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 <linux/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/crash_dump.h> 143 144 #include "net-sysfs.h" 145 146 /* Instead of increasing this, you should create a hash table. */ 147 #define MAX_GRO_SKBS 8 148 149 /* This should be increased if a protocol with a bigger head is added. */ 150 #define GRO_MAX_HEAD (MAX_HEADER + 128) 151 152 static DEFINE_SPINLOCK(ptype_lock); 153 static DEFINE_SPINLOCK(offload_lock); 154 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 155 struct list_head ptype_all __read_mostly; /* Taps */ 156 static struct list_head offload_base __read_mostly; 157 158 static int netif_rx_internal(struct sk_buff *skb); 159 static int call_netdevice_notifiers_info(unsigned long val, 160 struct net_device *dev, 161 struct netdev_notifier_info *info); 162 163 /* 164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 165 * semaphore. 166 * 167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 168 * 169 * Writers must hold the rtnl semaphore while they loop through the 170 * dev_base_head list, and hold dev_base_lock for writing when they do the 171 * actual updates. This allows pure readers to access the list even 172 * while a writer is preparing to update it. 173 * 174 * To put it another way, dev_base_lock is held for writing only to 175 * protect against pure readers; the rtnl semaphore provides the 176 * protection against other writers. 177 * 178 * See, for example usages, register_netdevice() and 179 * unregister_netdevice(), which must be called with the rtnl 180 * semaphore held. 181 */ 182 DEFINE_RWLOCK(dev_base_lock); 183 EXPORT_SYMBOL(dev_base_lock); 184 185 /* protects napi_hash addition/deletion and napi_gen_id */ 186 static DEFINE_SPINLOCK(napi_hash_lock); 187 188 static unsigned int napi_gen_id = NR_CPUS; 189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 190 191 static seqcount_t devnet_rename_seq; 192 193 static inline void dev_base_seq_inc(struct net *net) 194 { 195 while (++net->dev_base_seq == 0) 196 ; 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 /* 363 * Add a protocol ID to the list. Now that the input handler is 364 * smarter we can dispense with all the messy stuff that used to be 365 * here. 366 * 367 * BEWARE!!! Protocol handlers, mangling input packets, 368 * MUST BE last in hash buckets and checking protocol handlers 369 * MUST start from promiscuous ptype_all chain in net_bh. 370 * It is true now, do not change it. 371 * Explanation follows: if protocol handler, mangling packet, will 372 * be the first on list, it is not able to sense, that packet 373 * is cloned and should be copied-on-write, so that it will 374 * change it and subsequent readers will get broken packet. 375 * --ANK (980803) 376 */ 377 378 static inline struct list_head *ptype_head(const struct packet_type *pt) 379 { 380 if (pt->type == htons(ETH_P_ALL)) 381 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 382 else 383 return pt->dev ? &pt->dev->ptype_specific : 384 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 385 } 386 387 /** 388 * dev_add_pack - add packet handler 389 * @pt: packet type declaration 390 * 391 * Add a protocol handler to the networking stack. The passed &packet_type 392 * is linked into kernel lists and may not be freed until it has been 393 * removed from the kernel lists. 394 * 395 * This call does not sleep therefore it can not 396 * guarantee all CPU's that are in middle of receiving packets 397 * will see the new packet type (until the next received packet). 398 */ 399 400 void dev_add_pack(struct packet_type *pt) 401 { 402 struct list_head *head = ptype_head(pt); 403 404 spin_lock(&ptype_lock); 405 list_add_rcu(&pt->list, head); 406 spin_unlock(&ptype_lock); 407 } 408 EXPORT_SYMBOL(dev_add_pack); 409 410 /** 411 * __dev_remove_pack - remove packet handler 412 * @pt: packet type declaration 413 * 414 * Remove a protocol handler that was previously added to the kernel 415 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 416 * from the kernel lists and can be freed or reused once this function 417 * returns. 418 * 419 * The packet type might still be in use by receivers 420 * and must not be freed until after all the CPU's have gone 421 * through a quiescent state. 422 */ 423 void __dev_remove_pack(struct packet_type *pt) 424 { 425 struct list_head *head = ptype_head(pt); 426 struct packet_type *pt1; 427 428 spin_lock(&ptype_lock); 429 430 list_for_each_entry(pt1, head, list) { 431 if (pt == pt1) { 432 list_del_rcu(&pt->list); 433 goto out; 434 } 435 } 436 437 pr_warn("dev_remove_pack: %p not found\n", pt); 438 out: 439 spin_unlock(&ptype_lock); 440 } 441 EXPORT_SYMBOL(__dev_remove_pack); 442 443 /** 444 * dev_remove_pack - remove packet handler 445 * @pt: packet type declaration 446 * 447 * Remove a protocol handler that was previously added to the kernel 448 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 449 * from the kernel lists and can be freed or reused once this function 450 * returns. 451 * 452 * This call sleeps to guarantee that no CPU is looking at the packet 453 * type after return. 454 */ 455 void dev_remove_pack(struct packet_type *pt) 456 { 457 __dev_remove_pack(pt); 458 459 synchronize_net(); 460 } 461 EXPORT_SYMBOL(dev_remove_pack); 462 463 464 /** 465 * dev_add_offload - register offload handlers 466 * @po: protocol offload declaration 467 * 468 * Add protocol offload handlers to the networking stack. The passed 469 * &proto_offload is linked into kernel lists and may not be freed until 470 * it has been removed from the kernel lists. 471 * 472 * This call does not sleep therefore it can not 473 * guarantee all CPU's that are in middle of receiving packets 474 * will see the new offload handlers (until the next received packet). 475 */ 476 void dev_add_offload(struct packet_offload *po) 477 { 478 struct packet_offload *elem; 479 480 spin_lock(&offload_lock); 481 list_for_each_entry(elem, &offload_base, list) { 482 if (po->priority < elem->priority) 483 break; 484 } 485 list_add_rcu(&po->list, elem->list.prev); 486 spin_unlock(&offload_lock); 487 } 488 EXPORT_SYMBOL(dev_add_offload); 489 490 /** 491 * __dev_remove_offload - remove offload handler 492 * @po: packet offload declaration 493 * 494 * Remove a protocol offload handler that was previously added to the 495 * kernel offload handlers by dev_add_offload(). The passed &offload_type 496 * is removed from the kernel lists and can be freed or reused once this 497 * function returns. 498 * 499 * The packet type might still be in use by receivers 500 * and must not be freed until after all the CPU's have gone 501 * through a quiescent state. 502 */ 503 static void __dev_remove_offload(struct packet_offload *po) 504 { 505 struct list_head *head = &offload_base; 506 struct packet_offload *po1; 507 508 spin_lock(&offload_lock); 509 510 list_for_each_entry(po1, head, list) { 511 if (po == po1) { 512 list_del_rcu(&po->list); 513 goto out; 514 } 515 } 516 517 pr_warn("dev_remove_offload: %p not found\n", po); 518 out: 519 spin_unlock(&offload_lock); 520 } 521 522 /** 523 * dev_remove_offload - remove packet offload handler 524 * @po: packet offload declaration 525 * 526 * Remove a packet offload handler that was previously added to the kernel 527 * offload handlers by dev_add_offload(). The passed &offload_type is 528 * removed from the kernel lists and can be freed or reused once this 529 * function returns. 530 * 531 * This call sleeps to guarantee that no CPU is looking at the packet 532 * type after return. 533 */ 534 void dev_remove_offload(struct packet_offload *po) 535 { 536 __dev_remove_offload(po); 537 538 synchronize_net(); 539 } 540 EXPORT_SYMBOL(dev_remove_offload); 541 542 /****************************************************************************** 543 * 544 * Device Boot-time Settings Routines 545 * 546 ******************************************************************************/ 547 548 /* Boot time configuration table */ 549 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 550 551 /** 552 * netdev_boot_setup_add - add new setup entry 553 * @name: name of the device 554 * @map: configured settings for the device 555 * 556 * Adds new setup entry to the dev_boot_setup list. The function 557 * returns 0 on error and 1 on success. This is a generic routine to 558 * all netdevices. 559 */ 560 static int netdev_boot_setup_add(char *name, struct ifmap *map) 561 { 562 struct netdev_boot_setup *s; 563 int i; 564 565 s = dev_boot_setup; 566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 567 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 568 memset(s[i].name, 0, sizeof(s[i].name)); 569 strlcpy(s[i].name, name, IFNAMSIZ); 570 memcpy(&s[i].map, map, sizeof(s[i].map)); 571 break; 572 } 573 } 574 575 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 576 } 577 578 /** 579 * netdev_boot_setup_check - check boot time settings 580 * @dev: the netdevice 581 * 582 * Check boot time settings for the device. 583 * The found settings are set for the device to be used 584 * later in the device probing. 585 * Returns 0 if no settings found, 1 if they are. 586 */ 587 int netdev_boot_setup_check(struct net_device *dev) 588 { 589 struct netdev_boot_setup *s = dev_boot_setup; 590 int i; 591 592 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 593 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 594 !strcmp(dev->name, s[i].name)) { 595 dev->irq = s[i].map.irq; 596 dev->base_addr = s[i].map.base_addr; 597 dev->mem_start = s[i].map.mem_start; 598 dev->mem_end = s[i].map.mem_end; 599 return 1; 600 } 601 } 602 return 0; 603 } 604 EXPORT_SYMBOL(netdev_boot_setup_check); 605 606 607 /** 608 * netdev_boot_base - get address from boot time settings 609 * @prefix: prefix for network device 610 * @unit: id for network device 611 * 612 * Check boot time settings for the base address of device. 613 * The found settings are set for the device to be used 614 * later in the device probing. 615 * Returns 0 if no settings found. 616 */ 617 unsigned long netdev_boot_base(const char *prefix, int unit) 618 { 619 const struct netdev_boot_setup *s = dev_boot_setup; 620 char name[IFNAMSIZ]; 621 int i; 622 623 sprintf(name, "%s%d", prefix, unit); 624 625 /* 626 * If device already registered then return base of 1 627 * to indicate not to probe for this interface 628 */ 629 if (__dev_get_by_name(&init_net, name)) 630 return 1; 631 632 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 633 if (!strcmp(name, s[i].name)) 634 return s[i].map.base_addr; 635 return 0; 636 } 637 638 /* 639 * Saves at boot time configured settings for any netdevice. 640 */ 641 int __init netdev_boot_setup(char *str) 642 { 643 int ints[5]; 644 struct ifmap map; 645 646 str = get_options(str, ARRAY_SIZE(ints), ints); 647 if (!str || !*str) 648 return 0; 649 650 /* Save settings */ 651 memset(&map, 0, sizeof(map)); 652 if (ints[0] > 0) 653 map.irq = ints[1]; 654 if (ints[0] > 1) 655 map.base_addr = ints[2]; 656 if (ints[0] > 2) 657 map.mem_start = ints[3]; 658 if (ints[0] > 3) 659 map.mem_end = ints[4]; 660 661 /* Add new entry to the list */ 662 return netdev_boot_setup_add(str, &map); 663 } 664 665 __setup("netdev=", netdev_boot_setup); 666 667 /******************************************************************************* 668 * 669 * Device Interface Subroutines 670 * 671 *******************************************************************************/ 672 673 /** 674 * dev_get_iflink - get 'iflink' value of a interface 675 * @dev: targeted interface 676 * 677 * Indicates the ifindex the interface is linked to. 678 * Physical interfaces have the same 'ifindex' and 'iflink' values. 679 */ 680 681 int dev_get_iflink(const struct net_device *dev) 682 { 683 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 684 return dev->netdev_ops->ndo_get_iflink(dev); 685 686 return dev->ifindex; 687 } 688 EXPORT_SYMBOL(dev_get_iflink); 689 690 /** 691 * dev_fill_metadata_dst - Retrieve tunnel egress information. 692 * @dev: targeted interface 693 * @skb: The packet. 694 * 695 * For better visibility of tunnel traffic OVS needs to retrieve 696 * egress tunnel information for a packet. Following API allows 697 * user to get this info. 698 */ 699 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 700 { 701 struct ip_tunnel_info *info; 702 703 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 704 return -EINVAL; 705 706 info = skb_tunnel_info_unclone(skb); 707 if (!info) 708 return -ENOMEM; 709 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 710 return -EINVAL; 711 712 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 713 } 714 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 715 716 /** 717 * __dev_get_by_name - find a device by its name 718 * @net: the applicable net namespace 719 * @name: name to find 720 * 721 * Find an interface by name. Must be called under RTNL semaphore 722 * or @dev_base_lock. If the name is found a pointer to the device 723 * is returned. If the name is not found then %NULL is returned. The 724 * reference counters are not incremented so the caller must be 725 * careful with locks. 726 */ 727 728 struct net_device *__dev_get_by_name(struct net *net, const char *name) 729 { 730 struct net_device *dev; 731 struct hlist_head *head = dev_name_hash(net, name); 732 733 hlist_for_each_entry(dev, head, name_hlist) 734 if (!strncmp(dev->name, name, IFNAMSIZ)) 735 return dev; 736 737 return NULL; 738 } 739 EXPORT_SYMBOL(__dev_get_by_name); 740 741 /** 742 * dev_get_by_name_rcu - find a device by its name 743 * @net: the applicable net namespace 744 * @name: name to find 745 * 746 * Find an interface by name. 747 * If the name is found a pointer to the device is returned. 748 * If the name is not found then %NULL is returned. 749 * The reference counters are not incremented so the caller must be 750 * careful with locks. The caller must hold RCU lock. 751 */ 752 753 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 754 { 755 struct net_device *dev; 756 struct hlist_head *head = dev_name_hash(net, name); 757 758 hlist_for_each_entry_rcu(dev, head, name_hlist) 759 if (!strncmp(dev->name, name, IFNAMSIZ)) 760 return dev; 761 762 return NULL; 763 } 764 EXPORT_SYMBOL(dev_get_by_name_rcu); 765 766 /** 767 * dev_get_by_name - find a device by its name 768 * @net: the applicable net namespace 769 * @name: name to find 770 * 771 * Find an interface by name. This can be called from any 772 * context and does its own locking. The returned handle has 773 * the usage count incremented and the caller must use dev_put() to 774 * release it when it is no longer needed. %NULL is returned if no 775 * matching device is found. 776 */ 777 778 struct net_device *dev_get_by_name(struct net *net, const char *name) 779 { 780 struct net_device *dev; 781 782 rcu_read_lock(); 783 dev = dev_get_by_name_rcu(net, name); 784 if (dev) 785 dev_hold(dev); 786 rcu_read_unlock(); 787 return dev; 788 } 789 EXPORT_SYMBOL(dev_get_by_name); 790 791 /** 792 * __dev_get_by_index - find a device by its ifindex 793 * @net: the applicable net namespace 794 * @ifindex: index of device 795 * 796 * Search for an interface by index. Returns %NULL if the device 797 * is not found or a pointer to the device. The device has not 798 * had its reference counter increased so the caller must be careful 799 * about locking. The caller must hold either the RTNL semaphore 800 * or @dev_base_lock. 801 */ 802 803 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 804 { 805 struct net_device *dev; 806 struct hlist_head *head = dev_index_hash(net, ifindex); 807 808 hlist_for_each_entry(dev, head, index_hlist) 809 if (dev->ifindex == ifindex) 810 return dev; 811 812 return NULL; 813 } 814 EXPORT_SYMBOL(__dev_get_by_index); 815 816 /** 817 * dev_get_by_index_rcu - find a device by its ifindex 818 * @net: the applicable net namespace 819 * @ifindex: index of device 820 * 821 * Search for an interface by index. Returns %NULL if the device 822 * is not found or a pointer to the device. The device has not 823 * had its reference counter increased so the caller must be careful 824 * about locking. The caller must hold RCU lock. 825 */ 826 827 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 828 { 829 struct net_device *dev; 830 struct hlist_head *head = dev_index_hash(net, ifindex); 831 832 hlist_for_each_entry_rcu(dev, head, index_hlist) 833 if (dev->ifindex == ifindex) 834 return dev; 835 836 return NULL; 837 } 838 EXPORT_SYMBOL(dev_get_by_index_rcu); 839 840 841 /** 842 * dev_get_by_index - find a device by its ifindex 843 * @net: the applicable net namespace 844 * @ifindex: index of device 845 * 846 * Search for an interface by index. Returns NULL if the device 847 * is not found or a pointer to the device. The device returned has 848 * had a reference added and the pointer is safe until the user calls 849 * dev_put to indicate they have finished with it. 850 */ 851 852 struct net_device *dev_get_by_index(struct net *net, int ifindex) 853 { 854 struct net_device *dev; 855 856 rcu_read_lock(); 857 dev = dev_get_by_index_rcu(net, ifindex); 858 if (dev) 859 dev_hold(dev); 860 rcu_read_unlock(); 861 return dev; 862 } 863 EXPORT_SYMBOL(dev_get_by_index); 864 865 /** 866 * netdev_get_name - get a netdevice name, knowing its ifindex. 867 * @net: network namespace 868 * @name: a pointer to the buffer where the name will be stored. 869 * @ifindex: the ifindex of the interface to get the name from. 870 * 871 * The use of raw_seqcount_begin() and cond_resched() before 872 * retrying is required as we want to give the writers a chance 873 * to complete when CONFIG_PREEMPT is not set. 874 */ 875 int netdev_get_name(struct net *net, char *name, int ifindex) 876 { 877 struct net_device *dev; 878 unsigned int seq; 879 880 retry: 881 seq = raw_seqcount_begin(&devnet_rename_seq); 882 rcu_read_lock(); 883 dev = dev_get_by_index_rcu(net, ifindex); 884 if (!dev) { 885 rcu_read_unlock(); 886 return -ENODEV; 887 } 888 889 strcpy(name, dev->name); 890 rcu_read_unlock(); 891 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 892 cond_resched(); 893 goto retry; 894 } 895 896 return 0; 897 } 898 899 /** 900 * dev_getbyhwaddr_rcu - find a device by its hardware address 901 * @net: the applicable net namespace 902 * @type: media type of device 903 * @ha: hardware address 904 * 905 * Search for an interface by MAC address. Returns NULL if the device 906 * is not found or a pointer to the device. 907 * The caller must hold RCU or RTNL. 908 * The returned device has not had its ref count increased 909 * and the caller must therefore be careful about locking 910 * 911 */ 912 913 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 914 const char *ha) 915 { 916 struct net_device *dev; 917 918 for_each_netdev_rcu(net, dev) 919 if (dev->type == type && 920 !memcmp(dev->dev_addr, ha, dev->addr_len)) 921 return dev; 922 923 return NULL; 924 } 925 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 926 927 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 928 { 929 struct net_device *dev; 930 931 ASSERT_RTNL(); 932 for_each_netdev(net, dev) 933 if (dev->type == type) 934 return dev; 935 936 return NULL; 937 } 938 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 939 940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 941 { 942 struct net_device *dev, *ret = NULL; 943 944 rcu_read_lock(); 945 for_each_netdev_rcu(net, dev) 946 if (dev->type == type) { 947 dev_hold(dev); 948 ret = dev; 949 break; 950 } 951 rcu_read_unlock(); 952 return ret; 953 } 954 EXPORT_SYMBOL(dev_getfirstbyhwtype); 955 956 /** 957 * __dev_get_by_flags - find any device with given flags 958 * @net: the applicable net namespace 959 * @if_flags: IFF_* values 960 * @mask: bitmask of bits in if_flags to check 961 * 962 * Search for any interface with the given flags. Returns NULL if a device 963 * is not found or a pointer to the device. Must be called inside 964 * rtnl_lock(), and result refcount is unchanged. 965 */ 966 967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 968 unsigned short mask) 969 { 970 struct net_device *dev, *ret; 971 972 ASSERT_RTNL(); 973 974 ret = NULL; 975 for_each_netdev(net, dev) { 976 if (((dev->flags ^ if_flags) & mask) == 0) { 977 ret = dev; 978 break; 979 } 980 } 981 return ret; 982 } 983 EXPORT_SYMBOL(__dev_get_by_flags); 984 985 /** 986 * dev_valid_name - check if name is okay for network device 987 * @name: name string 988 * 989 * Network device names need to be valid file names to 990 * to allow sysfs to work. We also disallow any kind of 991 * whitespace. 992 */ 993 bool dev_valid_name(const char *name) 994 { 995 if (*name == '\0') 996 return false; 997 if (strlen(name) >= IFNAMSIZ) 998 return false; 999 if (!strcmp(name, ".") || !strcmp(name, "..")) 1000 return false; 1001 1002 while (*name) { 1003 if (*name == '/' || *name == ':' || isspace(*name)) 1004 return false; 1005 name++; 1006 } 1007 return true; 1008 } 1009 EXPORT_SYMBOL(dev_valid_name); 1010 1011 /** 1012 * __dev_alloc_name - allocate a name for a device 1013 * @net: network namespace to allocate the device name in 1014 * @name: name format string 1015 * @buf: scratch buffer and result name string 1016 * 1017 * Passed a format string - eg "lt%d" it will try and find a suitable 1018 * id. It scans list of devices to build up a free map, then chooses 1019 * the first empty slot. The caller must hold the dev_base or rtnl lock 1020 * while allocating the name and adding the device in order to avoid 1021 * duplicates. 1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1023 * Returns the number of the unit assigned or a negative errno code. 1024 */ 1025 1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1027 { 1028 int i = 0; 1029 const char *p; 1030 const int max_netdevices = 8*PAGE_SIZE; 1031 unsigned long *inuse; 1032 struct net_device *d; 1033 1034 p = strnchr(name, IFNAMSIZ-1, '%'); 1035 if (p) { 1036 /* 1037 * Verify the string as this thing may have come from 1038 * the user. There must be either one "%d" and no other "%" 1039 * characters. 1040 */ 1041 if (p[1] != 'd' || strchr(p + 2, '%')) 1042 return -EINVAL; 1043 1044 /* Use one page as a bit array of possible slots */ 1045 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1046 if (!inuse) 1047 return -ENOMEM; 1048 1049 for_each_netdev(net, d) { 1050 if (!sscanf(d->name, name, &i)) 1051 continue; 1052 if (i < 0 || i >= max_netdevices) 1053 continue; 1054 1055 /* avoid cases where sscanf is not exact inverse of printf */ 1056 snprintf(buf, IFNAMSIZ, name, i); 1057 if (!strncmp(buf, d->name, IFNAMSIZ)) 1058 set_bit(i, inuse); 1059 } 1060 1061 i = find_first_zero_bit(inuse, max_netdevices); 1062 free_page((unsigned long) inuse); 1063 } 1064 1065 if (buf != name) 1066 snprintf(buf, IFNAMSIZ, name, i); 1067 if (!__dev_get_by_name(net, buf)) 1068 return i; 1069 1070 /* It is possible to run out of possible slots 1071 * when the name is long and there isn't enough space left 1072 * for the digits, or if all bits are used. 1073 */ 1074 return -ENFILE; 1075 } 1076 1077 /** 1078 * dev_alloc_name - allocate a name for a device 1079 * @dev: device 1080 * @name: name format string 1081 * 1082 * Passed a format string - eg "lt%d" it will try and find a suitable 1083 * id. It scans list of devices to build up a free map, then chooses 1084 * the first empty slot. The caller must hold the dev_base or rtnl lock 1085 * while allocating the name and adding the device in order to avoid 1086 * duplicates. 1087 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1088 * Returns the number of the unit assigned or a negative errno code. 1089 */ 1090 1091 int dev_alloc_name(struct net_device *dev, const char *name) 1092 { 1093 char buf[IFNAMSIZ]; 1094 struct net *net; 1095 int ret; 1096 1097 BUG_ON(!dev_net(dev)); 1098 net = dev_net(dev); 1099 ret = __dev_alloc_name(net, name, buf); 1100 if (ret >= 0) 1101 strlcpy(dev->name, buf, IFNAMSIZ); 1102 return ret; 1103 } 1104 EXPORT_SYMBOL(dev_alloc_name); 1105 1106 static int dev_alloc_name_ns(struct net *net, 1107 struct net_device *dev, 1108 const char *name) 1109 { 1110 char buf[IFNAMSIZ]; 1111 int ret; 1112 1113 ret = __dev_alloc_name(net, name, buf); 1114 if (ret >= 0) 1115 strlcpy(dev->name, buf, IFNAMSIZ); 1116 return ret; 1117 } 1118 1119 static int dev_get_valid_name(struct net *net, 1120 struct net_device *dev, 1121 const char *name) 1122 { 1123 BUG_ON(!net); 1124 1125 if (!dev_valid_name(name)) 1126 return -EINVAL; 1127 1128 if (strchr(name, '%')) 1129 return dev_alloc_name_ns(net, dev, name); 1130 else if (__dev_get_by_name(net, name)) 1131 return -EEXIST; 1132 else if (dev->name != name) 1133 strlcpy(dev->name, name, IFNAMSIZ); 1134 1135 return 0; 1136 } 1137 1138 /** 1139 * dev_change_name - change name of a device 1140 * @dev: device 1141 * @newname: name (or format string) must be at least IFNAMSIZ 1142 * 1143 * Change name of a device, can pass format strings "eth%d". 1144 * for wildcarding. 1145 */ 1146 int dev_change_name(struct net_device *dev, const char *newname) 1147 { 1148 unsigned char old_assign_type; 1149 char oldname[IFNAMSIZ]; 1150 int err = 0; 1151 int ret; 1152 struct net *net; 1153 1154 ASSERT_RTNL(); 1155 BUG_ON(!dev_net(dev)); 1156 1157 net = dev_net(dev); 1158 if (dev->flags & IFF_UP) 1159 return -EBUSY; 1160 1161 write_seqcount_begin(&devnet_rename_seq); 1162 1163 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1164 write_seqcount_end(&devnet_rename_seq); 1165 return 0; 1166 } 1167 1168 memcpy(oldname, dev->name, IFNAMSIZ); 1169 1170 err = dev_get_valid_name(net, dev, newname); 1171 if (err < 0) { 1172 write_seqcount_end(&devnet_rename_seq); 1173 return err; 1174 } 1175 1176 if (oldname[0] && !strchr(oldname, '%')) 1177 netdev_info(dev, "renamed from %s\n", oldname); 1178 1179 old_assign_type = dev->name_assign_type; 1180 dev->name_assign_type = NET_NAME_RENAMED; 1181 1182 rollback: 1183 ret = device_rename(&dev->dev, dev->name); 1184 if (ret) { 1185 memcpy(dev->name, oldname, IFNAMSIZ); 1186 dev->name_assign_type = old_assign_type; 1187 write_seqcount_end(&devnet_rename_seq); 1188 return ret; 1189 } 1190 1191 write_seqcount_end(&devnet_rename_seq); 1192 1193 netdev_adjacent_rename_links(dev, oldname); 1194 1195 write_lock_bh(&dev_base_lock); 1196 hlist_del_rcu(&dev->name_hlist); 1197 write_unlock_bh(&dev_base_lock); 1198 1199 synchronize_rcu(); 1200 1201 write_lock_bh(&dev_base_lock); 1202 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1203 write_unlock_bh(&dev_base_lock); 1204 1205 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1206 ret = notifier_to_errno(ret); 1207 1208 if (ret) { 1209 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1210 if (err >= 0) { 1211 err = ret; 1212 write_seqcount_begin(&devnet_rename_seq); 1213 memcpy(dev->name, oldname, IFNAMSIZ); 1214 memcpy(oldname, newname, IFNAMSIZ); 1215 dev->name_assign_type = old_assign_type; 1216 old_assign_type = NET_NAME_RENAMED; 1217 goto rollback; 1218 } else { 1219 pr_err("%s: name change rollback failed: %d\n", 1220 dev->name, ret); 1221 } 1222 } 1223 1224 return err; 1225 } 1226 1227 /** 1228 * dev_set_alias - change ifalias of a device 1229 * @dev: device 1230 * @alias: name up to IFALIASZ 1231 * @len: limit of bytes to copy from info 1232 * 1233 * Set ifalias for a device, 1234 */ 1235 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1236 { 1237 char *new_ifalias; 1238 1239 ASSERT_RTNL(); 1240 1241 if (len >= IFALIASZ) 1242 return -EINVAL; 1243 1244 if (!len) { 1245 kfree(dev->ifalias); 1246 dev->ifalias = NULL; 1247 return 0; 1248 } 1249 1250 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); 1251 if (!new_ifalias) 1252 return -ENOMEM; 1253 dev->ifalias = new_ifalias; 1254 1255 strlcpy(dev->ifalias, alias, len+1); 1256 return len; 1257 } 1258 1259 1260 /** 1261 * netdev_features_change - device changes features 1262 * @dev: device to cause notification 1263 * 1264 * Called to indicate a device has changed features. 1265 */ 1266 void netdev_features_change(struct net_device *dev) 1267 { 1268 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1269 } 1270 EXPORT_SYMBOL(netdev_features_change); 1271 1272 /** 1273 * netdev_state_change - device changes state 1274 * @dev: device to cause notification 1275 * 1276 * Called to indicate a device has changed state. This function calls 1277 * the notifier chains for netdev_chain and sends a NEWLINK message 1278 * to the routing socket. 1279 */ 1280 void netdev_state_change(struct net_device *dev) 1281 { 1282 if (dev->flags & IFF_UP) { 1283 struct netdev_notifier_change_info change_info; 1284 1285 change_info.flags_changed = 0; 1286 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 1287 &change_info.info); 1288 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1289 } 1290 } 1291 EXPORT_SYMBOL(netdev_state_change); 1292 1293 /** 1294 * netdev_notify_peers - notify network peers about existence of @dev 1295 * @dev: network device 1296 * 1297 * Generate traffic such that interested network peers are aware of 1298 * @dev, such as by generating a gratuitous ARP. This may be used when 1299 * a device wants to inform the rest of the network about some sort of 1300 * reconfiguration such as a failover event or virtual machine 1301 * migration. 1302 */ 1303 void netdev_notify_peers(struct net_device *dev) 1304 { 1305 rtnl_lock(); 1306 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1307 rtnl_unlock(); 1308 } 1309 EXPORT_SYMBOL(netdev_notify_peers); 1310 1311 static int __dev_open(struct net_device *dev) 1312 { 1313 const struct net_device_ops *ops = dev->netdev_ops; 1314 int ret; 1315 1316 ASSERT_RTNL(); 1317 1318 if (!netif_device_present(dev)) 1319 return -ENODEV; 1320 1321 /* Block netpoll from trying to do any rx path servicing. 1322 * If we don't do this there is a chance ndo_poll_controller 1323 * or ndo_poll may be running while we open the device 1324 */ 1325 netpoll_poll_disable(dev); 1326 1327 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); 1328 ret = notifier_to_errno(ret); 1329 if (ret) 1330 return ret; 1331 1332 set_bit(__LINK_STATE_START, &dev->state); 1333 1334 if (ops->ndo_validate_addr) 1335 ret = ops->ndo_validate_addr(dev); 1336 1337 if (!ret && ops->ndo_open) 1338 ret = ops->ndo_open(dev); 1339 1340 netpoll_poll_enable(dev); 1341 1342 if (ret) 1343 clear_bit(__LINK_STATE_START, &dev->state); 1344 else { 1345 dev->flags |= IFF_UP; 1346 dev_set_rx_mode(dev); 1347 dev_activate(dev); 1348 add_device_randomness(dev->dev_addr, dev->addr_len); 1349 } 1350 1351 return ret; 1352 } 1353 1354 /** 1355 * dev_open - prepare an interface for use. 1356 * @dev: device to open 1357 * 1358 * Takes a device from down to up state. The device's private open 1359 * function is invoked and then the multicast lists are loaded. Finally 1360 * the device is moved into the up state and a %NETDEV_UP message is 1361 * sent to the netdev notifier chain. 1362 * 1363 * Calling this function on an active interface is a nop. On a failure 1364 * a negative errno code is returned. 1365 */ 1366 int dev_open(struct net_device *dev) 1367 { 1368 int ret; 1369 1370 if (dev->flags & IFF_UP) 1371 return 0; 1372 1373 ret = __dev_open(dev); 1374 if (ret < 0) 1375 return ret; 1376 1377 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1378 call_netdevice_notifiers(NETDEV_UP, dev); 1379 1380 return ret; 1381 } 1382 EXPORT_SYMBOL(dev_open); 1383 1384 static int __dev_close_many(struct list_head *head) 1385 { 1386 struct net_device *dev; 1387 1388 ASSERT_RTNL(); 1389 might_sleep(); 1390 1391 list_for_each_entry(dev, head, close_list) { 1392 /* Temporarily disable netpoll until the interface is down */ 1393 netpoll_poll_disable(dev); 1394 1395 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1396 1397 clear_bit(__LINK_STATE_START, &dev->state); 1398 1399 /* Synchronize to scheduled poll. We cannot touch poll list, it 1400 * can be even on different cpu. So just clear netif_running(). 1401 * 1402 * dev->stop() will invoke napi_disable() on all of it's 1403 * napi_struct instances on this device. 1404 */ 1405 smp_mb__after_atomic(); /* Commit netif_running(). */ 1406 } 1407 1408 dev_deactivate_many(head); 1409 1410 list_for_each_entry(dev, head, close_list) { 1411 const struct net_device_ops *ops = dev->netdev_ops; 1412 1413 /* 1414 * Call the device specific close. This cannot fail. 1415 * Only if device is UP 1416 * 1417 * We allow it to be called even after a DETACH hot-plug 1418 * event. 1419 */ 1420 if (ops->ndo_stop) 1421 ops->ndo_stop(dev); 1422 1423 dev->flags &= ~IFF_UP; 1424 netpoll_poll_enable(dev); 1425 } 1426 1427 return 0; 1428 } 1429 1430 static int __dev_close(struct net_device *dev) 1431 { 1432 int retval; 1433 LIST_HEAD(single); 1434 1435 list_add(&dev->close_list, &single); 1436 retval = __dev_close_many(&single); 1437 list_del(&single); 1438 1439 return retval; 1440 } 1441 1442 int dev_close_many(struct list_head *head, bool unlink) 1443 { 1444 struct net_device *dev, *tmp; 1445 1446 /* Remove the devices that don't need to be closed */ 1447 list_for_each_entry_safe(dev, tmp, head, close_list) 1448 if (!(dev->flags & IFF_UP)) 1449 list_del_init(&dev->close_list); 1450 1451 __dev_close_many(head); 1452 1453 list_for_each_entry_safe(dev, tmp, head, close_list) { 1454 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1455 call_netdevice_notifiers(NETDEV_DOWN, dev); 1456 if (unlink) 1457 list_del_init(&dev->close_list); 1458 } 1459 1460 return 0; 1461 } 1462 EXPORT_SYMBOL(dev_close_many); 1463 1464 /** 1465 * dev_close - shutdown an interface. 1466 * @dev: device to shutdown 1467 * 1468 * This function moves an active device into down state. A 1469 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1470 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1471 * chain. 1472 */ 1473 int dev_close(struct net_device *dev) 1474 { 1475 if (dev->flags & IFF_UP) { 1476 LIST_HEAD(single); 1477 1478 list_add(&dev->close_list, &single); 1479 dev_close_many(&single, true); 1480 list_del(&single); 1481 } 1482 return 0; 1483 } 1484 EXPORT_SYMBOL(dev_close); 1485 1486 1487 /** 1488 * dev_disable_lro - disable Large Receive Offload on a device 1489 * @dev: device 1490 * 1491 * Disable Large Receive Offload (LRO) on a net device. Must be 1492 * called under RTNL. This is needed if received packets may be 1493 * forwarded to another interface. 1494 */ 1495 void dev_disable_lro(struct net_device *dev) 1496 { 1497 struct net_device *lower_dev; 1498 struct list_head *iter; 1499 1500 dev->wanted_features &= ~NETIF_F_LRO; 1501 netdev_update_features(dev); 1502 1503 if (unlikely(dev->features & NETIF_F_LRO)) 1504 netdev_WARN(dev, "failed to disable LRO!\n"); 1505 1506 netdev_for_each_lower_dev(dev, lower_dev, iter) 1507 dev_disable_lro(lower_dev); 1508 } 1509 EXPORT_SYMBOL(dev_disable_lro); 1510 1511 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1512 struct net_device *dev) 1513 { 1514 struct netdev_notifier_info info; 1515 1516 netdev_notifier_info_init(&info, dev); 1517 return nb->notifier_call(nb, val, &info); 1518 } 1519 1520 static int dev_boot_phase = 1; 1521 1522 /** 1523 * register_netdevice_notifier - register a network notifier block 1524 * @nb: notifier 1525 * 1526 * Register a notifier to be called when network device events occur. 1527 * The notifier passed is linked into the kernel structures and must 1528 * not be reused until it has been unregistered. A negative errno code 1529 * is returned on a failure. 1530 * 1531 * When registered all registration and up events are replayed 1532 * to the new notifier to allow device to have a race free 1533 * view of the network device list. 1534 */ 1535 1536 int register_netdevice_notifier(struct notifier_block *nb) 1537 { 1538 struct net_device *dev; 1539 struct net_device *last; 1540 struct net *net; 1541 int err; 1542 1543 rtnl_lock(); 1544 err = raw_notifier_chain_register(&netdev_chain, nb); 1545 if (err) 1546 goto unlock; 1547 if (dev_boot_phase) 1548 goto unlock; 1549 for_each_net(net) { 1550 for_each_netdev(net, dev) { 1551 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1552 err = notifier_to_errno(err); 1553 if (err) 1554 goto rollback; 1555 1556 if (!(dev->flags & IFF_UP)) 1557 continue; 1558 1559 call_netdevice_notifier(nb, NETDEV_UP, dev); 1560 } 1561 } 1562 1563 unlock: 1564 rtnl_unlock(); 1565 return err; 1566 1567 rollback: 1568 last = dev; 1569 for_each_net(net) { 1570 for_each_netdev(net, dev) { 1571 if (dev == last) 1572 goto outroll; 1573 1574 if (dev->flags & IFF_UP) { 1575 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1576 dev); 1577 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1578 } 1579 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1580 } 1581 } 1582 1583 outroll: 1584 raw_notifier_chain_unregister(&netdev_chain, nb); 1585 goto unlock; 1586 } 1587 EXPORT_SYMBOL(register_netdevice_notifier); 1588 1589 /** 1590 * unregister_netdevice_notifier - unregister a network notifier block 1591 * @nb: notifier 1592 * 1593 * Unregister a notifier previously registered by 1594 * register_netdevice_notifier(). The notifier is unlinked into the 1595 * kernel structures and may then be reused. A negative errno code 1596 * is returned on a failure. 1597 * 1598 * After unregistering unregister and down device events are synthesized 1599 * for all devices on the device list to the removed notifier to remove 1600 * the need for special case cleanup code. 1601 */ 1602 1603 int unregister_netdevice_notifier(struct notifier_block *nb) 1604 { 1605 struct net_device *dev; 1606 struct net *net; 1607 int err; 1608 1609 rtnl_lock(); 1610 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1611 if (err) 1612 goto unlock; 1613 1614 for_each_net(net) { 1615 for_each_netdev(net, dev) { 1616 if (dev->flags & IFF_UP) { 1617 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1618 dev); 1619 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1620 } 1621 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1622 } 1623 } 1624 unlock: 1625 rtnl_unlock(); 1626 return err; 1627 } 1628 EXPORT_SYMBOL(unregister_netdevice_notifier); 1629 1630 /** 1631 * call_netdevice_notifiers_info - call all network notifier blocks 1632 * @val: value passed unmodified to notifier function 1633 * @dev: net_device pointer passed unmodified to notifier function 1634 * @info: notifier information data 1635 * 1636 * Call all network notifier blocks. Parameters and return value 1637 * are as for raw_notifier_call_chain(). 1638 */ 1639 1640 static int call_netdevice_notifiers_info(unsigned long val, 1641 struct net_device *dev, 1642 struct netdev_notifier_info *info) 1643 { 1644 ASSERT_RTNL(); 1645 netdev_notifier_info_init(info, dev); 1646 return raw_notifier_call_chain(&netdev_chain, val, info); 1647 } 1648 1649 /** 1650 * call_netdevice_notifiers - call all network notifier blocks 1651 * @val: value passed unmodified to notifier function 1652 * @dev: net_device pointer passed unmodified to notifier function 1653 * 1654 * Call all network notifier blocks. Parameters and return value 1655 * are as for raw_notifier_call_chain(). 1656 */ 1657 1658 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1659 { 1660 struct netdev_notifier_info info; 1661 1662 return call_netdevice_notifiers_info(val, dev, &info); 1663 } 1664 EXPORT_SYMBOL(call_netdevice_notifiers); 1665 1666 #ifdef CONFIG_NET_INGRESS 1667 static struct static_key ingress_needed __read_mostly; 1668 1669 void net_inc_ingress_queue(void) 1670 { 1671 static_key_slow_inc(&ingress_needed); 1672 } 1673 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1674 1675 void net_dec_ingress_queue(void) 1676 { 1677 static_key_slow_dec(&ingress_needed); 1678 } 1679 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1680 #endif 1681 1682 #ifdef CONFIG_NET_EGRESS 1683 static struct static_key egress_needed __read_mostly; 1684 1685 void net_inc_egress_queue(void) 1686 { 1687 static_key_slow_inc(&egress_needed); 1688 } 1689 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 1690 1691 void net_dec_egress_queue(void) 1692 { 1693 static_key_slow_dec(&egress_needed); 1694 } 1695 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 1696 #endif 1697 1698 static struct static_key netstamp_needed __read_mostly; 1699 #ifdef HAVE_JUMP_LABEL 1700 static atomic_t netstamp_needed_deferred; 1701 static void netstamp_clear(struct work_struct *work) 1702 { 1703 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1704 1705 while (deferred--) 1706 static_key_slow_dec(&netstamp_needed); 1707 } 1708 static DECLARE_WORK(netstamp_work, netstamp_clear); 1709 #endif 1710 1711 void net_enable_timestamp(void) 1712 { 1713 static_key_slow_inc(&netstamp_needed); 1714 } 1715 EXPORT_SYMBOL(net_enable_timestamp); 1716 1717 void net_disable_timestamp(void) 1718 { 1719 #ifdef HAVE_JUMP_LABEL 1720 /* net_disable_timestamp() can be called from non process context */ 1721 atomic_inc(&netstamp_needed_deferred); 1722 schedule_work(&netstamp_work); 1723 #else 1724 static_key_slow_dec(&netstamp_needed); 1725 #endif 1726 } 1727 EXPORT_SYMBOL(net_disable_timestamp); 1728 1729 static inline void net_timestamp_set(struct sk_buff *skb) 1730 { 1731 skb->tstamp = 0; 1732 if (static_key_false(&netstamp_needed)) 1733 __net_timestamp(skb); 1734 } 1735 1736 #define net_timestamp_check(COND, SKB) \ 1737 if (static_key_false(&netstamp_needed)) { \ 1738 if ((COND) && !(SKB)->tstamp) \ 1739 __net_timestamp(SKB); \ 1740 } \ 1741 1742 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 1743 { 1744 unsigned int len; 1745 1746 if (!(dev->flags & IFF_UP)) 1747 return false; 1748 1749 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1750 if (skb->len <= len) 1751 return true; 1752 1753 /* if TSO is enabled, we don't care about the length as the packet 1754 * could be forwarded without being segmented before 1755 */ 1756 if (skb_is_gso(skb)) 1757 return true; 1758 1759 return false; 1760 } 1761 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1762 1763 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1764 { 1765 int ret = ____dev_forward_skb(dev, skb); 1766 1767 if (likely(!ret)) { 1768 skb->protocol = eth_type_trans(skb, dev); 1769 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1770 } 1771 1772 return ret; 1773 } 1774 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1775 1776 /** 1777 * dev_forward_skb - loopback an skb to another netif 1778 * 1779 * @dev: destination network device 1780 * @skb: buffer to forward 1781 * 1782 * return values: 1783 * NET_RX_SUCCESS (no congestion) 1784 * NET_RX_DROP (packet was dropped, but freed) 1785 * 1786 * dev_forward_skb can be used for injecting an skb from the 1787 * start_xmit function of one device into the receive queue 1788 * of another device. 1789 * 1790 * The receiving device may be in another namespace, so 1791 * we have to clear all information in the skb that could 1792 * impact namespace isolation. 1793 */ 1794 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1795 { 1796 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1797 } 1798 EXPORT_SYMBOL_GPL(dev_forward_skb); 1799 1800 static inline int deliver_skb(struct sk_buff *skb, 1801 struct packet_type *pt_prev, 1802 struct net_device *orig_dev) 1803 { 1804 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 1805 return -ENOMEM; 1806 atomic_inc(&skb->users); 1807 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1808 } 1809 1810 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 1811 struct packet_type **pt, 1812 struct net_device *orig_dev, 1813 __be16 type, 1814 struct list_head *ptype_list) 1815 { 1816 struct packet_type *ptype, *pt_prev = *pt; 1817 1818 list_for_each_entry_rcu(ptype, ptype_list, list) { 1819 if (ptype->type != type) 1820 continue; 1821 if (pt_prev) 1822 deliver_skb(skb, pt_prev, orig_dev); 1823 pt_prev = ptype; 1824 } 1825 *pt = pt_prev; 1826 } 1827 1828 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1829 { 1830 if (!ptype->af_packet_priv || !skb->sk) 1831 return false; 1832 1833 if (ptype->id_match) 1834 return ptype->id_match(ptype, skb->sk); 1835 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1836 return true; 1837 1838 return false; 1839 } 1840 1841 /* 1842 * Support routine. Sends outgoing frames to any network 1843 * taps currently in use. 1844 */ 1845 1846 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1847 { 1848 struct packet_type *ptype; 1849 struct sk_buff *skb2 = NULL; 1850 struct packet_type *pt_prev = NULL; 1851 struct list_head *ptype_list = &ptype_all; 1852 1853 rcu_read_lock(); 1854 again: 1855 list_for_each_entry_rcu(ptype, ptype_list, list) { 1856 /* Never send packets back to the socket 1857 * they originated from - MvS (miquels@drinkel.ow.org) 1858 */ 1859 if (skb_loop_sk(ptype, skb)) 1860 continue; 1861 1862 if (pt_prev) { 1863 deliver_skb(skb2, pt_prev, skb->dev); 1864 pt_prev = ptype; 1865 continue; 1866 } 1867 1868 /* need to clone skb, done only once */ 1869 skb2 = skb_clone(skb, GFP_ATOMIC); 1870 if (!skb2) 1871 goto out_unlock; 1872 1873 net_timestamp_set(skb2); 1874 1875 /* skb->nh should be correctly 1876 * set by sender, so that the second statement is 1877 * just protection against buggy protocols. 1878 */ 1879 skb_reset_mac_header(skb2); 1880 1881 if (skb_network_header(skb2) < skb2->data || 1882 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 1883 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 1884 ntohs(skb2->protocol), 1885 dev->name); 1886 skb_reset_network_header(skb2); 1887 } 1888 1889 skb2->transport_header = skb2->network_header; 1890 skb2->pkt_type = PACKET_OUTGOING; 1891 pt_prev = ptype; 1892 } 1893 1894 if (ptype_list == &ptype_all) { 1895 ptype_list = &dev->ptype_all; 1896 goto again; 1897 } 1898 out_unlock: 1899 if (pt_prev) 1900 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 1901 rcu_read_unlock(); 1902 } 1903 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 1904 1905 /** 1906 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 1907 * @dev: Network device 1908 * @txq: number of queues available 1909 * 1910 * If real_num_tx_queues is changed the tc mappings may no longer be 1911 * valid. To resolve this verify the tc mapping remains valid and if 1912 * not NULL the mapping. With no priorities mapping to this 1913 * offset/count pair it will no longer be used. In the worst case TC0 1914 * is invalid nothing can be done so disable priority mappings. If is 1915 * expected that drivers will fix this mapping if they can before 1916 * calling netif_set_real_num_tx_queues. 1917 */ 1918 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 1919 { 1920 int i; 1921 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 1922 1923 /* If TC0 is invalidated disable TC mapping */ 1924 if (tc->offset + tc->count > txq) { 1925 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 1926 dev->num_tc = 0; 1927 return; 1928 } 1929 1930 /* Invalidated prio to tc mappings set to TC0 */ 1931 for (i = 1; i < TC_BITMASK + 1; i++) { 1932 int q = netdev_get_prio_tc_map(dev, i); 1933 1934 tc = &dev->tc_to_txq[q]; 1935 if (tc->offset + tc->count > txq) { 1936 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 1937 i, q); 1938 netdev_set_prio_tc_map(dev, i, 0); 1939 } 1940 } 1941 } 1942 1943 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 1944 { 1945 if (dev->num_tc) { 1946 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 1947 int i; 1948 1949 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 1950 if ((txq - tc->offset) < tc->count) 1951 return i; 1952 } 1953 1954 return -1; 1955 } 1956 1957 return 0; 1958 } 1959 1960 #ifdef CONFIG_XPS 1961 static DEFINE_MUTEX(xps_map_mutex); 1962 #define xmap_dereference(P) \ 1963 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 1964 1965 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 1966 int tci, u16 index) 1967 { 1968 struct xps_map *map = NULL; 1969 int pos; 1970 1971 if (dev_maps) 1972 map = xmap_dereference(dev_maps->cpu_map[tci]); 1973 if (!map) 1974 return false; 1975 1976 for (pos = map->len; pos--;) { 1977 if (map->queues[pos] != index) 1978 continue; 1979 1980 if (map->len > 1) { 1981 map->queues[pos] = map->queues[--map->len]; 1982 break; 1983 } 1984 1985 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); 1986 kfree_rcu(map, rcu); 1987 return false; 1988 } 1989 1990 return true; 1991 } 1992 1993 static bool remove_xps_queue_cpu(struct net_device *dev, 1994 struct xps_dev_maps *dev_maps, 1995 int cpu, u16 offset, u16 count) 1996 { 1997 int num_tc = dev->num_tc ? : 1; 1998 bool active = false; 1999 int tci; 2000 2001 for (tci = cpu * num_tc; num_tc--; tci++) { 2002 int i, j; 2003 2004 for (i = count, j = offset; i--; j++) { 2005 if (!remove_xps_queue(dev_maps, cpu, j)) 2006 break; 2007 } 2008 2009 active |= i < 0; 2010 } 2011 2012 return active; 2013 } 2014 2015 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2016 u16 count) 2017 { 2018 struct xps_dev_maps *dev_maps; 2019 int cpu, i; 2020 bool active = false; 2021 2022 mutex_lock(&xps_map_mutex); 2023 dev_maps = xmap_dereference(dev->xps_maps); 2024 2025 if (!dev_maps) 2026 goto out_no_maps; 2027 2028 for_each_possible_cpu(cpu) 2029 active |= remove_xps_queue_cpu(dev, dev_maps, cpu, 2030 offset, count); 2031 2032 if (!active) { 2033 RCU_INIT_POINTER(dev->xps_maps, NULL); 2034 kfree_rcu(dev_maps, rcu); 2035 } 2036 2037 for (i = offset + (count - 1); count--; i--) 2038 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), 2039 NUMA_NO_NODE); 2040 2041 out_no_maps: 2042 mutex_unlock(&xps_map_mutex); 2043 } 2044 2045 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2046 { 2047 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2048 } 2049 2050 static struct xps_map *expand_xps_map(struct xps_map *map, 2051 int cpu, u16 index) 2052 { 2053 struct xps_map *new_map; 2054 int alloc_len = XPS_MIN_MAP_ALLOC; 2055 int i, pos; 2056 2057 for (pos = 0; map && pos < map->len; pos++) { 2058 if (map->queues[pos] != index) 2059 continue; 2060 return map; 2061 } 2062 2063 /* Need to add queue to this CPU's existing map */ 2064 if (map) { 2065 if (pos < map->alloc_len) 2066 return map; 2067 2068 alloc_len = map->alloc_len * 2; 2069 } 2070 2071 /* Need to allocate new map to store queue on this CPU's map */ 2072 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2073 cpu_to_node(cpu)); 2074 if (!new_map) 2075 return NULL; 2076 2077 for (i = 0; i < pos; i++) 2078 new_map->queues[i] = map->queues[i]; 2079 new_map->alloc_len = alloc_len; 2080 new_map->len = pos; 2081 2082 return new_map; 2083 } 2084 2085 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2086 u16 index) 2087 { 2088 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2089 int i, cpu, tci, numa_node_id = -2; 2090 int maps_sz, num_tc = 1, tc = 0; 2091 struct xps_map *map, *new_map; 2092 bool active = false; 2093 2094 if (dev->num_tc) { 2095 num_tc = dev->num_tc; 2096 tc = netdev_txq_to_tc(dev, index); 2097 if (tc < 0) 2098 return -EINVAL; 2099 } 2100 2101 maps_sz = XPS_DEV_MAPS_SIZE(num_tc); 2102 if (maps_sz < L1_CACHE_BYTES) 2103 maps_sz = L1_CACHE_BYTES; 2104 2105 mutex_lock(&xps_map_mutex); 2106 2107 dev_maps = xmap_dereference(dev->xps_maps); 2108 2109 /* allocate memory for queue storage */ 2110 for_each_cpu_and(cpu, cpu_online_mask, mask) { 2111 if (!new_dev_maps) 2112 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2113 if (!new_dev_maps) { 2114 mutex_unlock(&xps_map_mutex); 2115 return -ENOMEM; 2116 } 2117 2118 tci = cpu * num_tc + tc; 2119 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : 2120 NULL; 2121 2122 map = expand_xps_map(map, cpu, index); 2123 if (!map) 2124 goto error; 2125 2126 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2127 } 2128 2129 if (!new_dev_maps) 2130 goto out_no_new_maps; 2131 2132 for_each_possible_cpu(cpu) { 2133 /* copy maps belonging to foreign traffic classes */ 2134 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { 2135 /* fill in the new device map from the old device map */ 2136 map = xmap_dereference(dev_maps->cpu_map[tci]); 2137 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2138 } 2139 2140 /* We need to explicitly update tci as prevous loop 2141 * could break out early if dev_maps is NULL. 2142 */ 2143 tci = cpu * num_tc + tc; 2144 2145 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { 2146 /* add queue to CPU maps */ 2147 int pos = 0; 2148 2149 map = xmap_dereference(new_dev_maps->cpu_map[tci]); 2150 while ((pos < map->len) && (map->queues[pos] != index)) 2151 pos++; 2152 2153 if (pos == map->len) 2154 map->queues[map->len++] = index; 2155 #ifdef CONFIG_NUMA 2156 if (numa_node_id == -2) 2157 numa_node_id = cpu_to_node(cpu); 2158 else if (numa_node_id != cpu_to_node(cpu)) 2159 numa_node_id = -1; 2160 #endif 2161 } else if (dev_maps) { 2162 /* fill in the new device map from the old device map */ 2163 map = xmap_dereference(dev_maps->cpu_map[tci]); 2164 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2165 } 2166 2167 /* copy maps belonging to foreign traffic classes */ 2168 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { 2169 /* fill in the new device map from the old device map */ 2170 map = xmap_dereference(dev_maps->cpu_map[tci]); 2171 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2172 } 2173 } 2174 2175 rcu_assign_pointer(dev->xps_maps, new_dev_maps); 2176 2177 /* Cleanup old maps */ 2178 if (!dev_maps) 2179 goto out_no_old_maps; 2180 2181 for_each_possible_cpu(cpu) { 2182 for (i = num_tc, tci = cpu * num_tc; i--; tci++) { 2183 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); 2184 map = xmap_dereference(dev_maps->cpu_map[tci]); 2185 if (map && map != new_map) 2186 kfree_rcu(map, rcu); 2187 } 2188 } 2189 2190 kfree_rcu(dev_maps, rcu); 2191 2192 out_no_old_maps: 2193 dev_maps = new_dev_maps; 2194 active = true; 2195 2196 out_no_new_maps: 2197 /* update Tx queue numa node */ 2198 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2199 (numa_node_id >= 0) ? numa_node_id : 2200 NUMA_NO_NODE); 2201 2202 if (!dev_maps) 2203 goto out_no_maps; 2204 2205 /* removes queue from unused CPUs */ 2206 for_each_possible_cpu(cpu) { 2207 for (i = tc, tci = cpu * num_tc; i--; tci++) 2208 active |= remove_xps_queue(dev_maps, tci, index); 2209 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) 2210 active |= remove_xps_queue(dev_maps, tci, index); 2211 for (i = num_tc - tc, tci++; --i; tci++) 2212 active |= remove_xps_queue(dev_maps, tci, index); 2213 } 2214 2215 /* free map if not active */ 2216 if (!active) { 2217 RCU_INIT_POINTER(dev->xps_maps, NULL); 2218 kfree_rcu(dev_maps, rcu); 2219 } 2220 2221 out_no_maps: 2222 mutex_unlock(&xps_map_mutex); 2223 2224 return 0; 2225 error: 2226 /* remove any maps that we added */ 2227 for_each_possible_cpu(cpu) { 2228 for (i = num_tc, tci = cpu * num_tc; i--; tci++) { 2229 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); 2230 map = dev_maps ? 2231 xmap_dereference(dev_maps->cpu_map[tci]) : 2232 NULL; 2233 if (new_map && new_map != map) 2234 kfree(new_map); 2235 } 2236 } 2237 2238 mutex_unlock(&xps_map_mutex); 2239 2240 kfree(new_dev_maps); 2241 return -ENOMEM; 2242 } 2243 EXPORT_SYMBOL(netif_set_xps_queue); 2244 2245 #endif 2246 void netdev_reset_tc(struct net_device *dev) 2247 { 2248 #ifdef CONFIG_XPS 2249 netif_reset_xps_queues_gt(dev, 0); 2250 #endif 2251 dev->num_tc = 0; 2252 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2253 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2254 } 2255 EXPORT_SYMBOL(netdev_reset_tc); 2256 2257 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2258 { 2259 if (tc >= dev->num_tc) 2260 return -EINVAL; 2261 2262 #ifdef CONFIG_XPS 2263 netif_reset_xps_queues(dev, offset, count); 2264 #endif 2265 dev->tc_to_txq[tc].count = count; 2266 dev->tc_to_txq[tc].offset = offset; 2267 return 0; 2268 } 2269 EXPORT_SYMBOL(netdev_set_tc_queue); 2270 2271 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2272 { 2273 if (num_tc > TC_MAX_QUEUE) 2274 return -EINVAL; 2275 2276 #ifdef CONFIG_XPS 2277 netif_reset_xps_queues_gt(dev, 0); 2278 #endif 2279 dev->num_tc = num_tc; 2280 return 0; 2281 } 2282 EXPORT_SYMBOL(netdev_set_num_tc); 2283 2284 /* 2285 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2286 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. 2287 */ 2288 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2289 { 2290 int rc; 2291 2292 if (txq < 1 || txq > dev->num_tx_queues) 2293 return -EINVAL; 2294 2295 if (dev->reg_state == NETREG_REGISTERED || 2296 dev->reg_state == NETREG_UNREGISTERING) { 2297 ASSERT_RTNL(); 2298 2299 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2300 txq); 2301 if (rc) 2302 return rc; 2303 2304 if (dev->num_tc) 2305 netif_setup_tc(dev, txq); 2306 2307 if (txq < dev->real_num_tx_queues) { 2308 qdisc_reset_all_tx_gt(dev, txq); 2309 #ifdef CONFIG_XPS 2310 netif_reset_xps_queues_gt(dev, txq); 2311 #endif 2312 } 2313 } 2314 2315 dev->real_num_tx_queues = txq; 2316 return 0; 2317 } 2318 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2319 2320 #ifdef CONFIG_SYSFS 2321 /** 2322 * netif_set_real_num_rx_queues - set actual number of RX queues used 2323 * @dev: Network device 2324 * @rxq: Actual number of RX queues 2325 * 2326 * This must be called either with the rtnl_lock held or before 2327 * registration of the net device. Returns 0 on success, or a 2328 * negative error code. If called before registration, it always 2329 * succeeds. 2330 */ 2331 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2332 { 2333 int rc; 2334 2335 if (rxq < 1 || rxq > dev->num_rx_queues) 2336 return -EINVAL; 2337 2338 if (dev->reg_state == NETREG_REGISTERED) { 2339 ASSERT_RTNL(); 2340 2341 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2342 rxq); 2343 if (rc) 2344 return rc; 2345 } 2346 2347 dev->real_num_rx_queues = rxq; 2348 return 0; 2349 } 2350 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2351 #endif 2352 2353 /** 2354 * netif_get_num_default_rss_queues - default number of RSS queues 2355 * 2356 * This routine should set an upper limit on the number of RSS queues 2357 * used by default by multiqueue devices. 2358 */ 2359 int netif_get_num_default_rss_queues(void) 2360 { 2361 return is_kdump_kernel() ? 2362 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2363 } 2364 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2365 2366 static void __netif_reschedule(struct Qdisc *q) 2367 { 2368 struct softnet_data *sd; 2369 unsigned long flags; 2370 2371 local_irq_save(flags); 2372 sd = this_cpu_ptr(&softnet_data); 2373 q->next_sched = NULL; 2374 *sd->output_queue_tailp = q; 2375 sd->output_queue_tailp = &q->next_sched; 2376 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2377 local_irq_restore(flags); 2378 } 2379 2380 void __netif_schedule(struct Qdisc *q) 2381 { 2382 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2383 __netif_reschedule(q); 2384 } 2385 EXPORT_SYMBOL(__netif_schedule); 2386 2387 struct dev_kfree_skb_cb { 2388 enum skb_free_reason reason; 2389 }; 2390 2391 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2392 { 2393 return (struct dev_kfree_skb_cb *)skb->cb; 2394 } 2395 2396 void netif_schedule_queue(struct netdev_queue *txq) 2397 { 2398 rcu_read_lock(); 2399 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2400 struct Qdisc *q = rcu_dereference(txq->qdisc); 2401 2402 __netif_schedule(q); 2403 } 2404 rcu_read_unlock(); 2405 } 2406 EXPORT_SYMBOL(netif_schedule_queue); 2407 2408 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2409 { 2410 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2411 struct Qdisc *q; 2412 2413 rcu_read_lock(); 2414 q = rcu_dereference(dev_queue->qdisc); 2415 __netif_schedule(q); 2416 rcu_read_unlock(); 2417 } 2418 } 2419 EXPORT_SYMBOL(netif_tx_wake_queue); 2420 2421 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2422 { 2423 unsigned long flags; 2424 2425 if (likely(atomic_read(&skb->users) == 1)) { 2426 smp_rmb(); 2427 atomic_set(&skb->users, 0); 2428 } else if (likely(!atomic_dec_and_test(&skb->users))) { 2429 return; 2430 } 2431 get_kfree_skb_cb(skb)->reason = reason; 2432 local_irq_save(flags); 2433 skb->next = __this_cpu_read(softnet_data.completion_queue); 2434 __this_cpu_write(softnet_data.completion_queue, skb); 2435 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2436 local_irq_restore(flags); 2437 } 2438 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2439 2440 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2441 { 2442 if (in_irq() || irqs_disabled()) 2443 __dev_kfree_skb_irq(skb, reason); 2444 else 2445 dev_kfree_skb(skb); 2446 } 2447 EXPORT_SYMBOL(__dev_kfree_skb_any); 2448 2449 2450 /** 2451 * netif_device_detach - mark device as removed 2452 * @dev: network device 2453 * 2454 * Mark device as removed from system and therefore no longer available. 2455 */ 2456 void netif_device_detach(struct net_device *dev) 2457 { 2458 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2459 netif_running(dev)) { 2460 netif_tx_stop_all_queues(dev); 2461 } 2462 } 2463 EXPORT_SYMBOL(netif_device_detach); 2464 2465 /** 2466 * netif_device_attach - mark device as attached 2467 * @dev: network device 2468 * 2469 * Mark device as attached from system and restart if needed. 2470 */ 2471 void netif_device_attach(struct net_device *dev) 2472 { 2473 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2474 netif_running(dev)) { 2475 netif_tx_wake_all_queues(dev); 2476 __netdev_watchdog_up(dev); 2477 } 2478 } 2479 EXPORT_SYMBOL(netif_device_attach); 2480 2481 /* 2482 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 2483 * to be used as a distribution range. 2484 */ 2485 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 2486 unsigned int num_tx_queues) 2487 { 2488 u32 hash; 2489 u16 qoffset = 0; 2490 u16 qcount = num_tx_queues; 2491 2492 if (skb_rx_queue_recorded(skb)) { 2493 hash = skb_get_rx_queue(skb); 2494 while (unlikely(hash >= num_tx_queues)) 2495 hash -= num_tx_queues; 2496 return hash; 2497 } 2498 2499 if (dev->num_tc) { 2500 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 2501 2502 qoffset = dev->tc_to_txq[tc].offset; 2503 qcount = dev->tc_to_txq[tc].count; 2504 } 2505 2506 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 2507 } 2508 EXPORT_SYMBOL(__skb_tx_hash); 2509 2510 static void skb_warn_bad_offload(const struct sk_buff *skb) 2511 { 2512 static const netdev_features_t null_features; 2513 struct net_device *dev = skb->dev; 2514 const char *name = ""; 2515 2516 if (!net_ratelimit()) 2517 return; 2518 2519 if (dev) { 2520 if (dev->dev.parent) 2521 name = dev_driver_string(dev->dev.parent); 2522 else 2523 name = netdev_name(dev); 2524 } 2525 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2526 "gso_type=%d ip_summed=%d\n", 2527 name, dev ? &dev->features : &null_features, 2528 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2529 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2530 skb_shinfo(skb)->gso_type, skb->ip_summed); 2531 } 2532 2533 /* 2534 * Invalidate hardware checksum when packet is to be mangled, and 2535 * complete checksum manually on outgoing path. 2536 */ 2537 int skb_checksum_help(struct sk_buff *skb) 2538 { 2539 __wsum csum; 2540 int ret = 0, offset; 2541 2542 if (skb->ip_summed == CHECKSUM_COMPLETE) 2543 goto out_set_summed; 2544 2545 if (unlikely(skb_shinfo(skb)->gso_size)) { 2546 skb_warn_bad_offload(skb); 2547 return -EINVAL; 2548 } 2549 2550 /* Before computing a checksum, we should make sure no frag could 2551 * be modified by an external entity : checksum could be wrong. 2552 */ 2553 if (skb_has_shared_frag(skb)) { 2554 ret = __skb_linearize(skb); 2555 if (ret) 2556 goto out; 2557 } 2558 2559 offset = skb_checksum_start_offset(skb); 2560 BUG_ON(offset >= skb_headlen(skb)); 2561 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2562 2563 offset += skb->csum_offset; 2564 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2565 2566 if (skb_cloned(skb) && 2567 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2568 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2569 if (ret) 2570 goto out; 2571 } 2572 2573 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 2574 out_set_summed: 2575 skb->ip_summed = CHECKSUM_NONE; 2576 out: 2577 return ret; 2578 } 2579 EXPORT_SYMBOL(skb_checksum_help); 2580 2581 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2582 { 2583 __be16 type = skb->protocol; 2584 2585 /* Tunnel gso handlers can set protocol to ethernet. */ 2586 if (type == htons(ETH_P_TEB)) { 2587 struct ethhdr *eth; 2588 2589 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2590 return 0; 2591 2592 eth = (struct ethhdr *)skb_mac_header(skb); 2593 type = eth->h_proto; 2594 } 2595 2596 return __vlan_get_protocol(skb, type, depth); 2597 } 2598 2599 /** 2600 * skb_mac_gso_segment - mac layer segmentation handler. 2601 * @skb: buffer to segment 2602 * @features: features for the output path (see dev->features) 2603 */ 2604 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 2605 netdev_features_t features) 2606 { 2607 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 2608 struct packet_offload *ptype; 2609 int vlan_depth = skb->mac_len; 2610 __be16 type = skb_network_protocol(skb, &vlan_depth); 2611 2612 if (unlikely(!type)) 2613 return ERR_PTR(-EINVAL); 2614 2615 __skb_pull(skb, vlan_depth); 2616 2617 rcu_read_lock(); 2618 list_for_each_entry_rcu(ptype, &offload_base, list) { 2619 if (ptype->type == type && ptype->callbacks.gso_segment) { 2620 segs = ptype->callbacks.gso_segment(skb, features); 2621 break; 2622 } 2623 } 2624 rcu_read_unlock(); 2625 2626 __skb_push(skb, skb->data - skb_mac_header(skb)); 2627 2628 return segs; 2629 } 2630 EXPORT_SYMBOL(skb_mac_gso_segment); 2631 2632 2633 /* openvswitch calls this on rx path, so we need a different check. 2634 */ 2635 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 2636 { 2637 if (tx_path) 2638 return skb->ip_summed != CHECKSUM_PARTIAL && 2639 skb->ip_summed != CHECKSUM_NONE; 2640 2641 return skb->ip_summed == CHECKSUM_NONE; 2642 } 2643 2644 /** 2645 * __skb_gso_segment - Perform segmentation on skb. 2646 * @skb: buffer to segment 2647 * @features: features for the output path (see dev->features) 2648 * @tx_path: whether it is called in TX path 2649 * 2650 * This function segments the given skb and returns a list of segments. 2651 * 2652 * It may return NULL if the skb requires no segmentation. This is 2653 * only possible when GSO is used for verifying header integrity. 2654 * 2655 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. 2656 */ 2657 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 2658 netdev_features_t features, bool tx_path) 2659 { 2660 struct sk_buff *segs; 2661 2662 if (unlikely(skb_needs_check(skb, tx_path))) { 2663 int err; 2664 2665 /* We're going to init ->check field in TCP or UDP header */ 2666 err = skb_cow_head(skb, 0); 2667 if (err < 0) 2668 return ERR_PTR(err); 2669 } 2670 2671 /* Only report GSO partial support if it will enable us to 2672 * support segmentation on this frame without needing additional 2673 * work. 2674 */ 2675 if (features & NETIF_F_GSO_PARTIAL) { 2676 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 2677 struct net_device *dev = skb->dev; 2678 2679 partial_features |= dev->features & dev->gso_partial_features; 2680 if (!skb_gso_ok(skb, features | partial_features)) 2681 features &= ~NETIF_F_GSO_PARTIAL; 2682 } 2683 2684 BUILD_BUG_ON(SKB_SGO_CB_OFFSET + 2685 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 2686 2687 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 2688 SKB_GSO_CB(skb)->encap_level = 0; 2689 2690 skb_reset_mac_header(skb); 2691 skb_reset_mac_len(skb); 2692 2693 segs = skb_mac_gso_segment(skb, features); 2694 2695 if (unlikely(skb_needs_check(skb, tx_path))) 2696 skb_warn_bad_offload(skb); 2697 2698 return segs; 2699 } 2700 EXPORT_SYMBOL(__skb_gso_segment); 2701 2702 /* Take action when hardware reception checksum errors are detected. */ 2703 #ifdef CONFIG_BUG 2704 void netdev_rx_csum_fault(struct net_device *dev) 2705 { 2706 if (net_ratelimit()) { 2707 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 2708 dump_stack(); 2709 } 2710 } 2711 EXPORT_SYMBOL(netdev_rx_csum_fault); 2712 #endif 2713 2714 /* Actually, we should eliminate this check as soon as we know, that: 2715 * 1. IOMMU is present and allows to map all the memory. 2716 * 2. No high memory really exists on this machine. 2717 */ 2718 2719 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 2720 { 2721 #ifdef CONFIG_HIGHMEM 2722 int i; 2723 2724 if (!(dev->features & NETIF_F_HIGHDMA)) { 2725 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2726 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2727 2728 if (PageHighMem(skb_frag_page(frag))) 2729 return 1; 2730 } 2731 } 2732 2733 if (PCI_DMA_BUS_IS_PHYS) { 2734 struct device *pdev = dev->dev.parent; 2735 2736 if (!pdev) 2737 return 0; 2738 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2739 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2740 dma_addr_t addr = page_to_phys(skb_frag_page(frag)); 2741 2742 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) 2743 return 1; 2744 } 2745 } 2746 #endif 2747 return 0; 2748 } 2749 2750 /* If MPLS offload request, verify we are testing hardware MPLS features 2751 * instead of standard features for the netdev. 2752 */ 2753 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 2754 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2755 netdev_features_t features, 2756 __be16 type) 2757 { 2758 if (eth_p_mpls(type)) 2759 features &= skb->dev->mpls_features; 2760 2761 return features; 2762 } 2763 #else 2764 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2765 netdev_features_t features, 2766 __be16 type) 2767 { 2768 return features; 2769 } 2770 #endif 2771 2772 static netdev_features_t harmonize_features(struct sk_buff *skb, 2773 netdev_features_t features) 2774 { 2775 int tmp; 2776 __be16 type; 2777 2778 type = skb_network_protocol(skb, &tmp); 2779 features = net_mpls_features(skb, features, type); 2780 2781 if (skb->ip_summed != CHECKSUM_NONE && 2782 !can_checksum_protocol(features, type)) { 2783 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 2784 } 2785 if (illegal_highdma(skb->dev, skb)) 2786 features &= ~NETIF_F_SG; 2787 2788 return features; 2789 } 2790 2791 netdev_features_t passthru_features_check(struct sk_buff *skb, 2792 struct net_device *dev, 2793 netdev_features_t features) 2794 { 2795 return features; 2796 } 2797 EXPORT_SYMBOL(passthru_features_check); 2798 2799 static netdev_features_t dflt_features_check(const struct sk_buff *skb, 2800 struct net_device *dev, 2801 netdev_features_t features) 2802 { 2803 return vlan_features_check(skb, features); 2804 } 2805 2806 static netdev_features_t gso_features_check(const struct sk_buff *skb, 2807 struct net_device *dev, 2808 netdev_features_t features) 2809 { 2810 u16 gso_segs = skb_shinfo(skb)->gso_segs; 2811 2812 if (gso_segs > dev->gso_max_segs) 2813 return features & ~NETIF_F_GSO_MASK; 2814 2815 /* Support for GSO partial features requires software 2816 * intervention before we can actually process the packets 2817 * so we need to strip support for any partial features now 2818 * and we can pull them back in after we have partially 2819 * segmented the frame. 2820 */ 2821 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 2822 features &= ~dev->gso_partial_features; 2823 2824 /* Make sure to clear the IPv4 ID mangling feature if the 2825 * IPv4 header has the potential to be fragmented. 2826 */ 2827 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 2828 struct iphdr *iph = skb->encapsulation ? 2829 inner_ip_hdr(skb) : ip_hdr(skb); 2830 2831 if (!(iph->frag_off & htons(IP_DF))) 2832 features &= ~NETIF_F_TSO_MANGLEID; 2833 } 2834 2835 return features; 2836 } 2837 2838 netdev_features_t netif_skb_features(struct sk_buff *skb) 2839 { 2840 struct net_device *dev = skb->dev; 2841 netdev_features_t features = dev->features; 2842 2843 if (skb_is_gso(skb)) 2844 features = gso_features_check(skb, dev, features); 2845 2846 /* If encapsulation offload request, verify we are testing 2847 * hardware encapsulation features instead of standard 2848 * features for the netdev 2849 */ 2850 if (skb->encapsulation) 2851 features &= dev->hw_enc_features; 2852 2853 if (skb_vlan_tagged(skb)) 2854 features = netdev_intersect_features(features, 2855 dev->vlan_features | 2856 NETIF_F_HW_VLAN_CTAG_TX | 2857 NETIF_F_HW_VLAN_STAG_TX); 2858 2859 if (dev->netdev_ops->ndo_features_check) 2860 features &= dev->netdev_ops->ndo_features_check(skb, dev, 2861 features); 2862 else 2863 features &= dflt_features_check(skb, dev, features); 2864 2865 return harmonize_features(skb, features); 2866 } 2867 EXPORT_SYMBOL(netif_skb_features); 2868 2869 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 2870 struct netdev_queue *txq, bool more) 2871 { 2872 unsigned int len; 2873 int rc; 2874 2875 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) 2876 dev_queue_xmit_nit(skb, dev); 2877 2878 len = skb->len; 2879 trace_net_dev_start_xmit(skb, dev); 2880 rc = netdev_start_xmit(skb, dev, txq, more); 2881 trace_net_dev_xmit(skb, rc, dev, len); 2882 2883 return rc; 2884 } 2885 2886 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 2887 struct netdev_queue *txq, int *ret) 2888 { 2889 struct sk_buff *skb = first; 2890 int rc = NETDEV_TX_OK; 2891 2892 while (skb) { 2893 struct sk_buff *next = skb->next; 2894 2895 skb->next = NULL; 2896 rc = xmit_one(skb, dev, txq, next != NULL); 2897 if (unlikely(!dev_xmit_complete(rc))) { 2898 skb->next = next; 2899 goto out; 2900 } 2901 2902 skb = next; 2903 if (netif_xmit_stopped(txq) && skb) { 2904 rc = NETDEV_TX_BUSY; 2905 break; 2906 } 2907 } 2908 2909 out: 2910 *ret = rc; 2911 return skb; 2912 } 2913 2914 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 2915 netdev_features_t features) 2916 { 2917 if (skb_vlan_tag_present(skb) && 2918 !vlan_hw_offload_capable(features, skb->vlan_proto)) 2919 skb = __vlan_hwaccel_push_inside(skb); 2920 return skb; 2921 } 2922 2923 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) 2924 { 2925 netdev_features_t features; 2926 2927 features = netif_skb_features(skb); 2928 skb = validate_xmit_vlan(skb, features); 2929 if (unlikely(!skb)) 2930 goto out_null; 2931 2932 if (netif_needs_gso(skb, features)) { 2933 struct sk_buff *segs; 2934 2935 segs = skb_gso_segment(skb, features); 2936 if (IS_ERR(segs)) { 2937 goto out_kfree_skb; 2938 } else if (segs) { 2939 consume_skb(skb); 2940 skb = segs; 2941 } 2942 } else { 2943 if (skb_needs_linearize(skb, features) && 2944 __skb_linearize(skb)) 2945 goto out_kfree_skb; 2946 2947 /* If packet is not checksummed and device does not 2948 * support checksumming for this protocol, complete 2949 * checksumming here. 2950 */ 2951 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2952 if (skb->encapsulation) 2953 skb_set_inner_transport_header(skb, 2954 skb_checksum_start_offset(skb)); 2955 else 2956 skb_set_transport_header(skb, 2957 skb_checksum_start_offset(skb)); 2958 if (!(features & NETIF_F_CSUM_MASK) && 2959 skb_checksum_help(skb)) 2960 goto out_kfree_skb; 2961 } 2962 } 2963 2964 return skb; 2965 2966 out_kfree_skb: 2967 kfree_skb(skb); 2968 out_null: 2969 atomic_long_inc(&dev->tx_dropped); 2970 return NULL; 2971 } 2972 2973 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) 2974 { 2975 struct sk_buff *next, *head = NULL, *tail; 2976 2977 for (; skb != NULL; skb = next) { 2978 next = skb->next; 2979 skb->next = NULL; 2980 2981 /* in case skb wont be segmented, point to itself */ 2982 skb->prev = skb; 2983 2984 skb = validate_xmit_skb(skb, dev); 2985 if (!skb) 2986 continue; 2987 2988 if (!head) 2989 head = skb; 2990 else 2991 tail->next = skb; 2992 /* If skb was segmented, skb->prev points to 2993 * the last segment. If not, it still contains skb. 2994 */ 2995 tail = skb->prev; 2996 } 2997 return head; 2998 } 2999 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3000 3001 static void qdisc_pkt_len_init(struct sk_buff *skb) 3002 { 3003 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3004 3005 qdisc_skb_cb(skb)->pkt_len = skb->len; 3006 3007 /* To get more precise estimation of bytes sent on wire, 3008 * we add to pkt_len the headers size of all segments 3009 */ 3010 if (shinfo->gso_size) { 3011 unsigned int hdr_len; 3012 u16 gso_segs = shinfo->gso_segs; 3013 3014 /* mac layer + network layer */ 3015 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3016 3017 /* + transport layer */ 3018 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 3019 hdr_len += tcp_hdrlen(skb); 3020 else 3021 hdr_len += sizeof(struct udphdr); 3022 3023 if (shinfo->gso_type & SKB_GSO_DODGY) 3024 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3025 shinfo->gso_size); 3026 3027 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3028 } 3029 } 3030 3031 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3032 struct net_device *dev, 3033 struct netdev_queue *txq) 3034 { 3035 spinlock_t *root_lock = qdisc_lock(q); 3036 struct sk_buff *to_free = NULL; 3037 bool contended; 3038 int rc; 3039 3040 qdisc_calculate_pkt_len(skb, q); 3041 /* 3042 * Heuristic to force contended enqueues to serialize on a 3043 * separate lock before trying to get qdisc main lock. 3044 * This permits qdisc->running owner to get the lock more 3045 * often and dequeue packets faster. 3046 */ 3047 contended = qdisc_is_running(q); 3048 if (unlikely(contended)) 3049 spin_lock(&q->busylock); 3050 3051 spin_lock(root_lock); 3052 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3053 __qdisc_drop(skb, &to_free); 3054 rc = NET_XMIT_DROP; 3055 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3056 qdisc_run_begin(q)) { 3057 /* 3058 * This is a work-conserving queue; there are no old skbs 3059 * waiting to be sent out; and the qdisc is not running - 3060 * xmit the skb directly. 3061 */ 3062 3063 qdisc_bstats_update(q, skb); 3064 3065 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3066 if (unlikely(contended)) { 3067 spin_unlock(&q->busylock); 3068 contended = false; 3069 } 3070 __qdisc_run(q); 3071 } else 3072 qdisc_run_end(q); 3073 3074 rc = NET_XMIT_SUCCESS; 3075 } else { 3076 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3077 if (qdisc_run_begin(q)) { 3078 if (unlikely(contended)) { 3079 spin_unlock(&q->busylock); 3080 contended = false; 3081 } 3082 __qdisc_run(q); 3083 } 3084 } 3085 spin_unlock(root_lock); 3086 if (unlikely(to_free)) 3087 kfree_skb_list(to_free); 3088 if (unlikely(contended)) 3089 spin_unlock(&q->busylock); 3090 return rc; 3091 } 3092 3093 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3094 static void skb_update_prio(struct sk_buff *skb) 3095 { 3096 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); 3097 3098 if (!skb->priority && skb->sk && map) { 3099 unsigned int prioidx = 3100 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); 3101 3102 if (prioidx < map->priomap_len) 3103 skb->priority = map->priomap[prioidx]; 3104 } 3105 } 3106 #else 3107 #define skb_update_prio(skb) 3108 #endif 3109 3110 DEFINE_PER_CPU(int, xmit_recursion); 3111 EXPORT_SYMBOL(xmit_recursion); 3112 3113 /** 3114 * dev_loopback_xmit - loop back @skb 3115 * @net: network namespace this loopback is happening in 3116 * @sk: sk needed to be a netfilter okfn 3117 * @skb: buffer to transmit 3118 */ 3119 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3120 { 3121 skb_reset_mac_header(skb); 3122 __skb_pull(skb, skb_network_offset(skb)); 3123 skb->pkt_type = PACKET_LOOPBACK; 3124 skb->ip_summed = CHECKSUM_UNNECESSARY; 3125 WARN_ON(!skb_dst(skb)); 3126 skb_dst_force(skb); 3127 netif_rx_ni(skb); 3128 return 0; 3129 } 3130 EXPORT_SYMBOL(dev_loopback_xmit); 3131 3132 #ifdef CONFIG_NET_EGRESS 3133 static struct sk_buff * 3134 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3135 { 3136 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); 3137 struct tcf_result cl_res; 3138 3139 if (!cl) 3140 return skb; 3141 3142 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3143 qdisc_bstats_cpu_update(cl->q, skb); 3144 3145 switch (tc_classify(skb, cl, &cl_res, false)) { 3146 case TC_ACT_OK: 3147 case TC_ACT_RECLASSIFY: 3148 skb->tc_index = TC_H_MIN(cl_res.classid); 3149 break; 3150 case TC_ACT_SHOT: 3151 qdisc_qstats_cpu_drop(cl->q); 3152 *ret = NET_XMIT_DROP; 3153 kfree_skb(skb); 3154 return NULL; 3155 case TC_ACT_STOLEN: 3156 case TC_ACT_QUEUED: 3157 *ret = NET_XMIT_SUCCESS; 3158 consume_skb(skb); 3159 return NULL; 3160 case TC_ACT_REDIRECT: 3161 /* No need to push/pop skb's mac_header here on egress! */ 3162 skb_do_redirect(skb); 3163 *ret = NET_XMIT_SUCCESS; 3164 return NULL; 3165 default: 3166 break; 3167 } 3168 3169 return skb; 3170 } 3171 #endif /* CONFIG_NET_EGRESS */ 3172 3173 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) 3174 { 3175 #ifdef CONFIG_XPS 3176 struct xps_dev_maps *dev_maps; 3177 struct xps_map *map; 3178 int queue_index = -1; 3179 3180 rcu_read_lock(); 3181 dev_maps = rcu_dereference(dev->xps_maps); 3182 if (dev_maps) { 3183 unsigned int tci = skb->sender_cpu - 1; 3184 3185 if (dev->num_tc) { 3186 tci *= dev->num_tc; 3187 tci += netdev_get_prio_tc_map(dev, skb->priority); 3188 } 3189 3190 map = rcu_dereference(dev_maps->cpu_map[tci]); 3191 if (map) { 3192 if (map->len == 1) 3193 queue_index = map->queues[0]; 3194 else 3195 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), 3196 map->len)]; 3197 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3198 queue_index = -1; 3199 } 3200 } 3201 rcu_read_unlock(); 3202 3203 return queue_index; 3204 #else 3205 return -1; 3206 #endif 3207 } 3208 3209 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) 3210 { 3211 struct sock *sk = skb->sk; 3212 int queue_index = sk_tx_queue_get(sk); 3213 3214 if (queue_index < 0 || skb->ooo_okay || 3215 queue_index >= dev->real_num_tx_queues) { 3216 int new_index = get_xps_queue(dev, skb); 3217 3218 if (new_index < 0) 3219 new_index = skb_tx_hash(dev, skb); 3220 3221 if (queue_index != new_index && sk && 3222 sk_fullsock(sk) && 3223 rcu_access_pointer(sk->sk_dst_cache)) 3224 sk_tx_queue_set(sk, new_index); 3225 3226 queue_index = new_index; 3227 } 3228 3229 return queue_index; 3230 } 3231 3232 struct netdev_queue *netdev_pick_tx(struct net_device *dev, 3233 struct sk_buff *skb, 3234 void *accel_priv) 3235 { 3236 int queue_index = 0; 3237 3238 #ifdef CONFIG_XPS 3239 u32 sender_cpu = skb->sender_cpu - 1; 3240 3241 if (sender_cpu >= (u32)NR_CPUS) 3242 skb->sender_cpu = raw_smp_processor_id() + 1; 3243 #endif 3244 3245 if (dev->real_num_tx_queues != 1) { 3246 const struct net_device_ops *ops = dev->netdev_ops; 3247 3248 if (ops->ndo_select_queue) 3249 queue_index = ops->ndo_select_queue(dev, skb, accel_priv, 3250 __netdev_pick_tx); 3251 else 3252 queue_index = __netdev_pick_tx(dev, skb); 3253 3254 if (!accel_priv) 3255 queue_index = netdev_cap_txqueue(dev, queue_index); 3256 } 3257 3258 skb_set_queue_mapping(skb, queue_index); 3259 return netdev_get_tx_queue(dev, queue_index); 3260 } 3261 3262 /** 3263 * __dev_queue_xmit - transmit a buffer 3264 * @skb: buffer to transmit 3265 * @accel_priv: private data used for L2 forwarding offload 3266 * 3267 * Queue a buffer for transmission to a network device. The caller must 3268 * have set the device and priority and built the buffer before calling 3269 * this function. The function can be called from an interrupt. 3270 * 3271 * A negative errno code is returned on a failure. A success does not 3272 * guarantee the frame will be transmitted as it may be dropped due 3273 * to congestion or traffic shaping. 3274 * 3275 * ----------------------------------------------------------------------------------- 3276 * I notice this method can also return errors from the queue disciplines, 3277 * including NET_XMIT_DROP, which is a positive value. So, errors can also 3278 * be positive. 3279 * 3280 * Regardless of the return value, the skb is consumed, so it is currently 3281 * difficult to retry a send to this method. (You can bump the ref count 3282 * before sending to hold a reference for retry if you are careful.) 3283 * 3284 * When calling this method, interrupts MUST be enabled. This is because 3285 * the BH enable code must have IRQs enabled so that it will not deadlock. 3286 * --BLG 3287 */ 3288 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) 3289 { 3290 struct net_device *dev = skb->dev; 3291 struct netdev_queue *txq; 3292 struct Qdisc *q; 3293 int rc = -ENOMEM; 3294 3295 skb_reset_mac_header(skb); 3296 3297 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 3298 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 3299 3300 /* Disable soft irqs for various locks below. Also 3301 * stops preemption for RCU. 3302 */ 3303 rcu_read_lock_bh(); 3304 3305 skb_update_prio(skb); 3306 3307 qdisc_pkt_len_init(skb); 3308 #ifdef CONFIG_NET_CLS_ACT 3309 skb->tc_at_ingress = 0; 3310 # ifdef CONFIG_NET_EGRESS 3311 if (static_key_false(&egress_needed)) { 3312 skb = sch_handle_egress(skb, &rc, dev); 3313 if (!skb) 3314 goto out; 3315 } 3316 # endif 3317 #endif 3318 /* If device/qdisc don't need skb->dst, release it right now while 3319 * its hot in this cpu cache. 3320 */ 3321 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 3322 skb_dst_drop(skb); 3323 else 3324 skb_dst_force(skb); 3325 3326 txq = netdev_pick_tx(dev, skb, accel_priv); 3327 q = rcu_dereference_bh(txq->qdisc); 3328 3329 trace_net_dev_queue(skb); 3330 if (q->enqueue) { 3331 rc = __dev_xmit_skb(skb, q, dev, txq); 3332 goto out; 3333 } 3334 3335 /* The device has no queue. Common case for software devices: 3336 * loopback, all the sorts of tunnels... 3337 3338 * Really, it is unlikely that netif_tx_lock protection is necessary 3339 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 3340 * counters.) 3341 * However, it is possible, that they rely on protection 3342 * made by us here. 3343 3344 * Check this and shot the lock. It is not prone from deadlocks. 3345 *Either shot noqueue qdisc, it is even simpler 8) 3346 */ 3347 if (dev->flags & IFF_UP) { 3348 int cpu = smp_processor_id(); /* ok because BHs are off */ 3349 3350 if (txq->xmit_lock_owner != cpu) { 3351 if (unlikely(__this_cpu_read(xmit_recursion) > 3352 XMIT_RECURSION_LIMIT)) 3353 goto recursion_alert; 3354 3355 skb = validate_xmit_skb(skb, dev); 3356 if (!skb) 3357 goto out; 3358 3359 HARD_TX_LOCK(dev, txq, cpu); 3360 3361 if (!netif_xmit_stopped(txq)) { 3362 __this_cpu_inc(xmit_recursion); 3363 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 3364 __this_cpu_dec(xmit_recursion); 3365 if (dev_xmit_complete(rc)) { 3366 HARD_TX_UNLOCK(dev, txq); 3367 goto out; 3368 } 3369 } 3370 HARD_TX_UNLOCK(dev, txq); 3371 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 3372 dev->name); 3373 } else { 3374 /* Recursion is detected! It is possible, 3375 * unfortunately 3376 */ 3377 recursion_alert: 3378 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 3379 dev->name); 3380 } 3381 } 3382 3383 rc = -ENETDOWN; 3384 rcu_read_unlock_bh(); 3385 3386 atomic_long_inc(&dev->tx_dropped); 3387 kfree_skb_list(skb); 3388 return rc; 3389 out: 3390 rcu_read_unlock_bh(); 3391 return rc; 3392 } 3393 3394 int dev_queue_xmit(struct sk_buff *skb) 3395 { 3396 return __dev_queue_xmit(skb, NULL); 3397 } 3398 EXPORT_SYMBOL(dev_queue_xmit); 3399 3400 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) 3401 { 3402 return __dev_queue_xmit(skb, accel_priv); 3403 } 3404 EXPORT_SYMBOL(dev_queue_xmit_accel); 3405 3406 3407 /************************************************************************* 3408 * Receiver routines 3409 *************************************************************************/ 3410 3411 int netdev_max_backlog __read_mostly = 1000; 3412 EXPORT_SYMBOL(netdev_max_backlog); 3413 3414 int netdev_tstamp_prequeue __read_mostly = 1; 3415 int netdev_budget __read_mostly = 300; 3416 int weight_p __read_mostly = 64; /* old backlog weight */ 3417 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 3418 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 3419 int dev_rx_weight __read_mostly = 64; 3420 int dev_tx_weight __read_mostly = 64; 3421 3422 /* Called with irq disabled */ 3423 static inline void ____napi_schedule(struct softnet_data *sd, 3424 struct napi_struct *napi) 3425 { 3426 list_add_tail(&napi->poll_list, &sd->poll_list); 3427 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3428 } 3429 3430 #ifdef CONFIG_RPS 3431 3432 /* One global table that all flow-based protocols share. */ 3433 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3434 EXPORT_SYMBOL(rps_sock_flow_table); 3435 u32 rps_cpu_mask __read_mostly; 3436 EXPORT_SYMBOL(rps_cpu_mask); 3437 3438 struct static_key rps_needed __read_mostly; 3439 EXPORT_SYMBOL(rps_needed); 3440 struct static_key rfs_needed __read_mostly; 3441 EXPORT_SYMBOL(rfs_needed); 3442 3443 static struct rps_dev_flow * 3444 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3445 struct rps_dev_flow *rflow, u16 next_cpu) 3446 { 3447 if (next_cpu < nr_cpu_ids) { 3448 #ifdef CONFIG_RFS_ACCEL 3449 struct netdev_rx_queue *rxqueue; 3450 struct rps_dev_flow_table *flow_table; 3451 struct rps_dev_flow *old_rflow; 3452 u32 flow_id; 3453 u16 rxq_index; 3454 int rc; 3455 3456 /* Should we steer this flow to a different hardware queue? */ 3457 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3458 !(dev->features & NETIF_F_NTUPLE)) 3459 goto out; 3460 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 3461 if (rxq_index == skb_get_rx_queue(skb)) 3462 goto out; 3463 3464 rxqueue = dev->_rx + rxq_index; 3465 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3466 if (!flow_table) 3467 goto out; 3468 flow_id = skb_get_hash(skb) & flow_table->mask; 3469 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 3470 rxq_index, flow_id); 3471 if (rc < 0) 3472 goto out; 3473 old_rflow = rflow; 3474 rflow = &flow_table->flows[flow_id]; 3475 rflow->filter = rc; 3476 if (old_rflow->filter == rflow->filter) 3477 old_rflow->filter = RPS_NO_FILTER; 3478 out: 3479 #endif 3480 rflow->last_qtail = 3481 per_cpu(softnet_data, next_cpu).input_queue_head; 3482 } 3483 3484 rflow->cpu = next_cpu; 3485 return rflow; 3486 } 3487 3488 /* 3489 * get_rps_cpu is called from netif_receive_skb and returns the target 3490 * CPU from the RPS map of the receiving queue for a given skb. 3491 * rcu_read_lock must be held on entry. 3492 */ 3493 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3494 struct rps_dev_flow **rflowp) 3495 { 3496 const struct rps_sock_flow_table *sock_flow_table; 3497 struct netdev_rx_queue *rxqueue = dev->_rx; 3498 struct rps_dev_flow_table *flow_table; 3499 struct rps_map *map; 3500 int cpu = -1; 3501 u32 tcpu; 3502 u32 hash; 3503 3504 if (skb_rx_queue_recorded(skb)) { 3505 u16 index = skb_get_rx_queue(skb); 3506 3507 if (unlikely(index >= dev->real_num_rx_queues)) { 3508 WARN_ONCE(dev->real_num_rx_queues > 1, 3509 "%s received packet on queue %u, but number " 3510 "of RX queues is %u\n", 3511 dev->name, index, dev->real_num_rx_queues); 3512 goto done; 3513 } 3514 rxqueue += index; 3515 } 3516 3517 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 3518 3519 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3520 map = rcu_dereference(rxqueue->rps_map); 3521 if (!flow_table && !map) 3522 goto done; 3523 3524 skb_reset_network_header(skb); 3525 hash = skb_get_hash(skb); 3526 if (!hash) 3527 goto done; 3528 3529 sock_flow_table = rcu_dereference(rps_sock_flow_table); 3530 if (flow_table && sock_flow_table) { 3531 struct rps_dev_flow *rflow; 3532 u32 next_cpu; 3533 u32 ident; 3534 3535 /* First check into global flow table if there is a match */ 3536 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 3537 if ((ident ^ hash) & ~rps_cpu_mask) 3538 goto try_rps; 3539 3540 next_cpu = ident & rps_cpu_mask; 3541 3542 /* OK, now we know there is a match, 3543 * we can look at the local (per receive queue) flow table 3544 */ 3545 rflow = &flow_table->flows[hash & flow_table->mask]; 3546 tcpu = rflow->cpu; 3547 3548 /* 3549 * If the desired CPU (where last recvmsg was done) is 3550 * different from current CPU (one in the rx-queue flow 3551 * table entry), switch if one of the following holds: 3552 * - Current CPU is unset (>= nr_cpu_ids). 3553 * - Current CPU is offline. 3554 * - The current CPU's queue tail has advanced beyond the 3555 * last packet that was enqueued using this table entry. 3556 * This guarantees that all previous packets for the flow 3557 * have been dequeued, thus preserving in order delivery. 3558 */ 3559 if (unlikely(tcpu != next_cpu) && 3560 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 3561 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 3562 rflow->last_qtail)) >= 0)) { 3563 tcpu = next_cpu; 3564 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 3565 } 3566 3567 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 3568 *rflowp = rflow; 3569 cpu = tcpu; 3570 goto done; 3571 } 3572 } 3573 3574 try_rps: 3575 3576 if (map) { 3577 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 3578 if (cpu_online(tcpu)) { 3579 cpu = tcpu; 3580 goto done; 3581 } 3582 } 3583 3584 done: 3585 return cpu; 3586 } 3587 3588 #ifdef CONFIG_RFS_ACCEL 3589 3590 /** 3591 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 3592 * @dev: Device on which the filter was set 3593 * @rxq_index: RX queue index 3594 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 3595 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 3596 * 3597 * Drivers that implement ndo_rx_flow_steer() should periodically call 3598 * this function for each installed filter and remove the filters for 3599 * which it returns %true. 3600 */ 3601 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 3602 u32 flow_id, u16 filter_id) 3603 { 3604 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 3605 struct rps_dev_flow_table *flow_table; 3606 struct rps_dev_flow *rflow; 3607 bool expire = true; 3608 unsigned int cpu; 3609 3610 rcu_read_lock(); 3611 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3612 if (flow_table && flow_id <= flow_table->mask) { 3613 rflow = &flow_table->flows[flow_id]; 3614 cpu = ACCESS_ONCE(rflow->cpu); 3615 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 3616 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 3617 rflow->last_qtail) < 3618 (int)(10 * flow_table->mask))) 3619 expire = false; 3620 } 3621 rcu_read_unlock(); 3622 return expire; 3623 } 3624 EXPORT_SYMBOL(rps_may_expire_flow); 3625 3626 #endif /* CONFIG_RFS_ACCEL */ 3627 3628 /* Called from hardirq (IPI) context */ 3629 static void rps_trigger_softirq(void *data) 3630 { 3631 struct softnet_data *sd = data; 3632 3633 ____napi_schedule(sd, &sd->backlog); 3634 sd->received_rps++; 3635 } 3636 3637 #endif /* CONFIG_RPS */ 3638 3639 /* 3640 * Check if this softnet_data structure is another cpu one 3641 * If yes, queue it to our IPI list and return 1 3642 * If no, return 0 3643 */ 3644 static int rps_ipi_queued(struct softnet_data *sd) 3645 { 3646 #ifdef CONFIG_RPS 3647 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 3648 3649 if (sd != mysd) { 3650 sd->rps_ipi_next = mysd->rps_ipi_list; 3651 mysd->rps_ipi_list = sd; 3652 3653 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3654 return 1; 3655 } 3656 #endif /* CONFIG_RPS */ 3657 return 0; 3658 } 3659 3660 #ifdef CONFIG_NET_FLOW_LIMIT 3661 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 3662 #endif 3663 3664 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 3665 { 3666 #ifdef CONFIG_NET_FLOW_LIMIT 3667 struct sd_flow_limit *fl; 3668 struct softnet_data *sd; 3669 unsigned int old_flow, new_flow; 3670 3671 if (qlen < (netdev_max_backlog >> 1)) 3672 return false; 3673 3674 sd = this_cpu_ptr(&softnet_data); 3675 3676 rcu_read_lock(); 3677 fl = rcu_dereference(sd->flow_limit); 3678 if (fl) { 3679 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 3680 old_flow = fl->history[fl->history_head]; 3681 fl->history[fl->history_head] = new_flow; 3682 3683 fl->history_head++; 3684 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 3685 3686 if (likely(fl->buckets[old_flow])) 3687 fl->buckets[old_flow]--; 3688 3689 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 3690 fl->count++; 3691 rcu_read_unlock(); 3692 return true; 3693 } 3694 } 3695 rcu_read_unlock(); 3696 #endif 3697 return false; 3698 } 3699 3700 /* 3701 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 3702 * queue (may be a remote CPU queue). 3703 */ 3704 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 3705 unsigned int *qtail) 3706 { 3707 struct softnet_data *sd; 3708 unsigned long flags; 3709 unsigned int qlen; 3710 3711 sd = &per_cpu(softnet_data, cpu); 3712 3713 local_irq_save(flags); 3714 3715 rps_lock(sd); 3716 if (!netif_running(skb->dev)) 3717 goto drop; 3718 qlen = skb_queue_len(&sd->input_pkt_queue); 3719 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 3720 if (qlen) { 3721 enqueue: 3722 __skb_queue_tail(&sd->input_pkt_queue, skb); 3723 input_queue_tail_incr_save(sd, qtail); 3724 rps_unlock(sd); 3725 local_irq_restore(flags); 3726 return NET_RX_SUCCESS; 3727 } 3728 3729 /* Schedule NAPI for backlog device 3730 * We can use non atomic operation since we own the queue lock 3731 */ 3732 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 3733 if (!rps_ipi_queued(sd)) 3734 ____napi_schedule(sd, &sd->backlog); 3735 } 3736 goto enqueue; 3737 } 3738 3739 drop: 3740 sd->dropped++; 3741 rps_unlock(sd); 3742 3743 local_irq_restore(flags); 3744 3745 atomic_long_inc(&skb->dev->rx_dropped); 3746 kfree_skb(skb); 3747 return NET_RX_DROP; 3748 } 3749 3750 static int netif_rx_internal(struct sk_buff *skb) 3751 { 3752 int ret; 3753 3754 net_timestamp_check(netdev_tstamp_prequeue, skb); 3755 3756 trace_netif_rx(skb); 3757 #ifdef CONFIG_RPS 3758 if (static_key_false(&rps_needed)) { 3759 struct rps_dev_flow voidflow, *rflow = &voidflow; 3760 int cpu; 3761 3762 preempt_disable(); 3763 rcu_read_lock(); 3764 3765 cpu = get_rps_cpu(skb->dev, skb, &rflow); 3766 if (cpu < 0) 3767 cpu = smp_processor_id(); 3768 3769 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 3770 3771 rcu_read_unlock(); 3772 preempt_enable(); 3773 } else 3774 #endif 3775 { 3776 unsigned int qtail; 3777 3778 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 3779 put_cpu(); 3780 } 3781 return ret; 3782 } 3783 3784 /** 3785 * netif_rx - post buffer to the network code 3786 * @skb: buffer to post 3787 * 3788 * This function receives a packet from a device driver and queues it for 3789 * the upper (protocol) levels to process. It always succeeds. The buffer 3790 * may be dropped during processing for congestion control or by the 3791 * protocol layers. 3792 * 3793 * return values: 3794 * NET_RX_SUCCESS (no congestion) 3795 * NET_RX_DROP (packet was dropped) 3796 * 3797 */ 3798 3799 int netif_rx(struct sk_buff *skb) 3800 { 3801 trace_netif_rx_entry(skb); 3802 3803 return netif_rx_internal(skb); 3804 } 3805 EXPORT_SYMBOL(netif_rx); 3806 3807 int netif_rx_ni(struct sk_buff *skb) 3808 { 3809 int err; 3810 3811 trace_netif_rx_ni_entry(skb); 3812 3813 preempt_disable(); 3814 err = netif_rx_internal(skb); 3815 if (local_softirq_pending()) 3816 do_softirq(); 3817 preempt_enable(); 3818 3819 return err; 3820 } 3821 EXPORT_SYMBOL(netif_rx_ni); 3822 3823 static __latent_entropy void net_tx_action(struct softirq_action *h) 3824 { 3825 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 3826 3827 if (sd->completion_queue) { 3828 struct sk_buff *clist; 3829 3830 local_irq_disable(); 3831 clist = sd->completion_queue; 3832 sd->completion_queue = NULL; 3833 local_irq_enable(); 3834 3835 while (clist) { 3836 struct sk_buff *skb = clist; 3837 3838 clist = clist->next; 3839 3840 WARN_ON(atomic_read(&skb->users)); 3841 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 3842 trace_consume_skb(skb); 3843 else 3844 trace_kfree_skb(skb, net_tx_action); 3845 3846 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 3847 __kfree_skb(skb); 3848 else 3849 __kfree_skb_defer(skb); 3850 } 3851 3852 __kfree_skb_flush(); 3853 } 3854 3855 if (sd->output_queue) { 3856 struct Qdisc *head; 3857 3858 local_irq_disable(); 3859 head = sd->output_queue; 3860 sd->output_queue = NULL; 3861 sd->output_queue_tailp = &sd->output_queue; 3862 local_irq_enable(); 3863 3864 while (head) { 3865 struct Qdisc *q = head; 3866 spinlock_t *root_lock; 3867 3868 head = head->next_sched; 3869 3870 root_lock = qdisc_lock(q); 3871 spin_lock(root_lock); 3872 /* We need to make sure head->next_sched is read 3873 * before clearing __QDISC_STATE_SCHED 3874 */ 3875 smp_mb__before_atomic(); 3876 clear_bit(__QDISC_STATE_SCHED, &q->state); 3877 qdisc_run(q); 3878 spin_unlock(root_lock); 3879 } 3880 } 3881 } 3882 3883 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 3884 /* This hook is defined here for ATM LANE */ 3885 int (*br_fdb_test_addr_hook)(struct net_device *dev, 3886 unsigned char *addr) __read_mostly; 3887 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 3888 #endif 3889 3890 static inline struct sk_buff * 3891 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3892 struct net_device *orig_dev) 3893 { 3894 #ifdef CONFIG_NET_CLS_ACT 3895 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); 3896 struct tcf_result cl_res; 3897 3898 /* If there's at least one ingress present somewhere (so 3899 * we get here via enabled static key), remaining devices 3900 * that are not configured with an ingress qdisc will bail 3901 * out here. 3902 */ 3903 if (!cl) 3904 return skb; 3905 if (*pt_prev) { 3906 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3907 *pt_prev = NULL; 3908 } 3909 3910 qdisc_skb_cb(skb)->pkt_len = skb->len; 3911 skb->tc_at_ingress = 1; 3912 qdisc_bstats_cpu_update(cl->q, skb); 3913 3914 switch (tc_classify(skb, cl, &cl_res, false)) { 3915 case TC_ACT_OK: 3916 case TC_ACT_RECLASSIFY: 3917 skb->tc_index = TC_H_MIN(cl_res.classid); 3918 break; 3919 case TC_ACT_SHOT: 3920 qdisc_qstats_cpu_drop(cl->q); 3921 kfree_skb(skb); 3922 return NULL; 3923 case TC_ACT_STOLEN: 3924 case TC_ACT_QUEUED: 3925 consume_skb(skb); 3926 return NULL; 3927 case TC_ACT_REDIRECT: 3928 /* skb_mac_header check was done by cls/act_bpf, so 3929 * we can safely push the L2 header back before 3930 * redirecting to another netdev 3931 */ 3932 __skb_push(skb, skb->mac_len); 3933 skb_do_redirect(skb); 3934 return NULL; 3935 default: 3936 break; 3937 } 3938 #endif /* CONFIG_NET_CLS_ACT */ 3939 return skb; 3940 } 3941 3942 /** 3943 * netdev_is_rx_handler_busy - check if receive handler is registered 3944 * @dev: device to check 3945 * 3946 * Check if a receive handler is already registered for a given device. 3947 * Return true if there one. 3948 * 3949 * The caller must hold the rtnl_mutex. 3950 */ 3951 bool netdev_is_rx_handler_busy(struct net_device *dev) 3952 { 3953 ASSERT_RTNL(); 3954 return dev && rtnl_dereference(dev->rx_handler); 3955 } 3956 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 3957 3958 /** 3959 * netdev_rx_handler_register - register receive handler 3960 * @dev: device to register a handler for 3961 * @rx_handler: receive handler to register 3962 * @rx_handler_data: data pointer that is used by rx handler 3963 * 3964 * Register a receive handler for a device. This handler will then be 3965 * called from __netif_receive_skb. A negative errno code is returned 3966 * on a failure. 3967 * 3968 * The caller must hold the rtnl_mutex. 3969 * 3970 * For a general description of rx_handler, see enum rx_handler_result. 3971 */ 3972 int netdev_rx_handler_register(struct net_device *dev, 3973 rx_handler_func_t *rx_handler, 3974 void *rx_handler_data) 3975 { 3976 if (netdev_is_rx_handler_busy(dev)) 3977 return -EBUSY; 3978 3979 /* Note: rx_handler_data must be set before rx_handler */ 3980 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 3981 rcu_assign_pointer(dev->rx_handler, rx_handler); 3982 3983 return 0; 3984 } 3985 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 3986 3987 /** 3988 * netdev_rx_handler_unregister - unregister receive handler 3989 * @dev: device to unregister a handler from 3990 * 3991 * Unregister a receive handler from a device. 3992 * 3993 * The caller must hold the rtnl_mutex. 3994 */ 3995 void netdev_rx_handler_unregister(struct net_device *dev) 3996 { 3997 3998 ASSERT_RTNL(); 3999 RCU_INIT_POINTER(dev->rx_handler, NULL); 4000 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 4001 * section has a guarantee to see a non NULL rx_handler_data 4002 * as well. 4003 */ 4004 synchronize_net(); 4005 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 4006 } 4007 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 4008 4009 /* 4010 * Limit the use of PFMEMALLOC reserves to those protocols that implement 4011 * the special handling of PFMEMALLOC skbs. 4012 */ 4013 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 4014 { 4015 switch (skb->protocol) { 4016 case htons(ETH_P_ARP): 4017 case htons(ETH_P_IP): 4018 case htons(ETH_P_IPV6): 4019 case htons(ETH_P_8021Q): 4020 case htons(ETH_P_8021AD): 4021 return true; 4022 default: 4023 return false; 4024 } 4025 } 4026 4027 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 4028 int *ret, struct net_device *orig_dev) 4029 { 4030 #ifdef CONFIG_NETFILTER_INGRESS 4031 if (nf_hook_ingress_active(skb)) { 4032 int ingress_retval; 4033 4034 if (*pt_prev) { 4035 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4036 *pt_prev = NULL; 4037 } 4038 4039 rcu_read_lock(); 4040 ingress_retval = nf_hook_ingress(skb); 4041 rcu_read_unlock(); 4042 return ingress_retval; 4043 } 4044 #endif /* CONFIG_NETFILTER_INGRESS */ 4045 return 0; 4046 } 4047 4048 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) 4049 { 4050 struct packet_type *ptype, *pt_prev; 4051 rx_handler_func_t *rx_handler; 4052 struct net_device *orig_dev; 4053 bool deliver_exact = false; 4054 int ret = NET_RX_DROP; 4055 __be16 type; 4056 4057 net_timestamp_check(!netdev_tstamp_prequeue, skb); 4058 4059 trace_netif_receive_skb(skb); 4060 4061 orig_dev = skb->dev; 4062 4063 skb_reset_network_header(skb); 4064 if (!skb_transport_header_was_set(skb)) 4065 skb_reset_transport_header(skb); 4066 skb_reset_mac_len(skb); 4067 4068 pt_prev = NULL; 4069 4070 another_round: 4071 skb->skb_iif = skb->dev->ifindex; 4072 4073 __this_cpu_inc(softnet_data.processed); 4074 4075 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 4076 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 4077 skb = skb_vlan_untag(skb); 4078 if (unlikely(!skb)) 4079 goto out; 4080 } 4081 4082 if (skb_skip_tc_classify(skb)) 4083 goto skip_classify; 4084 4085 if (pfmemalloc) 4086 goto skip_taps; 4087 4088 list_for_each_entry_rcu(ptype, &ptype_all, list) { 4089 if (pt_prev) 4090 ret = deliver_skb(skb, pt_prev, orig_dev); 4091 pt_prev = ptype; 4092 } 4093 4094 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 4095 if (pt_prev) 4096 ret = deliver_skb(skb, pt_prev, orig_dev); 4097 pt_prev = ptype; 4098 } 4099 4100 skip_taps: 4101 #ifdef CONFIG_NET_INGRESS 4102 if (static_key_false(&ingress_needed)) { 4103 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); 4104 if (!skb) 4105 goto out; 4106 4107 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 4108 goto out; 4109 } 4110 #endif 4111 skb_reset_tc(skb); 4112 skip_classify: 4113 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 4114 goto drop; 4115 4116 if (skb_vlan_tag_present(skb)) { 4117 if (pt_prev) { 4118 ret = deliver_skb(skb, pt_prev, orig_dev); 4119 pt_prev = NULL; 4120 } 4121 if (vlan_do_receive(&skb)) 4122 goto another_round; 4123 else if (unlikely(!skb)) 4124 goto out; 4125 } 4126 4127 rx_handler = rcu_dereference(skb->dev->rx_handler); 4128 if (rx_handler) { 4129 if (pt_prev) { 4130 ret = deliver_skb(skb, pt_prev, orig_dev); 4131 pt_prev = NULL; 4132 } 4133 switch (rx_handler(&skb)) { 4134 case RX_HANDLER_CONSUMED: 4135 ret = NET_RX_SUCCESS; 4136 goto out; 4137 case RX_HANDLER_ANOTHER: 4138 goto another_round; 4139 case RX_HANDLER_EXACT: 4140 deliver_exact = true; 4141 case RX_HANDLER_PASS: 4142 break; 4143 default: 4144 BUG(); 4145 } 4146 } 4147 4148 if (unlikely(skb_vlan_tag_present(skb))) { 4149 if (skb_vlan_tag_get_id(skb)) 4150 skb->pkt_type = PACKET_OTHERHOST; 4151 /* Note: we might in the future use prio bits 4152 * and set skb->priority like in vlan_do_receive() 4153 * For the time being, just ignore Priority Code Point 4154 */ 4155 skb->vlan_tci = 0; 4156 } 4157 4158 type = skb->protocol; 4159 4160 /* deliver only exact match when indicated */ 4161 if (likely(!deliver_exact)) { 4162 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4163 &ptype_base[ntohs(type) & 4164 PTYPE_HASH_MASK]); 4165 } 4166 4167 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4168 &orig_dev->ptype_specific); 4169 4170 if (unlikely(skb->dev != orig_dev)) { 4171 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4172 &skb->dev->ptype_specific); 4173 } 4174 4175 if (pt_prev) { 4176 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 4177 goto drop; 4178 else 4179 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 4180 } else { 4181 drop: 4182 if (!deliver_exact) 4183 atomic_long_inc(&skb->dev->rx_dropped); 4184 else 4185 atomic_long_inc(&skb->dev->rx_nohandler); 4186 kfree_skb(skb); 4187 /* Jamal, now you will not able to escape explaining 4188 * me how you were going to use this. :-) 4189 */ 4190 ret = NET_RX_DROP; 4191 } 4192 4193 out: 4194 return ret; 4195 } 4196 4197 static int __netif_receive_skb(struct sk_buff *skb) 4198 { 4199 int ret; 4200 4201 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 4202 unsigned long pflags = current->flags; 4203 4204 /* 4205 * PFMEMALLOC skbs are special, they should 4206 * - be delivered to SOCK_MEMALLOC sockets only 4207 * - stay away from userspace 4208 * - have bounded memory usage 4209 * 4210 * Use PF_MEMALLOC as this saves us from propagating the allocation 4211 * context down to all allocation sites. 4212 */ 4213 current->flags |= PF_MEMALLOC; 4214 ret = __netif_receive_skb_core(skb, true); 4215 tsk_restore_flags(current, pflags, PF_MEMALLOC); 4216 } else 4217 ret = __netif_receive_skb_core(skb, false); 4218 4219 return ret; 4220 } 4221 4222 static int netif_receive_skb_internal(struct sk_buff *skb) 4223 { 4224 int ret; 4225 4226 net_timestamp_check(netdev_tstamp_prequeue, skb); 4227 4228 if (skb_defer_rx_timestamp(skb)) 4229 return NET_RX_SUCCESS; 4230 4231 rcu_read_lock(); 4232 4233 #ifdef CONFIG_RPS 4234 if (static_key_false(&rps_needed)) { 4235 struct rps_dev_flow voidflow, *rflow = &voidflow; 4236 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 4237 4238 if (cpu >= 0) { 4239 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4240 rcu_read_unlock(); 4241 return ret; 4242 } 4243 } 4244 #endif 4245 ret = __netif_receive_skb(skb); 4246 rcu_read_unlock(); 4247 return ret; 4248 } 4249 4250 /** 4251 * netif_receive_skb - process receive buffer from network 4252 * @skb: buffer to process 4253 * 4254 * netif_receive_skb() is the main receive data processing function. 4255 * It always succeeds. The buffer may be dropped during processing 4256 * for congestion control or by the protocol layers. 4257 * 4258 * This function may only be called from softirq context and interrupts 4259 * should be enabled. 4260 * 4261 * Return values (usually ignored): 4262 * NET_RX_SUCCESS: no congestion 4263 * NET_RX_DROP: packet was dropped 4264 */ 4265 int netif_receive_skb(struct sk_buff *skb) 4266 { 4267 trace_netif_receive_skb_entry(skb); 4268 4269 return netif_receive_skb_internal(skb); 4270 } 4271 EXPORT_SYMBOL(netif_receive_skb); 4272 4273 DEFINE_PER_CPU(struct work_struct, flush_works); 4274 4275 /* Network device is going away, flush any packets still pending */ 4276 static void flush_backlog(struct work_struct *work) 4277 { 4278 struct sk_buff *skb, *tmp; 4279 struct softnet_data *sd; 4280 4281 local_bh_disable(); 4282 sd = this_cpu_ptr(&softnet_data); 4283 4284 local_irq_disable(); 4285 rps_lock(sd); 4286 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 4287 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 4288 __skb_unlink(skb, &sd->input_pkt_queue); 4289 kfree_skb(skb); 4290 input_queue_head_incr(sd); 4291 } 4292 } 4293 rps_unlock(sd); 4294 local_irq_enable(); 4295 4296 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 4297 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 4298 __skb_unlink(skb, &sd->process_queue); 4299 kfree_skb(skb); 4300 input_queue_head_incr(sd); 4301 } 4302 } 4303 local_bh_enable(); 4304 } 4305 4306 static void flush_all_backlogs(void) 4307 { 4308 unsigned int cpu; 4309 4310 get_online_cpus(); 4311 4312 for_each_online_cpu(cpu) 4313 queue_work_on(cpu, system_highpri_wq, 4314 per_cpu_ptr(&flush_works, cpu)); 4315 4316 for_each_online_cpu(cpu) 4317 flush_work(per_cpu_ptr(&flush_works, cpu)); 4318 4319 put_online_cpus(); 4320 } 4321 4322 static int napi_gro_complete(struct sk_buff *skb) 4323 { 4324 struct packet_offload *ptype; 4325 __be16 type = skb->protocol; 4326 struct list_head *head = &offload_base; 4327 int err = -ENOENT; 4328 4329 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 4330 4331 if (NAPI_GRO_CB(skb)->count == 1) { 4332 skb_shinfo(skb)->gso_size = 0; 4333 goto out; 4334 } 4335 4336 rcu_read_lock(); 4337 list_for_each_entry_rcu(ptype, head, list) { 4338 if (ptype->type != type || !ptype->callbacks.gro_complete) 4339 continue; 4340 4341 err = ptype->callbacks.gro_complete(skb, 0); 4342 break; 4343 } 4344 rcu_read_unlock(); 4345 4346 if (err) { 4347 WARN_ON(&ptype->list == head); 4348 kfree_skb(skb); 4349 return NET_RX_SUCCESS; 4350 } 4351 4352 out: 4353 return netif_receive_skb_internal(skb); 4354 } 4355 4356 /* napi->gro_list contains packets ordered by age. 4357 * youngest packets at the head of it. 4358 * Complete skbs in reverse order to reduce latencies. 4359 */ 4360 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 4361 { 4362 struct sk_buff *skb, *prev = NULL; 4363 4364 /* scan list and build reverse chain */ 4365 for (skb = napi->gro_list; skb != NULL; skb = skb->next) { 4366 skb->prev = prev; 4367 prev = skb; 4368 } 4369 4370 for (skb = prev; skb; skb = prev) { 4371 skb->next = NULL; 4372 4373 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 4374 return; 4375 4376 prev = skb->prev; 4377 napi_gro_complete(skb); 4378 napi->gro_count--; 4379 } 4380 4381 napi->gro_list = NULL; 4382 } 4383 EXPORT_SYMBOL(napi_gro_flush); 4384 4385 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) 4386 { 4387 struct sk_buff *p; 4388 unsigned int maclen = skb->dev->hard_header_len; 4389 u32 hash = skb_get_hash_raw(skb); 4390 4391 for (p = napi->gro_list; p; p = p->next) { 4392 unsigned long diffs; 4393 4394 NAPI_GRO_CB(p)->flush = 0; 4395 4396 if (hash != skb_get_hash_raw(p)) { 4397 NAPI_GRO_CB(p)->same_flow = 0; 4398 continue; 4399 } 4400 4401 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 4402 diffs |= p->vlan_tci ^ skb->vlan_tci; 4403 diffs |= skb_metadata_dst_cmp(p, skb); 4404 if (maclen == ETH_HLEN) 4405 diffs |= compare_ether_header(skb_mac_header(p), 4406 skb_mac_header(skb)); 4407 else if (!diffs) 4408 diffs = memcmp(skb_mac_header(p), 4409 skb_mac_header(skb), 4410 maclen); 4411 NAPI_GRO_CB(p)->same_flow = !diffs; 4412 } 4413 } 4414 4415 static void skb_gro_reset_offset(struct sk_buff *skb) 4416 { 4417 const struct skb_shared_info *pinfo = skb_shinfo(skb); 4418 const skb_frag_t *frag0 = &pinfo->frags[0]; 4419 4420 NAPI_GRO_CB(skb)->data_offset = 0; 4421 NAPI_GRO_CB(skb)->frag0 = NULL; 4422 NAPI_GRO_CB(skb)->frag0_len = 0; 4423 4424 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 4425 pinfo->nr_frags && 4426 !PageHighMem(skb_frag_page(frag0))) { 4427 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 4428 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 4429 skb_frag_size(frag0), 4430 skb->end - skb->tail); 4431 } 4432 } 4433 4434 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 4435 { 4436 struct skb_shared_info *pinfo = skb_shinfo(skb); 4437 4438 BUG_ON(skb->end - skb->tail < grow); 4439 4440 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 4441 4442 skb->data_len -= grow; 4443 skb->tail += grow; 4444 4445 pinfo->frags[0].page_offset += grow; 4446 skb_frag_size_sub(&pinfo->frags[0], grow); 4447 4448 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 4449 skb_frag_unref(skb, 0); 4450 memmove(pinfo->frags, pinfo->frags + 1, 4451 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 4452 } 4453 } 4454 4455 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4456 { 4457 struct sk_buff **pp = NULL; 4458 struct packet_offload *ptype; 4459 __be16 type = skb->protocol; 4460 struct list_head *head = &offload_base; 4461 int same_flow; 4462 enum gro_result ret; 4463 int grow; 4464 4465 if (!(skb->dev->features & NETIF_F_GRO)) 4466 goto normal; 4467 4468 if (skb->csum_bad) 4469 goto normal; 4470 4471 gro_list_prepare(napi, skb); 4472 4473 rcu_read_lock(); 4474 list_for_each_entry_rcu(ptype, head, list) { 4475 if (ptype->type != type || !ptype->callbacks.gro_receive) 4476 continue; 4477 4478 skb_set_network_header(skb, skb_gro_offset(skb)); 4479 skb_reset_mac_len(skb); 4480 NAPI_GRO_CB(skb)->same_flow = 0; 4481 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 4482 NAPI_GRO_CB(skb)->free = 0; 4483 NAPI_GRO_CB(skb)->encap_mark = 0; 4484 NAPI_GRO_CB(skb)->recursion_counter = 0; 4485 NAPI_GRO_CB(skb)->is_fou = 0; 4486 NAPI_GRO_CB(skb)->is_atomic = 1; 4487 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 4488 4489 /* Setup for GRO checksum validation */ 4490 switch (skb->ip_summed) { 4491 case CHECKSUM_COMPLETE: 4492 NAPI_GRO_CB(skb)->csum = skb->csum; 4493 NAPI_GRO_CB(skb)->csum_valid = 1; 4494 NAPI_GRO_CB(skb)->csum_cnt = 0; 4495 break; 4496 case CHECKSUM_UNNECESSARY: 4497 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 4498 NAPI_GRO_CB(skb)->csum_valid = 0; 4499 break; 4500 default: 4501 NAPI_GRO_CB(skb)->csum_cnt = 0; 4502 NAPI_GRO_CB(skb)->csum_valid = 0; 4503 } 4504 4505 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); 4506 break; 4507 } 4508 rcu_read_unlock(); 4509 4510 if (&ptype->list == head) 4511 goto normal; 4512 4513 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { 4514 ret = GRO_CONSUMED; 4515 goto ok; 4516 } 4517 4518 same_flow = NAPI_GRO_CB(skb)->same_flow; 4519 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 4520 4521 if (pp) { 4522 struct sk_buff *nskb = *pp; 4523 4524 *pp = nskb->next; 4525 nskb->next = NULL; 4526 napi_gro_complete(nskb); 4527 napi->gro_count--; 4528 } 4529 4530 if (same_flow) 4531 goto ok; 4532 4533 if (NAPI_GRO_CB(skb)->flush) 4534 goto normal; 4535 4536 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { 4537 struct sk_buff *nskb = napi->gro_list; 4538 4539 /* locate the end of the list to select the 'oldest' flow */ 4540 while (nskb->next) { 4541 pp = &nskb->next; 4542 nskb = *pp; 4543 } 4544 *pp = NULL; 4545 nskb->next = NULL; 4546 napi_gro_complete(nskb); 4547 } else { 4548 napi->gro_count++; 4549 } 4550 NAPI_GRO_CB(skb)->count = 1; 4551 NAPI_GRO_CB(skb)->age = jiffies; 4552 NAPI_GRO_CB(skb)->last = skb; 4553 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 4554 skb->next = napi->gro_list; 4555 napi->gro_list = skb; 4556 ret = GRO_HELD; 4557 4558 pull: 4559 grow = skb_gro_offset(skb) - skb_headlen(skb); 4560 if (grow > 0) 4561 gro_pull_from_frag0(skb, grow); 4562 ok: 4563 return ret; 4564 4565 normal: 4566 ret = GRO_NORMAL; 4567 goto pull; 4568 } 4569 4570 struct packet_offload *gro_find_receive_by_type(__be16 type) 4571 { 4572 struct list_head *offload_head = &offload_base; 4573 struct packet_offload *ptype; 4574 4575 list_for_each_entry_rcu(ptype, offload_head, list) { 4576 if (ptype->type != type || !ptype->callbacks.gro_receive) 4577 continue; 4578 return ptype; 4579 } 4580 return NULL; 4581 } 4582 EXPORT_SYMBOL(gro_find_receive_by_type); 4583 4584 struct packet_offload *gro_find_complete_by_type(__be16 type) 4585 { 4586 struct list_head *offload_head = &offload_base; 4587 struct packet_offload *ptype; 4588 4589 list_for_each_entry_rcu(ptype, offload_head, list) { 4590 if (ptype->type != type || !ptype->callbacks.gro_complete) 4591 continue; 4592 return ptype; 4593 } 4594 return NULL; 4595 } 4596 EXPORT_SYMBOL(gro_find_complete_by_type); 4597 4598 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 4599 { 4600 switch (ret) { 4601 case GRO_NORMAL: 4602 if (netif_receive_skb_internal(skb)) 4603 ret = GRO_DROP; 4604 break; 4605 4606 case GRO_DROP: 4607 kfree_skb(skb); 4608 break; 4609 4610 case GRO_MERGED_FREE: 4611 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) { 4612 skb_dst_drop(skb); 4613 secpath_reset(skb); 4614 kmem_cache_free(skbuff_head_cache, skb); 4615 } else { 4616 __kfree_skb(skb); 4617 } 4618 break; 4619 4620 case GRO_HELD: 4621 case GRO_MERGED: 4622 case GRO_CONSUMED: 4623 break; 4624 } 4625 4626 return ret; 4627 } 4628 4629 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4630 { 4631 skb_mark_napi_id(skb, napi); 4632 trace_napi_gro_receive_entry(skb); 4633 4634 skb_gro_reset_offset(skb); 4635 4636 return napi_skb_finish(dev_gro_receive(napi, skb), skb); 4637 } 4638 EXPORT_SYMBOL(napi_gro_receive); 4639 4640 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 4641 { 4642 if (unlikely(skb->pfmemalloc)) { 4643 consume_skb(skb); 4644 return; 4645 } 4646 __skb_pull(skb, skb_headlen(skb)); 4647 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 4648 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 4649 skb->vlan_tci = 0; 4650 skb->dev = napi->dev; 4651 skb->skb_iif = 0; 4652 skb->encapsulation = 0; 4653 skb_shinfo(skb)->gso_type = 0; 4654 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 4655 secpath_reset(skb); 4656 4657 napi->skb = skb; 4658 } 4659 4660 struct sk_buff *napi_get_frags(struct napi_struct *napi) 4661 { 4662 struct sk_buff *skb = napi->skb; 4663 4664 if (!skb) { 4665 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 4666 if (skb) { 4667 napi->skb = skb; 4668 skb_mark_napi_id(skb, napi); 4669 } 4670 } 4671 return skb; 4672 } 4673 EXPORT_SYMBOL(napi_get_frags); 4674 4675 static gro_result_t napi_frags_finish(struct napi_struct *napi, 4676 struct sk_buff *skb, 4677 gro_result_t ret) 4678 { 4679 switch (ret) { 4680 case GRO_NORMAL: 4681 case GRO_HELD: 4682 __skb_push(skb, ETH_HLEN); 4683 skb->protocol = eth_type_trans(skb, skb->dev); 4684 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 4685 ret = GRO_DROP; 4686 break; 4687 4688 case GRO_DROP: 4689 case GRO_MERGED_FREE: 4690 napi_reuse_skb(napi, skb); 4691 break; 4692 4693 case GRO_MERGED: 4694 case GRO_CONSUMED: 4695 break; 4696 } 4697 4698 return ret; 4699 } 4700 4701 /* Upper GRO stack assumes network header starts at gro_offset=0 4702 * Drivers could call both napi_gro_frags() and napi_gro_receive() 4703 * We copy ethernet header into skb->data to have a common layout. 4704 */ 4705 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 4706 { 4707 struct sk_buff *skb = napi->skb; 4708 const struct ethhdr *eth; 4709 unsigned int hlen = sizeof(*eth); 4710 4711 napi->skb = NULL; 4712 4713 skb_reset_mac_header(skb); 4714 skb_gro_reset_offset(skb); 4715 4716 eth = skb_gro_header_fast(skb, 0); 4717 if (unlikely(skb_gro_header_hard(skb, hlen))) { 4718 eth = skb_gro_header_slow(skb, hlen, 0); 4719 if (unlikely(!eth)) { 4720 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 4721 __func__, napi->dev->name); 4722 napi_reuse_skb(napi, skb); 4723 return NULL; 4724 } 4725 } else { 4726 gro_pull_from_frag0(skb, hlen); 4727 NAPI_GRO_CB(skb)->frag0 += hlen; 4728 NAPI_GRO_CB(skb)->frag0_len -= hlen; 4729 } 4730 __skb_pull(skb, hlen); 4731 4732 /* 4733 * This works because the only protocols we care about don't require 4734 * special handling. 4735 * We'll fix it up properly in napi_frags_finish() 4736 */ 4737 skb->protocol = eth->h_proto; 4738 4739 return skb; 4740 } 4741 4742 gro_result_t napi_gro_frags(struct napi_struct *napi) 4743 { 4744 struct sk_buff *skb = napi_frags_skb(napi); 4745 4746 if (!skb) 4747 return GRO_DROP; 4748 4749 trace_napi_gro_frags_entry(skb); 4750 4751 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 4752 } 4753 EXPORT_SYMBOL(napi_gro_frags); 4754 4755 /* Compute the checksum from gro_offset and return the folded value 4756 * after adding in any pseudo checksum. 4757 */ 4758 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 4759 { 4760 __wsum wsum; 4761 __sum16 sum; 4762 4763 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 4764 4765 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 4766 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 4767 if (likely(!sum)) { 4768 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 4769 !skb->csum_complete_sw) 4770 netdev_rx_csum_fault(skb->dev); 4771 } 4772 4773 NAPI_GRO_CB(skb)->csum = wsum; 4774 NAPI_GRO_CB(skb)->csum_valid = 1; 4775 4776 return sum; 4777 } 4778 EXPORT_SYMBOL(__skb_gro_checksum_complete); 4779 4780 /* 4781 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 4782 * Note: called with local irq disabled, but exits with local irq enabled. 4783 */ 4784 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 4785 { 4786 #ifdef CONFIG_RPS 4787 struct softnet_data *remsd = sd->rps_ipi_list; 4788 4789 if (remsd) { 4790 sd->rps_ipi_list = NULL; 4791 4792 local_irq_enable(); 4793 4794 /* Send pending IPI's to kick RPS processing on remote cpus. */ 4795 while (remsd) { 4796 struct softnet_data *next = remsd->rps_ipi_next; 4797 4798 if (cpu_online(remsd->cpu)) 4799 smp_call_function_single_async(remsd->cpu, 4800 &remsd->csd); 4801 remsd = next; 4802 } 4803 } else 4804 #endif 4805 local_irq_enable(); 4806 } 4807 4808 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 4809 { 4810 #ifdef CONFIG_RPS 4811 return sd->rps_ipi_list != NULL; 4812 #else 4813 return false; 4814 #endif 4815 } 4816 4817 static int process_backlog(struct napi_struct *napi, int quota) 4818 { 4819 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 4820 bool again = true; 4821 int work = 0; 4822 4823 /* Check if we have pending ipi, its better to send them now, 4824 * not waiting net_rx_action() end. 4825 */ 4826 if (sd_has_rps_ipi_waiting(sd)) { 4827 local_irq_disable(); 4828 net_rps_action_and_irq_enable(sd); 4829 } 4830 4831 napi->weight = dev_rx_weight; 4832 while (again) { 4833 struct sk_buff *skb; 4834 4835 while ((skb = __skb_dequeue(&sd->process_queue))) { 4836 rcu_read_lock(); 4837 __netif_receive_skb(skb); 4838 rcu_read_unlock(); 4839 input_queue_head_incr(sd); 4840 if (++work >= quota) 4841 return work; 4842 4843 } 4844 4845 local_irq_disable(); 4846 rps_lock(sd); 4847 if (skb_queue_empty(&sd->input_pkt_queue)) { 4848 /* 4849 * Inline a custom version of __napi_complete(). 4850 * only current cpu owns and manipulates this napi, 4851 * and NAPI_STATE_SCHED is the only possible flag set 4852 * on backlog. 4853 * We can use a plain write instead of clear_bit(), 4854 * and we dont need an smp_mb() memory barrier. 4855 */ 4856 napi->state = 0; 4857 again = false; 4858 } else { 4859 skb_queue_splice_tail_init(&sd->input_pkt_queue, 4860 &sd->process_queue); 4861 } 4862 rps_unlock(sd); 4863 local_irq_enable(); 4864 } 4865 4866 return work; 4867 } 4868 4869 /** 4870 * __napi_schedule - schedule for receive 4871 * @n: entry to schedule 4872 * 4873 * The entry's receive function will be scheduled to run. 4874 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 4875 */ 4876 void __napi_schedule(struct napi_struct *n) 4877 { 4878 unsigned long flags; 4879 4880 local_irq_save(flags); 4881 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4882 local_irq_restore(flags); 4883 } 4884 EXPORT_SYMBOL(__napi_schedule); 4885 4886 /** 4887 * __napi_schedule_irqoff - schedule for receive 4888 * @n: entry to schedule 4889 * 4890 * Variant of __napi_schedule() assuming hard irqs are masked 4891 */ 4892 void __napi_schedule_irqoff(struct napi_struct *n) 4893 { 4894 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 4895 } 4896 EXPORT_SYMBOL(__napi_schedule_irqoff); 4897 4898 bool napi_complete_done(struct napi_struct *n, int work_done) 4899 { 4900 unsigned long flags; 4901 4902 /* 4903 * 1) Don't let napi dequeue from the cpu poll list 4904 * just in case its running on a different cpu. 4905 * 2) If we are busy polling, do nothing here, we have 4906 * the guarantee we will be called later. 4907 */ 4908 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 4909 NAPIF_STATE_IN_BUSY_POLL))) 4910 return false; 4911 4912 if (n->gro_list) { 4913 unsigned long timeout = 0; 4914 4915 if (work_done) 4916 timeout = n->dev->gro_flush_timeout; 4917 4918 if (timeout) 4919 hrtimer_start(&n->timer, ns_to_ktime(timeout), 4920 HRTIMER_MODE_REL_PINNED); 4921 else 4922 napi_gro_flush(n, false); 4923 } 4924 if (unlikely(!list_empty(&n->poll_list))) { 4925 /* If n->poll_list is not empty, we need to mask irqs */ 4926 local_irq_save(flags); 4927 list_del_init(&n->poll_list); 4928 local_irq_restore(flags); 4929 } 4930 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); 4931 return true; 4932 } 4933 EXPORT_SYMBOL(napi_complete_done); 4934 4935 /* must be called under rcu_read_lock(), as we dont take a reference */ 4936 static struct napi_struct *napi_by_id(unsigned int napi_id) 4937 { 4938 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 4939 struct napi_struct *napi; 4940 4941 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 4942 if (napi->napi_id == napi_id) 4943 return napi; 4944 4945 return NULL; 4946 } 4947 4948 #if defined(CONFIG_NET_RX_BUSY_POLL) 4949 4950 #define BUSY_POLL_BUDGET 8 4951 4952 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) 4953 { 4954 int rc; 4955 4956 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 4957 4958 local_bh_disable(); 4959 4960 /* All we really want here is to re-enable device interrupts. 4961 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 4962 */ 4963 rc = napi->poll(napi, BUSY_POLL_BUDGET); 4964 netpoll_poll_unlock(have_poll_lock); 4965 if (rc == BUSY_POLL_BUDGET) 4966 __napi_schedule(napi); 4967 local_bh_enable(); 4968 if (local_softirq_pending()) 4969 do_softirq(); 4970 } 4971 4972 bool sk_busy_loop(struct sock *sk, int nonblock) 4973 { 4974 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0; 4975 int (*napi_poll)(struct napi_struct *napi, int budget); 4976 void *have_poll_lock = NULL; 4977 struct napi_struct *napi; 4978 int rc; 4979 4980 restart: 4981 rc = false; 4982 napi_poll = NULL; 4983 4984 rcu_read_lock(); 4985 4986 napi = napi_by_id(sk->sk_napi_id); 4987 if (!napi) 4988 goto out; 4989 4990 preempt_disable(); 4991 for (;;) { 4992 rc = 0; 4993 local_bh_disable(); 4994 if (!napi_poll) { 4995 unsigned long val = READ_ONCE(napi->state); 4996 4997 /* If multiple threads are competing for this napi, 4998 * we avoid dirtying napi->state as much as we can. 4999 */ 5000 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 5001 NAPIF_STATE_IN_BUSY_POLL)) 5002 goto count; 5003 if (cmpxchg(&napi->state, val, 5004 val | NAPIF_STATE_IN_BUSY_POLL | 5005 NAPIF_STATE_SCHED) != val) 5006 goto count; 5007 have_poll_lock = netpoll_poll_lock(napi); 5008 napi_poll = napi->poll; 5009 } 5010 rc = napi_poll(napi, BUSY_POLL_BUDGET); 5011 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); 5012 count: 5013 if (rc > 0) 5014 __NET_ADD_STATS(sock_net(sk), 5015 LINUX_MIB_BUSYPOLLRXPACKETS, rc); 5016 local_bh_enable(); 5017 5018 if (nonblock || !skb_queue_empty(&sk->sk_receive_queue) || 5019 busy_loop_timeout(end_time)) 5020 break; 5021 5022 if (unlikely(need_resched())) { 5023 if (napi_poll) 5024 busy_poll_stop(napi, have_poll_lock); 5025 preempt_enable(); 5026 rcu_read_unlock(); 5027 cond_resched(); 5028 rc = !skb_queue_empty(&sk->sk_receive_queue); 5029 if (rc || busy_loop_timeout(end_time)) 5030 return rc; 5031 goto restart; 5032 } 5033 cpu_relax(); 5034 } 5035 if (napi_poll) 5036 busy_poll_stop(napi, have_poll_lock); 5037 preempt_enable(); 5038 rc = !skb_queue_empty(&sk->sk_receive_queue); 5039 out: 5040 rcu_read_unlock(); 5041 return rc; 5042 } 5043 EXPORT_SYMBOL(sk_busy_loop); 5044 5045 #endif /* CONFIG_NET_RX_BUSY_POLL */ 5046 5047 static void napi_hash_add(struct napi_struct *napi) 5048 { 5049 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || 5050 test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) 5051 return; 5052 5053 spin_lock(&napi_hash_lock); 5054 5055 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */ 5056 do { 5057 if (unlikely(++napi_gen_id < NR_CPUS + 1)) 5058 napi_gen_id = NR_CPUS + 1; 5059 } while (napi_by_id(napi_gen_id)); 5060 napi->napi_id = napi_gen_id; 5061 5062 hlist_add_head_rcu(&napi->napi_hash_node, 5063 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 5064 5065 spin_unlock(&napi_hash_lock); 5066 } 5067 5068 /* Warning : caller is responsible to make sure rcu grace period 5069 * is respected before freeing memory containing @napi 5070 */ 5071 bool napi_hash_del(struct napi_struct *napi) 5072 { 5073 bool rcu_sync_needed = false; 5074 5075 spin_lock(&napi_hash_lock); 5076 5077 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { 5078 rcu_sync_needed = true; 5079 hlist_del_rcu(&napi->napi_hash_node); 5080 } 5081 spin_unlock(&napi_hash_lock); 5082 return rcu_sync_needed; 5083 } 5084 EXPORT_SYMBOL_GPL(napi_hash_del); 5085 5086 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 5087 { 5088 struct napi_struct *napi; 5089 5090 napi = container_of(timer, struct napi_struct, timer); 5091 if (napi->gro_list) 5092 napi_schedule_irqoff(napi); 5093 5094 return HRTIMER_NORESTART; 5095 } 5096 5097 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 5098 int (*poll)(struct napi_struct *, int), int weight) 5099 { 5100 INIT_LIST_HEAD(&napi->poll_list); 5101 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 5102 napi->timer.function = napi_watchdog; 5103 napi->gro_count = 0; 5104 napi->gro_list = NULL; 5105 napi->skb = NULL; 5106 napi->poll = poll; 5107 if (weight > NAPI_POLL_WEIGHT) 5108 pr_err_once("netif_napi_add() called with weight %d on device %s\n", 5109 weight, dev->name); 5110 napi->weight = weight; 5111 list_add(&napi->dev_list, &dev->napi_list); 5112 napi->dev = dev; 5113 #ifdef CONFIG_NETPOLL 5114 napi->poll_owner = -1; 5115 #endif 5116 set_bit(NAPI_STATE_SCHED, &napi->state); 5117 napi_hash_add(napi); 5118 } 5119 EXPORT_SYMBOL(netif_napi_add); 5120 5121 void napi_disable(struct napi_struct *n) 5122 { 5123 might_sleep(); 5124 set_bit(NAPI_STATE_DISABLE, &n->state); 5125 5126 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 5127 msleep(1); 5128 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 5129 msleep(1); 5130 5131 hrtimer_cancel(&n->timer); 5132 5133 clear_bit(NAPI_STATE_DISABLE, &n->state); 5134 } 5135 EXPORT_SYMBOL(napi_disable); 5136 5137 /* Must be called in process context */ 5138 void netif_napi_del(struct napi_struct *napi) 5139 { 5140 might_sleep(); 5141 if (napi_hash_del(napi)) 5142 synchronize_net(); 5143 list_del_init(&napi->dev_list); 5144 napi_free_frags(napi); 5145 5146 kfree_skb_list(napi->gro_list); 5147 napi->gro_list = NULL; 5148 napi->gro_count = 0; 5149 } 5150 EXPORT_SYMBOL(netif_napi_del); 5151 5152 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 5153 { 5154 void *have; 5155 int work, weight; 5156 5157 list_del_init(&n->poll_list); 5158 5159 have = netpoll_poll_lock(n); 5160 5161 weight = n->weight; 5162 5163 /* This NAPI_STATE_SCHED test is for avoiding a race 5164 * with netpoll's poll_napi(). Only the entity which 5165 * obtains the lock and sees NAPI_STATE_SCHED set will 5166 * actually make the ->poll() call. Therefore we avoid 5167 * accidentally calling ->poll() when NAPI is not scheduled. 5168 */ 5169 work = 0; 5170 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 5171 work = n->poll(n, weight); 5172 trace_napi_poll(n, work, weight); 5173 } 5174 5175 WARN_ON_ONCE(work > weight); 5176 5177 if (likely(work < weight)) 5178 goto out_unlock; 5179 5180 /* Drivers must not modify the NAPI state if they 5181 * consume the entire weight. In such cases this code 5182 * still "owns" the NAPI instance and therefore can 5183 * move the instance around on the list at-will. 5184 */ 5185 if (unlikely(napi_disable_pending(n))) { 5186 napi_complete(n); 5187 goto out_unlock; 5188 } 5189 5190 if (n->gro_list) { 5191 /* flush too old packets 5192 * If HZ < 1000, flush all packets. 5193 */ 5194 napi_gro_flush(n, HZ >= 1000); 5195 } 5196 5197 /* Some drivers may have called napi_schedule 5198 * prior to exhausting their budget. 5199 */ 5200 if (unlikely(!list_empty(&n->poll_list))) { 5201 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 5202 n->dev ? n->dev->name : "backlog"); 5203 goto out_unlock; 5204 } 5205 5206 list_add_tail(&n->poll_list, repoll); 5207 5208 out_unlock: 5209 netpoll_poll_unlock(have); 5210 5211 return work; 5212 } 5213 5214 static __latent_entropy void net_rx_action(struct softirq_action *h) 5215 { 5216 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5217 unsigned long time_limit = jiffies + 2; 5218 int budget = netdev_budget; 5219 LIST_HEAD(list); 5220 LIST_HEAD(repoll); 5221 5222 local_irq_disable(); 5223 list_splice_init(&sd->poll_list, &list); 5224 local_irq_enable(); 5225 5226 for (;;) { 5227 struct napi_struct *n; 5228 5229 if (list_empty(&list)) { 5230 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 5231 goto out; 5232 break; 5233 } 5234 5235 n = list_first_entry(&list, struct napi_struct, poll_list); 5236 budget -= napi_poll(n, &repoll); 5237 5238 /* If softirq window is exhausted then punt. 5239 * Allow this to run for 2 jiffies since which will allow 5240 * an average latency of 1.5/HZ. 5241 */ 5242 if (unlikely(budget <= 0 || 5243 time_after_eq(jiffies, time_limit))) { 5244 sd->time_squeeze++; 5245 break; 5246 } 5247 } 5248 5249 local_irq_disable(); 5250 5251 list_splice_tail_init(&sd->poll_list, &list); 5252 list_splice_tail(&repoll, &list); 5253 list_splice(&list, &sd->poll_list); 5254 if (!list_empty(&sd->poll_list)) 5255 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 5256 5257 net_rps_action_and_irq_enable(sd); 5258 out: 5259 __kfree_skb_flush(); 5260 } 5261 5262 struct netdev_adjacent { 5263 struct net_device *dev; 5264 5265 /* upper master flag, there can only be one master device per list */ 5266 bool master; 5267 5268 /* counter for the number of times this device was added to us */ 5269 u16 ref_nr; 5270 5271 /* private field for the users */ 5272 void *private; 5273 5274 struct list_head list; 5275 struct rcu_head rcu; 5276 }; 5277 5278 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 5279 struct list_head *adj_list) 5280 { 5281 struct netdev_adjacent *adj; 5282 5283 list_for_each_entry(adj, adj_list, list) { 5284 if (adj->dev == adj_dev) 5285 return adj; 5286 } 5287 return NULL; 5288 } 5289 5290 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) 5291 { 5292 struct net_device *dev = data; 5293 5294 return upper_dev == dev; 5295 } 5296 5297 /** 5298 * netdev_has_upper_dev - Check if device is linked to an upper device 5299 * @dev: device 5300 * @upper_dev: upper device to check 5301 * 5302 * Find out if a device is linked to specified upper device and return true 5303 * in case it is. Note that this checks only immediate upper device, 5304 * not through a complete stack of devices. The caller must hold the RTNL lock. 5305 */ 5306 bool netdev_has_upper_dev(struct net_device *dev, 5307 struct net_device *upper_dev) 5308 { 5309 ASSERT_RTNL(); 5310 5311 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 5312 upper_dev); 5313 } 5314 EXPORT_SYMBOL(netdev_has_upper_dev); 5315 5316 /** 5317 * netdev_has_upper_dev_all - Check if device is linked to an upper device 5318 * @dev: device 5319 * @upper_dev: upper device to check 5320 * 5321 * Find out if a device is linked to specified upper device and return true 5322 * in case it is. Note that this checks the entire upper device chain. 5323 * The caller must hold rcu lock. 5324 */ 5325 5326 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 5327 struct net_device *upper_dev) 5328 { 5329 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 5330 upper_dev); 5331 } 5332 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 5333 5334 /** 5335 * netdev_has_any_upper_dev - Check if device is linked to some device 5336 * @dev: device 5337 * 5338 * Find out if a device is linked to an upper device and return true in case 5339 * it is. The caller must hold the RTNL lock. 5340 */ 5341 static bool netdev_has_any_upper_dev(struct net_device *dev) 5342 { 5343 ASSERT_RTNL(); 5344 5345 return !list_empty(&dev->adj_list.upper); 5346 } 5347 5348 /** 5349 * netdev_master_upper_dev_get - Get master upper device 5350 * @dev: device 5351 * 5352 * Find a master upper device and return pointer to it or NULL in case 5353 * it's not there. The caller must hold the RTNL lock. 5354 */ 5355 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 5356 { 5357 struct netdev_adjacent *upper; 5358 5359 ASSERT_RTNL(); 5360 5361 if (list_empty(&dev->adj_list.upper)) 5362 return NULL; 5363 5364 upper = list_first_entry(&dev->adj_list.upper, 5365 struct netdev_adjacent, list); 5366 if (likely(upper->master)) 5367 return upper->dev; 5368 return NULL; 5369 } 5370 EXPORT_SYMBOL(netdev_master_upper_dev_get); 5371 5372 /** 5373 * netdev_has_any_lower_dev - Check if device is linked to some device 5374 * @dev: device 5375 * 5376 * Find out if a device is linked to a lower device and return true in case 5377 * it is. The caller must hold the RTNL lock. 5378 */ 5379 static bool netdev_has_any_lower_dev(struct net_device *dev) 5380 { 5381 ASSERT_RTNL(); 5382 5383 return !list_empty(&dev->adj_list.lower); 5384 } 5385 5386 void *netdev_adjacent_get_private(struct list_head *adj_list) 5387 { 5388 struct netdev_adjacent *adj; 5389 5390 adj = list_entry(adj_list, struct netdev_adjacent, list); 5391 5392 return adj->private; 5393 } 5394 EXPORT_SYMBOL(netdev_adjacent_get_private); 5395 5396 /** 5397 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 5398 * @dev: device 5399 * @iter: list_head ** of the current position 5400 * 5401 * Gets the next device from the dev's upper list, starting from iter 5402 * position. The caller must hold RCU read lock. 5403 */ 5404 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 5405 struct list_head **iter) 5406 { 5407 struct netdev_adjacent *upper; 5408 5409 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 5410 5411 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5412 5413 if (&upper->list == &dev->adj_list.upper) 5414 return NULL; 5415 5416 *iter = &upper->list; 5417 5418 return upper->dev; 5419 } 5420 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 5421 5422 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 5423 struct list_head **iter) 5424 { 5425 struct netdev_adjacent *upper; 5426 5427 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 5428 5429 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5430 5431 if (&upper->list == &dev->adj_list.upper) 5432 return NULL; 5433 5434 *iter = &upper->list; 5435 5436 return upper->dev; 5437 } 5438 5439 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 5440 int (*fn)(struct net_device *dev, 5441 void *data), 5442 void *data) 5443 { 5444 struct net_device *udev; 5445 struct list_head *iter; 5446 int ret; 5447 5448 for (iter = &dev->adj_list.upper, 5449 udev = netdev_next_upper_dev_rcu(dev, &iter); 5450 udev; 5451 udev = netdev_next_upper_dev_rcu(dev, &iter)) { 5452 /* first is the upper device itself */ 5453 ret = fn(udev, data); 5454 if (ret) 5455 return ret; 5456 5457 /* then look at all of its upper devices */ 5458 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); 5459 if (ret) 5460 return ret; 5461 } 5462 5463 return 0; 5464 } 5465 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 5466 5467 /** 5468 * netdev_lower_get_next_private - Get the next ->private from the 5469 * lower neighbour list 5470 * @dev: device 5471 * @iter: list_head ** of the current position 5472 * 5473 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5474 * list, starting from iter position. The caller must hold either hold the 5475 * RTNL lock or its own locking that guarantees that the neighbour lower 5476 * list will remain unchanged. 5477 */ 5478 void *netdev_lower_get_next_private(struct net_device *dev, 5479 struct list_head **iter) 5480 { 5481 struct netdev_adjacent *lower; 5482 5483 lower = list_entry(*iter, struct netdev_adjacent, list); 5484 5485 if (&lower->list == &dev->adj_list.lower) 5486 return NULL; 5487 5488 *iter = lower->list.next; 5489 5490 return lower->private; 5491 } 5492 EXPORT_SYMBOL(netdev_lower_get_next_private); 5493 5494 /** 5495 * netdev_lower_get_next_private_rcu - Get the next ->private from the 5496 * lower neighbour list, RCU 5497 * variant 5498 * @dev: device 5499 * @iter: list_head ** of the current position 5500 * 5501 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5502 * list, starting from iter position. The caller must hold RCU read lock. 5503 */ 5504 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 5505 struct list_head **iter) 5506 { 5507 struct netdev_adjacent *lower; 5508 5509 WARN_ON_ONCE(!rcu_read_lock_held()); 5510 5511 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5512 5513 if (&lower->list == &dev->adj_list.lower) 5514 return NULL; 5515 5516 *iter = &lower->list; 5517 5518 return lower->private; 5519 } 5520 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 5521 5522 /** 5523 * netdev_lower_get_next - Get the next device from the lower neighbour 5524 * list 5525 * @dev: device 5526 * @iter: list_head ** of the current position 5527 * 5528 * Gets the next netdev_adjacent from the dev's lower neighbour 5529 * list, starting from iter position. The caller must hold RTNL lock or 5530 * its own locking that guarantees that the neighbour lower 5531 * list will remain unchanged. 5532 */ 5533 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 5534 { 5535 struct netdev_adjacent *lower; 5536 5537 lower = list_entry(*iter, struct netdev_adjacent, list); 5538 5539 if (&lower->list == &dev->adj_list.lower) 5540 return NULL; 5541 5542 *iter = lower->list.next; 5543 5544 return lower->dev; 5545 } 5546 EXPORT_SYMBOL(netdev_lower_get_next); 5547 5548 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 5549 struct list_head **iter) 5550 { 5551 struct netdev_adjacent *lower; 5552 5553 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 5554 5555 if (&lower->list == &dev->adj_list.lower) 5556 return NULL; 5557 5558 *iter = &lower->list; 5559 5560 return lower->dev; 5561 } 5562 5563 int netdev_walk_all_lower_dev(struct net_device *dev, 5564 int (*fn)(struct net_device *dev, 5565 void *data), 5566 void *data) 5567 { 5568 struct net_device *ldev; 5569 struct list_head *iter; 5570 int ret; 5571 5572 for (iter = &dev->adj_list.lower, 5573 ldev = netdev_next_lower_dev(dev, &iter); 5574 ldev; 5575 ldev = netdev_next_lower_dev(dev, &iter)) { 5576 /* first is the lower device itself */ 5577 ret = fn(ldev, data); 5578 if (ret) 5579 return ret; 5580 5581 /* then look at all of its lower devices */ 5582 ret = netdev_walk_all_lower_dev(ldev, fn, data); 5583 if (ret) 5584 return ret; 5585 } 5586 5587 return 0; 5588 } 5589 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 5590 5591 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 5592 struct list_head **iter) 5593 { 5594 struct netdev_adjacent *lower; 5595 5596 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5597 if (&lower->list == &dev->adj_list.lower) 5598 return NULL; 5599 5600 *iter = &lower->list; 5601 5602 return lower->dev; 5603 } 5604 5605 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 5606 int (*fn)(struct net_device *dev, 5607 void *data), 5608 void *data) 5609 { 5610 struct net_device *ldev; 5611 struct list_head *iter; 5612 int ret; 5613 5614 for (iter = &dev->adj_list.lower, 5615 ldev = netdev_next_lower_dev_rcu(dev, &iter); 5616 ldev; 5617 ldev = netdev_next_lower_dev_rcu(dev, &iter)) { 5618 /* first is the lower device itself */ 5619 ret = fn(ldev, data); 5620 if (ret) 5621 return ret; 5622 5623 /* then look at all of its lower devices */ 5624 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); 5625 if (ret) 5626 return ret; 5627 } 5628 5629 return 0; 5630 } 5631 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 5632 5633 /** 5634 * netdev_lower_get_first_private_rcu - Get the first ->private from the 5635 * lower neighbour list, RCU 5636 * variant 5637 * @dev: device 5638 * 5639 * Gets the first netdev_adjacent->private from the dev's lower neighbour 5640 * list. The caller must hold RCU read lock. 5641 */ 5642 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 5643 { 5644 struct netdev_adjacent *lower; 5645 5646 lower = list_first_or_null_rcu(&dev->adj_list.lower, 5647 struct netdev_adjacent, list); 5648 if (lower) 5649 return lower->private; 5650 return NULL; 5651 } 5652 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 5653 5654 /** 5655 * netdev_master_upper_dev_get_rcu - Get master upper device 5656 * @dev: device 5657 * 5658 * Find a master upper device and return pointer to it or NULL in case 5659 * it's not there. The caller must hold the RCU read lock. 5660 */ 5661 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 5662 { 5663 struct netdev_adjacent *upper; 5664 5665 upper = list_first_or_null_rcu(&dev->adj_list.upper, 5666 struct netdev_adjacent, list); 5667 if (upper && likely(upper->master)) 5668 return upper->dev; 5669 return NULL; 5670 } 5671 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 5672 5673 static int netdev_adjacent_sysfs_add(struct net_device *dev, 5674 struct net_device *adj_dev, 5675 struct list_head *dev_list) 5676 { 5677 char linkname[IFNAMSIZ+7]; 5678 5679 sprintf(linkname, dev_list == &dev->adj_list.upper ? 5680 "upper_%s" : "lower_%s", adj_dev->name); 5681 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 5682 linkname); 5683 } 5684 static void netdev_adjacent_sysfs_del(struct net_device *dev, 5685 char *name, 5686 struct list_head *dev_list) 5687 { 5688 char linkname[IFNAMSIZ+7]; 5689 5690 sprintf(linkname, dev_list == &dev->adj_list.upper ? 5691 "upper_%s" : "lower_%s", name); 5692 sysfs_remove_link(&(dev->dev.kobj), linkname); 5693 } 5694 5695 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 5696 struct net_device *adj_dev, 5697 struct list_head *dev_list) 5698 { 5699 return (dev_list == &dev->adj_list.upper || 5700 dev_list == &dev->adj_list.lower) && 5701 net_eq(dev_net(dev), dev_net(adj_dev)); 5702 } 5703 5704 static int __netdev_adjacent_dev_insert(struct net_device *dev, 5705 struct net_device *adj_dev, 5706 struct list_head *dev_list, 5707 void *private, bool master) 5708 { 5709 struct netdev_adjacent *adj; 5710 int ret; 5711 5712 adj = __netdev_find_adj(adj_dev, dev_list); 5713 5714 if (adj) { 5715 adj->ref_nr += 1; 5716 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 5717 dev->name, adj_dev->name, adj->ref_nr); 5718 5719 return 0; 5720 } 5721 5722 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 5723 if (!adj) 5724 return -ENOMEM; 5725 5726 adj->dev = adj_dev; 5727 adj->master = master; 5728 adj->ref_nr = 1; 5729 adj->private = private; 5730 dev_hold(adj_dev); 5731 5732 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 5733 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 5734 5735 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 5736 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 5737 if (ret) 5738 goto free_adj; 5739 } 5740 5741 /* Ensure that master link is always the first item in list. */ 5742 if (master) { 5743 ret = sysfs_create_link(&(dev->dev.kobj), 5744 &(adj_dev->dev.kobj), "master"); 5745 if (ret) 5746 goto remove_symlinks; 5747 5748 list_add_rcu(&adj->list, dev_list); 5749 } else { 5750 list_add_tail_rcu(&adj->list, dev_list); 5751 } 5752 5753 return 0; 5754 5755 remove_symlinks: 5756 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 5757 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 5758 free_adj: 5759 kfree(adj); 5760 dev_put(adj_dev); 5761 5762 return ret; 5763 } 5764 5765 static void __netdev_adjacent_dev_remove(struct net_device *dev, 5766 struct net_device *adj_dev, 5767 u16 ref_nr, 5768 struct list_head *dev_list) 5769 { 5770 struct netdev_adjacent *adj; 5771 5772 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 5773 dev->name, adj_dev->name, ref_nr); 5774 5775 adj = __netdev_find_adj(adj_dev, dev_list); 5776 5777 if (!adj) { 5778 pr_err("Adjacency does not exist for device %s from %s\n", 5779 dev->name, adj_dev->name); 5780 WARN_ON(1); 5781 return; 5782 } 5783 5784 if (adj->ref_nr > ref_nr) { 5785 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 5786 dev->name, adj_dev->name, ref_nr, 5787 adj->ref_nr - ref_nr); 5788 adj->ref_nr -= ref_nr; 5789 return; 5790 } 5791 5792 if (adj->master) 5793 sysfs_remove_link(&(dev->dev.kobj), "master"); 5794 5795 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 5796 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 5797 5798 list_del_rcu(&adj->list); 5799 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 5800 adj_dev->name, dev->name, adj_dev->name); 5801 dev_put(adj_dev); 5802 kfree_rcu(adj, rcu); 5803 } 5804 5805 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 5806 struct net_device *upper_dev, 5807 struct list_head *up_list, 5808 struct list_head *down_list, 5809 void *private, bool master) 5810 { 5811 int ret; 5812 5813 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 5814 private, master); 5815 if (ret) 5816 return ret; 5817 5818 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 5819 private, false); 5820 if (ret) { 5821 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 5822 return ret; 5823 } 5824 5825 return 0; 5826 } 5827 5828 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 5829 struct net_device *upper_dev, 5830 u16 ref_nr, 5831 struct list_head *up_list, 5832 struct list_head *down_list) 5833 { 5834 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 5835 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 5836 } 5837 5838 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 5839 struct net_device *upper_dev, 5840 void *private, bool master) 5841 { 5842 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 5843 &dev->adj_list.upper, 5844 &upper_dev->adj_list.lower, 5845 private, master); 5846 } 5847 5848 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 5849 struct net_device *upper_dev) 5850 { 5851 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 5852 &dev->adj_list.upper, 5853 &upper_dev->adj_list.lower); 5854 } 5855 5856 static int __netdev_upper_dev_link(struct net_device *dev, 5857 struct net_device *upper_dev, bool master, 5858 void *upper_priv, void *upper_info) 5859 { 5860 struct netdev_notifier_changeupper_info changeupper_info; 5861 int ret = 0; 5862 5863 ASSERT_RTNL(); 5864 5865 if (dev == upper_dev) 5866 return -EBUSY; 5867 5868 /* To prevent loops, check if dev is not upper device to upper_dev. */ 5869 if (netdev_has_upper_dev(upper_dev, dev)) 5870 return -EBUSY; 5871 5872 if (netdev_has_upper_dev(dev, upper_dev)) 5873 return -EEXIST; 5874 5875 if (master && netdev_master_upper_dev_get(dev)) 5876 return -EBUSY; 5877 5878 changeupper_info.upper_dev = upper_dev; 5879 changeupper_info.master = master; 5880 changeupper_info.linking = true; 5881 changeupper_info.upper_info = upper_info; 5882 5883 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, 5884 &changeupper_info.info); 5885 ret = notifier_to_errno(ret); 5886 if (ret) 5887 return ret; 5888 5889 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 5890 master); 5891 if (ret) 5892 return ret; 5893 5894 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, 5895 &changeupper_info.info); 5896 ret = notifier_to_errno(ret); 5897 if (ret) 5898 goto rollback; 5899 5900 return 0; 5901 5902 rollback: 5903 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5904 5905 return ret; 5906 } 5907 5908 /** 5909 * netdev_upper_dev_link - Add a link to the upper device 5910 * @dev: device 5911 * @upper_dev: new upper device 5912 * 5913 * Adds a link to device which is upper to this one. The caller must hold 5914 * the RTNL lock. On a failure a negative errno code is returned. 5915 * On success the reference counts are adjusted and the function 5916 * returns zero. 5917 */ 5918 int netdev_upper_dev_link(struct net_device *dev, 5919 struct net_device *upper_dev) 5920 { 5921 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); 5922 } 5923 EXPORT_SYMBOL(netdev_upper_dev_link); 5924 5925 /** 5926 * netdev_master_upper_dev_link - Add a master link to the upper device 5927 * @dev: device 5928 * @upper_dev: new upper device 5929 * @upper_priv: upper device private 5930 * @upper_info: upper info to be passed down via notifier 5931 * 5932 * Adds a link to device which is upper to this one. In this case, only 5933 * one master upper device can be linked, although other non-master devices 5934 * might be linked as well. The caller must hold the RTNL lock. 5935 * On a failure a negative errno code is returned. On success the reference 5936 * counts are adjusted and the function returns zero. 5937 */ 5938 int netdev_master_upper_dev_link(struct net_device *dev, 5939 struct net_device *upper_dev, 5940 void *upper_priv, void *upper_info) 5941 { 5942 return __netdev_upper_dev_link(dev, upper_dev, true, 5943 upper_priv, upper_info); 5944 } 5945 EXPORT_SYMBOL(netdev_master_upper_dev_link); 5946 5947 /** 5948 * netdev_upper_dev_unlink - Removes a link to upper device 5949 * @dev: device 5950 * @upper_dev: new upper device 5951 * 5952 * Removes a link to device which is upper to this one. The caller must hold 5953 * the RTNL lock. 5954 */ 5955 void netdev_upper_dev_unlink(struct net_device *dev, 5956 struct net_device *upper_dev) 5957 { 5958 struct netdev_notifier_changeupper_info changeupper_info; 5959 5960 ASSERT_RTNL(); 5961 5962 changeupper_info.upper_dev = upper_dev; 5963 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 5964 changeupper_info.linking = false; 5965 5966 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, 5967 &changeupper_info.info); 5968 5969 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 5970 5971 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, 5972 &changeupper_info.info); 5973 } 5974 EXPORT_SYMBOL(netdev_upper_dev_unlink); 5975 5976 /** 5977 * netdev_bonding_info_change - Dispatch event about slave change 5978 * @dev: device 5979 * @bonding_info: info to dispatch 5980 * 5981 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 5982 * The caller must hold the RTNL lock. 5983 */ 5984 void netdev_bonding_info_change(struct net_device *dev, 5985 struct netdev_bonding_info *bonding_info) 5986 { 5987 struct netdev_notifier_bonding_info info; 5988 5989 memcpy(&info.bonding_info, bonding_info, 5990 sizeof(struct netdev_bonding_info)); 5991 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, 5992 &info.info); 5993 } 5994 EXPORT_SYMBOL(netdev_bonding_info_change); 5995 5996 static void netdev_adjacent_add_links(struct net_device *dev) 5997 { 5998 struct netdev_adjacent *iter; 5999 6000 struct net *net = dev_net(dev); 6001 6002 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6003 if (!net_eq(net, dev_net(iter->dev))) 6004 continue; 6005 netdev_adjacent_sysfs_add(iter->dev, dev, 6006 &iter->dev->adj_list.lower); 6007 netdev_adjacent_sysfs_add(dev, iter->dev, 6008 &dev->adj_list.upper); 6009 } 6010 6011 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6012 if (!net_eq(net, dev_net(iter->dev))) 6013 continue; 6014 netdev_adjacent_sysfs_add(iter->dev, dev, 6015 &iter->dev->adj_list.upper); 6016 netdev_adjacent_sysfs_add(dev, iter->dev, 6017 &dev->adj_list.lower); 6018 } 6019 } 6020 6021 static void netdev_adjacent_del_links(struct net_device *dev) 6022 { 6023 struct netdev_adjacent *iter; 6024 6025 struct net *net = dev_net(dev); 6026 6027 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6028 if (!net_eq(net, dev_net(iter->dev))) 6029 continue; 6030 netdev_adjacent_sysfs_del(iter->dev, dev->name, 6031 &iter->dev->adj_list.lower); 6032 netdev_adjacent_sysfs_del(dev, iter->dev->name, 6033 &dev->adj_list.upper); 6034 } 6035 6036 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6037 if (!net_eq(net, dev_net(iter->dev))) 6038 continue; 6039 netdev_adjacent_sysfs_del(iter->dev, dev->name, 6040 &iter->dev->adj_list.upper); 6041 netdev_adjacent_sysfs_del(dev, iter->dev->name, 6042 &dev->adj_list.lower); 6043 } 6044 } 6045 6046 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 6047 { 6048 struct netdev_adjacent *iter; 6049 6050 struct net *net = dev_net(dev); 6051 6052 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6053 if (!net_eq(net, dev_net(iter->dev))) 6054 continue; 6055 netdev_adjacent_sysfs_del(iter->dev, oldname, 6056 &iter->dev->adj_list.lower); 6057 netdev_adjacent_sysfs_add(iter->dev, dev, 6058 &iter->dev->adj_list.lower); 6059 } 6060 6061 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6062 if (!net_eq(net, dev_net(iter->dev))) 6063 continue; 6064 netdev_adjacent_sysfs_del(iter->dev, oldname, 6065 &iter->dev->adj_list.upper); 6066 netdev_adjacent_sysfs_add(iter->dev, dev, 6067 &iter->dev->adj_list.upper); 6068 } 6069 } 6070 6071 void *netdev_lower_dev_get_private(struct net_device *dev, 6072 struct net_device *lower_dev) 6073 { 6074 struct netdev_adjacent *lower; 6075 6076 if (!lower_dev) 6077 return NULL; 6078 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 6079 if (!lower) 6080 return NULL; 6081 6082 return lower->private; 6083 } 6084 EXPORT_SYMBOL(netdev_lower_dev_get_private); 6085 6086 6087 int dev_get_nest_level(struct net_device *dev) 6088 { 6089 struct net_device *lower = NULL; 6090 struct list_head *iter; 6091 int max_nest = -1; 6092 int nest; 6093 6094 ASSERT_RTNL(); 6095 6096 netdev_for_each_lower_dev(dev, lower, iter) { 6097 nest = dev_get_nest_level(lower); 6098 if (max_nest < nest) 6099 max_nest = nest; 6100 } 6101 6102 return max_nest + 1; 6103 } 6104 EXPORT_SYMBOL(dev_get_nest_level); 6105 6106 /** 6107 * netdev_lower_change - Dispatch event about lower device state change 6108 * @lower_dev: device 6109 * @lower_state_info: state to dispatch 6110 * 6111 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 6112 * The caller must hold the RTNL lock. 6113 */ 6114 void netdev_lower_state_changed(struct net_device *lower_dev, 6115 void *lower_state_info) 6116 { 6117 struct netdev_notifier_changelowerstate_info changelowerstate_info; 6118 6119 ASSERT_RTNL(); 6120 changelowerstate_info.lower_state_info = lower_state_info; 6121 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, 6122 &changelowerstate_info.info); 6123 } 6124 EXPORT_SYMBOL(netdev_lower_state_changed); 6125 6126 static void dev_change_rx_flags(struct net_device *dev, int flags) 6127 { 6128 const struct net_device_ops *ops = dev->netdev_ops; 6129 6130 if (ops->ndo_change_rx_flags) 6131 ops->ndo_change_rx_flags(dev, flags); 6132 } 6133 6134 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 6135 { 6136 unsigned int old_flags = dev->flags; 6137 kuid_t uid; 6138 kgid_t gid; 6139 6140 ASSERT_RTNL(); 6141 6142 dev->flags |= IFF_PROMISC; 6143 dev->promiscuity += inc; 6144 if (dev->promiscuity == 0) { 6145 /* 6146 * Avoid overflow. 6147 * If inc causes overflow, untouch promisc and return error. 6148 */ 6149 if (inc < 0) 6150 dev->flags &= ~IFF_PROMISC; 6151 else { 6152 dev->promiscuity -= inc; 6153 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 6154 dev->name); 6155 return -EOVERFLOW; 6156 } 6157 } 6158 if (dev->flags != old_flags) { 6159 pr_info("device %s %s promiscuous mode\n", 6160 dev->name, 6161 dev->flags & IFF_PROMISC ? "entered" : "left"); 6162 if (audit_enabled) { 6163 current_uid_gid(&uid, &gid); 6164 audit_log(current->audit_context, GFP_ATOMIC, 6165 AUDIT_ANOM_PROMISCUOUS, 6166 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 6167 dev->name, (dev->flags & IFF_PROMISC), 6168 (old_flags & IFF_PROMISC), 6169 from_kuid(&init_user_ns, audit_get_loginuid(current)), 6170 from_kuid(&init_user_ns, uid), 6171 from_kgid(&init_user_ns, gid), 6172 audit_get_sessionid(current)); 6173 } 6174 6175 dev_change_rx_flags(dev, IFF_PROMISC); 6176 } 6177 if (notify) 6178 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 6179 return 0; 6180 } 6181 6182 /** 6183 * dev_set_promiscuity - update promiscuity count on a device 6184 * @dev: device 6185 * @inc: modifier 6186 * 6187 * Add or remove promiscuity from a device. While the count in the device 6188 * remains above zero the interface remains promiscuous. Once it hits zero 6189 * the device reverts back to normal filtering operation. A negative inc 6190 * value is used to drop promiscuity on the device. 6191 * Return 0 if successful or a negative errno code on error. 6192 */ 6193 int dev_set_promiscuity(struct net_device *dev, int inc) 6194 { 6195 unsigned int old_flags = dev->flags; 6196 int err; 6197 6198 err = __dev_set_promiscuity(dev, inc, true); 6199 if (err < 0) 6200 return err; 6201 if (dev->flags != old_flags) 6202 dev_set_rx_mode(dev); 6203 return err; 6204 } 6205 EXPORT_SYMBOL(dev_set_promiscuity); 6206 6207 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 6208 { 6209 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 6210 6211 ASSERT_RTNL(); 6212 6213 dev->flags |= IFF_ALLMULTI; 6214 dev->allmulti += inc; 6215 if (dev->allmulti == 0) { 6216 /* 6217 * Avoid overflow. 6218 * If inc causes overflow, untouch allmulti and return error. 6219 */ 6220 if (inc < 0) 6221 dev->flags &= ~IFF_ALLMULTI; 6222 else { 6223 dev->allmulti -= inc; 6224 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 6225 dev->name); 6226 return -EOVERFLOW; 6227 } 6228 } 6229 if (dev->flags ^ old_flags) { 6230 dev_change_rx_flags(dev, IFF_ALLMULTI); 6231 dev_set_rx_mode(dev); 6232 if (notify) 6233 __dev_notify_flags(dev, old_flags, 6234 dev->gflags ^ old_gflags); 6235 } 6236 return 0; 6237 } 6238 6239 /** 6240 * dev_set_allmulti - update allmulti count on a device 6241 * @dev: device 6242 * @inc: modifier 6243 * 6244 * Add or remove reception of all multicast frames to a device. While the 6245 * count in the device remains above zero the interface remains listening 6246 * to all interfaces. Once it hits zero the device reverts back to normal 6247 * filtering operation. A negative @inc value is used to drop the counter 6248 * when releasing a resource needing all multicasts. 6249 * Return 0 if successful or a negative errno code on error. 6250 */ 6251 6252 int dev_set_allmulti(struct net_device *dev, int inc) 6253 { 6254 return __dev_set_allmulti(dev, inc, true); 6255 } 6256 EXPORT_SYMBOL(dev_set_allmulti); 6257 6258 /* 6259 * Upload unicast and multicast address lists to device and 6260 * configure RX filtering. When the device doesn't support unicast 6261 * filtering it is put in promiscuous mode while unicast addresses 6262 * are present. 6263 */ 6264 void __dev_set_rx_mode(struct net_device *dev) 6265 { 6266 const struct net_device_ops *ops = dev->netdev_ops; 6267 6268 /* dev_open will call this function so the list will stay sane. */ 6269 if (!(dev->flags&IFF_UP)) 6270 return; 6271 6272 if (!netif_device_present(dev)) 6273 return; 6274 6275 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 6276 /* Unicast addresses changes may only happen under the rtnl, 6277 * therefore calling __dev_set_promiscuity here is safe. 6278 */ 6279 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 6280 __dev_set_promiscuity(dev, 1, false); 6281 dev->uc_promisc = true; 6282 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 6283 __dev_set_promiscuity(dev, -1, false); 6284 dev->uc_promisc = false; 6285 } 6286 } 6287 6288 if (ops->ndo_set_rx_mode) 6289 ops->ndo_set_rx_mode(dev); 6290 } 6291 6292 void dev_set_rx_mode(struct net_device *dev) 6293 { 6294 netif_addr_lock_bh(dev); 6295 __dev_set_rx_mode(dev); 6296 netif_addr_unlock_bh(dev); 6297 } 6298 6299 /** 6300 * dev_get_flags - get flags reported to userspace 6301 * @dev: device 6302 * 6303 * Get the combination of flag bits exported through APIs to userspace. 6304 */ 6305 unsigned int dev_get_flags(const struct net_device *dev) 6306 { 6307 unsigned int flags; 6308 6309 flags = (dev->flags & ~(IFF_PROMISC | 6310 IFF_ALLMULTI | 6311 IFF_RUNNING | 6312 IFF_LOWER_UP | 6313 IFF_DORMANT)) | 6314 (dev->gflags & (IFF_PROMISC | 6315 IFF_ALLMULTI)); 6316 6317 if (netif_running(dev)) { 6318 if (netif_oper_up(dev)) 6319 flags |= IFF_RUNNING; 6320 if (netif_carrier_ok(dev)) 6321 flags |= IFF_LOWER_UP; 6322 if (netif_dormant(dev)) 6323 flags |= IFF_DORMANT; 6324 } 6325 6326 return flags; 6327 } 6328 EXPORT_SYMBOL(dev_get_flags); 6329 6330 int __dev_change_flags(struct net_device *dev, unsigned int flags) 6331 { 6332 unsigned int old_flags = dev->flags; 6333 int ret; 6334 6335 ASSERT_RTNL(); 6336 6337 /* 6338 * Set the flags on our device. 6339 */ 6340 6341 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 6342 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 6343 IFF_AUTOMEDIA)) | 6344 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 6345 IFF_ALLMULTI)); 6346 6347 /* 6348 * Load in the correct multicast list now the flags have changed. 6349 */ 6350 6351 if ((old_flags ^ flags) & IFF_MULTICAST) 6352 dev_change_rx_flags(dev, IFF_MULTICAST); 6353 6354 dev_set_rx_mode(dev); 6355 6356 /* 6357 * Have we downed the interface. We handle IFF_UP ourselves 6358 * according to user attempts to set it, rather than blindly 6359 * setting it. 6360 */ 6361 6362 ret = 0; 6363 if ((old_flags ^ flags) & IFF_UP) 6364 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); 6365 6366 if ((flags ^ dev->gflags) & IFF_PROMISC) { 6367 int inc = (flags & IFF_PROMISC) ? 1 : -1; 6368 unsigned int old_flags = dev->flags; 6369 6370 dev->gflags ^= IFF_PROMISC; 6371 6372 if (__dev_set_promiscuity(dev, inc, false) >= 0) 6373 if (dev->flags != old_flags) 6374 dev_set_rx_mode(dev); 6375 } 6376 6377 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 6378 * is important. Some (broken) drivers set IFF_PROMISC, when 6379 * IFF_ALLMULTI is requested not asking us and not reporting. 6380 */ 6381 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 6382 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 6383 6384 dev->gflags ^= IFF_ALLMULTI; 6385 __dev_set_allmulti(dev, inc, false); 6386 } 6387 6388 return ret; 6389 } 6390 6391 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 6392 unsigned int gchanges) 6393 { 6394 unsigned int changes = dev->flags ^ old_flags; 6395 6396 if (gchanges) 6397 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 6398 6399 if (changes & IFF_UP) { 6400 if (dev->flags & IFF_UP) 6401 call_netdevice_notifiers(NETDEV_UP, dev); 6402 else 6403 call_netdevice_notifiers(NETDEV_DOWN, dev); 6404 } 6405 6406 if (dev->flags & IFF_UP && 6407 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 6408 struct netdev_notifier_change_info change_info; 6409 6410 change_info.flags_changed = changes; 6411 call_netdevice_notifiers_info(NETDEV_CHANGE, dev, 6412 &change_info.info); 6413 } 6414 } 6415 6416 /** 6417 * dev_change_flags - change device settings 6418 * @dev: device 6419 * @flags: device state flags 6420 * 6421 * Change settings on device based state flags. The flags are 6422 * in the userspace exported format. 6423 */ 6424 int dev_change_flags(struct net_device *dev, unsigned int flags) 6425 { 6426 int ret; 6427 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 6428 6429 ret = __dev_change_flags(dev, flags); 6430 if (ret < 0) 6431 return ret; 6432 6433 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 6434 __dev_notify_flags(dev, old_flags, changes); 6435 return ret; 6436 } 6437 EXPORT_SYMBOL(dev_change_flags); 6438 6439 static int __dev_set_mtu(struct net_device *dev, int new_mtu) 6440 { 6441 const struct net_device_ops *ops = dev->netdev_ops; 6442 6443 if (ops->ndo_change_mtu) 6444 return ops->ndo_change_mtu(dev, new_mtu); 6445 6446 dev->mtu = new_mtu; 6447 return 0; 6448 } 6449 6450 /** 6451 * dev_set_mtu - Change maximum transfer unit 6452 * @dev: device 6453 * @new_mtu: new transfer unit 6454 * 6455 * Change the maximum transfer size of the network device. 6456 */ 6457 int dev_set_mtu(struct net_device *dev, int new_mtu) 6458 { 6459 int err, orig_mtu; 6460 6461 if (new_mtu == dev->mtu) 6462 return 0; 6463 6464 /* MTU must be positive, and in range */ 6465 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 6466 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", 6467 dev->name, new_mtu, dev->min_mtu); 6468 return -EINVAL; 6469 } 6470 6471 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 6472 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", 6473 dev->name, new_mtu, dev->max_mtu); 6474 return -EINVAL; 6475 } 6476 6477 if (!netif_device_present(dev)) 6478 return -ENODEV; 6479 6480 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 6481 err = notifier_to_errno(err); 6482 if (err) 6483 return err; 6484 6485 orig_mtu = dev->mtu; 6486 err = __dev_set_mtu(dev, new_mtu); 6487 6488 if (!err) { 6489 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 6490 err = notifier_to_errno(err); 6491 if (err) { 6492 /* setting mtu back and notifying everyone again, 6493 * so that they have a chance to revert changes. 6494 */ 6495 __dev_set_mtu(dev, orig_mtu); 6496 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 6497 } 6498 } 6499 return err; 6500 } 6501 EXPORT_SYMBOL(dev_set_mtu); 6502 6503 /** 6504 * dev_set_group - Change group this device belongs to 6505 * @dev: device 6506 * @new_group: group this device should belong to 6507 */ 6508 void dev_set_group(struct net_device *dev, int new_group) 6509 { 6510 dev->group = new_group; 6511 } 6512 EXPORT_SYMBOL(dev_set_group); 6513 6514 /** 6515 * dev_set_mac_address - Change Media Access Control Address 6516 * @dev: device 6517 * @sa: new address 6518 * 6519 * Change the hardware (MAC) address of the device 6520 */ 6521 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 6522 { 6523 const struct net_device_ops *ops = dev->netdev_ops; 6524 int err; 6525 6526 if (!ops->ndo_set_mac_address) 6527 return -EOPNOTSUPP; 6528 if (sa->sa_family != dev->type) 6529 return -EINVAL; 6530 if (!netif_device_present(dev)) 6531 return -ENODEV; 6532 err = ops->ndo_set_mac_address(dev, sa); 6533 if (err) 6534 return err; 6535 dev->addr_assign_type = NET_ADDR_SET; 6536 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 6537 add_device_randomness(dev->dev_addr, dev->addr_len); 6538 return 0; 6539 } 6540 EXPORT_SYMBOL(dev_set_mac_address); 6541 6542 /** 6543 * dev_change_carrier - Change device carrier 6544 * @dev: device 6545 * @new_carrier: new value 6546 * 6547 * Change device carrier 6548 */ 6549 int dev_change_carrier(struct net_device *dev, bool new_carrier) 6550 { 6551 const struct net_device_ops *ops = dev->netdev_ops; 6552 6553 if (!ops->ndo_change_carrier) 6554 return -EOPNOTSUPP; 6555 if (!netif_device_present(dev)) 6556 return -ENODEV; 6557 return ops->ndo_change_carrier(dev, new_carrier); 6558 } 6559 EXPORT_SYMBOL(dev_change_carrier); 6560 6561 /** 6562 * dev_get_phys_port_id - Get device physical port ID 6563 * @dev: device 6564 * @ppid: port ID 6565 * 6566 * Get device physical port ID 6567 */ 6568 int dev_get_phys_port_id(struct net_device *dev, 6569 struct netdev_phys_item_id *ppid) 6570 { 6571 const struct net_device_ops *ops = dev->netdev_ops; 6572 6573 if (!ops->ndo_get_phys_port_id) 6574 return -EOPNOTSUPP; 6575 return ops->ndo_get_phys_port_id(dev, ppid); 6576 } 6577 EXPORT_SYMBOL(dev_get_phys_port_id); 6578 6579 /** 6580 * dev_get_phys_port_name - Get device physical port name 6581 * @dev: device 6582 * @name: port name 6583 * @len: limit of bytes to copy to name 6584 * 6585 * Get device physical port name 6586 */ 6587 int dev_get_phys_port_name(struct net_device *dev, 6588 char *name, size_t len) 6589 { 6590 const struct net_device_ops *ops = dev->netdev_ops; 6591 6592 if (!ops->ndo_get_phys_port_name) 6593 return -EOPNOTSUPP; 6594 return ops->ndo_get_phys_port_name(dev, name, len); 6595 } 6596 EXPORT_SYMBOL(dev_get_phys_port_name); 6597 6598 /** 6599 * dev_change_proto_down - update protocol port state information 6600 * @dev: device 6601 * @proto_down: new value 6602 * 6603 * This info can be used by switch drivers to set the phys state of the 6604 * port. 6605 */ 6606 int dev_change_proto_down(struct net_device *dev, bool proto_down) 6607 { 6608 const struct net_device_ops *ops = dev->netdev_ops; 6609 6610 if (!ops->ndo_change_proto_down) 6611 return -EOPNOTSUPP; 6612 if (!netif_device_present(dev)) 6613 return -ENODEV; 6614 return ops->ndo_change_proto_down(dev, proto_down); 6615 } 6616 EXPORT_SYMBOL(dev_change_proto_down); 6617 6618 /** 6619 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 6620 * @dev: device 6621 * @fd: new program fd or negative value to clear 6622 * @flags: xdp-related flags 6623 * 6624 * Set or clear a bpf program for a device 6625 */ 6626 int dev_change_xdp_fd(struct net_device *dev, int fd, u32 flags) 6627 { 6628 const struct net_device_ops *ops = dev->netdev_ops; 6629 struct bpf_prog *prog = NULL; 6630 struct netdev_xdp xdp; 6631 int err; 6632 6633 ASSERT_RTNL(); 6634 6635 if (!ops->ndo_xdp) 6636 return -EOPNOTSUPP; 6637 if (fd >= 0) { 6638 if (flags & XDP_FLAGS_UPDATE_IF_NOEXIST) { 6639 memset(&xdp, 0, sizeof(xdp)); 6640 xdp.command = XDP_QUERY_PROG; 6641 6642 err = ops->ndo_xdp(dev, &xdp); 6643 if (err < 0) 6644 return err; 6645 if (xdp.prog_attached) 6646 return -EBUSY; 6647 } 6648 6649 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); 6650 if (IS_ERR(prog)) 6651 return PTR_ERR(prog); 6652 } 6653 6654 memset(&xdp, 0, sizeof(xdp)); 6655 xdp.command = XDP_SETUP_PROG; 6656 xdp.prog = prog; 6657 6658 err = ops->ndo_xdp(dev, &xdp); 6659 if (err < 0 && prog) 6660 bpf_prog_put(prog); 6661 6662 return err; 6663 } 6664 EXPORT_SYMBOL(dev_change_xdp_fd); 6665 6666 /** 6667 * dev_new_index - allocate an ifindex 6668 * @net: the applicable net namespace 6669 * 6670 * Returns a suitable unique value for a new device interface 6671 * number. The caller must hold the rtnl semaphore or the 6672 * dev_base_lock to be sure it remains unique. 6673 */ 6674 static int dev_new_index(struct net *net) 6675 { 6676 int ifindex = net->ifindex; 6677 6678 for (;;) { 6679 if (++ifindex <= 0) 6680 ifindex = 1; 6681 if (!__dev_get_by_index(net, ifindex)) 6682 return net->ifindex = ifindex; 6683 } 6684 } 6685 6686 /* Delayed registration/unregisteration */ 6687 static LIST_HEAD(net_todo_list); 6688 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 6689 6690 static void net_set_todo(struct net_device *dev) 6691 { 6692 list_add_tail(&dev->todo_list, &net_todo_list); 6693 dev_net(dev)->dev_unreg_count++; 6694 } 6695 6696 static void rollback_registered_many(struct list_head *head) 6697 { 6698 struct net_device *dev, *tmp; 6699 LIST_HEAD(close_head); 6700 6701 BUG_ON(dev_boot_phase); 6702 ASSERT_RTNL(); 6703 6704 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 6705 /* Some devices call without registering 6706 * for initialization unwind. Remove those 6707 * devices and proceed with the remaining. 6708 */ 6709 if (dev->reg_state == NETREG_UNINITIALIZED) { 6710 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 6711 dev->name, dev); 6712 6713 WARN_ON(1); 6714 list_del(&dev->unreg_list); 6715 continue; 6716 } 6717 dev->dismantle = true; 6718 BUG_ON(dev->reg_state != NETREG_REGISTERED); 6719 } 6720 6721 /* If device is running, close it first. */ 6722 list_for_each_entry(dev, head, unreg_list) 6723 list_add_tail(&dev->close_list, &close_head); 6724 dev_close_many(&close_head, true); 6725 6726 list_for_each_entry(dev, head, unreg_list) { 6727 /* And unlink it from device chain. */ 6728 unlist_netdevice(dev); 6729 6730 dev->reg_state = NETREG_UNREGISTERING; 6731 } 6732 flush_all_backlogs(); 6733 6734 synchronize_net(); 6735 6736 list_for_each_entry(dev, head, unreg_list) { 6737 struct sk_buff *skb = NULL; 6738 6739 /* Shutdown queueing discipline. */ 6740 dev_shutdown(dev); 6741 6742 6743 /* Notify protocols, that we are about to destroy 6744 * this device. They should clean all the things. 6745 */ 6746 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 6747 6748 if (!dev->rtnl_link_ops || 6749 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 6750 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 6751 GFP_KERNEL); 6752 6753 /* 6754 * Flush the unicast and multicast chains 6755 */ 6756 dev_uc_flush(dev); 6757 dev_mc_flush(dev); 6758 6759 if (dev->netdev_ops->ndo_uninit) 6760 dev->netdev_ops->ndo_uninit(dev); 6761 6762 if (skb) 6763 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 6764 6765 /* Notifier chain MUST detach us all upper devices. */ 6766 WARN_ON(netdev_has_any_upper_dev(dev)); 6767 WARN_ON(netdev_has_any_lower_dev(dev)); 6768 6769 /* Remove entries from kobject tree */ 6770 netdev_unregister_kobject(dev); 6771 #ifdef CONFIG_XPS 6772 /* Remove XPS queueing entries */ 6773 netif_reset_xps_queues_gt(dev, 0); 6774 #endif 6775 } 6776 6777 synchronize_net(); 6778 6779 list_for_each_entry(dev, head, unreg_list) 6780 dev_put(dev); 6781 } 6782 6783 static void rollback_registered(struct net_device *dev) 6784 { 6785 LIST_HEAD(single); 6786 6787 list_add(&dev->unreg_list, &single); 6788 rollback_registered_many(&single); 6789 list_del(&single); 6790 } 6791 6792 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 6793 struct net_device *upper, netdev_features_t features) 6794 { 6795 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 6796 netdev_features_t feature; 6797 int feature_bit; 6798 6799 for_each_netdev_feature(&upper_disables, feature_bit) { 6800 feature = __NETIF_F_BIT(feature_bit); 6801 if (!(upper->wanted_features & feature) 6802 && (features & feature)) { 6803 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 6804 &feature, upper->name); 6805 features &= ~feature; 6806 } 6807 } 6808 6809 return features; 6810 } 6811 6812 static void netdev_sync_lower_features(struct net_device *upper, 6813 struct net_device *lower, netdev_features_t features) 6814 { 6815 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 6816 netdev_features_t feature; 6817 int feature_bit; 6818 6819 for_each_netdev_feature(&upper_disables, feature_bit) { 6820 feature = __NETIF_F_BIT(feature_bit); 6821 if (!(features & feature) && (lower->features & feature)) { 6822 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 6823 &feature, lower->name); 6824 lower->wanted_features &= ~feature; 6825 netdev_update_features(lower); 6826 6827 if (unlikely(lower->features & feature)) 6828 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 6829 &feature, lower->name); 6830 } 6831 } 6832 } 6833 6834 static netdev_features_t netdev_fix_features(struct net_device *dev, 6835 netdev_features_t features) 6836 { 6837 /* Fix illegal checksum combinations */ 6838 if ((features & NETIF_F_HW_CSUM) && 6839 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 6840 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 6841 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 6842 } 6843 6844 /* TSO requires that SG is present as well. */ 6845 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 6846 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 6847 features &= ~NETIF_F_ALL_TSO; 6848 } 6849 6850 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 6851 !(features & NETIF_F_IP_CSUM)) { 6852 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 6853 features &= ~NETIF_F_TSO; 6854 features &= ~NETIF_F_TSO_ECN; 6855 } 6856 6857 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 6858 !(features & NETIF_F_IPV6_CSUM)) { 6859 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 6860 features &= ~NETIF_F_TSO6; 6861 } 6862 6863 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 6864 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 6865 features &= ~NETIF_F_TSO_MANGLEID; 6866 6867 /* TSO ECN requires that TSO is present as well. */ 6868 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 6869 features &= ~NETIF_F_TSO_ECN; 6870 6871 /* Software GSO depends on SG. */ 6872 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 6873 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 6874 features &= ~NETIF_F_GSO; 6875 } 6876 6877 /* UFO needs SG and checksumming */ 6878 if (features & NETIF_F_UFO) { 6879 /* maybe split UFO into V4 and V6? */ 6880 if (!(features & NETIF_F_HW_CSUM) && 6881 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != 6882 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { 6883 netdev_dbg(dev, 6884 "Dropping NETIF_F_UFO since no checksum offload features.\n"); 6885 features &= ~NETIF_F_UFO; 6886 } 6887 6888 if (!(features & NETIF_F_SG)) { 6889 netdev_dbg(dev, 6890 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); 6891 features &= ~NETIF_F_UFO; 6892 } 6893 } 6894 6895 /* GSO partial features require GSO partial be set */ 6896 if ((features & dev->gso_partial_features) && 6897 !(features & NETIF_F_GSO_PARTIAL)) { 6898 netdev_dbg(dev, 6899 "Dropping partially supported GSO features since no GSO partial.\n"); 6900 features &= ~dev->gso_partial_features; 6901 } 6902 6903 return features; 6904 } 6905 6906 int __netdev_update_features(struct net_device *dev) 6907 { 6908 struct net_device *upper, *lower; 6909 netdev_features_t features; 6910 struct list_head *iter; 6911 int err = -1; 6912 6913 ASSERT_RTNL(); 6914 6915 features = netdev_get_wanted_features(dev); 6916 6917 if (dev->netdev_ops->ndo_fix_features) 6918 features = dev->netdev_ops->ndo_fix_features(dev, features); 6919 6920 /* driver might be less strict about feature dependencies */ 6921 features = netdev_fix_features(dev, features); 6922 6923 /* some features can't be enabled if they're off an an upper device */ 6924 netdev_for_each_upper_dev_rcu(dev, upper, iter) 6925 features = netdev_sync_upper_features(dev, upper, features); 6926 6927 if (dev->features == features) 6928 goto sync_lower; 6929 6930 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 6931 &dev->features, &features); 6932 6933 if (dev->netdev_ops->ndo_set_features) 6934 err = dev->netdev_ops->ndo_set_features(dev, features); 6935 else 6936 err = 0; 6937 6938 if (unlikely(err < 0)) { 6939 netdev_err(dev, 6940 "set_features() failed (%d); wanted %pNF, left %pNF\n", 6941 err, &features, &dev->features); 6942 /* return non-0 since some features might have changed and 6943 * it's better to fire a spurious notification than miss it 6944 */ 6945 return -1; 6946 } 6947 6948 sync_lower: 6949 /* some features must be disabled on lower devices when disabled 6950 * on an upper device (think: bonding master or bridge) 6951 */ 6952 netdev_for_each_lower_dev(dev, lower, iter) 6953 netdev_sync_lower_features(dev, lower, features); 6954 6955 if (!err) 6956 dev->features = features; 6957 6958 return err < 0 ? 0 : 1; 6959 } 6960 6961 /** 6962 * netdev_update_features - recalculate device features 6963 * @dev: the device to check 6964 * 6965 * Recalculate dev->features set and send notifications if it 6966 * has changed. Should be called after driver or hardware dependent 6967 * conditions might have changed that influence the features. 6968 */ 6969 void netdev_update_features(struct net_device *dev) 6970 { 6971 if (__netdev_update_features(dev)) 6972 netdev_features_change(dev); 6973 } 6974 EXPORT_SYMBOL(netdev_update_features); 6975 6976 /** 6977 * netdev_change_features - recalculate device features 6978 * @dev: the device to check 6979 * 6980 * Recalculate dev->features set and send notifications even 6981 * if they have not changed. Should be called instead of 6982 * netdev_update_features() if also dev->vlan_features might 6983 * have changed to allow the changes to be propagated to stacked 6984 * VLAN devices. 6985 */ 6986 void netdev_change_features(struct net_device *dev) 6987 { 6988 __netdev_update_features(dev); 6989 netdev_features_change(dev); 6990 } 6991 EXPORT_SYMBOL(netdev_change_features); 6992 6993 /** 6994 * netif_stacked_transfer_operstate - transfer operstate 6995 * @rootdev: the root or lower level device to transfer state from 6996 * @dev: the device to transfer operstate to 6997 * 6998 * Transfer operational state from root to device. This is normally 6999 * called when a stacking relationship exists between the root 7000 * device and the device(a leaf device). 7001 */ 7002 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 7003 struct net_device *dev) 7004 { 7005 if (rootdev->operstate == IF_OPER_DORMANT) 7006 netif_dormant_on(dev); 7007 else 7008 netif_dormant_off(dev); 7009 7010 if (netif_carrier_ok(rootdev)) { 7011 if (!netif_carrier_ok(dev)) 7012 netif_carrier_on(dev); 7013 } else { 7014 if (netif_carrier_ok(dev)) 7015 netif_carrier_off(dev); 7016 } 7017 } 7018 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 7019 7020 #ifdef CONFIG_SYSFS 7021 static int netif_alloc_rx_queues(struct net_device *dev) 7022 { 7023 unsigned int i, count = dev->num_rx_queues; 7024 struct netdev_rx_queue *rx; 7025 size_t sz = count * sizeof(*rx); 7026 7027 BUG_ON(count < 1); 7028 7029 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 7030 if (!rx) { 7031 rx = vzalloc(sz); 7032 if (!rx) 7033 return -ENOMEM; 7034 } 7035 dev->_rx = rx; 7036 7037 for (i = 0; i < count; i++) 7038 rx[i].dev = dev; 7039 return 0; 7040 } 7041 #endif 7042 7043 static void netdev_init_one_queue(struct net_device *dev, 7044 struct netdev_queue *queue, void *_unused) 7045 { 7046 /* Initialize queue lock */ 7047 spin_lock_init(&queue->_xmit_lock); 7048 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 7049 queue->xmit_lock_owner = -1; 7050 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 7051 queue->dev = dev; 7052 #ifdef CONFIG_BQL 7053 dql_init(&queue->dql, HZ); 7054 #endif 7055 } 7056 7057 static void netif_free_tx_queues(struct net_device *dev) 7058 { 7059 kvfree(dev->_tx); 7060 } 7061 7062 static int netif_alloc_netdev_queues(struct net_device *dev) 7063 { 7064 unsigned int count = dev->num_tx_queues; 7065 struct netdev_queue *tx; 7066 size_t sz = count * sizeof(*tx); 7067 7068 if (count < 1 || count > 0xffff) 7069 return -EINVAL; 7070 7071 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 7072 if (!tx) { 7073 tx = vzalloc(sz); 7074 if (!tx) 7075 return -ENOMEM; 7076 } 7077 dev->_tx = tx; 7078 7079 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 7080 spin_lock_init(&dev->tx_global_lock); 7081 7082 return 0; 7083 } 7084 7085 void netif_tx_stop_all_queues(struct net_device *dev) 7086 { 7087 unsigned int i; 7088 7089 for (i = 0; i < dev->num_tx_queues; i++) { 7090 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 7091 7092 netif_tx_stop_queue(txq); 7093 } 7094 } 7095 EXPORT_SYMBOL(netif_tx_stop_all_queues); 7096 7097 /** 7098 * register_netdevice - register a network device 7099 * @dev: device to register 7100 * 7101 * Take a completed network device structure and add it to the kernel 7102 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 7103 * chain. 0 is returned on success. A negative errno code is returned 7104 * on a failure to set up the device, or if the name is a duplicate. 7105 * 7106 * Callers must hold the rtnl semaphore. You may want 7107 * register_netdev() instead of this. 7108 * 7109 * BUGS: 7110 * The locking appears insufficient to guarantee two parallel registers 7111 * will not get the same name. 7112 */ 7113 7114 int register_netdevice(struct net_device *dev) 7115 { 7116 int ret; 7117 struct net *net = dev_net(dev); 7118 7119 BUG_ON(dev_boot_phase); 7120 ASSERT_RTNL(); 7121 7122 might_sleep(); 7123 7124 /* When net_device's are persistent, this will be fatal. */ 7125 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 7126 BUG_ON(!net); 7127 7128 spin_lock_init(&dev->addr_list_lock); 7129 netdev_set_addr_lockdep_class(dev); 7130 7131 ret = dev_get_valid_name(net, dev, dev->name); 7132 if (ret < 0) 7133 goto out; 7134 7135 /* Init, if this function is available */ 7136 if (dev->netdev_ops->ndo_init) { 7137 ret = dev->netdev_ops->ndo_init(dev); 7138 if (ret) { 7139 if (ret > 0) 7140 ret = -EIO; 7141 goto out; 7142 } 7143 } 7144 7145 if (((dev->hw_features | dev->features) & 7146 NETIF_F_HW_VLAN_CTAG_FILTER) && 7147 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 7148 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 7149 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 7150 ret = -EINVAL; 7151 goto err_uninit; 7152 } 7153 7154 ret = -EBUSY; 7155 if (!dev->ifindex) 7156 dev->ifindex = dev_new_index(net); 7157 else if (__dev_get_by_index(net, dev->ifindex)) 7158 goto err_uninit; 7159 7160 /* Transfer changeable features to wanted_features and enable 7161 * software offloads (GSO and GRO). 7162 */ 7163 dev->hw_features |= NETIF_F_SOFT_FEATURES; 7164 dev->features |= NETIF_F_SOFT_FEATURES; 7165 dev->wanted_features = dev->features & dev->hw_features; 7166 7167 if (!(dev->flags & IFF_LOOPBACK)) 7168 dev->hw_features |= NETIF_F_NOCACHE_COPY; 7169 7170 /* If IPv4 TCP segmentation offload is supported we should also 7171 * allow the device to enable segmenting the frame with the option 7172 * of ignoring a static IP ID value. This doesn't enable the 7173 * feature itself but allows the user to enable it later. 7174 */ 7175 if (dev->hw_features & NETIF_F_TSO) 7176 dev->hw_features |= NETIF_F_TSO_MANGLEID; 7177 if (dev->vlan_features & NETIF_F_TSO) 7178 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 7179 if (dev->mpls_features & NETIF_F_TSO) 7180 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 7181 if (dev->hw_enc_features & NETIF_F_TSO) 7182 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 7183 7184 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 7185 */ 7186 dev->vlan_features |= NETIF_F_HIGHDMA; 7187 7188 /* Make NETIF_F_SG inheritable to tunnel devices. 7189 */ 7190 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 7191 7192 /* Make NETIF_F_SG inheritable to MPLS. 7193 */ 7194 dev->mpls_features |= NETIF_F_SG; 7195 7196 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 7197 ret = notifier_to_errno(ret); 7198 if (ret) 7199 goto err_uninit; 7200 7201 ret = netdev_register_kobject(dev); 7202 if (ret) 7203 goto err_uninit; 7204 dev->reg_state = NETREG_REGISTERED; 7205 7206 __netdev_update_features(dev); 7207 7208 /* 7209 * Default initial state at registry is that the 7210 * device is present. 7211 */ 7212 7213 set_bit(__LINK_STATE_PRESENT, &dev->state); 7214 7215 linkwatch_init_dev(dev); 7216 7217 dev_init_scheduler(dev); 7218 dev_hold(dev); 7219 list_netdevice(dev); 7220 add_device_randomness(dev->dev_addr, dev->addr_len); 7221 7222 /* If the device has permanent device address, driver should 7223 * set dev_addr and also addr_assign_type should be set to 7224 * NET_ADDR_PERM (default value). 7225 */ 7226 if (dev->addr_assign_type == NET_ADDR_PERM) 7227 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 7228 7229 /* Notify protocols, that a new device appeared. */ 7230 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 7231 ret = notifier_to_errno(ret); 7232 if (ret) { 7233 rollback_registered(dev); 7234 dev->reg_state = NETREG_UNREGISTERED; 7235 } 7236 /* 7237 * Prevent userspace races by waiting until the network 7238 * device is fully setup before sending notifications. 7239 */ 7240 if (!dev->rtnl_link_ops || 7241 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 7242 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 7243 7244 out: 7245 return ret; 7246 7247 err_uninit: 7248 if (dev->netdev_ops->ndo_uninit) 7249 dev->netdev_ops->ndo_uninit(dev); 7250 goto out; 7251 } 7252 EXPORT_SYMBOL(register_netdevice); 7253 7254 /** 7255 * init_dummy_netdev - init a dummy network device for NAPI 7256 * @dev: device to init 7257 * 7258 * This takes a network device structure and initialize the minimum 7259 * amount of fields so it can be used to schedule NAPI polls without 7260 * registering a full blown interface. This is to be used by drivers 7261 * that need to tie several hardware interfaces to a single NAPI 7262 * poll scheduler due to HW limitations. 7263 */ 7264 int init_dummy_netdev(struct net_device *dev) 7265 { 7266 /* Clear everything. Note we don't initialize spinlocks 7267 * are they aren't supposed to be taken by any of the 7268 * NAPI code and this dummy netdev is supposed to be 7269 * only ever used for NAPI polls 7270 */ 7271 memset(dev, 0, sizeof(struct net_device)); 7272 7273 /* make sure we BUG if trying to hit standard 7274 * register/unregister code path 7275 */ 7276 dev->reg_state = NETREG_DUMMY; 7277 7278 /* NAPI wants this */ 7279 INIT_LIST_HEAD(&dev->napi_list); 7280 7281 /* a dummy interface is started by default */ 7282 set_bit(__LINK_STATE_PRESENT, &dev->state); 7283 set_bit(__LINK_STATE_START, &dev->state); 7284 7285 /* Note : We dont allocate pcpu_refcnt for dummy devices, 7286 * because users of this 'device' dont need to change 7287 * its refcount. 7288 */ 7289 7290 return 0; 7291 } 7292 EXPORT_SYMBOL_GPL(init_dummy_netdev); 7293 7294 7295 /** 7296 * register_netdev - register a network device 7297 * @dev: device to register 7298 * 7299 * Take a completed network device structure and add it to the kernel 7300 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 7301 * chain. 0 is returned on success. A negative errno code is returned 7302 * on a failure to set up the device, or if the name is a duplicate. 7303 * 7304 * This is a wrapper around register_netdevice that takes the rtnl semaphore 7305 * and expands the device name if you passed a format string to 7306 * alloc_netdev. 7307 */ 7308 int register_netdev(struct net_device *dev) 7309 { 7310 int err; 7311 7312 rtnl_lock(); 7313 err = register_netdevice(dev); 7314 rtnl_unlock(); 7315 return err; 7316 } 7317 EXPORT_SYMBOL(register_netdev); 7318 7319 int netdev_refcnt_read(const struct net_device *dev) 7320 { 7321 int i, refcnt = 0; 7322 7323 for_each_possible_cpu(i) 7324 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 7325 return refcnt; 7326 } 7327 EXPORT_SYMBOL(netdev_refcnt_read); 7328 7329 /** 7330 * netdev_wait_allrefs - wait until all references are gone. 7331 * @dev: target net_device 7332 * 7333 * This is called when unregistering network devices. 7334 * 7335 * Any protocol or device that holds a reference should register 7336 * for netdevice notification, and cleanup and put back the 7337 * reference if they receive an UNREGISTER event. 7338 * We can get stuck here if buggy protocols don't correctly 7339 * call dev_put. 7340 */ 7341 static void netdev_wait_allrefs(struct net_device *dev) 7342 { 7343 unsigned long rebroadcast_time, warning_time; 7344 int refcnt; 7345 7346 linkwatch_forget_dev(dev); 7347 7348 rebroadcast_time = warning_time = jiffies; 7349 refcnt = netdev_refcnt_read(dev); 7350 7351 while (refcnt != 0) { 7352 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 7353 rtnl_lock(); 7354 7355 /* Rebroadcast unregister notification */ 7356 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7357 7358 __rtnl_unlock(); 7359 rcu_barrier(); 7360 rtnl_lock(); 7361 7362 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7363 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 7364 &dev->state)) { 7365 /* We must not have linkwatch events 7366 * pending on unregister. If this 7367 * happens, we simply run the queue 7368 * unscheduled, resulting in a noop 7369 * for this device. 7370 */ 7371 linkwatch_run_queue(); 7372 } 7373 7374 __rtnl_unlock(); 7375 7376 rebroadcast_time = jiffies; 7377 } 7378 7379 msleep(250); 7380 7381 refcnt = netdev_refcnt_read(dev); 7382 7383 if (time_after(jiffies, warning_time + 10 * HZ)) { 7384 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 7385 dev->name, refcnt); 7386 warning_time = jiffies; 7387 } 7388 } 7389 } 7390 7391 /* The sequence is: 7392 * 7393 * rtnl_lock(); 7394 * ... 7395 * register_netdevice(x1); 7396 * register_netdevice(x2); 7397 * ... 7398 * unregister_netdevice(y1); 7399 * unregister_netdevice(y2); 7400 * ... 7401 * rtnl_unlock(); 7402 * free_netdev(y1); 7403 * free_netdev(y2); 7404 * 7405 * We are invoked by rtnl_unlock(). 7406 * This allows us to deal with problems: 7407 * 1) We can delete sysfs objects which invoke hotplug 7408 * without deadlocking with linkwatch via keventd. 7409 * 2) Since we run with the RTNL semaphore not held, we can sleep 7410 * safely in order to wait for the netdev refcnt to drop to zero. 7411 * 7412 * We must not return until all unregister events added during 7413 * the interval the lock was held have been completed. 7414 */ 7415 void netdev_run_todo(void) 7416 { 7417 struct list_head list; 7418 7419 /* Snapshot list, allow later requests */ 7420 list_replace_init(&net_todo_list, &list); 7421 7422 __rtnl_unlock(); 7423 7424 7425 /* Wait for rcu callbacks to finish before next phase */ 7426 if (!list_empty(&list)) 7427 rcu_barrier(); 7428 7429 while (!list_empty(&list)) { 7430 struct net_device *dev 7431 = list_first_entry(&list, struct net_device, todo_list); 7432 list_del(&dev->todo_list); 7433 7434 rtnl_lock(); 7435 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7436 __rtnl_unlock(); 7437 7438 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 7439 pr_err("network todo '%s' but state %d\n", 7440 dev->name, dev->reg_state); 7441 dump_stack(); 7442 continue; 7443 } 7444 7445 dev->reg_state = NETREG_UNREGISTERED; 7446 7447 netdev_wait_allrefs(dev); 7448 7449 /* paranoia */ 7450 BUG_ON(netdev_refcnt_read(dev)); 7451 BUG_ON(!list_empty(&dev->ptype_all)); 7452 BUG_ON(!list_empty(&dev->ptype_specific)); 7453 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 7454 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 7455 WARN_ON(dev->dn_ptr); 7456 7457 if (dev->destructor) 7458 dev->destructor(dev); 7459 7460 /* Report a network device has been unregistered */ 7461 rtnl_lock(); 7462 dev_net(dev)->dev_unreg_count--; 7463 __rtnl_unlock(); 7464 wake_up(&netdev_unregistering_wq); 7465 7466 /* Free network device */ 7467 kobject_put(&dev->dev.kobj); 7468 } 7469 } 7470 7471 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 7472 * all the same fields in the same order as net_device_stats, with only 7473 * the type differing, but rtnl_link_stats64 may have additional fields 7474 * at the end for newer counters. 7475 */ 7476 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 7477 const struct net_device_stats *netdev_stats) 7478 { 7479 #if BITS_PER_LONG == 64 7480 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 7481 memcpy(stats64, netdev_stats, sizeof(*stats64)); 7482 /* zero out counters that only exist in rtnl_link_stats64 */ 7483 memset((char *)stats64 + sizeof(*netdev_stats), 0, 7484 sizeof(*stats64) - sizeof(*netdev_stats)); 7485 #else 7486 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 7487 const unsigned long *src = (const unsigned long *)netdev_stats; 7488 u64 *dst = (u64 *)stats64; 7489 7490 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 7491 for (i = 0; i < n; i++) 7492 dst[i] = src[i]; 7493 /* zero out counters that only exist in rtnl_link_stats64 */ 7494 memset((char *)stats64 + n * sizeof(u64), 0, 7495 sizeof(*stats64) - n * sizeof(u64)); 7496 #endif 7497 } 7498 EXPORT_SYMBOL(netdev_stats_to_stats64); 7499 7500 /** 7501 * dev_get_stats - get network device statistics 7502 * @dev: device to get statistics from 7503 * @storage: place to store stats 7504 * 7505 * Get network statistics from device. Return @storage. 7506 * The device driver may provide its own method by setting 7507 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 7508 * otherwise the internal statistics structure is used. 7509 */ 7510 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 7511 struct rtnl_link_stats64 *storage) 7512 { 7513 const struct net_device_ops *ops = dev->netdev_ops; 7514 7515 if (ops->ndo_get_stats64) { 7516 memset(storage, 0, sizeof(*storage)); 7517 ops->ndo_get_stats64(dev, storage); 7518 } else if (ops->ndo_get_stats) { 7519 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 7520 } else { 7521 netdev_stats_to_stats64(storage, &dev->stats); 7522 } 7523 storage->rx_dropped += atomic_long_read(&dev->rx_dropped); 7524 storage->tx_dropped += atomic_long_read(&dev->tx_dropped); 7525 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler); 7526 return storage; 7527 } 7528 EXPORT_SYMBOL(dev_get_stats); 7529 7530 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 7531 { 7532 struct netdev_queue *queue = dev_ingress_queue(dev); 7533 7534 #ifdef CONFIG_NET_CLS_ACT 7535 if (queue) 7536 return queue; 7537 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 7538 if (!queue) 7539 return NULL; 7540 netdev_init_one_queue(dev, queue, NULL); 7541 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 7542 queue->qdisc_sleeping = &noop_qdisc; 7543 rcu_assign_pointer(dev->ingress_queue, queue); 7544 #endif 7545 return queue; 7546 } 7547 7548 static const struct ethtool_ops default_ethtool_ops; 7549 7550 void netdev_set_default_ethtool_ops(struct net_device *dev, 7551 const struct ethtool_ops *ops) 7552 { 7553 if (dev->ethtool_ops == &default_ethtool_ops) 7554 dev->ethtool_ops = ops; 7555 } 7556 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 7557 7558 void netdev_freemem(struct net_device *dev) 7559 { 7560 char *addr = (char *)dev - dev->padded; 7561 7562 kvfree(addr); 7563 } 7564 7565 /** 7566 * alloc_netdev_mqs - allocate network device 7567 * @sizeof_priv: size of private data to allocate space for 7568 * @name: device name format string 7569 * @name_assign_type: origin of device name 7570 * @setup: callback to initialize device 7571 * @txqs: the number of TX subqueues to allocate 7572 * @rxqs: the number of RX subqueues to allocate 7573 * 7574 * Allocates a struct net_device with private data area for driver use 7575 * and performs basic initialization. Also allocates subqueue structs 7576 * for each queue on the device. 7577 */ 7578 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 7579 unsigned char name_assign_type, 7580 void (*setup)(struct net_device *), 7581 unsigned int txqs, unsigned int rxqs) 7582 { 7583 struct net_device *dev; 7584 size_t alloc_size; 7585 struct net_device *p; 7586 7587 BUG_ON(strlen(name) >= sizeof(dev->name)); 7588 7589 if (txqs < 1) { 7590 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 7591 return NULL; 7592 } 7593 7594 #ifdef CONFIG_SYSFS 7595 if (rxqs < 1) { 7596 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 7597 return NULL; 7598 } 7599 #endif 7600 7601 alloc_size = sizeof(struct net_device); 7602 if (sizeof_priv) { 7603 /* ensure 32-byte alignment of private area */ 7604 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 7605 alloc_size += sizeof_priv; 7606 } 7607 /* ensure 32-byte alignment of whole construct */ 7608 alloc_size += NETDEV_ALIGN - 1; 7609 7610 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 7611 if (!p) 7612 p = vzalloc(alloc_size); 7613 if (!p) 7614 return NULL; 7615 7616 dev = PTR_ALIGN(p, NETDEV_ALIGN); 7617 dev->padded = (char *)dev - (char *)p; 7618 7619 dev->pcpu_refcnt = alloc_percpu(int); 7620 if (!dev->pcpu_refcnt) 7621 goto free_dev; 7622 7623 if (dev_addr_init(dev)) 7624 goto free_pcpu; 7625 7626 dev_mc_init(dev); 7627 dev_uc_init(dev); 7628 7629 dev_net_set(dev, &init_net); 7630 7631 dev->gso_max_size = GSO_MAX_SIZE; 7632 dev->gso_max_segs = GSO_MAX_SEGS; 7633 7634 INIT_LIST_HEAD(&dev->napi_list); 7635 INIT_LIST_HEAD(&dev->unreg_list); 7636 INIT_LIST_HEAD(&dev->close_list); 7637 INIT_LIST_HEAD(&dev->link_watch_list); 7638 INIT_LIST_HEAD(&dev->adj_list.upper); 7639 INIT_LIST_HEAD(&dev->adj_list.lower); 7640 INIT_LIST_HEAD(&dev->ptype_all); 7641 INIT_LIST_HEAD(&dev->ptype_specific); 7642 #ifdef CONFIG_NET_SCHED 7643 hash_init(dev->qdisc_hash); 7644 #endif 7645 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 7646 setup(dev); 7647 7648 if (!dev->tx_queue_len) { 7649 dev->priv_flags |= IFF_NO_QUEUE; 7650 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 7651 } 7652 7653 dev->num_tx_queues = txqs; 7654 dev->real_num_tx_queues = txqs; 7655 if (netif_alloc_netdev_queues(dev)) 7656 goto free_all; 7657 7658 #ifdef CONFIG_SYSFS 7659 dev->num_rx_queues = rxqs; 7660 dev->real_num_rx_queues = rxqs; 7661 if (netif_alloc_rx_queues(dev)) 7662 goto free_all; 7663 #endif 7664 7665 strcpy(dev->name, name); 7666 dev->name_assign_type = name_assign_type; 7667 dev->group = INIT_NETDEV_GROUP; 7668 if (!dev->ethtool_ops) 7669 dev->ethtool_ops = &default_ethtool_ops; 7670 7671 nf_hook_ingress_init(dev); 7672 7673 return dev; 7674 7675 free_all: 7676 free_netdev(dev); 7677 return NULL; 7678 7679 free_pcpu: 7680 free_percpu(dev->pcpu_refcnt); 7681 free_dev: 7682 netdev_freemem(dev); 7683 return NULL; 7684 } 7685 EXPORT_SYMBOL(alloc_netdev_mqs); 7686 7687 /** 7688 * free_netdev - free network device 7689 * @dev: device 7690 * 7691 * This function does the last stage of destroying an allocated device 7692 * interface. The reference to the device object is released. If this 7693 * is the last reference then it will be freed.Must be called in process 7694 * context. 7695 */ 7696 void free_netdev(struct net_device *dev) 7697 { 7698 struct napi_struct *p, *n; 7699 7700 might_sleep(); 7701 netif_free_tx_queues(dev); 7702 #ifdef CONFIG_SYSFS 7703 kvfree(dev->_rx); 7704 #endif 7705 7706 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 7707 7708 /* Flush device addresses */ 7709 dev_addr_flush(dev); 7710 7711 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 7712 netif_napi_del(p); 7713 7714 free_percpu(dev->pcpu_refcnt); 7715 dev->pcpu_refcnt = NULL; 7716 7717 /* Compatibility with error handling in drivers */ 7718 if (dev->reg_state == NETREG_UNINITIALIZED) { 7719 netdev_freemem(dev); 7720 return; 7721 } 7722 7723 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 7724 dev->reg_state = NETREG_RELEASED; 7725 7726 /* will free via device release */ 7727 put_device(&dev->dev); 7728 } 7729 EXPORT_SYMBOL(free_netdev); 7730 7731 /** 7732 * synchronize_net - Synchronize with packet receive processing 7733 * 7734 * Wait for packets currently being received to be done. 7735 * Does not block later packets from starting. 7736 */ 7737 void synchronize_net(void) 7738 { 7739 might_sleep(); 7740 if (rtnl_is_locked()) 7741 synchronize_rcu_expedited(); 7742 else 7743 synchronize_rcu(); 7744 } 7745 EXPORT_SYMBOL(synchronize_net); 7746 7747 /** 7748 * unregister_netdevice_queue - remove device from the kernel 7749 * @dev: device 7750 * @head: list 7751 * 7752 * This function shuts down a device interface and removes it 7753 * from the kernel tables. 7754 * If head not NULL, device is queued to be unregistered later. 7755 * 7756 * Callers must hold the rtnl semaphore. You may want 7757 * unregister_netdev() instead of this. 7758 */ 7759 7760 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 7761 { 7762 ASSERT_RTNL(); 7763 7764 if (head) { 7765 list_move_tail(&dev->unreg_list, head); 7766 } else { 7767 rollback_registered(dev); 7768 /* Finish processing unregister after unlock */ 7769 net_set_todo(dev); 7770 } 7771 } 7772 EXPORT_SYMBOL(unregister_netdevice_queue); 7773 7774 /** 7775 * unregister_netdevice_many - unregister many devices 7776 * @head: list of devices 7777 * 7778 * Note: As most callers use a stack allocated list_head, 7779 * we force a list_del() to make sure stack wont be corrupted later. 7780 */ 7781 void unregister_netdevice_many(struct list_head *head) 7782 { 7783 struct net_device *dev; 7784 7785 if (!list_empty(head)) { 7786 rollback_registered_many(head); 7787 list_for_each_entry(dev, head, unreg_list) 7788 net_set_todo(dev); 7789 list_del(head); 7790 } 7791 } 7792 EXPORT_SYMBOL(unregister_netdevice_many); 7793 7794 /** 7795 * unregister_netdev - remove device from the kernel 7796 * @dev: device 7797 * 7798 * This function shuts down a device interface and removes it 7799 * from the kernel tables. 7800 * 7801 * This is just a wrapper for unregister_netdevice that takes 7802 * the rtnl semaphore. In general you want to use this and not 7803 * unregister_netdevice. 7804 */ 7805 void unregister_netdev(struct net_device *dev) 7806 { 7807 rtnl_lock(); 7808 unregister_netdevice(dev); 7809 rtnl_unlock(); 7810 } 7811 EXPORT_SYMBOL(unregister_netdev); 7812 7813 /** 7814 * dev_change_net_namespace - move device to different nethost namespace 7815 * @dev: device 7816 * @net: network namespace 7817 * @pat: If not NULL name pattern to try if the current device name 7818 * is already taken in the destination network namespace. 7819 * 7820 * This function shuts down a device interface and moves it 7821 * to a new network namespace. On success 0 is returned, on 7822 * a failure a netagive errno code is returned. 7823 * 7824 * Callers must hold the rtnl semaphore. 7825 */ 7826 7827 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 7828 { 7829 int err; 7830 7831 ASSERT_RTNL(); 7832 7833 /* Don't allow namespace local devices to be moved. */ 7834 err = -EINVAL; 7835 if (dev->features & NETIF_F_NETNS_LOCAL) 7836 goto out; 7837 7838 /* Ensure the device has been registrered */ 7839 if (dev->reg_state != NETREG_REGISTERED) 7840 goto out; 7841 7842 /* Get out if there is nothing todo */ 7843 err = 0; 7844 if (net_eq(dev_net(dev), net)) 7845 goto out; 7846 7847 /* Pick the destination device name, and ensure 7848 * we can use it in the destination network namespace. 7849 */ 7850 err = -EEXIST; 7851 if (__dev_get_by_name(net, dev->name)) { 7852 /* We get here if we can't use the current device name */ 7853 if (!pat) 7854 goto out; 7855 if (dev_get_valid_name(net, dev, pat) < 0) 7856 goto out; 7857 } 7858 7859 /* 7860 * And now a mini version of register_netdevice unregister_netdevice. 7861 */ 7862 7863 /* If device is running close it first. */ 7864 dev_close(dev); 7865 7866 /* And unlink it from device chain */ 7867 err = -ENODEV; 7868 unlist_netdevice(dev); 7869 7870 synchronize_net(); 7871 7872 /* Shutdown queueing discipline. */ 7873 dev_shutdown(dev); 7874 7875 /* Notify protocols, that we are about to destroy 7876 * this device. They should clean all the things. 7877 * 7878 * Note that dev->reg_state stays at NETREG_REGISTERED. 7879 * This is wanted because this way 8021q and macvlan know 7880 * the device is just moving and can keep their slaves up. 7881 */ 7882 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7883 rcu_barrier(); 7884 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7885 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); 7886 7887 /* 7888 * Flush the unicast and multicast chains 7889 */ 7890 dev_uc_flush(dev); 7891 dev_mc_flush(dev); 7892 7893 /* Send a netdev-removed uevent to the old namespace */ 7894 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 7895 netdev_adjacent_del_links(dev); 7896 7897 /* Actually switch the network namespace */ 7898 dev_net_set(dev, net); 7899 7900 /* If there is an ifindex conflict assign a new one */ 7901 if (__dev_get_by_index(net, dev->ifindex)) 7902 dev->ifindex = dev_new_index(net); 7903 7904 /* Send a netdev-add uevent to the new namespace */ 7905 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 7906 netdev_adjacent_add_links(dev); 7907 7908 /* Fixup kobjects */ 7909 err = device_rename(&dev->dev, dev->name); 7910 WARN_ON(err); 7911 7912 /* Add the device back in the hashes */ 7913 list_netdevice(dev); 7914 7915 /* Notify protocols, that a new device appeared. */ 7916 call_netdevice_notifiers(NETDEV_REGISTER, dev); 7917 7918 /* 7919 * Prevent userspace races by waiting until the network 7920 * device is fully setup before sending notifications. 7921 */ 7922 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 7923 7924 synchronize_net(); 7925 err = 0; 7926 out: 7927 return err; 7928 } 7929 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 7930 7931 static int dev_cpu_dead(unsigned int oldcpu) 7932 { 7933 struct sk_buff **list_skb; 7934 struct sk_buff *skb; 7935 unsigned int cpu; 7936 struct softnet_data *sd, *oldsd; 7937 7938 local_irq_disable(); 7939 cpu = smp_processor_id(); 7940 sd = &per_cpu(softnet_data, cpu); 7941 oldsd = &per_cpu(softnet_data, oldcpu); 7942 7943 /* Find end of our completion_queue. */ 7944 list_skb = &sd->completion_queue; 7945 while (*list_skb) 7946 list_skb = &(*list_skb)->next; 7947 /* Append completion queue from offline CPU. */ 7948 *list_skb = oldsd->completion_queue; 7949 oldsd->completion_queue = NULL; 7950 7951 /* Append output queue from offline CPU. */ 7952 if (oldsd->output_queue) { 7953 *sd->output_queue_tailp = oldsd->output_queue; 7954 sd->output_queue_tailp = oldsd->output_queue_tailp; 7955 oldsd->output_queue = NULL; 7956 oldsd->output_queue_tailp = &oldsd->output_queue; 7957 } 7958 /* Append NAPI poll list from offline CPU, with one exception : 7959 * process_backlog() must be called by cpu owning percpu backlog. 7960 * We properly handle process_queue & input_pkt_queue later. 7961 */ 7962 while (!list_empty(&oldsd->poll_list)) { 7963 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 7964 struct napi_struct, 7965 poll_list); 7966 7967 list_del_init(&napi->poll_list); 7968 if (napi->poll == process_backlog) 7969 napi->state = 0; 7970 else 7971 ____napi_schedule(sd, napi); 7972 } 7973 7974 raise_softirq_irqoff(NET_TX_SOFTIRQ); 7975 local_irq_enable(); 7976 7977 /* Process offline CPU's input_pkt_queue */ 7978 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 7979 netif_rx_ni(skb); 7980 input_queue_head_incr(oldsd); 7981 } 7982 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 7983 netif_rx_ni(skb); 7984 input_queue_head_incr(oldsd); 7985 } 7986 7987 return 0; 7988 } 7989 7990 /** 7991 * netdev_increment_features - increment feature set by one 7992 * @all: current feature set 7993 * @one: new feature set 7994 * @mask: mask feature set 7995 * 7996 * Computes a new feature set after adding a device with feature set 7997 * @one to the master device with current feature set @all. Will not 7998 * enable anything that is off in @mask. Returns the new feature set. 7999 */ 8000 netdev_features_t netdev_increment_features(netdev_features_t all, 8001 netdev_features_t one, netdev_features_t mask) 8002 { 8003 if (mask & NETIF_F_HW_CSUM) 8004 mask |= NETIF_F_CSUM_MASK; 8005 mask |= NETIF_F_VLAN_CHALLENGED; 8006 8007 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 8008 all &= one | ~NETIF_F_ALL_FOR_ALL; 8009 8010 /* If one device supports hw checksumming, set for all. */ 8011 if (all & NETIF_F_HW_CSUM) 8012 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 8013 8014 return all; 8015 } 8016 EXPORT_SYMBOL(netdev_increment_features); 8017 8018 static struct hlist_head * __net_init netdev_create_hash(void) 8019 { 8020 int i; 8021 struct hlist_head *hash; 8022 8023 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); 8024 if (hash != NULL) 8025 for (i = 0; i < NETDEV_HASHENTRIES; i++) 8026 INIT_HLIST_HEAD(&hash[i]); 8027 8028 return hash; 8029 } 8030 8031 /* Initialize per network namespace state */ 8032 static int __net_init netdev_init(struct net *net) 8033 { 8034 if (net != &init_net) 8035 INIT_LIST_HEAD(&net->dev_base_head); 8036 8037 net->dev_name_head = netdev_create_hash(); 8038 if (net->dev_name_head == NULL) 8039 goto err_name; 8040 8041 net->dev_index_head = netdev_create_hash(); 8042 if (net->dev_index_head == NULL) 8043 goto err_idx; 8044 8045 return 0; 8046 8047 err_idx: 8048 kfree(net->dev_name_head); 8049 err_name: 8050 return -ENOMEM; 8051 } 8052 8053 /** 8054 * netdev_drivername - network driver for the device 8055 * @dev: network device 8056 * 8057 * Determine network driver for device. 8058 */ 8059 const char *netdev_drivername(const struct net_device *dev) 8060 { 8061 const struct device_driver *driver; 8062 const struct device *parent; 8063 const char *empty = ""; 8064 8065 parent = dev->dev.parent; 8066 if (!parent) 8067 return empty; 8068 8069 driver = parent->driver; 8070 if (driver && driver->name) 8071 return driver->name; 8072 return empty; 8073 } 8074 8075 static void __netdev_printk(const char *level, const struct net_device *dev, 8076 struct va_format *vaf) 8077 { 8078 if (dev && dev->dev.parent) { 8079 dev_printk_emit(level[1] - '0', 8080 dev->dev.parent, 8081 "%s %s %s%s: %pV", 8082 dev_driver_string(dev->dev.parent), 8083 dev_name(dev->dev.parent), 8084 netdev_name(dev), netdev_reg_state(dev), 8085 vaf); 8086 } else if (dev) { 8087 printk("%s%s%s: %pV", 8088 level, netdev_name(dev), netdev_reg_state(dev), vaf); 8089 } else { 8090 printk("%s(NULL net_device): %pV", level, vaf); 8091 } 8092 } 8093 8094 void netdev_printk(const char *level, const struct net_device *dev, 8095 const char *format, ...) 8096 { 8097 struct va_format vaf; 8098 va_list args; 8099 8100 va_start(args, format); 8101 8102 vaf.fmt = format; 8103 vaf.va = &args; 8104 8105 __netdev_printk(level, dev, &vaf); 8106 8107 va_end(args); 8108 } 8109 EXPORT_SYMBOL(netdev_printk); 8110 8111 #define define_netdev_printk_level(func, level) \ 8112 void func(const struct net_device *dev, const char *fmt, ...) \ 8113 { \ 8114 struct va_format vaf; \ 8115 va_list args; \ 8116 \ 8117 va_start(args, fmt); \ 8118 \ 8119 vaf.fmt = fmt; \ 8120 vaf.va = &args; \ 8121 \ 8122 __netdev_printk(level, dev, &vaf); \ 8123 \ 8124 va_end(args); \ 8125 } \ 8126 EXPORT_SYMBOL(func); 8127 8128 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 8129 define_netdev_printk_level(netdev_alert, KERN_ALERT); 8130 define_netdev_printk_level(netdev_crit, KERN_CRIT); 8131 define_netdev_printk_level(netdev_err, KERN_ERR); 8132 define_netdev_printk_level(netdev_warn, KERN_WARNING); 8133 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 8134 define_netdev_printk_level(netdev_info, KERN_INFO); 8135 8136 static void __net_exit netdev_exit(struct net *net) 8137 { 8138 kfree(net->dev_name_head); 8139 kfree(net->dev_index_head); 8140 } 8141 8142 static struct pernet_operations __net_initdata netdev_net_ops = { 8143 .init = netdev_init, 8144 .exit = netdev_exit, 8145 }; 8146 8147 static void __net_exit default_device_exit(struct net *net) 8148 { 8149 struct net_device *dev, *aux; 8150 /* 8151 * Push all migratable network devices back to the 8152 * initial network namespace 8153 */ 8154 rtnl_lock(); 8155 for_each_netdev_safe(net, dev, aux) { 8156 int err; 8157 char fb_name[IFNAMSIZ]; 8158 8159 /* Ignore unmoveable devices (i.e. loopback) */ 8160 if (dev->features & NETIF_F_NETNS_LOCAL) 8161 continue; 8162 8163 /* Leave virtual devices for the generic cleanup */ 8164 if (dev->rtnl_link_ops) 8165 continue; 8166 8167 /* Push remaining network devices to init_net */ 8168 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 8169 err = dev_change_net_namespace(dev, &init_net, fb_name); 8170 if (err) { 8171 pr_emerg("%s: failed to move %s to init_net: %d\n", 8172 __func__, dev->name, err); 8173 BUG(); 8174 } 8175 } 8176 rtnl_unlock(); 8177 } 8178 8179 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 8180 { 8181 /* Return with the rtnl_lock held when there are no network 8182 * devices unregistering in any network namespace in net_list. 8183 */ 8184 struct net *net; 8185 bool unregistering; 8186 DEFINE_WAIT_FUNC(wait, woken_wake_function); 8187 8188 add_wait_queue(&netdev_unregistering_wq, &wait); 8189 for (;;) { 8190 unregistering = false; 8191 rtnl_lock(); 8192 list_for_each_entry(net, net_list, exit_list) { 8193 if (net->dev_unreg_count > 0) { 8194 unregistering = true; 8195 break; 8196 } 8197 } 8198 if (!unregistering) 8199 break; 8200 __rtnl_unlock(); 8201 8202 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 8203 } 8204 remove_wait_queue(&netdev_unregistering_wq, &wait); 8205 } 8206 8207 static void __net_exit default_device_exit_batch(struct list_head *net_list) 8208 { 8209 /* At exit all network devices most be removed from a network 8210 * namespace. Do this in the reverse order of registration. 8211 * Do this across as many network namespaces as possible to 8212 * improve batching efficiency. 8213 */ 8214 struct net_device *dev; 8215 struct net *net; 8216 LIST_HEAD(dev_kill_list); 8217 8218 /* To prevent network device cleanup code from dereferencing 8219 * loopback devices or network devices that have been freed 8220 * wait here for all pending unregistrations to complete, 8221 * before unregistring the loopback device and allowing the 8222 * network namespace be freed. 8223 * 8224 * The netdev todo list containing all network devices 8225 * unregistrations that happen in default_device_exit_batch 8226 * will run in the rtnl_unlock() at the end of 8227 * default_device_exit_batch. 8228 */ 8229 rtnl_lock_unregistering(net_list); 8230 list_for_each_entry(net, net_list, exit_list) { 8231 for_each_netdev_reverse(net, dev) { 8232 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 8233 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 8234 else 8235 unregister_netdevice_queue(dev, &dev_kill_list); 8236 } 8237 } 8238 unregister_netdevice_many(&dev_kill_list); 8239 rtnl_unlock(); 8240 } 8241 8242 static struct pernet_operations __net_initdata default_device_ops = { 8243 .exit = default_device_exit, 8244 .exit_batch = default_device_exit_batch, 8245 }; 8246 8247 /* 8248 * Initialize the DEV module. At boot time this walks the device list and 8249 * unhooks any devices that fail to initialise (normally hardware not 8250 * present) and leaves us with a valid list of present and active devices. 8251 * 8252 */ 8253 8254 /* 8255 * This is called single threaded during boot, so no need 8256 * to take the rtnl semaphore. 8257 */ 8258 static int __init net_dev_init(void) 8259 { 8260 int i, rc = -ENOMEM; 8261 8262 BUG_ON(!dev_boot_phase); 8263 8264 if (dev_proc_init()) 8265 goto out; 8266 8267 if (netdev_kobject_init()) 8268 goto out; 8269 8270 INIT_LIST_HEAD(&ptype_all); 8271 for (i = 0; i < PTYPE_HASH_SIZE; i++) 8272 INIT_LIST_HEAD(&ptype_base[i]); 8273 8274 INIT_LIST_HEAD(&offload_base); 8275 8276 if (register_pernet_subsys(&netdev_net_ops)) 8277 goto out; 8278 8279 /* 8280 * Initialise the packet receive queues. 8281 */ 8282 8283 for_each_possible_cpu(i) { 8284 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 8285 struct softnet_data *sd = &per_cpu(softnet_data, i); 8286 8287 INIT_WORK(flush, flush_backlog); 8288 8289 skb_queue_head_init(&sd->input_pkt_queue); 8290 skb_queue_head_init(&sd->process_queue); 8291 INIT_LIST_HEAD(&sd->poll_list); 8292 sd->output_queue_tailp = &sd->output_queue; 8293 #ifdef CONFIG_RPS 8294 sd->csd.func = rps_trigger_softirq; 8295 sd->csd.info = sd; 8296 sd->cpu = i; 8297 #endif 8298 8299 sd->backlog.poll = process_backlog; 8300 sd->backlog.weight = weight_p; 8301 } 8302 8303 dev_boot_phase = 0; 8304 8305 /* The loopback device is special if any other network devices 8306 * is present in a network namespace the loopback device must 8307 * be present. Since we now dynamically allocate and free the 8308 * loopback device ensure this invariant is maintained by 8309 * keeping the loopback device as the first device on the 8310 * list of network devices. Ensuring the loopback devices 8311 * is the first device that appears and the last network device 8312 * that disappears. 8313 */ 8314 if (register_pernet_device(&loopback_net_ops)) 8315 goto out; 8316 8317 if (register_pernet_device(&default_device_ops)) 8318 goto out; 8319 8320 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 8321 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 8322 8323 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 8324 NULL, dev_cpu_dead); 8325 WARN_ON(rc < 0); 8326 dst_subsys_init(); 8327 rc = 0; 8328 out: 8329 return rc; 8330 } 8331 8332 subsys_initcall(net_dev_init); 8333