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/sched/mm.h> 85 #include <linux/mutex.h> 86 #include <linux/string.h> 87 #include <linux/mm.h> 88 #include <linux/socket.h> 89 #include <linux/sockios.h> 90 #include <linux/errno.h> 91 #include <linux/interrupt.h> 92 #include <linux/if_ether.h> 93 #include <linux/netdevice.h> 94 #include <linux/etherdevice.h> 95 #include <linux/ethtool.h> 96 #include <linux/notifier.h> 97 #include <linux/skbuff.h> 98 #include <linux/bpf.h> 99 #include <linux/bpf_trace.h> 100 #include <net/net_namespace.h> 101 #include <net/sock.h> 102 #include <net/busy_poll.h> 103 #include <linux/rtnetlink.h> 104 #include <linux/stat.h> 105 #include <net/dst.h> 106 #include <net/dst_metadata.h> 107 #include <net/pkt_sched.h> 108 #include <net/pkt_cls.h> 109 #include <net/checksum.h> 110 #include <net/xfrm.h> 111 #include <linux/highmem.h> 112 #include <linux/init.h> 113 #include <linux/module.h> 114 #include <linux/netpoll.h> 115 #include <linux/rcupdate.h> 116 #include <linux/delay.h> 117 #include <net/iw_handler.h> 118 #include <asm/current.h> 119 #include <linux/audit.h> 120 #include <linux/dmaengine.h> 121 #include <linux/err.h> 122 #include <linux/ctype.h> 123 #include <linux/if_arp.h> 124 #include <linux/if_vlan.h> 125 #include <linux/ip.h> 126 #include <net/ip.h> 127 #include <net/mpls.h> 128 #include <linux/ipv6.h> 129 #include <linux/in.h> 130 #include <linux/jhash.h> 131 #include <linux/random.h> 132 #include <trace/events/napi.h> 133 #include <trace/events/net.h> 134 #include <trace/events/skb.h> 135 #include <linux/pci.h> 136 #include <linux/inetdevice.h> 137 #include <linux/cpu_rmap.h> 138 #include <linux/static_key.h> 139 #include <linux/hashtable.h> 140 #include <linux/vmalloc.h> 141 #include <linux/if_macvlan.h> 142 #include <linux/errqueue.h> 143 #include <linux/hrtimer.h> 144 #include <linux/netfilter_ingress.h> 145 #include <linux/crash_dump.h> 146 #include <linux/sctp.h> 147 #include <net/udp_tunnel.h> 148 #include <linux/net_namespace.h> 149 150 #include "net-sysfs.h" 151 152 /* Instead of increasing this, you should create a hash table. */ 153 #define MAX_GRO_SKBS 8 154 155 /* This should be increased if a protocol with a bigger head is added. */ 156 #define GRO_MAX_HEAD (MAX_HEADER + 128) 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 static DEFINE_SPINLOCK(offload_lock); 160 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 161 struct list_head ptype_all __read_mostly; /* Taps */ 162 static struct list_head offload_base __read_mostly; 163 164 static int netif_rx_internal(struct sk_buff *skb); 165 static int call_netdevice_notifiers_info(unsigned long val, 166 struct netdev_notifier_info *info); 167 static struct napi_struct *napi_by_id(unsigned int napi_id); 168 169 /* 170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 171 * semaphore. 172 * 173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 174 * 175 * Writers must hold the rtnl semaphore while they loop through the 176 * dev_base_head list, and hold dev_base_lock for writing when they do the 177 * actual updates. This allows pure readers to access the list even 178 * while a writer is preparing to update it. 179 * 180 * To put it another way, dev_base_lock is held for writing only to 181 * protect against pure readers; the rtnl semaphore provides the 182 * protection against other writers. 183 * 184 * See, for example usages, register_netdevice() and 185 * unregister_netdevice(), which must be called with the rtnl 186 * semaphore held. 187 */ 188 DEFINE_RWLOCK(dev_base_lock); 189 EXPORT_SYMBOL(dev_base_lock); 190 191 static DEFINE_MUTEX(ifalias_mutex); 192 193 /* protects napi_hash addition/deletion and napi_gen_id */ 194 static DEFINE_SPINLOCK(napi_hash_lock); 195 196 static unsigned int napi_gen_id = NR_CPUS; 197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 198 199 static seqcount_t devnet_rename_seq; 200 201 static inline void dev_base_seq_inc(struct net *net) 202 { 203 while (++net->dev_base_seq == 0) 204 ; 205 } 206 207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 208 { 209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 210 211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 212 } 213 214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 215 { 216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 217 } 218 219 static inline void rps_lock(struct softnet_data *sd) 220 { 221 #ifdef CONFIG_RPS 222 spin_lock(&sd->input_pkt_queue.lock); 223 #endif 224 } 225 226 static inline void rps_unlock(struct softnet_data *sd) 227 { 228 #ifdef CONFIG_RPS 229 spin_unlock(&sd->input_pkt_queue.lock); 230 #endif 231 } 232 233 /* Device list insertion */ 234 static void list_netdevice(struct net_device *dev) 235 { 236 struct net *net = dev_net(dev); 237 238 ASSERT_RTNL(); 239 240 write_lock_bh(&dev_base_lock); 241 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 242 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 243 hlist_add_head_rcu(&dev->index_hlist, 244 dev_index_hash(net, dev->ifindex)); 245 write_unlock_bh(&dev_base_lock); 246 247 dev_base_seq_inc(net); 248 } 249 250 /* Device list removal 251 * caller must respect a RCU grace period before freeing/reusing dev 252 */ 253 static void unlist_netdevice(struct net_device *dev) 254 { 255 ASSERT_RTNL(); 256 257 /* Unlink dev from the device chain */ 258 write_lock_bh(&dev_base_lock); 259 list_del_rcu(&dev->dev_list); 260 hlist_del_rcu(&dev->name_hlist); 261 hlist_del_rcu(&dev->index_hlist); 262 write_unlock_bh(&dev_base_lock); 263 264 dev_base_seq_inc(dev_net(dev)); 265 } 266 267 /* 268 * Our notifier list 269 */ 270 271 static RAW_NOTIFIER_HEAD(netdev_chain); 272 273 /* 274 * Device drivers call our routines to queue packets here. We empty the 275 * queue in the local softnet handler. 276 */ 277 278 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 279 EXPORT_PER_CPU_SYMBOL(softnet_data); 280 281 #ifdef CONFIG_LOCKDEP 282 /* 283 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 284 * according to dev->type 285 */ 286 static const unsigned short netdev_lock_type[] = { 287 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 288 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 289 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 290 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 291 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 292 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 293 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 294 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 295 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 296 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 297 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 298 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 299 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 300 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 301 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 302 303 static const char *const netdev_lock_name[] = { 304 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 305 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 306 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 307 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 308 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 309 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 310 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 311 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 312 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 313 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 314 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 315 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 316 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 317 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 318 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 319 320 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 321 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 322 323 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 324 { 325 int i; 326 327 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 328 if (netdev_lock_type[i] == dev_type) 329 return i; 330 /* the last key is used by default */ 331 return ARRAY_SIZE(netdev_lock_type) - 1; 332 } 333 334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 335 unsigned short dev_type) 336 { 337 int i; 338 339 i = netdev_lock_pos(dev_type); 340 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 341 netdev_lock_name[i]); 342 } 343 344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 345 { 346 int i; 347 348 i = netdev_lock_pos(dev->type); 349 lockdep_set_class_and_name(&dev->addr_list_lock, 350 &netdev_addr_lock_key[i], 351 netdev_lock_name[i]); 352 } 353 #else 354 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 355 unsigned short dev_type) 356 { 357 } 358 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 359 { 360 } 361 #endif 362 363 /******************************************************************************* 364 * 365 * Protocol management and registration routines 366 * 367 *******************************************************************************/ 368 369 370 /* 371 * Add a protocol ID to the list. Now that the input handler is 372 * smarter we can dispense with all the messy stuff that used to be 373 * here. 374 * 375 * BEWARE!!! Protocol handlers, mangling input packets, 376 * MUST BE last in hash buckets and checking protocol handlers 377 * MUST start from promiscuous ptype_all chain in net_bh. 378 * It is true now, do not change it. 379 * Explanation follows: if protocol handler, mangling packet, will 380 * be the first on list, it is not able to sense, that packet 381 * is cloned and should be copied-on-write, so that it will 382 * change it and subsequent readers will get broken packet. 383 * --ANK (980803) 384 */ 385 386 static inline struct list_head *ptype_head(const struct packet_type *pt) 387 { 388 if (pt->type == htons(ETH_P_ALL)) 389 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 390 else 391 return pt->dev ? &pt->dev->ptype_specific : 392 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 393 } 394 395 /** 396 * dev_add_pack - add packet handler 397 * @pt: packet type declaration 398 * 399 * Add a protocol handler to the networking stack. The passed &packet_type 400 * is linked into kernel lists and may not be freed until it has been 401 * removed from the kernel lists. 402 * 403 * This call does not sleep therefore it can not 404 * guarantee all CPU's that are in middle of receiving packets 405 * will see the new packet type (until the next received packet). 406 */ 407 408 void dev_add_pack(struct packet_type *pt) 409 { 410 struct list_head *head = ptype_head(pt); 411 412 spin_lock(&ptype_lock); 413 list_add_rcu(&pt->list, head); 414 spin_unlock(&ptype_lock); 415 } 416 EXPORT_SYMBOL(dev_add_pack); 417 418 /** 419 * __dev_remove_pack - remove packet handler 420 * @pt: packet type declaration 421 * 422 * Remove a protocol handler that was previously added to the kernel 423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 424 * from the kernel lists and can be freed or reused once this function 425 * returns. 426 * 427 * The packet type might still be in use by receivers 428 * and must not be freed until after all the CPU's have gone 429 * through a quiescent state. 430 */ 431 void __dev_remove_pack(struct packet_type *pt) 432 { 433 struct list_head *head = ptype_head(pt); 434 struct packet_type *pt1; 435 436 spin_lock(&ptype_lock); 437 438 list_for_each_entry(pt1, head, list) { 439 if (pt == pt1) { 440 list_del_rcu(&pt->list); 441 goto out; 442 } 443 } 444 445 pr_warn("dev_remove_pack: %p not found\n", pt); 446 out: 447 spin_unlock(&ptype_lock); 448 } 449 EXPORT_SYMBOL(__dev_remove_pack); 450 451 /** 452 * dev_remove_pack - remove packet handler 453 * @pt: packet type declaration 454 * 455 * Remove a protocol handler that was previously added to the kernel 456 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 457 * from the kernel lists and can be freed or reused once this function 458 * returns. 459 * 460 * This call sleeps to guarantee that no CPU is looking at the packet 461 * type after return. 462 */ 463 void dev_remove_pack(struct packet_type *pt) 464 { 465 __dev_remove_pack(pt); 466 467 synchronize_net(); 468 } 469 EXPORT_SYMBOL(dev_remove_pack); 470 471 472 /** 473 * dev_add_offload - register offload handlers 474 * @po: protocol offload declaration 475 * 476 * Add protocol offload handlers to the networking stack. The passed 477 * &proto_offload is linked into kernel lists and may not be freed until 478 * it has been removed from the kernel lists. 479 * 480 * This call does not sleep therefore it can not 481 * guarantee all CPU's that are in middle of receiving packets 482 * will see the new offload handlers (until the next received packet). 483 */ 484 void dev_add_offload(struct packet_offload *po) 485 { 486 struct packet_offload *elem; 487 488 spin_lock(&offload_lock); 489 list_for_each_entry(elem, &offload_base, list) { 490 if (po->priority < elem->priority) 491 break; 492 } 493 list_add_rcu(&po->list, elem->list.prev); 494 spin_unlock(&offload_lock); 495 } 496 EXPORT_SYMBOL(dev_add_offload); 497 498 /** 499 * __dev_remove_offload - remove offload handler 500 * @po: packet offload declaration 501 * 502 * Remove a protocol offload handler that was previously added to the 503 * kernel offload handlers by dev_add_offload(). The passed &offload_type 504 * is removed from the kernel lists and can be freed or reused once this 505 * function returns. 506 * 507 * The packet type might still be in use by receivers 508 * and must not be freed until after all the CPU's have gone 509 * through a quiescent state. 510 */ 511 static void __dev_remove_offload(struct packet_offload *po) 512 { 513 struct list_head *head = &offload_base; 514 struct packet_offload *po1; 515 516 spin_lock(&offload_lock); 517 518 list_for_each_entry(po1, head, list) { 519 if (po == po1) { 520 list_del_rcu(&po->list); 521 goto out; 522 } 523 } 524 525 pr_warn("dev_remove_offload: %p not found\n", po); 526 out: 527 spin_unlock(&offload_lock); 528 } 529 530 /** 531 * dev_remove_offload - remove packet offload handler 532 * @po: packet offload declaration 533 * 534 * Remove a packet offload handler that was previously added to the kernel 535 * offload handlers by dev_add_offload(). The passed &offload_type is 536 * removed from the kernel lists and can be freed or reused once this 537 * function returns. 538 * 539 * This call sleeps to guarantee that no CPU is looking at the packet 540 * type after return. 541 */ 542 void dev_remove_offload(struct packet_offload *po) 543 { 544 __dev_remove_offload(po); 545 546 synchronize_net(); 547 } 548 EXPORT_SYMBOL(dev_remove_offload); 549 550 /****************************************************************************** 551 * 552 * Device Boot-time Settings Routines 553 * 554 ******************************************************************************/ 555 556 /* Boot time configuration table */ 557 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 558 559 /** 560 * netdev_boot_setup_add - add new setup entry 561 * @name: name of the device 562 * @map: configured settings for the device 563 * 564 * Adds new setup entry to the dev_boot_setup list. The function 565 * returns 0 on error and 1 on success. This is a generic routine to 566 * all netdevices. 567 */ 568 static int netdev_boot_setup_add(char *name, struct ifmap *map) 569 { 570 struct netdev_boot_setup *s; 571 int i; 572 573 s = dev_boot_setup; 574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 575 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 576 memset(s[i].name, 0, sizeof(s[i].name)); 577 strlcpy(s[i].name, name, IFNAMSIZ); 578 memcpy(&s[i].map, map, sizeof(s[i].map)); 579 break; 580 } 581 } 582 583 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 584 } 585 586 /** 587 * netdev_boot_setup_check - check boot time settings 588 * @dev: the netdevice 589 * 590 * Check boot time settings for the device. 591 * The found settings are set for the device to be used 592 * later in the device probing. 593 * Returns 0 if no settings found, 1 if they are. 594 */ 595 int netdev_boot_setup_check(struct net_device *dev) 596 { 597 struct netdev_boot_setup *s = dev_boot_setup; 598 int i; 599 600 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 601 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 602 !strcmp(dev->name, s[i].name)) { 603 dev->irq = s[i].map.irq; 604 dev->base_addr = s[i].map.base_addr; 605 dev->mem_start = s[i].map.mem_start; 606 dev->mem_end = s[i].map.mem_end; 607 return 1; 608 } 609 } 610 return 0; 611 } 612 EXPORT_SYMBOL(netdev_boot_setup_check); 613 614 615 /** 616 * netdev_boot_base - get address from boot time settings 617 * @prefix: prefix for network device 618 * @unit: id for network device 619 * 620 * Check boot time settings for the base address of device. 621 * The found settings are set for the device to be used 622 * later in the device probing. 623 * Returns 0 if no settings found. 624 */ 625 unsigned long netdev_boot_base(const char *prefix, int unit) 626 { 627 const struct netdev_boot_setup *s = dev_boot_setup; 628 char name[IFNAMSIZ]; 629 int i; 630 631 sprintf(name, "%s%d", prefix, unit); 632 633 /* 634 * If device already registered then return base of 1 635 * to indicate not to probe for this interface 636 */ 637 if (__dev_get_by_name(&init_net, name)) 638 return 1; 639 640 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 641 if (!strcmp(name, s[i].name)) 642 return s[i].map.base_addr; 643 return 0; 644 } 645 646 /* 647 * Saves at boot time configured settings for any netdevice. 648 */ 649 int __init netdev_boot_setup(char *str) 650 { 651 int ints[5]; 652 struct ifmap map; 653 654 str = get_options(str, ARRAY_SIZE(ints), ints); 655 if (!str || !*str) 656 return 0; 657 658 /* Save settings */ 659 memset(&map, 0, sizeof(map)); 660 if (ints[0] > 0) 661 map.irq = ints[1]; 662 if (ints[0] > 1) 663 map.base_addr = ints[2]; 664 if (ints[0] > 2) 665 map.mem_start = ints[3]; 666 if (ints[0] > 3) 667 map.mem_end = ints[4]; 668 669 /* Add new entry to the list */ 670 return netdev_boot_setup_add(str, &map); 671 } 672 673 __setup("netdev=", netdev_boot_setup); 674 675 /******************************************************************************* 676 * 677 * Device Interface Subroutines 678 * 679 *******************************************************************************/ 680 681 /** 682 * dev_get_iflink - get 'iflink' value of a interface 683 * @dev: targeted interface 684 * 685 * Indicates the ifindex the interface is linked to. 686 * Physical interfaces have the same 'ifindex' and 'iflink' values. 687 */ 688 689 int dev_get_iflink(const struct net_device *dev) 690 { 691 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 692 return dev->netdev_ops->ndo_get_iflink(dev); 693 694 return dev->ifindex; 695 } 696 EXPORT_SYMBOL(dev_get_iflink); 697 698 /** 699 * dev_fill_metadata_dst - Retrieve tunnel egress information. 700 * @dev: targeted interface 701 * @skb: The packet. 702 * 703 * For better visibility of tunnel traffic OVS needs to retrieve 704 * egress tunnel information for a packet. Following API allows 705 * user to get this info. 706 */ 707 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 708 { 709 struct ip_tunnel_info *info; 710 711 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 712 return -EINVAL; 713 714 info = skb_tunnel_info_unclone(skb); 715 if (!info) 716 return -ENOMEM; 717 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 718 return -EINVAL; 719 720 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 721 } 722 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 723 724 /** 725 * __dev_get_by_name - find a device by its name 726 * @net: the applicable net namespace 727 * @name: name to find 728 * 729 * Find an interface by name. Must be called under RTNL semaphore 730 * or @dev_base_lock. If the name is found a pointer to the device 731 * is returned. If the name is not found then %NULL is returned. The 732 * reference counters are not incremented so the caller must be 733 * careful with locks. 734 */ 735 736 struct net_device *__dev_get_by_name(struct net *net, const char *name) 737 { 738 struct net_device *dev; 739 struct hlist_head *head = dev_name_hash(net, name); 740 741 hlist_for_each_entry(dev, head, name_hlist) 742 if (!strncmp(dev->name, name, IFNAMSIZ)) 743 return dev; 744 745 return NULL; 746 } 747 EXPORT_SYMBOL(__dev_get_by_name); 748 749 /** 750 * dev_get_by_name_rcu - find a device by its name 751 * @net: the applicable net namespace 752 * @name: name to find 753 * 754 * Find an interface by name. 755 * If the name is found a pointer to the device is returned. 756 * If the name is not found then %NULL is returned. 757 * The reference counters are not incremented so the caller must be 758 * careful with locks. The caller must hold RCU lock. 759 */ 760 761 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 762 { 763 struct net_device *dev; 764 struct hlist_head *head = dev_name_hash(net, name); 765 766 hlist_for_each_entry_rcu(dev, head, name_hlist) 767 if (!strncmp(dev->name, name, IFNAMSIZ)) 768 return dev; 769 770 return NULL; 771 } 772 EXPORT_SYMBOL(dev_get_by_name_rcu); 773 774 /** 775 * dev_get_by_name - find a device by its name 776 * @net: the applicable net namespace 777 * @name: name to find 778 * 779 * Find an interface by name. This can be called from any 780 * context and does its own locking. The returned handle has 781 * the usage count incremented and the caller must use dev_put() to 782 * release it when it is no longer needed. %NULL is returned if no 783 * matching device is found. 784 */ 785 786 struct net_device *dev_get_by_name(struct net *net, const char *name) 787 { 788 struct net_device *dev; 789 790 rcu_read_lock(); 791 dev = dev_get_by_name_rcu(net, name); 792 if (dev) 793 dev_hold(dev); 794 rcu_read_unlock(); 795 return dev; 796 } 797 EXPORT_SYMBOL(dev_get_by_name); 798 799 /** 800 * __dev_get_by_index - find a device by its ifindex 801 * @net: the applicable net namespace 802 * @ifindex: index of device 803 * 804 * Search for an interface by index. Returns %NULL if the device 805 * is not found or a pointer to the device. The device has not 806 * had its reference counter increased so the caller must be careful 807 * about locking. The caller must hold either the RTNL semaphore 808 * or @dev_base_lock. 809 */ 810 811 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 812 { 813 struct net_device *dev; 814 struct hlist_head *head = dev_index_hash(net, ifindex); 815 816 hlist_for_each_entry(dev, head, index_hlist) 817 if (dev->ifindex == ifindex) 818 return dev; 819 820 return NULL; 821 } 822 EXPORT_SYMBOL(__dev_get_by_index); 823 824 /** 825 * dev_get_by_index_rcu - find a device by its ifindex 826 * @net: the applicable net namespace 827 * @ifindex: index of device 828 * 829 * Search for an interface by index. Returns %NULL if the device 830 * is not found or a pointer to the device. The device has not 831 * had its reference counter increased so the caller must be careful 832 * about locking. The caller must hold RCU lock. 833 */ 834 835 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 836 { 837 struct net_device *dev; 838 struct hlist_head *head = dev_index_hash(net, ifindex); 839 840 hlist_for_each_entry_rcu(dev, head, index_hlist) 841 if (dev->ifindex == ifindex) 842 return dev; 843 844 return NULL; 845 } 846 EXPORT_SYMBOL(dev_get_by_index_rcu); 847 848 849 /** 850 * dev_get_by_index - find a device by its ifindex 851 * @net: the applicable net namespace 852 * @ifindex: index of device 853 * 854 * Search for an interface by index. Returns NULL if the device 855 * is not found or a pointer to the device. The device returned has 856 * had a reference added and the pointer is safe until the user calls 857 * dev_put to indicate they have finished with it. 858 */ 859 860 struct net_device *dev_get_by_index(struct net *net, int ifindex) 861 { 862 struct net_device *dev; 863 864 rcu_read_lock(); 865 dev = dev_get_by_index_rcu(net, ifindex); 866 if (dev) 867 dev_hold(dev); 868 rcu_read_unlock(); 869 return dev; 870 } 871 EXPORT_SYMBOL(dev_get_by_index); 872 873 /** 874 * dev_get_by_napi_id - find a device by napi_id 875 * @napi_id: ID of the NAPI struct 876 * 877 * Search for an interface by NAPI ID. Returns %NULL if the device 878 * is not found or a pointer to the device. The device has not had 879 * its reference counter increased so the caller must be careful 880 * about locking. The caller must hold RCU lock. 881 */ 882 883 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 884 { 885 struct napi_struct *napi; 886 887 WARN_ON_ONCE(!rcu_read_lock_held()); 888 889 if (napi_id < MIN_NAPI_ID) 890 return NULL; 891 892 napi = napi_by_id(napi_id); 893 894 return napi ? napi->dev : NULL; 895 } 896 EXPORT_SYMBOL(dev_get_by_napi_id); 897 898 /** 899 * netdev_get_name - get a netdevice name, knowing its ifindex. 900 * @net: network namespace 901 * @name: a pointer to the buffer where the name will be stored. 902 * @ifindex: the ifindex of the interface to get the name from. 903 * 904 * The use of raw_seqcount_begin() and cond_resched() before 905 * retrying is required as we want to give the writers a chance 906 * to complete when CONFIG_PREEMPT is not set. 907 */ 908 int netdev_get_name(struct net *net, char *name, int ifindex) 909 { 910 struct net_device *dev; 911 unsigned int seq; 912 913 retry: 914 seq = raw_seqcount_begin(&devnet_rename_seq); 915 rcu_read_lock(); 916 dev = dev_get_by_index_rcu(net, ifindex); 917 if (!dev) { 918 rcu_read_unlock(); 919 return -ENODEV; 920 } 921 922 strcpy(name, dev->name); 923 rcu_read_unlock(); 924 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 925 cond_resched(); 926 goto retry; 927 } 928 929 return 0; 930 } 931 932 /** 933 * dev_getbyhwaddr_rcu - find a device by its hardware address 934 * @net: the applicable net namespace 935 * @type: media type of device 936 * @ha: hardware address 937 * 938 * Search for an interface by MAC address. Returns NULL if the device 939 * is not found or a pointer to the device. 940 * The caller must hold RCU or RTNL. 941 * The returned device has not had its ref count increased 942 * and the caller must therefore be careful about locking 943 * 944 */ 945 946 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 947 const char *ha) 948 { 949 struct net_device *dev; 950 951 for_each_netdev_rcu(net, dev) 952 if (dev->type == type && 953 !memcmp(dev->dev_addr, ha, dev->addr_len)) 954 return dev; 955 956 return NULL; 957 } 958 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 959 960 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 961 { 962 struct net_device *dev; 963 964 ASSERT_RTNL(); 965 for_each_netdev(net, dev) 966 if (dev->type == type) 967 return dev; 968 969 return NULL; 970 } 971 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 972 973 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 974 { 975 struct net_device *dev, *ret = NULL; 976 977 rcu_read_lock(); 978 for_each_netdev_rcu(net, dev) 979 if (dev->type == type) { 980 dev_hold(dev); 981 ret = dev; 982 break; 983 } 984 rcu_read_unlock(); 985 return ret; 986 } 987 EXPORT_SYMBOL(dev_getfirstbyhwtype); 988 989 /** 990 * __dev_get_by_flags - find any device with given flags 991 * @net: the applicable net namespace 992 * @if_flags: IFF_* values 993 * @mask: bitmask of bits in if_flags to check 994 * 995 * Search for any interface with the given flags. Returns NULL if a device 996 * is not found or a pointer to the device. Must be called inside 997 * rtnl_lock(), and result refcount is unchanged. 998 */ 999 1000 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1001 unsigned short mask) 1002 { 1003 struct net_device *dev, *ret; 1004 1005 ASSERT_RTNL(); 1006 1007 ret = NULL; 1008 for_each_netdev(net, dev) { 1009 if (((dev->flags ^ if_flags) & mask) == 0) { 1010 ret = dev; 1011 break; 1012 } 1013 } 1014 return ret; 1015 } 1016 EXPORT_SYMBOL(__dev_get_by_flags); 1017 1018 /** 1019 * dev_valid_name - check if name is okay for network device 1020 * @name: name string 1021 * 1022 * Network device names need to be valid file names to 1023 * to allow sysfs to work. We also disallow any kind of 1024 * whitespace. 1025 */ 1026 bool dev_valid_name(const char *name) 1027 { 1028 if (*name == '\0') 1029 return false; 1030 if (strlen(name) >= IFNAMSIZ) 1031 return false; 1032 if (!strcmp(name, ".") || !strcmp(name, "..")) 1033 return false; 1034 1035 while (*name) { 1036 if (*name == '/' || *name == ':' || isspace(*name)) 1037 return false; 1038 name++; 1039 } 1040 return true; 1041 } 1042 EXPORT_SYMBOL(dev_valid_name); 1043 1044 /** 1045 * __dev_alloc_name - allocate a name for a device 1046 * @net: network namespace to allocate the device name in 1047 * @name: name format string 1048 * @buf: scratch buffer and result name string 1049 * 1050 * Passed a format string - eg "lt%d" it will try and find a suitable 1051 * id. It scans list of devices to build up a free map, then chooses 1052 * the first empty slot. The caller must hold the dev_base or rtnl lock 1053 * while allocating the name and adding the device in order to avoid 1054 * duplicates. 1055 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1056 * Returns the number of the unit assigned or a negative errno code. 1057 */ 1058 1059 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1060 { 1061 int i = 0; 1062 const char *p; 1063 const int max_netdevices = 8*PAGE_SIZE; 1064 unsigned long *inuse; 1065 struct net_device *d; 1066 1067 if (!dev_valid_name(name)) 1068 return -EINVAL; 1069 1070 p = strchr(name, '%'); 1071 if (p) { 1072 /* 1073 * Verify the string as this thing may have come from 1074 * the user. There must be either one "%d" and no other "%" 1075 * characters. 1076 */ 1077 if (p[1] != 'd' || strchr(p + 2, '%')) 1078 return -EINVAL; 1079 1080 /* Use one page as a bit array of possible slots */ 1081 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1082 if (!inuse) 1083 return -ENOMEM; 1084 1085 for_each_netdev(net, d) { 1086 if (!sscanf(d->name, name, &i)) 1087 continue; 1088 if (i < 0 || i >= max_netdevices) 1089 continue; 1090 1091 /* avoid cases where sscanf is not exact inverse of printf */ 1092 snprintf(buf, IFNAMSIZ, name, i); 1093 if (!strncmp(buf, d->name, IFNAMSIZ)) 1094 set_bit(i, inuse); 1095 } 1096 1097 i = find_first_zero_bit(inuse, max_netdevices); 1098 free_page((unsigned long) inuse); 1099 } 1100 1101 snprintf(buf, IFNAMSIZ, name, i); 1102 if (!__dev_get_by_name(net, buf)) 1103 return i; 1104 1105 /* It is possible to run out of possible slots 1106 * when the name is long and there isn't enough space left 1107 * for the digits, or if all bits are used. 1108 */ 1109 return p ? -ENFILE : -EEXIST; 1110 } 1111 1112 static int dev_alloc_name_ns(struct net *net, 1113 struct net_device *dev, 1114 const char *name) 1115 { 1116 char buf[IFNAMSIZ]; 1117 int ret; 1118 1119 BUG_ON(!net); 1120 ret = __dev_alloc_name(net, name, buf); 1121 if (ret >= 0) 1122 strlcpy(dev->name, buf, IFNAMSIZ); 1123 return ret; 1124 } 1125 1126 /** 1127 * dev_alloc_name - allocate a name for a device 1128 * @dev: device 1129 * @name: name format string 1130 * 1131 * Passed a format string - eg "lt%d" it will try and find a suitable 1132 * id. It scans list of devices to build up a free map, then chooses 1133 * the first empty slot. The caller must hold the dev_base or rtnl lock 1134 * while allocating the name and adding the device in order to avoid 1135 * duplicates. 1136 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1137 * Returns the number of the unit assigned or a negative errno code. 1138 */ 1139 1140 int dev_alloc_name(struct net_device *dev, const char *name) 1141 { 1142 return dev_alloc_name_ns(dev_net(dev), dev, name); 1143 } 1144 EXPORT_SYMBOL(dev_alloc_name); 1145 1146 int dev_get_valid_name(struct net *net, struct net_device *dev, 1147 const char *name) 1148 { 1149 return dev_alloc_name_ns(net, dev, name); 1150 } 1151 EXPORT_SYMBOL(dev_get_valid_name); 1152 1153 /** 1154 * dev_change_name - change name of a device 1155 * @dev: device 1156 * @newname: name (or format string) must be at least IFNAMSIZ 1157 * 1158 * Change name of a device, can pass format strings "eth%d". 1159 * for wildcarding. 1160 */ 1161 int dev_change_name(struct net_device *dev, const char *newname) 1162 { 1163 unsigned char old_assign_type; 1164 char oldname[IFNAMSIZ]; 1165 int err = 0; 1166 int ret; 1167 struct net *net; 1168 1169 ASSERT_RTNL(); 1170 BUG_ON(!dev_net(dev)); 1171 1172 net = dev_net(dev); 1173 if (dev->flags & IFF_UP) 1174 return -EBUSY; 1175 1176 write_seqcount_begin(&devnet_rename_seq); 1177 1178 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1179 write_seqcount_end(&devnet_rename_seq); 1180 return 0; 1181 } 1182 1183 memcpy(oldname, dev->name, IFNAMSIZ); 1184 1185 err = dev_get_valid_name(net, dev, newname); 1186 if (err < 0) { 1187 write_seqcount_end(&devnet_rename_seq); 1188 return err; 1189 } 1190 1191 if (oldname[0] && !strchr(oldname, '%')) 1192 netdev_info(dev, "renamed from %s\n", oldname); 1193 1194 old_assign_type = dev->name_assign_type; 1195 dev->name_assign_type = NET_NAME_RENAMED; 1196 1197 rollback: 1198 ret = device_rename(&dev->dev, dev->name); 1199 if (ret) { 1200 memcpy(dev->name, oldname, IFNAMSIZ); 1201 dev->name_assign_type = old_assign_type; 1202 write_seqcount_end(&devnet_rename_seq); 1203 return ret; 1204 } 1205 1206 write_seqcount_end(&devnet_rename_seq); 1207 1208 netdev_adjacent_rename_links(dev, oldname); 1209 1210 write_lock_bh(&dev_base_lock); 1211 hlist_del_rcu(&dev->name_hlist); 1212 write_unlock_bh(&dev_base_lock); 1213 1214 synchronize_rcu(); 1215 1216 write_lock_bh(&dev_base_lock); 1217 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1218 write_unlock_bh(&dev_base_lock); 1219 1220 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1221 ret = notifier_to_errno(ret); 1222 1223 if (ret) { 1224 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1225 if (err >= 0) { 1226 err = ret; 1227 write_seqcount_begin(&devnet_rename_seq); 1228 memcpy(dev->name, oldname, IFNAMSIZ); 1229 memcpy(oldname, newname, IFNAMSIZ); 1230 dev->name_assign_type = old_assign_type; 1231 old_assign_type = NET_NAME_RENAMED; 1232 goto rollback; 1233 } else { 1234 pr_err("%s: name change rollback failed: %d\n", 1235 dev->name, ret); 1236 } 1237 } 1238 1239 return err; 1240 } 1241 1242 /** 1243 * dev_set_alias - change ifalias of a device 1244 * @dev: device 1245 * @alias: name up to IFALIASZ 1246 * @len: limit of bytes to copy from info 1247 * 1248 * Set ifalias for a device, 1249 */ 1250 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1251 { 1252 struct dev_ifalias *new_alias = NULL; 1253 1254 if (len >= IFALIASZ) 1255 return -EINVAL; 1256 1257 if (len) { 1258 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1259 if (!new_alias) 1260 return -ENOMEM; 1261 1262 memcpy(new_alias->ifalias, alias, len); 1263 new_alias->ifalias[len] = 0; 1264 } 1265 1266 mutex_lock(&ifalias_mutex); 1267 rcu_swap_protected(dev->ifalias, new_alias, 1268 mutex_is_locked(&ifalias_mutex)); 1269 mutex_unlock(&ifalias_mutex); 1270 1271 if (new_alias) 1272 kfree_rcu(new_alias, rcuhead); 1273 1274 return len; 1275 } 1276 1277 /** 1278 * dev_get_alias - get ifalias of a device 1279 * @dev: device 1280 * @name: buffer to store name of ifalias 1281 * @len: size of buffer 1282 * 1283 * get ifalias for a device. Caller must make sure dev cannot go 1284 * away, e.g. rcu read lock or own a reference count to device. 1285 */ 1286 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1287 { 1288 const struct dev_ifalias *alias; 1289 int ret = 0; 1290 1291 rcu_read_lock(); 1292 alias = rcu_dereference(dev->ifalias); 1293 if (alias) 1294 ret = snprintf(name, len, "%s", alias->ifalias); 1295 rcu_read_unlock(); 1296 1297 return ret; 1298 } 1299 1300 /** 1301 * netdev_features_change - device changes features 1302 * @dev: device to cause notification 1303 * 1304 * Called to indicate a device has changed features. 1305 */ 1306 void netdev_features_change(struct net_device *dev) 1307 { 1308 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1309 } 1310 EXPORT_SYMBOL(netdev_features_change); 1311 1312 /** 1313 * netdev_state_change - device changes state 1314 * @dev: device to cause notification 1315 * 1316 * Called to indicate a device has changed state. This function calls 1317 * the notifier chains for netdev_chain and sends a NEWLINK message 1318 * to the routing socket. 1319 */ 1320 void netdev_state_change(struct net_device *dev) 1321 { 1322 if (dev->flags & IFF_UP) { 1323 struct netdev_notifier_change_info change_info = { 1324 .info.dev = dev, 1325 }; 1326 1327 call_netdevice_notifiers_info(NETDEV_CHANGE, 1328 &change_info.info); 1329 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1330 } 1331 } 1332 EXPORT_SYMBOL(netdev_state_change); 1333 1334 /** 1335 * netdev_notify_peers - notify network peers about existence of @dev 1336 * @dev: network device 1337 * 1338 * Generate traffic such that interested network peers are aware of 1339 * @dev, such as by generating a gratuitous ARP. This may be used when 1340 * a device wants to inform the rest of the network about some sort of 1341 * reconfiguration such as a failover event or virtual machine 1342 * migration. 1343 */ 1344 void netdev_notify_peers(struct net_device *dev) 1345 { 1346 rtnl_lock(); 1347 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1348 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1349 rtnl_unlock(); 1350 } 1351 EXPORT_SYMBOL(netdev_notify_peers); 1352 1353 static int __dev_open(struct net_device *dev) 1354 { 1355 const struct net_device_ops *ops = dev->netdev_ops; 1356 int ret; 1357 1358 ASSERT_RTNL(); 1359 1360 if (!netif_device_present(dev)) 1361 return -ENODEV; 1362 1363 /* Block netpoll from trying to do any rx path servicing. 1364 * If we don't do this there is a chance ndo_poll_controller 1365 * or ndo_poll may be running while we open the device 1366 */ 1367 netpoll_poll_disable(dev); 1368 1369 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); 1370 ret = notifier_to_errno(ret); 1371 if (ret) 1372 return ret; 1373 1374 set_bit(__LINK_STATE_START, &dev->state); 1375 1376 if (ops->ndo_validate_addr) 1377 ret = ops->ndo_validate_addr(dev); 1378 1379 if (!ret && ops->ndo_open) 1380 ret = ops->ndo_open(dev); 1381 1382 netpoll_poll_enable(dev); 1383 1384 if (ret) 1385 clear_bit(__LINK_STATE_START, &dev->state); 1386 else { 1387 dev->flags |= IFF_UP; 1388 dev_set_rx_mode(dev); 1389 dev_activate(dev); 1390 add_device_randomness(dev->dev_addr, dev->addr_len); 1391 } 1392 1393 return ret; 1394 } 1395 1396 /** 1397 * dev_open - prepare an interface for use. 1398 * @dev: device to open 1399 * 1400 * Takes a device from down to up state. The device's private open 1401 * function is invoked and then the multicast lists are loaded. Finally 1402 * the device is moved into the up state and a %NETDEV_UP message is 1403 * sent to the netdev notifier chain. 1404 * 1405 * Calling this function on an active interface is a nop. On a failure 1406 * a negative errno code is returned. 1407 */ 1408 int dev_open(struct net_device *dev) 1409 { 1410 int ret; 1411 1412 if (dev->flags & IFF_UP) 1413 return 0; 1414 1415 ret = __dev_open(dev); 1416 if (ret < 0) 1417 return ret; 1418 1419 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1420 call_netdevice_notifiers(NETDEV_UP, dev); 1421 1422 return ret; 1423 } 1424 EXPORT_SYMBOL(dev_open); 1425 1426 static void __dev_close_many(struct list_head *head) 1427 { 1428 struct net_device *dev; 1429 1430 ASSERT_RTNL(); 1431 might_sleep(); 1432 1433 list_for_each_entry(dev, head, close_list) { 1434 /* Temporarily disable netpoll until the interface is down */ 1435 netpoll_poll_disable(dev); 1436 1437 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1438 1439 clear_bit(__LINK_STATE_START, &dev->state); 1440 1441 /* Synchronize to scheduled poll. We cannot touch poll list, it 1442 * can be even on different cpu. So just clear netif_running(). 1443 * 1444 * dev->stop() will invoke napi_disable() on all of it's 1445 * napi_struct instances on this device. 1446 */ 1447 smp_mb__after_atomic(); /* Commit netif_running(). */ 1448 } 1449 1450 dev_deactivate_many(head); 1451 1452 list_for_each_entry(dev, head, close_list) { 1453 const struct net_device_ops *ops = dev->netdev_ops; 1454 1455 /* 1456 * Call the device specific close. This cannot fail. 1457 * Only if device is UP 1458 * 1459 * We allow it to be called even after a DETACH hot-plug 1460 * event. 1461 */ 1462 if (ops->ndo_stop) 1463 ops->ndo_stop(dev); 1464 1465 dev->flags &= ~IFF_UP; 1466 netpoll_poll_enable(dev); 1467 } 1468 } 1469 1470 static void __dev_close(struct net_device *dev) 1471 { 1472 LIST_HEAD(single); 1473 1474 list_add(&dev->close_list, &single); 1475 __dev_close_many(&single); 1476 list_del(&single); 1477 } 1478 1479 void dev_close_many(struct list_head *head, bool unlink) 1480 { 1481 struct net_device *dev, *tmp; 1482 1483 /* Remove the devices that don't need to be closed */ 1484 list_for_each_entry_safe(dev, tmp, head, close_list) 1485 if (!(dev->flags & IFF_UP)) 1486 list_del_init(&dev->close_list); 1487 1488 __dev_close_many(head); 1489 1490 list_for_each_entry_safe(dev, tmp, head, close_list) { 1491 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1492 call_netdevice_notifiers(NETDEV_DOWN, dev); 1493 if (unlink) 1494 list_del_init(&dev->close_list); 1495 } 1496 } 1497 EXPORT_SYMBOL(dev_close_many); 1498 1499 /** 1500 * dev_close - shutdown an interface. 1501 * @dev: device to shutdown 1502 * 1503 * This function moves an active device into down state. A 1504 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1505 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1506 * chain. 1507 */ 1508 void dev_close(struct net_device *dev) 1509 { 1510 if (dev->flags & IFF_UP) { 1511 LIST_HEAD(single); 1512 1513 list_add(&dev->close_list, &single); 1514 dev_close_many(&single, true); 1515 list_del(&single); 1516 } 1517 } 1518 EXPORT_SYMBOL(dev_close); 1519 1520 1521 /** 1522 * dev_disable_lro - disable Large Receive Offload on a device 1523 * @dev: device 1524 * 1525 * Disable Large Receive Offload (LRO) on a net device. Must be 1526 * called under RTNL. This is needed if received packets may be 1527 * forwarded to another interface. 1528 */ 1529 void dev_disable_lro(struct net_device *dev) 1530 { 1531 struct net_device *lower_dev; 1532 struct list_head *iter; 1533 1534 dev->wanted_features &= ~NETIF_F_LRO; 1535 netdev_update_features(dev); 1536 1537 if (unlikely(dev->features & NETIF_F_LRO)) 1538 netdev_WARN(dev, "failed to disable LRO!\n"); 1539 1540 netdev_for_each_lower_dev(dev, lower_dev, iter) 1541 dev_disable_lro(lower_dev); 1542 } 1543 EXPORT_SYMBOL(dev_disable_lro); 1544 1545 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1546 struct net_device *dev) 1547 { 1548 struct netdev_notifier_info info = { 1549 .dev = dev, 1550 }; 1551 1552 return nb->notifier_call(nb, val, &info); 1553 } 1554 1555 static int dev_boot_phase = 1; 1556 1557 /** 1558 * register_netdevice_notifier - register a network notifier block 1559 * @nb: notifier 1560 * 1561 * Register a notifier to be called when network device events occur. 1562 * The notifier passed is linked into the kernel structures and must 1563 * not be reused until it has been unregistered. A negative errno code 1564 * is returned on a failure. 1565 * 1566 * When registered all registration and up events are replayed 1567 * to the new notifier to allow device to have a race free 1568 * view of the network device list. 1569 */ 1570 1571 int register_netdevice_notifier(struct notifier_block *nb) 1572 { 1573 struct net_device *dev; 1574 struct net_device *last; 1575 struct net *net; 1576 int err; 1577 1578 rtnl_lock(); 1579 err = raw_notifier_chain_register(&netdev_chain, nb); 1580 if (err) 1581 goto unlock; 1582 if (dev_boot_phase) 1583 goto unlock; 1584 for_each_net(net) { 1585 for_each_netdev(net, dev) { 1586 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1587 err = notifier_to_errno(err); 1588 if (err) 1589 goto rollback; 1590 1591 if (!(dev->flags & IFF_UP)) 1592 continue; 1593 1594 call_netdevice_notifier(nb, NETDEV_UP, dev); 1595 } 1596 } 1597 1598 unlock: 1599 rtnl_unlock(); 1600 return err; 1601 1602 rollback: 1603 last = dev; 1604 for_each_net(net) { 1605 for_each_netdev(net, dev) { 1606 if (dev == last) 1607 goto outroll; 1608 1609 if (dev->flags & IFF_UP) { 1610 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1611 dev); 1612 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1613 } 1614 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1615 } 1616 } 1617 1618 outroll: 1619 raw_notifier_chain_unregister(&netdev_chain, nb); 1620 goto unlock; 1621 } 1622 EXPORT_SYMBOL(register_netdevice_notifier); 1623 1624 /** 1625 * unregister_netdevice_notifier - unregister a network notifier block 1626 * @nb: notifier 1627 * 1628 * Unregister a notifier previously registered by 1629 * register_netdevice_notifier(). The notifier is unlinked into the 1630 * kernel structures and may then be reused. A negative errno code 1631 * is returned on a failure. 1632 * 1633 * After unregistering unregister and down device events are synthesized 1634 * for all devices on the device list to the removed notifier to remove 1635 * the need for special case cleanup code. 1636 */ 1637 1638 int unregister_netdevice_notifier(struct notifier_block *nb) 1639 { 1640 struct net_device *dev; 1641 struct net *net; 1642 int err; 1643 1644 rtnl_lock(); 1645 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1646 if (err) 1647 goto unlock; 1648 1649 for_each_net(net) { 1650 for_each_netdev(net, dev) { 1651 if (dev->flags & IFF_UP) { 1652 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1653 dev); 1654 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1655 } 1656 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1657 } 1658 } 1659 unlock: 1660 rtnl_unlock(); 1661 return err; 1662 } 1663 EXPORT_SYMBOL(unregister_netdevice_notifier); 1664 1665 /** 1666 * call_netdevice_notifiers_info - call all network notifier blocks 1667 * @val: value passed unmodified to notifier function 1668 * @dev: net_device pointer passed unmodified to notifier function 1669 * @info: notifier information data 1670 * 1671 * Call all network notifier blocks. Parameters and return value 1672 * are as for raw_notifier_call_chain(). 1673 */ 1674 1675 static int call_netdevice_notifiers_info(unsigned long val, 1676 struct netdev_notifier_info *info) 1677 { 1678 ASSERT_RTNL(); 1679 return raw_notifier_call_chain(&netdev_chain, val, info); 1680 } 1681 1682 /** 1683 * call_netdevice_notifiers - call all network notifier blocks 1684 * @val: value passed unmodified to notifier function 1685 * @dev: net_device pointer passed unmodified to notifier function 1686 * 1687 * Call all network notifier blocks. Parameters and return value 1688 * are as for raw_notifier_call_chain(). 1689 */ 1690 1691 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1692 { 1693 struct netdev_notifier_info info = { 1694 .dev = dev, 1695 }; 1696 1697 return call_netdevice_notifiers_info(val, &info); 1698 } 1699 EXPORT_SYMBOL(call_netdevice_notifiers); 1700 1701 #ifdef CONFIG_NET_INGRESS 1702 static struct static_key ingress_needed __read_mostly; 1703 1704 void net_inc_ingress_queue(void) 1705 { 1706 static_key_slow_inc(&ingress_needed); 1707 } 1708 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1709 1710 void net_dec_ingress_queue(void) 1711 { 1712 static_key_slow_dec(&ingress_needed); 1713 } 1714 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1715 #endif 1716 1717 #ifdef CONFIG_NET_EGRESS 1718 static struct static_key egress_needed __read_mostly; 1719 1720 void net_inc_egress_queue(void) 1721 { 1722 static_key_slow_inc(&egress_needed); 1723 } 1724 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 1725 1726 void net_dec_egress_queue(void) 1727 { 1728 static_key_slow_dec(&egress_needed); 1729 } 1730 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 1731 #endif 1732 1733 static struct static_key netstamp_needed __read_mostly; 1734 #ifdef HAVE_JUMP_LABEL 1735 static atomic_t netstamp_needed_deferred; 1736 static atomic_t netstamp_wanted; 1737 static void netstamp_clear(struct work_struct *work) 1738 { 1739 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1740 int wanted; 1741 1742 wanted = atomic_add_return(deferred, &netstamp_wanted); 1743 if (wanted > 0) 1744 static_key_enable(&netstamp_needed); 1745 else 1746 static_key_disable(&netstamp_needed); 1747 } 1748 static DECLARE_WORK(netstamp_work, netstamp_clear); 1749 #endif 1750 1751 void net_enable_timestamp(void) 1752 { 1753 #ifdef HAVE_JUMP_LABEL 1754 int wanted; 1755 1756 while (1) { 1757 wanted = atomic_read(&netstamp_wanted); 1758 if (wanted <= 0) 1759 break; 1760 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 1761 return; 1762 } 1763 atomic_inc(&netstamp_needed_deferred); 1764 schedule_work(&netstamp_work); 1765 #else 1766 static_key_slow_inc(&netstamp_needed); 1767 #endif 1768 } 1769 EXPORT_SYMBOL(net_enable_timestamp); 1770 1771 void net_disable_timestamp(void) 1772 { 1773 #ifdef HAVE_JUMP_LABEL 1774 int wanted; 1775 1776 while (1) { 1777 wanted = atomic_read(&netstamp_wanted); 1778 if (wanted <= 1) 1779 break; 1780 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 1781 return; 1782 } 1783 atomic_dec(&netstamp_needed_deferred); 1784 schedule_work(&netstamp_work); 1785 #else 1786 static_key_slow_dec(&netstamp_needed); 1787 #endif 1788 } 1789 EXPORT_SYMBOL(net_disable_timestamp); 1790 1791 static inline void net_timestamp_set(struct sk_buff *skb) 1792 { 1793 skb->tstamp = 0; 1794 if (static_key_false(&netstamp_needed)) 1795 __net_timestamp(skb); 1796 } 1797 1798 #define net_timestamp_check(COND, SKB) \ 1799 if (static_key_false(&netstamp_needed)) { \ 1800 if ((COND) && !(SKB)->tstamp) \ 1801 __net_timestamp(SKB); \ 1802 } \ 1803 1804 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 1805 { 1806 unsigned int len; 1807 1808 if (!(dev->flags & IFF_UP)) 1809 return false; 1810 1811 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1812 if (skb->len <= len) 1813 return true; 1814 1815 /* if TSO is enabled, we don't care about the length as the packet 1816 * could be forwarded without being segmented before 1817 */ 1818 if (skb_is_gso(skb)) 1819 return true; 1820 1821 return false; 1822 } 1823 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1824 1825 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1826 { 1827 int ret = ____dev_forward_skb(dev, skb); 1828 1829 if (likely(!ret)) { 1830 skb->protocol = eth_type_trans(skb, dev); 1831 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1832 } 1833 1834 return ret; 1835 } 1836 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1837 1838 /** 1839 * dev_forward_skb - loopback an skb to another netif 1840 * 1841 * @dev: destination network device 1842 * @skb: buffer to forward 1843 * 1844 * return values: 1845 * NET_RX_SUCCESS (no congestion) 1846 * NET_RX_DROP (packet was dropped, but freed) 1847 * 1848 * dev_forward_skb can be used for injecting an skb from the 1849 * start_xmit function of one device into the receive queue 1850 * of another device. 1851 * 1852 * The receiving device may be in another namespace, so 1853 * we have to clear all information in the skb that could 1854 * impact namespace isolation. 1855 */ 1856 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1857 { 1858 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1859 } 1860 EXPORT_SYMBOL_GPL(dev_forward_skb); 1861 1862 static inline int deliver_skb(struct sk_buff *skb, 1863 struct packet_type *pt_prev, 1864 struct net_device *orig_dev) 1865 { 1866 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 1867 return -ENOMEM; 1868 refcount_inc(&skb->users); 1869 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1870 } 1871 1872 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 1873 struct packet_type **pt, 1874 struct net_device *orig_dev, 1875 __be16 type, 1876 struct list_head *ptype_list) 1877 { 1878 struct packet_type *ptype, *pt_prev = *pt; 1879 1880 list_for_each_entry_rcu(ptype, ptype_list, list) { 1881 if (ptype->type != type) 1882 continue; 1883 if (pt_prev) 1884 deliver_skb(skb, pt_prev, orig_dev); 1885 pt_prev = ptype; 1886 } 1887 *pt = pt_prev; 1888 } 1889 1890 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1891 { 1892 if (!ptype->af_packet_priv || !skb->sk) 1893 return false; 1894 1895 if (ptype->id_match) 1896 return ptype->id_match(ptype, skb->sk); 1897 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1898 return true; 1899 1900 return false; 1901 } 1902 1903 /* 1904 * Support routine. Sends outgoing frames to any network 1905 * taps currently in use. 1906 */ 1907 1908 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1909 { 1910 struct packet_type *ptype; 1911 struct sk_buff *skb2 = NULL; 1912 struct packet_type *pt_prev = NULL; 1913 struct list_head *ptype_list = &ptype_all; 1914 1915 rcu_read_lock(); 1916 again: 1917 list_for_each_entry_rcu(ptype, ptype_list, list) { 1918 /* Never send packets back to the socket 1919 * they originated from - MvS (miquels@drinkel.ow.org) 1920 */ 1921 if (skb_loop_sk(ptype, skb)) 1922 continue; 1923 1924 if (pt_prev) { 1925 deliver_skb(skb2, pt_prev, skb->dev); 1926 pt_prev = ptype; 1927 continue; 1928 } 1929 1930 /* need to clone skb, done only once */ 1931 skb2 = skb_clone(skb, GFP_ATOMIC); 1932 if (!skb2) 1933 goto out_unlock; 1934 1935 net_timestamp_set(skb2); 1936 1937 /* skb->nh should be correctly 1938 * set by sender, so that the second statement is 1939 * just protection against buggy protocols. 1940 */ 1941 skb_reset_mac_header(skb2); 1942 1943 if (skb_network_header(skb2) < skb2->data || 1944 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 1945 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 1946 ntohs(skb2->protocol), 1947 dev->name); 1948 skb_reset_network_header(skb2); 1949 } 1950 1951 skb2->transport_header = skb2->network_header; 1952 skb2->pkt_type = PACKET_OUTGOING; 1953 pt_prev = ptype; 1954 } 1955 1956 if (ptype_list == &ptype_all) { 1957 ptype_list = &dev->ptype_all; 1958 goto again; 1959 } 1960 out_unlock: 1961 if (pt_prev) { 1962 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 1963 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 1964 else 1965 kfree_skb(skb2); 1966 } 1967 rcu_read_unlock(); 1968 } 1969 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 1970 1971 /** 1972 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 1973 * @dev: Network device 1974 * @txq: number of queues available 1975 * 1976 * If real_num_tx_queues is changed the tc mappings may no longer be 1977 * valid. To resolve this verify the tc mapping remains valid and if 1978 * not NULL the mapping. With no priorities mapping to this 1979 * offset/count pair it will no longer be used. In the worst case TC0 1980 * is invalid nothing can be done so disable priority mappings. If is 1981 * expected that drivers will fix this mapping if they can before 1982 * calling netif_set_real_num_tx_queues. 1983 */ 1984 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 1985 { 1986 int i; 1987 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 1988 1989 /* If TC0 is invalidated disable TC mapping */ 1990 if (tc->offset + tc->count > txq) { 1991 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 1992 dev->num_tc = 0; 1993 return; 1994 } 1995 1996 /* Invalidated prio to tc mappings set to TC0 */ 1997 for (i = 1; i < TC_BITMASK + 1; i++) { 1998 int q = netdev_get_prio_tc_map(dev, i); 1999 2000 tc = &dev->tc_to_txq[q]; 2001 if (tc->offset + tc->count > txq) { 2002 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2003 i, q); 2004 netdev_set_prio_tc_map(dev, i, 0); 2005 } 2006 } 2007 } 2008 2009 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2010 { 2011 if (dev->num_tc) { 2012 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2013 int i; 2014 2015 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2016 if ((txq - tc->offset) < tc->count) 2017 return i; 2018 } 2019 2020 return -1; 2021 } 2022 2023 return 0; 2024 } 2025 EXPORT_SYMBOL(netdev_txq_to_tc); 2026 2027 #ifdef CONFIG_XPS 2028 static DEFINE_MUTEX(xps_map_mutex); 2029 #define xmap_dereference(P) \ 2030 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2031 2032 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2033 int tci, u16 index) 2034 { 2035 struct xps_map *map = NULL; 2036 int pos; 2037 2038 if (dev_maps) 2039 map = xmap_dereference(dev_maps->cpu_map[tci]); 2040 if (!map) 2041 return false; 2042 2043 for (pos = map->len; pos--;) { 2044 if (map->queues[pos] != index) 2045 continue; 2046 2047 if (map->len > 1) { 2048 map->queues[pos] = map->queues[--map->len]; 2049 break; 2050 } 2051 2052 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); 2053 kfree_rcu(map, rcu); 2054 return false; 2055 } 2056 2057 return true; 2058 } 2059 2060 static bool remove_xps_queue_cpu(struct net_device *dev, 2061 struct xps_dev_maps *dev_maps, 2062 int cpu, u16 offset, u16 count) 2063 { 2064 int num_tc = dev->num_tc ? : 1; 2065 bool active = false; 2066 int tci; 2067 2068 for (tci = cpu * num_tc; num_tc--; tci++) { 2069 int i, j; 2070 2071 for (i = count, j = offset; i--; j++) { 2072 if (!remove_xps_queue(dev_maps, cpu, j)) 2073 break; 2074 } 2075 2076 active |= i < 0; 2077 } 2078 2079 return active; 2080 } 2081 2082 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2083 u16 count) 2084 { 2085 struct xps_dev_maps *dev_maps; 2086 int cpu, i; 2087 bool active = false; 2088 2089 mutex_lock(&xps_map_mutex); 2090 dev_maps = xmap_dereference(dev->xps_maps); 2091 2092 if (!dev_maps) 2093 goto out_no_maps; 2094 2095 for_each_possible_cpu(cpu) 2096 active |= remove_xps_queue_cpu(dev, dev_maps, cpu, 2097 offset, count); 2098 2099 if (!active) { 2100 RCU_INIT_POINTER(dev->xps_maps, NULL); 2101 kfree_rcu(dev_maps, rcu); 2102 } 2103 2104 for (i = offset + (count - 1); count--; i--) 2105 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), 2106 NUMA_NO_NODE); 2107 2108 out_no_maps: 2109 mutex_unlock(&xps_map_mutex); 2110 } 2111 2112 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2113 { 2114 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2115 } 2116 2117 static struct xps_map *expand_xps_map(struct xps_map *map, 2118 int cpu, u16 index) 2119 { 2120 struct xps_map *new_map; 2121 int alloc_len = XPS_MIN_MAP_ALLOC; 2122 int i, pos; 2123 2124 for (pos = 0; map && pos < map->len; pos++) { 2125 if (map->queues[pos] != index) 2126 continue; 2127 return map; 2128 } 2129 2130 /* Need to add queue to this CPU's existing map */ 2131 if (map) { 2132 if (pos < map->alloc_len) 2133 return map; 2134 2135 alloc_len = map->alloc_len * 2; 2136 } 2137 2138 /* Need to allocate new map to store queue on this CPU's map */ 2139 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2140 cpu_to_node(cpu)); 2141 if (!new_map) 2142 return NULL; 2143 2144 for (i = 0; i < pos; i++) 2145 new_map->queues[i] = map->queues[i]; 2146 new_map->alloc_len = alloc_len; 2147 new_map->len = pos; 2148 2149 return new_map; 2150 } 2151 2152 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2153 u16 index) 2154 { 2155 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2156 int i, cpu, tci, numa_node_id = -2; 2157 int maps_sz, num_tc = 1, tc = 0; 2158 struct xps_map *map, *new_map; 2159 bool active = false; 2160 2161 if (dev->num_tc) { 2162 num_tc = dev->num_tc; 2163 tc = netdev_txq_to_tc(dev, index); 2164 if (tc < 0) 2165 return -EINVAL; 2166 } 2167 2168 maps_sz = XPS_DEV_MAPS_SIZE(num_tc); 2169 if (maps_sz < L1_CACHE_BYTES) 2170 maps_sz = L1_CACHE_BYTES; 2171 2172 mutex_lock(&xps_map_mutex); 2173 2174 dev_maps = xmap_dereference(dev->xps_maps); 2175 2176 /* allocate memory for queue storage */ 2177 for_each_cpu_and(cpu, cpu_online_mask, mask) { 2178 if (!new_dev_maps) 2179 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2180 if (!new_dev_maps) { 2181 mutex_unlock(&xps_map_mutex); 2182 return -ENOMEM; 2183 } 2184 2185 tci = cpu * num_tc + tc; 2186 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : 2187 NULL; 2188 2189 map = expand_xps_map(map, cpu, index); 2190 if (!map) 2191 goto error; 2192 2193 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2194 } 2195 2196 if (!new_dev_maps) 2197 goto out_no_new_maps; 2198 2199 for_each_possible_cpu(cpu) { 2200 /* copy maps belonging to foreign traffic classes */ 2201 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { 2202 /* fill in the new device map from the old device map */ 2203 map = xmap_dereference(dev_maps->cpu_map[tci]); 2204 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2205 } 2206 2207 /* We need to explicitly update tci as prevous loop 2208 * could break out early if dev_maps is NULL. 2209 */ 2210 tci = cpu * num_tc + tc; 2211 2212 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { 2213 /* add queue to CPU maps */ 2214 int pos = 0; 2215 2216 map = xmap_dereference(new_dev_maps->cpu_map[tci]); 2217 while ((pos < map->len) && (map->queues[pos] != index)) 2218 pos++; 2219 2220 if (pos == map->len) 2221 map->queues[map->len++] = index; 2222 #ifdef CONFIG_NUMA 2223 if (numa_node_id == -2) 2224 numa_node_id = cpu_to_node(cpu); 2225 else if (numa_node_id != cpu_to_node(cpu)) 2226 numa_node_id = -1; 2227 #endif 2228 } else if (dev_maps) { 2229 /* fill in the new device map from the old device map */ 2230 map = xmap_dereference(dev_maps->cpu_map[tci]); 2231 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2232 } 2233 2234 /* copy maps belonging to foreign traffic classes */ 2235 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { 2236 /* fill in the new device map from the old device map */ 2237 map = xmap_dereference(dev_maps->cpu_map[tci]); 2238 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); 2239 } 2240 } 2241 2242 rcu_assign_pointer(dev->xps_maps, new_dev_maps); 2243 2244 /* Cleanup old maps */ 2245 if (!dev_maps) 2246 goto out_no_old_maps; 2247 2248 for_each_possible_cpu(cpu) { 2249 for (i = num_tc, tci = cpu * num_tc; i--; tci++) { 2250 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); 2251 map = xmap_dereference(dev_maps->cpu_map[tci]); 2252 if (map && map != new_map) 2253 kfree_rcu(map, rcu); 2254 } 2255 } 2256 2257 kfree_rcu(dev_maps, rcu); 2258 2259 out_no_old_maps: 2260 dev_maps = new_dev_maps; 2261 active = true; 2262 2263 out_no_new_maps: 2264 /* update Tx queue numa node */ 2265 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2266 (numa_node_id >= 0) ? numa_node_id : 2267 NUMA_NO_NODE); 2268 2269 if (!dev_maps) 2270 goto out_no_maps; 2271 2272 /* removes queue from unused CPUs */ 2273 for_each_possible_cpu(cpu) { 2274 for (i = tc, tci = cpu * num_tc; i--; tci++) 2275 active |= remove_xps_queue(dev_maps, tci, index); 2276 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) 2277 active |= remove_xps_queue(dev_maps, tci, index); 2278 for (i = num_tc - tc, tci++; --i; tci++) 2279 active |= remove_xps_queue(dev_maps, tci, index); 2280 } 2281 2282 /* free map if not active */ 2283 if (!active) { 2284 RCU_INIT_POINTER(dev->xps_maps, NULL); 2285 kfree_rcu(dev_maps, rcu); 2286 } 2287 2288 out_no_maps: 2289 mutex_unlock(&xps_map_mutex); 2290 2291 return 0; 2292 error: 2293 /* remove any maps that we added */ 2294 for_each_possible_cpu(cpu) { 2295 for (i = num_tc, tci = cpu * num_tc; i--; tci++) { 2296 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); 2297 map = dev_maps ? 2298 xmap_dereference(dev_maps->cpu_map[tci]) : 2299 NULL; 2300 if (new_map && new_map != map) 2301 kfree(new_map); 2302 } 2303 } 2304 2305 mutex_unlock(&xps_map_mutex); 2306 2307 kfree(new_dev_maps); 2308 return -ENOMEM; 2309 } 2310 EXPORT_SYMBOL(netif_set_xps_queue); 2311 2312 #endif 2313 void netdev_reset_tc(struct net_device *dev) 2314 { 2315 #ifdef CONFIG_XPS 2316 netif_reset_xps_queues_gt(dev, 0); 2317 #endif 2318 dev->num_tc = 0; 2319 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2320 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2321 } 2322 EXPORT_SYMBOL(netdev_reset_tc); 2323 2324 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2325 { 2326 if (tc >= dev->num_tc) 2327 return -EINVAL; 2328 2329 #ifdef CONFIG_XPS 2330 netif_reset_xps_queues(dev, offset, count); 2331 #endif 2332 dev->tc_to_txq[tc].count = count; 2333 dev->tc_to_txq[tc].offset = offset; 2334 return 0; 2335 } 2336 EXPORT_SYMBOL(netdev_set_tc_queue); 2337 2338 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2339 { 2340 if (num_tc > TC_MAX_QUEUE) 2341 return -EINVAL; 2342 2343 #ifdef CONFIG_XPS 2344 netif_reset_xps_queues_gt(dev, 0); 2345 #endif 2346 dev->num_tc = num_tc; 2347 return 0; 2348 } 2349 EXPORT_SYMBOL(netdev_set_num_tc); 2350 2351 /* 2352 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2353 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. 2354 */ 2355 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2356 { 2357 int rc; 2358 2359 if (txq < 1 || txq > dev->num_tx_queues) 2360 return -EINVAL; 2361 2362 if (dev->reg_state == NETREG_REGISTERED || 2363 dev->reg_state == NETREG_UNREGISTERING) { 2364 ASSERT_RTNL(); 2365 2366 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2367 txq); 2368 if (rc) 2369 return rc; 2370 2371 if (dev->num_tc) 2372 netif_setup_tc(dev, txq); 2373 2374 if (txq < dev->real_num_tx_queues) { 2375 qdisc_reset_all_tx_gt(dev, txq); 2376 #ifdef CONFIG_XPS 2377 netif_reset_xps_queues_gt(dev, txq); 2378 #endif 2379 } 2380 } 2381 2382 dev->real_num_tx_queues = txq; 2383 return 0; 2384 } 2385 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2386 2387 #ifdef CONFIG_SYSFS 2388 /** 2389 * netif_set_real_num_rx_queues - set actual number of RX queues used 2390 * @dev: Network device 2391 * @rxq: Actual number of RX queues 2392 * 2393 * This must be called either with the rtnl_lock held or before 2394 * registration of the net device. Returns 0 on success, or a 2395 * negative error code. If called before registration, it always 2396 * succeeds. 2397 */ 2398 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2399 { 2400 int rc; 2401 2402 if (rxq < 1 || rxq > dev->num_rx_queues) 2403 return -EINVAL; 2404 2405 if (dev->reg_state == NETREG_REGISTERED) { 2406 ASSERT_RTNL(); 2407 2408 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2409 rxq); 2410 if (rc) 2411 return rc; 2412 } 2413 2414 dev->real_num_rx_queues = rxq; 2415 return 0; 2416 } 2417 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2418 #endif 2419 2420 /** 2421 * netif_get_num_default_rss_queues - default number of RSS queues 2422 * 2423 * This routine should set an upper limit on the number of RSS queues 2424 * used by default by multiqueue devices. 2425 */ 2426 int netif_get_num_default_rss_queues(void) 2427 { 2428 return is_kdump_kernel() ? 2429 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2430 } 2431 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2432 2433 static void __netif_reschedule(struct Qdisc *q) 2434 { 2435 struct softnet_data *sd; 2436 unsigned long flags; 2437 2438 local_irq_save(flags); 2439 sd = this_cpu_ptr(&softnet_data); 2440 q->next_sched = NULL; 2441 *sd->output_queue_tailp = q; 2442 sd->output_queue_tailp = &q->next_sched; 2443 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2444 local_irq_restore(flags); 2445 } 2446 2447 void __netif_schedule(struct Qdisc *q) 2448 { 2449 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2450 __netif_reschedule(q); 2451 } 2452 EXPORT_SYMBOL(__netif_schedule); 2453 2454 struct dev_kfree_skb_cb { 2455 enum skb_free_reason reason; 2456 }; 2457 2458 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2459 { 2460 return (struct dev_kfree_skb_cb *)skb->cb; 2461 } 2462 2463 void netif_schedule_queue(struct netdev_queue *txq) 2464 { 2465 rcu_read_lock(); 2466 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2467 struct Qdisc *q = rcu_dereference(txq->qdisc); 2468 2469 __netif_schedule(q); 2470 } 2471 rcu_read_unlock(); 2472 } 2473 EXPORT_SYMBOL(netif_schedule_queue); 2474 2475 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2476 { 2477 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2478 struct Qdisc *q; 2479 2480 rcu_read_lock(); 2481 q = rcu_dereference(dev_queue->qdisc); 2482 __netif_schedule(q); 2483 rcu_read_unlock(); 2484 } 2485 } 2486 EXPORT_SYMBOL(netif_tx_wake_queue); 2487 2488 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2489 { 2490 unsigned long flags; 2491 2492 if (unlikely(!skb)) 2493 return; 2494 2495 if (likely(refcount_read(&skb->users) == 1)) { 2496 smp_rmb(); 2497 refcount_set(&skb->users, 0); 2498 } else if (likely(!refcount_dec_and_test(&skb->users))) { 2499 return; 2500 } 2501 get_kfree_skb_cb(skb)->reason = reason; 2502 local_irq_save(flags); 2503 skb->next = __this_cpu_read(softnet_data.completion_queue); 2504 __this_cpu_write(softnet_data.completion_queue, skb); 2505 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2506 local_irq_restore(flags); 2507 } 2508 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2509 2510 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2511 { 2512 if (in_irq() || irqs_disabled()) 2513 __dev_kfree_skb_irq(skb, reason); 2514 else 2515 dev_kfree_skb(skb); 2516 } 2517 EXPORT_SYMBOL(__dev_kfree_skb_any); 2518 2519 2520 /** 2521 * netif_device_detach - mark device as removed 2522 * @dev: network device 2523 * 2524 * Mark device as removed from system and therefore no longer available. 2525 */ 2526 void netif_device_detach(struct net_device *dev) 2527 { 2528 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2529 netif_running(dev)) { 2530 netif_tx_stop_all_queues(dev); 2531 } 2532 } 2533 EXPORT_SYMBOL(netif_device_detach); 2534 2535 /** 2536 * netif_device_attach - mark device as attached 2537 * @dev: network device 2538 * 2539 * Mark device as attached from system and restart if needed. 2540 */ 2541 void netif_device_attach(struct net_device *dev) 2542 { 2543 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2544 netif_running(dev)) { 2545 netif_tx_wake_all_queues(dev); 2546 __netdev_watchdog_up(dev); 2547 } 2548 } 2549 EXPORT_SYMBOL(netif_device_attach); 2550 2551 /* 2552 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 2553 * to be used as a distribution range. 2554 */ 2555 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 2556 unsigned int num_tx_queues) 2557 { 2558 u32 hash; 2559 u16 qoffset = 0; 2560 u16 qcount = num_tx_queues; 2561 2562 if (skb_rx_queue_recorded(skb)) { 2563 hash = skb_get_rx_queue(skb); 2564 while (unlikely(hash >= num_tx_queues)) 2565 hash -= num_tx_queues; 2566 return hash; 2567 } 2568 2569 if (dev->num_tc) { 2570 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 2571 2572 qoffset = dev->tc_to_txq[tc].offset; 2573 qcount = dev->tc_to_txq[tc].count; 2574 } 2575 2576 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 2577 } 2578 EXPORT_SYMBOL(__skb_tx_hash); 2579 2580 static void skb_warn_bad_offload(const struct sk_buff *skb) 2581 { 2582 static const netdev_features_t null_features; 2583 struct net_device *dev = skb->dev; 2584 const char *name = ""; 2585 2586 if (!net_ratelimit()) 2587 return; 2588 2589 if (dev) { 2590 if (dev->dev.parent) 2591 name = dev_driver_string(dev->dev.parent); 2592 else 2593 name = netdev_name(dev); 2594 } 2595 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2596 "gso_type=%d ip_summed=%d\n", 2597 name, dev ? &dev->features : &null_features, 2598 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2599 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2600 skb_shinfo(skb)->gso_type, skb->ip_summed); 2601 } 2602 2603 /* 2604 * Invalidate hardware checksum when packet is to be mangled, and 2605 * complete checksum manually on outgoing path. 2606 */ 2607 int skb_checksum_help(struct sk_buff *skb) 2608 { 2609 __wsum csum; 2610 int ret = 0, offset; 2611 2612 if (skb->ip_summed == CHECKSUM_COMPLETE) 2613 goto out_set_summed; 2614 2615 if (unlikely(skb_shinfo(skb)->gso_size)) { 2616 skb_warn_bad_offload(skb); 2617 return -EINVAL; 2618 } 2619 2620 /* Before computing a checksum, we should make sure no frag could 2621 * be modified by an external entity : checksum could be wrong. 2622 */ 2623 if (skb_has_shared_frag(skb)) { 2624 ret = __skb_linearize(skb); 2625 if (ret) 2626 goto out; 2627 } 2628 2629 offset = skb_checksum_start_offset(skb); 2630 BUG_ON(offset >= skb_headlen(skb)); 2631 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2632 2633 offset += skb->csum_offset; 2634 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2635 2636 if (skb_cloned(skb) && 2637 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2638 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2639 if (ret) 2640 goto out; 2641 } 2642 2643 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 2644 out_set_summed: 2645 skb->ip_summed = CHECKSUM_NONE; 2646 out: 2647 return ret; 2648 } 2649 EXPORT_SYMBOL(skb_checksum_help); 2650 2651 int skb_crc32c_csum_help(struct sk_buff *skb) 2652 { 2653 __le32 crc32c_csum; 2654 int ret = 0, offset, start; 2655 2656 if (skb->ip_summed != CHECKSUM_PARTIAL) 2657 goto out; 2658 2659 if (unlikely(skb_is_gso(skb))) 2660 goto out; 2661 2662 /* Before computing a checksum, we should make sure no frag could 2663 * be modified by an external entity : checksum could be wrong. 2664 */ 2665 if (unlikely(skb_has_shared_frag(skb))) { 2666 ret = __skb_linearize(skb); 2667 if (ret) 2668 goto out; 2669 } 2670 start = skb_checksum_start_offset(skb); 2671 offset = start + offsetof(struct sctphdr, checksum); 2672 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 2673 ret = -EINVAL; 2674 goto out; 2675 } 2676 if (skb_cloned(skb) && 2677 !skb_clone_writable(skb, offset + sizeof(__le32))) { 2678 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2679 if (ret) 2680 goto out; 2681 } 2682 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 2683 skb->len - start, ~(__u32)0, 2684 crc32c_csum_stub)); 2685 *(__le32 *)(skb->data + offset) = crc32c_csum; 2686 skb->ip_summed = CHECKSUM_NONE; 2687 skb->csum_not_inet = 0; 2688 out: 2689 return ret; 2690 } 2691 2692 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2693 { 2694 __be16 type = skb->protocol; 2695 2696 /* Tunnel gso handlers can set protocol to ethernet. */ 2697 if (type == htons(ETH_P_TEB)) { 2698 struct ethhdr *eth; 2699 2700 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2701 return 0; 2702 2703 eth = (struct ethhdr *)skb_mac_header(skb); 2704 type = eth->h_proto; 2705 } 2706 2707 return __vlan_get_protocol(skb, type, depth); 2708 } 2709 2710 /** 2711 * skb_mac_gso_segment - mac layer segmentation handler. 2712 * @skb: buffer to segment 2713 * @features: features for the output path (see dev->features) 2714 */ 2715 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 2716 netdev_features_t features) 2717 { 2718 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 2719 struct packet_offload *ptype; 2720 int vlan_depth = skb->mac_len; 2721 __be16 type = skb_network_protocol(skb, &vlan_depth); 2722 2723 if (unlikely(!type)) 2724 return ERR_PTR(-EINVAL); 2725 2726 __skb_pull(skb, vlan_depth); 2727 2728 rcu_read_lock(); 2729 list_for_each_entry_rcu(ptype, &offload_base, list) { 2730 if (ptype->type == type && ptype->callbacks.gso_segment) { 2731 segs = ptype->callbacks.gso_segment(skb, features); 2732 break; 2733 } 2734 } 2735 rcu_read_unlock(); 2736 2737 __skb_push(skb, skb->data - skb_mac_header(skb)); 2738 2739 return segs; 2740 } 2741 EXPORT_SYMBOL(skb_mac_gso_segment); 2742 2743 2744 /* openvswitch calls this on rx path, so we need a different check. 2745 */ 2746 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 2747 { 2748 if (tx_path) 2749 return skb->ip_summed != CHECKSUM_PARTIAL; 2750 2751 return skb->ip_summed == CHECKSUM_NONE; 2752 } 2753 2754 /** 2755 * __skb_gso_segment - Perform segmentation on skb. 2756 * @skb: buffer to segment 2757 * @features: features for the output path (see dev->features) 2758 * @tx_path: whether it is called in TX path 2759 * 2760 * This function segments the given skb and returns a list of segments. 2761 * 2762 * It may return NULL if the skb requires no segmentation. This is 2763 * only possible when GSO is used for verifying header integrity. 2764 * 2765 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. 2766 */ 2767 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 2768 netdev_features_t features, bool tx_path) 2769 { 2770 struct sk_buff *segs; 2771 2772 if (unlikely(skb_needs_check(skb, tx_path))) { 2773 int err; 2774 2775 /* We're going to init ->check field in TCP or UDP header */ 2776 err = skb_cow_head(skb, 0); 2777 if (err < 0) 2778 return ERR_PTR(err); 2779 } 2780 2781 /* Only report GSO partial support if it will enable us to 2782 * support segmentation on this frame without needing additional 2783 * work. 2784 */ 2785 if (features & NETIF_F_GSO_PARTIAL) { 2786 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 2787 struct net_device *dev = skb->dev; 2788 2789 partial_features |= dev->features & dev->gso_partial_features; 2790 if (!skb_gso_ok(skb, features | partial_features)) 2791 features &= ~NETIF_F_GSO_PARTIAL; 2792 } 2793 2794 BUILD_BUG_ON(SKB_SGO_CB_OFFSET + 2795 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 2796 2797 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 2798 SKB_GSO_CB(skb)->encap_level = 0; 2799 2800 skb_reset_mac_header(skb); 2801 skb_reset_mac_len(skb); 2802 2803 segs = skb_mac_gso_segment(skb, features); 2804 2805 if (unlikely(skb_needs_check(skb, tx_path))) 2806 skb_warn_bad_offload(skb); 2807 2808 return segs; 2809 } 2810 EXPORT_SYMBOL(__skb_gso_segment); 2811 2812 /* Take action when hardware reception checksum errors are detected. */ 2813 #ifdef CONFIG_BUG 2814 void netdev_rx_csum_fault(struct net_device *dev) 2815 { 2816 if (net_ratelimit()) { 2817 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 2818 dump_stack(); 2819 } 2820 } 2821 EXPORT_SYMBOL(netdev_rx_csum_fault); 2822 #endif 2823 2824 /* Actually, we should eliminate this check as soon as we know, that: 2825 * 1. IOMMU is present and allows to map all the memory. 2826 * 2. No high memory really exists on this machine. 2827 */ 2828 2829 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 2830 { 2831 #ifdef CONFIG_HIGHMEM 2832 int i; 2833 2834 if (!(dev->features & NETIF_F_HIGHDMA)) { 2835 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2836 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2837 2838 if (PageHighMem(skb_frag_page(frag))) 2839 return 1; 2840 } 2841 } 2842 2843 if (PCI_DMA_BUS_IS_PHYS) { 2844 struct device *pdev = dev->dev.parent; 2845 2846 if (!pdev) 2847 return 0; 2848 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2849 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2850 dma_addr_t addr = page_to_phys(skb_frag_page(frag)); 2851 2852 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) 2853 return 1; 2854 } 2855 } 2856 #endif 2857 return 0; 2858 } 2859 2860 /* If MPLS offload request, verify we are testing hardware MPLS features 2861 * instead of standard features for the netdev. 2862 */ 2863 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 2864 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2865 netdev_features_t features, 2866 __be16 type) 2867 { 2868 if (eth_p_mpls(type)) 2869 features &= skb->dev->mpls_features; 2870 2871 return features; 2872 } 2873 #else 2874 static netdev_features_t net_mpls_features(struct sk_buff *skb, 2875 netdev_features_t features, 2876 __be16 type) 2877 { 2878 return features; 2879 } 2880 #endif 2881 2882 static netdev_features_t harmonize_features(struct sk_buff *skb, 2883 netdev_features_t features) 2884 { 2885 int tmp; 2886 __be16 type; 2887 2888 type = skb_network_protocol(skb, &tmp); 2889 features = net_mpls_features(skb, features, type); 2890 2891 if (skb->ip_summed != CHECKSUM_NONE && 2892 !can_checksum_protocol(features, type)) { 2893 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 2894 } 2895 if (illegal_highdma(skb->dev, skb)) 2896 features &= ~NETIF_F_SG; 2897 2898 return features; 2899 } 2900 2901 netdev_features_t passthru_features_check(struct sk_buff *skb, 2902 struct net_device *dev, 2903 netdev_features_t features) 2904 { 2905 return features; 2906 } 2907 EXPORT_SYMBOL(passthru_features_check); 2908 2909 static netdev_features_t dflt_features_check(const struct sk_buff *skb, 2910 struct net_device *dev, 2911 netdev_features_t features) 2912 { 2913 return vlan_features_check(skb, features); 2914 } 2915 2916 static netdev_features_t gso_features_check(const struct sk_buff *skb, 2917 struct net_device *dev, 2918 netdev_features_t features) 2919 { 2920 u16 gso_segs = skb_shinfo(skb)->gso_segs; 2921 2922 if (gso_segs > dev->gso_max_segs) 2923 return features & ~NETIF_F_GSO_MASK; 2924 2925 /* Support for GSO partial features requires software 2926 * intervention before we can actually process the packets 2927 * so we need to strip support for any partial features now 2928 * and we can pull them back in after we have partially 2929 * segmented the frame. 2930 */ 2931 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 2932 features &= ~dev->gso_partial_features; 2933 2934 /* Make sure to clear the IPv4 ID mangling feature if the 2935 * IPv4 header has the potential to be fragmented. 2936 */ 2937 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 2938 struct iphdr *iph = skb->encapsulation ? 2939 inner_ip_hdr(skb) : ip_hdr(skb); 2940 2941 if (!(iph->frag_off & htons(IP_DF))) 2942 features &= ~NETIF_F_TSO_MANGLEID; 2943 } 2944 2945 return features; 2946 } 2947 2948 netdev_features_t netif_skb_features(struct sk_buff *skb) 2949 { 2950 struct net_device *dev = skb->dev; 2951 netdev_features_t features = dev->features; 2952 2953 if (skb_is_gso(skb)) 2954 features = gso_features_check(skb, dev, features); 2955 2956 /* If encapsulation offload request, verify we are testing 2957 * hardware encapsulation features instead of standard 2958 * features for the netdev 2959 */ 2960 if (skb->encapsulation) 2961 features &= dev->hw_enc_features; 2962 2963 if (skb_vlan_tagged(skb)) 2964 features = netdev_intersect_features(features, 2965 dev->vlan_features | 2966 NETIF_F_HW_VLAN_CTAG_TX | 2967 NETIF_F_HW_VLAN_STAG_TX); 2968 2969 if (dev->netdev_ops->ndo_features_check) 2970 features &= dev->netdev_ops->ndo_features_check(skb, dev, 2971 features); 2972 else 2973 features &= dflt_features_check(skb, dev, features); 2974 2975 return harmonize_features(skb, features); 2976 } 2977 EXPORT_SYMBOL(netif_skb_features); 2978 2979 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 2980 struct netdev_queue *txq, bool more) 2981 { 2982 unsigned int len; 2983 int rc; 2984 2985 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) 2986 dev_queue_xmit_nit(skb, dev); 2987 2988 len = skb->len; 2989 trace_net_dev_start_xmit(skb, dev); 2990 rc = netdev_start_xmit(skb, dev, txq, more); 2991 trace_net_dev_xmit(skb, rc, dev, len); 2992 2993 return rc; 2994 } 2995 2996 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 2997 struct netdev_queue *txq, int *ret) 2998 { 2999 struct sk_buff *skb = first; 3000 int rc = NETDEV_TX_OK; 3001 3002 while (skb) { 3003 struct sk_buff *next = skb->next; 3004 3005 skb->next = NULL; 3006 rc = xmit_one(skb, dev, txq, next != NULL); 3007 if (unlikely(!dev_xmit_complete(rc))) { 3008 skb->next = next; 3009 goto out; 3010 } 3011 3012 skb = next; 3013 if (netif_xmit_stopped(txq) && skb) { 3014 rc = NETDEV_TX_BUSY; 3015 break; 3016 } 3017 } 3018 3019 out: 3020 *ret = rc; 3021 return skb; 3022 } 3023 3024 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3025 netdev_features_t features) 3026 { 3027 if (skb_vlan_tag_present(skb) && 3028 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3029 skb = __vlan_hwaccel_push_inside(skb); 3030 return skb; 3031 } 3032 3033 int skb_csum_hwoffload_help(struct sk_buff *skb, 3034 const netdev_features_t features) 3035 { 3036 if (unlikely(skb->csum_not_inet)) 3037 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3038 skb_crc32c_csum_help(skb); 3039 3040 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); 3041 } 3042 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3043 3044 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) 3045 { 3046 netdev_features_t features; 3047 3048 features = netif_skb_features(skb); 3049 skb = validate_xmit_vlan(skb, features); 3050 if (unlikely(!skb)) 3051 goto out_null; 3052 3053 if (netif_needs_gso(skb, features)) { 3054 struct sk_buff *segs; 3055 3056 segs = skb_gso_segment(skb, features); 3057 if (IS_ERR(segs)) { 3058 goto out_kfree_skb; 3059 } else if (segs) { 3060 consume_skb(skb); 3061 skb = segs; 3062 } 3063 } else { 3064 if (skb_needs_linearize(skb, features) && 3065 __skb_linearize(skb)) 3066 goto out_kfree_skb; 3067 3068 if (validate_xmit_xfrm(skb, features)) 3069 goto out_kfree_skb; 3070 3071 /* If packet is not checksummed and device does not 3072 * support checksumming for this protocol, complete 3073 * checksumming here. 3074 */ 3075 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3076 if (skb->encapsulation) 3077 skb_set_inner_transport_header(skb, 3078 skb_checksum_start_offset(skb)); 3079 else 3080 skb_set_transport_header(skb, 3081 skb_checksum_start_offset(skb)); 3082 if (skb_csum_hwoffload_help(skb, features)) 3083 goto out_kfree_skb; 3084 } 3085 } 3086 3087 return skb; 3088 3089 out_kfree_skb: 3090 kfree_skb(skb); 3091 out_null: 3092 atomic_long_inc(&dev->tx_dropped); 3093 return NULL; 3094 } 3095 3096 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) 3097 { 3098 struct sk_buff *next, *head = NULL, *tail; 3099 3100 for (; skb != NULL; skb = next) { 3101 next = skb->next; 3102 skb->next = NULL; 3103 3104 /* in case skb wont be segmented, point to itself */ 3105 skb->prev = skb; 3106 3107 skb = validate_xmit_skb(skb, dev); 3108 if (!skb) 3109 continue; 3110 3111 if (!head) 3112 head = skb; 3113 else 3114 tail->next = skb; 3115 /* If skb was segmented, skb->prev points to 3116 * the last segment. If not, it still contains skb. 3117 */ 3118 tail = skb->prev; 3119 } 3120 return head; 3121 } 3122 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3123 3124 static void qdisc_pkt_len_init(struct sk_buff *skb) 3125 { 3126 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3127 3128 qdisc_skb_cb(skb)->pkt_len = skb->len; 3129 3130 /* To get more precise estimation of bytes sent on wire, 3131 * we add to pkt_len the headers size of all segments 3132 */ 3133 if (shinfo->gso_size) { 3134 unsigned int hdr_len; 3135 u16 gso_segs = shinfo->gso_segs; 3136 3137 /* mac layer + network layer */ 3138 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3139 3140 /* + transport layer */ 3141 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 3142 hdr_len += tcp_hdrlen(skb); 3143 else 3144 hdr_len += sizeof(struct udphdr); 3145 3146 if (shinfo->gso_type & SKB_GSO_DODGY) 3147 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3148 shinfo->gso_size); 3149 3150 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3151 } 3152 } 3153 3154 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3155 struct net_device *dev, 3156 struct netdev_queue *txq) 3157 { 3158 spinlock_t *root_lock = qdisc_lock(q); 3159 struct sk_buff *to_free = NULL; 3160 bool contended; 3161 int rc; 3162 3163 qdisc_calculate_pkt_len(skb, q); 3164 /* 3165 * Heuristic to force contended enqueues to serialize on a 3166 * separate lock before trying to get qdisc main lock. 3167 * This permits qdisc->running owner to get the lock more 3168 * often and dequeue packets faster. 3169 */ 3170 contended = qdisc_is_running(q); 3171 if (unlikely(contended)) 3172 spin_lock(&q->busylock); 3173 3174 spin_lock(root_lock); 3175 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3176 __qdisc_drop(skb, &to_free); 3177 rc = NET_XMIT_DROP; 3178 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3179 qdisc_run_begin(q)) { 3180 /* 3181 * This is a work-conserving queue; there are no old skbs 3182 * waiting to be sent out; and the qdisc is not running - 3183 * xmit the skb directly. 3184 */ 3185 3186 qdisc_bstats_update(q, skb); 3187 3188 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3189 if (unlikely(contended)) { 3190 spin_unlock(&q->busylock); 3191 contended = false; 3192 } 3193 __qdisc_run(q); 3194 } else 3195 qdisc_run_end(q); 3196 3197 rc = NET_XMIT_SUCCESS; 3198 } else { 3199 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3200 if (qdisc_run_begin(q)) { 3201 if (unlikely(contended)) { 3202 spin_unlock(&q->busylock); 3203 contended = false; 3204 } 3205 __qdisc_run(q); 3206 } 3207 } 3208 spin_unlock(root_lock); 3209 if (unlikely(to_free)) 3210 kfree_skb_list(to_free); 3211 if (unlikely(contended)) 3212 spin_unlock(&q->busylock); 3213 return rc; 3214 } 3215 3216 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3217 static void skb_update_prio(struct sk_buff *skb) 3218 { 3219 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); 3220 3221 if (!skb->priority && skb->sk && map) { 3222 unsigned int prioidx = 3223 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); 3224 3225 if (prioidx < map->priomap_len) 3226 skb->priority = map->priomap[prioidx]; 3227 } 3228 } 3229 #else 3230 #define skb_update_prio(skb) 3231 #endif 3232 3233 DEFINE_PER_CPU(int, xmit_recursion); 3234 EXPORT_SYMBOL(xmit_recursion); 3235 3236 /** 3237 * dev_loopback_xmit - loop back @skb 3238 * @net: network namespace this loopback is happening in 3239 * @sk: sk needed to be a netfilter okfn 3240 * @skb: buffer to transmit 3241 */ 3242 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3243 { 3244 skb_reset_mac_header(skb); 3245 __skb_pull(skb, skb_network_offset(skb)); 3246 skb->pkt_type = PACKET_LOOPBACK; 3247 skb->ip_summed = CHECKSUM_UNNECESSARY; 3248 WARN_ON(!skb_dst(skb)); 3249 skb_dst_force(skb); 3250 netif_rx_ni(skb); 3251 return 0; 3252 } 3253 EXPORT_SYMBOL(dev_loopback_xmit); 3254 3255 #ifdef CONFIG_NET_EGRESS 3256 static struct sk_buff * 3257 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3258 { 3259 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3260 struct tcf_result cl_res; 3261 3262 if (!miniq) 3263 return skb; 3264 3265 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3266 mini_qdisc_bstats_cpu_update(miniq, skb); 3267 3268 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3269 case TC_ACT_OK: 3270 case TC_ACT_RECLASSIFY: 3271 skb->tc_index = TC_H_MIN(cl_res.classid); 3272 break; 3273 case TC_ACT_SHOT: 3274 mini_qdisc_qstats_cpu_drop(miniq); 3275 *ret = NET_XMIT_DROP; 3276 kfree_skb(skb); 3277 return NULL; 3278 case TC_ACT_STOLEN: 3279 case TC_ACT_QUEUED: 3280 case TC_ACT_TRAP: 3281 *ret = NET_XMIT_SUCCESS; 3282 consume_skb(skb); 3283 return NULL; 3284 case TC_ACT_REDIRECT: 3285 /* No need to push/pop skb's mac_header here on egress! */ 3286 skb_do_redirect(skb); 3287 *ret = NET_XMIT_SUCCESS; 3288 return NULL; 3289 default: 3290 break; 3291 } 3292 3293 return skb; 3294 } 3295 #endif /* CONFIG_NET_EGRESS */ 3296 3297 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) 3298 { 3299 #ifdef CONFIG_XPS 3300 struct xps_dev_maps *dev_maps; 3301 struct xps_map *map; 3302 int queue_index = -1; 3303 3304 rcu_read_lock(); 3305 dev_maps = rcu_dereference(dev->xps_maps); 3306 if (dev_maps) { 3307 unsigned int tci = skb->sender_cpu - 1; 3308 3309 if (dev->num_tc) { 3310 tci *= dev->num_tc; 3311 tci += netdev_get_prio_tc_map(dev, skb->priority); 3312 } 3313 3314 map = rcu_dereference(dev_maps->cpu_map[tci]); 3315 if (map) { 3316 if (map->len == 1) 3317 queue_index = map->queues[0]; 3318 else 3319 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), 3320 map->len)]; 3321 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3322 queue_index = -1; 3323 } 3324 } 3325 rcu_read_unlock(); 3326 3327 return queue_index; 3328 #else 3329 return -1; 3330 #endif 3331 } 3332 3333 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) 3334 { 3335 struct sock *sk = skb->sk; 3336 int queue_index = sk_tx_queue_get(sk); 3337 3338 if (queue_index < 0 || skb->ooo_okay || 3339 queue_index >= dev->real_num_tx_queues) { 3340 int new_index = get_xps_queue(dev, skb); 3341 3342 if (new_index < 0) 3343 new_index = skb_tx_hash(dev, skb); 3344 3345 if (queue_index != new_index && sk && 3346 sk_fullsock(sk) && 3347 rcu_access_pointer(sk->sk_dst_cache)) 3348 sk_tx_queue_set(sk, new_index); 3349 3350 queue_index = new_index; 3351 } 3352 3353 return queue_index; 3354 } 3355 3356 struct netdev_queue *netdev_pick_tx(struct net_device *dev, 3357 struct sk_buff *skb, 3358 void *accel_priv) 3359 { 3360 int queue_index = 0; 3361 3362 #ifdef CONFIG_XPS 3363 u32 sender_cpu = skb->sender_cpu - 1; 3364 3365 if (sender_cpu >= (u32)NR_CPUS) 3366 skb->sender_cpu = raw_smp_processor_id() + 1; 3367 #endif 3368 3369 if (dev->real_num_tx_queues != 1) { 3370 const struct net_device_ops *ops = dev->netdev_ops; 3371 3372 if (ops->ndo_select_queue) 3373 queue_index = ops->ndo_select_queue(dev, skb, accel_priv, 3374 __netdev_pick_tx); 3375 else 3376 queue_index = __netdev_pick_tx(dev, skb); 3377 3378 if (!accel_priv) 3379 queue_index = netdev_cap_txqueue(dev, queue_index); 3380 } 3381 3382 skb_set_queue_mapping(skb, queue_index); 3383 return netdev_get_tx_queue(dev, queue_index); 3384 } 3385 3386 /** 3387 * __dev_queue_xmit - transmit a buffer 3388 * @skb: buffer to transmit 3389 * @accel_priv: private data used for L2 forwarding offload 3390 * 3391 * Queue a buffer for transmission to a network device. The caller must 3392 * have set the device and priority and built the buffer before calling 3393 * this function. The function can be called from an interrupt. 3394 * 3395 * A negative errno code is returned on a failure. A success does not 3396 * guarantee the frame will be transmitted as it may be dropped due 3397 * to congestion or traffic shaping. 3398 * 3399 * ----------------------------------------------------------------------------------- 3400 * I notice this method can also return errors from the queue disciplines, 3401 * including NET_XMIT_DROP, which is a positive value. So, errors can also 3402 * be positive. 3403 * 3404 * Regardless of the return value, the skb is consumed, so it is currently 3405 * difficult to retry a send to this method. (You can bump the ref count 3406 * before sending to hold a reference for retry if you are careful.) 3407 * 3408 * When calling this method, interrupts MUST be enabled. This is because 3409 * the BH enable code must have IRQs enabled so that it will not deadlock. 3410 * --BLG 3411 */ 3412 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) 3413 { 3414 struct net_device *dev = skb->dev; 3415 struct netdev_queue *txq; 3416 struct Qdisc *q; 3417 int rc = -ENOMEM; 3418 3419 skb_reset_mac_header(skb); 3420 3421 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 3422 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 3423 3424 /* Disable soft irqs for various locks below. Also 3425 * stops preemption for RCU. 3426 */ 3427 rcu_read_lock_bh(); 3428 3429 skb_update_prio(skb); 3430 3431 qdisc_pkt_len_init(skb); 3432 #ifdef CONFIG_NET_CLS_ACT 3433 skb->tc_at_ingress = 0; 3434 # ifdef CONFIG_NET_EGRESS 3435 if (static_key_false(&egress_needed)) { 3436 skb = sch_handle_egress(skb, &rc, dev); 3437 if (!skb) 3438 goto out; 3439 } 3440 # endif 3441 #endif 3442 /* If device/qdisc don't need skb->dst, release it right now while 3443 * its hot in this cpu cache. 3444 */ 3445 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 3446 skb_dst_drop(skb); 3447 else 3448 skb_dst_force(skb); 3449 3450 txq = netdev_pick_tx(dev, skb, accel_priv); 3451 q = rcu_dereference_bh(txq->qdisc); 3452 3453 trace_net_dev_queue(skb); 3454 if (q->enqueue) { 3455 rc = __dev_xmit_skb(skb, q, dev, txq); 3456 goto out; 3457 } 3458 3459 /* The device has no queue. Common case for software devices: 3460 * loopback, all the sorts of tunnels... 3461 3462 * Really, it is unlikely that netif_tx_lock protection is necessary 3463 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 3464 * counters.) 3465 * However, it is possible, that they rely on protection 3466 * made by us here. 3467 3468 * Check this and shot the lock. It is not prone from deadlocks. 3469 *Either shot noqueue qdisc, it is even simpler 8) 3470 */ 3471 if (dev->flags & IFF_UP) { 3472 int cpu = smp_processor_id(); /* ok because BHs are off */ 3473 3474 if (txq->xmit_lock_owner != cpu) { 3475 if (unlikely(__this_cpu_read(xmit_recursion) > 3476 XMIT_RECURSION_LIMIT)) 3477 goto recursion_alert; 3478 3479 skb = validate_xmit_skb(skb, dev); 3480 if (!skb) 3481 goto out; 3482 3483 HARD_TX_LOCK(dev, txq, cpu); 3484 3485 if (!netif_xmit_stopped(txq)) { 3486 __this_cpu_inc(xmit_recursion); 3487 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 3488 __this_cpu_dec(xmit_recursion); 3489 if (dev_xmit_complete(rc)) { 3490 HARD_TX_UNLOCK(dev, txq); 3491 goto out; 3492 } 3493 } 3494 HARD_TX_UNLOCK(dev, txq); 3495 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 3496 dev->name); 3497 } else { 3498 /* Recursion is detected! It is possible, 3499 * unfortunately 3500 */ 3501 recursion_alert: 3502 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 3503 dev->name); 3504 } 3505 } 3506 3507 rc = -ENETDOWN; 3508 rcu_read_unlock_bh(); 3509 3510 atomic_long_inc(&dev->tx_dropped); 3511 kfree_skb_list(skb); 3512 return rc; 3513 out: 3514 rcu_read_unlock_bh(); 3515 return rc; 3516 } 3517 3518 int dev_queue_xmit(struct sk_buff *skb) 3519 { 3520 return __dev_queue_xmit(skb, NULL); 3521 } 3522 EXPORT_SYMBOL(dev_queue_xmit); 3523 3524 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) 3525 { 3526 return __dev_queue_xmit(skb, accel_priv); 3527 } 3528 EXPORT_SYMBOL(dev_queue_xmit_accel); 3529 3530 3531 /************************************************************************* 3532 * Receiver routines 3533 *************************************************************************/ 3534 3535 int netdev_max_backlog __read_mostly = 1000; 3536 EXPORT_SYMBOL(netdev_max_backlog); 3537 3538 int netdev_tstamp_prequeue __read_mostly = 1; 3539 int netdev_budget __read_mostly = 300; 3540 unsigned int __read_mostly netdev_budget_usecs = 2000; 3541 int weight_p __read_mostly = 64; /* old backlog weight */ 3542 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 3543 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 3544 int dev_rx_weight __read_mostly = 64; 3545 int dev_tx_weight __read_mostly = 64; 3546 3547 /* Called with irq disabled */ 3548 static inline void ____napi_schedule(struct softnet_data *sd, 3549 struct napi_struct *napi) 3550 { 3551 list_add_tail(&napi->poll_list, &sd->poll_list); 3552 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3553 } 3554 3555 #ifdef CONFIG_RPS 3556 3557 /* One global table that all flow-based protocols share. */ 3558 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3559 EXPORT_SYMBOL(rps_sock_flow_table); 3560 u32 rps_cpu_mask __read_mostly; 3561 EXPORT_SYMBOL(rps_cpu_mask); 3562 3563 struct static_key rps_needed __read_mostly; 3564 EXPORT_SYMBOL(rps_needed); 3565 struct static_key rfs_needed __read_mostly; 3566 EXPORT_SYMBOL(rfs_needed); 3567 3568 static struct rps_dev_flow * 3569 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3570 struct rps_dev_flow *rflow, u16 next_cpu) 3571 { 3572 if (next_cpu < nr_cpu_ids) { 3573 #ifdef CONFIG_RFS_ACCEL 3574 struct netdev_rx_queue *rxqueue; 3575 struct rps_dev_flow_table *flow_table; 3576 struct rps_dev_flow *old_rflow; 3577 u32 flow_id; 3578 u16 rxq_index; 3579 int rc; 3580 3581 /* Should we steer this flow to a different hardware queue? */ 3582 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3583 !(dev->features & NETIF_F_NTUPLE)) 3584 goto out; 3585 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 3586 if (rxq_index == skb_get_rx_queue(skb)) 3587 goto out; 3588 3589 rxqueue = dev->_rx + rxq_index; 3590 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3591 if (!flow_table) 3592 goto out; 3593 flow_id = skb_get_hash(skb) & flow_table->mask; 3594 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 3595 rxq_index, flow_id); 3596 if (rc < 0) 3597 goto out; 3598 old_rflow = rflow; 3599 rflow = &flow_table->flows[flow_id]; 3600 rflow->filter = rc; 3601 if (old_rflow->filter == rflow->filter) 3602 old_rflow->filter = RPS_NO_FILTER; 3603 out: 3604 #endif 3605 rflow->last_qtail = 3606 per_cpu(softnet_data, next_cpu).input_queue_head; 3607 } 3608 3609 rflow->cpu = next_cpu; 3610 return rflow; 3611 } 3612 3613 /* 3614 * get_rps_cpu is called from netif_receive_skb and returns the target 3615 * CPU from the RPS map of the receiving queue for a given skb. 3616 * rcu_read_lock must be held on entry. 3617 */ 3618 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3619 struct rps_dev_flow **rflowp) 3620 { 3621 const struct rps_sock_flow_table *sock_flow_table; 3622 struct netdev_rx_queue *rxqueue = dev->_rx; 3623 struct rps_dev_flow_table *flow_table; 3624 struct rps_map *map; 3625 int cpu = -1; 3626 u32 tcpu; 3627 u32 hash; 3628 3629 if (skb_rx_queue_recorded(skb)) { 3630 u16 index = skb_get_rx_queue(skb); 3631 3632 if (unlikely(index >= dev->real_num_rx_queues)) { 3633 WARN_ONCE(dev->real_num_rx_queues > 1, 3634 "%s received packet on queue %u, but number " 3635 "of RX queues is %u\n", 3636 dev->name, index, dev->real_num_rx_queues); 3637 goto done; 3638 } 3639 rxqueue += index; 3640 } 3641 3642 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 3643 3644 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3645 map = rcu_dereference(rxqueue->rps_map); 3646 if (!flow_table && !map) 3647 goto done; 3648 3649 skb_reset_network_header(skb); 3650 hash = skb_get_hash(skb); 3651 if (!hash) 3652 goto done; 3653 3654 sock_flow_table = rcu_dereference(rps_sock_flow_table); 3655 if (flow_table && sock_flow_table) { 3656 struct rps_dev_flow *rflow; 3657 u32 next_cpu; 3658 u32 ident; 3659 3660 /* First check into global flow table if there is a match */ 3661 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 3662 if ((ident ^ hash) & ~rps_cpu_mask) 3663 goto try_rps; 3664 3665 next_cpu = ident & rps_cpu_mask; 3666 3667 /* OK, now we know there is a match, 3668 * we can look at the local (per receive queue) flow table 3669 */ 3670 rflow = &flow_table->flows[hash & flow_table->mask]; 3671 tcpu = rflow->cpu; 3672 3673 /* 3674 * If the desired CPU (where last recvmsg was done) is 3675 * different from current CPU (one in the rx-queue flow 3676 * table entry), switch if one of the following holds: 3677 * - Current CPU is unset (>= nr_cpu_ids). 3678 * - Current CPU is offline. 3679 * - The current CPU's queue tail has advanced beyond the 3680 * last packet that was enqueued using this table entry. 3681 * This guarantees that all previous packets for the flow 3682 * have been dequeued, thus preserving in order delivery. 3683 */ 3684 if (unlikely(tcpu != next_cpu) && 3685 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 3686 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 3687 rflow->last_qtail)) >= 0)) { 3688 tcpu = next_cpu; 3689 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 3690 } 3691 3692 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 3693 *rflowp = rflow; 3694 cpu = tcpu; 3695 goto done; 3696 } 3697 } 3698 3699 try_rps: 3700 3701 if (map) { 3702 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 3703 if (cpu_online(tcpu)) { 3704 cpu = tcpu; 3705 goto done; 3706 } 3707 } 3708 3709 done: 3710 return cpu; 3711 } 3712 3713 #ifdef CONFIG_RFS_ACCEL 3714 3715 /** 3716 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 3717 * @dev: Device on which the filter was set 3718 * @rxq_index: RX queue index 3719 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 3720 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 3721 * 3722 * Drivers that implement ndo_rx_flow_steer() should periodically call 3723 * this function for each installed filter and remove the filters for 3724 * which it returns %true. 3725 */ 3726 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 3727 u32 flow_id, u16 filter_id) 3728 { 3729 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 3730 struct rps_dev_flow_table *flow_table; 3731 struct rps_dev_flow *rflow; 3732 bool expire = true; 3733 unsigned int cpu; 3734 3735 rcu_read_lock(); 3736 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3737 if (flow_table && flow_id <= flow_table->mask) { 3738 rflow = &flow_table->flows[flow_id]; 3739 cpu = READ_ONCE(rflow->cpu); 3740 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 3741 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 3742 rflow->last_qtail) < 3743 (int)(10 * flow_table->mask))) 3744 expire = false; 3745 } 3746 rcu_read_unlock(); 3747 return expire; 3748 } 3749 EXPORT_SYMBOL(rps_may_expire_flow); 3750 3751 #endif /* CONFIG_RFS_ACCEL */ 3752 3753 /* Called from hardirq (IPI) context */ 3754 static void rps_trigger_softirq(void *data) 3755 { 3756 struct softnet_data *sd = data; 3757 3758 ____napi_schedule(sd, &sd->backlog); 3759 sd->received_rps++; 3760 } 3761 3762 #endif /* CONFIG_RPS */ 3763 3764 /* 3765 * Check if this softnet_data structure is another cpu one 3766 * If yes, queue it to our IPI list and return 1 3767 * If no, return 0 3768 */ 3769 static int rps_ipi_queued(struct softnet_data *sd) 3770 { 3771 #ifdef CONFIG_RPS 3772 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 3773 3774 if (sd != mysd) { 3775 sd->rps_ipi_next = mysd->rps_ipi_list; 3776 mysd->rps_ipi_list = sd; 3777 3778 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3779 return 1; 3780 } 3781 #endif /* CONFIG_RPS */ 3782 return 0; 3783 } 3784 3785 #ifdef CONFIG_NET_FLOW_LIMIT 3786 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 3787 #endif 3788 3789 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 3790 { 3791 #ifdef CONFIG_NET_FLOW_LIMIT 3792 struct sd_flow_limit *fl; 3793 struct softnet_data *sd; 3794 unsigned int old_flow, new_flow; 3795 3796 if (qlen < (netdev_max_backlog >> 1)) 3797 return false; 3798 3799 sd = this_cpu_ptr(&softnet_data); 3800 3801 rcu_read_lock(); 3802 fl = rcu_dereference(sd->flow_limit); 3803 if (fl) { 3804 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 3805 old_flow = fl->history[fl->history_head]; 3806 fl->history[fl->history_head] = new_flow; 3807 3808 fl->history_head++; 3809 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 3810 3811 if (likely(fl->buckets[old_flow])) 3812 fl->buckets[old_flow]--; 3813 3814 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 3815 fl->count++; 3816 rcu_read_unlock(); 3817 return true; 3818 } 3819 } 3820 rcu_read_unlock(); 3821 #endif 3822 return false; 3823 } 3824 3825 /* 3826 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 3827 * queue (may be a remote CPU queue). 3828 */ 3829 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 3830 unsigned int *qtail) 3831 { 3832 struct softnet_data *sd; 3833 unsigned long flags; 3834 unsigned int qlen; 3835 3836 sd = &per_cpu(softnet_data, cpu); 3837 3838 local_irq_save(flags); 3839 3840 rps_lock(sd); 3841 if (!netif_running(skb->dev)) 3842 goto drop; 3843 qlen = skb_queue_len(&sd->input_pkt_queue); 3844 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 3845 if (qlen) { 3846 enqueue: 3847 __skb_queue_tail(&sd->input_pkt_queue, skb); 3848 input_queue_tail_incr_save(sd, qtail); 3849 rps_unlock(sd); 3850 local_irq_restore(flags); 3851 return NET_RX_SUCCESS; 3852 } 3853 3854 /* Schedule NAPI for backlog device 3855 * We can use non atomic operation since we own the queue lock 3856 */ 3857 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 3858 if (!rps_ipi_queued(sd)) 3859 ____napi_schedule(sd, &sd->backlog); 3860 } 3861 goto enqueue; 3862 } 3863 3864 drop: 3865 sd->dropped++; 3866 rps_unlock(sd); 3867 3868 local_irq_restore(flags); 3869 3870 atomic_long_inc(&skb->dev->rx_dropped); 3871 kfree_skb(skb); 3872 return NET_RX_DROP; 3873 } 3874 3875 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 3876 struct bpf_prog *xdp_prog) 3877 { 3878 u32 metalen, act = XDP_DROP; 3879 struct xdp_buff xdp; 3880 void *orig_data; 3881 int hlen, off; 3882 u32 mac_len; 3883 3884 /* Reinjected packets coming from act_mirred or similar should 3885 * not get XDP generic processing. 3886 */ 3887 if (skb_cloned(skb)) 3888 return XDP_PASS; 3889 3890 /* XDP packets must be linear and must have sufficient headroom 3891 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 3892 * native XDP provides, thus we need to do it here as well. 3893 */ 3894 if (skb_is_nonlinear(skb) || 3895 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 3896 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 3897 int troom = skb->tail + skb->data_len - skb->end; 3898 3899 /* In case we have to go down the path and also linearize, 3900 * then lets do the pskb_expand_head() work just once here. 3901 */ 3902 if (pskb_expand_head(skb, 3903 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 3904 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 3905 goto do_drop; 3906 if (troom > 0 && __skb_linearize(skb)) 3907 goto do_drop; 3908 } 3909 3910 /* The XDP program wants to see the packet starting at the MAC 3911 * header. 3912 */ 3913 mac_len = skb->data - skb_mac_header(skb); 3914 hlen = skb_headlen(skb) + mac_len; 3915 xdp.data = skb->data - mac_len; 3916 xdp.data_meta = xdp.data; 3917 xdp.data_end = xdp.data + hlen; 3918 xdp.data_hard_start = skb->data - skb_headroom(skb); 3919 orig_data = xdp.data; 3920 3921 act = bpf_prog_run_xdp(xdp_prog, &xdp); 3922 3923 off = xdp.data - orig_data; 3924 if (off > 0) 3925 __skb_pull(skb, off); 3926 else if (off < 0) 3927 __skb_push(skb, -off); 3928 skb->mac_header += off; 3929 3930 switch (act) { 3931 case XDP_REDIRECT: 3932 case XDP_TX: 3933 __skb_push(skb, mac_len); 3934 break; 3935 case XDP_PASS: 3936 metalen = xdp.data - xdp.data_meta; 3937 if (metalen) 3938 skb_metadata_set(skb, metalen); 3939 break; 3940 default: 3941 bpf_warn_invalid_xdp_action(act); 3942 /* fall through */ 3943 case XDP_ABORTED: 3944 trace_xdp_exception(skb->dev, xdp_prog, act); 3945 /* fall through */ 3946 case XDP_DROP: 3947 do_drop: 3948 kfree_skb(skb); 3949 break; 3950 } 3951 3952 return act; 3953 } 3954 3955 /* When doing generic XDP we have to bypass the qdisc layer and the 3956 * network taps in order to match in-driver-XDP behavior. 3957 */ 3958 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 3959 { 3960 struct net_device *dev = skb->dev; 3961 struct netdev_queue *txq; 3962 bool free_skb = true; 3963 int cpu, rc; 3964 3965 txq = netdev_pick_tx(dev, skb, NULL); 3966 cpu = smp_processor_id(); 3967 HARD_TX_LOCK(dev, txq, cpu); 3968 if (!netif_xmit_stopped(txq)) { 3969 rc = netdev_start_xmit(skb, dev, txq, 0); 3970 if (dev_xmit_complete(rc)) 3971 free_skb = false; 3972 } 3973 HARD_TX_UNLOCK(dev, txq); 3974 if (free_skb) { 3975 trace_xdp_exception(dev, xdp_prog, XDP_TX); 3976 kfree_skb(skb); 3977 } 3978 } 3979 EXPORT_SYMBOL_GPL(generic_xdp_tx); 3980 3981 static struct static_key generic_xdp_needed __read_mostly; 3982 3983 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 3984 { 3985 if (xdp_prog) { 3986 u32 act = netif_receive_generic_xdp(skb, xdp_prog); 3987 int err; 3988 3989 if (act != XDP_PASS) { 3990 switch (act) { 3991 case XDP_REDIRECT: 3992 err = xdp_do_generic_redirect(skb->dev, skb, 3993 xdp_prog); 3994 if (err) 3995 goto out_redir; 3996 /* fallthru to submit skb */ 3997 case XDP_TX: 3998 generic_xdp_tx(skb, xdp_prog); 3999 break; 4000 } 4001 return XDP_DROP; 4002 } 4003 } 4004 return XDP_PASS; 4005 out_redir: 4006 kfree_skb(skb); 4007 return XDP_DROP; 4008 } 4009 EXPORT_SYMBOL_GPL(do_xdp_generic); 4010 4011 static int netif_rx_internal(struct sk_buff *skb) 4012 { 4013 int ret; 4014 4015 net_timestamp_check(netdev_tstamp_prequeue, skb); 4016 4017 trace_netif_rx(skb); 4018 4019 if (static_key_false(&generic_xdp_needed)) { 4020 int ret; 4021 4022 preempt_disable(); 4023 rcu_read_lock(); 4024 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 4025 rcu_read_unlock(); 4026 preempt_enable(); 4027 4028 /* Consider XDP consuming the packet a success from 4029 * the netdev point of view we do not want to count 4030 * this as an error. 4031 */ 4032 if (ret != XDP_PASS) 4033 return NET_RX_SUCCESS; 4034 } 4035 4036 #ifdef CONFIG_RPS 4037 if (static_key_false(&rps_needed)) { 4038 struct rps_dev_flow voidflow, *rflow = &voidflow; 4039 int cpu; 4040 4041 preempt_disable(); 4042 rcu_read_lock(); 4043 4044 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4045 if (cpu < 0) 4046 cpu = smp_processor_id(); 4047 4048 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4049 4050 rcu_read_unlock(); 4051 preempt_enable(); 4052 } else 4053 #endif 4054 { 4055 unsigned int qtail; 4056 4057 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4058 put_cpu(); 4059 } 4060 return ret; 4061 } 4062 4063 /** 4064 * netif_rx - post buffer to the network code 4065 * @skb: buffer to post 4066 * 4067 * This function receives a packet from a device driver and queues it for 4068 * the upper (protocol) levels to process. It always succeeds. The buffer 4069 * may be dropped during processing for congestion control or by the 4070 * protocol layers. 4071 * 4072 * return values: 4073 * NET_RX_SUCCESS (no congestion) 4074 * NET_RX_DROP (packet was dropped) 4075 * 4076 */ 4077 4078 int netif_rx(struct sk_buff *skb) 4079 { 4080 trace_netif_rx_entry(skb); 4081 4082 return netif_rx_internal(skb); 4083 } 4084 EXPORT_SYMBOL(netif_rx); 4085 4086 int netif_rx_ni(struct sk_buff *skb) 4087 { 4088 int err; 4089 4090 trace_netif_rx_ni_entry(skb); 4091 4092 preempt_disable(); 4093 err = netif_rx_internal(skb); 4094 if (local_softirq_pending()) 4095 do_softirq(); 4096 preempt_enable(); 4097 4098 return err; 4099 } 4100 EXPORT_SYMBOL(netif_rx_ni); 4101 4102 static __latent_entropy void net_tx_action(struct softirq_action *h) 4103 { 4104 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4105 4106 if (sd->completion_queue) { 4107 struct sk_buff *clist; 4108 4109 local_irq_disable(); 4110 clist = sd->completion_queue; 4111 sd->completion_queue = NULL; 4112 local_irq_enable(); 4113 4114 while (clist) { 4115 struct sk_buff *skb = clist; 4116 4117 clist = clist->next; 4118 4119 WARN_ON(refcount_read(&skb->users)); 4120 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4121 trace_consume_skb(skb); 4122 else 4123 trace_kfree_skb(skb, net_tx_action); 4124 4125 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4126 __kfree_skb(skb); 4127 else 4128 __kfree_skb_defer(skb); 4129 } 4130 4131 __kfree_skb_flush(); 4132 } 4133 4134 if (sd->output_queue) { 4135 struct Qdisc *head; 4136 4137 local_irq_disable(); 4138 head = sd->output_queue; 4139 sd->output_queue = NULL; 4140 sd->output_queue_tailp = &sd->output_queue; 4141 local_irq_enable(); 4142 4143 while (head) { 4144 struct Qdisc *q = head; 4145 spinlock_t *root_lock; 4146 4147 head = head->next_sched; 4148 4149 root_lock = qdisc_lock(q); 4150 spin_lock(root_lock); 4151 /* We need to make sure head->next_sched is read 4152 * before clearing __QDISC_STATE_SCHED 4153 */ 4154 smp_mb__before_atomic(); 4155 clear_bit(__QDISC_STATE_SCHED, &q->state); 4156 qdisc_run(q); 4157 spin_unlock(root_lock); 4158 } 4159 } 4160 } 4161 4162 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4163 /* This hook is defined here for ATM LANE */ 4164 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4165 unsigned char *addr) __read_mostly; 4166 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4167 #endif 4168 4169 static inline struct sk_buff * 4170 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4171 struct net_device *orig_dev) 4172 { 4173 #ifdef CONFIG_NET_CLS_ACT 4174 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4175 struct tcf_result cl_res; 4176 4177 /* If there's at least one ingress present somewhere (so 4178 * we get here via enabled static key), remaining devices 4179 * that are not configured with an ingress qdisc will bail 4180 * out here. 4181 */ 4182 if (!miniq) 4183 return skb; 4184 4185 if (*pt_prev) { 4186 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4187 *pt_prev = NULL; 4188 } 4189 4190 qdisc_skb_cb(skb)->pkt_len = skb->len; 4191 skb->tc_at_ingress = 1; 4192 mini_qdisc_bstats_cpu_update(miniq, skb); 4193 4194 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 4195 case TC_ACT_OK: 4196 case TC_ACT_RECLASSIFY: 4197 skb->tc_index = TC_H_MIN(cl_res.classid); 4198 break; 4199 case TC_ACT_SHOT: 4200 mini_qdisc_qstats_cpu_drop(miniq); 4201 kfree_skb(skb); 4202 return NULL; 4203 case TC_ACT_STOLEN: 4204 case TC_ACT_QUEUED: 4205 case TC_ACT_TRAP: 4206 consume_skb(skb); 4207 return NULL; 4208 case TC_ACT_REDIRECT: 4209 /* skb_mac_header check was done by cls/act_bpf, so 4210 * we can safely push the L2 header back before 4211 * redirecting to another netdev 4212 */ 4213 __skb_push(skb, skb->mac_len); 4214 skb_do_redirect(skb); 4215 return NULL; 4216 default: 4217 break; 4218 } 4219 #endif /* CONFIG_NET_CLS_ACT */ 4220 return skb; 4221 } 4222 4223 /** 4224 * netdev_is_rx_handler_busy - check if receive handler is registered 4225 * @dev: device to check 4226 * 4227 * Check if a receive handler is already registered for a given device. 4228 * Return true if there one. 4229 * 4230 * The caller must hold the rtnl_mutex. 4231 */ 4232 bool netdev_is_rx_handler_busy(struct net_device *dev) 4233 { 4234 ASSERT_RTNL(); 4235 return dev && rtnl_dereference(dev->rx_handler); 4236 } 4237 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 4238 4239 /** 4240 * netdev_rx_handler_register - register receive handler 4241 * @dev: device to register a handler for 4242 * @rx_handler: receive handler to register 4243 * @rx_handler_data: data pointer that is used by rx handler 4244 * 4245 * Register a receive handler for a device. This handler will then be 4246 * called from __netif_receive_skb. A negative errno code is returned 4247 * on a failure. 4248 * 4249 * The caller must hold the rtnl_mutex. 4250 * 4251 * For a general description of rx_handler, see enum rx_handler_result. 4252 */ 4253 int netdev_rx_handler_register(struct net_device *dev, 4254 rx_handler_func_t *rx_handler, 4255 void *rx_handler_data) 4256 { 4257 if (netdev_is_rx_handler_busy(dev)) 4258 return -EBUSY; 4259 4260 /* Note: rx_handler_data must be set before rx_handler */ 4261 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 4262 rcu_assign_pointer(dev->rx_handler, rx_handler); 4263 4264 return 0; 4265 } 4266 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 4267 4268 /** 4269 * netdev_rx_handler_unregister - unregister receive handler 4270 * @dev: device to unregister a handler from 4271 * 4272 * Unregister a receive handler from a device. 4273 * 4274 * The caller must hold the rtnl_mutex. 4275 */ 4276 void netdev_rx_handler_unregister(struct net_device *dev) 4277 { 4278 4279 ASSERT_RTNL(); 4280 RCU_INIT_POINTER(dev->rx_handler, NULL); 4281 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 4282 * section has a guarantee to see a non NULL rx_handler_data 4283 * as well. 4284 */ 4285 synchronize_net(); 4286 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 4287 } 4288 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 4289 4290 /* 4291 * Limit the use of PFMEMALLOC reserves to those protocols that implement 4292 * the special handling of PFMEMALLOC skbs. 4293 */ 4294 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 4295 { 4296 switch (skb->protocol) { 4297 case htons(ETH_P_ARP): 4298 case htons(ETH_P_IP): 4299 case htons(ETH_P_IPV6): 4300 case htons(ETH_P_8021Q): 4301 case htons(ETH_P_8021AD): 4302 return true; 4303 default: 4304 return false; 4305 } 4306 } 4307 4308 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 4309 int *ret, struct net_device *orig_dev) 4310 { 4311 #ifdef CONFIG_NETFILTER_INGRESS 4312 if (nf_hook_ingress_active(skb)) { 4313 int ingress_retval; 4314 4315 if (*pt_prev) { 4316 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4317 *pt_prev = NULL; 4318 } 4319 4320 rcu_read_lock(); 4321 ingress_retval = nf_hook_ingress(skb); 4322 rcu_read_unlock(); 4323 return ingress_retval; 4324 } 4325 #endif /* CONFIG_NETFILTER_INGRESS */ 4326 return 0; 4327 } 4328 4329 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) 4330 { 4331 struct packet_type *ptype, *pt_prev; 4332 rx_handler_func_t *rx_handler; 4333 struct net_device *orig_dev; 4334 bool deliver_exact = false; 4335 int ret = NET_RX_DROP; 4336 __be16 type; 4337 4338 net_timestamp_check(!netdev_tstamp_prequeue, skb); 4339 4340 trace_netif_receive_skb(skb); 4341 4342 orig_dev = skb->dev; 4343 4344 skb_reset_network_header(skb); 4345 if (!skb_transport_header_was_set(skb)) 4346 skb_reset_transport_header(skb); 4347 skb_reset_mac_len(skb); 4348 4349 pt_prev = NULL; 4350 4351 another_round: 4352 skb->skb_iif = skb->dev->ifindex; 4353 4354 __this_cpu_inc(softnet_data.processed); 4355 4356 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 4357 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 4358 skb = skb_vlan_untag(skb); 4359 if (unlikely(!skb)) 4360 goto out; 4361 } 4362 4363 if (skb_skip_tc_classify(skb)) 4364 goto skip_classify; 4365 4366 if (pfmemalloc) 4367 goto skip_taps; 4368 4369 list_for_each_entry_rcu(ptype, &ptype_all, list) { 4370 if (pt_prev) 4371 ret = deliver_skb(skb, pt_prev, orig_dev); 4372 pt_prev = ptype; 4373 } 4374 4375 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 4376 if (pt_prev) 4377 ret = deliver_skb(skb, pt_prev, orig_dev); 4378 pt_prev = ptype; 4379 } 4380 4381 skip_taps: 4382 #ifdef CONFIG_NET_INGRESS 4383 if (static_key_false(&ingress_needed)) { 4384 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); 4385 if (!skb) 4386 goto out; 4387 4388 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 4389 goto out; 4390 } 4391 #endif 4392 skb_reset_tc(skb); 4393 skip_classify: 4394 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 4395 goto drop; 4396 4397 if (skb_vlan_tag_present(skb)) { 4398 if (pt_prev) { 4399 ret = deliver_skb(skb, pt_prev, orig_dev); 4400 pt_prev = NULL; 4401 } 4402 if (vlan_do_receive(&skb)) 4403 goto another_round; 4404 else if (unlikely(!skb)) 4405 goto out; 4406 } 4407 4408 rx_handler = rcu_dereference(skb->dev->rx_handler); 4409 if (rx_handler) { 4410 if (pt_prev) { 4411 ret = deliver_skb(skb, pt_prev, orig_dev); 4412 pt_prev = NULL; 4413 } 4414 switch (rx_handler(&skb)) { 4415 case RX_HANDLER_CONSUMED: 4416 ret = NET_RX_SUCCESS; 4417 goto out; 4418 case RX_HANDLER_ANOTHER: 4419 goto another_round; 4420 case RX_HANDLER_EXACT: 4421 deliver_exact = true; 4422 case RX_HANDLER_PASS: 4423 break; 4424 default: 4425 BUG(); 4426 } 4427 } 4428 4429 if (unlikely(skb_vlan_tag_present(skb))) { 4430 if (skb_vlan_tag_get_id(skb)) 4431 skb->pkt_type = PACKET_OTHERHOST; 4432 /* Note: we might in the future use prio bits 4433 * and set skb->priority like in vlan_do_receive() 4434 * For the time being, just ignore Priority Code Point 4435 */ 4436 skb->vlan_tci = 0; 4437 } 4438 4439 type = skb->protocol; 4440 4441 /* deliver only exact match when indicated */ 4442 if (likely(!deliver_exact)) { 4443 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4444 &ptype_base[ntohs(type) & 4445 PTYPE_HASH_MASK]); 4446 } 4447 4448 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4449 &orig_dev->ptype_specific); 4450 4451 if (unlikely(skb->dev != orig_dev)) { 4452 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4453 &skb->dev->ptype_specific); 4454 } 4455 4456 if (pt_prev) { 4457 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 4458 goto drop; 4459 else 4460 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 4461 } else { 4462 drop: 4463 if (!deliver_exact) 4464 atomic_long_inc(&skb->dev->rx_dropped); 4465 else 4466 atomic_long_inc(&skb->dev->rx_nohandler); 4467 kfree_skb(skb); 4468 /* Jamal, now you will not able to escape explaining 4469 * me how you were going to use this. :-) 4470 */ 4471 ret = NET_RX_DROP; 4472 } 4473 4474 out: 4475 return ret; 4476 } 4477 4478 /** 4479 * netif_receive_skb_core - special purpose version of netif_receive_skb 4480 * @skb: buffer to process 4481 * 4482 * More direct receive version of netif_receive_skb(). It should 4483 * only be used by callers that have a need to skip RPS and Generic XDP. 4484 * Caller must also take care of handling if (page_is_)pfmemalloc. 4485 * 4486 * This function may only be called from softirq context and interrupts 4487 * should be enabled. 4488 * 4489 * Return values (usually ignored): 4490 * NET_RX_SUCCESS: no congestion 4491 * NET_RX_DROP: packet was dropped 4492 */ 4493 int netif_receive_skb_core(struct sk_buff *skb) 4494 { 4495 int ret; 4496 4497 rcu_read_lock(); 4498 ret = __netif_receive_skb_core(skb, false); 4499 rcu_read_unlock(); 4500 4501 return ret; 4502 } 4503 EXPORT_SYMBOL(netif_receive_skb_core); 4504 4505 static int __netif_receive_skb(struct sk_buff *skb) 4506 { 4507 int ret; 4508 4509 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 4510 unsigned int noreclaim_flag; 4511 4512 /* 4513 * PFMEMALLOC skbs are special, they should 4514 * - be delivered to SOCK_MEMALLOC sockets only 4515 * - stay away from userspace 4516 * - have bounded memory usage 4517 * 4518 * Use PF_MEMALLOC as this saves us from propagating the allocation 4519 * context down to all allocation sites. 4520 */ 4521 noreclaim_flag = memalloc_noreclaim_save(); 4522 ret = __netif_receive_skb_core(skb, true); 4523 memalloc_noreclaim_restore(noreclaim_flag); 4524 } else 4525 ret = __netif_receive_skb_core(skb, false); 4526 4527 return ret; 4528 } 4529 4530 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 4531 { 4532 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 4533 struct bpf_prog *new = xdp->prog; 4534 int ret = 0; 4535 4536 switch (xdp->command) { 4537 case XDP_SETUP_PROG: 4538 rcu_assign_pointer(dev->xdp_prog, new); 4539 if (old) 4540 bpf_prog_put(old); 4541 4542 if (old && !new) { 4543 static_key_slow_dec(&generic_xdp_needed); 4544 } else if (new && !old) { 4545 static_key_slow_inc(&generic_xdp_needed); 4546 dev_disable_lro(dev); 4547 } 4548 break; 4549 4550 case XDP_QUERY_PROG: 4551 xdp->prog_attached = !!old; 4552 xdp->prog_id = old ? old->aux->id : 0; 4553 break; 4554 4555 default: 4556 ret = -EINVAL; 4557 break; 4558 } 4559 4560 return ret; 4561 } 4562 4563 static int netif_receive_skb_internal(struct sk_buff *skb) 4564 { 4565 int ret; 4566 4567 net_timestamp_check(netdev_tstamp_prequeue, skb); 4568 4569 if (skb_defer_rx_timestamp(skb)) 4570 return NET_RX_SUCCESS; 4571 4572 if (static_key_false(&generic_xdp_needed)) { 4573 int ret; 4574 4575 preempt_disable(); 4576 rcu_read_lock(); 4577 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 4578 rcu_read_unlock(); 4579 preempt_enable(); 4580 4581 if (ret != XDP_PASS) 4582 return NET_RX_DROP; 4583 } 4584 4585 rcu_read_lock(); 4586 #ifdef CONFIG_RPS 4587 if (static_key_false(&rps_needed)) { 4588 struct rps_dev_flow voidflow, *rflow = &voidflow; 4589 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 4590 4591 if (cpu >= 0) { 4592 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4593 rcu_read_unlock(); 4594 return ret; 4595 } 4596 } 4597 #endif 4598 ret = __netif_receive_skb(skb); 4599 rcu_read_unlock(); 4600 return ret; 4601 } 4602 4603 /** 4604 * netif_receive_skb - process receive buffer from network 4605 * @skb: buffer to process 4606 * 4607 * netif_receive_skb() is the main receive data processing function. 4608 * It always succeeds. The buffer may be dropped during processing 4609 * for congestion control or by the protocol layers. 4610 * 4611 * This function may only be called from softirq context and interrupts 4612 * should be enabled. 4613 * 4614 * Return values (usually ignored): 4615 * NET_RX_SUCCESS: no congestion 4616 * NET_RX_DROP: packet was dropped 4617 */ 4618 int netif_receive_skb(struct sk_buff *skb) 4619 { 4620 trace_netif_receive_skb_entry(skb); 4621 4622 return netif_receive_skb_internal(skb); 4623 } 4624 EXPORT_SYMBOL(netif_receive_skb); 4625 4626 DEFINE_PER_CPU(struct work_struct, flush_works); 4627 4628 /* Network device is going away, flush any packets still pending */ 4629 static void flush_backlog(struct work_struct *work) 4630 { 4631 struct sk_buff *skb, *tmp; 4632 struct softnet_data *sd; 4633 4634 local_bh_disable(); 4635 sd = this_cpu_ptr(&softnet_data); 4636 4637 local_irq_disable(); 4638 rps_lock(sd); 4639 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 4640 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 4641 __skb_unlink(skb, &sd->input_pkt_queue); 4642 kfree_skb(skb); 4643 input_queue_head_incr(sd); 4644 } 4645 } 4646 rps_unlock(sd); 4647 local_irq_enable(); 4648 4649 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 4650 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 4651 __skb_unlink(skb, &sd->process_queue); 4652 kfree_skb(skb); 4653 input_queue_head_incr(sd); 4654 } 4655 } 4656 local_bh_enable(); 4657 } 4658 4659 static void flush_all_backlogs(void) 4660 { 4661 unsigned int cpu; 4662 4663 get_online_cpus(); 4664 4665 for_each_online_cpu(cpu) 4666 queue_work_on(cpu, system_highpri_wq, 4667 per_cpu_ptr(&flush_works, cpu)); 4668 4669 for_each_online_cpu(cpu) 4670 flush_work(per_cpu_ptr(&flush_works, cpu)); 4671 4672 put_online_cpus(); 4673 } 4674 4675 static int napi_gro_complete(struct sk_buff *skb) 4676 { 4677 struct packet_offload *ptype; 4678 __be16 type = skb->protocol; 4679 struct list_head *head = &offload_base; 4680 int err = -ENOENT; 4681 4682 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 4683 4684 if (NAPI_GRO_CB(skb)->count == 1) { 4685 skb_shinfo(skb)->gso_size = 0; 4686 goto out; 4687 } 4688 4689 rcu_read_lock(); 4690 list_for_each_entry_rcu(ptype, head, list) { 4691 if (ptype->type != type || !ptype->callbacks.gro_complete) 4692 continue; 4693 4694 err = ptype->callbacks.gro_complete(skb, 0); 4695 break; 4696 } 4697 rcu_read_unlock(); 4698 4699 if (err) { 4700 WARN_ON(&ptype->list == head); 4701 kfree_skb(skb); 4702 return NET_RX_SUCCESS; 4703 } 4704 4705 out: 4706 return netif_receive_skb_internal(skb); 4707 } 4708 4709 /* napi->gro_list contains packets ordered by age. 4710 * youngest packets at the head of it. 4711 * Complete skbs in reverse order to reduce latencies. 4712 */ 4713 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 4714 { 4715 struct sk_buff *skb, *prev = NULL; 4716 4717 /* scan list and build reverse chain */ 4718 for (skb = napi->gro_list; skb != NULL; skb = skb->next) { 4719 skb->prev = prev; 4720 prev = skb; 4721 } 4722 4723 for (skb = prev; skb; skb = prev) { 4724 skb->next = NULL; 4725 4726 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 4727 return; 4728 4729 prev = skb->prev; 4730 napi_gro_complete(skb); 4731 napi->gro_count--; 4732 } 4733 4734 napi->gro_list = NULL; 4735 } 4736 EXPORT_SYMBOL(napi_gro_flush); 4737 4738 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) 4739 { 4740 struct sk_buff *p; 4741 unsigned int maclen = skb->dev->hard_header_len; 4742 u32 hash = skb_get_hash_raw(skb); 4743 4744 for (p = napi->gro_list; p; p = p->next) { 4745 unsigned long diffs; 4746 4747 NAPI_GRO_CB(p)->flush = 0; 4748 4749 if (hash != skb_get_hash_raw(p)) { 4750 NAPI_GRO_CB(p)->same_flow = 0; 4751 continue; 4752 } 4753 4754 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 4755 diffs |= p->vlan_tci ^ skb->vlan_tci; 4756 diffs |= skb_metadata_dst_cmp(p, skb); 4757 diffs |= skb_metadata_differs(p, skb); 4758 if (maclen == ETH_HLEN) 4759 diffs |= compare_ether_header(skb_mac_header(p), 4760 skb_mac_header(skb)); 4761 else if (!diffs) 4762 diffs = memcmp(skb_mac_header(p), 4763 skb_mac_header(skb), 4764 maclen); 4765 NAPI_GRO_CB(p)->same_flow = !diffs; 4766 } 4767 } 4768 4769 static void skb_gro_reset_offset(struct sk_buff *skb) 4770 { 4771 const struct skb_shared_info *pinfo = skb_shinfo(skb); 4772 const skb_frag_t *frag0 = &pinfo->frags[0]; 4773 4774 NAPI_GRO_CB(skb)->data_offset = 0; 4775 NAPI_GRO_CB(skb)->frag0 = NULL; 4776 NAPI_GRO_CB(skb)->frag0_len = 0; 4777 4778 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 4779 pinfo->nr_frags && 4780 !PageHighMem(skb_frag_page(frag0))) { 4781 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 4782 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 4783 skb_frag_size(frag0), 4784 skb->end - skb->tail); 4785 } 4786 } 4787 4788 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 4789 { 4790 struct skb_shared_info *pinfo = skb_shinfo(skb); 4791 4792 BUG_ON(skb->end - skb->tail < grow); 4793 4794 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 4795 4796 skb->data_len -= grow; 4797 skb->tail += grow; 4798 4799 pinfo->frags[0].page_offset += grow; 4800 skb_frag_size_sub(&pinfo->frags[0], grow); 4801 4802 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 4803 skb_frag_unref(skb, 0); 4804 memmove(pinfo->frags, pinfo->frags + 1, 4805 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 4806 } 4807 } 4808 4809 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4810 { 4811 struct sk_buff **pp = NULL; 4812 struct packet_offload *ptype; 4813 __be16 type = skb->protocol; 4814 struct list_head *head = &offload_base; 4815 int same_flow; 4816 enum gro_result ret; 4817 int grow; 4818 4819 if (netif_elide_gro(skb->dev)) 4820 goto normal; 4821 4822 gro_list_prepare(napi, skb); 4823 4824 rcu_read_lock(); 4825 list_for_each_entry_rcu(ptype, head, list) { 4826 if (ptype->type != type || !ptype->callbacks.gro_receive) 4827 continue; 4828 4829 skb_set_network_header(skb, skb_gro_offset(skb)); 4830 skb_reset_mac_len(skb); 4831 NAPI_GRO_CB(skb)->same_flow = 0; 4832 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 4833 NAPI_GRO_CB(skb)->free = 0; 4834 NAPI_GRO_CB(skb)->encap_mark = 0; 4835 NAPI_GRO_CB(skb)->recursion_counter = 0; 4836 NAPI_GRO_CB(skb)->is_fou = 0; 4837 NAPI_GRO_CB(skb)->is_atomic = 1; 4838 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 4839 4840 /* Setup for GRO checksum validation */ 4841 switch (skb->ip_summed) { 4842 case CHECKSUM_COMPLETE: 4843 NAPI_GRO_CB(skb)->csum = skb->csum; 4844 NAPI_GRO_CB(skb)->csum_valid = 1; 4845 NAPI_GRO_CB(skb)->csum_cnt = 0; 4846 break; 4847 case CHECKSUM_UNNECESSARY: 4848 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 4849 NAPI_GRO_CB(skb)->csum_valid = 0; 4850 break; 4851 default: 4852 NAPI_GRO_CB(skb)->csum_cnt = 0; 4853 NAPI_GRO_CB(skb)->csum_valid = 0; 4854 } 4855 4856 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); 4857 break; 4858 } 4859 rcu_read_unlock(); 4860 4861 if (&ptype->list == head) 4862 goto normal; 4863 4864 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { 4865 ret = GRO_CONSUMED; 4866 goto ok; 4867 } 4868 4869 same_flow = NAPI_GRO_CB(skb)->same_flow; 4870 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 4871 4872 if (pp) { 4873 struct sk_buff *nskb = *pp; 4874 4875 *pp = nskb->next; 4876 nskb->next = NULL; 4877 napi_gro_complete(nskb); 4878 napi->gro_count--; 4879 } 4880 4881 if (same_flow) 4882 goto ok; 4883 4884 if (NAPI_GRO_CB(skb)->flush) 4885 goto normal; 4886 4887 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { 4888 struct sk_buff *nskb = napi->gro_list; 4889 4890 /* locate the end of the list to select the 'oldest' flow */ 4891 while (nskb->next) { 4892 pp = &nskb->next; 4893 nskb = *pp; 4894 } 4895 *pp = NULL; 4896 nskb->next = NULL; 4897 napi_gro_complete(nskb); 4898 } else { 4899 napi->gro_count++; 4900 } 4901 NAPI_GRO_CB(skb)->count = 1; 4902 NAPI_GRO_CB(skb)->age = jiffies; 4903 NAPI_GRO_CB(skb)->last = skb; 4904 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 4905 skb->next = napi->gro_list; 4906 napi->gro_list = skb; 4907 ret = GRO_HELD; 4908 4909 pull: 4910 grow = skb_gro_offset(skb) - skb_headlen(skb); 4911 if (grow > 0) 4912 gro_pull_from_frag0(skb, grow); 4913 ok: 4914 return ret; 4915 4916 normal: 4917 ret = GRO_NORMAL; 4918 goto pull; 4919 } 4920 4921 struct packet_offload *gro_find_receive_by_type(__be16 type) 4922 { 4923 struct list_head *offload_head = &offload_base; 4924 struct packet_offload *ptype; 4925 4926 list_for_each_entry_rcu(ptype, offload_head, list) { 4927 if (ptype->type != type || !ptype->callbacks.gro_receive) 4928 continue; 4929 return ptype; 4930 } 4931 return NULL; 4932 } 4933 EXPORT_SYMBOL(gro_find_receive_by_type); 4934 4935 struct packet_offload *gro_find_complete_by_type(__be16 type) 4936 { 4937 struct list_head *offload_head = &offload_base; 4938 struct packet_offload *ptype; 4939 4940 list_for_each_entry_rcu(ptype, offload_head, list) { 4941 if (ptype->type != type || !ptype->callbacks.gro_complete) 4942 continue; 4943 return ptype; 4944 } 4945 return NULL; 4946 } 4947 EXPORT_SYMBOL(gro_find_complete_by_type); 4948 4949 static void napi_skb_free_stolen_head(struct sk_buff *skb) 4950 { 4951 skb_dst_drop(skb); 4952 secpath_reset(skb); 4953 kmem_cache_free(skbuff_head_cache, skb); 4954 } 4955 4956 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 4957 { 4958 switch (ret) { 4959 case GRO_NORMAL: 4960 if (netif_receive_skb_internal(skb)) 4961 ret = GRO_DROP; 4962 break; 4963 4964 case GRO_DROP: 4965 kfree_skb(skb); 4966 break; 4967 4968 case GRO_MERGED_FREE: 4969 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 4970 napi_skb_free_stolen_head(skb); 4971 else 4972 __kfree_skb(skb); 4973 break; 4974 4975 case GRO_HELD: 4976 case GRO_MERGED: 4977 case GRO_CONSUMED: 4978 break; 4979 } 4980 4981 return ret; 4982 } 4983 4984 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 4985 { 4986 skb_mark_napi_id(skb, napi); 4987 trace_napi_gro_receive_entry(skb); 4988 4989 skb_gro_reset_offset(skb); 4990 4991 return napi_skb_finish(dev_gro_receive(napi, skb), skb); 4992 } 4993 EXPORT_SYMBOL(napi_gro_receive); 4994 4995 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 4996 { 4997 if (unlikely(skb->pfmemalloc)) { 4998 consume_skb(skb); 4999 return; 5000 } 5001 __skb_pull(skb, skb_headlen(skb)); 5002 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 5003 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 5004 skb->vlan_tci = 0; 5005 skb->dev = napi->dev; 5006 skb->skb_iif = 0; 5007 skb->encapsulation = 0; 5008 skb_shinfo(skb)->gso_type = 0; 5009 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5010 secpath_reset(skb); 5011 5012 napi->skb = skb; 5013 } 5014 5015 struct sk_buff *napi_get_frags(struct napi_struct *napi) 5016 { 5017 struct sk_buff *skb = napi->skb; 5018 5019 if (!skb) { 5020 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 5021 if (skb) { 5022 napi->skb = skb; 5023 skb_mark_napi_id(skb, napi); 5024 } 5025 } 5026 return skb; 5027 } 5028 EXPORT_SYMBOL(napi_get_frags); 5029 5030 static gro_result_t napi_frags_finish(struct napi_struct *napi, 5031 struct sk_buff *skb, 5032 gro_result_t ret) 5033 { 5034 switch (ret) { 5035 case GRO_NORMAL: 5036 case GRO_HELD: 5037 __skb_push(skb, ETH_HLEN); 5038 skb->protocol = eth_type_trans(skb, skb->dev); 5039 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 5040 ret = GRO_DROP; 5041 break; 5042 5043 case GRO_DROP: 5044 napi_reuse_skb(napi, skb); 5045 break; 5046 5047 case GRO_MERGED_FREE: 5048 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 5049 napi_skb_free_stolen_head(skb); 5050 else 5051 napi_reuse_skb(napi, skb); 5052 break; 5053 5054 case GRO_MERGED: 5055 case GRO_CONSUMED: 5056 break; 5057 } 5058 5059 return ret; 5060 } 5061 5062 /* Upper GRO stack assumes network header starts at gro_offset=0 5063 * Drivers could call both napi_gro_frags() and napi_gro_receive() 5064 * We copy ethernet header into skb->data to have a common layout. 5065 */ 5066 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 5067 { 5068 struct sk_buff *skb = napi->skb; 5069 const struct ethhdr *eth; 5070 unsigned int hlen = sizeof(*eth); 5071 5072 napi->skb = NULL; 5073 5074 skb_reset_mac_header(skb); 5075 skb_gro_reset_offset(skb); 5076 5077 eth = skb_gro_header_fast(skb, 0); 5078 if (unlikely(skb_gro_header_hard(skb, hlen))) { 5079 eth = skb_gro_header_slow(skb, hlen, 0); 5080 if (unlikely(!eth)) { 5081 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 5082 __func__, napi->dev->name); 5083 napi_reuse_skb(napi, skb); 5084 return NULL; 5085 } 5086 } else { 5087 gro_pull_from_frag0(skb, hlen); 5088 NAPI_GRO_CB(skb)->frag0 += hlen; 5089 NAPI_GRO_CB(skb)->frag0_len -= hlen; 5090 } 5091 __skb_pull(skb, hlen); 5092 5093 /* 5094 * This works because the only protocols we care about don't require 5095 * special handling. 5096 * We'll fix it up properly in napi_frags_finish() 5097 */ 5098 skb->protocol = eth->h_proto; 5099 5100 return skb; 5101 } 5102 5103 gro_result_t napi_gro_frags(struct napi_struct *napi) 5104 { 5105 struct sk_buff *skb = napi_frags_skb(napi); 5106 5107 if (!skb) 5108 return GRO_DROP; 5109 5110 trace_napi_gro_frags_entry(skb); 5111 5112 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 5113 } 5114 EXPORT_SYMBOL(napi_gro_frags); 5115 5116 /* Compute the checksum from gro_offset and return the folded value 5117 * after adding in any pseudo checksum. 5118 */ 5119 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 5120 { 5121 __wsum wsum; 5122 __sum16 sum; 5123 5124 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 5125 5126 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 5127 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 5128 if (likely(!sum)) { 5129 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 5130 !skb->csum_complete_sw) 5131 netdev_rx_csum_fault(skb->dev); 5132 } 5133 5134 NAPI_GRO_CB(skb)->csum = wsum; 5135 NAPI_GRO_CB(skb)->csum_valid = 1; 5136 5137 return sum; 5138 } 5139 EXPORT_SYMBOL(__skb_gro_checksum_complete); 5140 5141 static void net_rps_send_ipi(struct softnet_data *remsd) 5142 { 5143 #ifdef CONFIG_RPS 5144 while (remsd) { 5145 struct softnet_data *next = remsd->rps_ipi_next; 5146 5147 if (cpu_online(remsd->cpu)) 5148 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5149 remsd = next; 5150 } 5151 #endif 5152 } 5153 5154 /* 5155 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5156 * Note: called with local irq disabled, but exits with local irq enabled. 5157 */ 5158 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5159 { 5160 #ifdef CONFIG_RPS 5161 struct softnet_data *remsd = sd->rps_ipi_list; 5162 5163 if (remsd) { 5164 sd->rps_ipi_list = NULL; 5165 5166 local_irq_enable(); 5167 5168 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5169 net_rps_send_ipi(remsd); 5170 } else 5171 #endif 5172 local_irq_enable(); 5173 } 5174 5175 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5176 { 5177 #ifdef CONFIG_RPS 5178 return sd->rps_ipi_list != NULL; 5179 #else 5180 return false; 5181 #endif 5182 } 5183 5184 static int process_backlog(struct napi_struct *napi, int quota) 5185 { 5186 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5187 bool again = true; 5188 int work = 0; 5189 5190 /* Check if we have pending ipi, its better to send them now, 5191 * not waiting net_rx_action() end. 5192 */ 5193 if (sd_has_rps_ipi_waiting(sd)) { 5194 local_irq_disable(); 5195 net_rps_action_and_irq_enable(sd); 5196 } 5197 5198 napi->weight = dev_rx_weight; 5199 while (again) { 5200 struct sk_buff *skb; 5201 5202 while ((skb = __skb_dequeue(&sd->process_queue))) { 5203 rcu_read_lock(); 5204 __netif_receive_skb(skb); 5205 rcu_read_unlock(); 5206 input_queue_head_incr(sd); 5207 if (++work >= quota) 5208 return work; 5209 5210 } 5211 5212 local_irq_disable(); 5213 rps_lock(sd); 5214 if (skb_queue_empty(&sd->input_pkt_queue)) { 5215 /* 5216 * Inline a custom version of __napi_complete(). 5217 * only current cpu owns and manipulates this napi, 5218 * and NAPI_STATE_SCHED is the only possible flag set 5219 * on backlog. 5220 * We can use a plain write instead of clear_bit(), 5221 * and we dont need an smp_mb() memory barrier. 5222 */ 5223 napi->state = 0; 5224 again = false; 5225 } else { 5226 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5227 &sd->process_queue); 5228 } 5229 rps_unlock(sd); 5230 local_irq_enable(); 5231 } 5232 5233 return work; 5234 } 5235 5236 /** 5237 * __napi_schedule - schedule for receive 5238 * @n: entry to schedule 5239 * 5240 * The entry's receive function will be scheduled to run. 5241 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5242 */ 5243 void __napi_schedule(struct napi_struct *n) 5244 { 5245 unsigned long flags; 5246 5247 local_irq_save(flags); 5248 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5249 local_irq_restore(flags); 5250 } 5251 EXPORT_SYMBOL(__napi_schedule); 5252 5253 /** 5254 * napi_schedule_prep - check if napi can be scheduled 5255 * @n: napi context 5256 * 5257 * Test if NAPI routine is already running, and if not mark 5258 * it as running. This is used as a condition variable 5259 * insure only one NAPI poll instance runs. We also make 5260 * sure there is no pending NAPI disable. 5261 */ 5262 bool napi_schedule_prep(struct napi_struct *n) 5263 { 5264 unsigned long val, new; 5265 5266 do { 5267 val = READ_ONCE(n->state); 5268 if (unlikely(val & NAPIF_STATE_DISABLE)) 5269 return false; 5270 new = val | NAPIF_STATE_SCHED; 5271 5272 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5273 * This was suggested by Alexander Duyck, as compiler 5274 * emits better code than : 5275 * if (val & NAPIF_STATE_SCHED) 5276 * new |= NAPIF_STATE_MISSED; 5277 */ 5278 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5279 NAPIF_STATE_MISSED; 5280 } while (cmpxchg(&n->state, val, new) != val); 5281 5282 return !(val & NAPIF_STATE_SCHED); 5283 } 5284 EXPORT_SYMBOL(napi_schedule_prep); 5285 5286 /** 5287 * __napi_schedule_irqoff - schedule for receive 5288 * @n: entry to schedule 5289 * 5290 * Variant of __napi_schedule() assuming hard irqs are masked 5291 */ 5292 void __napi_schedule_irqoff(struct napi_struct *n) 5293 { 5294 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5295 } 5296 EXPORT_SYMBOL(__napi_schedule_irqoff); 5297 5298 bool napi_complete_done(struct napi_struct *n, int work_done) 5299 { 5300 unsigned long flags, val, new; 5301 5302 /* 5303 * 1) Don't let napi dequeue from the cpu poll list 5304 * just in case its running on a different cpu. 5305 * 2) If we are busy polling, do nothing here, we have 5306 * the guarantee we will be called later. 5307 */ 5308 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5309 NAPIF_STATE_IN_BUSY_POLL))) 5310 return false; 5311 5312 if (n->gro_list) { 5313 unsigned long timeout = 0; 5314 5315 if (work_done) 5316 timeout = n->dev->gro_flush_timeout; 5317 5318 if (timeout) 5319 hrtimer_start(&n->timer, ns_to_ktime(timeout), 5320 HRTIMER_MODE_REL_PINNED); 5321 else 5322 napi_gro_flush(n, false); 5323 } 5324 if (unlikely(!list_empty(&n->poll_list))) { 5325 /* If n->poll_list is not empty, we need to mask irqs */ 5326 local_irq_save(flags); 5327 list_del_init(&n->poll_list); 5328 local_irq_restore(flags); 5329 } 5330 5331 do { 5332 val = READ_ONCE(n->state); 5333 5334 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 5335 5336 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); 5337 5338 /* If STATE_MISSED was set, leave STATE_SCHED set, 5339 * because we will call napi->poll() one more time. 5340 * This C code was suggested by Alexander Duyck to help gcc. 5341 */ 5342 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 5343 NAPIF_STATE_SCHED; 5344 } while (cmpxchg(&n->state, val, new) != val); 5345 5346 if (unlikely(val & NAPIF_STATE_MISSED)) { 5347 __napi_schedule(n); 5348 return false; 5349 } 5350 5351 return true; 5352 } 5353 EXPORT_SYMBOL(napi_complete_done); 5354 5355 /* must be called under rcu_read_lock(), as we dont take a reference */ 5356 static struct napi_struct *napi_by_id(unsigned int napi_id) 5357 { 5358 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 5359 struct napi_struct *napi; 5360 5361 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 5362 if (napi->napi_id == napi_id) 5363 return napi; 5364 5365 return NULL; 5366 } 5367 5368 #if defined(CONFIG_NET_RX_BUSY_POLL) 5369 5370 #define BUSY_POLL_BUDGET 8 5371 5372 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) 5373 { 5374 int rc; 5375 5376 /* Busy polling means there is a high chance device driver hard irq 5377 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 5378 * set in napi_schedule_prep(). 5379 * Since we are about to call napi->poll() once more, we can safely 5380 * clear NAPI_STATE_MISSED. 5381 * 5382 * Note: x86 could use a single "lock and ..." instruction 5383 * to perform these two clear_bit() 5384 */ 5385 clear_bit(NAPI_STATE_MISSED, &napi->state); 5386 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 5387 5388 local_bh_disable(); 5389 5390 /* All we really want here is to re-enable device interrupts. 5391 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 5392 */ 5393 rc = napi->poll(napi, BUSY_POLL_BUDGET); 5394 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); 5395 netpoll_poll_unlock(have_poll_lock); 5396 if (rc == BUSY_POLL_BUDGET) 5397 __napi_schedule(napi); 5398 local_bh_enable(); 5399 } 5400 5401 void napi_busy_loop(unsigned int napi_id, 5402 bool (*loop_end)(void *, unsigned long), 5403 void *loop_end_arg) 5404 { 5405 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 5406 int (*napi_poll)(struct napi_struct *napi, int budget); 5407 void *have_poll_lock = NULL; 5408 struct napi_struct *napi; 5409 5410 restart: 5411 napi_poll = NULL; 5412 5413 rcu_read_lock(); 5414 5415 napi = napi_by_id(napi_id); 5416 if (!napi) 5417 goto out; 5418 5419 preempt_disable(); 5420 for (;;) { 5421 int work = 0; 5422 5423 local_bh_disable(); 5424 if (!napi_poll) { 5425 unsigned long val = READ_ONCE(napi->state); 5426 5427 /* If multiple threads are competing for this napi, 5428 * we avoid dirtying napi->state as much as we can. 5429 */ 5430 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 5431 NAPIF_STATE_IN_BUSY_POLL)) 5432 goto count; 5433 if (cmpxchg(&napi->state, val, 5434 val | NAPIF_STATE_IN_BUSY_POLL | 5435 NAPIF_STATE_SCHED) != val) 5436 goto count; 5437 have_poll_lock = netpoll_poll_lock(napi); 5438 napi_poll = napi->poll; 5439 } 5440 work = napi_poll(napi, BUSY_POLL_BUDGET); 5441 trace_napi_poll(napi, work, BUSY_POLL_BUDGET); 5442 count: 5443 if (work > 0) 5444 __NET_ADD_STATS(dev_net(napi->dev), 5445 LINUX_MIB_BUSYPOLLRXPACKETS, work); 5446 local_bh_enable(); 5447 5448 if (!loop_end || loop_end(loop_end_arg, start_time)) 5449 break; 5450 5451 if (unlikely(need_resched())) { 5452 if (napi_poll) 5453 busy_poll_stop(napi, have_poll_lock); 5454 preempt_enable(); 5455 rcu_read_unlock(); 5456 cond_resched(); 5457 if (loop_end(loop_end_arg, start_time)) 5458 return; 5459 goto restart; 5460 } 5461 cpu_relax(); 5462 } 5463 if (napi_poll) 5464 busy_poll_stop(napi, have_poll_lock); 5465 preempt_enable(); 5466 out: 5467 rcu_read_unlock(); 5468 } 5469 EXPORT_SYMBOL(napi_busy_loop); 5470 5471 #endif /* CONFIG_NET_RX_BUSY_POLL */ 5472 5473 static void napi_hash_add(struct napi_struct *napi) 5474 { 5475 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || 5476 test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) 5477 return; 5478 5479 spin_lock(&napi_hash_lock); 5480 5481 /* 0..NR_CPUS range is reserved for sender_cpu use */ 5482 do { 5483 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 5484 napi_gen_id = MIN_NAPI_ID; 5485 } while (napi_by_id(napi_gen_id)); 5486 napi->napi_id = napi_gen_id; 5487 5488 hlist_add_head_rcu(&napi->napi_hash_node, 5489 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 5490 5491 spin_unlock(&napi_hash_lock); 5492 } 5493 5494 /* Warning : caller is responsible to make sure rcu grace period 5495 * is respected before freeing memory containing @napi 5496 */ 5497 bool napi_hash_del(struct napi_struct *napi) 5498 { 5499 bool rcu_sync_needed = false; 5500 5501 spin_lock(&napi_hash_lock); 5502 5503 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { 5504 rcu_sync_needed = true; 5505 hlist_del_rcu(&napi->napi_hash_node); 5506 } 5507 spin_unlock(&napi_hash_lock); 5508 return rcu_sync_needed; 5509 } 5510 EXPORT_SYMBOL_GPL(napi_hash_del); 5511 5512 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 5513 { 5514 struct napi_struct *napi; 5515 5516 napi = container_of(timer, struct napi_struct, timer); 5517 5518 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 5519 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 5520 */ 5521 if (napi->gro_list && !napi_disable_pending(napi) && 5522 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) 5523 __napi_schedule_irqoff(napi); 5524 5525 return HRTIMER_NORESTART; 5526 } 5527 5528 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 5529 int (*poll)(struct napi_struct *, int), int weight) 5530 { 5531 INIT_LIST_HEAD(&napi->poll_list); 5532 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 5533 napi->timer.function = napi_watchdog; 5534 napi->gro_count = 0; 5535 napi->gro_list = NULL; 5536 napi->skb = NULL; 5537 napi->poll = poll; 5538 if (weight > NAPI_POLL_WEIGHT) 5539 pr_err_once("netif_napi_add() called with weight %d on device %s\n", 5540 weight, dev->name); 5541 napi->weight = weight; 5542 list_add(&napi->dev_list, &dev->napi_list); 5543 napi->dev = dev; 5544 #ifdef CONFIG_NETPOLL 5545 napi->poll_owner = -1; 5546 #endif 5547 set_bit(NAPI_STATE_SCHED, &napi->state); 5548 napi_hash_add(napi); 5549 } 5550 EXPORT_SYMBOL(netif_napi_add); 5551 5552 void napi_disable(struct napi_struct *n) 5553 { 5554 might_sleep(); 5555 set_bit(NAPI_STATE_DISABLE, &n->state); 5556 5557 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 5558 msleep(1); 5559 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 5560 msleep(1); 5561 5562 hrtimer_cancel(&n->timer); 5563 5564 clear_bit(NAPI_STATE_DISABLE, &n->state); 5565 } 5566 EXPORT_SYMBOL(napi_disable); 5567 5568 /* Must be called in process context */ 5569 void netif_napi_del(struct napi_struct *napi) 5570 { 5571 might_sleep(); 5572 if (napi_hash_del(napi)) 5573 synchronize_net(); 5574 list_del_init(&napi->dev_list); 5575 napi_free_frags(napi); 5576 5577 kfree_skb_list(napi->gro_list); 5578 napi->gro_list = NULL; 5579 napi->gro_count = 0; 5580 } 5581 EXPORT_SYMBOL(netif_napi_del); 5582 5583 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 5584 { 5585 void *have; 5586 int work, weight; 5587 5588 list_del_init(&n->poll_list); 5589 5590 have = netpoll_poll_lock(n); 5591 5592 weight = n->weight; 5593 5594 /* This NAPI_STATE_SCHED test is for avoiding a race 5595 * with netpoll's poll_napi(). Only the entity which 5596 * obtains the lock and sees NAPI_STATE_SCHED set will 5597 * actually make the ->poll() call. Therefore we avoid 5598 * accidentally calling ->poll() when NAPI is not scheduled. 5599 */ 5600 work = 0; 5601 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 5602 work = n->poll(n, weight); 5603 trace_napi_poll(n, work, weight); 5604 } 5605 5606 WARN_ON_ONCE(work > weight); 5607 5608 if (likely(work < weight)) 5609 goto out_unlock; 5610 5611 /* Drivers must not modify the NAPI state if they 5612 * consume the entire weight. In such cases this code 5613 * still "owns" the NAPI instance and therefore can 5614 * move the instance around on the list at-will. 5615 */ 5616 if (unlikely(napi_disable_pending(n))) { 5617 napi_complete(n); 5618 goto out_unlock; 5619 } 5620 5621 if (n->gro_list) { 5622 /* flush too old packets 5623 * If HZ < 1000, flush all packets. 5624 */ 5625 napi_gro_flush(n, HZ >= 1000); 5626 } 5627 5628 /* Some drivers may have called napi_schedule 5629 * prior to exhausting their budget. 5630 */ 5631 if (unlikely(!list_empty(&n->poll_list))) { 5632 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 5633 n->dev ? n->dev->name : "backlog"); 5634 goto out_unlock; 5635 } 5636 5637 list_add_tail(&n->poll_list, repoll); 5638 5639 out_unlock: 5640 netpoll_poll_unlock(have); 5641 5642 return work; 5643 } 5644 5645 static __latent_entropy void net_rx_action(struct softirq_action *h) 5646 { 5647 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5648 unsigned long time_limit = jiffies + 5649 usecs_to_jiffies(netdev_budget_usecs); 5650 int budget = netdev_budget; 5651 LIST_HEAD(list); 5652 LIST_HEAD(repoll); 5653 5654 local_irq_disable(); 5655 list_splice_init(&sd->poll_list, &list); 5656 local_irq_enable(); 5657 5658 for (;;) { 5659 struct napi_struct *n; 5660 5661 if (list_empty(&list)) { 5662 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 5663 goto out; 5664 break; 5665 } 5666 5667 n = list_first_entry(&list, struct napi_struct, poll_list); 5668 budget -= napi_poll(n, &repoll); 5669 5670 /* If softirq window is exhausted then punt. 5671 * Allow this to run for 2 jiffies since which will allow 5672 * an average latency of 1.5/HZ. 5673 */ 5674 if (unlikely(budget <= 0 || 5675 time_after_eq(jiffies, time_limit))) { 5676 sd->time_squeeze++; 5677 break; 5678 } 5679 } 5680 5681 local_irq_disable(); 5682 5683 list_splice_tail_init(&sd->poll_list, &list); 5684 list_splice_tail(&repoll, &list); 5685 list_splice(&list, &sd->poll_list); 5686 if (!list_empty(&sd->poll_list)) 5687 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 5688 5689 net_rps_action_and_irq_enable(sd); 5690 out: 5691 __kfree_skb_flush(); 5692 } 5693 5694 struct netdev_adjacent { 5695 struct net_device *dev; 5696 5697 /* upper master flag, there can only be one master device per list */ 5698 bool master; 5699 5700 /* counter for the number of times this device was added to us */ 5701 u16 ref_nr; 5702 5703 /* private field for the users */ 5704 void *private; 5705 5706 struct list_head list; 5707 struct rcu_head rcu; 5708 }; 5709 5710 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 5711 struct list_head *adj_list) 5712 { 5713 struct netdev_adjacent *adj; 5714 5715 list_for_each_entry(adj, adj_list, list) { 5716 if (adj->dev == adj_dev) 5717 return adj; 5718 } 5719 return NULL; 5720 } 5721 5722 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) 5723 { 5724 struct net_device *dev = data; 5725 5726 return upper_dev == dev; 5727 } 5728 5729 /** 5730 * netdev_has_upper_dev - Check if device is linked to an upper device 5731 * @dev: device 5732 * @upper_dev: upper device to check 5733 * 5734 * Find out if a device is linked to specified upper device and return true 5735 * in case it is. Note that this checks only immediate upper device, 5736 * not through a complete stack of devices. The caller must hold the RTNL lock. 5737 */ 5738 bool netdev_has_upper_dev(struct net_device *dev, 5739 struct net_device *upper_dev) 5740 { 5741 ASSERT_RTNL(); 5742 5743 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 5744 upper_dev); 5745 } 5746 EXPORT_SYMBOL(netdev_has_upper_dev); 5747 5748 /** 5749 * netdev_has_upper_dev_all - Check if device is linked to an upper device 5750 * @dev: device 5751 * @upper_dev: upper device to check 5752 * 5753 * Find out if a device is linked to specified upper device and return true 5754 * in case it is. Note that this checks the entire upper device chain. 5755 * The caller must hold rcu lock. 5756 */ 5757 5758 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 5759 struct net_device *upper_dev) 5760 { 5761 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 5762 upper_dev); 5763 } 5764 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 5765 5766 /** 5767 * netdev_has_any_upper_dev - Check if device is linked to some device 5768 * @dev: device 5769 * 5770 * Find out if a device is linked to an upper device and return true in case 5771 * it is. The caller must hold the RTNL lock. 5772 */ 5773 bool netdev_has_any_upper_dev(struct net_device *dev) 5774 { 5775 ASSERT_RTNL(); 5776 5777 return !list_empty(&dev->adj_list.upper); 5778 } 5779 EXPORT_SYMBOL(netdev_has_any_upper_dev); 5780 5781 /** 5782 * netdev_master_upper_dev_get - Get master upper device 5783 * @dev: device 5784 * 5785 * Find a master upper device and return pointer to it or NULL in case 5786 * it's not there. The caller must hold the RTNL lock. 5787 */ 5788 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 5789 { 5790 struct netdev_adjacent *upper; 5791 5792 ASSERT_RTNL(); 5793 5794 if (list_empty(&dev->adj_list.upper)) 5795 return NULL; 5796 5797 upper = list_first_entry(&dev->adj_list.upper, 5798 struct netdev_adjacent, list); 5799 if (likely(upper->master)) 5800 return upper->dev; 5801 return NULL; 5802 } 5803 EXPORT_SYMBOL(netdev_master_upper_dev_get); 5804 5805 /** 5806 * netdev_has_any_lower_dev - Check if device is linked to some device 5807 * @dev: device 5808 * 5809 * Find out if a device is linked to a lower device and return true in case 5810 * it is. The caller must hold the RTNL lock. 5811 */ 5812 static bool netdev_has_any_lower_dev(struct net_device *dev) 5813 { 5814 ASSERT_RTNL(); 5815 5816 return !list_empty(&dev->adj_list.lower); 5817 } 5818 5819 void *netdev_adjacent_get_private(struct list_head *adj_list) 5820 { 5821 struct netdev_adjacent *adj; 5822 5823 adj = list_entry(adj_list, struct netdev_adjacent, list); 5824 5825 return adj->private; 5826 } 5827 EXPORT_SYMBOL(netdev_adjacent_get_private); 5828 5829 /** 5830 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 5831 * @dev: device 5832 * @iter: list_head ** of the current position 5833 * 5834 * Gets the next device from the dev's upper list, starting from iter 5835 * position. The caller must hold RCU read lock. 5836 */ 5837 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 5838 struct list_head **iter) 5839 { 5840 struct netdev_adjacent *upper; 5841 5842 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 5843 5844 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5845 5846 if (&upper->list == &dev->adj_list.upper) 5847 return NULL; 5848 5849 *iter = &upper->list; 5850 5851 return upper->dev; 5852 } 5853 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 5854 5855 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 5856 struct list_head **iter) 5857 { 5858 struct netdev_adjacent *upper; 5859 5860 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 5861 5862 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5863 5864 if (&upper->list == &dev->adj_list.upper) 5865 return NULL; 5866 5867 *iter = &upper->list; 5868 5869 return upper->dev; 5870 } 5871 5872 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 5873 int (*fn)(struct net_device *dev, 5874 void *data), 5875 void *data) 5876 { 5877 struct net_device *udev; 5878 struct list_head *iter; 5879 int ret; 5880 5881 for (iter = &dev->adj_list.upper, 5882 udev = netdev_next_upper_dev_rcu(dev, &iter); 5883 udev; 5884 udev = netdev_next_upper_dev_rcu(dev, &iter)) { 5885 /* first is the upper device itself */ 5886 ret = fn(udev, data); 5887 if (ret) 5888 return ret; 5889 5890 /* then look at all of its upper devices */ 5891 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); 5892 if (ret) 5893 return ret; 5894 } 5895 5896 return 0; 5897 } 5898 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 5899 5900 /** 5901 * netdev_lower_get_next_private - Get the next ->private from the 5902 * lower neighbour list 5903 * @dev: device 5904 * @iter: list_head ** of the current position 5905 * 5906 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5907 * list, starting from iter position. The caller must hold either hold the 5908 * RTNL lock or its own locking that guarantees that the neighbour lower 5909 * list will remain unchanged. 5910 */ 5911 void *netdev_lower_get_next_private(struct net_device *dev, 5912 struct list_head **iter) 5913 { 5914 struct netdev_adjacent *lower; 5915 5916 lower = list_entry(*iter, struct netdev_adjacent, list); 5917 5918 if (&lower->list == &dev->adj_list.lower) 5919 return NULL; 5920 5921 *iter = lower->list.next; 5922 5923 return lower->private; 5924 } 5925 EXPORT_SYMBOL(netdev_lower_get_next_private); 5926 5927 /** 5928 * netdev_lower_get_next_private_rcu - Get the next ->private from the 5929 * lower neighbour list, RCU 5930 * variant 5931 * @dev: device 5932 * @iter: list_head ** of the current position 5933 * 5934 * Gets the next netdev_adjacent->private from the dev's lower neighbour 5935 * list, starting from iter position. The caller must hold RCU read lock. 5936 */ 5937 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 5938 struct list_head **iter) 5939 { 5940 struct netdev_adjacent *lower; 5941 5942 WARN_ON_ONCE(!rcu_read_lock_held()); 5943 5944 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 5945 5946 if (&lower->list == &dev->adj_list.lower) 5947 return NULL; 5948 5949 *iter = &lower->list; 5950 5951 return lower->private; 5952 } 5953 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 5954 5955 /** 5956 * netdev_lower_get_next - Get the next device from the lower neighbour 5957 * list 5958 * @dev: device 5959 * @iter: list_head ** of the current position 5960 * 5961 * Gets the next netdev_adjacent from the dev's lower neighbour 5962 * list, starting from iter position. The caller must hold RTNL lock or 5963 * its own locking that guarantees that the neighbour lower 5964 * list will remain unchanged. 5965 */ 5966 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 5967 { 5968 struct netdev_adjacent *lower; 5969 5970 lower = list_entry(*iter, struct netdev_adjacent, list); 5971 5972 if (&lower->list == &dev->adj_list.lower) 5973 return NULL; 5974 5975 *iter = lower->list.next; 5976 5977 return lower->dev; 5978 } 5979 EXPORT_SYMBOL(netdev_lower_get_next); 5980 5981 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 5982 struct list_head **iter) 5983 { 5984 struct netdev_adjacent *lower; 5985 5986 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 5987 5988 if (&lower->list == &dev->adj_list.lower) 5989 return NULL; 5990 5991 *iter = &lower->list; 5992 5993 return lower->dev; 5994 } 5995 5996 int netdev_walk_all_lower_dev(struct net_device *dev, 5997 int (*fn)(struct net_device *dev, 5998 void *data), 5999 void *data) 6000 { 6001 struct net_device *ldev; 6002 struct list_head *iter; 6003 int ret; 6004 6005 for (iter = &dev->adj_list.lower, 6006 ldev = netdev_next_lower_dev(dev, &iter); 6007 ldev; 6008 ldev = netdev_next_lower_dev(dev, &iter)) { 6009 /* first is the lower device itself */ 6010 ret = fn(ldev, data); 6011 if (ret) 6012 return ret; 6013 6014 /* then look at all of its lower devices */ 6015 ret = netdev_walk_all_lower_dev(ldev, fn, data); 6016 if (ret) 6017 return ret; 6018 } 6019 6020 return 0; 6021 } 6022 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 6023 6024 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 6025 struct list_head **iter) 6026 { 6027 struct netdev_adjacent *lower; 6028 6029 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6030 if (&lower->list == &dev->adj_list.lower) 6031 return NULL; 6032 6033 *iter = &lower->list; 6034 6035 return lower->dev; 6036 } 6037 6038 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 6039 int (*fn)(struct net_device *dev, 6040 void *data), 6041 void *data) 6042 { 6043 struct net_device *ldev; 6044 struct list_head *iter; 6045 int ret; 6046 6047 for (iter = &dev->adj_list.lower, 6048 ldev = netdev_next_lower_dev_rcu(dev, &iter); 6049 ldev; 6050 ldev = netdev_next_lower_dev_rcu(dev, &iter)) { 6051 /* first is the lower device itself */ 6052 ret = fn(ldev, data); 6053 if (ret) 6054 return ret; 6055 6056 /* then look at all of its lower devices */ 6057 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); 6058 if (ret) 6059 return ret; 6060 } 6061 6062 return 0; 6063 } 6064 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 6065 6066 /** 6067 * netdev_lower_get_first_private_rcu - Get the first ->private from the 6068 * lower neighbour list, RCU 6069 * variant 6070 * @dev: device 6071 * 6072 * Gets the first netdev_adjacent->private from the dev's lower neighbour 6073 * list. The caller must hold RCU read lock. 6074 */ 6075 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 6076 { 6077 struct netdev_adjacent *lower; 6078 6079 lower = list_first_or_null_rcu(&dev->adj_list.lower, 6080 struct netdev_adjacent, list); 6081 if (lower) 6082 return lower->private; 6083 return NULL; 6084 } 6085 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 6086 6087 /** 6088 * netdev_master_upper_dev_get_rcu - Get master upper device 6089 * @dev: device 6090 * 6091 * Find a master upper device and return pointer to it or NULL in case 6092 * it's not there. The caller must hold the RCU read lock. 6093 */ 6094 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 6095 { 6096 struct netdev_adjacent *upper; 6097 6098 upper = list_first_or_null_rcu(&dev->adj_list.upper, 6099 struct netdev_adjacent, list); 6100 if (upper && likely(upper->master)) 6101 return upper->dev; 6102 return NULL; 6103 } 6104 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 6105 6106 static int netdev_adjacent_sysfs_add(struct net_device *dev, 6107 struct net_device *adj_dev, 6108 struct list_head *dev_list) 6109 { 6110 char linkname[IFNAMSIZ+7]; 6111 6112 sprintf(linkname, dev_list == &dev->adj_list.upper ? 6113 "upper_%s" : "lower_%s", adj_dev->name); 6114 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 6115 linkname); 6116 } 6117 static void netdev_adjacent_sysfs_del(struct net_device *dev, 6118 char *name, 6119 struct list_head *dev_list) 6120 { 6121 char linkname[IFNAMSIZ+7]; 6122 6123 sprintf(linkname, dev_list == &dev->adj_list.upper ? 6124 "upper_%s" : "lower_%s", name); 6125 sysfs_remove_link(&(dev->dev.kobj), linkname); 6126 } 6127 6128 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 6129 struct net_device *adj_dev, 6130 struct list_head *dev_list) 6131 { 6132 return (dev_list == &dev->adj_list.upper || 6133 dev_list == &dev->adj_list.lower) && 6134 net_eq(dev_net(dev), dev_net(adj_dev)); 6135 } 6136 6137 static int __netdev_adjacent_dev_insert(struct net_device *dev, 6138 struct net_device *adj_dev, 6139 struct list_head *dev_list, 6140 void *private, bool master) 6141 { 6142 struct netdev_adjacent *adj; 6143 int ret; 6144 6145 adj = __netdev_find_adj(adj_dev, dev_list); 6146 6147 if (adj) { 6148 adj->ref_nr += 1; 6149 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 6150 dev->name, adj_dev->name, adj->ref_nr); 6151 6152 return 0; 6153 } 6154 6155 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 6156 if (!adj) 6157 return -ENOMEM; 6158 6159 adj->dev = adj_dev; 6160 adj->master = master; 6161 adj->ref_nr = 1; 6162 adj->private = private; 6163 dev_hold(adj_dev); 6164 6165 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 6166 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 6167 6168 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 6169 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 6170 if (ret) 6171 goto free_adj; 6172 } 6173 6174 /* Ensure that master link is always the first item in list. */ 6175 if (master) { 6176 ret = sysfs_create_link(&(dev->dev.kobj), 6177 &(adj_dev->dev.kobj), "master"); 6178 if (ret) 6179 goto remove_symlinks; 6180 6181 list_add_rcu(&adj->list, dev_list); 6182 } else { 6183 list_add_tail_rcu(&adj->list, dev_list); 6184 } 6185 6186 return 0; 6187 6188 remove_symlinks: 6189 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 6190 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 6191 free_adj: 6192 kfree(adj); 6193 dev_put(adj_dev); 6194 6195 return ret; 6196 } 6197 6198 static void __netdev_adjacent_dev_remove(struct net_device *dev, 6199 struct net_device *adj_dev, 6200 u16 ref_nr, 6201 struct list_head *dev_list) 6202 { 6203 struct netdev_adjacent *adj; 6204 6205 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 6206 dev->name, adj_dev->name, ref_nr); 6207 6208 adj = __netdev_find_adj(adj_dev, dev_list); 6209 6210 if (!adj) { 6211 pr_err("Adjacency does not exist for device %s from %s\n", 6212 dev->name, adj_dev->name); 6213 WARN_ON(1); 6214 return; 6215 } 6216 6217 if (adj->ref_nr > ref_nr) { 6218 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 6219 dev->name, adj_dev->name, ref_nr, 6220 adj->ref_nr - ref_nr); 6221 adj->ref_nr -= ref_nr; 6222 return; 6223 } 6224 6225 if (adj->master) 6226 sysfs_remove_link(&(dev->dev.kobj), "master"); 6227 6228 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 6229 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 6230 6231 list_del_rcu(&adj->list); 6232 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 6233 adj_dev->name, dev->name, adj_dev->name); 6234 dev_put(adj_dev); 6235 kfree_rcu(adj, rcu); 6236 } 6237 6238 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 6239 struct net_device *upper_dev, 6240 struct list_head *up_list, 6241 struct list_head *down_list, 6242 void *private, bool master) 6243 { 6244 int ret; 6245 6246 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 6247 private, master); 6248 if (ret) 6249 return ret; 6250 6251 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 6252 private, false); 6253 if (ret) { 6254 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 6255 return ret; 6256 } 6257 6258 return 0; 6259 } 6260 6261 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 6262 struct net_device *upper_dev, 6263 u16 ref_nr, 6264 struct list_head *up_list, 6265 struct list_head *down_list) 6266 { 6267 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 6268 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 6269 } 6270 6271 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 6272 struct net_device *upper_dev, 6273 void *private, bool master) 6274 { 6275 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 6276 &dev->adj_list.upper, 6277 &upper_dev->adj_list.lower, 6278 private, master); 6279 } 6280 6281 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 6282 struct net_device *upper_dev) 6283 { 6284 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 6285 &dev->adj_list.upper, 6286 &upper_dev->adj_list.lower); 6287 } 6288 6289 static int __netdev_upper_dev_link(struct net_device *dev, 6290 struct net_device *upper_dev, bool master, 6291 void *upper_priv, void *upper_info, 6292 struct netlink_ext_ack *extack) 6293 { 6294 struct netdev_notifier_changeupper_info changeupper_info = { 6295 .info = { 6296 .dev = dev, 6297 .extack = extack, 6298 }, 6299 .upper_dev = upper_dev, 6300 .master = master, 6301 .linking = true, 6302 .upper_info = upper_info, 6303 }; 6304 int ret = 0; 6305 6306 ASSERT_RTNL(); 6307 6308 if (dev == upper_dev) 6309 return -EBUSY; 6310 6311 /* To prevent loops, check if dev is not upper device to upper_dev. */ 6312 if (netdev_has_upper_dev(upper_dev, dev)) 6313 return -EBUSY; 6314 6315 if (netdev_has_upper_dev(dev, upper_dev)) 6316 return -EEXIST; 6317 6318 if (master && netdev_master_upper_dev_get(dev)) 6319 return -EBUSY; 6320 6321 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 6322 &changeupper_info.info); 6323 ret = notifier_to_errno(ret); 6324 if (ret) 6325 return ret; 6326 6327 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 6328 master); 6329 if (ret) 6330 return ret; 6331 6332 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 6333 &changeupper_info.info); 6334 ret = notifier_to_errno(ret); 6335 if (ret) 6336 goto rollback; 6337 6338 return 0; 6339 6340 rollback: 6341 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 6342 6343 return ret; 6344 } 6345 6346 /** 6347 * netdev_upper_dev_link - Add a link to the upper device 6348 * @dev: device 6349 * @upper_dev: new upper device 6350 * 6351 * Adds a link to device which is upper to this one. The caller must hold 6352 * the RTNL lock. On a failure a negative errno code is returned. 6353 * On success the reference counts are adjusted and the function 6354 * returns zero. 6355 */ 6356 int netdev_upper_dev_link(struct net_device *dev, 6357 struct net_device *upper_dev, 6358 struct netlink_ext_ack *extack) 6359 { 6360 return __netdev_upper_dev_link(dev, upper_dev, false, 6361 NULL, NULL, extack); 6362 } 6363 EXPORT_SYMBOL(netdev_upper_dev_link); 6364 6365 /** 6366 * netdev_master_upper_dev_link - Add a master link to the upper device 6367 * @dev: device 6368 * @upper_dev: new upper device 6369 * @upper_priv: upper device private 6370 * @upper_info: upper info to be passed down via notifier 6371 * 6372 * Adds a link to device which is upper to this one. In this case, only 6373 * one master upper device can be linked, although other non-master devices 6374 * might be linked as well. The caller must hold the RTNL lock. 6375 * On a failure a negative errno code is returned. On success the reference 6376 * counts are adjusted and the function returns zero. 6377 */ 6378 int netdev_master_upper_dev_link(struct net_device *dev, 6379 struct net_device *upper_dev, 6380 void *upper_priv, void *upper_info, 6381 struct netlink_ext_ack *extack) 6382 { 6383 return __netdev_upper_dev_link(dev, upper_dev, true, 6384 upper_priv, upper_info, extack); 6385 } 6386 EXPORT_SYMBOL(netdev_master_upper_dev_link); 6387 6388 /** 6389 * netdev_upper_dev_unlink - Removes a link to upper device 6390 * @dev: device 6391 * @upper_dev: new upper device 6392 * 6393 * Removes a link to device which is upper to this one. The caller must hold 6394 * the RTNL lock. 6395 */ 6396 void netdev_upper_dev_unlink(struct net_device *dev, 6397 struct net_device *upper_dev) 6398 { 6399 struct netdev_notifier_changeupper_info changeupper_info = { 6400 .info = { 6401 .dev = dev, 6402 }, 6403 .upper_dev = upper_dev, 6404 .linking = false, 6405 }; 6406 6407 ASSERT_RTNL(); 6408 6409 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 6410 6411 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 6412 &changeupper_info.info); 6413 6414 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 6415 6416 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 6417 &changeupper_info.info); 6418 } 6419 EXPORT_SYMBOL(netdev_upper_dev_unlink); 6420 6421 /** 6422 * netdev_bonding_info_change - Dispatch event about slave change 6423 * @dev: device 6424 * @bonding_info: info to dispatch 6425 * 6426 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 6427 * The caller must hold the RTNL lock. 6428 */ 6429 void netdev_bonding_info_change(struct net_device *dev, 6430 struct netdev_bonding_info *bonding_info) 6431 { 6432 struct netdev_notifier_bonding_info info = { 6433 .info.dev = dev, 6434 }; 6435 6436 memcpy(&info.bonding_info, bonding_info, 6437 sizeof(struct netdev_bonding_info)); 6438 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 6439 &info.info); 6440 } 6441 EXPORT_SYMBOL(netdev_bonding_info_change); 6442 6443 static void netdev_adjacent_add_links(struct net_device *dev) 6444 { 6445 struct netdev_adjacent *iter; 6446 6447 struct net *net = dev_net(dev); 6448 6449 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6450 if (!net_eq(net, dev_net(iter->dev))) 6451 continue; 6452 netdev_adjacent_sysfs_add(iter->dev, dev, 6453 &iter->dev->adj_list.lower); 6454 netdev_adjacent_sysfs_add(dev, iter->dev, 6455 &dev->adj_list.upper); 6456 } 6457 6458 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6459 if (!net_eq(net, dev_net(iter->dev))) 6460 continue; 6461 netdev_adjacent_sysfs_add(iter->dev, dev, 6462 &iter->dev->adj_list.upper); 6463 netdev_adjacent_sysfs_add(dev, iter->dev, 6464 &dev->adj_list.lower); 6465 } 6466 } 6467 6468 static void netdev_adjacent_del_links(struct net_device *dev) 6469 { 6470 struct netdev_adjacent *iter; 6471 6472 struct net *net = dev_net(dev); 6473 6474 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6475 if (!net_eq(net, dev_net(iter->dev))) 6476 continue; 6477 netdev_adjacent_sysfs_del(iter->dev, dev->name, 6478 &iter->dev->adj_list.lower); 6479 netdev_adjacent_sysfs_del(dev, iter->dev->name, 6480 &dev->adj_list.upper); 6481 } 6482 6483 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6484 if (!net_eq(net, dev_net(iter->dev))) 6485 continue; 6486 netdev_adjacent_sysfs_del(iter->dev, dev->name, 6487 &iter->dev->adj_list.upper); 6488 netdev_adjacent_sysfs_del(dev, iter->dev->name, 6489 &dev->adj_list.lower); 6490 } 6491 } 6492 6493 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 6494 { 6495 struct netdev_adjacent *iter; 6496 6497 struct net *net = dev_net(dev); 6498 6499 list_for_each_entry(iter, &dev->adj_list.upper, list) { 6500 if (!net_eq(net, dev_net(iter->dev))) 6501 continue; 6502 netdev_adjacent_sysfs_del(iter->dev, oldname, 6503 &iter->dev->adj_list.lower); 6504 netdev_adjacent_sysfs_add(iter->dev, dev, 6505 &iter->dev->adj_list.lower); 6506 } 6507 6508 list_for_each_entry(iter, &dev->adj_list.lower, list) { 6509 if (!net_eq(net, dev_net(iter->dev))) 6510 continue; 6511 netdev_adjacent_sysfs_del(iter->dev, oldname, 6512 &iter->dev->adj_list.upper); 6513 netdev_adjacent_sysfs_add(iter->dev, dev, 6514 &iter->dev->adj_list.upper); 6515 } 6516 } 6517 6518 void *netdev_lower_dev_get_private(struct net_device *dev, 6519 struct net_device *lower_dev) 6520 { 6521 struct netdev_adjacent *lower; 6522 6523 if (!lower_dev) 6524 return NULL; 6525 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 6526 if (!lower) 6527 return NULL; 6528 6529 return lower->private; 6530 } 6531 EXPORT_SYMBOL(netdev_lower_dev_get_private); 6532 6533 6534 int dev_get_nest_level(struct net_device *dev) 6535 { 6536 struct net_device *lower = NULL; 6537 struct list_head *iter; 6538 int max_nest = -1; 6539 int nest; 6540 6541 ASSERT_RTNL(); 6542 6543 netdev_for_each_lower_dev(dev, lower, iter) { 6544 nest = dev_get_nest_level(lower); 6545 if (max_nest < nest) 6546 max_nest = nest; 6547 } 6548 6549 return max_nest + 1; 6550 } 6551 EXPORT_SYMBOL(dev_get_nest_level); 6552 6553 /** 6554 * netdev_lower_change - Dispatch event about lower device state change 6555 * @lower_dev: device 6556 * @lower_state_info: state to dispatch 6557 * 6558 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 6559 * The caller must hold the RTNL lock. 6560 */ 6561 void netdev_lower_state_changed(struct net_device *lower_dev, 6562 void *lower_state_info) 6563 { 6564 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 6565 .info.dev = lower_dev, 6566 }; 6567 6568 ASSERT_RTNL(); 6569 changelowerstate_info.lower_state_info = lower_state_info; 6570 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 6571 &changelowerstate_info.info); 6572 } 6573 EXPORT_SYMBOL(netdev_lower_state_changed); 6574 6575 static void dev_change_rx_flags(struct net_device *dev, int flags) 6576 { 6577 const struct net_device_ops *ops = dev->netdev_ops; 6578 6579 if (ops->ndo_change_rx_flags) 6580 ops->ndo_change_rx_flags(dev, flags); 6581 } 6582 6583 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 6584 { 6585 unsigned int old_flags = dev->flags; 6586 kuid_t uid; 6587 kgid_t gid; 6588 6589 ASSERT_RTNL(); 6590 6591 dev->flags |= IFF_PROMISC; 6592 dev->promiscuity += inc; 6593 if (dev->promiscuity == 0) { 6594 /* 6595 * Avoid overflow. 6596 * If inc causes overflow, untouch promisc and return error. 6597 */ 6598 if (inc < 0) 6599 dev->flags &= ~IFF_PROMISC; 6600 else { 6601 dev->promiscuity -= inc; 6602 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 6603 dev->name); 6604 return -EOVERFLOW; 6605 } 6606 } 6607 if (dev->flags != old_flags) { 6608 pr_info("device %s %s promiscuous mode\n", 6609 dev->name, 6610 dev->flags & IFF_PROMISC ? "entered" : "left"); 6611 if (audit_enabled) { 6612 current_uid_gid(&uid, &gid); 6613 audit_log(current->audit_context, GFP_ATOMIC, 6614 AUDIT_ANOM_PROMISCUOUS, 6615 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 6616 dev->name, (dev->flags & IFF_PROMISC), 6617 (old_flags & IFF_PROMISC), 6618 from_kuid(&init_user_ns, audit_get_loginuid(current)), 6619 from_kuid(&init_user_ns, uid), 6620 from_kgid(&init_user_ns, gid), 6621 audit_get_sessionid(current)); 6622 } 6623 6624 dev_change_rx_flags(dev, IFF_PROMISC); 6625 } 6626 if (notify) 6627 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 6628 return 0; 6629 } 6630 6631 /** 6632 * dev_set_promiscuity - update promiscuity count on a device 6633 * @dev: device 6634 * @inc: modifier 6635 * 6636 * Add or remove promiscuity from a device. While the count in the device 6637 * remains above zero the interface remains promiscuous. Once it hits zero 6638 * the device reverts back to normal filtering operation. A negative inc 6639 * value is used to drop promiscuity on the device. 6640 * Return 0 if successful or a negative errno code on error. 6641 */ 6642 int dev_set_promiscuity(struct net_device *dev, int inc) 6643 { 6644 unsigned int old_flags = dev->flags; 6645 int err; 6646 6647 err = __dev_set_promiscuity(dev, inc, true); 6648 if (err < 0) 6649 return err; 6650 if (dev->flags != old_flags) 6651 dev_set_rx_mode(dev); 6652 return err; 6653 } 6654 EXPORT_SYMBOL(dev_set_promiscuity); 6655 6656 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 6657 { 6658 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 6659 6660 ASSERT_RTNL(); 6661 6662 dev->flags |= IFF_ALLMULTI; 6663 dev->allmulti += inc; 6664 if (dev->allmulti == 0) { 6665 /* 6666 * Avoid overflow. 6667 * If inc causes overflow, untouch allmulti and return error. 6668 */ 6669 if (inc < 0) 6670 dev->flags &= ~IFF_ALLMULTI; 6671 else { 6672 dev->allmulti -= inc; 6673 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 6674 dev->name); 6675 return -EOVERFLOW; 6676 } 6677 } 6678 if (dev->flags ^ old_flags) { 6679 dev_change_rx_flags(dev, IFF_ALLMULTI); 6680 dev_set_rx_mode(dev); 6681 if (notify) 6682 __dev_notify_flags(dev, old_flags, 6683 dev->gflags ^ old_gflags); 6684 } 6685 return 0; 6686 } 6687 6688 /** 6689 * dev_set_allmulti - update allmulti count on a device 6690 * @dev: device 6691 * @inc: modifier 6692 * 6693 * Add or remove reception of all multicast frames to a device. While the 6694 * count in the device remains above zero the interface remains listening 6695 * to all interfaces. Once it hits zero the device reverts back to normal 6696 * filtering operation. A negative @inc value is used to drop the counter 6697 * when releasing a resource needing all multicasts. 6698 * Return 0 if successful or a negative errno code on error. 6699 */ 6700 6701 int dev_set_allmulti(struct net_device *dev, int inc) 6702 { 6703 return __dev_set_allmulti(dev, inc, true); 6704 } 6705 EXPORT_SYMBOL(dev_set_allmulti); 6706 6707 /* 6708 * Upload unicast and multicast address lists to device and 6709 * configure RX filtering. When the device doesn't support unicast 6710 * filtering it is put in promiscuous mode while unicast addresses 6711 * are present. 6712 */ 6713 void __dev_set_rx_mode(struct net_device *dev) 6714 { 6715 const struct net_device_ops *ops = dev->netdev_ops; 6716 6717 /* dev_open will call this function so the list will stay sane. */ 6718 if (!(dev->flags&IFF_UP)) 6719 return; 6720 6721 if (!netif_device_present(dev)) 6722 return; 6723 6724 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 6725 /* Unicast addresses changes may only happen under the rtnl, 6726 * therefore calling __dev_set_promiscuity here is safe. 6727 */ 6728 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 6729 __dev_set_promiscuity(dev, 1, false); 6730 dev->uc_promisc = true; 6731 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 6732 __dev_set_promiscuity(dev, -1, false); 6733 dev->uc_promisc = false; 6734 } 6735 } 6736 6737 if (ops->ndo_set_rx_mode) 6738 ops->ndo_set_rx_mode(dev); 6739 } 6740 6741 void dev_set_rx_mode(struct net_device *dev) 6742 { 6743 netif_addr_lock_bh(dev); 6744 __dev_set_rx_mode(dev); 6745 netif_addr_unlock_bh(dev); 6746 } 6747 6748 /** 6749 * dev_get_flags - get flags reported to userspace 6750 * @dev: device 6751 * 6752 * Get the combination of flag bits exported through APIs to userspace. 6753 */ 6754 unsigned int dev_get_flags(const struct net_device *dev) 6755 { 6756 unsigned int flags; 6757 6758 flags = (dev->flags & ~(IFF_PROMISC | 6759 IFF_ALLMULTI | 6760 IFF_RUNNING | 6761 IFF_LOWER_UP | 6762 IFF_DORMANT)) | 6763 (dev->gflags & (IFF_PROMISC | 6764 IFF_ALLMULTI)); 6765 6766 if (netif_running(dev)) { 6767 if (netif_oper_up(dev)) 6768 flags |= IFF_RUNNING; 6769 if (netif_carrier_ok(dev)) 6770 flags |= IFF_LOWER_UP; 6771 if (netif_dormant(dev)) 6772 flags |= IFF_DORMANT; 6773 } 6774 6775 return flags; 6776 } 6777 EXPORT_SYMBOL(dev_get_flags); 6778 6779 int __dev_change_flags(struct net_device *dev, unsigned int flags) 6780 { 6781 unsigned int old_flags = dev->flags; 6782 int ret; 6783 6784 ASSERT_RTNL(); 6785 6786 /* 6787 * Set the flags on our device. 6788 */ 6789 6790 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 6791 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 6792 IFF_AUTOMEDIA)) | 6793 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 6794 IFF_ALLMULTI)); 6795 6796 /* 6797 * Load in the correct multicast list now the flags have changed. 6798 */ 6799 6800 if ((old_flags ^ flags) & IFF_MULTICAST) 6801 dev_change_rx_flags(dev, IFF_MULTICAST); 6802 6803 dev_set_rx_mode(dev); 6804 6805 /* 6806 * Have we downed the interface. We handle IFF_UP ourselves 6807 * according to user attempts to set it, rather than blindly 6808 * setting it. 6809 */ 6810 6811 ret = 0; 6812 if ((old_flags ^ flags) & IFF_UP) { 6813 if (old_flags & IFF_UP) 6814 __dev_close(dev); 6815 else 6816 ret = __dev_open(dev); 6817 } 6818 6819 if ((flags ^ dev->gflags) & IFF_PROMISC) { 6820 int inc = (flags & IFF_PROMISC) ? 1 : -1; 6821 unsigned int old_flags = dev->flags; 6822 6823 dev->gflags ^= IFF_PROMISC; 6824 6825 if (__dev_set_promiscuity(dev, inc, false) >= 0) 6826 if (dev->flags != old_flags) 6827 dev_set_rx_mode(dev); 6828 } 6829 6830 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 6831 * is important. Some (broken) drivers set IFF_PROMISC, when 6832 * IFF_ALLMULTI is requested not asking us and not reporting. 6833 */ 6834 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 6835 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 6836 6837 dev->gflags ^= IFF_ALLMULTI; 6838 __dev_set_allmulti(dev, inc, false); 6839 } 6840 6841 return ret; 6842 } 6843 6844 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 6845 unsigned int gchanges) 6846 { 6847 unsigned int changes = dev->flags ^ old_flags; 6848 6849 if (gchanges) 6850 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 6851 6852 if (changes & IFF_UP) { 6853 if (dev->flags & IFF_UP) 6854 call_netdevice_notifiers(NETDEV_UP, dev); 6855 else 6856 call_netdevice_notifiers(NETDEV_DOWN, dev); 6857 } 6858 6859 if (dev->flags & IFF_UP && 6860 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 6861 struct netdev_notifier_change_info change_info = { 6862 .info = { 6863 .dev = dev, 6864 }, 6865 .flags_changed = changes, 6866 }; 6867 6868 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 6869 } 6870 } 6871 6872 /** 6873 * dev_change_flags - change device settings 6874 * @dev: device 6875 * @flags: device state flags 6876 * 6877 * Change settings on device based state flags. The flags are 6878 * in the userspace exported format. 6879 */ 6880 int dev_change_flags(struct net_device *dev, unsigned int flags) 6881 { 6882 int ret; 6883 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 6884 6885 ret = __dev_change_flags(dev, flags); 6886 if (ret < 0) 6887 return ret; 6888 6889 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 6890 __dev_notify_flags(dev, old_flags, changes); 6891 return ret; 6892 } 6893 EXPORT_SYMBOL(dev_change_flags); 6894 6895 int __dev_set_mtu(struct net_device *dev, int new_mtu) 6896 { 6897 const struct net_device_ops *ops = dev->netdev_ops; 6898 6899 if (ops->ndo_change_mtu) 6900 return ops->ndo_change_mtu(dev, new_mtu); 6901 6902 dev->mtu = new_mtu; 6903 return 0; 6904 } 6905 EXPORT_SYMBOL(__dev_set_mtu); 6906 6907 /** 6908 * dev_set_mtu - Change maximum transfer unit 6909 * @dev: device 6910 * @new_mtu: new transfer unit 6911 * 6912 * Change the maximum transfer size of the network device. 6913 */ 6914 int dev_set_mtu(struct net_device *dev, int new_mtu) 6915 { 6916 int err, orig_mtu; 6917 6918 if (new_mtu == dev->mtu) 6919 return 0; 6920 6921 /* MTU must be positive, and in range */ 6922 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 6923 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", 6924 dev->name, new_mtu, dev->min_mtu); 6925 return -EINVAL; 6926 } 6927 6928 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 6929 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", 6930 dev->name, new_mtu, dev->max_mtu); 6931 return -EINVAL; 6932 } 6933 6934 if (!netif_device_present(dev)) 6935 return -ENODEV; 6936 6937 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 6938 err = notifier_to_errno(err); 6939 if (err) 6940 return err; 6941 6942 orig_mtu = dev->mtu; 6943 err = __dev_set_mtu(dev, new_mtu); 6944 6945 if (!err) { 6946 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 6947 err = notifier_to_errno(err); 6948 if (err) { 6949 /* setting mtu back and notifying everyone again, 6950 * so that they have a chance to revert changes. 6951 */ 6952 __dev_set_mtu(dev, orig_mtu); 6953 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 6954 } 6955 } 6956 return err; 6957 } 6958 EXPORT_SYMBOL(dev_set_mtu); 6959 6960 /** 6961 * dev_set_group - Change group this device belongs to 6962 * @dev: device 6963 * @new_group: group this device should belong to 6964 */ 6965 void dev_set_group(struct net_device *dev, int new_group) 6966 { 6967 dev->group = new_group; 6968 } 6969 EXPORT_SYMBOL(dev_set_group); 6970 6971 /** 6972 * dev_set_mac_address - Change Media Access Control Address 6973 * @dev: device 6974 * @sa: new address 6975 * 6976 * Change the hardware (MAC) address of the device 6977 */ 6978 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 6979 { 6980 const struct net_device_ops *ops = dev->netdev_ops; 6981 int err; 6982 6983 if (!ops->ndo_set_mac_address) 6984 return -EOPNOTSUPP; 6985 if (sa->sa_family != dev->type) 6986 return -EINVAL; 6987 if (!netif_device_present(dev)) 6988 return -ENODEV; 6989 err = ops->ndo_set_mac_address(dev, sa); 6990 if (err) 6991 return err; 6992 dev->addr_assign_type = NET_ADDR_SET; 6993 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 6994 add_device_randomness(dev->dev_addr, dev->addr_len); 6995 return 0; 6996 } 6997 EXPORT_SYMBOL(dev_set_mac_address); 6998 6999 /** 7000 * dev_change_carrier - Change device carrier 7001 * @dev: device 7002 * @new_carrier: new value 7003 * 7004 * Change device carrier 7005 */ 7006 int dev_change_carrier(struct net_device *dev, bool new_carrier) 7007 { 7008 const struct net_device_ops *ops = dev->netdev_ops; 7009 7010 if (!ops->ndo_change_carrier) 7011 return -EOPNOTSUPP; 7012 if (!netif_device_present(dev)) 7013 return -ENODEV; 7014 return ops->ndo_change_carrier(dev, new_carrier); 7015 } 7016 EXPORT_SYMBOL(dev_change_carrier); 7017 7018 /** 7019 * dev_get_phys_port_id - Get device physical port ID 7020 * @dev: device 7021 * @ppid: port ID 7022 * 7023 * Get device physical port ID 7024 */ 7025 int dev_get_phys_port_id(struct net_device *dev, 7026 struct netdev_phys_item_id *ppid) 7027 { 7028 const struct net_device_ops *ops = dev->netdev_ops; 7029 7030 if (!ops->ndo_get_phys_port_id) 7031 return -EOPNOTSUPP; 7032 return ops->ndo_get_phys_port_id(dev, ppid); 7033 } 7034 EXPORT_SYMBOL(dev_get_phys_port_id); 7035 7036 /** 7037 * dev_get_phys_port_name - Get device physical port name 7038 * @dev: device 7039 * @name: port name 7040 * @len: limit of bytes to copy to name 7041 * 7042 * Get device physical port name 7043 */ 7044 int dev_get_phys_port_name(struct net_device *dev, 7045 char *name, size_t len) 7046 { 7047 const struct net_device_ops *ops = dev->netdev_ops; 7048 7049 if (!ops->ndo_get_phys_port_name) 7050 return -EOPNOTSUPP; 7051 return ops->ndo_get_phys_port_name(dev, name, len); 7052 } 7053 EXPORT_SYMBOL(dev_get_phys_port_name); 7054 7055 /** 7056 * dev_change_proto_down - update protocol port state information 7057 * @dev: device 7058 * @proto_down: new value 7059 * 7060 * This info can be used by switch drivers to set the phys state of the 7061 * port. 7062 */ 7063 int dev_change_proto_down(struct net_device *dev, bool proto_down) 7064 { 7065 const struct net_device_ops *ops = dev->netdev_ops; 7066 7067 if (!ops->ndo_change_proto_down) 7068 return -EOPNOTSUPP; 7069 if (!netif_device_present(dev)) 7070 return -ENODEV; 7071 return ops->ndo_change_proto_down(dev, proto_down); 7072 } 7073 EXPORT_SYMBOL(dev_change_proto_down); 7074 7075 u8 __dev_xdp_attached(struct net_device *dev, bpf_op_t bpf_op, u32 *prog_id) 7076 { 7077 struct netdev_bpf xdp; 7078 7079 memset(&xdp, 0, sizeof(xdp)); 7080 xdp.command = XDP_QUERY_PROG; 7081 7082 /* Query must always succeed. */ 7083 WARN_ON(bpf_op(dev, &xdp) < 0); 7084 if (prog_id) 7085 *prog_id = xdp.prog_id; 7086 7087 return xdp.prog_attached; 7088 } 7089 7090 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, 7091 struct netlink_ext_ack *extack, u32 flags, 7092 struct bpf_prog *prog) 7093 { 7094 struct netdev_bpf xdp; 7095 7096 memset(&xdp, 0, sizeof(xdp)); 7097 if (flags & XDP_FLAGS_HW_MODE) 7098 xdp.command = XDP_SETUP_PROG_HW; 7099 else 7100 xdp.command = XDP_SETUP_PROG; 7101 xdp.extack = extack; 7102 xdp.flags = flags; 7103 xdp.prog = prog; 7104 7105 return bpf_op(dev, &xdp); 7106 } 7107 7108 /** 7109 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 7110 * @dev: device 7111 * @extack: netlink extended ack 7112 * @fd: new program fd or negative value to clear 7113 * @flags: xdp-related flags 7114 * 7115 * Set or clear a bpf program for a device 7116 */ 7117 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 7118 int fd, u32 flags) 7119 { 7120 const struct net_device_ops *ops = dev->netdev_ops; 7121 struct bpf_prog *prog = NULL; 7122 bpf_op_t bpf_op, bpf_chk; 7123 int err; 7124 7125 ASSERT_RTNL(); 7126 7127 bpf_op = bpf_chk = ops->ndo_bpf; 7128 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) 7129 return -EOPNOTSUPP; 7130 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) 7131 bpf_op = generic_xdp_install; 7132 if (bpf_op == bpf_chk) 7133 bpf_chk = generic_xdp_install; 7134 7135 if (fd >= 0) { 7136 if (bpf_chk && __dev_xdp_attached(dev, bpf_chk, NULL)) 7137 return -EEXIST; 7138 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && 7139 __dev_xdp_attached(dev, bpf_op, NULL)) 7140 return -EBUSY; 7141 7142 if (bpf_op == ops->ndo_bpf) 7143 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 7144 dev); 7145 else 7146 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); 7147 if (IS_ERR(prog)) 7148 return PTR_ERR(prog); 7149 } 7150 7151 err = dev_xdp_install(dev, bpf_op, extack, flags, prog); 7152 if (err < 0 && prog) 7153 bpf_prog_put(prog); 7154 7155 return err; 7156 } 7157 7158 /** 7159 * dev_new_index - allocate an ifindex 7160 * @net: the applicable net namespace 7161 * 7162 * Returns a suitable unique value for a new device interface 7163 * number. The caller must hold the rtnl semaphore or the 7164 * dev_base_lock to be sure it remains unique. 7165 */ 7166 static int dev_new_index(struct net *net) 7167 { 7168 int ifindex = net->ifindex; 7169 7170 for (;;) { 7171 if (++ifindex <= 0) 7172 ifindex = 1; 7173 if (!__dev_get_by_index(net, ifindex)) 7174 return net->ifindex = ifindex; 7175 } 7176 } 7177 7178 /* Delayed registration/unregisteration */ 7179 static LIST_HEAD(net_todo_list); 7180 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 7181 7182 static void net_set_todo(struct net_device *dev) 7183 { 7184 list_add_tail(&dev->todo_list, &net_todo_list); 7185 dev_net(dev)->dev_unreg_count++; 7186 } 7187 7188 static void rollback_registered_many(struct list_head *head) 7189 { 7190 struct net_device *dev, *tmp; 7191 LIST_HEAD(close_head); 7192 7193 BUG_ON(dev_boot_phase); 7194 ASSERT_RTNL(); 7195 7196 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 7197 /* Some devices call without registering 7198 * for initialization unwind. Remove those 7199 * devices and proceed with the remaining. 7200 */ 7201 if (dev->reg_state == NETREG_UNINITIALIZED) { 7202 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 7203 dev->name, dev); 7204 7205 WARN_ON(1); 7206 list_del(&dev->unreg_list); 7207 continue; 7208 } 7209 dev->dismantle = true; 7210 BUG_ON(dev->reg_state != NETREG_REGISTERED); 7211 } 7212 7213 /* If device is running, close it first. */ 7214 list_for_each_entry(dev, head, unreg_list) 7215 list_add_tail(&dev->close_list, &close_head); 7216 dev_close_many(&close_head, true); 7217 7218 list_for_each_entry(dev, head, unreg_list) { 7219 /* And unlink it from device chain. */ 7220 unlist_netdevice(dev); 7221 7222 dev->reg_state = NETREG_UNREGISTERING; 7223 } 7224 flush_all_backlogs(); 7225 7226 synchronize_net(); 7227 7228 list_for_each_entry(dev, head, unreg_list) { 7229 struct sk_buff *skb = NULL; 7230 7231 /* Shutdown queueing discipline. */ 7232 dev_shutdown(dev); 7233 7234 7235 /* Notify protocols, that we are about to destroy 7236 * this device. They should clean all the things. 7237 */ 7238 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7239 7240 if (!dev->rtnl_link_ops || 7241 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 7242 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 7243 GFP_KERNEL, NULL); 7244 7245 /* 7246 * Flush the unicast and multicast chains 7247 */ 7248 dev_uc_flush(dev); 7249 dev_mc_flush(dev); 7250 7251 if (dev->netdev_ops->ndo_uninit) 7252 dev->netdev_ops->ndo_uninit(dev); 7253 7254 if (skb) 7255 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 7256 7257 /* Notifier chain MUST detach us all upper devices. */ 7258 WARN_ON(netdev_has_any_upper_dev(dev)); 7259 WARN_ON(netdev_has_any_lower_dev(dev)); 7260 7261 /* Remove entries from kobject tree */ 7262 netdev_unregister_kobject(dev); 7263 #ifdef CONFIG_XPS 7264 /* Remove XPS queueing entries */ 7265 netif_reset_xps_queues_gt(dev, 0); 7266 #endif 7267 } 7268 7269 synchronize_net(); 7270 7271 list_for_each_entry(dev, head, unreg_list) 7272 dev_put(dev); 7273 } 7274 7275 static void rollback_registered(struct net_device *dev) 7276 { 7277 LIST_HEAD(single); 7278 7279 list_add(&dev->unreg_list, &single); 7280 rollback_registered_many(&single); 7281 list_del(&single); 7282 } 7283 7284 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 7285 struct net_device *upper, netdev_features_t features) 7286 { 7287 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 7288 netdev_features_t feature; 7289 int feature_bit; 7290 7291 for_each_netdev_feature(&upper_disables, feature_bit) { 7292 feature = __NETIF_F_BIT(feature_bit); 7293 if (!(upper->wanted_features & feature) 7294 && (features & feature)) { 7295 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 7296 &feature, upper->name); 7297 features &= ~feature; 7298 } 7299 } 7300 7301 return features; 7302 } 7303 7304 static void netdev_sync_lower_features(struct net_device *upper, 7305 struct net_device *lower, netdev_features_t features) 7306 { 7307 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 7308 netdev_features_t feature; 7309 int feature_bit; 7310 7311 for_each_netdev_feature(&upper_disables, feature_bit) { 7312 feature = __NETIF_F_BIT(feature_bit); 7313 if (!(features & feature) && (lower->features & feature)) { 7314 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 7315 &feature, lower->name); 7316 lower->wanted_features &= ~feature; 7317 netdev_update_features(lower); 7318 7319 if (unlikely(lower->features & feature)) 7320 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 7321 &feature, lower->name); 7322 } 7323 } 7324 } 7325 7326 static netdev_features_t netdev_fix_features(struct net_device *dev, 7327 netdev_features_t features) 7328 { 7329 /* Fix illegal checksum combinations */ 7330 if ((features & NETIF_F_HW_CSUM) && 7331 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 7332 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 7333 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 7334 } 7335 7336 /* TSO requires that SG is present as well. */ 7337 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 7338 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 7339 features &= ~NETIF_F_ALL_TSO; 7340 } 7341 7342 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 7343 !(features & NETIF_F_IP_CSUM)) { 7344 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 7345 features &= ~NETIF_F_TSO; 7346 features &= ~NETIF_F_TSO_ECN; 7347 } 7348 7349 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 7350 !(features & NETIF_F_IPV6_CSUM)) { 7351 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 7352 features &= ~NETIF_F_TSO6; 7353 } 7354 7355 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 7356 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 7357 features &= ~NETIF_F_TSO_MANGLEID; 7358 7359 /* TSO ECN requires that TSO is present as well. */ 7360 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 7361 features &= ~NETIF_F_TSO_ECN; 7362 7363 /* Software GSO depends on SG. */ 7364 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 7365 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 7366 features &= ~NETIF_F_GSO; 7367 } 7368 7369 /* GSO partial features require GSO partial be set */ 7370 if ((features & dev->gso_partial_features) && 7371 !(features & NETIF_F_GSO_PARTIAL)) { 7372 netdev_dbg(dev, 7373 "Dropping partially supported GSO features since no GSO partial.\n"); 7374 features &= ~dev->gso_partial_features; 7375 } 7376 7377 return features; 7378 } 7379 7380 int __netdev_update_features(struct net_device *dev) 7381 { 7382 struct net_device *upper, *lower; 7383 netdev_features_t features; 7384 struct list_head *iter; 7385 int err = -1; 7386 7387 ASSERT_RTNL(); 7388 7389 features = netdev_get_wanted_features(dev); 7390 7391 if (dev->netdev_ops->ndo_fix_features) 7392 features = dev->netdev_ops->ndo_fix_features(dev, features); 7393 7394 /* driver might be less strict about feature dependencies */ 7395 features = netdev_fix_features(dev, features); 7396 7397 /* some features can't be enabled if they're off an an upper device */ 7398 netdev_for_each_upper_dev_rcu(dev, upper, iter) 7399 features = netdev_sync_upper_features(dev, upper, features); 7400 7401 if (dev->features == features) 7402 goto sync_lower; 7403 7404 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 7405 &dev->features, &features); 7406 7407 if (dev->netdev_ops->ndo_set_features) 7408 err = dev->netdev_ops->ndo_set_features(dev, features); 7409 else 7410 err = 0; 7411 7412 if (unlikely(err < 0)) { 7413 netdev_err(dev, 7414 "set_features() failed (%d); wanted %pNF, left %pNF\n", 7415 err, &features, &dev->features); 7416 /* return non-0 since some features might have changed and 7417 * it's better to fire a spurious notification than miss it 7418 */ 7419 return -1; 7420 } 7421 7422 sync_lower: 7423 /* some features must be disabled on lower devices when disabled 7424 * on an upper device (think: bonding master or bridge) 7425 */ 7426 netdev_for_each_lower_dev(dev, lower, iter) 7427 netdev_sync_lower_features(dev, lower, features); 7428 7429 if (!err) { 7430 netdev_features_t diff = features ^ dev->features; 7431 7432 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 7433 /* udp_tunnel_{get,drop}_rx_info both need 7434 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 7435 * device, or they won't do anything. 7436 * Thus we need to update dev->features 7437 * *before* calling udp_tunnel_get_rx_info, 7438 * but *after* calling udp_tunnel_drop_rx_info. 7439 */ 7440 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 7441 dev->features = features; 7442 udp_tunnel_get_rx_info(dev); 7443 } else { 7444 udp_tunnel_drop_rx_info(dev); 7445 } 7446 } 7447 7448 dev->features = features; 7449 } 7450 7451 return err < 0 ? 0 : 1; 7452 } 7453 7454 /** 7455 * netdev_update_features - recalculate device features 7456 * @dev: the device to check 7457 * 7458 * Recalculate dev->features set and send notifications if it 7459 * has changed. Should be called after driver or hardware dependent 7460 * conditions might have changed that influence the features. 7461 */ 7462 void netdev_update_features(struct net_device *dev) 7463 { 7464 if (__netdev_update_features(dev)) 7465 netdev_features_change(dev); 7466 } 7467 EXPORT_SYMBOL(netdev_update_features); 7468 7469 /** 7470 * netdev_change_features - recalculate device features 7471 * @dev: the device to check 7472 * 7473 * Recalculate dev->features set and send notifications even 7474 * if they have not changed. Should be called instead of 7475 * netdev_update_features() if also dev->vlan_features might 7476 * have changed to allow the changes to be propagated to stacked 7477 * VLAN devices. 7478 */ 7479 void netdev_change_features(struct net_device *dev) 7480 { 7481 __netdev_update_features(dev); 7482 netdev_features_change(dev); 7483 } 7484 EXPORT_SYMBOL(netdev_change_features); 7485 7486 /** 7487 * netif_stacked_transfer_operstate - transfer operstate 7488 * @rootdev: the root or lower level device to transfer state from 7489 * @dev: the device to transfer operstate to 7490 * 7491 * Transfer operational state from root to device. This is normally 7492 * called when a stacking relationship exists between the root 7493 * device and the device(a leaf device). 7494 */ 7495 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 7496 struct net_device *dev) 7497 { 7498 if (rootdev->operstate == IF_OPER_DORMANT) 7499 netif_dormant_on(dev); 7500 else 7501 netif_dormant_off(dev); 7502 7503 if (netif_carrier_ok(rootdev)) 7504 netif_carrier_on(dev); 7505 else 7506 netif_carrier_off(dev); 7507 } 7508 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 7509 7510 #ifdef CONFIG_SYSFS 7511 static int netif_alloc_rx_queues(struct net_device *dev) 7512 { 7513 unsigned int i, count = dev->num_rx_queues; 7514 struct netdev_rx_queue *rx; 7515 size_t sz = count * sizeof(*rx); 7516 7517 BUG_ON(count < 1); 7518 7519 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 7520 if (!rx) 7521 return -ENOMEM; 7522 7523 dev->_rx = rx; 7524 7525 for (i = 0; i < count; i++) 7526 rx[i].dev = dev; 7527 return 0; 7528 } 7529 #endif 7530 7531 static void netdev_init_one_queue(struct net_device *dev, 7532 struct netdev_queue *queue, void *_unused) 7533 { 7534 /* Initialize queue lock */ 7535 spin_lock_init(&queue->_xmit_lock); 7536 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 7537 queue->xmit_lock_owner = -1; 7538 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 7539 queue->dev = dev; 7540 #ifdef CONFIG_BQL 7541 dql_init(&queue->dql, HZ); 7542 #endif 7543 } 7544 7545 static void netif_free_tx_queues(struct net_device *dev) 7546 { 7547 kvfree(dev->_tx); 7548 } 7549 7550 static int netif_alloc_netdev_queues(struct net_device *dev) 7551 { 7552 unsigned int count = dev->num_tx_queues; 7553 struct netdev_queue *tx; 7554 size_t sz = count * sizeof(*tx); 7555 7556 if (count < 1 || count > 0xffff) 7557 return -EINVAL; 7558 7559 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 7560 if (!tx) 7561 return -ENOMEM; 7562 7563 dev->_tx = tx; 7564 7565 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 7566 spin_lock_init(&dev->tx_global_lock); 7567 7568 return 0; 7569 } 7570 7571 void netif_tx_stop_all_queues(struct net_device *dev) 7572 { 7573 unsigned int i; 7574 7575 for (i = 0; i < dev->num_tx_queues; i++) { 7576 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 7577 7578 netif_tx_stop_queue(txq); 7579 } 7580 } 7581 EXPORT_SYMBOL(netif_tx_stop_all_queues); 7582 7583 /** 7584 * register_netdevice - register a network device 7585 * @dev: device to register 7586 * 7587 * Take a completed network device structure and add it to the kernel 7588 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 7589 * chain. 0 is returned on success. A negative errno code is returned 7590 * on a failure to set up the device, or if the name is a duplicate. 7591 * 7592 * Callers must hold the rtnl semaphore. You may want 7593 * register_netdev() instead of this. 7594 * 7595 * BUGS: 7596 * The locking appears insufficient to guarantee two parallel registers 7597 * will not get the same name. 7598 */ 7599 7600 int register_netdevice(struct net_device *dev) 7601 { 7602 int ret; 7603 struct net *net = dev_net(dev); 7604 7605 BUG_ON(dev_boot_phase); 7606 ASSERT_RTNL(); 7607 7608 might_sleep(); 7609 7610 /* When net_device's are persistent, this will be fatal. */ 7611 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 7612 BUG_ON(!net); 7613 7614 spin_lock_init(&dev->addr_list_lock); 7615 netdev_set_addr_lockdep_class(dev); 7616 7617 ret = dev_get_valid_name(net, dev, dev->name); 7618 if (ret < 0) 7619 goto out; 7620 7621 /* Init, if this function is available */ 7622 if (dev->netdev_ops->ndo_init) { 7623 ret = dev->netdev_ops->ndo_init(dev); 7624 if (ret) { 7625 if (ret > 0) 7626 ret = -EIO; 7627 goto out; 7628 } 7629 } 7630 7631 if (((dev->hw_features | dev->features) & 7632 NETIF_F_HW_VLAN_CTAG_FILTER) && 7633 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 7634 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 7635 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 7636 ret = -EINVAL; 7637 goto err_uninit; 7638 } 7639 7640 ret = -EBUSY; 7641 if (!dev->ifindex) 7642 dev->ifindex = dev_new_index(net); 7643 else if (__dev_get_by_index(net, dev->ifindex)) 7644 goto err_uninit; 7645 7646 /* Transfer changeable features to wanted_features and enable 7647 * software offloads (GSO and GRO). 7648 */ 7649 dev->hw_features |= NETIF_F_SOFT_FEATURES; 7650 dev->features |= NETIF_F_SOFT_FEATURES; 7651 7652 if (dev->netdev_ops->ndo_udp_tunnel_add) { 7653 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 7654 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 7655 } 7656 7657 dev->wanted_features = dev->features & dev->hw_features; 7658 7659 if (!(dev->flags & IFF_LOOPBACK)) 7660 dev->hw_features |= NETIF_F_NOCACHE_COPY; 7661 7662 /* If IPv4 TCP segmentation offload is supported we should also 7663 * allow the device to enable segmenting the frame with the option 7664 * of ignoring a static IP ID value. This doesn't enable the 7665 * feature itself but allows the user to enable it later. 7666 */ 7667 if (dev->hw_features & NETIF_F_TSO) 7668 dev->hw_features |= NETIF_F_TSO_MANGLEID; 7669 if (dev->vlan_features & NETIF_F_TSO) 7670 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 7671 if (dev->mpls_features & NETIF_F_TSO) 7672 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 7673 if (dev->hw_enc_features & NETIF_F_TSO) 7674 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 7675 7676 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 7677 */ 7678 dev->vlan_features |= NETIF_F_HIGHDMA; 7679 7680 /* Make NETIF_F_SG inheritable to tunnel devices. 7681 */ 7682 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 7683 7684 /* Make NETIF_F_SG inheritable to MPLS. 7685 */ 7686 dev->mpls_features |= NETIF_F_SG; 7687 7688 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 7689 ret = notifier_to_errno(ret); 7690 if (ret) 7691 goto err_uninit; 7692 7693 ret = netdev_register_kobject(dev); 7694 if (ret) 7695 goto err_uninit; 7696 dev->reg_state = NETREG_REGISTERED; 7697 7698 __netdev_update_features(dev); 7699 7700 /* 7701 * Default initial state at registry is that the 7702 * device is present. 7703 */ 7704 7705 set_bit(__LINK_STATE_PRESENT, &dev->state); 7706 7707 linkwatch_init_dev(dev); 7708 7709 dev_init_scheduler(dev); 7710 dev_hold(dev); 7711 list_netdevice(dev); 7712 add_device_randomness(dev->dev_addr, dev->addr_len); 7713 7714 /* If the device has permanent device address, driver should 7715 * set dev_addr and also addr_assign_type should be set to 7716 * NET_ADDR_PERM (default value). 7717 */ 7718 if (dev->addr_assign_type == NET_ADDR_PERM) 7719 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 7720 7721 /* Notify protocols, that a new device appeared. */ 7722 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 7723 ret = notifier_to_errno(ret); 7724 if (ret) { 7725 rollback_registered(dev); 7726 dev->reg_state = NETREG_UNREGISTERED; 7727 } 7728 /* 7729 * Prevent userspace races by waiting until the network 7730 * device is fully setup before sending notifications. 7731 */ 7732 if (!dev->rtnl_link_ops || 7733 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 7734 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 7735 7736 out: 7737 return ret; 7738 7739 err_uninit: 7740 if (dev->netdev_ops->ndo_uninit) 7741 dev->netdev_ops->ndo_uninit(dev); 7742 if (dev->priv_destructor) 7743 dev->priv_destructor(dev); 7744 goto out; 7745 } 7746 EXPORT_SYMBOL(register_netdevice); 7747 7748 /** 7749 * init_dummy_netdev - init a dummy network device for NAPI 7750 * @dev: device to init 7751 * 7752 * This takes a network device structure and initialize the minimum 7753 * amount of fields so it can be used to schedule NAPI polls without 7754 * registering a full blown interface. This is to be used by drivers 7755 * that need to tie several hardware interfaces to a single NAPI 7756 * poll scheduler due to HW limitations. 7757 */ 7758 int init_dummy_netdev(struct net_device *dev) 7759 { 7760 /* Clear everything. Note we don't initialize spinlocks 7761 * are they aren't supposed to be taken by any of the 7762 * NAPI code and this dummy netdev is supposed to be 7763 * only ever used for NAPI polls 7764 */ 7765 memset(dev, 0, sizeof(struct net_device)); 7766 7767 /* make sure we BUG if trying to hit standard 7768 * register/unregister code path 7769 */ 7770 dev->reg_state = NETREG_DUMMY; 7771 7772 /* NAPI wants this */ 7773 INIT_LIST_HEAD(&dev->napi_list); 7774 7775 /* a dummy interface is started by default */ 7776 set_bit(__LINK_STATE_PRESENT, &dev->state); 7777 set_bit(__LINK_STATE_START, &dev->state); 7778 7779 /* Note : We dont allocate pcpu_refcnt for dummy devices, 7780 * because users of this 'device' dont need to change 7781 * its refcount. 7782 */ 7783 7784 return 0; 7785 } 7786 EXPORT_SYMBOL_GPL(init_dummy_netdev); 7787 7788 7789 /** 7790 * register_netdev - register a network device 7791 * @dev: device to register 7792 * 7793 * Take a completed network device structure and add it to the kernel 7794 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 7795 * chain. 0 is returned on success. A negative errno code is returned 7796 * on a failure to set up the device, or if the name is a duplicate. 7797 * 7798 * This is a wrapper around register_netdevice that takes the rtnl semaphore 7799 * and expands the device name if you passed a format string to 7800 * alloc_netdev. 7801 */ 7802 int register_netdev(struct net_device *dev) 7803 { 7804 int err; 7805 7806 rtnl_lock(); 7807 err = register_netdevice(dev); 7808 rtnl_unlock(); 7809 return err; 7810 } 7811 EXPORT_SYMBOL(register_netdev); 7812 7813 int netdev_refcnt_read(const struct net_device *dev) 7814 { 7815 int i, refcnt = 0; 7816 7817 for_each_possible_cpu(i) 7818 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 7819 return refcnt; 7820 } 7821 EXPORT_SYMBOL(netdev_refcnt_read); 7822 7823 /** 7824 * netdev_wait_allrefs - wait until all references are gone. 7825 * @dev: target net_device 7826 * 7827 * This is called when unregistering network devices. 7828 * 7829 * Any protocol or device that holds a reference should register 7830 * for netdevice notification, and cleanup and put back the 7831 * reference if they receive an UNREGISTER event. 7832 * We can get stuck here if buggy protocols don't correctly 7833 * call dev_put. 7834 */ 7835 static void netdev_wait_allrefs(struct net_device *dev) 7836 { 7837 unsigned long rebroadcast_time, warning_time; 7838 int refcnt; 7839 7840 linkwatch_forget_dev(dev); 7841 7842 rebroadcast_time = warning_time = jiffies; 7843 refcnt = netdev_refcnt_read(dev); 7844 7845 while (refcnt != 0) { 7846 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 7847 rtnl_lock(); 7848 7849 /* Rebroadcast unregister notification */ 7850 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7851 7852 __rtnl_unlock(); 7853 rcu_barrier(); 7854 rtnl_lock(); 7855 7856 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7857 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 7858 &dev->state)) { 7859 /* We must not have linkwatch events 7860 * pending on unregister. If this 7861 * happens, we simply run the queue 7862 * unscheduled, resulting in a noop 7863 * for this device. 7864 */ 7865 linkwatch_run_queue(); 7866 } 7867 7868 __rtnl_unlock(); 7869 7870 rebroadcast_time = jiffies; 7871 } 7872 7873 msleep(250); 7874 7875 refcnt = netdev_refcnt_read(dev); 7876 7877 if (time_after(jiffies, warning_time + 10 * HZ)) { 7878 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 7879 dev->name, refcnt); 7880 warning_time = jiffies; 7881 } 7882 } 7883 } 7884 7885 /* The sequence is: 7886 * 7887 * rtnl_lock(); 7888 * ... 7889 * register_netdevice(x1); 7890 * register_netdevice(x2); 7891 * ... 7892 * unregister_netdevice(y1); 7893 * unregister_netdevice(y2); 7894 * ... 7895 * rtnl_unlock(); 7896 * free_netdev(y1); 7897 * free_netdev(y2); 7898 * 7899 * We are invoked by rtnl_unlock(). 7900 * This allows us to deal with problems: 7901 * 1) We can delete sysfs objects which invoke hotplug 7902 * without deadlocking with linkwatch via keventd. 7903 * 2) Since we run with the RTNL semaphore not held, we can sleep 7904 * safely in order to wait for the netdev refcnt to drop to zero. 7905 * 7906 * We must not return until all unregister events added during 7907 * the interval the lock was held have been completed. 7908 */ 7909 void netdev_run_todo(void) 7910 { 7911 struct list_head list; 7912 7913 /* Snapshot list, allow later requests */ 7914 list_replace_init(&net_todo_list, &list); 7915 7916 __rtnl_unlock(); 7917 7918 7919 /* Wait for rcu callbacks to finish before next phase */ 7920 if (!list_empty(&list)) 7921 rcu_barrier(); 7922 7923 while (!list_empty(&list)) { 7924 struct net_device *dev 7925 = list_first_entry(&list, struct net_device, todo_list); 7926 list_del(&dev->todo_list); 7927 7928 rtnl_lock(); 7929 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 7930 __rtnl_unlock(); 7931 7932 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 7933 pr_err("network todo '%s' but state %d\n", 7934 dev->name, dev->reg_state); 7935 dump_stack(); 7936 continue; 7937 } 7938 7939 dev->reg_state = NETREG_UNREGISTERED; 7940 7941 netdev_wait_allrefs(dev); 7942 7943 /* paranoia */ 7944 BUG_ON(netdev_refcnt_read(dev)); 7945 BUG_ON(!list_empty(&dev->ptype_all)); 7946 BUG_ON(!list_empty(&dev->ptype_specific)); 7947 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 7948 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 7949 WARN_ON(dev->dn_ptr); 7950 7951 if (dev->priv_destructor) 7952 dev->priv_destructor(dev); 7953 if (dev->needs_free_netdev) 7954 free_netdev(dev); 7955 7956 /* Report a network device has been unregistered */ 7957 rtnl_lock(); 7958 dev_net(dev)->dev_unreg_count--; 7959 __rtnl_unlock(); 7960 wake_up(&netdev_unregistering_wq); 7961 7962 /* Free network device */ 7963 kobject_put(&dev->dev.kobj); 7964 } 7965 } 7966 7967 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 7968 * all the same fields in the same order as net_device_stats, with only 7969 * the type differing, but rtnl_link_stats64 may have additional fields 7970 * at the end for newer counters. 7971 */ 7972 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 7973 const struct net_device_stats *netdev_stats) 7974 { 7975 #if BITS_PER_LONG == 64 7976 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 7977 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 7978 /* zero out counters that only exist in rtnl_link_stats64 */ 7979 memset((char *)stats64 + sizeof(*netdev_stats), 0, 7980 sizeof(*stats64) - sizeof(*netdev_stats)); 7981 #else 7982 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 7983 const unsigned long *src = (const unsigned long *)netdev_stats; 7984 u64 *dst = (u64 *)stats64; 7985 7986 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 7987 for (i = 0; i < n; i++) 7988 dst[i] = src[i]; 7989 /* zero out counters that only exist in rtnl_link_stats64 */ 7990 memset((char *)stats64 + n * sizeof(u64), 0, 7991 sizeof(*stats64) - n * sizeof(u64)); 7992 #endif 7993 } 7994 EXPORT_SYMBOL(netdev_stats_to_stats64); 7995 7996 /** 7997 * dev_get_stats - get network device statistics 7998 * @dev: device to get statistics from 7999 * @storage: place to store stats 8000 * 8001 * Get network statistics from device. Return @storage. 8002 * The device driver may provide its own method by setting 8003 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 8004 * otherwise the internal statistics structure is used. 8005 */ 8006 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 8007 struct rtnl_link_stats64 *storage) 8008 { 8009 const struct net_device_ops *ops = dev->netdev_ops; 8010 8011 if (ops->ndo_get_stats64) { 8012 memset(storage, 0, sizeof(*storage)); 8013 ops->ndo_get_stats64(dev, storage); 8014 } else if (ops->ndo_get_stats) { 8015 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 8016 } else { 8017 netdev_stats_to_stats64(storage, &dev->stats); 8018 } 8019 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 8020 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 8021 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 8022 return storage; 8023 } 8024 EXPORT_SYMBOL(dev_get_stats); 8025 8026 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 8027 { 8028 struct netdev_queue *queue = dev_ingress_queue(dev); 8029 8030 #ifdef CONFIG_NET_CLS_ACT 8031 if (queue) 8032 return queue; 8033 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 8034 if (!queue) 8035 return NULL; 8036 netdev_init_one_queue(dev, queue, NULL); 8037 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 8038 queue->qdisc_sleeping = &noop_qdisc; 8039 rcu_assign_pointer(dev->ingress_queue, queue); 8040 #endif 8041 return queue; 8042 } 8043 8044 static const struct ethtool_ops default_ethtool_ops; 8045 8046 void netdev_set_default_ethtool_ops(struct net_device *dev, 8047 const struct ethtool_ops *ops) 8048 { 8049 if (dev->ethtool_ops == &default_ethtool_ops) 8050 dev->ethtool_ops = ops; 8051 } 8052 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 8053 8054 void netdev_freemem(struct net_device *dev) 8055 { 8056 char *addr = (char *)dev - dev->padded; 8057 8058 kvfree(addr); 8059 } 8060 8061 /** 8062 * alloc_netdev_mqs - allocate network device 8063 * @sizeof_priv: size of private data to allocate space for 8064 * @name: device name format string 8065 * @name_assign_type: origin of device name 8066 * @setup: callback to initialize device 8067 * @txqs: the number of TX subqueues to allocate 8068 * @rxqs: the number of RX subqueues to allocate 8069 * 8070 * Allocates a struct net_device with private data area for driver use 8071 * and performs basic initialization. Also allocates subqueue structs 8072 * for each queue on the device. 8073 */ 8074 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 8075 unsigned char name_assign_type, 8076 void (*setup)(struct net_device *), 8077 unsigned int txqs, unsigned int rxqs) 8078 { 8079 struct net_device *dev; 8080 unsigned int alloc_size; 8081 struct net_device *p; 8082 8083 BUG_ON(strlen(name) >= sizeof(dev->name)); 8084 8085 if (txqs < 1) { 8086 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 8087 return NULL; 8088 } 8089 8090 #ifdef CONFIG_SYSFS 8091 if (rxqs < 1) { 8092 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 8093 return NULL; 8094 } 8095 #endif 8096 8097 alloc_size = sizeof(struct net_device); 8098 if (sizeof_priv) { 8099 /* ensure 32-byte alignment of private area */ 8100 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 8101 alloc_size += sizeof_priv; 8102 } 8103 /* ensure 32-byte alignment of whole construct */ 8104 alloc_size += NETDEV_ALIGN - 1; 8105 8106 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 8107 if (!p) 8108 return NULL; 8109 8110 dev = PTR_ALIGN(p, NETDEV_ALIGN); 8111 dev->padded = (char *)dev - (char *)p; 8112 8113 dev->pcpu_refcnt = alloc_percpu(int); 8114 if (!dev->pcpu_refcnt) 8115 goto free_dev; 8116 8117 if (dev_addr_init(dev)) 8118 goto free_pcpu; 8119 8120 dev_mc_init(dev); 8121 dev_uc_init(dev); 8122 8123 dev_net_set(dev, &init_net); 8124 8125 dev->gso_max_size = GSO_MAX_SIZE; 8126 dev->gso_max_segs = GSO_MAX_SEGS; 8127 8128 INIT_LIST_HEAD(&dev->napi_list); 8129 INIT_LIST_HEAD(&dev->unreg_list); 8130 INIT_LIST_HEAD(&dev->close_list); 8131 INIT_LIST_HEAD(&dev->link_watch_list); 8132 INIT_LIST_HEAD(&dev->adj_list.upper); 8133 INIT_LIST_HEAD(&dev->adj_list.lower); 8134 INIT_LIST_HEAD(&dev->ptype_all); 8135 INIT_LIST_HEAD(&dev->ptype_specific); 8136 #ifdef CONFIG_NET_SCHED 8137 hash_init(dev->qdisc_hash); 8138 #endif 8139 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 8140 setup(dev); 8141 8142 if (!dev->tx_queue_len) { 8143 dev->priv_flags |= IFF_NO_QUEUE; 8144 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 8145 } 8146 8147 dev->num_tx_queues = txqs; 8148 dev->real_num_tx_queues = txqs; 8149 if (netif_alloc_netdev_queues(dev)) 8150 goto free_all; 8151 8152 #ifdef CONFIG_SYSFS 8153 dev->num_rx_queues = rxqs; 8154 dev->real_num_rx_queues = rxqs; 8155 if (netif_alloc_rx_queues(dev)) 8156 goto free_all; 8157 #endif 8158 8159 strcpy(dev->name, name); 8160 dev->name_assign_type = name_assign_type; 8161 dev->group = INIT_NETDEV_GROUP; 8162 if (!dev->ethtool_ops) 8163 dev->ethtool_ops = &default_ethtool_ops; 8164 8165 nf_hook_ingress_init(dev); 8166 8167 return dev; 8168 8169 free_all: 8170 free_netdev(dev); 8171 return NULL; 8172 8173 free_pcpu: 8174 free_percpu(dev->pcpu_refcnt); 8175 free_dev: 8176 netdev_freemem(dev); 8177 return NULL; 8178 } 8179 EXPORT_SYMBOL(alloc_netdev_mqs); 8180 8181 /** 8182 * free_netdev - free network device 8183 * @dev: device 8184 * 8185 * This function does the last stage of destroying an allocated device 8186 * interface. The reference to the device object is released. If this 8187 * is the last reference then it will be freed.Must be called in process 8188 * context. 8189 */ 8190 void free_netdev(struct net_device *dev) 8191 { 8192 struct napi_struct *p, *n; 8193 struct bpf_prog *prog; 8194 8195 might_sleep(); 8196 netif_free_tx_queues(dev); 8197 #ifdef CONFIG_SYSFS 8198 kvfree(dev->_rx); 8199 #endif 8200 8201 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 8202 8203 /* Flush device addresses */ 8204 dev_addr_flush(dev); 8205 8206 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 8207 netif_napi_del(p); 8208 8209 free_percpu(dev->pcpu_refcnt); 8210 dev->pcpu_refcnt = NULL; 8211 8212 prog = rcu_dereference_protected(dev->xdp_prog, 1); 8213 if (prog) { 8214 bpf_prog_put(prog); 8215 static_key_slow_dec(&generic_xdp_needed); 8216 } 8217 8218 /* Compatibility with error handling in drivers */ 8219 if (dev->reg_state == NETREG_UNINITIALIZED) { 8220 netdev_freemem(dev); 8221 return; 8222 } 8223 8224 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 8225 dev->reg_state = NETREG_RELEASED; 8226 8227 /* will free via device release */ 8228 put_device(&dev->dev); 8229 } 8230 EXPORT_SYMBOL(free_netdev); 8231 8232 /** 8233 * synchronize_net - Synchronize with packet receive processing 8234 * 8235 * Wait for packets currently being received to be done. 8236 * Does not block later packets from starting. 8237 */ 8238 void synchronize_net(void) 8239 { 8240 might_sleep(); 8241 if (rtnl_is_locked()) 8242 synchronize_rcu_expedited(); 8243 else 8244 synchronize_rcu(); 8245 } 8246 EXPORT_SYMBOL(synchronize_net); 8247 8248 /** 8249 * unregister_netdevice_queue - remove device from the kernel 8250 * @dev: device 8251 * @head: list 8252 * 8253 * This function shuts down a device interface and removes it 8254 * from the kernel tables. 8255 * If head not NULL, device is queued to be unregistered later. 8256 * 8257 * Callers must hold the rtnl semaphore. You may want 8258 * unregister_netdev() instead of this. 8259 */ 8260 8261 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 8262 { 8263 ASSERT_RTNL(); 8264 8265 if (head) { 8266 list_move_tail(&dev->unreg_list, head); 8267 } else { 8268 rollback_registered(dev); 8269 /* Finish processing unregister after unlock */ 8270 net_set_todo(dev); 8271 } 8272 } 8273 EXPORT_SYMBOL(unregister_netdevice_queue); 8274 8275 /** 8276 * unregister_netdevice_many - unregister many devices 8277 * @head: list of devices 8278 * 8279 * Note: As most callers use a stack allocated list_head, 8280 * we force a list_del() to make sure stack wont be corrupted later. 8281 */ 8282 void unregister_netdevice_many(struct list_head *head) 8283 { 8284 struct net_device *dev; 8285 8286 if (!list_empty(head)) { 8287 rollback_registered_many(head); 8288 list_for_each_entry(dev, head, unreg_list) 8289 net_set_todo(dev); 8290 list_del(head); 8291 } 8292 } 8293 EXPORT_SYMBOL(unregister_netdevice_many); 8294 8295 /** 8296 * unregister_netdev - remove device from the kernel 8297 * @dev: device 8298 * 8299 * This function shuts down a device interface and removes it 8300 * from the kernel tables. 8301 * 8302 * This is just a wrapper for unregister_netdevice that takes 8303 * the rtnl semaphore. In general you want to use this and not 8304 * unregister_netdevice. 8305 */ 8306 void unregister_netdev(struct net_device *dev) 8307 { 8308 rtnl_lock(); 8309 unregister_netdevice(dev); 8310 rtnl_unlock(); 8311 } 8312 EXPORT_SYMBOL(unregister_netdev); 8313 8314 /** 8315 * dev_change_net_namespace - move device to different nethost namespace 8316 * @dev: device 8317 * @net: network namespace 8318 * @pat: If not NULL name pattern to try if the current device name 8319 * is already taken in the destination network namespace. 8320 * 8321 * This function shuts down a device interface and moves it 8322 * to a new network namespace. On success 0 is returned, on 8323 * a failure a netagive errno code is returned. 8324 * 8325 * Callers must hold the rtnl semaphore. 8326 */ 8327 8328 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 8329 { 8330 int err, new_nsid; 8331 8332 ASSERT_RTNL(); 8333 8334 /* Don't allow namespace local devices to be moved. */ 8335 err = -EINVAL; 8336 if (dev->features & NETIF_F_NETNS_LOCAL) 8337 goto out; 8338 8339 /* Ensure the device has been registrered */ 8340 if (dev->reg_state != NETREG_REGISTERED) 8341 goto out; 8342 8343 /* Get out if there is nothing todo */ 8344 err = 0; 8345 if (net_eq(dev_net(dev), net)) 8346 goto out; 8347 8348 /* Pick the destination device name, and ensure 8349 * we can use it in the destination network namespace. 8350 */ 8351 err = -EEXIST; 8352 if (__dev_get_by_name(net, dev->name)) { 8353 /* We get here if we can't use the current device name */ 8354 if (!pat) 8355 goto out; 8356 if (dev_get_valid_name(net, dev, pat) < 0) 8357 goto out; 8358 } 8359 8360 /* 8361 * And now a mini version of register_netdevice unregister_netdevice. 8362 */ 8363 8364 /* If device is running close it first. */ 8365 dev_close(dev); 8366 8367 /* And unlink it from device chain */ 8368 err = -ENODEV; 8369 unlist_netdevice(dev); 8370 8371 synchronize_net(); 8372 8373 /* Shutdown queueing discipline. */ 8374 dev_shutdown(dev); 8375 8376 /* Notify protocols, that we are about to destroy 8377 * this device. They should clean all the things. 8378 * 8379 * Note that dev->reg_state stays at NETREG_REGISTERED. 8380 * This is wanted because this way 8021q and macvlan know 8381 * the device is just moving and can keep their slaves up. 8382 */ 8383 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 8384 rcu_barrier(); 8385 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); 8386 if (dev->rtnl_link_ops && dev->rtnl_link_ops->get_link_net) 8387 new_nsid = peernet2id_alloc(dev_net(dev), net); 8388 else 8389 new_nsid = peernet2id(dev_net(dev), net); 8390 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid); 8391 8392 /* 8393 * Flush the unicast and multicast chains 8394 */ 8395 dev_uc_flush(dev); 8396 dev_mc_flush(dev); 8397 8398 /* Send a netdev-removed uevent to the old namespace */ 8399 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 8400 netdev_adjacent_del_links(dev); 8401 8402 /* Actually switch the network namespace */ 8403 dev_net_set(dev, net); 8404 8405 /* If there is an ifindex conflict assign a new one */ 8406 if (__dev_get_by_index(net, dev->ifindex)) 8407 dev->ifindex = dev_new_index(net); 8408 8409 /* Send a netdev-add uevent to the new namespace */ 8410 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 8411 netdev_adjacent_add_links(dev); 8412 8413 /* Fixup kobjects */ 8414 err = device_rename(&dev->dev, dev->name); 8415 WARN_ON(err); 8416 8417 /* Add the device back in the hashes */ 8418 list_netdevice(dev); 8419 8420 /* Notify protocols, that a new device appeared. */ 8421 call_netdevice_notifiers(NETDEV_REGISTER, dev); 8422 8423 /* 8424 * Prevent userspace races by waiting until the network 8425 * device is fully setup before sending notifications. 8426 */ 8427 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 8428 8429 synchronize_net(); 8430 err = 0; 8431 out: 8432 return err; 8433 } 8434 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 8435 8436 static int dev_cpu_dead(unsigned int oldcpu) 8437 { 8438 struct sk_buff **list_skb; 8439 struct sk_buff *skb; 8440 unsigned int cpu; 8441 struct softnet_data *sd, *oldsd, *remsd = NULL; 8442 8443 local_irq_disable(); 8444 cpu = smp_processor_id(); 8445 sd = &per_cpu(softnet_data, cpu); 8446 oldsd = &per_cpu(softnet_data, oldcpu); 8447 8448 /* Find end of our completion_queue. */ 8449 list_skb = &sd->completion_queue; 8450 while (*list_skb) 8451 list_skb = &(*list_skb)->next; 8452 /* Append completion queue from offline CPU. */ 8453 *list_skb = oldsd->completion_queue; 8454 oldsd->completion_queue = NULL; 8455 8456 /* Append output queue from offline CPU. */ 8457 if (oldsd->output_queue) { 8458 *sd->output_queue_tailp = oldsd->output_queue; 8459 sd->output_queue_tailp = oldsd->output_queue_tailp; 8460 oldsd->output_queue = NULL; 8461 oldsd->output_queue_tailp = &oldsd->output_queue; 8462 } 8463 /* Append NAPI poll list from offline CPU, with one exception : 8464 * process_backlog() must be called by cpu owning percpu backlog. 8465 * We properly handle process_queue & input_pkt_queue later. 8466 */ 8467 while (!list_empty(&oldsd->poll_list)) { 8468 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 8469 struct napi_struct, 8470 poll_list); 8471 8472 list_del_init(&napi->poll_list); 8473 if (napi->poll == process_backlog) 8474 napi->state = 0; 8475 else 8476 ____napi_schedule(sd, napi); 8477 } 8478 8479 raise_softirq_irqoff(NET_TX_SOFTIRQ); 8480 local_irq_enable(); 8481 8482 #ifdef CONFIG_RPS 8483 remsd = oldsd->rps_ipi_list; 8484 oldsd->rps_ipi_list = NULL; 8485 #endif 8486 /* send out pending IPI's on offline CPU */ 8487 net_rps_send_ipi(remsd); 8488 8489 /* Process offline CPU's input_pkt_queue */ 8490 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 8491 netif_rx_ni(skb); 8492 input_queue_head_incr(oldsd); 8493 } 8494 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 8495 netif_rx_ni(skb); 8496 input_queue_head_incr(oldsd); 8497 } 8498 8499 return 0; 8500 } 8501 8502 /** 8503 * netdev_increment_features - increment feature set by one 8504 * @all: current feature set 8505 * @one: new feature set 8506 * @mask: mask feature set 8507 * 8508 * Computes a new feature set after adding a device with feature set 8509 * @one to the master device with current feature set @all. Will not 8510 * enable anything that is off in @mask. Returns the new feature set. 8511 */ 8512 netdev_features_t netdev_increment_features(netdev_features_t all, 8513 netdev_features_t one, netdev_features_t mask) 8514 { 8515 if (mask & NETIF_F_HW_CSUM) 8516 mask |= NETIF_F_CSUM_MASK; 8517 mask |= NETIF_F_VLAN_CHALLENGED; 8518 8519 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 8520 all &= one | ~NETIF_F_ALL_FOR_ALL; 8521 8522 /* If one device supports hw checksumming, set for all. */ 8523 if (all & NETIF_F_HW_CSUM) 8524 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 8525 8526 return all; 8527 } 8528 EXPORT_SYMBOL(netdev_increment_features); 8529 8530 static struct hlist_head * __net_init netdev_create_hash(void) 8531 { 8532 int i; 8533 struct hlist_head *hash; 8534 8535 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); 8536 if (hash != NULL) 8537 for (i = 0; i < NETDEV_HASHENTRIES; i++) 8538 INIT_HLIST_HEAD(&hash[i]); 8539 8540 return hash; 8541 } 8542 8543 /* Initialize per network namespace state */ 8544 static int __net_init netdev_init(struct net *net) 8545 { 8546 if (net != &init_net) 8547 INIT_LIST_HEAD(&net->dev_base_head); 8548 8549 net->dev_name_head = netdev_create_hash(); 8550 if (net->dev_name_head == NULL) 8551 goto err_name; 8552 8553 net->dev_index_head = netdev_create_hash(); 8554 if (net->dev_index_head == NULL) 8555 goto err_idx; 8556 8557 return 0; 8558 8559 err_idx: 8560 kfree(net->dev_name_head); 8561 err_name: 8562 return -ENOMEM; 8563 } 8564 8565 /** 8566 * netdev_drivername - network driver for the device 8567 * @dev: network device 8568 * 8569 * Determine network driver for device. 8570 */ 8571 const char *netdev_drivername(const struct net_device *dev) 8572 { 8573 const struct device_driver *driver; 8574 const struct device *parent; 8575 const char *empty = ""; 8576 8577 parent = dev->dev.parent; 8578 if (!parent) 8579 return empty; 8580 8581 driver = parent->driver; 8582 if (driver && driver->name) 8583 return driver->name; 8584 return empty; 8585 } 8586 8587 static void __netdev_printk(const char *level, const struct net_device *dev, 8588 struct va_format *vaf) 8589 { 8590 if (dev && dev->dev.parent) { 8591 dev_printk_emit(level[1] - '0', 8592 dev->dev.parent, 8593 "%s %s %s%s: %pV", 8594 dev_driver_string(dev->dev.parent), 8595 dev_name(dev->dev.parent), 8596 netdev_name(dev), netdev_reg_state(dev), 8597 vaf); 8598 } else if (dev) { 8599 printk("%s%s%s: %pV", 8600 level, netdev_name(dev), netdev_reg_state(dev), vaf); 8601 } else { 8602 printk("%s(NULL net_device): %pV", level, vaf); 8603 } 8604 } 8605 8606 void netdev_printk(const char *level, const struct net_device *dev, 8607 const char *format, ...) 8608 { 8609 struct va_format vaf; 8610 va_list args; 8611 8612 va_start(args, format); 8613 8614 vaf.fmt = format; 8615 vaf.va = &args; 8616 8617 __netdev_printk(level, dev, &vaf); 8618 8619 va_end(args); 8620 } 8621 EXPORT_SYMBOL(netdev_printk); 8622 8623 #define define_netdev_printk_level(func, level) \ 8624 void func(const struct net_device *dev, const char *fmt, ...) \ 8625 { \ 8626 struct va_format vaf; \ 8627 va_list args; \ 8628 \ 8629 va_start(args, fmt); \ 8630 \ 8631 vaf.fmt = fmt; \ 8632 vaf.va = &args; \ 8633 \ 8634 __netdev_printk(level, dev, &vaf); \ 8635 \ 8636 va_end(args); \ 8637 } \ 8638 EXPORT_SYMBOL(func); 8639 8640 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 8641 define_netdev_printk_level(netdev_alert, KERN_ALERT); 8642 define_netdev_printk_level(netdev_crit, KERN_CRIT); 8643 define_netdev_printk_level(netdev_err, KERN_ERR); 8644 define_netdev_printk_level(netdev_warn, KERN_WARNING); 8645 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 8646 define_netdev_printk_level(netdev_info, KERN_INFO); 8647 8648 static void __net_exit netdev_exit(struct net *net) 8649 { 8650 kfree(net->dev_name_head); 8651 kfree(net->dev_index_head); 8652 if (net != &init_net) 8653 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 8654 } 8655 8656 static struct pernet_operations __net_initdata netdev_net_ops = { 8657 .init = netdev_init, 8658 .exit = netdev_exit, 8659 }; 8660 8661 static void __net_exit default_device_exit(struct net *net) 8662 { 8663 struct net_device *dev, *aux; 8664 /* 8665 * Push all migratable network devices back to the 8666 * initial network namespace 8667 */ 8668 rtnl_lock(); 8669 for_each_netdev_safe(net, dev, aux) { 8670 int err; 8671 char fb_name[IFNAMSIZ]; 8672 8673 /* Ignore unmoveable devices (i.e. loopback) */ 8674 if (dev->features & NETIF_F_NETNS_LOCAL) 8675 continue; 8676 8677 /* Leave virtual devices for the generic cleanup */ 8678 if (dev->rtnl_link_ops) 8679 continue; 8680 8681 /* Push remaining network devices to init_net */ 8682 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 8683 err = dev_change_net_namespace(dev, &init_net, fb_name); 8684 if (err) { 8685 pr_emerg("%s: failed to move %s to init_net: %d\n", 8686 __func__, dev->name, err); 8687 BUG(); 8688 } 8689 } 8690 rtnl_unlock(); 8691 } 8692 8693 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 8694 { 8695 /* Return with the rtnl_lock held when there are no network 8696 * devices unregistering in any network namespace in net_list. 8697 */ 8698 struct net *net; 8699 bool unregistering; 8700 DEFINE_WAIT_FUNC(wait, woken_wake_function); 8701 8702 add_wait_queue(&netdev_unregistering_wq, &wait); 8703 for (;;) { 8704 unregistering = false; 8705 rtnl_lock(); 8706 list_for_each_entry(net, net_list, exit_list) { 8707 if (net->dev_unreg_count > 0) { 8708 unregistering = true; 8709 break; 8710 } 8711 } 8712 if (!unregistering) 8713 break; 8714 __rtnl_unlock(); 8715 8716 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 8717 } 8718 remove_wait_queue(&netdev_unregistering_wq, &wait); 8719 } 8720 8721 static void __net_exit default_device_exit_batch(struct list_head *net_list) 8722 { 8723 /* At exit all network devices most be removed from a network 8724 * namespace. Do this in the reverse order of registration. 8725 * Do this across as many network namespaces as possible to 8726 * improve batching efficiency. 8727 */ 8728 struct net_device *dev; 8729 struct net *net; 8730 LIST_HEAD(dev_kill_list); 8731 8732 /* To prevent network device cleanup code from dereferencing 8733 * loopback devices or network devices that have been freed 8734 * wait here for all pending unregistrations to complete, 8735 * before unregistring the loopback device and allowing the 8736 * network namespace be freed. 8737 * 8738 * The netdev todo list containing all network devices 8739 * unregistrations that happen in default_device_exit_batch 8740 * will run in the rtnl_unlock() at the end of 8741 * default_device_exit_batch. 8742 */ 8743 rtnl_lock_unregistering(net_list); 8744 list_for_each_entry(net, net_list, exit_list) { 8745 for_each_netdev_reverse(net, dev) { 8746 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 8747 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 8748 else 8749 unregister_netdevice_queue(dev, &dev_kill_list); 8750 } 8751 } 8752 unregister_netdevice_many(&dev_kill_list); 8753 rtnl_unlock(); 8754 } 8755 8756 static struct pernet_operations __net_initdata default_device_ops = { 8757 .exit = default_device_exit, 8758 .exit_batch = default_device_exit_batch, 8759 }; 8760 8761 /* 8762 * Initialize the DEV module. At boot time this walks the device list and 8763 * unhooks any devices that fail to initialise (normally hardware not 8764 * present) and leaves us with a valid list of present and active devices. 8765 * 8766 */ 8767 8768 /* 8769 * This is called single threaded during boot, so no need 8770 * to take the rtnl semaphore. 8771 */ 8772 static int __init net_dev_init(void) 8773 { 8774 int i, rc = -ENOMEM; 8775 8776 BUG_ON(!dev_boot_phase); 8777 8778 if (dev_proc_init()) 8779 goto out; 8780 8781 if (netdev_kobject_init()) 8782 goto out; 8783 8784 INIT_LIST_HEAD(&ptype_all); 8785 for (i = 0; i < PTYPE_HASH_SIZE; i++) 8786 INIT_LIST_HEAD(&ptype_base[i]); 8787 8788 INIT_LIST_HEAD(&offload_base); 8789 8790 if (register_pernet_subsys(&netdev_net_ops)) 8791 goto out; 8792 8793 /* 8794 * Initialise the packet receive queues. 8795 */ 8796 8797 for_each_possible_cpu(i) { 8798 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 8799 struct softnet_data *sd = &per_cpu(softnet_data, i); 8800 8801 INIT_WORK(flush, flush_backlog); 8802 8803 skb_queue_head_init(&sd->input_pkt_queue); 8804 skb_queue_head_init(&sd->process_queue); 8805 INIT_LIST_HEAD(&sd->poll_list); 8806 sd->output_queue_tailp = &sd->output_queue; 8807 #ifdef CONFIG_RPS 8808 sd->csd.func = rps_trigger_softirq; 8809 sd->csd.info = sd; 8810 sd->cpu = i; 8811 #endif 8812 8813 sd->backlog.poll = process_backlog; 8814 sd->backlog.weight = weight_p; 8815 } 8816 8817 dev_boot_phase = 0; 8818 8819 /* The loopback device is special if any other network devices 8820 * is present in a network namespace the loopback device must 8821 * be present. Since we now dynamically allocate and free the 8822 * loopback device ensure this invariant is maintained by 8823 * keeping the loopback device as the first device on the 8824 * list of network devices. Ensuring the loopback devices 8825 * is the first device that appears and the last network device 8826 * that disappears. 8827 */ 8828 if (register_pernet_device(&loopback_net_ops)) 8829 goto out; 8830 8831 if (register_pernet_device(&default_device_ops)) 8832 goto out; 8833 8834 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 8835 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 8836 8837 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 8838 NULL, dev_cpu_dead); 8839 WARN_ON(rc < 0); 8840 rc = 0; 8841 out: 8842 return rc; 8843 } 8844 8845 subsys_initcall(net_dev_init); 8846