev->flags & IFF_UP)) 1666 return false; 1667 1668 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1669 if (skb->len <= len) 1670 return true; 1671 1672 /* if TSO is enabled, we don't care about the length as the packet 1673 * could be forwarded without being segmented before 1674 */ 1675 if (skb_is_gso(skb)) 1676 return true; 1677 1678 return false; 1679 } 1680 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1681 1682 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1683 { 1684 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 1685 if (skb_copy_ubufs(skb, GFP_ATOMIC)) { 1686 atomic_long_inc(&dev->rx_dropped); 1687 kfree_skb(skb); 1688 return NET_RX_DROP; 1689 } 1690 } 1691 1692 if (unlikely(!is_skb_forwardable(dev, skb))) { 1693 atomic_long_inc(&dev->rx_dropped); 1694 kfree_skb(skb); 1695 return NET_RX_DROP; 1696 } 1697 1698 skb_scrub_packet(skb, true); 1699 skb->protocol = eth_type_trans(skb, dev); 1700 1701 return 0; 1702 } 1703 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1704 1705 /** 1706 * dev_forward_skb - loopback an skb to another netif 1707 * 1708 * @dev: destination network device 1709 * @skb: buffer to forward 1710 * 1711 * return values: 1712 * NET_RX_SUCCESS (no congestion) 1713 * NET_RX_DROP (packet was dropped, but freed) 1714 * 1715 * dev_forward_skb can be used for injecting an skb from the 1716 * start_xmit function of one device into the receive queue 1717 * of another device. 1718 * 1719 * The receiving device may be in another namespace, so 1720 * we have to clear all information in the skb that could 1721 * impact namespace isolation. 1722 */ 1723 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1724 { 1725 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1726 } 1727 EXPORT_SYMBOL_GPL(dev_forward_skb); 1728 1729 static inline int deliver_skb(struct sk_buff *skb, 1730 struct packet_type *pt_prev, 1731 struct net_device *orig_dev) 1732 { 1733 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 1734 return -ENOMEM; 1735 atomic_inc(&skb->users); 1736 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1737 } 1738 1739 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1740 { 1741 if (!ptype->af_packet_priv || !skb->sk) 1742 return false; 1743 1744 if (ptype->id_match) 1745 return ptype->id_match(ptype, skb->sk); 1746 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1747 return true; 1748 1749 return false; 1750 } 1751 1752 /* 1753 * Support routine. Sends outgoing frames to any network 1754 * taps currently in use. 1755 */ 1756 1757 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1758 { 1759 struct packet_type *ptype; 1760 struct sk_buff *skb2 = NULL; 1761 struct packet_type *pt_prev = NULL; 1762 1763 rcu_read_lock(); 1764 list_for_each_entry_rcu(ptype, &ptype_all, list) { 1765 /* Never send packets back to the socket 1766 * they originated from - MvS (miquels@drinkel.ow.org) 1767 */ 1768 if ((ptype->dev == dev || !ptype->dev) && 1769 (!skb_loop_sk(ptype, skb))) { 1770 if (pt_prev) { 1771 deliver_skb(skb2, pt_prev, skb->dev); 1772 pt_prev = ptype; 1773 continue; 1774 } 1775 1776 skb2 = skb_clone(skb, GFP_ATOMIC); 1777 if (!skb2) 1778 break; 1779 1780 net_timestamp_set(skb2); 1781 1782 /* skb->nh should be correctly 1783 set by sender, so that the second statement is 1784 just protection against buggy protocols. 1785 */ 1786 skb_reset_mac_header(skb2); 1787 1788 if (skb_network_header(skb2) < skb2->data || 1789 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 1790 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 1791 ntohs(skb2->protocol), 1792 dev->name); 1793 skb_reset_network_header(skb2); 1794 } 1795 1796 skb2->transport_header = skb2->network_header; 1797 skb2->pkt_type = PACKET_OUTGOING; 1798 pt_prev = ptype; 1799 } 1800 } 1801 if (pt_prev) 1802 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 1803 rcu_read_unlock(); 1804 } 1805 1806 /** 1807 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 1808 * @dev: Network device 1809 * @txq: number of queues available 1810 * 1811 * If real_num_tx_queues is changed the tc mappings may no longer be 1812 * valid. To resolve this verify the tc mapping remains valid and if 1813 * not NULL the mapping. With no priorities mapping to this 1814 * offset/count pair it will no longer be used. In the worst case TC0 1815 * is invalid nothing can be done so disable priority mappings. If is 1816 * expected that drivers will fix this mapping if they can before 1817 * calling netif_set_real_num_tx_queues. 1818 */ 1819 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 1820 { 1821 int i; 1822 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 1823 1824 /* If TC0 is invalidated disable TC mapping */ 1825 if (tc->offset + tc->count > txq) { 1826 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 1827 dev->num_tc = 0; 1828 return; 1829 } 1830 1831 /* Invalidated prio to tc mappings set to TC0 */ 1832 for (i = 1; i < TC_BITMASK + 1; i++) { 1833 int q = netdev_get_prio_tc_map(dev, i); 1834 1835 tc = &dev->tc_to_txq[q]; 1836 if (tc->offset + tc->count > txq) { 1837 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 1838 i, q); 1839 netdev_set_prio_tc_map(dev, i, 0); 1840 } 1841 } 1842 } 1843 1844 #ifdef CONFIG_XPS 1845 static DEFINE_MUTEX(xps_map_mutex); 1846 #define xmap_dereference(P) \ 1847 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 1848 1849 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, 1850 int cpu, u16 index) 1851 { 1852 struct xps_map *map = NULL; 1853 int pos; 1854 1855 if (dev_maps) 1856 map = xmap_dereference(dev_maps->cpu_map[cpu]); 1857 1858 for (pos = 0; map && pos < map->len; pos++) { 1859 if (map->queues[pos] == index) { 1860 if (map->len > 1) { 1861 map->queues[pos] = map->queues[--map->len]; 1862 } else { 1863 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); 1864 kfree_rcu(map, rcu); 1865 map = NULL; 1866 } 1867 break; 1868 } 1869 } 1870 1871 return map; 1872 } 1873 1874 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 1875 { 1876 struct xps_dev_maps *dev_maps; 1877 int cpu, i; 1878 bool active = false; 1879 1880 mutex_lock(&xps_map_mutex); 1881 dev_maps = xmap_dereference(dev->xps_maps); 1882 1883 if (!dev_maps) 1884 goto out_no_maps; 1885 1886 for_each_possible_cpu(cpu) { 1887 for (i = index; i < dev->num_tx_queues; i++) { 1888 if (!remove_xps_queue(dev_maps, cpu, i)) 1889 break; 1890 } 1891 if (i == dev->num_tx_queues) 1892 active = true; 1893 } 1894 1895 if (!active) { 1896 RCU_INIT_POINTER(dev->xps_maps, NULL); 1897 kfree_rcu(dev_maps, rcu); 1898 } 1899 1900 for (i = index; i < dev->num_tx_queues; i++) 1901 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), 1902 NUMA_NO_NODE); 1903 1904 out_no_maps: 1905 mutex_unlock(&xps_map_mutex); 1906 } 1907 1908 static struct xps_map *expand_xps_map(struct xps_map *map, 1909 int cpu, u16 index) 1910 { 1911 struct xps_map *new_map; 1912 int alloc_len = XPS_MIN_MAP_ALLOC; 1913 int i, pos; 1914 1915 for (pos = 0; map && pos < map->len; pos++) { 1916 if (map->queues[pos] != index) 1917 continue; 1918 return map; 1919 } 1920 1921 /* Need to add queue to this CPU's existing map */ 1922 if (map) { 1923 if (pos < map->alloc_len) 1924 return map; 1925 1926 alloc_len = map->alloc_len * 2; 1927 } 1928 1929 /* Need to allocate new map to store queue on this CPU's map */ 1930 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 1931 cpu_to_node(cpu)); 1932 if (!new_map) 1933 return NULL; 1934 1935 for (i = 0; i < pos; i++) 1936 new_map->queues[i] = map->queues[i]; 1937 new_map->alloc_len = alloc_len; 1938 new_map->len = pos; 1939 1940 return new_map; 1941 } 1942 1943 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 1944 u16 index) 1945 { 1946 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 1947 struct xps_map *map, *new_map; 1948 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); 1949 int cpu, numa_node_id = -2; 1950 bool active = false; 1951 1952 mutex_lock(&xps_map_mutex); 1953 1954 dev_maps = xmap_dereference(dev->xps_maps); 1955 1956 /* allocate memory for queue storage */ 1957 for_each_online_cpu(cpu) { 1958 if (!cpumask_test_cpu(cpu, mask)) 1959 continue; 1960 1961 if (!new_dev_maps) 1962 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 1963 if (!new_dev_maps) { 1964 mutex_unlock(&xps_map_mutex); 1965 return -ENOMEM; 1966 } 1967 1968 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : 1969 NULL; 1970 1971 map = expand_xps_map(map, cpu, index); 1972 if (!map) 1973 goto error; 1974 1975 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); 1976 } 1977 1978 if (!new_dev_maps) 1979 goto out_no_new_maps; 1980 1981 for_each_possible_cpu(cpu) { 1982 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { 1983 /* add queue to CPU maps */ 1984 int pos = 0; 1985 1986 map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 1987 while ((pos < map->len) && (map->queues[pos] != index)) 1988 pos++; 1989 1990 if (pos == map->len) 1991 map->queues[map->len++] = index; 1992 #ifdef CONFIG_NUMA 1993 if (numa_node_id == -2) 1994 numa_node_id = cpu_to_node(cpu); 1995 else if (numa_node_id != cpu_to_node(cpu)) 1996 numa_node_id = -1; 1997 #endif 1998 } else if (dev_maps) { 1999 /* fill in the new device map from the old device map */ 2000 map = xmap_dereference(dev_maps->cpu_map[cpu]); 2001 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); 2002 } 2003 2004 } 2005 2006 rcu_assign_pointer(dev->xps_maps, new_dev_maps); 2007 2008 /* Cleanup old maps */ 2009 if (dev_maps) { 2010 for_each_possible_cpu(cpu) { 2011 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2012 map = xmap_dereference(dev_maps->cpu_map[cpu]); 2013 if (map && map != new_map) 2014 kfree_rcu(map, rcu); 2015 } 2016 2017 kfree_rcu(dev_maps, rcu); 2018 } 2019 2020 dev_maps = new_dev_maps; 2021 active = true; 2022 2023 out_no_new_maps: 2024 /* update Tx queue numa node */ 2025 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2026 (numa_node_id >= 0) ? numa_node_id : 2027 NUMA_NO_NODE); 2028 2029 if (!dev_maps) 2030 goto out_no_maps; 2031 2032 /* removes queue from unused CPUs */ 2033 for_each_possible_cpu(cpu) { 2034 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) 2035 continue; 2036 2037 if (remove_xps_queue(dev_maps, cpu, index)) 2038 active = true; 2039 } 2040 2041 /* free map if not active */ 2042 if (!active) { 2043 RCU_INIT_POINTER(dev->xps_maps, NULL); 2044 kfree_rcu(dev_maps, rcu); 2045 } 2046 2047 out_no_maps: 2048 mutex_unlock(&xps_map_mutex); 2049 2050 return 0; 2051 error: 2052 /* remove any maps that we added */ 2053 for_each_possible_cpu(cpu) { 2054 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2055 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : 2056 NULL; 2057 if (new_map && new_map != map) 2058 kfree(new_map); 2059 } 2060 2061 mutex_unlock(&xps_map_mutex); 2062 2063 kfree(new_dev_maps); 2064 return -ENOMEM; 2065 } 2066 EXPORT_SYMBOL(netif_set_xps_queue); 2067 2068 #endif 2069 /* 2070 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2071 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. 2072 */ 2073 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2074 { 2075 int rc; 2076 2077 if (txq < 1 || txq > dev->num_tx_queues) 2078 return -EINVAL; 2079 2080 if (dev->reg_state == NETREG_REGISTERED || 2081 dev->reg_state == NETREG_UNREGISTERING) { 2082 ASSERT_RTNL(); 2083 2084 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2085 txq); 2086 if (rc) 2087 return rc; 2088 2089 if (dev->num_tc) 2090 netif_setup_tc(dev, txq); 2091 2092 if (txq < dev->real_num_tx_queues) { 2093 qdisc_reset_all_tx_gt(dev, txq); 2094 #ifdef CONFIG_XPS 2095 netif_reset_xps_queues_gt(dev, txq); 2096 #endif 2097 } 2098 } 2099 2100 dev->real_num_tx_queues = txq; 2101 return 0; 2102 } 2103 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2104 2105 #ifdef CONFIG_SYSFS 2106 /** 2107 * netif_set_real_num_rx_queues - set actual number of RX queues used 2108 * @dev: Network device 2109 * @rxq: Actual number of RX queues 2110 * 2111 * This must be called either with the rtnl_lock held or before 2112 * registration of the net device. Returns 0 on success, or a 2113 * negative error code. If called before registration, it always 2114 * succeeds. 2115 */ 2116 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2117 { 2118 int rc; 2119 2120 if (rxq < 1 || rxq > dev->num_rx_queues) 2121 return -EINVAL; 2122 2123 if (dev->reg_state == NETREG_REGISTERED) { 2124 ASSERT_RTNL(); 2125 2126 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2127 rxq); 2128 if (rc) 2129 return rc; 2130 } 2131 2132 dev->real_num_rx_queues = rxq; 2133 return 0; 2134 } 2135 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2136 #endif 2137 2138 /** 2139 * netif_get_num_default_rss_queues - default number of RSS queues 2140 * 2141 * This routine should set an upper limit on the number of RSS queues 2142 * used by default by multiqueue devices. 2143 */ 2144 int netif_get_num_default_rss_queues(void) 2145 { 2146 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2147 } 2148 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2149 2150 static inline void __netif_reschedule(struct Qdisc *q) 2151 { 2152 struct softnet_data *sd; 2153 unsigned long flags; 2154 2155 local_irq_save(flags); 2156 sd = this_cpu_ptr(&softnet_data); 2157 q->next_sched = NULL; 2158 *sd->output_queue_tailp = q; 2159 sd->output_queue_tailp = &q->next_sched; 2160 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2161 local_irq_restore(flags); 2162 } 2163 2164 void __netif_schedule(struct Qdisc *q) 2165 { 2166 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2167 __netif_reschedule(q); 2168 } 2169 EXPORT_SYMBOL(__netif_schedule); 2170 2171 struct dev_kfree_skb_cb { 2172 enum skb_free_reason reason; 2173 }; 2174 2175 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2176 { 2177 return (struct dev_kfree_skb_cb *)skb->cb; 2178 } 2179 2180 void netif_schedule_queue(struct netdev_queue *txq) 2181 { 2182 rcu_read_lock(); 2183 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2184 struct Qdisc *q = rcu_dereference(txq->qdisc); 2185 2186 __netif_schedule(q); 2187 } 2188 rcu_read_unlock(); 2189 } 2190 EXPORT_SYMBOL(netif_schedule_queue); 2191 2192 /** 2193 * netif_wake_subqueue - allow sending packets on subqueue 2194 * @dev: network device 2195 * @queue_index: sub queue index 2196 * 2197 * Resume individual transmit queue of a device with multiple transmit queues. 2198 */ 2199 void netif_wake_subqueue(struct net_device *dev, u16 queue_index) 2200 { 2201 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); 2202 2203 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { 2204 struct Qdisc *q; 2205 2206 rcu_read_lock(); 2207 q = rcu_dereference(txq->qdisc); 2208 __netif_schedule(q); 2209 rcu_read_unlock(); 2210 } 2211 } 2212 EXPORT_SYMBOL(netif_wake_subqueue); 2213 2214 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2215 { 2216 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2217 struct Qdisc *q; 2218 2219 rcu_read_lock(); 2220 q = rcu_dereference(dev_queue->qdisc); 2221 __netif_schedule(q); 2222 rcu_read_unlock(); 2223 } 2224 } 2225 EXPORT_SYMBOL(netif_tx_wake_queue); 2226 2227 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2228 { 2229 unsigned long flags; 2230 2231 if (likely(atomic_read(&skb->users) == 1)) { 2232 smp_rmb(); 2233 atomic_set(&skb->users, 0); 2234 } else if (likely(!atomic_dec_and_test(&skb->users))) { 2235 return; 2236 } 2237 get_kfree_skb_cb(skb)->reason = reason; 2238 local_irq_save(flags); 2239 skb->next = __this_cpu_read(softnet_data.completion_queue); 2240 __this_cpu_write(softnet_data.completion_queue, skb); 2241 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2242 local_irq_restore(flags); 2243 } 2244 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2245 2246 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2247 { 2248 if (in_irq() || irqs_disabled()) 2249 __dev_kfree_skb_irq(skb, reason); 2250 else 2251 dev_kfree_skb(skb); 2252 } 2253 EXPORT_SYMBOL(__dev_kfree_skb_any); 2254 2255 2256 /** 2257 * netif_device_detach - mark device as removed 2258 * @dev: network device 2259 * 2260 * Mark device as removed from system and therefore no longer available. 2261 */ 2262 void netif_device_detach(struct net_device *dev) 2263 { 2264 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2265 netif_running(dev)) { 2266 netif_tx_stop_all_queues(dev); 2267 } 2268 } 2269 EXPORT_SYMBOL(netif_device_detach); 2270 2271 /** 2272 * netif_device_attach - mark device as attached 2273 * @dev: network device 2274 * 2275 * Mark device as attached from system and restart if needed. 2276 */ 2277 void netif_device_attach(struct net_device *dev) 2278 { 2279 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2280 netif_running(dev)) { 2281 netif_tx_wake_all_queues(dev); 2282 __netdev_watchdog_up(dev); 2283 } 2284 } 2285 EXPORT_SYMBOL(netif_device_attach); 2286 2287 static void skb_warn_bad_offload(const struct sk_buff *skb) 2288 { 2289 static const netdev_features_t null_features = 0; 2290 struct net_device *dev = skb->dev; 2291 const char *driver = ""; 2292 2293 if (!net_ratelimit()) 2294 return; 2295 2296 if (dev && dev->dev.parent) 2297 driver = dev_driver_string(dev->dev.parent); 2298 2299 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2300 "gso_type=%d ip_summed=%d\n", 2301 driver, dev ? &dev->features : &null_features, 2302 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2303 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2304 skb_shinfo(skb)->gso_type, skb->ip_summed); 2305 } 2306 2307 /* 2308 * Invalidate hardware checksum when packet is to be mangled, and 2309 * complete checksum manually on outgoing path. 2310 */ 2311 int skb_checksum_help(struct sk_buff *skb) 2312 { 2313 __wsum csum; 2314 int ret = 0, offset; 2315 2316 if (skb->ip_summed == CHECKSUM_COMPLETE) 2317 goto out_set_summed; 2318 2319 if (unlikely(skb_shinfo(skb)->gso_size)) { 2320 skb_warn_bad_offload(skb); 2321 return -EINVAL; 2322 } 2323 2324 /* Before computing a checksum, we should make sure no frag could 2325 * be modified by an external entity : checksum could be wrong. 2326 */ 2327 if (skb_has_shared_frag(skb)) { 2328 ret = __skb_linearize(skb); 2329 if (ret) 2330 goto out; 2331 } 2332 2333 offset = skb_checksum_start_offset(skb); 2334 BUG_ON(offset >= skb_headlen(skb)); 2335 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2336 2337 offset += skb->csum_offset; 2338 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2339 2340 if (skb_cloned(skb) && 2341 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2342 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2343 if (ret) 2344 goto out; 2345 } 2346 2347 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2348 out_set_summed: 2349 skb->ip_summed = CHECKSUM_NONE; 2350 out: 2351 return ret; 2352 } 2353 EXPORT_SYMBOL(skb_checksum_help); 2354 2355 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2356 { 2357 unsigned int vlan_depth = skb->mac_len; 2358 __be16 type = skb->protocol; 2359 2360 /* Tunnel gso handlers can set protocol to ethernet. */ 2361 if (type == htons(ETH_P_TEB)) { 2362 struct ethhdr *eth; 2363 2364 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2365 return 0; 2366 2367 eth = (struct ethhdr *)skb_mac_header(skb); 2368 type = eth->h_proto; 2369 } 2370 2371 /* if skb->protocol is 802.1Q/AD then the header should already be 2372 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at 2373 * ETH_HLEN otherwise 2374 */ 2375 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) { 2376 if (vlan_depth) { 2377 if (WARN_ON(vlan_depth < VLAN_HLEN)) 2378 return 0; 2379 vlan_depth -= VLAN_HLEN; 2380 } else { 2381 vlan_depth = ETH_HLEN; 2382 } 2383 do { 2384 struct vlan_hdr *vh; 2385 2386 if (unlikely(!pskb_may_pull(skb, 2387 vlan_depth + VLAN_HLEN))) 2388 return 0; 2389 2390 vh = (struct vlan_hdr *)(skb->data + vlan_depth); 2391 type = vh->h_vlan_encapsulated_proto; 2392 vlan_depth += VLAN_HLEN; 2393 } while (type == htons(ETH_P_8021Q) || 2394 type == htons(ETH_P_8021AD)); 2395 } 2396 2397 *depth = vlan_depth; 2398 2399 return type; 2400 } 2401 2402 /** 2403 * skb_mac_gso_segment - mac layer segmentation handler. 2404 * @skb: buffer to segment 2405 * @features: features for the output path (see dev->features) 2406 */ 2407 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 2408 netdev_features_t features) 2409 { 2410 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 2411 struct packet_offload *ptype; 2412 int vlan_depth = skb->mac_len; 2413 __be16 type = skb_network_protocol(skb, &vlan_depth); 2414 2415 if (unlikely(!type)) 2416 return ERR_PTR(-EINVAL); 2417 2418 __skb_pull(skb, vlan_depth); 2419 2420 rcu_read_lock(); 2421 list_for_each_entry_rcu(ptype, &offload_base, list) { 2422 if (ptype->type == type && ptype->callbacks.gso_segment) { 2423 segs = ptype->callbacks.gso_segment(skb, features); 2424 break; 2425 } 2426 } 2427 rcu_read_unlock(); 2428 2429 __skb_push(skb, skb->data - skb_mac_header(skb)); 2430 2431 return segs; 2432 } 2433 EXPORT_SYMBOL(skb_mac_gso_segment); 2434 2435 2436 /* openvswitch calls this on rx path, so we need a different check. 2437 */ 2438 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 2439 { 2440 if (tx_path) 2441 return skb->ip_summed != CHECKSUM_PARTIAL; 2442 else 2443 return skb->ip_summed == CHECKSUM_NONE; 2444 } 2445 2446 /** 2447 * __skb_gso_segment - Perform segmentation on skb. 2448 * @skb: buffer to segment 2449 * @features: features for the output path (see dev->features) 2450 * @tx_path: whether it is called in TX path 2451 * 2452 * This function segments the given skb and returns a list of segments. 2453 * 2454 * It may return NULL if the skb requires no segmentation. This is 2455 * only possible when GSO is used for verifying header integrity. 2456 */ 2457 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 2458 netdev_features_t features, bool tx_path) 2459 { 2460 if (unlikely(skb_needs_check(skb, tx_path))) { 2461 int err; 2462 2463 skb_warn_bad_offload(skb); 2464 2465 err = skb_cow_head(skb, 0); 2466 if (err < 0) 2467 return ERR_PTR(err); 2468 } 2469 2470 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 2471 SKB_GSO_CB(skb)->encap_level = 0; 2472 2473 skb_reset_mac_header(skb); 2474 skb_reset_mac_len(skb); 2475 2476 return skb_mac_gso_segment(skb, features); 2477 } 2478 EXPORT_SYMBOL(__skb_gso_segment); 2479 2480 /* Take action when hardware reception checksum errors are detected. */ 2481 #ifdef CONFIG_BUG 2482 void netdev_rx_csum_fault(struct net_device *dev) 2483 { 2484 if (net_ratelimit()) { 2485 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 2486 dump_stack(); 2487 } 2488 } 2489 EXPORT_SYMBOL(netdev_rx_csum_fault); 2490 #endif 2491 2492 /* Actually, we should eliminate this check as soon as we know, that: 2493 * 1. IOMMU is present and allows to map all the memory. 2494 * 2. No high memory really exists on this machine. 2495 */ 2496 2497 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 2498 { 2499 #ifdef CONFIG_HIGHMEM 2500 int i; 2501 if (!(dev->features & NETIF_F_HIGHDMA)) { 2502 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2503 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2504 if (PageHighMem(skb_frag_page(frag))) 2505 return 1; 2506 } 2507 } 2508 2509 if (PCI_DMA_BUS_IS_PHYS) { 2510 struct device *pdev = dev->dev.parent; 2511 2512 if (!pdev) 2513 return 0; 2514 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2515 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2516 dma_addr_t addr = page_to_phys(skb_frag_page(frag)); 2517 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) 2518 return 1; 2519 } 2520 } 2521 #endif 2522 return 0; 2523 } 2524 2525 /* If MPLS offload request, verify we are testing hardware MPLS features 2526 * instead of standard features for the netdev. 2527 */ 2528 #ifdef CONFIG_NET_MPLS_GSO 2529 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2530 netdev_features_t features, 2531 __be16 type) 2532 { 2533 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC)) 2534 features &= skb->dev->mpls_features; 2535 2536 return features; 2537 } 2538 #else 2539 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2540 netdev_features_t features, 2541 __be16 type) 2542 { 2543 return features; 2544 } 2545 #endif 2546 2547 static netdev_features_t harmonize_features(struct sk_buff *skb, 2548 netdev_features_t features) 2549 { 2550 int tmp; 2551 __be16 type; 2552 2553 type = skb_network_protocol(skb, &tmp); 2554 features = net_mpls_features(skb, features, type); 2555 2556 if (skb->ip_summed != CHECKSUM_NONE && 2557 !can_checksum_protocol(features, type)) { 2558 features &= ~NETIF_F_ALL_CSUM; 2559 } else if (illegal_highdma(skb->dev, skb)) { 2560 features &= ~NETIF_F_SG; 2561 } 2562 2563 return features; 2564 } 2565 2566 netdev_features_t netif_skb_features(struct sk_buff *skb) 2567 { 2568 const struct net_device *dev = skb->dev; 2569 netdev_features_t features = dev->features; 2570 u16 gso_segs = skb_shinfo(skb)->gso_segs; 2571 __be16 protocol = skb->protocol; 2572 2573 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs) 2574 features &= ~NETIF_F_GSO_MASK; 2575 2576 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) { 2577 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; 2578 protocol = veh->h_vlan_encapsulated_proto; 2579 } else if (!vlan_tx_tag_present(skb)) { 2580 return harmonize_features(skb, features); 2581 } 2582 2583 features = netdev_intersect_features(features, 2584 dev->vlan_features | 2585 NETIF_F_HW_VLAN_CTAG_TX | 2586 NETIF_F_HW_VLAN_STAG_TX); 2587 2588 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) 2589 features = netdev_intersect_features(features, 2590 NETIF_F_SG | 2591 NETIF_F_HIGHDMA | 2592 NETIF_F_FRAGLIST | 2593 NETIF_F_GEN_CSUM | 2594 NETIF_F_HW_VLAN_CTAG_TX | 2595 NETIF_F_HW_VLAN_STAG_TX); 2596 2597 return harmonize_features(skb, features); 2598 } 2599 EXPORT_SYMBOL(netif_skb_features); 2600 2601 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 2602 struct netdev_queue *txq, bool more) 2603 { 2604 unsigned int len; 2605 int rc; 2606 2607 if (!list_empty(&ptype_all)) 2608 dev_queue_xmit_nit(skb, dev); 2609 2610 len = skb->len; 2611 trace_net_dev_start_xmit(skb, dev); 2612 rc = netdev_start_xmit(skb, dev, txq, more); 2613 trace_net_dev_xmit(skb, rc, dev, len); 2614 2615 return rc; 2616 } 2617 2618 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 2619 struct netdev_queue *txq, int *ret) 2620 { 2621 struct sk_buff *skb = first; 2622 int rc = NETDEV_TX_OK; 2623 2624 while (skb) { 2625 struct sk_buff *next = skb->next; 2626 2627 skb->next = NULL; 2628 rc = xmit_one(skb, dev, txq, next != NULL); 2629 if (unlikely(!dev_xmit_complete(rc))) { 2630 skb->next = next; 2631 goto out; 2632 } 2633 2634 skb = next; 2635 if (netif_xmit_stopped(txq) && skb) { 2636 rc = NETDEV_TX_BUSY; 2637 break; 2638 } 2639 } 2640 2641 out: 2642 *ret = rc; 2643 return skb; 2644 } 2645 2646 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 2647 netdev_features_t features) 2648 { 2649 if (vlan_tx_tag_present(skb) && 2650 !vlan_hw_offload_capable(features, skb->vlan_proto)) { 2651 skb = __vlan_put_tag(skb, skb->vlan_proto, 2652 vlan_tx_tag_get(skb)); 2653 if (skb) 2654 skb->vlan_tci = 0; 2655 } 2656 return skb; 2657 } 2658 2659 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) 2660 { 2661 netdev_features_t features; 2662 2663 if (skb->next) 2664 return skb; 2665 2666 features = netif_skb_features(skb); 2667 skb = validate_xmit_vlan(skb, features); 2668 if (unlikely(!skb)) 2669 goto out_null; 2670 2671 /* If encapsulation offload request, verify we are testing 2672 * hardware encapsulation features instead of standard 2673 * features for the netdev 2674 */ 2675 if (skb->encapsulation) 2676 features &= dev->hw_enc_features; 2677 2678 if (netif_needs_gso(dev, skb, features)) { 2679 struct sk_buff *segs; 2680 2681 segs = skb_gso_segment(skb, features); 2682 if (IS_ERR(segs)) { 2683 segs = NULL; 2684 } else if (segs) { 2685 consume_skb(skb); 2686 skb = segs; 2687 } 2688 } else { 2689 if (skb_needs_linearize(skb, features) && 2690 __skb_linearize(skb)) 2691 goto out_kfree_skb; 2692 2693 /* If packet is not checksummed and device does not 2694 * support checksumming for this protocol, complete 2695 * checksumming here. 2696 */ 2697 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2698 if (skb->encapsulation) 2699 skb_set_inner_transport_header(skb, 2700 skb_checksum_start_offset(skb)); 2701 else 2702 skb_set_transport_header(skb, 2703 skb_checksum_start_offset(skb)); 2704 if (!(features & NETIF_F_ALL_CSUM) && 2705 skb_checksum_help(skb)) 2706 goto out_kfree_skb; 2707 } 2708 } 2709 2710 return skb; 2711 2712 out_kfree_skb: 2713 kfree_skb(skb); 2714 out_null: 2715 return NULL; 2716 } 2717 2718 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) 2719 { 2720 struct sk_buff *next, *head = NULL, *tail; 2721 2722 for (; skb != NULL; skb = next) { 2723 next = skb->next; 2724 skb->next = NULL; 2725 2726 /* in case skb wont be segmented, point to itself */ 2727 skb->prev = skb; 2728 2729 skb = validate_xmit_skb(skb, dev); 2730 if (!skb) 2731 continue; 2732 2733 if (!head) 2734 head = skb; 2735 else 2736 tail->next = skb; 2737 /* If skb was segmented, skb->prev points to 2738 * the last segment. If not, it still contains skb. 2739 */ 2740 tail = skb->prev; 2741 } 2742 return head; 2743 } 2744 2745 static void qdisc_pkt_len_init(struct sk_buff *skb) 2746 { 2747 const struct skb_shared_info *shinfo = skb_shinfo(skb); 2748 2749 qdisc_skb_cb(skb)->pkt_len = skb->len; 2750 2751 /* To get more precise estimation of bytes sent on wire, 2752 * we add to pkt_len the headers size of all segments 2753 */ 2754 if (shinfo->gso_size) { 2755 unsigned int hdr_len; 2756 u16 gso_segs = shinfo->gso_segs; 2757 2758 /* mac layer + network layer */ 2759 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 2760 2761 /* + transport layer */ 2762 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 2763 hdr_len += tcp_hdrlen(skb); 2764 else 2765 hdr_len += sizeof(struct udphdr); 2766 2767 if (shinfo->gso_type & SKB_GSO_DODGY) 2768 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 2769 shinfo->gso_size); 2770 2771 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 2772 } 2773 } 2774 2775 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 2776 struct net_device *dev, 2777 struct netdev_queue *txq) 2778 { 2779 spinlock_t *root_lock = qdisc_lock(q); 2780 bool contended; 2781 int rc; 2782 2783 qdisc_pkt_len_init(skb); 2784 qdisc_calculate_pkt_len(skb, q); 2785 /* 2786 * Heuristic to force contended enqueues to serialize on a 2787 * separate lock before trying to get qdisc main lock. 2788 * This permits __QDISC___STATE_RUNNING owner to get the lock more 2789 * often and dequeue packets faster. 2790 */ 2791 contended = qdisc_is_running(q); 2792 if (unlikely(contended)) 2793 spin_lock(&q->busylock); 2794 2795 spin_lock(root_lock); 2796 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 2797 kfree_skb(skb); 2798 rc = NET_XMIT_DROP; 2799 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 2800 qdisc_run_begin(q)) { 2801 /* 2802 * This is a work-conserving queue; there are no old skbs 2803 * waiting to be sent out; and the qdisc is not running - 2804 * xmit the skb directly. 2805 */ 2806 2807 qdisc_bstats_update(q, skb); 2808 2809 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 2810 if (unlikely(contended)) { 2811 spin_unlock(&q->busylock); 2812 contended = false; 2813 } 2814 __qdisc_run(q); 2815 } else 2816 qdisc_run_end(q); 2817 2818 rc = NET_XMIT_SUCCESS; 2819 } else { 2820 rc = q->enqueue(skb, q) & NET_XMIT_MASK; 2821 if (qdisc_run_begin(q)) { 2822 if (unlikely(contended)) { 2823 spin_unlock(&q->busylock); 2824 contended = false; 2825 } 2826 __qdisc_run(q); 2827 } 2828 } 2829 spin_unlock(root_lock); 2830 if (unlikely(contended)) 2831 spin_unlock(&q->busylock); 2832 return rc; 2833 } 2834 2835 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 2836 static void skb_update_prio(struct sk_buff *skb) 2837 { 2838 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); 2839 2840 if (!skb->priority && skb->sk && map) { 2841 unsigned int prioidx = skb->sk->sk_cgrp_prioidx; 2842 2843 if (prioidx < map->priomap_len) 2844 skb->priority = map->priomap[prioidx]; 2845 } 2846 } 2847 #else 2848 #define skb_update_prio(skb) 2849 #endif 2850 2851 static DEFINE_PER_CPU(int, xmit_recursion); 2852 #define RECURSION_LIMIT 10 2853 2854 /** 2855 * dev_loopback_xmit - loop back @skb 2856 * @skb: buffer to transmit 2857 */ 2858 int dev_loopback_xmit(struct sk_buff *skb) 2859 { 2860 skb_reset_mac_header(skb); 2861 __skb_pull(skb, skb_network_offset(skb)); 2862 skb->pkt_type = PACKET_LOOPBACK; 2863 skb->ip_summed = CHECKSUM_UNNECESSARY; 2864 WARN_ON(!skb_dst(skb)); 2865 skb_dst_force(skb); 2866 netif_rx_ni(skb); 2867 return 0; 2868 } 2869 EXPORT_SYMBOL(dev_loopback_xmit); 2870 2871 /** 2872 * __dev_queue_xmit - transmit a buffer 2873 * @skb: buffer to transmit 2874 * @accel_priv: private data used for L2 forwarding offload 2875 * 2876 * Queue a buffer for transmission to a network device. The caller must 2877 * have set the device and priority and built the buffer before calling 2878 * this function. The function can be called from an interrupt. 2879 * 2880 * A negative errno code is returned on a failure. A success does not 2881 * guarantee the frame will be transmitted as it may be dropped due 2882 * to congestion or traffic shaping. 2883 * 2884 * ----------------------------------------------------------------------------------- 2885 * I notice this method can also return errors from the queue disciplines, 2886 * including NET_XMIT_DROP, which is a positive value. So, errors can also 2887 * be positive. 2888 * 2889 * Regardless of the return value, the skb is consumed, so it is currently 2890 * difficult to retry a send to this method. (You can bump the ref count 2891 * before sending to hold a reference for retry if you are careful.) 2892 * 2893 * When calling this method, interrupts MUST be enabled. This is because 2894 * the BH enable code must have IRQs enabled so that it will not deadlock. 2895 * --BLG 2896 */ 2897 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) 2898 { 2899 struct net_device *dev = skb->dev; 2900 struct netdev_queue *txq; 2901 struct Qdisc *q; 2902 int rc = -ENOMEM; 2903 2904 skb_reset_mac_header(skb); 2905 2906 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 2907 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 2908 2909 /* Disable soft irqs for various locks below. Also 2910 * stops preemption for RCU. 2911 */ 2912 rcu_read_lock_bh(); 2913 2914 skb_update_prio(skb); 2915 2916 /* If device/qdisc don't need skb->dst, release it right now while 2917 * its hot in this cpu cache. 2918 */ 2919 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 2920 skb_dst_drop(skb); 2921 else 2922 skb_dst_force(skb); 2923 2924 txq = netdev_pick_tx(dev, skb, accel_priv); 2925 q = rcu_dereference_bh(txq->qdisc); 2926 2927 #ifdef CONFIG_NET_CLS_ACT 2928 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); 2929 #endif 2930 trace_net_dev_queue(skb); 2931 if (q->enqueue) { 2932 rc = __dev_xmit_skb(skb, q, dev, txq); 2933 goto out; 2934 } 2935 2936 /* The device has no queue. Common case for software devices: 2937 loopback, all the sorts of tunnels... 2938 2939 Really, it is unlikely that netif_tx_lock protection is necessary 2940 here. (f.e. loopback and IP tunnels are clean ignoring statistics 2941 counters.) 2942 However, it is possible, that they rely on protection 2943 made by us here. 2944 2945 Check this and shot the lock. It is not prone from deadlocks. 2946 Either shot noqueue qdisc, it is even simpler 8) 2947 */ 2948 if (dev->flags & IFF_UP) { 2949 int cpu = smp_processor_id(); /* ok because BHs are off */ 2950 2951 if (txq->xmit_lock_owner != cpu) { 2952 2953 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) 2954 goto recursion_alert; 2955 2956 skb = validate_xmit_skb(skb, dev); 2957 if (!skb) 2958 goto drop; 2959 2960 HARD_TX_LOCK(dev, txq, cpu); 2961 2962 if (!netif_xmit_stopped(txq)) { 2963 __this_cpu_inc(xmit_recursion); 2964 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 2965 __this_cpu_dec(xmit_recursion); 2966 if (dev_xmit_complete(rc)) { 2967 HARD_TX_UNLOCK(dev, txq); 2968 goto out; 2969 } 2970 } 2971 HARD_TX_UNLOCK(dev, txq); 2972 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 2973 dev->name); 2974 } else { 2975 /* Recursion is detected! It is possible, 2976 * unfortunately 2977 */ 2978 recursion_alert: 2979 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 2980 dev->name); 2981 } 2982 } 2983 2984 rc = -ENETDOWN; 2985 drop: 2986 rcu_read_unlock_bh(); 2987 2988 atomic_long_inc(&dev->tx_dropped); 2989 kfree_skb_list(skb); 2990 return rc; 2991 out: 2992 rcu_read_unlock_bh(); 2993 return rc; 2994 } 2995 2996 int dev_queue_xmit(struct sk_buff *skb) 2997 { 2998 return __dev_queue_xmit(skb, NULL); 2999 } 3000 EXPORT_SYMBOL(dev_queue_xmit); 3001 3002 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) 3003 { 3004 return __dev_queue_xmit(skb, accel_priv); 3005 } 3006 EXPORT_SYMBOL(dev_queue_xmit_accel); 3007 3008 3009 /*======================================================================= 3010 Receiver routines 3011 =======================================================================*/ 3012 3013 int netdev_max_backlog __read_mostly = 1000; 3014 EXPORT_SYMBOL(netdev_max_backlog); 3015 3016 int netdev_tstamp_prequeue __read_mostly = 1; 3017 int netdev_budget __read_mostly = 300; 3018 int weight_p __read_mostly = 64; /* old backlog weight */ 3019 3020 /* Called with irq disabled */ 3021 static inline void ____napi_schedule(struct softnet_data *sd, 3022 struct napi_struct *napi) 3023 { 3024 list_add_tail(&napi->poll_list, &sd->poll_list); 3025 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3026 } 3027 3028 #ifdef CONFIG_RPS 3029 3030 /* One global table that all flow-based protocols share. */ 3031 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3032 EXPORT_SYMBOL(rps_sock_flow_table); 3033 3034 struct static_key rps_needed __read_mostly; 3035 3036 static struct rps_dev_flow * 3037 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3038 struct rps_dev_flow *rflow, u16 next_cpu) 3039 { 3040 if (next_cpu != RPS_NO_CPU) { 3041 #ifdef CONFIG_RFS_ACCEL 3042 struct netdev_rx_queue *rxqueue; 3043 struct rps_dev_flow_table *flow_table; 3044 struct rps_dev_flow *old_rflow; 3045 u32 flow_id; 3046 u16 rxq_index; 3047 int rc; 3048 3049 /* Should we steer this flow to a different hardware queue? */ 3050 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3051 !(dev->features & NETIF_F_NTUPLE)) 3052 goto out; 3053 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 3054 if (rxq_index == skb_get_rx_queue(skb)) 3055 goto out; 3056 3057 rxqueue = dev->_rx + rxq_index; 3058 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3059 if (!flow_table) 3060 goto out; 3061 flow_id = skb_get_hash(skb) & flow_table->mask; 3062 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 3063 rxq_index, flow_id); 3064 if (rc < 0) 3065 goto out; 3066 old_rflow = rflow; 3067 rflow = &flow_table->flows[flow_id]; 3068 rflow->filter = rc; 3069 if (old_rflow->filter == rflow->filter) 3070 old_rflow->filter = RPS_NO_FILTER; 3071 out: 3072 #endif 3073 rflow->last_qtail = 3074 per_cpu(softnet_data, next_cpu).input_queue_head; 3075 } 3076 3077 rflow->cpu = next_cpu; 3078 return rflow; 3079 } 3080 3081 /* 3082 * get_rps_cpu is called from netif_receive_skb and returns the target 3083 * CPU from the RPS map of the receiving queue for a given skb. 3084 * rcu_read_lock must be held on entry. 3085 */ 3086 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3087 struct rps_dev_flow **rflowp) 3088 { 3089 struct netdev_rx_queue *rxqueue; 3090 struct rps_map *map; 3091 struct rps_dev_flow_table *flow_table; 3092 struct rps_sock_flow_table *sock_flow_table; 3093 int cpu = -1; 3094 u16 tcpu; 3095 u32 hash; 3096 3097 if (skb_rx_queue_recorded(skb)) { 3098 u16 index = skb_get_rx_queue(skb); 3099 if (unlikely(index >= dev->real_num_rx_queues)) { 3100 WARN_ONCE(dev->real_num_rx_queues > 1, 3101 "%s received packet on queue %u, but number " 3102 "of RX queues is %u\n", 3103 dev->name, index, dev->real_num_rx_queues); 3104 goto done; 3105 } 3106 rxqueue = dev->_rx + index; 3107 } else 3108 rxqueue = dev->_rx; 3109 3110 map = rcu_dereference(rxqueue->rps_map); 3111 if (map) { 3112 if (map->len == 1 && 3113 !rcu_access_pointer(rxqueue->rps_flow_table)) { 3114 tcpu = map->cpus[0]; 3115 if (cpu_online(tcpu)) 3116 cpu = tcpu; 3117 goto done; 3118 } 3119 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) { 3120 goto done; 3121 } 3122 3123 skb_reset_network_header(skb); 3124 hash = skb_get_hash(skb); 3125 if (!hash) 3126 goto done; 3127 3128 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3129 sock_flow_table = rcu_dereference(rps_sock_flow_table); 3130 if (flow_table && sock_flow_table) { 3131 u16 next_cpu; 3132 struct rps_dev_flow *rflow; 3133 3134 rflow = &flow_table->flows[hash & flow_table->mask]; 3135 tcpu = rflow->cpu; 3136 3137 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask]; 3138 3139 /* 3140 * If the desired CPU (where last recvmsg was done) is 3141 * different from current CPU (one in the rx-queue flow 3142 * table entry), switch if one of the following holds: 3143 * - Current CPU is unset (equal to RPS_NO_CPU). 3144 * - Current CPU is offline. 3145 * - The current CPU's queue tail has advanced beyond the 3146 * last packet that was enqueued using this table entry. 3147 * This guarantees that all previous packets for the flow 3148 * have been dequeued, thus preserving in order delivery. 3149 */ 3150 if (unlikely(tcpu != next_cpu) && 3151 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || 3152 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 3153 rflow->last_qtail)) >= 0)) { 3154 tcpu = next_cpu; 3155 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 3156 } 3157 3158 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { 3159 *rflowp = rflow; 3160 cpu = tcpu; 3161 goto done; 3162 } 3163 } 3164 3165 if (map) { 3166 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 3167 if (cpu_online(tcpu)) { 3168 cpu = tcpu; 3169 goto done; 3170 } 3171 } 3172 3173 done: 3174 return cpu; 3175 } 3176 3177 #ifdef CONFIG_RFS_ACCEL 3178 3179 /** 3180 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 3181 * @dev: Device on which the filter was set 3182 * @rxq_index: RX queue index 3183 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 3184 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 3185 * 3186 * Drivers that implement ndo_rx_flow_steer() should periodically call 3187 * this function for each installed filter and remove the filters for 3188 * which it returns %true. 3189 */ 3190 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 3191 u32 flow_id, u16 filter_id) 3192 { 3193 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 3194 struct rps_dev_flow_table *flow_table; 3195 struct rps_dev_flow *rflow; 3196 bool expire = true; 3197 int cpu; 3198 3199 rcu_read_lock(); 3200 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3201 if (flow_table && flow_id <= flow_table->mask) { 3202 rflow = &flow_table->flows[flow_id]; 3203 cpu = ACCESS_ONCE(rflow->cpu); 3204 if (rflow->filter == filter_id && cpu != RPS_NO_CPU && 3205 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 3206 rflow->last_qtail) < 3207 (int)(10 * flow_table->mask))) 3208 expire = false; 3209 } 3210 rcu_read_unlock(); 3211 return expire; 3212 } 3213 EXPORT_SYMBOL(rps_may_expire_flow); 3214 3215 #endif /* CONFIG_RFS_ACCEL */ 3216 3217 /* Called from hardirq (IPI) context */ 3218 static void rps_trigger_softirq(void *data) 3219 { 3220 struct softnet_data *sd = data; 3221 3222 ____napi_schedule(sd, &sd->backlog); 3223 sd->received_rps++; 3224 } 3225 3226 #endif /* CONFIG_RPS */ 3227 3228 /* 3229 * Check if this softnet_data structure is another cpu one 3230 * If yes, queue it to our IPI list and return 1 3231 * If no, return 0 3232 */ 3233 static int rps_ipi_queued(struct softnet_data *sd) 3234 { 3235 #ifdef CONFIG_RPS 3236 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 3237 3238 if (sd != mysd) { 3239 sd->rps_ipi_next = mysd->rps_ipi_list; 3240 mysd->rps_ipi_list = sd; 3241 3242 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3243 return 1; 3244 } 3245 #endif /* CONFIG_RPS */ 3246 return 0; 3247 } 3248 3249 #ifdef CONFIG_NET_FLOW_LIMIT 3250 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 3251 #endif 3252 3253 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 3254 { 3255 #ifdef CONFIG_NET_FLOW_LIMIT 3256 struct sd_flow_limit *fl; 3257 struct softnet_data *sd; 3258 unsigned int old_flow, new_flow; 3259 3260 if (qlen < (netdev_max_backlog >> 1)) 3261 return false; 3262 3263 sd = this_cpu_ptr(&softnet_data); 3264 3265 rcu_read_lock(); 3266 fl = rcu_dereference(sd->flow_limit); 3267 if (fl) { 3268 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 3269 old_flow = fl->history[fl->history_head]; 3270 fl->history[fl->history_head] = new_flow; 3271 3272 fl->history_head++; 3273 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 3274 3275 if (likely(fl->buckets[old_flow])) 3276 fl->buckets[old_flow]--; 3277 3278 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 3279 fl->count++; 3280 rcu_read_unlock(); 3281 return true; 3282 } 3283 } 3284 rcu_read_unlock(); 3285 #endif 3286 return false; 3287 } 3288 3289 /* 3290 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 3291 * queue (may be a remote CPU queue). 3292 */ 3293 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 3294 unsigned int *qtail) 3295 { 3296 struct softnet_data *sd; 3297 unsigned long flags; 3298 unsigned int qlen; 3299 3300 sd = &per_cpu(softnet_data, cpu); 3301 3302 local_irq_save(flags); 3303 3304 rps_lock(sd); 3305 qlen = skb_queue_len(&sd->input_pkt_queue); 3306 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 3307 if (skb_queue_len(&sd->input_pkt_queue)) { 3308 enqueue: 3309 __skb_queue_tail(&sd->input_pkt_queue, skb); 3310 input_queue_tail_incr_save(sd, qtail); 3311 rps_unlock(sd); 3312 local_irq_restore(flags); 3313 return NET_RX_SUCCESS; 3314 } 3315 3316 /* Schedule NAPI for backlog device 3317 * We can use non atomic operation since we own the queue lock 3318 */ 3319 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 3320 if (!rps_ipi_queued(sd)) 3321 ____napi_schedule(sd, &sd->backlog); 3322 } 3323 goto enqueue; 3324 } 3325 3326 sd->dropped++; 3327 rps_unlock(sd); 3328 3329 local_irq_restore(flags); 3330 3331 atomic_long_inc(&skb->dev->rx_dropped); 3332 kfree_skb(skb); 3333 return NET_RX_DROP; 3334 } 3335 3336 static int netif_rx_internal(struct sk_buff *skb) 3337 { 3338 int ret; 3339 3340 net_timestamp_check(netdev_tstamp_prequeue, skb); 3341 3342 trace_netif_rx(skb); 3343 #ifdef CONFIG_RPS 3344 if (static_key_false(&rps_needed)) { 3345 struct rps_dev_flow voidflow, *rflow = &voidflow; 3346 int cpu; 3347 3348 preempt_disable(); 3349 rcu_read_lock(); 3350 3351 cpu = get_rps_cpu(skb->dev, skb, &rflow); 3352 if (cpu < 0) 3353 cpu = smp_processor_id(); 3354 3355 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 3356 3357 rcu_read_unlock(); 3358 preempt_enable(); 3359 } else 3360 #endif 3361 { 3362 unsigned int qtail; 3363 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 3364 put_cpu(); 3365 } 3366 return ret; 3367 } 3368 3369 /** 3370 * netif_rx - post buffer to the network code 3371 * @skb: buffer to post 3372 * 3373 * This function receives a packet from a device driver and queues it for 3374 * the upper (protocol) levels to process. It always succeeds. The buffer 3375 * may be dropped during processing for congestion control or by the 3376 * protocol layers. 3377 * 3378 * return values: 3379 * NET_RX_SUCCESS (no congestion) 3380 * NET_RX_DROP (packet was dropped) 3381 * 3382 */ 3383 3384 int netif_rx(struct sk_buff *skb) 3385 { 3386 trace_netif_rx_entry(skb); 3387 3388 return netif_rx_internal(skb); 3389 } 3390 EXPORT_SYMBOL(netif_rx); 3391 3392 int netif_rx_ni(struct sk_buff *skb) 3393 { 3394 int err; 3395 3396 trace_netif_rx_ni_entry(skb); 3397 3398 preempt_disable(); 3399 err = netif_rx_internal(skb); 3400 if (local_softirq_pending()) 3401 do_softirq(); 3402 preempt_enable(); 3403 3404 return err; 3405 } 3406 EXPORT_SYMBOL(netif_rx_ni); 3407 3408 static void net_tx_action(struct softirq_action *h) 3409 { 3410 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 3411 3412 if (sd->completion_queue) { 3413 struct sk_buff *clist; 3414 3415 local_irq_disable(); 3416 clist = sd->completion_queue; 3417 sd->completion_queue = NULL; 3418 local_irq_enable(); 3419 3420 while (clist) { 3421 struct sk_buff *skb = clist; 3422 clist = clist->next; 3423 3424 WARN_ON(atomic_read(&skb->users)); 3425 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 3426 trace_consume_skb(skb); 3427 else 3428 trace_kfree_skb(skb, net_tx_action); 3429 __kfree_skb(skb); 3430 } 3431 } 3432 3433 if (sd->output_queue) { 3434 struct Qdisc *head; 3435 3436 local_irq_disable(); 3437 head = sd->output_queue; 3438 sd->output_queue = NULL; 3439 sd->output_queue_tailp = &sd->output_queue; 3440 local_irq_enable(); 3441 3442 while (head) { 3443 struct Qdisc *q = head; 3444 spinlock_t *root_lock; 3445 3446 head = head->next_sched; 3447 3448 root_lock = qdisc_lock(q); 3449 if (spin_trylock(root_lock)) { 3450 smp_mb__before_atomic(); 3451 clear_bit(__QDISC_STATE_SCHED, 3452 &q->state); 3453 qdisc_run(q); 3454 spin_unlock(root_lock); 3455 } else { 3456 if (!test_bit(__QDISC_STATE_DEACTIVATED, 3457 &q->state)) { 3458 __netif_reschedule(q); 3459 } else { 3460 smp_mb__before_atomic(); 3461 clear_bit(__QDISC_STATE_SCHED, 3462 &q->state); 3463 } 3464 } 3465 } 3466 } 3467 } 3468 3469 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ 3470 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) 3471 /* This hook is defined here for ATM LANE */ 3472 int (*br_fdb_test_addr_hook)(struct net_device *dev, 3473 unsigned char *addr) __read_mostly; 3474 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 3475 #endif 3476 3477 #ifdef CONFIG_NET_CLS_ACT 3478 /* TODO: Maybe we should just force sch_ingress to be compiled in 3479 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions 3480 * a compare and 2 stores extra right now if we dont have it on 3481 * but have CONFIG_NET_CLS_ACT 3482 * NOTE: This doesn't stop any functionality; if you dont have 3483 * the ingress scheduler, you just can't add policies on ingress. 3484 * 3485 */ 3486 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq) 3487 { 3488 struct net_device *dev = skb->dev; 3489 u32 ttl = G_TC_RTTL(skb->tc_verd); 3490 int result = TC_ACT_OK; 3491 struct Qdisc *q; 3492 3493 if (unlikely(MAX_RED_LOOP < ttl++)) { 3494 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n", 3495 skb->skb_iif, dev->ifindex); 3496 return TC_ACT_SHOT; 3497 } 3498 3499 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); 3500 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); 3501 3502 q = rcu_dereference(rxq->qdisc); 3503 if (q != &noop_qdisc) { 3504 spin_lock(qdisc_lock(q)); 3505 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) 3506 result = qdisc_enqueue_root(skb, q); 3507 spin_unlock(qdisc_lock(q)); 3508 } 3509 3510 return result; 3511 } 3512 3513 static inline struct sk_buff *handle_ing(struct sk_buff *skb, 3514 struct packet_type **pt_prev, 3515 int *ret, struct net_device *orig_dev) 3516 { 3517 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue); 3518 3519 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc) 3520 goto out; 3521 3522 if (*pt_prev) { 3523 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3524 *pt_prev = NULL; 3525 } 3526 3527 switch (ing_filter(skb, rxq)) { 3528 case TC_ACT_SHOT: 3529 case TC_ACT_STOLEN: 3530 kfree_skb(skb); 3531 return NULL; 3532 } 3533 3534 out: 3535 skb->tc_verd = 0; 3536 return skb; 3537 } 3538 #endif 3539 3540 /** 3541 * netdev_rx_handler_register - register receive handler 3542 * @dev: device to register a handler for 3543 * @rx_handler: receive handler to register 3544 * @rx_handler_data: data pointer that is used by rx handler 3545 * 3546 * Register a receive handler for a device. This handler will then be 3547 * called from __netif_receive_skb. A negative errno code is returned 3548 * on a failure. 3549 * 3550 * The caller must hold the rtnl_mutex. 3551 * 3552 * For a general description of rx_handler, see enum rx_handler_result. 3553 */ 3554 int netdev_rx_handler_register(struct net_device *dev, 3555 rx_handler_func_t *rx_handler, 3556 void *rx_handler_data) 3557 { 3558 ASSERT_RTNL(); 3559 3560 if (dev->rx_handler) 3561 return -EBUSY; 3562 3563 /* Note: rx_handler_data must be set before rx_handler */ 3564 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 3565 rcu_assign_pointer(dev->rx_handler, rx_handler); 3566 3567 return 0; 3568 } 3569 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 3570 3571 /** 3572 * netdev_rx_handler_unregister - unregister receive handler 3573 * @dev: device to unregister a handler from 3574 * 3575 * Unregister a receive handler from a device. 3576 * 3577 * The caller must hold the rtnl_mutex. 3578 */ 3579 void netdev_rx_handler_unregister(struct net_device *dev) 3580 { 3581 3582 ASSERT_RTNL(); 3583 RCU_INIT_POINTER(dev->rx_handler, NULL); 3584 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 3585 * section has a guarantee to see a non NULL rx_handler_data 3586 * as well. 3587 */ 3588 synchronize_net(); 3589 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 3590 } 3591 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 3592 3593 /* 3594 * Limit the use of PFMEMALLOC reserves to those protocols that implement 3595 * the special handling of PFMEMALLOC skbs. 3596 */ 3597 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 3598 { 3599 switch (skb->protocol) { 3600 case htons(ETH_P_ARP): 3601 case htons(ETH_P_IP): 3602 case htons(ETH_P_IPV6): 3603 case htons(ETH_P_8021Q): 3604 case htons(ETH_P_8021AD): 3605 return true; 3606 default: 3607 return false; 3608 } 3609 } 3610 3611 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) 3612 { 3613 struct packet_type *ptype, *pt_prev; 3614 rx_handler_func_t *rx_handler; 3615 struct net_device *orig_dev; 3616 struct net_device *null_or_dev; 3617 bool deliver_exact = false; 3618 int ret = NET_RX_DROP; 3619 __be16 type; 3620 3621 net_timestamp_check(!netdev_tstamp_prequeue, skb); 3622 3623 trace_netif_receive_skb(skb); 3624 3625 orig_dev = skb->dev; 3626 3627 skb_reset_network_header(skb); 3628 if (!skb_transport_header_was_set(skb)) 3629 skb_reset_transport_header(skb); 3630 skb_reset_mac_len(skb); 3631 3632 pt_prev = NULL; 3633 3634 rcu_read_lock(); 3635 3636 another_round: 3637 skb->skb_iif = skb->dev->ifindex; 3638 3639 __this_cpu_inc(softnet_data.processed); 3640 3641 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 3642 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 3643 skb = skb_vlan_untag(skb); 3644 if (unlikely(!skb)) 3645 goto unlock; 3646 } 3647 3648 #ifdef CONFIG_NET_CLS_ACT 3649 if (skb->tc_verd & TC_NCLS) { 3650 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); 3651 goto ncls; 3652 } 3653 #endif 3654 3655 if (pfmemalloc) 3656 goto skip_taps; 3657 3658 list_for_each_entry_rcu(ptype, &ptype_all, list) { 3659 if (!ptype->dev || ptype->dev == skb->dev) { 3660 if (pt_prev) 3661 ret = deliver_skb(skb, pt_prev, orig_dev); 3662 pt_prev = ptype; 3663 } 3664 } 3665 3666 skip_taps: 3667 #ifdef CONFIG_NET_CLS_ACT 3668 skb = handle_ing(skb, &pt_prev, &ret, orig_dev); 3669 if (!skb) 3670 goto unlock; 3671 ncls: 3672 #endif 3673 3674 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 3675 goto drop; 3676 3677 if (vlan_tx_tag_present(skb)) { 3678 if (pt_prev) { 3679 ret = deliver_skb(skb, pt_prev, orig_dev); 3680 pt_prev = NULL; 3681 } 3682 if (vlan_do_receive(&skb)) 3683 goto another_round; 3684 else if (unlikely(!skb)) 3685 goto unlock; 3686 } 3687 3688 rx_handler = rcu_dereference(skb->dev->rx_handler); 3689 if (rx_handler) { 3690 if (pt_prev) { 3691 ret = deliver_skb(skb, pt_prev, orig_dev); 3692 pt_prev = NULL; 3693 } 3694 switch (rx_handler(&skb)) { 3695 case RX_HANDLER_CONSUMED: 3696 ret = NET_RX_SUCCESS; 3697 goto unlock; 3698 case RX_HANDLER_ANOTHER: 3699 goto another_round; 3700 case RX_HANDLER_EXACT: 3701 deliver_exact = true; 3702 case RX_HANDLER_PASS: 3703 break; 3704 default: 3705 BUG(); 3706 } 3707 } 3708 3709 if (unlikely(vlan_tx_tag_present(skb))) { 3710 if (vlan_tx_tag_get_id(skb)) 3711 skb->pkt_type = PACKET_OTHERHOST; 3712 /* Note: we might in the future use prio bits 3713 * and set skb->priority like in vlan_do_receive() 3714 * For the time being, just ignore Priority Code Point 3715 */ 3716 skb->vlan_tci = 0; 3717 } 3718 3719 /* deliver only exact match when indicated */ 3720 null_or_dev = deliver_exact ? skb->dev : NULL; 3721 3722 type = skb->protocol; 3723 list_for_each_entry_rcu(ptype, 3724 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { 3725 if (ptype->type == type && 3726 (ptype->dev == null_or_dev || ptype->dev == skb->dev || 3727 ptype->dev == orig_dev)) { 3728 if (pt_prev) 3729 ret = deliver_skb(skb, pt_prev, orig_dev); 3730 pt_prev = ptype; 3731 } 3732 } 3733 3734 if (pt_prev) { 3735 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 3736 goto drop; 3737 else 3738 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 3739 } else { 3740 drop: 3741 atomic_long_inc(&skb->dev->rx_dropped); 3742 kfree_skb(skb); 3743 /* Jamal, now you will not able to escape explaining 3744 * me how you were going to use this. :-) 3745 */ 3746 ret = NET_RX_DROP; 3747 } 3748 3749 unlock: 3750 rcu_read_unlock(); 3751 return ret; 3752 } 3753 3754 static int __netif_receive_skb(struct sk_buff *skb) 3755 { 3756 int ret; 3757 3758 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 3759 unsigned long pflags = current->flags; 3760 3761 /* 3762 * PFMEMALLOC skbs are special, they should 3763 * - be delivered to SOCK_MEMALLOC sockets only 3764 * - stay away from userspace 3765 * - have bounded memory usage 3766 * 3767 * Use PF_MEMALLOC as this saves us from propagating the allocation 3768 * context down to all allocation sites. 3769 */ 3770 current->flags |= PF_MEMALLOC; 3771 ret = __netif_receive_skb_core(skb, true); 3772 tsk_restore_flags(current, pflags, PF_MEMALLOC); 3773 } else 3774 ret = __netif_receive_skb_core(skb, false); 3775 3776 return ret; 3777 } 3778 3779 static int netif_receive_skb_internal(struct sk_buff *skb) 3780 { 3781 net_timestamp_check(netdev_tstamp_prequeue, skb); 3782 3783 if (skb_defer_rx_timestamp(skb)) 3784 return NET_RX_SUCCESS; 3785 3786 #ifdef CONFIG_RPS 3787 if (static_key_false(&rps_needed)) { 3788 struct rps_dev_flow voidflow, *rflow = &voidflow; 3789 int cpu, ret; 3790 3791 rcu_read_lock(); 3792 3793 cpu = get_rps_cpu(skb->dev, skb, &rflow); 3794 3795 if (cpu >= 0) { 3796 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 3797 rcu_read_unlock(); 3798 return ret; 3799 } 3800 rcu_read_unlock(); 3801 } 3802 #endif 3803 return __netif_receive_skb(skb); 3804 } 3805 3806 /** 3807 * netif_receive_skb - process receive buffer from network 3808 * @skb: buffer to process 3809 * 3810 * netif_receive_skb() is the main receive data processing function. 3811 * It always succeeds. The buffer may be dropped during processing 3812 * for congestion control or by the protocol layers. 3813 * 3814 * This function may only be called from softirq context and interrupts 3815 * should be enabled. 3816 * 3817 * Return values (usually ignored): 3818 * NET_RX_SUCCESS: no congestion 3819 * NET_RX_DROP: packet was dropped 3820 */ 3821 int netif_receive_skb(struct sk_buff *skb) 3822 { 3823 trace_netif_receive_skb_entry(skb); 3824 3825 return netif_receive_skb_internal(skb); 3826 } 3827 EXPORT_SYMBOL(netif_receive_skb); 3828 3829 /* Network device is going away, flush any packets still pending 3830 * Called with irqs disabled. 3831 */ 3832 static void flush_backlog(void *arg) 3833 { 3834 struct net_device *dev = arg; 3835 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 3836 struct sk_buff *skb, *tmp; 3837 3838 rps_lock(sd); 3839 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 3840 if (skb->dev == dev) { 3841 __skb_unlink(skb, &sd->input_pkt_queue); 3842 kfree_skb(skb); 3843 input_queue_head_incr(sd); 3844 } 3845 } 3846 rps_unlock(sd); 3847 3848 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 3849 if (skb->dev == dev) { 3850 __skb_unlink(skb, &sd->process_queue); 3851 kfree_skb(skb); 3852 input_queue_head_incr(sd); 3853 } 3854 } 3855 } 3856 3857 static int napi_gro_complete(struct sk_buff *skb) 3858 { 3859 struct packet_offload *ptype; 3860 __be16 type = skb->protocol; 3861 struct list_head *head = &offload_base; 3862 int err = -ENOENT; 3863 3864 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 3865 3866 if (NAPI_GRO_CB(skb)->count == 1) { 3867 skb_shinfo(skb)->gso_size = 0; 3868 goto out; 3869 } 3870 3871 rcu_read_lock(); 3872 list_for_each_entry_rcu(ptype, head, list) { 3873 if (ptype->type != type || !ptype->callbacks.gro_complete) 3874 continue; 3875 3876 err = ptype->callbacks.gro_complete(skb, 0); 3877 break; 3878 } 3879 rcu_read_unlock(); 3880 3881 if (err) { 3882 WARN_ON(&ptype->list == head); 3883 kfree_skb(skb); 3884 return NET_RX_SUCCESS; 3885 } 3886 3887 out: 3888 return netif_receive_skb_internal(skb); 3889 } 3890 3891 /* napi->gro_list contains packets ordered by age. 3892 * youngest packets at the head of it. 3893 * Complete skbs in reverse order to reduce latencies. 3894 */ 3895 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 3896 { 3897 struct sk_buff *skb, *prev = NULL; 3898 3899 /* scan list and build reverse chain */ 3900 for (skb = napi->gro_list; skb != NULL; skb = skb->next) { 3901 skb->prev = prev; 3902 prev = skb; 3903 } 3904 3905 for (skb = prev; skb; skb = prev) { 3906 skb->next = NULL; 3907 3908 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 3909 return; 3910 3911 prev = skb->prev; 3912 napi_gro_complete(skb); 3913 napi->gro_count--; 3914 } 3915 3916 napi->gro_list = NULL; 3917 } 3918 EXPORT_SYMBOL(napi_gro_flush); 3919 3920 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) 3921 { 3922 struct sk_buff *p; 3923 unsigned int maclen = skb->dev->hard_header_len; 3924 u32 hash = skb_get_hash_raw(skb); 3925 3926 for (p = napi->gro_list; p; p = p->next) { 3927 unsigned long diffs; 3928 3929 NAPI_GRO_CB(p)->flush = 0; 3930 3931 if (hash != skb_get_hash_raw(p)) { 3932 NAPI_GRO_CB(p)->same_flow = 0; 3933 continue; 3934 } 3935 3936 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 3937 diffs |= p->vlan_tci ^ skb->vlan_tci; 3938 if (maclen == ETH_HLEN) 3939 diffs |= compare_ether_header(skb_mac_header(p), 3940 skb_mac_header(skb)); 3941 else if (!diffs) 3942 diffs = memcmp(skb_mac_header(p), 3943 skb_mac_header(skb), 3944 maclen); 3945 NAPI_GRO_CB(p)->same_flow = !diffs; 3946 } 3947 } 3948 3949 static void skb_gro_reset_offset(struct sk_buff *skb) 3950 { 3951 const struct skb_shared_info *pinfo = skb_shinfo(skb); 3952 const skb_frag_t *frag0 = &pinfo->frags[0]; 3953 3954 NAPI_GRO_CB(skb)->data_offset = 0; 3955 NAPI_GRO_CB(skb)->frag0 = NULL; 3956 NAPI_GRO_CB(skb)->frag0_len = 0; 3957 3958 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 3959 pinfo->nr_frags && 3960 !PageHighMem(skb_frag_page(frag0))) { 3961 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 3962 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); 3963 } 3964 } 3965 3966 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 3967 { 3968 struct skb_shared_info *pinfo = skb_shinfo(skb); 3969 3970 BUG_ON(skb->end - skb->tail < grow); 3971 3972 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 3973 3974 skb->data_len -= grow; 3975 skb->tail += grow; 3976 3977 pinfo->frags[0].page_offset += grow; 3978 skb_frag_size_sub(&pinfo->frags[0], grow); 3979 3980 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 3981 skb_frag_unref(skb, 0); 3982 memmove(pinfo->frags, pinfo->frags + 1, 3983 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 3984 } 3985 } 3986 3987 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 3988 { 3989 struct sk_buff **pp = NULL; 3990 struct packet_offload *ptype; 3991 __be16 type = skb->protocol; 3992 struct list_head *head = &offload_base; 3993 int same_flow; 3994 enum gro_result ret; 3995 int grow; 3996 3997 if (!(skb->dev->features & NETIF_F_GRO)) 3998 goto normal; 3999 4000 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) 4001 goto normal; 4002 4003 gro_list_prepare(napi, skb); 4004 4005 rcu_read_lock(); 4006 list_for_each_entry_rcu(ptype, head, list) { 4007 if (ptype->type != type || !ptype->callbacks.gro_receive) 4008 continue; 4009 4010 skb_set_network_header(skb, skb_gro_offset(skb)); 4011 skb_reset_mac_len(skb); 4012 NAPI_GRO_CB(skb)->same_flow = 0; 4013 NAPI_GRO_CB(skb)->flush = 0; 4014 NAPI_GRO_CB(skb)->free = 0; 4015 NAPI_GRO_CB(skb)->udp_mark = 0; 4016 4017 /* Setup for GRO checksum validation */ 4018 switch (skb->ip_summed) { 4019 case CHECKSUM_COMPLETE: 4020 NAPI_GRO_CB(skb)->csum = skb->csum; 4021 NAPI_GRO_CB(skb)->csum_valid = 1; 4022 NAPI_GRO_CB(skb)->csum_cnt = 0; 4023 break; 4024 case CHECKSUM_UNNECESSARY: 4025 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 4026 NAPI_GRO_CB(skb)->csum_valid = 0; 4027 break; 4028 default: 4029 NAPI_GRO_CB(skb)->csum_cnt = 0; 4030 NAPI_GRO_CB(skb)->csum_valid = 0; 4031 } 4032 4033 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); 4034 break; 4035 } 4036 rcu_read_unlock(); 4037 4038 if (&ptype->list == head) 4039 goto normal; 4040 4041 same_flow = NAPI_GRO_CB(skb)->same_flow; 4042 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 4043 4044 if (pp) { 4045 struct sk_buff *nskb = *pp; 4046 4047 *pp = nskb->next; 4048 nskb->next = NULL; 4049 napi_gro_complete(nskb); 4050 napi->gro_count--; 4051 } 4052 4053 if (same_flow) 4054 goto ok; 4055 4056 if (NAPI_GRO_CB(skb)->flush) 4057 goto normal; 4058 4059 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { 4060 struct sk_buff *nskb = napi->gro_list; 4061 4062 /* locate the end of the list to select the 'oldest' flow */ 4063 while (nskb->next) { 4064 pp = &nskb->next; 4065 nskb = *pp; 4066 } 4067 *pp = NULL; 4068 nskb->next = NULL; 4069 napi_gro_complete(nskb); 4070 } else { 4071 napi->gro_count++; 4072 } 4073 NAPI_GRO_CB(skb)->count = 1; 4074 NAPI_GRO_CB(skb)->age = jiffies; 4075 NAPI_GRO_CB(skb)->last = skb; 4076 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 4077 skb->next = napi->gro_list; 4078 napi->gro_list = skb; 4079 ret = GRO_HELD; 4080 4081 pull: 4082 grow = skb_gro_offset(skb) - skb_headlen(skb); 4083 if (grow > 0) 4084 gro_pull_from_frag0(skb, grow); 4085 ok: 4086 return ret; 4087 4088 normal: 4089 ret = GRO_NORMAL; 4090 goto pull; 4091 } 4092 4093 struct packet_offload *gro_find_receive_by_type(__be16 type) 4094 { 4095 struct list_head *offload_head = &offload_base; 4096 struct packet_offload *ptype; 4097 4098 list_for_each_entry_rcu(ptype, offload_head, list) { 4099 if (ptype->type != type || !ptype->callbacks.gro_receive) 4100 continue; 4101 return ptype; 4102 } 4103 return NULL; 4104 } 4105 EXPORT_SYMBOL(gro_find_receive_by_type); 4106 4107 struct packet_offload *gro_find_complete_by_type(__be16 type) 4108 { 4109 struct list_head *offload_head = &offload_base; 4110 struct packet_offload *ptype; 4111 4112 list_for_each_entry_rcu(ptype, offload_head, list) { 4113 if (ptype->type != type || !ptype->callbacks.gro_complete) 4114 continue; 4115 return ptype; 4116 } 4117 return NULL; 4118 } 4119 EXPORT_SYMBOL(gro_find_complete_by_type); 4120 4121 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 4122 { 4123 switch (ret) { 4124 case GRO_NORMAL: 4125 if (netif_receive_skb_internal(skb)) 4126 ret = GRO_DROP; 4127 break; 4128 4129 case GRO_DROP: 4130 kfree_skb(skb); 4131 break; 4132 4133 case GRO_MERGED_FREE: 4134 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 4135 kmem_cache_free(skbuff_head_cache, skb); 4136 else 4137 __kfree_skb(skb); 4138 break; 4139 4140 case GRO_HELD: 4141 case GRO_MERGED: 4142 break; 4143 } 4144 4145 return ret; 4146 } 4147 4148 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4149 { 4150 trace_napi_gro_receive_entry(skb); 4151 4152 skb_gro_reset_offset(skb); 4153 4154 return napi_skb_finish(dev_gro_receive(napi, skb), skb); 4155 } 4156 EXPORT_SYMBOL(napi_gro_receive); 4157 4158 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 4159 { 4160 if (unlikely(skb->pfmemalloc)) { 4161 consume_skb(skb); 4162 return; 4163 } 4164 __skb_pull(skb, skb_headlen(skb)); 4165 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 4166 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 4167 skb->vlan_tci = 0; 4168 skb->dev = napi->dev; 4169 skb->skb_iif = 0; 4170 skb->encapsulation = 0; 4171 skb_shinfo(skb)->gso_type = 0; 4172 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 4173 4174 napi->skb = skb; 4175 } 4176 4177 struct sk_buff *napi_get_frags(struct napi_struct *napi) 4178 { 4179 struct sk_buff *skb = napi->skb; 4180 4181 if (!skb) { 4182 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); 4183 napi->skb = skb; 4184 } 4185 return skb; 4186 } 4187 EXPORT_SYMBOL(napi_get_frags); 4188 4189 static gro_result_t napi_frags_finish(struct napi_struct *napi, 4190 struct sk_buff *skb, 4191 gro_result_t ret) 4192 { 4193 switch (ret) { 4194 case GRO_NORMAL: 4195 case GRO_HELD: 4196 __skb_push(skb, ETH_HLEN); 4197 skb->protocol = eth_type_trans(skb, skb->dev); 4198 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 4199 ret = GRO_DROP; 4200 break; 4201 4202 case GRO_DROP: 4203 case GRO_MERGED_FREE: 4204 napi_reuse_skb(napi, skb); 4205 break; 4206 4207 case GRO_MERGED: 4208 break; 4209 } 4210 4211 return ret; 4212 } 4213 4214 /* Upper GRO stack assumes network header starts at gro_offset=0 4215 * Drivers could call both napi_gro_frags() and napi_gro_receive() 4216 * We copy ethernet header into skb->data to have a common layout. 4217 */ 4218 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 4219 { 4220 struct sk_buff *skb = napi->skb; 4221 const struct ethhdr *eth; 4222 unsigned int hlen = sizeof(*eth); 4223 4224 napi->skb = NULL; 4225 4226 skb_reset_mac_header(skb); 4227 skb_gro_reset_offset(skb); 4228 4229 eth = skb_gro_header_fast(skb, 0); 4230 if (unlikely(skb_gro_header_hard(skb, hlen))) { 4231 eth = skb_gro_header_slow(skb, hlen, 0); 4232 if (unlikely(!eth)) { 4233 napi_reuse_skb(napi, skb); 4234 return NULL; 4235 } 4236 } else { 4237 gro_pull_from_frag0(skb, hlen); 4238 NAPI_GRO_CB(skb)->frag0 += hlen; 4239 NAPI_GRO_CB(skb)->frag0_len -= hlen; 4240 } 4241 __skb_pull(skb, hlen); 4242 4243 /* 4244 * This works because the only protocols we care about don't require 4245 * special handling. 4246 * We'll fix it up properly in napi_frags_finish() 4247 */ 4248 skb->protocol = eth->h_proto; 4249 4250 return skb; 4251 } 4252 4253 gro_result_t napi_gro_frags(struct napi_struct *napi) 4254 { 4255 struct sk_buff *skb = napi_frags_skb(napi); 4256 4257 if (!skb) 4258 return GRO_DROP; 4259 4260 trace_napi_gro_frags_entry(skb); 4261 4262 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 4263 } 4264 EXPORT_SYMBOL(napi_gro_frags); 4265 4266 /* Compute the checksum from gro_offset and return the folded value 4267 * after adding in any pseudo checksum. 4268 */ 4269 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 4270 { 4271 __wsum wsum; 4272 __sum16 sum; 4273 4274 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 4275 4276 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 4277 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 4278 if (likely(!sum)) { 4279 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 4280 !skb->csum_complete_sw) 4281 netdev_rx_csum_fault(skb->dev); 4282 } 4283 4284 NAPI_GRO_CB(skb)->csum = wsum; 4285 NAPI_GRO_CB(skb)->csum_valid = 1; 4286 4287 return sum; 4288 } 4289 EXPORT_SYMBOL(__skb_gro_checksum_complete); 4290 4291 /* 4292 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 4293 * Note: called with local irq disabled, but exits with local irq enabled. 4294 */ 4295 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 4296 { 4297 #ifdef CONFIG_RPS 4298 struct softnet_data *remsd = sd->rps_ipi_list; 4299 4300 if (remsd) { 4301 sd->rps_ipi_list = NULL; 4302 4303 local_irq_enable(); 4304 4305 /* Send pending IPI's to kick RPS processing on remote cpus. */ 4306 while (remsd) { 4307 struct softnet_data *next = remsd->rps_ipi_next; 4308 4309 if (cpu_online(remsd->cpu)) 4310 smp_call_function_single_async(remsd->cpu, 4311 &remsd->csd); 4312 remsd = next; 4313 } 4314 } else 4315 #endif 4316 local_irq_enable(); 4317 } 4318 4319 static int process_backlog(struct napi_struct *napi, int quota) 4320 { 4321 int work = 0; 4322 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 4323 4324 #ifdef CONFIG_RPS 4325 /* Check if we have pending ipi, its better to send them now, 4326 * not waiting net_rx_action() end. 4327 */ 4328 if (sd->rps_ipi_list) { 4329 local_irq_disable(); 4330 net_rps_action_and_irq_enable(sd); 4331 } 4332 #endif 4333 napi->weight = weight_p; 4334 local_irq_disable(); 4335 while (1) { 4336 struct sk_buff *skb; 4337 4338 while ((skb = __skb_dequeue(&sd->process_queue))) { 4339 local_irq_enable(); 4340 __netif_receive_skb(skb); 4341 local_irq_disable(); 4342 input_queue_head_incr(sd); 4343 if (++work >= quota) { 4344 local_irq_enable(); 4345 return work; 4346 } 4347 } 4348 4349 rps_lock(sd); 4350 if (skb_queue_empty(&sd->input_pkt_queue)) { 4351 /* 4352 * Inline a custom version of __napi_complete(). 4353 * only current cpu owns and manipulates this napi, 4354 * and NAPI_STATE_SCHED is the only possible flag set 4355 * on backlog. 4356 * We can use a plain write instead of clear_bit(), 4357 * and we dont need an smp_mb() memory barrier. 4358 */ 4359 list_del(&napi->poll_list); 4360 napi->state = 0; 4361 rps_unlock(sd); 4362 4363 break; 4364 } 4365 4366 skb_queue_splice_tail_init(&sd->input_pkt_queue, 4367 &sd->process_queue); 4368 rps_unlock(sd); 4369 } 4370 local_irq_enable(); 4371 4372 return work; 4373 } 4374 4375 /** 4376 * __napi_schedule - schedule for receive 4377 * @n: entry to schedule 4378 * 4379 * The entry's receive function will be scheduled to run 4380 */ 4381 void __napi_schedule(struct napi_struct *n) 4382 { 4383 unsigned long flags; 4384 4385 local_irq_save(flags); 4386 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4387 local_irq_restore(flags); 4388 } 4389 EXPORT_SYMBOL(__napi_schedule); 4390 4391 void __napi_complete(struct napi_struct *n) 4392 { 4393 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 4394 BUG_ON(n->gro_list); 4395 4396 list_del(&n->poll_list); 4397 smp_mb__before_atomic(); 4398 clear_bit(NAPI_STATE_SCHED, &n->state); 4399 } 4400 EXPORT_SYMBOL(__napi_complete); 4401 4402 void napi_complete(struct napi_struct *n) 4403 { 4404 unsigned long flags; 4405 4406 /* 4407 * don't let napi dequeue from the cpu poll list 4408 * just in case its running on a different cpu 4409 */ 4410 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) 4411 return; 4412 4413 napi_gro_flush(n, false); 4414 local_irq_save(flags); 4415 __napi_complete(n); 4416 local_irq_restore(flags); 4417 } 4418 EXPORT_SYMBOL(napi_complete); 4419 4420 /* must be called under rcu_read_lock(), as we dont take a reference */ 4421 struct napi_struct *napi_by_id(unsigned int napi_id) 4422 { 4423 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 4424 struct napi_struct *napi; 4425 4426 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 4427 if (napi->napi_id == napi_id) 4428 return napi; 4429 4430 return NULL; 4431 } 4432 EXPORT_SYMBOL_GPL(napi_by_id); 4433 4434 void napi_hash_add(struct napi_struct *napi) 4435 { 4436 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { 4437 4438 spin_lock(&napi_hash_lock); 4439 4440 /* 0 is not a valid id, we also skip an id that is taken 4441 * we expect both events to be extremely rare 4442 */ 4443 napi->napi_id = 0; 4444 while (!napi->napi_id) { 4445 napi->napi_id = ++napi_gen_id; 4446 if (napi_by_id(napi->napi_id)) 4447 napi->napi_id = 0; 4448 } 4449 4450 hlist_add_head_rcu(&napi->napi_hash_node, 4451 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 4452 4453 spin_unlock(&napi_hash_lock); 4454 } 4455 } 4456 EXPORT_SYMBOL_GPL(napi_hash_add); 4457 4458 /* Warning : caller is responsible to make sure rcu grace period 4459 * is respected before freeing memory containing @napi 4460 */ 4461 void napi_hash_del(struct napi_struct *napi) 4462 { 4463 spin_lock(&napi_hash_lock); 4464 4465 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) 4466 hlist_del_rcu(&napi->napi_hash_node); 4467 4468 spin_unlock(&napi_hash_lock); 4469 } 4470 EXPORT_SYMBOL_GPL(napi_hash_del); 4471 4472 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 4473 int (*poll)(struct napi_struct *, int), int weight) 4474 { 4475 INIT_LIST_HEAD(&napi->poll_list); 4476 napi->gro_count = 0; 4477 napi->gro_list = NULL; 4478 napi->skb = NULL; 4479 napi->poll = poll; 4480 if (weight > NAPI_POLL_WEIGHT) 4481 pr_err_once("netif_napi_add() called with weight %d on device %s\n", 4482 weight, dev->name); 4483 napi->weight = weight; 4484 list_add(&napi->dev_list, &dev->napi_list); 4485 napi->dev = dev; 4486 #ifdef CONFIG_NETPOLL 4487 spin_lock_init(&napi->poll_lock); 4488 napi->poll_owner = -1; 4489 #endif 4490 set_bit(NAPI_STATE_SCHED, &napi->state); 4491 } 4492 EXPORT_SYMBOL(netif_napi_add); 4493 4494 void netif_napi_del(struct napi_struct *napi) 4495 { 4496 list_del_init(&napi->dev_list); 4497 napi_free_frags(napi); 4498 4499 kfree_skb_list(napi->gro_list); 4500 napi->gro_list = NULL; 4501 napi->gro_count = 0; 4502 } 4503 EXPORT_SYMBOL(netif_napi_del); 4504 4505 static void net_rx_action(struct softirq_action *h) 4506 { 4507 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4508 unsigned long time_limit = jiffies + 2; 4509 int budget = netdev_budget; 4510 void *have; 4511 4512 local_irq_disable(); 4513 4514 while (!list_empty(&sd->poll_list)) { 4515 struct napi_struct *n; 4516 int work, weight; 4517 4518 /* If softirq window is exhuasted then punt. 4519 * Allow this to run for 2 jiffies since which will allow 4520 * an average latency of 1.5/HZ. 4521 */ 4522 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit))) 4523 goto softnet_break; 4524 4525 local_irq_enable(); 4526 4527 /* Even though interrupts have been re-enabled, this 4528 * access is safe because interrupts can only add new 4529 * entries to the tail of this list, and only ->poll() 4530 * calls can remove this head entry from the list. 4531 */ 4532 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); 4533 4534 have = netpoll_poll_lock(n); 4535 4536 weight = n->weight; 4537 4538 /* This NAPI_STATE_SCHED test is for avoiding a race 4539 * with netpoll's poll_napi(). Only the entity which 4540 * obtains the lock and sees NAPI_STATE_SCHED set will 4541 * actually make the ->poll() call. Therefore we avoid 4542 * accidentally calling ->poll() when NAPI is not scheduled. 4543 */ 4544 work = 0; 4545 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 4546 work = n->poll(n, weight); 4547 trace_napi_poll(n); 4548 } 4549 4550 WARN_ON_ONCE(work > weight); 4551 4552 budget -= work; 4553 4554 local_irq_disable(); 4555 4556 /* Drivers must not modify the NAPI state if they 4557 * consume the entire weight. In such cases this code 4558 * still "owns" the NAPI instance and therefore can 4559 * move the instance around on the list at-will. 4560 */ 4561 if (unlikely(work == weight)) { 4562 if (unlikely(napi_disable_pending(n))) { 4563 local_irq_enable(); 4564 napi_complete(n); 4565 local_irq_disable(); 4566 } else { 4567 if (n->gro_list) { 4568 /* flush too old packets 4569 * If HZ < 1000, flush all packets. 4570 */ 4571 local_irq_enable(); 4572 napi_gro_flush(n, HZ >= 1000); 4573 local_irq_disable(); 4574 } 4575 list_move_tail(&n->poll_list, &sd->poll_list); 4576 } 4577 } 4578 4579 netpoll_poll_unlock(have); 4580 } 4581 out: 4582 net_rps_action_and_irq_enable(sd); 4583 4584 return; 4585 4586 softnet_break: 4587 sd->time_squeeze++; 4588 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4589 goto out; 4590 } 4591 4592 struct netdev_adjacent { 4593 struct net_device *dev; 4594 4595 /* upper master flag, there can only be one master device per list */ 4596 bool master; 4597 4598 /* counter for the number of times this device was added to us */ 4599 u16 ref_nr; 4600 4601 /* private field for the users */ 4602 void *private; 4603 4604 struct list_head list; 4605 struct rcu_head rcu; 4606 }; 4607 4608 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, 4609 struct net_device *adj_dev, 4610 struct list_head *adj_list) 4611 { 4612 struct netdev_adjacent *adj; 4613 4614 list_for_each_entry(adj, adj_list, list) { 4615 if (adj->dev == adj_dev) 4616 return adj; 4617 } 4618 return NULL; 4619 } 4620 4621 /** 4622 * netdev_has_upper_dev - Check if device is linked to an upper device 4623 * @dev: device 4624 * @upper_dev: upper device to check 4625 * 4626 * Find out if a device is linked to specified upper device and return true 4627 * in case it is. Note that this checks only immediate upper device, 4628 * not through a complete stack of devices. The caller must hold the RTNL lock. 4629 */ 4630 bool netdev_has_upper_dev(struct net_device *dev, 4631 struct net_device *upper_dev) 4632 { 4633 ASSERT_RTNL(); 4634 4635 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); 4636 } 4637 EXPORT_SYMBOL(netdev_has_upper_dev); 4638 4639 /** 4640 * netdev_has_any_upper_dev - Check if device is linked to some device 4641 * @dev: device 4642 * 4643 * Find out if a device is linked to an upper device and return true in case 4644 * it is. The caller must hold the RTNL lock. 4645 */ 4646 static bool netdev_has_any_upper_dev(struct net_device *dev) 4647 { 4648 ASSERT_RTNL(); 4649 4650 return !list_empty(&dev->all_adj_list.upper); 4651 } 4652 4653 /** 4654 * netdev_master_upper_dev_get - Get master upper device 4655 * @dev: device 4656 * 4657 * Find a master upper device and return pointer to it or NULL in case 4658 * it's not there. The caller must hold the RTNL lock. 4659 */ 4660 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 4661 { 4662 struct netdev_adjacent *upper; 4663 4664 ASSERT_RTNL(); 4665 4666 if (list_empty(&dev->adj_list.upper)) 4667 return NULL; 4668 4669 upper = list_first_entry(&dev->adj_list.upper, 4670 struct netdev_adjacent, list); 4671 if (likely(upper->master)) 4672 return upper->dev; 4673 return NULL; 4674 } 4675 EXPORT_SYMBOL(netdev_master_upper_dev_get); 4676 4677 void *netdev_adjacent_get_private(struct list_head *adj_list) 4678 { 4679 struct netdev_adjacent *adj; 4680 4681 adj = list_entry(adj_list, struct netdev_adjacent, list); 4682 4683 return adj->private; 4684 } 4685 EXPORT_SYMBOL(netdev_adjacent_get_private); 4686 4687 /** 4688 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 4689 * @dev: device 4690 * @iter: list_head ** of the current position 4691 * 4692 * Gets the next device from the dev's upper list, starting from iter 4693 * position. The caller must hold RCU read lock. 4694 */ 4695 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 4696 struct list_head **iter) 4697 { 4698 struct netdev_adjacent *upper; 4699 4700 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 4701 4702 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 4703 4704 if (&upper->list == &dev->adj_list.upper) 4705 return NULL; 4706 4707 *iter = &upper->list; 4708 4709 return upper->dev; 4710 } 4711 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 4712 4713 /** 4714 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list 4715 * @dev: device 4716 * @iter: list_head ** of the current position 4717 * 4718 * Gets the next device from the dev's upper list, starting from iter 4719 * position. The caller must hold RCU read lock. 4720 */ 4721 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, 4722 struct list_head **iter) 4723 { 4724 struct netdev_adjacent *upper; 4725 4726 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 4727 4728 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 4729 4730 if (&upper->list == &dev->all_adj_list.upper) 4731 return NULL; 4732 4733 *iter = &upper->list; 4734 4735 return upper->dev; 4736 } 4737 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); 4738 4739 /** 4740 * netdev_lower_get_next_private - Get the next ->private from the 4741 * lower neighbour list 4742 * @dev: device 4743 * @iter: list_head ** of the current position 4744 * 4745 * Gets the next netdev_adjacent->private from the dev's lower neighbour 4746 * list, starting from iter position. The caller must hold either hold the 4747 * RTNL lock or its own locking that guarantees that the neighbour lower 4748 * list will remain unchainged. 4749 */ 4750 void *netdev_lower_get_next_private(struct net_device *dev, 4751 struct list_head **iter) 4752 { 4753 struct netdev_adjacent *lower; 4754 4755 lower = list_entry(*iter, struct netdev_adjacent, list); 4756 4757 if (&lower->list == &dev->adj_list.lower) 4758 return NULL; 4759 4760 *iter = lower->list.next; 4761 4762 return lower->private; 4763 } 4764 EXPORT_SYMBOL(netdev_lower_get_next_private); 4765 4766 /** 4767 * netdev_lower_get_next_private_rcu - Get the next ->private from the 4768 * lower neighbour list, RCU 4769 * variant 4770 * @dev: device 4771 * @iter: list_head ** of the current position 4772 * 4773 * Gets the next netdev_adjacent->private from the dev's lower neighbour 4774 * list, starting from iter position. The caller must hold RCU read lock. 4775 */ 4776 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 4777 struct list_head **iter) 4778 { 4779 struct netdev_adjacent *lower; 4780 4781 WARN_ON_ONCE(!rcu_read_lock_held()); 4782 4783 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 4784 4785 if (&lower->list == &dev->adj_list.lower) 4786 return NULL; 4787 4788 *iter = &lower->list; 4789 4790 return lower->private; 4791 } 4792 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 4793 4794 /** 4795 * netdev_lower_get_next - Get the next device from the lower neighbour 4796 * list 4797 * @dev: device 4798 * @iter: list_head ** of the current position 4799 * 4800 * Gets the next netdev_adjacent from the dev's lower neighbour 4801 * list, starting from iter position. The caller must hold RTNL lock or 4802 * its own locking that guarantees that the neighbour lower 4803 * list will remain unchainged. 4804 */ 4805 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 4806 { 4807 struct netdev_adjacent *lower; 4808 4809 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 4810 4811 if (&lower->list == &dev->adj_list.lower) 4812 return NULL; 4813 4814 *iter = &lower->list; 4815 4816 return lower->dev; 4817 } 4818 EXPORT_SYMBOL(netdev_lower_get_next); 4819 4820 /** 4821 * netdev_lower_get_first_private_rcu - Get the first ->private from the 4822 * lower neighbour list, RCU 4823 * variant 4824 * @dev: device 4825 * 4826 * Gets the first netdev_adjacent->private from the dev's lower neighbour 4827 * list. The caller must hold RCU read lock. 4828 */ 4829 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 4830 { 4831 struct netdev_adjacent *lower; 4832 4833 lower = list_first_or_null_rcu(&dev->adj_list.lower, 4834 struct netdev_adjacent, list); 4835 if (lower) 4836 return lower->private; 4837 return NULL; 4838 } 4839 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 4840 4841 /** 4842 * netdev_master_upper_dev_get_rcu - Get master upper device 4843 * @dev: device 4844 * 4845 * Find a master upper device and return pointer to it or NULL in case 4846 * it's not there. The caller must hold the RCU read lock. 4847 */ 4848 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 4849 { 4850 struct netdev_adjacent *upper; 4851 4852 upper = list_first_or_null_rcu(&dev->adj_list.upper, 4853 struct netdev_adjacent, list); 4854 if (upper && likely(upper->master)) 4855 return upper->dev; 4856 return NULL; 4857 } 4858 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 4859 4860 static int netdev_adjacent_sysfs_add(struct net_device *dev, 4861 struct net_device *adj_dev, 4862 struct list_head *dev_list) 4863 { 4864 char linkname[IFNAMSIZ+7]; 4865 sprintf(linkname, dev_list == &dev->adj_list.upper ? 4866 "upper_%s" : "lower_%s", adj_dev->name); 4867 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 4868 linkname); 4869 } 4870 static void netdev_adjacent_sysfs_del(struct net_device *dev, 4871 char *name, 4872 struct list_head *dev_list) 4873 { 4874 char linkname[IFNAMSIZ+7]; 4875 sprintf(linkname, dev_list == &dev->adj_list.upper ? 4876 "upper_%s" : "lower_%s", name); 4877 sysfs_remove_link(&(dev->dev.kobj), linkname); 4878 } 4879 4880 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 4881 struct net_device *adj_dev, 4882 struct list_head *dev_list) 4883 { 4884 return (dev_list == &dev->adj_list.upper || 4885 dev_list == &dev->adj_list.lower) && 4886 net_eq(dev_net(dev), dev_net(adj_dev)); 4887 } 4888 4889 static int __netdev_adjacent_dev_insert(struct net_device *dev, 4890 struct net_device *adj_dev, 4891 struct list_head *dev_list, 4892 void *private, bool master) 4893 { 4894 struct netdev_adjacent *adj; 4895 int ret; 4896 4897 adj = __netdev_find_adj(dev, adj_dev, dev_list); 4898 4899 if (adj) { 4900 adj->ref_nr++; 4901 return 0; 4902 } 4903 4904 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 4905 if (!adj) 4906 return -ENOMEM; 4907 4908 adj->dev = adj_dev; 4909 adj->master = master; 4910 adj->ref_nr = 1; 4911 adj->private = private; 4912 dev_hold(adj_dev); 4913 4914 pr_debug("dev_hold for %s, because of link added from %s to %s\n", 4915 adj_dev->name, dev->name, adj_dev->name); 4916 4917 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 4918 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 4919 if (ret) 4920 goto free_adj; 4921 } 4922 4923 /* Ensure that master link is always the first item in list. */ 4924 if (master) { 4925 ret = sysfs_create_link(&(dev->dev.kobj), 4926 &(adj_dev->dev.kobj), "master"); 4927 if (ret) 4928 goto remove_symlinks; 4929 4930 list_add_rcu(&adj->list, dev_list); 4931 } else { 4932 list_add_tail_rcu(&adj->list, dev_list); 4933 } 4934 4935 return 0; 4936 4937 remove_symlinks: 4938 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 4939 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 4940 free_adj: 4941 kfree(adj); 4942 dev_put(adj_dev); 4943 4944 return ret; 4945 } 4946 4947 static void __netdev_adjacent_dev_remove(struct net_device *dev, 4948 struct net_device *adj_dev, 4949 struct list_head *dev_list) 4950 { 4951 struct netdev_adjacent *adj; 4952 4953 adj = __netdev_find_adj(dev, adj_dev, dev_list); 4954 4955 if (!adj) { 4956 pr_err("tried to remove device %s from %s\n", 4957 dev->name, adj_dev->name); 4958 BUG(); 4959 } 4960 4961 if (adj->ref_nr > 1) { 4962 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, 4963 adj->ref_nr-1); 4964 adj->ref_nr--; 4965 return; 4966 } 4967 4968 if (adj->master) 4969 sysfs_remove_link(&(dev->dev.kobj), "master"); 4970 4971 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 4972 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 4973 4974 list_del_rcu(&adj->list); 4975 pr_debug("dev_put for %s, because link removed from %s to %s\n", 4976 adj_dev->name, dev->name, adj_dev->name); 4977 dev_put(adj_dev); 4978 kfree_rcu(adj, rcu); 4979 } 4980 4981 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 4982 struct net_device *upper_dev, 4983 struct list_head *up_list, 4984 struct list_head *down_list, 4985 void *private, bool master) 4986 { 4987 int ret; 4988 4989 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, 4990 master); 4991 if (ret) 4992 return ret; 4993 4994 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, 4995 false); 4996 if (ret) { 4997 __netdev_adjacent_dev_remove(dev, upper_dev, up_list); 4998 return ret; 4999 } 5000 5001 return 0; 5002 } 5003 5004 static int __netdev_adjacent_dev_link(struct net_device *dev, 5005 struct net_device *upper_dev) 5006 { 5007 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 5008 &dev->all_adj_list.upper, 5009 &upper_dev->all_adj_list.lower, 5010 NULL, false); 5011 } 5012 5013 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 5014 struct net_device *upper_dev, 5015 struct list_head *up_list, 5016 struct list_head *down_list) 5017 { 5018 __netdev_adjacent_dev_remove(dev, upper_dev, up_list); 5019 __netdev_adjacent_dev_remove(upper_dev, dev, down_list); 5020 } 5021 5022 static void __netdev_adjacent_dev_unlink(struct net_device *dev, 5023 struct net_device *upper_dev) 5024 { 5025 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 5026 &dev->all_adj_list.upper, 5027 &upper_dev->all_adj_list.lower); 5028 } 5029 5030 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 5031 struct net_device *upper_dev, 5032 void *private, bool master) 5033 { 5034 int ret = __netdev_adjacent_dev_link(dev, upper_dev); 5035 5036 if (ret) 5037 return ret; 5038 5039 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 5040 &dev->adj_list.upper, 5041 &upper_dev->adj_list.lower, 5042 private, master); 5043 if (ret) { 5044 __netdev_adjacent_dev_unlink(dev, upper_dev); 5045 return ret; 5046 } 5047 5048 return 0; 5049 } 5050 5051 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 5052 struct net_device *upper_dev) 5053 { 5054 __netdev_adjacent_dev_unlink(dev, upper_dev); 5055 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 5056 &dev->adj_list.upper, 5057 &upper_dev->adj_list.lower); 5058 } 5059 5060 static int __netdev_upper_dev_link(struct net_device *dev, 5061 struct net_device *upper_dev, bool master, 5062 void *private) 5063 { 5064 struct netdev_adjacent *i, *j, *to_i, *to_j; 5065 int ret = 0; 5066 5067 ASSERT_RTNL(); 5068 5069 if (dev == upper_dev) 5070 return -EBUSY; 5071 5072 /* To prevent loops, check if dev is not upper device to upper_dev. */ 5073 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) 5074 return -EBUSY; 5075 5076 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper)) 5077 return -EEXIST; 5078 5079 if (master && netdev_master_upper_dev_get(dev)) 5080 return -EBUSY; 5081 5082 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, 5083 master); 5084 if (ret) 5085 return ret; 5086 5087 /* Now that we linked these devs, make all the upper_dev's 5088 * all_adj_list.upper visible to every dev's all_adj_list.lower an 5089 * versa, and don't forget the devices itself. All of these 5090 * links are non-neighbours. 5091 */ 5092 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5093 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { 5094 pr_debug("Interlinking %s with %s, non-neighbour\n", 5095 i->dev->name, j->dev->name); 5096 ret = __netdev_adjacent_dev_link(i->dev, j->dev); 5097 if (ret) 5098 goto rollback_mesh; 5099 } 5100 } 5101 5102 /* add dev to every upper_dev's upper device */ 5103 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { 5104 pr_debug("linking %s's upper device %s with %s\n", 5105 upper_dev->name, i->dev->name, dev->name); 5106 ret = __netdev_adjacent_dev_link(dev, i->dev); 5107 if (ret) 5108 goto rollback_upper_mesh; 5109 } 5110 5111 /* add upper_dev to every dev's lower device */ 5112 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5113 pr_debug("linking %s's lower device %s with %s\n", dev->name, 5114 i->dev->name, upper_dev->name); 5115 ret = __netdev_adjacent_dev_link(i->dev, upper_dev); 5116 if (ret) 5117 goto rollback_lower_mesh; 5118 } 5119 5120 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); 5121 return 0; 5122 5123 rollback_lower_mesh: 5124 to_i = i; 5125 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5126 if (i == to_i) 5127 break; 5128 __netdev_adjacent_dev_unlink(i->dev, upper_dev); 5129 } 5130 5131 i = NULL; 5132 5133 rollback_upper_mesh: 5134 to_i = i; 5135 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { 5136 if (i == to_i) 5137 break; 5138 __netdev_adjacent_dev_unlink(dev, i->dev); 5139 } 5140 5141 i = j = NULL; 5142 5143 rollback_mesh: 5144 to_i = i; 5145 to_j = j; 5146 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5147 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { 5148 if (i == to_i && j == to_j) 5149 break; 5150 __netdev_adjacent_dev_unlink(i->dev, j->dev); 5151 } 5152 if (i == to_i) 5153 break; 5154 } 5155 5156 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5157 5158 return ret; 5159 } 5160 5161 /** 5162 * netdev_upper_dev_link - Add a link to the upper device 5163 * @dev: device 5164 * @upper_dev: new upper device 5165 * 5166 * Adds a link to device which is upper to this one. The caller must hold 5167 * the RTNL lock. On a failure a negative errno code is returned. 5168 * On success the reference counts are adjusted and the function 5169 * returns zero. 5170 */ 5171 int netdev_upper_dev_link(struct net_device *dev, 5172 struct net_device *upper_dev) 5173 { 5174 return __netdev_upper_dev_link(dev, upper_dev, false, NULL); 5175 } 5176 EXPORT_SYMBOL(netdev_upper_dev_link); 5177 5178 /** 5179 * netdev_master_upper_dev_link - Add a master link to the upper device 5180 * @dev: device 5181 * @upper_dev: new upper device 5182 * 5183 * Adds a link to device which is upper to this one. In this case, only 5184 * one master upper device can be linked, although other non-master devices 5185 * might be linked as well. The caller must hold the RTNL lock. 5186 * On a failure a negative errno code is returned. On success the reference 5187 * counts are adjusted and the function returns zero. 5188 */ 5189 int netdev_master_upper_dev_link(struct net_device *dev, 5190 struct net_device *upper_dev) 5191 { 5192 return __netdev_upper_dev_link(dev, upper_dev, true, NULL); 5193 } 5194 EXPORT_SYMBOL(netdev_master_upper_dev_link); 5195 5196 int netdev_master_upper_dev_link_private(struct net_device *dev, 5197 struct net_device *upper_dev, 5198 void *private) 5199 { 5200 return __netdev_upper_dev_link(dev, upper_dev, true, private); 5201 } 5202 EXPORT_SYMBOL(netdev_master_upper_dev_link_private); 5203 5204 /** 5205 * netdev_upper_dev_unlink - Removes a link to upper device 5206 * @dev: device 5207 * @upper_dev: new upper device 5208 * 5209 * Removes a link to device which is upper to this one. The caller must hold 5210 * the RTNL lock. 5211 */ 5212 void netdev_upper_dev_unlink(struct net_device *dev, 5213 struct net_device *upper_dev) 5214 { 5215 struct netdev_adjacent *i, *j; 5216 ASSERT_RTNL(); 5217 5218 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5219 5220 /* Here is the tricky part. We must remove all dev's lower 5221 * devices from all upper_dev's upper devices and vice 5222 * versa, to maintain the graph relationship. 5223 */ 5224 list_for_each_entry(i, &dev->all_adj_list.lower, list) 5225 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) 5226 __netdev_adjacent_dev_unlink(i->dev, j->dev); 5227 5228 /* remove also the devices itself from lower/upper device 5229 * list 5230 */ 5231 list_for_each_entry(i, &dev->all_adj_list.lower, list) 5232 __netdev_adjacent_dev_unlink(i->dev, upper_dev); 5233 5234 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) 5235 __netdev_adjacent_dev_unlink(dev, i->dev); 5236 5237 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); 5238 } 5239 EXPORT_SYMBOL(netdev_upper_dev_unlink); 5240 5241 void netdev_adjacent_add_links(struct net_device *dev) 5242 { 5243 struct netdev_adjacent *iter; 5244 5245 struct net *net = dev_net(dev); 5246 5247 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5248 if (!net_eq(net,dev_net(iter->dev))) 5249 continue; 5250 netdev_adjacent_sysfs_add(iter->dev, dev, 5251 &iter->dev->adj_list.lower); 5252 netdev_adjacent_sysfs_add(dev, iter->dev, 5253 &dev->adj_list.upper); 5254 } 5255 5256 list_for_each_entry(iter, &dev->adj_list.lower, list) { 5257 if (!net_eq(net,dev_net(iter->dev))) 5258 continue; 5259 netdev_adjacent_sysfs_add(iter->dev, dev, 5260 &iter->dev->adj_list.upper); 5261 netdev_adjacent_sysfs_add(dev, iter->dev, 5262 &dev->adj_list.lower); 5263 } 5264 } 5265 5266 void netdev_adjacent_del_links(struct net_device *dev) 5267 { 5268 struct netdev_adjacent *iter; 5269 5270 struct net *net = dev_net(dev); 5271 5272 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5273 if (!net_eq(net,dev_net(iter->dev))) 5274 continue; 5275 netdev_adjacent_sysfs_del(iter->dev, dev->name, 5276 &iter->dev->adj_list.lower); 5277 netdev_adjacent_sysfs_del(dev, iter->dev->name, 5278 &dev->adj_list.upper); 5279 } 5280 5281 list_for_each_entry(iter, &dev->adj_list.lower, list) { 5282 if (!net_eq(net,dev_net(iter->dev))) 5283 continue; 5284 netdev_adjacent_sysfs_del(iter->dev, dev->name, 5285 &iter->dev->adj_list.upper); 5286 netdev_adjacent_sysfs_del(dev, iter->dev->name, 5287 &dev->adj_list.lower); 5288 } 5289 } 5290 5291 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 5292 { 5293 struct netdev_adjacent *iter; 5294 5295 struct net *net = dev_net(dev); 5296 5297 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5298 if (!net_eq(net,dev_net(iter->dev))) 5299 continue; 5300 netdev_adjacent_sysfs_del(iter->dev, oldname, 5301 &iter->dev->adj_list.lower); 5302 netdev_adjacent_sysfs_add(iter->dev, dev, 5303 &iter->dev->adj_list.lower); 5304 } 5305 5306 list_for_each_entry(iter, &dev->adj_list.lower, list) { 5307 if (!net_eq(net,dev_net(iter->dev))) 5308 continue; 5309 netdev_adjacent_sysfs_del(iter->dev, oldname, 5310 &iter->dev->adj_list.upper); 5311 netdev_adjacent_sysfs_add(iter->dev, dev, 5312 &iter->dev->adj_list.upper); 5313 } 5314 } 5315 5316 void *netdev_lower_dev_get_private(struct net_device *dev, 5317 struct net_device *lower_dev) 5318 { 5319 struct netdev_adjacent *lower; 5320 5321 if (!lower_dev) 5322 return NULL; 5323 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); 5324 if (!lower) 5325 return NULL; 5326 5327 return lower->private; 5328 } 5329 EXPORT_SYMBOL(netdev_lower_dev_get_private); 5330 5331 5332 int dev_get_nest_level(struct net_device *dev, 5333 bool (*type_check)(struct net_device *dev)) 5334 { 5335 struct net_device *lower = NULL; 5336 struct list_head *iter; 5337 int max_nest = -1; 5338 int nest; 5339 5340 ASSERT_RTNL(); 5341 5342 netdev_for_each_lower_dev(dev, lower, iter) { 5343 nest = dev_get_nest_level(lower, type_check); 5344 if (max_nest < nest) 5345 max_nest = nest; 5346 } 5347 5348 if (type_check(dev)) 5349 max_nest++; 5350 5351 return max_nest; 5352 } 5353 EXPORT_SYMBOL(dev_get_nest_level); 5354 5355 static void dev_change_rx_flags(struct net_device *dev, int flags) 5356 { 5357 const struct net_device_ops *ops = dev->netdev_ops; 5358 5359 if (ops->ndo_change_rx_flags) 5360 ops->ndo_change_rx_flags(dev, flags); 5361 } 5362 5363 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 5364 { 5365 unsigned int old_flags = dev->flags; 5366 kuid_t uid; 5367 kgid_t gid; 5368 5369 ASSERT_RTNL(); 5370 5371 dev->flags |= IFF_PROMISC; 5372 dev->promiscuity += inc; 5373 if (dev->promiscuity == 0) { 5374 /* 5375 * Avoid overflow. 5376 * If inc causes overflow, untouch promisc and return error. 5377 */ 5378 if (inc < 0) 5379 dev->flags &= ~IFF_PROMISC; 5380 else { 5381 dev->promiscuity -= inc; 5382 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 5383 dev->name); 5384 return -EOVERFLOW; 5385 } 5386 } 5387 if (dev->flags != old_flags) { 5388 pr_info("device %s %s promiscuous mode\n", 5389 dev->name, 5390 dev->flags & IFF_PROMISC ? "entered" : "left"); 5391 if (audit_enabled) { 5392 current_uid_gid(&uid, &gid); 5393 audit_log(current->audit_context, GFP_ATOMIC, 5394 AUDIT_ANOM_PROMISCUOUS, 5395 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 5396 dev->name, (dev->flags & IFF_PROMISC), 5397 (old_flags & IFF_PROMISC), 5398 from_kuid(&init_user_ns, audit_get_loginuid(current)), 5399 from_kuid(&init_user_ns, uid), 5400 from_kgid(&init_user_ns, gid), 5401 audit_get_sessionid(current)); 5402 } 5403 5404 dev_change_rx_flags(dev, IFF_PROMISC); 5405 } 5406 if (notify) 5407 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 5408 return 0; 5409 } 5410 5411 /** 5412 * dev_set_promiscuity - update promiscuity count on a device 5413 * @dev: device 5414 * @inc: modifier 5415 * 5416 * Add or remove promiscuity from a device. While the count in the device 5417 * remains above zero the interface remains promiscuous. Once it hits zero 5418 * the device reverts back to normal filtering operation. A negative inc 5419 * value is used to drop promiscuity on the device. 5420 * Return 0 if successful or a negative errno code on error. 5421 */ 5422 int dev_set_promiscuity(struct net_device *dev, int inc) 5423 { 5424 unsigned int old_flags = dev->flags; 5425 int err; 5426 5427 err = __dev_set_promiscuity(dev, inc, true); 5428 if (err < 0) 5429 return err; 5430 if (dev->flags != old_flags) 5431 dev_set_rx_mode(dev); 5432 return err; 5433 } 5434 EXPORT_SYMBOL(dev_set_promiscuity); 5435 5436 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 5437 { 5438 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 5439 5440 ASSERT_RTNL(); 5441 5442 dev->flags |= IFF_ALLMULTI; 5443 dev->allmulti += inc; 5444 if (dev->allmulti == 0) { 5445 /* 5446 * Avoid overflow. 5447 * If inc causes overflow, untouch allmulti and return error. 5448 */ 5449 if (inc < 0) 5450 dev->flags &= ~IFF_ALLMULTI; 5451 else { 5452 dev->allmulti -= inc; 5453 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 5454 dev->name); 5455 return -EOVERFLOW; 5456 } 5457 } 5458 if (dev->flags ^ old_flags) { 5459 dev_change_rx_flags(dev, IFF_ALLMULTI); 5460 dev_set_rx_mode(dev); 5461 if (notify) 5462 __dev_notify_flags(dev, old_flags, 5463 dev->gflags ^ old_gflags); 5464 } 5465 return 0; 5466 } 5467 5468 /** 5469 * dev_set_allmulti - update allmulti count on a device 5470 * @dev: device 5471 * @inc: modifier 5472 * 5473 * Add or remove reception of all multicast frames to a device. While the 5474 * count in the device remains above zero the interface remains listening 5475 * to all interfaces. Once it hits zero the device reverts back to normal 5476 * filtering operation. A negative @inc value is used to drop the counter 5477 * when releasing a resource needing all multicasts. 5478 * Return 0 if successful or a negative errno code on error. 5479 */ 5480 5481 int dev_set_allmulti(struct net_device *dev, int inc) 5482 { 5483 return __dev_set_allmulti(dev, inc, true); 5484 } 5485 EXPORT_SYMBOL(dev_set_allmulti); 5486 5487 /* 5488 * Upload unicast and multicast address lists to device and 5489 * configure RX filtering. When the device doesn't support unicast 5490 * filtering it is put in promiscuous mode while unicast addresses 5491 * are present. 5492 */ 5493 void __dev_set_rx_mode(struct net_device *dev) 5494 { 5495 const struct net_device_ops *ops = dev->netdev_ops; 5496 5497 /* dev_open will call this function so the list will stay sane. */ 5498 if (!(dev->flags&IFF_UP)) 5499 return; 5500 5501 if (!netif_device_present(dev)) 5502 return; 5503 5504 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 5505 /* Unicast addresses changes may only happen under the rtnl, 5506 * therefore calling __dev_set_promiscuity here is safe. 5507 */ 5508 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 5509 __dev_set_promiscuity(dev, 1, false); 5510 dev->uc_promisc = true; 5511 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 5512 __dev_set_promiscuity(dev, -1, false); 5513 dev->uc_promisc = false; 5514 } 5515 } 5516 5517 if (ops->ndo_set_rx_mode) 5518 ops->ndo_set_rx_mode(dev); 5519 } 5520 5521 void dev_set_rx_mode(struct net_device *dev) 5522 { 5523 netif_addr_lock_bh(dev); 5524 __dev_set_rx_mode(dev); 5525 netif_addr_unlock_bh(dev); 5526 } 5527 5528 /** 5529 * dev_get_flags - get flags reported to userspace 5530 * @dev: device 5531 * 5532 * Get the combination of flag bits exported through APIs to userspace. 5533 */ 5534 unsigned int dev_get_flags(const struct net_device *dev) 5535 { 5536 unsigned int flags; 5537 5538 flags = (dev->flags & ~(IFF_PROMISC | 5539 IFF_ALLMULTI | 5540 IFF_RUNNING | 5541 IFF_LOWER_UP | 5542 IFF_DORMANT)) | 5543 (dev->gflags & (IFF_PROMISC | 5544 IFF_ALLMULTI)); 5545 5546 if (netif_running(dev)) { 5547 if (netif_oper_up(dev)) 5548 flags |= IFF_RUNNING; 5549 if (netif_carrier_ok(dev)) 5550 flags |= IFF_LOWER_UP; 5551 if (netif_dormant(dev)) 5552 flags |= IFF_DORMANT; 5553 } 5554 5555 return flags; 5556 } 5557 EXPORT_SYMBOL(dev_get_flags); 5558 5559 int __dev_change_flags(struct net_device *dev, unsigned int flags) 5560 { 5561 unsigned int old_flags = dev->flags; 5562 int ret; 5563 5564 ASSERT_RTNL(); 5565 5566 /* 5567 * Set the flags on our device. 5568 */ 5569 5570 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 5571 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 5572 IFF_AUTOMEDIA)) | 5573 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 5574 IFF_ALLMULTI)); 5575 5576 /* 5577 * Load in the correct multicast list now the flags have changed. 5578 */ 5579 5580 if ((old_flags ^ flags) & IFF_MULTICAST) 5581 dev_change_rx_flags(dev, IFF_MULTICAST); 5582 5583 dev_set_rx_mode(dev); 5584 5585 /* 5586 * Have we downed the interface. We handle IFF_UP ourselves 5587 * according to user attempts to set it, rather than blindly 5588 * setting it. 5589 */ 5590 5591 ret = 0; 5592 if ((old_flags ^ flags) & IFF_UP) 5593 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); 5594 5595 if ((flags ^ dev->gflags) & IFF_PROMISC) { 5596 int inc = (flags & IFF_PROMISC) ? 1 : -1; 5597 unsigned int old_flags = dev->flags; 5598 5599 dev->gflags ^= IFF_PROMISC; 5600 5601 if (__dev_set_promiscuity(dev, inc, false) >= 0) 5602 if (dev->flags != old_flags) 5603 dev_set_rx_mode(dev); 5604 } 5605 5606 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 5607 is important. Some (broken) drivers set IFF_PROMISC, when 5608 IFF_ALLMULTI is requested not asking us and not reporting. 5609 */ 5610 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 5611 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 5612 5613 dev->gflags ^= IFF_ALLMULTI; 5614 __dev_set_allmulti(dev, inc, false); 5615 } 5616 5617 return ret; 5618 } 5619 5620 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 5621 unsigned int gchanges) 5622 { 5623 unsigned int changes = dev->flags ^ old_flags; 5624 5625 if (gchanges) 5626 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 5627 5628 if (changes & IFF_UP) { 5629 if (dev->flags & IFF_UP) 5630 call_netdevice_notifiers(NETDEV_UP, dev); 5631 else 5632 call_netdevice_notifiers(NETDEV_DOWN, dev); 5633 } 5634 5635 if (dev->flags & IFF_UP && 5636 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 5637 struct netdev_notifier_change_info change_info; 5638 5639 change_info.flags_changed = changes; 5640 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 5641 &change_info.info); 5642 } 5643 } 5644 5645 /** 5646 * dev_change_flags - change device settings 5647 * @dev: device 5648 * @flags: device state flags 5649 * 5650 * Change settings on device based state flags. The flags are 5651 * in the userspace exported format. 5652 */ 5653 int dev_change_flags(struct net_device *dev, unsigned int flags) 5654 { 5655 int ret; 5656 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 5657 5658 ret = __dev_change_flags(dev, flags); 5659 if (ret < 0) 5660 return ret; 5661 5662 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 5663 __dev_notify_flags(dev, old_flags, changes); 5664 return ret; 5665 } 5666 EXPORT_SYMBOL(dev_change_flags); 5667 5668 static int __dev_set_mtu(struct net_device *dev, int new_mtu) 5669 { 5670 const struct net_device_ops *ops = dev->netdev_ops; 5671 5672 if (ops->ndo_change_mtu) 5673 return ops->ndo_change_mtu(dev, new_mtu); 5674 5675 dev->mtu = new_mtu; 5676 return 0; 5677 } 5678 5679 /** 5680 * dev_set_mtu - Change maximum transfer unit 5681 * @dev: device 5682 * @new_mtu: new transfer unit 5683 * 5684 * Change the maximum transfer size of the network device. 5685 */ 5686 int dev_set_mtu(struct net_device *dev, int new_mtu) 5687 { 5688 int err, orig_mtu; 5689 5690 if (new_mtu == dev->mtu) 5691 return 0; 5692 5693 /* MTU must be positive. */ 5694 if (new_mtu < 0) 5695 return -EINVAL; 5696 5697 if (!netif_device_present(dev)) 5698 return -ENODEV; 5699 5700 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 5701 err = notifier_to_errno(err); 5702 if (err) 5703 return err; 5704 5705 orig_mtu = dev->mtu; 5706 err = __dev_set_mtu(dev, new_mtu); 5707 5708 if (!err) { 5709 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 5710 err = notifier_to_errno(err); 5711 if (err) { 5712 /* setting mtu back and notifying everyone again, 5713 * so that they have a chance to revert changes. 5714 */ 5715 __dev_set_mtu(dev, orig_mtu); 5716 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 5717 } 5718 } 5719 return err; 5720 } 5721 EXPORT_SYMBOL(dev_set_mtu); 5722 5723 /** 5724 * dev_set_group - Change group this device belongs to 5725 * @dev: device 5726 * @new_group: group this device should belong to 5727 */ 5728 void dev_set_group(struct net_device *dev, int new_group) 5729 { 5730 dev->group = new_group; 5731 } 5732 EXPORT_SYMBOL(dev_set_group); 5733 5734 /** 5735 * dev_set_mac_address - Change Media Access Control Address 5736 * @dev: device 5737 * @sa: new address 5738 * 5739 * Change the hardware (MAC) address of the device 5740 */ 5741 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 5742 { 5743 const struct net_device_ops *ops = dev->netdev_ops; 5744 int err; 5745 5746 if (!ops->ndo_set_mac_address) 5747 return -EOPNOTSUPP; 5748 if (sa->sa_family != dev->type) 5749 return -EINVAL; 5750 if (!netif_device_present(dev)) 5751 return -ENODEV; 5752 err = ops->ndo_set_mac_address(dev, sa); 5753 if (err) 5754 return err; 5755 dev->addr_assign_type = NET_ADDR_SET; 5756 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 5757 add_device_randomness(dev->dev_addr, dev->addr_len); 5758 return 0; 5759 } 5760 EXPORT_SYMBOL(dev_set_mac_address); 5761 5762 /** 5763 * dev_change_carrier - Change device carrier 5764 * @dev: device 5765 * @new_carrier: new value 5766 * 5767 * Change device carrier 5768 */ 5769 int dev_change_carrier(struct net_device *dev, bool new_carrier) 5770 { 5771 const struct net_device_ops *ops = dev->netdev_ops; 5772 5773 if (!ops->ndo_change_carrier) 5774 return -EOPNOTSUPP; 5775 if (!netif_device_present(dev)) 5776 return -ENODEV; 5777 return ops->ndo_change_carrier(dev, new_carrier); 5778 } 5779 EXPORT_SYMBOL(dev_change_carrier); 5780 5781 /** 5782 * dev_get_phys_port_id - Get device physical port ID 5783 * @dev: device 5784 * @ppid: port ID 5785 * 5786 * Get device physical port ID 5787 */ 5788 int dev_get_phys_port_id(struct net_device *dev, 5789 struct netdev_phys_port_id *ppid) 5790 { 5791 const struct net_device_ops *ops = dev->netdev_ops; 5792 5793 if (!ops->ndo_get_phys_port_id) 5794 return -EOPNOTSUPP; 5795 return ops->ndo_get_phys_port_id(dev, ppid); 5796 } 5797 EXPORT_SYMBOL(dev_get_phys_port_id); 5798 5799 /** 5800 * dev_new_index - allocate an ifindex 5801 * @net: the applicable net namespace 5802 * 5803 * Returns a suitable unique value for a new device interface 5804 * number. The caller must hold the rtnl semaphore or the 5805 * dev_base_lock to be sure it remains unique. 5806 */ 5807 static int dev_new_index(struct net *net) 5808 { 5809 int ifindex = net->ifindex; 5810 for (;;) { 5811 if (++ifindex <= 0) 5812 ifindex = 1; 5813 if (!__dev_get_by_index(net, ifindex)) 5814 return net->ifindex = ifindex; 5815 } 5816 } 5817 5818 /* Delayed registration/unregisteration */ 5819 static LIST_HEAD(net_todo_list); 5820 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 5821 5822 static void net_set_todo(struct net_device *dev) 5823 { 5824 list_add_tail(&dev->todo_list, &net_todo_list); 5825 dev_net(dev)->dev_unreg_count++; 5826 } 5827 5828 static void rollback_registered_many(struct list_head *head) 5829 { 5830 struct net_device *dev, *tmp; 5831 LIST_HEAD(close_head); 5832 5833 BUG_ON(dev_boot_phase); 5834 ASSERT_RTNL(); 5835 5836 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 5837 /* Some devices call without registering 5838 * for initialization unwind. Remove those 5839 * devices and proceed with the remaining. 5840 */ 5841 if (dev->reg_state == NETREG_UNINITIALIZED) { 5842 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 5843 dev->name, dev); 5844 5845 WARN_ON(1); 5846 list_del(&dev->unreg_list); 5847 continue; 5848 } 5849 dev->dismantle = true; 5850 BUG_ON(dev->reg_state != NETREG_REGISTERED); 5851 } 5852 5853 /* If device is running, close it first. */ 5854 list_for_each_entry(dev, head, unreg_list) 5855 list_add_tail(&dev->close_list, &close_head); 5856 dev_close_many(&close_head); 5857 5858 list_for_each_entry(dev, head, unreg_list) { 5859 /* And unlink it from device chain. */ 5860 unlist_netdevice(dev); 5861 5862 dev->reg_state = NETREG_UNREGISTERING; 5863 } 5864 5865 synchronize_net(); 5866 5867 list_for_each_entry(dev, head, unreg_list) { 5868 /* Shutdown queueing discipline. */ 5869 dev_shutdown(dev); 5870 5871 5872 /* Notify protocols, that we are about to destroy 5873 this device. They should clean all the things. 5874 */ 5875 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 5876 5877 /* 5878 * Flush the unicast and multicast chains 5879 */ 5880 dev_uc_flush(dev); 5881 dev_mc_flush(dev); 5882 5883 if (dev->netdev_ops->ndo_uninit) 5884 dev->netdev_ops->ndo_uninit(dev); 5885 5886 if (!dev->rtnl_link_ops || 5887 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 5888 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); 5889 5890 /* Notifier chain MUST detach us all upper devices. */ 5891 WARN_ON(netdev_has_any_upper_dev(dev)); 5892 5893 /* Remove entries from kobject tree */ 5894 netdev_unregister_kobject(dev); 5895 #ifdef CONFIG_XPS 5896 /* Remove XPS queueing entries */ 5897 netif_reset_xps_queues_gt(dev, 0); 5898 #endif 5899 } 5900 5901 synchronize_net(); 5902 5903 list_for_each_entry(dev, head, unreg_list) 5904 dev_put(dev); 5905 } 5906 5907 static void rollback_registered(struct net_device *dev) 5908 { 5909 LIST_HEAD(single); 5910 5911 list_add(&dev->unreg_list, &single); 5912 rollback_registered_many(&single); 5913 list_del(&single); 5914 } 5915 5916 static netdev_features_t netdev_fix_features(struct net_device *dev, 5917 netdev_features_t features) 5918 { 5919 /* Fix illegal checksum combinations */ 5920 if ((features & NETIF_F_HW_CSUM) && 5921 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 5922 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 5923 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 5924 } 5925 5926 /* TSO requires that SG is present as well. */ 5927 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 5928 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 5929 features &= ~NETIF_F_ALL_TSO; 5930 } 5931 5932 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 5933 !(features & NETIF_F_IP_CSUM)) { 5934 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 5935 features &= ~NETIF_F_TSO; 5936 features &= ~NETIF_F_TSO_ECN; 5937 } 5938 5939 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 5940 !(features & NETIF_F_IPV6_CSUM)) { 5941 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 5942 features &= ~NETIF_F_TSO6; 5943 } 5944 5945 /* TSO ECN requires that TSO is present as well. */ 5946 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 5947 features &= ~NETIF_F_TSO_ECN; 5948 5949 /* Software GSO depends on SG. */ 5950 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 5951 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 5952 features &= ~NETIF_F_GSO; 5953 } 5954 5955 /* UFO needs SG and checksumming */ 5956 if (features & NETIF_F_UFO) { 5957 /* maybe split UFO into V4 and V6? */ 5958 if (!((features & NETIF_F_GEN_CSUM) || 5959 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) 5960 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 5961 netdev_dbg(dev, 5962 "Dropping NETIF_F_UFO since no checksum offload features.\n"); 5963 features &= ~NETIF_F_UFO; 5964 } 5965 5966 if (!(features & NETIF_F_SG)) { 5967 netdev_dbg(dev, 5968 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); 5969 features &= ~NETIF_F_UFO; 5970 } 5971 } 5972 5973 #ifdef CONFIG_NET_RX_BUSY_POLL 5974 if (dev->netdev_ops->ndo_busy_poll) 5975 features |= NETIF_F_BUSY_POLL; 5976 else 5977 #endif 5978 features &= ~NETIF_F_BUSY_POLL; 5979 5980 return features; 5981 } 5982 5983 int __netdev_update_features(struct net_device *dev) 5984 { 5985 netdev_features_t features; 5986 int err = 0; 5987 5988 ASSERT_RTNL(); 5989 5990 features = netdev_get_wanted_features(dev); 5991 5992 if (dev->netdev_ops->ndo_fix_features) 5993 features = dev->netdev_ops->ndo_fix_features(dev, features); 5994 5995 /* driver might be less strict about feature dependencies */ 5996 features = netdev_fix_features(dev, features); 5997 5998 if (dev->features == features) 5999 return 0; 6000 6001 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 6002 &dev->features, &features); 6003 6004 if (dev->netdev_ops->ndo_set_features) 6005 err = dev->netdev_ops->ndo_set_features(dev, features); 6006 6007 if (unlikely(err < 0)) { 6008 netdev_err(dev, 6009 "set_features() failed (%d); wanted %pNF, left %pNF\n", 6010 err, &features, &dev->features); 6011 return -1; 6012 } 6013 6014 if (!err) 6015 dev->features = features; 6016 6017 return 1; 6018 } 6019 6020 /** 6021 * netdev_update_features - recalculate device features 6022 * @dev: the device to check 6023 * 6024 * Recalculate dev->features set and send notifications if it 6025 * has changed. Should be called after driver or hardware dependent 6026 * conditions might have changed that influence the features. 6027 */ 6028 void netdev_update_features(struct net_device *dev) 6029 { 6030 if (__netdev_update_features(dev)) 6031 netdev_features_change(dev); 6032 } 6033 EXPORT_SYMBOL(netdev_update_features); 6034 6035 /** 6036 * netdev_change_features - recalculate device features 6037 * @dev: the device to check 6038 * 6039 * Recalculate dev->features set and send notifications even 6040 * if they have not changed. Should be called instead of 6041 * netdev_update_features() if also dev->vlan_features might 6042 * have changed to allow the changes to be propagated to stacked 6043 * VLAN devices. 6044 */ 6045 void netdev_change_features(struct net_device *dev) 6046 { 6047 __netdev_update_features(dev); 6048 netdev_features_change(dev); 6049 } 6050 EXPORT_SYMBOL(netdev_change_features); 6051 6052 /** 6053 * netif_stacked_transfer_operstate - transfer operstate 6054 * @rootdev: the root or lower level device to transfer state from 6055 * @dev: the device to transfer operstate to 6056 * 6057 * Transfer operational state from root to device. This is normally 6058 * called when a stacking relationship exists between the root 6059 * device and the device(a leaf device). 6060 */ 6061 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 6062 struct net_device *dev) 6063 { 6064 if (rootdev->operstate == IF_OPER_DORMANT) 6065 netif_dormant_on(dev); 6066 else 6067 netif_dormant_off(dev); 6068 6069 if (netif_carrier_ok(rootdev)) { 6070 if (!netif_carrier_ok(dev)) 6071 netif_carrier_on(dev); 6072 } else { 6073 if (netif_carrier_ok(dev)) 6074 netif_carrier_off(dev); 6075 } 6076 } 6077 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 6078 6079 #ifdef CONFIG_SYSFS 6080 static int netif_alloc_rx_queues(struct net_device *dev) 6081 { 6082 unsigned int i, count = dev->num_rx_queues; 6083 struct netdev_rx_queue *rx; 6084 6085 BUG_ON(count < 1); 6086 6087 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL); 6088 if (!rx) 6089 return -ENOMEM; 6090 6091 dev->_rx = rx; 6092 6093 for (i = 0; i < count; i++) 6094 rx[i].dev = dev; 6095 return 0; 6096 } 6097 #endif 6098 6099 static void netdev_init_one_queue(struct net_device *dev, 6100 struct netdev_queue *queue, void *_unused) 6101 { 6102 /* Initialize queue lock */ 6103 spin_lock_init(&queue->_xmit_lock); 6104 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 6105 queue->xmit_lock_owner = -1; 6106 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 6107 queue->dev = dev; 6108 #ifdef CONFIG_BQL 6109 dql_init(&queue->dql, HZ); 6110 #endif 6111 } 6112 6113 static void netif_free_tx_queues(struct net_device *dev) 6114 { 6115 kvfree(dev->_tx); 6116 } 6117 6118 static int netif_alloc_netdev_queues(struct net_device *dev) 6119 { 6120 unsigned int count = dev->num_tx_queues; 6121 struct netdev_queue *tx; 6122 size_t sz = count * sizeof(*tx); 6123 6124 BUG_ON(count < 1 || count > 0xffff); 6125 6126 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 6127 if (!tx) { 6128 tx = vzalloc(sz); 6129 if (!tx) 6130 return -ENOMEM; 6131 } 6132 dev->_tx = tx; 6133 6134 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 6135 spin_lock_init(&dev->tx_global_lock); 6136 6137 return 0; 6138 } 6139 6140 /** 6141 * register_netdevice - register a network device 6142 * @dev: device to register 6143 * 6144 * Take a completed network device structure and add it to the kernel 6145 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 6146 * chain. 0 is returned on success. A negative errno code is returned 6147 * on a failure to set up the device, or if the name is a duplicate. 6148 * 6149 * Callers must hold the rtnl semaphore. You may want 6150 * register_netdev() instead of this. 6151 * 6152 * BUGS: 6153 * The locking appears insufficient to guarantee two parallel registers 6154 * will not get the same name. 6155 */ 6156 6157 int register_netdevice(struct net_device *dev) 6158 { 6159 int ret; 6160 struct net *net = dev_net(dev); 6161 6162 BUG_ON(dev_boot_phase); 6163 ASSERT_RTNL(); 6164 6165 might_sleep(); 6166 6167 /* When net_device's are persistent, this will be fatal. */ 6168 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 6169 BUG_ON(!net); 6170 6171 spin_lock_init(&dev->addr_list_lock); 6172 netdev_set_addr_lockdep_class(dev); 6173 6174 dev->iflink = -1; 6175 6176 ret = dev_get_valid_name(net, dev, dev->name); 6177 if (ret < 0) 6178 goto out; 6179 6180 /* Init, if this function is available */ 6181 if (dev->netdev_ops->ndo_init) { 6182 ret = dev->netdev_ops->ndo_init(dev); 6183 if (ret) { 6184 if (ret > 0) 6185 ret = -EIO; 6186 goto out; 6187 } 6188 } 6189 6190 if (((dev->hw_features | dev->features) & 6191 NETIF_F_HW_VLAN_CTAG_FILTER) && 6192 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 6193 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 6194 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 6195 ret = -EINVAL; 6196 goto err_uninit; 6197 } 6198 6199 ret = -EBUSY; 6200 if (!dev->ifindex) 6201 dev->ifindex = dev_new_index(net); 6202 else if (__dev_get_by_index(net, dev->ifindex)) 6203 goto err_uninit; 6204 6205 if (dev->iflink == -1) 6206 dev->iflink = dev->ifindex; 6207 6208 /* Transfer changeable features to wanted_features and enable 6209 * software offloads (GSO and GRO). 6210 */ 6211 dev->hw_features |= NETIF_F_SOFT_FEATURES; 6212 dev->features |= NETIF_F_SOFT_FEATURES; 6213 dev->wanted_features = dev->features & dev->hw_features; 6214 6215 if (!(dev->flags & IFF_LOOPBACK)) { 6216 dev->hw_features |= NETIF_F_NOCACHE_COPY; 6217 } 6218 6219 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 6220 */ 6221 dev->vlan_features |= NETIF_F_HIGHDMA; 6222 6223 /* Make NETIF_F_SG inheritable to tunnel devices. 6224 */ 6225 dev->hw_enc_features |= NETIF_F_SG; 6226 6227 /* Make NETIF_F_SG inheritable to MPLS. 6228 */ 6229 dev->mpls_features |= NETIF_F_SG; 6230 6231 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 6232 ret = notifier_to_errno(ret); 6233 if (ret) 6234 goto err_uninit; 6235 6236 ret = netdev_register_kobject(dev); 6237 if (ret) 6238 goto err_uninit; 6239 dev->reg_state = NETREG_REGISTERED; 6240 6241 __netdev_update_features(dev); 6242 6243 /* 6244 * Default initial state at registry is that the 6245 * device is present. 6246 */ 6247 6248 set_bit(__LINK_STATE_PRESENT, &dev->state); 6249 6250 linkwatch_init_dev(dev); 6251 6252 dev_init_scheduler(dev); 6253 dev_hold(dev); 6254 list_netdevice(dev); 6255 add_device_randomness(dev->dev_addr, dev->addr_len); 6256 6257 /* If the device has permanent device address, driver should 6258 * set dev_addr and also addr_assign_type should be set to 6259 * NET_ADDR_PERM (default value). 6260 */ 6261 if (dev->addr_assign_type == NET_ADDR_PERM) 6262 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 6263 6264 /* Notify protocols, that a new device appeared. */ 6265 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 6266 ret = notifier_to_errno(ret); 6267 if (ret) { 6268 rollback_registered(dev); 6269 dev->reg_state = NETREG_UNREGISTERED; 6270 } 6271 /* 6272 * Prevent userspace races by waiting until the network 6273 * device is fully setup before sending notifications. 6274 */ 6275 if (!dev->rtnl_link_ops || 6276 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 6277 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 6278 6279 out: 6280 return ret; 6281 6282 err_uninit: 6283 if (dev->netdev_ops->ndo_uninit) 6284 dev->netdev_ops->ndo_uninit(dev); 6285 goto out; 6286 } 6287 EXPORT_SYMBOL(register_netdevice); 6288 6289 /** 6290 * init_dummy_netdev - init a dummy network device for NAPI 6291 * @dev: device to init 6292 * 6293 * This takes a network device structure and initialize the minimum 6294 * amount of fields so it can be used to schedule NAPI polls without 6295 * registering a full blown interface. This is to be used by drivers 6296 * that need to tie several hardware interfaces to a single NAPI 6297 * poll scheduler due to HW limitations. 6298 */ 6299 int init_dummy_netdev(struct net_device *dev) 6300 { 6301 /* Clear everything. Note we don't initialize spinlocks 6302 * are they aren't supposed to be taken by any of the 6303 * NAPI code and this dummy netdev is supposed to be 6304 * only ever used for NAPI polls 6305 */ 6306 memset(dev, 0, sizeof(struct net_device)); 6307 6308 /* make sure we BUG if trying to hit standard 6309 * register/unregister code path 6310 */ 6311 dev->reg_state = NETREG_DUMMY; 6312 6313 /* NAPI wants this */ 6314 INIT_LIST_HEAD(&dev->napi_list); 6315 6316 /* a dummy interface is started by default */ 6317 set_bit(__LINK_STATE_PRESENT, &dev->state); 6318 set_bit(__LINK_STATE_START, &dev->state); 6319 6320 /* Note : We dont allocate pcpu_refcnt for dummy devices, 6321 * because users of this 'device' dont need to change 6322 * its refcount. 6323 */ 6324 6325 return 0; 6326 } 6327 EXPORT_SYMBOL_GPL(init_dummy_netdev); 6328 6329 6330 /** 6331 * register_netdev - register a network device 6332 * @dev: device to register 6333 * 6334 * Take a completed network device structure and add it to the kernel 6335 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 6336 * chain. 0 is returned on success. A negative errno code is returned 6337 * on a failure to set up the device, or if the name is a duplicate. 6338 * 6339 * This is a wrapper around register_netdevice that takes the rtnl semaphore 6340 * and expands the device name if you passed a format string to 6341 * alloc_netdev. 6342 */ 6343 int register_netdev(struct net_device *dev) 6344 { 6345 int err; 6346 6347 rtnl_lock(); 6348 err = register_netdevice(dev); 6349 rtnl_unlock(); 6350 return err; 6351 } 6352 EXPORT_SYMBOL(register_netdev); 6353 6354 int netdev_refcnt_read(const struct net_device *dev) 6355 { 6356 int i, refcnt = 0; 6357 6358 for_each_possible_cpu(i) 6359 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 6360 return refcnt; 6361 } 6362 EXPORT_SYMBOL(netdev_refcnt_read); 6363 6364 /** 6365 * netdev_wait_allrefs - wait until all references are gone. 6366 * @dev: target net_device 6367 * 6368 * This is called when unregistering network devices. 6369 * 6370 * Any protocol or device that holds a reference should register 6371 * for netdevice notification, and cleanup and put back the 6372 * reference if they receive an UNREGISTER event. 6373 * We can get stuck here if buggy protocols don't correctly 6374 * call dev_put. 6375 */ 6376 static void netdev_wait_allrefs(struct net_device *dev) 6377 { 6378 unsigned long rebroadcast_time, warning_time; 6379 int refcnt; 6380 6381 linkwatch_forget_dev(dev); 6382 6383 rebroadcast_time = warning_time = jiffies; 6384 refcnt = netdev_refcnt_read(dev); 6385 6386 while (refcnt != 0) { 6387 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 6388 rtnl_lock(); 6389 6390 /* Rebroadcast unregister notification */ 6391 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 6392 6393 __rtnl_unlock(); 6394 rcu_barrier(); 6395 rtnl_lock(); 6396 6397 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 6398 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 6399 &dev->state)) { 6400 /* We must not have linkwatch events 6401 * pending on unregister. If this 6402 * happens, we simply run the queue 6403 * unscheduled, resulting in a noop 6404 * for this device. 6405 */ 6406 linkwatch_run_queue(); 6407 } 6408 6409 __rtnl_unlock(); 6410 6411 rebroadcast_time = jiffies; 6412 } 6413 6414 msleep(250); 6415 6416 refcnt = netdev_refcnt_read(dev); 6417 6418 if (time_after(jiffies, warning_time + 10 * HZ)) { 6419 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 6420 dev->name, refcnt); 6421 warning_time = jiffies; 6422 } 6423 } 6424 } 6425 6426 /* The sequence is: 6427 * 6428 * rtnl_lock(); 6429 * ... 6430 * register_netdevice(x1); 6431 * register_netdevice(x2); 6432 * ... 6433 * unregister_netdevice(y1); 6434 * unregister_netdevice(y2); 6435 * ... 6436 * rtnl_unlock(); 6437 * free_netdev(y1); 6438 * free_netdev(y2); 6439 * 6440 * We are invoked by rtnl_unlock(). 6441 * This allows us to deal with problems: 6442 * 1) We can delete sysfs objects which invoke hotplug 6443 * without deadlocking with linkwatch via keventd. 6444 * 2) Since we run with the RTNL semaphore not held, we can sleep 6445 * safely in order to wait for the netdev refcnt to drop to zero. 6446 * 6447 * We must not return until all unregister events added during 6448 * the interval the lock was held have been completed. 6449 */ 6450 void netdev_run_todo(void) 6451 { 6452 struct list_head list; 6453 6454 /* Snapshot list, allow later requests */ 6455 list_replace_init(&net_todo_list, &list); 6456 6457 __rtnl_unlock(); 6458 6459 6460 /* Wait for rcu callbacks to finish before next phase */ 6461 if (!list_empty(&list)) 6462 rcu_barrier(); 6463 6464 while (!list_empty(&list)) { 6465 struct net_device *dev 6466 = list_first_entry(&list, struct net_device, todo_list); 6467 list_del(&dev->todo_list); 6468 6469 rtnl_lock(); 6470 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 6471 __rtnl_unlock(); 6472 6473 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 6474 pr_err("network todo '%s' but state %d\n", 6475 dev->name, dev->reg_state); 6476 dump_stack(); 6477 continue; 6478 } 6479 6480 dev->reg_state = NETREG_UNREGISTERED; 6481 6482 on_each_cpu(flush_backlog, dev, 1); 6483 6484 netdev_wait_allrefs(dev); 6485 6486 /* paranoia */ 6487 BUG_ON(netdev_refcnt_read(dev)); 6488 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 6489 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 6490 WARN_ON(dev->dn_ptr); 6491 6492 if (dev->destructor) 6493 dev->destructor(dev); 6494 6495 /* Report a network device has been unregistered */ 6496 rtnl_lock(); 6497 dev_net(dev)->dev_unreg_count--; 6498 __rtnl_unlock(); 6499 wake_up(&netdev_unregistering_wq); 6500 6501 /* Free network device */ 6502 kobject_put(&dev->dev.kobj); 6503 } 6504 } 6505 6506 /* Convert net_device_stats to rtnl_link_stats64. They have the same 6507 * fields in the same order, with only the type differing. 6508 */ 6509 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 6510 const struct net_device_stats *netdev_stats) 6511 { 6512 #if BITS_PER_LONG == 64 6513 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); 6514 memcpy(stats64, netdev_stats, sizeof(*stats64)); 6515 #else 6516 size_t i, n = sizeof(*stats64) / sizeof(u64); 6517 const unsigned long *src = (const unsigned long *)netdev_stats; 6518 u64 *dst = (u64 *)stats64; 6519 6520 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != 6521 sizeof(*stats64) / sizeof(u64)); 6522 for (i = 0; i < n; i++) 6523 dst[i] = src[i]; 6524 #endif 6525 } 6526 EXPORT_SYMBOL(netdev_stats_to_stats64); 6527 6528 /** 6529 * dev_get_stats - get network device statistics 6530 * @dev: device to get statistics from 6531 * @storage: place to store stats 6532 * 6533 * Get network statistics from device. Return @storage. 6534 * The device driver may provide its own method by setting 6535 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 6536 * otherwise the internal statistics structure is used. 6537 */ 6538 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 6539 struct rtnl_link_stats64 *storage) 6540 { 6541 const struct net_device_ops *ops = dev->netdev_ops; 6542 6543 if (ops->ndo_get_stats64) { 6544 memset(storage, 0, sizeof(*storage)); 6545 ops->ndo_get_stats64(dev, storage); 6546 } else if (ops->ndo_get_stats) { 6547 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 6548 } else { 6549 netdev_stats_to_stats64(storage, &dev->stats); 6550 } 6551 storage->rx_dropped += atomic_long_read(&dev->rx_dropped); 6552 storage->tx_dropped += atomic_long_read(&dev->tx_dropped); 6553 return storage; 6554 } 6555 EXPORT_SYMBOL(dev_get_stats); 6556 6557 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 6558 { 6559 struct netdev_queue *queue = dev_ingress_queue(dev); 6560 6561 #ifdef CONFIG_NET_CLS_ACT 6562 if (queue) 6563 return queue; 6564 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 6565 if (!queue) 6566 return NULL; 6567 netdev_init_one_queue(dev, queue, NULL); 6568 queue->qdisc = &noop_qdisc; 6569 queue->qdisc_sleeping = &noop_qdisc; 6570 rcu_assign_pointer(dev->ingress_queue, queue); 6571 #endif 6572 return queue; 6573 } 6574 6575 static const struct ethtool_ops default_ethtool_ops; 6576 6577 void netdev_set_default_ethtool_ops(struct net_device *dev, 6578 const struct ethtool_ops *ops) 6579 { 6580 if (dev->ethtool_ops == &default_ethtool_ops) 6581 dev->ethtool_ops = ops; 6582 } 6583 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 6584 6585 void netdev_freemem(struct net_device *dev) 6586 { 6587 char *addr = (char *)dev - dev->padded; 6588 6589 kvfree(addr); 6590 } 6591 6592 /** 6593 * alloc_netdev_mqs - allocate network device 6594 * @sizeof_priv: size of private data to allocate space for 6595 * @name: device name format string 6596 * @name_assign_type: origin of device name 6597 * @setup: callback to initialize device 6598 * @txqs: the number of TX subqueues to allocate 6599 * @rxqs: the number of RX subqueues to allocate 6600 * 6601 * Allocates a struct net_device with private data area for driver use 6602 * and performs basic initialization. Also allocates subqueue structs 6603 * for each queue on the device. 6604 */ 6605 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 6606 unsigned char name_assign_type, 6607 void (*setup)(struct net_device *), 6608 unsigned int txqs, unsigned int rxqs) 6609 { 6610 struct net_device *dev; 6611 size_t alloc_size; 6612 struct net_device *p; 6613 6614 BUG_ON(strlen(name) >= sizeof(dev->name)); 6615 6616 if (txqs < 1) { 6617 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 6618 return NULL; 6619 } 6620 6621 #ifdef CONFIG_SYSFS 6622 if (rxqs < 1) { 6623 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 6624 return NULL; 6625 } 6626 #endif 6627 6628 alloc_size = sizeof(struct net_device); 6629 if (sizeof_priv) { 6630 /* ensure 32-byte alignment of private area */ 6631 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 6632 alloc_size += sizeof_priv; 6633 } 6634 /* ensure 32-byte alignment of whole construct */ 6635 alloc_size += NETDEV_ALIGN - 1; 6636 6637 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 6638 if (!p) 6639 p = vzalloc(alloc_size); 6640 if (!p) 6641 return NULL; 6642 6643 dev = PTR_ALIGN(p, NETDEV_ALIGN); 6644 dev->padded = (char *)dev - (char *)p; 6645 6646 dev->pcpu_refcnt = alloc_percpu(int); 6647 if (!dev->pcpu_refcnt) 6648 goto free_dev; 6649 6650 if (dev_addr_init(dev)) 6651 goto free_pcpu; 6652 6653 dev_mc_init(dev); 6654 dev_uc_init(dev); 6655 6656 dev_net_set(dev, &init_net); 6657 6658 dev->gso_max_size = GSO_MAX_SIZE; 6659 dev->gso_max_segs = GSO_MAX_SEGS; 6660 dev->gso_min_segs = 0; 6661 6662 INIT_LIST_HEAD(&dev->napi_list); 6663 INIT_LIST_HEAD(&dev->unreg_list); 6664 INIT_LIST_HEAD(&dev->close_list); 6665 INIT_LIST_HEAD(&dev->link_watch_list); 6666 INIT_LIST_HEAD(&dev->adj_list.upper); 6667 INIT_LIST_HEAD(&dev->adj_list.lower); 6668 INIT_LIST_HEAD(&dev->all_adj_list.upper); 6669 INIT_LIST_HEAD(&dev->all_adj_list.lower); 6670 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 6671 setup(dev); 6672 6673 dev->num_tx_queues = txqs; 6674 dev->real_num_tx_queues = txqs; 6675 if (netif_alloc_netdev_queues(dev)) 6676 goto free_all; 6677 6678 #ifdef CONFIG_SYSFS 6679 dev->num_rx_queues = rxqs; 6680 dev->real_num_rx_queues = rxqs; 6681 if (netif_alloc_rx_queues(dev)) 6682 goto free_all; 6683 #endif 6684 6685 strcpy(dev->name, name); 6686 dev->name_assign_type = name_assign_type; 6687 dev->group = INIT_NETDEV_GROUP; 6688 if (!dev->ethtool_ops) 6689 dev->ethtool_ops = &default_ethtool_ops; 6690 return dev; 6691 6692 free_all: 6693 free_netdev(dev); 6694 return NULL; 6695 6696 free_pcpu: 6697 free_percpu(dev->pcpu_refcnt); 6698 free_dev: 6699 netdev_freemem(dev); 6700 return NULL; 6701 } 6702 EXPORT_SYMBOL(alloc_netdev_mqs); 6703 6704 /** 6705 * free_netdev - free network device 6706 * @dev: device 6707 * 6708 * This function does the last stage of destroying an allocated device 6709 * interface. The reference to the device object is released. 6710 * If this is the last reference then it will be freed. 6711 */ 6712 void free_netdev(struct net_device *dev) 6713 { 6714 struct napi_struct *p, *n; 6715 6716 release_net(dev_net(dev)); 6717 6718 netif_free_tx_queues(dev); 6719 #ifdef CONFIG_SYSFS 6720 kfree(dev->_rx); 6721 #endif 6722 6723 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 6724 6725 /* Flush device addresses */ 6726 dev_addr_flush(dev); 6727 6728 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 6729 netif_napi_del(p); 6730 6731 free_percpu(dev->pcpu_refcnt); 6732 dev->pcpu_refcnt = NULL; 6733 6734 /* Compatibility with error handling in drivers */ 6735 if (dev->reg_state == NETREG_UNINITIALIZED) { 6736 netdev_freemem(dev); 6737 return; 6738 } 6739 6740 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 6741 dev->reg_state = NETREG_RELEASED; 6742 6743 /* will free via device release */ 6744 put_device(&dev->dev); 6745 } 6746 EXPORT_SYMBOL(free_netdev); 6747 6748 /** 6749 * synchronize_net - Synchronize with packet receive processing 6750 * 6751 * Wait for packets currently being received to be done. 6752 * Does not block later packets from starting. 6753 */ 6754 void synchronize_net(void) 6755 { 6756 might_sleep(); 6757 if (rtnl_is_locked()) 6758 synchronize_rcu_expedited(); 6759 else 6760 synchronize_rcu(); 6761 } 6762 EXPORT_SYMBOL(synchronize_net); 6763 6764 /** 6765 * unregister_netdevice_queue - remove device from the kernel 6766 * @dev: device 6767 * @head: list 6768 * 6769 * This function shuts down a device interface and removes it 6770 * from the kernel tables. 6771 * If head not NULL, device is queued to be unregistered later. 6772 * 6773 * Callers must hold the rtnl semaphore. You may want 6774 * unregister_netdev() instead of this. 6775 */ 6776 6777 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 6778 { 6779 ASSERT_RTNL(); 6780 6781 if (head) { 6782 list_move_tail(&dev->unreg_list, head); 6783 } else { 6784 rollback_registered(dev); 6785 /* Finish processing unregister after unlock */ 6786 net_set_todo(dev); 6787 } 6788 } 6789 EXPORT_SYMBOL(unregister_netdevice_queue); 6790 6791 /** 6792 * unregister_netdevice_many - unregister many devices 6793 * @head: list of devices 6794 * 6795 * Note: As most callers use a stack allocated list_head, 6796 * we force a list_del() to make sure stack wont be corrupted later. 6797 */ 6798 void unregister_netdevice_many(struct list_head *head) 6799 { 6800 struct net_device *dev; 6801 6802 if (!list_empty(head)) { 6803 rollback_registered_many(head); 6804 list_for_each_entry(dev, head, unreg_list) 6805 net_set_todo(dev); 6806 list_del(head); 6807 } 6808 } 6809 EXPORT_SYMBOL(unregister_netdevice_many); 6810 6811 /** 6812 * unregister_netdev - remove device from the kernel 6813 * @dev: device 6814 * 6815 * This function shuts down a device interface and removes it 6816 * from the kernel tables. 6817 * 6818 * This is just a wrapper for unregister_netdevice that takes 6819 * the rtnl semaphore. In general you want to use this and not 6820 * unregister_netdevice. 6821 */ 6822 void unregister_netdev(struct net_device *dev) 6823 { 6824 rtnl_lock(); 6825 unregister_netdevice(dev); 6826 rtnl_unlock(); 6827 } 6828 EXPORT_SYMBOL(unregister_netdev); 6829 6830 /** 6831 * dev_change_net_namespace - move device to different nethost namespace 6832 * @dev: device 6833 * @net: network namespace 6834 * @pat: If not NULL name pattern to try if the current device name 6835 * is already taken in the destination network namespace. 6836 * 6837 * This function shuts down a device interface and moves it 6838 * to a new network namespace. On success 0 is returned, on 6839 * a failure a netagive errno code is returned. 6840 * 6841 * Callers must hold the rtnl semaphore. 6842 */ 6843 6844 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 6845 { 6846 int err; 6847 6848 ASSERT_RTNL(); 6849 6850 /* Don't allow namespace local devices to be moved. */ 6851 err = -EINVAL; 6852 if (dev->features & NETIF_F_NETNS_LOCAL) 6853 goto out; 6854 6855 /* Ensure the device has been registrered */ 6856 if (dev->reg_state != NETREG_REGISTERED) 6857 goto out; 6858 6859 /* Get out if there is nothing todo */ 6860 err = 0; 6861 if (net_eq(dev_net(dev), net)) 6862 goto out; 6863 6864 /* Pick the destination device name, and ensure 6865 * we can use it in the destination network namespace. 6866 */ 6867 err = -EEXIST; 6868 if (__dev_get_by_name(net, dev->name)) { 6869 /* We get here if we can't use the current device name */ 6870 if (!pat) 6871 goto out; 6872 if (dev_get_valid_name(net, dev, pat) < 0) 6873 goto out; 6874 } 6875 6876 /* 6877 * And now a mini version of register_netdevice unregister_netdevice. 6878 */ 6879 6880 /* If device is running close it first. */ 6881 dev_close(dev); 6882 6883 /* And unlink it from device chain */ 6884 err = -ENODEV; 6885 unlist_netdevice(dev); 6886 6887 synchronize_net(); 6888 6889 /* Shutdown queueing discipline. */ 6890 dev_shutdown(dev); 6891 6892 /* Notify protocols, that we are about to destroy 6893 this device. They should clean all the things. 6894 6895 Note that dev->reg_state stays at NETREG_REGISTERED. 6896 This is wanted because this way 8021q and macvlan know 6897 the device is just moving and can keep their slaves up. 6898 */ 6899 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 6900 rcu_barrier(); 6901 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 6902 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); 6903 6904 /* 6905 * Flush the unicast and multicast chains 6906 */ 6907 dev_uc_flush(dev); 6908 dev_mc_flush(dev); 6909 6910 /* Send a netdev-removed uevent to the old namespace */ 6911 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 6912 netdev_adjacent_del_links(dev); 6913 6914 /* Actually switch the network namespace */ 6915 dev_net_set(dev, net); 6916 6917 /* If there is an ifindex conflict assign a new one */ 6918 if (__dev_get_by_index(net, dev->ifindex)) { 6919 int iflink = (dev->iflink == dev->ifindex); 6920 dev->ifindex = dev_new_index(net); 6921 if (iflink) 6922 dev->iflink = dev->ifindex; 6923 } 6924 6925 /* Send a netdev-add uevent to the new namespace */ 6926 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 6927 netdev_adjacent_add_links(dev); 6928 6929 /* Fixup kobjects */ 6930 err = device_rename(&dev->dev, dev->name); 6931 WARN_ON(err); 6932 6933 /* Add the device back in the hashes */ 6934 list_netdevice(dev); 6935 6936 /* Notify protocols, that a new device appeared. */ 6937 call_netdevice_notifiers(NETDEV_REGISTER, dev); 6938 6939 /* 6940 * Prevent userspace races by waiting until the network 6941 * device is fully setup before sending notifications. 6942 */ 6943 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 6944 6945 synchronize_net(); 6946 err = 0; 6947 out: 6948 return err; 6949 } 6950 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 6951 6952 static int dev_cpu_callback(struct notifier_block *nfb, 6953 unsigned long action, 6954 void *ocpu) 6955 { 6956 struct sk_buff **list_skb; 6957 struct sk_buff *skb; 6958 unsigned int cpu, oldcpu = (unsigned long)ocpu; 6959 struct softnet_data *sd, *oldsd; 6960 6961 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) 6962 return NOTIFY_OK; 6963 6964 local_irq_disable(); 6965 cpu = smp_processor_id(); 6966 sd = &per_cpu(softnet_data, cpu); 6967 oldsd = &per_cpu(softnet_data, oldcpu); 6968 6969 /* Find end of our completion_queue. */ 6970 list_skb = &sd->completion_queue; 6971 while (*list_skb) 6972 list_skb = &(*list_skb)->next; 6973 /* Append completion queue from offline CPU. */ 6974 *list_skb = oldsd->completion_queue; 6975 oldsd->completion_queue = NULL; 6976 6977 /* Append output queue from offline CPU. */ 6978 if (oldsd->output_queue) { 6979 *sd->output_queue_tailp = oldsd->output_queue; 6980 sd->output_queue_tailp = oldsd->output_queue_tailp; 6981 oldsd->output_queue = NULL; 6982 oldsd->output_queue_tailp = &oldsd->output_queue; 6983 } 6984 /* Append NAPI poll list from offline CPU. */ 6985 if (!list_empty(&oldsd->poll_list)) { 6986 list_splice_init(&oldsd->poll_list, &sd->poll_list); 6987 raise_softirq_irqoff(NET_RX_SOFTIRQ); 6988 } 6989 6990 raise_softirq_irqoff(NET_TX_SOFTIRQ); 6991 local_irq_enable(); 6992 6993 /* Process offline CPU's input_pkt_queue */ 6994 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 6995 netif_rx_internal(skb); 6996 input_queue_head_incr(oldsd); 6997 } 6998 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { 6999 netif_rx_internal(skb); 7000 input_queue_head_incr(oldsd); 7001 } 7002 7003 return NOTIFY_OK; 7004 } 7005 7006 7007 /** 7008 * netdev_increment_features - increment feature set by one 7009 * @all: current feature set 7010 * @one: new feature set 7011 * @mask: mask feature set 7012 * 7013 * Computes a new feature set after adding a device with feature set 7014 * @one to the master device with current feature set @all. Will not 7015 * enable anything that is off in @mask. Returns the new feature set. 7016 */ 7017 netdev_features_t netdev_increment_features(netdev_features_t all, 7018 netdev_features_t one, netdev_features_t mask) 7019 { 7020 if (mask & NETIF_F_GEN_CSUM) 7021 mask |= NETIF_F_ALL_CSUM; 7022 mask |= NETIF_F_VLAN_CHALLENGED; 7023 7024 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; 7025 all &= one | ~NETIF_F_ALL_FOR_ALL; 7026 7027 /* If one device supports hw checksumming, set for all. */ 7028 if (all & NETIF_F_GEN_CSUM) 7029 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); 7030 7031 return all; 7032 } 7033 EXPORT_SYMBOL(netdev_increment_features); 7034 7035 static struct hlist_head * __net_init netdev_create_hash(void) 7036 { 7037 int i; 7038 struct hlist_head *hash; 7039 7040 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); 7041 if (hash != NULL) 7042 for (i = 0; i < NETDEV_HASHENTRIES; i++) 7043 INIT_HLIST_HEAD(&hash[i]); 7044 7045 return hash; 7046 } 7047 7048 /* Initialize per network namespace state */ 7049 static int __net_init netdev_init(struct net *net) 7050 { 7051 if (net != &init_net) 7052 INIT_LIST_HEAD(&net->dev_base_head); 7053 7054 net->dev_name_head = netdev_create_hash(); 7055 if (net->dev_name_head == NULL) 7056 goto err_name; 7057 7058 net->dev_index_head = netdev_create_hash(); 7059 if (net->dev_index_head == NULL) 7060 goto err_idx; 7061 7062 return 0; 7063 7064 err_idx: 7065 kfree(net->dev_name_head); 7066 err_name: 7067 return -ENOMEM; 7068 } 7069 7070 /** 7071 * netdev_drivername - network driver for the device 7072 * @dev: network device 7073 * 7074 * Determine network driver for device. 7075 */ 7076 const char *netdev_drivername(const struct net_device *dev) 7077 { 7078 const struct device_driver *driver; 7079 const struct device *parent; 7080 const char *empty = ""; 7081 7082 parent = dev->dev.parent; 7083 if (!parent) 7084 return empty; 7085 7086 driver = parent->driver; 7087 if (driver && driver->name) 7088 return driver->name; 7089 return empty; 7090 } 7091 7092 static void __netdev_printk(const char *level, const struct net_device *dev, 7093 struct va_format *vaf) 7094 { 7095 if (dev && dev->dev.parent) { 7096 dev_printk_emit(level[1] - '0', 7097 dev->dev.parent, 7098 "%s %s %s%s: %pV", 7099 dev_driver_string(dev->dev.parent), 7100 dev_name(dev->dev.parent), 7101 netdev_name(dev), netdev_reg_state(dev), 7102 vaf); 7103 } else if (dev) { 7104 printk("%s%s%s: %pV", 7105 level, netdev_name(dev), netdev_reg_state(dev), vaf); 7106 } else { 7107 printk("%s(NULL net_device): %pV", level, vaf); 7108 } 7109 } 7110 7111 void netdev_printk(const char *level, const struct net_device *dev, 7112 const char *format, ...) 7113 { 7114 struct va_format vaf; 7115 va_list args; 7116 7117 va_start(args, format); 7118 7119 vaf.fmt = format; 7120 vaf.va = &args; 7121 7122 __netdev_printk(level, dev, &vaf); 7123 7124 va_end(args); 7125 } 7126 EXPORT_SYMBOL(netdev_printk); 7127 7128 #define define_netdev_printk_level(func, level) \ 7129 void func(const struct net_device *dev, const char *fmt, ...) \ 7130 { \ 7131 struct va_format vaf; \ 7132 va_list args; \ 7133 \ 7134 va_start(args, fmt); \ 7135 \ 7136 vaf.fmt = fmt; \ 7137 vaf.va = &args; \ 7138 \ 7139 __netdev_printk(level, dev, &vaf); \ 7140 \ 7141 va_end(args); \ 7142 } \ 7143 EXPORT_SYMBOL(func); 7144 7145 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 7146 define_netdev_printk_level(netdev_alert, KERN_ALERT); 7147 define_netdev_printk_level(netdev_crit, KERN_CRIT); 7148 define_netdev_printk_level(netdev_err, KERN_ERR); 7149 define_netdev_printk_level(netdev_warn, KERN_WARNING); 7150 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 7151 define_netdev_printk_level(netdev_info, KERN_INFO); 7152 7153 static void __net_exit netdev_exit(struct net *net) 7154 { 7155 kfree(net->dev_name_head); 7156 kfree(net->dev_index_head); 7157 } 7158 7159 static struct pernet_operations __net_initdata netdev_net_ops = { 7160 .init = netdev_init, 7161 .exit = netdev_exit, 7162 }; 7163 7164 static void __net_exit default_device_exit(struct net *net) 7165 { 7166 struct net_device *dev, *aux; 7167 /* 7168 * Push all migratable network devices back to the 7169 * initial network namespace 7170 */ 7171 rtnl_lock(); 7172 for_each_netdev_safe(net, dev, aux) { 7173 int err; 7174 char fb_name[IFNAMSIZ]; 7175 7176 /* Ignore unmoveable devices (i.e. loopback) */ 7177 if (dev->features & NETIF_F_NETNS_LOCAL) 7178 continue; 7179 7180 /* Leave virtual devices for the generic cleanup */ 7181 if (dev->rtnl_link_ops) 7182 continue; 7183 7184 /* Push remaining network devices to init_net */ 7185 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 7186 err = dev_change_net_namespace(dev, &init_net, fb_name); 7187 if (err) { 7188 pr_emerg("%s: failed to move %s to init_net: %d\n", 7189 __func__, dev->name, err); 7190 BUG(); 7191 } 7192 } 7193 rtnl_unlock(); 7194 } 7195 7196 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 7197 { 7198 /* Return with the rtnl_lock held when there are no network 7199 * devices unregistering in any network namespace in net_list. 7200 */ 7201 struct net *net; 7202 bool unregistering; 7203 DEFINE_WAIT(wait); 7204 7205 for (;;) { 7206 prepare_to_wait(&netdev_unregistering_wq, &wait, 7207 TASK_UNINTERRUPTIBLE); 7208 unregistering = false; 7209 rtnl_lock(); 7210 list_for_each_entry(net, net_list, exit_list) { 7211 if (net->dev_unreg_count > 0) { 7212 unregistering = true; 7213 break; 7214 } 7215 } 7216 if (!unregistering) 7217 break; 7218 __rtnl_unlock(); 7219 schedule(); 7220 } 7221 finish_wait(&netdev_unregistering_wq, &wait); 7222 } 7223 7224 static void __net_exit default_device_exit_batch(struct list_head *net_list) 7225 { 7226 /* At exit all network devices most be removed from a network 7227 * namespace. Do this in the reverse order of registration. 7228 * Do this across as many network namespaces as possible to 7229 * improve batching efficiency. 7230 */ 7231 struct net_device *dev; 7232 struct net *net; 7233 LIST_HEAD(dev_kill_list); 7234 7235 /* To prevent network device cleanup code from dereferencing 7236 * loopback devices or network devices that have been freed 7237 * wait here for all pending unregistrations to complete, 7238 * before unregistring the loopback device and allowing the 7239 * network namespace be freed. 7240 * 7241 * The netdev todo list containing all network devices 7242 * unregistrations that happen in default_device_exit_batch 7243 * will run in the rtnl_unlock() at the end of 7244 * default_device_exit_batch. 7245 */ 7246 rtnl_lock_unregistering(net_list); 7247 list_for_each_entry(net, net_list, exit_list) { 7248 for_each_netdev_reverse(net, dev) { 7249 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 7250 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 7251 else 7252 unregister_netdevice_queue(dev, &dev_kill_list); 7253 } 7254 } 7255 unregister_netdevice_many(&dev_kill_list); 7256 rtnl_unlock(); 7257 } 7258 7259 static struct pernet_operations __net_initdata default_device_ops = { 7260 .exit = default_device_exit, 7261 .exit_batch = default_device_exit_batch, 7262 }; 7263 7264 /* 7265 * Initialize the DEV module. At boot time this walks the device list and 7266 * unhooks any devices that fail to initialise (normally hardware not 7267 * present) and leaves us with a valid list of present and active devices. 7268 * 7269 */ 7270 7271 /* 7272 * This is called single threaded during boot, so no need 7273 * to take the rtnl semaphore. 7274 */ 7275 static int __init net_dev_init(void) 7276 { 7277 int i, rc = -ENOMEM; 7278 7279 BUG_ON(!dev_boot_phase); 7280 7281 if (dev_proc_init()) 7282 goto out; 7283 7284 if (netdev_kobject_init()) 7285 goto out; 7286 7287 INIT_LIST_HEAD(&ptype_all); 7288 for (i = 0; i < PTYPE_HASH_SIZE; i++) 7289 INIT_LIST_HEAD(&ptype_base[i]); 7290 7291 INIT_LIST_HEAD(&offload_base); 7292 7293 if (register_pernet_subsys(&netdev_net_ops)) 7294 goto out; 7295 7296 /* 7297 * Initialise the packet receive queues. 7298 */ 7299 7300 for_each_possible_cpu(i) { 7301 struct softnet_data *sd = &per_cpu(softnet_data, i); 7302 7303 skb_queue_head_init(&sd->input_pkt_queue); 7304 skb_queue_head_init(&sd->process_queue); 7305 INIT_LIST_HEAD(&sd->poll_list); 7306 sd->output_queue_tailp = &sd->output_queue; 7307 #ifdef CONFIG_RPS 7308 sd->csd.func = rps_trigger_softirq; 7309 sd->csd.info = sd; 7310 sd->cpu = i; 7311 #endif 7312 7313 sd->backlog.poll = process_backlog; 7314 sd->backlog.weight = weight_p; 7315 } 7316 7317 dev_boot_phase = 0; 7318 7319 /* The loopback device is special if any other network devices 7320 * is present in a network namespace the loopback device must 7321 * be present. Since we now dynamically allocate and free the 7322 * loopback device ensure this invariant is maintained by 7323 * keeping the loopback device as the first device on the 7324 * list of network devices. Ensuring the loopback devices 7325 * is the first device that appears and the last network device 7326 * that disappears. 7327 */ 7328 if (register_pernet_device(&loopback_net_ops)) 7329 goto out; 7330 7331 if (register_pernet_device(&default_device_ops)) 7332 goto out; 7333 7334 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 7335 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 7336 7337 hotcpu_notifier(dev_cpu_callback, 0); 7338 dst_init(); 7339 rc = 0; 7340 out: 7341 return rc; 7342 } 7343 7344 subsys_initcall(net_dev_init); 7345