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 <linux/bpf.h> 98 #include <net/net_namespace.h> 99 #include <net/sock.h> 100 #include <net/busy_poll.h> 101 #include <linux/rtnetlink.h> 102 #include <linux/stat.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/pkt_sched.h> 106 #include <net/checksum.h> 107 #include <net/xfrm.h> 108 #include <linux/highmem.h> 109 #include <linux/init.h> 110 #include <linux/module.h> 111 #include <linux/netpoll.h> 112 #include <linux/rcupdate.h> 113 #include <linux/delay.h> 114 #include <net/iw_handler.h> 115 #include <asm/current.h> 116 #include <linux/audit.h> 117 #include <linux/dmaengine.h> 118 #include <linux/err.h> 119 #include <linux/ctype.h> 120 #include <linux/if_arp.h> 121 #include <linux/if_vlan.h> 122 #include <linux/ip.h> 123 #include <net/ip.h> 124 #include <net/mpls.h> 125 #include <linux/ipv6.h> 126 #include <linux/in.h> 127 #include <linux/jhash.h> 128 #include <linux/random.h> 129 #include <trace/events/napi.h> 130 #include <trace/events/net.h> 131 #include <trace/events/skb.h> 132 #include <linux/pci.h> 133 #include <linux/inetdevice.h> 134 #include <linux/cpu_rmap.h> 135 #include <linux/static_key.h> 136 #include <linux/hashtable.h> 137 #include <linux/vmalloc.h> 138 #include <linux/if_macvlan.h> 139 #include <linux/errqueue.h> 140 #include <linux/hrtimer.h> 141 #include <linux/netfilter_ingress.h> 142 #include <linux/sctp.h> 143 #include <linux/crash_dump.h> 144 145 #include "net-sysfs.h" 146 147 /* Instead of increasing this, you should create a hash table. */ 148 #define MAX_GRO_SKBS 8 149 150 /* This should be increased if a protocol with a bigger head is added. */ 151 #define GRO_MAX_HEAD (MAX_HEADER + 128) 152 153 static DEFINE_SPINLOCK(ptype_lock); 154 static DEFINE_SPINLOCK(offload_lock); 155 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 156 struct list_head ptype_all __read_mostly; /* Taps */ 157 static struct list_head offload_base __read_mostly; 158 159 static int netif_rx_internal(struct sk_buff *skb); 160 static int call_netdevice_notifiers_info(unsigned long val, 161 struct net_device *dev, 162 struct netdev_notifier_info *info); 163 164 /* 165 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 166 * semaphore. 167 * 168 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 169 * 170 * Writers must hold the rtnl semaphore while they loop through the 171 * dev_base_head list, and hold dev_base_lock for writing when they do the 172 * actual updates. This allows pure readers to access the list even 173 * while a writer is preparing to update it. 174 * 175 * To put it another way, dev_base_lock is held for writing only to 176 * protect against pure readers; the rtnl semaphore provides the 177 * protection against other writers. 178 * 179 * See, for example usages, register_netdevice() and 180 * unregister_netdevice(), which must be called with the rtnl 181 * semaphore held. 182 */ 183 DEFINE_RWLOCK(dev_base_lock); 184 EXPORT_SYMBOL(dev_base_lock); 185 186 /* protects napi_hash addition/deletion and napi_gen_id */ 187 static DEFINE_SPINLOCK(napi_hash_lock); 188 189 static unsigned int napi_gen_id = NR_CPUS; 190 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 191 192 static seqcount_t devnet_rename_seq; 193 194 static inline void dev_base_seq_inc(struct net *net) 195 { 196 while (++net->dev_base_seq == 0); 197 } 198 199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 200 { 201 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 202 203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 204 } 205 206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 207 { 208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 209 } 210 211 static inline void rps_lock(struct softnet_data *sd) 212 { 213 #ifdef CONFIG_RPS 214 spin_lock(&sd->input_pkt_queue.lock); 215 #endif 216 } 217 218 static inline void rps_unlock(struct softnet_data *sd) 219 { 220 #ifdef CONFIG_RPS 221 spin_unlock(&sd->input_pkt_queue.lock); 222 #endif 223 } 224 225 /* Device list insertion */ 226 static void list_netdevice(struct net_device *dev) 227 { 228 struct net *net = dev_net(dev); 229 230 ASSERT_RTNL(); 231 232 write_lock_bh(&dev_base_lock); 233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 235 hlist_add_head_rcu(&dev->index_hlist, 236 dev_index_hash(net, dev->ifindex)); 237 write_unlock_bh(&dev_base_lock); 238 239 dev_base_seq_inc(net); 240 } 241 242 /* Device list removal 243 * caller must respect a RCU grace period before freeing/reusing dev 244 */ 245 static void unlist_netdevice(struct net_device *dev) 246 { 247 ASSERT_RTNL(); 248 249 /* Unlink dev from the device chain */ 250 write_lock_bh(&dev_base_lock); 251 list_del_rcu(&dev->dev_list); 252 hlist_del_rcu(&dev->name_hlist); 253 hlist_del_rcu(&dev->index_hlist); 254 write_unlock_bh(&dev_base_lock); 255 256 dev_base_seq_inc(dev_net(dev)); 257 } 258 259 /* 260 * Our notifier list 261 */ 262 263 static RAW_NOTIFIER_HEAD(netdev_chain); 264 265 /* 266 * Device drivers call our routines to queue packets here. We empty the 267 * queue in the local softnet handler. 268 */ 269 270 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 271 EXPORT_PER_CPU_SYMBOL(softnet_data); 272 273 #ifdef CONFIG_LOCKDEP 274 /* 275 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 276 * according to dev->type 277 */ 278 static const unsigned short netdev_lock_type[] = 279 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 280 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 281 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 282 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 283 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 284 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 285 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 286 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 287 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 288 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 289 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 290 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 291 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 292 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 293 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 294 295 static const char *const netdev_lock_name[] = 296 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 308 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 309 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 310 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 311 312 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 313 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 314 315 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 316 { 317 int i; 318 319 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 320 if (netdev_lock_type[i] == dev_type) 321 return i; 322 /* the last key is used by default */ 323 return ARRAY_SIZE(netdev_lock_type) - 1; 324 } 325 326 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 327 unsigned short dev_type) 328 { 329 int i; 330 331 i = netdev_lock_pos(dev_type); 332 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 333 netdev_lock_name[i]); 334 } 335 336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 337 { 338 int i; 339 340 i = netdev_lock_pos(dev->type); 341 lockdep_set_class_and_name(&dev->addr_list_lock, 342 &netdev_addr_lock_key[i], 343 netdev_lock_name[i]); 344 } 345 #else 346 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 347 unsigned short dev_type) 348 { 349 } 350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 351 { 352 } 353 #endif 354 355 /******************************************************************************* 356 357 Protocol management and registration routines 358 359 *******************************************************************************/ 360 361 /* 362 * Add a protocol ID to the list. Now that the input handler is 363 * smarter we can dispense with all the messy stuff that used to be 364 * here. 365 * 366 * BEWARE!!! Protocol handlers, mangling input packets, 367 * MUST BE last in hash buckets and checking protocol handlers 368 * MUST start from promiscuous ptype_all chain in net_bh. 369 * It is true now, do not change it. 370 * Explanation follows: if protocol handler, mangling packet, will 371 * be the first on list, it is not able to sense, that packet 372 * is cloned and should be copied-on-write, so that it will 373 * change it and subsequent readers will get broken packet. 374 * --ANK (980803) 375 */ 376 377 static inline struct list_head *ptype_head(const struct packet_type *pt) 378 { 379 if (pt->type == htons(ETH_P_ALL)) 380 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 381 else 382 return pt->dev ? &pt->dev->ptype_specific : 383 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 384 } 385 386 /** 387 * dev_add_pack - add packet handler 388 * @pt: packet type declaration 389 * 390 * Add a protocol handler to the networking stack. The passed &packet_type 391 * is linked into kernel lists and may not be freed until it has been 392 * removed from the kernel lists. 393 * 394 * This call does not sleep therefore it can not 395 * guarantee all CPU's that are in middle of receiving packets 396 * will see the new packet type (until the next received packet). 397 */ 398 399 void dev_add_pack(struct packet_type *pt) 400 { 401 struct list_head *head = ptype_head(pt); 402 403 spin_lock(&ptype_lock); 404 list_add_rcu(&pt->list, head); 405 spin_unlock(&ptype_lock); 406 } 407 EXPORT_SYMBOL(dev_add_pack); 408 409 /** 410 * __dev_remove_pack - remove packet handler 411 * @pt: packet type declaration 412 * 413 * Remove a protocol handler that was previously added to the kernel 414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 415 * from the kernel lists and can be freed or reused once this function 416 * returns. 417 * 418 * The packet type might still be in use by receivers 419 * and must not be freed until after all the CPU's have gone 420 * through a quiescent state. 421 */ 422 void __dev_remove_pack(struct packet_type *pt) 423 { 424 struct list_head *head = ptype_head(pt); 425 struct packet_type *pt1; 426 427 spin_lock(&ptype_lock); 428 429 list_for_each_entry(pt1, head, list) { 430 if (pt == pt1) { 431 list_del_rcu(&pt->list); 432 goto out; 433 } 434 } 435 436 pr_warn("dev_remove_pack: %p not found\n", pt); 437 out: 438 spin_unlock(&ptype_lock); 439 } 440 EXPORT_SYMBOL(__dev_remove_pack); 441 442 /** 443 * dev_remove_pack - remove packet handler 444 * @pt: packet type declaration 445 * 446 * Remove a protocol handler that was previously added to the kernel 447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 448 * from the kernel lists and can be freed or reused once this function 449 * returns. 450 * 451 * This call sleeps to guarantee that no CPU is looking at the packet 452 * type after return. 453 */ 454 void dev_remove_pack(struct packet_type *pt) 455 { 456 __dev_remove_pack(pt); 457 458 synchronize_net(); 459 } 460 EXPORT_SYMBOL(dev_remove_pack); 461 462 463 /** 464 * dev_add_offload - register offload handlers 465 * @po: protocol offload declaration 466 * 467 * Add protocol offload handlers to the networking stack. The passed 468 * &proto_offload is linked into kernel lists and may not be freed until 469 * it has been removed from the kernel lists. 470 * 471 * This call does not sleep therefore it can not 472 * guarantee all CPU's that are in middle of receiving packets 473 * will see the new offload handlers (until the next received packet). 474 */ 475 void dev_add_offload(struct packet_offload *po) 476 { 477 struct packet_offload *elem; 478 479 spin_lock(&offload_lock); 480 list_for_each_entry(elem, &offload_base, list) { 481 if (po->priority < elem->priority) 482 break; 483 } 484 list_add_rcu(&po->list, elem->list.prev); 485 spin_unlock(&offload_lock); 486 } 487 EXPORT_SYMBOL(dev_add_offload); 488 489 /** 490 * __dev_remove_offload - remove offload handler 491 * @po: packet offload declaration 492 * 493 * Remove a protocol offload handler that was previously added to the 494 * kernel offload handlers by dev_add_offload(). The passed &offload_type 495 * is removed from the kernel lists and can be freed or reused once this 496 * function returns. 497 * 498 * The packet type might still be in use by receivers 499 * and must not be freed until after all the CPU's have gone 500 * through a quiescent state. 501 */ 502 static void __dev_remove_offload(struct packet_offload *po) 503 { 504 struct list_head *head = &offload_base; 505 struct packet_offload *po1; 506 507 spin_lock(&offload_lock); 508 509 list_for_each_entry(po1, head, list) { 510 if (po == po1) { 511 list_del_rcu(&po->list); 512 goto out; 513 } 514 } 515 516 pr_warn("dev_remove_offload: %p not found\n", po); 517 out: 518 spin_unlock(&offload_lock); 519 } 520 521 /** 522 * dev_remove_offload - remove packet offload handler 523 * @po: packet offload declaration 524 * 525 * Remove a packet offload handler that was previously added to the kernel 526 * offload handlers by dev_add_offload(). The passed &offload_type is 527 * removed from the kernel lists and can be freed or reused once this 528 * function returns. 529 * 530 * This call sleeps to guarantee that no CPU is looking at the packet 531 * type after return. 532 */ 533 void dev_remove_offload(struct packet_offload *po) 534 { 535 __dev_remove_offload(po); 536 537 synchronize_net(); 538 } 539 EXPORT_SYMBOL(dev_remove_offload); 540 541 /****************************************************************************** 542 543 Device Boot-time Settings Routines 544 545 *******************************************************************************/ 546 547 /* Boot time configuration table */ 548 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 549 550 /** 551 * netdev_boot_setup_add - add new setup entry 552 * @name: name of the device 553 * @map: configured settings for the device 554 * 555 * Adds new setup entry to the dev_boot_setup list. The function 556 * returns 0 on error and 1 on success. This is a generic routine to 557 * all netdevices. 558 */ 559 static int netdev_boot_setup_add(char *name, struct ifmap *map) 560 { 561 struct netdev_boot_setup *s; 562 int i; 563 564 s = dev_boot_setup; 565 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 566 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 567 memset(s[i].name, 0, sizeof(s[i].name)); 568 strlcpy(s[i].name, name, IFNAMSIZ); 569 memcpy(&s[i].map, map, sizeof(s[i].map)); 570 break; 571 } 572 } 573 574 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 575 } 576 577 /** 578 * netdev_boot_setup_check - check boot time settings 579 * @dev: the netdevice 580 * 581 * Check boot time settings for the device. 582 * The found settings are set for the device to be used 583 * later in the device probing. 584 * Returns 0 if no settings found, 1 if they are. 585 */ 586 int netdev_boot_setup_check(struct net_device *dev) 587 { 588 struct netdev_boot_setup *s = dev_boot_setup; 589 int i; 590 591 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 592 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 593 !strcmp(dev->name, s[i].name)) { 594 dev->irq = s[i].map.irq; 595 dev->base_addr = s[i].map.base_addr; 596 dev->mem_start = s[i].map.mem_start; 597 dev->mem_end = s[i].map.mem_end; 598 return 1; 599 } 600 } 601 return 0; 602 } 603 EXPORT_SYMBOL(netdev_boot_setup_check); 604 605 606 /** 607 * netdev_boot_base - get address from boot time settings 608 * @prefix: prefix for network device 609 * @unit: id for network device 610 * 611 * Check boot time settings for the base address of device. 612 * The found settings are set for the device to be used 613 * later in the device probing. 614 * Returns 0 if no settings found. 615 */ 616 unsigned long netdev_boot_base(const char *prefix, int unit) 617 { 618 const struct netdev_boot_setup *s = dev_boot_setup; 619 char name[IFNAMSIZ]; 620 int i; 621 622 sprintf(name, "%s%d", prefix, unit); 623 624 /* 625 * If device already registered then return base of 1 626 * to indicate not to probe for this interface 627 */ 628 if (__dev_get_by_name(&init_net, name)) 629 return 1; 630 631 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 632 if (!strcmp(name, s[i].name)) 633 return s[i].map.base_addr; 634 return 0; 635 } 636 637 /* 638 * Saves at boot time configured settings for any netdevice. 639 */ 640 int __init netdev_boot_setup(char *str) 641 { 642 int ints[5]; 643 struct ifmap map; 644 645 str = get_options(str, ARRAY_SIZE(ints), ints); 646 if (!str || !*str) 647 return 0; 648 649 /* Save settings */ 650 memset(&map, 0, sizeof(map)); 651 if (ints[0] > 0) 652 map.irq = ints[1]; 653 if (ints[0] > 1) 654 map.base_addr = ints[2]; 655 if (ints[0] > 2) 656 map.mem_start = ints[3]; 657 if (ints[0] > 3) 658 map.mem_end = ints[4]; 659 660 /* Add new entry to the list */ 661 return netdev_boot_setup_add(str, &map); 662 } 663 664 __setup("netdev=", netdev_boot_setup); 665 666 /******************************************************************************* 667 668 Device Interface Subroutines 669 670 *******************************************************************************/ 671 672 /** 673 * dev_get_iflink - get 'iflink' value of a interface 674 * @dev: targeted interface 675 * 676 * Indicates the ifindex the interface is linked to. 677 * Physical interfaces have the same 'ifindex' and 'iflink' values. 678 */ 679 680 int dev_get_iflink(const struct net_device *dev) 681 { 682 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 683 return dev->netdev_ops->ndo_get_iflink(dev); 684 685 return dev->ifindex; 686 } 687 EXPORT_SYMBOL(dev_get_iflink); 688 689 /** 690 * dev_fill_metadata_dst - Retrieve tunnel egress information. 691 * @dev: targeted interface 692 * @skb: The packet. 693 * 694 * For better visibility of tunnel traffic OVS needs to retrieve 695 * egress tunnel information for a packet. Following API allows 696 * user to get this info. 697 */ 698 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 699 { 700 struct ip_tunnel_info *info; 701 702 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 703 return -EINVAL; 704 705 info = skb_tunnel_info_unclone(skb); 706 if (!info) 707 return -ENOMEM; 708 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 709 return -EINVAL; 710 711 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 712 } 713 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 714 715 /** 716 * __dev_get_by_name - find a device by its name 717 * @net: the applicable net namespace 718 * @name: name to find 719 * 720 * Find an interface by name. Must be called under RTNL semaphore 721 * or @dev_base_lock. If the name is found a pointer to the device 722 * is returned. If the name is not found then %NULL is returned. The 723 * reference counters are not incremented so the caller must be 724 * careful with locks. 725 */ 726 727 struct net_device *__dev_get_by_name(struct net *net, const char *name) 728 { 729 struct net_device *dev; 730 struct hlist_head *head = dev_name_hash(net, name); 731 732 hlist_for_each_entry(dev, head, name_hlist) 733 if (!strncmp(dev->name, name, IFNAMSIZ)) 734 return dev; 735 736 return NULL; 737 } 738 EXPORT_SYMBOL(__dev_get_by_name); 739 740 /** 741 * dev_get_by_name_rcu - find a device by its name 742 * @net: the applicable net namespace 743 * @name: name to find 744 * 745 * Find an interface by name. 746 * If the name is found a pointer to the device is returned. 747 * If the name is not found then %NULL is returned. 748 * The reference counters are not incremented so the caller must be 749 * careful with locks. The caller must hold RCU lock. 750 */ 751 752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 753 { 754 struct net_device *dev; 755 struct hlist_head *head = dev_name_hash(net, name); 756 757 hlist_for_each_entry_rcu(dev, head, name_hlist) 758 if (!strncmp(dev->name, name, IFNAMSIZ)) 759 return dev; 760 761 return NULL; 762 } 763 EXPORT_SYMBOL(dev_get_by_name_rcu); 764 765 /** 766 * dev_get_by_name - find a device by its name 767 * @net: the applicable net namespace 768 * @name: name to find 769 * 770 * Find an interface by name. This can be called from any 771 * context and does its own locking. The returned handle has 772 * the usage count incremented and the caller must use dev_put() to 773 * release it when it is no longer needed. %NULL is returned if no 774 * matching device is found. 775 */ 776 777 struct net_device *dev_get_by_name(struct net *net, const char *name) 778 { 779 struct net_device *dev; 780 781 rcu_read_lock(); 782 dev = dev_get_by_name_rcu(net, name); 783 if (dev) 784 dev_hold(dev); 785 rcu_read_unlock(); 786 return dev; 787 } 788 EXPORT_SYMBOL(dev_get_by_name); 789 790 /** 791 * __dev_get_by_index - find a device by its ifindex 792 * @net: the applicable net namespace 793 * @ifindex: index of device 794 * 795 * Search for an interface by index. Returns %NULL if the device 796 * is not found or a pointer to the device. The device has not 797 * had its reference counter increased so the caller must be careful 798 * about locking. The caller must hold either the RTNL semaphore 799 * or @dev_base_lock. 800 */ 801 802 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 803 { 804 struct net_device *dev; 805 struct hlist_head *head = dev_index_hash(net, ifindex); 806 807 hlist_for_each_entry(dev, head, index_hlist) 808 if (dev->ifindex == ifindex) 809 return dev; 810 811 return NULL; 812 } 813 EXPORT_SYMBOL(__dev_get_by_index); 814 815 /** 816 * dev_get_by_index_rcu - find a device by its ifindex 817 * @net: the applicable net namespace 818 * @ifindex: index of device 819 * 820 * Search for an interface by index. Returns %NULL if the device 821 * is not found or a pointer to the device. The device has not 822 * had its reference counter increased so the caller must be careful 823 * about locking. The caller must hold RCU lock. 824 */ 825 826 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 827 { 828 struct net_device *dev; 829 struct hlist_head *head = dev_index_hash(net, ifindex); 830 831 hlist_for_each_entry_rcu(dev, head, index_hlist) 832 if (dev->ifindex == ifindex) 833 return dev; 834 835 return NULL; 836 } 837 EXPORT_SYMBOL(dev_get_by_index_rcu); 838 839 840 /** 841 * dev_get_by_index - find a device by its ifindex 842 * @net: the applicable net namespace 843 * @ifindex: index of device 844 * 845 * Search for an interface by index. Returns NULL if the device 846 * is not found or a pointer to the device. The device returned has 847 * had a reference added and the pointer is safe until the user calls 848 * dev_put to indicate they have finished with it. 849 */ 850 851 struct net_device *dev_get_by_index(struct net *net, int ifindex) 852 { 853 struct net_device *dev; 854 855 rcu_read_lock(); 856 dev = dev_get_by_index_rcu(net, ifindex); 857 if (dev) 858 dev_hold(dev); 859 rcu_read_unlock(); 860 return dev; 861 } 862 EXPORT_SYMBOL(dev_get_by_index); 863 864 /** 865 * netdev_get_name - get a netdevice name, knowing its ifindex. 866 * @net: network namespace 867 * @name: a pointer to the buffer where the name will be stored. 868 * @ifindex: the ifindex of the interface to get the name from. 869 * 870 * The use of raw_seqcount_begin() and cond_resched() before 871 * retrying is required as we want to give the writers a chance 872 * to complete when CONFIG_PREEMPT is not set. 873 */ 874 int netdev_get_name(struct net *net, char *name, int ifindex) 875 { 876 struct net_device *dev; 877 unsigned int seq; 878 879 retry: 880 seq = raw_seqcount_begin(&devnet_rename_seq); 881 rcu_read_lock(); 882 dev = dev_get_by_index_rcu(net, ifindex); 883 if (!dev) { 884 rcu_read_unlock(); 885 return -ENODEV; 886 } 887 888 strcpy(name, dev->name); 889 rcu_read_unlock(); 890 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 891 cond_resched(); 892 goto retry; 893 } 894 895 return 0; 896 } 897 898 /** 899 * dev_getbyhwaddr_rcu - find a device by its hardware address 900 * @net: the applicable net namespace 901 * @type: media type of device 902 * @ha: hardware address 903 * 904 * Search for an interface by MAC address. Returns NULL if the device 905 * is not found or a pointer to the device. 906 * The caller must hold RCU or RTNL. 907 * The returned device has not had its ref count increased 908 * and the caller must therefore be careful about locking 909 * 910 */ 911 912 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 913 const char *ha) 914 { 915 struct net_device *dev; 916 917 for_each_netdev_rcu(net, dev) 918 if (dev->type == type && 919 !memcmp(dev->dev_addr, ha, dev->addr_len)) 920 return dev; 921 922 return NULL; 923 } 924 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 925 926 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 927 { 928 struct net_device *dev; 929 930 ASSERT_RTNL(); 931 for_each_netdev(net, dev) 932 if (dev->type == type) 933 return dev; 934 935 return NULL; 936 } 937 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 938 939 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 940 { 941 struct net_device *dev, *ret = NULL; 942 943 rcu_read_lock(); 944 for_each_netdev_rcu(net, dev) 945 if (dev->type == type) { 946 dev_hold(dev); 947 ret = dev; 948 break; 949 } 950 rcu_read_unlock(); 951 return ret; 952 } 953 EXPORT_SYMBOL(dev_getfirstbyhwtype); 954 955 /** 956 * __dev_get_by_flags - find any device with given flags 957 * @net: the applicable net namespace 958 * @if_flags: IFF_* values 959 * @mask: bitmask of bits in if_flags to check 960 * 961 * Search for any interface with the given flags. Returns NULL if a device 962 * is not found or a pointer to the device. Must be called inside 963 * rtnl_lock(), and result refcount is unchanged. 964 */ 965 966 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 967 unsigned short mask) 968 { 969 struct net_device *dev, *ret; 970 971 ASSERT_RTNL(); 972 973 ret = NULL; 974 for_each_netdev(net, dev) { 975 if (((dev->flags ^ if_flags) & mask) == 0) { 976 ret = dev; 977 break; 978 } 979 } 980 return ret; 981 } 982 EXPORT_SYMBOL(__dev_get_by_flags); 983 984 /** 985 * dev_valid_name - check if name is okay for network device 986 * @name: name string 987 * 988 * Network device names need to be valid file names to 989 * to allow sysfs to work. We also disallow any kind of 990 * whitespace. 991 */ 992 bool dev_valid_name(const char *name) 993 { 994 if (*name == '\0') 995 return false; 996 if (strlen(name) >= IFNAMSIZ) 997 return false; 998 if (!strcmp(name, ".") || !strcmp(name, "..")) 999 return false; 1000 1001 while (*name) { 1002 if (*name == '/' || *name == ':' || isspace(*name)) 1003 return false; 1004 name++; 1005 } 1006 return true; 1007 } 1008 EXPORT_SYMBOL(dev_valid_name); 1009 1010 /** 1011 * __dev_alloc_name - allocate a name for a device 1012 * @net: network namespace to allocate the device name in 1013 * @name: name format string 1014 * @buf: scratch buffer and result name string 1015 * 1016 * Passed a format string - eg "lt%d" it will try and find a suitable 1017 * id. It scans list of devices to build up a free map, then chooses 1018 * the first empty slot. The caller must hold the dev_base or rtnl lock 1019 * while allocating the name and adding the device in order to avoid 1020 * duplicates. 1021 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1022 * Returns the number of the unit assigned or a negative errno code. 1023 */ 1024 1025 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1026 { 1027 int i = 0; 1028 const char *p; 1029 const int max_netdevices = 8*PAGE_SIZE; 1030 unsigned long *inuse; 1031 struct net_device *d; 1032 1033 p = strnchr(name, IFNAMSIZ-1, '%'); 1034 if (p) { 1035 /* 1036 * Verify the string as this thing may have come from 1037 * the user. There must be either one "%d" and no other "%" 1038 * characters. 1039 */ 1040 if (p[1] != 'd' || strchr(p + 2, '%')) 1041 return -EINVAL; 1042 1043 /* Use one page as a bit array of possible slots */ 1044 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1045 if (!inuse) 1046 return -ENOMEM; 1047 1048 for_each_netdev(net, d) { 1049 if (!sscanf(d->name, name, &i)) 1050 continue; 1051 if (i < 0 || i >= max_netdevices) 1052 continue; 1053 1054 /* avoid cases where sscanf is not exact inverse of printf */ 1055 snprintf(buf, IFNAMSIZ, name, i); 1056 if (!strncmp(buf, d->name, IFNAMSIZ)) 1057 set_bit(i, inuse); 1058 } 1059 1060 i = find_first_zero_bit(inuse, max_netdevices); 1061 free_page((unsigned long) inuse); 1062 } 1063 1064 if (buf != name) 1065 snprintf(buf, IFNAMSIZ, name, i); 1066 if (!__dev_get_by_name(net, buf)) 1067 return i; 1068 1069 /* It is possible to run out of possible slots 1070 * when the name is long and there isn't enough space left 1071 * for the digits, or if all bits are used. 1072 */ 1073 return -ENFILE; 1074 } 1075 1076 /** 1077 * dev_alloc_name - allocate a name for a device 1078 * @dev: device 1079 * @name: name format string 1080 * 1081 * Passed a format string - eg "lt%d" it will try and find a suitable 1082 * id. It scans list of devices to build up a free map, then chooses 1083 * the first empty slot. The caller must hold the dev_base or rtnl lock 1084 * while allocating the name and adding the device in order to avoid 1085 * duplicates. 1086 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1087 * Returns the number of the unit assigned or a negative errno code. 1088 */ 1089 1090 int dev_alloc_name(struct net_device *dev, const char *name) 1091 { 1092 char buf[IFNAMSIZ]; 1093 struct net *net; 1094 int ret; 1095 1096 BUG_ON(!dev_net(dev)); 1097 net = dev_net(dev); 1098 ret = __dev_alloc_name(net, name, buf); 1099 if (ret >= 0) 1100 strlcpy(dev->name, buf, IFNAMSIZ); 1101 return ret; 1102 } 1103 EXPORT_SYMBOL(dev_alloc_name); 1104 1105 static int dev_alloc_name_ns(struct net *net, 1106 struct net_device *dev, 1107 const char *name) 1108 { 1109 char buf[IFNAMSIZ]; 1110 int ret; 1111 1112 ret = __dev_alloc_name(net, name, buf); 1113 if (ret >= 0) 1114 strlcpy(dev->name, buf, IFNAMSIZ); 1115 return ret; 1116 } 1117 1118 static int dev_get_valid_name(struct net *net, 1119 struct net_device *dev, 1120 const char *name) 1121 { 1122 BUG_ON(!net); 1123 1124 if (!dev_valid_name(name)) 1125 return -EINVAL; 1126 1127 if (strchr(name, '%')) 1128 return dev_alloc_name_ns(net, dev, name); 1129 else if (__dev_get_by_name(net, name)) 1130 return -EEXIST; 1131 else if (dev->name != name) 1132 strlcpy(dev->name, name, IFNAMSIZ); 1133 1134 return 0; 1135 } 1136 1137 /** 1138 * dev_change_name - change name of a device 1139 * @dev: device 1140 * @newname: name (or format string) must be at least IFNAMSIZ 1141 * 1142 * Change name of a device, can pass format strings "eth%d". 1143 * for wildcarding. 1144 */ 1145 int dev_change_name(struct net_device *dev, const char *newname) 1146 { 1147 unsigned char old_assign_type; 1148 char oldname[IFNAMSIZ]; 1149 int err = 0; 1150 int ret; 1151 struct net *net; 1152 1153 ASSERT_RTNL(); 1154 BUG_ON(!dev_net(dev)); 1155 1156 net = dev_net(dev); 1157 if (dev->flags & IFF_UP) 1158 return -EBUSY; 1159 1160 write_seqcount_begin(&devnet_rename_seq); 1161 1162 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1163 write_seqcount_end(&devnet_rename_seq); 1164 return 0; 1165 } 1166 1167 memcpy(oldname, dev->name, IFNAMSIZ); 1168 1169 err = dev_get_valid_name(net, dev, newname); 1170 if (err < 0) { 1171 write_seqcount_end(&devnet_rename_seq); 1172 return err; 1173 } 1174 1175 if (oldname[0] && !strchr(oldname, '%')) 1176 netdev_info(dev, "renamed from %s\n", oldname); 1177 1178 old_assign_type = dev->name_assign_type; 1179 dev->name_assign_type = NET_NAME_RENAMED; 1180 1181 rollback: 1182 ret = device_rename(&dev->dev, dev->name); 1183 if (ret) { 1184 memcpy(dev->name, oldname, IFNAMSIZ); 1185 dev->name_assign_type = old_assign_type; 1186 write_seqcount_end(&devnet_rename_seq); 1187 return ret; 1188 } 1189 1190 write_seqcount_end(&devnet_rename_seq); 1191 1192 netdev_adjacent_rename_links(dev, oldname); 1193 1194 write_lock_bh(&dev_base_lock); 1195 hlist_del_rcu(&dev->name_hlist); 1196 write_unlock_bh(&dev_base_lock); 1197 1198 synchronize_rcu(); 1199 1200 write_lock_bh(&dev_base_lock); 1201 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1202 write_unlock_bh(&dev_base_lock); 1203 1204 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1205 ret = notifier_to_errno(ret); 1206 1207 if (ret) { 1208 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1209 if (err >= 0) { 1210 err = ret; 1211 write_seqcount_begin(&devnet_rename_seq); 1212 memcpy(dev->name, oldname, IFNAMSIZ); 1213 memcpy(oldname, newname, IFNAMSIZ); 1214 dev->name_assign_type = old_assign_type; 1215 old_assign_type = NET_NAME_RENAMED; 1216 goto rollback; 1217 } else { 1218 pr_err("%s: name change rollback failed: %d\n", 1219 dev->name, ret); 1220 } 1221 } 1222 1223 return err; 1224 } 1225 1226 /** 1227 * dev_set_alias - change ifalias of a device 1228 * @dev: device 1229 * @alias: name up to IFALIASZ 1230 * @len: limit of bytes to copy from info 1231 * 1232 * Set ifalias for a device, 1233 */ 1234 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1235 { 1236 char *new_ifalias; 1237 1238 ASSERT_RTNL(); 1239 1240 if (len >= IFALIASZ) 1241 return -EINVAL; 1242 1243 if (!len) { 1244 kfree(dev->ifalias); 1245 dev->ifalias = NULL; 1246 return 0; 1247 } 1248 1249 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); 1250 if (!new_ifalias) 1251 return -ENOMEM; 1252 dev->ifalias = new_ifalias; 1253 1254 strlcpy(dev->ifalias, alias, len+1); 1255 return len; 1256 } 1257 1258 1259 /** 1260 * netdev_features_change - device changes features 1261 * @dev: device to cause notification 1262 * 1263 * Called to indicate a device has changed features. 1264 */ 1265 void netdev_features_change(struct net_device *dev) 1266 { 1267 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1268 } 1269 EXPORT_SYMBOL(netdev_features_change); 1270 1271 /** 1272 * netdev_state_change - device changes state 1273 * @dev: device to cause notification 1274 * 1275 * Called to indicate a device has changed state. This function calls 1276 * the notifier chains for netdev_chain and sends a NEWLINK message 1277 * to the routing socket. 1278 */ 1279 void netdev_state_change(struct net_device *dev) 1280 { 1281 if (dev->flags & IFF_UP) { 1282 struct netdev_notifier_change_info change_info; 1283 1284 change_info.flags_changed = 0; 1285 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 1286 &change_info.info); 1287 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1288 } 1289 } 1290 EXPORT_SYMBOL(netdev_state_change); 1291 1292 /** 1293 * netdev_notify_peers - notify network peers about existence of @dev 1294 * @dev: network device 1295 * 1296 * Generate traffic such that interested network peers are aware of 1297 * @dev, such as by generating a gratuitous ARP. This may be used when 1298 * a device wants to inform the rest of the network about some sort of 1299 * reconfiguration such as a failover event or virtual machine 1300 * migration. 1301 */ 1302 void netdev_notify_peers(struct net_device *dev) 1303 { 1304 rtnl_lock(); 1305 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1306 rtnl_unlock(); 1307 } 1308 EXPORT_SYMBOL(netdev_notify_peers); 1309 1310 static int __dev_open(struct net_device *dev) 1311 { 1312 const struct net_device_ops *ops = dev->netdev_ops; 1313 int ret; 1314 1315 ASSERT_RTNL(); 1316 1317 if (!netif_device_present(dev)) 1318 return -ENODEV; 1319 1320 /* Block netpoll from trying to do any rx path servicing. 1321 * If we don't do this there is a chance ndo_poll_controller 1322 * or ndo_poll may be running while we open the device 1323 */ 1324 netpoll_poll_disable(dev); 1325 1326 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); 1327 ret = notifier_to_errno(ret); 1328 if (ret) 1329 return ret; 1330 1331 set_bit(__LINK_STATE_START, &dev->state); 1332 1333 if (ops->ndo_validate_addr) 1334 ret = ops->ndo_validate_addr(dev); 1335 1336 if (!ret && ops->ndo_open) 1337 ret = ops->ndo_open(dev); 1338 1339 netpoll_poll_enable(dev); 1340 1341 if (ret) 1342 clear_bit(__LINK_STATE_START, &dev->state); 1343 else { 1344 dev->flags |= IFF_UP; 1345 dev_set_rx_mode(dev); 1346 dev_activate(dev); 1347 add_device_randomness(dev->dev_addr, dev->addr_len); 1348 } 1349 1350 return ret; 1351 } 1352 1353 /** 1354 * dev_open - prepare an interface for use. 1355 * @dev: device to open 1356 * 1357 * Takes a device from down to up state. The device's private open 1358 * function is invoked and then the multicast lists are loaded. Finally 1359 * the device is moved into the up state and a %NETDEV_UP message is 1360 * sent to the netdev notifier chain. 1361 * 1362 * Calling this function on an active interface is a nop. On a failure 1363 * a negative errno code is returned. 1364 */ 1365 int dev_open(struct net_device *dev) 1366 { 1367 int ret; 1368 1369 if (dev->flags & IFF_UP) 1370 return 0; 1371 1372 ret = __dev_open(dev); 1373 if (ret < 0) 1374 return ret; 1375 1376 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1377 call_netdevice_notifiers(NETDEV_UP, dev); 1378 1379 return ret; 1380 } 1381 EXPORT_SYMBOL(dev_open); 1382 1383 static int __dev_close_many(struct list_head *head) 1384 { 1385 struct net_device *dev; 1386 1387 ASSERT_RTNL(); 1388 might_sleep(); 1389 1390 list_for_each_entry(dev, head, close_list) { 1391 /* Temporarily disable netpoll until the interface is down */ 1392 netpoll_poll_disable(dev); 1393 1394 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1395 1396 clear_bit(__LINK_STATE_START, &dev->state); 1397 1398 /* Synchronize to scheduled poll. We cannot touch poll list, it 1399 * can be even on different cpu. So just clear netif_running(). 1400 * 1401 * dev->stop() will invoke napi_disable() on all of it's 1402 * napi_struct instances on this device. 1403 */ 1404 smp_mb__after_atomic(); /* Commit netif_running(). */ 1405 } 1406 1407 dev_deactivate_many(head); 1408 1409 list_for_each_entry(dev, head, close_list) { 1410 const struct net_device_ops *ops = dev->netdev_ops; 1411 1412 /* 1413 * Call the device specific close. This cannot fail. 1414 * Only if device is UP 1415 * 1416 * We allow it to be called even after a DETACH hot-plug 1417 * event. 1418 */ 1419 if (ops->ndo_stop) 1420 ops->ndo_stop(dev); 1421 1422 dev->flags &= ~IFF_UP; 1423 netpoll_poll_enable(dev); 1424 } 1425 1426 return 0; 1427 } 1428 1429 static int __dev_close(struct net_device *dev) 1430 { 1431 int retval; 1432 LIST_HEAD(single); 1433 1434 list_add(&dev->close_list, &single); 1435 retval = __dev_close_many(&single); 1436 list_del(&single); 1437 1438 return retval; 1439 } 1440 1441 int dev_close_many(struct list_head *head, bool unlink) 1442 { 1443 struct net_device *dev, *tmp; 1444 1445 /* Remove the devices that don't need to be closed */ 1446 list_for_each_entry_safe(dev, tmp, head, close_list) 1447 if (!(dev->flags & IFF_UP)) 1448 list_del_init(&dev->close_list); 1449 1450 __dev_close_many(head); 1451 1452 list_for_each_entry_safe(dev, tmp, head, close_list) { 1453 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1454 call_netdevice_notifiers(NETDEV_DOWN, dev); 1455 if (unlink) 1456 list_del_init(&dev->close_list); 1457 } 1458 1459 return 0; 1460 } 1461 EXPORT_SYMBOL(dev_close_many); 1462 1463 /** 1464 * dev_close - shutdown an interface. 1465 * @dev: device to shutdown 1466 * 1467 * This function moves an active device into down state. A 1468 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1469 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1470 * chain. 1471 */ 1472 int dev_close(struct net_device *dev) 1473 { 1474 if (dev->flags & IFF_UP) { 1475 LIST_HEAD(single); 1476 1477 list_add(&dev->close_list, &single); 1478 dev_close_many(&single, true); 1479 list_del(&single); 1480 } 1481 return 0; 1482 } 1483 EXPORT_SYMBOL(dev_close); 1484 1485 1486 /** 1487 * dev_disable_lro - disable Large Receive Offload on a device 1488 * @dev: device 1489 * 1490 * Disable Large Receive Offload (LRO) on a net device. Must be 1491 * called under RTNL. This is needed if received packets may be 1492 * forwarded to another interface. 1493 */ 1494 void dev_disable_lro(struct net_device *dev) 1495 { 1496 struct net_device *lower_dev; 1497 struct list_head *iter; 1498 1499 dev->wanted_features &= ~NETIF_F_LRO; 1500 netdev_update_features(dev); 1501 1502 if (unlikely(dev->features & NETIF_F_LRO)) 1503 netdev_WARN(dev, "failed to disable LRO!\n"); 1504 1505 netdev_for_each_lower_dev(dev, lower_dev, iter) 1506 dev_disable_lro(lower_dev); 1507 } 1508 EXPORT_SYMBOL(dev_disable_lro); 1509 1510 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1511 struct net_device *dev) 1512 { 1513 struct netdev_notifier_info info; 1514 1515 netdev_notifier_info_init(&info, dev); 1516 return nb->notifier_call(nb, val, &info); 1517 } 1518 1519 static int dev_boot_phase = 1; 1520 1521 /** 1522 * register_netdevice_notifier - register a network notifier block 1523 * @nb: notifier 1524 * 1525 * Register a notifier to be called when network device events occur. 1526 * The notifier passed is linked into the kernel structures and must 1527 * not be reused until it has been unregistered. A negative errno code 1528 * is returned on a failure. 1529 * 1530 * When registered all registration and up events are replayed 1531 * to the new notifier to allow device to have a race free 1532 * view of the network device list. 1533 */ 1534 1535 int register_netdevice_notifier(struct notifier_block *nb) 1536 { 1537 struct net_device *dev; 1538 struct net_device *last; 1539 struct net *net; 1540 int err; 1541 1542 rtnl_lock(); 1543 err = raw_notifier_chain_register(&netdev_chain, nb); 1544 if (err) 1545 goto unlock; 1546 if (dev_boot_phase) 1547 goto unlock; 1548 for_each_net(net) { 1549 for_each_netdev(net, dev) { 1550 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1551 err = notifier_to_errno(err); 1552 if (err) 1553 goto rollback; 1554 1555 if (!(dev->flags & IFF_UP)) 1556 continue; 1557 1558 call_netdevice_notifier(nb, NETDEV_UP, dev); 1559 } 1560 } 1561 1562 unlock: 1563 rtnl_unlock(); 1564 return err; 1565 1566 rollback: 1567 last = dev; 1568 for_each_net(net) { 1569 for_each_netdev(net, dev) { 1570 if (dev == last) 1571 goto outroll; 1572 1573 if (dev->flags & IFF_UP) { 1574 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1575 dev); 1576 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1577 } 1578 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1579 } 1580 } 1581 1582 outroll: 1583 raw_notifier_chain_unregister(&netdev_chain, nb); 1584 goto unlock; 1585 } 1586 EXPORT_SYMBOL(register_netdevice_notifier); 1587 1588 /** 1589 * unregister_netdevice_notifier - unregister a network notifier block 1590 * @nb: notifier 1591 * 1592 * Unregister a notifier previously registered by 1593 * register_netdevice_notifier(). The notifier is unlinked into the 1594 * kernel structures and may then be reused. A negative errno code 1595 * is returned on a failure. 1596 * 1597 * After unregistering unregister and down device events are synthesized 1598 * for all devices on the device list to the removed notifier to remove 1599 * the need for special case cleanup code. 1600 */ 1601 1602 int unregister_netdevice_notifier(struct notifier_block *nb) 1603 { 1604 struct net_device *dev; 1605 struct net *net; 1606 int err; 1607 1608 rtnl_lock(); 1609 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1610 if (err) 1611 goto unlock; 1612 1613 for_each_net(net) { 1614 for_each_netdev(net, dev) { 1615 if (dev->flags & IFF_UP) { 1616 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1617 dev); 1618 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1619 } 1620 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1621 } 1622 } 1623 unlock: 1624 rtnl_unlock(); 1625 return err; 1626 } 1627 EXPORT_SYMBOL(unregister_netdevice_notifier); 1628 1629 /** 1630 * call_netdevice_notifiers_info - call all network notifier blocks 1631 * @val: value passed unmodified to notifier function 1632 * @dev: net_device pointer passed unmodified to notifier function 1633 * @info: notifier information data 1634 * 1635 * Call all network notifier blocks. Parameters and return value 1636 * are as for raw_notifier_call_chain(). 1637 */ 1638 1639 static int call_netdevice_notifiers_info(unsigned long val, 1640 struct net_device *dev, 1641 struct netdev_notifier_info *info) 1642 { 1643 ASSERT_RTNL(); 1644 netdev_notifier_info_init(info, dev); 1645 return raw_notifier_call_chain(&netdev_chain, val, info); 1646 } 1647 1648 /** 1649 * call_netdevice_notifiers - call all network notifier blocks 1650 * @val: value passed unmodified to notifier function 1651 * @dev: net_device pointer passed unmodified to notifier function 1652 * 1653 * Call all network notifier blocks. Parameters and return value 1654 * are as for raw_notifier_call_chain(). 1655 */ 1656 1657 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1658 { 1659 struct netdev_notifier_info info; 1660 1661 return call_netdevice_notifiers_info(val, dev, &info); 1662 } 1663 EXPORT_SYMBOL(call_netdevice_notifiers); 1664 1665 #ifdef CONFIG_NET_INGRESS 1666 static struct static_key ingress_needed __read_mostly; 1667 1668 void net_inc_ingress_queue(void) 1669 { 1670 static_key_slow_inc(&ingress_needed); 1671 } 1672 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1673 1674 void net_dec_ingress_queue(void) 1675 { 1676 static_key_slow_dec(&ingress_needed); 1677 } 1678 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1679 #endif 1680 1681 #ifdef CONFIG_NET_EGRESS 1682 static struct static_key egress_needed __read_mostly; 1683 1684 void net_inc_egress_queue(void) 1685 { 1686 static_key_slow_inc(&egress_needed); 1687 } 1688 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 1689 1690 void net_dec_egress_queue(void) 1691 { 1692 static_key_slow_dec(&egress_needed); 1693 } 1694 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 1695 #endif 1696 1697 static struct static_key netstamp_needed __read_mostly; 1698 #ifdef HAVE_JUMP_LABEL 1699 /* We are not allowed to call static_key_slow_dec() from irq context 1700 * If net_disable_timestamp() is called from irq context, defer the 1701 * static_key_slow_dec() calls. 1702 */ 1703 static atomic_t netstamp_needed_deferred; 1704 #endif 1705 1706 void net_enable_timestamp(void) 1707 { 1708 #ifdef HAVE_JUMP_LABEL 1709 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1710 1711 if (deferred) { 1712 while (--deferred) 1713 static_key_slow_dec(&netstamp_needed); 1714 return; 1715 } 1716 #endif 1717 static_key_slow_inc(&netstamp_needed); 1718 } 1719 EXPORT_SYMBOL(net_enable_timestamp); 1720 1721 void net_disable_timestamp(void) 1722 { 1723 #ifdef HAVE_JUMP_LABEL 1724 if (in_interrupt()) { 1725 atomic_inc(&netstamp_needed_deferred); 1726 return; 1727 } 1728 #endif 1729 static_key_slow_dec(&netstamp_needed); 1730 } 1731 EXPORT_SYMBOL(net_disable_timestamp); 1732 1733 static inline void net_timestamp_set(struct sk_buff *skb) 1734 { 1735 skb->tstamp.tv64 = 0; 1736 if (static_key_false(&netstamp_needed)) 1737 __net_timestamp(skb); 1738 } 1739 1740 #define net_timestamp_check(COND, SKB) \ 1741 if (static_key_false(&netstamp_needed)) { \ 1742 if ((COND) && !(SKB)->tstamp.tv64) \ 1743 __net_timestamp(SKB); \ 1744 } \ 1745 1746 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 1747 { 1748 unsigned int len; 1749 1750 if (!(dev->flags & IFF_UP)) 1751 return false; 1752 1753 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1754 if (skb->len <= len) 1755 return true; 1756 1757 /* if TSO is enabled, we don't care about the length as the packet 1758 * could be forwarded without being segmented before 1759 */ 1760 if (skb_is_gso(skb)) 1761 return true; 1762 1763 return false; 1764 } 1765 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1766 1767 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1768 { 1769 if (skb_orphan_frags(skb, GFP_ATOMIC) || 1770 unlikely(!is_skb_forwardable(dev, skb))) { 1771 atomic_long_inc(&dev->rx_dropped); 1772 kfree_skb(skb); 1773 return NET_RX_DROP; 1774 } 1775 1776 skb_scrub_packet(skb, true); 1777 skb->priority = 0; 1778 skb->protocol = eth_type_trans(skb, dev); 1779 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1780 1781 return 0; 1782 } 1783 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1784 1785 /** 1786 * dev_forward_skb - loopback an skb to another netif 1787 * 1788 * @dev: destination network device 1789 * @skb: buffer to forward 1790 * 1791 * return values: 1792 * NET_RX_SUCCESS (no congestion) 1793 * NET_RX_DROP (packet was dropped, but freed) 1794 * 1795 * dev_forward_skb can be used for injecting an skb from the 1796 * start_xmit function of one device into the receive queue 1797 * of another device. 1798 * 1799 * The receiving device may be in another namespace, so 1800 * we have to clear all information in the skb that could 1801 * impact namespace isolation. 1802 */ 1803 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1804 { 1805 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1806 } 1807 EXPORT_SYMBOL_GPL(dev_forward_skb); 1808 1809 static inline int deliver_skb(struct sk_buff *skb, 1810 struct packet_type *pt_prev, 1811 struct net_device *orig_dev) 1812 { 1813 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 1814 return -ENOMEM; 1815 atomic_inc(&skb->users); 1816 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1817 } 1818 1819 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 1820 struct packet_type **pt, 1821 struct net_device *orig_dev, 1822 __be16 type, 1823 struct list_head *ptype_list) 1824 { 1825 struct packet_type *ptype, *pt_prev = *pt; 1826 1827 list_for_each_entry_rcu(ptype, ptype_list, list) { 1828 if (ptype->type != type) 1829 continue; 1830 if (pt_prev) 1831 deliver_skb(skb, pt_prev, orig_dev); 1832 pt_prev = ptype; 1833 } 1834 *pt = pt_prev; 1835 } 1836 1837 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1838 { 1839 if (!ptype->af_packet_priv || !skb->sk) 1840 return false; 1841 1842 if (ptype->id_match) 1843 return ptype->id_match(ptype, skb->sk); 1844 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1845 return true; 1846 1847 return false; 1848 } 1849 1850 /* 1851 * Support routine. Sends outgoing frames to any network 1852 * taps currently in use. 1853 */ 1854 1855 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1856 { 1857 struct packet_type *ptype; 1858 struct sk_buff *skb2 = NULL; 1859 struct packet_type *pt_prev = NULL; 1860 struct list_head *ptype_list = &ptype_all; 1861 1862 rcu_read_lock(); 1863 again: 1864 list_for_each_entry_rcu(ptype, ptype_list, list) { 1865 /* Never send packets back to the socket 1866 * they originated from - MvS (miquels@drinkel.ow.org) 1867 */ 1868 if (skb_loop_sk(ptype, skb)) 1869 continue; 1870 1871 if (pt_prev) { 1872 deliver_skb(skb2, pt_prev, skb->dev); 1873 pt_prev = ptype; 1874 continue; 1875 } 1876 1877 /* need to clone skb, done only once */ 1878 skb2 = skb_clone(skb, GFP_ATOMIC); 1879 if (!skb2) 1880 goto out_unlock; 1881 1882 net_timestamp_set(skb2); 1883 1884 /* skb->nh should be correctly 1885 * set by sender, so that the second statement is 1886 * just protection against buggy protocols. 1887 */ 1888 skb_reset_mac_header(skb2); 1889 1890 if (skb_network_header(skb2) < skb2->data || 1891 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 1892 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 1893 ntohs(skb2->protocol), 1894 dev->name); 1895 skb_reset_network_header(skb2); 1896 } 1897 1898 skb2->transport_header = skb2->network_header; 1899 skb2->pkt_type = PACKET_OUTGOING; 1900 pt_prev = ptype; 1901 } 1902 1903 if (ptype_list == &ptype_all) { 1904 ptype_list = &dev->ptype_all; 1905 goto again; 1906 } 1907 out_unlock: 1908 if (pt_prev) 1909 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 1910 rcu_read_unlock(); 1911 } 1912 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 1913 1914 /** 1915 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 1916 * @dev: Network device 1917 * @txq: number of queues available 1918 * 1919 * If real_num_tx_queues is changed the tc mappings may no longer be 1920 * valid. To resolve this verify the tc mapping remains valid and if 1921 * not NULL the mapping. With no priorities mapping to this 1922 * offset/count pair it will no longer be used. In the worst case TC0 1923 * is invalid nothing can be done so disable priority mappings. If is 1924 * expected that drivers will fix this mapping if they can before 1925 * calling netif_set_real_num_tx_queues. 1926 */ 1927 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 1928 { 1929 int i; 1930 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 1931 1932 /* If TC0 is invalidated disable TC mapping */ 1933 if (tc->offset + tc->count > txq) { 1934 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 1935 dev->num_tc = 0; 1936 return; 1937 } 1938 1939 /* Invalidated prio to tc mappings set to TC0 */ 1940 for (i = 1; i < TC_BITMASK + 1; i++) { 1941 int q = netdev_get_prio_tc_map(dev, i); 1942 1943 tc = &dev->tc_to_txq[q]; 1944 if (tc->offset + tc->count > txq) { 1945 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 1946 i, q); 1947 netdev_set_prio_tc_map(dev, i, 0); 1948 } 1949 } 1950 } 1951 1952 #ifdef CONFIG_XPS 1953 static DEFINE_MUTEX(xps_map_mutex); 1954 #define xmap_dereference(P) \ 1955 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 1956 1957 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, 1958 int cpu, u16 index) 1959 { 1960 struct xps_map *map = NULL; 1961 int pos; 1962 1963 if (dev_maps) 1964 map = xmap_dereference(dev_maps->cpu_map[cpu]); 1965 1966 for (pos = 0; map && pos < map->len; pos++) { 1967 if (map->queues[pos] == index) { 1968 if (map->len > 1) { 1969 map->queues[pos] = map->queues[--map->len]; 1970 } else { 1971 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); 1972 kfree_rcu(map, rcu); 1973 map = NULL; 1974 } 1975 break; 1976 } 1977 } 1978 1979 return map; 1980 } 1981 1982 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 1983 { 1984 struct xps_dev_maps *dev_maps; 1985 int cpu, i; 1986 bool active = false; 1987 1988 mutex_lock(&xps_map_mutex); 1989 dev_maps = xmap_dereference(dev->xps_maps); 1990 1991 if (!dev_maps) 1992 goto out_no_maps; 1993 1994 for_each_possible_cpu(cpu) { 1995 for (i = index; i < dev->num_tx_queues; i++) { 1996 if (!remove_xps_queue(dev_maps, cpu, i)) 1997 break; 1998 } 1999 if (i == dev->num_tx_queues) 2000 active = true; 2001 } 2002 2003 if (!active) { 2004 RCU_INIT_POINTER(dev->xps_maps, NULL); 2005 kfree_rcu(dev_maps, rcu); 2006 } 2007 2008 for (i = index; i < dev->num_tx_queues; i++) 2009 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), 2010 NUMA_NO_NODE); 2011 2012 out_no_maps: 2013 mutex_unlock(&xps_map_mutex); 2014 } 2015 2016 static struct xps_map *expand_xps_map(struct xps_map *map, 2017 int cpu, u16 index) 2018 { 2019 struct xps_map *new_map; 2020 int alloc_len = XPS_MIN_MAP_ALLOC; 2021 int i, pos; 2022 2023 for (pos = 0; map && pos < map->len; pos++) { 2024 if (map->queues[pos] != index) 2025 continue; 2026 return map; 2027 } 2028 2029 /* Need to add queue to this CPU's existing map */ 2030 if (map) { 2031 if (pos < map->alloc_len) 2032 return map; 2033 2034 alloc_len = map->alloc_len * 2; 2035 } 2036 2037 /* Need to allocate new map to store queue on this CPU's map */ 2038 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2039 cpu_to_node(cpu)); 2040 if (!new_map) 2041 return NULL; 2042 2043 for (i = 0; i < pos; i++) 2044 new_map->queues[i] = map->queues[i]; 2045 new_map->alloc_len = alloc_len; 2046 new_map->len = pos; 2047 2048 return new_map; 2049 } 2050 2051 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2052 u16 index) 2053 { 2054 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2055 struct xps_map *map, *new_map; 2056 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); 2057 int cpu, numa_node_id = -2; 2058 bool active = false; 2059 2060 mutex_lock(&xps_map_mutex); 2061 2062 dev_maps = xmap_dereference(dev->xps_maps); 2063 2064 /* allocate memory for queue storage */ 2065 for_each_online_cpu(cpu) { 2066 if (!cpumask_test_cpu(cpu, mask)) 2067 continue; 2068 2069 if (!new_dev_maps) 2070 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2071 if (!new_dev_maps) { 2072 mutex_unlock(&xps_map_mutex); 2073 return -ENOMEM; 2074 } 2075 2076 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : 2077 NULL; 2078 2079 map = expand_xps_map(map, cpu, index); 2080 if (!map) 2081 goto error; 2082 2083 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); 2084 } 2085 2086 if (!new_dev_maps) 2087 goto out_no_new_maps; 2088 2089 for_each_possible_cpu(cpu) { 2090 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { 2091 /* add queue to CPU maps */ 2092 int pos = 0; 2093 2094 map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2095 while ((pos < map->len) && (map->queues[pos] != index)) 2096 pos++; 2097 2098 if (pos == map->len) 2099 map->queues[map->len++] = index; 2100 #ifdef CONFIG_NUMA 2101 if (numa_node_id == -2) 2102 numa_node_id = cpu_to_node(cpu); 2103 else if (numa_node_id != cpu_to_node(cpu)) 2104 numa_node_id = -1; 2105 #endif 2106 } else if (dev_maps) { 2107 /* fill in the new device map from the old device map */ 2108 map = xmap_dereference(dev_maps->cpu_map[cpu]); 2109 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); 2110 } 2111 2112 } 2113 2114 rcu_assign_pointer(dev->xps_maps, new_dev_maps); 2115 2116 /* Cleanup old maps */ 2117 if (dev_maps) { 2118 for_each_possible_cpu(cpu) { 2119 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2120 map = xmap_dereference(dev_maps->cpu_map[cpu]); 2121 if (map && map != new_map) 2122 kfree_rcu(map, rcu); 2123 } 2124 2125 kfree_rcu(dev_maps, rcu); 2126 } 2127 2128 dev_maps = new_dev_maps; 2129 active = true; 2130 2131 out_no_new_maps: 2132 /* update Tx queue numa node */ 2133 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2134 (numa_node_id >= 0) ? numa_node_id : 2135 NUMA_NO_NODE); 2136 2137 if (!dev_maps) 2138 goto out_no_maps; 2139 2140 /* removes queue from unused CPUs */ 2141 for_each_possible_cpu(cpu) { 2142 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) 2143 continue; 2144 2145 if (remove_xps_queue(dev_maps, cpu, index)) 2146 active = true; 2147 } 2148 2149 /* free map if not active */ 2150 if (!active) { 2151 RCU_INIT_POINTER(dev->xps_maps, NULL); 2152 kfree_rcu(dev_maps, rcu); 2153 } 2154 2155 out_no_maps: 2156 mutex_unlock(&xps_map_mutex); 2157 2158 return 0; 2159 error: 2160 /* remove any maps that we added */ 2161 for_each_possible_cpu(cpu) { 2162 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); 2163 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : 2164 NULL; 2165 if (new_map && new_map != map) 2166 kfree(new_map); 2167 } 2168 2169 mutex_unlock(&xps_map_mutex); 2170 2171 kfree(new_dev_maps); 2172 return -ENOMEM; 2173 } 2174 EXPORT_SYMBOL(netif_set_xps_queue); 2175 2176 #endif 2177 /* 2178 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2179 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. 2180 */ 2181 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2182 { 2183 int rc; 2184 2185 if (txq < 1 || txq > dev->num_tx_queues) 2186 return -EINVAL; 2187 2188 if (dev->reg_state == NETREG_REGISTERED || 2189 dev->reg_state == NETREG_UNREGISTERING) { 2190 ASSERT_RTNL(); 2191 2192 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2193 txq); 2194 if (rc) 2195 return rc; 2196 2197 if (dev->num_tc) 2198 netif_setup_tc(dev, txq); 2199 2200 if (txq < dev->real_num_tx_queues) { 2201 qdisc_reset_all_tx_gt(dev, txq); 2202 #ifdef CONFIG_XPS 2203 netif_reset_xps_queues_gt(dev, txq); 2204 #endif 2205 } 2206 } 2207 2208 dev->real_num_tx_queues = txq; 2209 return 0; 2210 } 2211 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2212 2213 #ifdef CONFIG_SYSFS 2214 /** 2215 * netif_set_real_num_rx_queues - set actual number of RX queues used 2216 * @dev: Network device 2217 * @rxq: Actual number of RX queues 2218 * 2219 * This must be called either with the rtnl_lock held or before 2220 * registration of the net device. Returns 0 on success, or a 2221 * negative error code. If called before registration, it always 2222 * succeeds. 2223 */ 2224 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2225 { 2226 int rc; 2227 2228 if (rxq < 1 || rxq > dev->num_rx_queues) 2229 return -EINVAL; 2230 2231 if (dev->reg_state == NETREG_REGISTERED) { 2232 ASSERT_RTNL(); 2233 2234 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2235 rxq); 2236 if (rc) 2237 return rc; 2238 } 2239 2240 dev->real_num_rx_queues = rxq; 2241 return 0; 2242 } 2243 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2244 #endif 2245 2246 /** 2247 * netif_get_num_default_rss_queues - default number of RSS queues 2248 * 2249 * This routine should set an upper limit on the number of RSS queues 2250 * used by default by multiqueue devices. 2251 */ 2252 int netif_get_num_default_rss_queues(void) 2253 { 2254 return is_kdump_kernel() ? 2255 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2256 } 2257 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2258 2259 static void __netif_reschedule(struct Qdisc *q) 2260 { 2261 struct softnet_data *sd; 2262 unsigned long flags; 2263 2264 local_irq_save(flags); 2265 sd = this_cpu_ptr(&softnet_data); 2266 q->next_sched = NULL; 2267 *sd->output_queue_tailp = q; 2268 sd->output_queue_tailp = &q->next_sched; 2269 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2270 local_irq_restore(flags); 2271 } 2272 2273 void __netif_schedule(struct Qdisc *q) 2274 { 2275 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2276 __netif_reschedule(q); 2277 } 2278 EXPORT_SYMBOL(__netif_schedule); 2279 2280 struct dev_kfree_skb_cb { 2281 enum skb_free_reason reason; 2282 }; 2283 2284 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2285 { 2286 return (struct dev_kfree_skb_cb *)skb->cb; 2287 } 2288 2289 void netif_schedule_queue(struct netdev_queue *txq) 2290 { 2291 rcu_read_lock(); 2292 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2293 struct Qdisc *q = rcu_dereference(txq->qdisc); 2294 2295 __netif_schedule(q); 2296 } 2297 rcu_read_unlock(); 2298 } 2299 EXPORT_SYMBOL(netif_schedule_queue); 2300 2301 /** 2302 * netif_wake_subqueue - allow sending packets on subqueue 2303 * @dev: network device 2304 * @queue_index: sub queue index 2305 * 2306 * Resume individual transmit queue of a device with multiple transmit queues. 2307 */ 2308 void netif_wake_subqueue(struct net_device *dev, u16 queue_index) 2309 { 2310 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); 2311 2312 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { 2313 struct Qdisc *q; 2314 2315 rcu_read_lock(); 2316 q = rcu_dereference(txq->qdisc); 2317 __netif_schedule(q); 2318 rcu_read_unlock(); 2319 } 2320 } 2321 EXPORT_SYMBOL(netif_wake_subqueue); 2322 2323 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2324 { 2325 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2326 struct Qdisc *q; 2327 2328 rcu_read_lock(); 2329 q = rcu_dereference(dev_queue->qdisc); 2330 __netif_schedule(q); 2331 rcu_read_unlock(); 2332 } 2333 } 2334 EXPORT_SYMBOL(netif_tx_wake_queue); 2335 2336 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2337 { 2338 unsigned long flags; 2339 2340 if (likely(atomic_read(&skb->users) == 1)) { 2341 smp_rmb(); 2342 atomic_set(&skb->users, 0); 2343 } else if (likely(!atomic_dec_and_test(&skb->users))) { 2344 return; 2345 } 2346 get_kfree_skb_cb(skb)->reason = reason; 2347 local_irq_save(flags); 2348 skb->next = __this_cpu_read(softnet_data.completion_queue); 2349 __this_cpu_write(softnet_data.completion_queue, skb); 2350 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2351 local_irq_restore(flags); 2352 } 2353 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2354 2355 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2356 { 2357 if (in_irq() || irqs_disabled()) 2358 __dev_kfree_skb_irq(skb, reason); 2359 else 2360 dev_kfree_skb(skb); 2361 } 2362 EXPORT_SYMBOL(__dev_kfree_skb_any); 2363 2364 2365 /** 2366 * netif_device_detach - mark device as removed 2367 * @dev: network device 2368 * 2369 * Mark device as removed from system and therefore no longer available. 2370 */ 2371 void netif_device_detach(struct net_device *dev) 2372 { 2373 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2374 netif_running(dev)) { 2375 netif_tx_stop_all_queues(dev); 2376 } 2377 } 2378 EXPORT_SYMBOL(netif_device_detach); 2379 2380 /** 2381 * netif_device_attach - mark device as attached 2382 * @dev: network device 2383 * 2384 * Mark device as attached from system and restart if needed. 2385 */ 2386 void netif_device_attach(struct net_device *dev) 2387 { 2388 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2389 netif_running(dev)) { 2390 netif_tx_wake_all_queues(dev); 2391 __netdev_watchdog_up(dev); 2392 } 2393 } 2394 EXPORT_SYMBOL(netif_device_attach); 2395 2396 /* 2397 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 2398 * to be used as a distribution range. 2399 */ 2400 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 2401 unsigned int num_tx_queues) 2402 { 2403 u32 hash; 2404 u16 qoffset = 0; 2405 u16 qcount = num_tx_queues; 2406 2407 if (skb_rx_queue_recorded(skb)) { 2408 hash = skb_get_rx_queue(skb); 2409 while (unlikely(hash >= num_tx_queues)) 2410 hash -= num_tx_queues; 2411 return hash; 2412 } 2413 2414 if (dev->num_tc) { 2415 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 2416 qoffset = dev->tc_to_txq[tc].offset; 2417 qcount = dev->tc_to_txq[tc].count; 2418 } 2419 2420 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 2421 } 2422 EXPORT_SYMBOL(__skb_tx_hash); 2423 2424 static void skb_warn_bad_offload(const struct sk_buff *skb) 2425 { 2426 static const netdev_features_t null_features; 2427 struct net_device *dev = skb->dev; 2428 const char *name = ""; 2429 2430 if (!net_ratelimit()) 2431 return; 2432 2433 if (dev) { 2434 if (dev->dev.parent) 2435 name = dev_driver_string(dev->dev.parent); 2436 else 2437 name = netdev_name(dev); 2438 } 2439 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2440 "gso_type=%d ip_summed=%d\n", 2441 name, dev ? &dev->features : &null_features, 2442 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2443 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2444 skb_shinfo(skb)->gso_type, skb->ip_summed); 2445 } 2446 2447 /* 2448 * Invalidate hardware checksum when packet is to be mangled, and 2449 * complete checksum manually on outgoing path. 2450 */ 2451 int skb_checksum_help(struct sk_buff *skb) 2452 { 2453 __wsum csum; 2454 int ret = 0, offset; 2455 2456 if (skb->ip_summed == CHECKSUM_COMPLETE) 2457 goto out_set_summed; 2458 2459 if (unlikely(skb_shinfo(skb)->gso_size)) { 2460 skb_warn_bad_offload(skb); 2461 return -EINVAL; 2462 } 2463 2464 /* Before computing a checksum, we should make sure no frag could 2465 * be modified by an external entity : checksum could be wrong. 2466 */ 2467 if (skb_has_shared_frag(skb)) { 2468 ret = __skb_linearize(skb); 2469 if (ret) 2470 goto out; 2471 } 2472 2473 offset = skb_checksum_start_offset(skb); 2474 BUG_ON(offset >= skb_headlen(skb)); 2475 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2476 2477 offset += skb->csum_offset; 2478 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2479 2480 if (skb_cloned(skb) && 2481 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2482 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2483 if (ret) 2484 goto out; 2485 } 2486 2487 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2488 out_set_summed: 2489 skb->ip_summed = CHECKSUM_NONE; 2490 out: 2491 return ret; 2492 } 2493 EXPORT_SYMBOL(skb_checksum_help); 2494 2495 /* skb_csum_offload_check - Driver helper function to determine if a device 2496 * with limited checksum offload capabilities is able to offload the checksum 2497 * for a given packet. 2498 * 2499 * Arguments: 2500 * skb - sk_buff for the packet in question 2501 * spec - contains the description of what device can offload 2502 * csum_encapped - returns true if the checksum being offloaded is 2503 * encpasulated. That is it is checksum for the transport header 2504 * in the inner headers. 2505 * checksum_help - when set indicates that helper function should 2506 * call skb_checksum_help if offload checks fail 2507 * 2508 * Returns: 2509 * true: Packet has passed the checksum checks and should be offloadable to 2510 * the device (a driver may still need to check for additional 2511 * restrictions of its device) 2512 * false: Checksum is not offloadable. If checksum_help was set then 2513 * skb_checksum_help was called to resolve checksum for non-GSO 2514 * packets and when IP protocol is not SCTP 2515 */ 2516 bool __skb_csum_offload_chk(struct sk_buff *skb, 2517 const struct skb_csum_offl_spec *spec, 2518 bool *csum_encapped, 2519 bool csum_help) 2520 { 2521 struct iphdr *iph; 2522 struct ipv6hdr *ipv6; 2523 void *nhdr; 2524 int protocol; 2525 u8 ip_proto; 2526 2527 if (skb->protocol == htons(ETH_P_8021Q) || 2528 skb->protocol == htons(ETH_P_8021AD)) { 2529 if (!spec->vlan_okay) 2530 goto need_help; 2531 } 2532 2533 /* We check whether the checksum refers to a transport layer checksum in 2534 * the outermost header or an encapsulated transport layer checksum that 2535 * corresponds to the inner headers of the skb. If the checksum is for 2536 * something else in the packet we need help. 2537 */ 2538 if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) { 2539 /* Non-encapsulated checksum */ 2540 protocol = eproto_to_ipproto(vlan_get_protocol(skb)); 2541 nhdr = skb_network_header(skb); 2542 *csum_encapped = false; 2543 if (spec->no_not_encapped) 2544 goto need_help; 2545 } else if (skb->encapsulation && spec->encap_okay && 2546 skb_checksum_start_offset(skb) == 2547 skb_inner_transport_offset(skb)) { 2548 /* Encapsulated checksum */ 2549 *csum_encapped = true; 2550 switch (skb->inner_protocol_type) { 2551 case ENCAP_TYPE_ETHER: 2552 protocol = eproto_to_ipproto(skb->inner_protocol); 2553 break; 2554 case ENCAP_TYPE_IPPROTO: 2555 protocol = skb->inner_protocol; 2556 break; 2557 } 2558 nhdr = skb_inner_network_header(skb); 2559 } else { 2560 goto need_help; 2561 } 2562 2563 switch (protocol) { 2564 case IPPROTO_IP: 2565 if (!spec->ipv4_okay) 2566 goto need_help; 2567 iph = nhdr; 2568 ip_proto = iph->protocol; 2569 if (iph->ihl != 5 && !spec->ip_options_okay) 2570 goto need_help; 2571 break; 2572 case IPPROTO_IPV6: 2573 if (!spec->ipv6_okay) 2574 goto need_help; 2575 if (spec->no_encapped_ipv6 && *csum_encapped) 2576 goto need_help; 2577 ipv6 = nhdr; 2578 nhdr += sizeof(*ipv6); 2579 ip_proto = ipv6->nexthdr; 2580 break; 2581 default: 2582 goto need_help; 2583 } 2584 2585 ip_proto_again: 2586 switch (ip_proto) { 2587 case IPPROTO_TCP: 2588 if (!spec->tcp_okay || 2589 skb->csum_offset != offsetof(struct tcphdr, check)) 2590 goto need_help; 2591 break; 2592 case IPPROTO_UDP: 2593 if (!spec->udp_okay || 2594 skb->csum_offset != offsetof(struct udphdr, check)) 2595 goto need_help; 2596 break; 2597 case IPPROTO_SCTP: 2598 if (!spec->sctp_okay || 2599 skb->csum_offset != offsetof(struct sctphdr, checksum)) 2600 goto cant_help; 2601 break; 2602 case NEXTHDR_HOP: 2603 case NEXTHDR_ROUTING: 2604 case NEXTHDR_DEST: { 2605 u8 *opthdr = nhdr; 2606 2607 if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay) 2608 goto need_help; 2609 2610 ip_proto = opthdr[0]; 2611 nhdr += (opthdr[1] + 1) << 3; 2612 2613 goto ip_proto_again; 2614 } 2615 default: 2616 goto need_help; 2617 } 2618 2619 /* Passed the tests for offloading checksum */ 2620 return true; 2621 2622 need_help: 2623 if (csum_help && !skb_shinfo(skb)->gso_size) 2624 skb_checksum_help(skb); 2625 cant_help: 2626 return false; 2627 } 2628 EXPORT_SYMBOL(__skb_csum_offload_chk); 2629 2630 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2631 { 2632 __be16 type = skb->protocol; 2633 2634 /* Tunnel gso handlers can set protocol to ethernet. */ 2635 if (type == htons(ETH_P_TEB)) { 2636 struct ethhdr *eth; 2637 2638 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2639 return 0; 2640 2641 eth = (struct ethhdr *)skb_mac_header(skb); 2642 type = eth->h_proto; 2643 } 2644 2645 return __vlan_get_protocol(skb, type, depth); 2646 } 2647 2648 /** 2649 * skb_mac_gso_segment - mac layer segmentation handler. 2650 * @skb: buffer to segment 2651 * @features: features for the output path (see dev->features) 2652 */ 2653 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 2654 netdev_features_t features) 2655 { 2656 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 2657 struct packet_offload *ptype; 2658 int vlan_depth = skb->mac_len; 2659 __be16 type = skb_network_protocol(skb, &vlan_depth); 2660 2661 if (unlikely(!type)) 2662 return ERR_PTR(-EINVAL); 2663 2664 __skb_pull(skb, vlan_depth); 2665 2666 rcu_read_lock(); 2667 list_for_each_entry_rcu(ptype, &offload_base, list) { 2668 if (ptype->type == type && ptype->callbacks.gso_segment) { 2669 segs = ptype->callbacks.gso_segment(skb, features); 2670 break; 2671 } 2672 } 2673 rcu_read_unlock(); 2674 2675 __skb_push(skb, skb->data - skb_mac_header(skb)); 2676 2677 return segs; 2678 } 2679 EXPORT_SYMBOL(skb_mac_gso_segment); 2680 2681 2682 /* openvswitch calls this on rx path, so we need a different check. 2683 */ 2684 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 2685 { 2686 if (tx_path) 2687 return skb->ip_summed != CHECKSUM_PARTIAL; 2688 else 2689 return skb->ip_summed == CHECKSUM_NONE; 2690 } 2691 2692 /** 2693 * __skb_gso_segment - Perform segmentation on skb. 2694 * @skb: buffer to segment 2695 * @features: features for the output path (see dev->features) 2696 * @tx_path: whether it is called in TX path 2697 * 2698 * This function segments the given skb and returns a list of segments. 2699 * 2700 * It may return NULL if the skb requires no segmentation. This is 2701 * only possible when GSO is used for verifying header integrity. 2702 * 2703 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. 2704 */ 2705 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 2706 netdev_features_t features, bool tx_path) 2707 { 2708 if (unlikely(skb_needs_check(skb, tx_path))) { 2709 int err; 2710 2711 skb_warn_bad_offload(skb); 2712 2713 err = skb_cow_head(skb, 0); 2714 if (err < 0) 2715 return ERR_PTR(err); 2716 } 2717 2718 /* Only report GSO partial support if it will enable us to 2719 * support segmentation on this frame without needing additional 2720 * work. 2721 */ 2722 if (features & NETIF_F_GSO_PARTIAL) { 2723 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 2724 struct net_device *dev = skb->dev; 2725 2726 partial_features |= dev->features & dev->gso_partial_features; 2727 if (!skb_gso_ok(skb, features | partial_features)) 2728 features &= ~NETIF_F_GSO_PARTIAL; 2729 } 2730 2731 BUILD_BUG_ON(SKB_SGO_CB_OFFSET + 2732 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 2733 2734 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 2735 SKB_GSO_CB(skb)->encap_level = 0; 2736 2737 skb_reset_mac_header(skb); 2738 skb_reset_mac_len(skb); 2739 2740 return skb_mac_gso_segment(skb, features); 2741 } 2742 EXPORT_SYMBOL(__skb_gso_segment); 2743 2744 /* Take action when hardware reception checksum errors are detected. */ 2745 #ifdef CONFIG_BUG 2746 void netdev_rx_csum_fault(struct net_device *dev) 2747 { 2748 if (net_ratelimit()) { 2749 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 2750 dump_stack(); 2751 } 2752 } 2753 EXPORT_SYMBOL(netdev_rx_csum_fault); 2754 #endif 2755 2756 /* Actually, we should eliminate this check as soon as we know, that: 2757 * 1. IOMMU is present and allows to map all the memory. 2758 * 2. No high memory really exists on this machine. 2759 */ 2760 2761 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 2762 { 2763 #ifdef CONFIG_HIGHMEM 2764 int i; 2765 if (!(dev->features & NETIF_F_HIGHDMA)) { 2766 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2767 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2768 if (PageHighMem(skb_frag_page(frag))) 2769 return 1; 2770 } 2771 } 2772 2773 if (PCI_DMA_BUS_IS_PHYS) { 2774 struct device *pdev = dev->dev.parent; 2775 2776 if (!pdev) 2777 return 0; 2778 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2779 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2780 dma_addr_t addr = page_to_phys(skb_frag_page(frag)); 2781 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) 2782 return 1; 2783 } 2784 } 2785 #endif 2786 return 0; 2787 } 2788 2789 /* If MPLS offload request, verify we are testing hardware MPLS features 2790 * instead of standard features for the netdev. 2791 */ 2792 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 2793 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2794 netdev_features_t features, 2795 __be16 type) 2796 { 2797 if (eth_p_mpls(type)) 2798 features &= skb->dev->mpls_features; 2799 2800 return features; 2801 } 2802 #else 2803 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2804 netdev_features_t features, 2805 __be16 type) 2806 { 2807 return features; 2808 } 2809 #endif 2810 2811 static netdev_features_t harmonize_features(struct sk_buff *skb, 2812 netdev_features_t features) 2813 { 2814 int tmp; 2815 __be16 type; 2816 2817 type = skb_network_protocol(skb, &tmp); 2818 features = net_mpls_features(skb, features, type); 2819 2820 if (skb->ip_summed != CHECKSUM_NONE && 2821 !can_checksum_protocol(features, type)) { 2822 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 2823 } else if (illegal_highdma(skb->dev, skb)) { 2824 features &= ~NETIF_F_SG; 2825 } 2826 2827 return features; 2828 } 2829 2830 netdev_features_t passthru_features_check(struct sk_buff *skb, 2831 struct net_device *dev, 2832 netdev_features_t features) 2833 { 2834 return features; 2835 } 2836 EXPORT_SYMBOL(passthru_features_check); 2837 2838 static netdev_features_t dflt_features_check(const struct sk_buff *skb, 2839 struct net_device *dev, 2840 netdev_features_t features) 2841 { 2842 return vlan_features_check(skb, features); 2843 } 2844 2845 static netdev_features_t gso_features_check(const struct sk_buff *skb, 2846 struct net_device *dev, 2847 netdev_features_t features) 2848 { 2849 u16 gso_segs = skb_shinfo(skb)->gso_segs; 2850 2851 if (gso_segs > dev->gso_max_segs) 2852 return features & ~NETIF_F_GSO_MASK; 2853 2854 /* Support for GSO partial features requires software 2855 * intervention before we can actually process the packets 2856 * so we need to strip support for any partial features now 2857 * and we can pull them back in after we have partially 2858 * segmented the frame. 2859 */ 2860 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 2861 features &= ~dev->gso_partial_features; 2862 2863 /* Make sure to clear the IPv4 ID mangling feature if the 2864 * IPv4 header has the potential to be fragmented. 2865 */ 2866 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 2867 struct iphdr *iph = skb->encapsulation ? 2868 inner_ip_hdr(skb) : ip_hdr(skb); 2869 2870 if (!(iph->frag_off & htons(IP_DF))) 2871 features &= ~NETIF_F_TSO_MANGLEID; 2872 } 2873 2874 return features; 2875 } 2876 2877 netdev_features_t netif_skb_features(struct sk_buff *skb) 2878 { 2879 struct net_device *dev = skb->dev; 2880 netdev_features_t features = dev->features; 2881 2882 if (skb_is_gso(skb)) 2883 features = gso_features_check(skb, dev, features); 2884 2885 /* If encapsulation offload request, verify we are testing 2886 * hardware encapsulation features instead of standard 2887 * features for the netdev 2888 */ 2889 if (skb->encapsulation) 2890 features &= dev->hw_enc_features; 2891 2892 if (skb_vlan_tagged(skb)) 2893 features = netdev_intersect_features(features, 2894 dev->vlan_features | 2895 NETIF_F_HW_VLAN_CTAG_TX | 2896 NETIF_F_HW_VLAN_STAG_TX); 2897 2898 if (dev->netdev_ops->ndo_features_check) 2899 features &= dev->netdev_ops->ndo_features_check(skb, dev, 2900 features); 2901 else 2902 features &= dflt_features_check(skb, dev, features); 2903 2904 return harmonize_features(skb, features); 2905 } 2906 EXPORT_SYMBOL(netif_skb_features); 2907 2908 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 2909 struct netdev_queue *txq, bool more) 2910 { 2911 unsigned int len; 2912 int rc; 2913 2914 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) 2915 dev_queue_xmit_nit(skb, dev); 2916 2917 len = skb->len; 2918 trace_net_dev_start_xmit(skb, dev); 2919 rc = netdev_start_xmit(skb, dev, txq, more); 2920 trace_net_dev_xmit(skb, rc, dev, len); 2921 2922 return rc; 2923 } 2924 2925 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 2926 struct netdev_queue *txq, int *ret) 2927 { 2928 struct sk_buff *skb = first; 2929 int rc = NETDEV_TX_OK; 2930 2931 while (skb) { 2932 struct sk_buff *next = skb->next; 2933 2934 skb->next = NULL; 2935 rc = xmit_one(skb, dev, txq, next != NULL); 2936 if (unlikely(!dev_xmit_complete(rc))) { 2937 skb->next = next; 2938 goto out; 2939 } 2940 2941 skb = next; 2942 if (netif_xmit_stopped(txq) && skb) { 2943 rc = NETDEV_TX_BUSY; 2944 break; 2945 } 2946 } 2947 2948 out: 2949 *ret = rc; 2950 return skb; 2951 } 2952 2953 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 2954 netdev_features_t features) 2955 { 2956 if (skb_vlan_tag_present(skb) && 2957 !vlan_hw_offload_capable(features, skb->vlan_proto)) 2958 skb = __vlan_hwaccel_push_inside(skb); 2959 return skb; 2960 } 2961 2962 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) 2963 { 2964 netdev_features_t features; 2965 2966 features = netif_skb_features(skb); 2967 skb = validate_xmit_vlan(skb, features); 2968 if (unlikely(!skb)) 2969 goto out_null; 2970 2971 if (netif_needs_gso(skb, features)) { 2972 struct sk_buff *segs; 2973 2974 segs = skb_gso_segment(skb, features); 2975 if (IS_ERR(segs)) { 2976 goto out_kfree_skb; 2977 } else if (segs) { 2978 consume_skb(skb); 2979 skb = segs; 2980 } 2981 } else { 2982 if (skb_needs_linearize(skb, features) && 2983 __skb_linearize(skb)) 2984 goto out_kfree_skb; 2985 2986 /* If packet is not checksummed and device does not 2987 * support checksumming for this protocol, complete 2988 * checksumming here. 2989 */ 2990 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2991 if (skb->encapsulation) 2992 skb_set_inner_transport_header(skb, 2993 skb_checksum_start_offset(skb)); 2994 else 2995 skb_set_transport_header(skb, 2996 skb_checksum_start_offset(skb)); 2997 if (!(features & NETIF_F_CSUM_MASK) && 2998 skb_checksum_help(skb)) 2999 goto out_kfree_skb; 3000 } 3001 } 3002 3003 return skb; 3004 3005 out_kfree_skb: 3006 kfree_skb(skb); 3007 out_null: 3008 atomic_long_inc(&dev->tx_dropped); 3009 return NULL; 3010 } 3011 3012 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) 3013 { 3014 struct sk_buff *next, *head = NULL, *tail; 3015 3016 for (; skb != NULL; skb = next) { 3017 next = skb->next; 3018 skb->next = NULL; 3019 3020 /* in case skb wont be segmented, point to itself */ 3021 skb->prev = skb; 3022 3023 skb = validate_xmit_skb(skb, dev); 3024 if (!skb) 3025 continue; 3026 3027 if (!head) 3028 head = skb; 3029 else 3030 tail->next = skb; 3031 /* If skb was segmented, skb->prev points to 3032 * the last segment. If not, it still contains skb. 3033 */ 3034 tail = skb->prev; 3035 } 3036 return head; 3037 } 3038 3039 static void qdisc_pkt_len_init(struct sk_buff *skb) 3040 { 3041 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3042 3043 qdisc_skb_cb(skb)->pkt_len = skb->len; 3044 3045 /* To get more precise estimation of bytes sent on wire, 3046 * we add to pkt_len the headers size of all segments 3047 */ 3048 if (shinfo->gso_size) { 3049 unsigned int hdr_len; 3050 u16 gso_segs = shinfo->gso_segs; 3051 3052 /* mac layer + network layer */ 3053 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3054 3055 /* + transport layer */ 3056 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 3057 hdr_len += tcp_hdrlen(skb); 3058 else 3059 hdr_len += sizeof(struct udphdr); 3060 3061 if (shinfo->gso_type & SKB_GSO_DODGY) 3062 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3063 shinfo->gso_size); 3064 3065 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3066 } 3067 } 3068 3069 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3070 struct net_device *dev, 3071 struct netdev_queue *txq) 3072 { 3073 spinlock_t *root_lock = qdisc_lock(q); 3074 struct sk_buff *to_free = NULL; 3075 bool contended; 3076 int rc; 3077 3078 qdisc_calculate_pkt_len(skb, q); 3079 /* 3080 * Heuristic to force contended enqueues to serialize on a 3081 * separate lock before trying to get qdisc main lock. 3082 * This permits qdisc->running owner to get the lock more 3083 * often and dequeue packets faster. 3084 */ 3085 contended = qdisc_is_running(q); 3086 if (unlikely(contended)) 3087 spin_lock(&q->busylock); 3088 3089 spin_lock(root_lock); 3090 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3091 __qdisc_drop(skb, &to_free); 3092 rc = NET_XMIT_DROP; 3093 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3094 qdisc_run_begin(q)) { 3095 /* 3096 * This is a work-conserving queue; there are no old skbs 3097 * waiting to be sent out; and the qdisc is not running - 3098 * xmit the skb directly. 3099 */ 3100 3101 qdisc_bstats_update(q, skb); 3102 3103 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3104 if (unlikely(contended)) { 3105 spin_unlock(&q->busylock); 3106 contended = false; 3107 } 3108 __qdisc_run(q); 3109 } else 3110 qdisc_run_end(q); 3111 3112 rc = NET_XMIT_SUCCESS; 3113 } else { 3114 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3115 if (qdisc_run_begin(q)) { 3116 if (unlikely(contended)) { 3117 spin_unlock(&q->busylock); 3118 contended = false; 3119 } 3120 __qdisc_run(q); 3121 } 3122 } 3123 spin_unlock(root_lock); 3124 if (unlikely(to_free)) 3125 kfree_skb_list(to_free); 3126 if (unlikely(contended)) 3127 spin_unlock(&q->busylock); 3128 return rc; 3129 } 3130 3131 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3132 static void skb_update_prio(struct sk_buff *skb) 3133 { 3134 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); 3135 3136 if (!skb->priority && skb->sk && map) { 3137 unsigned int prioidx = 3138 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); 3139 3140 if (prioidx < map->priomap_len) 3141 skb->priority = map->priomap[prioidx]; 3142 } 3143 } 3144 #else 3145 #define skb_update_prio(skb) 3146 #endif 3147 3148 DEFINE_PER_CPU(int, xmit_recursion); 3149 EXPORT_SYMBOL(xmit_recursion); 3150 3151 /** 3152 * dev_loopback_xmit - loop back @skb 3153 * @net: network namespace this loopback is happening in 3154 * @sk: sk needed to be a netfilter okfn 3155 * @skb: buffer to transmit 3156 */ 3157 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3158 { 3159 skb_reset_mac_header(skb); 3160 __skb_pull(skb, skb_network_offset(skb)); 3161 skb->pkt_type = PACKET_LOOPBACK; 3162 skb->ip_summed = CHECKSUM_UNNECESSARY; 3163 WARN_ON(!skb_dst(skb)); 3164 skb_dst_force(skb); 3165 netif_rx_ni(skb); 3166 return 0; 3167 } 3168 EXPORT_SYMBOL(dev_loopback_xmit); 3169 3170 #ifdef CONFIG_NET_EGRESS 3171 static struct sk_buff * 3172 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3173 { 3174 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); 3175 struct tcf_result cl_res; 3176 3177 if (!cl) 3178 return skb; 3179 3180 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set 3181 * earlier by the caller. 3182 */ 3183 qdisc_bstats_cpu_update(cl->q, skb); 3184 3185 switch (tc_classify(skb, cl, &cl_res, false)) { 3186 case TC_ACT_OK: 3187 case TC_ACT_RECLASSIFY: 3188 skb->tc_index = TC_H_MIN(cl_res.classid); 3189 break; 3190 case TC_ACT_SHOT: 3191 qdisc_qstats_cpu_drop(cl->q); 3192 *ret = NET_XMIT_DROP; 3193 kfree_skb(skb); 3194 return NULL; 3195 case TC_ACT_STOLEN: 3196 case TC_ACT_QUEUED: 3197 *ret = NET_XMIT_SUCCESS; 3198 consume_skb(skb); 3199 return NULL; 3200 case TC_ACT_REDIRECT: 3201 /* No need to push/pop skb's mac_header here on egress! */ 3202 skb_do_redirect(skb); 3203 *ret = NET_XMIT_SUCCESS; 3204 return NULL; 3205 default: 3206 break; 3207 } 3208 3209 return skb; 3210 } 3211 #endif /* CONFIG_NET_EGRESS */ 3212 3213 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) 3214 { 3215 #ifdef CONFIG_XPS 3216 struct xps_dev_maps *dev_maps; 3217 struct xps_map *map; 3218 int queue_index = -1; 3219 3220 rcu_read_lock(); 3221 dev_maps = rcu_dereference(dev->xps_maps); 3222 if (dev_maps) { 3223 map = rcu_dereference( 3224 dev_maps->cpu_map[skb->sender_cpu - 1]); 3225 if (map) { 3226 if (map->len == 1) 3227 queue_index = map->queues[0]; 3228 else 3229 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), 3230 map->len)]; 3231 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3232 queue_index = -1; 3233 } 3234 } 3235 rcu_read_unlock(); 3236 3237 return queue_index; 3238 #else 3239 return -1; 3240 #endif 3241 } 3242 3243 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) 3244 { 3245 struct sock *sk = skb->sk; 3246 int queue_index = sk_tx_queue_get(sk); 3247 3248 if (queue_index < 0 || skb->ooo_okay || 3249 queue_index >= dev->real_num_tx_queues) { 3250 int new_index = get_xps_queue(dev, skb); 3251 if (new_index < 0) 3252 new_index = skb_tx_hash(dev, skb); 3253 3254 if (queue_index != new_index && sk && 3255 sk_fullsock(sk) && 3256 rcu_access_pointer(sk->sk_dst_cache)) 3257 sk_tx_queue_set(sk, new_index); 3258 3259 queue_index = new_index; 3260 } 3261 3262 return queue_index; 3263 } 3264 3265 struct netdev_queue *netdev_pick_tx(struct net_device *dev, 3266 struct sk_buff *skb, 3267 void *accel_priv) 3268 { 3269 int queue_index = 0; 3270 3271 #ifdef CONFIG_XPS 3272 u32 sender_cpu = skb->sender_cpu - 1; 3273 3274 if (sender_cpu >= (u32)NR_CPUS) 3275 skb->sender_cpu = raw_smp_processor_id() + 1; 3276 #endif 3277 3278 if (dev->real_num_tx_queues != 1) { 3279 const struct net_device_ops *ops = dev->netdev_ops; 3280 if (ops->ndo_select_queue) 3281 queue_index = ops->ndo_select_queue(dev, skb, accel_priv, 3282 __netdev_pick_tx); 3283 else 3284 queue_index = __netdev_pick_tx(dev, skb); 3285 3286 if (!accel_priv) 3287 queue_index = netdev_cap_txqueue(dev, queue_index); 3288 } 3289 3290 skb_set_queue_mapping(skb, queue_index); 3291 return netdev_get_tx_queue(dev, queue_index); 3292 } 3293 3294 /** 3295 * __dev_queue_xmit - transmit a buffer 3296 * @skb: buffer to transmit 3297 * @accel_priv: private data used for L2 forwarding offload 3298 * 3299 * Queue a buffer for transmission to a network device. The caller must 3300 * have set the device and priority and built the buffer before calling 3301 * this function. The function can be called from an interrupt. 3302 * 3303 * A negative errno code is returned on a failure. A success does not 3304 * guarantee the frame will be transmitted as it may be dropped due 3305 * to congestion or traffic shaping. 3306 * 3307 * ----------------------------------------------------------------------------------- 3308 * I notice this method can also return errors from the queue disciplines, 3309 * including NET_XMIT_DROP, which is a positive value. So, errors can also 3310 * be positive. 3311 * 3312 * Regardless of the return value, the skb is consumed, so it is currently 3313 * difficult to retry a send to this method. (You can bump the ref count 3314 * before sending to hold a reference for retry if you are careful.) 3315 * 3316 * When calling this method, interrupts MUST be enabled. This is because 3317 * the BH enable code must have IRQs enabled so that it will not deadlock. 3318 * --BLG 3319 */ 3320 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) 3321 { 3322 struct net_device *dev = skb->dev; 3323 struct netdev_queue *txq; 3324 struct Qdisc *q; 3325 int rc = -ENOMEM; 3326 3327 skb_reset_mac_header(skb); 3328 3329 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 3330 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 3331 3332 /* Disable soft irqs for various locks below. Also 3333 * stops preemption for RCU. 3334 */ 3335 rcu_read_lock_bh(); 3336 3337 skb_update_prio(skb); 3338 3339 qdisc_pkt_len_init(skb); 3340 #ifdef CONFIG_NET_CLS_ACT 3341 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); 3342 # ifdef CONFIG_NET_EGRESS 3343 if (static_key_false(&egress_needed)) { 3344 skb = sch_handle_egress(skb, &rc, dev); 3345 if (!skb) 3346 goto out; 3347 } 3348 # endif 3349 #endif 3350 /* If device/qdisc don't need skb->dst, release it right now while 3351 * its hot in this cpu cache. 3352 */ 3353 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 3354 skb_dst_drop(skb); 3355 else 3356 skb_dst_force(skb); 3357 3358 txq = netdev_pick_tx(dev, skb, accel_priv); 3359 q = rcu_dereference_bh(txq->qdisc); 3360 3361 trace_net_dev_queue(skb); 3362 if (q->enqueue) { 3363 rc = __dev_xmit_skb(skb, q, dev, txq); 3364 goto out; 3365 } 3366 3367 /* The device has no queue. Common case for software devices: 3368 loopback, all the sorts of tunnels... 3369 3370 Really, it is unlikely that netif_tx_lock protection is necessary 3371 here. (f.e. loopback and IP tunnels are clean ignoring statistics 3372 counters.) 3373 However, it is possible, that they rely on protection 3374 made by us here. 3375 3376 Check this and shot the lock. It is not prone from deadlocks. 3377 Either shot noqueue qdisc, it is even simpler 8) 3378 */ 3379 if (dev->flags & IFF_UP) { 3380 int cpu = smp_processor_id(); /* ok because BHs are off */ 3381 3382 if (txq->xmit_lock_owner != cpu) { 3383 if (unlikely(__this_cpu_read(xmit_recursion) > 3384 XMIT_RECURSION_LIMIT)) 3385 goto recursion_alert; 3386 3387 skb = validate_xmit_skb(skb, dev); 3388 if (!skb) 3389 goto out; 3390 3391 HARD_TX_LOCK(dev, txq, cpu); 3392 3393 if (!netif_xmit_stopped(txq)) { 3394 __this_cpu_inc(xmit_recursion); 3395 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 3396 __this_cpu_dec(xmit_recursion); 3397 if (dev_xmit_complete(rc)) { 3398 HARD_TX_UNLOCK(dev, txq); 3399 goto out; 3400 } 3401 } 3402 HARD_TX_UNLOCK(dev, txq); 3403 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 3404 dev->name); 3405 } else { 3406 /* Recursion is detected! It is possible, 3407 * unfortunately 3408 */ 3409 recursion_alert: 3410 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 3411 dev->name); 3412 } 3413 } 3414 3415 rc = -ENETDOWN; 3416 rcu_read_unlock_bh(); 3417 3418 atomic_long_inc(&dev->tx_dropped); 3419 kfree_skb_list(skb); 3420 return rc; 3421 out: 3422 rcu_read_unlock_bh(); 3423 return rc; 3424 } 3425 3426 int dev_queue_xmit(struct sk_buff *skb) 3427 { 3428 return __dev_queue_xmit(skb, NULL); 3429 } 3430 EXPORT_SYMBOL(dev_queue_xmit); 3431 3432 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) 3433 { 3434 return __dev_queue_xmit(skb, accel_priv); 3435 } 3436 EXPORT_SYMBOL(dev_queue_xmit_accel); 3437 3438 3439 /*======================================================================= 3440 Receiver routines 3441 =======================================================================*/ 3442 3443 int netdev_max_backlog __read_mostly = 1000; 3444 EXPORT_SYMBOL(netdev_max_backlog); 3445 3446 int netdev_tstamp_prequeue __read_mostly = 1; 3447 int netdev_budget __read_mostly = 300; 3448 int weight_p __read_mostly = 64; /* old backlog weight */ 3449 3450 /* Called with irq disabled */ 3451 static inline void ____napi_schedule(struct softnet_data *sd, 3452 struct napi_struct *napi) 3453 { 3454 list_add_tail(&napi->poll_list, &sd->poll_list); 3455 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3456 } 3457 3458 #ifdef CONFIG_RPS 3459 3460 /* One global table that all flow-based protocols share. */ 3461 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3462 EXPORT_SYMBOL(rps_sock_flow_table); 3463 u32 rps_cpu_mask __read_mostly; 3464 EXPORT_SYMBOL(rps_cpu_mask); 3465 3466 struct static_key rps_needed __read_mostly; 3467 EXPORT_SYMBOL(rps_needed); 3468 3469 static struct rps_dev_flow * 3470 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3471 struct rps_dev_flow *rflow, u16 next_cpu) 3472 { 3473 if (next_cpu < nr_cpu_ids) { 3474 #ifdef CONFIG_RFS_ACCEL 3475 struct netdev_rx_queue *rxqueue; 3476 struct rps_dev_flow_table *flow_table; 3477 struct rps_dev_flow *old_rflow; 3478 u32 flow_id; 3479 u16 rxq_index; 3480 int rc; 3481 3482 /* Should we steer this flow to a different hardware queue? */ 3483 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3484 !(dev->features & NETIF_F_NTUPLE)) 3485 goto out; 3486 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 3487 if (rxq_index == skb_get_rx_queue(skb)) 3488 goto out; 3489 3490 rxqueue = dev->_rx + rxq_index; 3491 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3492 if (!flow_table) 3493 goto out; 3494 flow_id = skb_get_hash(skb) & flow_table->mask; 3495 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 3496 rxq_index, flow_id); 3497 if (rc < 0) 3498 goto out; 3499 old_rflow = rflow; 3500 rflow = &flow_table->flows[flow_id]; 3501 rflow->filter = rc; 3502 if (old_rflow->filter == rflow->filter) 3503 old_rflow->filter = RPS_NO_FILTER; 3504 out: 3505 #endif 3506 rflow->last_qtail = 3507 per_cpu(softnet_data, next_cpu).input_queue_head; 3508 } 3509 3510 rflow->cpu = next_cpu; 3511 return rflow; 3512 } 3513 3514 /* 3515 * get_rps_cpu is called from netif_receive_skb and returns the target 3516 * CPU from the RPS map of the receiving queue for a given skb. 3517 * rcu_read_lock must be held on entry. 3518 */ 3519 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3520 struct rps_dev_flow **rflowp) 3521 { 3522 const struct rps_sock_flow_table *sock_flow_table; 3523 struct netdev_rx_queue *rxqueue = dev->_rx; 3524 struct rps_dev_flow_table *flow_table; 3525 struct rps_map *map; 3526 int cpu = -1; 3527 u32 tcpu; 3528 u32 hash; 3529 3530 if (skb_rx_queue_recorded(skb)) { 3531 u16 index = skb_get_rx_queue(skb); 3532 3533 if (unlikely(index >= dev->real_num_rx_queues)) { 3534 WARN_ONCE(dev->real_num_rx_queues > 1, 3535 "%s received packet on queue %u, but number " 3536 "of RX queues is %u\n", 3537 dev->name, index, dev->real_num_rx_queues); 3538 goto done; 3539 } 3540 rxqueue += index; 3541 } 3542 3543 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 3544 3545 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3546 map = rcu_dereference(rxqueue->rps_map); 3547 if (!flow_table && !map) 3548 goto done; 3549 3550 skb_reset_network_header(skb); 3551 hash = skb_get_hash(skb); 3552 if (!hash) 3553 goto done; 3554 3555 sock_flow_table = rcu_dereference(rps_sock_flow_table); 3556 if (flow_table && sock_flow_table) { 3557 struct rps_dev_flow *rflow; 3558 u32 next_cpu; 3559 u32 ident; 3560 3561 /* First check into global flow table if there is a match */ 3562 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 3563 if ((ident ^ hash) & ~rps_cpu_mask) 3564 goto try_rps; 3565 3566 next_cpu = ident & rps_cpu_mask; 3567 3568 /* OK, now we know there is a match, 3569 * we can look at the local (per receive queue) flow table 3570 */ 3571 rflow = &flow_table->flows[hash & flow_table->mask]; 3572 tcpu = rflow->cpu; 3573 3574 /* 3575 * If the desired CPU (where last recvmsg was done) is 3576 * different from current CPU (one in the rx-queue flow 3577 * table entry), switch if one of the following holds: 3578 * - Current CPU is unset (>= nr_cpu_ids). 3579 * - Current CPU is offline. 3580 * - The current CPU's queue tail has advanced beyond the 3581 * last packet that was enqueued using this table entry. 3582 * This guarantees that all previous packets for the flow 3583 * have been dequeued, thus preserving in order delivery. 3584 */ 3585 if (unlikely(tcpu != next_cpu) && 3586 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 3587 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 3588 rflow->last_qtail)) >= 0)) { 3589 tcpu = next_cpu; 3590 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 3591 } 3592 3593 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 3594 *rflowp = rflow; 3595 cpu = tcpu; 3596 goto done; 3597 } 3598 } 3599 3600 try_rps: 3601 3602 if (map) { 3603 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 3604 if (cpu_online(tcpu)) { 3605 cpu = tcpu; 3606 goto done; 3607 } 3608 } 3609 3610 done: 3611 return cpu; 3612 } 3613 3614 #ifdef CONFIG_RFS_ACCEL 3615 3616 /** 3617 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 3618 * @dev: Device on which the filter was set 3619 * @rxq_index: RX queue index 3620 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 3621 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 3622 * 3623 * Drivers that implement ndo_rx_flow_steer() should periodically call 3624 * this function for each installed filter and remove the filters for 3625 * which it returns %true. 3626 */ 3627 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 3628 u32 flow_id, u16 filter_id) 3629 { 3630 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 3631 struct rps_dev_flow_table *flow_table; 3632 struct rps_dev_flow *rflow; 3633 bool expire = true; 3634 unsigned int cpu; 3635 3636 rcu_read_lock(); 3637 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3638 if (flow_table && flow_id <= flow_table->mask) { 3639 rflow = &flow_table->flows[flow_id]; 3640 cpu = ACCESS_ONCE(rflow->cpu); 3641 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 3642 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 3643 rflow->last_qtail) < 3644 (int)(10 * flow_table->mask))) 3645 expire = false; 3646 } 3647 rcu_read_unlock(); 3648 return expire; 3649 } 3650 EXPORT_SYMBOL(rps_may_expire_flow); 3651 3652 #endif /* CONFIG_RFS_ACCEL */ 3653 3654 /* Called from hardirq (IPI) context */ 3655 static void rps_trigger_softirq(void *data) 3656 { 3657 struct softnet_data *sd = data; 3658 3659 ____napi_schedule(sd, &sd->backlog); 3660 sd->received_rps++; 3661 } 3662 3663 #endif /* CONFIG_RPS */ 3664 3665 /* 3666 * Check if this softnet_data structure is another cpu one 3667 * If yes, queue it to our IPI list and return 1 3668 * If no, return 0 3669 */ 3670 static int rps_ipi_queued(struct softnet_data *sd) 3671 { 3672 #ifdef CONFIG_RPS 3673 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 3674 3675 if (sd != mysd) { 3676 sd->rps_ipi_next = mysd->rps_ipi_list; 3677 mysd->rps_ipi_list = sd; 3678 3679 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3680 return 1; 3681 } 3682 #endif /* CONFIG_RPS */ 3683 return 0; 3684 } 3685 3686 #ifdef CONFIG_NET_FLOW_LIMIT 3687 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 3688 #endif 3689 3690 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 3691 { 3692 #ifdef CONFIG_NET_FLOW_LIMIT 3693 struct sd_flow_limit *fl; 3694 struct softnet_data *sd; 3695 unsigned int old_flow, new_flow; 3696 3697 if (qlen < (netdev_max_backlog >> 1)) 3698 return false; 3699 3700 sd = this_cpu_ptr(&softnet_data); 3701 3702 rcu_read_lock(); 3703 fl = rcu_dereference(sd->flow_limit); 3704 if (fl) { 3705 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 3706 old_flow = fl->history[fl->history_head]; 3707 fl->history[fl->history_head] = new_flow; 3708 3709 fl->history_head++; 3710 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 3711 3712 if (likely(fl->buckets[old_flow])) 3713 fl->buckets[old_flow]--; 3714 3715 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 3716 fl->count++; 3717 rcu_read_unlock(); 3718 return true; 3719 } 3720 } 3721 rcu_read_unlock(); 3722 #endif 3723 return false; 3724 } 3725 3726 /* 3727 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 3728 * queue (may be a remote CPU queue). 3729 */ 3730 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 3731 unsigned int *qtail) 3732 { 3733 struct softnet_data *sd; 3734 unsigned long flags; 3735 unsigned int qlen; 3736 3737 sd = &per_cpu(softnet_data, cpu); 3738 3739 local_irq_save(flags); 3740 3741 rps_lock(sd); 3742 if (!netif_running(skb->dev)) 3743 goto drop; 3744 qlen = skb_queue_len(&sd->input_pkt_queue); 3745 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 3746 if (qlen) { 3747 enqueue: 3748 __skb_queue_tail(&sd->input_pkt_queue, skb); 3749 input_queue_tail_incr_save(sd, qtail); 3750 rps_unlock(sd); 3751 local_irq_restore(flags); 3752 return NET_RX_SUCCESS; 3753 } 3754 3755 /* Schedule NAPI for backlog device 3756 * We can use non atomic operation since we own the queue lock 3757 */ 3758 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 3759 if (!rps_ipi_queued(sd)) 3760 ____napi_schedule(sd, &sd->backlog); 3761 } 3762 goto enqueue; 3763 } 3764 3765 drop: 3766 sd->dropped++; 3767 rps_unlock(sd); 3768 3769 local_irq_restore(flags); 3770 3771 atomic_long_inc(&skb->dev->rx_dropped); 3772 kfree_skb(skb); 3773 return NET_RX_DROP; 3774 } 3775 3776 static int netif_rx_internal(struct sk_buff *skb) 3777 { 3778 int ret; 3779 3780 net_timestamp_check(netdev_tstamp_prequeue, skb); 3781 3782 trace_netif_rx(skb); 3783 #ifdef CONFIG_RPS 3784 if (static_key_false(&rps_needed)) { 3785 struct rps_dev_flow voidflow, *rflow = &voidflow; 3786 int cpu; 3787 3788 preempt_disable(); 3789 rcu_read_lock(); 3790 3791 cpu = get_rps_cpu(skb->dev, skb, &rflow); 3792 if (cpu < 0) 3793 cpu = smp_processor_id(); 3794 3795 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 3796 3797 rcu_read_unlock(); 3798 preempt_enable(); 3799 } else 3800 #endif 3801 { 3802 unsigned int qtail; 3803 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 3804 put_cpu(); 3805 } 3806 return ret; 3807 } 3808 3809 /** 3810 * netif_rx - post buffer to the network code 3811 * @skb: buffer to post 3812 * 3813 * This function receives a packet from a device driver and queues it for 3814 * the upper (protocol) levels to process. It always succeeds. The buffer 3815 * may be dropped during processing for congestion control or by the 3816 * protocol layers. 3817 * 3818 * return values: 3819 * NET_RX_SUCCESS (no congestion) 3820 * NET_RX_DROP (packet was dropped) 3821 * 3822 */ 3823 3824 int netif_rx(struct sk_buff *skb) 3825 { 3826 trace_netif_rx_entry(skb); 3827 3828 return netif_rx_internal(skb); 3829 } 3830 EXPORT_SYMBOL(netif_rx); 3831 3832 int netif_rx_ni(struct sk_buff *skb) 3833 { 3834 int err; 3835 3836 trace_netif_rx_ni_entry(skb); 3837 3838 preempt_disable(); 3839 err = netif_rx_internal(skb); 3840 if (local_softirq_pending()) 3841 do_softirq(); 3842 preempt_enable(); 3843 3844 return err; 3845 } 3846 EXPORT_SYMBOL(netif_rx_ni); 3847 3848 static __latent_entropy void net_tx_action(struct softirq_action *h) 3849 { 3850 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 3851 3852 if (sd->completion_queue) { 3853 struct sk_buff *clist; 3854 3855 local_irq_disable(); 3856 clist = sd->completion_queue; 3857 sd->completion_queue = NULL; 3858 local_irq_enable(); 3859 3860 while (clist) { 3861 struct sk_buff *skb = clist; 3862 clist = clist->next; 3863 3864 WARN_ON(atomic_read(&skb->users)); 3865 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 3866 trace_consume_skb(skb); 3867 else 3868 trace_kfree_skb(skb, net_tx_action); 3869 3870 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 3871 __kfree_skb(skb); 3872 else 3873 __kfree_skb_defer(skb); 3874 } 3875 3876 __kfree_skb_flush(); 3877 } 3878 3879 if (sd->output_queue) { 3880 struct Qdisc *head; 3881 3882 local_irq_disable(); 3883 head = sd->output_queue; 3884 sd->output_queue = NULL; 3885 sd->output_queue_tailp = &sd->output_queue; 3886 local_irq_enable(); 3887 3888 while (head) { 3889 struct Qdisc *q = head; 3890 spinlock_t *root_lock; 3891 3892 head = head->next_sched; 3893 3894 root_lock = qdisc_lock(q); 3895 spin_lock(root_lock); 3896 /* We need to make sure head->next_sched is read 3897 * before clearing __QDISC_STATE_SCHED 3898 */ 3899 smp_mb__before_atomic(); 3900 clear_bit(__QDISC_STATE_SCHED, &q->state); 3901 qdisc_run(q); 3902 spin_unlock(root_lock); 3903 } 3904 } 3905 } 3906 3907 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 3908 /* This hook is defined here for ATM LANE */ 3909 int (*br_fdb_test_addr_hook)(struct net_device *dev, 3910 unsigned char *addr) __read_mostly; 3911 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 3912 #endif 3913 3914 static inline struct sk_buff * 3915 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3916 struct net_device *orig_dev) 3917 { 3918 #ifdef CONFIG_NET_CLS_ACT 3919 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); 3920 struct tcf_result cl_res; 3921 3922 /* If there's at least one ingress present somewhere (so 3923 * we get here via enabled static key), remaining devices 3924 * that are not configured with an ingress qdisc will bail 3925 * out here. 3926 */ 3927 if (!cl) 3928 return skb; 3929 if (*pt_prev) { 3930 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3931 *pt_prev = NULL; 3932 } 3933 3934 qdisc_skb_cb(skb)->pkt_len = skb->len; 3935 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); 3936 qdisc_bstats_cpu_update(cl->q, skb); 3937 3938 switch (tc_classify(skb, cl, &cl_res, false)) { 3939 case TC_ACT_OK: 3940 case TC_ACT_RECLASSIFY: 3941 skb->tc_index = TC_H_MIN(cl_res.classid); 3942 break; 3943 case TC_ACT_SHOT: 3944 qdisc_qstats_cpu_drop(cl->q); 3945 kfree_skb(skb); 3946 return NULL; 3947 case TC_ACT_STOLEN: 3948 case TC_ACT_QUEUED: 3949 consume_skb(skb); 3950 return NULL; 3951 case TC_ACT_REDIRECT: 3952 /* skb_mac_header check was done by cls/act_bpf, so 3953 * we can safely push the L2 header back before 3954 * redirecting to another netdev 3955 */ 3956 __skb_push(skb, skb->mac_len); 3957 skb_do_redirect(skb); 3958 return NULL; 3959 default: 3960 break; 3961 } 3962 #endif /* CONFIG_NET_CLS_ACT */ 3963 return skb; 3964 } 3965 3966 /** 3967 * netdev_is_rx_handler_busy - check if receive handler is registered 3968 * @dev: device to check 3969 * 3970 * Check if a receive handler is already registered for a given device. 3971 * Return true if there one. 3972 * 3973 * The caller must hold the rtnl_mutex. 3974 */ 3975 bool netdev_is_rx_handler_busy(struct net_device *dev) 3976 { 3977 ASSERT_RTNL(); 3978 return dev && rtnl_dereference(dev->rx_handler); 3979 } 3980 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 3981 3982 /** 3983 * netdev_rx_handler_register - register receive handler 3984 * @dev: device to register a handler for 3985 * @rx_handler: receive handler to register 3986 * @rx_handler_data: data pointer that is used by rx handler 3987 * 3988 * Register a receive handler for a device. This handler will then be 3989 * called from __netif_receive_skb. A negative errno code is returned 3990 * on a failure. 3991 * 3992 * The caller must hold the rtnl_mutex. 3993 * 3994 * For a general description of rx_handler, see enum rx_handler_result. 3995 */ 3996 int netdev_rx_handler_register(struct net_device *dev, 3997 rx_handler_func_t *rx_handler, 3998 void *rx_handler_data) 3999 { 4000 ASSERT_RTNL(); 4001 4002 if (dev->rx_handler) 4003 return -EBUSY; 4004 4005 /* Note: rx_handler_data must be set before rx_handler */ 4006 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 4007 rcu_assign_pointer(dev->rx_handler, rx_handler); 4008 4009 return 0; 4010 } 4011 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 4012 4013 /** 4014 * netdev_rx_handler_unregister - unregister receive handler 4015 * @dev: device to unregister a handler from 4016 * 4017 * Unregister a receive handler from a device. 4018 * 4019 * The caller must hold the rtnl_mutex. 4020 */ 4021 void netdev_rx_handler_unregister(struct net_device *dev) 4022 { 4023 4024 ASSERT_RTNL(); 4025 RCU_INIT_POINTER(dev->rx_handler, NULL); 4026 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 4027 * section has a guarantee to see a non NULL rx_handler_data 4028 * as well. 4029 */ 4030 synchronize_net(); 4031 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 4032 } 4033 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 4034 4035 /* 4036 * Limit the use of PFMEMALLOC reserves to those protocols that implement 4037 * the special handling of PFMEMALLOC skbs. 4038 */ 4039 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 4040 { 4041 switch (skb->protocol) { 4042 case htons(ETH_P_ARP): 4043 case htons(ETH_P_IP): 4044 case htons(ETH_P_IPV6): 4045 case htons(ETH_P_8021Q): 4046 case htons(ETH_P_8021AD): 4047 return true; 4048 default: 4049 return false; 4050 } 4051 } 4052 4053 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 4054 int *ret, struct net_device *orig_dev) 4055 { 4056 #ifdef CONFIG_NETFILTER_INGRESS 4057 if (nf_hook_ingress_active(skb)) { 4058 int ingress_retval; 4059 4060 if (*pt_prev) { 4061 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4062 *pt_prev = NULL; 4063 } 4064 4065 rcu_read_lock(); 4066 ingress_retval = nf_hook_ingress(skb); 4067 rcu_read_unlock(); 4068 return ingress_retval; 4069 } 4070 #endif /* CONFIG_NETFILTER_INGRESS */ 4071 return 0; 4072 } 4073 4074 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) 4075 { 4076 struct packet_type *ptype, *pt_prev; 4077 rx_handler_func_t *rx_handler; 4078 struct net_device *orig_dev; 4079 bool deliver_exact = false; 4080 int ret = NET_RX_DROP; 4081 __be16 type; 4082 4083 net_timestamp_check(!netdev_tstamp_prequeue, skb); 4084 4085 trace_netif_receive_skb(skb); 4086 4087 orig_dev = skb->dev; 4088 4089 skb_reset_network_header(skb); 4090 if (!skb_transport_header_was_set(skb)) 4091 skb_reset_transport_header(skb); 4092 skb_reset_mac_len(skb); 4093 4094 pt_prev = NULL; 4095 4096 another_round: 4097 skb->skb_iif = skb->dev->ifindex; 4098 4099 __this_cpu_inc(softnet_data.processed); 4100 4101 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 4102 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 4103 skb = skb_vlan_untag(skb); 4104 if (unlikely(!skb)) 4105 goto out; 4106 } 4107 4108 #ifdef CONFIG_NET_CLS_ACT 4109 if (skb->tc_verd & TC_NCLS) { 4110 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); 4111 goto ncls; 4112 } 4113 #endif 4114 4115 if (pfmemalloc) 4116 goto skip_taps; 4117 4118 list_for_each_entry_rcu(ptype, &ptype_all, list) { 4119 if (pt_prev) 4120 ret = deliver_skb(skb, pt_prev, orig_dev); 4121 pt_prev = ptype; 4122 } 4123 4124 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 4125 if (pt_prev) 4126 ret = deliver_skb(skb, pt_prev, orig_dev); 4127 pt_prev = ptype; 4128 } 4129 4130 skip_taps: 4131 #ifdef CONFIG_NET_INGRESS 4132 if (static_key_false(&ingress_needed)) { 4133 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); 4134 if (!skb) 4135 goto out; 4136 4137 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 4138 goto out; 4139 } 4140 #endif 4141 #ifdef CONFIG_NET_CLS_ACT 4142 skb->tc_verd = 0; 4143 ncls: 4144 #endif 4145 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 4146 goto drop; 4147 4148 if (skb_vlan_tag_present(skb)) { 4149 if (pt_prev) { 4150 ret = deliver_skb(skb, pt_prev, orig_dev); 4151 pt_prev = NULL; 4152 } 4153 if (vlan_do_receive(&skb)) 4154 goto another_round; 4155 else if (unlikely(!skb)) 4156 goto out; 4157 } 4158 4159 rx_handler = rcu_dereference(skb->dev->rx_handler); 4160 if (rx_handler) { 4161 if (pt_prev) { 4162 ret = deliver_skb(skb, pt_prev, orig_dev); 4163 pt_prev = NULL; 4164 } 4165 switch (rx_handler(&skb)) { 4166 case RX_HANDLER_CONSUMED: 4167 ret = NET_RX_SUCCESS; 4168 goto out; 4169 case RX_HANDLER_ANOTHER: 4170 goto another_round; 4171 case RX_HANDLER_EXACT: 4172 deliver_exact = true; 4173 case RX_HANDLER_PASS: 4174 break; 4175 default: 4176 BUG(); 4177 } 4178 } 4179 4180 if (unlikely(skb_vlan_tag_present(skb))) { 4181 if (skb_vlan_tag_get_id(skb)) 4182 skb->pkt_type = PACKET_OTHERHOST; 4183 /* Note: we might in the future use prio bits 4184 * and set skb->priority like in vlan_do_receive() 4185 * For the time being, just ignore Priority Code Point 4186 */ 4187 skb->vlan_tci = 0; 4188 } 4189 4190 type = skb->protocol; 4191 4192 /* deliver only exact match when indicated */ 4193 if (likely(!deliver_exact)) { 4194 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4195 &ptype_base[ntohs(type) & 4196 PTYPE_HASH_MASK]); 4197 } 4198 4199 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4200 &orig_dev->ptype_specific); 4201 4202 if (unlikely(skb->dev != orig_dev)) { 4203 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4204 &skb->dev->ptype_specific); 4205 } 4206 4207 if (pt_prev) { 4208 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 4209 goto drop; 4210 else 4211 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 4212 } else { 4213 drop: 4214 if (!deliver_exact) 4215 atomic_long_inc(&skb->dev->rx_dropped); 4216 else 4217 atomic_long_inc(&skb->dev->rx_nohandler); 4218 kfree_skb(skb); 4219 /* Jamal, now you will not able to escape explaining 4220 * me how you were going to use this. :-) 4221 */ 4222 ret = NET_RX_DROP; 4223 } 4224 4225 out: 4226 return ret; 4227 } 4228 4229 static int __netif_receive_skb(struct sk_buff *skb) 4230 { 4231 int ret; 4232 4233 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 4234 unsigned long pflags = current->flags; 4235 4236 /* 4237 * PFMEMALLOC skbs are special, they should 4238 * - be delivered to SOCK_MEMALLOC sockets only 4239 * - stay away from userspace 4240 * - have bounded memory usage 4241 * 4242 * Use PF_MEMALLOC as this saves us from propagating the allocation 4243 * context down to all allocation sites. 4244 */ 4245 current->flags |= PF_MEMALLOC; 4246 ret = __netif_receive_skb_core(skb, true); 4247 tsk_restore_flags(current, pflags, PF_MEMALLOC); 4248 } else 4249 ret = __netif_receive_skb_core(skb, false); 4250 4251 return ret; 4252 } 4253 4254 static int netif_receive_skb_internal(struct sk_buff *skb) 4255 { 4256 int ret; 4257 4258 net_timestamp_check(netdev_tstamp_prequeue, skb); 4259 4260 if (skb_defer_rx_timestamp(skb)) 4261 return NET_RX_SUCCESS; 4262 4263 rcu_read_lock(); 4264 4265 #ifdef CONFIG_RPS 4266 if (static_key_false(&rps_needed)) { 4267 struct rps_dev_flow voidflow, *rflow = &voidflow; 4268 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 4269 4270 if (cpu >= 0) { 4271 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4272 rcu_read_unlock(); 4273 return ret; 4274 } 4275 } 4276 #endif 4277 ret = __netif_receive_skb(skb); 4278 rcu_read_unlock(); 4279 return ret; 4280 } 4281 4282 /** 4283 * netif_receive_skb - process receive buffer from network 4284 * @skb: buffer to process 4285 * 4286 * netif_receive_skb() is the main receive data processing function. 4287 * It always succeeds. The buffer may be dropped during processing 4288 * for congestion control or by the protocol layers. 4289 * 4290 * This function may only be called from softirq context and interrupts 4291 * should be enabled. 4292 * 4293 * Return values (usually ignored): 4294 * NET_RX_SUCCESS: no congestion 4295 * NET_RX_DROP: packet was dropped 4296 */ 4297 int netif_receive_skb(struct sk_buff *skb) 4298 { 4299 trace_netif_receive_skb_entry(skb); 4300 4301 return netif_receive_skb_internal(skb); 4302 } 4303 EXPORT_SYMBOL(netif_receive_skb); 4304 4305 DEFINE_PER_CPU(struct work_struct, flush_works); 4306 4307 /* Network device is going away, flush any packets still pending */ 4308 static void flush_backlog(struct work_struct *work) 4309 { 4310 struct sk_buff *skb, *tmp; 4311 struct softnet_data *sd; 4312 4313 local_bh_disable(); 4314 sd = this_cpu_ptr(&softnet_data); 4315 4316 local_irq_disable(); 4317 rps_lock(sd); 4318 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 4319 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 4320 __skb_unlink(skb, &sd->input_pkt_queue); 4321 kfree_skb(skb); 4322 input_queue_head_incr(sd); 4323 } 4324 } 4325 rps_unlock(sd); 4326 local_irq_enable(); 4327 4328 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 4329 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 4330 __skb_unlink(skb, &sd->process_queue); 4331 kfree_skb(skb); 4332 input_queue_head_incr(sd); 4333 } 4334 } 4335 local_bh_enable(); 4336 } 4337 4338 static void flush_all_backlogs(void) 4339 { 4340 unsigned int cpu; 4341 4342 get_online_cpus(); 4343 4344 for_each_online_cpu(cpu) 4345 queue_work_on(cpu, system_highpri_wq, 4346 per_cpu_ptr(&flush_works, cpu)); 4347 4348 for_each_online_cpu(cpu) 4349 flush_work(per_cpu_ptr(&flush_works, cpu)); 4350 4351 put_online_cpus(); 4352 } 4353 4354 static int napi_gro_complete(struct sk_buff *skb) 4355 { 4356 struct packet_offload *ptype; 4357 __be16 type = skb->protocol; 4358 struct list_head *head = &offload_base; 4359 int err = -ENOENT; 4360 4361 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 4362 4363 if (NAPI_GRO_CB(skb)->count == 1) { 4364 skb_shinfo(skb)->gso_size = 0; 4365 goto out; 4366 } 4367 4368 rcu_read_lock(); 4369 list_for_each_entry_rcu(ptype, head, list) { 4370 if (ptype->type != type || !ptype->callbacks.gro_complete) 4371 continue; 4372 4373 err = ptype->callbacks.gro_complete(skb, 0); 4374 break; 4375 } 4376 rcu_read_unlock(); 4377 4378 if (err) { 4379 WARN_ON(&ptype->list == head); 4380 kfree_skb(skb); 4381 return NET_RX_SUCCESS; 4382 } 4383 4384 out: 4385 return netif_receive_skb_internal(skb); 4386 } 4387 4388 /* napi->gro_list contains packets ordered by age. 4389 * youngest packets at the head of it. 4390 * Complete skbs in reverse order to reduce latencies. 4391 */ 4392 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 4393 { 4394 struct sk_buff *skb, *prev = NULL; 4395 4396 /* scan list and build reverse chain */ 4397 for (skb = napi->gro_list; skb != NULL; skb = skb->next) { 4398 skb->prev = prev; 4399 prev = skb; 4400 } 4401 4402 for (skb = prev; skb; skb = prev) { 4403 skb->next = NULL; 4404 4405 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 4406 return; 4407 4408 prev = skb->prev; 4409 napi_gro_complete(skb); 4410 napi->gro_count--; 4411 } 4412 4413 napi->gro_list = NULL; 4414 } 4415 EXPORT_SYMBOL(napi_gro_flush); 4416 4417 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) 4418 { 4419 struct sk_buff *p; 4420 unsigned int maclen = skb->dev->hard_header_len; 4421 u32 hash = skb_get_hash_raw(skb); 4422 4423 for (p = napi->gro_list; p; p = p->next) { 4424 unsigned long diffs; 4425 4426 NAPI_GRO_CB(p)->flush = 0; 4427 4428 if (hash != skb_get_hash_raw(p)) { 4429 NAPI_GRO_CB(p)->same_flow = 0; 4430 continue; 4431 } 4432 4433 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 4434 diffs |= p->vlan_tci ^ skb->vlan_tci; 4435 diffs |= skb_metadata_dst_cmp(p, skb); 4436 if (maclen == ETH_HLEN) 4437 diffs |= compare_ether_header(skb_mac_header(p), 4438 skb_mac_header(skb)); 4439 else if (!diffs) 4440 diffs = memcmp(skb_mac_header(p), 4441 skb_mac_header(skb), 4442 maclen); 4443 NAPI_GRO_CB(p)->same_flow = !diffs; 4444 } 4445 } 4446 4447 static void skb_gro_reset_offset(struct sk_buff *skb) 4448 { 4449 const struct skb_shared_info *pinfo = skb_shinfo(skb); 4450 const skb_frag_t *frag0 = &pinfo->frags[0]; 4451 4452 NAPI_GRO_CB(skb)->data_offset = 0; 4453 NAPI_GRO_CB(skb)->frag0 = NULL; 4454 NAPI_GRO_CB(skb)->frag0_len = 0; 4455 4456 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 4457 pinfo->nr_frags && 4458 !PageHighMem(skb_frag_page(frag0))) { 4459 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 4460 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); 4461 } 4462 } 4463 4464 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 4465 { 4466 struct skb_shared_info *pinfo = skb_shinfo(skb); 4467 4468 BUG_ON(skb->end - skb->tail < grow); 4469 4470 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 4471 4472 skb->data_len -= grow; 4473 skb->tail += grow; 4474 4475 pinfo->frags[0].page_offset += grow; 4476 skb_frag_size_sub(&pinfo->frags[0], grow); 4477 4478 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 4479 skb_frag_unref(skb, 0); 4480 memmove(pinfo->frags, pinfo->frags + 1, 4481 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 4482 } 4483 } 4484 4485 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4486 { 4487 struct sk_buff **pp = NULL; 4488 struct packet_offload *ptype; 4489 __be16 type = skb->protocol; 4490 struct list_head *head = &offload_base; 4491 int same_flow; 4492 enum gro_result ret; 4493 int grow; 4494 4495 if (!(skb->dev->features & NETIF_F_GRO)) 4496 goto normal; 4497 4498 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) 4499 goto normal; 4500 4501 gro_list_prepare(napi, skb); 4502 4503 rcu_read_lock(); 4504 list_for_each_entry_rcu(ptype, head, list) { 4505 if (ptype->type != type || !ptype->callbacks.gro_receive) 4506 continue; 4507 4508 skb_set_network_header(skb, skb_gro_offset(skb)); 4509 skb_reset_mac_len(skb); 4510 NAPI_GRO_CB(skb)->same_flow = 0; 4511 NAPI_GRO_CB(skb)->flush = 0; 4512 NAPI_GRO_CB(skb)->free = 0; 4513 NAPI_GRO_CB(skb)->encap_mark = 0; 4514 NAPI_GRO_CB(skb)->is_fou = 0; 4515 NAPI_GRO_CB(skb)->is_atomic = 1; 4516 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 4517 4518 /* Setup for GRO checksum validation */ 4519 switch (skb->ip_summed) { 4520 case CHECKSUM_COMPLETE: 4521 NAPI_GRO_CB(skb)->csum = skb->csum; 4522 NAPI_GRO_CB(skb)->csum_valid = 1; 4523 NAPI_GRO_CB(skb)->csum_cnt = 0; 4524 break; 4525 case CHECKSUM_UNNECESSARY: 4526 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 4527 NAPI_GRO_CB(skb)->csum_valid = 0; 4528 break; 4529 default: 4530 NAPI_GRO_CB(skb)->csum_cnt = 0; 4531 NAPI_GRO_CB(skb)->csum_valid = 0; 4532 } 4533 4534 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); 4535 break; 4536 } 4537 rcu_read_unlock(); 4538 4539 if (&ptype->list == head) 4540 goto normal; 4541 4542 same_flow = NAPI_GRO_CB(skb)->same_flow; 4543 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 4544 4545 if (pp) { 4546 struct sk_buff *nskb = *pp; 4547 4548 *pp = nskb->next; 4549 nskb->next = NULL; 4550 napi_gro_complete(nskb); 4551 napi->gro_count--; 4552 } 4553 4554 if (same_flow) 4555 goto ok; 4556 4557 if (NAPI_GRO_CB(skb)->flush) 4558 goto normal; 4559 4560 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { 4561 struct sk_buff *nskb = napi->gro_list; 4562 4563 /* locate the end of the list to select the 'oldest' flow */ 4564 while (nskb->next) { 4565 pp = &nskb->next; 4566 nskb = *pp; 4567 } 4568 *pp = NULL; 4569 nskb->next = NULL; 4570 napi_gro_complete(nskb); 4571 } else { 4572 napi->gro_count++; 4573 } 4574 NAPI_GRO_CB(skb)->count = 1; 4575 NAPI_GRO_CB(skb)->age = jiffies; 4576 NAPI_GRO_CB(skb)->last = skb; 4577 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 4578 skb->next = napi->gro_list; 4579 napi->gro_list = skb; 4580 ret = GRO_HELD; 4581 4582 pull: 4583 grow = skb_gro_offset(skb) - skb_headlen(skb); 4584 if (grow > 0) 4585 gro_pull_from_frag0(skb, grow); 4586 ok: 4587 return ret; 4588 4589 normal: 4590 ret = GRO_NORMAL; 4591 goto pull; 4592 } 4593 4594 struct packet_offload *gro_find_receive_by_type(__be16 type) 4595 { 4596 struct list_head *offload_head = &offload_base; 4597 struct packet_offload *ptype; 4598 4599 list_for_each_entry_rcu(ptype, offload_head, list) { 4600 if (ptype->type != type || !ptype->callbacks.gro_receive) 4601 continue; 4602 return ptype; 4603 } 4604 return NULL; 4605 } 4606 EXPORT_SYMBOL(gro_find_receive_by_type); 4607 4608 struct packet_offload *gro_find_complete_by_type(__be16 type) 4609 { 4610 struct list_head *offload_head = &offload_base; 4611 struct packet_offload *ptype; 4612 4613 list_for_each_entry_rcu(ptype, offload_head, list) { 4614 if (ptype->type != type || !ptype->callbacks.gro_complete) 4615 continue; 4616 return ptype; 4617 } 4618 return NULL; 4619 } 4620 EXPORT_SYMBOL(gro_find_complete_by_type); 4621 4622 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 4623 { 4624 switch (ret) { 4625 case GRO_NORMAL: 4626 if (netif_receive_skb_internal(skb)) 4627 ret = GRO_DROP; 4628 break; 4629 4630 case GRO_DROP: 4631 kfree_skb(skb); 4632 break; 4633 4634 case GRO_MERGED_FREE: 4635 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) { 4636 skb_dst_drop(skb); 4637 kmem_cache_free(skbuff_head_cache, skb); 4638 } else { 4639 __kfree_skb(skb); 4640 } 4641 break; 4642 4643 case GRO_HELD: 4644 case GRO_MERGED: 4645 break; 4646 } 4647 4648 return ret; 4649 } 4650 4651 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4652 { 4653 skb_mark_napi_id(skb, napi); 4654 trace_napi_gro_receive_entry(skb); 4655 4656 skb_gro_reset_offset(skb); 4657 4658 return napi_skb_finish(dev_gro_receive(napi, skb), skb); 4659 } 4660 EXPORT_SYMBOL(napi_gro_receive); 4661 4662 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 4663 { 4664 if (unlikely(skb->pfmemalloc)) { 4665 consume_skb(skb); 4666 return; 4667 } 4668 __skb_pull(skb, skb_headlen(skb)); 4669 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 4670 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 4671 skb->vlan_tci = 0; 4672 skb->dev = napi->dev; 4673 skb->skb_iif = 0; 4674 skb->encapsulation = 0; 4675 skb_shinfo(skb)->gso_type = 0; 4676 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 4677 4678 napi->skb = skb; 4679 } 4680 4681 struct sk_buff *napi_get_frags(struct napi_struct *napi) 4682 { 4683 struct sk_buff *skb = napi->skb; 4684 4685 if (!skb) { 4686 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 4687 if (skb) { 4688 napi->skb = skb; 4689 skb_mark_napi_id(skb, napi); 4690 } 4691 } 4692 return skb; 4693 } 4694 EXPORT_SYMBOL(napi_get_frags); 4695 4696 static gro_result_t napi_frags_finish(struct napi_struct *napi, 4697 struct sk_buff *skb, 4698 gro_result_t ret) 4699 { 4700 switch (ret) { 4701 case GRO_NORMAL: 4702 case GRO_HELD: 4703 __skb_push(skb, ETH_HLEN); 4704 skb->protocol = eth_type_trans(skb, skb->dev); 4705 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 4706 ret = GRO_DROP; 4707 break; 4708 4709 case GRO_DROP: 4710 case GRO_MERGED_FREE: 4711 napi_reuse_skb(napi, skb); 4712 break; 4713 4714 case GRO_MERGED: 4715 break; 4716 } 4717 4718 return ret; 4719 } 4720 4721 /* Upper GRO stack assumes network header starts at gro_offset=0 4722 * Drivers could call both napi_gro_frags() and napi_gro_receive() 4723 * We copy ethernet header into skb->data to have a common layout. 4724 */ 4725 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 4726 { 4727 struct sk_buff *skb = napi->skb; 4728 const struct ethhdr *eth; 4729 unsigned int hlen = sizeof(*eth); 4730 4731 napi->skb = NULL; 4732 4733 skb_reset_mac_header(skb); 4734 skb_gro_reset_offset(skb); 4735 4736 eth = skb_gro_header_fast(skb, 0); 4737 if (unlikely(skb_gro_header_hard(skb, hlen))) { 4738 eth = skb_gro_header_slow(skb, hlen, 0); 4739 if (unlikely(!eth)) { 4740 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 4741 __func__, napi->dev->name); 4742 napi_reuse_skb(napi, skb); 4743 return NULL; 4744 } 4745 } else { 4746 gro_pull_from_frag0(skb, hlen); 4747 NAPI_GRO_CB(skb)->frag0 += hlen; 4748 NAPI_GRO_CB(skb)->frag0_len -= hlen; 4749 } 4750 __skb_pull(skb, hlen); 4751 4752 /* 4753 * This works because the only protocols we care about don't require 4754 * special handling. 4755 * We'll fix it up properly in napi_frags_finish() 4756 */ 4757 skb->protocol = eth->h_proto; 4758 4759 return skb; 4760 } 4761 4762 gro_result_t napi_gro_frags(struct napi_struct *napi) 4763 { 4764 struct sk_buff *skb = napi_frags_skb(napi); 4765 4766 if (!skb) 4767 return GRO_DROP; 4768 4769 trace_napi_gro_frags_entry(skb); 4770 4771 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 4772 } 4773 EXPORT_SYMBOL(napi_gro_frags); 4774 4775 /* Compute the checksum from gro_offset and return the folded value 4776 * after adding in any pseudo checksum. 4777 */ 4778 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 4779 { 4780 __wsum wsum; 4781 __sum16 sum; 4782 4783 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 4784 4785 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 4786 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 4787 if (likely(!sum)) { 4788 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 4789 !skb->csum_complete_sw) 4790 netdev_rx_csum_fault(skb->dev); 4791 } 4792 4793 NAPI_GRO_CB(skb)->csum = wsum; 4794 NAPI_GRO_CB(skb)->csum_valid = 1; 4795 4796 return sum; 4797 } 4798 EXPORT_SYMBOL(__skb_gro_checksum_complete); 4799 4800 /* 4801 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 4802 * Note: called with local irq disabled, but exits with local irq enabled. 4803 */ 4804 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 4805 { 4806 #ifdef CONFIG_RPS 4807 struct softnet_data *remsd = sd->rps_ipi_list; 4808 4809 if (remsd) { 4810 sd->rps_ipi_list = NULL; 4811 4812 local_irq_enable(); 4813 4814 /* Send pending IPI's to kick RPS processing on remote cpus. */ 4815 while (remsd) { 4816 struct softnet_data *next = remsd->rps_ipi_next; 4817 4818 if (cpu_online(remsd->cpu)) 4819 smp_call_function_single_async(remsd->cpu, 4820 &remsd->csd); 4821 remsd = next; 4822 } 4823 } else 4824 #endif 4825 local_irq_enable(); 4826 } 4827 4828 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 4829 { 4830 #ifdef CONFIG_RPS 4831 return sd->rps_ipi_list != NULL; 4832 #else 4833 return false; 4834 #endif 4835 } 4836 4837 static int process_backlog(struct napi_struct *napi, int quota) 4838 { 4839 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 4840 bool again = true; 4841 int work = 0; 4842 4843 /* Check if we have pending ipi, its better to send them now, 4844 * not waiting net_rx_action() end. 4845 */ 4846 if (sd_has_rps_ipi_waiting(sd)) { 4847 local_irq_disable(); 4848 net_rps_action_and_irq_enable(sd); 4849 } 4850 4851 napi->weight = weight_p; 4852 while (again) { 4853 struct sk_buff *skb; 4854 4855 while ((skb = __skb_dequeue(&sd->process_queue))) { 4856 rcu_read_lock(); 4857 __netif_receive_skb(skb); 4858 rcu_read_unlock(); 4859 input_queue_head_incr(sd); 4860 if (++work >= quota) 4861 return work; 4862 4863 } 4864 4865 local_irq_disable(); 4866 rps_lock(sd); 4867 if (skb_queue_empty(&sd->input_pkt_queue)) { 4868 /* 4869 * Inline a custom version of __napi_complete(). 4870 * only current cpu owns and manipulates this napi, 4871 * and NAPI_STATE_SCHED is the only possible flag set 4872 * on backlog. 4873 * We can use a plain write instead of clear_bit(), 4874 * and we dont need an smp_mb() memory barrier. 4875 */ 4876 napi->state = 0; 4877 again = false; 4878 } else { 4879 skb_queue_splice_tail_init(&sd->input_pkt_queue, 4880 &sd->process_queue); 4881 } 4882 rps_unlock(sd); 4883 local_irq_enable(); 4884 } 4885 4886 return work; 4887 } 4888 4889 /** 4890 * __napi_schedule - schedule for receive 4891 * @n: entry to schedule 4892 * 4893 * The entry's receive function will be scheduled to run. 4894 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 4895 */ 4896 void __napi_schedule(struct napi_struct *n) 4897 { 4898 unsigned long flags; 4899 4900 local_irq_save(flags); 4901 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4902 local_irq_restore(flags); 4903 } 4904 EXPORT_SYMBOL(__napi_schedule); 4905 4906 /** 4907 * __napi_schedule_irqoff - schedule for receive 4908 * @n: entry to schedule 4909 * 4910 * Variant of __napi_schedule() assuming hard irqs are masked 4911 */ 4912 void __napi_schedule_irqoff(struct napi_struct *n) 4913 { 4914 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4915 } 4916 EXPORT_SYMBOL(__napi_schedule_irqoff); 4917 4918 void __napi_complete(struct napi_struct *n) 4919 { 4920 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 4921 4922 list_del_init(&n->poll_list); 4923 smp_mb__before_atomic(); 4924 clear_bit(NAPI_STATE_SCHED, &n->state); 4925 } 4926 EXPORT_SYMBOL(__napi_complete); 4927 4928 void napi_complete_done(struct napi_struct *n, int work_done) 4929 { 4930 unsigned long flags; 4931 4932 /* 4933 * don't let napi dequeue from the cpu poll list 4934 * just in case its running on a different cpu 4935 */ 4936 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) 4937 return; 4938 4939 if (n->gro_list) { 4940 unsigned long timeout = 0; 4941 4942 if (work_done) 4943 timeout = n->dev->gro_flush_timeout; 4944 4945 if (timeout) 4946 hrtimer_start(&n->timer, ns_to_ktime(timeout), 4947 HRTIMER_MODE_REL_PINNED); 4948 else 4949 napi_gro_flush(n, false); 4950 } 4951 if (likely(list_empty(&n->poll_list))) { 4952 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); 4953 } else { 4954 /* If n->poll_list is not empty, we need to mask irqs */ 4955 local_irq_save(flags); 4956 __napi_complete(n); 4957 local_irq_restore(flags); 4958 } 4959 } 4960 EXPORT_SYMBOL(napi_complete_done); 4961 4962 /* must be called under rcu_read_lock(), as we dont take a reference */ 4963 static struct napi_struct *napi_by_id(unsigned int napi_id) 4964 { 4965 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 4966 struct napi_struct *napi; 4967 4968 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 4969 if (napi->napi_id == napi_id) 4970 return napi; 4971 4972 return NULL; 4973 } 4974 4975 #if defined(CONFIG_NET_RX_BUSY_POLL) 4976 #define BUSY_POLL_BUDGET 8 4977 bool sk_busy_loop(struct sock *sk, int nonblock) 4978 { 4979 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0; 4980 int (*busy_poll)(struct napi_struct *dev); 4981 struct napi_struct *napi; 4982 int rc = false; 4983 4984 rcu_read_lock(); 4985 4986 napi = napi_by_id(sk->sk_napi_id); 4987 if (!napi) 4988 goto out; 4989 4990 /* Note: ndo_busy_poll method is optional in linux-4.5 */ 4991 busy_poll = napi->dev->netdev_ops->ndo_busy_poll; 4992 4993 do { 4994 rc = 0; 4995 local_bh_disable(); 4996 if (busy_poll) { 4997 rc = busy_poll(napi); 4998 } else if (napi_schedule_prep(napi)) { 4999 void *have = netpoll_poll_lock(napi); 5000 5001 if (test_bit(NAPI_STATE_SCHED, &napi->state)) { 5002 rc = napi->poll(napi, BUSY_POLL_BUDGET); 5003 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); 5004 if (rc == BUSY_POLL_BUDGET) { 5005 napi_complete_done(napi, rc); 5006 napi_schedule(napi); 5007 } 5008 } 5009 netpoll_poll_unlock(have); 5010 } 5011 if (rc > 0) 5012 __NET_ADD_STATS(sock_net(sk), 5013 LINUX_MIB_BUSYPOLLRXPACKETS, rc); 5014 local_bh_enable(); 5015 5016 if (rc == LL_FLUSH_FAILED) 5017 break; /* permanent failure */ 5018 5019 cpu_relax(); 5020 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) && 5021 !need_resched() && !busy_loop_timeout(end_time)); 5022 5023 rc = !skb_queue_empty(&sk->sk_receive_queue); 5024 out: 5025 rcu_read_unlock(); 5026 return rc; 5027 } 5028 EXPORT_SYMBOL(sk_busy_loop); 5029 5030 #endif /* CONFIG_NET_RX_BUSY_POLL */ 5031 5032 void napi_hash_add(struct napi_struct *napi) 5033 { 5034 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || 5035 test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) 5036 return; 5037 5038 spin_lock(&napi_hash_lock); 5039 5040 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */ 5041 do { 5042 if (unlikely(++napi_gen_id < NR_CPUS + 1)) 5043 napi_gen_id = NR_CPUS + 1; 5044 } while (napi_by_id(napi_gen_id)); 5045 napi->napi_id = napi_gen_id; 5046 5047 hlist_add_head_rcu(&napi->napi_hash_node, 5048 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 5049 5050 spin_unlock(&napi_hash_lock); 5051 } 5052 EXPORT_SYMBOL_GPL(napi_hash_add); 5053 5054 /* Warning : caller is responsible to make sure rcu grace period 5055 * is respected before freeing memory containing @napi 5056 */ 5057 bool napi_hash_del(struct napi_struct *napi) 5058 { 5059 bool rcu_sync_needed = false; 5060 5061 spin_lock(&napi_hash_lock); 5062 5063 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { 5064 rcu_sync_needed = true; 5065 hlist_del_rcu(&napi->napi_hash_node); 5066 } 5067 spin_unlock(&napi_hash_lock); 5068 return rcu_sync_needed; 5069 } 5070 EXPORT_SYMBOL_GPL(napi_hash_del); 5071 5072 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 5073 { 5074 struct napi_struct *napi; 5075 5076 napi = container_of(timer, struct napi_struct, timer); 5077 if (napi->gro_list) 5078 napi_schedule(napi); 5079 5080 return HRTIMER_NORESTART; 5081 } 5082 5083 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 5084 int (*poll)(struct napi_struct *, int), int weight) 5085 { 5086 INIT_LIST_HEAD(&napi->poll_list); 5087 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 5088 napi->timer.function = napi_watchdog; 5089 napi->gro_count = 0; 5090 napi->gro_list = NULL; 5091 napi->skb = NULL; 5092 napi->poll = poll; 5093 if (weight > NAPI_POLL_WEIGHT) 5094 pr_err_once("netif_napi_add() called with weight %d on device %s\n", 5095 weight, dev->name); 5096 napi->weight = weight; 5097 list_add(&napi->dev_list, &dev->napi_list); 5098 napi->dev = dev; 5099 #ifdef CONFIG_NETPOLL 5100 spin_lock_init(&napi->poll_lock); 5101 napi->poll_owner = -1; 5102 #endif 5103 set_bit(NAPI_STATE_SCHED, &napi->state); 5104 napi_hash_add(napi); 5105 } 5106 EXPORT_SYMBOL(netif_napi_add); 5107 5108 void napi_disable(struct napi_struct *n) 5109 { 5110 might_sleep(); 5111 set_bit(NAPI_STATE_DISABLE, &n->state); 5112 5113 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 5114 msleep(1); 5115 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 5116 msleep(1); 5117 5118 hrtimer_cancel(&n->timer); 5119 5120 clear_bit(NAPI_STATE_DISABLE, &n->state); 5121 } 5122 EXPORT_SYMBOL(napi_disable); 5123 5124 /* Must be called in process context */ 5125 void netif_napi_del(struct napi_struct *napi) 5126 { 5127 might_sleep(); 5128 if (napi_hash_del(napi)) 5129 synchronize_net(); 5130 list_del_init(&napi->dev_list); 5131 napi_free_frags(napi); 5132 5133 kfree_skb_list(napi->gro_list); 5134 napi->gro_list = NULL; 5135 napi->gro_count = 0; 5136 } 5137 EXPORT_SYMBOL(netif_napi_del); 5138 5139 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 5140 { 5141 void *have; 5142 int work, weight; 5143 5144 list_del_init(&n->poll_list); 5145 5146 have = netpoll_poll_lock(n); 5147 5148 weight = n->weight; 5149 5150 /* This NAPI_STATE_SCHED test is for avoiding a race 5151 * with netpoll's poll_napi(). Only the entity which 5152 * obtains the lock and sees NAPI_STATE_SCHED set will 5153 * actually make the ->poll() call. Therefore we avoid 5154 * accidentally calling ->poll() when NAPI is not scheduled. 5155 */ 5156 work = 0; 5157 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 5158 work = n->poll(n, weight); 5159 trace_napi_poll(n, work, weight); 5160 } 5161 5162 WARN_ON_ONCE(work > weight); 5163 5164 if (likely(work < weight)) 5165 goto out_unlock; 5166 5167 /* Drivers must not modify the NAPI state if they 5168 * consume the entire weight. In such cases this code 5169 * still "owns" the NAPI instance and therefore can 5170 * move the instance around on the list at-will. 5171 */ 5172 if (unlikely(napi_disable_pending(n))) { 5173 napi_complete(n); 5174 goto out_unlock; 5175 } 5176 5177 if (n->gro_list) { 5178 /* flush too old packets 5179 * If HZ < 1000, flush all packets. 5180 */ 5181 napi_gro_flush(n, HZ >= 1000); 5182 } 5183 5184 /* Some drivers may have called napi_schedule 5185 * prior to exhausting their budget. 5186 */ 5187 if (unlikely(!list_empty(&n->poll_list))) { 5188 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 5189 n->dev ? n->dev->name : "backlog"); 5190 goto out_unlock; 5191 } 5192 5193 list_add_tail(&n->poll_list, repoll); 5194 5195 out_unlock: 5196 netpoll_poll_unlock(have); 5197 5198 return work; 5199 } 5200 5201 static __latent_entropy void net_rx_action(struct softirq_action *h) 5202 { 5203 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5204 unsigned long time_limit = jiffies + 2; 5205 int budget = netdev_budget; 5206 LIST_HEAD(list); 5207 LIST_HEAD(repoll); 5208 5209 local_irq_disable(); 5210 list_splice_init(&sd->poll_list, &list); 5211 local_irq_enable(); 5212 5213 for (;;) { 5214 struct napi_struct *n; 5215 5216 if (list_empty(&list)) { 5217 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 5218 return; 5219 break; 5220 } 5221 5222 n = list_first_entry(&list, struct napi_struct, poll_list); 5223 budget -= napi_poll(n, &repoll); 5224 5225 /* If softirq window is exhausted then punt. 5226 * Allow this to run for 2 jiffies since which will allow 5227 * an average latency of 1.5/HZ. 5228 */ 5229 if (unlikely(budget <= 0 || 5230 time_after_eq(jiffies, time_limit))) { 5231 sd->time_squeeze++; 5232 break; 5233 } 5234 } 5235 5236 __kfree_skb_flush(); 5237 local_irq_disable(); 5238 5239 list_splice_tail_init(&sd->poll_list, &list); 5240 list_splice_tail(&repoll, &list); 5241 list_splice(&list, &sd->poll_list); 5242 if (!list_empty(&sd->poll_list)) 5243 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 5244 5245 net_rps_action_and_irq_enable(sd); 5246 } 5247 5248 struct netdev_adjacent { 5249 struct net_device *dev; 5250 5251 /* upper master flag, there can only be one master device per list */ 5252 bool master; 5253 5254 /* counter for the number of times this device was added to us */ 5255 u16 ref_nr; 5256 5257 /* private field for the users */ 5258 void *private; 5259 5260 struct list_head list; 5261 struct rcu_head rcu; 5262 }; 5263 5264 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 5265 struct list_head *adj_list) 5266 { 5267 struct netdev_adjacent *adj; 5268 5269 list_for_each_entry(adj, adj_list, list) { 5270 if (adj->dev == adj_dev) 5271 return adj; 5272 } 5273 return NULL; 5274 } 5275 5276 /** 5277 * netdev_has_upper_dev - Check if device is linked to an upper device 5278 * @dev: device 5279 * @upper_dev: upper device to check 5280 * 5281 * Find out if a device is linked to specified upper device and return true 5282 * in case it is. Note that this checks only immediate upper device, 5283 * not through a complete stack of devices. The caller must hold the RTNL lock. 5284 */ 5285 bool netdev_has_upper_dev(struct net_device *dev, 5286 struct net_device *upper_dev) 5287 { 5288 ASSERT_RTNL(); 5289 5290 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper); 5291 } 5292 EXPORT_SYMBOL(netdev_has_upper_dev); 5293 5294 /** 5295 * netdev_has_any_upper_dev - Check if device is linked to some device 5296 * @dev: device 5297 * 5298 * Find out if a device is linked to an upper device and return true in case 5299 * it is. The caller must hold the RTNL lock. 5300 */ 5301 static bool netdev_has_any_upper_dev(struct net_device *dev) 5302 { 5303 ASSERT_RTNL(); 5304 5305 return !list_empty(&dev->all_adj_list.upper); 5306 } 5307 5308 /** 5309 * netdev_master_upper_dev_get - Get master upper device 5310 * @dev: device 5311 * 5312 * Find a master upper device and return pointer to it or NULL in case 5313 * it's not there. The caller must hold the RTNL lock. 5314 */ 5315 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 5316 { 5317 struct netdev_adjacent *upper; 5318 5319 ASSERT_RTNL(); 5320 5321 if (list_empty(&dev->adj_list.upper)) 5322 return NULL; 5323 5324 upper = list_first_entry(&dev->adj_list.upper, 5325 struct netdev_adjacent, list); 5326 if (likely(upper->master)) 5327 return upper->dev; 5328 return NULL; 5329 } 5330 EXPORT_SYMBOL(netdev_master_upper_dev_get); 5331 5332 void *netdev_adjacent_get_private(struct list_head *adj_list) 5333 { 5334 struct netdev_adjacent *adj; 5335 5336 adj = list_entry(adj_list, struct netdev_adjacent, list); 5337 5338 return adj->private; 5339 } 5340 EXPORT_SYMBOL(netdev_adjacent_get_private); 5341 5342 /** 5343 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 5344 * @dev: device 5345 * @iter: list_head ** of the current position 5346 * 5347 * Gets the next device from the dev's upper list, starting from iter 5348 * position. The caller must hold RCU read lock. 5349 */ 5350 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 5351 struct list_head **iter) 5352 { 5353 struct netdev_adjacent *upper; 5354 5355 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 5356 5357 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5358 5359 if (&upper->list == &dev->adj_list.upper) 5360 return NULL; 5361 5362 *iter = &upper->list; 5363 5364 return upper->dev; 5365 } 5366 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 5367 5368 /** 5369 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list 5370 * @dev: device 5371 * @iter: list_head ** of the current position 5372 * 5373 * Gets the next device from the dev's upper list, starting from iter 5374 * position. The caller must hold RCU read lock. 5375 */ 5376 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, 5377 struct list_head **iter) 5378 { 5379 struct netdev_adjacent *upper; 5380 5381 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 5382 5383 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5384 5385 if (&upper->list == &dev->all_adj_list.upper) 5386 return NULL; 5387 5388 *iter = &upper->list; 5389 5390 return upper->dev; 5391 } 5392 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); 5393 5394 /** 5395 * netdev_lower_get_next_private - Get the next ->private from the 5396 * lower neighbour list 5397 * @dev: device 5398 * @iter: list_head ** of the current position 5399 * 5400 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5401 * list, starting from iter position. The caller must hold either hold the 5402 * RTNL lock or its own locking that guarantees that the neighbour lower 5403 * list will remain unchanged. 5404 */ 5405 void *netdev_lower_get_next_private(struct net_device *dev, 5406 struct list_head **iter) 5407 { 5408 struct netdev_adjacent *lower; 5409 5410 lower = list_entry(*iter, struct netdev_adjacent, list); 5411 5412 if (&lower->list == &dev->adj_list.lower) 5413 return NULL; 5414 5415 *iter = lower->list.next; 5416 5417 return lower->private; 5418 } 5419 EXPORT_SYMBOL(netdev_lower_get_next_private); 5420 5421 /** 5422 * netdev_lower_get_next_private_rcu - Get the next ->private from the 5423 * lower neighbour list, RCU 5424 * variant 5425 * @dev: device 5426 * @iter: list_head ** of the current position 5427 * 5428 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5429 * list, starting from iter position. The caller must hold RCU read lock. 5430 */ 5431 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 5432 struct list_head **iter) 5433 { 5434 struct netdev_adjacent *lower; 5435 5436 WARN_ON_ONCE(!rcu_read_lock_held()); 5437 5438 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5439 5440 if (&lower->list == &dev->adj_list.lower) 5441 return NULL; 5442 5443 *iter = &lower->list; 5444 5445 return lower->private; 5446 } 5447 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 5448 5449 /** 5450 * netdev_lower_get_next - Get the next device from the lower neighbour 5451 * list 5452 * @dev: device 5453 * @iter: list_head ** of the current position 5454 * 5455 * Gets the next netdev_adjacent from the dev's lower neighbour 5456 * list, starting from iter position. The caller must hold RTNL lock or 5457 * its own locking that guarantees that the neighbour lower 5458 * list will remain unchanged. 5459 */ 5460 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 5461 { 5462 struct netdev_adjacent *lower; 5463 5464 lower = list_entry(*iter, struct netdev_adjacent, list); 5465 5466 if (&lower->list == &dev->adj_list.lower) 5467 return NULL; 5468 5469 *iter = lower->list.next; 5470 5471 return lower->dev; 5472 } 5473 EXPORT_SYMBOL(netdev_lower_get_next); 5474 5475 /** 5476 * netdev_all_lower_get_next - Get the next device from all lower neighbour list 5477 * @dev: device 5478 * @iter: list_head ** of the current position 5479 * 5480 * Gets the next netdev_adjacent from the dev's all lower neighbour 5481 * list, starting from iter position. The caller must hold RTNL lock or 5482 * its own locking that guarantees that the neighbour all lower 5483 * list will remain unchanged. 5484 */ 5485 struct net_device *netdev_all_lower_get_next(struct net_device *dev, struct list_head **iter) 5486 { 5487 struct netdev_adjacent *lower; 5488 5489 lower = list_entry(*iter, struct netdev_adjacent, list); 5490 5491 if (&lower->list == &dev->all_adj_list.lower) 5492 return NULL; 5493 5494 *iter = lower->list.next; 5495 5496 return lower->dev; 5497 } 5498 EXPORT_SYMBOL(netdev_all_lower_get_next); 5499 5500 /** 5501 * netdev_all_lower_get_next_rcu - Get the next device from all 5502 * lower neighbour list, RCU variant 5503 * @dev: device 5504 * @iter: list_head ** of the current position 5505 * 5506 * Gets the next netdev_adjacent from the dev's all lower neighbour 5507 * list, starting from iter position. The caller must hold RCU read lock. 5508 */ 5509 struct net_device *netdev_all_lower_get_next_rcu(struct net_device *dev, 5510 struct list_head **iter) 5511 { 5512 struct netdev_adjacent *lower; 5513 5514 lower = list_first_or_null_rcu(&dev->all_adj_list.lower, 5515 struct netdev_adjacent, list); 5516 5517 return lower ? lower->dev : NULL; 5518 } 5519 EXPORT_SYMBOL(netdev_all_lower_get_next_rcu); 5520 5521 /** 5522 * netdev_lower_get_first_private_rcu - Get the first ->private from the 5523 * lower neighbour list, RCU 5524 * variant 5525 * @dev: device 5526 * 5527 * Gets the first netdev_adjacent->private from the dev's lower neighbour 5528 * list. The caller must hold RCU read lock. 5529 */ 5530 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 5531 { 5532 struct netdev_adjacent *lower; 5533 5534 lower = list_first_or_null_rcu(&dev->adj_list.lower, 5535 struct netdev_adjacent, list); 5536 if (lower) 5537 return lower->private; 5538 return NULL; 5539 } 5540 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 5541 5542 /** 5543 * netdev_master_upper_dev_get_rcu - Get master upper device 5544 * @dev: device 5545 * 5546 * Find a master upper device and return pointer to it or NULL in case 5547 * it's not there. The caller must hold the RCU read lock. 5548 */ 5549 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 5550 { 5551 struct netdev_adjacent *upper; 5552 5553 upper = list_first_or_null_rcu(&dev->adj_list.upper, 5554 struct netdev_adjacent, list); 5555 if (upper && likely(upper->master)) 5556 return upper->dev; 5557 return NULL; 5558 } 5559 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 5560 5561 static int netdev_adjacent_sysfs_add(struct net_device *dev, 5562 struct net_device *adj_dev, 5563 struct list_head *dev_list) 5564 { 5565 char linkname[IFNAMSIZ+7]; 5566 sprintf(linkname, dev_list == &dev->adj_list.upper ? 5567 "upper_%s" : "lower_%s", adj_dev->name); 5568 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 5569 linkname); 5570 } 5571 static void netdev_adjacent_sysfs_del(struct net_device *dev, 5572 char *name, 5573 struct list_head *dev_list) 5574 { 5575 char linkname[IFNAMSIZ+7]; 5576 sprintf(linkname, dev_list == &dev->adj_list.upper ? 5577 "upper_%s" : "lower_%s", name); 5578 sysfs_remove_link(&(dev->dev.kobj), linkname); 5579 } 5580 5581 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 5582 struct net_device *adj_dev, 5583 struct list_head *dev_list) 5584 { 5585 return (dev_list == &dev->adj_list.upper || 5586 dev_list == &dev->adj_list.lower) && 5587 net_eq(dev_net(dev), dev_net(adj_dev)); 5588 } 5589 5590 static int __netdev_adjacent_dev_insert(struct net_device *dev, 5591 struct net_device *adj_dev, 5592 u16 ref_nr, 5593 struct list_head *dev_list, 5594 void *private, bool master) 5595 { 5596 struct netdev_adjacent *adj; 5597 int ret; 5598 5599 adj = __netdev_find_adj(adj_dev, dev_list); 5600 5601 if (adj) { 5602 adj->ref_nr += ref_nr; 5603 return 0; 5604 } 5605 5606 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 5607 if (!adj) 5608 return -ENOMEM; 5609 5610 adj->dev = adj_dev; 5611 adj->master = master; 5612 adj->ref_nr = ref_nr; 5613 adj->private = private; 5614 dev_hold(adj_dev); 5615 5616 pr_debug("dev_hold for %s, because of link added from %s to %s\n", 5617 adj_dev->name, dev->name, adj_dev->name); 5618 5619 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 5620 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 5621 if (ret) 5622 goto free_adj; 5623 } 5624 5625 /* Ensure that master link is always the first item in list. */ 5626 if (master) { 5627 ret = sysfs_create_link(&(dev->dev.kobj), 5628 &(adj_dev->dev.kobj), "master"); 5629 if (ret) 5630 goto remove_symlinks; 5631 5632 list_add_rcu(&adj->list, dev_list); 5633 } else { 5634 list_add_tail_rcu(&adj->list, dev_list); 5635 } 5636 5637 return 0; 5638 5639 remove_symlinks: 5640 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 5641 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 5642 free_adj: 5643 kfree(adj); 5644 dev_put(adj_dev); 5645 5646 return ret; 5647 } 5648 5649 static void __netdev_adjacent_dev_remove(struct net_device *dev, 5650 struct net_device *adj_dev, 5651 u16 ref_nr, 5652 struct list_head *dev_list) 5653 { 5654 struct netdev_adjacent *adj; 5655 5656 adj = __netdev_find_adj(adj_dev, dev_list); 5657 5658 if (!adj) { 5659 pr_err("tried to remove device %s from %s\n", 5660 dev->name, adj_dev->name); 5661 BUG(); 5662 } 5663 5664 if (adj->ref_nr > ref_nr) { 5665 pr_debug("%s to %s ref_nr-%d = %d\n", dev->name, adj_dev->name, 5666 ref_nr, adj->ref_nr-ref_nr); 5667 adj->ref_nr -= ref_nr; 5668 return; 5669 } 5670 5671 if (adj->master) 5672 sysfs_remove_link(&(dev->dev.kobj), "master"); 5673 5674 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 5675 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 5676 5677 list_del_rcu(&adj->list); 5678 pr_debug("dev_put for %s, because link removed from %s to %s\n", 5679 adj_dev->name, dev->name, adj_dev->name); 5680 dev_put(adj_dev); 5681 kfree_rcu(adj, rcu); 5682 } 5683 5684 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 5685 struct net_device *upper_dev, 5686 u16 ref_nr, 5687 struct list_head *up_list, 5688 struct list_head *down_list, 5689 void *private, bool master) 5690 { 5691 int ret; 5692 5693 ret = __netdev_adjacent_dev_insert(dev, upper_dev, ref_nr, up_list, 5694 private, master); 5695 if (ret) 5696 return ret; 5697 5698 ret = __netdev_adjacent_dev_insert(upper_dev, dev, ref_nr, down_list, 5699 private, false); 5700 if (ret) { 5701 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 5702 return ret; 5703 } 5704 5705 return 0; 5706 } 5707 5708 static int __netdev_adjacent_dev_link(struct net_device *dev, 5709 struct net_device *upper_dev, 5710 u16 ref_nr) 5711 { 5712 return __netdev_adjacent_dev_link_lists(dev, upper_dev, ref_nr, 5713 &dev->all_adj_list.upper, 5714 &upper_dev->all_adj_list.lower, 5715 NULL, false); 5716 } 5717 5718 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 5719 struct net_device *upper_dev, 5720 u16 ref_nr, 5721 struct list_head *up_list, 5722 struct list_head *down_list) 5723 { 5724 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 5725 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 5726 } 5727 5728 static void __netdev_adjacent_dev_unlink(struct net_device *dev, 5729 struct net_device *upper_dev, 5730 u16 ref_nr) 5731 { 5732 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, ref_nr, 5733 &dev->all_adj_list.upper, 5734 &upper_dev->all_adj_list.lower); 5735 } 5736 5737 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 5738 struct net_device *upper_dev, 5739 void *private, bool master) 5740 { 5741 int ret = __netdev_adjacent_dev_link(dev, upper_dev, 1); 5742 5743 if (ret) 5744 return ret; 5745 5746 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 1, 5747 &dev->adj_list.upper, 5748 &upper_dev->adj_list.lower, 5749 private, master); 5750 if (ret) { 5751 __netdev_adjacent_dev_unlink(dev, upper_dev, 1); 5752 return ret; 5753 } 5754 5755 return 0; 5756 } 5757 5758 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 5759 struct net_device *upper_dev) 5760 { 5761 __netdev_adjacent_dev_unlink(dev, upper_dev, 1); 5762 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 5763 &dev->adj_list.upper, 5764 &upper_dev->adj_list.lower); 5765 } 5766 5767 static int __netdev_upper_dev_link(struct net_device *dev, 5768 struct net_device *upper_dev, bool master, 5769 void *upper_priv, void *upper_info) 5770 { 5771 struct netdev_notifier_changeupper_info changeupper_info; 5772 struct netdev_adjacent *i, *j, *to_i, *to_j; 5773 int ret = 0; 5774 5775 ASSERT_RTNL(); 5776 5777 if (dev == upper_dev) 5778 return -EBUSY; 5779 5780 /* To prevent loops, check if dev is not upper device to upper_dev. */ 5781 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper)) 5782 return -EBUSY; 5783 5784 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper)) 5785 return -EEXIST; 5786 5787 if (master && netdev_master_upper_dev_get(dev)) 5788 return -EBUSY; 5789 5790 changeupper_info.upper_dev = upper_dev; 5791 changeupper_info.master = master; 5792 changeupper_info.linking = true; 5793 changeupper_info.upper_info = upper_info; 5794 5795 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, 5796 &changeupper_info.info); 5797 ret = notifier_to_errno(ret); 5798 if (ret) 5799 return ret; 5800 5801 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 5802 master); 5803 if (ret) 5804 return ret; 5805 5806 /* Now that we linked these devs, make all the upper_dev's 5807 * all_adj_list.upper visible to every dev's all_adj_list.lower an 5808 * versa, and don't forget the devices itself. All of these 5809 * links are non-neighbours. 5810 */ 5811 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5812 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { 5813 pr_debug("Interlinking %s with %s, non-neighbour\n", 5814 i->dev->name, j->dev->name); 5815 ret = __netdev_adjacent_dev_link(i->dev, j->dev, i->ref_nr); 5816 if (ret) 5817 goto rollback_mesh; 5818 } 5819 } 5820 5821 /* add dev to every upper_dev's upper device */ 5822 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { 5823 pr_debug("linking %s's upper device %s with %s\n", 5824 upper_dev->name, i->dev->name, dev->name); 5825 ret = __netdev_adjacent_dev_link(dev, i->dev, i->ref_nr); 5826 if (ret) 5827 goto rollback_upper_mesh; 5828 } 5829 5830 /* add upper_dev to every dev's lower device */ 5831 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5832 pr_debug("linking %s's lower device %s with %s\n", dev->name, 5833 i->dev->name, upper_dev->name); 5834 ret = __netdev_adjacent_dev_link(i->dev, upper_dev, i->ref_nr); 5835 if (ret) 5836 goto rollback_lower_mesh; 5837 } 5838 5839 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, 5840 &changeupper_info.info); 5841 ret = notifier_to_errno(ret); 5842 if (ret) 5843 goto rollback_lower_mesh; 5844 5845 return 0; 5846 5847 rollback_lower_mesh: 5848 to_i = i; 5849 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5850 if (i == to_i) 5851 break; 5852 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr); 5853 } 5854 5855 i = NULL; 5856 5857 rollback_upper_mesh: 5858 to_i = i; 5859 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { 5860 if (i == to_i) 5861 break; 5862 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr); 5863 } 5864 5865 i = j = NULL; 5866 5867 rollback_mesh: 5868 to_i = i; 5869 to_j = j; 5870 list_for_each_entry(i, &dev->all_adj_list.lower, list) { 5871 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { 5872 if (i == to_i && j == to_j) 5873 break; 5874 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr); 5875 } 5876 if (i == to_i) 5877 break; 5878 } 5879 5880 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5881 5882 return ret; 5883 } 5884 5885 /** 5886 * netdev_upper_dev_link - Add a link to the upper device 5887 * @dev: device 5888 * @upper_dev: new upper device 5889 * 5890 * Adds a link to device which is upper to this one. The caller must hold 5891 * the RTNL lock. On a failure a negative errno code is returned. 5892 * On success the reference counts are adjusted and the function 5893 * returns zero. 5894 */ 5895 int netdev_upper_dev_link(struct net_device *dev, 5896 struct net_device *upper_dev) 5897 { 5898 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); 5899 } 5900 EXPORT_SYMBOL(netdev_upper_dev_link); 5901 5902 /** 5903 * netdev_master_upper_dev_link - Add a master link to the upper device 5904 * @dev: device 5905 * @upper_dev: new upper device 5906 * @upper_priv: upper device private 5907 * @upper_info: upper info to be passed down via notifier 5908 * 5909 * Adds a link to device which is upper to this one. In this case, only 5910 * one master upper device can be linked, although other non-master devices 5911 * might be linked as well. The caller must hold the RTNL lock. 5912 * On a failure a negative errno code is returned. On success the reference 5913 * counts are adjusted and the function returns zero. 5914 */ 5915 int netdev_master_upper_dev_link(struct net_device *dev, 5916 struct net_device *upper_dev, 5917 void *upper_priv, void *upper_info) 5918 { 5919 return __netdev_upper_dev_link(dev, upper_dev, true, 5920 upper_priv, upper_info); 5921 } 5922 EXPORT_SYMBOL(netdev_master_upper_dev_link); 5923 5924 /** 5925 * netdev_upper_dev_unlink - Removes a link to upper device 5926 * @dev: device 5927 * @upper_dev: new upper device 5928 * 5929 * Removes a link to device which is upper to this one. The caller must hold 5930 * the RTNL lock. 5931 */ 5932 void netdev_upper_dev_unlink(struct net_device *dev, 5933 struct net_device *upper_dev) 5934 { 5935 struct netdev_notifier_changeupper_info changeupper_info; 5936 struct netdev_adjacent *i, *j; 5937 ASSERT_RTNL(); 5938 5939 changeupper_info.upper_dev = upper_dev; 5940 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 5941 changeupper_info.linking = false; 5942 5943 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, 5944 &changeupper_info.info); 5945 5946 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5947 5948 /* Here is the tricky part. We must remove all dev's lower 5949 * devices from all upper_dev's upper devices and vice 5950 * versa, to maintain the graph relationship. 5951 */ 5952 list_for_each_entry(i, &dev->all_adj_list.lower, list) 5953 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) 5954 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr); 5955 5956 /* remove also the devices itself from lower/upper device 5957 * list 5958 */ 5959 list_for_each_entry(i, &dev->all_adj_list.lower, list) 5960 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr); 5961 5962 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) 5963 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr); 5964 5965 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, 5966 &changeupper_info.info); 5967 } 5968 EXPORT_SYMBOL(netdev_upper_dev_unlink); 5969 5970 /** 5971 * netdev_bonding_info_change - Dispatch event about slave change 5972 * @dev: device 5973 * @bonding_info: info to dispatch 5974 * 5975 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 5976 * The caller must hold the RTNL lock. 5977 */ 5978 void netdev_bonding_info_change(struct net_device *dev, 5979 struct netdev_bonding_info *bonding_info) 5980 { 5981 struct netdev_notifier_bonding_info info; 5982 5983 memcpy(&info.bonding_info, bonding_info, 5984 sizeof(struct netdev_bonding_info)); 5985 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, 5986 &info.info); 5987 } 5988 EXPORT_SYMBOL(netdev_bonding_info_change); 5989 5990 static void netdev_adjacent_add_links(struct net_device *dev) 5991 { 5992 struct netdev_adjacent *iter; 5993 5994 struct net *net = dev_net(dev); 5995 5996 list_for_each_entry(iter, &dev->adj_list.upper, list) { 5997 if (!net_eq(net, dev_net(iter->dev))) 5998 continue; 5999 netdev_adjacent_sysfs_add(iter->dev, dev, 6000 &iter->dev->adj_list.lower); 6001 netdev_adjacent_sysfs_add(dev, iter->dev, 6002 &dev->adj_list.upper); 6003 } 6004 6005 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6006 if (!net_eq(net, dev_net(iter->dev))) 6007 continue; 6008 netdev_adjacent_sysfs_add(iter->dev, dev, 6009 &iter->dev->adj_list.upper); 6010 netdev_adjacent_sysfs_add(dev, iter->dev, 6011 &dev->adj_list.lower); 6012 } 6013 } 6014 6015 static void netdev_adjacent_del_links(struct net_device *dev) 6016 { 6017 struct netdev_adjacent *iter; 6018 6019 struct net *net = dev_net(dev); 6020 6021 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6022 if (!net_eq(net, dev_net(iter->dev))) 6023 continue; 6024 netdev_adjacent_sysfs_del(iter->dev, dev->name, 6025 &iter->dev->adj_list.lower); 6026 netdev_adjacent_sysfs_del(dev, iter->dev->name, 6027 &dev->adj_list.upper); 6028 } 6029 6030 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6031 if (!net_eq(net, dev_net(iter->dev))) 6032 continue; 6033 netdev_adjacent_sysfs_del(iter->dev, dev->name, 6034 &iter->dev->adj_list.upper); 6035 netdev_adjacent_sysfs_del(dev, iter->dev->name, 6036 &dev->adj_list.lower); 6037 } 6038 } 6039 6040 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 6041 { 6042 struct netdev_adjacent *iter; 6043 6044 struct net *net = dev_net(dev); 6045 6046 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6047 if (!net_eq(net, dev_net(iter->dev))) 6048 continue; 6049 netdev_adjacent_sysfs_del(iter->dev, oldname, 6050 &iter->dev->adj_list.lower); 6051 netdev_adjacent_sysfs_add(iter->dev, dev, 6052 &iter->dev->adj_list.lower); 6053 } 6054 6055 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6056 if (!net_eq(net, dev_net(iter->dev))) 6057 continue; 6058 netdev_adjacent_sysfs_del(iter->dev, oldname, 6059 &iter->dev->adj_list.upper); 6060 netdev_adjacent_sysfs_add(iter->dev, dev, 6061 &iter->dev->adj_list.upper); 6062 } 6063 } 6064 6065 void *netdev_lower_dev_get_private(struct net_device *dev, 6066 struct net_device *lower_dev) 6067 { 6068 struct netdev_adjacent *lower; 6069 6070 if (!lower_dev) 6071 return NULL; 6072 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 6073 if (!lower) 6074 return NULL; 6075 6076 return lower->private; 6077 } 6078 EXPORT_SYMBOL(netdev_lower_dev_get_private); 6079 6080 6081 int dev_get_nest_level(struct net_device *dev) 6082 { 6083 struct net_device *lower = NULL; 6084 struct list_head *iter; 6085 int max_nest = -1; 6086 int nest; 6087 6088 ASSERT_RTNL(); 6089 6090 netdev_for_each_lower_dev(dev, lower, iter) { 6091 nest = dev_get_nest_level(lower); 6092 if (max_nest < nest) 6093 max_nest = nest; 6094 } 6095 6096 return max_nest + 1; 6097 } 6098 EXPORT_SYMBOL(dev_get_nest_level); 6099 6100 /** 6101 * netdev_lower_change - Dispatch event about lower device state change 6102 * @lower_dev: device 6103 * @lower_state_info: state to dispatch 6104 * 6105 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 6106 * The caller must hold the RTNL lock. 6107 */ 6108 void netdev_lower_state_changed(struct net_device *lower_dev, 6109 void *lower_state_info) 6110 { 6111 struct netdev_notifier_changelowerstate_info changelowerstate_info; 6112 6113 ASSERT_RTNL(); 6114 changelowerstate_info.lower_state_info = lower_state_info; 6115 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, 6116 &changelowerstate_info.info); 6117 } 6118 EXPORT_SYMBOL(netdev_lower_state_changed); 6119 6120 int netdev_default_l2upper_neigh_construct(struct net_device *dev, 6121 struct neighbour *n) 6122 { 6123 struct net_device *lower_dev, *stop_dev; 6124 struct list_head *iter; 6125 int err; 6126 6127 netdev_for_each_lower_dev(dev, lower_dev, iter) { 6128 if (!lower_dev->netdev_ops->ndo_neigh_construct) 6129 continue; 6130 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n); 6131 if (err) { 6132 stop_dev = lower_dev; 6133 goto rollback; 6134 } 6135 } 6136 return 0; 6137 6138 rollback: 6139 netdev_for_each_lower_dev(dev, lower_dev, iter) { 6140 if (lower_dev == stop_dev) 6141 break; 6142 if (!lower_dev->netdev_ops->ndo_neigh_destroy) 6143 continue; 6144 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n); 6145 } 6146 return err; 6147 } 6148 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct); 6149 6150 void netdev_default_l2upper_neigh_destroy(struct net_device *dev, 6151 struct neighbour *n) 6152 { 6153 struct net_device *lower_dev; 6154 struct list_head *iter; 6155 6156 netdev_for_each_lower_dev(dev, lower_dev, iter) { 6157 if (!lower_dev->netdev_ops->ndo_neigh_destroy) 6158 continue; 6159 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n); 6160 } 6161 } 6162 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy); 6163 6164 static void dev_change_rx_flags(struct net_device *dev, int flags) 6165 { 6166 const struct net_device_ops *ops = dev->netdev_ops; 6167 6168 if (ops->ndo_change_rx_flags) 6169 ops->ndo_change_rx_flags(dev, flags); 6170 } 6171 6172 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 6173 { 6174 unsigned int old_flags = dev->flags; 6175 kuid_t uid; 6176 kgid_t gid; 6177 6178 ASSERT_RTNL(); 6179 6180 dev->flags |= IFF_PROMISC; 6181 dev->promiscuity += inc; 6182 if (dev->promiscuity == 0) { 6183 /* 6184 * Avoid overflow. 6185 * If inc causes overflow, untouch promisc and return error. 6186 */ 6187 if (inc < 0) 6188 dev->flags &= ~IFF_PROMISC; 6189 else { 6190 dev->promiscuity -= inc; 6191 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 6192 dev->name); 6193 return -EOVERFLOW; 6194 } 6195 } 6196 if (dev->flags != old_flags) { 6197 pr_info("device %s %s promiscuous mode\n", 6198 dev->name, 6199 dev->flags & IFF_PROMISC ? "entered" : "left"); 6200 if (audit_enabled) { 6201 current_uid_gid(&uid, &gid); 6202 audit_log(current->audit_context, GFP_ATOMIC, 6203 AUDIT_ANOM_PROMISCUOUS, 6204 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 6205 dev->name, (dev->flags & IFF_PROMISC), 6206 (old_flags & IFF_PROMISC), 6207 from_kuid(&init_user_ns, audit_get_loginuid(current)), 6208 from_kuid(&init_user_ns, uid), 6209 from_kgid(&init_user_ns, gid), 6210 audit_get_sessionid(current)); 6211 } 6212 6213 dev_change_rx_flags(dev, IFF_PROMISC); 6214 } 6215 if (notify) 6216 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 6217 return 0; 6218 } 6219 6220 /** 6221 * dev_set_promiscuity - update promiscuity count on a device 6222 * @dev: device 6223 * @inc: modifier 6224 * 6225 * Add or remove promiscuity from a device. While the count in the device 6226 * remains above zero the interface remains promiscuous. Once it hits zero 6227 * the device reverts back to normal filtering operation. A negative inc 6228 * value is used to drop promiscuity on the device. 6229 * Return 0 if successful or a negative errno code on error. 6230 */ 6231 int dev_set_promiscuity(struct net_device *dev, int inc) 6232 { 6233 unsigned int old_flags = dev->flags; 6234 int err; 6235 6236 err = __dev_set_promiscuity(dev, inc, true); 6237 if (err < 0) 6238 return err; 6239 if (dev->flags != old_flags) 6240 dev_set_rx_mode(dev); 6241 return err; 6242 } 6243 EXPORT_SYMBOL(dev_set_promiscuity); 6244 6245 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 6246 { 6247 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 6248 6249 ASSERT_RTNL(); 6250 6251 dev->flags |= IFF_ALLMULTI; 6252 dev->allmulti += inc; 6253 if (dev->allmulti == 0) { 6254 /* 6255 * Avoid overflow. 6256 * If inc causes overflow, untouch allmulti and return error. 6257 */ 6258 if (inc < 0) 6259 dev->flags &= ~IFF_ALLMULTI; 6260 else { 6261 dev->allmulti -= inc; 6262 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 6263 dev->name); 6264 return -EOVERFLOW; 6265 } 6266 } 6267 if (dev->flags ^ old_flags) { 6268 dev_change_rx_flags(dev, IFF_ALLMULTI); 6269 dev_set_rx_mode(dev); 6270 if (notify) 6271 __dev_notify_flags(dev, old_flags, 6272 dev->gflags ^ old_gflags); 6273 } 6274 return 0; 6275 } 6276 6277 /** 6278 * dev_set_allmulti - update allmulti count on a device 6279 * @dev: device 6280 * @inc: modifier 6281 * 6282 * Add or remove reception of all multicast frames to a device. While the 6283 * count in the device remains above zero the interface remains listening 6284 * to all interfaces. Once it hits zero the device reverts back to normal 6285 * filtering operation. A negative @inc value is used to drop the counter 6286 * when releasing a resource needing all multicasts. 6287 * Return 0 if successful or a negative errno code on error. 6288 */ 6289 6290 int dev_set_allmulti(struct net_device *dev, int inc) 6291 { 6292 return __dev_set_allmulti(dev, inc, true); 6293 } 6294 EXPORT_SYMBOL(dev_set_allmulti); 6295 6296 /* 6297 * Upload unicast and multicast address lists to device and 6298 * configure RX filtering. When the device doesn't support unicast 6299 * filtering it is put in promiscuous mode while unicast addresses 6300 * are present. 6301 */ 6302 void __dev_set_rx_mode(struct net_device *dev) 6303 { 6304 const struct net_device_ops *ops = dev->netdev_ops; 6305 6306 /* dev_open will call this function so the list will stay sane. */ 6307 if (!(dev->flags&IFF_UP)) 6308 return; 6309 6310 if (!netif_device_present(dev)) 6311 return; 6312 6313 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 6314 /* Unicast addresses changes may only happen under the rtnl, 6315 * therefore calling __dev_set_promiscuity here is safe. 6316 */ 6317 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 6318 __dev_set_promiscuity(dev, 1, false); 6319 dev->uc_promisc = true; 6320 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 6321 __dev_set_promiscuity(dev, -1, false); 6322 dev->uc_promisc = false; 6323 } 6324 } 6325 6326 if (ops->ndo_set_rx_mode) 6327 ops->ndo_set_rx_mode(dev); 6328 } 6329 6330 void dev_set_rx_mode(struct net_device *dev) 6331 { 6332 netif_addr_lock_bh(dev); 6333 __dev_set_rx_mode(dev); 6334 netif_addr_unlock_bh(dev); 6335 } 6336 6337 /** 6338 * dev_get_flags - get flags reported to userspace 6339 * @dev: device 6340 * 6341 * Get the combination of flag bits exported through APIs to userspace. 6342 */ 6343 unsigned int dev_get_flags(const struct net_device *dev) 6344 { 6345 unsigned int flags; 6346 6347 flags = (dev->flags & ~(IFF_PROMISC | 6348 IFF_ALLMULTI | 6349 IFF_RUNNING | 6350 IFF_LOWER_UP | 6351 IFF_DORMANT)) | 6352 (dev->gflags & (IFF_PROMISC | 6353 IFF_ALLMULTI)); 6354 6355 if (netif_running(dev)) { 6356 if (netif_oper_up(dev)) 6357 flags |= IFF_RUNNING; 6358 if (netif_carrier_ok(dev)) 6359 flags |= IFF_LOWER_UP; 6360 if (netif_dormant(dev)) 6361 flags |= IFF_DORMANT; 6362 } 6363 6364 return flags; 6365 } 6366 EXPORT_SYMBOL(dev_get_flags); 6367 6368 int __dev_change_flags(struct net_device *dev, unsigned int flags) 6369 { 6370 unsigned int old_flags = dev->flags; 6371 int ret; 6372 6373 ASSERT_RTNL(); 6374 6375 /* 6376 * Set the flags on our device. 6377 */ 6378 6379 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 6380 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 6381 IFF_AUTOMEDIA)) | 6382 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 6383 IFF_ALLMULTI)); 6384 6385 /* 6386 * Load in the correct multicast list now the flags have changed. 6387 */ 6388 6389 if ((old_flags ^ flags) & IFF_MULTICAST) 6390 dev_change_rx_flags(dev, IFF_MULTICAST); 6391 6392 dev_set_rx_mode(dev); 6393 6394 /* 6395 * Have we downed the interface. We handle IFF_UP ourselves 6396 * according to user attempts to set it, rather than blindly 6397 * setting it. 6398 */ 6399 6400 ret = 0; 6401 if ((old_flags ^ flags) & IFF_UP) 6402 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); 6403 6404 if ((flags ^ dev->gflags) & IFF_PROMISC) { 6405 int inc = (flags & IFF_PROMISC) ? 1 : -1; 6406 unsigned int old_flags = dev->flags; 6407 6408 dev->gflags ^= IFF_PROMISC; 6409 6410 if (__dev_set_promiscuity(dev, inc, false) >= 0) 6411 if (dev->flags != old_flags) 6412 dev_set_rx_mode(dev); 6413 } 6414 6415 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 6416 is important. Some (broken) drivers set IFF_PROMISC, when 6417 IFF_ALLMULTI is requested not asking us and not reporting. 6418 */ 6419 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 6420 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 6421 6422 dev->gflags ^= IFF_ALLMULTI; 6423 __dev_set_allmulti(dev, inc, false); 6424 } 6425 6426 return ret; 6427 } 6428 6429 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 6430 unsigned int gchanges) 6431 { 6432 unsigned int changes = dev->flags ^ old_flags; 6433 6434 if (gchanges) 6435 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 6436 6437 if (changes & IFF_UP) { 6438 if (dev->flags & IFF_UP) 6439 call_netdevice_notifiers(NETDEV_UP, dev); 6440 else 6441 call_netdevice_notifiers(NETDEV_DOWN, dev); 6442 } 6443 6444 if (dev->flags & IFF_UP && 6445 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 6446 struct netdev_notifier_change_info change_info; 6447 6448 change_info.flags_changed = changes; 6449 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 6450 &change_info.info); 6451 } 6452 } 6453 6454 /** 6455 * dev_change_flags - change device settings 6456 * @dev: device 6457 * @flags: device state flags 6458 * 6459 * Change settings on device based state flags. The flags are 6460 * in the userspace exported format. 6461 */ 6462 int dev_change_flags(struct net_device *dev, unsigned int flags) 6463 { 6464 int ret; 6465 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 6466 6467 ret = __dev_change_flags(dev, flags); 6468 if (ret < 0) 6469 return ret; 6470 6471 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 6472 __dev_notify_flags(dev, old_flags, changes); 6473 return ret; 6474 } 6475 EXPORT_SYMBOL(dev_change_flags); 6476 6477 static int __dev_set_mtu(struct net_device *dev, int new_mtu) 6478 { 6479 const struct net_device_ops *ops = dev->netdev_ops; 6480 6481 if (ops->ndo_change_mtu) 6482 return ops->ndo_change_mtu(dev, new_mtu); 6483 6484 dev->mtu = new_mtu; 6485 return 0; 6486 } 6487 6488 /** 6489 * dev_set_mtu - Change maximum transfer unit 6490 * @dev: device 6491 * @new_mtu: new transfer unit 6492 * 6493 * Change the maximum transfer size of the network device. 6494 */ 6495 int dev_set_mtu(struct net_device *dev, int new_mtu) 6496 { 6497 int err, orig_mtu; 6498 6499 if (new_mtu == dev->mtu) 6500 return 0; 6501 6502 /* MTU must be positive. */ 6503 if (new_mtu < 0) 6504 return -EINVAL; 6505 6506 if (!netif_device_present(dev)) 6507 return -ENODEV; 6508 6509 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 6510 err = notifier_to_errno(err); 6511 if (err) 6512 return err; 6513 6514 orig_mtu = dev->mtu; 6515 err = __dev_set_mtu(dev, new_mtu); 6516 6517 if (!err) { 6518 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 6519 err = notifier_to_errno(err); 6520 if (err) { 6521 /* setting mtu back and notifying everyone again, 6522 * so that they have a chance to revert changes. 6523 */ 6524 __dev_set_mtu(dev, orig_mtu); 6525 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 6526 } 6527 } 6528 return err; 6529 } 6530 EXPORT_SYMBOL(dev_set_mtu); 6531 6532 /** 6533 * dev_set_group - Change group this device belongs to 6534 * @dev: device 6535 * @new_group: group this device should belong to 6536 */ 6537 void dev_set_group(struct net_device *dev, int new_group) 6538 { 6539 dev->group = new_group; 6540 } 6541 EXPORT_SYMBOL(dev_set_group); 6542 6543 /** 6544 * dev_set_mac_address - Change Media Access Control Address 6545 * @dev: device 6546 * @sa: new address 6547 * 6548 * Change the hardware (MAC) address of the device 6549 */ 6550 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 6551 { 6552 const struct net_device_ops *ops = dev->netdev_ops; 6553 int err; 6554 6555 if (!ops->ndo_set_mac_address) 6556 return -EOPNOTSUPP; 6557 if (sa->sa_family != dev->type) 6558 return -EINVAL; 6559 if (!netif_device_present(dev)) 6560 return -ENODEV; 6561 err = ops->ndo_set_mac_address(dev, sa); 6562 if (err) 6563 return err; 6564 dev->addr_assign_type = NET_ADDR_SET; 6565 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 6566 add_device_randomness(dev->dev_addr, dev->addr_len); 6567 return 0; 6568 } 6569 EXPORT_SYMBOL(dev_set_mac_address); 6570 6571 /** 6572 * dev_change_carrier - Change device carrier 6573 * @dev: device 6574 * @new_carrier: new value 6575 * 6576 * Change device carrier 6577 */ 6578 int dev_change_carrier(struct net_device *dev, bool new_carrier) 6579 { 6580 const struct net_device_ops *ops = dev->netdev_ops; 6581 6582 if (!ops->ndo_change_carrier) 6583 return -EOPNOTSUPP; 6584 if (!netif_device_present(dev)) 6585 return -ENODEV; 6586 return ops->ndo_change_carrier(dev, new_carrier); 6587 } 6588 EXPORT_SYMBOL(dev_change_carrier); 6589 6590 /** 6591 * dev_get_phys_port_id - Get device physical port ID 6592 * @dev: device 6593 * @ppid: port ID 6594 * 6595 * Get device physical port ID 6596 */ 6597 int dev_get_phys_port_id(struct net_device *dev, 6598 struct netdev_phys_item_id *ppid) 6599 { 6600 const struct net_device_ops *ops = dev->netdev_ops; 6601 6602 if (!ops->ndo_get_phys_port_id) 6603 return -EOPNOTSUPP; 6604 return ops->ndo_get_phys_port_id(dev, ppid); 6605 } 6606 EXPORT_SYMBOL(dev_get_phys_port_id); 6607 6608 /** 6609 * dev_get_phys_port_name - Get device physical port name 6610 * @dev: device 6611 * @name: port name 6612 * @len: limit of bytes to copy to name 6613 * 6614 * Get device physical port name 6615 */ 6616 int dev_get_phys_port_name(struct net_device *dev, 6617 char *name, size_t len) 6618 { 6619 const struct net_device_ops *ops = dev->netdev_ops; 6620 6621 if (!ops->ndo_get_phys_port_name) 6622 return -EOPNOTSUPP; 6623 return ops->ndo_get_phys_port_name(dev, name, len); 6624 } 6625 EXPORT_SYMBOL(dev_get_phys_port_name); 6626 6627 /** 6628 * dev_change_proto_down - update protocol port state information 6629 * @dev: device 6630 * @proto_down: new value 6631 * 6632 * This info can be used by switch drivers to set the phys state of the 6633 * port. 6634 */ 6635 int dev_change_proto_down(struct net_device *dev, bool proto_down) 6636 { 6637 const struct net_device_ops *ops = dev->netdev_ops; 6638 6639 if (!ops->ndo_change_proto_down) 6640 return -EOPNOTSUPP; 6641 if (!netif_device_present(dev)) 6642 return -ENODEV; 6643 return ops->ndo_change_proto_down(dev, proto_down); 6644 } 6645 EXPORT_SYMBOL(dev_change_proto_down); 6646 6647 /** 6648 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 6649 * @dev: device 6650 * @fd: new program fd or negative value to clear 6651 * 6652 * Set or clear a bpf program for a device 6653 */ 6654 int dev_change_xdp_fd(struct net_device *dev, int fd) 6655 { 6656 const struct net_device_ops *ops = dev->netdev_ops; 6657 struct bpf_prog *prog = NULL; 6658 struct netdev_xdp xdp = {}; 6659 int err; 6660 6661 if (!ops->ndo_xdp) 6662 return -EOPNOTSUPP; 6663 if (fd >= 0) { 6664 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); 6665 if (IS_ERR(prog)) 6666 return PTR_ERR(prog); 6667 } 6668 6669 xdp.command = XDP_SETUP_PROG; 6670 xdp.prog = prog; 6671 err = ops->ndo_xdp(dev, &xdp); 6672 if (err < 0 && prog) 6673 bpf_prog_put(prog); 6674 6675 return err; 6676 } 6677 EXPORT_SYMBOL(dev_change_xdp_fd); 6678 6679 /** 6680 * dev_new_index - allocate an ifindex 6681 * @net: the applicable net namespace 6682 * 6683 * Returns a suitable unique value for a new device interface 6684 * number. The caller must hold the rtnl semaphore or the 6685 * dev_base_lock to be sure it remains unique. 6686 */ 6687 static int dev_new_index(struct net *net) 6688 { 6689 int ifindex = net->ifindex; 6690 for (;;) { 6691 if (++ifindex <= 0) 6692 ifindex = 1; 6693 if (!__dev_get_by_index(net, ifindex)) 6694 return net->ifindex = ifindex; 6695 } 6696 } 6697 6698 /* Delayed registration/unregisteration */ 6699 static LIST_HEAD(net_todo_list); 6700 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 6701 6702 static void net_set_todo(struct net_device *dev) 6703 { 6704 list_add_tail(&dev->todo_list, &net_todo_list); 6705 dev_net(dev)->dev_unreg_count++; 6706 } 6707 6708 static void rollback_registered_many(struct list_head *head) 6709 { 6710 struct net_device *dev, *tmp; 6711 LIST_HEAD(close_head); 6712 6713 BUG_ON(dev_boot_phase); 6714 ASSERT_RTNL(); 6715 6716 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 6717 /* Some devices call without registering 6718 * for initialization unwind. Remove those 6719 * devices and proceed with the remaining. 6720 */ 6721 if (dev->reg_state == NETREG_UNINITIALIZED) { 6722 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 6723 dev->name, dev); 6724 6725 WARN_ON(1); 6726 list_del(&dev->unreg_list); 6727 continue; 6728 } 6729 dev->dismantle = true; 6730 BUG_ON(dev->reg_state != NETREG_REGISTERED); 6731 } 6732 6733 /* If device is running, close it first. */ 6734 list_for_each_entry(dev, head, unreg_list) 6735 list_add_tail(&dev->close_list, &close_head); 6736 dev_close_many(&close_head, true); 6737 6738 list_for_each_entry(dev, head, unreg_list) { 6739 /* And unlink it from device chain. */ 6740 unlist_netdevice(dev); 6741 6742 dev->reg_state = NETREG_UNREGISTERING; 6743 } 6744 flush_all_backlogs(); 6745 6746 synchronize_net(); 6747 6748 list_for_each_entry(dev, head, unreg_list) { 6749 struct sk_buff *skb = NULL; 6750 6751 /* Shutdown queueing discipline. */ 6752 dev_shutdown(dev); 6753 6754 6755 /* Notify protocols, that we are about to destroy 6756 this device. They should clean all the things. 6757 */ 6758 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 6759 6760 if (!dev->rtnl_link_ops || 6761 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 6762 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 6763 GFP_KERNEL); 6764 6765 /* 6766 * Flush the unicast and multicast chains 6767 */ 6768 dev_uc_flush(dev); 6769 dev_mc_flush(dev); 6770 6771 if (dev->netdev_ops->ndo_uninit) 6772 dev->netdev_ops->ndo_uninit(dev); 6773 6774 if (skb) 6775 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 6776 6777 /* Notifier chain MUST detach us all upper devices. */ 6778 WARN_ON(netdev_has_any_upper_dev(dev)); 6779 6780 /* Remove entries from kobject tree */ 6781 netdev_unregister_kobject(dev); 6782 #ifdef CONFIG_XPS 6783 /* Remove XPS queueing entries */ 6784 netif_reset_xps_queues_gt(dev, 0); 6785 #endif 6786 } 6787 6788 synchronize_net(); 6789 6790 list_for_each_entry(dev, head, unreg_list) 6791 dev_put(dev); 6792 } 6793 6794 static void rollback_registered(struct net_device *dev) 6795 { 6796 LIST_HEAD(single); 6797 6798 list_add(&dev->unreg_list, &single); 6799 rollback_registered_many(&single); 6800 list_del(&single); 6801 } 6802 6803 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 6804 struct net_device *upper, netdev_features_t features) 6805 { 6806 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 6807 netdev_features_t feature; 6808 int feature_bit; 6809 6810 for_each_netdev_feature(&upper_disables, feature_bit) { 6811 feature = __NETIF_F_BIT(feature_bit); 6812 if (!(upper->wanted_features & feature) 6813 && (features & feature)) { 6814 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 6815 &feature, upper->name); 6816 features &= ~feature; 6817 } 6818 } 6819 6820 return features; 6821 } 6822 6823 static void netdev_sync_lower_features(struct net_device *upper, 6824 struct net_device *lower, netdev_features_t features) 6825 { 6826 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 6827 netdev_features_t feature; 6828 int feature_bit; 6829 6830 for_each_netdev_feature(&upper_disables, feature_bit) { 6831 feature = __NETIF_F_BIT(feature_bit); 6832 if (!(features & feature) && (lower->features & feature)) { 6833 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 6834 &feature, lower->name); 6835 lower->wanted_features &= ~feature; 6836 netdev_update_features(lower); 6837 6838 if (unlikely(lower->features & feature)) 6839 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 6840 &feature, lower->name); 6841 } 6842 } 6843 } 6844 6845 static netdev_features_t netdev_fix_features(struct net_device *dev, 6846 netdev_features_t features) 6847 { 6848 /* Fix illegal checksum combinations */ 6849 if ((features & NETIF_F_HW_CSUM) && 6850 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 6851 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 6852 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 6853 } 6854 6855 /* TSO requires that SG is present as well. */ 6856 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 6857 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 6858 features &= ~NETIF_F_ALL_TSO; 6859 } 6860 6861 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 6862 !(features & NETIF_F_IP_CSUM)) { 6863 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 6864 features &= ~NETIF_F_TSO; 6865 features &= ~NETIF_F_TSO_ECN; 6866 } 6867 6868 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 6869 !(features & NETIF_F_IPV6_CSUM)) { 6870 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 6871 features &= ~NETIF_F_TSO6; 6872 } 6873 6874 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 6875 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 6876 features &= ~NETIF_F_TSO_MANGLEID; 6877 6878 /* TSO ECN requires that TSO is present as well. */ 6879 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 6880 features &= ~NETIF_F_TSO_ECN; 6881 6882 /* Software GSO depends on SG. */ 6883 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 6884 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 6885 features &= ~NETIF_F_GSO; 6886 } 6887 6888 /* UFO needs SG and checksumming */ 6889 if (features & NETIF_F_UFO) { 6890 /* maybe split UFO into V4 and V6? */ 6891 if (!(features & NETIF_F_HW_CSUM) && 6892 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != 6893 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { 6894 netdev_dbg(dev, 6895 "Dropping NETIF_F_UFO since no checksum offload features.\n"); 6896 features &= ~NETIF_F_UFO; 6897 } 6898 6899 if (!(features & NETIF_F_SG)) { 6900 netdev_dbg(dev, 6901 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); 6902 features &= ~NETIF_F_UFO; 6903 } 6904 } 6905 6906 /* GSO partial features require GSO partial be set */ 6907 if ((features & dev->gso_partial_features) && 6908 !(features & NETIF_F_GSO_PARTIAL)) { 6909 netdev_dbg(dev, 6910 "Dropping partially supported GSO features since no GSO partial.\n"); 6911 features &= ~dev->gso_partial_features; 6912 } 6913 6914 #ifdef CONFIG_NET_RX_BUSY_POLL 6915 if (dev->netdev_ops->ndo_busy_poll) 6916 features |= NETIF_F_BUSY_POLL; 6917 else 6918 #endif 6919 features &= ~NETIF_F_BUSY_POLL; 6920 6921 return features; 6922 } 6923 6924 int __netdev_update_features(struct net_device *dev) 6925 { 6926 struct net_device *upper, *lower; 6927 netdev_features_t features; 6928 struct list_head *iter; 6929 int err = -1; 6930 6931 ASSERT_RTNL(); 6932 6933 features = netdev_get_wanted_features(dev); 6934 6935 if (dev->netdev_ops->ndo_fix_features) 6936 features = dev->netdev_ops->ndo_fix_features(dev, features); 6937 6938 /* driver might be less strict about feature dependencies */ 6939 features = netdev_fix_features(dev, features); 6940 6941 /* some features can't be enabled if they're off an an upper device */ 6942 netdev_for_each_upper_dev_rcu(dev, upper, iter) 6943 features = netdev_sync_upper_features(dev, upper, features); 6944 6945 if (dev->features == features) 6946 goto sync_lower; 6947 6948 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 6949 &dev->features, &features); 6950 6951 if (dev->netdev_ops->ndo_set_features) 6952 err = dev->netdev_ops->ndo_set_features(dev, features); 6953 else 6954 err = 0; 6955 6956 if (unlikely(err < 0)) { 6957 netdev_err(dev, 6958 "set_features() failed (%d); wanted %pNF, left %pNF\n", 6959 err, &features, &dev->features); 6960 /* return non-0 since some features might have changed and 6961 * it's better to fire a spurious notification than miss it 6962 */ 6963 return -1; 6964 } 6965 6966 sync_lower: 6967 /* some features must be disabled on lower devices when disabled 6968 * on an upper device (think: bonding master or bridge) 6969 */ 6970 netdev_for_each_lower_dev(dev, lower, iter) 6971 netdev_sync_lower_features(dev, lower, features); 6972 6973 if (!err) 6974 dev->features = features; 6975 6976 return err < 0 ? 0 : 1; 6977 } 6978 6979 /** 6980 * netdev_update_features - recalculate device features 6981 * @dev: the device to check 6982 * 6983 * Recalculate dev->features set and send notifications if it 6984 * has changed. Should be called after driver or hardware dependent 6985 * conditions might have changed that influence the features. 6986 */ 6987 void netdev_update_features(struct net_device *dev) 6988 { 6989 if (__netdev_update_features(dev)) 6990 netdev_features_change(dev); 6991 } 6992 EXPORT_SYMBOL(netdev_update_features); 6993 6994 /** 6995 * netdev_change_features - recalculate device features 6996 * @dev: the device to check 6997 * 6998 * Recalculate dev->features set and send notifications even 6999 * if they have not changed. Should be called instead of 7000 * netdev_update_features() if also dev->vlan_features might 7001 * have changed to allow the changes to be propagated to stacked 7002 * VLAN devices. 7003 */ 7004 void netdev_change_features(struct net_device *dev) 7005 { 7006 __netdev_update_features(dev); 7007 netdev_features_change(dev); 7008 } 7009 EXPORT_SYMBOL(netdev_change_features); 7010 7011 /** 7012 * netif_stacked_transfer_operstate - transfer operstate 7013 * @rootdev: the root or lower level device to transfer state from 7014 * @dev: the device to transfer operstate to 7015 * 7016 * Transfer operational state from root to device. This is normally 7017 * called when a stacking relationship exists between the root 7018 * device and the device(a leaf device). 7019 */ 7020 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 7021 struct net_device *dev) 7022 { 7023 if (rootdev->operstate == IF_OPER_DORMANT) 7024 netif_dormant_on(dev); 7025 else 7026 netif_dormant_off(dev); 7027 7028 if (netif_carrier_ok(rootdev)) { 7029 if (!netif_carrier_ok(dev)) 7030 netif_carrier_on(dev); 7031 } else { 7032 if (netif_carrier_ok(dev)) 7033 netif_carrier_off(dev); 7034 } 7035 } 7036 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 7037 7038 #ifdef CONFIG_SYSFS 7039 static int netif_alloc_rx_queues(struct net_device *dev) 7040 { 7041 unsigned int i, count = dev->num_rx_queues; 7042 struct netdev_rx_queue *rx; 7043 size_t sz = count * sizeof(*rx); 7044 7045 BUG_ON(count < 1); 7046 7047 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 7048 if (!rx) { 7049 rx = vzalloc(sz); 7050 if (!rx) 7051 return -ENOMEM; 7052 } 7053 dev->_rx = rx; 7054 7055 for (i = 0; i < count; i++) 7056 rx[i].dev = dev; 7057 return 0; 7058 } 7059 #endif 7060 7061 static void netdev_init_one_queue(struct net_device *dev, 7062 struct netdev_queue *queue, void *_unused) 7063 { 7064 /* Initialize queue lock */ 7065 spin_lock_init(&queue->_xmit_lock); 7066 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 7067 queue->xmit_lock_owner = -1; 7068 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 7069 queue->dev = dev; 7070 #ifdef CONFIG_BQL 7071 dql_init(&queue->dql, HZ); 7072 #endif 7073 } 7074 7075 static void netif_free_tx_queues(struct net_device *dev) 7076 { 7077 kvfree(dev->_tx); 7078 } 7079 7080 static int netif_alloc_netdev_queues(struct net_device *dev) 7081 { 7082 unsigned int count = dev->num_tx_queues; 7083 struct netdev_queue *tx; 7084 size_t sz = count * sizeof(*tx); 7085 7086 if (count < 1 || count > 0xffff) 7087 return -EINVAL; 7088 7089 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 7090 if (!tx) { 7091 tx = vzalloc(sz); 7092 if (!tx) 7093 return -ENOMEM; 7094 } 7095 dev->_tx = tx; 7096 7097 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 7098 spin_lock_init(&dev->tx_global_lock); 7099 7100 return 0; 7101 } 7102 7103 void netif_tx_stop_all_queues(struct net_device *dev) 7104 { 7105 unsigned int i; 7106 7107 for (i = 0; i < dev->num_tx_queues; i++) { 7108 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 7109 netif_tx_stop_queue(txq); 7110 } 7111 } 7112 EXPORT_SYMBOL(netif_tx_stop_all_queues); 7113 7114 /** 7115 * register_netdevice - register a network device 7116 * @dev: device to register 7117 * 7118 * Take a completed network device structure and add it to the kernel 7119 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 7120 * chain. 0 is returned on success. A negative errno code is returned 7121 * on a failure to set up the device, or if the name is a duplicate. 7122 * 7123 * Callers must hold the rtnl semaphore. You may want 7124 * register_netdev() instead of this. 7125 * 7126 * BUGS: 7127 * The locking appears insufficient to guarantee two parallel registers 7128 * will not get the same name. 7129 */ 7130 7131 int register_netdevice(struct net_device *dev) 7132 { 7133 int ret; 7134 struct net *net = dev_net(dev); 7135 7136 BUG_ON(dev_boot_phase); 7137 ASSERT_RTNL(); 7138 7139 might_sleep(); 7140 7141 /* When net_device's are persistent, this will be fatal. */ 7142 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 7143 BUG_ON(!net); 7144 7145 spin_lock_init(&dev->addr_list_lock); 7146 netdev_set_addr_lockdep_class(dev); 7147 7148 ret = dev_get_valid_name(net, dev, dev->name); 7149 if (ret < 0) 7150 goto out; 7151 7152 /* Init, if this function is available */ 7153 if (dev->netdev_ops->ndo_init) { 7154 ret = dev->netdev_ops->ndo_init(dev); 7155 if (ret) { 7156 if (ret > 0) 7157 ret = -EIO; 7158 goto out; 7159 } 7160 } 7161 7162 if (((dev->hw_features | dev->features) & 7163 NETIF_F_HW_VLAN_CTAG_FILTER) && 7164 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 7165 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 7166 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 7167 ret = -EINVAL; 7168 goto err_uninit; 7169 } 7170 7171 ret = -EBUSY; 7172 if (!dev->ifindex) 7173 dev->ifindex = dev_new_index(net); 7174 else if (__dev_get_by_index(net, dev->ifindex)) 7175 goto err_uninit; 7176 7177 /* Transfer changeable features to wanted_features and enable 7178 * software offloads (GSO and GRO). 7179 */ 7180 dev->hw_features |= NETIF_F_SOFT_FEATURES; 7181 dev->features |= NETIF_F_SOFT_FEATURES; 7182 dev->wanted_features = dev->features & dev->hw_features; 7183 7184 if (!(dev->flags & IFF_LOOPBACK)) 7185 dev->hw_features |= NETIF_F_NOCACHE_COPY; 7186 7187 /* If IPv4 TCP segmentation offload is supported we should also 7188 * allow the device to enable segmenting the frame with the option 7189 * of ignoring a static IP ID value. This doesn't enable the 7190 * feature itself but allows the user to enable it later. 7191 */ 7192 if (dev->hw_features & NETIF_F_TSO) 7193 dev->hw_features |= NETIF_F_TSO_MANGLEID; 7194 if (dev->vlan_features & NETIF_F_TSO) 7195 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 7196 if (dev->mpls_features & NETIF_F_TSO) 7197 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 7198 if (dev->hw_enc_features & NETIF_F_TSO) 7199 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 7200 7201 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 7202 */ 7203 dev->vlan_features |= NETIF_F_HIGHDMA; 7204 7205 /* Make NETIF_F_SG inheritable to tunnel devices. 7206 */ 7207 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 7208 7209 /* Make NETIF_F_SG inheritable to MPLS. 7210 */ 7211 dev->mpls_features |= NETIF_F_SG; 7212 7213 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 7214 ret = notifier_to_errno(ret); 7215 if (ret) 7216 goto err_uninit; 7217 7218 ret = netdev_register_kobject(dev); 7219 if (ret) 7220 goto err_uninit; 7221 dev->reg_state = NETREG_REGISTERED; 7222 7223 __netdev_update_features(dev); 7224 7225 /* 7226 * Default initial state at registry is that the 7227 * device is present. 7228 */ 7229 7230 set_bit(__LINK_STATE_PRESENT, &dev->state); 7231 7232 linkwatch_init_dev(dev); 7233 7234 dev_init_scheduler(dev); 7235 dev_hold(dev); 7236 list_netdevice(dev); 7237 add_device_randomness(dev->dev_addr, dev->addr_len); 7238 7239 /* If the device has permanent device address, driver should 7240 * set dev_addr and also addr_assign_type should be set to 7241 * NET_ADDR_PERM (default value). 7242 */ 7243 if (dev->addr_assign_type == NET_ADDR_PERM) 7244 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 7245 7246 /* Notify protocols, that a new device appeared. */ 7247 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 7248 ret = notifier_to_errno(ret); 7249 if (ret) { 7250 rollback_registered(dev); 7251 dev->reg_state = NETREG_UNREGISTERED; 7252 } 7253 /* 7254 * Prevent userspace races by waiting until the network 7255 * device is fully setup before sending notifications. 7256 */ 7257 if (!dev->rtnl_link_ops || 7258 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 7259 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 7260 7261 out: 7262 return ret; 7263 7264 err_uninit: 7265 if (dev->netdev_ops->ndo_uninit) 7266 dev->netdev_ops->ndo_uninit(dev); 7267 goto out; 7268 } 7269 EXPORT_SYMBOL(register_netdevice); 7270 7271 /** 7272 * init_dummy_netdev - init a dummy network device for NAPI 7273 * @dev: device to init 7274 * 7275 * This takes a network device structure and initialize the minimum 7276 * amount of fields so it can be used to schedule NAPI polls without 7277 * registering a full blown interface. This is to be used by drivers 7278 * that need to tie several hardware interfaces to a single NAPI 7279 * poll scheduler due to HW limitations. 7280 */ 7281 int init_dummy_netdev(struct net_device *dev) 7282 { 7283 /* Clear everything. Note we don't initialize spinlocks 7284 * are they aren't supposed to be taken by any of the 7285 * NAPI code and this dummy netdev is supposed to be 7286 * only ever used for NAPI polls 7287 */ 7288 memset(dev, 0, sizeof(struct net_device)); 7289 7290 /* make sure we BUG if trying to hit standard 7291 * register/unregister code path 7292 */ 7293 dev->reg_state = NETREG_DUMMY; 7294 7295 /* NAPI wants this */ 7296 INIT_LIST_HEAD(&dev->napi_list); 7297 7298 /* a dummy interface is started by default */ 7299 set_bit(__LINK_STATE_PRESENT, &dev->state); 7300 set_bit(__LINK_STATE_START, &dev->state); 7301 7302 /* Note : We dont allocate pcpu_refcnt for dummy devices, 7303 * because users of this 'device' dont need to change 7304 * its refcount. 7305 */ 7306 7307 return 0; 7308 } 7309 EXPORT_SYMBOL_GPL(init_dummy_netdev); 7310 7311 7312 /** 7313 * register_netdev - register a network device 7314 * @dev: device to register 7315 * 7316 * Take a completed network device structure and add it to the kernel 7317 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 7318 * chain. 0 is returned on success. A negative errno code is returned 7319 * on a failure to set up the device, or if the name is a duplicate. 7320 * 7321 * This is a wrapper around register_netdevice that takes the rtnl semaphore 7322 * and expands the device name if you passed a format string to 7323 * alloc_netdev. 7324 */ 7325 int register_netdev(struct net_device *dev) 7326 { 7327 int err; 7328 7329 rtnl_lock(); 7330 err = register_netdevice(dev); 7331 rtnl_unlock(); 7332 return err; 7333 } 7334 EXPORT_SYMBOL(register_netdev); 7335 7336 int netdev_refcnt_read(const struct net_device *dev) 7337 { 7338 int i, refcnt = 0; 7339 7340 for_each_possible_cpu(i) 7341 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 7342 return refcnt; 7343 } 7344 EXPORT_SYMBOL(netdev_refcnt_read); 7345 7346 /** 7347 * netdev_wait_allrefs - wait until all references are gone. 7348 * @dev: target net_device 7349 * 7350 * This is called when unregistering network devices. 7351 * 7352 * Any protocol or device that holds a reference should register 7353 * for netdevice notification, and cleanup and put back the 7354 * reference if they receive an UNREGISTER event. 7355 * We can get stuck here if buggy protocols don't correctly 7356 * call dev_put. 7357 */ 7358 static void netdev_wait_allrefs(struct net_device *dev) 7359 { 7360 unsigned long rebroadcast_time, warning_time; 7361 int refcnt; 7362 7363 linkwatch_forget_dev(dev); 7364 7365 rebroadcast_time = warning_time = jiffies; 7366 refcnt = netdev_refcnt_read(dev); 7367 7368 while (refcnt != 0) { 7369 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 7370 rtnl_lock(); 7371 7372 /* Rebroadcast unregister notification */ 7373 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7374 7375 __rtnl_unlock(); 7376 rcu_barrier(); 7377 rtnl_lock(); 7378 7379 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7380 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 7381 &dev->state)) { 7382 /* We must not have linkwatch events 7383 * pending on unregister. If this 7384 * happens, we simply run the queue 7385 * unscheduled, resulting in a noop 7386 * for this device. 7387 */ 7388 linkwatch_run_queue(); 7389 } 7390 7391 __rtnl_unlock(); 7392 7393 rebroadcast_time = jiffies; 7394 } 7395 7396 msleep(250); 7397 7398 refcnt = netdev_refcnt_read(dev); 7399 7400 if (time_after(jiffies, warning_time + 10 * HZ)) { 7401 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 7402 dev->name, refcnt); 7403 warning_time = jiffies; 7404 } 7405 } 7406 } 7407 7408 /* The sequence is: 7409 * 7410 * rtnl_lock(); 7411 * ... 7412 * register_netdevice(x1); 7413 * register_netdevice(x2); 7414 * ... 7415 * unregister_netdevice(y1); 7416 * unregister_netdevice(y2); 7417 * ... 7418 * rtnl_unlock(); 7419 * free_netdev(y1); 7420 * free_netdev(y2); 7421 * 7422 * We are invoked by rtnl_unlock(). 7423 * This allows us to deal with problems: 7424 * 1) We can delete sysfs objects which invoke hotplug 7425 * without deadlocking with linkwatch via keventd. 7426 * 2) Since we run with the RTNL semaphore not held, we can sleep 7427 * safely in order to wait for the netdev refcnt to drop to zero. 7428 * 7429 * We must not return until all unregister events added during 7430 * the interval the lock was held have been completed. 7431 */ 7432 void netdev_run_todo(void) 7433 { 7434 struct list_head list; 7435 7436 /* Snapshot list, allow later requests */ 7437 list_replace_init(&net_todo_list, &list); 7438 7439 __rtnl_unlock(); 7440 7441 7442 /* Wait for rcu callbacks to finish before next phase */ 7443 if (!list_empty(&list)) 7444 rcu_barrier(); 7445 7446 while (!list_empty(&list)) { 7447 struct net_device *dev 7448 = list_first_entry(&list, struct net_device, todo_list); 7449 list_del(&dev->todo_list); 7450 7451 rtnl_lock(); 7452 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7453 __rtnl_unlock(); 7454 7455 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 7456 pr_err("network todo '%s' but state %d\n", 7457 dev->name, dev->reg_state); 7458 dump_stack(); 7459 continue; 7460 } 7461 7462 dev->reg_state = NETREG_UNREGISTERED; 7463 7464 netdev_wait_allrefs(dev); 7465 7466 /* paranoia */ 7467 BUG_ON(netdev_refcnt_read(dev)); 7468 BUG_ON(!list_empty(&dev->ptype_all)); 7469 BUG_ON(!list_empty(&dev->ptype_specific)); 7470 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 7471 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 7472 WARN_ON(dev->dn_ptr); 7473 7474 if (dev->destructor) 7475 dev->destructor(dev); 7476 7477 /* Report a network device has been unregistered */ 7478 rtnl_lock(); 7479 dev_net(dev)->dev_unreg_count--; 7480 __rtnl_unlock(); 7481 wake_up(&netdev_unregistering_wq); 7482 7483 /* Free network device */ 7484 kobject_put(&dev->dev.kobj); 7485 } 7486 } 7487 7488 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 7489 * all the same fields in the same order as net_device_stats, with only 7490 * the type differing, but rtnl_link_stats64 may have additional fields 7491 * at the end for newer counters. 7492 */ 7493 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 7494 const struct net_device_stats *netdev_stats) 7495 { 7496 #if BITS_PER_LONG == 64 7497 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 7498 memcpy(stats64, netdev_stats, sizeof(*stats64)); 7499 /* zero out counters that only exist in rtnl_link_stats64 */ 7500 memset((char *)stats64 + sizeof(*netdev_stats), 0, 7501 sizeof(*stats64) - sizeof(*netdev_stats)); 7502 #else 7503 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 7504 const unsigned long *src = (const unsigned long *)netdev_stats; 7505 u64 *dst = (u64 *)stats64; 7506 7507 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 7508 for (i = 0; i < n; i++) 7509 dst[i] = src[i]; 7510 /* zero out counters that only exist in rtnl_link_stats64 */ 7511 memset((char *)stats64 + n * sizeof(u64), 0, 7512 sizeof(*stats64) - n * sizeof(u64)); 7513 #endif 7514 } 7515 EXPORT_SYMBOL(netdev_stats_to_stats64); 7516 7517 /** 7518 * dev_get_stats - get network device statistics 7519 * @dev: device to get statistics from 7520 * @storage: place to store stats 7521 * 7522 * Get network statistics from device. Return @storage. 7523 * The device driver may provide its own method by setting 7524 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 7525 * otherwise the internal statistics structure is used. 7526 */ 7527 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 7528 struct rtnl_link_stats64 *storage) 7529 { 7530 const struct net_device_ops *ops = dev->netdev_ops; 7531 7532 if (ops->ndo_get_stats64) { 7533 memset(storage, 0, sizeof(*storage)); 7534 ops->ndo_get_stats64(dev, storage); 7535 } else if (ops->ndo_get_stats) { 7536 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 7537 } else { 7538 netdev_stats_to_stats64(storage, &dev->stats); 7539 } 7540 storage->rx_dropped += atomic_long_read(&dev->rx_dropped); 7541 storage->tx_dropped += atomic_long_read(&dev->tx_dropped); 7542 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler); 7543 return storage; 7544 } 7545 EXPORT_SYMBOL(dev_get_stats); 7546 7547 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 7548 { 7549 struct netdev_queue *queue = dev_ingress_queue(dev); 7550 7551 #ifdef CONFIG_NET_CLS_ACT 7552 if (queue) 7553 return queue; 7554 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 7555 if (!queue) 7556 return NULL; 7557 netdev_init_one_queue(dev, queue, NULL); 7558 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 7559 queue->qdisc_sleeping = &noop_qdisc; 7560 rcu_assign_pointer(dev->ingress_queue, queue); 7561 #endif 7562 return queue; 7563 } 7564 7565 static const struct ethtool_ops default_ethtool_ops; 7566 7567 void netdev_set_default_ethtool_ops(struct net_device *dev, 7568 const struct ethtool_ops *ops) 7569 { 7570 if (dev->ethtool_ops == &default_ethtool_ops) 7571 dev->ethtool_ops = ops; 7572 } 7573 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 7574 7575 void netdev_freemem(struct net_device *dev) 7576 { 7577 char *addr = (char *)dev - dev->padded; 7578 7579 kvfree(addr); 7580 } 7581 7582 /** 7583 * alloc_netdev_mqs - allocate network device 7584 * @sizeof_priv: size of private data to allocate space for 7585 * @name: device name format string 7586 * @name_assign_type: origin of device name 7587 * @setup: callback to initialize device 7588 * @txqs: the number of TX subqueues to allocate 7589 * @rxqs: the number of RX subqueues to allocate 7590 * 7591 * Allocates a struct net_device with private data area for driver use 7592 * and performs basic initialization. Also allocates subqueue structs 7593 * for each queue on the device. 7594 */ 7595 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 7596 unsigned char name_assign_type, 7597 void (*setup)(struct net_device *), 7598 unsigned int txqs, unsigned int rxqs) 7599 { 7600 struct net_device *dev; 7601 size_t alloc_size; 7602 struct net_device *p; 7603 7604 BUG_ON(strlen(name) >= sizeof(dev->name)); 7605 7606 if (txqs < 1) { 7607 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 7608 return NULL; 7609 } 7610 7611 #ifdef CONFIG_SYSFS 7612 if (rxqs < 1) { 7613 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 7614 return NULL; 7615 } 7616 #endif 7617 7618 alloc_size = sizeof(struct net_device); 7619 if (sizeof_priv) { 7620 /* ensure 32-byte alignment of private area */ 7621 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 7622 alloc_size += sizeof_priv; 7623 } 7624 /* ensure 32-byte alignment of whole construct */ 7625 alloc_size += NETDEV_ALIGN - 1; 7626 7627 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 7628 if (!p) 7629 p = vzalloc(alloc_size); 7630 if (!p) 7631 return NULL; 7632 7633 dev = PTR_ALIGN(p, NETDEV_ALIGN); 7634 dev->padded = (char *)dev - (char *)p; 7635 7636 dev->pcpu_refcnt = alloc_percpu(int); 7637 if (!dev->pcpu_refcnt) 7638 goto free_dev; 7639 7640 if (dev_addr_init(dev)) 7641 goto free_pcpu; 7642 7643 dev_mc_init(dev); 7644 dev_uc_init(dev); 7645 7646 dev_net_set(dev, &init_net); 7647 7648 dev->gso_max_size = GSO_MAX_SIZE; 7649 dev->gso_max_segs = GSO_MAX_SEGS; 7650 7651 INIT_LIST_HEAD(&dev->napi_list); 7652 INIT_LIST_HEAD(&dev->unreg_list); 7653 INIT_LIST_HEAD(&dev->close_list); 7654 INIT_LIST_HEAD(&dev->link_watch_list); 7655 INIT_LIST_HEAD(&dev->adj_list.upper); 7656 INIT_LIST_HEAD(&dev->adj_list.lower); 7657 INIT_LIST_HEAD(&dev->all_adj_list.upper); 7658 INIT_LIST_HEAD(&dev->all_adj_list.lower); 7659 INIT_LIST_HEAD(&dev->ptype_all); 7660 INIT_LIST_HEAD(&dev->ptype_specific); 7661 #ifdef CONFIG_NET_SCHED 7662 hash_init(dev->qdisc_hash); 7663 #endif 7664 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 7665 setup(dev); 7666 7667 if (!dev->tx_queue_len) { 7668 dev->priv_flags |= IFF_NO_QUEUE; 7669 dev->tx_queue_len = 1; 7670 } 7671 7672 dev->num_tx_queues = txqs; 7673 dev->real_num_tx_queues = txqs; 7674 if (netif_alloc_netdev_queues(dev)) 7675 goto free_all; 7676 7677 #ifdef CONFIG_SYSFS 7678 dev->num_rx_queues = rxqs; 7679 dev->real_num_rx_queues = rxqs; 7680 if (netif_alloc_rx_queues(dev)) 7681 goto free_all; 7682 #endif 7683 7684 strcpy(dev->name, name); 7685 dev->name_assign_type = name_assign_type; 7686 dev->group = INIT_NETDEV_GROUP; 7687 if (!dev->ethtool_ops) 7688 dev->ethtool_ops = &default_ethtool_ops; 7689 7690 nf_hook_ingress_init(dev); 7691 7692 return dev; 7693 7694 free_all: 7695 free_netdev(dev); 7696 return NULL; 7697 7698 free_pcpu: 7699 free_percpu(dev->pcpu_refcnt); 7700 free_dev: 7701 netdev_freemem(dev); 7702 return NULL; 7703 } 7704 EXPORT_SYMBOL(alloc_netdev_mqs); 7705 7706 /** 7707 * free_netdev - free network device 7708 * @dev: device 7709 * 7710 * This function does the last stage of destroying an allocated device 7711 * interface. The reference to the device object is released. 7712 * If this is the last reference then it will be freed. 7713 * Must be called in process context. 7714 */ 7715 void free_netdev(struct net_device *dev) 7716 { 7717 struct napi_struct *p, *n; 7718 7719 might_sleep(); 7720 netif_free_tx_queues(dev); 7721 #ifdef CONFIG_SYSFS 7722 kvfree(dev->_rx); 7723 #endif 7724 7725 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 7726 7727 /* Flush device addresses */ 7728 dev_addr_flush(dev); 7729 7730 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 7731 netif_napi_del(p); 7732 7733 free_percpu(dev->pcpu_refcnt); 7734 dev->pcpu_refcnt = NULL; 7735 7736 /* Compatibility with error handling in drivers */ 7737 if (dev->reg_state == NETREG_UNINITIALIZED) { 7738 netdev_freemem(dev); 7739 return; 7740 } 7741 7742 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 7743 dev->reg_state = NETREG_RELEASED; 7744 7745 /* will free via device release */ 7746 put_device(&dev->dev); 7747 } 7748 EXPORT_SYMBOL(free_netdev); 7749 7750 /** 7751 * synchronize_net - Synchronize with packet receive processing 7752 * 7753 * Wait for packets currently being received to be done. 7754 * Does not block later packets from starting. 7755 */ 7756 void synchronize_net(void) 7757 { 7758 might_sleep(); 7759 if (rtnl_is_locked()) 7760 synchronize_rcu_expedited(); 7761 else 7762 synchronize_rcu(); 7763 } 7764 EXPORT_SYMBOL(synchronize_net); 7765 7766 /** 7767 * unregister_netdevice_queue - remove device from the kernel 7768 * @dev: device 7769 * @head: list 7770 * 7771 * This function shuts down a device interface and removes it 7772 * from the kernel tables. 7773 * If head not NULL, device is queued to be unregistered later. 7774 * 7775 * Callers must hold the rtnl semaphore. You may want 7776 * unregister_netdev() instead of this. 7777 */ 7778 7779 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 7780 { 7781 ASSERT_RTNL(); 7782 7783 if (head) { 7784 list_move_tail(&dev->unreg_list, head); 7785 } else { 7786 rollback_registered(dev); 7787 /* Finish processing unregister after unlock */ 7788 net_set_todo(dev); 7789 } 7790 } 7791 EXPORT_SYMBOL(unregister_netdevice_queue); 7792 7793 /** 7794 * unregister_netdevice_many - unregister many devices 7795 * @head: list of devices 7796 * 7797 * Note: As most callers use a stack allocated list_head, 7798 * we force a list_del() to make sure stack wont be corrupted later. 7799 */ 7800 void unregister_netdevice_many(struct list_head *head) 7801 { 7802 struct net_device *dev; 7803 7804 if (!list_empty(head)) { 7805 rollback_registered_many(head); 7806 list_for_each_entry(dev, head, unreg_list) 7807 net_set_todo(dev); 7808 list_del(head); 7809 } 7810 } 7811 EXPORT_SYMBOL(unregister_netdevice_many); 7812 7813 /** 7814 * unregister_netdev - remove device from the kernel 7815 * @dev: device 7816 * 7817 * This function shuts down a device interface and removes it 7818 * from the kernel tables. 7819 * 7820 * This is just a wrapper for unregister_netdevice that takes 7821 * the rtnl semaphore. In general you want to use this and not 7822 * unregister_netdevice. 7823 */ 7824 void unregister_netdev(struct net_device *dev) 7825 { 7826 rtnl_lock(); 7827 unregister_netdevice(dev); 7828 rtnl_unlock(); 7829 } 7830 EXPORT_SYMBOL(unregister_netdev); 7831 7832 /** 7833 * dev_change_net_namespace - move device to different nethost namespace 7834 * @dev: device 7835 * @net: network namespace 7836 * @pat: If not NULL name pattern to try if the current device name 7837 * is already taken in the destination network namespace. 7838 * 7839 * This function shuts down a device interface and moves it 7840 * to a new network namespace. On success 0 is returned, on 7841 * a failure a netagive errno code is returned. 7842 * 7843 * Callers must hold the rtnl semaphore. 7844 */ 7845 7846 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 7847 { 7848 int err; 7849 7850 ASSERT_RTNL(); 7851 7852 /* Don't allow namespace local devices to be moved. */ 7853 err = -EINVAL; 7854 if (dev->features & NETIF_F_NETNS_LOCAL) 7855 goto out; 7856 7857 /* Ensure the device has been registrered */ 7858 if (dev->reg_state != NETREG_REGISTERED) 7859 goto out; 7860 7861 /* Get out if there is nothing todo */ 7862 err = 0; 7863 if (net_eq(dev_net(dev), net)) 7864 goto out; 7865 7866 /* Pick the destination device name, and ensure 7867 * we can use it in the destination network namespace. 7868 */ 7869 err = -EEXIST; 7870 if (__dev_get_by_name(net, dev->name)) { 7871 /* We get here if we can't use the current device name */ 7872 if (!pat) 7873 goto out; 7874 if (dev_get_valid_name(net, dev, pat) < 0) 7875 goto out; 7876 } 7877 7878 /* 7879 * And now a mini version of register_netdevice unregister_netdevice. 7880 */ 7881 7882 /* If device is running close it first. */ 7883 dev_close(dev); 7884 7885 /* And unlink it from device chain */ 7886 err = -ENODEV; 7887 unlist_netdevice(dev); 7888 7889 synchronize_net(); 7890 7891 /* Shutdown queueing discipline. */ 7892 dev_shutdown(dev); 7893 7894 /* Notify protocols, that we are about to destroy 7895 this device. They should clean all the things. 7896 7897 Note that dev->reg_state stays at NETREG_REGISTERED. 7898 This is wanted because this way 8021q and macvlan know 7899 the device is just moving and can keep their slaves up. 7900 */ 7901 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7902 rcu_barrier(); 7903 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7904 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); 7905 7906 /* 7907 * Flush the unicast and multicast chains 7908 */ 7909 dev_uc_flush(dev); 7910 dev_mc_flush(dev); 7911 7912 /* Send a netdev-removed uevent to the old namespace */ 7913 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 7914 netdev_adjacent_del_links(dev); 7915 7916 /* Actually switch the network namespace */ 7917 dev_net_set(dev, net); 7918 7919 /* If there is an ifindex conflict assign a new one */ 7920 if (__dev_get_by_index(net, dev->ifindex)) 7921 dev->ifindex = dev_new_index(net); 7922 7923 /* Send a netdev-add uevent to the new namespace */ 7924 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 7925 netdev_adjacent_add_links(dev); 7926 7927 /* Fixup kobjects */ 7928 err = device_rename(&dev->dev, dev->name); 7929 WARN_ON(err); 7930 7931 /* Add the device back in the hashes */ 7932 list_netdevice(dev); 7933 7934 /* Notify protocols, that a new device appeared. */ 7935 call_netdevice_notifiers(NETDEV_REGISTER, dev); 7936 7937 /* 7938 * Prevent userspace races by waiting until the network 7939 * device is fully setup before sending notifications. 7940 */ 7941 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 7942 7943 synchronize_net(); 7944 err = 0; 7945 out: 7946 return err; 7947 } 7948 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 7949 7950 static int dev_cpu_callback(struct notifier_block *nfb, 7951 unsigned long action, 7952 void *ocpu) 7953 { 7954 struct sk_buff **list_skb; 7955 struct sk_buff *skb; 7956 unsigned int cpu, oldcpu = (unsigned long)ocpu; 7957 struct softnet_data *sd, *oldsd; 7958 7959 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) 7960 return NOTIFY_OK; 7961 7962 local_irq_disable(); 7963 cpu = smp_processor_id(); 7964 sd = &per_cpu(softnet_data, cpu); 7965 oldsd = &per_cpu(softnet_data, oldcpu); 7966 7967 /* Find end of our completion_queue. */ 7968 list_skb = &sd->completion_queue; 7969 while (*list_skb) 7970 list_skb = &(*list_skb)->next; 7971 /* Append completion queue from offline CPU. */ 7972 *list_skb = oldsd->completion_queue; 7973 oldsd->completion_queue = NULL; 7974 7975 /* Append output queue from offline CPU. */ 7976 if (oldsd->output_queue) { 7977 *sd->output_queue_tailp = oldsd->output_queue; 7978 sd->output_queue_tailp = oldsd->output_queue_tailp; 7979 oldsd->output_queue = NULL; 7980 oldsd->output_queue_tailp = &oldsd->output_queue; 7981 } 7982 /* Append NAPI poll list from offline CPU, with one exception : 7983 * process_backlog() must be called by cpu owning percpu backlog. 7984 * We properly handle process_queue & input_pkt_queue later. 7985 */ 7986 while (!list_empty(&oldsd->poll_list)) { 7987 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 7988 struct napi_struct, 7989 poll_list); 7990 7991 list_del_init(&napi->poll_list); 7992 if (napi->poll == process_backlog) 7993 napi->state = 0; 7994 else 7995 ____napi_schedule(sd, napi); 7996 } 7997 7998 raise_softirq_irqoff(NET_TX_SOFTIRQ); 7999 local_irq_enable(); 8000 8001 /* Process offline CPU's input_pkt_queue */ 8002 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 8003 netif_rx_ni(skb); 8004 input_queue_head_incr(oldsd); 8005 } 8006 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 8007 netif_rx_ni(skb); 8008 input_queue_head_incr(oldsd); 8009 } 8010 8011 return NOTIFY_OK; 8012 } 8013 8014 8015 /** 8016 * netdev_increment_features - increment feature set by one 8017 * @all: current feature set 8018 * @one: new feature set 8019 * @mask: mask feature set 8020 * 8021 * Computes a new feature set after adding a device with feature set 8022 * @one to the master device with current feature set @all. Will not 8023 * enable anything that is off in @mask. Returns the new feature set. 8024 */ 8025 netdev_features_t netdev_increment_features(netdev_features_t all, 8026 netdev_features_t one, netdev_features_t mask) 8027 { 8028 if (mask & NETIF_F_HW_CSUM) 8029 mask |= NETIF_F_CSUM_MASK; 8030 mask |= NETIF_F_VLAN_CHALLENGED; 8031 8032 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 8033 all &= one | ~NETIF_F_ALL_FOR_ALL; 8034 8035 /* If one device supports hw checksumming, set for all. */ 8036 if (all & NETIF_F_HW_CSUM) 8037 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 8038 8039 return all; 8040 } 8041 EXPORT_SYMBOL(netdev_increment_features); 8042 8043 static struct hlist_head * __net_init netdev_create_hash(void) 8044 { 8045 int i; 8046 struct hlist_head *hash; 8047 8048 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); 8049 if (hash != NULL) 8050 for (i = 0; i < NETDEV_HASHENTRIES; i++) 8051 INIT_HLIST_HEAD(&hash[i]); 8052 8053 return hash; 8054 } 8055 8056 /* Initialize per network namespace state */ 8057 static int __net_init netdev_init(struct net *net) 8058 { 8059 if (net != &init_net) 8060 INIT_LIST_HEAD(&net->dev_base_head); 8061 8062 net->dev_name_head = netdev_create_hash(); 8063 if (net->dev_name_head == NULL) 8064 goto err_name; 8065 8066 net->dev_index_head = netdev_create_hash(); 8067 if (net->dev_index_head == NULL) 8068 goto err_idx; 8069 8070 return 0; 8071 8072 err_idx: 8073 kfree(net->dev_name_head); 8074 err_name: 8075 return -ENOMEM; 8076 } 8077 8078 /** 8079 * netdev_drivername - network driver for the device 8080 * @dev: network device 8081 * 8082 * Determine network driver for device. 8083 */ 8084 const char *netdev_drivername(const struct net_device *dev) 8085 { 8086 const struct device_driver *driver; 8087 const struct device *parent; 8088 const char *empty = ""; 8089 8090 parent = dev->dev.parent; 8091 if (!parent) 8092 return empty; 8093 8094 driver = parent->driver; 8095 if (driver && driver->name) 8096 return driver->name; 8097 return empty; 8098 } 8099 8100 static void __netdev_printk(const char *level, const struct net_device *dev, 8101 struct va_format *vaf) 8102 { 8103 if (dev && dev->dev.parent) { 8104 dev_printk_emit(level[1] - '0', 8105 dev->dev.parent, 8106 "%s %s %s%s: %pV", 8107 dev_driver_string(dev->dev.parent), 8108 dev_name(dev->dev.parent), 8109 netdev_name(dev), netdev_reg_state(dev), 8110 vaf); 8111 } else if (dev) { 8112 printk("%s%s%s: %pV", 8113 level, netdev_name(dev), netdev_reg_state(dev), vaf); 8114 } else { 8115 printk("%s(NULL net_device): %pV", level, vaf); 8116 } 8117 } 8118 8119 void netdev_printk(const char *level, const struct net_device *dev, 8120 const char *format, ...) 8121 { 8122 struct va_format vaf; 8123 va_list args; 8124 8125 va_start(args, format); 8126 8127 vaf.fmt = format; 8128 vaf.va = &args; 8129 8130 __netdev_printk(level, dev, &vaf); 8131 8132 va_end(args); 8133 } 8134 EXPORT_SYMBOL(netdev_printk); 8135 8136 #define define_netdev_printk_level(func, level) \ 8137 void func(const struct net_device *dev, const char *fmt, ...) \ 8138 { \ 8139 struct va_format vaf; \ 8140 va_list args; \ 8141 \ 8142 va_start(args, fmt); \ 8143 \ 8144 vaf.fmt = fmt; \ 8145 vaf.va = &args; \ 8146 \ 8147 __netdev_printk(level, dev, &vaf); \ 8148 \ 8149 va_end(args); \ 8150 } \ 8151 EXPORT_SYMBOL(func); 8152 8153 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 8154 define_netdev_printk_level(netdev_alert, KERN_ALERT); 8155 define_netdev_printk_level(netdev_crit, KERN_CRIT); 8156 define_netdev_printk_level(netdev_err, KERN_ERR); 8157 define_netdev_printk_level(netdev_warn, KERN_WARNING); 8158 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 8159 define_netdev_printk_level(netdev_info, KERN_INFO); 8160 8161 static void __net_exit netdev_exit(struct net *net) 8162 { 8163 kfree(net->dev_name_head); 8164 kfree(net->dev_index_head); 8165 } 8166 8167 static struct pernet_operations __net_initdata netdev_net_ops = { 8168 .init = netdev_init, 8169 .exit = netdev_exit, 8170 }; 8171 8172 static void __net_exit default_device_exit(struct net *net) 8173 { 8174 struct net_device *dev, *aux; 8175 /* 8176 * Push all migratable network devices back to the 8177 * initial network namespace 8178 */ 8179 rtnl_lock(); 8180 for_each_netdev_safe(net, dev, aux) { 8181 int err; 8182 char fb_name[IFNAMSIZ]; 8183 8184 /* Ignore unmoveable devices (i.e. loopback) */ 8185 if (dev->features & NETIF_F_NETNS_LOCAL) 8186 continue; 8187 8188 /* Leave virtual devices for the generic cleanup */ 8189 if (dev->rtnl_link_ops) 8190 continue; 8191 8192 /* Push remaining network devices to init_net */ 8193 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 8194 err = dev_change_net_namespace(dev, &init_net, fb_name); 8195 if (err) { 8196 pr_emerg("%s: failed to move %s to init_net: %d\n", 8197 __func__, dev->name, err); 8198 BUG(); 8199 } 8200 } 8201 rtnl_unlock(); 8202 } 8203 8204 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 8205 { 8206 /* Return with the rtnl_lock held when there are no network 8207 * devices unregistering in any network namespace in net_list. 8208 */ 8209 struct net *net; 8210 bool unregistering; 8211 DEFINE_WAIT_FUNC(wait, woken_wake_function); 8212 8213 add_wait_queue(&netdev_unregistering_wq, &wait); 8214 for (;;) { 8215 unregistering = false; 8216 rtnl_lock(); 8217 list_for_each_entry(net, net_list, exit_list) { 8218 if (net->dev_unreg_count > 0) { 8219 unregistering = true; 8220 break; 8221 } 8222 } 8223 if (!unregistering) 8224 break; 8225 __rtnl_unlock(); 8226 8227 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 8228 } 8229 remove_wait_queue(&netdev_unregistering_wq, &wait); 8230 } 8231 8232 static void __net_exit default_device_exit_batch(struct list_head *net_list) 8233 { 8234 /* At exit all network devices most be removed from a network 8235 * namespace. Do this in the reverse order of registration. 8236 * Do this across as many network namespaces as possible to 8237 * improve batching efficiency. 8238 */ 8239 struct net_device *dev; 8240 struct net *net; 8241 LIST_HEAD(dev_kill_list); 8242 8243 /* To prevent network device cleanup code from dereferencing 8244 * loopback devices or network devices that have been freed 8245 * wait here for all pending unregistrations to complete, 8246 * before unregistring the loopback device and allowing the 8247 * network namespace be freed. 8248 * 8249 * The netdev todo list containing all network devices 8250 * unregistrations that happen in default_device_exit_batch 8251 * will run in the rtnl_unlock() at the end of 8252 * default_device_exit_batch. 8253 */ 8254 rtnl_lock_unregistering(net_list); 8255 list_for_each_entry(net, net_list, exit_list) { 8256 for_each_netdev_reverse(net, dev) { 8257 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 8258 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 8259 else 8260 unregister_netdevice_queue(dev, &dev_kill_list); 8261 } 8262 } 8263 unregister_netdevice_many(&dev_kill_list); 8264 rtnl_unlock(); 8265 } 8266 8267 static struct pernet_operations __net_initdata default_device_ops = { 8268 .exit = default_device_exit, 8269 .exit_batch = default_device_exit_batch, 8270 }; 8271 8272 /* 8273 * Initialize the DEV module. At boot time this walks the device list and 8274 * unhooks any devices that fail to initialise (normally hardware not 8275 * present) and leaves us with a valid list of present and active devices. 8276 * 8277 */ 8278 8279 /* 8280 * This is called single threaded during boot, so no need 8281 * to take the rtnl semaphore. 8282 */ 8283 static int __init net_dev_init(void) 8284 { 8285 int i, rc = -ENOMEM; 8286 8287 BUG_ON(!dev_boot_phase); 8288 8289 if (dev_proc_init()) 8290 goto out; 8291 8292 if (netdev_kobject_init()) 8293 goto out; 8294 8295 INIT_LIST_HEAD(&ptype_all); 8296 for (i = 0; i < PTYPE_HASH_SIZE; i++) 8297 INIT_LIST_HEAD(&ptype_base[i]); 8298 8299 INIT_LIST_HEAD(&offload_base); 8300 8301 if (register_pernet_subsys(&netdev_net_ops)) 8302 goto out; 8303 8304 /* 8305 * Initialise the packet receive queues. 8306 */ 8307 8308 for_each_possible_cpu(i) { 8309 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 8310 struct softnet_data *sd = &per_cpu(softnet_data, i); 8311 8312 INIT_WORK(flush, flush_backlog); 8313 8314 skb_queue_head_init(&sd->input_pkt_queue); 8315 skb_queue_head_init(&sd->process_queue); 8316 INIT_LIST_HEAD(&sd->poll_list); 8317 sd->output_queue_tailp = &sd->output_queue; 8318 #ifdef CONFIG_RPS 8319 sd->csd.func = rps_trigger_softirq; 8320 sd->csd.info = sd; 8321 sd->cpu = i; 8322 #endif 8323 8324 sd->backlog.poll = process_backlog; 8325 sd->backlog.weight = weight_p; 8326 } 8327 8328 dev_boot_phase = 0; 8329 8330 /* The loopback device is special if any other network devices 8331 * is present in a network namespace the loopback device must 8332 * be present. Since we now dynamically allocate and free the 8333 * loopback device ensure this invariant is maintained by 8334 * keeping the loopback device as the first device on the 8335 * list of network devices. Ensuring the loopback devices 8336 * is the first device that appears and the last network device 8337 * that disappears. 8338 */ 8339 if (register_pernet_device(&loopback_net_ops)) 8340 goto out; 8341 8342 if (register_pernet_device(&default_device_ops)) 8343 goto out; 8344 8345 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 8346 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 8347 8348 hotcpu_notifier(dev_cpu_callback, 0); 8349 dst_subsys_init(); 8350 rc = 0; 8351 out: 8352 return rc; 8353 } 8354 8355 subsys_initcall(net_dev_init); 8356