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 <net/mpls.h> 122 #include <linux/ipv6.h> 123 #include <linux/in.h> 124 #include <linux/jhash.h> 125 #include <linux/random.h> 126 #include <trace/events/napi.h> 127 #include <trace/events/net.h> 128 #include <trace/events/skb.h> 129 #include <linux/pci.h> 130 #include <linux/inetdevice.h> 131 #include <linux/cpu_rmap.h> 132 #include <linux/static_key.h> 133 #include <linux/hashtable.h> 134 #include <linux/vmalloc.h> 135 #include <linux/if_macvlan.h> 136 #include <linux/errqueue.h> 137 #include <linux/hrtimer.h> 138 #include <linux/netfilter_ingress.h> 139 140 #include "net-sysfs.h" 141 142 /* Instead of increasing this, you should create a hash table. */ 143 #define MAX_GRO_SKBS 8 144 145 /* This should be increased if a protocol with a bigger head is added. */ 146 #define GRO_MAX_HEAD (MAX_HEADER + 128) 147 148 static DEFINE_SPINLOCK(ptype_lock); 149 static DEFINE_SPINLOCK(offload_lock); 150 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 151 struct list_head ptype_all __read_mostly; /* Taps */ 152 static struct list_head offload_base __read_mostly; 153 154 static int netif_rx_internal(struct sk_buff *skb); 155 static int call_netdevice_notifiers_info(unsigned long val, 156 struct net_device *dev, 157 struct netdev_notifier_info *info); 158 159 /* 160 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 161 * semaphore. 162 * 163 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 164 * 165 * Writers must hold the rtnl semaphore while they loop through the 166 * dev_base_head list, and hold dev_base_lock for writing when they do the 167 * actual updates. This allows pure readers to access the list even 168 * while a writer is preparing to update it. 169 * 170 * To put it another way, dev_base_lock is held for writing only to 171 * protect against pure readers; the rtnl semaphore provides the 172 * protection against other writers. 173 * 174 * See, for example usages, register_netdevice() and 175 * unregister_netdevice(), which must be called with the rtnl 176 * semaphore held. 177 */ 178 DEFINE_RWLOCK(dev_base_lock); 179 EXPORT_SYMBOL(dev_base_lock); 180 181 /* protects napi_hash addition/deletion and napi_gen_id */ 182 static DEFINE_SPINLOCK(napi_hash_lock); 183 184 static unsigned int napi_gen_id; 185 static DEFINE_HASHTABLE(napi_hash, 8); 186 187 static seqcount_t devnet_rename_seq; 188 189 static inline void dev_base_seq_inc(struct net *net) 190 { 191 while (++net->dev_base_seq == 0); 192 } 193 194 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 195 { 196 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); 197 198 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 199 } 200 201 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 202 { 203 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 204 } 205 206 static inline void rps_lock(struct softnet_data *sd) 207 { 208 #ifdef CONFIG_RPS 209 spin_lock(&sd->input_pkt_queue.lock); 210 #endif 211 } 212 213 static inline void rps_unlock(struct softnet_data *sd) 214 { 215 #ifdef CONFIG_RPS 216 spin_unlock(&sd->input_pkt_queue.lock); 217 #endif 218 } 219 220 /* Device list insertion */ 221 static void list_netdevice(struct net_device *dev) 222 { 223 struct net *net = dev_net(dev); 224 225 ASSERT_RTNL(); 226 227 write_lock_bh(&dev_base_lock); 228 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 229 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 230 hlist_add_head_rcu(&dev->index_hlist, 231 dev_index_hash(net, dev->ifindex)); 232 write_unlock_bh(&dev_base_lock); 233 234 dev_base_seq_inc(net); 235 } 236 237 /* Device list removal 238 * caller must respect a RCU grace period before freeing/reusing dev 239 */ 240 static void unlist_netdevice(struct net_device *dev) 241 { 242 ASSERT_RTNL(); 243 244 /* Unlink dev from the device chain */ 245 write_lock_bh(&dev_base_lock); 246 list_del_rcu(&dev->dev_list); 247 hlist_del_rcu(&dev->name_hlist); 248 hlist_del_rcu(&dev->index_hlist); 249 write_unlock_bh(&dev_base_lock); 250 251 dev_base_seq_inc(dev_net(dev)); 252 } 253 254 /* 255 * Our notifier list 256 */ 257 258 static RAW_NOTIFIER_HEAD(netdev_chain); 259 260 /* 261 * Device drivers call our routines to queue packets here. We empty the 262 * queue in the local softnet handler. 263 */ 264 265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 266 EXPORT_PER_CPU_SYMBOL(softnet_data); 267 268 #ifdef CONFIG_LOCKDEP 269 /* 270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 271 * according to dev->type 272 */ 273 static const unsigned short netdev_lock_type[] = 274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 286 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 287 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 288 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 289 290 static const char *const netdev_lock_name[] = 291 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 292 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 293 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 294 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 295 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 296 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 297 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 298 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 299 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 300 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 301 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 302 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 303 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 304 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 305 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 306 307 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 308 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 309 310 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 311 { 312 int i; 313 314 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 315 if (netdev_lock_type[i] == dev_type) 316 return i; 317 /* the last key is used by default */ 318 return ARRAY_SIZE(netdev_lock_type) - 1; 319 } 320 321 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 322 unsigned short dev_type) 323 { 324 int i; 325 326 i = netdev_lock_pos(dev_type); 327 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 328 netdev_lock_name[i]); 329 } 330 331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 332 { 333 int i; 334 335 i = netdev_lock_pos(dev->type); 336 lockdep_set_class_and_name(&dev->addr_list_lock, 337 &netdev_addr_lock_key[i], 338 netdev_lock_name[i]); 339 } 340 #else 341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 342 unsigned short dev_type) 343 { 344 } 345 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 346 { 347 } 348 #endif 349 350 /******************************************************************************* 351 352 Protocol management and registration routines 353 354 *******************************************************************************/ 355 356 /* 357 * Add a protocol ID to the list. Now that the input handler is 358 * smarter we can dispense with all the messy stuff that used to be 359 * here. 360 * 361 * BEWARE!!! Protocol handlers, mangling input packets, 362 * MUST BE last in hash buckets and checking protocol handlers 363 * MUST start from promiscuous ptype_all chain in net_bh. 364 * It is true now, do not change it. 365 * Explanation follows: if protocol handler, mangling packet, will 366 * be the first on list, it is not able to sense, that packet 367 * is cloned and should be copied-on-write, so that it will 368 * change it and subsequent readers will get broken packet. 369 * --ANK (980803) 370 */ 371 372 static inline struct list_head *ptype_head(const struct packet_type *pt) 373 { 374 if (pt->type == htons(ETH_P_ALL)) 375 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 376 else 377 return pt->dev ? &pt->dev->ptype_specific : 378 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 379 } 380 381 /** 382 * dev_add_pack - add packet handler 383 * @pt: packet type declaration 384 * 385 * Add a protocol handler to the networking stack. The passed &packet_type 386 * is linked into kernel lists and may not be freed until it has been 387 * removed from the kernel lists. 388 * 389 * This call does not sleep therefore it can not 390 * guarantee all CPU's that are in middle of receiving packets 391 * will see the new packet type (until the next received packet). 392 */ 393 394 void dev_add_pack(struct packet_type *pt) 395 { 396 struct list_head *head = ptype_head(pt); 397 398 spin_lock(&ptype_lock); 399 list_add_rcu(&pt->list, head); 400 spin_unlock(&ptype_lock); 401 } 402 EXPORT_SYMBOL(dev_add_pack); 403 404 /** 405 * __dev_remove_pack - remove packet handler 406 * @pt: packet type declaration 407 * 408 * Remove a protocol handler that was previously added to the kernel 409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 410 * from the kernel lists and can be freed or reused once this function 411 * returns. 412 * 413 * The packet type might still be in use by receivers 414 * and must not be freed until after all the CPU's have gone 415 * through a quiescent state. 416 */ 417 void __dev_remove_pack(struct packet_type *pt) 418 { 419 struct list_head *head = ptype_head(pt); 420 struct packet_type *pt1; 421 422 spin_lock(&ptype_lock); 423 424 list_for_each_entry(pt1, head, list) { 425 if (pt == pt1) { 426 list_del_rcu(&pt->list); 427 goto out; 428 } 429 } 430 431 pr_warn("dev_remove_pack: %p not found\n", pt); 432 out: 433 spin_unlock(&ptype_lock); 434 } 435 EXPORT_SYMBOL(__dev_remove_pack); 436 437 /** 438 * dev_remove_pack - remove packet handler 439 * @pt: packet type declaration 440 * 441 * Remove a protocol handler that was previously added to the kernel 442 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 443 * from the kernel lists and can be freed or reused once this function 444 * returns. 445 * 446 * This call sleeps to guarantee that no CPU is looking at the packet 447 * type after return. 448 */ 449 void dev_remove_pack(struct packet_type *pt) 450 { 451 __dev_remove_pack(pt); 452 453 synchronize_net(); 454 } 455 EXPORT_SYMBOL(dev_remove_pack); 456 457 458 /** 459 * dev_add_offload - register offload handlers 460 * @po: protocol offload declaration 461 * 462 * Add protocol offload handlers to the networking stack. The passed 463 * &proto_offload is linked into kernel lists and may not be freed until 464 * it has been removed from the kernel lists. 465 * 466 * This call does not sleep therefore it can not 467 * guarantee all CPU's that are in middle of receiving packets 468 * will see the new offload handlers (until the next received packet). 469 */ 470 void dev_add_offload(struct packet_offload *po) 471 { 472 struct packet_offload *elem; 473 474 spin_lock(&offload_lock); 475 list_for_each_entry(elem, &offload_base, list) { 476 if (po->priority < elem->priority) 477 break; 478 } 479 list_add_rcu(&po->list, elem->list.prev); 480 spin_unlock(&offload_lock); 481 } 482 EXPORT_SYMBOL(dev_add_offload); 483 484 /** 485 * __dev_remove_offload - remove offload handler 486 * @po: packet offload declaration 487 * 488 * Remove a protocol offload handler that was previously added to the 489 * kernel offload handlers by dev_add_offload(). The passed &offload_type 490 * is removed from the kernel lists and can be freed or reused once this 491 * function returns. 492 * 493 * The packet type might still be in use by receivers 494 * and must not be freed until after all the CPU's have gone 495 * through a quiescent state. 496 */ 497 static void __dev_remove_offload(struct packet_offload *po) 498 { 499 struct list_head *head = &offload_base; 500 struct packet_offload *po1; 501 502 spin_lock(&offload_lock); 503 504 list_for_each_entry(po1, head, list) { 505 if (po == po1) { 506 list_del_rcu(&po->list); 507 goto out; 508 } 509 } 510 511 pr_warn("dev_remove_offload: %p not found\n", po); 512 out: 513 spin_unlock(&offload_lock); 514 } 515 516 /** 517 * dev_remove_offload - remove packet offload handler 518 * @po: packet offload declaration 519 * 520 * Remove a packet offload handler that was previously added to the kernel 521 * offload handlers by dev_add_offload(). The passed &offload_type is 522 * removed from the kernel lists and can be freed or reused once this 523 * function returns. 524 * 525 * This call sleeps to guarantee that no CPU is looking at the packet 526 * type after return. 527 */ 528 void dev_remove_offload(struct packet_offload *po) 529 { 530 __dev_remove_offload(po); 531 532 synchronize_net(); 533 } 534 EXPORT_SYMBOL(dev_remove_offload); 535 536 /****************************************************************************** 537 538 Device Boot-time Settings Routines 539 540 *******************************************************************************/ 541 542 /* Boot time configuration table */ 543 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 544 545 /** 546 * netdev_boot_setup_add - add new setup entry 547 * @name: name of the device 548 * @map: configured settings for the device 549 * 550 * Adds new setup entry to the dev_boot_setup list. The function 551 * returns 0 on error and 1 on success. This is a generic routine to 552 * all netdevices. 553 */ 554 static int netdev_boot_setup_add(char *name, struct ifmap *map) 555 { 556 struct netdev_boot_setup *s; 557 int i; 558 559 s = dev_boot_setup; 560 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 561 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 562 memset(s[i].name, 0, sizeof(s[i].name)); 563 strlcpy(s[i].name, name, IFNAMSIZ); 564 memcpy(&s[i].map, map, sizeof(s[i].map)); 565 break; 566 } 567 } 568 569 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 570 } 571 572 /** 573 * netdev_boot_setup_check - check boot time settings 574 * @dev: the netdevice 575 * 576 * Check boot time settings for the device. 577 * The found settings are set for the device to be used 578 * later in the device probing. 579 * Returns 0 if no settings found, 1 if they are. 580 */ 581 int netdev_boot_setup_check(struct net_device *dev) 582 { 583 struct netdev_boot_setup *s = dev_boot_setup; 584 int i; 585 586 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 587 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 588 !strcmp(dev->name, s[i].name)) { 589 dev->irq = s[i].map.irq; 590 dev->base_addr = s[i].map.base_addr; 591 dev->mem_start = s[i].map.mem_start; 592 dev->mem_end = s[i].map.mem_end; 593 return 1; 594 } 595 } 596 return 0; 597 } 598 EXPORT_SYMBOL(netdev_boot_setup_check); 599 600 601 /** 602 * netdev_boot_base - get address from boot time settings 603 * @prefix: prefix for network device 604 * @unit: id for network device 605 * 606 * Check boot time settings for the base address of device. 607 * The found settings are set for the device to be used 608 * later in the device probing. 609 * Returns 0 if no settings found. 610 */ 611 unsigned long netdev_boot_base(const char *prefix, int unit) 612 { 613 const struct netdev_boot_setup *s = dev_boot_setup; 614 char name[IFNAMSIZ]; 615 int i; 616 617 sprintf(name, "%s%d", prefix, unit); 618 619 /* 620 * If device already registered then return base of 1 621 * to indicate not to probe for this interface 622 */ 623 if (__dev_get_by_name(&init_net, name)) 624 return 1; 625 626 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 627 if (!strcmp(name, s[i].name)) 628 return s[i].map.base_addr; 629 return 0; 630 } 631 632 /* 633 * Saves at boot time configured settings for any netdevice. 634 */ 635 int __init netdev_boot_setup(char *str) 636 { 637 int ints[5]; 638 struct ifmap map; 639 640 str = get_options(str, ARRAY_SIZE(ints), ints); 641 if (!str || !*str) 642 return 0; 643 644 /* Save settings */ 645 memset(&map, 0, sizeof(map)); 646 if (ints[0] > 0) 647 map.irq = ints[1]; 648 if (ints[0] > 1) 649 map.base_addr = ints[2]; 650 if (ints[0] > 2) 651 map.mem_start = ints[3]; 652 if (ints[0] > 3) 653 map.mem_end = ints[4]; 654 655 /* Add new entry to the list */ 656 return netdev_boot_setup_add(str, &map); 657 } 658 659 __setup("netdev=", netdev_boot_setup); 660 661 /******************************************************************************* 662 663 Device Interface Subroutines 664 665 *******************************************************************************/ 666 667 /** 668 * dev_get_iflink - get 'iflink' value of a interface 669 * @dev: targeted interface 670 * 671 * Indicates the ifindex the interface is linked to. 672 * Physical interfaces have the same 'ifindex' and 'iflink' values. 673 */ 674 675 int dev_get_iflink(const struct net_device *dev) 676 { 677 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 678 return dev->netdev_ops->ndo_get_iflink(dev); 679 680 /* If dev->rtnl_link_ops is set, it's a virtual interface. */ 681 if (dev->rtnl_link_ops) 682 return 0; 683 684 return dev->ifindex; 685 } 686 EXPORT_SYMBOL(dev_get_iflink); 687 688 /** 689 * __dev_get_by_name - find a device by its name 690 * @net: the applicable net namespace 691 * @name: name to find 692 * 693 * Find an interface by name. Must be called under RTNL semaphore 694 * or @dev_base_lock. If the name is found a pointer to the device 695 * is returned. If the name is not found then %NULL is returned. The 696 * reference counters are not incremented so the caller must be 697 * careful with locks. 698 */ 699 700 struct net_device *__dev_get_by_name(struct net *net, const char *name) 701 { 702 struct net_device *dev; 703 struct hlist_head *head = dev_name_hash(net, name); 704 705 hlist_for_each_entry(dev, head, name_hlist) 706 if (!strncmp(dev->name, name, IFNAMSIZ)) 707 return dev; 708 709 return NULL; 710 } 711 EXPORT_SYMBOL(__dev_get_by_name); 712 713 /** 714 * dev_get_by_name_rcu - find a device by its name 715 * @net: the applicable net namespace 716 * @name: name to find 717 * 718 * Find an interface by name. 719 * If the name is found a pointer to the device is returned. 720 * If the name is not found then %NULL is returned. 721 * The reference counters are not incremented so the caller must be 722 * careful with locks. The caller must hold RCU lock. 723 */ 724 725 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 726 { 727 struct net_device *dev; 728 struct hlist_head *head = dev_name_hash(net, name); 729 730 hlist_for_each_entry_rcu(dev, head, name_hlist) 731 if (!strncmp(dev->name, name, IFNAMSIZ)) 732 return dev; 733 734 return NULL; 735 } 736 EXPORT_SYMBOL(dev_get_by_name_rcu); 737 738 /** 739 * dev_get_by_name - find a device by its name 740 * @net: the applicable net namespace 741 * @name: name to find 742 * 743 * Find an interface by name. This can be called from any 744 * context and does its own locking. The returned handle has 745 * the usage count incremented and the caller must use dev_put() to 746 * release it when it is no longer needed. %NULL is returned if no 747 * matching device is found. 748 */ 749 750 struct net_device *dev_get_by_name(struct net *net, const char *name) 751 { 752 struct net_device *dev; 753 754 rcu_read_lock(); 755 dev = dev_get_by_name_rcu(net, name); 756 if (dev) 757 dev_hold(dev); 758 rcu_read_unlock(); 759 return dev; 760 } 761 EXPORT_SYMBOL(dev_get_by_name); 762 763 /** 764 * __dev_get_by_index - find a device by its ifindex 765 * @net: the applicable net namespace 766 * @ifindex: index of device 767 * 768 * Search for an interface by index. Returns %NULL if the device 769 * is not found or a pointer to the device. The device has not 770 * had its reference counter increased so the caller must be careful 771 * about locking. The caller must hold either the RTNL semaphore 772 * or @dev_base_lock. 773 */ 774 775 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 776 { 777 struct net_device *dev; 778 struct hlist_head *head = dev_index_hash(net, ifindex); 779 780 hlist_for_each_entry(dev, head, index_hlist) 781 if (dev->ifindex == ifindex) 782 return dev; 783 784 return NULL; 785 } 786 EXPORT_SYMBOL(__dev_get_by_index); 787 788 /** 789 * dev_get_by_index_rcu - find a device by its ifindex 790 * @net: the applicable net namespace 791 * @ifindex: index of device 792 * 793 * Search for an interface by index. Returns %NULL if the device 794 * is not found or a pointer to the device. The device has not 795 * had its reference counter increased so the caller must be careful 796 * about locking. The caller must hold RCU lock. 797 */ 798 799 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 800 { 801 struct net_device *dev; 802 struct hlist_head *head = dev_index_hash(net, ifindex); 803 804 hlist_for_each_entry_rcu(dev, head, index_hlist) 805 if (dev->ifindex == ifindex) 806 return dev; 807 808 return NULL; 809 } 810 EXPORT_SYMBOL(dev_get_by_index_rcu); 811 812 813 /** 814 * dev_get_by_index - find a device by its ifindex 815 * @net: the applicable net namespace 816 * @ifindex: index of device 817 * 818 * Search for an interface by index. Returns NULL if the device 819 * is not found or a pointer to the device. The device returned has 820 * had a reference added and the pointer is safe until the user calls 821 * dev_put to indicate they have finished with it. 822 */ 823 824 struct net_device *dev_get_by_index(struct net *net, int ifindex) 825 { 826 struct net_device *dev; 827 828 rcu_read_lock(); 829 dev = dev_get_by_index_rcu(net, ifindex); 830 if (dev) 831 dev_hold(dev); 832 rcu_read_unlock(); 833 return dev; 834 } 835 EXPORT_SYMBOL(dev_get_by_index); 836 837 /** 838 * netdev_get_name - get a netdevice name, knowing its ifindex. 839 * @net: network namespace 840 * @name: a pointer to the buffer where the name will be stored. 841 * @ifindex: the ifindex of the interface to get the name from. 842 * 843 * The use of raw_seqcount_begin() and cond_resched() before 844 * retrying is required as we want to give the writers a chance 845 * to complete when CONFIG_PREEMPT is not set. 846 */ 847 int netdev_get_name(struct net *net, char *name, int ifindex) 848 { 849 struct net_device *dev; 850 unsigned int seq; 851 852 retry: 853 seq = raw_seqcount_begin(&devnet_rename_seq); 854 rcu_read_lock(); 855 dev = dev_get_by_index_rcu(net, ifindex); 856 if (!dev) { 857 rcu_read_unlock(); 858 return -ENODEV; 859 } 860 861 strcpy(name, dev->name); 862 rcu_read_unlock(); 863 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 864 cond_resched(); 865 goto retry; 866 } 867 868 return 0; 869 } 870 871 /** 872 * dev_getbyhwaddr_rcu - find a device by its hardware address 873 * @net: the applicable net namespace 874 * @type: media type of device 875 * @ha: hardware address 876 * 877 * Search for an interface by MAC address. Returns NULL if the device 878 * is not found or a pointer to the device. 879 * The caller must hold RCU or RTNL. 880 * The returned device has not had its ref count increased 881 * and the caller must therefore be careful about locking 882 * 883 */ 884 885 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 886 const char *ha) 887 { 888 struct net_device *dev; 889 890 for_each_netdev_rcu(net, dev) 891 if (dev->type == type && 892 !memcmp(dev->dev_addr, ha, dev->addr_len)) 893 return dev; 894 895 return NULL; 896 } 897 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 898 899 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 900 { 901 struct net_device *dev; 902 903 ASSERT_RTNL(); 904 for_each_netdev(net, dev) 905 if (dev->type == type) 906 return dev; 907 908 return NULL; 909 } 910 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 911 912 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 913 { 914 struct net_device *dev, *ret = NULL; 915 916 rcu_read_lock(); 917 for_each_netdev_rcu(net, dev) 918 if (dev->type == type) { 919 dev_hold(dev); 920 ret = dev; 921 break; 922 } 923 rcu_read_unlock(); 924 return ret; 925 } 926 EXPORT_SYMBOL(dev_getfirstbyhwtype); 927 928 /** 929 * __dev_get_by_flags - find any device with given flags 930 * @net: the applicable net namespace 931 * @if_flags: IFF_* values 932 * @mask: bitmask of bits in if_flags to check 933 * 934 * Search for any interface with the given flags. Returns NULL if a device 935 * is not found or a pointer to the device. Must be called inside 936 * rtnl_lock(), and result refcount is unchanged. 937 */ 938 939 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 940 unsigned short mask) 941 { 942 struct net_device *dev, *ret; 943 944 ASSERT_RTNL(); 945 946 ret = NULL; 947 for_each_netdev(net, dev) { 948 if (((dev->flags ^ if_flags) & mask) == 0) { 949 ret = dev; 950 break; 951 } 952 } 953 return ret; 954 } 955 EXPORT_SYMBOL(__dev_get_by_flags); 956 957 /** 958 * dev_valid_name - check if name is okay for network device 959 * @name: name string 960 * 961 * Network device names need to be valid file names to 962 * to allow sysfs to work. We also disallow any kind of 963 * whitespace. 964 */ 965 bool dev_valid_name(const char *name) 966 { 967 if (*name == '\0') 968 return false; 969 if (strlen(name) >= IFNAMSIZ) 970 return false; 971 if (!strcmp(name, ".") || !strcmp(name, "..")) 972 return false; 973 974 while (*name) { 975 if (*name == '/' || *name == ':' || isspace(*name)) 976 return false; 977 name++; 978 } 979 return true; 980 } 981 EXPORT_SYMBOL(dev_valid_name); 982 983 /** 984 * __dev_alloc_name - allocate a name for a device 985 * @net: network namespace to allocate the device name in 986 * @name: name format string 987 * @buf: scratch buffer and result name string 988 * 989 * Passed a format string - eg "lt%d" it will try and find a suitable 990 * id. It scans list of devices to build up a free map, then chooses 991 * the first empty slot. The caller must hold the dev_base or rtnl lock 992 * while allocating the name and adding the device in order to avoid 993 * duplicates. 994 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 995 * Returns the number of the unit assigned or a negative errno code. 996 */ 997 998 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 999 { 1000 int i = 0; 1001 const char *p; 1002 const int max_netdevices = 8*PAGE_SIZE; 1003 unsigned long *inuse; 1004 struct net_device *d; 1005 1006 p = strnchr(name, IFNAMSIZ-1, '%'); 1007 if (p) { 1008 /* 1009 * Verify the string as this thing may have come from 1010 * the user. There must be either one "%d" and no other "%" 1011 * characters. 1012 */ 1013 if (p[1] != 'd' || strchr(p + 2, '%')) 1014 return -EINVAL; 1015 1016 /* Use one page as a bit array of possible slots */ 1017 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1018 if (!inuse) 1019 return -ENOMEM; 1020 1021 for_each_netdev(net, d) { 1022 if (!sscanf(d->name, name, &i)) 1023 continue; 1024 if (i < 0 || i >= max_netdevices) 1025 continue; 1026 1027 /* avoid cases where sscanf is not exact inverse of printf */ 1028 snprintf(buf, IFNAMSIZ, name, i); 1029 if (!strncmp(buf, d->name, IFNAMSIZ)) 1030 set_bit(i, inuse); 1031 } 1032 1033 i = find_first_zero_bit(inuse, max_netdevices); 1034 free_page((unsigned long) inuse); 1035 } 1036 1037 if (buf != name) 1038 snprintf(buf, IFNAMSIZ, name, i); 1039 if (!__dev_get_by_name(net, buf)) 1040 return i; 1041 1042 /* It is possible to run out of possible slots 1043 * when the name is long and there isn't enough space left 1044 * for the digits, or if all bits are used. 1045 */ 1046 return -ENFILE; 1047 } 1048 1049 /** 1050 * dev_alloc_name - allocate a name for a device 1051 * @dev: device 1052 * @name: name format string 1053 * 1054 * Passed a format string - eg "lt%d" it will try and find a suitable 1055 * id. It scans list of devices to build up a free map, then chooses 1056 * the first empty slot. The caller must hold the dev_base or rtnl lock 1057 * while allocating the name and adding the device in order to avoid 1058 * duplicates. 1059 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1060 * Returns the number of the unit assigned or a negative errno code. 1061 */ 1062 1063 int dev_alloc_name(struct net_device *dev, const char *name) 1064 { 1065 char buf[IFNAMSIZ]; 1066 struct net *net; 1067 int ret; 1068 1069 BUG_ON(!dev_net(dev)); 1070 net = dev_net(dev); 1071 ret = __dev_alloc_name(net, name, buf); 1072 if (ret >= 0) 1073 strlcpy(dev->name, buf, IFNAMSIZ); 1074 return ret; 1075 } 1076 EXPORT_SYMBOL(dev_alloc_name); 1077 1078 static int dev_alloc_name_ns(struct net *net, 1079 struct net_device *dev, 1080 const char *name) 1081 { 1082 char buf[IFNAMSIZ]; 1083 int ret; 1084 1085 ret = __dev_alloc_name(net, name, buf); 1086 if (ret >= 0) 1087 strlcpy(dev->name, buf, IFNAMSIZ); 1088 return ret; 1089 } 1090 1091 static int dev_get_valid_name(struct net *net, 1092 struct net_device *dev, 1093 const char *name) 1094 { 1095 BUG_ON(!net); 1096 1097 if (!dev_valid_name(name)) 1098 return -EINVAL; 1099 1100 if (strchr(name, '%')) 1101 return dev_alloc_name_ns(net, dev, name); 1102 else if (__dev_get_by_name(net, name)) 1103 return -EEXIST; 1104 else if (dev->name != name) 1105 strlcpy(dev->name, name, IFNAMSIZ); 1106 1107 return 0; 1108 } 1109 1110 /** 1111 * dev_change_name - change name of a device 1112 * @dev: device 1113 * @newname: name (or format string) must be at least IFNAMSIZ 1114 * 1115 * Change name of a device, can pass format strings "eth%d". 1116 * for wildcarding. 1117 */ 1118 int dev_change_name(struct net_device *dev, const char *newname) 1119 { 1120 unsigned char old_assign_type; 1121 char oldname[IFNAMSIZ]; 1122 int err = 0; 1123 int ret; 1124 struct net *net; 1125 1126 ASSERT_RTNL(); 1127 BUG_ON(!dev_net(dev)); 1128 1129 net = dev_net(dev); 1130 if (dev->flags & IFF_UP) 1131 return -EBUSY; 1132 1133 write_seqcount_begin(&devnet_rename_seq); 1134 1135 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1136 write_seqcount_end(&devnet_rename_seq); 1137 return 0; 1138 } 1139 1140 memcpy(oldname, dev->name, IFNAMSIZ); 1141 1142 err = dev_get_valid_name(net, dev, newname); 1143 if (err < 0) { 1144 write_seqcount_end(&devnet_rename_seq); 1145 return err; 1146 } 1147 1148 if (oldname[0] && !strchr(oldname, '%')) 1149 netdev_info(dev, "renamed from %s\n", oldname); 1150 1151 old_assign_type = dev->name_assign_type; 1152 dev->name_assign_type = NET_NAME_RENAMED; 1153 1154 rollback: 1155 ret = device_rename(&dev->dev, dev->name); 1156 if (ret) { 1157 memcpy(dev->name, oldname, IFNAMSIZ); 1158 dev->name_assign_type = old_assign_type; 1159 write_seqcount_end(&devnet_rename_seq); 1160 return ret; 1161 } 1162 1163 write_seqcount_end(&devnet_rename_seq); 1164 1165 netdev_adjacent_rename_links(dev, oldname); 1166 1167 write_lock_bh(&dev_base_lock); 1168 hlist_del_rcu(&dev->name_hlist); 1169 write_unlock_bh(&dev_base_lock); 1170 1171 synchronize_rcu(); 1172 1173 write_lock_bh(&dev_base_lock); 1174 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1175 write_unlock_bh(&dev_base_lock); 1176 1177 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1178 ret = notifier_to_errno(ret); 1179 1180 if (ret) { 1181 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1182 if (err >= 0) { 1183 err = ret; 1184 write_seqcount_begin(&devnet_rename_seq); 1185 memcpy(dev->name, oldname, IFNAMSIZ); 1186 memcpy(oldname, newname, IFNAMSIZ); 1187 dev->name_assign_type = old_assign_type; 1188 old_assign_type = NET_NAME_RENAMED; 1189 goto rollback; 1190 } else { 1191 pr_err("%s: name change rollback failed: %d\n", 1192 dev->name, ret); 1193 } 1194 } 1195 1196 return err; 1197 } 1198 1199 /** 1200 * dev_set_alias - change ifalias of a device 1201 * @dev: device 1202 * @alias: name up to IFALIASZ 1203 * @len: limit of bytes to copy from info 1204 * 1205 * Set ifalias for a device, 1206 */ 1207 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1208 { 1209 char *new_ifalias; 1210 1211 ASSERT_RTNL(); 1212 1213 if (len >= IFALIASZ) 1214 return -EINVAL; 1215 1216 if (!len) { 1217 kfree(dev->ifalias); 1218 dev->ifalias = NULL; 1219 return 0; 1220 } 1221 1222 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); 1223 if (!new_ifalias) 1224 return -ENOMEM; 1225 dev->ifalias = new_ifalias; 1226 1227 strlcpy(dev->ifalias, alias, len+1); 1228 return len; 1229 } 1230 1231 1232 /** 1233 * netdev_features_change - device changes features 1234 * @dev: device to cause notification 1235 * 1236 * Called to indicate a device has changed features. 1237 */ 1238 void netdev_features_change(struct net_device *dev) 1239 { 1240 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1241 } 1242 EXPORT_SYMBOL(netdev_features_change); 1243 1244 /** 1245 * netdev_state_change - device changes state 1246 * @dev: device to cause notification 1247 * 1248 * Called to indicate a device has changed state. This function calls 1249 * the notifier chains for netdev_chain and sends a NEWLINK message 1250 * to the routing socket. 1251 */ 1252 void netdev_state_change(struct net_device *dev) 1253 { 1254 if (dev->flags & IFF_UP) { 1255 struct netdev_notifier_change_info change_info; 1256 1257 change_info.flags_changed = 0; 1258 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 1259 &change_info.info); 1260 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1261 } 1262 } 1263 EXPORT_SYMBOL(netdev_state_change); 1264 1265 /** 1266 * netdev_notify_peers - notify network peers about existence of @dev 1267 * @dev: network device 1268 * 1269 * Generate traffic such that interested network peers are aware of 1270 * @dev, such as by generating a gratuitous ARP. This may be used when 1271 * a device wants to inform the rest of the network about some sort of 1272 * reconfiguration such as a failover event or virtual machine 1273 * migration. 1274 */ 1275 void netdev_notify_peers(struct net_device *dev) 1276 { 1277 rtnl_lock(); 1278 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1279 rtnl_unlock(); 1280 } 1281 EXPORT_SYMBOL(netdev_notify_peers); 1282 1283 static int __dev_open(struct net_device *dev) 1284 { 1285 const struct net_device_ops *ops = dev->netdev_ops; 1286 int ret; 1287 1288 ASSERT_RTNL(); 1289 1290 if (!netif_device_present(dev)) 1291 return -ENODEV; 1292 1293 /* Block netpoll from trying to do any rx path servicing. 1294 * If we don't do this there is a chance ndo_poll_controller 1295 * or ndo_poll may be running while we open the device 1296 */ 1297 netpoll_poll_disable(dev); 1298 1299 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); 1300 ret = notifier_to_errno(ret); 1301 if (ret) 1302 return ret; 1303 1304 set_bit(__LINK_STATE_START, &dev->state); 1305 1306 if (ops->ndo_validate_addr) 1307 ret = ops->ndo_validate_addr(dev); 1308 1309 if (!ret && ops->ndo_open) 1310 ret = ops->ndo_open(dev); 1311 1312 netpoll_poll_enable(dev); 1313 1314 if (ret) 1315 clear_bit(__LINK_STATE_START, &dev->state); 1316 else { 1317 dev->flags |= IFF_UP; 1318 dev_set_rx_mode(dev); 1319 dev_activate(dev); 1320 add_device_randomness(dev->dev_addr, dev->addr_len); 1321 } 1322 1323 return ret; 1324 } 1325 1326 /** 1327 * dev_open - prepare an interface for use. 1328 * @dev: device to open 1329 * 1330 * Takes a device from down to up state. The device's private open 1331 * function is invoked and then the multicast lists are loaded. Finally 1332 * the device is moved into the up state and a %NETDEV_UP message is 1333 * sent to the netdev notifier chain. 1334 * 1335 * Calling this function on an active interface is a nop. On a failure 1336 * a negative errno code is returned. 1337 */ 1338 int dev_open(struct net_device *dev) 1339 { 1340 int ret; 1341 1342 if (dev->flags & IFF_UP) 1343 return 0; 1344 1345 ret = __dev_open(dev); 1346 if (ret < 0) 1347 return ret; 1348 1349 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1350 call_netdevice_notifiers(NETDEV_UP, dev); 1351 1352 return ret; 1353 } 1354 EXPORT_SYMBOL(dev_open); 1355 1356 static int __dev_close_many(struct list_head *head) 1357 { 1358 struct net_device *dev; 1359 1360 ASSERT_RTNL(); 1361 might_sleep(); 1362 1363 list_for_each_entry(dev, head, close_list) { 1364 /* Temporarily disable netpoll until the interface is down */ 1365 netpoll_poll_disable(dev); 1366 1367 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1368 1369 clear_bit(__LINK_STATE_START, &dev->state); 1370 1371 /* Synchronize to scheduled poll. We cannot touch poll list, it 1372 * can be even on different cpu. So just clear netif_running(). 1373 * 1374 * dev->stop() will invoke napi_disable() on all of it's 1375 * napi_struct instances on this device. 1376 */ 1377 smp_mb__after_atomic(); /* Commit netif_running(). */ 1378 } 1379 1380 dev_deactivate_many(head); 1381 1382 list_for_each_entry(dev, head, close_list) { 1383 const struct net_device_ops *ops = dev->netdev_ops; 1384 1385 /* 1386 * Call the device specific close. This cannot fail. 1387 * Only if device is UP 1388 * 1389 * We allow it to be called even after a DETACH hot-plug 1390 * event. 1391 */ 1392 if (ops->ndo_stop) 1393 ops->ndo_stop(dev); 1394 1395 dev->flags &= ~IFF_UP; 1396 netpoll_poll_enable(dev); 1397 } 1398 1399 return 0; 1400 } 1401 1402 static int __dev_close(struct net_device *dev) 1403 { 1404 int retval; 1405 LIST_HEAD(single); 1406 1407 list_add(&dev->close_list, &single); 1408 retval = __dev_close_many(&single); 1409 list_del(&single); 1410 1411 return retval; 1412 } 1413 1414 int dev_close_many(struct list_head *head, bool unlink) 1415 { 1416 struct net_device *dev, *tmp; 1417 1418 /* Remove the devices that don't need to be closed */ 1419 list_for_each_entry_safe(dev, tmp, head, close_list) 1420 if (!(dev->flags & IFF_UP)) 1421 list_del_init(&dev->close_list); 1422 1423 __dev_close_many(head); 1424 1425 list_for_each_entry_safe(dev, tmp, head, close_list) { 1426 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1427 call_netdevice_notifiers(NETDEV_DOWN, dev); 1428 if (unlink) 1429 list_del_init(&dev->close_list); 1430 } 1431 1432 return 0; 1433 } 1434 EXPORT_SYMBOL(dev_close_many); 1435 1436 /** 1437 * dev_close - shutdown an interface. 1438 * @dev: device to shutdown 1439 * 1440 * This function moves an active device into down state. A 1441 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1442 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1443 * chain. 1444 */ 1445 int dev_close(struct net_device *dev) 1446 { 1447 if (dev->flags & IFF_UP) { 1448 LIST_HEAD(single); 1449 1450 list_add(&dev->close_list, &single); 1451 dev_close_many(&single, true); 1452 list_del(&single); 1453 } 1454 return 0; 1455 } 1456 EXPORT_SYMBOL(dev_close); 1457 1458 1459 /** 1460 * dev_disable_lro - disable Large Receive Offload on a device 1461 * @dev: device 1462 * 1463 * Disable Large Receive Offload (LRO) on a net device. Must be 1464 * called under RTNL. This is needed if received packets may be 1465 * forwarded to another interface. 1466 */ 1467 void dev_disable_lro(struct net_device *dev) 1468 { 1469 struct net_device *lower_dev; 1470 struct list_head *iter; 1471 1472 dev->wanted_features &= ~NETIF_F_LRO; 1473 netdev_update_features(dev); 1474 1475 if (unlikely(dev->features & NETIF_F_LRO)) 1476 netdev_WARN(dev, "failed to disable LRO!\n"); 1477 1478 netdev_for_each_lower_dev(dev, lower_dev, iter) 1479 dev_disable_lro(lower_dev); 1480 } 1481 EXPORT_SYMBOL(dev_disable_lro); 1482 1483 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1484 struct net_device *dev) 1485 { 1486 struct netdev_notifier_info info; 1487 1488 netdev_notifier_info_init(&info, dev); 1489 return nb->notifier_call(nb, val, &info); 1490 } 1491 1492 static int dev_boot_phase = 1; 1493 1494 /** 1495 * register_netdevice_notifier - register a network notifier block 1496 * @nb: notifier 1497 * 1498 * Register a notifier to be called when network device events occur. 1499 * The notifier passed is linked into the kernel structures and must 1500 * not be reused until it has been unregistered. A negative errno code 1501 * is returned on a failure. 1502 * 1503 * When registered all registration and up events are replayed 1504 * to the new notifier to allow device to have a race free 1505 * view of the network device list. 1506 */ 1507 1508 int register_netdevice_notifier(struct notifier_block *nb) 1509 { 1510 struct net_device *dev; 1511 struct net_device *last; 1512 struct net *net; 1513 int err; 1514 1515 rtnl_lock(); 1516 err = raw_notifier_chain_register(&netdev_chain, nb); 1517 if (err) 1518 goto unlock; 1519 if (dev_boot_phase) 1520 goto unlock; 1521 for_each_net(net) { 1522 for_each_netdev(net, dev) { 1523 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1524 err = notifier_to_errno(err); 1525 if (err) 1526 goto rollback; 1527 1528 if (!(dev->flags & IFF_UP)) 1529 continue; 1530 1531 call_netdevice_notifier(nb, NETDEV_UP, dev); 1532 } 1533 } 1534 1535 unlock: 1536 rtnl_unlock(); 1537 return err; 1538 1539 rollback: 1540 last = dev; 1541 for_each_net(net) { 1542 for_each_netdev(net, dev) { 1543 if (dev == last) 1544 goto outroll; 1545 1546 if (dev->flags & IFF_UP) { 1547 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1548 dev); 1549 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1550 } 1551 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1552 } 1553 } 1554 1555 outroll: 1556 raw_notifier_chain_unregister(&netdev_chain, nb); 1557 goto unlock; 1558 } 1559 EXPORT_SYMBOL(register_netdevice_notifier); 1560 1561 /** 1562 * unregister_netdevice_notifier - unregister a network notifier block 1563 * @nb: notifier 1564 * 1565 * Unregister a notifier previously registered by 1566 * register_netdevice_notifier(). The notifier is unlinked into the 1567 * kernel structures and may then be reused. A negative errno code 1568 * is returned on a failure. 1569 * 1570 * After unregistering unregister and down device events are synthesized 1571 * for all devices on the device list to the removed notifier to remove 1572 * the need for special case cleanup code. 1573 */ 1574 1575 int unregister_netdevice_notifier(struct notifier_block *nb) 1576 { 1577 struct net_device *dev; 1578 struct net *net; 1579 int err; 1580 1581 rtnl_lock(); 1582 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1583 if (err) 1584 goto unlock; 1585 1586 for_each_net(net) { 1587 for_each_netdev(net, dev) { 1588 if (dev->flags & IFF_UP) { 1589 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1590 dev); 1591 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1592 } 1593 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1594 } 1595 } 1596 unlock: 1597 rtnl_unlock(); 1598 return err; 1599 } 1600 EXPORT_SYMBOL(unregister_netdevice_notifier); 1601 1602 /** 1603 * call_netdevice_notifiers_info - call all network notifier blocks 1604 * @val: value passed unmodified to notifier function 1605 * @dev: net_device pointer passed unmodified to notifier function 1606 * @info: notifier information data 1607 * 1608 * Call all network notifier blocks. Parameters and return value 1609 * are as for raw_notifier_call_chain(). 1610 */ 1611 1612 static int call_netdevice_notifiers_info(unsigned long val, 1613 struct net_device *dev, 1614 struct netdev_notifier_info *info) 1615 { 1616 ASSERT_RTNL(); 1617 netdev_notifier_info_init(info, dev); 1618 return raw_notifier_call_chain(&netdev_chain, val, info); 1619 } 1620 1621 /** 1622 * call_netdevice_notifiers - call all network notifier blocks 1623 * @val: value passed unmodified to notifier function 1624 * @dev: net_device pointer passed unmodified to notifier function 1625 * 1626 * Call all network notifier blocks. Parameters and return value 1627 * are as for raw_notifier_call_chain(). 1628 */ 1629 1630 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1631 { 1632 struct netdev_notifier_info info; 1633 1634 return call_netdevice_notifiers_info(val, dev, &info); 1635 } 1636 EXPORT_SYMBOL(call_netdevice_notifiers); 1637 1638 #ifdef CONFIG_NET_INGRESS 1639 static struct static_key ingress_needed __read_mostly; 1640 1641 void net_inc_ingress_queue(void) 1642 { 1643 static_key_slow_inc(&ingress_needed); 1644 } 1645 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1646 1647 void net_dec_ingress_queue(void) 1648 { 1649 static_key_slow_dec(&ingress_needed); 1650 } 1651 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1652 #endif 1653 1654 static struct static_key netstamp_needed __read_mostly; 1655 #ifdef HAVE_JUMP_LABEL 1656 /* We are not allowed to call static_key_slow_dec() from irq context 1657 * If net_disable_timestamp() is called from irq context, defer the 1658 * static_key_slow_dec() calls. 1659 */ 1660 static atomic_t netstamp_needed_deferred; 1661 #endif 1662 1663 void net_enable_timestamp(void) 1664 { 1665 #ifdef HAVE_JUMP_LABEL 1666 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1667 1668 if (deferred) { 1669 while (--deferred) 1670 static_key_slow_dec(&netstamp_needed); 1671 return; 1672 } 1673 #endif 1674 static_key_slow_inc(&netstamp_needed); 1675 } 1676 EXPORT_SYMBOL(net_enable_timestamp); 1677 1678 void net_disable_timestamp(void) 1679 { 1680 #ifdef HAVE_JUMP_LABEL 1681 if (in_interrupt()) { 1682 atomic_inc(&netstamp_needed_deferred); 1683 return; 1684 } 1685 #endif 1686 static_key_slow_dec(&netstamp_needed); 1687 } 1688 EXPORT_SYMBOL(net_disable_timestamp); 1689 1690 static inline void net_timestamp_set(struct sk_buff *skb) 1691 { 1692 skb->tstamp.tv64 = 0; 1693 if (static_key_false(&netstamp_needed)) 1694 __net_timestamp(skb); 1695 } 1696 1697 #define net_timestamp_check(COND, SKB) \ 1698 if (static_key_false(&netstamp_needed)) { \ 1699 if ((COND) && !(SKB)->tstamp.tv64) \ 1700 __net_timestamp(SKB); \ 1701 } \ 1702 1703 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) 1704 { 1705 unsigned int len; 1706 1707 if (!(dev->flags & IFF_UP)) 1708 return false; 1709 1710 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1711 if (skb->len <= len) 1712 return true; 1713 1714 /* if TSO is enabled, we don't care about the length as the packet 1715 * could be forwarded without being segmented before 1716 */ 1717 if (skb_is_gso(skb)) 1718 return true; 1719 1720 return false; 1721 } 1722 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1723 1724 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1725 { 1726 if (skb_orphan_frags(skb, GFP_ATOMIC) || 1727 unlikely(!is_skb_forwardable(dev, skb))) { 1728 atomic_long_inc(&dev->rx_dropped); 1729 kfree_skb(skb); 1730 return NET_RX_DROP; 1731 } 1732 1733 skb_scrub_packet(skb, true); 1734 skb->priority = 0; 1735 skb->protocol = eth_type_trans(skb, dev); 1736 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1737 1738 return 0; 1739 } 1740 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1741 1742 /** 1743 * dev_forward_skb - loopback an skb to another netif 1744 * 1745 * @dev: destination network device 1746 * @skb: buffer to forward 1747 * 1748 * return values: 1749 * NET_RX_SUCCESS (no congestion) 1750 * NET_RX_DROP (packet was dropped, but freed) 1751 * 1752 * dev_forward_skb can be used for injecting an skb from the 1753 * start_xmit function of one device into the receive queue 1754 * of another device. 1755 * 1756 * The receiving device may be in another namespace, so 1757 * we have to clear all information in the skb that could 1758 * impact namespace isolation. 1759 */ 1760 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1761 { 1762 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1763 } 1764 EXPORT_SYMBOL_GPL(dev_forward_skb); 1765 1766 static inline int deliver_skb(struct sk_buff *skb, 1767 struct packet_type *pt_prev, 1768 struct net_device *orig_dev) 1769 { 1770 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 1771 return -ENOMEM; 1772 atomic_inc(&skb->users); 1773 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1774 } 1775 1776 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 1777 struct packet_type **pt, 1778 struct net_device *orig_dev, 1779 __be16 type, 1780 struct list_head *ptype_list) 1781 { 1782 struct packet_type *ptype, *pt_prev = *pt; 1783 1784 list_for_each_entry_rcu(ptype, ptype_list, list) { 1785 if (ptype->type != type) 1786 continue; 1787 if (pt_prev) 1788 deliver_skb(skb, pt_prev, orig_dev); 1789 pt_prev = ptype; 1790 } 1791 *pt = pt_prev; 1792 } 1793 1794 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1795 { 1796 if (!ptype->af_packet_priv || !skb->sk) 1797 return false; 1798 1799 if (ptype->id_match) 1800 return ptype->id_match(ptype, skb->sk); 1801 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1802 return true; 1803 1804 return false; 1805 } 1806 1807 /* 1808 * Support routine. Sends outgoing frames to any network 1809 * taps currently in use. 1810 */ 1811 1812 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1813 { 1814 struct packet_type *ptype; 1815 struct sk_buff *skb2 = NULL; 1816 struct packet_type *pt_prev = NULL; 1817 struct list_head *ptype_list = &ptype_all; 1818 1819 rcu_read_lock(); 1820 again: 1821 list_for_each_entry_rcu(ptype, ptype_list, list) { 1822 /* Never send packets back to the socket 1823 * they originated from - MvS (miquels@drinkel.ow.org) 1824 */ 1825 if (skb_loop_sk(ptype, skb)) 1826 continue; 1827 1828 if (pt_prev) { 1829 deliver_skb(skb2, pt_prev, skb->dev); 1830 pt_prev = ptype; 1831 continue; 1832 } 1833 1834 /* need to clone skb, done only once */ 1835 skb2 = skb_clone(skb, GFP_ATOMIC); 1836 if (!skb2) 1837 goto out_unlock; 1838 1839 net_timestamp_set(skb2); 1840 1841 /* skb->nh should be correctly 1842 * set by sender, so that the second statement is 1843 * just protection against buggy protocols. 1844 */ 1845 skb_reset_mac_header(skb2); 1846 1847 if (skb_network_header(skb2) < skb2->data || 1848 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 1849 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 1850 ntohs(skb2->protocol), 1851 dev->name); 1852 skb_reset_network_header(skb2); 1853 } 1854 1855 skb2->transport_header = skb2->network_header; 1856 skb2->pkt_type = PACKET_OUTGOING; 1857 pt_prev = ptype; 1858 } 1859 1860 if (ptype_list == &ptype_all) { 1861 ptype_list = &dev->ptype_all; 1862 goto again; 1863 } 1864 out_unlock: 1865 if (pt_prev) 1866 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 1867 rcu_read_unlock(); 1868 } 1869 1870 /** 1871 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 1872 * @dev: Network device 1873 * @txq: number of queues available 1874 * 1875 * If real_num_tx_queues is changed the tc mappings may no longer be 1876 * valid. To resolve this verify the tc mapping remains valid and if 1877 * not NULL the mapping. With no priorities mapping to this 1878 * offset/count pair it will no longer be used. In the worst case TC0 1879 * is invalid nothing can be done so disable priority mappings. If is 1880 * expected that drivers will fix this mapping if they can before 1881 * calling netif_set_real_num_tx_queues. 1882 */ 1883 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 1884 { 1885 int i; 1886 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 1887 1888 /* If TC0 is invalidated disable TC mapping */ 1889 if (tc->offset + tc->count > txq) { 1890 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 1891 dev->num_tc = 0; 1892 return; 1893 } 1894 1895 /* Invalidated prio to tc mappings set to TC0 */ 1896 for (i = 1; i < TC_BITMASK + 1; i++) { 1897 int q = netdev_get_prio_tc_map(dev, i); 1898 1899 tc = &dev->tc_to_txq[q]; 1900 if (tc->offset + tc->count > txq) { 1901 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 1902 i, q); 1903 netdev_set_prio_tc_map(dev, i, 0); 1904 } 1905 } 1906 } 1907 1908 #ifdef CONFIG_XPS 1909 static DEFINE_MUTEX(xps_map_mutex); 1910 #define xmap_dereference(P) \ 1911 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 1912 1913 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, 1914 int cpu, u16 index) 1915 { 1916 struct xps_map *map = NULL; 1917 int pos; 1918 1919 if (dev_maps) 1920 map = xmap_dereference(dev_maps->cpu_map[cpu]); 1921 1922 for (pos = 0; map && pos < map->len; pos++) { 1923 if (map->queues[pos] == index) { 1924 if (map->len > 1) { 1925 map->queues[pos] = map->queues[--map->len]; 1926 } else { 1927 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); 1928 kfree_rcu(map, rcu); 1929 map = NULL; 1930 } 1931 break; 1932 } 1933 } 1934 1935 return map; 1936 } 1937 1938 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 1939 { 1940 struct xps_dev_maps *dev_maps; 1941 int cpu, i; 1942 bool active = false; 1943 1944 mutex_lock(&xps_map_mutex); 1945 dev_maps = xmap_dereference(dev->xps_maps); 1946 1947 if (!dev_maps) 1948 goto out_no_maps; 1949 1950 for_each_possible_cpu(cpu) { 1951 for (i = index; i < dev->num_tx_queues; i++) { 1952 if (!remove_xps_queue(dev_maps, cpu, i)) 1953 break; 1954 } 1955 if (i == dev->num_tx_queues) 1956 active = true; 1957 } 1958 1959 if (!active) { 1960 RCU_INIT_POINTER(dev->xps_maps, NULL); 1961 kfree_rcu(dev_maps, rcu); 1962 } 1963 1964 for (i = index; i < dev->num_tx_queues; i++) 1965 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), 1966 NUMA_NO_NODE); 1967 1968 out_no_maps: 1969 mutex_unlock(&xps_map_mutex); 1970 } 1971 1972 static struct xps_map *expand_xps_map(struct xps_map *map, 1973 int cpu, u16 index) 1974 { 1975 struct xps_map *new_map; 1976 int alloc_len = XPS_MIN_MAP_ALLOC; 1977 int i, pos; 1978 1979 for (pos = 0; map && pos < map->len; pos++) { 1980 if (map->queues[pos] != index) 1981 continue; 1982 return map; 1983 } 1984 1985 /* Need to add queue to this CPU's existing map */ 1986 if (map) { 1987 if (pos < map->alloc_len) 1988 return map; 1989 1990 alloc_len = map->alloc_len * 2; 1991 } 1992 1993 /* Need to allocate new map to store queue on this CPU's map */ 1994 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 1995 cpu_to_node(cpu)); 1996 if (!new_map) 1997 return NULL; 1998 1999 for (i = 0; i < pos; i++) 2000 new_map->queues[i] = map->queues[i]; 2001 new_map->alloc_len = alloc_len; 2002 new_map->len = pos; 2003 2004 return new_map; 2005 } 2006 2007 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2008 u16 index) 2009 { 2010 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2011 struct xps_map *map, *new_map; 2012 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); 2013 int cpu, numa_node_id = -2; 2014 bool active = false; 2015 2016 mutex_lock(&xps_map_mutex); 2017 2018 dev_maps = xmap_dereference(dev->xps_maps); 2019 2020 /* allocate memory for queue storage */ 2021 for_each_online_cpu(cpu) { 2022 if (!cpumask_test_cpu(cpu, mask)) 2023 continue; 2024 2025 if (!new_dev_maps) 2026 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2027 if (!new_dev_maps) { 2028 mutex_unlock(&xps_map_mutex); 2029 return -ENOMEM; 2030 } 2031 2032 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : 2033 NULL; 2034 2035 map = expand_xps_map(map, cpu, index); 2036 if (!map) 2037 goto error; 2038 2039 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); 2040 } 2041 2042 if (!new_dev_maps) 2043 goto out_no_new_maps; 2044 2045 for_each_possible_cpu(cpu) { 2046 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { 2047 /* add queue to CPU maps */ 2048 int pos = 0; 2049 2050 map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2051 while ((pos < map->len) && (map->queues[pos] != index)) 2052 pos++; 2053 2054 if (pos == map->len) 2055 map->queues[map->len++] = index; 2056 #ifdef CONFIG_NUMA 2057 if (numa_node_id == -2) 2058 numa_node_id = cpu_to_node(cpu); 2059 else if (numa_node_id != cpu_to_node(cpu)) 2060 numa_node_id = -1; 2061 #endif 2062 } else if (dev_maps) { 2063 /* fill in the new device map from the old device map */ 2064 map = xmap_dereference(dev_maps->cpu_map[cpu]); 2065 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); 2066 } 2067 2068 } 2069 2070 rcu_assign_pointer(dev->xps_maps, new_dev_maps); 2071 2072 /* Cleanup old maps */ 2073 if (dev_maps) { 2074 for_each_possible_cpu(cpu) { 2075 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2076 map = xmap_dereference(dev_maps->cpu_map[cpu]); 2077 if (map && map != new_map) 2078 kfree_rcu(map, rcu); 2079 } 2080 2081 kfree_rcu(dev_maps, rcu); 2082 } 2083 2084 dev_maps = new_dev_maps; 2085 active = true; 2086 2087 out_no_new_maps: 2088 /* update Tx queue numa node */ 2089 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2090 (numa_node_id >= 0) ? numa_node_id : 2091 NUMA_NO_NODE); 2092 2093 if (!dev_maps) 2094 goto out_no_maps; 2095 2096 /* removes queue from unused CPUs */ 2097 for_each_possible_cpu(cpu) { 2098 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) 2099 continue; 2100 2101 if (remove_xps_queue(dev_maps, cpu, index)) 2102 active = true; 2103 } 2104 2105 /* free map if not active */ 2106 if (!active) { 2107 RCU_INIT_POINTER(dev->xps_maps, NULL); 2108 kfree_rcu(dev_maps, rcu); 2109 } 2110 2111 out_no_maps: 2112 mutex_unlock(&xps_map_mutex); 2113 2114 return 0; 2115 error: 2116 /* remove any maps that we added */ 2117 for_each_possible_cpu(cpu) { 2118 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2119 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : 2120 NULL; 2121 if (new_map && new_map != map) 2122 kfree(new_map); 2123 } 2124 2125 mutex_unlock(&xps_map_mutex); 2126 2127 kfree(new_dev_maps); 2128 return -ENOMEM; 2129 } 2130 EXPORT_SYMBOL(netif_set_xps_queue); 2131 2132 #endif 2133 /* 2134 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2135 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. 2136 */ 2137 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2138 { 2139 int rc; 2140 2141 if (txq < 1 || txq > dev->num_tx_queues) 2142 return -EINVAL; 2143 2144 if (dev->reg_state == NETREG_REGISTERED || 2145 dev->reg_state == NETREG_UNREGISTERING) { 2146 ASSERT_RTNL(); 2147 2148 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2149 txq); 2150 if (rc) 2151 return rc; 2152 2153 if (dev->num_tc) 2154 netif_setup_tc(dev, txq); 2155 2156 if (txq < dev->real_num_tx_queues) { 2157 qdisc_reset_all_tx_gt(dev, txq); 2158 #ifdef CONFIG_XPS 2159 netif_reset_xps_queues_gt(dev, txq); 2160 #endif 2161 } 2162 } 2163 2164 dev->real_num_tx_queues = txq; 2165 return 0; 2166 } 2167 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2168 2169 #ifdef CONFIG_SYSFS 2170 /** 2171 * netif_set_real_num_rx_queues - set actual number of RX queues used 2172 * @dev: Network device 2173 * @rxq: Actual number of RX queues 2174 * 2175 * This must be called either with the rtnl_lock held or before 2176 * registration of the net device. Returns 0 on success, or a 2177 * negative error code. If called before registration, it always 2178 * succeeds. 2179 */ 2180 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2181 { 2182 int rc; 2183 2184 if (rxq < 1 || rxq > dev->num_rx_queues) 2185 return -EINVAL; 2186 2187 if (dev->reg_state == NETREG_REGISTERED) { 2188 ASSERT_RTNL(); 2189 2190 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2191 rxq); 2192 if (rc) 2193 return rc; 2194 } 2195 2196 dev->real_num_rx_queues = rxq; 2197 return 0; 2198 } 2199 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2200 #endif 2201 2202 /** 2203 * netif_get_num_default_rss_queues - default number of RSS queues 2204 * 2205 * This routine should set an upper limit on the number of RSS queues 2206 * used by default by multiqueue devices. 2207 */ 2208 int netif_get_num_default_rss_queues(void) 2209 { 2210 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2211 } 2212 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2213 2214 static inline void __netif_reschedule(struct Qdisc *q) 2215 { 2216 struct softnet_data *sd; 2217 unsigned long flags; 2218 2219 local_irq_save(flags); 2220 sd = this_cpu_ptr(&softnet_data); 2221 q->next_sched = NULL; 2222 *sd->output_queue_tailp = q; 2223 sd->output_queue_tailp = &q->next_sched; 2224 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2225 local_irq_restore(flags); 2226 } 2227 2228 void __netif_schedule(struct Qdisc *q) 2229 { 2230 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2231 __netif_reschedule(q); 2232 } 2233 EXPORT_SYMBOL(__netif_schedule); 2234 2235 struct dev_kfree_skb_cb { 2236 enum skb_free_reason reason; 2237 }; 2238 2239 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2240 { 2241 return (struct dev_kfree_skb_cb *)skb->cb; 2242 } 2243 2244 void netif_schedule_queue(struct netdev_queue *txq) 2245 { 2246 rcu_read_lock(); 2247 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2248 struct Qdisc *q = rcu_dereference(txq->qdisc); 2249 2250 __netif_schedule(q); 2251 } 2252 rcu_read_unlock(); 2253 } 2254 EXPORT_SYMBOL(netif_schedule_queue); 2255 2256 /** 2257 * netif_wake_subqueue - allow sending packets on subqueue 2258 * @dev: network device 2259 * @queue_index: sub queue index 2260 * 2261 * Resume individual transmit queue of a device with multiple transmit queues. 2262 */ 2263 void netif_wake_subqueue(struct net_device *dev, u16 queue_index) 2264 { 2265 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); 2266 2267 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { 2268 struct Qdisc *q; 2269 2270 rcu_read_lock(); 2271 q = rcu_dereference(txq->qdisc); 2272 __netif_schedule(q); 2273 rcu_read_unlock(); 2274 } 2275 } 2276 EXPORT_SYMBOL(netif_wake_subqueue); 2277 2278 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2279 { 2280 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2281 struct Qdisc *q; 2282 2283 rcu_read_lock(); 2284 q = rcu_dereference(dev_queue->qdisc); 2285 __netif_schedule(q); 2286 rcu_read_unlock(); 2287 } 2288 } 2289 EXPORT_SYMBOL(netif_tx_wake_queue); 2290 2291 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2292 { 2293 unsigned long flags; 2294 2295 if (likely(atomic_read(&skb->users) == 1)) { 2296 smp_rmb(); 2297 atomic_set(&skb->users, 0); 2298 } else if (likely(!atomic_dec_and_test(&skb->users))) { 2299 return; 2300 } 2301 get_kfree_skb_cb(skb)->reason = reason; 2302 local_irq_save(flags); 2303 skb->next = __this_cpu_read(softnet_data.completion_queue); 2304 __this_cpu_write(softnet_data.completion_queue, skb); 2305 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2306 local_irq_restore(flags); 2307 } 2308 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2309 2310 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2311 { 2312 if (in_irq() || irqs_disabled()) 2313 __dev_kfree_skb_irq(skb, reason); 2314 else 2315 dev_kfree_skb(skb); 2316 } 2317 EXPORT_SYMBOL(__dev_kfree_skb_any); 2318 2319 2320 /** 2321 * netif_device_detach - mark device as removed 2322 * @dev: network device 2323 * 2324 * Mark device as removed from system and therefore no longer available. 2325 */ 2326 void netif_device_detach(struct net_device *dev) 2327 { 2328 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2329 netif_running(dev)) { 2330 netif_tx_stop_all_queues(dev); 2331 } 2332 } 2333 EXPORT_SYMBOL(netif_device_detach); 2334 2335 /** 2336 * netif_device_attach - mark device as attached 2337 * @dev: network device 2338 * 2339 * Mark device as attached from system and restart if needed. 2340 */ 2341 void netif_device_attach(struct net_device *dev) 2342 { 2343 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2344 netif_running(dev)) { 2345 netif_tx_wake_all_queues(dev); 2346 __netdev_watchdog_up(dev); 2347 } 2348 } 2349 EXPORT_SYMBOL(netif_device_attach); 2350 2351 /* 2352 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 2353 * to be used as a distribution range. 2354 */ 2355 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 2356 unsigned int num_tx_queues) 2357 { 2358 u32 hash; 2359 u16 qoffset = 0; 2360 u16 qcount = num_tx_queues; 2361 2362 if (skb_rx_queue_recorded(skb)) { 2363 hash = skb_get_rx_queue(skb); 2364 while (unlikely(hash >= num_tx_queues)) 2365 hash -= num_tx_queues; 2366 return hash; 2367 } 2368 2369 if (dev->num_tc) { 2370 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 2371 qoffset = dev->tc_to_txq[tc].offset; 2372 qcount = dev->tc_to_txq[tc].count; 2373 } 2374 2375 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 2376 } 2377 EXPORT_SYMBOL(__skb_tx_hash); 2378 2379 static void skb_warn_bad_offload(const struct sk_buff *skb) 2380 { 2381 static const netdev_features_t null_features = 0; 2382 struct net_device *dev = skb->dev; 2383 const char *driver = ""; 2384 2385 if (!net_ratelimit()) 2386 return; 2387 2388 if (dev && dev->dev.parent) 2389 driver = dev_driver_string(dev->dev.parent); 2390 2391 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2392 "gso_type=%d ip_summed=%d\n", 2393 driver, dev ? &dev->features : &null_features, 2394 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2395 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2396 skb_shinfo(skb)->gso_type, skb->ip_summed); 2397 } 2398 2399 /* 2400 * Invalidate hardware checksum when packet is to be mangled, and 2401 * complete checksum manually on outgoing path. 2402 */ 2403 int skb_checksum_help(struct sk_buff *skb) 2404 { 2405 __wsum csum; 2406 int ret = 0, offset; 2407 2408 if (skb->ip_summed == CHECKSUM_COMPLETE) 2409 goto out_set_summed; 2410 2411 if (unlikely(skb_shinfo(skb)->gso_size)) { 2412 skb_warn_bad_offload(skb); 2413 return -EINVAL; 2414 } 2415 2416 /* Before computing a checksum, we should make sure no frag could 2417 * be modified by an external entity : checksum could be wrong. 2418 */ 2419 if (skb_has_shared_frag(skb)) { 2420 ret = __skb_linearize(skb); 2421 if (ret) 2422 goto out; 2423 } 2424 2425 offset = skb_checksum_start_offset(skb); 2426 BUG_ON(offset >= skb_headlen(skb)); 2427 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2428 2429 offset += skb->csum_offset; 2430 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2431 2432 if (skb_cloned(skb) && 2433 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2434 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2435 if (ret) 2436 goto out; 2437 } 2438 2439 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2440 out_set_summed: 2441 skb->ip_summed = CHECKSUM_NONE; 2442 out: 2443 return ret; 2444 } 2445 EXPORT_SYMBOL(skb_checksum_help); 2446 2447 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2448 { 2449 __be16 type = skb->protocol; 2450 2451 /* Tunnel gso handlers can set protocol to ethernet. */ 2452 if (type == htons(ETH_P_TEB)) { 2453 struct ethhdr *eth; 2454 2455 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2456 return 0; 2457 2458 eth = (struct ethhdr *)skb_mac_header(skb); 2459 type = eth->h_proto; 2460 } 2461 2462 return __vlan_get_protocol(skb, type, depth); 2463 } 2464 2465 /** 2466 * skb_mac_gso_segment - mac layer segmentation handler. 2467 * @skb: buffer to segment 2468 * @features: features for the output path (see dev->features) 2469 */ 2470 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 2471 netdev_features_t features) 2472 { 2473 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 2474 struct packet_offload *ptype; 2475 int vlan_depth = skb->mac_len; 2476 __be16 type = skb_network_protocol(skb, &vlan_depth); 2477 2478 if (unlikely(!type)) 2479 return ERR_PTR(-EINVAL); 2480 2481 __skb_pull(skb, vlan_depth); 2482 2483 rcu_read_lock(); 2484 list_for_each_entry_rcu(ptype, &offload_base, list) { 2485 if (ptype->type == type && ptype->callbacks.gso_segment) { 2486 segs = ptype->callbacks.gso_segment(skb, features); 2487 break; 2488 } 2489 } 2490 rcu_read_unlock(); 2491 2492 __skb_push(skb, skb->data - skb_mac_header(skb)); 2493 2494 return segs; 2495 } 2496 EXPORT_SYMBOL(skb_mac_gso_segment); 2497 2498 2499 /* openvswitch calls this on rx path, so we need a different check. 2500 */ 2501 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 2502 { 2503 if (tx_path) 2504 return skb->ip_summed != CHECKSUM_PARTIAL; 2505 else 2506 return skb->ip_summed == CHECKSUM_NONE; 2507 } 2508 2509 /** 2510 * __skb_gso_segment - Perform segmentation on skb. 2511 * @skb: buffer to segment 2512 * @features: features for the output path (see dev->features) 2513 * @tx_path: whether it is called in TX path 2514 * 2515 * This function segments the given skb and returns a list of segments. 2516 * 2517 * It may return NULL if the skb requires no segmentation. This is 2518 * only possible when GSO is used for verifying header integrity. 2519 */ 2520 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 2521 netdev_features_t features, bool tx_path) 2522 { 2523 if (unlikely(skb_needs_check(skb, tx_path))) { 2524 int err; 2525 2526 skb_warn_bad_offload(skb); 2527 2528 err = skb_cow_head(skb, 0); 2529 if (err < 0) 2530 return ERR_PTR(err); 2531 } 2532 2533 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 2534 SKB_GSO_CB(skb)->encap_level = 0; 2535 2536 skb_reset_mac_header(skb); 2537 skb_reset_mac_len(skb); 2538 2539 return skb_mac_gso_segment(skb, features); 2540 } 2541 EXPORT_SYMBOL(__skb_gso_segment); 2542 2543 /* Take action when hardware reception checksum errors are detected. */ 2544 #ifdef CONFIG_BUG 2545 void netdev_rx_csum_fault(struct net_device *dev) 2546 { 2547 if (net_ratelimit()) { 2548 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 2549 dump_stack(); 2550 } 2551 } 2552 EXPORT_SYMBOL(netdev_rx_csum_fault); 2553 #endif 2554 2555 /* Actually, we should eliminate this check as soon as we know, that: 2556 * 1. IOMMU is present and allows to map all the memory. 2557 * 2. No high memory really exists on this machine. 2558 */ 2559 2560 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 2561 { 2562 #ifdef CONFIG_HIGHMEM 2563 int i; 2564 if (!(dev->features & NETIF_F_HIGHDMA)) { 2565 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2566 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2567 if (PageHighMem(skb_frag_page(frag))) 2568 return 1; 2569 } 2570 } 2571 2572 if (PCI_DMA_BUS_IS_PHYS) { 2573 struct device *pdev = dev->dev.parent; 2574 2575 if (!pdev) 2576 return 0; 2577 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2578 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2579 dma_addr_t addr = page_to_phys(skb_frag_page(frag)); 2580 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) 2581 return 1; 2582 } 2583 } 2584 #endif 2585 return 0; 2586 } 2587 2588 /* If MPLS offload request, verify we are testing hardware MPLS features 2589 * instead of standard features for the netdev. 2590 */ 2591 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 2592 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2593 netdev_features_t features, 2594 __be16 type) 2595 { 2596 if (eth_p_mpls(type)) 2597 features &= skb->dev->mpls_features; 2598 2599 return features; 2600 } 2601 #else 2602 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2603 netdev_features_t features, 2604 __be16 type) 2605 { 2606 return features; 2607 } 2608 #endif 2609 2610 static netdev_features_t harmonize_features(struct sk_buff *skb, 2611 netdev_features_t features) 2612 { 2613 int tmp; 2614 __be16 type; 2615 2616 type = skb_network_protocol(skb, &tmp); 2617 features = net_mpls_features(skb, features, type); 2618 2619 if (skb->ip_summed != CHECKSUM_NONE && 2620 !can_checksum_protocol(features, type)) { 2621 features &= ~NETIF_F_ALL_CSUM; 2622 } else if (illegal_highdma(skb->dev, skb)) { 2623 features &= ~NETIF_F_SG; 2624 } 2625 2626 return features; 2627 } 2628 2629 netdev_features_t passthru_features_check(struct sk_buff *skb, 2630 struct net_device *dev, 2631 netdev_features_t features) 2632 { 2633 return features; 2634 } 2635 EXPORT_SYMBOL(passthru_features_check); 2636 2637 static netdev_features_t dflt_features_check(const struct sk_buff *skb, 2638 struct net_device *dev, 2639 netdev_features_t features) 2640 { 2641 return vlan_features_check(skb, features); 2642 } 2643 2644 netdev_features_t netif_skb_features(struct sk_buff *skb) 2645 { 2646 struct net_device *dev = skb->dev; 2647 netdev_features_t features = dev->features; 2648 u16 gso_segs = skb_shinfo(skb)->gso_segs; 2649 2650 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs) 2651 features &= ~NETIF_F_GSO_MASK; 2652 2653 /* If encapsulation offload request, verify we are testing 2654 * hardware encapsulation features instead of standard 2655 * features for the netdev 2656 */ 2657 if (skb->encapsulation) 2658 features &= dev->hw_enc_features; 2659 2660 if (skb_vlan_tagged(skb)) 2661 features = netdev_intersect_features(features, 2662 dev->vlan_features | 2663 NETIF_F_HW_VLAN_CTAG_TX | 2664 NETIF_F_HW_VLAN_STAG_TX); 2665 2666 if (dev->netdev_ops->ndo_features_check) 2667 features &= dev->netdev_ops->ndo_features_check(skb, dev, 2668 features); 2669 else 2670 features &= dflt_features_check(skb, dev, features); 2671 2672 return harmonize_features(skb, features); 2673 } 2674 EXPORT_SYMBOL(netif_skb_features); 2675 2676 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 2677 struct netdev_queue *txq, bool more) 2678 { 2679 unsigned int len; 2680 int rc; 2681 2682 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) 2683 dev_queue_xmit_nit(skb, dev); 2684 2685 len = skb->len; 2686 trace_net_dev_start_xmit(skb, dev); 2687 rc = netdev_start_xmit(skb, dev, txq, more); 2688 trace_net_dev_xmit(skb, rc, dev, len); 2689 2690 return rc; 2691 } 2692 2693 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 2694 struct netdev_queue *txq, int *ret) 2695 { 2696 struct sk_buff *skb = first; 2697 int rc = NETDEV_TX_OK; 2698 2699 while (skb) { 2700 struct sk_buff *next = skb->next; 2701 2702 skb->next = NULL; 2703 rc = xmit_one(skb, dev, txq, next != NULL); 2704 if (unlikely(!dev_xmit_complete(rc))) { 2705 skb->next = next; 2706 goto out; 2707 } 2708 2709 skb = next; 2710 if (netif_xmit_stopped(txq) && skb) { 2711 rc = NETDEV_TX_BUSY; 2712 break; 2713 } 2714 } 2715 2716 out: 2717 *ret = rc; 2718 return skb; 2719 } 2720 2721 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 2722 netdev_features_t features) 2723 { 2724 if (skb_vlan_tag_present(skb) && 2725 !vlan_hw_offload_capable(features, skb->vlan_proto)) 2726 skb = __vlan_hwaccel_push_inside(skb); 2727 return skb; 2728 } 2729 2730 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) 2731 { 2732 netdev_features_t features; 2733 2734 if (skb->next) 2735 return skb; 2736 2737 features = netif_skb_features(skb); 2738 skb = validate_xmit_vlan(skb, features); 2739 if (unlikely(!skb)) 2740 goto out_null; 2741 2742 if (netif_needs_gso(skb, features)) { 2743 struct sk_buff *segs; 2744 2745 segs = skb_gso_segment(skb, features); 2746 if (IS_ERR(segs)) { 2747 goto out_kfree_skb; 2748 } else if (segs) { 2749 consume_skb(skb); 2750 skb = segs; 2751 } 2752 } else { 2753 if (skb_needs_linearize(skb, features) && 2754 __skb_linearize(skb)) 2755 goto out_kfree_skb; 2756 2757 /* If packet is not checksummed and device does not 2758 * support checksumming for this protocol, complete 2759 * checksumming here. 2760 */ 2761 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2762 if (skb->encapsulation) 2763 skb_set_inner_transport_header(skb, 2764 skb_checksum_start_offset(skb)); 2765 else 2766 skb_set_transport_header(skb, 2767 skb_checksum_start_offset(skb)); 2768 if (!(features & NETIF_F_ALL_CSUM) && 2769 skb_checksum_help(skb)) 2770 goto out_kfree_skb; 2771 } 2772 } 2773 2774 return skb; 2775 2776 out_kfree_skb: 2777 kfree_skb(skb); 2778 out_null: 2779 return NULL; 2780 } 2781 2782 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) 2783 { 2784 struct sk_buff *next, *head = NULL, *tail; 2785 2786 for (; skb != NULL; skb = next) { 2787 next = skb->next; 2788 skb->next = NULL; 2789 2790 /* in case skb wont be segmented, point to itself */ 2791 skb->prev = skb; 2792 2793 skb = validate_xmit_skb(skb, dev); 2794 if (!skb) 2795 continue; 2796 2797 if (!head) 2798 head = skb; 2799 else 2800 tail->next = skb; 2801 /* If skb was segmented, skb->prev points to 2802 * the last segment. If not, it still contains skb. 2803 */ 2804 tail = skb->prev; 2805 } 2806 return head; 2807 } 2808 2809 static void qdisc_pkt_len_init(struct sk_buff *skb) 2810 { 2811 const struct skb_shared_info *shinfo = skb_shinfo(skb); 2812 2813 qdisc_skb_cb(skb)->pkt_len = skb->len; 2814 2815 /* To get more precise estimation of bytes sent on wire, 2816 * we add to pkt_len the headers size of all segments 2817 */ 2818 if (shinfo->gso_size) { 2819 unsigned int hdr_len; 2820 u16 gso_segs = shinfo->gso_segs; 2821 2822 /* mac layer + network layer */ 2823 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 2824 2825 /* + transport layer */ 2826 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 2827 hdr_len += tcp_hdrlen(skb); 2828 else 2829 hdr_len += sizeof(struct udphdr); 2830 2831 if (shinfo->gso_type & SKB_GSO_DODGY) 2832 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 2833 shinfo->gso_size); 2834 2835 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 2836 } 2837 } 2838 2839 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 2840 struct net_device *dev, 2841 struct netdev_queue *txq) 2842 { 2843 spinlock_t *root_lock = qdisc_lock(q); 2844 bool contended; 2845 int rc; 2846 2847 qdisc_pkt_len_init(skb); 2848 qdisc_calculate_pkt_len(skb, q); 2849 /* 2850 * Heuristic to force contended enqueues to serialize on a 2851 * separate lock before trying to get qdisc main lock. 2852 * This permits __QDISC___STATE_RUNNING owner to get the lock more 2853 * often and dequeue packets faster. 2854 */ 2855 contended = qdisc_is_running(q); 2856 if (unlikely(contended)) 2857 spin_lock(&q->busylock); 2858 2859 spin_lock(root_lock); 2860 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 2861 kfree_skb(skb); 2862 rc = NET_XMIT_DROP; 2863 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 2864 qdisc_run_begin(q)) { 2865 /* 2866 * This is a work-conserving queue; there are no old skbs 2867 * waiting to be sent out; and the qdisc is not running - 2868 * xmit the skb directly. 2869 */ 2870 2871 qdisc_bstats_update(q, skb); 2872 2873 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 2874 if (unlikely(contended)) { 2875 spin_unlock(&q->busylock); 2876 contended = false; 2877 } 2878 __qdisc_run(q); 2879 } else 2880 qdisc_run_end(q); 2881 2882 rc = NET_XMIT_SUCCESS; 2883 } else { 2884 rc = q->enqueue(skb, q) & NET_XMIT_MASK; 2885 if (qdisc_run_begin(q)) { 2886 if (unlikely(contended)) { 2887 spin_unlock(&q->busylock); 2888 contended = false; 2889 } 2890 __qdisc_run(q); 2891 } 2892 } 2893 spin_unlock(root_lock); 2894 if (unlikely(contended)) 2895 spin_unlock(&q->busylock); 2896 return rc; 2897 } 2898 2899 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 2900 static void skb_update_prio(struct sk_buff *skb) 2901 { 2902 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); 2903 2904 if (!skb->priority && skb->sk && map) { 2905 unsigned int prioidx = skb->sk->sk_cgrp_prioidx; 2906 2907 if (prioidx < map->priomap_len) 2908 skb->priority = map->priomap[prioidx]; 2909 } 2910 } 2911 #else 2912 #define skb_update_prio(skb) 2913 #endif 2914 2915 DEFINE_PER_CPU(int, xmit_recursion); 2916 EXPORT_SYMBOL(xmit_recursion); 2917 2918 #define RECURSION_LIMIT 10 2919 2920 /** 2921 * dev_loopback_xmit - loop back @skb 2922 * @skb: buffer to transmit 2923 */ 2924 int dev_loopback_xmit(struct sock *sk, struct sk_buff *skb) 2925 { 2926 skb_reset_mac_header(skb); 2927 __skb_pull(skb, skb_network_offset(skb)); 2928 skb->pkt_type = PACKET_LOOPBACK; 2929 skb->ip_summed = CHECKSUM_UNNECESSARY; 2930 WARN_ON(!skb_dst(skb)); 2931 skb_dst_force(skb); 2932 netif_rx_ni(skb); 2933 return 0; 2934 } 2935 EXPORT_SYMBOL(dev_loopback_xmit); 2936 2937 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) 2938 { 2939 #ifdef CONFIG_XPS 2940 struct xps_dev_maps *dev_maps; 2941 struct xps_map *map; 2942 int queue_index = -1; 2943 2944 rcu_read_lock(); 2945 dev_maps = rcu_dereference(dev->xps_maps); 2946 if (dev_maps) { 2947 map = rcu_dereference( 2948 dev_maps->cpu_map[skb->sender_cpu - 1]); 2949 if (map) { 2950 if (map->len == 1) 2951 queue_index = map->queues[0]; 2952 else 2953 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), 2954 map->len)]; 2955 if (unlikely(queue_index >= dev->real_num_tx_queues)) 2956 queue_index = -1; 2957 } 2958 } 2959 rcu_read_unlock(); 2960 2961 return queue_index; 2962 #else 2963 return -1; 2964 #endif 2965 } 2966 2967 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) 2968 { 2969 struct sock *sk = skb->sk; 2970 int queue_index = sk_tx_queue_get(sk); 2971 2972 if (queue_index < 0 || skb->ooo_okay || 2973 queue_index >= dev->real_num_tx_queues) { 2974 int new_index = get_xps_queue(dev, skb); 2975 if (new_index < 0) 2976 new_index = skb_tx_hash(dev, skb); 2977 2978 if (queue_index != new_index && sk && 2979 rcu_access_pointer(sk->sk_dst_cache)) 2980 sk_tx_queue_set(sk, new_index); 2981 2982 queue_index = new_index; 2983 } 2984 2985 return queue_index; 2986 } 2987 2988 struct netdev_queue *netdev_pick_tx(struct net_device *dev, 2989 struct sk_buff *skb, 2990 void *accel_priv) 2991 { 2992 int queue_index = 0; 2993 2994 #ifdef CONFIG_XPS 2995 if (skb->sender_cpu == 0) 2996 skb->sender_cpu = raw_smp_processor_id() + 1; 2997 #endif 2998 2999 if (dev->real_num_tx_queues != 1) { 3000 const struct net_device_ops *ops = dev->netdev_ops; 3001 if (ops->ndo_select_queue) 3002 queue_index = ops->ndo_select_queue(dev, skb, accel_priv, 3003 __netdev_pick_tx); 3004 else 3005 queue_index = __netdev_pick_tx(dev, skb); 3006 3007 if (!accel_priv) 3008 queue_index = netdev_cap_txqueue(dev, queue_index); 3009 } 3010 3011 skb_set_queue_mapping(skb, queue_index); 3012 return netdev_get_tx_queue(dev, queue_index); 3013 } 3014 3015 /** 3016 * __dev_queue_xmit - transmit a buffer 3017 * @skb: buffer to transmit 3018 * @accel_priv: private data used for L2 forwarding offload 3019 * 3020 * Queue a buffer for transmission to a network device. The caller must 3021 * have set the device and priority and built the buffer before calling 3022 * this function. The function can be called from an interrupt. 3023 * 3024 * A negative errno code is returned on a failure. A success does not 3025 * guarantee the frame will be transmitted as it may be dropped due 3026 * to congestion or traffic shaping. 3027 * 3028 * ----------------------------------------------------------------------------------- 3029 * I notice this method can also return errors from the queue disciplines, 3030 * including NET_XMIT_DROP, which is a positive value. So, errors can also 3031 * be positive. 3032 * 3033 * Regardless of the return value, the skb is consumed, so it is currently 3034 * difficult to retry a send to this method. (You can bump the ref count 3035 * before sending to hold a reference for retry if you are careful.) 3036 * 3037 * When calling this method, interrupts MUST be enabled. This is because 3038 * the BH enable code must have IRQs enabled so that it will not deadlock. 3039 * --BLG 3040 */ 3041 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) 3042 { 3043 struct net_device *dev = skb->dev; 3044 struct netdev_queue *txq; 3045 struct Qdisc *q; 3046 int rc = -ENOMEM; 3047 3048 skb_reset_mac_header(skb); 3049 3050 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 3051 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 3052 3053 /* Disable soft irqs for various locks below. Also 3054 * stops preemption for RCU. 3055 */ 3056 rcu_read_lock_bh(); 3057 3058 skb_update_prio(skb); 3059 3060 /* If device/qdisc don't need skb->dst, release it right now while 3061 * its hot in this cpu cache. 3062 */ 3063 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 3064 skb_dst_drop(skb); 3065 else 3066 skb_dst_force(skb); 3067 3068 txq = netdev_pick_tx(dev, skb, accel_priv); 3069 q = rcu_dereference_bh(txq->qdisc); 3070 3071 #ifdef CONFIG_NET_CLS_ACT 3072 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); 3073 #endif 3074 trace_net_dev_queue(skb); 3075 if (q->enqueue) { 3076 rc = __dev_xmit_skb(skb, q, dev, txq); 3077 goto out; 3078 } 3079 3080 /* The device has no queue. Common case for software devices: 3081 loopback, all the sorts of tunnels... 3082 3083 Really, it is unlikely that netif_tx_lock protection is necessary 3084 here. (f.e. loopback and IP tunnels are clean ignoring statistics 3085 counters.) 3086 However, it is possible, that they rely on protection 3087 made by us here. 3088 3089 Check this and shot the lock. It is not prone from deadlocks. 3090 Either shot noqueue qdisc, it is even simpler 8) 3091 */ 3092 if (dev->flags & IFF_UP) { 3093 int cpu = smp_processor_id(); /* ok because BHs are off */ 3094 3095 if (txq->xmit_lock_owner != cpu) { 3096 3097 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) 3098 goto recursion_alert; 3099 3100 skb = validate_xmit_skb(skb, dev); 3101 if (!skb) 3102 goto drop; 3103 3104 HARD_TX_LOCK(dev, txq, cpu); 3105 3106 if (!netif_xmit_stopped(txq)) { 3107 __this_cpu_inc(xmit_recursion); 3108 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 3109 __this_cpu_dec(xmit_recursion); 3110 if (dev_xmit_complete(rc)) { 3111 HARD_TX_UNLOCK(dev, txq); 3112 goto out; 3113 } 3114 } 3115 HARD_TX_UNLOCK(dev, txq); 3116 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 3117 dev->name); 3118 } else { 3119 /* Recursion is detected! It is possible, 3120 * unfortunately 3121 */ 3122 recursion_alert: 3123 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 3124 dev->name); 3125 } 3126 } 3127 3128 rc = -ENETDOWN; 3129 drop: 3130 rcu_read_unlock_bh(); 3131 3132 atomic_long_inc(&dev->tx_dropped); 3133 kfree_skb_list(skb); 3134 return rc; 3135 out: 3136 rcu_read_unlock_bh(); 3137 return rc; 3138 } 3139 3140 int dev_queue_xmit_sk(struct sock *sk, struct sk_buff *skb) 3141 { 3142 return __dev_queue_xmit(skb, NULL); 3143 } 3144 EXPORT_SYMBOL(dev_queue_xmit_sk); 3145 3146 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) 3147 { 3148 return __dev_queue_xmit(skb, accel_priv); 3149 } 3150 EXPORT_SYMBOL(dev_queue_xmit_accel); 3151 3152 3153 /*======================================================================= 3154 Receiver routines 3155 =======================================================================*/ 3156 3157 int netdev_max_backlog __read_mostly = 1000; 3158 EXPORT_SYMBOL(netdev_max_backlog); 3159 3160 int netdev_tstamp_prequeue __read_mostly = 1; 3161 int netdev_budget __read_mostly = 300; 3162 int weight_p __read_mostly = 64; /* old backlog weight */ 3163 3164 /* Called with irq disabled */ 3165 static inline void ____napi_schedule(struct softnet_data *sd, 3166 struct napi_struct *napi) 3167 { 3168 list_add_tail(&napi->poll_list, &sd->poll_list); 3169 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3170 } 3171 3172 #ifdef CONFIG_RPS 3173 3174 /* One global table that all flow-based protocols share. */ 3175 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3176 EXPORT_SYMBOL(rps_sock_flow_table); 3177 u32 rps_cpu_mask __read_mostly; 3178 EXPORT_SYMBOL(rps_cpu_mask); 3179 3180 struct static_key rps_needed __read_mostly; 3181 3182 static struct rps_dev_flow * 3183 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3184 struct rps_dev_flow *rflow, u16 next_cpu) 3185 { 3186 if (next_cpu < nr_cpu_ids) { 3187 #ifdef CONFIG_RFS_ACCEL 3188 struct netdev_rx_queue *rxqueue; 3189 struct rps_dev_flow_table *flow_table; 3190 struct rps_dev_flow *old_rflow; 3191 u32 flow_id; 3192 u16 rxq_index; 3193 int rc; 3194 3195 /* Should we steer this flow to a different hardware queue? */ 3196 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3197 !(dev->features & NETIF_F_NTUPLE)) 3198 goto out; 3199 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 3200 if (rxq_index == skb_get_rx_queue(skb)) 3201 goto out; 3202 3203 rxqueue = dev->_rx + rxq_index; 3204 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3205 if (!flow_table) 3206 goto out; 3207 flow_id = skb_get_hash(skb) & flow_table->mask; 3208 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 3209 rxq_index, flow_id); 3210 if (rc < 0) 3211 goto out; 3212 old_rflow = rflow; 3213 rflow = &flow_table->flows[flow_id]; 3214 rflow->filter = rc; 3215 if (old_rflow->filter == rflow->filter) 3216 old_rflow->filter = RPS_NO_FILTER; 3217 out: 3218 #endif 3219 rflow->last_qtail = 3220 per_cpu(softnet_data, next_cpu).input_queue_head; 3221 } 3222 3223 rflow->cpu = next_cpu; 3224 return rflow; 3225 } 3226 3227 /* 3228 * get_rps_cpu is called from netif_receive_skb and returns the target 3229 * CPU from the RPS map of the receiving queue for a given skb. 3230 * rcu_read_lock must be held on entry. 3231 */ 3232 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3233 struct rps_dev_flow **rflowp) 3234 { 3235 const struct rps_sock_flow_table *sock_flow_table; 3236 struct netdev_rx_queue *rxqueue = dev->_rx; 3237 struct rps_dev_flow_table *flow_table; 3238 struct rps_map *map; 3239 int cpu = -1; 3240 u32 tcpu; 3241 u32 hash; 3242 3243 if (skb_rx_queue_recorded(skb)) { 3244 u16 index = skb_get_rx_queue(skb); 3245 3246 if (unlikely(index >= dev->real_num_rx_queues)) { 3247 WARN_ONCE(dev->real_num_rx_queues > 1, 3248 "%s received packet on queue %u, but number " 3249 "of RX queues is %u\n", 3250 dev->name, index, dev->real_num_rx_queues); 3251 goto done; 3252 } 3253 rxqueue += index; 3254 } 3255 3256 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 3257 3258 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3259 map = rcu_dereference(rxqueue->rps_map); 3260 if (!flow_table && !map) 3261 goto done; 3262 3263 skb_reset_network_header(skb); 3264 hash = skb_get_hash(skb); 3265 if (!hash) 3266 goto done; 3267 3268 sock_flow_table = rcu_dereference(rps_sock_flow_table); 3269 if (flow_table && sock_flow_table) { 3270 struct rps_dev_flow *rflow; 3271 u32 next_cpu; 3272 u32 ident; 3273 3274 /* First check into global flow table if there is a match */ 3275 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 3276 if ((ident ^ hash) & ~rps_cpu_mask) 3277 goto try_rps; 3278 3279 next_cpu = ident & rps_cpu_mask; 3280 3281 /* OK, now we know there is a match, 3282 * we can look at the local (per receive queue) flow table 3283 */ 3284 rflow = &flow_table->flows[hash & flow_table->mask]; 3285 tcpu = rflow->cpu; 3286 3287 /* 3288 * If the desired CPU (where last recvmsg was done) is 3289 * different from current CPU (one in the rx-queue flow 3290 * table entry), switch if one of the following holds: 3291 * - Current CPU is unset (>= nr_cpu_ids). 3292 * - Current CPU is offline. 3293 * - The current CPU's queue tail has advanced beyond the 3294 * last packet that was enqueued using this table entry. 3295 * This guarantees that all previous packets for the flow 3296 * have been dequeued, thus preserving in order delivery. 3297 */ 3298 if (unlikely(tcpu != next_cpu) && 3299 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 3300 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 3301 rflow->last_qtail)) >= 0)) { 3302 tcpu = next_cpu; 3303 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 3304 } 3305 3306 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 3307 *rflowp = rflow; 3308 cpu = tcpu; 3309 goto done; 3310 } 3311 } 3312 3313 try_rps: 3314 3315 if (map) { 3316 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 3317 if (cpu_online(tcpu)) { 3318 cpu = tcpu; 3319 goto done; 3320 } 3321 } 3322 3323 done: 3324 return cpu; 3325 } 3326 3327 #ifdef CONFIG_RFS_ACCEL 3328 3329 /** 3330 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 3331 * @dev: Device on which the filter was set 3332 * @rxq_index: RX queue index 3333 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 3334 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 3335 * 3336 * Drivers that implement ndo_rx_flow_steer() should periodically call 3337 * this function for each installed filter and remove the filters for 3338 * which it returns %true. 3339 */ 3340 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 3341 u32 flow_id, u16 filter_id) 3342 { 3343 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 3344 struct rps_dev_flow_table *flow_table; 3345 struct rps_dev_flow *rflow; 3346 bool expire = true; 3347 unsigned int cpu; 3348 3349 rcu_read_lock(); 3350 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3351 if (flow_table && flow_id <= flow_table->mask) { 3352 rflow = &flow_table->flows[flow_id]; 3353 cpu = ACCESS_ONCE(rflow->cpu); 3354 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 3355 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 3356 rflow->last_qtail) < 3357 (int)(10 * flow_table->mask))) 3358 expire = false; 3359 } 3360 rcu_read_unlock(); 3361 return expire; 3362 } 3363 EXPORT_SYMBOL(rps_may_expire_flow); 3364 3365 #endif /* CONFIG_RFS_ACCEL */ 3366 3367 /* Called from hardirq (IPI) context */ 3368 static void rps_trigger_softirq(void *data) 3369 { 3370 struct softnet_data *sd = data; 3371 3372 ____napi_schedule(sd, &sd->backlog); 3373 sd->received_rps++; 3374 } 3375 3376 #endif /* CONFIG_RPS */ 3377 3378 /* 3379 * Check if this softnet_data structure is another cpu one 3380 * If yes, queue it to our IPI list and return 1 3381 * If no, return 0 3382 */ 3383 static int rps_ipi_queued(struct softnet_data *sd) 3384 { 3385 #ifdef CONFIG_RPS 3386 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 3387 3388 if (sd != mysd) { 3389 sd->rps_ipi_next = mysd->rps_ipi_list; 3390 mysd->rps_ipi_list = sd; 3391 3392 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3393 return 1; 3394 } 3395 #endif /* CONFIG_RPS */ 3396 return 0; 3397 } 3398 3399 #ifdef CONFIG_NET_FLOW_LIMIT 3400 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 3401 #endif 3402 3403 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 3404 { 3405 #ifdef CONFIG_NET_FLOW_LIMIT 3406 struct sd_flow_limit *fl; 3407 struct softnet_data *sd; 3408 unsigned int old_flow, new_flow; 3409 3410 if (qlen < (netdev_max_backlog >> 1)) 3411 return false; 3412 3413 sd = this_cpu_ptr(&softnet_data); 3414 3415 rcu_read_lock(); 3416 fl = rcu_dereference(sd->flow_limit); 3417 if (fl) { 3418 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 3419 old_flow = fl->history[fl->history_head]; 3420 fl->history[fl->history_head] = new_flow; 3421 3422 fl->history_head++; 3423 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 3424 3425 if (likely(fl->buckets[old_flow])) 3426 fl->buckets[old_flow]--; 3427 3428 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 3429 fl->count++; 3430 rcu_read_unlock(); 3431 return true; 3432 } 3433 } 3434 rcu_read_unlock(); 3435 #endif 3436 return false; 3437 } 3438 3439 /* 3440 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 3441 * queue (may be a remote CPU queue). 3442 */ 3443 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 3444 unsigned int *qtail) 3445 { 3446 struct softnet_data *sd; 3447 unsigned long flags; 3448 unsigned int qlen; 3449 3450 sd = &per_cpu(softnet_data, cpu); 3451 3452 local_irq_save(flags); 3453 3454 rps_lock(sd); 3455 qlen = skb_queue_len(&sd->input_pkt_queue); 3456 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 3457 if (qlen) { 3458 enqueue: 3459 __skb_queue_tail(&sd->input_pkt_queue, skb); 3460 input_queue_tail_incr_save(sd, qtail); 3461 rps_unlock(sd); 3462 local_irq_restore(flags); 3463 return NET_RX_SUCCESS; 3464 } 3465 3466 /* Schedule NAPI for backlog device 3467 * We can use non atomic operation since we own the queue lock 3468 */ 3469 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 3470 if (!rps_ipi_queued(sd)) 3471 ____napi_schedule(sd, &sd->backlog); 3472 } 3473 goto enqueue; 3474 } 3475 3476 sd->dropped++; 3477 rps_unlock(sd); 3478 3479 local_irq_restore(flags); 3480 3481 atomic_long_inc(&skb->dev->rx_dropped); 3482 kfree_skb(skb); 3483 return NET_RX_DROP; 3484 } 3485 3486 static int netif_rx_internal(struct sk_buff *skb) 3487 { 3488 int ret; 3489 3490 net_timestamp_check(netdev_tstamp_prequeue, skb); 3491 3492 trace_netif_rx(skb); 3493 #ifdef CONFIG_RPS 3494 if (static_key_false(&rps_needed)) { 3495 struct rps_dev_flow voidflow, *rflow = &voidflow; 3496 int cpu; 3497 3498 preempt_disable(); 3499 rcu_read_lock(); 3500 3501 cpu = get_rps_cpu(skb->dev, skb, &rflow); 3502 if (cpu < 0) 3503 cpu = smp_processor_id(); 3504 3505 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 3506 3507 rcu_read_unlock(); 3508 preempt_enable(); 3509 } else 3510 #endif 3511 { 3512 unsigned int qtail; 3513 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 3514 put_cpu(); 3515 } 3516 return ret; 3517 } 3518 3519 /** 3520 * netif_rx - post buffer to the network code 3521 * @skb: buffer to post 3522 * 3523 * This function receives a packet from a device driver and queues it for 3524 * the upper (protocol) levels to process. It always succeeds. The buffer 3525 * may be dropped during processing for congestion control or by the 3526 * protocol layers. 3527 * 3528 * return values: 3529 * NET_RX_SUCCESS (no congestion) 3530 * NET_RX_DROP (packet was dropped) 3531 * 3532 */ 3533 3534 int netif_rx(struct sk_buff *skb) 3535 { 3536 trace_netif_rx_entry(skb); 3537 3538 return netif_rx_internal(skb); 3539 } 3540 EXPORT_SYMBOL(netif_rx); 3541 3542 int netif_rx_ni(struct sk_buff *skb) 3543 { 3544 int err; 3545 3546 trace_netif_rx_ni_entry(skb); 3547 3548 preempt_disable(); 3549 err = netif_rx_internal(skb); 3550 if (local_softirq_pending()) 3551 do_softirq(); 3552 preempt_enable(); 3553 3554 return err; 3555 } 3556 EXPORT_SYMBOL(netif_rx_ni); 3557 3558 static void net_tx_action(struct softirq_action *h) 3559 { 3560 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 3561 3562 if (sd->completion_queue) { 3563 struct sk_buff *clist; 3564 3565 local_irq_disable(); 3566 clist = sd->completion_queue; 3567 sd->completion_queue = NULL; 3568 local_irq_enable(); 3569 3570 while (clist) { 3571 struct sk_buff *skb = clist; 3572 clist = clist->next; 3573 3574 WARN_ON(atomic_read(&skb->users)); 3575 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 3576 trace_consume_skb(skb); 3577 else 3578 trace_kfree_skb(skb, net_tx_action); 3579 __kfree_skb(skb); 3580 } 3581 } 3582 3583 if (sd->output_queue) { 3584 struct Qdisc *head; 3585 3586 local_irq_disable(); 3587 head = sd->output_queue; 3588 sd->output_queue = NULL; 3589 sd->output_queue_tailp = &sd->output_queue; 3590 local_irq_enable(); 3591 3592 while (head) { 3593 struct Qdisc *q = head; 3594 spinlock_t *root_lock; 3595 3596 head = head->next_sched; 3597 3598 root_lock = qdisc_lock(q); 3599 if (spin_trylock(root_lock)) { 3600 smp_mb__before_atomic(); 3601 clear_bit(__QDISC_STATE_SCHED, 3602 &q->state); 3603 qdisc_run(q); 3604 spin_unlock(root_lock); 3605 } else { 3606 if (!test_bit(__QDISC_STATE_DEACTIVATED, 3607 &q->state)) { 3608 __netif_reschedule(q); 3609 } else { 3610 smp_mb__before_atomic(); 3611 clear_bit(__QDISC_STATE_SCHED, 3612 &q->state); 3613 } 3614 } 3615 } 3616 } 3617 } 3618 3619 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ 3620 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) 3621 /* This hook is defined here for ATM LANE */ 3622 int (*br_fdb_test_addr_hook)(struct net_device *dev, 3623 unsigned char *addr) __read_mostly; 3624 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 3625 #endif 3626 3627 static inline struct sk_buff *handle_ing(struct sk_buff *skb, 3628 struct packet_type **pt_prev, 3629 int *ret, struct net_device *orig_dev) 3630 { 3631 #ifdef CONFIG_NET_CLS_ACT 3632 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); 3633 struct tcf_result cl_res; 3634 3635 /* If there's at least one ingress present somewhere (so 3636 * we get here via enabled static key), remaining devices 3637 * that are not configured with an ingress qdisc will bail 3638 * out here. 3639 */ 3640 if (!cl) 3641 return skb; 3642 if (*pt_prev) { 3643 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3644 *pt_prev = NULL; 3645 } 3646 3647 qdisc_skb_cb(skb)->pkt_len = skb->len; 3648 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); 3649 qdisc_bstats_update_cpu(cl->q, skb); 3650 3651 switch (tc_classify(skb, cl, &cl_res)) { 3652 case TC_ACT_OK: 3653 case TC_ACT_RECLASSIFY: 3654 skb->tc_index = TC_H_MIN(cl_res.classid); 3655 break; 3656 case TC_ACT_SHOT: 3657 qdisc_qstats_drop_cpu(cl->q); 3658 case TC_ACT_STOLEN: 3659 case TC_ACT_QUEUED: 3660 kfree_skb(skb); 3661 return NULL; 3662 default: 3663 break; 3664 } 3665 #endif /* CONFIG_NET_CLS_ACT */ 3666 return skb; 3667 } 3668 3669 /** 3670 * netdev_rx_handler_register - register receive handler 3671 * @dev: device to register a handler for 3672 * @rx_handler: receive handler to register 3673 * @rx_handler_data: data pointer that is used by rx handler 3674 * 3675 * Register a receive handler for a device. This handler will then be 3676 * called from __netif_receive_skb. A negative errno code is returned 3677 * on a failure. 3678 * 3679 * The caller must hold the rtnl_mutex. 3680 * 3681 * For a general description of rx_handler, see enum rx_handler_result. 3682 */ 3683 int netdev_rx_handler_register(struct net_device *dev, 3684 rx_handler_func_t *rx_handler, 3685 void *rx_handler_data) 3686 { 3687 ASSERT_RTNL(); 3688 3689 if (dev->rx_handler) 3690 return -EBUSY; 3691 3692 /* Note: rx_handler_data must be set before rx_handler */ 3693 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 3694 rcu_assign_pointer(dev->rx_handler, rx_handler); 3695 3696 return 0; 3697 } 3698 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 3699 3700 /** 3701 * netdev_rx_handler_unregister - unregister receive handler 3702 * @dev: device to unregister a handler from 3703 * 3704 * Unregister a receive handler from a device. 3705 * 3706 * The caller must hold the rtnl_mutex. 3707 */ 3708 void netdev_rx_handler_unregister(struct net_device *dev) 3709 { 3710 3711 ASSERT_RTNL(); 3712 RCU_INIT_POINTER(dev->rx_handler, NULL); 3713 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 3714 * section has a guarantee to see a non NULL rx_handler_data 3715 * as well. 3716 */ 3717 synchronize_net(); 3718 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 3719 } 3720 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 3721 3722 /* 3723 * Limit the use of PFMEMALLOC reserves to those protocols that implement 3724 * the special handling of PFMEMALLOC skbs. 3725 */ 3726 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 3727 { 3728 switch (skb->protocol) { 3729 case htons(ETH_P_ARP): 3730 case htons(ETH_P_IP): 3731 case htons(ETH_P_IPV6): 3732 case htons(ETH_P_8021Q): 3733 case htons(ETH_P_8021AD): 3734 return true; 3735 default: 3736 return false; 3737 } 3738 } 3739 3740 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 3741 int *ret, struct net_device *orig_dev) 3742 { 3743 #ifdef CONFIG_NETFILTER_INGRESS 3744 if (nf_hook_ingress_active(skb)) { 3745 if (*pt_prev) { 3746 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3747 *pt_prev = NULL; 3748 } 3749 3750 return nf_hook_ingress(skb); 3751 } 3752 #endif /* CONFIG_NETFILTER_INGRESS */ 3753 return 0; 3754 } 3755 3756 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) 3757 { 3758 struct packet_type *ptype, *pt_prev; 3759 rx_handler_func_t *rx_handler; 3760 struct net_device *orig_dev; 3761 bool deliver_exact = false; 3762 int ret = NET_RX_DROP; 3763 __be16 type; 3764 3765 net_timestamp_check(!netdev_tstamp_prequeue, skb); 3766 3767 trace_netif_receive_skb(skb); 3768 3769 orig_dev = skb->dev; 3770 3771 skb_reset_network_header(skb); 3772 if (!skb_transport_header_was_set(skb)) 3773 skb_reset_transport_header(skb); 3774 skb_reset_mac_len(skb); 3775 3776 pt_prev = NULL; 3777 3778 rcu_read_lock(); 3779 3780 another_round: 3781 skb->skb_iif = skb->dev->ifindex; 3782 3783 __this_cpu_inc(softnet_data.processed); 3784 3785 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 3786 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 3787 skb = skb_vlan_untag(skb); 3788 if (unlikely(!skb)) 3789 goto unlock; 3790 } 3791 3792 #ifdef CONFIG_NET_CLS_ACT 3793 if (skb->tc_verd & TC_NCLS) { 3794 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); 3795 goto ncls; 3796 } 3797 #endif 3798 3799 if (pfmemalloc) 3800 goto skip_taps; 3801 3802 list_for_each_entry_rcu(ptype, &ptype_all, list) { 3803 if (pt_prev) 3804 ret = deliver_skb(skb, pt_prev, orig_dev); 3805 pt_prev = ptype; 3806 } 3807 3808 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 3809 if (pt_prev) 3810 ret = deliver_skb(skb, pt_prev, orig_dev); 3811 pt_prev = ptype; 3812 } 3813 3814 skip_taps: 3815 #ifdef CONFIG_NET_INGRESS 3816 if (static_key_false(&ingress_needed)) { 3817 skb = handle_ing(skb, &pt_prev, &ret, orig_dev); 3818 if (!skb) 3819 goto unlock; 3820 3821 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 3822 goto unlock; 3823 } 3824 #endif 3825 #ifdef CONFIG_NET_CLS_ACT 3826 skb->tc_verd = 0; 3827 ncls: 3828 #endif 3829 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 3830 goto drop; 3831 3832 if (skb_vlan_tag_present(skb)) { 3833 if (pt_prev) { 3834 ret = deliver_skb(skb, pt_prev, orig_dev); 3835 pt_prev = NULL; 3836 } 3837 if (vlan_do_receive(&skb)) 3838 goto another_round; 3839 else if (unlikely(!skb)) 3840 goto unlock; 3841 } 3842 3843 rx_handler = rcu_dereference(skb->dev->rx_handler); 3844 if (rx_handler) { 3845 if (pt_prev) { 3846 ret = deliver_skb(skb, pt_prev, orig_dev); 3847 pt_prev = NULL; 3848 } 3849 switch (rx_handler(&skb)) { 3850 case RX_HANDLER_CONSUMED: 3851 ret = NET_RX_SUCCESS; 3852 goto unlock; 3853 case RX_HANDLER_ANOTHER: 3854 goto another_round; 3855 case RX_HANDLER_EXACT: 3856 deliver_exact = true; 3857 case RX_HANDLER_PASS: 3858 break; 3859 default: 3860 BUG(); 3861 } 3862 } 3863 3864 if (unlikely(skb_vlan_tag_present(skb))) { 3865 if (skb_vlan_tag_get_id(skb)) 3866 skb->pkt_type = PACKET_OTHERHOST; 3867 /* Note: we might in the future use prio bits 3868 * and set skb->priority like in vlan_do_receive() 3869 * For the time being, just ignore Priority Code Point 3870 */ 3871 skb->vlan_tci = 0; 3872 } 3873 3874 type = skb->protocol; 3875 3876 /* deliver only exact match when indicated */ 3877 if (likely(!deliver_exact)) { 3878 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 3879 &ptype_base[ntohs(type) & 3880 PTYPE_HASH_MASK]); 3881 } 3882 3883 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 3884 &orig_dev->ptype_specific); 3885 3886 if (unlikely(skb->dev != orig_dev)) { 3887 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 3888 &skb->dev->ptype_specific); 3889 } 3890 3891 if (pt_prev) { 3892 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 3893 goto drop; 3894 else 3895 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 3896 } else { 3897 drop: 3898 atomic_long_inc(&skb->dev->rx_dropped); 3899 kfree_skb(skb); 3900 /* Jamal, now you will not able to escape explaining 3901 * me how you were going to use this. :-) 3902 */ 3903 ret = NET_RX_DROP; 3904 } 3905 3906 unlock: 3907 rcu_read_unlock(); 3908 return ret; 3909 } 3910 3911 static int __netif_receive_skb(struct sk_buff *skb) 3912 { 3913 int ret; 3914 3915 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 3916 unsigned long pflags = current->flags; 3917 3918 /* 3919 * PFMEMALLOC skbs are special, they should 3920 * - be delivered to SOCK_MEMALLOC sockets only 3921 * - stay away from userspace 3922 * - have bounded memory usage 3923 * 3924 * Use PF_MEMALLOC as this saves us from propagating the allocation 3925 * context down to all allocation sites. 3926 */ 3927 current->flags |= PF_MEMALLOC; 3928 ret = __netif_receive_skb_core(skb, true); 3929 tsk_restore_flags(current, pflags, PF_MEMALLOC); 3930 } else 3931 ret = __netif_receive_skb_core(skb, false); 3932 3933 return ret; 3934 } 3935 3936 static int netif_receive_skb_internal(struct sk_buff *skb) 3937 { 3938 net_timestamp_check(netdev_tstamp_prequeue, skb); 3939 3940 if (skb_defer_rx_timestamp(skb)) 3941 return NET_RX_SUCCESS; 3942 3943 #ifdef CONFIG_RPS 3944 if (static_key_false(&rps_needed)) { 3945 struct rps_dev_flow voidflow, *rflow = &voidflow; 3946 int cpu, ret; 3947 3948 rcu_read_lock(); 3949 3950 cpu = get_rps_cpu(skb->dev, skb, &rflow); 3951 3952 if (cpu >= 0) { 3953 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 3954 rcu_read_unlock(); 3955 return ret; 3956 } 3957 rcu_read_unlock(); 3958 } 3959 #endif 3960 return __netif_receive_skb(skb); 3961 } 3962 3963 /** 3964 * netif_receive_skb - process receive buffer from network 3965 * @skb: buffer to process 3966 * 3967 * netif_receive_skb() is the main receive data processing function. 3968 * It always succeeds. The buffer may be dropped during processing 3969 * for congestion control or by the protocol layers. 3970 * 3971 * This function may only be called from softirq context and interrupts 3972 * should be enabled. 3973 * 3974 * Return values (usually ignored): 3975 * NET_RX_SUCCESS: no congestion 3976 * NET_RX_DROP: packet was dropped 3977 */ 3978 int netif_receive_skb_sk(struct sock *sk, struct sk_buff *skb) 3979 { 3980 trace_netif_receive_skb_entry(skb); 3981 3982 return netif_receive_skb_internal(skb); 3983 } 3984 EXPORT_SYMBOL(netif_receive_skb_sk); 3985 3986 /* Network device is going away, flush any packets still pending 3987 * Called with irqs disabled. 3988 */ 3989 static void flush_backlog(void *arg) 3990 { 3991 struct net_device *dev = arg; 3992 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 3993 struct sk_buff *skb, *tmp; 3994 3995 rps_lock(sd); 3996 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 3997 if (skb->dev == dev) { 3998 __skb_unlink(skb, &sd->input_pkt_queue); 3999 kfree_skb(skb); 4000 input_queue_head_incr(sd); 4001 } 4002 } 4003 rps_unlock(sd); 4004 4005 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 4006 if (skb->dev == dev) { 4007 __skb_unlink(skb, &sd->process_queue); 4008 kfree_skb(skb); 4009 input_queue_head_incr(sd); 4010 } 4011 } 4012 } 4013 4014 static int napi_gro_complete(struct sk_buff *skb) 4015 { 4016 struct packet_offload *ptype; 4017 __be16 type = skb->protocol; 4018 struct list_head *head = &offload_base; 4019 int err = -ENOENT; 4020 4021 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 4022 4023 if (NAPI_GRO_CB(skb)->count == 1) { 4024 skb_shinfo(skb)->gso_size = 0; 4025 goto out; 4026 } 4027 4028 rcu_read_lock(); 4029 list_for_each_entry_rcu(ptype, head, list) { 4030 if (ptype->type != type || !ptype->callbacks.gro_complete) 4031 continue; 4032 4033 err = ptype->callbacks.gro_complete(skb, 0); 4034 break; 4035 } 4036 rcu_read_unlock(); 4037 4038 if (err) { 4039 WARN_ON(&ptype->list == head); 4040 kfree_skb(skb); 4041 return NET_RX_SUCCESS; 4042 } 4043 4044 out: 4045 return netif_receive_skb_internal(skb); 4046 } 4047 4048 /* napi->gro_list contains packets ordered by age. 4049 * youngest packets at the head of it. 4050 * Complete skbs in reverse order to reduce latencies. 4051 */ 4052 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 4053 { 4054 struct sk_buff *skb, *prev = NULL; 4055 4056 /* scan list and build reverse chain */ 4057 for (skb = napi->gro_list; skb != NULL; skb = skb->next) { 4058 skb->prev = prev; 4059 prev = skb; 4060 } 4061 4062 for (skb = prev; skb; skb = prev) { 4063 skb->next = NULL; 4064 4065 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 4066 return; 4067 4068 prev = skb->prev; 4069 napi_gro_complete(skb); 4070 napi->gro_count--; 4071 } 4072 4073 napi->gro_list = NULL; 4074 } 4075 EXPORT_SYMBOL(napi_gro_flush); 4076 4077 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) 4078 { 4079 struct sk_buff *p; 4080 unsigned int maclen = skb->dev->hard_header_len; 4081 u32 hash = skb_get_hash_raw(skb); 4082 4083 for (p = napi->gro_list; p; p = p->next) { 4084 unsigned long diffs; 4085 4086 NAPI_GRO_CB(p)->flush = 0; 4087 4088 if (hash != skb_get_hash_raw(p)) { 4089 NAPI_GRO_CB(p)->same_flow = 0; 4090 continue; 4091 } 4092 4093 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 4094 diffs |= p->vlan_tci ^ skb->vlan_tci; 4095 if (maclen == ETH_HLEN) 4096 diffs |= compare_ether_header(skb_mac_header(p), 4097 skb_mac_header(skb)); 4098 else if (!diffs) 4099 diffs = memcmp(skb_mac_header(p), 4100 skb_mac_header(skb), 4101 maclen); 4102 NAPI_GRO_CB(p)->same_flow = !diffs; 4103 } 4104 } 4105 4106 static void skb_gro_reset_offset(struct sk_buff *skb) 4107 { 4108 const struct skb_shared_info *pinfo = skb_shinfo(skb); 4109 const skb_frag_t *frag0 = &pinfo->frags[0]; 4110 4111 NAPI_GRO_CB(skb)->data_offset = 0; 4112 NAPI_GRO_CB(skb)->frag0 = NULL; 4113 NAPI_GRO_CB(skb)->frag0_len = 0; 4114 4115 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 4116 pinfo->nr_frags && 4117 !PageHighMem(skb_frag_page(frag0))) { 4118 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 4119 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); 4120 } 4121 } 4122 4123 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 4124 { 4125 struct skb_shared_info *pinfo = skb_shinfo(skb); 4126 4127 BUG_ON(skb->end - skb->tail < grow); 4128 4129 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 4130 4131 skb->data_len -= grow; 4132 skb->tail += grow; 4133 4134 pinfo->frags[0].page_offset += grow; 4135 skb_frag_size_sub(&pinfo->frags[0], grow); 4136 4137 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 4138 skb_frag_unref(skb, 0); 4139 memmove(pinfo->frags, pinfo->frags + 1, 4140 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 4141 } 4142 } 4143 4144 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4145 { 4146 struct sk_buff **pp = NULL; 4147 struct packet_offload *ptype; 4148 __be16 type = skb->protocol; 4149 struct list_head *head = &offload_base; 4150 int same_flow; 4151 enum gro_result ret; 4152 int grow; 4153 4154 if (!(skb->dev->features & NETIF_F_GRO)) 4155 goto normal; 4156 4157 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) 4158 goto normal; 4159 4160 gro_list_prepare(napi, skb); 4161 4162 rcu_read_lock(); 4163 list_for_each_entry_rcu(ptype, head, list) { 4164 if (ptype->type != type || !ptype->callbacks.gro_receive) 4165 continue; 4166 4167 skb_set_network_header(skb, skb_gro_offset(skb)); 4168 skb_reset_mac_len(skb); 4169 NAPI_GRO_CB(skb)->same_flow = 0; 4170 NAPI_GRO_CB(skb)->flush = 0; 4171 NAPI_GRO_CB(skb)->free = 0; 4172 NAPI_GRO_CB(skb)->udp_mark = 0; 4173 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 4174 4175 /* Setup for GRO checksum validation */ 4176 switch (skb->ip_summed) { 4177 case CHECKSUM_COMPLETE: 4178 NAPI_GRO_CB(skb)->csum = skb->csum; 4179 NAPI_GRO_CB(skb)->csum_valid = 1; 4180 NAPI_GRO_CB(skb)->csum_cnt = 0; 4181 break; 4182 case CHECKSUM_UNNECESSARY: 4183 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 4184 NAPI_GRO_CB(skb)->csum_valid = 0; 4185 break; 4186 default: 4187 NAPI_GRO_CB(skb)->csum_cnt = 0; 4188 NAPI_GRO_CB(skb)->csum_valid = 0; 4189 } 4190 4191 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); 4192 break; 4193 } 4194 rcu_read_unlock(); 4195 4196 if (&ptype->list == head) 4197 goto normal; 4198 4199 same_flow = NAPI_GRO_CB(skb)->same_flow; 4200 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 4201 4202 if (pp) { 4203 struct sk_buff *nskb = *pp; 4204 4205 *pp = nskb->next; 4206 nskb->next = NULL; 4207 napi_gro_complete(nskb); 4208 napi->gro_count--; 4209 } 4210 4211 if (same_flow) 4212 goto ok; 4213 4214 if (NAPI_GRO_CB(skb)->flush) 4215 goto normal; 4216 4217 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { 4218 struct sk_buff *nskb = napi->gro_list; 4219 4220 /* locate the end of the list to select the 'oldest' flow */ 4221 while (nskb->next) { 4222 pp = &nskb->next; 4223 nskb = *pp; 4224 } 4225 *pp = NULL; 4226 nskb->next = NULL; 4227 napi_gro_complete(nskb); 4228 } else { 4229 napi->gro_count++; 4230 } 4231 NAPI_GRO_CB(skb)->count = 1; 4232 NAPI_GRO_CB(skb)->age = jiffies; 4233 NAPI_GRO_CB(skb)->last = skb; 4234 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 4235 skb->next = napi->gro_list; 4236 napi->gro_list = skb; 4237 ret = GRO_HELD; 4238 4239 pull: 4240 grow = skb_gro_offset(skb) - skb_headlen(skb); 4241 if (grow > 0) 4242 gro_pull_from_frag0(skb, grow); 4243 ok: 4244 return ret; 4245 4246 normal: 4247 ret = GRO_NORMAL; 4248 goto pull; 4249 } 4250 4251 struct packet_offload *gro_find_receive_by_type(__be16 type) 4252 { 4253 struct list_head *offload_head = &offload_base; 4254 struct packet_offload *ptype; 4255 4256 list_for_each_entry_rcu(ptype, offload_head, list) { 4257 if (ptype->type != type || !ptype->callbacks.gro_receive) 4258 continue; 4259 return ptype; 4260 } 4261 return NULL; 4262 } 4263 EXPORT_SYMBOL(gro_find_receive_by_type); 4264 4265 struct packet_offload *gro_find_complete_by_type(__be16 type) 4266 { 4267 struct list_head *offload_head = &offload_base; 4268 struct packet_offload *ptype; 4269 4270 list_for_each_entry_rcu(ptype, offload_head, list) { 4271 if (ptype->type != type || !ptype->callbacks.gro_complete) 4272 continue; 4273 return ptype; 4274 } 4275 return NULL; 4276 } 4277 EXPORT_SYMBOL(gro_find_complete_by_type); 4278 4279 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 4280 { 4281 switch (ret) { 4282 case GRO_NORMAL: 4283 if (netif_receive_skb_internal(skb)) 4284 ret = GRO_DROP; 4285 break; 4286 4287 case GRO_DROP: 4288 kfree_skb(skb); 4289 break; 4290 4291 case GRO_MERGED_FREE: 4292 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 4293 kmem_cache_free(skbuff_head_cache, skb); 4294 else 4295 __kfree_skb(skb); 4296 break; 4297 4298 case GRO_HELD: 4299 case GRO_MERGED: 4300 break; 4301 } 4302 4303 return ret; 4304 } 4305 4306 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4307 { 4308 trace_napi_gro_receive_entry(skb); 4309 4310 skb_gro_reset_offset(skb); 4311 4312 return napi_skb_finish(dev_gro_receive(napi, skb), skb); 4313 } 4314 EXPORT_SYMBOL(napi_gro_receive); 4315 4316 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 4317 { 4318 if (unlikely(skb->pfmemalloc)) { 4319 consume_skb(skb); 4320 return; 4321 } 4322 __skb_pull(skb, skb_headlen(skb)); 4323 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 4324 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 4325 skb->vlan_tci = 0; 4326 skb->dev = napi->dev; 4327 skb->skb_iif = 0; 4328 skb->encapsulation = 0; 4329 skb_shinfo(skb)->gso_type = 0; 4330 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 4331 4332 napi->skb = skb; 4333 } 4334 4335 struct sk_buff *napi_get_frags(struct napi_struct *napi) 4336 { 4337 struct sk_buff *skb = napi->skb; 4338 4339 if (!skb) { 4340 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 4341 napi->skb = skb; 4342 } 4343 return skb; 4344 } 4345 EXPORT_SYMBOL(napi_get_frags); 4346 4347 static gro_result_t napi_frags_finish(struct napi_struct *napi, 4348 struct sk_buff *skb, 4349 gro_result_t ret) 4350 { 4351 switch (ret) { 4352 case GRO_NORMAL: 4353 case GRO_HELD: 4354 __skb_push(skb, ETH_HLEN); 4355 skb->protocol = eth_type_trans(skb, skb->dev); 4356 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 4357 ret = GRO_DROP; 4358 break; 4359 4360 case GRO_DROP: 4361 case GRO_MERGED_FREE: 4362 napi_reuse_skb(napi, skb); 4363 break; 4364 4365 case GRO_MERGED: 4366 break; 4367 } 4368 4369 return ret; 4370 } 4371 4372 /* Upper GRO stack assumes network header starts at gro_offset=0 4373 * Drivers could call both napi_gro_frags() and napi_gro_receive() 4374 * We copy ethernet header into skb->data to have a common layout. 4375 */ 4376 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 4377 { 4378 struct sk_buff *skb = napi->skb; 4379 const struct ethhdr *eth; 4380 unsigned int hlen = sizeof(*eth); 4381 4382 napi->skb = NULL; 4383 4384 skb_reset_mac_header(skb); 4385 skb_gro_reset_offset(skb); 4386 4387 eth = skb_gro_header_fast(skb, 0); 4388 if (unlikely(skb_gro_header_hard(skb, hlen))) { 4389 eth = skb_gro_header_slow(skb, hlen, 0); 4390 if (unlikely(!eth)) { 4391 napi_reuse_skb(napi, skb); 4392 return NULL; 4393 } 4394 } else { 4395 gro_pull_from_frag0(skb, hlen); 4396 NAPI_GRO_CB(skb)->frag0 += hlen; 4397 NAPI_GRO_CB(skb)->frag0_len -= hlen; 4398 } 4399 __skb_pull(skb, hlen); 4400 4401 /* 4402 * This works because the only protocols we care about don't require 4403 * special handling. 4404 * We'll fix it up properly in napi_frags_finish() 4405 */ 4406 skb->protocol = eth->h_proto; 4407 4408 return skb; 4409 } 4410 4411 gro_result_t napi_gro_frags(struct napi_struct *napi) 4412 { 4413 struct sk_buff *skb = napi_frags_skb(napi); 4414 4415 if (!skb) 4416 return GRO_DROP; 4417 4418 trace_napi_gro_frags_entry(skb); 4419 4420 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 4421 } 4422 EXPORT_SYMBOL(napi_gro_frags); 4423 4424 /* Compute the checksum from gro_offset and return the folded value 4425 * after adding in any pseudo checksum. 4426 */ 4427 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 4428 { 4429 __wsum wsum; 4430 __sum16 sum; 4431 4432 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 4433 4434 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 4435 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 4436 if (likely(!sum)) { 4437 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 4438 !skb->csum_complete_sw) 4439 netdev_rx_csum_fault(skb->dev); 4440 } 4441 4442 NAPI_GRO_CB(skb)->csum = wsum; 4443 NAPI_GRO_CB(skb)->csum_valid = 1; 4444 4445 return sum; 4446 } 4447 EXPORT_SYMBOL(__skb_gro_checksum_complete); 4448 4449 /* 4450 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 4451 * Note: called with local irq disabled, but exits with local irq enabled. 4452 */ 4453 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 4454 { 4455 #ifdef CONFIG_RPS 4456 struct softnet_data *remsd = sd->rps_ipi_list; 4457 4458 if (remsd) { 4459 sd->rps_ipi_list = NULL; 4460 4461 local_irq_enable(); 4462 4463 /* Send pending IPI's to kick RPS processing on remote cpus. */ 4464 while (remsd) { 4465 struct softnet_data *next = remsd->rps_ipi_next; 4466 4467 if (cpu_online(remsd->cpu)) 4468 smp_call_function_single_async(remsd->cpu, 4469 &remsd->csd); 4470 remsd = next; 4471 } 4472 } else 4473 #endif 4474 local_irq_enable(); 4475 } 4476 4477 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 4478 { 4479 #ifdef CONFIG_RPS 4480 return sd->rps_ipi_list != NULL; 4481 #else 4482 return false; 4483 #endif 4484 } 4485 4486 static int process_backlog(struct napi_struct *napi, int quota) 4487 { 4488 int work = 0; 4489 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 4490 4491 /* Check if we have pending ipi, its better to send them now, 4492 * not waiting net_rx_action() end. 4493 */ 4494 if (sd_has_rps_ipi_waiting(sd)) { 4495 local_irq_disable(); 4496 net_rps_action_and_irq_enable(sd); 4497 } 4498 4499 napi->weight = weight_p; 4500 local_irq_disable(); 4501 while (1) { 4502 struct sk_buff *skb; 4503 4504 while ((skb = __skb_dequeue(&sd->process_queue))) { 4505 local_irq_enable(); 4506 __netif_receive_skb(skb); 4507 local_irq_disable(); 4508 input_queue_head_incr(sd); 4509 if (++work >= quota) { 4510 local_irq_enable(); 4511 return work; 4512 } 4513 } 4514 4515 rps_lock(sd); 4516 if (skb_queue_empty(&sd->input_pkt_queue)) { 4517 /* 4518 * Inline a custom version of __napi_complete(). 4519 * only current cpu owns and manipulates this napi, 4520 * and NAPI_STATE_SCHED is the only possible flag set 4521 * on backlog. 4522 * We can use a plain write instead of clear_bit(), 4523 * and we dont need an smp_mb() memory barrier. 4524 */ 4525 napi->state = 0; 4526 rps_unlock(sd); 4527 4528 break; 4529 } 4530 4531 skb_queue_splice_tail_init(&sd->input_pkt_queue, 4532 &sd->process_queue); 4533 rps_unlock(sd); 4534 } 4535 local_irq_enable(); 4536 4537 return work; 4538 } 4539 4540 /** 4541 * __napi_schedule - schedule for receive 4542 * @n: entry to schedule 4543 * 4544 * The entry's receive function will be scheduled to run. 4545 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 4546 */ 4547 void __napi_schedule(struct napi_struct *n) 4548 { 4549 unsigned long flags; 4550 4551 local_irq_save(flags); 4552 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4553 local_irq_restore(flags); 4554 } 4555 EXPORT_SYMBOL(__napi_schedule); 4556 4557 /** 4558 * __napi_schedule_irqoff - schedule for receive 4559 * @n: entry to schedule 4560 * 4561 * Variant of __napi_schedule() assuming hard irqs are masked 4562 */ 4563 void __napi_schedule_irqoff(struct napi_struct *n) 4564 { 4565 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4566 } 4567 EXPORT_SYMBOL(__napi_schedule_irqoff); 4568 4569 void __napi_complete(struct napi_struct *n) 4570 { 4571 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 4572 4573 list_del_init(&n->poll_list); 4574 smp_mb__before_atomic(); 4575 clear_bit(NAPI_STATE_SCHED, &n->state); 4576 } 4577 EXPORT_SYMBOL(__napi_complete); 4578 4579 void napi_complete_done(struct napi_struct *n, int work_done) 4580 { 4581 unsigned long flags; 4582 4583 /* 4584 * don't let napi dequeue from the cpu poll list 4585 * just in case its running on a different cpu 4586 */ 4587 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) 4588 return; 4589 4590 if (n->gro_list) { 4591 unsigned long timeout = 0; 4592 4593 if (work_done) 4594 timeout = n->dev->gro_flush_timeout; 4595 4596 if (timeout) 4597 hrtimer_start(&n->timer, ns_to_ktime(timeout), 4598 HRTIMER_MODE_REL_PINNED); 4599 else 4600 napi_gro_flush(n, false); 4601 } 4602 if (likely(list_empty(&n->poll_list))) { 4603 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); 4604 } else { 4605 /* If n->poll_list is not empty, we need to mask irqs */ 4606 local_irq_save(flags); 4607 __napi_complete(n); 4608 local_irq_restore(flags); 4609 } 4610 } 4611 EXPORT_SYMBOL(napi_complete_done); 4612 4613 /* must be called under rcu_read_lock(), as we dont take a reference */ 4614 struct napi_struct *napi_by_id(unsigned int napi_id) 4615 { 4616 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 4617 struct napi_struct *napi; 4618 4619 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 4620 if (napi->napi_id == napi_id) 4621 return napi; 4622 4623 return NULL; 4624 } 4625 EXPORT_SYMBOL_GPL(napi_by_id); 4626 4627 void napi_hash_add(struct napi_struct *napi) 4628 { 4629 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) { 4630 4631 spin_lock(&napi_hash_lock); 4632 4633 /* 0 is not a valid id, we also skip an id that is taken 4634 * we expect both events to be extremely rare 4635 */ 4636 napi->napi_id = 0; 4637 while (!napi->napi_id) { 4638 napi->napi_id = ++napi_gen_id; 4639 if (napi_by_id(napi->napi_id)) 4640 napi->napi_id = 0; 4641 } 4642 4643 hlist_add_head_rcu(&napi->napi_hash_node, 4644 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 4645 4646 spin_unlock(&napi_hash_lock); 4647 } 4648 } 4649 EXPORT_SYMBOL_GPL(napi_hash_add); 4650 4651 /* Warning : caller is responsible to make sure rcu grace period 4652 * is respected before freeing memory containing @napi 4653 */ 4654 void napi_hash_del(struct napi_struct *napi) 4655 { 4656 spin_lock(&napi_hash_lock); 4657 4658 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) 4659 hlist_del_rcu(&napi->napi_hash_node); 4660 4661 spin_unlock(&napi_hash_lock); 4662 } 4663 EXPORT_SYMBOL_GPL(napi_hash_del); 4664 4665 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 4666 { 4667 struct napi_struct *napi; 4668 4669 napi = container_of(timer, struct napi_struct, timer); 4670 if (napi->gro_list) 4671 napi_schedule(napi); 4672 4673 return HRTIMER_NORESTART; 4674 } 4675 4676 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 4677 int (*poll)(struct napi_struct *, int), int weight) 4678 { 4679 INIT_LIST_HEAD(&napi->poll_list); 4680 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 4681 napi->timer.function = napi_watchdog; 4682 napi->gro_count = 0; 4683 napi->gro_list = NULL; 4684 napi->skb = NULL; 4685 napi->poll = poll; 4686 if (weight > NAPI_POLL_WEIGHT) 4687 pr_err_once("netif_napi_add() called with weight %d on device %s\n", 4688 weight, dev->name); 4689 napi->weight = weight; 4690 list_add(&napi->dev_list, &dev->napi_list); 4691 napi->dev = dev; 4692 #ifdef CONFIG_NETPOLL 4693 spin_lock_init(&napi->poll_lock); 4694 napi->poll_owner = -1; 4695 #endif 4696 set_bit(NAPI_STATE_SCHED, &napi->state); 4697 } 4698 EXPORT_SYMBOL(netif_napi_add); 4699 4700 void napi_disable(struct napi_struct *n) 4701 { 4702 might_sleep(); 4703 set_bit(NAPI_STATE_DISABLE, &n->state); 4704 4705 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 4706 msleep(1); 4707 4708 hrtimer_cancel(&n->timer); 4709 4710 clear_bit(NAPI_STATE_DISABLE, &n->state); 4711 } 4712 EXPORT_SYMBOL(napi_disable); 4713 4714 void netif_napi_del(struct napi_struct *napi) 4715 { 4716 list_del_init(&napi->dev_list); 4717 napi_free_frags(napi); 4718 4719 kfree_skb_list(napi->gro_list); 4720 napi->gro_list = NULL; 4721 napi->gro_count = 0; 4722 } 4723 EXPORT_SYMBOL(netif_napi_del); 4724 4725 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 4726 { 4727 void *have; 4728 int work, weight; 4729 4730 list_del_init(&n->poll_list); 4731 4732 have = netpoll_poll_lock(n); 4733 4734 weight = n->weight; 4735 4736 /* This NAPI_STATE_SCHED test is for avoiding a race 4737 * with netpoll's poll_napi(). Only the entity which 4738 * obtains the lock and sees NAPI_STATE_SCHED set will 4739 * actually make the ->poll() call. Therefore we avoid 4740 * accidentally calling ->poll() when NAPI is not scheduled. 4741 */ 4742 work = 0; 4743 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 4744 work = n->poll(n, weight); 4745 trace_napi_poll(n); 4746 } 4747 4748 WARN_ON_ONCE(work > weight); 4749 4750 if (likely(work < weight)) 4751 goto out_unlock; 4752 4753 /* Drivers must not modify the NAPI state if they 4754 * consume the entire weight. In such cases this code 4755 * still "owns" the NAPI instance and therefore can 4756 * move the instance around on the list at-will. 4757 */ 4758 if (unlikely(napi_disable_pending(n))) { 4759 napi_complete(n); 4760 goto out_unlock; 4761 } 4762 4763 if (n->gro_list) { 4764 /* flush too old packets 4765 * If HZ < 1000, flush all packets. 4766 */ 4767 napi_gro_flush(n, HZ >= 1000); 4768 } 4769 4770 /* Some drivers may have called napi_schedule 4771 * prior to exhausting their budget. 4772 */ 4773 if (unlikely(!list_empty(&n->poll_list))) { 4774 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 4775 n->dev ? n->dev->name : "backlog"); 4776 goto out_unlock; 4777 } 4778 4779 list_add_tail(&n->poll_list, repoll); 4780 4781 out_unlock: 4782 netpoll_poll_unlock(have); 4783 4784 return work; 4785 } 4786 4787 static void net_rx_action(struct softirq_action *h) 4788 { 4789 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4790 unsigned long time_limit = jiffies + 2; 4791 int budget = netdev_budget; 4792 LIST_HEAD(list); 4793 LIST_HEAD(repoll); 4794 4795 local_irq_disable(); 4796 list_splice_init(&sd->poll_list, &list); 4797 local_irq_enable(); 4798 4799 for (;;) { 4800 struct napi_struct *n; 4801 4802 if (list_empty(&list)) { 4803 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 4804 return; 4805 break; 4806 } 4807 4808 n = list_first_entry(&list, struct napi_struct, poll_list); 4809 budget -= napi_poll(n, &repoll); 4810 4811 /* If softirq window is exhausted then punt. 4812 * Allow this to run for 2 jiffies since which will allow 4813 * an average latency of 1.5/HZ. 4814 */ 4815 if (unlikely(budget <= 0 || 4816 time_after_eq(jiffies, time_limit))) { 4817 sd->time_squeeze++; 4818 break; 4819 } 4820 } 4821 4822 local_irq_disable(); 4823 4824 list_splice_tail_init(&sd->poll_list, &list); 4825 list_splice_tail(&repoll, &list); 4826 list_splice(&list, &sd->poll_list); 4827 if (!list_empty(&sd->poll_list)) 4828 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4829 4830 net_rps_action_and_irq_enable(sd); 4831 } 4832 4833 struct netdev_adjacent { 4834 struct net_device *dev; 4835 4836 /* upper master flag, there can only be one master device per list */ 4837 bool master; 4838 4839 /* counter for the number of times this device was added to us */ 4840 u16 ref_nr; 4841 4842 /* private field for the users */ 4843 void *private; 4844 4845 struct list_head list; 4846 struct rcu_head rcu; 4847 }; 4848 4849 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev, 4850 struct net_device *adj_dev, 4851 struct list_head *adj_list) 4852 { 4853 struct netdev_adjacent *adj; 4854 4855 list_for_each_entry(adj, adj_list, list) { 4856 if (adj->dev == adj_dev) 4857 return adj; 4858 } 4859 return NULL; 4860 } 4861 4862 /** 4863 * netdev_has_upper_dev - Check if device is linked to an upper device 4864 * @dev: device 4865 * @upper_dev: upper device to check 4866 * 4867 * Find out if a device is linked to specified upper device and return true 4868 * in case it is. Note that this checks only immediate upper device, 4869 * not through a complete stack of devices. The caller must hold the RTNL lock. 4870 */ 4871 bool netdev_has_upper_dev(struct net_device *dev, 4872 struct net_device *upper_dev) 4873 { 4874 ASSERT_RTNL(); 4875 4876 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper); 4877 } 4878 EXPORT_SYMBOL(netdev_has_upper_dev); 4879 4880 /** 4881 * netdev_has_any_upper_dev - Check if device is linked to some device 4882 * @dev: device 4883 * 4884 * Find out if a device is linked to an upper device and return true in case 4885 * it is. The caller must hold the RTNL lock. 4886 */ 4887 static bool netdev_has_any_upper_dev(struct net_device *dev) 4888 { 4889 ASSERT_RTNL(); 4890 4891 return !list_empty(&dev->all_adj_list.upper); 4892 } 4893 4894 /** 4895 * netdev_master_upper_dev_get - Get master upper device 4896 * @dev: device 4897 * 4898 * Find a master upper device and return pointer to it or NULL in case 4899 * it's not there. The caller must hold the RTNL lock. 4900 */ 4901 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 4902 { 4903 struct netdev_adjacent *upper; 4904 4905 ASSERT_RTNL(); 4906 4907 if (list_empty(&dev->adj_list.upper)) 4908 return NULL; 4909 4910 upper = list_first_entry(&dev->adj_list.upper, 4911 struct netdev_adjacent, list); 4912 if (likely(upper->master)) 4913 return upper->dev; 4914 return NULL; 4915 } 4916 EXPORT_SYMBOL(netdev_master_upper_dev_get); 4917 4918 void *netdev_adjacent_get_private(struct list_head *adj_list) 4919 { 4920 struct netdev_adjacent *adj; 4921 4922 adj = list_entry(adj_list, struct netdev_adjacent, list); 4923 4924 return adj->private; 4925 } 4926 EXPORT_SYMBOL(netdev_adjacent_get_private); 4927 4928 /** 4929 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 4930 * @dev: device 4931 * @iter: list_head ** of the current position 4932 * 4933 * Gets the next device from the dev's upper list, starting from iter 4934 * position. The caller must hold RCU read lock. 4935 */ 4936 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 4937 struct list_head **iter) 4938 { 4939 struct netdev_adjacent *upper; 4940 4941 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 4942 4943 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 4944 4945 if (&upper->list == &dev->adj_list.upper) 4946 return NULL; 4947 4948 *iter = &upper->list; 4949 4950 return upper->dev; 4951 } 4952 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 4953 4954 /** 4955 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list 4956 * @dev: device 4957 * @iter: list_head ** of the current position 4958 * 4959 * Gets the next device from the dev's upper list, starting from iter 4960 * position. The caller must hold RCU read lock. 4961 */ 4962 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, 4963 struct list_head **iter) 4964 { 4965 struct netdev_adjacent *upper; 4966 4967 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 4968 4969 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 4970 4971 if (&upper->list == &dev->all_adj_list.upper) 4972 return NULL; 4973 4974 *iter = &upper->list; 4975 4976 return upper->dev; 4977 } 4978 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); 4979 4980 /** 4981 * netdev_lower_get_next_private - Get the next ->private from the 4982 * lower neighbour list 4983 * @dev: device 4984 * @iter: list_head ** of the current position 4985 * 4986 * Gets the next netdev_adjacent->private from the dev's lower neighbour 4987 * list, starting from iter position. The caller must hold either hold the 4988 * RTNL lock or its own locking that guarantees that the neighbour lower 4989 * list will remain unchainged. 4990 */ 4991 void *netdev_lower_get_next_private(struct net_device *dev, 4992 struct list_head **iter) 4993 { 4994 struct netdev_adjacent *lower; 4995 4996 lower = list_entry(*iter, struct netdev_adjacent, list); 4997 4998 if (&lower->list == &dev->adj_list.lower) 4999 return NULL; 5000 5001 *iter = lower->list.next; 5002 5003 return lower->private; 5004 } 5005 EXPORT_SYMBOL(netdev_lower_get_next_private); 5006 5007 /** 5008 * netdev_lower_get_next_private_rcu - Get the next ->private from the 5009 * lower neighbour list, RCU 5010 * variant 5011 * @dev: device 5012 * @iter: list_head ** of the current position 5013 * 5014 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5015 * list, starting from iter position. The caller must hold RCU read lock. 5016 */ 5017 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 5018 struct list_head **iter) 5019 { 5020 struct netdev_adjacent *lower; 5021 5022 WARN_ON_ONCE(!rcu_read_lock_held()); 5023 5024 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5025 5026 if (&lower->list == &dev->adj_list.lower) 5027 return NULL; 5028 5029 *iter = &lower->list; 5030 5031 return lower->private; 5032 } 5033 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 5034 5035 /** 5036 * netdev_lower_get_next - Get the next device from the lower neighbour 5037 * list 5038 * @dev: device 5039 * @iter: list_head ** of the current position 5040 * 5041 * Gets the next netdev_adjacent from the dev's lower neighbour 5042 * list, starting from iter position. The caller must hold RTNL lock or 5043 * its own locking that guarantees that the neighbour lower 5044 * list will remain unchainged. 5045 */ 5046 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 5047 { 5048 struct netdev_adjacent *lower; 5049 5050 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 5051 5052 if (&lower->list == &dev->adj_list.lower) 5053 return NULL; 5054 5055 *iter = &lower->list; 5056 5057 return lower->dev; 5058 } 5059 EXPORT_SYMBOL(netdev_lower_get_next); 5060 5061 /** 5062 * netdev_lower_get_first_private_rcu - Get the first ->private from the 5063 * lower neighbour list, RCU 5064 * variant 5065 * @dev: device 5066 * 5067 * Gets the first netdev_adjacent->private from the dev's lower neighbour 5068 * list. The caller must hold RCU read lock. 5069 */ 5070 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 5071 { 5072 struct netdev_adjacent *lower; 5073 5074 lower = list_first_or_null_rcu(&dev->adj_list.lower, 5075 struct netdev_adjacent, list); 5076 if (lower) 5077 return lower->private; 5078 return NULL; 5079 } 5080 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 5081 5082 /** 5083 * netdev_master_upper_dev_get_rcu - Get master upper device 5084 * @dev: device 5085 * 5086 * Find a master upper device and return pointer to it or NULL in case 5087 * it's not there. The caller must hold the RCU read lock. 5088 */ 5089 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 5090 { 5091 struct netdev_adjacent *upper; 5092 5093 upper = list_first_or_null_rcu(&dev->adj_list.upper, 5094 struct netdev_adjacent, list); 5095 if (upper && likely(upper->master)) 5096 return upper->dev; 5097 return NULL; 5098 } 5099 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 5100 5101 static int netdev_adjacent_sysfs_add(struct net_device *dev, 5102 struct net_device *adj_dev, 5103 struct list_head *dev_list) 5104 { 5105 char linkname[IFNAMSIZ+7]; 5106 sprintf(linkname, dev_list == &dev->adj_list.upper ? 5107 "upper_%s" : "lower_%s", adj_dev->name); 5108 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 5109 linkname); 5110 } 5111 static void netdev_adjacent_sysfs_del(struct net_device *dev, 5112 char *name, 5113 struct list_head *dev_list) 5114 { 5115 char linkname[IFNAMSIZ+7]; 5116 sprintf(linkname, dev_list == &dev->adj_list.upper ? 5117 "upper_%s" : "lower_%s", name); 5118 sysfs_remove_link(&(dev->dev.kobj), linkname); 5119 } 5120 5121 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 5122 struct net_device *adj_dev, 5123 struct list_head *dev_list) 5124 { 5125 return (dev_list == &dev->adj_list.upper || 5126 dev_list == &dev->adj_list.lower) && 5127 net_eq(dev_net(dev), dev_net(adj_dev)); 5128 } 5129 5130 static int __netdev_adjacent_dev_insert(struct net_device *dev, 5131 struct net_device *adj_dev, 5132 struct list_head *dev_list, 5133 void *private, bool master) 5134 { 5135 struct netdev_adjacent *adj; 5136 int ret; 5137 5138 adj = __netdev_find_adj(dev, adj_dev, dev_list); 5139 5140 if (adj) { 5141 adj->ref_nr++; 5142 return 0; 5143 } 5144 5145 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 5146 if (!adj) 5147 return -ENOMEM; 5148 5149 adj->dev = adj_dev; 5150 adj->master = master; 5151 adj->ref_nr = 1; 5152 adj->private = private; 5153 dev_hold(adj_dev); 5154 5155 pr_debug("dev_hold for %s, because of link added from %s to %s\n", 5156 adj_dev->name, dev->name, adj_dev->name); 5157 5158 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 5159 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 5160 if (ret) 5161 goto free_adj; 5162 } 5163 5164 /* Ensure that master link is always the first item in list. */ 5165 if (master) { 5166 ret = sysfs_create_link(&(dev->dev.kobj), 5167 &(adj_dev->dev.kobj), "master"); 5168 if (ret) 5169 goto remove_symlinks; 5170 5171 list_add_rcu(&adj->list, dev_list); 5172 } else { 5173 list_add_tail_rcu(&adj->list, dev_list); 5174 } 5175 5176 return 0; 5177 5178 remove_symlinks: 5179 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 5180 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 5181 free_adj: 5182 kfree(adj); 5183 dev_put(adj_dev); 5184 5185 return ret; 5186 } 5187 5188 static void __netdev_adjacent_dev_remove(struct net_device *dev, 5189 struct net_device *adj_dev, 5190 struct list_head *dev_list) 5191 { 5192 struct netdev_adjacent *adj; 5193 5194 adj = __netdev_find_adj(dev, adj_dev, dev_list); 5195 5196 if (!adj) { 5197 pr_err("tried to remove device %s from %s\n", 5198 dev->name, adj_dev->name); 5199 BUG(); 5200 } 5201 5202 if (adj->ref_nr > 1) { 5203 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, 5204 adj->ref_nr-1); 5205 adj->ref_nr--; 5206 return; 5207 } 5208 5209 if (adj->master) 5210 sysfs_remove_link(&(dev->dev.kobj), "master"); 5211 5212 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 5213 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 5214 5215 list_del_rcu(&adj->list); 5216 pr_debug("dev_put for %s, because link removed from %s to %s\n", 5217 adj_dev->name, dev->name, adj_dev->name); 5218 dev_put(adj_dev); 5219 kfree_rcu(adj, rcu); 5220 } 5221 5222 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 5223 struct net_device *upper_dev, 5224 struct list_head *up_list, 5225 struct list_head *down_list, 5226 void *private, bool master) 5227 { 5228 int ret; 5229 5230 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, 5231 master); 5232 if (ret) 5233 return ret; 5234 5235 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, 5236 false); 5237 if (ret) { 5238 __netdev_adjacent_dev_remove(dev, upper_dev, up_list); 5239 return ret; 5240 } 5241 5242 return 0; 5243 } 5244 5245 static int __netdev_adjacent_dev_link(struct net_device *dev, 5246 struct net_device *upper_dev) 5247 { 5248 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 5249 &dev->all_adj_list.upper, 5250 &upper_dev->all_adj_list.lower, 5251 NULL, false); 5252 } 5253 5254 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 5255 struct net_device *upper_dev, 5256 struct list_head *up_list, 5257 struct list_head *down_list) 5258 { 5259 __netdev_adjacent_dev_remove(dev, upper_dev, up_list); 5260 __netdev_adjacent_dev_remove(upper_dev, dev, down_list); 5261 } 5262 5263 static void __netdev_adjacent_dev_unlink(struct net_device *dev, 5264 struct net_device *upper_dev) 5265 { 5266 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 5267 &dev->all_adj_list.upper, 5268 &upper_dev->all_adj_list.lower); 5269 } 5270 5271 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 5272 struct net_device *upper_dev, 5273 void *private, bool master) 5274 { 5275 int ret = __netdev_adjacent_dev_link(dev, upper_dev); 5276 5277 if (ret) 5278 return ret; 5279 5280 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 5281 &dev->adj_list.upper, 5282 &upper_dev->adj_list.lower, 5283 private, master); 5284 if (ret) { 5285 __netdev_adjacent_dev_unlink(dev, upper_dev); 5286 return ret; 5287 } 5288 5289 return 0; 5290 } 5291 5292 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 5293 struct net_device *upper_dev) 5294 { 5295 __netdev_adjacent_dev_unlink(dev, upper_dev); 5296 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 5297 &dev->adj_list.upper, 5298 &upper_dev->adj_list.lower); 5299 } 5300 5301 static int __netdev_upper_dev_link(struct net_device *dev, 5302 struct net_device *upper_dev, bool master, 5303 void *private) 5304 { 5305 struct netdev_adjacent *i, *j, *to_i, *to_j; 5306 int ret = 0; 5307 5308 ASSERT_RTNL(); 5309 5310 if (dev == upper_dev) 5311 return -EBUSY; 5312 5313 /* To prevent loops, check if dev is not upper device to upper_dev. */ 5314 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper)) 5315 return -EBUSY; 5316 5317 if (__netdev_find_adj(dev, upper_dev, &dev->adj_list.upper)) 5318 return -EEXIST; 5319 5320 if (master && netdev_master_upper_dev_get(dev)) 5321 return -EBUSY; 5322 5323 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private, 5324 master); 5325 if (ret) 5326 return ret; 5327 5328 /* Now that we linked these devs, make all the upper_dev's 5329 * all_adj_list.upper visible to every dev's all_adj_list.lower an 5330 * versa, and don't forget the devices itself. All of these 5331 * links are non-neighbours. 5332 */ 5333 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5334 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { 5335 pr_debug("Interlinking %s with %s, non-neighbour\n", 5336 i->dev->name, j->dev->name); 5337 ret = __netdev_adjacent_dev_link(i->dev, j->dev); 5338 if (ret) 5339 goto rollback_mesh; 5340 } 5341 } 5342 5343 /* add dev to every upper_dev's upper device */ 5344 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { 5345 pr_debug("linking %s's upper device %s with %s\n", 5346 upper_dev->name, i->dev->name, dev->name); 5347 ret = __netdev_adjacent_dev_link(dev, i->dev); 5348 if (ret) 5349 goto rollback_upper_mesh; 5350 } 5351 5352 /* add upper_dev to every dev's lower device */ 5353 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5354 pr_debug("linking %s's lower device %s with %s\n", dev->name, 5355 i->dev->name, upper_dev->name); 5356 ret = __netdev_adjacent_dev_link(i->dev, upper_dev); 5357 if (ret) 5358 goto rollback_lower_mesh; 5359 } 5360 5361 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); 5362 return 0; 5363 5364 rollback_lower_mesh: 5365 to_i = i; 5366 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5367 if (i == to_i) 5368 break; 5369 __netdev_adjacent_dev_unlink(i->dev, upper_dev); 5370 } 5371 5372 i = NULL; 5373 5374 rollback_upper_mesh: 5375 to_i = i; 5376 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { 5377 if (i == to_i) 5378 break; 5379 __netdev_adjacent_dev_unlink(dev, i->dev); 5380 } 5381 5382 i = j = NULL; 5383 5384 rollback_mesh: 5385 to_i = i; 5386 to_j = j; 5387 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5388 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { 5389 if (i == to_i && j == to_j) 5390 break; 5391 __netdev_adjacent_dev_unlink(i->dev, j->dev); 5392 } 5393 if (i == to_i) 5394 break; 5395 } 5396 5397 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5398 5399 return ret; 5400 } 5401 5402 /** 5403 * netdev_upper_dev_link - Add a link to the upper device 5404 * @dev: device 5405 * @upper_dev: new upper device 5406 * 5407 * Adds a link to device which is upper to this one. The caller must hold 5408 * the RTNL lock. On a failure a negative errno code is returned. 5409 * On success the reference counts are adjusted and the function 5410 * returns zero. 5411 */ 5412 int netdev_upper_dev_link(struct net_device *dev, 5413 struct net_device *upper_dev) 5414 { 5415 return __netdev_upper_dev_link(dev, upper_dev, false, NULL); 5416 } 5417 EXPORT_SYMBOL(netdev_upper_dev_link); 5418 5419 /** 5420 * netdev_master_upper_dev_link - Add a master link to the upper device 5421 * @dev: device 5422 * @upper_dev: new upper device 5423 * 5424 * Adds a link to device which is upper to this one. In this case, only 5425 * one master upper device can be linked, although other non-master devices 5426 * might be linked as well. The caller must hold the RTNL lock. 5427 * On a failure a negative errno code is returned. On success the reference 5428 * counts are adjusted and the function returns zero. 5429 */ 5430 int netdev_master_upper_dev_link(struct net_device *dev, 5431 struct net_device *upper_dev) 5432 { 5433 return __netdev_upper_dev_link(dev, upper_dev, true, NULL); 5434 } 5435 EXPORT_SYMBOL(netdev_master_upper_dev_link); 5436 5437 int netdev_master_upper_dev_link_private(struct net_device *dev, 5438 struct net_device *upper_dev, 5439 void *private) 5440 { 5441 return __netdev_upper_dev_link(dev, upper_dev, true, private); 5442 } 5443 EXPORT_SYMBOL(netdev_master_upper_dev_link_private); 5444 5445 /** 5446 * netdev_upper_dev_unlink - Removes a link to upper device 5447 * @dev: device 5448 * @upper_dev: new upper device 5449 * 5450 * Removes a link to device which is upper to this one. The caller must hold 5451 * the RTNL lock. 5452 */ 5453 void netdev_upper_dev_unlink(struct net_device *dev, 5454 struct net_device *upper_dev) 5455 { 5456 struct netdev_adjacent *i, *j; 5457 ASSERT_RTNL(); 5458 5459 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5460 5461 /* Here is the tricky part. We must remove all dev's lower 5462 * devices from all upper_dev's upper devices and vice 5463 * versa, to maintain the graph relationship. 5464 */ 5465 list_for_each_entry(i, &dev->all_adj_list.lower, list) 5466 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) 5467 __netdev_adjacent_dev_unlink(i->dev, j->dev); 5468 5469 /* remove also the devices itself from lower/upper device 5470 * list 5471 */ 5472 list_for_each_entry(i, &dev->all_adj_list.lower, list) 5473 __netdev_adjacent_dev_unlink(i->dev, upper_dev); 5474 5475 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) 5476 __netdev_adjacent_dev_unlink(dev, i->dev); 5477 5478 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev); 5479 } 5480 EXPORT_SYMBOL(netdev_upper_dev_unlink); 5481 5482 /** 5483 * netdev_bonding_info_change - Dispatch event about slave change 5484 * @dev: device 5485 * @bonding_info: info to dispatch 5486 * 5487 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 5488 * The caller must hold the RTNL lock. 5489 */ 5490 void netdev_bonding_info_change(struct net_device *dev, 5491 struct netdev_bonding_info *bonding_info) 5492 { 5493 struct netdev_notifier_bonding_info info; 5494 5495 memcpy(&info.bonding_info, bonding_info, 5496 sizeof(struct netdev_bonding_info)); 5497 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, 5498 &info.info); 5499 } 5500 EXPORT_SYMBOL(netdev_bonding_info_change); 5501 5502 static void netdev_adjacent_add_links(struct net_device *dev) 5503 { 5504 struct netdev_adjacent *iter; 5505 5506 struct net *net = dev_net(dev); 5507 5508 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5509 if (!net_eq(net,dev_net(iter->dev))) 5510 continue; 5511 netdev_adjacent_sysfs_add(iter->dev, dev, 5512 &iter->dev->adj_list.lower); 5513 netdev_adjacent_sysfs_add(dev, iter->dev, 5514 &dev->adj_list.upper); 5515 } 5516 5517 list_for_each_entry(iter, &dev->adj_list.lower, list) { 5518 if (!net_eq(net,dev_net(iter->dev))) 5519 continue; 5520 netdev_adjacent_sysfs_add(iter->dev, dev, 5521 &iter->dev->adj_list.upper); 5522 netdev_adjacent_sysfs_add(dev, iter->dev, 5523 &dev->adj_list.lower); 5524 } 5525 } 5526 5527 static void netdev_adjacent_del_links(struct net_device *dev) 5528 { 5529 struct netdev_adjacent *iter; 5530 5531 struct net *net = dev_net(dev); 5532 5533 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5534 if (!net_eq(net,dev_net(iter->dev))) 5535 continue; 5536 netdev_adjacent_sysfs_del(iter->dev, dev->name, 5537 &iter->dev->adj_list.lower); 5538 netdev_adjacent_sysfs_del(dev, iter->dev->name, 5539 &dev->adj_list.upper); 5540 } 5541 5542 list_for_each_entry(iter, &dev->adj_list.lower, list) { 5543 if (!net_eq(net,dev_net(iter->dev))) 5544 continue; 5545 netdev_adjacent_sysfs_del(iter->dev, dev->name, 5546 &iter->dev->adj_list.upper); 5547 netdev_adjacent_sysfs_del(dev, iter->dev->name, 5548 &dev->adj_list.lower); 5549 } 5550 } 5551 5552 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 5553 { 5554 struct netdev_adjacent *iter; 5555 5556 struct net *net = dev_net(dev); 5557 5558 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5559 if (!net_eq(net,dev_net(iter->dev))) 5560 continue; 5561 netdev_adjacent_sysfs_del(iter->dev, oldname, 5562 &iter->dev->adj_list.lower); 5563 netdev_adjacent_sysfs_add(iter->dev, dev, 5564 &iter->dev->adj_list.lower); 5565 } 5566 5567 list_for_each_entry(iter, &dev->adj_list.lower, list) { 5568 if (!net_eq(net,dev_net(iter->dev))) 5569 continue; 5570 netdev_adjacent_sysfs_del(iter->dev, oldname, 5571 &iter->dev->adj_list.upper); 5572 netdev_adjacent_sysfs_add(iter->dev, dev, 5573 &iter->dev->adj_list.upper); 5574 } 5575 } 5576 5577 void *netdev_lower_dev_get_private(struct net_device *dev, 5578 struct net_device *lower_dev) 5579 { 5580 struct netdev_adjacent *lower; 5581 5582 if (!lower_dev) 5583 return NULL; 5584 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower); 5585 if (!lower) 5586 return NULL; 5587 5588 return lower->private; 5589 } 5590 EXPORT_SYMBOL(netdev_lower_dev_get_private); 5591 5592 5593 int dev_get_nest_level(struct net_device *dev, 5594 bool (*type_check)(struct net_device *dev)) 5595 { 5596 struct net_device *lower = NULL; 5597 struct list_head *iter; 5598 int max_nest = -1; 5599 int nest; 5600 5601 ASSERT_RTNL(); 5602 5603 netdev_for_each_lower_dev(dev, lower, iter) { 5604 nest = dev_get_nest_level(lower, type_check); 5605 if (max_nest < nest) 5606 max_nest = nest; 5607 } 5608 5609 if (type_check(dev)) 5610 max_nest++; 5611 5612 return max_nest; 5613 } 5614 EXPORT_SYMBOL(dev_get_nest_level); 5615 5616 static void dev_change_rx_flags(struct net_device *dev, int flags) 5617 { 5618 const struct net_device_ops *ops = dev->netdev_ops; 5619 5620 if (ops->ndo_change_rx_flags) 5621 ops->ndo_change_rx_flags(dev, flags); 5622 } 5623 5624 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 5625 { 5626 unsigned int old_flags = dev->flags; 5627 kuid_t uid; 5628 kgid_t gid; 5629 5630 ASSERT_RTNL(); 5631 5632 dev->flags |= IFF_PROMISC; 5633 dev->promiscuity += inc; 5634 if (dev->promiscuity == 0) { 5635 /* 5636 * Avoid overflow. 5637 * If inc causes overflow, untouch promisc and return error. 5638 */ 5639 if (inc < 0) 5640 dev->flags &= ~IFF_PROMISC; 5641 else { 5642 dev->promiscuity -= inc; 5643 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 5644 dev->name); 5645 return -EOVERFLOW; 5646 } 5647 } 5648 if (dev->flags != old_flags) { 5649 pr_info("device %s %s promiscuous mode\n", 5650 dev->name, 5651 dev->flags & IFF_PROMISC ? "entered" : "left"); 5652 if (audit_enabled) { 5653 current_uid_gid(&uid, &gid); 5654 audit_log(current->audit_context, GFP_ATOMIC, 5655 AUDIT_ANOM_PROMISCUOUS, 5656 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 5657 dev->name, (dev->flags & IFF_PROMISC), 5658 (old_flags & IFF_PROMISC), 5659 from_kuid(&init_user_ns, audit_get_loginuid(current)), 5660 from_kuid(&init_user_ns, uid), 5661 from_kgid(&init_user_ns, gid), 5662 audit_get_sessionid(current)); 5663 } 5664 5665 dev_change_rx_flags(dev, IFF_PROMISC); 5666 } 5667 if (notify) 5668 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 5669 return 0; 5670 } 5671 5672 /** 5673 * dev_set_promiscuity - update promiscuity count on a device 5674 * @dev: device 5675 * @inc: modifier 5676 * 5677 * Add or remove promiscuity from a device. While the count in the device 5678 * remains above zero the interface remains promiscuous. Once it hits zero 5679 * the device reverts back to normal filtering operation. A negative inc 5680 * value is used to drop promiscuity on the device. 5681 * Return 0 if successful or a negative errno code on error. 5682 */ 5683 int dev_set_promiscuity(struct net_device *dev, int inc) 5684 { 5685 unsigned int old_flags = dev->flags; 5686 int err; 5687 5688 err = __dev_set_promiscuity(dev, inc, true); 5689 if (err < 0) 5690 return err; 5691 if (dev->flags != old_flags) 5692 dev_set_rx_mode(dev); 5693 return err; 5694 } 5695 EXPORT_SYMBOL(dev_set_promiscuity); 5696 5697 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 5698 { 5699 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 5700 5701 ASSERT_RTNL(); 5702 5703 dev->flags |= IFF_ALLMULTI; 5704 dev->allmulti += inc; 5705 if (dev->allmulti == 0) { 5706 /* 5707 * Avoid overflow. 5708 * If inc causes overflow, untouch allmulti and return error. 5709 */ 5710 if (inc < 0) 5711 dev->flags &= ~IFF_ALLMULTI; 5712 else { 5713 dev->allmulti -= inc; 5714 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 5715 dev->name); 5716 return -EOVERFLOW; 5717 } 5718 } 5719 if (dev->flags ^ old_flags) { 5720 dev_change_rx_flags(dev, IFF_ALLMULTI); 5721 dev_set_rx_mode(dev); 5722 if (notify) 5723 __dev_notify_flags(dev, old_flags, 5724 dev->gflags ^ old_gflags); 5725 } 5726 return 0; 5727 } 5728 5729 /** 5730 * dev_set_allmulti - update allmulti count on a device 5731 * @dev: device 5732 * @inc: modifier 5733 * 5734 * Add or remove reception of all multicast frames to a device. While the 5735 * count in the device remains above zero the interface remains listening 5736 * to all interfaces. Once it hits zero the device reverts back to normal 5737 * filtering operation. A negative @inc value is used to drop the counter 5738 * when releasing a resource needing all multicasts. 5739 * Return 0 if successful or a negative errno code on error. 5740 */ 5741 5742 int dev_set_allmulti(struct net_device *dev, int inc) 5743 { 5744 return __dev_set_allmulti(dev, inc, true); 5745 } 5746 EXPORT_SYMBOL(dev_set_allmulti); 5747 5748 /* 5749 * Upload unicast and multicast address lists to device and 5750 * configure RX filtering. When the device doesn't support unicast 5751 * filtering it is put in promiscuous mode while unicast addresses 5752 * are present. 5753 */ 5754 void __dev_set_rx_mode(struct net_device *dev) 5755 { 5756 const struct net_device_ops *ops = dev->netdev_ops; 5757 5758 /* dev_open will call this function so the list will stay sane. */ 5759 if (!(dev->flags&IFF_UP)) 5760 return; 5761 5762 if (!netif_device_present(dev)) 5763 return; 5764 5765 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 5766 /* Unicast addresses changes may only happen under the rtnl, 5767 * therefore calling __dev_set_promiscuity here is safe. 5768 */ 5769 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 5770 __dev_set_promiscuity(dev, 1, false); 5771 dev->uc_promisc = true; 5772 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 5773 __dev_set_promiscuity(dev, -1, false); 5774 dev->uc_promisc = false; 5775 } 5776 } 5777 5778 if (ops->ndo_set_rx_mode) 5779 ops->ndo_set_rx_mode(dev); 5780 } 5781 5782 void dev_set_rx_mode(struct net_device *dev) 5783 { 5784 netif_addr_lock_bh(dev); 5785 __dev_set_rx_mode(dev); 5786 netif_addr_unlock_bh(dev); 5787 } 5788 5789 /** 5790 * dev_get_flags - get flags reported to userspace 5791 * @dev: device 5792 * 5793 * Get the combination of flag bits exported through APIs to userspace. 5794 */ 5795 unsigned int dev_get_flags(const struct net_device *dev) 5796 { 5797 unsigned int flags; 5798 5799 flags = (dev->flags & ~(IFF_PROMISC | 5800 IFF_ALLMULTI | 5801 IFF_RUNNING | 5802 IFF_LOWER_UP | 5803 IFF_DORMANT)) | 5804 (dev->gflags & (IFF_PROMISC | 5805 IFF_ALLMULTI)); 5806 5807 if (netif_running(dev)) { 5808 if (netif_oper_up(dev)) 5809 flags |= IFF_RUNNING; 5810 if (netif_carrier_ok(dev)) 5811 flags |= IFF_LOWER_UP; 5812 if (netif_dormant(dev)) 5813 flags |= IFF_DORMANT; 5814 } 5815 5816 return flags; 5817 } 5818 EXPORT_SYMBOL(dev_get_flags); 5819 5820 int __dev_change_flags(struct net_device *dev, unsigned int flags) 5821 { 5822 unsigned int old_flags = dev->flags; 5823 int ret; 5824 5825 ASSERT_RTNL(); 5826 5827 /* 5828 * Set the flags on our device. 5829 */ 5830 5831 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 5832 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 5833 IFF_AUTOMEDIA)) | 5834 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 5835 IFF_ALLMULTI)); 5836 5837 /* 5838 * Load in the correct multicast list now the flags have changed. 5839 */ 5840 5841 if ((old_flags ^ flags) & IFF_MULTICAST) 5842 dev_change_rx_flags(dev, IFF_MULTICAST); 5843 5844 dev_set_rx_mode(dev); 5845 5846 /* 5847 * Have we downed the interface. We handle IFF_UP ourselves 5848 * according to user attempts to set it, rather than blindly 5849 * setting it. 5850 */ 5851 5852 ret = 0; 5853 if ((old_flags ^ flags) & IFF_UP) 5854 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); 5855 5856 if ((flags ^ dev->gflags) & IFF_PROMISC) { 5857 int inc = (flags & IFF_PROMISC) ? 1 : -1; 5858 unsigned int old_flags = dev->flags; 5859 5860 dev->gflags ^= IFF_PROMISC; 5861 5862 if (__dev_set_promiscuity(dev, inc, false) >= 0) 5863 if (dev->flags != old_flags) 5864 dev_set_rx_mode(dev); 5865 } 5866 5867 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 5868 is important. Some (broken) drivers set IFF_PROMISC, when 5869 IFF_ALLMULTI is requested not asking us and not reporting. 5870 */ 5871 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 5872 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 5873 5874 dev->gflags ^= IFF_ALLMULTI; 5875 __dev_set_allmulti(dev, inc, false); 5876 } 5877 5878 return ret; 5879 } 5880 5881 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 5882 unsigned int gchanges) 5883 { 5884 unsigned int changes = dev->flags ^ old_flags; 5885 5886 if (gchanges) 5887 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 5888 5889 if (changes & IFF_UP) { 5890 if (dev->flags & IFF_UP) 5891 call_netdevice_notifiers(NETDEV_UP, dev); 5892 else 5893 call_netdevice_notifiers(NETDEV_DOWN, dev); 5894 } 5895 5896 if (dev->flags & IFF_UP && 5897 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 5898 struct netdev_notifier_change_info change_info; 5899 5900 change_info.flags_changed = changes; 5901 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 5902 &change_info.info); 5903 } 5904 } 5905 5906 /** 5907 * dev_change_flags - change device settings 5908 * @dev: device 5909 * @flags: device state flags 5910 * 5911 * Change settings on device based state flags. The flags are 5912 * in the userspace exported format. 5913 */ 5914 int dev_change_flags(struct net_device *dev, unsigned int flags) 5915 { 5916 int ret; 5917 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 5918 5919 ret = __dev_change_flags(dev, flags); 5920 if (ret < 0) 5921 return ret; 5922 5923 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 5924 __dev_notify_flags(dev, old_flags, changes); 5925 return ret; 5926 } 5927 EXPORT_SYMBOL(dev_change_flags); 5928 5929 static int __dev_set_mtu(struct net_device *dev, int new_mtu) 5930 { 5931 const struct net_device_ops *ops = dev->netdev_ops; 5932 5933 if (ops->ndo_change_mtu) 5934 return ops->ndo_change_mtu(dev, new_mtu); 5935 5936 dev->mtu = new_mtu; 5937 return 0; 5938 } 5939 5940 /** 5941 * dev_set_mtu - Change maximum transfer unit 5942 * @dev: device 5943 * @new_mtu: new transfer unit 5944 * 5945 * Change the maximum transfer size of the network device. 5946 */ 5947 int dev_set_mtu(struct net_device *dev, int new_mtu) 5948 { 5949 int err, orig_mtu; 5950 5951 if (new_mtu == dev->mtu) 5952 return 0; 5953 5954 /* MTU must be positive. */ 5955 if (new_mtu < 0) 5956 return -EINVAL; 5957 5958 if (!netif_device_present(dev)) 5959 return -ENODEV; 5960 5961 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 5962 err = notifier_to_errno(err); 5963 if (err) 5964 return err; 5965 5966 orig_mtu = dev->mtu; 5967 err = __dev_set_mtu(dev, new_mtu); 5968 5969 if (!err) { 5970 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 5971 err = notifier_to_errno(err); 5972 if (err) { 5973 /* setting mtu back and notifying everyone again, 5974 * so that they have a chance to revert changes. 5975 */ 5976 __dev_set_mtu(dev, orig_mtu); 5977 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 5978 } 5979 } 5980 return err; 5981 } 5982 EXPORT_SYMBOL(dev_set_mtu); 5983 5984 /** 5985 * dev_set_group - Change group this device belongs to 5986 * @dev: device 5987 * @new_group: group this device should belong to 5988 */ 5989 void dev_set_group(struct net_device *dev, int new_group) 5990 { 5991 dev->group = new_group; 5992 } 5993 EXPORT_SYMBOL(dev_set_group); 5994 5995 /** 5996 * dev_set_mac_address - Change Media Access Control Address 5997 * @dev: device 5998 * @sa: new address 5999 * 6000 * Change the hardware (MAC) address of the device 6001 */ 6002 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 6003 { 6004 const struct net_device_ops *ops = dev->netdev_ops; 6005 int err; 6006 6007 if (!ops->ndo_set_mac_address) 6008 return -EOPNOTSUPP; 6009 if (sa->sa_family != dev->type) 6010 return -EINVAL; 6011 if (!netif_device_present(dev)) 6012 return -ENODEV; 6013 err = ops->ndo_set_mac_address(dev, sa); 6014 if (err) 6015 return err; 6016 dev->addr_assign_type = NET_ADDR_SET; 6017 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 6018 add_device_randomness(dev->dev_addr, dev->addr_len); 6019 return 0; 6020 } 6021 EXPORT_SYMBOL(dev_set_mac_address); 6022 6023 /** 6024 * dev_change_carrier - Change device carrier 6025 * @dev: device 6026 * @new_carrier: new value 6027 * 6028 * Change device carrier 6029 */ 6030 int dev_change_carrier(struct net_device *dev, bool new_carrier) 6031 { 6032 const struct net_device_ops *ops = dev->netdev_ops; 6033 6034 if (!ops->ndo_change_carrier) 6035 return -EOPNOTSUPP; 6036 if (!netif_device_present(dev)) 6037 return -ENODEV; 6038 return ops->ndo_change_carrier(dev, new_carrier); 6039 } 6040 EXPORT_SYMBOL(dev_change_carrier); 6041 6042 /** 6043 * dev_get_phys_port_id - Get device physical port ID 6044 * @dev: device 6045 * @ppid: port ID 6046 * 6047 * Get device physical port ID 6048 */ 6049 int dev_get_phys_port_id(struct net_device *dev, 6050 struct netdev_phys_item_id *ppid) 6051 { 6052 const struct net_device_ops *ops = dev->netdev_ops; 6053 6054 if (!ops->ndo_get_phys_port_id) 6055 return -EOPNOTSUPP; 6056 return ops->ndo_get_phys_port_id(dev, ppid); 6057 } 6058 EXPORT_SYMBOL(dev_get_phys_port_id); 6059 6060 /** 6061 * dev_get_phys_port_name - Get device physical port name 6062 * @dev: device 6063 * @name: port name 6064 * 6065 * Get device physical port name 6066 */ 6067 int dev_get_phys_port_name(struct net_device *dev, 6068 char *name, size_t len) 6069 { 6070 const struct net_device_ops *ops = dev->netdev_ops; 6071 6072 if (!ops->ndo_get_phys_port_name) 6073 return -EOPNOTSUPP; 6074 return ops->ndo_get_phys_port_name(dev, name, len); 6075 } 6076 EXPORT_SYMBOL(dev_get_phys_port_name); 6077 6078 /** 6079 * dev_new_index - allocate an ifindex 6080 * @net: the applicable net namespace 6081 * 6082 * Returns a suitable unique value for a new device interface 6083 * number. The caller must hold the rtnl semaphore or the 6084 * dev_base_lock to be sure it remains unique. 6085 */ 6086 static int dev_new_index(struct net *net) 6087 { 6088 int ifindex = net->ifindex; 6089 for (;;) { 6090 if (++ifindex <= 0) 6091 ifindex = 1; 6092 if (!__dev_get_by_index(net, ifindex)) 6093 return net->ifindex = ifindex; 6094 } 6095 } 6096 6097 /* Delayed registration/unregisteration */ 6098 static LIST_HEAD(net_todo_list); 6099 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 6100 6101 static void net_set_todo(struct net_device *dev) 6102 { 6103 list_add_tail(&dev->todo_list, &net_todo_list); 6104 dev_net(dev)->dev_unreg_count++; 6105 } 6106 6107 static void rollback_registered_many(struct list_head *head) 6108 { 6109 struct net_device *dev, *tmp; 6110 LIST_HEAD(close_head); 6111 6112 BUG_ON(dev_boot_phase); 6113 ASSERT_RTNL(); 6114 6115 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 6116 /* Some devices call without registering 6117 * for initialization unwind. Remove those 6118 * devices and proceed with the remaining. 6119 */ 6120 if (dev->reg_state == NETREG_UNINITIALIZED) { 6121 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 6122 dev->name, dev); 6123 6124 WARN_ON(1); 6125 list_del(&dev->unreg_list); 6126 continue; 6127 } 6128 dev->dismantle = true; 6129 BUG_ON(dev->reg_state != NETREG_REGISTERED); 6130 } 6131 6132 /* If device is running, close it first. */ 6133 list_for_each_entry(dev, head, unreg_list) 6134 list_add_tail(&dev->close_list, &close_head); 6135 dev_close_many(&close_head, true); 6136 6137 list_for_each_entry(dev, head, unreg_list) { 6138 /* And unlink it from device chain. */ 6139 unlist_netdevice(dev); 6140 6141 dev->reg_state = NETREG_UNREGISTERING; 6142 } 6143 6144 synchronize_net(); 6145 6146 list_for_each_entry(dev, head, unreg_list) { 6147 struct sk_buff *skb = NULL; 6148 6149 /* Shutdown queueing discipline. */ 6150 dev_shutdown(dev); 6151 6152 6153 /* Notify protocols, that we are about to destroy 6154 this device. They should clean all the things. 6155 */ 6156 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 6157 6158 if (!dev->rtnl_link_ops || 6159 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 6160 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 6161 GFP_KERNEL); 6162 6163 /* 6164 * Flush the unicast and multicast chains 6165 */ 6166 dev_uc_flush(dev); 6167 dev_mc_flush(dev); 6168 6169 if (dev->netdev_ops->ndo_uninit) 6170 dev->netdev_ops->ndo_uninit(dev); 6171 6172 if (skb) 6173 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 6174 6175 /* Notifier chain MUST detach us all upper devices. */ 6176 WARN_ON(netdev_has_any_upper_dev(dev)); 6177 6178 /* Remove entries from kobject tree */ 6179 netdev_unregister_kobject(dev); 6180 #ifdef CONFIG_XPS 6181 /* Remove XPS queueing entries */ 6182 netif_reset_xps_queues_gt(dev, 0); 6183 #endif 6184 } 6185 6186 synchronize_net(); 6187 6188 list_for_each_entry(dev, head, unreg_list) 6189 dev_put(dev); 6190 } 6191 6192 static void rollback_registered(struct net_device *dev) 6193 { 6194 LIST_HEAD(single); 6195 6196 list_add(&dev->unreg_list, &single); 6197 rollback_registered_many(&single); 6198 list_del(&single); 6199 } 6200 6201 static netdev_features_t netdev_fix_features(struct net_device *dev, 6202 netdev_features_t features) 6203 { 6204 /* Fix illegal checksum combinations */ 6205 if ((features & NETIF_F_HW_CSUM) && 6206 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 6207 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 6208 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 6209 } 6210 6211 /* TSO requires that SG is present as well. */ 6212 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 6213 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 6214 features &= ~NETIF_F_ALL_TSO; 6215 } 6216 6217 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 6218 !(features & NETIF_F_IP_CSUM)) { 6219 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 6220 features &= ~NETIF_F_TSO; 6221 features &= ~NETIF_F_TSO_ECN; 6222 } 6223 6224 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 6225 !(features & NETIF_F_IPV6_CSUM)) { 6226 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 6227 features &= ~NETIF_F_TSO6; 6228 } 6229 6230 /* TSO ECN requires that TSO is present as well. */ 6231 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 6232 features &= ~NETIF_F_TSO_ECN; 6233 6234 /* Software GSO depends on SG. */ 6235 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 6236 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 6237 features &= ~NETIF_F_GSO; 6238 } 6239 6240 /* UFO needs SG and checksumming */ 6241 if (features & NETIF_F_UFO) { 6242 /* maybe split UFO into V4 and V6? */ 6243 if (!((features & NETIF_F_GEN_CSUM) || 6244 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM)) 6245 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 6246 netdev_dbg(dev, 6247 "Dropping NETIF_F_UFO since no checksum offload features.\n"); 6248 features &= ~NETIF_F_UFO; 6249 } 6250 6251 if (!(features & NETIF_F_SG)) { 6252 netdev_dbg(dev, 6253 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); 6254 features &= ~NETIF_F_UFO; 6255 } 6256 } 6257 6258 #ifdef CONFIG_NET_RX_BUSY_POLL 6259 if (dev->netdev_ops->ndo_busy_poll) 6260 features |= NETIF_F_BUSY_POLL; 6261 else 6262 #endif 6263 features &= ~NETIF_F_BUSY_POLL; 6264 6265 return features; 6266 } 6267 6268 int __netdev_update_features(struct net_device *dev) 6269 { 6270 netdev_features_t features; 6271 int err = 0; 6272 6273 ASSERT_RTNL(); 6274 6275 features = netdev_get_wanted_features(dev); 6276 6277 if (dev->netdev_ops->ndo_fix_features) 6278 features = dev->netdev_ops->ndo_fix_features(dev, features); 6279 6280 /* driver might be less strict about feature dependencies */ 6281 features = netdev_fix_features(dev, features); 6282 6283 if (dev->features == features) 6284 return 0; 6285 6286 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 6287 &dev->features, &features); 6288 6289 if (dev->netdev_ops->ndo_set_features) 6290 err = dev->netdev_ops->ndo_set_features(dev, features); 6291 6292 if (unlikely(err < 0)) { 6293 netdev_err(dev, 6294 "set_features() failed (%d); wanted %pNF, left %pNF\n", 6295 err, &features, &dev->features); 6296 return -1; 6297 } 6298 6299 if (!err) 6300 dev->features = features; 6301 6302 return 1; 6303 } 6304 6305 /** 6306 * netdev_update_features - recalculate device features 6307 * @dev: the device to check 6308 * 6309 * Recalculate dev->features set and send notifications if it 6310 * has changed. Should be called after driver or hardware dependent 6311 * conditions might have changed that influence the features. 6312 */ 6313 void netdev_update_features(struct net_device *dev) 6314 { 6315 if (__netdev_update_features(dev)) 6316 netdev_features_change(dev); 6317 } 6318 EXPORT_SYMBOL(netdev_update_features); 6319 6320 /** 6321 * netdev_change_features - recalculate device features 6322 * @dev: the device to check 6323 * 6324 * Recalculate dev->features set and send notifications even 6325 * if they have not changed. Should be called instead of 6326 * netdev_update_features() if also dev->vlan_features might 6327 * have changed to allow the changes to be propagated to stacked 6328 * VLAN devices. 6329 */ 6330 void netdev_change_features(struct net_device *dev) 6331 { 6332 __netdev_update_features(dev); 6333 netdev_features_change(dev); 6334 } 6335 EXPORT_SYMBOL(netdev_change_features); 6336 6337 /** 6338 * netif_stacked_transfer_operstate - transfer operstate 6339 * @rootdev: the root or lower level device to transfer state from 6340 * @dev: the device to transfer operstate to 6341 * 6342 * Transfer operational state from root to device. This is normally 6343 * called when a stacking relationship exists between the root 6344 * device and the device(a leaf device). 6345 */ 6346 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 6347 struct net_device *dev) 6348 { 6349 if (rootdev->operstate == IF_OPER_DORMANT) 6350 netif_dormant_on(dev); 6351 else 6352 netif_dormant_off(dev); 6353 6354 if (netif_carrier_ok(rootdev)) { 6355 if (!netif_carrier_ok(dev)) 6356 netif_carrier_on(dev); 6357 } else { 6358 if (netif_carrier_ok(dev)) 6359 netif_carrier_off(dev); 6360 } 6361 } 6362 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 6363 6364 #ifdef CONFIG_SYSFS 6365 static int netif_alloc_rx_queues(struct net_device *dev) 6366 { 6367 unsigned int i, count = dev->num_rx_queues; 6368 struct netdev_rx_queue *rx; 6369 size_t sz = count * sizeof(*rx); 6370 6371 BUG_ON(count < 1); 6372 6373 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 6374 if (!rx) { 6375 rx = vzalloc(sz); 6376 if (!rx) 6377 return -ENOMEM; 6378 } 6379 dev->_rx = rx; 6380 6381 for (i = 0; i < count; i++) 6382 rx[i].dev = dev; 6383 return 0; 6384 } 6385 #endif 6386 6387 static void netdev_init_one_queue(struct net_device *dev, 6388 struct netdev_queue *queue, void *_unused) 6389 { 6390 /* Initialize queue lock */ 6391 spin_lock_init(&queue->_xmit_lock); 6392 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 6393 queue->xmit_lock_owner = -1; 6394 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 6395 queue->dev = dev; 6396 #ifdef CONFIG_BQL 6397 dql_init(&queue->dql, HZ); 6398 #endif 6399 } 6400 6401 static void netif_free_tx_queues(struct net_device *dev) 6402 { 6403 kvfree(dev->_tx); 6404 } 6405 6406 static int netif_alloc_netdev_queues(struct net_device *dev) 6407 { 6408 unsigned int count = dev->num_tx_queues; 6409 struct netdev_queue *tx; 6410 size_t sz = count * sizeof(*tx); 6411 6412 BUG_ON(count < 1 || count > 0xffff); 6413 6414 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 6415 if (!tx) { 6416 tx = vzalloc(sz); 6417 if (!tx) 6418 return -ENOMEM; 6419 } 6420 dev->_tx = tx; 6421 6422 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 6423 spin_lock_init(&dev->tx_global_lock); 6424 6425 return 0; 6426 } 6427 6428 void netif_tx_stop_all_queues(struct net_device *dev) 6429 { 6430 unsigned int i; 6431 6432 for (i = 0; i < dev->num_tx_queues; i++) { 6433 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 6434 netif_tx_stop_queue(txq); 6435 } 6436 } 6437 EXPORT_SYMBOL(netif_tx_stop_all_queues); 6438 6439 /** 6440 * register_netdevice - register a network device 6441 * @dev: device to register 6442 * 6443 * Take a completed network device structure and add it to the kernel 6444 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 6445 * chain. 0 is returned on success. A negative errno code is returned 6446 * on a failure to set up the device, or if the name is a duplicate. 6447 * 6448 * Callers must hold the rtnl semaphore. You may want 6449 * register_netdev() instead of this. 6450 * 6451 * BUGS: 6452 * The locking appears insufficient to guarantee two parallel registers 6453 * will not get the same name. 6454 */ 6455 6456 int register_netdevice(struct net_device *dev) 6457 { 6458 int ret; 6459 struct net *net = dev_net(dev); 6460 6461 BUG_ON(dev_boot_phase); 6462 ASSERT_RTNL(); 6463 6464 might_sleep(); 6465 6466 /* When net_device's are persistent, this will be fatal. */ 6467 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 6468 BUG_ON(!net); 6469 6470 spin_lock_init(&dev->addr_list_lock); 6471 netdev_set_addr_lockdep_class(dev); 6472 6473 ret = dev_get_valid_name(net, dev, dev->name); 6474 if (ret < 0) 6475 goto out; 6476 6477 /* Init, if this function is available */ 6478 if (dev->netdev_ops->ndo_init) { 6479 ret = dev->netdev_ops->ndo_init(dev); 6480 if (ret) { 6481 if (ret > 0) 6482 ret = -EIO; 6483 goto out; 6484 } 6485 } 6486 6487 if (((dev->hw_features | dev->features) & 6488 NETIF_F_HW_VLAN_CTAG_FILTER) && 6489 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 6490 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 6491 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 6492 ret = -EINVAL; 6493 goto err_uninit; 6494 } 6495 6496 ret = -EBUSY; 6497 if (!dev->ifindex) 6498 dev->ifindex = dev_new_index(net); 6499 else if (__dev_get_by_index(net, dev->ifindex)) 6500 goto err_uninit; 6501 6502 /* Transfer changeable features to wanted_features and enable 6503 * software offloads (GSO and GRO). 6504 */ 6505 dev->hw_features |= NETIF_F_SOFT_FEATURES; 6506 dev->features |= NETIF_F_SOFT_FEATURES; 6507 dev->wanted_features = dev->features & dev->hw_features; 6508 6509 if (!(dev->flags & IFF_LOOPBACK)) { 6510 dev->hw_features |= NETIF_F_NOCACHE_COPY; 6511 } 6512 6513 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 6514 */ 6515 dev->vlan_features |= NETIF_F_HIGHDMA; 6516 6517 /* Make NETIF_F_SG inheritable to tunnel devices. 6518 */ 6519 dev->hw_enc_features |= NETIF_F_SG; 6520 6521 /* Make NETIF_F_SG inheritable to MPLS. 6522 */ 6523 dev->mpls_features |= NETIF_F_SG; 6524 6525 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 6526 ret = notifier_to_errno(ret); 6527 if (ret) 6528 goto err_uninit; 6529 6530 ret = netdev_register_kobject(dev); 6531 if (ret) 6532 goto err_uninit; 6533 dev->reg_state = NETREG_REGISTERED; 6534 6535 __netdev_update_features(dev); 6536 6537 /* 6538 * Default initial state at registry is that the 6539 * device is present. 6540 */ 6541 6542 set_bit(__LINK_STATE_PRESENT, &dev->state); 6543 6544 linkwatch_init_dev(dev); 6545 6546 dev_init_scheduler(dev); 6547 dev_hold(dev); 6548 list_netdevice(dev); 6549 add_device_randomness(dev->dev_addr, dev->addr_len); 6550 6551 /* If the device has permanent device address, driver should 6552 * set dev_addr and also addr_assign_type should be set to 6553 * NET_ADDR_PERM (default value). 6554 */ 6555 if (dev->addr_assign_type == NET_ADDR_PERM) 6556 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 6557 6558 /* Notify protocols, that a new device appeared. */ 6559 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 6560 ret = notifier_to_errno(ret); 6561 if (ret) { 6562 rollback_registered(dev); 6563 dev->reg_state = NETREG_UNREGISTERED; 6564 } 6565 /* 6566 * Prevent userspace races by waiting until the network 6567 * device is fully setup before sending notifications. 6568 */ 6569 if (!dev->rtnl_link_ops || 6570 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 6571 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 6572 6573 out: 6574 return ret; 6575 6576 err_uninit: 6577 if (dev->netdev_ops->ndo_uninit) 6578 dev->netdev_ops->ndo_uninit(dev); 6579 goto out; 6580 } 6581 EXPORT_SYMBOL(register_netdevice); 6582 6583 /** 6584 * init_dummy_netdev - init a dummy network device for NAPI 6585 * @dev: device to init 6586 * 6587 * This takes a network device structure and initialize the minimum 6588 * amount of fields so it can be used to schedule NAPI polls without 6589 * registering a full blown interface. This is to be used by drivers 6590 * that need to tie several hardware interfaces to a single NAPI 6591 * poll scheduler due to HW limitations. 6592 */ 6593 int init_dummy_netdev(struct net_device *dev) 6594 { 6595 /* Clear everything. Note we don't initialize spinlocks 6596 * are they aren't supposed to be taken by any of the 6597 * NAPI code and this dummy netdev is supposed to be 6598 * only ever used for NAPI polls 6599 */ 6600 memset(dev, 0, sizeof(struct net_device)); 6601 6602 /* make sure we BUG if trying to hit standard 6603 * register/unregister code path 6604 */ 6605 dev->reg_state = NETREG_DUMMY; 6606 6607 /* NAPI wants this */ 6608 INIT_LIST_HEAD(&dev->napi_list); 6609 6610 /* a dummy interface is started by default */ 6611 set_bit(__LINK_STATE_PRESENT, &dev->state); 6612 set_bit(__LINK_STATE_START, &dev->state); 6613 6614 /* Note : We dont allocate pcpu_refcnt for dummy devices, 6615 * because users of this 'device' dont need to change 6616 * its refcount. 6617 */ 6618 6619 return 0; 6620 } 6621 EXPORT_SYMBOL_GPL(init_dummy_netdev); 6622 6623 6624 /** 6625 * register_netdev - register a network device 6626 * @dev: device to register 6627 * 6628 * Take a completed network device structure and add it to the kernel 6629 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 6630 * chain. 0 is returned on success. A negative errno code is returned 6631 * on a failure to set up the device, or if the name is a duplicate. 6632 * 6633 * This is a wrapper around register_netdevice that takes the rtnl semaphore 6634 * and expands the device name if you passed a format string to 6635 * alloc_netdev. 6636 */ 6637 int register_netdev(struct net_device *dev) 6638 { 6639 int err; 6640 6641 rtnl_lock(); 6642 err = register_netdevice(dev); 6643 rtnl_unlock(); 6644 return err; 6645 } 6646 EXPORT_SYMBOL(register_netdev); 6647 6648 int netdev_refcnt_read(const struct net_device *dev) 6649 { 6650 int i, refcnt = 0; 6651 6652 for_each_possible_cpu(i) 6653 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 6654 return refcnt; 6655 } 6656 EXPORT_SYMBOL(netdev_refcnt_read); 6657 6658 /** 6659 * netdev_wait_allrefs - wait until all references are gone. 6660 * @dev: target net_device 6661 * 6662 * This is called when unregistering network devices. 6663 * 6664 * Any protocol or device that holds a reference should register 6665 * for netdevice notification, and cleanup and put back the 6666 * reference if they receive an UNREGISTER event. 6667 * We can get stuck here if buggy protocols don't correctly 6668 * call dev_put. 6669 */ 6670 static void netdev_wait_allrefs(struct net_device *dev) 6671 { 6672 unsigned long rebroadcast_time, warning_time; 6673 int refcnt; 6674 6675 linkwatch_forget_dev(dev); 6676 6677 rebroadcast_time = warning_time = jiffies; 6678 refcnt = netdev_refcnt_read(dev); 6679 6680 while (refcnt != 0) { 6681 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 6682 rtnl_lock(); 6683 6684 /* Rebroadcast unregister notification */ 6685 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 6686 6687 __rtnl_unlock(); 6688 rcu_barrier(); 6689 rtnl_lock(); 6690 6691 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 6692 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 6693 &dev->state)) { 6694 /* We must not have linkwatch events 6695 * pending on unregister. If this 6696 * happens, we simply run the queue 6697 * unscheduled, resulting in a noop 6698 * for this device. 6699 */ 6700 linkwatch_run_queue(); 6701 } 6702 6703 __rtnl_unlock(); 6704 6705 rebroadcast_time = jiffies; 6706 } 6707 6708 msleep(250); 6709 6710 refcnt = netdev_refcnt_read(dev); 6711 6712 if (time_after(jiffies, warning_time + 10 * HZ)) { 6713 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 6714 dev->name, refcnt); 6715 warning_time = jiffies; 6716 } 6717 } 6718 } 6719 6720 /* The sequence is: 6721 * 6722 * rtnl_lock(); 6723 * ... 6724 * register_netdevice(x1); 6725 * register_netdevice(x2); 6726 * ... 6727 * unregister_netdevice(y1); 6728 * unregister_netdevice(y2); 6729 * ... 6730 * rtnl_unlock(); 6731 * free_netdev(y1); 6732 * free_netdev(y2); 6733 * 6734 * We are invoked by rtnl_unlock(). 6735 * This allows us to deal with problems: 6736 * 1) We can delete sysfs objects which invoke hotplug 6737 * without deadlocking with linkwatch via keventd. 6738 * 2) Since we run with the RTNL semaphore not held, we can sleep 6739 * safely in order to wait for the netdev refcnt to drop to zero. 6740 * 6741 * We must not return until all unregister events added during 6742 * the interval the lock was held have been completed. 6743 */ 6744 void netdev_run_todo(void) 6745 { 6746 struct list_head list; 6747 6748 /* Snapshot list, allow later requests */ 6749 list_replace_init(&net_todo_list, &list); 6750 6751 __rtnl_unlock(); 6752 6753 6754 /* Wait for rcu callbacks to finish before next phase */ 6755 if (!list_empty(&list)) 6756 rcu_barrier(); 6757 6758 while (!list_empty(&list)) { 6759 struct net_device *dev 6760 = list_first_entry(&list, struct net_device, todo_list); 6761 list_del(&dev->todo_list); 6762 6763 rtnl_lock(); 6764 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 6765 __rtnl_unlock(); 6766 6767 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 6768 pr_err("network todo '%s' but state %d\n", 6769 dev->name, dev->reg_state); 6770 dump_stack(); 6771 continue; 6772 } 6773 6774 dev->reg_state = NETREG_UNREGISTERED; 6775 6776 on_each_cpu(flush_backlog, dev, 1); 6777 6778 netdev_wait_allrefs(dev); 6779 6780 /* paranoia */ 6781 BUG_ON(netdev_refcnt_read(dev)); 6782 BUG_ON(!list_empty(&dev->ptype_all)); 6783 BUG_ON(!list_empty(&dev->ptype_specific)); 6784 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 6785 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 6786 WARN_ON(dev->dn_ptr); 6787 6788 if (dev->destructor) 6789 dev->destructor(dev); 6790 6791 /* Report a network device has been unregistered */ 6792 rtnl_lock(); 6793 dev_net(dev)->dev_unreg_count--; 6794 __rtnl_unlock(); 6795 wake_up(&netdev_unregistering_wq); 6796 6797 /* Free network device */ 6798 kobject_put(&dev->dev.kobj); 6799 } 6800 } 6801 6802 /* Convert net_device_stats to rtnl_link_stats64. They have the same 6803 * fields in the same order, with only the type differing. 6804 */ 6805 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 6806 const struct net_device_stats *netdev_stats) 6807 { 6808 #if BITS_PER_LONG == 64 6809 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats)); 6810 memcpy(stats64, netdev_stats, sizeof(*stats64)); 6811 #else 6812 size_t i, n = sizeof(*stats64) / sizeof(u64); 6813 const unsigned long *src = (const unsigned long *)netdev_stats; 6814 u64 *dst = (u64 *)stats64; 6815 6816 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) != 6817 sizeof(*stats64) / sizeof(u64)); 6818 for (i = 0; i < n; i++) 6819 dst[i] = src[i]; 6820 #endif 6821 } 6822 EXPORT_SYMBOL(netdev_stats_to_stats64); 6823 6824 /** 6825 * dev_get_stats - get network device statistics 6826 * @dev: device to get statistics from 6827 * @storage: place to store stats 6828 * 6829 * Get network statistics from device. Return @storage. 6830 * The device driver may provide its own method by setting 6831 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 6832 * otherwise the internal statistics structure is used. 6833 */ 6834 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 6835 struct rtnl_link_stats64 *storage) 6836 { 6837 const struct net_device_ops *ops = dev->netdev_ops; 6838 6839 if (ops->ndo_get_stats64) { 6840 memset(storage, 0, sizeof(*storage)); 6841 ops->ndo_get_stats64(dev, storage); 6842 } else if (ops->ndo_get_stats) { 6843 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 6844 } else { 6845 netdev_stats_to_stats64(storage, &dev->stats); 6846 } 6847 storage->rx_dropped += atomic_long_read(&dev->rx_dropped); 6848 storage->tx_dropped += atomic_long_read(&dev->tx_dropped); 6849 return storage; 6850 } 6851 EXPORT_SYMBOL(dev_get_stats); 6852 6853 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 6854 { 6855 struct netdev_queue *queue = dev_ingress_queue(dev); 6856 6857 #ifdef CONFIG_NET_CLS_ACT 6858 if (queue) 6859 return queue; 6860 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 6861 if (!queue) 6862 return NULL; 6863 netdev_init_one_queue(dev, queue, NULL); 6864 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 6865 queue->qdisc_sleeping = &noop_qdisc; 6866 rcu_assign_pointer(dev->ingress_queue, queue); 6867 #endif 6868 return queue; 6869 } 6870 6871 static const struct ethtool_ops default_ethtool_ops; 6872 6873 void netdev_set_default_ethtool_ops(struct net_device *dev, 6874 const struct ethtool_ops *ops) 6875 { 6876 if (dev->ethtool_ops == &default_ethtool_ops) 6877 dev->ethtool_ops = ops; 6878 } 6879 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 6880 6881 void netdev_freemem(struct net_device *dev) 6882 { 6883 char *addr = (char *)dev - dev->padded; 6884 6885 kvfree(addr); 6886 } 6887 6888 /** 6889 * alloc_netdev_mqs - allocate network device 6890 * @sizeof_priv: size of private data to allocate space for 6891 * @name: device name format string 6892 * @name_assign_type: origin of device name 6893 * @setup: callback to initialize device 6894 * @txqs: the number of TX subqueues to allocate 6895 * @rxqs: the number of RX subqueues to allocate 6896 * 6897 * Allocates a struct net_device with private data area for driver use 6898 * and performs basic initialization. Also allocates subqueue structs 6899 * for each queue on the device. 6900 */ 6901 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 6902 unsigned char name_assign_type, 6903 void (*setup)(struct net_device *), 6904 unsigned int txqs, unsigned int rxqs) 6905 { 6906 struct net_device *dev; 6907 size_t alloc_size; 6908 struct net_device *p; 6909 6910 BUG_ON(strlen(name) >= sizeof(dev->name)); 6911 6912 if (txqs < 1) { 6913 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 6914 return NULL; 6915 } 6916 6917 #ifdef CONFIG_SYSFS 6918 if (rxqs < 1) { 6919 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 6920 return NULL; 6921 } 6922 #endif 6923 6924 alloc_size = sizeof(struct net_device); 6925 if (sizeof_priv) { 6926 /* ensure 32-byte alignment of private area */ 6927 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 6928 alloc_size += sizeof_priv; 6929 } 6930 /* ensure 32-byte alignment of whole construct */ 6931 alloc_size += NETDEV_ALIGN - 1; 6932 6933 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 6934 if (!p) 6935 p = vzalloc(alloc_size); 6936 if (!p) 6937 return NULL; 6938 6939 dev = PTR_ALIGN(p, NETDEV_ALIGN); 6940 dev->padded = (char *)dev - (char *)p; 6941 6942 dev->pcpu_refcnt = alloc_percpu(int); 6943 if (!dev->pcpu_refcnt) 6944 goto free_dev; 6945 6946 if (dev_addr_init(dev)) 6947 goto free_pcpu; 6948 6949 dev_mc_init(dev); 6950 dev_uc_init(dev); 6951 6952 dev_net_set(dev, &init_net); 6953 6954 dev->gso_max_size = GSO_MAX_SIZE; 6955 dev->gso_max_segs = GSO_MAX_SEGS; 6956 dev->gso_min_segs = 0; 6957 6958 INIT_LIST_HEAD(&dev->napi_list); 6959 INIT_LIST_HEAD(&dev->unreg_list); 6960 INIT_LIST_HEAD(&dev->close_list); 6961 INIT_LIST_HEAD(&dev->link_watch_list); 6962 INIT_LIST_HEAD(&dev->adj_list.upper); 6963 INIT_LIST_HEAD(&dev->adj_list.lower); 6964 INIT_LIST_HEAD(&dev->all_adj_list.upper); 6965 INIT_LIST_HEAD(&dev->all_adj_list.lower); 6966 INIT_LIST_HEAD(&dev->ptype_all); 6967 INIT_LIST_HEAD(&dev->ptype_specific); 6968 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 6969 setup(dev); 6970 6971 dev->num_tx_queues = txqs; 6972 dev->real_num_tx_queues = txqs; 6973 if (netif_alloc_netdev_queues(dev)) 6974 goto free_all; 6975 6976 #ifdef CONFIG_SYSFS 6977 dev->num_rx_queues = rxqs; 6978 dev->real_num_rx_queues = rxqs; 6979 if (netif_alloc_rx_queues(dev)) 6980 goto free_all; 6981 #endif 6982 6983 strcpy(dev->name, name); 6984 dev->name_assign_type = name_assign_type; 6985 dev->group = INIT_NETDEV_GROUP; 6986 if (!dev->ethtool_ops) 6987 dev->ethtool_ops = &default_ethtool_ops; 6988 6989 nf_hook_ingress_init(dev); 6990 6991 return dev; 6992 6993 free_all: 6994 free_netdev(dev); 6995 return NULL; 6996 6997 free_pcpu: 6998 free_percpu(dev->pcpu_refcnt); 6999 free_dev: 7000 netdev_freemem(dev); 7001 return NULL; 7002 } 7003 EXPORT_SYMBOL(alloc_netdev_mqs); 7004 7005 /** 7006 * free_netdev - free network device 7007 * @dev: device 7008 * 7009 * This function does the last stage of destroying an allocated device 7010 * interface. The reference to the device object is released. 7011 * If this is the last reference then it will be freed. 7012 */ 7013 void free_netdev(struct net_device *dev) 7014 { 7015 struct napi_struct *p, *n; 7016 7017 netif_free_tx_queues(dev); 7018 #ifdef CONFIG_SYSFS 7019 kvfree(dev->_rx); 7020 #endif 7021 7022 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 7023 7024 /* Flush device addresses */ 7025 dev_addr_flush(dev); 7026 7027 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 7028 netif_napi_del(p); 7029 7030 free_percpu(dev->pcpu_refcnt); 7031 dev->pcpu_refcnt = NULL; 7032 7033 /* Compatibility with error handling in drivers */ 7034 if (dev->reg_state == NETREG_UNINITIALIZED) { 7035 netdev_freemem(dev); 7036 return; 7037 } 7038 7039 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 7040 dev->reg_state = NETREG_RELEASED; 7041 7042 /* will free via device release */ 7043 put_device(&dev->dev); 7044 } 7045 EXPORT_SYMBOL(free_netdev); 7046 7047 /** 7048 * synchronize_net - Synchronize with packet receive processing 7049 * 7050 * Wait for packets currently being received to be done. 7051 * Does not block later packets from starting. 7052 */ 7053 void synchronize_net(void) 7054 { 7055 might_sleep(); 7056 if (rtnl_is_locked()) 7057 synchronize_rcu_expedited(); 7058 else 7059 synchronize_rcu(); 7060 } 7061 EXPORT_SYMBOL(synchronize_net); 7062 7063 /** 7064 * unregister_netdevice_queue - remove device from the kernel 7065 * @dev: device 7066 * @head: list 7067 * 7068 * This function shuts down a device interface and removes it 7069 * from the kernel tables. 7070 * If head not NULL, device is queued to be unregistered later. 7071 * 7072 * Callers must hold the rtnl semaphore. You may want 7073 * unregister_netdev() instead of this. 7074 */ 7075 7076 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 7077 { 7078 ASSERT_RTNL(); 7079 7080 if (head) { 7081 list_move_tail(&dev->unreg_list, head); 7082 } else { 7083 rollback_registered(dev); 7084 /* Finish processing unregister after unlock */ 7085 net_set_todo(dev); 7086 } 7087 } 7088 EXPORT_SYMBOL(unregister_netdevice_queue); 7089 7090 /** 7091 * unregister_netdevice_many - unregister many devices 7092 * @head: list of devices 7093 * 7094 * Note: As most callers use a stack allocated list_head, 7095 * we force a list_del() to make sure stack wont be corrupted later. 7096 */ 7097 void unregister_netdevice_many(struct list_head *head) 7098 { 7099 struct net_device *dev; 7100 7101 if (!list_empty(head)) { 7102 rollback_registered_many(head); 7103 list_for_each_entry(dev, head, unreg_list) 7104 net_set_todo(dev); 7105 list_del(head); 7106 } 7107 } 7108 EXPORT_SYMBOL(unregister_netdevice_many); 7109 7110 /** 7111 * unregister_netdev - remove device from the kernel 7112 * @dev: device 7113 * 7114 * This function shuts down a device interface and removes it 7115 * from the kernel tables. 7116 * 7117 * This is just a wrapper for unregister_netdevice that takes 7118 * the rtnl semaphore. In general you want to use this and not 7119 * unregister_netdevice. 7120 */ 7121 void unregister_netdev(struct net_device *dev) 7122 { 7123 rtnl_lock(); 7124 unregister_netdevice(dev); 7125 rtnl_unlock(); 7126 } 7127 EXPORT_SYMBOL(unregister_netdev); 7128 7129 /** 7130 * dev_change_net_namespace - move device to different nethost namespace 7131 * @dev: device 7132 * @net: network namespace 7133 * @pat: If not NULL name pattern to try if the current device name 7134 * is already taken in the destination network namespace. 7135 * 7136 * This function shuts down a device interface and moves it 7137 * to a new network namespace. On success 0 is returned, on 7138 * a failure a netagive errno code is returned. 7139 * 7140 * Callers must hold the rtnl semaphore. 7141 */ 7142 7143 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 7144 { 7145 int err; 7146 7147 ASSERT_RTNL(); 7148 7149 /* Don't allow namespace local devices to be moved. */ 7150 err = -EINVAL; 7151 if (dev->features & NETIF_F_NETNS_LOCAL) 7152 goto out; 7153 7154 /* Ensure the device has been registrered */ 7155 if (dev->reg_state != NETREG_REGISTERED) 7156 goto out; 7157 7158 /* Get out if there is nothing todo */ 7159 err = 0; 7160 if (net_eq(dev_net(dev), net)) 7161 goto out; 7162 7163 /* Pick the destination device name, and ensure 7164 * we can use it in the destination network namespace. 7165 */ 7166 err = -EEXIST; 7167 if (__dev_get_by_name(net, dev->name)) { 7168 /* We get here if we can't use the current device name */ 7169 if (!pat) 7170 goto out; 7171 if (dev_get_valid_name(net, dev, pat) < 0) 7172 goto out; 7173 } 7174 7175 /* 7176 * And now a mini version of register_netdevice unregister_netdevice. 7177 */ 7178 7179 /* If device is running close it first. */ 7180 dev_close(dev); 7181 7182 /* And unlink it from device chain */ 7183 err = -ENODEV; 7184 unlist_netdevice(dev); 7185 7186 synchronize_net(); 7187 7188 /* Shutdown queueing discipline. */ 7189 dev_shutdown(dev); 7190 7191 /* Notify protocols, that we are about to destroy 7192 this device. They should clean all the things. 7193 7194 Note that dev->reg_state stays at NETREG_REGISTERED. 7195 This is wanted because this way 8021q and macvlan know 7196 the device is just moving and can keep their slaves up. 7197 */ 7198 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7199 rcu_barrier(); 7200 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7201 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); 7202 7203 /* 7204 * Flush the unicast and multicast chains 7205 */ 7206 dev_uc_flush(dev); 7207 dev_mc_flush(dev); 7208 7209 /* Send a netdev-removed uevent to the old namespace */ 7210 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 7211 netdev_adjacent_del_links(dev); 7212 7213 /* Actually switch the network namespace */ 7214 dev_net_set(dev, net); 7215 7216 /* If there is an ifindex conflict assign a new one */ 7217 if (__dev_get_by_index(net, dev->ifindex)) 7218 dev->ifindex = dev_new_index(net); 7219 7220 /* Send a netdev-add uevent to the new namespace */ 7221 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 7222 netdev_adjacent_add_links(dev); 7223 7224 /* Fixup kobjects */ 7225 err = device_rename(&dev->dev, dev->name); 7226 WARN_ON(err); 7227 7228 /* Add the device back in the hashes */ 7229 list_netdevice(dev); 7230 7231 /* Notify protocols, that a new device appeared. */ 7232 call_netdevice_notifiers(NETDEV_REGISTER, dev); 7233 7234 /* 7235 * Prevent userspace races by waiting until the network 7236 * device is fully setup before sending notifications. 7237 */ 7238 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 7239 7240 synchronize_net(); 7241 err = 0; 7242 out: 7243 return err; 7244 } 7245 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 7246 7247 static int dev_cpu_callback(struct notifier_block *nfb, 7248 unsigned long action, 7249 void *ocpu) 7250 { 7251 struct sk_buff **list_skb; 7252 struct sk_buff *skb; 7253 unsigned int cpu, oldcpu = (unsigned long)ocpu; 7254 struct softnet_data *sd, *oldsd; 7255 7256 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) 7257 return NOTIFY_OK; 7258 7259 local_irq_disable(); 7260 cpu = smp_processor_id(); 7261 sd = &per_cpu(softnet_data, cpu); 7262 oldsd = &per_cpu(softnet_data, oldcpu); 7263 7264 /* Find end of our completion_queue. */ 7265 list_skb = &sd->completion_queue; 7266 while (*list_skb) 7267 list_skb = &(*list_skb)->next; 7268 /* Append completion queue from offline CPU. */ 7269 *list_skb = oldsd->completion_queue; 7270 oldsd->completion_queue = NULL; 7271 7272 /* Append output queue from offline CPU. */ 7273 if (oldsd->output_queue) { 7274 *sd->output_queue_tailp = oldsd->output_queue; 7275 sd->output_queue_tailp = oldsd->output_queue_tailp; 7276 oldsd->output_queue = NULL; 7277 oldsd->output_queue_tailp = &oldsd->output_queue; 7278 } 7279 /* Append NAPI poll list from offline CPU, with one exception : 7280 * process_backlog() must be called by cpu owning percpu backlog. 7281 * We properly handle process_queue & input_pkt_queue later. 7282 */ 7283 while (!list_empty(&oldsd->poll_list)) { 7284 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 7285 struct napi_struct, 7286 poll_list); 7287 7288 list_del_init(&napi->poll_list); 7289 if (napi->poll == process_backlog) 7290 napi->state = 0; 7291 else 7292 ____napi_schedule(sd, napi); 7293 } 7294 7295 raise_softirq_irqoff(NET_TX_SOFTIRQ); 7296 local_irq_enable(); 7297 7298 /* Process offline CPU's input_pkt_queue */ 7299 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 7300 netif_rx_ni(skb); 7301 input_queue_head_incr(oldsd); 7302 } 7303 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 7304 netif_rx_ni(skb); 7305 input_queue_head_incr(oldsd); 7306 } 7307 7308 return NOTIFY_OK; 7309 } 7310 7311 7312 /** 7313 * netdev_increment_features - increment feature set by one 7314 * @all: current feature set 7315 * @one: new feature set 7316 * @mask: mask feature set 7317 * 7318 * Computes a new feature set after adding a device with feature set 7319 * @one to the master device with current feature set @all. Will not 7320 * enable anything that is off in @mask. Returns the new feature set. 7321 */ 7322 netdev_features_t netdev_increment_features(netdev_features_t all, 7323 netdev_features_t one, netdev_features_t mask) 7324 { 7325 if (mask & NETIF_F_GEN_CSUM) 7326 mask |= NETIF_F_ALL_CSUM; 7327 mask |= NETIF_F_VLAN_CHALLENGED; 7328 7329 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask; 7330 all &= one | ~NETIF_F_ALL_FOR_ALL; 7331 7332 /* If one device supports hw checksumming, set for all. */ 7333 if (all & NETIF_F_GEN_CSUM) 7334 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM); 7335 7336 return all; 7337 } 7338 EXPORT_SYMBOL(netdev_increment_features); 7339 7340 static struct hlist_head * __net_init netdev_create_hash(void) 7341 { 7342 int i; 7343 struct hlist_head *hash; 7344 7345 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); 7346 if (hash != NULL) 7347 for (i = 0; i < NETDEV_HASHENTRIES; i++) 7348 INIT_HLIST_HEAD(&hash[i]); 7349 7350 return hash; 7351 } 7352 7353 /* Initialize per network namespace state */ 7354 static int __net_init netdev_init(struct net *net) 7355 { 7356 if (net != &init_net) 7357 INIT_LIST_HEAD(&net->dev_base_head); 7358 7359 net->dev_name_head = netdev_create_hash(); 7360 if (net->dev_name_head == NULL) 7361 goto err_name; 7362 7363 net->dev_index_head = netdev_create_hash(); 7364 if (net->dev_index_head == NULL) 7365 goto err_idx; 7366 7367 return 0; 7368 7369 err_idx: 7370 kfree(net->dev_name_head); 7371 err_name: 7372 return -ENOMEM; 7373 } 7374 7375 /** 7376 * netdev_drivername - network driver for the device 7377 * @dev: network device 7378 * 7379 * Determine network driver for device. 7380 */ 7381 const char *netdev_drivername(const struct net_device *dev) 7382 { 7383 const struct device_driver *driver; 7384 const struct device *parent; 7385 const char *empty = ""; 7386 7387 parent = dev->dev.parent; 7388 if (!parent) 7389 return empty; 7390 7391 driver = parent->driver; 7392 if (driver && driver->name) 7393 return driver->name; 7394 return empty; 7395 } 7396 7397 static void __netdev_printk(const char *level, const struct net_device *dev, 7398 struct va_format *vaf) 7399 { 7400 if (dev && dev->dev.parent) { 7401 dev_printk_emit(level[1] - '0', 7402 dev->dev.parent, 7403 "%s %s %s%s: %pV", 7404 dev_driver_string(dev->dev.parent), 7405 dev_name(dev->dev.parent), 7406 netdev_name(dev), netdev_reg_state(dev), 7407 vaf); 7408 } else if (dev) { 7409 printk("%s%s%s: %pV", 7410 level, netdev_name(dev), netdev_reg_state(dev), vaf); 7411 } else { 7412 printk("%s(NULL net_device): %pV", level, vaf); 7413 } 7414 } 7415 7416 void netdev_printk(const char *level, const struct net_device *dev, 7417 const char *format, ...) 7418 { 7419 struct va_format vaf; 7420 va_list args; 7421 7422 va_start(args, format); 7423 7424 vaf.fmt = format; 7425 vaf.va = &args; 7426 7427 __netdev_printk(level, dev, &vaf); 7428 7429 va_end(args); 7430 } 7431 EXPORT_SYMBOL(netdev_printk); 7432 7433 #define define_netdev_printk_level(func, level) \ 7434 void func(const struct net_device *dev, const char *fmt, ...) \ 7435 { \ 7436 struct va_format vaf; \ 7437 va_list args; \ 7438 \ 7439 va_start(args, fmt); \ 7440 \ 7441 vaf.fmt = fmt; \ 7442 vaf.va = &args; \ 7443 \ 7444 __netdev_printk(level, dev, &vaf); \ 7445 \ 7446 va_end(args); \ 7447 } \ 7448 EXPORT_SYMBOL(func); 7449 7450 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 7451 define_netdev_printk_level(netdev_alert, KERN_ALERT); 7452 define_netdev_printk_level(netdev_crit, KERN_CRIT); 7453 define_netdev_printk_level(netdev_err, KERN_ERR); 7454 define_netdev_printk_level(netdev_warn, KERN_WARNING); 7455 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 7456 define_netdev_printk_level(netdev_info, KERN_INFO); 7457 7458 static void __net_exit netdev_exit(struct net *net) 7459 { 7460 kfree(net->dev_name_head); 7461 kfree(net->dev_index_head); 7462 } 7463 7464 static struct pernet_operations __net_initdata netdev_net_ops = { 7465 .init = netdev_init, 7466 .exit = netdev_exit, 7467 }; 7468 7469 static void __net_exit default_device_exit(struct net *net) 7470 { 7471 struct net_device *dev, *aux; 7472 /* 7473 * Push all migratable network devices back to the 7474 * initial network namespace 7475 */ 7476 rtnl_lock(); 7477 for_each_netdev_safe(net, dev, aux) { 7478 int err; 7479 char fb_name[IFNAMSIZ]; 7480 7481 /* Ignore unmoveable devices (i.e. loopback) */ 7482 if (dev->features & NETIF_F_NETNS_LOCAL) 7483 continue; 7484 7485 /* Leave virtual devices for the generic cleanup */ 7486 if (dev->rtnl_link_ops) 7487 continue; 7488 7489 /* Push remaining network devices to init_net */ 7490 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 7491 err = dev_change_net_namespace(dev, &init_net, fb_name); 7492 if (err) { 7493 pr_emerg("%s: failed to move %s to init_net: %d\n", 7494 __func__, dev->name, err); 7495 BUG(); 7496 } 7497 } 7498 rtnl_unlock(); 7499 } 7500 7501 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 7502 { 7503 /* Return with the rtnl_lock held when there are no network 7504 * devices unregistering in any network namespace in net_list. 7505 */ 7506 struct net *net; 7507 bool unregistering; 7508 DEFINE_WAIT_FUNC(wait, woken_wake_function); 7509 7510 add_wait_queue(&netdev_unregistering_wq, &wait); 7511 for (;;) { 7512 unregistering = false; 7513 rtnl_lock(); 7514 list_for_each_entry(net, net_list, exit_list) { 7515 if (net->dev_unreg_count > 0) { 7516 unregistering = true; 7517 break; 7518 } 7519 } 7520 if (!unregistering) 7521 break; 7522 __rtnl_unlock(); 7523 7524 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 7525 } 7526 remove_wait_queue(&netdev_unregistering_wq, &wait); 7527 } 7528 7529 static void __net_exit default_device_exit_batch(struct list_head *net_list) 7530 { 7531 /* At exit all network devices most be removed from a network 7532 * namespace. Do this in the reverse order of registration. 7533 * Do this across as many network namespaces as possible to 7534 * improve batching efficiency. 7535 */ 7536 struct net_device *dev; 7537 struct net *net; 7538 LIST_HEAD(dev_kill_list); 7539 7540 /* To prevent network device cleanup code from dereferencing 7541 * loopback devices or network devices that have been freed 7542 * wait here for all pending unregistrations to complete, 7543 * before unregistring the loopback device and allowing the 7544 * network namespace be freed. 7545 * 7546 * The netdev todo list containing all network devices 7547 * unregistrations that happen in default_device_exit_batch 7548 * will run in the rtnl_unlock() at the end of 7549 * default_device_exit_batch. 7550 */ 7551 rtnl_lock_unregistering(net_list); 7552 list_for_each_entry(net, net_list, exit_list) { 7553 for_each_netdev_reverse(net, dev) { 7554 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 7555 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 7556 else 7557 unregister_netdevice_queue(dev, &dev_kill_list); 7558 } 7559 } 7560 unregister_netdevice_many(&dev_kill_list); 7561 rtnl_unlock(); 7562 } 7563 7564 static struct pernet_operations __net_initdata default_device_ops = { 7565 .exit = default_device_exit, 7566 .exit_batch = default_device_exit_batch, 7567 }; 7568 7569 /* 7570 * Initialize the DEV module. At boot time this walks the device list and 7571 * unhooks any devices that fail to initialise (normally hardware not 7572 * present) and leaves us with a valid list of present and active devices. 7573 * 7574 */ 7575 7576 /* 7577 * This is called single threaded during boot, so no need 7578 * to take the rtnl semaphore. 7579 */ 7580 static int __init net_dev_init(void) 7581 { 7582 int i, rc = -ENOMEM; 7583 7584 BUG_ON(!dev_boot_phase); 7585 7586 if (dev_proc_init()) 7587 goto out; 7588 7589 if (netdev_kobject_init()) 7590 goto out; 7591 7592 INIT_LIST_HEAD(&ptype_all); 7593 for (i = 0; i < PTYPE_HASH_SIZE; i++) 7594 INIT_LIST_HEAD(&ptype_base[i]); 7595 7596 INIT_LIST_HEAD(&offload_base); 7597 7598 if (register_pernet_subsys(&netdev_net_ops)) 7599 goto out; 7600 7601 /* 7602 * Initialise the packet receive queues. 7603 */ 7604 7605 for_each_possible_cpu(i) { 7606 struct softnet_data *sd = &per_cpu(softnet_data, i); 7607 7608 skb_queue_head_init(&sd->input_pkt_queue); 7609 skb_queue_head_init(&sd->process_queue); 7610 INIT_LIST_HEAD(&sd->poll_list); 7611 sd->output_queue_tailp = &sd->output_queue; 7612 #ifdef CONFIG_RPS 7613 sd->csd.func = rps_trigger_softirq; 7614 sd->csd.info = sd; 7615 sd->cpu = i; 7616 #endif 7617 7618 sd->backlog.poll = process_backlog; 7619 sd->backlog.weight = weight_p; 7620 } 7621 7622 dev_boot_phase = 0; 7623 7624 /* The loopback device is special if any other network devices 7625 * is present in a network namespace the loopback device must 7626 * be present. Since we now dynamically allocate and free the 7627 * loopback device ensure this invariant is maintained by 7628 * keeping the loopback device as the first device on the 7629 * list of network devices. Ensuring the loopback devices 7630 * is the first device that appears and the last network device 7631 * that disappears. 7632 */ 7633 if (register_pernet_device(&loopback_net_ops)) 7634 goto out; 7635 7636 if (register_pernet_device(&default_device_ops)) 7637 goto out; 7638 7639 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 7640 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 7641 7642 hotcpu_notifier(dev_cpu_callback, 0); 7643 dst_init(); 7644 rc = 0; 7645 out: 7646 return rc; 7647 } 7648 7649 subsys_initcall(net_dev_init); 7650