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/skbuff.h> 97 #include <linux/bpf.h> 98 #include <linux/bpf_trace.h> 99 #include <net/net_namespace.h> 100 #include <net/sock.h> 101 #include <net/busy_poll.h> 102 #include <linux/rtnetlink.h> 103 #include <linux/stat.h> 104 #include <net/dst.h> 105 #include <net/dst_metadata.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <linux/pci.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_ingress.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 149 #include "net-sysfs.h" 150 151 #define MAX_GRO_SKBS 8 152 153 /* This should be increased if a protocol with a bigger head is added. */ 154 #define GRO_MAX_HEAD (MAX_HEADER + 128) 155 156 static DEFINE_SPINLOCK(ptype_lock); 157 static DEFINE_SPINLOCK(offload_lock); 158 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 159 struct list_head ptype_all __read_mostly; /* Taps */ 160 static struct list_head offload_base __read_mostly; 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_info(unsigned long val, 164 struct netdev_notifier_info *info); 165 static struct napi_struct *napi_by_id(unsigned int napi_id); 166 167 /* 168 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 169 * semaphore. 170 * 171 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 172 * 173 * Writers must hold the rtnl semaphore while they loop through the 174 * dev_base_head list, and hold dev_base_lock for writing when they do the 175 * actual updates. This allows pure readers to access the list even 176 * while a writer is preparing to update it. 177 * 178 * To put it another way, dev_base_lock is held for writing only to 179 * protect against pure readers; the rtnl semaphore provides the 180 * protection against other writers. 181 * 182 * See, for example usages, register_netdevice() and 183 * unregister_netdevice(), which must be called with the rtnl 184 * semaphore held. 185 */ 186 DEFINE_RWLOCK(dev_base_lock); 187 EXPORT_SYMBOL(dev_base_lock); 188 189 static DEFINE_MUTEX(ifalias_mutex); 190 191 /* protects napi_hash addition/deletion and napi_gen_id */ 192 static DEFINE_SPINLOCK(napi_hash_lock); 193 194 static unsigned int napi_gen_id = NR_CPUS; 195 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 196 197 static seqcount_t devnet_rename_seq; 198 199 static inline void dev_base_seq_inc(struct net *net) 200 { 201 while (++net->dev_base_seq == 0) 202 ; 203 } 204 205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 206 { 207 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 208 209 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 210 } 211 212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 213 { 214 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 215 } 216 217 static inline void rps_lock(struct softnet_data *sd) 218 { 219 #ifdef CONFIG_RPS 220 spin_lock(&sd->input_pkt_queue.lock); 221 #endif 222 } 223 224 static inline void rps_unlock(struct softnet_data *sd) 225 { 226 #ifdef CONFIG_RPS 227 spin_unlock(&sd->input_pkt_queue.lock); 228 #endif 229 } 230 231 /* Device list insertion */ 232 static void list_netdevice(struct net_device *dev) 233 { 234 struct net *net = dev_net(dev); 235 236 ASSERT_RTNL(); 237 238 write_lock_bh(&dev_base_lock); 239 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 240 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 241 hlist_add_head_rcu(&dev->index_hlist, 242 dev_index_hash(net, dev->ifindex)); 243 write_unlock_bh(&dev_base_lock); 244 245 dev_base_seq_inc(net); 246 } 247 248 /* Device list removal 249 * caller must respect a RCU grace period before freeing/reusing dev 250 */ 251 static void unlist_netdevice(struct net_device *dev) 252 { 253 ASSERT_RTNL(); 254 255 /* Unlink dev from the device chain */ 256 write_lock_bh(&dev_base_lock); 257 list_del_rcu(&dev->dev_list); 258 hlist_del_rcu(&dev->name_hlist); 259 hlist_del_rcu(&dev->index_hlist); 260 write_unlock_bh(&dev_base_lock); 261 262 dev_base_seq_inc(dev_net(dev)); 263 } 264 265 /* 266 * Our notifier list 267 */ 268 269 static RAW_NOTIFIER_HEAD(netdev_chain); 270 271 /* 272 * Device drivers call our routines to queue packets here. We empty the 273 * queue in the local softnet handler. 274 */ 275 276 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 277 EXPORT_PER_CPU_SYMBOL(softnet_data); 278 279 #ifdef CONFIG_LOCKDEP 280 /* 281 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 282 * according to dev->type 283 */ 284 static const unsigned short netdev_lock_type[] = { 285 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 286 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 287 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 288 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 289 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 290 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 291 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 292 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 293 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 294 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 295 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 296 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 297 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 298 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 299 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 300 301 static const char *const netdev_lock_name[] = { 302 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 303 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 304 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 305 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 306 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 307 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 308 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 309 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 310 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 311 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 312 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 313 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 314 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 315 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 316 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 317 318 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 319 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 320 321 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 322 { 323 int i; 324 325 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 326 if (netdev_lock_type[i] == dev_type) 327 return i; 328 /* the last key is used by default */ 329 return ARRAY_SIZE(netdev_lock_type) - 1; 330 } 331 332 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 333 unsigned short dev_type) 334 { 335 int i; 336 337 i = netdev_lock_pos(dev_type); 338 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 339 netdev_lock_name[i]); 340 } 341 342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 343 { 344 int i; 345 346 i = netdev_lock_pos(dev->type); 347 lockdep_set_class_and_name(&dev->addr_list_lock, 348 &netdev_addr_lock_key[i], 349 netdev_lock_name[i]); 350 } 351 #else 352 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 353 unsigned short dev_type) 354 { 355 } 356 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 357 { 358 } 359 #endif 360 361 /******************************************************************************* 362 * 363 * Protocol management and registration routines 364 * 365 *******************************************************************************/ 366 367 368 /* 369 * Add a protocol ID to the list. Now that the input handler is 370 * smarter we can dispense with all the messy stuff that used to be 371 * here. 372 * 373 * BEWARE!!! Protocol handlers, mangling input packets, 374 * MUST BE last in hash buckets and checking protocol handlers 375 * MUST start from promiscuous ptype_all chain in net_bh. 376 * It is true now, do not change it. 377 * Explanation follows: if protocol handler, mangling packet, will 378 * be the first on list, it is not able to sense, that packet 379 * is cloned and should be copied-on-write, so that it will 380 * change it and subsequent readers will get broken packet. 381 * --ANK (980803) 382 */ 383 384 static inline struct list_head *ptype_head(const struct packet_type *pt) 385 { 386 if (pt->type == htons(ETH_P_ALL)) 387 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 388 else 389 return pt->dev ? &pt->dev->ptype_specific : 390 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 391 } 392 393 /** 394 * dev_add_pack - add packet handler 395 * @pt: packet type declaration 396 * 397 * Add a protocol handler to the networking stack. The passed &packet_type 398 * is linked into kernel lists and may not be freed until it has been 399 * removed from the kernel lists. 400 * 401 * This call does not sleep therefore it can not 402 * guarantee all CPU's that are in middle of receiving packets 403 * will see the new packet type (until the next received packet). 404 */ 405 406 void dev_add_pack(struct packet_type *pt) 407 { 408 struct list_head *head = ptype_head(pt); 409 410 spin_lock(&ptype_lock); 411 list_add_rcu(&pt->list, head); 412 spin_unlock(&ptype_lock); 413 } 414 EXPORT_SYMBOL(dev_add_pack); 415 416 /** 417 * __dev_remove_pack - remove packet handler 418 * @pt: packet type declaration 419 * 420 * Remove a protocol handler that was previously added to the kernel 421 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 422 * from the kernel lists and can be freed or reused once this function 423 * returns. 424 * 425 * The packet type might still be in use by receivers 426 * and must not be freed until after all the CPU's have gone 427 * through a quiescent state. 428 */ 429 void __dev_remove_pack(struct packet_type *pt) 430 { 431 struct list_head *head = ptype_head(pt); 432 struct packet_type *pt1; 433 434 spin_lock(&ptype_lock); 435 436 list_for_each_entry(pt1, head, list) { 437 if (pt == pt1) { 438 list_del_rcu(&pt->list); 439 goto out; 440 } 441 } 442 443 pr_warn("dev_remove_pack: %p not found\n", pt); 444 out: 445 spin_unlock(&ptype_lock); 446 } 447 EXPORT_SYMBOL(__dev_remove_pack); 448 449 /** 450 * dev_remove_pack - remove packet handler 451 * @pt: packet type declaration 452 * 453 * Remove a protocol handler that was previously added to the kernel 454 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 455 * from the kernel lists and can be freed or reused once this function 456 * returns. 457 * 458 * This call sleeps to guarantee that no CPU is looking at the packet 459 * type after return. 460 */ 461 void dev_remove_pack(struct packet_type *pt) 462 { 463 __dev_remove_pack(pt); 464 465 synchronize_net(); 466 } 467 EXPORT_SYMBOL(dev_remove_pack); 468 469 470 /** 471 * dev_add_offload - register offload handlers 472 * @po: protocol offload declaration 473 * 474 * Add protocol offload handlers to the networking stack. The passed 475 * &proto_offload is linked into kernel lists and may not be freed until 476 * it has been removed from the kernel lists. 477 * 478 * This call does not sleep therefore it can not 479 * guarantee all CPU's that are in middle of receiving packets 480 * will see the new offload handlers (until the next received packet). 481 */ 482 void dev_add_offload(struct packet_offload *po) 483 { 484 struct packet_offload *elem; 485 486 spin_lock(&offload_lock); 487 list_for_each_entry(elem, &offload_base, list) { 488 if (po->priority < elem->priority) 489 break; 490 } 491 list_add_rcu(&po->list, elem->list.prev); 492 spin_unlock(&offload_lock); 493 } 494 EXPORT_SYMBOL(dev_add_offload); 495 496 /** 497 * __dev_remove_offload - remove offload handler 498 * @po: packet offload declaration 499 * 500 * Remove a protocol offload handler that was previously added to the 501 * kernel offload handlers by dev_add_offload(). The passed &offload_type 502 * is removed from the kernel lists and can be freed or reused once this 503 * function returns. 504 * 505 * The packet type might still be in use by receivers 506 * and must not be freed until after all the CPU's have gone 507 * through a quiescent state. 508 */ 509 static void __dev_remove_offload(struct packet_offload *po) 510 { 511 struct list_head *head = &offload_base; 512 struct packet_offload *po1; 513 514 spin_lock(&offload_lock); 515 516 list_for_each_entry(po1, head, list) { 517 if (po == po1) { 518 list_del_rcu(&po->list); 519 goto out; 520 } 521 } 522 523 pr_warn("dev_remove_offload: %p not found\n", po); 524 out: 525 spin_unlock(&offload_lock); 526 } 527 528 /** 529 * dev_remove_offload - remove packet offload handler 530 * @po: packet offload declaration 531 * 532 * Remove a packet offload handler that was previously added to the kernel 533 * offload handlers by dev_add_offload(). The passed &offload_type is 534 * removed from the kernel lists and can be freed or reused once this 535 * function returns. 536 * 537 * This call sleeps to guarantee that no CPU is looking at the packet 538 * type after return. 539 */ 540 void dev_remove_offload(struct packet_offload *po) 541 { 542 __dev_remove_offload(po); 543 544 synchronize_net(); 545 } 546 EXPORT_SYMBOL(dev_remove_offload); 547 548 /****************************************************************************** 549 * 550 * Device Boot-time Settings Routines 551 * 552 ******************************************************************************/ 553 554 /* Boot time configuration table */ 555 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 556 557 /** 558 * netdev_boot_setup_add - add new setup entry 559 * @name: name of the device 560 * @map: configured settings for the device 561 * 562 * Adds new setup entry to the dev_boot_setup list. The function 563 * returns 0 on error and 1 on success. This is a generic routine to 564 * all netdevices. 565 */ 566 static int netdev_boot_setup_add(char *name, struct ifmap *map) 567 { 568 struct netdev_boot_setup *s; 569 int i; 570 571 s = dev_boot_setup; 572 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 573 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 574 memset(s[i].name, 0, sizeof(s[i].name)); 575 strlcpy(s[i].name, name, IFNAMSIZ); 576 memcpy(&s[i].map, map, sizeof(s[i].map)); 577 break; 578 } 579 } 580 581 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 582 } 583 584 /** 585 * netdev_boot_setup_check - check boot time settings 586 * @dev: the netdevice 587 * 588 * Check boot time settings for the device. 589 * The found settings are set for the device to be used 590 * later in the device probing. 591 * Returns 0 if no settings found, 1 if they are. 592 */ 593 int netdev_boot_setup_check(struct net_device *dev) 594 { 595 struct netdev_boot_setup *s = dev_boot_setup; 596 int i; 597 598 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 599 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 600 !strcmp(dev->name, s[i].name)) { 601 dev->irq = s[i].map.irq; 602 dev->base_addr = s[i].map.base_addr; 603 dev->mem_start = s[i].map.mem_start; 604 dev->mem_end = s[i].map.mem_end; 605 return 1; 606 } 607 } 608 return 0; 609 } 610 EXPORT_SYMBOL(netdev_boot_setup_check); 611 612 613 /** 614 * netdev_boot_base - get address from boot time settings 615 * @prefix: prefix for network device 616 * @unit: id for network device 617 * 618 * Check boot time settings for the base address of device. 619 * The found settings are set for the device to be used 620 * later in the device probing. 621 * Returns 0 if no settings found. 622 */ 623 unsigned long netdev_boot_base(const char *prefix, int unit) 624 { 625 const struct netdev_boot_setup *s = dev_boot_setup; 626 char name[IFNAMSIZ]; 627 int i; 628 629 sprintf(name, "%s%d", prefix, unit); 630 631 /* 632 * If device already registered then return base of 1 633 * to indicate not to probe for this interface 634 */ 635 if (__dev_get_by_name(&init_net, name)) 636 return 1; 637 638 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 639 if (!strcmp(name, s[i].name)) 640 return s[i].map.base_addr; 641 return 0; 642 } 643 644 /* 645 * Saves at boot time configured settings for any netdevice. 646 */ 647 int __init netdev_boot_setup(char *str) 648 { 649 int ints[5]; 650 struct ifmap map; 651 652 str = get_options(str, ARRAY_SIZE(ints), ints); 653 if (!str || !*str) 654 return 0; 655 656 /* Save settings */ 657 memset(&map, 0, sizeof(map)); 658 if (ints[0] > 0) 659 map.irq = ints[1]; 660 if (ints[0] > 1) 661 map.base_addr = ints[2]; 662 if (ints[0] > 2) 663 map.mem_start = ints[3]; 664 if (ints[0] > 3) 665 map.mem_end = ints[4]; 666 667 /* Add new entry to the list */ 668 return netdev_boot_setup_add(str, &map); 669 } 670 671 __setup("netdev=", netdev_boot_setup); 672 673 /******************************************************************************* 674 * 675 * Device Interface Subroutines 676 * 677 *******************************************************************************/ 678 679 /** 680 * dev_get_iflink - get 'iflink' value of a interface 681 * @dev: targeted interface 682 * 683 * Indicates the ifindex the interface is linked to. 684 * Physical interfaces have the same 'ifindex' and 'iflink' values. 685 */ 686 687 int dev_get_iflink(const struct net_device *dev) 688 { 689 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 690 return dev->netdev_ops->ndo_get_iflink(dev); 691 692 return dev->ifindex; 693 } 694 EXPORT_SYMBOL(dev_get_iflink); 695 696 /** 697 * dev_fill_metadata_dst - Retrieve tunnel egress information. 698 * @dev: targeted interface 699 * @skb: The packet. 700 * 701 * For better visibility of tunnel traffic OVS needs to retrieve 702 * egress tunnel information for a packet. Following API allows 703 * user to get this info. 704 */ 705 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 706 { 707 struct ip_tunnel_info *info; 708 709 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 710 return -EINVAL; 711 712 info = skb_tunnel_info_unclone(skb); 713 if (!info) 714 return -ENOMEM; 715 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 716 return -EINVAL; 717 718 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 719 } 720 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 721 722 /** 723 * __dev_get_by_name - find a device by its name 724 * @net: the applicable net namespace 725 * @name: name to find 726 * 727 * Find an interface by name. Must be called under RTNL semaphore 728 * or @dev_base_lock. If the name is found a pointer to the device 729 * is returned. If the name is not found then %NULL is returned. The 730 * reference counters are not incremented so the caller must be 731 * careful with locks. 732 */ 733 734 struct net_device *__dev_get_by_name(struct net *net, const char *name) 735 { 736 struct net_device *dev; 737 struct hlist_head *head = dev_name_hash(net, name); 738 739 hlist_for_each_entry(dev, head, name_hlist) 740 if (!strncmp(dev->name, name, IFNAMSIZ)) 741 return dev; 742 743 return NULL; 744 } 745 EXPORT_SYMBOL(__dev_get_by_name); 746 747 /** 748 * dev_get_by_name_rcu - find a device by its name 749 * @net: the applicable net namespace 750 * @name: name to find 751 * 752 * Find an interface by name. 753 * If the name is found a pointer to the device is returned. 754 * If the name is not found then %NULL is returned. 755 * The reference counters are not incremented so the caller must be 756 * careful with locks. The caller must hold RCU lock. 757 */ 758 759 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 760 { 761 struct net_device *dev; 762 struct hlist_head *head = dev_name_hash(net, name); 763 764 hlist_for_each_entry_rcu(dev, head, name_hlist) 765 if (!strncmp(dev->name, name, IFNAMSIZ)) 766 return dev; 767 768 return NULL; 769 } 770 EXPORT_SYMBOL(dev_get_by_name_rcu); 771 772 /** 773 * dev_get_by_name - find a device by its name 774 * @net: the applicable net namespace 775 * @name: name to find 776 * 777 * Find an interface by name. This can be called from any 778 * context and does its own locking. The returned handle has 779 * the usage count incremented and the caller must use dev_put() to 780 * release it when it is no longer needed. %NULL is returned if no 781 * matching device is found. 782 */ 783 784 struct net_device *dev_get_by_name(struct net *net, const char *name) 785 { 786 struct net_device *dev; 787 788 rcu_read_lock(); 789 dev = dev_get_by_name_rcu(net, name); 790 if (dev) 791 dev_hold(dev); 792 rcu_read_unlock(); 793 return dev; 794 } 795 EXPORT_SYMBOL(dev_get_by_name); 796 797 /** 798 * __dev_get_by_index - find a device by its ifindex 799 * @net: the applicable net namespace 800 * @ifindex: index of device 801 * 802 * Search for an interface by index. Returns %NULL if the device 803 * is not found or a pointer to the device. The device has not 804 * had its reference counter increased so the caller must be careful 805 * about locking. The caller must hold either the RTNL semaphore 806 * or @dev_base_lock. 807 */ 808 809 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 810 { 811 struct net_device *dev; 812 struct hlist_head *head = dev_index_hash(net, ifindex); 813 814 hlist_for_each_entry(dev, head, index_hlist) 815 if (dev->ifindex == ifindex) 816 return dev; 817 818 return NULL; 819 } 820 EXPORT_SYMBOL(__dev_get_by_index); 821 822 /** 823 * dev_get_by_index_rcu - find a device by its ifindex 824 * @net: the applicable net namespace 825 * @ifindex: index of device 826 * 827 * Search for an interface by index. Returns %NULL if the device 828 * is not found or a pointer to the device. The device has not 829 * had its reference counter increased so the caller must be careful 830 * about locking. The caller must hold RCU lock. 831 */ 832 833 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 834 { 835 struct net_device *dev; 836 struct hlist_head *head = dev_index_hash(net, ifindex); 837 838 hlist_for_each_entry_rcu(dev, head, index_hlist) 839 if (dev->ifindex == ifindex) 840 return dev; 841 842 return NULL; 843 } 844 EXPORT_SYMBOL(dev_get_by_index_rcu); 845 846 847 /** 848 * dev_get_by_index - find a device by its ifindex 849 * @net: the applicable net namespace 850 * @ifindex: index of device 851 * 852 * Search for an interface by index. Returns NULL if the device 853 * is not found or a pointer to the device. The device returned has 854 * had a reference added and the pointer is safe until the user calls 855 * dev_put to indicate they have finished with it. 856 */ 857 858 struct net_device *dev_get_by_index(struct net *net, int ifindex) 859 { 860 struct net_device *dev; 861 862 rcu_read_lock(); 863 dev = dev_get_by_index_rcu(net, ifindex); 864 if (dev) 865 dev_hold(dev); 866 rcu_read_unlock(); 867 return dev; 868 } 869 EXPORT_SYMBOL(dev_get_by_index); 870 871 /** 872 * dev_get_by_napi_id - find a device by napi_id 873 * @napi_id: ID of the NAPI struct 874 * 875 * Search for an interface by NAPI ID. Returns %NULL if the device 876 * is not found or a pointer to the device. The device has not had 877 * its reference counter increased so the caller must be careful 878 * about locking. The caller must hold RCU lock. 879 */ 880 881 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 882 { 883 struct napi_struct *napi; 884 885 WARN_ON_ONCE(!rcu_read_lock_held()); 886 887 if (napi_id < MIN_NAPI_ID) 888 return NULL; 889 890 napi = napi_by_id(napi_id); 891 892 return napi ? napi->dev : NULL; 893 } 894 EXPORT_SYMBOL(dev_get_by_napi_id); 895 896 /** 897 * netdev_get_name - get a netdevice name, knowing its ifindex. 898 * @net: network namespace 899 * @name: a pointer to the buffer where the name will be stored. 900 * @ifindex: the ifindex of the interface to get the name from. 901 * 902 * The use of raw_seqcount_begin() and cond_resched() before 903 * retrying is required as we want to give the writers a chance 904 * to complete when CONFIG_PREEMPT is not set. 905 */ 906 int netdev_get_name(struct net *net, char *name, int ifindex) 907 { 908 struct net_device *dev; 909 unsigned int seq; 910 911 retry: 912 seq = raw_seqcount_begin(&devnet_rename_seq); 913 rcu_read_lock(); 914 dev = dev_get_by_index_rcu(net, ifindex); 915 if (!dev) { 916 rcu_read_unlock(); 917 return -ENODEV; 918 } 919 920 strcpy(name, dev->name); 921 rcu_read_unlock(); 922 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 923 cond_resched(); 924 goto retry; 925 } 926 927 return 0; 928 } 929 930 /** 931 * dev_getbyhwaddr_rcu - find a device by its hardware address 932 * @net: the applicable net namespace 933 * @type: media type of device 934 * @ha: hardware address 935 * 936 * Search for an interface by MAC address. Returns NULL if the device 937 * is not found or a pointer to the device. 938 * The caller must hold RCU or RTNL. 939 * The returned device has not had its ref count increased 940 * and the caller must therefore be careful about locking 941 * 942 */ 943 944 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 945 const char *ha) 946 { 947 struct net_device *dev; 948 949 for_each_netdev_rcu(net, dev) 950 if (dev->type == type && 951 !memcmp(dev->dev_addr, ha, dev->addr_len)) 952 return dev; 953 954 return NULL; 955 } 956 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 957 958 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 959 { 960 struct net_device *dev; 961 962 ASSERT_RTNL(); 963 for_each_netdev(net, dev) 964 if (dev->type == type) 965 return dev; 966 967 return NULL; 968 } 969 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 970 971 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 972 { 973 struct net_device *dev, *ret = NULL; 974 975 rcu_read_lock(); 976 for_each_netdev_rcu(net, dev) 977 if (dev->type == type) { 978 dev_hold(dev); 979 ret = dev; 980 break; 981 } 982 rcu_read_unlock(); 983 return ret; 984 } 985 EXPORT_SYMBOL(dev_getfirstbyhwtype); 986 987 /** 988 * __dev_get_by_flags - find any device with given flags 989 * @net: the applicable net namespace 990 * @if_flags: IFF_* values 991 * @mask: bitmask of bits in if_flags to check 992 * 993 * Search for any interface with the given flags. Returns NULL if a device 994 * is not found or a pointer to the device. Must be called inside 995 * rtnl_lock(), and result refcount is unchanged. 996 */ 997 998 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 999 unsigned short mask) 1000 { 1001 struct net_device *dev, *ret; 1002 1003 ASSERT_RTNL(); 1004 1005 ret = NULL; 1006 for_each_netdev(net, dev) { 1007 if (((dev->flags ^ if_flags) & mask) == 0) { 1008 ret = dev; 1009 break; 1010 } 1011 } 1012 return ret; 1013 } 1014 EXPORT_SYMBOL(__dev_get_by_flags); 1015 1016 /** 1017 * dev_valid_name - check if name is okay for network device 1018 * @name: name string 1019 * 1020 * Network device names need to be valid file names to 1021 * to allow sysfs to work. We also disallow any kind of 1022 * whitespace. 1023 */ 1024 bool dev_valid_name(const char *name) 1025 { 1026 if (*name == '\0') 1027 return false; 1028 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1029 return false; 1030 if (!strcmp(name, ".") || !strcmp(name, "..")) 1031 return false; 1032 1033 while (*name) { 1034 if (*name == '/' || *name == ':' || isspace(*name)) 1035 return false; 1036 name++; 1037 } 1038 return true; 1039 } 1040 EXPORT_SYMBOL(dev_valid_name); 1041 1042 /** 1043 * __dev_alloc_name - allocate a name for a device 1044 * @net: network namespace to allocate the device name in 1045 * @name: name format string 1046 * @buf: scratch buffer and result name string 1047 * 1048 * Passed a format string - eg "lt%d" it will try and find a suitable 1049 * id. It scans list of devices to build up a free map, then chooses 1050 * the first empty slot. The caller must hold the dev_base or rtnl lock 1051 * while allocating the name and adding the device in order to avoid 1052 * duplicates. 1053 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1054 * Returns the number of the unit assigned or a negative errno code. 1055 */ 1056 1057 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1058 { 1059 int i = 0; 1060 const char *p; 1061 const int max_netdevices = 8*PAGE_SIZE; 1062 unsigned long *inuse; 1063 struct net_device *d; 1064 1065 if (!dev_valid_name(name)) 1066 return -EINVAL; 1067 1068 p = strchr(name, '%'); 1069 if (p) { 1070 /* 1071 * Verify the string as this thing may have come from 1072 * the user. There must be either one "%d" and no other "%" 1073 * characters. 1074 */ 1075 if (p[1] != 'd' || strchr(p + 2, '%')) 1076 return -EINVAL; 1077 1078 /* Use one page as a bit array of possible slots */ 1079 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1080 if (!inuse) 1081 return -ENOMEM; 1082 1083 for_each_netdev(net, d) { 1084 if (!sscanf(d->name, name, &i)) 1085 continue; 1086 if (i < 0 || i >= max_netdevices) 1087 continue; 1088 1089 /* avoid cases where sscanf is not exact inverse of printf */ 1090 snprintf(buf, IFNAMSIZ, name, i); 1091 if (!strncmp(buf, d->name, IFNAMSIZ)) 1092 set_bit(i, inuse); 1093 } 1094 1095 i = find_first_zero_bit(inuse, max_netdevices); 1096 free_page((unsigned long) inuse); 1097 } 1098 1099 snprintf(buf, IFNAMSIZ, name, i); 1100 if (!__dev_get_by_name(net, buf)) 1101 return i; 1102 1103 /* It is possible to run out of possible slots 1104 * when the name is long and there isn't enough space left 1105 * for the digits, or if all bits are used. 1106 */ 1107 return -ENFILE; 1108 } 1109 1110 static int dev_alloc_name_ns(struct net *net, 1111 struct net_device *dev, 1112 const char *name) 1113 { 1114 char buf[IFNAMSIZ]; 1115 int ret; 1116 1117 BUG_ON(!net); 1118 ret = __dev_alloc_name(net, name, buf); 1119 if (ret >= 0) 1120 strlcpy(dev->name, buf, IFNAMSIZ); 1121 return ret; 1122 } 1123 1124 /** 1125 * dev_alloc_name - allocate a name for a device 1126 * @dev: device 1127 * @name: name format string 1128 * 1129 * Passed a format string - eg "lt%d" it will try and find a suitable 1130 * id. It scans list of devices to build up a free map, then chooses 1131 * the first empty slot. The caller must hold the dev_base or rtnl lock 1132 * while allocating the name and adding the device in order to avoid 1133 * duplicates. 1134 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1135 * Returns the number of the unit assigned or a negative errno code. 1136 */ 1137 1138 int dev_alloc_name(struct net_device *dev, const char *name) 1139 { 1140 return dev_alloc_name_ns(dev_net(dev), dev, name); 1141 } 1142 EXPORT_SYMBOL(dev_alloc_name); 1143 1144 int dev_get_valid_name(struct net *net, struct net_device *dev, 1145 const char *name) 1146 { 1147 BUG_ON(!net); 1148 1149 if (!dev_valid_name(name)) 1150 return -EINVAL; 1151 1152 if (strchr(name, '%')) 1153 return dev_alloc_name_ns(net, dev, name); 1154 else if (__dev_get_by_name(net, name)) 1155 return -EEXIST; 1156 else if (dev->name != name) 1157 strlcpy(dev->name, name, IFNAMSIZ); 1158 1159 return 0; 1160 } 1161 EXPORT_SYMBOL(dev_get_valid_name); 1162 1163 /** 1164 * dev_change_name - change name of a device 1165 * @dev: device 1166 * @newname: name (or format string) must be at least IFNAMSIZ 1167 * 1168 * Change name of a device, can pass format strings "eth%d". 1169 * for wildcarding. 1170 */ 1171 int dev_change_name(struct net_device *dev, const char *newname) 1172 { 1173 unsigned char old_assign_type; 1174 char oldname[IFNAMSIZ]; 1175 int err = 0; 1176 int ret; 1177 struct net *net; 1178 1179 ASSERT_RTNL(); 1180 BUG_ON(!dev_net(dev)); 1181 1182 net = dev_net(dev); 1183 if (dev->flags & IFF_UP) 1184 return -EBUSY; 1185 1186 write_seqcount_begin(&devnet_rename_seq); 1187 1188 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1189 write_seqcount_end(&devnet_rename_seq); 1190 return 0; 1191 } 1192 1193 memcpy(oldname, dev->name, IFNAMSIZ); 1194 1195 err = dev_get_valid_name(net, dev, newname); 1196 if (err < 0) { 1197 write_seqcount_end(&devnet_rename_seq); 1198 return err; 1199 } 1200 1201 if (oldname[0] && !strchr(oldname, '%')) 1202 netdev_info(dev, "renamed from %s\n", oldname); 1203 1204 old_assign_type = dev->name_assign_type; 1205 dev->name_assign_type = NET_NAME_RENAMED; 1206 1207 rollback: 1208 ret = device_rename(&dev->dev, dev->name); 1209 if (ret) { 1210 memcpy(dev->name, oldname, IFNAMSIZ); 1211 dev->name_assign_type = old_assign_type; 1212 write_seqcount_end(&devnet_rename_seq); 1213 return ret; 1214 } 1215 1216 write_seqcount_end(&devnet_rename_seq); 1217 1218 netdev_adjacent_rename_links(dev, oldname); 1219 1220 write_lock_bh(&dev_base_lock); 1221 hlist_del_rcu(&dev->name_hlist); 1222 write_unlock_bh(&dev_base_lock); 1223 1224 synchronize_rcu(); 1225 1226 write_lock_bh(&dev_base_lock); 1227 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1228 write_unlock_bh(&dev_base_lock); 1229 1230 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1231 ret = notifier_to_errno(ret); 1232 1233 if (ret) { 1234 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1235 if (err >= 0) { 1236 err = ret; 1237 write_seqcount_begin(&devnet_rename_seq); 1238 memcpy(dev->name, oldname, IFNAMSIZ); 1239 memcpy(oldname, newname, IFNAMSIZ); 1240 dev->name_assign_type = old_assign_type; 1241 old_assign_type = NET_NAME_RENAMED; 1242 goto rollback; 1243 } else { 1244 pr_err("%s: name change rollback failed: %d\n", 1245 dev->name, ret); 1246 } 1247 } 1248 1249 return err; 1250 } 1251 1252 /** 1253 * dev_set_alias - change ifalias of a device 1254 * @dev: device 1255 * @alias: name up to IFALIASZ 1256 * @len: limit of bytes to copy from info 1257 * 1258 * Set ifalias for a device, 1259 */ 1260 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1261 { 1262 struct dev_ifalias *new_alias = NULL; 1263 1264 if (len >= IFALIASZ) 1265 return -EINVAL; 1266 1267 if (len) { 1268 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1269 if (!new_alias) 1270 return -ENOMEM; 1271 1272 memcpy(new_alias->ifalias, alias, len); 1273 new_alias->ifalias[len] = 0; 1274 } 1275 1276 mutex_lock(&ifalias_mutex); 1277 rcu_swap_protected(dev->ifalias, new_alias, 1278 mutex_is_locked(&ifalias_mutex)); 1279 mutex_unlock(&ifalias_mutex); 1280 1281 if (new_alias) 1282 kfree_rcu(new_alias, rcuhead); 1283 1284 return len; 1285 } 1286 EXPORT_SYMBOL(dev_set_alias); 1287 1288 /** 1289 * dev_get_alias - get ifalias of a device 1290 * @dev: device 1291 * @name: buffer to store name of ifalias 1292 * @len: size of buffer 1293 * 1294 * get ifalias for a device. Caller must make sure dev cannot go 1295 * away, e.g. rcu read lock or own a reference count to device. 1296 */ 1297 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1298 { 1299 const struct dev_ifalias *alias; 1300 int ret = 0; 1301 1302 rcu_read_lock(); 1303 alias = rcu_dereference(dev->ifalias); 1304 if (alias) 1305 ret = snprintf(name, len, "%s", alias->ifalias); 1306 rcu_read_unlock(); 1307 1308 return ret; 1309 } 1310 1311 /** 1312 * netdev_features_change - device changes features 1313 * @dev: device to cause notification 1314 * 1315 * Called to indicate a device has changed features. 1316 */ 1317 void netdev_features_change(struct net_device *dev) 1318 { 1319 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1320 } 1321 EXPORT_SYMBOL(netdev_features_change); 1322 1323 /** 1324 * netdev_state_change - device changes state 1325 * @dev: device to cause notification 1326 * 1327 * Called to indicate a device has changed state. This function calls 1328 * the notifier chains for netdev_chain and sends a NEWLINK message 1329 * to the routing socket. 1330 */ 1331 void netdev_state_change(struct net_device *dev) 1332 { 1333 if (dev->flags & IFF_UP) { 1334 struct netdev_notifier_change_info change_info = { 1335 .info.dev = dev, 1336 }; 1337 1338 call_netdevice_notifiers_info(NETDEV_CHANGE, 1339 &change_info.info); 1340 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1341 } 1342 } 1343 EXPORT_SYMBOL(netdev_state_change); 1344 1345 /** 1346 * netdev_notify_peers - notify network peers about existence of @dev 1347 * @dev: network device 1348 * 1349 * Generate traffic such that interested network peers are aware of 1350 * @dev, such as by generating a gratuitous ARP. This may be used when 1351 * a device wants to inform the rest of the network about some sort of 1352 * reconfiguration such as a failover event or virtual machine 1353 * migration. 1354 */ 1355 void netdev_notify_peers(struct net_device *dev) 1356 { 1357 rtnl_lock(); 1358 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1359 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1360 rtnl_unlock(); 1361 } 1362 EXPORT_SYMBOL(netdev_notify_peers); 1363 1364 static int __dev_open(struct net_device *dev) 1365 { 1366 const struct net_device_ops *ops = dev->netdev_ops; 1367 int ret; 1368 1369 ASSERT_RTNL(); 1370 1371 if (!netif_device_present(dev)) 1372 return -ENODEV; 1373 1374 /* Block netpoll from trying to do any rx path servicing. 1375 * If we don't do this there is a chance ndo_poll_controller 1376 * or ndo_poll may be running while we open the device 1377 */ 1378 netpoll_poll_disable(dev); 1379 1380 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); 1381 ret = notifier_to_errno(ret); 1382 if (ret) 1383 return ret; 1384 1385 set_bit(__LINK_STATE_START, &dev->state); 1386 1387 if (ops->ndo_validate_addr) 1388 ret = ops->ndo_validate_addr(dev); 1389 1390 if (!ret && ops->ndo_open) 1391 ret = ops->ndo_open(dev); 1392 1393 netpoll_poll_enable(dev); 1394 1395 if (ret) 1396 clear_bit(__LINK_STATE_START, &dev->state); 1397 else { 1398 dev->flags |= IFF_UP; 1399 dev_set_rx_mode(dev); 1400 dev_activate(dev); 1401 add_device_randomness(dev->dev_addr, dev->addr_len); 1402 } 1403 1404 return ret; 1405 } 1406 1407 /** 1408 * dev_open - prepare an interface for use. 1409 * @dev: device to open 1410 * 1411 * Takes a device from down to up state. The device's private open 1412 * function is invoked and then the multicast lists are loaded. Finally 1413 * the device is moved into the up state and a %NETDEV_UP message is 1414 * sent to the netdev notifier chain. 1415 * 1416 * Calling this function on an active interface is a nop. On a failure 1417 * a negative errno code is returned. 1418 */ 1419 int dev_open(struct net_device *dev) 1420 { 1421 int ret; 1422 1423 if (dev->flags & IFF_UP) 1424 return 0; 1425 1426 ret = __dev_open(dev); 1427 if (ret < 0) 1428 return ret; 1429 1430 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1431 call_netdevice_notifiers(NETDEV_UP, dev); 1432 1433 return ret; 1434 } 1435 EXPORT_SYMBOL(dev_open); 1436 1437 static void __dev_close_many(struct list_head *head) 1438 { 1439 struct net_device *dev; 1440 1441 ASSERT_RTNL(); 1442 might_sleep(); 1443 1444 list_for_each_entry(dev, head, close_list) { 1445 /* Temporarily disable netpoll until the interface is down */ 1446 netpoll_poll_disable(dev); 1447 1448 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1449 1450 clear_bit(__LINK_STATE_START, &dev->state); 1451 1452 /* Synchronize to scheduled poll. We cannot touch poll list, it 1453 * can be even on different cpu. So just clear netif_running(). 1454 * 1455 * dev->stop() will invoke napi_disable() on all of it's 1456 * napi_struct instances on this device. 1457 */ 1458 smp_mb__after_atomic(); /* Commit netif_running(). */ 1459 } 1460 1461 dev_deactivate_many(head); 1462 1463 list_for_each_entry(dev, head, close_list) { 1464 const struct net_device_ops *ops = dev->netdev_ops; 1465 1466 /* 1467 * Call the device specific close. This cannot fail. 1468 * Only if device is UP 1469 * 1470 * We allow it to be called even after a DETACH hot-plug 1471 * event. 1472 */ 1473 if (ops->ndo_stop) 1474 ops->ndo_stop(dev); 1475 1476 dev->flags &= ~IFF_UP; 1477 netpoll_poll_enable(dev); 1478 } 1479 } 1480 1481 static void __dev_close(struct net_device *dev) 1482 { 1483 LIST_HEAD(single); 1484 1485 list_add(&dev->close_list, &single); 1486 __dev_close_many(&single); 1487 list_del(&single); 1488 } 1489 1490 void dev_close_many(struct list_head *head, bool unlink) 1491 { 1492 struct net_device *dev, *tmp; 1493 1494 /* Remove the devices that don't need to be closed */ 1495 list_for_each_entry_safe(dev, tmp, head, close_list) 1496 if (!(dev->flags & IFF_UP)) 1497 list_del_init(&dev->close_list); 1498 1499 __dev_close_many(head); 1500 1501 list_for_each_entry_safe(dev, tmp, head, close_list) { 1502 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1503 call_netdevice_notifiers(NETDEV_DOWN, dev); 1504 if (unlink) 1505 list_del_init(&dev->close_list); 1506 } 1507 } 1508 EXPORT_SYMBOL(dev_close_many); 1509 1510 /** 1511 * dev_close - shutdown an interface. 1512 * @dev: device to shutdown 1513 * 1514 * This function moves an active device into down state. A 1515 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1516 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1517 * chain. 1518 */ 1519 void dev_close(struct net_device *dev) 1520 { 1521 if (dev->flags & IFF_UP) { 1522 LIST_HEAD(single); 1523 1524 list_add(&dev->close_list, &single); 1525 dev_close_many(&single, true); 1526 list_del(&single); 1527 } 1528 } 1529 EXPORT_SYMBOL(dev_close); 1530 1531 1532 /** 1533 * dev_disable_lro - disable Large Receive Offload on a device 1534 * @dev: device 1535 * 1536 * Disable Large Receive Offload (LRO) on a net device. Must be 1537 * called under RTNL. This is needed if received packets may be 1538 * forwarded to another interface. 1539 */ 1540 void dev_disable_lro(struct net_device *dev) 1541 { 1542 struct net_device *lower_dev; 1543 struct list_head *iter; 1544 1545 dev->wanted_features &= ~NETIF_F_LRO; 1546 netdev_update_features(dev); 1547 1548 if (unlikely(dev->features & NETIF_F_LRO)) 1549 netdev_WARN(dev, "failed to disable LRO!\n"); 1550 1551 netdev_for_each_lower_dev(dev, lower_dev, iter) 1552 dev_disable_lro(lower_dev); 1553 } 1554 EXPORT_SYMBOL(dev_disable_lro); 1555 1556 /** 1557 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1558 * @dev: device 1559 * 1560 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1561 * called under RTNL. This is needed if Generic XDP is installed on 1562 * the device. 1563 */ 1564 static void dev_disable_gro_hw(struct net_device *dev) 1565 { 1566 dev->wanted_features &= ~NETIF_F_GRO_HW; 1567 netdev_update_features(dev); 1568 1569 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1570 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1571 } 1572 1573 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1574 { 1575 #define N(val) \ 1576 case NETDEV_##val: \ 1577 return "NETDEV_" __stringify(val); 1578 switch (cmd) { 1579 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1580 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1581 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1582 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1583 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1584 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1585 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1586 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1587 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1588 } 1589 #undef N 1590 return "UNKNOWN_NETDEV_EVENT"; 1591 } 1592 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1593 1594 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1595 struct net_device *dev) 1596 { 1597 struct netdev_notifier_info info = { 1598 .dev = dev, 1599 }; 1600 1601 return nb->notifier_call(nb, val, &info); 1602 } 1603 1604 static int dev_boot_phase = 1; 1605 1606 /** 1607 * register_netdevice_notifier - register a network notifier block 1608 * @nb: notifier 1609 * 1610 * Register a notifier to be called when network device events occur. 1611 * The notifier passed is linked into the kernel structures and must 1612 * not be reused until it has been unregistered. A negative errno code 1613 * is returned on a failure. 1614 * 1615 * When registered all registration and up events are replayed 1616 * to the new notifier to allow device to have a race free 1617 * view of the network device list. 1618 */ 1619 1620 int register_netdevice_notifier(struct notifier_block *nb) 1621 { 1622 struct net_device *dev; 1623 struct net_device *last; 1624 struct net *net; 1625 int err; 1626 1627 /* Close race with setup_net() and cleanup_net() */ 1628 down_write(&pernet_ops_rwsem); 1629 rtnl_lock(); 1630 err = raw_notifier_chain_register(&netdev_chain, nb); 1631 if (err) 1632 goto unlock; 1633 if (dev_boot_phase) 1634 goto unlock; 1635 for_each_net(net) { 1636 for_each_netdev(net, dev) { 1637 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1638 err = notifier_to_errno(err); 1639 if (err) 1640 goto rollback; 1641 1642 if (!(dev->flags & IFF_UP)) 1643 continue; 1644 1645 call_netdevice_notifier(nb, NETDEV_UP, dev); 1646 } 1647 } 1648 1649 unlock: 1650 rtnl_unlock(); 1651 up_write(&pernet_ops_rwsem); 1652 return err; 1653 1654 rollback: 1655 last = dev; 1656 for_each_net(net) { 1657 for_each_netdev(net, dev) { 1658 if (dev == last) 1659 goto outroll; 1660 1661 if (dev->flags & IFF_UP) { 1662 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1663 dev); 1664 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1665 } 1666 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1667 } 1668 } 1669 1670 outroll: 1671 raw_notifier_chain_unregister(&netdev_chain, nb); 1672 goto unlock; 1673 } 1674 EXPORT_SYMBOL(register_netdevice_notifier); 1675 1676 /** 1677 * unregister_netdevice_notifier - unregister a network notifier block 1678 * @nb: notifier 1679 * 1680 * Unregister a notifier previously registered by 1681 * register_netdevice_notifier(). The notifier is unlinked into the 1682 * kernel structures and may then be reused. A negative errno code 1683 * is returned on a failure. 1684 * 1685 * After unregistering unregister and down device events are synthesized 1686 * for all devices on the device list to the removed notifier to remove 1687 * the need for special case cleanup code. 1688 */ 1689 1690 int unregister_netdevice_notifier(struct notifier_block *nb) 1691 { 1692 struct net_device *dev; 1693 struct net *net; 1694 int err; 1695 1696 /* Close race with setup_net() and cleanup_net() */ 1697 down_write(&pernet_ops_rwsem); 1698 rtnl_lock(); 1699 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1700 if (err) 1701 goto unlock; 1702 1703 for_each_net(net) { 1704 for_each_netdev(net, dev) { 1705 if (dev->flags & IFF_UP) { 1706 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1707 dev); 1708 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1709 } 1710 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1711 } 1712 } 1713 unlock: 1714 rtnl_unlock(); 1715 up_write(&pernet_ops_rwsem); 1716 return err; 1717 } 1718 EXPORT_SYMBOL(unregister_netdevice_notifier); 1719 1720 /** 1721 * call_netdevice_notifiers_info - call all network notifier blocks 1722 * @val: value passed unmodified to notifier function 1723 * @info: notifier information data 1724 * 1725 * Call all network notifier blocks. Parameters and return value 1726 * are as for raw_notifier_call_chain(). 1727 */ 1728 1729 static int call_netdevice_notifiers_info(unsigned long val, 1730 struct netdev_notifier_info *info) 1731 { 1732 ASSERT_RTNL(); 1733 return raw_notifier_call_chain(&netdev_chain, val, info); 1734 } 1735 1736 /** 1737 * call_netdevice_notifiers - call all network notifier blocks 1738 * @val: value passed unmodified to notifier function 1739 * @dev: net_device pointer passed unmodified to notifier function 1740 * 1741 * Call all network notifier blocks. Parameters and return value 1742 * are as for raw_notifier_call_chain(). 1743 */ 1744 1745 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1746 { 1747 struct netdev_notifier_info info = { 1748 .dev = dev, 1749 }; 1750 1751 return call_netdevice_notifiers_info(val, &info); 1752 } 1753 EXPORT_SYMBOL(call_netdevice_notifiers); 1754 1755 #ifdef CONFIG_NET_INGRESS 1756 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 1757 1758 void net_inc_ingress_queue(void) 1759 { 1760 static_branch_inc(&ingress_needed_key); 1761 } 1762 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1763 1764 void net_dec_ingress_queue(void) 1765 { 1766 static_branch_dec(&ingress_needed_key); 1767 } 1768 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1769 #endif 1770 1771 #ifdef CONFIG_NET_EGRESS 1772 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 1773 1774 void net_inc_egress_queue(void) 1775 { 1776 static_branch_inc(&egress_needed_key); 1777 } 1778 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 1779 1780 void net_dec_egress_queue(void) 1781 { 1782 static_branch_dec(&egress_needed_key); 1783 } 1784 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 1785 #endif 1786 1787 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 1788 #ifdef HAVE_JUMP_LABEL 1789 static atomic_t netstamp_needed_deferred; 1790 static atomic_t netstamp_wanted; 1791 static void netstamp_clear(struct work_struct *work) 1792 { 1793 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1794 int wanted; 1795 1796 wanted = atomic_add_return(deferred, &netstamp_wanted); 1797 if (wanted > 0) 1798 static_branch_enable(&netstamp_needed_key); 1799 else 1800 static_branch_disable(&netstamp_needed_key); 1801 } 1802 static DECLARE_WORK(netstamp_work, netstamp_clear); 1803 #endif 1804 1805 void net_enable_timestamp(void) 1806 { 1807 #ifdef HAVE_JUMP_LABEL 1808 int wanted; 1809 1810 while (1) { 1811 wanted = atomic_read(&netstamp_wanted); 1812 if (wanted <= 0) 1813 break; 1814 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 1815 return; 1816 } 1817 atomic_inc(&netstamp_needed_deferred); 1818 schedule_work(&netstamp_work); 1819 #else 1820 static_branch_inc(&netstamp_needed_key); 1821 #endif 1822 } 1823 EXPORT_SYMBOL(net_enable_timestamp); 1824 1825 void net_disable_timestamp(void) 1826 { 1827 #ifdef HAVE_JUMP_LABEL 1828 int wanted; 1829 1830 while (1) { 1831 wanted = atomic_read(&netstamp_wanted); 1832 if (wanted <= 1) 1833 break; 1834 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 1835 return; 1836 } 1837 atomic_dec(&netstamp_needed_deferred); 1838 schedule_work(&netstamp_work); 1839 #else 1840 static_branch_dec(&netstamp_needed_key); 1841 #endif 1842 } 1843 EXPORT_SYMBOL(net_disable_timestamp); 1844 1845 static inline void net_timestamp_set(struct sk_buff *skb) 1846 { 1847 skb->tstamp = 0; 1848 if (static_branch_unlikely(&netstamp_needed_key)) 1849 __net_timestamp(skb); 1850 } 1851 1852 #define net_timestamp_check(COND, SKB) \ 1853 if (static_branch_unlikely(&netstamp_needed_key)) { \ 1854 if ((COND) && !(SKB)->tstamp) \ 1855 __net_timestamp(SKB); \ 1856 } \ 1857 1858 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 1859 { 1860 unsigned int len; 1861 1862 if (!(dev->flags & IFF_UP)) 1863 return false; 1864 1865 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1866 if (skb->len <= len) 1867 return true; 1868 1869 /* if TSO is enabled, we don't care about the length as the packet 1870 * could be forwarded without being segmented before 1871 */ 1872 if (skb_is_gso(skb)) 1873 return true; 1874 1875 return false; 1876 } 1877 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1878 1879 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1880 { 1881 int ret = ____dev_forward_skb(dev, skb); 1882 1883 if (likely(!ret)) { 1884 skb->protocol = eth_type_trans(skb, dev); 1885 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1886 } 1887 1888 return ret; 1889 } 1890 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1891 1892 /** 1893 * dev_forward_skb - loopback an skb to another netif 1894 * 1895 * @dev: destination network device 1896 * @skb: buffer to forward 1897 * 1898 * return values: 1899 * NET_RX_SUCCESS (no congestion) 1900 * NET_RX_DROP (packet was dropped, but freed) 1901 * 1902 * dev_forward_skb can be used for injecting an skb from the 1903 * start_xmit function of one device into the receive queue 1904 * of another device. 1905 * 1906 * The receiving device may be in another namespace, so 1907 * we have to clear all information in the skb that could 1908 * impact namespace isolation. 1909 */ 1910 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1911 { 1912 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1913 } 1914 EXPORT_SYMBOL_GPL(dev_forward_skb); 1915 1916 static inline int deliver_skb(struct sk_buff *skb, 1917 struct packet_type *pt_prev, 1918 struct net_device *orig_dev) 1919 { 1920 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 1921 return -ENOMEM; 1922 refcount_inc(&skb->users); 1923 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1924 } 1925 1926 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 1927 struct packet_type **pt, 1928 struct net_device *orig_dev, 1929 __be16 type, 1930 struct list_head *ptype_list) 1931 { 1932 struct packet_type *ptype, *pt_prev = *pt; 1933 1934 list_for_each_entry_rcu(ptype, ptype_list, list) { 1935 if (ptype->type != type) 1936 continue; 1937 if (pt_prev) 1938 deliver_skb(skb, pt_prev, orig_dev); 1939 pt_prev = ptype; 1940 } 1941 *pt = pt_prev; 1942 } 1943 1944 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1945 { 1946 if (!ptype->af_packet_priv || !skb->sk) 1947 return false; 1948 1949 if (ptype->id_match) 1950 return ptype->id_match(ptype, skb->sk); 1951 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1952 return true; 1953 1954 return false; 1955 } 1956 1957 /* 1958 * Support routine. Sends outgoing frames to any network 1959 * taps currently in use. 1960 */ 1961 1962 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 1963 { 1964 struct packet_type *ptype; 1965 struct sk_buff *skb2 = NULL; 1966 struct packet_type *pt_prev = NULL; 1967 struct list_head *ptype_list = &ptype_all; 1968 1969 rcu_read_lock(); 1970 again: 1971 list_for_each_entry_rcu(ptype, ptype_list, list) { 1972 if (ptype->ignore_outgoing) 1973 continue; 1974 1975 /* Never send packets back to the socket 1976 * they originated from - MvS (miquels@drinkel.ow.org) 1977 */ 1978 if (skb_loop_sk(ptype, skb)) 1979 continue; 1980 1981 if (pt_prev) { 1982 deliver_skb(skb2, pt_prev, skb->dev); 1983 pt_prev = ptype; 1984 continue; 1985 } 1986 1987 /* need to clone skb, done only once */ 1988 skb2 = skb_clone(skb, GFP_ATOMIC); 1989 if (!skb2) 1990 goto out_unlock; 1991 1992 net_timestamp_set(skb2); 1993 1994 /* skb->nh should be correctly 1995 * set by sender, so that the second statement is 1996 * just protection against buggy protocols. 1997 */ 1998 skb_reset_mac_header(skb2); 1999 2000 if (skb_network_header(skb2) < skb2->data || 2001 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2002 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2003 ntohs(skb2->protocol), 2004 dev->name); 2005 skb_reset_network_header(skb2); 2006 } 2007 2008 skb2->transport_header = skb2->network_header; 2009 skb2->pkt_type = PACKET_OUTGOING; 2010 pt_prev = ptype; 2011 } 2012 2013 if (ptype_list == &ptype_all) { 2014 ptype_list = &dev->ptype_all; 2015 goto again; 2016 } 2017 out_unlock: 2018 if (pt_prev) { 2019 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2020 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2021 else 2022 kfree_skb(skb2); 2023 } 2024 rcu_read_unlock(); 2025 } 2026 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2027 2028 /** 2029 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2030 * @dev: Network device 2031 * @txq: number of queues available 2032 * 2033 * If real_num_tx_queues is changed the tc mappings may no longer be 2034 * valid. To resolve this verify the tc mapping remains valid and if 2035 * not NULL the mapping. With no priorities mapping to this 2036 * offset/count pair it will no longer be used. In the worst case TC0 2037 * is invalid nothing can be done so disable priority mappings. If is 2038 * expected that drivers will fix this mapping if they can before 2039 * calling netif_set_real_num_tx_queues. 2040 */ 2041 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2042 { 2043 int i; 2044 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2045 2046 /* If TC0 is invalidated disable TC mapping */ 2047 if (tc->offset + tc->count > txq) { 2048 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2049 dev->num_tc = 0; 2050 return; 2051 } 2052 2053 /* Invalidated prio to tc mappings set to TC0 */ 2054 for (i = 1; i < TC_BITMASK + 1; i++) { 2055 int q = netdev_get_prio_tc_map(dev, i); 2056 2057 tc = &dev->tc_to_txq[q]; 2058 if (tc->offset + tc->count > txq) { 2059 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2060 i, q); 2061 netdev_set_prio_tc_map(dev, i, 0); 2062 } 2063 } 2064 } 2065 2066 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2067 { 2068 if (dev->num_tc) { 2069 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2070 int i; 2071 2072 /* walk through the TCs and see if it falls into any of them */ 2073 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2074 if ((txq - tc->offset) < tc->count) 2075 return i; 2076 } 2077 2078 /* didn't find it, just return -1 to indicate no match */ 2079 return -1; 2080 } 2081 2082 return 0; 2083 } 2084 EXPORT_SYMBOL(netdev_txq_to_tc); 2085 2086 #ifdef CONFIG_XPS 2087 struct static_key xps_needed __read_mostly; 2088 EXPORT_SYMBOL(xps_needed); 2089 struct static_key xps_rxqs_needed __read_mostly; 2090 EXPORT_SYMBOL(xps_rxqs_needed); 2091 static DEFINE_MUTEX(xps_map_mutex); 2092 #define xmap_dereference(P) \ 2093 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2094 2095 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2096 int tci, u16 index) 2097 { 2098 struct xps_map *map = NULL; 2099 int pos; 2100 2101 if (dev_maps) 2102 map = xmap_dereference(dev_maps->attr_map[tci]); 2103 if (!map) 2104 return false; 2105 2106 for (pos = map->len; pos--;) { 2107 if (map->queues[pos] != index) 2108 continue; 2109 2110 if (map->len > 1) { 2111 map->queues[pos] = map->queues[--map->len]; 2112 break; 2113 } 2114 2115 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2116 kfree_rcu(map, rcu); 2117 return false; 2118 } 2119 2120 return true; 2121 } 2122 2123 static bool remove_xps_queue_cpu(struct net_device *dev, 2124 struct xps_dev_maps *dev_maps, 2125 int cpu, u16 offset, u16 count) 2126 { 2127 int num_tc = dev->num_tc ? : 1; 2128 bool active = false; 2129 int tci; 2130 2131 for (tci = cpu * num_tc; num_tc--; tci++) { 2132 int i, j; 2133 2134 for (i = count, j = offset; i--; j++) { 2135 if (!remove_xps_queue(dev_maps, tci, j)) 2136 break; 2137 } 2138 2139 active |= i < 0; 2140 } 2141 2142 return active; 2143 } 2144 2145 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, 2146 struct xps_dev_maps *dev_maps, unsigned int nr_ids, 2147 u16 offset, u16 count, bool is_rxqs_map) 2148 { 2149 bool active = false; 2150 int i, j; 2151 2152 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), 2153 j < nr_ids;) 2154 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, 2155 count); 2156 if (!active) { 2157 if (is_rxqs_map) { 2158 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); 2159 } else { 2160 RCU_INIT_POINTER(dev->xps_cpus_map, NULL); 2161 2162 for (i = offset + (count - 1); count--; i--) 2163 netdev_queue_numa_node_write( 2164 netdev_get_tx_queue(dev, i), 2165 NUMA_NO_NODE); 2166 } 2167 kfree_rcu(dev_maps, rcu); 2168 } 2169 } 2170 2171 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2172 u16 count) 2173 { 2174 const unsigned long *possible_mask = NULL; 2175 struct xps_dev_maps *dev_maps; 2176 unsigned int nr_ids; 2177 2178 if (!static_key_false(&xps_needed)) 2179 return; 2180 2181 cpus_read_lock(); 2182 mutex_lock(&xps_map_mutex); 2183 2184 if (static_key_false(&xps_rxqs_needed)) { 2185 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2186 if (dev_maps) { 2187 nr_ids = dev->num_rx_queues; 2188 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, 2189 offset, count, true); 2190 } 2191 } 2192 2193 dev_maps = xmap_dereference(dev->xps_cpus_map); 2194 if (!dev_maps) 2195 goto out_no_maps; 2196 2197 if (num_possible_cpus() > 1) 2198 possible_mask = cpumask_bits(cpu_possible_mask); 2199 nr_ids = nr_cpu_ids; 2200 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, 2201 false); 2202 2203 out_no_maps: 2204 if (static_key_enabled(&xps_rxqs_needed)) 2205 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2206 2207 static_key_slow_dec_cpuslocked(&xps_needed); 2208 mutex_unlock(&xps_map_mutex); 2209 cpus_read_unlock(); 2210 } 2211 2212 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2213 { 2214 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2215 } 2216 2217 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2218 u16 index, bool is_rxqs_map) 2219 { 2220 struct xps_map *new_map; 2221 int alloc_len = XPS_MIN_MAP_ALLOC; 2222 int i, pos; 2223 2224 for (pos = 0; map && pos < map->len; pos++) { 2225 if (map->queues[pos] != index) 2226 continue; 2227 return map; 2228 } 2229 2230 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2231 if (map) { 2232 if (pos < map->alloc_len) 2233 return map; 2234 2235 alloc_len = map->alloc_len * 2; 2236 } 2237 2238 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2239 * map 2240 */ 2241 if (is_rxqs_map) 2242 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2243 else 2244 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2245 cpu_to_node(attr_index)); 2246 if (!new_map) 2247 return NULL; 2248 2249 for (i = 0; i < pos; i++) 2250 new_map->queues[i] = map->queues[i]; 2251 new_map->alloc_len = alloc_len; 2252 new_map->len = pos; 2253 2254 return new_map; 2255 } 2256 2257 /* Must be called under cpus_read_lock */ 2258 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2259 u16 index, bool is_rxqs_map) 2260 { 2261 const unsigned long *online_mask = NULL, *possible_mask = NULL; 2262 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2263 int i, j, tci, numa_node_id = -2; 2264 int maps_sz, num_tc = 1, tc = 0; 2265 struct xps_map *map, *new_map; 2266 bool active = false; 2267 unsigned int nr_ids; 2268 2269 if (dev->num_tc) { 2270 /* Do not allow XPS on subordinate device directly */ 2271 num_tc = dev->num_tc; 2272 if (num_tc < 0) 2273 return -EINVAL; 2274 2275 /* If queue belongs to subordinate dev use its map */ 2276 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2277 2278 tc = netdev_txq_to_tc(dev, index); 2279 if (tc < 0) 2280 return -EINVAL; 2281 } 2282 2283 mutex_lock(&xps_map_mutex); 2284 if (is_rxqs_map) { 2285 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2286 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2287 nr_ids = dev->num_rx_queues; 2288 } else { 2289 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2290 if (num_possible_cpus() > 1) { 2291 online_mask = cpumask_bits(cpu_online_mask); 2292 possible_mask = cpumask_bits(cpu_possible_mask); 2293 } 2294 dev_maps = xmap_dereference(dev->xps_cpus_map); 2295 nr_ids = nr_cpu_ids; 2296 } 2297 2298 if (maps_sz < L1_CACHE_BYTES) 2299 maps_sz = L1_CACHE_BYTES; 2300 2301 /* allocate memory for queue storage */ 2302 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2303 j < nr_ids;) { 2304 if (!new_dev_maps) 2305 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2306 if (!new_dev_maps) { 2307 mutex_unlock(&xps_map_mutex); 2308 return -ENOMEM; 2309 } 2310 2311 tci = j * num_tc + tc; 2312 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : 2313 NULL; 2314 2315 map = expand_xps_map(map, j, index, is_rxqs_map); 2316 if (!map) 2317 goto error; 2318 2319 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2320 } 2321 2322 if (!new_dev_maps) 2323 goto out_no_new_maps; 2324 2325 static_key_slow_inc_cpuslocked(&xps_needed); 2326 if (is_rxqs_map) 2327 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2328 2329 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2330 j < nr_ids;) { 2331 /* copy maps belonging to foreign traffic classes */ 2332 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { 2333 /* fill in the new device map from the old device map */ 2334 map = xmap_dereference(dev_maps->attr_map[tci]); 2335 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2336 } 2337 2338 /* We need to explicitly update tci as prevous loop 2339 * could break out early if dev_maps is NULL. 2340 */ 2341 tci = j * num_tc + tc; 2342 2343 if (netif_attr_test_mask(j, mask, nr_ids) && 2344 netif_attr_test_online(j, online_mask, nr_ids)) { 2345 /* add tx-queue to CPU/rx-queue maps */ 2346 int pos = 0; 2347 2348 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2349 while ((pos < map->len) && (map->queues[pos] != index)) 2350 pos++; 2351 2352 if (pos == map->len) 2353 map->queues[map->len++] = index; 2354 #ifdef CONFIG_NUMA 2355 if (!is_rxqs_map) { 2356 if (numa_node_id == -2) 2357 numa_node_id = cpu_to_node(j); 2358 else if (numa_node_id != cpu_to_node(j)) 2359 numa_node_id = -1; 2360 } 2361 #endif 2362 } else if (dev_maps) { 2363 /* fill in the new device map from the old device map */ 2364 map = xmap_dereference(dev_maps->attr_map[tci]); 2365 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2366 } 2367 2368 /* copy maps belonging to foreign traffic classes */ 2369 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { 2370 /* fill in the new device map from the old device map */ 2371 map = xmap_dereference(dev_maps->attr_map[tci]); 2372 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2373 } 2374 } 2375 2376 if (is_rxqs_map) 2377 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); 2378 else 2379 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); 2380 2381 /* Cleanup old maps */ 2382 if (!dev_maps) 2383 goto out_no_old_maps; 2384 2385 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2386 j < nr_ids;) { 2387 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2388 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2389 map = xmap_dereference(dev_maps->attr_map[tci]); 2390 if (map && map != new_map) 2391 kfree_rcu(map, rcu); 2392 } 2393 } 2394 2395 kfree_rcu(dev_maps, rcu); 2396 2397 out_no_old_maps: 2398 dev_maps = new_dev_maps; 2399 active = true; 2400 2401 out_no_new_maps: 2402 if (!is_rxqs_map) { 2403 /* update Tx queue numa node */ 2404 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2405 (numa_node_id >= 0) ? 2406 numa_node_id : NUMA_NO_NODE); 2407 } 2408 2409 if (!dev_maps) 2410 goto out_no_maps; 2411 2412 /* removes tx-queue from unused CPUs/rx-queues */ 2413 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2414 j < nr_ids;) { 2415 for (i = tc, tci = j * num_tc; i--; tci++) 2416 active |= remove_xps_queue(dev_maps, tci, index); 2417 if (!netif_attr_test_mask(j, mask, nr_ids) || 2418 !netif_attr_test_online(j, online_mask, nr_ids)) 2419 active |= remove_xps_queue(dev_maps, tci, index); 2420 for (i = num_tc - tc, tci++; --i; tci++) 2421 active |= remove_xps_queue(dev_maps, tci, index); 2422 } 2423 2424 /* free map if not active */ 2425 if (!active) { 2426 if (is_rxqs_map) 2427 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); 2428 else 2429 RCU_INIT_POINTER(dev->xps_cpus_map, NULL); 2430 kfree_rcu(dev_maps, rcu); 2431 } 2432 2433 out_no_maps: 2434 mutex_unlock(&xps_map_mutex); 2435 2436 return 0; 2437 error: 2438 /* remove any maps that we added */ 2439 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2440 j < nr_ids;) { 2441 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2442 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2443 map = dev_maps ? 2444 xmap_dereference(dev_maps->attr_map[tci]) : 2445 NULL; 2446 if (new_map && new_map != map) 2447 kfree(new_map); 2448 } 2449 } 2450 2451 mutex_unlock(&xps_map_mutex); 2452 2453 kfree(new_dev_maps); 2454 return -ENOMEM; 2455 } 2456 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2457 2458 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2459 u16 index) 2460 { 2461 int ret; 2462 2463 cpus_read_lock(); 2464 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); 2465 cpus_read_unlock(); 2466 2467 return ret; 2468 } 2469 EXPORT_SYMBOL(netif_set_xps_queue); 2470 2471 #endif 2472 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2473 { 2474 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2475 2476 /* Unbind any subordinate channels */ 2477 while (txq-- != &dev->_tx[0]) { 2478 if (txq->sb_dev) 2479 netdev_unbind_sb_channel(dev, txq->sb_dev); 2480 } 2481 } 2482 2483 void netdev_reset_tc(struct net_device *dev) 2484 { 2485 #ifdef CONFIG_XPS 2486 netif_reset_xps_queues_gt(dev, 0); 2487 #endif 2488 netdev_unbind_all_sb_channels(dev); 2489 2490 /* Reset TC configuration of device */ 2491 dev->num_tc = 0; 2492 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2493 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2494 } 2495 EXPORT_SYMBOL(netdev_reset_tc); 2496 2497 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2498 { 2499 if (tc >= dev->num_tc) 2500 return -EINVAL; 2501 2502 #ifdef CONFIG_XPS 2503 netif_reset_xps_queues(dev, offset, count); 2504 #endif 2505 dev->tc_to_txq[tc].count = count; 2506 dev->tc_to_txq[tc].offset = offset; 2507 return 0; 2508 } 2509 EXPORT_SYMBOL(netdev_set_tc_queue); 2510 2511 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2512 { 2513 if (num_tc > TC_MAX_QUEUE) 2514 return -EINVAL; 2515 2516 #ifdef CONFIG_XPS 2517 netif_reset_xps_queues_gt(dev, 0); 2518 #endif 2519 netdev_unbind_all_sb_channels(dev); 2520 2521 dev->num_tc = num_tc; 2522 return 0; 2523 } 2524 EXPORT_SYMBOL(netdev_set_num_tc); 2525 2526 void netdev_unbind_sb_channel(struct net_device *dev, 2527 struct net_device *sb_dev) 2528 { 2529 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2530 2531 #ifdef CONFIG_XPS 2532 netif_reset_xps_queues_gt(sb_dev, 0); 2533 #endif 2534 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2535 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2536 2537 while (txq-- != &dev->_tx[0]) { 2538 if (txq->sb_dev == sb_dev) 2539 txq->sb_dev = NULL; 2540 } 2541 } 2542 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2543 2544 int netdev_bind_sb_channel_queue(struct net_device *dev, 2545 struct net_device *sb_dev, 2546 u8 tc, u16 count, u16 offset) 2547 { 2548 /* Make certain the sb_dev and dev are already configured */ 2549 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2550 return -EINVAL; 2551 2552 /* We cannot hand out queues we don't have */ 2553 if ((offset + count) > dev->real_num_tx_queues) 2554 return -EINVAL; 2555 2556 /* Record the mapping */ 2557 sb_dev->tc_to_txq[tc].count = count; 2558 sb_dev->tc_to_txq[tc].offset = offset; 2559 2560 /* Provide a way for Tx queue to find the tc_to_txq map or 2561 * XPS map for itself. 2562 */ 2563 while (count--) 2564 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2565 2566 return 0; 2567 } 2568 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2569 2570 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2571 { 2572 /* Do not use a multiqueue device to represent a subordinate channel */ 2573 if (netif_is_multiqueue(dev)) 2574 return -ENODEV; 2575 2576 /* We allow channels 1 - 32767 to be used for subordinate channels. 2577 * Channel 0 is meant to be "native" mode and used only to represent 2578 * the main root device. We allow writing 0 to reset the device back 2579 * to normal mode after being used as a subordinate channel. 2580 */ 2581 if (channel > S16_MAX) 2582 return -EINVAL; 2583 2584 dev->num_tc = -channel; 2585 2586 return 0; 2587 } 2588 EXPORT_SYMBOL(netdev_set_sb_channel); 2589 2590 /* 2591 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2592 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2593 */ 2594 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2595 { 2596 bool disabling; 2597 int rc; 2598 2599 disabling = txq < dev->real_num_tx_queues; 2600 2601 if (txq < 1 || txq > dev->num_tx_queues) 2602 return -EINVAL; 2603 2604 if (dev->reg_state == NETREG_REGISTERED || 2605 dev->reg_state == NETREG_UNREGISTERING) { 2606 ASSERT_RTNL(); 2607 2608 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2609 txq); 2610 if (rc) 2611 return rc; 2612 2613 if (dev->num_tc) 2614 netif_setup_tc(dev, txq); 2615 2616 dev->real_num_tx_queues = txq; 2617 2618 if (disabling) { 2619 synchronize_net(); 2620 qdisc_reset_all_tx_gt(dev, txq); 2621 #ifdef CONFIG_XPS 2622 netif_reset_xps_queues_gt(dev, txq); 2623 #endif 2624 } 2625 } else { 2626 dev->real_num_tx_queues = txq; 2627 } 2628 2629 return 0; 2630 } 2631 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2632 2633 #ifdef CONFIG_SYSFS 2634 /** 2635 * netif_set_real_num_rx_queues - set actual number of RX queues used 2636 * @dev: Network device 2637 * @rxq: Actual number of RX queues 2638 * 2639 * This must be called either with the rtnl_lock held or before 2640 * registration of the net device. Returns 0 on success, or a 2641 * negative error code. If called before registration, it always 2642 * succeeds. 2643 */ 2644 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2645 { 2646 int rc; 2647 2648 if (rxq < 1 || rxq > dev->num_rx_queues) 2649 return -EINVAL; 2650 2651 if (dev->reg_state == NETREG_REGISTERED) { 2652 ASSERT_RTNL(); 2653 2654 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2655 rxq); 2656 if (rc) 2657 return rc; 2658 } 2659 2660 dev->real_num_rx_queues = rxq; 2661 return 0; 2662 } 2663 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2664 #endif 2665 2666 /** 2667 * netif_get_num_default_rss_queues - default number of RSS queues 2668 * 2669 * This routine should set an upper limit on the number of RSS queues 2670 * used by default by multiqueue devices. 2671 */ 2672 int netif_get_num_default_rss_queues(void) 2673 { 2674 return is_kdump_kernel() ? 2675 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2676 } 2677 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2678 2679 static void __netif_reschedule(struct Qdisc *q) 2680 { 2681 struct softnet_data *sd; 2682 unsigned long flags; 2683 2684 local_irq_save(flags); 2685 sd = this_cpu_ptr(&softnet_data); 2686 q->next_sched = NULL; 2687 *sd->output_queue_tailp = q; 2688 sd->output_queue_tailp = &q->next_sched; 2689 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2690 local_irq_restore(flags); 2691 } 2692 2693 void __netif_schedule(struct Qdisc *q) 2694 { 2695 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2696 __netif_reschedule(q); 2697 } 2698 EXPORT_SYMBOL(__netif_schedule); 2699 2700 struct dev_kfree_skb_cb { 2701 enum skb_free_reason reason; 2702 }; 2703 2704 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2705 { 2706 return (struct dev_kfree_skb_cb *)skb->cb; 2707 } 2708 2709 void netif_schedule_queue(struct netdev_queue *txq) 2710 { 2711 rcu_read_lock(); 2712 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2713 struct Qdisc *q = rcu_dereference(txq->qdisc); 2714 2715 __netif_schedule(q); 2716 } 2717 rcu_read_unlock(); 2718 } 2719 EXPORT_SYMBOL(netif_schedule_queue); 2720 2721 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2722 { 2723 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2724 struct Qdisc *q; 2725 2726 rcu_read_lock(); 2727 q = rcu_dereference(dev_queue->qdisc); 2728 __netif_schedule(q); 2729 rcu_read_unlock(); 2730 } 2731 } 2732 EXPORT_SYMBOL(netif_tx_wake_queue); 2733 2734 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2735 { 2736 unsigned long flags; 2737 2738 if (unlikely(!skb)) 2739 return; 2740 2741 if (likely(refcount_read(&skb->users) == 1)) { 2742 smp_rmb(); 2743 refcount_set(&skb->users, 0); 2744 } else if (likely(!refcount_dec_and_test(&skb->users))) { 2745 return; 2746 } 2747 get_kfree_skb_cb(skb)->reason = reason; 2748 local_irq_save(flags); 2749 skb->next = __this_cpu_read(softnet_data.completion_queue); 2750 __this_cpu_write(softnet_data.completion_queue, skb); 2751 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2752 local_irq_restore(flags); 2753 } 2754 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2755 2756 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2757 { 2758 if (in_irq() || irqs_disabled()) 2759 __dev_kfree_skb_irq(skb, reason); 2760 else 2761 dev_kfree_skb(skb); 2762 } 2763 EXPORT_SYMBOL(__dev_kfree_skb_any); 2764 2765 2766 /** 2767 * netif_device_detach - mark device as removed 2768 * @dev: network device 2769 * 2770 * Mark device as removed from system and therefore no longer available. 2771 */ 2772 void netif_device_detach(struct net_device *dev) 2773 { 2774 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2775 netif_running(dev)) { 2776 netif_tx_stop_all_queues(dev); 2777 } 2778 } 2779 EXPORT_SYMBOL(netif_device_detach); 2780 2781 /** 2782 * netif_device_attach - mark device as attached 2783 * @dev: network device 2784 * 2785 * Mark device as attached from system and restart if needed. 2786 */ 2787 void netif_device_attach(struct net_device *dev) 2788 { 2789 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2790 netif_running(dev)) { 2791 netif_tx_wake_all_queues(dev); 2792 __netdev_watchdog_up(dev); 2793 } 2794 } 2795 EXPORT_SYMBOL(netif_device_attach); 2796 2797 /* 2798 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 2799 * to be used as a distribution range. 2800 */ 2801 static u16 skb_tx_hash(const struct net_device *dev, 2802 const struct net_device *sb_dev, 2803 struct sk_buff *skb) 2804 { 2805 u32 hash; 2806 u16 qoffset = 0; 2807 u16 qcount = dev->real_num_tx_queues; 2808 2809 if (dev->num_tc) { 2810 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 2811 2812 qoffset = sb_dev->tc_to_txq[tc].offset; 2813 qcount = sb_dev->tc_to_txq[tc].count; 2814 } 2815 2816 if (skb_rx_queue_recorded(skb)) { 2817 hash = skb_get_rx_queue(skb); 2818 while (unlikely(hash >= qcount)) 2819 hash -= qcount; 2820 return hash + qoffset; 2821 } 2822 2823 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 2824 } 2825 2826 static void skb_warn_bad_offload(const struct sk_buff *skb) 2827 { 2828 static const netdev_features_t null_features; 2829 struct net_device *dev = skb->dev; 2830 const char *name = ""; 2831 2832 if (!net_ratelimit()) 2833 return; 2834 2835 if (dev) { 2836 if (dev->dev.parent) 2837 name = dev_driver_string(dev->dev.parent); 2838 else 2839 name = netdev_name(dev); 2840 } 2841 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2842 "gso_type=%d ip_summed=%d\n", 2843 name, dev ? &dev->features : &null_features, 2844 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2845 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2846 skb_shinfo(skb)->gso_type, skb->ip_summed); 2847 } 2848 2849 /* 2850 * Invalidate hardware checksum when packet is to be mangled, and 2851 * complete checksum manually on outgoing path. 2852 */ 2853 int skb_checksum_help(struct sk_buff *skb) 2854 { 2855 __wsum csum; 2856 int ret = 0, offset; 2857 2858 if (skb->ip_summed == CHECKSUM_COMPLETE) 2859 goto out_set_summed; 2860 2861 if (unlikely(skb_shinfo(skb)->gso_size)) { 2862 skb_warn_bad_offload(skb); 2863 return -EINVAL; 2864 } 2865 2866 /* Before computing a checksum, we should make sure no frag could 2867 * be modified by an external entity : checksum could be wrong. 2868 */ 2869 if (skb_has_shared_frag(skb)) { 2870 ret = __skb_linearize(skb); 2871 if (ret) 2872 goto out; 2873 } 2874 2875 offset = skb_checksum_start_offset(skb); 2876 BUG_ON(offset >= skb_headlen(skb)); 2877 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2878 2879 offset += skb->csum_offset; 2880 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2881 2882 if (skb_cloned(skb) && 2883 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2884 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2885 if (ret) 2886 goto out; 2887 } 2888 2889 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 2890 out_set_summed: 2891 skb->ip_summed = CHECKSUM_NONE; 2892 out: 2893 return ret; 2894 } 2895 EXPORT_SYMBOL(skb_checksum_help); 2896 2897 int skb_crc32c_csum_help(struct sk_buff *skb) 2898 { 2899 __le32 crc32c_csum; 2900 int ret = 0, offset, start; 2901 2902 if (skb->ip_summed != CHECKSUM_PARTIAL) 2903 goto out; 2904 2905 if (unlikely(skb_is_gso(skb))) 2906 goto out; 2907 2908 /* Before computing a checksum, we should make sure no frag could 2909 * be modified by an external entity : checksum could be wrong. 2910 */ 2911 if (unlikely(skb_has_shared_frag(skb))) { 2912 ret = __skb_linearize(skb); 2913 if (ret) 2914 goto out; 2915 } 2916 start = skb_checksum_start_offset(skb); 2917 offset = start + offsetof(struct sctphdr, checksum); 2918 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 2919 ret = -EINVAL; 2920 goto out; 2921 } 2922 if (skb_cloned(skb) && 2923 !skb_clone_writable(skb, offset + sizeof(__le32))) { 2924 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2925 if (ret) 2926 goto out; 2927 } 2928 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 2929 skb->len - start, ~(__u32)0, 2930 crc32c_csum_stub)); 2931 *(__le32 *)(skb->data + offset) = crc32c_csum; 2932 skb->ip_summed = CHECKSUM_NONE; 2933 skb->csum_not_inet = 0; 2934 out: 2935 return ret; 2936 } 2937 2938 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2939 { 2940 __be16 type = skb->protocol; 2941 2942 /* Tunnel gso handlers can set protocol to ethernet. */ 2943 if (type == htons(ETH_P_TEB)) { 2944 struct ethhdr *eth; 2945 2946 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2947 return 0; 2948 2949 eth = (struct ethhdr *)skb->data; 2950 type = eth->h_proto; 2951 } 2952 2953 return __vlan_get_protocol(skb, type, depth); 2954 } 2955 2956 /** 2957 * skb_mac_gso_segment - mac layer segmentation handler. 2958 * @skb: buffer to segment 2959 * @features: features for the output path (see dev->features) 2960 */ 2961 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 2962 netdev_features_t features) 2963 { 2964 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 2965 struct packet_offload *ptype; 2966 int vlan_depth = skb->mac_len; 2967 __be16 type = skb_network_protocol(skb, &vlan_depth); 2968 2969 if (unlikely(!type)) 2970 return ERR_PTR(-EINVAL); 2971 2972 __skb_pull(skb, vlan_depth); 2973 2974 rcu_read_lock(); 2975 list_for_each_entry_rcu(ptype, &offload_base, list) { 2976 if (ptype->type == type && ptype->callbacks.gso_segment) { 2977 segs = ptype->callbacks.gso_segment(skb, features); 2978 break; 2979 } 2980 } 2981 rcu_read_unlock(); 2982 2983 __skb_push(skb, skb->data - skb_mac_header(skb)); 2984 2985 return segs; 2986 } 2987 EXPORT_SYMBOL(skb_mac_gso_segment); 2988 2989 2990 /* openvswitch calls this on rx path, so we need a different check. 2991 */ 2992 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 2993 { 2994 if (tx_path) 2995 return skb->ip_summed != CHECKSUM_PARTIAL && 2996 skb->ip_summed != CHECKSUM_UNNECESSARY; 2997 2998 return skb->ip_summed == CHECKSUM_NONE; 2999 } 3000 3001 /** 3002 * __skb_gso_segment - Perform segmentation on skb. 3003 * @skb: buffer to segment 3004 * @features: features for the output path (see dev->features) 3005 * @tx_path: whether it is called in TX path 3006 * 3007 * This function segments the given skb and returns a list of segments. 3008 * 3009 * It may return NULL if the skb requires no segmentation. This is 3010 * only possible when GSO is used for verifying header integrity. 3011 * 3012 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. 3013 */ 3014 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3015 netdev_features_t features, bool tx_path) 3016 { 3017 struct sk_buff *segs; 3018 3019 if (unlikely(skb_needs_check(skb, tx_path))) { 3020 int err; 3021 3022 /* We're going to init ->check field in TCP or UDP header */ 3023 err = skb_cow_head(skb, 0); 3024 if (err < 0) 3025 return ERR_PTR(err); 3026 } 3027 3028 /* Only report GSO partial support if it will enable us to 3029 * support segmentation on this frame without needing additional 3030 * work. 3031 */ 3032 if (features & NETIF_F_GSO_PARTIAL) { 3033 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3034 struct net_device *dev = skb->dev; 3035 3036 partial_features |= dev->features & dev->gso_partial_features; 3037 if (!skb_gso_ok(skb, features | partial_features)) 3038 features &= ~NETIF_F_GSO_PARTIAL; 3039 } 3040 3041 BUILD_BUG_ON(SKB_SGO_CB_OFFSET + 3042 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3043 3044 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3045 SKB_GSO_CB(skb)->encap_level = 0; 3046 3047 skb_reset_mac_header(skb); 3048 skb_reset_mac_len(skb); 3049 3050 segs = skb_mac_gso_segment(skb, features); 3051 3052 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3053 skb_warn_bad_offload(skb); 3054 3055 return segs; 3056 } 3057 EXPORT_SYMBOL(__skb_gso_segment); 3058 3059 /* Take action when hardware reception checksum errors are detected. */ 3060 #ifdef CONFIG_BUG 3061 void netdev_rx_csum_fault(struct net_device *dev) 3062 { 3063 if (net_ratelimit()) { 3064 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3065 dump_stack(); 3066 } 3067 } 3068 EXPORT_SYMBOL(netdev_rx_csum_fault); 3069 #endif 3070 3071 /* XXX: check that highmem exists at all on the given machine. */ 3072 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3073 { 3074 #ifdef CONFIG_HIGHMEM 3075 int i; 3076 3077 if (!(dev->features & NETIF_F_HIGHDMA)) { 3078 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3079 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3080 3081 if (PageHighMem(skb_frag_page(frag))) 3082 return 1; 3083 } 3084 } 3085 #endif 3086 return 0; 3087 } 3088 3089 /* If MPLS offload request, verify we are testing hardware MPLS features 3090 * instead of standard features for the netdev. 3091 */ 3092 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3093 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3094 netdev_features_t features, 3095 __be16 type) 3096 { 3097 if (eth_p_mpls(type)) 3098 features &= skb->dev->mpls_features; 3099 3100 return features; 3101 } 3102 #else 3103 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3104 netdev_features_t features, 3105 __be16 type) 3106 { 3107 return features; 3108 } 3109 #endif 3110 3111 static netdev_features_t harmonize_features(struct sk_buff *skb, 3112 netdev_features_t features) 3113 { 3114 int tmp; 3115 __be16 type; 3116 3117 type = skb_network_protocol(skb, &tmp); 3118 features = net_mpls_features(skb, features, type); 3119 3120 if (skb->ip_summed != CHECKSUM_NONE && 3121 !can_checksum_protocol(features, type)) { 3122 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3123 } 3124 if (illegal_highdma(skb->dev, skb)) 3125 features &= ~NETIF_F_SG; 3126 3127 return features; 3128 } 3129 3130 netdev_features_t passthru_features_check(struct sk_buff *skb, 3131 struct net_device *dev, 3132 netdev_features_t features) 3133 { 3134 return features; 3135 } 3136 EXPORT_SYMBOL(passthru_features_check); 3137 3138 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3139 struct net_device *dev, 3140 netdev_features_t features) 3141 { 3142 return vlan_features_check(skb, features); 3143 } 3144 3145 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3146 struct net_device *dev, 3147 netdev_features_t features) 3148 { 3149 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3150 3151 if (gso_segs > dev->gso_max_segs) 3152 return features & ~NETIF_F_GSO_MASK; 3153 3154 /* Support for GSO partial features requires software 3155 * intervention before we can actually process the packets 3156 * so we need to strip support for any partial features now 3157 * and we can pull them back in after we have partially 3158 * segmented the frame. 3159 */ 3160 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3161 features &= ~dev->gso_partial_features; 3162 3163 /* Make sure to clear the IPv4 ID mangling feature if the 3164 * IPv4 header has the potential to be fragmented. 3165 */ 3166 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3167 struct iphdr *iph = skb->encapsulation ? 3168 inner_ip_hdr(skb) : ip_hdr(skb); 3169 3170 if (!(iph->frag_off & htons(IP_DF))) 3171 features &= ~NETIF_F_TSO_MANGLEID; 3172 } 3173 3174 return features; 3175 } 3176 3177 netdev_features_t netif_skb_features(struct sk_buff *skb) 3178 { 3179 struct net_device *dev = skb->dev; 3180 netdev_features_t features = dev->features; 3181 3182 if (skb_is_gso(skb)) 3183 features = gso_features_check(skb, dev, features); 3184 3185 /* If encapsulation offload request, verify we are testing 3186 * hardware encapsulation features instead of standard 3187 * features for the netdev 3188 */ 3189 if (skb->encapsulation) 3190 features &= dev->hw_enc_features; 3191 3192 if (skb_vlan_tagged(skb)) 3193 features = netdev_intersect_features(features, 3194 dev->vlan_features | 3195 NETIF_F_HW_VLAN_CTAG_TX | 3196 NETIF_F_HW_VLAN_STAG_TX); 3197 3198 if (dev->netdev_ops->ndo_features_check) 3199 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3200 features); 3201 else 3202 features &= dflt_features_check(skb, dev, features); 3203 3204 return harmonize_features(skb, features); 3205 } 3206 EXPORT_SYMBOL(netif_skb_features); 3207 3208 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3209 struct netdev_queue *txq, bool more) 3210 { 3211 unsigned int len; 3212 int rc; 3213 3214 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) 3215 dev_queue_xmit_nit(skb, dev); 3216 3217 len = skb->len; 3218 trace_net_dev_start_xmit(skb, dev); 3219 rc = netdev_start_xmit(skb, dev, txq, more); 3220 trace_net_dev_xmit(skb, rc, dev, len); 3221 3222 return rc; 3223 } 3224 3225 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3226 struct netdev_queue *txq, int *ret) 3227 { 3228 struct sk_buff *skb = first; 3229 int rc = NETDEV_TX_OK; 3230 3231 while (skb) { 3232 struct sk_buff *next = skb->next; 3233 3234 skb_mark_not_on_list(skb); 3235 rc = xmit_one(skb, dev, txq, next != NULL); 3236 if (unlikely(!dev_xmit_complete(rc))) { 3237 skb->next = next; 3238 goto out; 3239 } 3240 3241 skb = next; 3242 if (netif_xmit_stopped(txq) && skb) { 3243 rc = NETDEV_TX_BUSY; 3244 break; 3245 } 3246 } 3247 3248 out: 3249 *ret = rc; 3250 return skb; 3251 } 3252 3253 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3254 netdev_features_t features) 3255 { 3256 if (skb_vlan_tag_present(skb) && 3257 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3258 skb = __vlan_hwaccel_push_inside(skb); 3259 return skb; 3260 } 3261 3262 int skb_csum_hwoffload_help(struct sk_buff *skb, 3263 const netdev_features_t features) 3264 { 3265 if (unlikely(skb->csum_not_inet)) 3266 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3267 skb_crc32c_csum_help(skb); 3268 3269 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); 3270 } 3271 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3272 3273 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3274 { 3275 netdev_features_t features; 3276 3277 features = netif_skb_features(skb); 3278 skb = validate_xmit_vlan(skb, features); 3279 if (unlikely(!skb)) 3280 goto out_null; 3281 3282 skb = sk_validate_xmit_skb(skb, dev); 3283 if (unlikely(!skb)) 3284 goto out_null; 3285 3286 if (netif_needs_gso(skb, features)) { 3287 struct sk_buff *segs; 3288 3289 segs = skb_gso_segment(skb, features); 3290 if (IS_ERR(segs)) { 3291 goto out_kfree_skb; 3292 } else if (segs) { 3293 consume_skb(skb); 3294 skb = segs; 3295 } 3296 } else { 3297 if (skb_needs_linearize(skb, features) && 3298 __skb_linearize(skb)) 3299 goto out_kfree_skb; 3300 3301 /* If packet is not checksummed and device does not 3302 * support checksumming for this protocol, complete 3303 * checksumming here. 3304 */ 3305 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3306 if (skb->encapsulation) 3307 skb_set_inner_transport_header(skb, 3308 skb_checksum_start_offset(skb)); 3309 else 3310 skb_set_transport_header(skb, 3311 skb_checksum_start_offset(skb)); 3312 if (skb_csum_hwoffload_help(skb, features)) 3313 goto out_kfree_skb; 3314 } 3315 } 3316 3317 skb = validate_xmit_xfrm(skb, features, again); 3318 3319 return skb; 3320 3321 out_kfree_skb: 3322 kfree_skb(skb); 3323 out_null: 3324 atomic_long_inc(&dev->tx_dropped); 3325 return NULL; 3326 } 3327 3328 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3329 { 3330 struct sk_buff *next, *head = NULL, *tail; 3331 3332 for (; skb != NULL; skb = next) { 3333 next = skb->next; 3334 skb_mark_not_on_list(skb); 3335 3336 /* in case skb wont be segmented, point to itself */ 3337 skb->prev = skb; 3338 3339 skb = validate_xmit_skb(skb, dev, again); 3340 if (!skb) 3341 continue; 3342 3343 if (!head) 3344 head = skb; 3345 else 3346 tail->next = skb; 3347 /* If skb was segmented, skb->prev points to 3348 * the last segment. If not, it still contains skb. 3349 */ 3350 tail = skb->prev; 3351 } 3352 return head; 3353 } 3354 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3355 3356 static void qdisc_pkt_len_init(struct sk_buff *skb) 3357 { 3358 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3359 3360 qdisc_skb_cb(skb)->pkt_len = skb->len; 3361 3362 /* To get more precise estimation of bytes sent on wire, 3363 * we add to pkt_len the headers size of all segments 3364 */ 3365 if (shinfo->gso_size) { 3366 unsigned int hdr_len; 3367 u16 gso_segs = shinfo->gso_segs; 3368 3369 /* mac layer + network layer */ 3370 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3371 3372 /* + transport layer */ 3373 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3374 const struct tcphdr *th; 3375 struct tcphdr _tcphdr; 3376 3377 th = skb_header_pointer(skb, skb_transport_offset(skb), 3378 sizeof(_tcphdr), &_tcphdr); 3379 if (likely(th)) 3380 hdr_len += __tcp_hdrlen(th); 3381 } else { 3382 struct udphdr _udphdr; 3383 3384 if (skb_header_pointer(skb, skb_transport_offset(skb), 3385 sizeof(_udphdr), &_udphdr)) 3386 hdr_len += sizeof(struct udphdr); 3387 } 3388 3389 if (shinfo->gso_type & SKB_GSO_DODGY) 3390 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3391 shinfo->gso_size); 3392 3393 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3394 } 3395 } 3396 3397 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3398 struct net_device *dev, 3399 struct netdev_queue *txq) 3400 { 3401 spinlock_t *root_lock = qdisc_lock(q); 3402 struct sk_buff *to_free = NULL; 3403 bool contended; 3404 int rc; 3405 3406 qdisc_calculate_pkt_len(skb, q); 3407 3408 if (q->flags & TCQ_F_NOLOCK) { 3409 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3410 __qdisc_drop(skb, &to_free); 3411 rc = NET_XMIT_DROP; 3412 } else { 3413 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3414 qdisc_run(q); 3415 } 3416 3417 if (unlikely(to_free)) 3418 kfree_skb_list(to_free); 3419 return rc; 3420 } 3421 3422 /* 3423 * Heuristic to force contended enqueues to serialize on a 3424 * separate lock before trying to get qdisc main lock. 3425 * This permits qdisc->running owner to get the lock more 3426 * often and dequeue packets faster. 3427 */ 3428 contended = qdisc_is_running(q); 3429 if (unlikely(contended)) 3430 spin_lock(&q->busylock); 3431 3432 spin_lock(root_lock); 3433 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3434 __qdisc_drop(skb, &to_free); 3435 rc = NET_XMIT_DROP; 3436 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3437 qdisc_run_begin(q)) { 3438 /* 3439 * This is a work-conserving queue; there are no old skbs 3440 * waiting to be sent out; and the qdisc is not running - 3441 * xmit the skb directly. 3442 */ 3443 3444 qdisc_bstats_update(q, skb); 3445 3446 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3447 if (unlikely(contended)) { 3448 spin_unlock(&q->busylock); 3449 contended = false; 3450 } 3451 __qdisc_run(q); 3452 } 3453 3454 qdisc_run_end(q); 3455 rc = NET_XMIT_SUCCESS; 3456 } else { 3457 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3458 if (qdisc_run_begin(q)) { 3459 if (unlikely(contended)) { 3460 spin_unlock(&q->busylock); 3461 contended = false; 3462 } 3463 __qdisc_run(q); 3464 qdisc_run_end(q); 3465 } 3466 } 3467 spin_unlock(root_lock); 3468 if (unlikely(to_free)) 3469 kfree_skb_list(to_free); 3470 if (unlikely(contended)) 3471 spin_unlock(&q->busylock); 3472 return rc; 3473 } 3474 3475 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3476 static void skb_update_prio(struct sk_buff *skb) 3477 { 3478 const struct netprio_map *map; 3479 const struct sock *sk; 3480 unsigned int prioidx; 3481 3482 if (skb->priority) 3483 return; 3484 map = rcu_dereference_bh(skb->dev->priomap); 3485 if (!map) 3486 return; 3487 sk = skb_to_full_sk(skb); 3488 if (!sk) 3489 return; 3490 3491 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3492 3493 if (prioidx < map->priomap_len) 3494 skb->priority = map->priomap[prioidx]; 3495 } 3496 #else 3497 #define skb_update_prio(skb) 3498 #endif 3499 3500 DEFINE_PER_CPU(int, xmit_recursion); 3501 EXPORT_SYMBOL(xmit_recursion); 3502 3503 /** 3504 * dev_loopback_xmit - loop back @skb 3505 * @net: network namespace this loopback is happening in 3506 * @sk: sk needed to be a netfilter okfn 3507 * @skb: buffer to transmit 3508 */ 3509 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3510 { 3511 skb_reset_mac_header(skb); 3512 __skb_pull(skb, skb_network_offset(skb)); 3513 skb->pkt_type = PACKET_LOOPBACK; 3514 skb->ip_summed = CHECKSUM_UNNECESSARY; 3515 WARN_ON(!skb_dst(skb)); 3516 skb_dst_force(skb); 3517 netif_rx_ni(skb); 3518 return 0; 3519 } 3520 EXPORT_SYMBOL(dev_loopback_xmit); 3521 3522 #ifdef CONFIG_NET_EGRESS 3523 static struct sk_buff * 3524 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3525 { 3526 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3527 struct tcf_result cl_res; 3528 3529 if (!miniq) 3530 return skb; 3531 3532 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3533 mini_qdisc_bstats_cpu_update(miniq, skb); 3534 3535 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3536 case TC_ACT_OK: 3537 case TC_ACT_RECLASSIFY: 3538 skb->tc_index = TC_H_MIN(cl_res.classid); 3539 break; 3540 case TC_ACT_SHOT: 3541 mini_qdisc_qstats_cpu_drop(miniq); 3542 *ret = NET_XMIT_DROP; 3543 kfree_skb(skb); 3544 return NULL; 3545 case TC_ACT_STOLEN: 3546 case TC_ACT_QUEUED: 3547 case TC_ACT_TRAP: 3548 *ret = NET_XMIT_SUCCESS; 3549 consume_skb(skb); 3550 return NULL; 3551 case TC_ACT_REDIRECT: 3552 /* No need to push/pop skb's mac_header here on egress! */ 3553 skb_do_redirect(skb); 3554 *ret = NET_XMIT_SUCCESS; 3555 return NULL; 3556 default: 3557 break; 3558 } 3559 3560 return skb; 3561 } 3562 #endif /* CONFIG_NET_EGRESS */ 3563 3564 #ifdef CONFIG_XPS 3565 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3566 struct xps_dev_maps *dev_maps, unsigned int tci) 3567 { 3568 struct xps_map *map; 3569 int queue_index = -1; 3570 3571 if (dev->num_tc) { 3572 tci *= dev->num_tc; 3573 tci += netdev_get_prio_tc_map(dev, skb->priority); 3574 } 3575 3576 map = rcu_dereference(dev_maps->attr_map[tci]); 3577 if (map) { 3578 if (map->len == 1) 3579 queue_index = map->queues[0]; 3580 else 3581 queue_index = map->queues[reciprocal_scale( 3582 skb_get_hash(skb), map->len)]; 3583 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3584 queue_index = -1; 3585 } 3586 return queue_index; 3587 } 3588 #endif 3589 3590 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3591 struct sk_buff *skb) 3592 { 3593 #ifdef CONFIG_XPS 3594 struct xps_dev_maps *dev_maps; 3595 struct sock *sk = skb->sk; 3596 int queue_index = -1; 3597 3598 if (!static_key_false(&xps_needed)) 3599 return -1; 3600 3601 rcu_read_lock(); 3602 if (!static_key_false(&xps_rxqs_needed)) 3603 goto get_cpus_map; 3604 3605 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); 3606 if (dev_maps) { 3607 int tci = sk_rx_queue_get(sk); 3608 3609 if (tci >= 0 && tci < dev->num_rx_queues) 3610 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3611 tci); 3612 } 3613 3614 get_cpus_map: 3615 if (queue_index < 0) { 3616 dev_maps = rcu_dereference(sb_dev->xps_cpus_map); 3617 if (dev_maps) { 3618 unsigned int tci = skb->sender_cpu - 1; 3619 3620 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3621 tci); 3622 } 3623 } 3624 rcu_read_unlock(); 3625 3626 return queue_index; 3627 #else 3628 return -1; 3629 #endif 3630 } 3631 3632 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3633 struct net_device *sb_dev, 3634 select_queue_fallback_t fallback) 3635 { 3636 return 0; 3637 } 3638 EXPORT_SYMBOL(dev_pick_tx_zero); 3639 3640 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 3641 struct net_device *sb_dev, 3642 select_queue_fallback_t fallback) 3643 { 3644 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 3645 } 3646 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 3647 3648 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 3649 struct net_device *sb_dev) 3650 { 3651 struct sock *sk = skb->sk; 3652 int queue_index = sk_tx_queue_get(sk); 3653 3654 sb_dev = sb_dev ? : dev; 3655 3656 if (queue_index < 0 || skb->ooo_okay || 3657 queue_index >= dev->real_num_tx_queues) { 3658 int new_index = get_xps_queue(dev, sb_dev, skb); 3659 3660 if (new_index < 0) 3661 new_index = skb_tx_hash(dev, sb_dev, skb); 3662 3663 if (queue_index != new_index && sk && 3664 sk_fullsock(sk) && 3665 rcu_access_pointer(sk->sk_dst_cache)) 3666 sk_tx_queue_set(sk, new_index); 3667 3668 queue_index = new_index; 3669 } 3670 3671 return queue_index; 3672 } 3673 3674 struct netdev_queue *netdev_pick_tx(struct net_device *dev, 3675 struct sk_buff *skb, 3676 struct net_device *sb_dev) 3677 { 3678 int queue_index = 0; 3679 3680 #ifdef CONFIG_XPS 3681 u32 sender_cpu = skb->sender_cpu - 1; 3682 3683 if (sender_cpu >= (u32)NR_CPUS) 3684 skb->sender_cpu = raw_smp_processor_id() + 1; 3685 #endif 3686 3687 if (dev->real_num_tx_queues != 1) { 3688 const struct net_device_ops *ops = dev->netdev_ops; 3689 3690 if (ops->ndo_select_queue) 3691 queue_index = ops->ndo_select_queue(dev, skb, sb_dev, 3692 __netdev_pick_tx); 3693 else 3694 queue_index = __netdev_pick_tx(dev, skb, sb_dev); 3695 3696 queue_index = netdev_cap_txqueue(dev, queue_index); 3697 } 3698 3699 skb_set_queue_mapping(skb, queue_index); 3700 return netdev_get_tx_queue(dev, queue_index); 3701 } 3702 3703 /** 3704 * __dev_queue_xmit - transmit a buffer 3705 * @skb: buffer to transmit 3706 * @sb_dev: suboordinate device used for L2 forwarding offload 3707 * 3708 * Queue a buffer for transmission to a network device. The caller must 3709 * have set the device and priority and built the buffer before calling 3710 * this function. The function can be called from an interrupt. 3711 * 3712 * A negative errno code is returned on a failure. A success does not 3713 * guarantee the frame will be transmitted as it may be dropped due 3714 * to congestion or traffic shaping. 3715 * 3716 * ----------------------------------------------------------------------------------- 3717 * I notice this method can also return errors from the queue disciplines, 3718 * including NET_XMIT_DROP, which is a positive value. So, errors can also 3719 * be positive. 3720 * 3721 * Regardless of the return value, the skb is consumed, so it is currently 3722 * difficult to retry a send to this method. (You can bump the ref count 3723 * before sending to hold a reference for retry if you are careful.) 3724 * 3725 * When calling this method, interrupts MUST be enabled. This is because 3726 * the BH enable code must have IRQs enabled so that it will not deadlock. 3727 * --BLG 3728 */ 3729 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 3730 { 3731 struct net_device *dev = skb->dev; 3732 struct netdev_queue *txq; 3733 struct Qdisc *q; 3734 int rc = -ENOMEM; 3735 bool again = false; 3736 3737 skb_reset_mac_header(skb); 3738 3739 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 3740 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 3741 3742 /* Disable soft irqs for various locks below. Also 3743 * stops preemption for RCU. 3744 */ 3745 rcu_read_lock_bh(); 3746 3747 skb_update_prio(skb); 3748 3749 qdisc_pkt_len_init(skb); 3750 #ifdef CONFIG_NET_CLS_ACT 3751 skb->tc_at_ingress = 0; 3752 # ifdef CONFIG_NET_EGRESS 3753 if (static_branch_unlikely(&egress_needed_key)) { 3754 skb = sch_handle_egress(skb, &rc, dev); 3755 if (!skb) 3756 goto out; 3757 } 3758 # endif 3759 #endif 3760 /* If device/qdisc don't need skb->dst, release it right now while 3761 * its hot in this cpu cache. 3762 */ 3763 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 3764 skb_dst_drop(skb); 3765 else 3766 skb_dst_force(skb); 3767 3768 txq = netdev_pick_tx(dev, skb, sb_dev); 3769 q = rcu_dereference_bh(txq->qdisc); 3770 3771 trace_net_dev_queue(skb); 3772 if (q->enqueue) { 3773 rc = __dev_xmit_skb(skb, q, dev, txq); 3774 goto out; 3775 } 3776 3777 /* The device has no queue. Common case for software devices: 3778 * loopback, all the sorts of tunnels... 3779 3780 * Really, it is unlikely that netif_tx_lock protection is necessary 3781 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 3782 * counters.) 3783 * However, it is possible, that they rely on protection 3784 * made by us here. 3785 3786 * Check this and shot the lock. It is not prone from deadlocks. 3787 *Either shot noqueue qdisc, it is even simpler 8) 3788 */ 3789 if (dev->flags & IFF_UP) { 3790 int cpu = smp_processor_id(); /* ok because BHs are off */ 3791 3792 if (txq->xmit_lock_owner != cpu) { 3793 if (unlikely(__this_cpu_read(xmit_recursion) > 3794 XMIT_RECURSION_LIMIT)) 3795 goto recursion_alert; 3796 3797 skb = validate_xmit_skb(skb, dev, &again); 3798 if (!skb) 3799 goto out; 3800 3801 HARD_TX_LOCK(dev, txq, cpu); 3802 3803 if (!netif_xmit_stopped(txq)) { 3804 __this_cpu_inc(xmit_recursion); 3805 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 3806 __this_cpu_dec(xmit_recursion); 3807 if (dev_xmit_complete(rc)) { 3808 HARD_TX_UNLOCK(dev, txq); 3809 goto out; 3810 } 3811 } 3812 HARD_TX_UNLOCK(dev, txq); 3813 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 3814 dev->name); 3815 } else { 3816 /* Recursion is detected! It is possible, 3817 * unfortunately 3818 */ 3819 recursion_alert: 3820 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 3821 dev->name); 3822 } 3823 } 3824 3825 rc = -ENETDOWN; 3826 rcu_read_unlock_bh(); 3827 3828 atomic_long_inc(&dev->tx_dropped); 3829 kfree_skb_list(skb); 3830 return rc; 3831 out: 3832 rcu_read_unlock_bh(); 3833 return rc; 3834 } 3835 3836 int dev_queue_xmit(struct sk_buff *skb) 3837 { 3838 return __dev_queue_xmit(skb, NULL); 3839 } 3840 EXPORT_SYMBOL(dev_queue_xmit); 3841 3842 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 3843 { 3844 return __dev_queue_xmit(skb, sb_dev); 3845 } 3846 EXPORT_SYMBOL(dev_queue_xmit_accel); 3847 3848 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 3849 { 3850 struct net_device *dev = skb->dev; 3851 struct sk_buff *orig_skb = skb; 3852 struct netdev_queue *txq; 3853 int ret = NETDEV_TX_BUSY; 3854 bool again = false; 3855 3856 if (unlikely(!netif_running(dev) || 3857 !netif_carrier_ok(dev))) 3858 goto drop; 3859 3860 skb = validate_xmit_skb_list(skb, dev, &again); 3861 if (skb != orig_skb) 3862 goto drop; 3863 3864 skb_set_queue_mapping(skb, queue_id); 3865 txq = skb_get_tx_queue(dev, skb); 3866 3867 local_bh_disable(); 3868 3869 HARD_TX_LOCK(dev, txq, smp_processor_id()); 3870 if (!netif_xmit_frozen_or_drv_stopped(txq)) 3871 ret = netdev_start_xmit(skb, dev, txq, false); 3872 HARD_TX_UNLOCK(dev, txq); 3873 3874 local_bh_enable(); 3875 3876 if (!dev_xmit_complete(ret)) 3877 kfree_skb(skb); 3878 3879 return ret; 3880 drop: 3881 atomic_long_inc(&dev->tx_dropped); 3882 kfree_skb_list(skb); 3883 return NET_XMIT_DROP; 3884 } 3885 EXPORT_SYMBOL(dev_direct_xmit); 3886 3887 /************************************************************************* 3888 * Receiver routines 3889 *************************************************************************/ 3890 3891 int netdev_max_backlog __read_mostly = 1000; 3892 EXPORT_SYMBOL(netdev_max_backlog); 3893 3894 int netdev_tstamp_prequeue __read_mostly = 1; 3895 int netdev_budget __read_mostly = 300; 3896 unsigned int __read_mostly netdev_budget_usecs = 2000; 3897 int weight_p __read_mostly = 64; /* old backlog weight */ 3898 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 3899 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 3900 int dev_rx_weight __read_mostly = 64; 3901 int dev_tx_weight __read_mostly = 64; 3902 3903 /* Called with irq disabled */ 3904 static inline void ____napi_schedule(struct softnet_data *sd, 3905 struct napi_struct *napi) 3906 { 3907 list_add_tail(&napi->poll_list, &sd->poll_list); 3908 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3909 } 3910 3911 #ifdef CONFIG_RPS 3912 3913 /* One global table that all flow-based protocols share. */ 3914 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3915 EXPORT_SYMBOL(rps_sock_flow_table); 3916 u32 rps_cpu_mask __read_mostly; 3917 EXPORT_SYMBOL(rps_cpu_mask); 3918 3919 struct static_key rps_needed __read_mostly; 3920 EXPORT_SYMBOL(rps_needed); 3921 struct static_key rfs_needed __read_mostly; 3922 EXPORT_SYMBOL(rfs_needed); 3923 3924 static struct rps_dev_flow * 3925 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3926 struct rps_dev_flow *rflow, u16 next_cpu) 3927 { 3928 if (next_cpu < nr_cpu_ids) { 3929 #ifdef CONFIG_RFS_ACCEL 3930 struct netdev_rx_queue *rxqueue; 3931 struct rps_dev_flow_table *flow_table; 3932 struct rps_dev_flow *old_rflow; 3933 u32 flow_id; 3934 u16 rxq_index; 3935 int rc; 3936 3937 /* Should we steer this flow to a different hardware queue? */ 3938 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3939 !(dev->features & NETIF_F_NTUPLE)) 3940 goto out; 3941 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 3942 if (rxq_index == skb_get_rx_queue(skb)) 3943 goto out; 3944 3945 rxqueue = dev->_rx + rxq_index; 3946 flow_table = rcu_dereference(rxqueue->rps_flow_table); 3947 if (!flow_table) 3948 goto out; 3949 flow_id = skb_get_hash(skb) & flow_table->mask; 3950 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 3951 rxq_index, flow_id); 3952 if (rc < 0) 3953 goto out; 3954 old_rflow = rflow; 3955 rflow = &flow_table->flows[flow_id]; 3956 rflow->filter = rc; 3957 if (old_rflow->filter == rflow->filter) 3958 old_rflow->filter = RPS_NO_FILTER; 3959 out: 3960 #endif 3961 rflow->last_qtail = 3962 per_cpu(softnet_data, next_cpu).input_queue_head; 3963 } 3964 3965 rflow->cpu = next_cpu; 3966 return rflow; 3967 } 3968 3969 /* 3970 * get_rps_cpu is called from netif_receive_skb and returns the target 3971 * CPU from the RPS map of the receiving queue for a given skb. 3972 * rcu_read_lock must be held on entry. 3973 */ 3974 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3975 struct rps_dev_flow **rflowp) 3976 { 3977 const struct rps_sock_flow_table *sock_flow_table; 3978 struct netdev_rx_queue *rxqueue = dev->_rx; 3979 struct rps_dev_flow_table *flow_table; 3980 struct rps_map *map; 3981 int cpu = -1; 3982 u32 tcpu; 3983 u32 hash; 3984 3985 if (skb_rx_queue_recorded(skb)) { 3986 u16 index = skb_get_rx_queue(skb); 3987 3988 if (unlikely(index >= dev->real_num_rx_queues)) { 3989 WARN_ONCE(dev->real_num_rx_queues > 1, 3990 "%s received packet on queue %u, but number " 3991 "of RX queues is %u\n", 3992 dev->name, index, dev->real_num_rx_queues); 3993 goto done; 3994 } 3995 rxqueue += index; 3996 } 3997 3998 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 3999 4000 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4001 map = rcu_dereference(rxqueue->rps_map); 4002 if (!flow_table && !map) 4003 goto done; 4004 4005 skb_reset_network_header(skb); 4006 hash = skb_get_hash(skb); 4007 if (!hash) 4008 goto done; 4009 4010 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4011 if (flow_table && sock_flow_table) { 4012 struct rps_dev_flow *rflow; 4013 u32 next_cpu; 4014 u32 ident; 4015 4016 /* First check into global flow table if there is a match */ 4017 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4018 if ((ident ^ hash) & ~rps_cpu_mask) 4019 goto try_rps; 4020 4021 next_cpu = ident & rps_cpu_mask; 4022 4023 /* OK, now we know there is a match, 4024 * we can look at the local (per receive queue) flow table 4025 */ 4026 rflow = &flow_table->flows[hash & flow_table->mask]; 4027 tcpu = rflow->cpu; 4028 4029 /* 4030 * If the desired CPU (where last recvmsg was done) is 4031 * different from current CPU (one in the rx-queue flow 4032 * table entry), switch if one of the following holds: 4033 * - Current CPU is unset (>= nr_cpu_ids). 4034 * - Current CPU is offline. 4035 * - The current CPU's queue tail has advanced beyond the 4036 * last packet that was enqueued using this table entry. 4037 * This guarantees that all previous packets for the flow 4038 * have been dequeued, thus preserving in order delivery. 4039 */ 4040 if (unlikely(tcpu != next_cpu) && 4041 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4042 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4043 rflow->last_qtail)) >= 0)) { 4044 tcpu = next_cpu; 4045 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4046 } 4047 4048 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4049 *rflowp = rflow; 4050 cpu = tcpu; 4051 goto done; 4052 } 4053 } 4054 4055 try_rps: 4056 4057 if (map) { 4058 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4059 if (cpu_online(tcpu)) { 4060 cpu = tcpu; 4061 goto done; 4062 } 4063 } 4064 4065 done: 4066 return cpu; 4067 } 4068 4069 #ifdef CONFIG_RFS_ACCEL 4070 4071 /** 4072 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4073 * @dev: Device on which the filter was set 4074 * @rxq_index: RX queue index 4075 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4076 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4077 * 4078 * Drivers that implement ndo_rx_flow_steer() should periodically call 4079 * this function for each installed filter and remove the filters for 4080 * which it returns %true. 4081 */ 4082 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4083 u32 flow_id, u16 filter_id) 4084 { 4085 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4086 struct rps_dev_flow_table *flow_table; 4087 struct rps_dev_flow *rflow; 4088 bool expire = true; 4089 unsigned int cpu; 4090 4091 rcu_read_lock(); 4092 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4093 if (flow_table && flow_id <= flow_table->mask) { 4094 rflow = &flow_table->flows[flow_id]; 4095 cpu = READ_ONCE(rflow->cpu); 4096 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4097 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4098 rflow->last_qtail) < 4099 (int)(10 * flow_table->mask))) 4100 expire = false; 4101 } 4102 rcu_read_unlock(); 4103 return expire; 4104 } 4105 EXPORT_SYMBOL(rps_may_expire_flow); 4106 4107 #endif /* CONFIG_RFS_ACCEL */ 4108 4109 /* Called from hardirq (IPI) context */ 4110 static void rps_trigger_softirq(void *data) 4111 { 4112 struct softnet_data *sd = data; 4113 4114 ____napi_schedule(sd, &sd->backlog); 4115 sd->received_rps++; 4116 } 4117 4118 #endif /* CONFIG_RPS */ 4119 4120 /* 4121 * Check if this softnet_data structure is another cpu one 4122 * If yes, queue it to our IPI list and return 1 4123 * If no, return 0 4124 */ 4125 static int rps_ipi_queued(struct softnet_data *sd) 4126 { 4127 #ifdef CONFIG_RPS 4128 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4129 4130 if (sd != mysd) { 4131 sd->rps_ipi_next = mysd->rps_ipi_list; 4132 mysd->rps_ipi_list = sd; 4133 4134 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4135 return 1; 4136 } 4137 #endif /* CONFIG_RPS */ 4138 return 0; 4139 } 4140 4141 #ifdef CONFIG_NET_FLOW_LIMIT 4142 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4143 #endif 4144 4145 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4146 { 4147 #ifdef CONFIG_NET_FLOW_LIMIT 4148 struct sd_flow_limit *fl; 4149 struct softnet_data *sd; 4150 unsigned int old_flow, new_flow; 4151 4152 if (qlen < (netdev_max_backlog >> 1)) 4153 return false; 4154 4155 sd = this_cpu_ptr(&softnet_data); 4156 4157 rcu_read_lock(); 4158 fl = rcu_dereference(sd->flow_limit); 4159 if (fl) { 4160 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4161 old_flow = fl->history[fl->history_head]; 4162 fl->history[fl->history_head] = new_flow; 4163 4164 fl->history_head++; 4165 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4166 4167 if (likely(fl->buckets[old_flow])) 4168 fl->buckets[old_flow]--; 4169 4170 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4171 fl->count++; 4172 rcu_read_unlock(); 4173 return true; 4174 } 4175 } 4176 rcu_read_unlock(); 4177 #endif 4178 return false; 4179 } 4180 4181 /* 4182 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4183 * queue (may be a remote CPU queue). 4184 */ 4185 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4186 unsigned int *qtail) 4187 { 4188 struct softnet_data *sd; 4189 unsigned long flags; 4190 unsigned int qlen; 4191 4192 sd = &per_cpu(softnet_data, cpu); 4193 4194 local_irq_save(flags); 4195 4196 rps_lock(sd); 4197 if (!netif_running(skb->dev)) 4198 goto drop; 4199 qlen = skb_queue_len(&sd->input_pkt_queue); 4200 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4201 if (qlen) { 4202 enqueue: 4203 __skb_queue_tail(&sd->input_pkt_queue, skb); 4204 input_queue_tail_incr_save(sd, qtail); 4205 rps_unlock(sd); 4206 local_irq_restore(flags); 4207 return NET_RX_SUCCESS; 4208 } 4209 4210 /* Schedule NAPI for backlog device 4211 * We can use non atomic operation since we own the queue lock 4212 */ 4213 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4214 if (!rps_ipi_queued(sd)) 4215 ____napi_schedule(sd, &sd->backlog); 4216 } 4217 goto enqueue; 4218 } 4219 4220 drop: 4221 sd->dropped++; 4222 rps_unlock(sd); 4223 4224 local_irq_restore(flags); 4225 4226 atomic_long_inc(&skb->dev->rx_dropped); 4227 kfree_skb(skb); 4228 return NET_RX_DROP; 4229 } 4230 4231 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4232 { 4233 struct net_device *dev = skb->dev; 4234 struct netdev_rx_queue *rxqueue; 4235 4236 rxqueue = dev->_rx; 4237 4238 if (skb_rx_queue_recorded(skb)) { 4239 u16 index = skb_get_rx_queue(skb); 4240 4241 if (unlikely(index >= dev->real_num_rx_queues)) { 4242 WARN_ONCE(dev->real_num_rx_queues > 1, 4243 "%s received packet on queue %u, but number " 4244 "of RX queues is %u\n", 4245 dev->name, index, dev->real_num_rx_queues); 4246 4247 return rxqueue; /* Return first rxqueue */ 4248 } 4249 rxqueue += index; 4250 } 4251 return rxqueue; 4252 } 4253 4254 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4255 struct xdp_buff *xdp, 4256 struct bpf_prog *xdp_prog) 4257 { 4258 struct netdev_rx_queue *rxqueue; 4259 void *orig_data, *orig_data_end; 4260 u32 metalen, act = XDP_DROP; 4261 int hlen, off; 4262 u32 mac_len; 4263 4264 /* Reinjected packets coming from act_mirred or similar should 4265 * not get XDP generic processing. 4266 */ 4267 if (skb_cloned(skb) || skb_is_tc_redirected(skb)) 4268 return XDP_PASS; 4269 4270 /* XDP packets must be linear and must have sufficient headroom 4271 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4272 * native XDP provides, thus we need to do it here as well. 4273 */ 4274 if (skb_is_nonlinear(skb) || 4275 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4276 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4277 int troom = skb->tail + skb->data_len - skb->end; 4278 4279 /* In case we have to go down the path and also linearize, 4280 * then lets do the pskb_expand_head() work just once here. 4281 */ 4282 if (pskb_expand_head(skb, 4283 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4284 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4285 goto do_drop; 4286 if (skb_linearize(skb)) 4287 goto do_drop; 4288 } 4289 4290 /* The XDP program wants to see the packet starting at the MAC 4291 * header. 4292 */ 4293 mac_len = skb->data - skb_mac_header(skb); 4294 hlen = skb_headlen(skb) + mac_len; 4295 xdp->data = skb->data - mac_len; 4296 xdp->data_meta = xdp->data; 4297 xdp->data_end = xdp->data + hlen; 4298 xdp->data_hard_start = skb->data - skb_headroom(skb); 4299 orig_data_end = xdp->data_end; 4300 orig_data = xdp->data; 4301 4302 rxqueue = netif_get_rxqueue(skb); 4303 xdp->rxq = &rxqueue->xdp_rxq; 4304 4305 act = bpf_prog_run_xdp(xdp_prog, xdp); 4306 4307 off = xdp->data - orig_data; 4308 if (off > 0) 4309 __skb_pull(skb, off); 4310 else if (off < 0) 4311 __skb_push(skb, -off); 4312 skb->mac_header += off; 4313 4314 /* check if bpf_xdp_adjust_tail was used. it can only "shrink" 4315 * pckt. 4316 */ 4317 off = orig_data_end - xdp->data_end; 4318 if (off != 0) { 4319 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4320 skb->len -= off; 4321 4322 } 4323 4324 switch (act) { 4325 case XDP_REDIRECT: 4326 case XDP_TX: 4327 __skb_push(skb, mac_len); 4328 break; 4329 case XDP_PASS: 4330 metalen = xdp->data - xdp->data_meta; 4331 if (metalen) 4332 skb_metadata_set(skb, metalen); 4333 break; 4334 default: 4335 bpf_warn_invalid_xdp_action(act); 4336 /* fall through */ 4337 case XDP_ABORTED: 4338 trace_xdp_exception(skb->dev, xdp_prog, act); 4339 /* fall through */ 4340 case XDP_DROP: 4341 do_drop: 4342 kfree_skb(skb); 4343 break; 4344 } 4345 4346 return act; 4347 } 4348 4349 /* When doing generic XDP we have to bypass the qdisc layer and the 4350 * network taps in order to match in-driver-XDP behavior. 4351 */ 4352 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4353 { 4354 struct net_device *dev = skb->dev; 4355 struct netdev_queue *txq; 4356 bool free_skb = true; 4357 int cpu, rc; 4358 4359 txq = netdev_pick_tx(dev, skb, NULL); 4360 cpu = smp_processor_id(); 4361 HARD_TX_LOCK(dev, txq, cpu); 4362 if (!netif_xmit_stopped(txq)) { 4363 rc = netdev_start_xmit(skb, dev, txq, 0); 4364 if (dev_xmit_complete(rc)) 4365 free_skb = false; 4366 } 4367 HARD_TX_UNLOCK(dev, txq); 4368 if (free_skb) { 4369 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4370 kfree_skb(skb); 4371 } 4372 } 4373 EXPORT_SYMBOL_GPL(generic_xdp_tx); 4374 4375 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4376 4377 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4378 { 4379 if (xdp_prog) { 4380 struct xdp_buff xdp; 4381 u32 act; 4382 int err; 4383 4384 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4385 if (act != XDP_PASS) { 4386 switch (act) { 4387 case XDP_REDIRECT: 4388 err = xdp_do_generic_redirect(skb->dev, skb, 4389 &xdp, xdp_prog); 4390 if (err) 4391 goto out_redir; 4392 break; 4393 case XDP_TX: 4394 generic_xdp_tx(skb, xdp_prog); 4395 break; 4396 } 4397 return XDP_DROP; 4398 } 4399 } 4400 return XDP_PASS; 4401 out_redir: 4402 kfree_skb(skb); 4403 return XDP_DROP; 4404 } 4405 EXPORT_SYMBOL_GPL(do_xdp_generic); 4406 4407 static int netif_rx_internal(struct sk_buff *skb) 4408 { 4409 int ret; 4410 4411 net_timestamp_check(netdev_tstamp_prequeue, skb); 4412 4413 trace_netif_rx(skb); 4414 4415 if (static_branch_unlikely(&generic_xdp_needed_key)) { 4416 int ret; 4417 4418 preempt_disable(); 4419 rcu_read_lock(); 4420 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 4421 rcu_read_unlock(); 4422 preempt_enable(); 4423 4424 /* Consider XDP consuming the packet a success from 4425 * the netdev point of view we do not want to count 4426 * this as an error. 4427 */ 4428 if (ret != XDP_PASS) 4429 return NET_RX_SUCCESS; 4430 } 4431 4432 #ifdef CONFIG_RPS 4433 if (static_key_false(&rps_needed)) { 4434 struct rps_dev_flow voidflow, *rflow = &voidflow; 4435 int cpu; 4436 4437 preempt_disable(); 4438 rcu_read_lock(); 4439 4440 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4441 if (cpu < 0) 4442 cpu = smp_processor_id(); 4443 4444 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4445 4446 rcu_read_unlock(); 4447 preempt_enable(); 4448 } else 4449 #endif 4450 { 4451 unsigned int qtail; 4452 4453 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4454 put_cpu(); 4455 } 4456 return ret; 4457 } 4458 4459 /** 4460 * netif_rx - post buffer to the network code 4461 * @skb: buffer to post 4462 * 4463 * This function receives a packet from a device driver and queues it for 4464 * the upper (protocol) levels to process. It always succeeds. The buffer 4465 * may be dropped during processing for congestion control or by the 4466 * protocol layers. 4467 * 4468 * return values: 4469 * NET_RX_SUCCESS (no congestion) 4470 * NET_RX_DROP (packet was dropped) 4471 * 4472 */ 4473 4474 int netif_rx(struct sk_buff *skb) 4475 { 4476 trace_netif_rx_entry(skb); 4477 4478 return netif_rx_internal(skb); 4479 } 4480 EXPORT_SYMBOL(netif_rx); 4481 4482 int netif_rx_ni(struct sk_buff *skb) 4483 { 4484 int err; 4485 4486 trace_netif_rx_ni_entry(skb); 4487 4488 preempt_disable(); 4489 err = netif_rx_internal(skb); 4490 if (local_softirq_pending()) 4491 do_softirq(); 4492 preempt_enable(); 4493 4494 return err; 4495 } 4496 EXPORT_SYMBOL(netif_rx_ni); 4497 4498 static __latent_entropy void net_tx_action(struct softirq_action *h) 4499 { 4500 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4501 4502 if (sd->completion_queue) { 4503 struct sk_buff *clist; 4504 4505 local_irq_disable(); 4506 clist = sd->completion_queue; 4507 sd->completion_queue = NULL; 4508 local_irq_enable(); 4509 4510 while (clist) { 4511 struct sk_buff *skb = clist; 4512 4513 clist = clist->next; 4514 4515 WARN_ON(refcount_read(&skb->users)); 4516 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4517 trace_consume_skb(skb); 4518 else 4519 trace_kfree_skb(skb, net_tx_action); 4520 4521 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4522 __kfree_skb(skb); 4523 else 4524 __kfree_skb_defer(skb); 4525 } 4526 4527 __kfree_skb_flush(); 4528 } 4529 4530 if (sd->output_queue) { 4531 struct Qdisc *head; 4532 4533 local_irq_disable(); 4534 head = sd->output_queue; 4535 sd->output_queue = NULL; 4536 sd->output_queue_tailp = &sd->output_queue; 4537 local_irq_enable(); 4538 4539 while (head) { 4540 struct Qdisc *q = head; 4541 spinlock_t *root_lock = NULL; 4542 4543 head = head->next_sched; 4544 4545 if (!(q->flags & TCQ_F_NOLOCK)) { 4546 root_lock = qdisc_lock(q); 4547 spin_lock(root_lock); 4548 } 4549 /* We need to make sure head->next_sched is read 4550 * before clearing __QDISC_STATE_SCHED 4551 */ 4552 smp_mb__before_atomic(); 4553 clear_bit(__QDISC_STATE_SCHED, &q->state); 4554 qdisc_run(q); 4555 if (root_lock) 4556 spin_unlock(root_lock); 4557 } 4558 } 4559 4560 xfrm_dev_backlog(sd); 4561 } 4562 4563 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4564 /* This hook is defined here for ATM LANE */ 4565 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4566 unsigned char *addr) __read_mostly; 4567 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4568 #endif 4569 4570 static inline struct sk_buff * 4571 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4572 struct net_device *orig_dev) 4573 { 4574 #ifdef CONFIG_NET_CLS_ACT 4575 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4576 struct tcf_result cl_res; 4577 4578 /* If there's at least one ingress present somewhere (so 4579 * we get here via enabled static key), remaining devices 4580 * that are not configured with an ingress qdisc will bail 4581 * out here. 4582 */ 4583 if (!miniq) 4584 return skb; 4585 4586 if (*pt_prev) { 4587 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4588 *pt_prev = NULL; 4589 } 4590 4591 qdisc_skb_cb(skb)->pkt_len = skb->len; 4592 skb->tc_at_ingress = 1; 4593 mini_qdisc_bstats_cpu_update(miniq, skb); 4594 4595 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 4596 case TC_ACT_OK: 4597 case TC_ACT_RECLASSIFY: 4598 skb->tc_index = TC_H_MIN(cl_res.classid); 4599 break; 4600 case TC_ACT_SHOT: 4601 mini_qdisc_qstats_cpu_drop(miniq); 4602 kfree_skb(skb); 4603 return NULL; 4604 case TC_ACT_STOLEN: 4605 case TC_ACT_QUEUED: 4606 case TC_ACT_TRAP: 4607 consume_skb(skb); 4608 return NULL; 4609 case TC_ACT_REDIRECT: 4610 /* skb_mac_header check was done by cls/act_bpf, so 4611 * we can safely push the L2 header back before 4612 * redirecting to another netdev 4613 */ 4614 __skb_push(skb, skb->mac_len); 4615 skb_do_redirect(skb); 4616 return NULL; 4617 case TC_ACT_REINSERT: 4618 /* this does not scrub the packet, and updates stats on error */ 4619 skb_tc_reinsert(skb, &cl_res); 4620 return NULL; 4621 default: 4622 break; 4623 } 4624 #endif /* CONFIG_NET_CLS_ACT */ 4625 return skb; 4626 } 4627 4628 /** 4629 * netdev_is_rx_handler_busy - check if receive handler is registered 4630 * @dev: device to check 4631 * 4632 * Check if a receive handler is already registered for a given device. 4633 * Return true if there one. 4634 * 4635 * The caller must hold the rtnl_mutex. 4636 */ 4637 bool netdev_is_rx_handler_busy(struct net_device *dev) 4638 { 4639 ASSERT_RTNL(); 4640 return dev && rtnl_dereference(dev->rx_handler); 4641 } 4642 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 4643 4644 /** 4645 * netdev_rx_handler_register - register receive handler 4646 * @dev: device to register a handler for 4647 * @rx_handler: receive handler to register 4648 * @rx_handler_data: data pointer that is used by rx handler 4649 * 4650 * Register a receive handler for a device. This handler will then be 4651 * called from __netif_receive_skb. A negative errno code is returned 4652 * on a failure. 4653 * 4654 * The caller must hold the rtnl_mutex. 4655 * 4656 * For a general description of rx_handler, see enum rx_handler_result. 4657 */ 4658 int netdev_rx_handler_register(struct net_device *dev, 4659 rx_handler_func_t *rx_handler, 4660 void *rx_handler_data) 4661 { 4662 if (netdev_is_rx_handler_busy(dev)) 4663 return -EBUSY; 4664 4665 if (dev->priv_flags & IFF_NO_RX_HANDLER) 4666 return -EINVAL; 4667 4668 /* Note: rx_handler_data must be set before rx_handler */ 4669 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 4670 rcu_assign_pointer(dev->rx_handler, rx_handler); 4671 4672 return 0; 4673 } 4674 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 4675 4676 /** 4677 * netdev_rx_handler_unregister - unregister receive handler 4678 * @dev: device to unregister a handler from 4679 * 4680 * Unregister a receive handler from a device. 4681 * 4682 * The caller must hold the rtnl_mutex. 4683 */ 4684 void netdev_rx_handler_unregister(struct net_device *dev) 4685 { 4686 4687 ASSERT_RTNL(); 4688 RCU_INIT_POINTER(dev->rx_handler, NULL); 4689 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 4690 * section has a guarantee to see a non NULL rx_handler_data 4691 * as well. 4692 */ 4693 synchronize_net(); 4694 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 4695 } 4696 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 4697 4698 /* 4699 * Limit the use of PFMEMALLOC reserves to those protocols that implement 4700 * the special handling of PFMEMALLOC skbs. 4701 */ 4702 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 4703 { 4704 switch (skb->protocol) { 4705 case htons(ETH_P_ARP): 4706 case htons(ETH_P_IP): 4707 case htons(ETH_P_IPV6): 4708 case htons(ETH_P_8021Q): 4709 case htons(ETH_P_8021AD): 4710 return true; 4711 default: 4712 return false; 4713 } 4714 } 4715 4716 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 4717 int *ret, struct net_device *orig_dev) 4718 { 4719 #ifdef CONFIG_NETFILTER_INGRESS 4720 if (nf_hook_ingress_active(skb)) { 4721 int ingress_retval; 4722 4723 if (*pt_prev) { 4724 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4725 *pt_prev = NULL; 4726 } 4727 4728 rcu_read_lock(); 4729 ingress_retval = nf_hook_ingress(skb); 4730 rcu_read_unlock(); 4731 return ingress_retval; 4732 } 4733 #endif /* CONFIG_NETFILTER_INGRESS */ 4734 return 0; 4735 } 4736 4737 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, 4738 struct packet_type **ppt_prev) 4739 { 4740 struct packet_type *ptype, *pt_prev; 4741 rx_handler_func_t *rx_handler; 4742 struct net_device *orig_dev; 4743 bool deliver_exact = false; 4744 int ret = NET_RX_DROP; 4745 __be16 type; 4746 4747 net_timestamp_check(!netdev_tstamp_prequeue, skb); 4748 4749 trace_netif_receive_skb(skb); 4750 4751 orig_dev = skb->dev; 4752 4753 skb_reset_network_header(skb); 4754 if (!skb_transport_header_was_set(skb)) 4755 skb_reset_transport_header(skb); 4756 skb_reset_mac_len(skb); 4757 4758 pt_prev = NULL; 4759 4760 another_round: 4761 skb->skb_iif = skb->dev->ifindex; 4762 4763 __this_cpu_inc(softnet_data.processed); 4764 4765 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 4766 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 4767 skb = skb_vlan_untag(skb); 4768 if (unlikely(!skb)) 4769 goto out; 4770 } 4771 4772 if (skb_skip_tc_classify(skb)) 4773 goto skip_classify; 4774 4775 if (pfmemalloc) 4776 goto skip_taps; 4777 4778 list_for_each_entry_rcu(ptype, &ptype_all, list) { 4779 if (pt_prev) 4780 ret = deliver_skb(skb, pt_prev, orig_dev); 4781 pt_prev = ptype; 4782 } 4783 4784 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 4785 if (pt_prev) 4786 ret = deliver_skb(skb, pt_prev, orig_dev); 4787 pt_prev = ptype; 4788 } 4789 4790 skip_taps: 4791 #ifdef CONFIG_NET_INGRESS 4792 if (static_branch_unlikely(&ingress_needed_key)) { 4793 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); 4794 if (!skb) 4795 goto out; 4796 4797 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 4798 goto out; 4799 } 4800 #endif 4801 skb_reset_tc(skb); 4802 skip_classify: 4803 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 4804 goto drop; 4805 4806 if (skb_vlan_tag_present(skb)) { 4807 if (pt_prev) { 4808 ret = deliver_skb(skb, pt_prev, orig_dev); 4809 pt_prev = NULL; 4810 } 4811 if (vlan_do_receive(&skb)) 4812 goto another_round; 4813 else if (unlikely(!skb)) 4814 goto out; 4815 } 4816 4817 rx_handler = rcu_dereference(skb->dev->rx_handler); 4818 if (rx_handler) { 4819 if (pt_prev) { 4820 ret = deliver_skb(skb, pt_prev, orig_dev); 4821 pt_prev = NULL; 4822 } 4823 switch (rx_handler(&skb)) { 4824 case RX_HANDLER_CONSUMED: 4825 ret = NET_RX_SUCCESS; 4826 goto out; 4827 case RX_HANDLER_ANOTHER: 4828 goto another_round; 4829 case RX_HANDLER_EXACT: 4830 deliver_exact = true; 4831 case RX_HANDLER_PASS: 4832 break; 4833 default: 4834 BUG(); 4835 } 4836 } 4837 4838 if (unlikely(skb_vlan_tag_present(skb))) { 4839 if (skb_vlan_tag_get_id(skb)) 4840 skb->pkt_type = PACKET_OTHERHOST; 4841 /* Note: we might in the future use prio bits 4842 * and set skb->priority like in vlan_do_receive() 4843 * For the time being, just ignore Priority Code Point 4844 */ 4845 skb->vlan_tci = 0; 4846 } 4847 4848 type = skb->protocol; 4849 4850 /* deliver only exact match when indicated */ 4851 if (likely(!deliver_exact)) { 4852 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4853 &ptype_base[ntohs(type) & 4854 PTYPE_HASH_MASK]); 4855 } 4856 4857 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4858 &orig_dev->ptype_specific); 4859 4860 if (unlikely(skb->dev != orig_dev)) { 4861 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4862 &skb->dev->ptype_specific); 4863 } 4864 4865 if (pt_prev) { 4866 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 4867 goto drop; 4868 *ppt_prev = pt_prev; 4869 } else { 4870 drop: 4871 if (!deliver_exact) 4872 atomic_long_inc(&skb->dev->rx_dropped); 4873 else 4874 atomic_long_inc(&skb->dev->rx_nohandler); 4875 kfree_skb(skb); 4876 /* Jamal, now you will not able to escape explaining 4877 * me how you were going to use this. :-) 4878 */ 4879 ret = NET_RX_DROP; 4880 } 4881 4882 out: 4883 return ret; 4884 } 4885 4886 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 4887 { 4888 struct net_device *orig_dev = skb->dev; 4889 struct packet_type *pt_prev = NULL; 4890 int ret; 4891 4892 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); 4893 if (pt_prev) 4894 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 4895 return ret; 4896 } 4897 4898 /** 4899 * netif_receive_skb_core - special purpose version of netif_receive_skb 4900 * @skb: buffer to process 4901 * 4902 * More direct receive version of netif_receive_skb(). It should 4903 * only be used by callers that have a need to skip RPS and Generic XDP. 4904 * Caller must also take care of handling if (page_is_)pfmemalloc. 4905 * 4906 * This function may only be called from softirq context and interrupts 4907 * should be enabled. 4908 * 4909 * Return values (usually ignored): 4910 * NET_RX_SUCCESS: no congestion 4911 * NET_RX_DROP: packet was dropped 4912 */ 4913 int netif_receive_skb_core(struct sk_buff *skb) 4914 { 4915 int ret; 4916 4917 rcu_read_lock(); 4918 ret = __netif_receive_skb_one_core(skb, false); 4919 rcu_read_unlock(); 4920 4921 return ret; 4922 } 4923 EXPORT_SYMBOL(netif_receive_skb_core); 4924 4925 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 4926 struct packet_type *pt_prev, 4927 struct net_device *orig_dev) 4928 { 4929 struct sk_buff *skb, *next; 4930 4931 if (!pt_prev) 4932 return; 4933 if (list_empty(head)) 4934 return; 4935 if (pt_prev->list_func != NULL) 4936 pt_prev->list_func(head, pt_prev, orig_dev); 4937 else 4938 list_for_each_entry_safe(skb, next, head, list) 4939 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 4940 } 4941 4942 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 4943 { 4944 /* Fast-path assumptions: 4945 * - There is no RX handler. 4946 * - Only one packet_type matches. 4947 * If either of these fails, we will end up doing some per-packet 4948 * processing in-line, then handling the 'last ptype' for the whole 4949 * sublist. This can't cause out-of-order delivery to any single ptype, 4950 * because the 'last ptype' must be constant across the sublist, and all 4951 * other ptypes are handled per-packet. 4952 */ 4953 /* Current (common) ptype of sublist */ 4954 struct packet_type *pt_curr = NULL; 4955 /* Current (common) orig_dev of sublist */ 4956 struct net_device *od_curr = NULL; 4957 struct list_head sublist; 4958 struct sk_buff *skb, *next; 4959 4960 INIT_LIST_HEAD(&sublist); 4961 list_for_each_entry_safe(skb, next, head, list) { 4962 struct net_device *orig_dev = skb->dev; 4963 struct packet_type *pt_prev = NULL; 4964 4965 list_del(&skb->list); 4966 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); 4967 if (!pt_prev) 4968 continue; 4969 if (pt_curr != pt_prev || od_curr != orig_dev) { 4970 /* dispatch old sublist */ 4971 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 4972 /* start new sublist */ 4973 INIT_LIST_HEAD(&sublist); 4974 pt_curr = pt_prev; 4975 od_curr = orig_dev; 4976 } 4977 list_add_tail(&skb->list, &sublist); 4978 } 4979 4980 /* dispatch final sublist */ 4981 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 4982 } 4983 4984 static int __netif_receive_skb(struct sk_buff *skb) 4985 { 4986 int ret; 4987 4988 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 4989 unsigned int noreclaim_flag; 4990 4991 /* 4992 * PFMEMALLOC skbs are special, they should 4993 * - be delivered to SOCK_MEMALLOC sockets only 4994 * - stay away from userspace 4995 * - have bounded memory usage 4996 * 4997 * Use PF_MEMALLOC as this saves us from propagating the allocation 4998 * context down to all allocation sites. 4999 */ 5000 noreclaim_flag = memalloc_noreclaim_save(); 5001 ret = __netif_receive_skb_one_core(skb, true); 5002 memalloc_noreclaim_restore(noreclaim_flag); 5003 } else 5004 ret = __netif_receive_skb_one_core(skb, false); 5005 5006 return ret; 5007 } 5008 5009 static void __netif_receive_skb_list(struct list_head *head) 5010 { 5011 unsigned long noreclaim_flag = 0; 5012 struct sk_buff *skb, *next; 5013 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5014 5015 list_for_each_entry_safe(skb, next, head, list) { 5016 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5017 struct list_head sublist; 5018 5019 /* Handle the previous sublist */ 5020 list_cut_before(&sublist, head, &skb->list); 5021 if (!list_empty(&sublist)) 5022 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5023 pfmemalloc = !pfmemalloc; 5024 /* See comments in __netif_receive_skb */ 5025 if (pfmemalloc) 5026 noreclaim_flag = memalloc_noreclaim_save(); 5027 else 5028 memalloc_noreclaim_restore(noreclaim_flag); 5029 } 5030 } 5031 /* Handle the remaining sublist */ 5032 if (!list_empty(head)) 5033 __netif_receive_skb_list_core(head, pfmemalloc); 5034 /* Restore pflags */ 5035 if (pfmemalloc) 5036 memalloc_noreclaim_restore(noreclaim_flag); 5037 } 5038 5039 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5040 { 5041 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5042 struct bpf_prog *new = xdp->prog; 5043 int ret = 0; 5044 5045 switch (xdp->command) { 5046 case XDP_SETUP_PROG: 5047 rcu_assign_pointer(dev->xdp_prog, new); 5048 if (old) 5049 bpf_prog_put(old); 5050 5051 if (old && !new) { 5052 static_branch_dec(&generic_xdp_needed_key); 5053 } else if (new && !old) { 5054 static_branch_inc(&generic_xdp_needed_key); 5055 dev_disable_lro(dev); 5056 dev_disable_gro_hw(dev); 5057 } 5058 break; 5059 5060 case XDP_QUERY_PROG: 5061 xdp->prog_id = old ? old->aux->id : 0; 5062 break; 5063 5064 default: 5065 ret = -EINVAL; 5066 break; 5067 } 5068 5069 return ret; 5070 } 5071 5072 static int netif_receive_skb_internal(struct sk_buff *skb) 5073 { 5074 int ret; 5075 5076 net_timestamp_check(netdev_tstamp_prequeue, skb); 5077 5078 if (skb_defer_rx_timestamp(skb)) 5079 return NET_RX_SUCCESS; 5080 5081 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5082 int ret; 5083 5084 preempt_disable(); 5085 rcu_read_lock(); 5086 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5087 rcu_read_unlock(); 5088 preempt_enable(); 5089 5090 if (ret != XDP_PASS) 5091 return NET_RX_DROP; 5092 } 5093 5094 rcu_read_lock(); 5095 #ifdef CONFIG_RPS 5096 if (static_key_false(&rps_needed)) { 5097 struct rps_dev_flow voidflow, *rflow = &voidflow; 5098 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5099 5100 if (cpu >= 0) { 5101 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5102 rcu_read_unlock(); 5103 return ret; 5104 } 5105 } 5106 #endif 5107 ret = __netif_receive_skb(skb); 5108 rcu_read_unlock(); 5109 return ret; 5110 } 5111 5112 static void netif_receive_skb_list_internal(struct list_head *head) 5113 { 5114 struct bpf_prog *xdp_prog = NULL; 5115 struct sk_buff *skb, *next; 5116 struct list_head sublist; 5117 5118 INIT_LIST_HEAD(&sublist); 5119 list_for_each_entry_safe(skb, next, head, list) { 5120 net_timestamp_check(netdev_tstamp_prequeue, skb); 5121 list_del(&skb->list); 5122 if (!skb_defer_rx_timestamp(skb)) 5123 list_add_tail(&skb->list, &sublist); 5124 } 5125 list_splice_init(&sublist, head); 5126 5127 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5128 preempt_disable(); 5129 rcu_read_lock(); 5130 list_for_each_entry_safe(skb, next, head, list) { 5131 xdp_prog = rcu_dereference(skb->dev->xdp_prog); 5132 list_del(&skb->list); 5133 if (do_xdp_generic(xdp_prog, skb) == XDP_PASS) 5134 list_add_tail(&skb->list, &sublist); 5135 } 5136 rcu_read_unlock(); 5137 preempt_enable(); 5138 /* Put passed packets back on main list */ 5139 list_splice_init(&sublist, head); 5140 } 5141 5142 rcu_read_lock(); 5143 #ifdef CONFIG_RPS 5144 if (static_key_false(&rps_needed)) { 5145 list_for_each_entry_safe(skb, next, head, list) { 5146 struct rps_dev_flow voidflow, *rflow = &voidflow; 5147 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5148 5149 if (cpu >= 0) { 5150 /* Will be handled, remove from list */ 5151 list_del(&skb->list); 5152 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5153 } 5154 } 5155 } 5156 #endif 5157 __netif_receive_skb_list(head); 5158 rcu_read_unlock(); 5159 } 5160 5161 /** 5162 * netif_receive_skb - process receive buffer from network 5163 * @skb: buffer to process 5164 * 5165 * netif_receive_skb() is the main receive data processing function. 5166 * It always succeeds. The buffer may be dropped during processing 5167 * for congestion control or by the protocol layers. 5168 * 5169 * This function may only be called from softirq context and interrupts 5170 * should be enabled. 5171 * 5172 * Return values (usually ignored): 5173 * NET_RX_SUCCESS: no congestion 5174 * NET_RX_DROP: packet was dropped 5175 */ 5176 int netif_receive_skb(struct sk_buff *skb) 5177 { 5178 trace_netif_receive_skb_entry(skb); 5179 5180 return netif_receive_skb_internal(skb); 5181 } 5182 EXPORT_SYMBOL(netif_receive_skb); 5183 5184 /** 5185 * netif_receive_skb_list - process many receive buffers from network 5186 * @head: list of skbs to process. 5187 * 5188 * Since return value of netif_receive_skb() is normally ignored, and 5189 * wouldn't be meaningful for a list, this function returns void. 5190 * 5191 * This function may only be called from softirq context and interrupts 5192 * should be enabled. 5193 */ 5194 void netif_receive_skb_list(struct list_head *head) 5195 { 5196 struct sk_buff *skb; 5197 5198 if (list_empty(head)) 5199 return; 5200 list_for_each_entry(skb, head, list) 5201 trace_netif_receive_skb_list_entry(skb); 5202 netif_receive_skb_list_internal(head); 5203 } 5204 EXPORT_SYMBOL(netif_receive_skb_list); 5205 5206 DEFINE_PER_CPU(struct work_struct, flush_works); 5207 5208 /* Network device is going away, flush any packets still pending */ 5209 static void flush_backlog(struct work_struct *work) 5210 { 5211 struct sk_buff *skb, *tmp; 5212 struct softnet_data *sd; 5213 5214 local_bh_disable(); 5215 sd = this_cpu_ptr(&softnet_data); 5216 5217 local_irq_disable(); 5218 rps_lock(sd); 5219 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5220 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5221 __skb_unlink(skb, &sd->input_pkt_queue); 5222 kfree_skb(skb); 5223 input_queue_head_incr(sd); 5224 } 5225 } 5226 rps_unlock(sd); 5227 local_irq_enable(); 5228 5229 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5230 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5231 __skb_unlink(skb, &sd->process_queue); 5232 kfree_skb(skb); 5233 input_queue_head_incr(sd); 5234 } 5235 } 5236 local_bh_enable(); 5237 } 5238 5239 static void flush_all_backlogs(void) 5240 { 5241 unsigned int cpu; 5242 5243 get_online_cpus(); 5244 5245 for_each_online_cpu(cpu) 5246 queue_work_on(cpu, system_highpri_wq, 5247 per_cpu_ptr(&flush_works, cpu)); 5248 5249 for_each_online_cpu(cpu) 5250 flush_work(per_cpu_ptr(&flush_works, cpu)); 5251 5252 put_online_cpus(); 5253 } 5254 5255 static int napi_gro_complete(struct sk_buff *skb) 5256 { 5257 struct packet_offload *ptype; 5258 __be16 type = skb->protocol; 5259 struct list_head *head = &offload_base; 5260 int err = -ENOENT; 5261 5262 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5263 5264 if (NAPI_GRO_CB(skb)->count == 1) { 5265 skb_shinfo(skb)->gso_size = 0; 5266 goto out; 5267 } 5268 5269 rcu_read_lock(); 5270 list_for_each_entry_rcu(ptype, head, list) { 5271 if (ptype->type != type || !ptype->callbacks.gro_complete) 5272 continue; 5273 5274 err = ptype->callbacks.gro_complete(skb, 0); 5275 break; 5276 } 5277 rcu_read_unlock(); 5278 5279 if (err) { 5280 WARN_ON(&ptype->list == head); 5281 kfree_skb(skb); 5282 return NET_RX_SUCCESS; 5283 } 5284 5285 out: 5286 return netif_receive_skb_internal(skb); 5287 } 5288 5289 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5290 bool flush_old) 5291 { 5292 struct list_head *head = &napi->gro_hash[index].list; 5293 struct sk_buff *skb, *p; 5294 5295 list_for_each_entry_safe_reverse(skb, p, head, list) { 5296 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5297 return; 5298 skb_list_del_init(skb); 5299 napi_gro_complete(skb); 5300 napi->gro_hash[index].count--; 5301 } 5302 5303 if (!napi->gro_hash[index].count) 5304 __clear_bit(index, &napi->gro_bitmask); 5305 } 5306 5307 /* napi->gro_hash[].list contains packets ordered by age. 5308 * youngest packets at the head of it. 5309 * Complete skbs in reverse order to reduce latencies. 5310 */ 5311 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5312 { 5313 u32 i; 5314 5315 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 5316 if (test_bit(i, &napi->gro_bitmask)) 5317 __napi_gro_flush_chain(napi, i, flush_old); 5318 } 5319 } 5320 EXPORT_SYMBOL(napi_gro_flush); 5321 5322 static struct list_head *gro_list_prepare(struct napi_struct *napi, 5323 struct sk_buff *skb) 5324 { 5325 unsigned int maclen = skb->dev->hard_header_len; 5326 u32 hash = skb_get_hash_raw(skb); 5327 struct list_head *head; 5328 struct sk_buff *p; 5329 5330 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; 5331 list_for_each_entry(p, head, list) { 5332 unsigned long diffs; 5333 5334 NAPI_GRO_CB(p)->flush = 0; 5335 5336 if (hash != skb_get_hash_raw(p)) { 5337 NAPI_GRO_CB(p)->same_flow = 0; 5338 continue; 5339 } 5340 5341 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5342 diffs |= p->vlan_tci ^ skb->vlan_tci; 5343 diffs |= skb_metadata_dst_cmp(p, skb); 5344 diffs |= skb_metadata_differs(p, skb); 5345 if (maclen == ETH_HLEN) 5346 diffs |= compare_ether_header(skb_mac_header(p), 5347 skb_mac_header(skb)); 5348 else if (!diffs) 5349 diffs = memcmp(skb_mac_header(p), 5350 skb_mac_header(skb), 5351 maclen); 5352 NAPI_GRO_CB(p)->same_flow = !diffs; 5353 } 5354 5355 return head; 5356 } 5357 5358 static void skb_gro_reset_offset(struct sk_buff *skb) 5359 { 5360 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5361 const skb_frag_t *frag0 = &pinfo->frags[0]; 5362 5363 NAPI_GRO_CB(skb)->data_offset = 0; 5364 NAPI_GRO_CB(skb)->frag0 = NULL; 5365 NAPI_GRO_CB(skb)->frag0_len = 0; 5366 5367 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 5368 pinfo->nr_frags && 5369 !PageHighMem(skb_frag_page(frag0))) { 5370 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5371 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5372 skb_frag_size(frag0), 5373 skb->end - skb->tail); 5374 } 5375 } 5376 5377 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5378 { 5379 struct skb_shared_info *pinfo = skb_shinfo(skb); 5380 5381 BUG_ON(skb->end - skb->tail < grow); 5382 5383 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5384 5385 skb->data_len -= grow; 5386 skb->tail += grow; 5387 5388 pinfo->frags[0].page_offset += grow; 5389 skb_frag_size_sub(&pinfo->frags[0], grow); 5390 5391 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 5392 skb_frag_unref(skb, 0); 5393 memmove(pinfo->frags, pinfo->frags + 1, 5394 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 5395 } 5396 } 5397 5398 static void gro_flush_oldest(struct list_head *head) 5399 { 5400 struct sk_buff *oldest; 5401 5402 oldest = list_last_entry(head, struct sk_buff, list); 5403 5404 /* We are called with head length >= MAX_GRO_SKBS, so this is 5405 * impossible. 5406 */ 5407 if (WARN_ON_ONCE(!oldest)) 5408 return; 5409 5410 /* Do not adjust napi->gro_hash[].count, caller is adding a new 5411 * SKB to the chain. 5412 */ 5413 list_del(&oldest->list); 5414 napi_gro_complete(oldest); 5415 } 5416 5417 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5418 { 5419 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 5420 struct list_head *head = &offload_base; 5421 struct packet_offload *ptype; 5422 __be16 type = skb->protocol; 5423 struct list_head *gro_head; 5424 struct sk_buff *pp = NULL; 5425 enum gro_result ret; 5426 int same_flow; 5427 int grow; 5428 5429 if (netif_elide_gro(skb->dev)) 5430 goto normal; 5431 5432 gro_head = gro_list_prepare(napi, skb); 5433 5434 rcu_read_lock(); 5435 list_for_each_entry_rcu(ptype, head, list) { 5436 if (ptype->type != type || !ptype->callbacks.gro_receive) 5437 continue; 5438 5439 skb_set_network_header(skb, skb_gro_offset(skb)); 5440 skb_reset_mac_len(skb); 5441 NAPI_GRO_CB(skb)->same_flow = 0; 5442 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 5443 NAPI_GRO_CB(skb)->free = 0; 5444 NAPI_GRO_CB(skb)->encap_mark = 0; 5445 NAPI_GRO_CB(skb)->recursion_counter = 0; 5446 NAPI_GRO_CB(skb)->is_fou = 0; 5447 NAPI_GRO_CB(skb)->is_atomic = 1; 5448 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 5449 5450 /* Setup for GRO checksum validation */ 5451 switch (skb->ip_summed) { 5452 case CHECKSUM_COMPLETE: 5453 NAPI_GRO_CB(skb)->csum = skb->csum; 5454 NAPI_GRO_CB(skb)->csum_valid = 1; 5455 NAPI_GRO_CB(skb)->csum_cnt = 0; 5456 break; 5457 case CHECKSUM_UNNECESSARY: 5458 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 5459 NAPI_GRO_CB(skb)->csum_valid = 0; 5460 break; 5461 default: 5462 NAPI_GRO_CB(skb)->csum_cnt = 0; 5463 NAPI_GRO_CB(skb)->csum_valid = 0; 5464 } 5465 5466 pp = ptype->callbacks.gro_receive(gro_head, skb); 5467 break; 5468 } 5469 rcu_read_unlock(); 5470 5471 if (&ptype->list == head) 5472 goto normal; 5473 5474 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { 5475 ret = GRO_CONSUMED; 5476 goto ok; 5477 } 5478 5479 same_flow = NAPI_GRO_CB(skb)->same_flow; 5480 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 5481 5482 if (pp) { 5483 skb_list_del_init(pp); 5484 napi_gro_complete(pp); 5485 napi->gro_hash[hash].count--; 5486 } 5487 5488 if (same_flow) 5489 goto ok; 5490 5491 if (NAPI_GRO_CB(skb)->flush) 5492 goto normal; 5493 5494 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { 5495 gro_flush_oldest(gro_head); 5496 } else { 5497 napi->gro_hash[hash].count++; 5498 } 5499 NAPI_GRO_CB(skb)->count = 1; 5500 NAPI_GRO_CB(skb)->age = jiffies; 5501 NAPI_GRO_CB(skb)->last = skb; 5502 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 5503 list_add(&skb->list, gro_head); 5504 ret = GRO_HELD; 5505 5506 pull: 5507 grow = skb_gro_offset(skb) - skb_headlen(skb); 5508 if (grow > 0) 5509 gro_pull_from_frag0(skb, grow); 5510 ok: 5511 if (napi->gro_hash[hash].count) { 5512 if (!test_bit(hash, &napi->gro_bitmask)) 5513 __set_bit(hash, &napi->gro_bitmask); 5514 } else if (test_bit(hash, &napi->gro_bitmask)) { 5515 __clear_bit(hash, &napi->gro_bitmask); 5516 } 5517 5518 return ret; 5519 5520 normal: 5521 ret = GRO_NORMAL; 5522 goto pull; 5523 } 5524 5525 struct packet_offload *gro_find_receive_by_type(__be16 type) 5526 { 5527 struct list_head *offload_head = &offload_base; 5528 struct packet_offload *ptype; 5529 5530 list_for_each_entry_rcu(ptype, offload_head, list) { 5531 if (ptype->type != type || !ptype->callbacks.gro_receive) 5532 continue; 5533 return ptype; 5534 } 5535 return NULL; 5536 } 5537 EXPORT_SYMBOL(gro_find_receive_by_type); 5538 5539 struct packet_offload *gro_find_complete_by_type(__be16 type) 5540 { 5541 struct list_head *offload_head = &offload_base; 5542 struct packet_offload *ptype; 5543 5544 list_for_each_entry_rcu(ptype, offload_head, list) { 5545 if (ptype->type != type || !ptype->callbacks.gro_complete) 5546 continue; 5547 return ptype; 5548 } 5549 return NULL; 5550 } 5551 EXPORT_SYMBOL(gro_find_complete_by_type); 5552 5553 static void napi_skb_free_stolen_head(struct sk_buff *skb) 5554 { 5555 skb_dst_drop(skb); 5556 secpath_reset(skb); 5557 kmem_cache_free(skbuff_head_cache, skb); 5558 } 5559 5560 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 5561 { 5562 switch (ret) { 5563 case GRO_NORMAL: 5564 if (netif_receive_skb_internal(skb)) 5565 ret = GRO_DROP; 5566 break; 5567 5568 case GRO_DROP: 5569 kfree_skb(skb); 5570 break; 5571 5572 case GRO_MERGED_FREE: 5573 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 5574 napi_skb_free_stolen_head(skb); 5575 else 5576 __kfree_skb(skb); 5577 break; 5578 5579 case GRO_HELD: 5580 case GRO_MERGED: 5581 case GRO_CONSUMED: 5582 break; 5583 } 5584 5585 return ret; 5586 } 5587 5588 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5589 { 5590 skb_mark_napi_id(skb, napi); 5591 trace_napi_gro_receive_entry(skb); 5592 5593 skb_gro_reset_offset(skb); 5594 5595 return napi_skb_finish(dev_gro_receive(napi, skb), skb); 5596 } 5597 EXPORT_SYMBOL(napi_gro_receive); 5598 5599 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 5600 { 5601 if (unlikely(skb->pfmemalloc)) { 5602 consume_skb(skb); 5603 return; 5604 } 5605 __skb_pull(skb, skb_headlen(skb)); 5606 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 5607 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 5608 skb->vlan_tci = 0; 5609 skb->dev = napi->dev; 5610 skb->skb_iif = 0; 5611 skb->encapsulation = 0; 5612 skb_shinfo(skb)->gso_type = 0; 5613 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5614 secpath_reset(skb); 5615 5616 napi->skb = skb; 5617 } 5618 5619 struct sk_buff *napi_get_frags(struct napi_struct *napi) 5620 { 5621 struct sk_buff *skb = napi->skb; 5622 5623 if (!skb) { 5624 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 5625 if (skb) { 5626 napi->skb = skb; 5627 skb_mark_napi_id(skb, napi); 5628 } 5629 } 5630 return skb; 5631 } 5632 EXPORT_SYMBOL(napi_get_frags); 5633 5634 static gro_result_t napi_frags_finish(struct napi_struct *napi, 5635 struct sk_buff *skb, 5636 gro_result_t ret) 5637 { 5638 switch (ret) { 5639 case GRO_NORMAL: 5640 case GRO_HELD: 5641 __skb_push(skb, ETH_HLEN); 5642 skb->protocol = eth_type_trans(skb, skb->dev); 5643 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 5644 ret = GRO_DROP; 5645 break; 5646 5647 case GRO_DROP: 5648 napi_reuse_skb(napi, skb); 5649 break; 5650 5651 case GRO_MERGED_FREE: 5652 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 5653 napi_skb_free_stolen_head(skb); 5654 else 5655 napi_reuse_skb(napi, skb); 5656 break; 5657 5658 case GRO_MERGED: 5659 case GRO_CONSUMED: 5660 break; 5661 } 5662 5663 return ret; 5664 } 5665 5666 /* Upper GRO stack assumes network header starts at gro_offset=0 5667 * Drivers could call both napi_gro_frags() and napi_gro_receive() 5668 * We copy ethernet header into skb->data to have a common layout. 5669 */ 5670 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 5671 { 5672 struct sk_buff *skb = napi->skb; 5673 const struct ethhdr *eth; 5674 unsigned int hlen = sizeof(*eth); 5675 5676 napi->skb = NULL; 5677 5678 skb_reset_mac_header(skb); 5679 skb_gro_reset_offset(skb); 5680 5681 eth = skb_gro_header_fast(skb, 0); 5682 if (unlikely(skb_gro_header_hard(skb, hlen))) { 5683 eth = skb_gro_header_slow(skb, hlen, 0); 5684 if (unlikely(!eth)) { 5685 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 5686 __func__, napi->dev->name); 5687 napi_reuse_skb(napi, skb); 5688 return NULL; 5689 } 5690 } else { 5691 gro_pull_from_frag0(skb, hlen); 5692 NAPI_GRO_CB(skb)->frag0 += hlen; 5693 NAPI_GRO_CB(skb)->frag0_len -= hlen; 5694 } 5695 __skb_pull(skb, hlen); 5696 5697 /* 5698 * This works because the only protocols we care about don't require 5699 * special handling. 5700 * We'll fix it up properly in napi_frags_finish() 5701 */ 5702 skb->protocol = eth->h_proto; 5703 5704 return skb; 5705 } 5706 5707 gro_result_t napi_gro_frags(struct napi_struct *napi) 5708 { 5709 struct sk_buff *skb = napi_frags_skb(napi); 5710 5711 if (!skb) 5712 return GRO_DROP; 5713 5714 trace_napi_gro_frags_entry(skb); 5715 5716 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 5717 } 5718 EXPORT_SYMBOL(napi_gro_frags); 5719 5720 /* Compute the checksum from gro_offset and return the folded value 5721 * after adding in any pseudo checksum. 5722 */ 5723 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 5724 { 5725 __wsum wsum; 5726 __sum16 sum; 5727 5728 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 5729 5730 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 5731 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 5732 if (likely(!sum)) { 5733 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 5734 !skb->csum_complete_sw) 5735 netdev_rx_csum_fault(skb->dev); 5736 } 5737 5738 NAPI_GRO_CB(skb)->csum = wsum; 5739 NAPI_GRO_CB(skb)->csum_valid = 1; 5740 5741 return sum; 5742 } 5743 EXPORT_SYMBOL(__skb_gro_checksum_complete); 5744 5745 static void net_rps_send_ipi(struct softnet_data *remsd) 5746 { 5747 #ifdef CONFIG_RPS 5748 while (remsd) { 5749 struct softnet_data *next = remsd->rps_ipi_next; 5750 5751 if (cpu_online(remsd->cpu)) 5752 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5753 remsd = next; 5754 } 5755 #endif 5756 } 5757 5758 /* 5759 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5760 * Note: called with local irq disabled, but exits with local irq enabled. 5761 */ 5762 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5763 { 5764 #ifdef CONFIG_RPS 5765 struct softnet_data *remsd = sd->rps_ipi_list; 5766 5767 if (remsd) { 5768 sd->rps_ipi_list = NULL; 5769 5770 local_irq_enable(); 5771 5772 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5773 net_rps_send_ipi(remsd); 5774 } else 5775 #endif 5776 local_irq_enable(); 5777 } 5778 5779 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5780 { 5781 #ifdef CONFIG_RPS 5782 return sd->rps_ipi_list != NULL; 5783 #else 5784 return false; 5785 #endif 5786 } 5787 5788 static int process_backlog(struct napi_struct *napi, int quota) 5789 { 5790 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5791 bool again = true; 5792 int work = 0; 5793 5794 /* Check if we have pending ipi, its better to send them now, 5795 * not waiting net_rx_action() end. 5796 */ 5797 if (sd_has_rps_ipi_waiting(sd)) { 5798 local_irq_disable(); 5799 net_rps_action_and_irq_enable(sd); 5800 } 5801 5802 napi->weight = dev_rx_weight; 5803 while (again) { 5804 struct sk_buff *skb; 5805 5806 while ((skb = __skb_dequeue(&sd->process_queue))) { 5807 rcu_read_lock(); 5808 __netif_receive_skb(skb); 5809 rcu_read_unlock(); 5810 input_queue_head_incr(sd); 5811 if (++work >= quota) 5812 return work; 5813 5814 } 5815 5816 local_irq_disable(); 5817 rps_lock(sd); 5818 if (skb_queue_empty(&sd->input_pkt_queue)) { 5819 /* 5820 * Inline a custom version of __napi_complete(). 5821 * only current cpu owns and manipulates this napi, 5822 * and NAPI_STATE_SCHED is the only possible flag set 5823 * on backlog. 5824 * We can use a plain write instead of clear_bit(), 5825 * and we dont need an smp_mb() memory barrier. 5826 */ 5827 napi->state = 0; 5828 again = false; 5829 } else { 5830 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5831 &sd->process_queue); 5832 } 5833 rps_unlock(sd); 5834 local_irq_enable(); 5835 } 5836 5837 return work; 5838 } 5839 5840 /** 5841 * __napi_schedule - schedule for receive 5842 * @n: entry to schedule 5843 * 5844 * The entry's receive function will be scheduled to run. 5845 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5846 */ 5847 void __napi_schedule(struct napi_struct *n) 5848 { 5849 unsigned long flags; 5850 5851 local_irq_save(flags); 5852 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5853 local_irq_restore(flags); 5854 } 5855 EXPORT_SYMBOL(__napi_schedule); 5856 5857 /** 5858 * napi_schedule_prep - check if napi can be scheduled 5859 * @n: napi context 5860 * 5861 * Test if NAPI routine is already running, and if not mark 5862 * it as running. This is used as a condition variable 5863 * insure only one NAPI poll instance runs. We also make 5864 * sure there is no pending NAPI disable. 5865 */ 5866 bool napi_schedule_prep(struct napi_struct *n) 5867 { 5868 unsigned long val, new; 5869 5870 do { 5871 val = READ_ONCE(n->state); 5872 if (unlikely(val & NAPIF_STATE_DISABLE)) 5873 return false; 5874 new = val | NAPIF_STATE_SCHED; 5875 5876 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5877 * This was suggested by Alexander Duyck, as compiler 5878 * emits better code than : 5879 * if (val & NAPIF_STATE_SCHED) 5880 * new |= NAPIF_STATE_MISSED; 5881 */ 5882 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5883 NAPIF_STATE_MISSED; 5884 } while (cmpxchg(&n->state, val, new) != val); 5885 5886 return !(val & NAPIF_STATE_SCHED); 5887 } 5888 EXPORT_SYMBOL(napi_schedule_prep); 5889 5890 /** 5891 * __napi_schedule_irqoff - schedule for receive 5892 * @n: entry to schedule 5893 * 5894 * Variant of __napi_schedule() assuming hard irqs are masked 5895 */ 5896 void __napi_schedule_irqoff(struct napi_struct *n) 5897 { 5898 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5899 } 5900 EXPORT_SYMBOL(__napi_schedule_irqoff); 5901 5902 bool napi_complete_done(struct napi_struct *n, int work_done) 5903 { 5904 unsigned long flags, val, new; 5905 5906 /* 5907 * 1) Don't let napi dequeue from the cpu poll list 5908 * just in case its running on a different cpu. 5909 * 2) If we are busy polling, do nothing here, we have 5910 * the guarantee we will be called later. 5911 */ 5912 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5913 NAPIF_STATE_IN_BUSY_POLL))) 5914 return false; 5915 5916 if (n->gro_bitmask) { 5917 unsigned long timeout = 0; 5918 5919 if (work_done) 5920 timeout = n->dev->gro_flush_timeout; 5921 5922 if (timeout) 5923 hrtimer_start(&n->timer, ns_to_ktime(timeout), 5924 HRTIMER_MODE_REL_PINNED); 5925 else 5926 napi_gro_flush(n, false); 5927 } 5928 if (unlikely(!list_empty(&n->poll_list))) { 5929 /* If n->poll_list is not empty, we need to mask irqs */ 5930 local_irq_save(flags); 5931 list_del_init(&n->poll_list); 5932 local_irq_restore(flags); 5933 } 5934 5935 do { 5936 val = READ_ONCE(n->state); 5937 5938 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 5939 5940 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); 5941 5942 /* If STATE_MISSED was set, leave STATE_SCHED set, 5943 * because we will call napi->poll() one more time. 5944 * This C code was suggested by Alexander Duyck to help gcc. 5945 */ 5946 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 5947 NAPIF_STATE_SCHED; 5948 } while (cmpxchg(&n->state, val, new) != val); 5949 5950 if (unlikely(val & NAPIF_STATE_MISSED)) { 5951 __napi_schedule(n); 5952 return false; 5953 } 5954 5955 return true; 5956 } 5957 EXPORT_SYMBOL(napi_complete_done); 5958 5959 /* must be called under rcu_read_lock(), as we dont take a reference */ 5960 static struct napi_struct *napi_by_id(unsigned int napi_id) 5961 { 5962 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 5963 struct napi_struct *napi; 5964 5965 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 5966 if (napi->napi_id == napi_id) 5967 return napi; 5968 5969 return NULL; 5970 } 5971 5972 #if defined(CONFIG_NET_RX_BUSY_POLL) 5973 5974 #define BUSY_POLL_BUDGET 8 5975 5976 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) 5977 { 5978 int rc; 5979 5980 /* Busy polling means there is a high chance device driver hard irq 5981 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 5982 * set in napi_schedule_prep(). 5983 * Since we are about to call napi->poll() once more, we can safely 5984 * clear NAPI_STATE_MISSED. 5985 * 5986 * Note: x86 could use a single "lock and ..." instruction 5987 * to perform these two clear_bit() 5988 */ 5989 clear_bit(NAPI_STATE_MISSED, &napi->state); 5990 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 5991 5992 local_bh_disable(); 5993 5994 /* All we really want here is to re-enable device interrupts. 5995 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 5996 */ 5997 rc = napi->poll(napi, BUSY_POLL_BUDGET); 5998 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); 5999 netpoll_poll_unlock(have_poll_lock); 6000 if (rc == BUSY_POLL_BUDGET) 6001 __napi_schedule(napi); 6002 local_bh_enable(); 6003 } 6004 6005 void napi_busy_loop(unsigned int napi_id, 6006 bool (*loop_end)(void *, unsigned long), 6007 void *loop_end_arg) 6008 { 6009 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6010 int (*napi_poll)(struct napi_struct *napi, int budget); 6011 void *have_poll_lock = NULL; 6012 struct napi_struct *napi; 6013 6014 restart: 6015 napi_poll = NULL; 6016 6017 rcu_read_lock(); 6018 6019 napi = napi_by_id(napi_id); 6020 if (!napi) 6021 goto out; 6022 6023 preempt_disable(); 6024 for (;;) { 6025 int work = 0; 6026 6027 local_bh_disable(); 6028 if (!napi_poll) { 6029 unsigned long val = READ_ONCE(napi->state); 6030 6031 /* If multiple threads are competing for this napi, 6032 * we avoid dirtying napi->state as much as we can. 6033 */ 6034 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6035 NAPIF_STATE_IN_BUSY_POLL)) 6036 goto count; 6037 if (cmpxchg(&napi->state, val, 6038 val | NAPIF_STATE_IN_BUSY_POLL | 6039 NAPIF_STATE_SCHED) != val) 6040 goto count; 6041 have_poll_lock = netpoll_poll_lock(napi); 6042 napi_poll = napi->poll; 6043 } 6044 work = napi_poll(napi, BUSY_POLL_BUDGET); 6045 trace_napi_poll(napi, work, BUSY_POLL_BUDGET); 6046 count: 6047 if (work > 0) 6048 __NET_ADD_STATS(dev_net(napi->dev), 6049 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6050 local_bh_enable(); 6051 6052 if (!loop_end || loop_end(loop_end_arg, start_time)) 6053 break; 6054 6055 if (unlikely(need_resched())) { 6056 if (napi_poll) 6057 busy_poll_stop(napi, have_poll_lock); 6058 preempt_enable(); 6059 rcu_read_unlock(); 6060 cond_resched(); 6061 if (loop_end(loop_end_arg, start_time)) 6062 return; 6063 goto restart; 6064 } 6065 cpu_relax(); 6066 } 6067 if (napi_poll) 6068 busy_poll_stop(napi, have_poll_lock); 6069 preempt_enable(); 6070 out: 6071 rcu_read_unlock(); 6072 } 6073 EXPORT_SYMBOL(napi_busy_loop); 6074 6075 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6076 6077 static void napi_hash_add(struct napi_struct *napi) 6078 { 6079 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || 6080 test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) 6081 return; 6082 6083 spin_lock(&napi_hash_lock); 6084 6085 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6086 do { 6087 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6088 napi_gen_id = MIN_NAPI_ID; 6089 } while (napi_by_id(napi_gen_id)); 6090 napi->napi_id = napi_gen_id; 6091 6092 hlist_add_head_rcu(&napi->napi_hash_node, 6093 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6094 6095 spin_unlock(&napi_hash_lock); 6096 } 6097 6098 /* Warning : caller is responsible to make sure rcu grace period 6099 * is respected before freeing memory containing @napi 6100 */ 6101 bool napi_hash_del(struct napi_struct *napi) 6102 { 6103 bool rcu_sync_needed = false; 6104 6105 spin_lock(&napi_hash_lock); 6106 6107 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { 6108 rcu_sync_needed = true; 6109 hlist_del_rcu(&napi->napi_hash_node); 6110 } 6111 spin_unlock(&napi_hash_lock); 6112 return rcu_sync_needed; 6113 } 6114 EXPORT_SYMBOL_GPL(napi_hash_del); 6115 6116 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6117 { 6118 struct napi_struct *napi; 6119 6120 napi = container_of(timer, struct napi_struct, timer); 6121 6122 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6123 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6124 */ 6125 if (napi->gro_bitmask && !napi_disable_pending(napi) && 6126 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) 6127 __napi_schedule_irqoff(napi); 6128 6129 return HRTIMER_NORESTART; 6130 } 6131 6132 static void init_gro_hash(struct napi_struct *napi) 6133 { 6134 int i; 6135 6136 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6137 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6138 napi->gro_hash[i].count = 0; 6139 } 6140 napi->gro_bitmask = 0; 6141 } 6142 6143 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6144 int (*poll)(struct napi_struct *, int), int weight) 6145 { 6146 INIT_LIST_HEAD(&napi->poll_list); 6147 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6148 napi->timer.function = napi_watchdog; 6149 init_gro_hash(napi); 6150 napi->skb = NULL; 6151 napi->poll = poll; 6152 if (weight > NAPI_POLL_WEIGHT) 6153 pr_err_once("netif_napi_add() called with weight %d on device %s\n", 6154 weight, dev->name); 6155 napi->weight = weight; 6156 list_add(&napi->dev_list, &dev->napi_list); 6157 napi->dev = dev; 6158 #ifdef CONFIG_NETPOLL 6159 napi->poll_owner = -1; 6160 #endif 6161 set_bit(NAPI_STATE_SCHED, &napi->state); 6162 napi_hash_add(napi); 6163 } 6164 EXPORT_SYMBOL(netif_napi_add); 6165 6166 void napi_disable(struct napi_struct *n) 6167 { 6168 might_sleep(); 6169 set_bit(NAPI_STATE_DISABLE, &n->state); 6170 6171 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6172 msleep(1); 6173 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6174 msleep(1); 6175 6176 hrtimer_cancel(&n->timer); 6177 6178 clear_bit(NAPI_STATE_DISABLE, &n->state); 6179 } 6180 EXPORT_SYMBOL(napi_disable); 6181 6182 static void flush_gro_hash(struct napi_struct *napi) 6183 { 6184 int i; 6185 6186 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6187 struct sk_buff *skb, *n; 6188 6189 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6190 kfree_skb(skb); 6191 napi->gro_hash[i].count = 0; 6192 } 6193 } 6194 6195 /* Must be called in process context */ 6196 void netif_napi_del(struct napi_struct *napi) 6197 { 6198 might_sleep(); 6199 if (napi_hash_del(napi)) 6200 synchronize_net(); 6201 list_del_init(&napi->dev_list); 6202 napi_free_frags(napi); 6203 6204 flush_gro_hash(napi); 6205 napi->gro_bitmask = 0; 6206 } 6207 EXPORT_SYMBOL(netif_napi_del); 6208 6209 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6210 { 6211 void *have; 6212 int work, weight; 6213 6214 list_del_init(&n->poll_list); 6215 6216 have = netpoll_poll_lock(n); 6217 6218 weight = n->weight; 6219 6220 /* This NAPI_STATE_SCHED test is for avoiding a race 6221 * with netpoll's poll_napi(). Only the entity which 6222 * obtains the lock and sees NAPI_STATE_SCHED set will 6223 * actually make the ->poll() call. Therefore we avoid 6224 * accidentally calling ->poll() when NAPI is not scheduled. 6225 */ 6226 work = 0; 6227 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6228 work = n->poll(n, weight); 6229 trace_napi_poll(n, work, weight); 6230 } 6231 6232 WARN_ON_ONCE(work > weight); 6233 6234 if (likely(work < weight)) 6235 goto out_unlock; 6236 6237 /* Drivers must not modify the NAPI state if they 6238 * consume the entire weight. In such cases this code 6239 * still "owns" the NAPI instance and therefore can 6240 * move the instance around on the list at-will. 6241 */ 6242 if (unlikely(napi_disable_pending(n))) { 6243 napi_complete(n); 6244 goto out_unlock; 6245 } 6246 6247 if (n->gro_bitmask) { 6248 /* flush too old packets 6249 * If HZ < 1000, flush all packets. 6250 */ 6251 napi_gro_flush(n, HZ >= 1000); 6252 } 6253 6254 /* Some drivers may have called napi_schedule 6255 * prior to exhausting their budget. 6256 */ 6257 if (unlikely(!list_empty(&n->poll_list))) { 6258 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6259 n->dev ? n->dev->name : "backlog"); 6260 goto out_unlock; 6261 } 6262 6263 list_add_tail(&n->poll_list, repoll); 6264 6265 out_unlock: 6266 netpoll_poll_unlock(have); 6267 6268 return work; 6269 } 6270 6271 static __latent_entropy void net_rx_action(struct softirq_action *h) 6272 { 6273 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6274 unsigned long time_limit = jiffies + 6275 usecs_to_jiffies(netdev_budget_usecs); 6276 int budget = netdev_budget; 6277 LIST_HEAD(list); 6278 LIST_HEAD(repoll); 6279 6280 local_irq_disable(); 6281 list_splice_init(&sd->poll_list, &list); 6282 local_irq_enable(); 6283 6284 for (;;) { 6285 struct napi_struct *n; 6286 6287 if (list_empty(&list)) { 6288 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6289 goto out; 6290 break; 6291 } 6292 6293 n = list_first_entry(&list, struct napi_struct, poll_list); 6294 budget -= napi_poll(n, &repoll); 6295 6296 /* If softirq window is exhausted then punt. 6297 * Allow this to run for 2 jiffies since which will allow 6298 * an average latency of 1.5/HZ. 6299 */ 6300 if (unlikely(budget <= 0 || 6301 time_after_eq(jiffies, time_limit))) { 6302 sd->time_squeeze++; 6303 break; 6304 } 6305 } 6306 6307 local_irq_disable(); 6308 6309 list_splice_tail_init(&sd->poll_list, &list); 6310 list_splice_tail(&repoll, &list); 6311 list_splice(&list, &sd->poll_list); 6312 if (!list_empty(&sd->poll_list)) 6313 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6314 6315 net_rps_action_and_irq_enable(sd); 6316 out: 6317 __kfree_skb_flush(); 6318 } 6319 6320 struct netdev_adjacent { 6321 struct net_device *dev; 6322 6323 /* upper master flag, there can only be one master device per list */ 6324 bool master; 6325 6326 /* counter for the number of times this device was added to us */ 6327 u16 ref_nr; 6328 6329 /* private field for the users */ 6330 void *private; 6331 6332 struct list_head list; 6333 struct rcu_head rcu; 6334 }; 6335 6336 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6337 struct list_head *adj_list) 6338 { 6339 struct netdev_adjacent *adj; 6340 6341 list_for_each_entry(adj, adj_list, list) { 6342 if (adj->dev == adj_dev) 6343 return adj; 6344 } 6345 return NULL; 6346 } 6347 6348 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) 6349 { 6350 struct net_device *dev = data; 6351 6352 return upper_dev == dev; 6353 } 6354 6355 /** 6356 * netdev_has_upper_dev - Check if device is linked to an upper device 6357 * @dev: device 6358 * @upper_dev: upper device to check 6359 * 6360 * Find out if a device is linked to specified upper device and return true 6361 * in case it is. Note that this checks only immediate upper device, 6362 * not through a complete stack of devices. The caller must hold the RTNL lock. 6363 */ 6364 bool netdev_has_upper_dev(struct net_device *dev, 6365 struct net_device *upper_dev) 6366 { 6367 ASSERT_RTNL(); 6368 6369 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 6370 upper_dev); 6371 } 6372 EXPORT_SYMBOL(netdev_has_upper_dev); 6373 6374 /** 6375 * netdev_has_upper_dev_all - Check if device is linked to an upper device 6376 * @dev: device 6377 * @upper_dev: upper device to check 6378 * 6379 * Find out if a device is linked to specified upper device and return true 6380 * in case it is. Note that this checks the entire upper device chain. 6381 * The caller must hold rcu lock. 6382 */ 6383 6384 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6385 struct net_device *upper_dev) 6386 { 6387 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 6388 upper_dev); 6389 } 6390 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6391 6392 /** 6393 * netdev_has_any_upper_dev - Check if device is linked to some device 6394 * @dev: device 6395 * 6396 * Find out if a device is linked to an upper device and return true in case 6397 * it is. The caller must hold the RTNL lock. 6398 */ 6399 bool netdev_has_any_upper_dev(struct net_device *dev) 6400 { 6401 ASSERT_RTNL(); 6402 6403 return !list_empty(&dev->adj_list.upper); 6404 } 6405 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6406 6407 /** 6408 * netdev_master_upper_dev_get - Get master upper device 6409 * @dev: device 6410 * 6411 * Find a master upper device and return pointer to it or NULL in case 6412 * it's not there. The caller must hold the RTNL lock. 6413 */ 6414 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6415 { 6416 struct netdev_adjacent *upper; 6417 6418 ASSERT_RTNL(); 6419 6420 if (list_empty(&dev->adj_list.upper)) 6421 return NULL; 6422 6423 upper = list_first_entry(&dev->adj_list.upper, 6424 struct netdev_adjacent, list); 6425 if (likely(upper->master)) 6426 return upper->dev; 6427 return NULL; 6428 } 6429 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6430 6431 /** 6432 * netdev_has_any_lower_dev - Check if device is linked to some device 6433 * @dev: device 6434 * 6435 * Find out if a device is linked to a lower device and return true in case 6436 * it is. The caller must hold the RTNL lock. 6437 */ 6438 static bool netdev_has_any_lower_dev(struct net_device *dev) 6439 { 6440 ASSERT_RTNL(); 6441 6442 return !list_empty(&dev->adj_list.lower); 6443 } 6444 6445 void *netdev_adjacent_get_private(struct list_head *adj_list) 6446 { 6447 struct netdev_adjacent *adj; 6448 6449 adj = list_entry(adj_list, struct netdev_adjacent, list); 6450 6451 return adj->private; 6452 } 6453 EXPORT_SYMBOL(netdev_adjacent_get_private); 6454 6455 /** 6456 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6457 * @dev: device 6458 * @iter: list_head ** of the current position 6459 * 6460 * Gets the next device from the dev's upper list, starting from iter 6461 * position. The caller must hold RCU read lock. 6462 */ 6463 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6464 struct list_head **iter) 6465 { 6466 struct netdev_adjacent *upper; 6467 6468 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6469 6470 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6471 6472 if (&upper->list == &dev->adj_list.upper) 6473 return NULL; 6474 6475 *iter = &upper->list; 6476 6477 return upper->dev; 6478 } 6479 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6480 6481 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6482 struct list_head **iter) 6483 { 6484 struct netdev_adjacent *upper; 6485 6486 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6487 6488 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6489 6490 if (&upper->list == &dev->adj_list.upper) 6491 return NULL; 6492 6493 *iter = &upper->list; 6494 6495 return upper->dev; 6496 } 6497 6498 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6499 int (*fn)(struct net_device *dev, 6500 void *data), 6501 void *data) 6502 { 6503 struct net_device *udev; 6504 struct list_head *iter; 6505 int ret; 6506 6507 for (iter = &dev->adj_list.upper, 6508 udev = netdev_next_upper_dev_rcu(dev, &iter); 6509 udev; 6510 udev = netdev_next_upper_dev_rcu(dev, &iter)) { 6511 /* first is the upper device itself */ 6512 ret = fn(udev, data); 6513 if (ret) 6514 return ret; 6515 6516 /* then look at all of its upper devices */ 6517 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); 6518 if (ret) 6519 return ret; 6520 } 6521 6522 return 0; 6523 } 6524 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6525 6526 /** 6527 * netdev_lower_get_next_private - Get the next ->private from the 6528 * lower neighbour list 6529 * @dev: device 6530 * @iter: list_head ** of the current position 6531 * 6532 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6533 * list, starting from iter position. The caller must hold either hold the 6534 * RTNL lock or its own locking that guarantees that the neighbour lower 6535 * list will remain unchanged. 6536 */ 6537 void *netdev_lower_get_next_private(struct net_device *dev, 6538 struct list_head **iter) 6539 { 6540 struct netdev_adjacent *lower; 6541 6542 lower = list_entry(*iter, struct netdev_adjacent, list); 6543 6544 if (&lower->list == &dev->adj_list.lower) 6545 return NULL; 6546 6547 *iter = lower->list.next; 6548 6549 return lower->private; 6550 } 6551 EXPORT_SYMBOL(netdev_lower_get_next_private); 6552 6553 /** 6554 * netdev_lower_get_next_private_rcu - Get the next ->private from the 6555 * lower neighbour list, RCU 6556 * variant 6557 * @dev: device 6558 * @iter: list_head ** of the current position 6559 * 6560 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6561 * list, starting from iter position. The caller must hold RCU read lock. 6562 */ 6563 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 6564 struct list_head **iter) 6565 { 6566 struct netdev_adjacent *lower; 6567 6568 WARN_ON_ONCE(!rcu_read_lock_held()); 6569 6570 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6571 6572 if (&lower->list == &dev->adj_list.lower) 6573 return NULL; 6574 6575 *iter = &lower->list; 6576 6577 return lower->private; 6578 } 6579 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 6580 6581 /** 6582 * netdev_lower_get_next - Get the next device from the lower neighbour 6583 * list 6584 * @dev: device 6585 * @iter: list_head ** of the current position 6586 * 6587 * Gets the next netdev_adjacent from the dev's lower neighbour 6588 * list, starting from iter position. The caller must hold RTNL lock or 6589 * its own locking that guarantees that the neighbour lower 6590 * list will remain unchanged. 6591 */ 6592 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 6593 { 6594 struct netdev_adjacent *lower; 6595 6596 lower = list_entry(*iter, struct netdev_adjacent, list); 6597 6598 if (&lower->list == &dev->adj_list.lower) 6599 return NULL; 6600 6601 *iter = lower->list.next; 6602 6603 return lower->dev; 6604 } 6605 EXPORT_SYMBOL(netdev_lower_get_next); 6606 6607 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 6608 struct list_head **iter) 6609 { 6610 struct netdev_adjacent *lower; 6611 6612 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 6613 6614 if (&lower->list == &dev->adj_list.lower) 6615 return NULL; 6616 6617 *iter = &lower->list; 6618 6619 return lower->dev; 6620 } 6621 6622 int netdev_walk_all_lower_dev(struct net_device *dev, 6623 int (*fn)(struct net_device *dev, 6624 void *data), 6625 void *data) 6626 { 6627 struct net_device *ldev; 6628 struct list_head *iter; 6629 int ret; 6630 6631 for (iter = &dev->adj_list.lower, 6632 ldev = netdev_next_lower_dev(dev, &iter); 6633 ldev; 6634 ldev = netdev_next_lower_dev(dev, &iter)) { 6635 /* first is the lower device itself */ 6636 ret = fn(ldev, data); 6637 if (ret) 6638 return ret; 6639 6640 /* then look at all of its lower devices */ 6641 ret = netdev_walk_all_lower_dev(ldev, fn, data); 6642 if (ret) 6643 return ret; 6644 } 6645 6646 return 0; 6647 } 6648 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 6649 6650 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 6651 struct list_head **iter) 6652 { 6653 struct netdev_adjacent *lower; 6654 6655 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6656 if (&lower->list == &dev->adj_list.lower) 6657 return NULL; 6658 6659 *iter = &lower->list; 6660 6661 return lower->dev; 6662 } 6663 6664 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 6665 int (*fn)(struct net_device *dev, 6666 void *data), 6667 void *data) 6668 { 6669 struct net_device *ldev; 6670 struct list_head *iter; 6671 int ret; 6672 6673 for (iter = &dev->adj_list.lower, 6674 ldev = netdev_next_lower_dev_rcu(dev, &iter); 6675 ldev; 6676 ldev = netdev_next_lower_dev_rcu(dev, &iter)) { 6677 /* first is the lower device itself */ 6678 ret = fn(ldev, data); 6679 if (ret) 6680 return ret; 6681 6682 /* then look at all of its lower devices */ 6683 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); 6684 if (ret) 6685 return ret; 6686 } 6687 6688 return 0; 6689 } 6690 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 6691 6692 /** 6693 * netdev_lower_get_first_private_rcu - Get the first ->private from the 6694 * lower neighbour list, RCU 6695 * variant 6696 * @dev: device 6697 * 6698 * Gets the first netdev_adjacent->private from the dev's lower neighbour 6699 * list. The caller must hold RCU read lock. 6700 */ 6701 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 6702 { 6703 struct netdev_adjacent *lower; 6704 6705 lower = list_first_or_null_rcu(&dev->adj_list.lower, 6706 struct netdev_adjacent, list); 6707 if (lower) 6708 return lower->private; 6709 return NULL; 6710 } 6711 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 6712 6713 /** 6714 * netdev_master_upper_dev_get_rcu - Get master upper device 6715 * @dev: device 6716 * 6717 * Find a master upper device and return pointer to it or NULL in case 6718 * it's not there. The caller must hold the RCU read lock. 6719 */ 6720 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 6721 { 6722 struct netdev_adjacent *upper; 6723 6724 upper = list_first_or_null_rcu(&dev->adj_list.upper, 6725 struct netdev_adjacent, list); 6726 if (upper && likely(upper->master)) 6727 return upper->dev; 6728 return NULL; 6729 } 6730 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 6731 6732 static int netdev_adjacent_sysfs_add(struct net_device *dev, 6733 struct net_device *adj_dev, 6734 struct list_head *dev_list) 6735 { 6736 char linkname[IFNAMSIZ+7]; 6737 6738 sprintf(linkname, dev_list == &dev->adj_list.upper ? 6739 "upper_%s" : "lower_%s", adj_dev->name); 6740 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 6741 linkname); 6742 } 6743 static void netdev_adjacent_sysfs_del(struct net_device *dev, 6744 char *name, 6745 struct list_head *dev_list) 6746 { 6747 char linkname[IFNAMSIZ+7]; 6748 6749 sprintf(linkname, dev_list == &dev->adj_list.upper ? 6750 "upper_%s" : "lower_%s", name); 6751 sysfs_remove_link(&(dev->dev.kobj), linkname); 6752 } 6753 6754 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 6755 struct net_device *adj_dev, 6756 struct list_head *dev_list) 6757 { 6758 return (dev_list == &dev->adj_list.upper || 6759 dev_list == &dev->adj_list.lower) && 6760 net_eq(dev_net(dev), dev_net(adj_dev)); 6761 } 6762 6763 static int __netdev_adjacent_dev_insert(struct net_device *dev, 6764 struct net_device *adj_dev, 6765 struct list_head *dev_list, 6766 void *private, bool master) 6767 { 6768 struct netdev_adjacent *adj; 6769 int ret; 6770 6771 adj = __netdev_find_adj(adj_dev, dev_list); 6772 6773 if (adj) { 6774 adj->ref_nr += 1; 6775 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 6776 dev->name, adj_dev->name, adj->ref_nr); 6777 6778 return 0; 6779 } 6780 6781 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 6782 if (!adj) 6783 return -ENOMEM; 6784 6785 adj->dev = adj_dev; 6786 adj->master = master; 6787 adj->ref_nr = 1; 6788 adj->private = private; 6789 dev_hold(adj_dev); 6790 6791 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 6792 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 6793 6794 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 6795 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 6796 if (ret) 6797 goto free_adj; 6798 } 6799 6800 /* Ensure that master link is always the first item in list. */ 6801 if (master) { 6802 ret = sysfs_create_link(&(dev->dev.kobj), 6803 &(adj_dev->dev.kobj), "master"); 6804 if (ret) 6805 goto remove_symlinks; 6806 6807 list_add_rcu(&adj->list, dev_list); 6808 } else { 6809 list_add_tail_rcu(&adj->list, dev_list); 6810 } 6811 6812 return 0; 6813 6814 remove_symlinks: 6815 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 6816 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 6817 free_adj: 6818 kfree(adj); 6819 dev_put(adj_dev); 6820 6821 return ret; 6822 } 6823 6824 static void __netdev_adjacent_dev_remove(struct net_device *dev, 6825 struct net_device *adj_dev, 6826 u16 ref_nr, 6827 struct list_head *dev_list) 6828 { 6829 struct netdev_adjacent *adj; 6830 6831 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 6832 dev->name, adj_dev->name, ref_nr); 6833 6834 adj = __netdev_find_adj(adj_dev, dev_list); 6835 6836 if (!adj) { 6837 pr_err("Adjacency does not exist for device %s from %s\n", 6838 dev->name, adj_dev->name); 6839 WARN_ON(1); 6840 return; 6841 } 6842 6843 if (adj->ref_nr > ref_nr) { 6844 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 6845 dev->name, adj_dev->name, ref_nr, 6846 adj->ref_nr - ref_nr); 6847 adj->ref_nr -= ref_nr; 6848 return; 6849 } 6850 6851 if (adj->master) 6852 sysfs_remove_link(&(dev->dev.kobj), "master"); 6853 6854 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 6855 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 6856 6857 list_del_rcu(&adj->list); 6858 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 6859 adj_dev->name, dev->name, adj_dev->name); 6860 dev_put(adj_dev); 6861 kfree_rcu(adj, rcu); 6862 } 6863 6864 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 6865 struct net_device *upper_dev, 6866 struct list_head *up_list, 6867 struct list_head *down_list, 6868 void *private, bool master) 6869 { 6870 int ret; 6871 6872 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 6873 private, master); 6874 if (ret) 6875 return ret; 6876 6877 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 6878 private, false); 6879 if (ret) { 6880 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 6881 return ret; 6882 } 6883 6884 return 0; 6885 } 6886 6887 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 6888 struct net_device *upper_dev, 6889 u16 ref_nr, 6890 struct list_head *up_list, 6891 struct list_head *down_list) 6892 { 6893 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 6894 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 6895 } 6896 6897 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 6898 struct net_device *upper_dev, 6899 void *private, bool master) 6900 { 6901 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 6902 &dev->adj_list.upper, 6903 &upper_dev->adj_list.lower, 6904 private, master); 6905 } 6906 6907 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 6908 struct net_device *upper_dev) 6909 { 6910 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 6911 &dev->adj_list.upper, 6912 &upper_dev->adj_list.lower); 6913 } 6914 6915 static int __netdev_upper_dev_link(struct net_device *dev, 6916 struct net_device *upper_dev, bool master, 6917 void *upper_priv, void *upper_info, 6918 struct netlink_ext_ack *extack) 6919 { 6920 struct netdev_notifier_changeupper_info changeupper_info = { 6921 .info = { 6922 .dev = dev, 6923 .extack = extack, 6924 }, 6925 .upper_dev = upper_dev, 6926 .master = master, 6927 .linking = true, 6928 .upper_info = upper_info, 6929 }; 6930 struct net_device *master_dev; 6931 int ret = 0; 6932 6933 ASSERT_RTNL(); 6934 6935 if (dev == upper_dev) 6936 return -EBUSY; 6937 6938 /* To prevent loops, check if dev is not upper device to upper_dev. */ 6939 if (netdev_has_upper_dev(upper_dev, dev)) 6940 return -EBUSY; 6941 6942 if (!master) { 6943 if (netdev_has_upper_dev(dev, upper_dev)) 6944 return -EEXIST; 6945 } else { 6946 master_dev = netdev_master_upper_dev_get(dev); 6947 if (master_dev) 6948 return master_dev == upper_dev ? -EEXIST : -EBUSY; 6949 } 6950 6951 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 6952 &changeupper_info.info); 6953 ret = notifier_to_errno(ret); 6954 if (ret) 6955 return ret; 6956 6957 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 6958 master); 6959 if (ret) 6960 return ret; 6961 6962 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 6963 &changeupper_info.info); 6964 ret = notifier_to_errno(ret); 6965 if (ret) 6966 goto rollback; 6967 6968 return 0; 6969 6970 rollback: 6971 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 6972 6973 return ret; 6974 } 6975 6976 /** 6977 * netdev_upper_dev_link - Add a link to the upper device 6978 * @dev: device 6979 * @upper_dev: new upper device 6980 * @extack: netlink extended ack 6981 * 6982 * Adds a link to device which is upper to this one. The caller must hold 6983 * the RTNL lock. On a failure a negative errno code is returned. 6984 * On success the reference counts are adjusted and the function 6985 * returns zero. 6986 */ 6987 int netdev_upper_dev_link(struct net_device *dev, 6988 struct net_device *upper_dev, 6989 struct netlink_ext_ack *extack) 6990 { 6991 return __netdev_upper_dev_link(dev, upper_dev, false, 6992 NULL, NULL, extack); 6993 } 6994 EXPORT_SYMBOL(netdev_upper_dev_link); 6995 6996 /** 6997 * netdev_master_upper_dev_link - Add a master link to the upper device 6998 * @dev: device 6999 * @upper_dev: new upper device 7000 * @upper_priv: upper device private 7001 * @upper_info: upper info to be passed down via notifier 7002 * @extack: netlink extended ack 7003 * 7004 * Adds a link to device which is upper to this one. In this case, only 7005 * one master upper device can be linked, although other non-master devices 7006 * might be linked as well. The caller must hold the RTNL lock. 7007 * On a failure a negative errno code is returned. On success the reference 7008 * counts are adjusted and the function returns zero. 7009 */ 7010 int netdev_master_upper_dev_link(struct net_device *dev, 7011 struct net_device *upper_dev, 7012 void *upper_priv, void *upper_info, 7013 struct netlink_ext_ack *extack) 7014 { 7015 return __netdev_upper_dev_link(dev, upper_dev, true, 7016 upper_priv, upper_info, extack); 7017 } 7018 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7019 7020 /** 7021 * netdev_upper_dev_unlink - Removes a link to upper device 7022 * @dev: device 7023 * @upper_dev: new upper device 7024 * 7025 * Removes a link to device which is upper to this one. The caller must hold 7026 * the RTNL lock. 7027 */ 7028 void netdev_upper_dev_unlink(struct net_device *dev, 7029 struct net_device *upper_dev) 7030 { 7031 struct netdev_notifier_changeupper_info changeupper_info = { 7032 .info = { 7033 .dev = dev, 7034 }, 7035 .upper_dev = upper_dev, 7036 .linking = false, 7037 }; 7038 7039 ASSERT_RTNL(); 7040 7041 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7042 7043 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7044 &changeupper_info.info); 7045 7046 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7047 7048 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7049 &changeupper_info.info); 7050 } 7051 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7052 7053 /** 7054 * netdev_bonding_info_change - Dispatch event about slave change 7055 * @dev: device 7056 * @bonding_info: info to dispatch 7057 * 7058 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7059 * The caller must hold the RTNL lock. 7060 */ 7061 void netdev_bonding_info_change(struct net_device *dev, 7062 struct netdev_bonding_info *bonding_info) 7063 { 7064 struct netdev_notifier_bonding_info info = { 7065 .info.dev = dev, 7066 }; 7067 7068 memcpy(&info.bonding_info, bonding_info, 7069 sizeof(struct netdev_bonding_info)); 7070 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7071 &info.info); 7072 } 7073 EXPORT_SYMBOL(netdev_bonding_info_change); 7074 7075 static void netdev_adjacent_add_links(struct net_device *dev) 7076 { 7077 struct netdev_adjacent *iter; 7078 7079 struct net *net = dev_net(dev); 7080 7081 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7082 if (!net_eq(net, dev_net(iter->dev))) 7083 continue; 7084 netdev_adjacent_sysfs_add(iter->dev, dev, 7085 &iter->dev->adj_list.lower); 7086 netdev_adjacent_sysfs_add(dev, iter->dev, 7087 &dev->adj_list.upper); 7088 } 7089 7090 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7091 if (!net_eq(net, dev_net(iter->dev))) 7092 continue; 7093 netdev_adjacent_sysfs_add(iter->dev, dev, 7094 &iter->dev->adj_list.upper); 7095 netdev_adjacent_sysfs_add(dev, iter->dev, 7096 &dev->adj_list.lower); 7097 } 7098 } 7099 7100 static void netdev_adjacent_del_links(struct net_device *dev) 7101 { 7102 struct netdev_adjacent *iter; 7103 7104 struct net *net = dev_net(dev); 7105 7106 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7107 if (!net_eq(net, dev_net(iter->dev))) 7108 continue; 7109 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7110 &iter->dev->adj_list.lower); 7111 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7112 &dev->adj_list.upper); 7113 } 7114 7115 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7116 if (!net_eq(net, dev_net(iter->dev))) 7117 continue; 7118 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7119 &iter->dev->adj_list.upper); 7120 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7121 &dev->adj_list.lower); 7122 } 7123 } 7124 7125 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 7126 { 7127 struct netdev_adjacent *iter; 7128 7129 struct net *net = dev_net(dev); 7130 7131 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7132 if (!net_eq(net, dev_net(iter->dev))) 7133 continue; 7134 netdev_adjacent_sysfs_del(iter->dev, oldname, 7135 &iter->dev->adj_list.lower); 7136 netdev_adjacent_sysfs_add(iter->dev, dev, 7137 &iter->dev->adj_list.lower); 7138 } 7139 7140 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7141 if (!net_eq(net, dev_net(iter->dev))) 7142 continue; 7143 netdev_adjacent_sysfs_del(iter->dev, oldname, 7144 &iter->dev->adj_list.upper); 7145 netdev_adjacent_sysfs_add(iter->dev, dev, 7146 &iter->dev->adj_list.upper); 7147 } 7148 } 7149 7150 void *netdev_lower_dev_get_private(struct net_device *dev, 7151 struct net_device *lower_dev) 7152 { 7153 struct netdev_adjacent *lower; 7154 7155 if (!lower_dev) 7156 return NULL; 7157 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 7158 if (!lower) 7159 return NULL; 7160 7161 return lower->private; 7162 } 7163 EXPORT_SYMBOL(netdev_lower_dev_get_private); 7164 7165 7166 int dev_get_nest_level(struct net_device *dev) 7167 { 7168 struct net_device *lower = NULL; 7169 struct list_head *iter; 7170 int max_nest = -1; 7171 int nest; 7172 7173 ASSERT_RTNL(); 7174 7175 netdev_for_each_lower_dev(dev, lower, iter) { 7176 nest = dev_get_nest_level(lower); 7177 if (max_nest < nest) 7178 max_nest = nest; 7179 } 7180 7181 return max_nest + 1; 7182 } 7183 EXPORT_SYMBOL(dev_get_nest_level); 7184 7185 /** 7186 * netdev_lower_change - Dispatch event about lower device state change 7187 * @lower_dev: device 7188 * @lower_state_info: state to dispatch 7189 * 7190 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 7191 * The caller must hold the RTNL lock. 7192 */ 7193 void netdev_lower_state_changed(struct net_device *lower_dev, 7194 void *lower_state_info) 7195 { 7196 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 7197 .info.dev = lower_dev, 7198 }; 7199 7200 ASSERT_RTNL(); 7201 changelowerstate_info.lower_state_info = lower_state_info; 7202 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 7203 &changelowerstate_info.info); 7204 } 7205 EXPORT_SYMBOL(netdev_lower_state_changed); 7206 7207 static void dev_change_rx_flags(struct net_device *dev, int flags) 7208 { 7209 const struct net_device_ops *ops = dev->netdev_ops; 7210 7211 if (ops->ndo_change_rx_flags) 7212 ops->ndo_change_rx_flags(dev, flags); 7213 } 7214 7215 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 7216 { 7217 unsigned int old_flags = dev->flags; 7218 kuid_t uid; 7219 kgid_t gid; 7220 7221 ASSERT_RTNL(); 7222 7223 dev->flags |= IFF_PROMISC; 7224 dev->promiscuity += inc; 7225 if (dev->promiscuity == 0) { 7226 /* 7227 * Avoid overflow. 7228 * If inc causes overflow, untouch promisc and return error. 7229 */ 7230 if (inc < 0) 7231 dev->flags &= ~IFF_PROMISC; 7232 else { 7233 dev->promiscuity -= inc; 7234 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 7235 dev->name); 7236 return -EOVERFLOW; 7237 } 7238 } 7239 if (dev->flags != old_flags) { 7240 pr_info("device %s %s promiscuous mode\n", 7241 dev->name, 7242 dev->flags & IFF_PROMISC ? "entered" : "left"); 7243 if (audit_enabled) { 7244 current_uid_gid(&uid, &gid); 7245 audit_log(audit_context(), GFP_ATOMIC, 7246 AUDIT_ANOM_PROMISCUOUS, 7247 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 7248 dev->name, (dev->flags & IFF_PROMISC), 7249 (old_flags & IFF_PROMISC), 7250 from_kuid(&init_user_ns, audit_get_loginuid(current)), 7251 from_kuid(&init_user_ns, uid), 7252 from_kgid(&init_user_ns, gid), 7253 audit_get_sessionid(current)); 7254 } 7255 7256 dev_change_rx_flags(dev, IFF_PROMISC); 7257 } 7258 if (notify) 7259 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 7260 return 0; 7261 } 7262 7263 /** 7264 * dev_set_promiscuity - update promiscuity count on a device 7265 * @dev: device 7266 * @inc: modifier 7267 * 7268 * Add or remove promiscuity from a device. While the count in the device 7269 * remains above zero the interface remains promiscuous. Once it hits zero 7270 * the device reverts back to normal filtering operation. A negative inc 7271 * value is used to drop promiscuity on the device. 7272 * Return 0 if successful or a negative errno code on error. 7273 */ 7274 int dev_set_promiscuity(struct net_device *dev, int inc) 7275 { 7276 unsigned int old_flags = dev->flags; 7277 int err; 7278 7279 err = __dev_set_promiscuity(dev, inc, true); 7280 if (err < 0) 7281 return err; 7282 if (dev->flags != old_flags) 7283 dev_set_rx_mode(dev); 7284 return err; 7285 } 7286 EXPORT_SYMBOL(dev_set_promiscuity); 7287 7288 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 7289 { 7290 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 7291 7292 ASSERT_RTNL(); 7293 7294 dev->flags |= IFF_ALLMULTI; 7295 dev->allmulti += inc; 7296 if (dev->allmulti == 0) { 7297 /* 7298 * Avoid overflow. 7299 * If inc causes overflow, untouch allmulti and return error. 7300 */ 7301 if (inc < 0) 7302 dev->flags &= ~IFF_ALLMULTI; 7303 else { 7304 dev->allmulti -= inc; 7305 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 7306 dev->name); 7307 return -EOVERFLOW; 7308 } 7309 } 7310 if (dev->flags ^ old_flags) { 7311 dev_change_rx_flags(dev, IFF_ALLMULTI); 7312 dev_set_rx_mode(dev); 7313 if (notify) 7314 __dev_notify_flags(dev, old_flags, 7315 dev->gflags ^ old_gflags); 7316 } 7317 return 0; 7318 } 7319 7320 /** 7321 * dev_set_allmulti - update allmulti count on a device 7322 * @dev: device 7323 * @inc: modifier 7324 * 7325 * Add or remove reception of all multicast frames to a device. While the 7326 * count in the device remains above zero the interface remains listening 7327 * to all interfaces. Once it hits zero the device reverts back to normal 7328 * filtering operation. A negative @inc value is used to drop the counter 7329 * when releasing a resource needing all multicasts. 7330 * Return 0 if successful or a negative errno code on error. 7331 */ 7332 7333 int dev_set_allmulti(struct net_device *dev, int inc) 7334 { 7335 return __dev_set_allmulti(dev, inc, true); 7336 } 7337 EXPORT_SYMBOL(dev_set_allmulti); 7338 7339 /* 7340 * Upload unicast and multicast address lists to device and 7341 * configure RX filtering. When the device doesn't support unicast 7342 * filtering it is put in promiscuous mode while unicast addresses 7343 * are present. 7344 */ 7345 void __dev_set_rx_mode(struct net_device *dev) 7346 { 7347 const struct net_device_ops *ops = dev->netdev_ops; 7348 7349 /* dev_open will call this function so the list will stay sane. */ 7350 if (!(dev->flags&IFF_UP)) 7351 return; 7352 7353 if (!netif_device_present(dev)) 7354 return; 7355 7356 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 7357 /* Unicast addresses changes may only happen under the rtnl, 7358 * therefore calling __dev_set_promiscuity here is safe. 7359 */ 7360 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 7361 __dev_set_promiscuity(dev, 1, false); 7362 dev->uc_promisc = true; 7363 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 7364 __dev_set_promiscuity(dev, -1, false); 7365 dev->uc_promisc = false; 7366 } 7367 } 7368 7369 if (ops->ndo_set_rx_mode) 7370 ops->ndo_set_rx_mode(dev); 7371 } 7372 7373 void dev_set_rx_mode(struct net_device *dev) 7374 { 7375 netif_addr_lock_bh(dev); 7376 __dev_set_rx_mode(dev); 7377 netif_addr_unlock_bh(dev); 7378 } 7379 7380 /** 7381 * dev_get_flags - get flags reported to userspace 7382 * @dev: device 7383 * 7384 * Get the combination of flag bits exported through APIs to userspace. 7385 */ 7386 unsigned int dev_get_flags(const struct net_device *dev) 7387 { 7388 unsigned int flags; 7389 7390 flags = (dev->flags & ~(IFF_PROMISC | 7391 IFF_ALLMULTI | 7392 IFF_RUNNING | 7393 IFF_LOWER_UP | 7394 IFF_DORMANT)) | 7395 (dev->gflags & (IFF_PROMISC | 7396 IFF_ALLMULTI)); 7397 7398 if (netif_running(dev)) { 7399 if (netif_oper_up(dev)) 7400 flags |= IFF_RUNNING; 7401 if (netif_carrier_ok(dev)) 7402 flags |= IFF_LOWER_UP; 7403 if (netif_dormant(dev)) 7404 flags |= IFF_DORMANT; 7405 } 7406 7407 return flags; 7408 } 7409 EXPORT_SYMBOL(dev_get_flags); 7410 7411 int __dev_change_flags(struct net_device *dev, unsigned int flags) 7412 { 7413 unsigned int old_flags = dev->flags; 7414 int ret; 7415 7416 ASSERT_RTNL(); 7417 7418 /* 7419 * Set the flags on our device. 7420 */ 7421 7422 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 7423 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 7424 IFF_AUTOMEDIA)) | 7425 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 7426 IFF_ALLMULTI)); 7427 7428 /* 7429 * Load in the correct multicast list now the flags have changed. 7430 */ 7431 7432 if ((old_flags ^ flags) & IFF_MULTICAST) 7433 dev_change_rx_flags(dev, IFF_MULTICAST); 7434 7435 dev_set_rx_mode(dev); 7436 7437 /* 7438 * Have we downed the interface. We handle IFF_UP ourselves 7439 * according to user attempts to set it, rather than blindly 7440 * setting it. 7441 */ 7442 7443 ret = 0; 7444 if ((old_flags ^ flags) & IFF_UP) { 7445 if (old_flags & IFF_UP) 7446 __dev_close(dev); 7447 else 7448 ret = __dev_open(dev); 7449 } 7450 7451 if ((flags ^ dev->gflags) & IFF_PROMISC) { 7452 int inc = (flags & IFF_PROMISC) ? 1 : -1; 7453 unsigned int old_flags = dev->flags; 7454 7455 dev->gflags ^= IFF_PROMISC; 7456 7457 if (__dev_set_promiscuity(dev, inc, false) >= 0) 7458 if (dev->flags != old_flags) 7459 dev_set_rx_mode(dev); 7460 } 7461 7462 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 7463 * is important. Some (broken) drivers set IFF_PROMISC, when 7464 * IFF_ALLMULTI is requested not asking us and not reporting. 7465 */ 7466 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 7467 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 7468 7469 dev->gflags ^= IFF_ALLMULTI; 7470 __dev_set_allmulti(dev, inc, false); 7471 } 7472 7473 return ret; 7474 } 7475 7476 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 7477 unsigned int gchanges) 7478 { 7479 unsigned int changes = dev->flags ^ old_flags; 7480 7481 if (gchanges) 7482 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 7483 7484 if (changes & IFF_UP) { 7485 if (dev->flags & IFF_UP) 7486 call_netdevice_notifiers(NETDEV_UP, dev); 7487 else 7488 call_netdevice_notifiers(NETDEV_DOWN, dev); 7489 } 7490 7491 if (dev->flags & IFF_UP && 7492 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 7493 struct netdev_notifier_change_info change_info = { 7494 .info = { 7495 .dev = dev, 7496 }, 7497 .flags_changed = changes, 7498 }; 7499 7500 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 7501 } 7502 } 7503 7504 /** 7505 * dev_change_flags - change device settings 7506 * @dev: device 7507 * @flags: device state flags 7508 * 7509 * Change settings on device based state flags. The flags are 7510 * in the userspace exported format. 7511 */ 7512 int dev_change_flags(struct net_device *dev, unsigned int flags) 7513 { 7514 int ret; 7515 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 7516 7517 ret = __dev_change_flags(dev, flags); 7518 if (ret < 0) 7519 return ret; 7520 7521 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 7522 __dev_notify_flags(dev, old_flags, changes); 7523 return ret; 7524 } 7525 EXPORT_SYMBOL(dev_change_flags); 7526 7527 int __dev_set_mtu(struct net_device *dev, int new_mtu) 7528 { 7529 const struct net_device_ops *ops = dev->netdev_ops; 7530 7531 if (ops->ndo_change_mtu) 7532 return ops->ndo_change_mtu(dev, new_mtu); 7533 7534 dev->mtu = new_mtu; 7535 return 0; 7536 } 7537 EXPORT_SYMBOL(__dev_set_mtu); 7538 7539 /** 7540 * dev_set_mtu_ext - Change maximum transfer unit 7541 * @dev: device 7542 * @new_mtu: new transfer unit 7543 * @extack: netlink extended ack 7544 * 7545 * Change the maximum transfer size of the network device. 7546 */ 7547 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 7548 struct netlink_ext_ack *extack) 7549 { 7550 int err, orig_mtu; 7551 7552 if (new_mtu == dev->mtu) 7553 return 0; 7554 7555 /* MTU must be positive, and in range */ 7556 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 7557 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 7558 return -EINVAL; 7559 } 7560 7561 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 7562 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 7563 return -EINVAL; 7564 } 7565 7566 if (!netif_device_present(dev)) 7567 return -ENODEV; 7568 7569 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 7570 err = notifier_to_errno(err); 7571 if (err) 7572 return err; 7573 7574 orig_mtu = dev->mtu; 7575 err = __dev_set_mtu(dev, new_mtu); 7576 7577 if (!err) { 7578 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 7579 err = notifier_to_errno(err); 7580 if (err) { 7581 /* setting mtu back and notifying everyone again, 7582 * so that they have a chance to revert changes. 7583 */ 7584 __dev_set_mtu(dev, orig_mtu); 7585 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); 7586 } 7587 } 7588 return err; 7589 } 7590 7591 int dev_set_mtu(struct net_device *dev, int new_mtu) 7592 { 7593 struct netlink_ext_ack extack; 7594 int err; 7595 7596 memset(&extack, 0, sizeof(extack)); 7597 err = dev_set_mtu_ext(dev, new_mtu, &extack); 7598 if (err && extack._msg) 7599 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 7600 return err; 7601 } 7602 EXPORT_SYMBOL(dev_set_mtu); 7603 7604 /** 7605 * dev_change_tx_queue_len - Change TX queue length of a netdevice 7606 * @dev: device 7607 * @new_len: new tx queue length 7608 */ 7609 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 7610 { 7611 unsigned int orig_len = dev->tx_queue_len; 7612 int res; 7613 7614 if (new_len != (unsigned int)new_len) 7615 return -ERANGE; 7616 7617 if (new_len != orig_len) { 7618 dev->tx_queue_len = new_len; 7619 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 7620 res = notifier_to_errno(res); 7621 if (res) 7622 goto err_rollback; 7623 res = dev_qdisc_change_tx_queue_len(dev); 7624 if (res) 7625 goto err_rollback; 7626 } 7627 7628 return 0; 7629 7630 err_rollback: 7631 netdev_err(dev, "refused to change device tx_queue_len\n"); 7632 dev->tx_queue_len = orig_len; 7633 return res; 7634 } 7635 7636 /** 7637 * dev_set_group - Change group this device belongs to 7638 * @dev: device 7639 * @new_group: group this device should belong to 7640 */ 7641 void dev_set_group(struct net_device *dev, int new_group) 7642 { 7643 dev->group = new_group; 7644 } 7645 EXPORT_SYMBOL(dev_set_group); 7646 7647 /** 7648 * dev_set_mac_address - Change Media Access Control Address 7649 * @dev: device 7650 * @sa: new address 7651 * 7652 * Change the hardware (MAC) address of the device 7653 */ 7654 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) 7655 { 7656 const struct net_device_ops *ops = dev->netdev_ops; 7657 int err; 7658 7659 if (!ops->ndo_set_mac_address) 7660 return -EOPNOTSUPP; 7661 if (sa->sa_family != dev->type) 7662 return -EINVAL; 7663 if (!netif_device_present(dev)) 7664 return -ENODEV; 7665 err = ops->ndo_set_mac_address(dev, sa); 7666 if (err) 7667 return err; 7668 dev->addr_assign_type = NET_ADDR_SET; 7669 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 7670 add_device_randomness(dev->dev_addr, dev->addr_len); 7671 return 0; 7672 } 7673 EXPORT_SYMBOL(dev_set_mac_address); 7674 7675 /** 7676 * dev_change_carrier - Change device carrier 7677 * @dev: device 7678 * @new_carrier: new value 7679 * 7680 * Change device carrier 7681 */ 7682 int dev_change_carrier(struct net_device *dev, bool new_carrier) 7683 { 7684 const struct net_device_ops *ops = dev->netdev_ops; 7685 7686 if (!ops->ndo_change_carrier) 7687 return -EOPNOTSUPP; 7688 if (!netif_device_present(dev)) 7689 return -ENODEV; 7690 return ops->ndo_change_carrier(dev, new_carrier); 7691 } 7692 EXPORT_SYMBOL(dev_change_carrier); 7693 7694 /** 7695 * dev_get_phys_port_id - Get device physical port ID 7696 * @dev: device 7697 * @ppid: port ID 7698 * 7699 * Get device physical port ID 7700 */ 7701 int dev_get_phys_port_id(struct net_device *dev, 7702 struct netdev_phys_item_id *ppid) 7703 { 7704 const struct net_device_ops *ops = dev->netdev_ops; 7705 7706 if (!ops->ndo_get_phys_port_id) 7707 return -EOPNOTSUPP; 7708 return ops->ndo_get_phys_port_id(dev, ppid); 7709 } 7710 EXPORT_SYMBOL(dev_get_phys_port_id); 7711 7712 /** 7713 * dev_get_phys_port_name - Get device physical port name 7714 * @dev: device 7715 * @name: port name 7716 * @len: limit of bytes to copy to name 7717 * 7718 * Get device physical port name 7719 */ 7720 int dev_get_phys_port_name(struct net_device *dev, 7721 char *name, size_t len) 7722 { 7723 const struct net_device_ops *ops = dev->netdev_ops; 7724 7725 if (!ops->ndo_get_phys_port_name) 7726 return -EOPNOTSUPP; 7727 return ops->ndo_get_phys_port_name(dev, name, len); 7728 } 7729 EXPORT_SYMBOL(dev_get_phys_port_name); 7730 7731 /** 7732 * dev_change_proto_down - update protocol port state information 7733 * @dev: device 7734 * @proto_down: new value 7735 * 7736 * This info can be used by switch drivers to set the phys state of the 7737 * port. 7738 */ 7739 int dev_change_proto_down(struct net_device *dev, bool proto_down) 7740 { 7741 const struct net_device_ops *ops = dev->netdev_ops; 7742 7743 if (!ops->ndo_change_proto_down) 7744 return -EOPNOTSUPP; 7745 if (!netif_device_present(dev)) 7746 return -ENODEV; 7747 return ops->ndo_change_proto_down(dev, proto_down); 7748 } 7749 EXPORT_SYMBOL(dev_change_proto_down); 7750 7751 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op, 7752 enum bpf_netdev_command cmd) 7753 { 7754 struct netdev_bpf xdp; 7755 7756 if (!bpf_op) 7757 return 0; 7758 7759 memset(&xdp, 0, sizeof(xdp)); 7760 xdp.command = cmd; 7761 7762 /* Query must always succeed. */ 7763 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG); 7764 7765 return xdp.prog_id; 7766 } 7767 7768 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, 7769 struct netlink_ext_ack *extack, u32 flags, 7770 struct bpf_prog *prog) 7771 { 7772 struct netdev_bpf xdp; 7773 7774 memset(&xdp, 0, sizeof(xdp)); 7775 if (flags & XDP_FLAGS_HW_MODE) 7776 xdp.command = XDP_SETUP_PROG_HW; 7777 else 7778 xdp.command = XDP_SETUP_PROG; 7779 xdp.extack = extack; 7780 xdp.flags = flags; 7781 xdp.prog = prog; 7782 7783 return bpf_op(dev, &xdp); 7784 } 7785 7786 static void dev_xdp_uninstall(struct net_device *dev) 7787 { 7788 struct netdev_bpf xdp; 7789 bpf_op_t ndo_bpf; 7790 7791 /* Remove generic XDP */ 7792 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL)); 7793 7794 /* Remove from the driver */ 7795 ndo_bpf = dev->netdev_ops->ndo_bpf; 7796 if (!ndo_bpf) 7797 return; 7798 7799 memset(&xdp, 0, sizeof(xdp)); 7800 xdp.command = XDP_QUERY_PROG; 7801 WARN_ON(ndo_bpf(dev, &xdp)); 7802 if (xdp.prog_id) 7803 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, 7804 NULL)); 7805 7806 /* Remove HW offload */ 7807 memset(&xdp, 0, sizeof(xdp)); 7808 xdp.command = XDP_QUERY_PROG_HW; 7809 if (!ndo_bpf(dev, &xdp) && xdp.prog_id) 7810 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, 7811 NULL)); 7812 } 7813 7814 /** 7815 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 7816 * @dev: device 7817 * @extack: netlink extended ack 7818 * @fd: new program fd or negative value to clear 7819 * @flags: xdp-related flags 7820 * 7821 * Set or clear a bpf program for a device 7822 */ 7823 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 7824 int fd, u32 flags) 7825 { 7826 const struct net_device_ops *ops = dev->netdev_ops; 7827 enum bpf_netdev_command query; 7828 struct bpf_prog *prog = NULL; 7829 bpf_op_t bpf_op, bpf_chk; 7830 int err; 7831 7832 ASSERT_RTNL(); 7833 7834 query = flags & XDP_FLAGS_HW_MODE ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG; 7835 7836 bpf_op = bpf_chk = ops->ndo_bpf; 7837 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) 7838 return -EOPNOTSUPP; 7839 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) 7840 bpf_op = generic_xdp_install; 7841 if (bpf_op == bpf_chk) 7842 bpf_chk = generic_xdp_install; 7843 7844 if (fd >= 0) { 7845 if (__dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG) || 7846 __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG_HW)) 7847 return -EEXIST; 7848 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && 7849 __dev_xdp_query(dev, bpf_op, query)) 7850 return -EBUSY; 7851 7852 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 7853 bpf_op == ops->ndo_bpf); 7854 if (IS_ERR(prog)) 7855 return PTR_ERR(prog); 7856 7857 if (!(flags & XDP_FLAGS_HW_MODE) && 7858 bpf_prog_is_dev_bound(prog->aux)) { 7859 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported"); 7860 bpf_prog_put(prog); 7861 return -EINVAL; 7862 } 7863 } 7864 7865 err = dev_xdp_install(dev, bpf_op, extack, flags, prog); 7866 if (err < 0 && prog) 7867 bpf_prog_put(prog); 7868 7869 return err; 7870 } 7871 7872 /** 7873 * dev_new_index - allocate an ifindex 7874 * @net: the applicable net namespace 7875 * 7876 * Returns a suitable unique value for a new device interface 7877 * number. The caller must hold the rtnl semaphore or the 7878 * dev_base_lock to be sure it remains unique. 7879 */ 7880 static int dev_new_index(struct net *net) 7881 { 7882 int ifindex = net->ifindex; 7883 7884 for (;;) { 7885 if (++ifindex <= 0) 7886 ifindex = 1; 7887 if (!__dev_get_by_index(net, ifindex)) 7888 return net->ifindex = ifindex; 7889 } 7890 } 7891 7892 /* Delayed registration/unregisteration */ 7893 static LIST_HEAD(net_todo_list); 7894 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 7895 7896 static void net_set_todo(struct net_device *dev) 7897 { 7898 list_add_tail(&dev->todo_list, &net_todo_list); 7899 dev_net(dev)->dev_unreg_count++; 7900 } 7901 7902 static void rollback_registered_many(struct list_head *head) 7903 { 7904 struct net_device *dev, *tmp; 7905 LIST_HEAD(close_head); 7906 7907 BUG_ON(dev_boot_phase); 7908 ASSERT_RTNL(); 7909 7910 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 7911 /* Some devices call without registering 7912 * for initialization unwind. Remove those 7913 * devices and proceed with the remaining. 7914 */ 7915 if (dev->reg_state == NETREG_UNINITIALIZED) { 7916 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 7917 dev->name, dev); 7918 7919 WARN_ON(1); 7920 list_del(&dev->unreg_list); 7921 continue; 7922 } 7923 dev->dismantle = true; 7924 BUG_ON(dev->reg_state != NETREG_REGISTERED); 7925 } 7926 7927 /* If device is running, close it first. */ 7928 list_for_each_entry(dev, head, unreg_list) 7929 list_add_tail(&dev->close_list, &close_head); 7930 dev_close_many(&close_head, true); 7931 7932 list_for_each_entry(dev, head, unreg_list) { 7933 /* And unlink it from device chain. */ 7934 unlist_netdevice(dev); 7935 7936 dev->reg_state = NETREG_UNREGISTERING; 7937 } 7938 flush_all_backlogs(); 7939 7940 synchronize_net(); 7941 7942 list_for_each_entry(dev, head, unreg_list) { 7943 struct sk_buff *skb = NULL; 7944 7945 /* Shutdown queueing discipline. */ 7946 dev_shutdown(dev); 7947 7948 dev_xdp_uninstall(dev); 7949 7950 /* Notify protocols, that we are about to destroy 7951 * this device. They should clean all the things. 7952 */ 7953 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 7954 7955 if (!dev->rtnl_link_ops || 7956 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 7957 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 7958 GFP_KERNEL, NULL, 0); 7959 7960 /* 7961 * Flush the unicast and multicast chains 7962 */ 7963 dev_uc_flush(dev); 7964 dev_mc_flush(dev); 7965 7966 if (dev->netdev_ops->ndo_uninit) 7967 dev->netdev_ops->ndo_uninit(dev); 7968 7969 if (skb) 7970 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 7971 7972 /* Notifier chain MUST detach us all upper devices. */ 7973 WARN_ON(netdev_has_any_upper_dev(dev)); 7974 WARN_ON(netdev_has_any_lower_dev(dev)); 7975 7976 /* Remove entries from kobject tree */ 7977 netdev_unregister_kobject(dev); 7978 #ifdef CONFIG_XPS 7979 /* Remove XPS queueing entries */ 7980 netif_reset_xps_queues_gt(dev, 0); 7981 #endif 7982 } 7983 7984 synchronize_net(); 7985 7986 list_for_each_entry(dev, head, unreg_list) 7987 dev_put(dev); 7988 } 7989 7990 static void rollback_registered(struct net_device *dev) 7991 { 7992 LIST_HEAD(single); 7993 7994 list_add(&dev->unreg_list, &single); 7995 rollback_registered_many(&single); 7996 list_del(&single); 7997 } 7998 7999 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 8000 struct net_device *upper, netdev_features_t features) 8001 { 8002 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 8003 netdev_features_t feature; 8004 int feature_bit; 8005 8006 for_each_netdev_feature(&upper_disables, feature_bit) { 8007 feature = __NETIF_F_BIT(feature_bit); 8008 if (!(upper->wanted_features & feature) 8009 && (features & feature)) { 8010 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 8011 &feature, upper->name); 8012 features &= ~feature; 8013 } 8014 } 8015 8016 return features; 8017 } 8018 8019 static void netdev_sync_lower_features(struct net_device *upper, 8020 struct net_device *lower, netdev_features_t features) 8021 { 8022 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 8023 netdev_features_t feature; 8024 int feature_bit; 8025 8026 for_each_netdev_feature(&upper_disables, feature_bit) { 8027 feature = __NETIF_F_BIT(feature_bit); 8028 if (!(features & feature) && (lower->features & feature)) { 8029 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 8030 &feature, lower->name); 8031 lower->wanted_features &= ~feature; 8032 netdev_update_features(lower); 8033 8034 if (unlikely(lower->features & feature)) 8035 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 8036 &feature, lower->name); 8037 } 8038 } 8039 } 8040 8041 static netdev_features_t netdev_fix_features(struct net_device *dev, 8042 netdev_features_t features) 8043 { 8044 /* Fix illegal checksum combinations */ 8045 if ((features & NETIF_F_HW_CSUM) && 8046 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 8047 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 8048 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 8049 } 8050 8051 /* TSO requires that SG is present as well. */ 8052 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 8053 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 8054 features &= ~NETIF_F_ALL_TSO; 8055 } 8056 8057 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 8058 !(features & NETIF_F_IP_CSUM)) { 8059 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 8060 features &= ~NETIF_F_TSO; 8061 features &= ~NETIF_F_TSO_ECN; 8062 } 8063 8064 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 8065 !(features & NETIF_F_IPV6_CSUM)) { 8066 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 8067 features &= ~NETIF_F_TSO6; 8068 } 8069 8070 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 8071 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 8072 features &= ~NETIF_F_TSO_MANGLEID; 8073 8074 /* TSO ECN requires that TSO is present as well. */ 8075 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 8076 features &= ~NETIF_F_TSO_ECN; 8077 8078 /* Software GSO depends on SG. */ 8079 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 8080 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 8081 features &= ~NETIF_F_GSO; 8082 } 8083 8084 /* GSO partial features require GSO partial be set */ 8085 if ((features & dev->gso_partial_features) && 8086 !(features & NETIF_F_GSO_PARTIAL)) { 8087 netdev_dbg(dev, 8088 "Dropping partially supported GSO features since no GSO partial.\n"); 8089 features &= ~dev->gso_partial_features; 8090 } 8091 8092 if (!(features & NETIF_F_RXCSUM)) { 8093 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 8094 * successfully merged by hardware must also have the 8095 * checksum verified by hardware. If the user does not 8096 * want to enable RXCSUM, logically, we should disable GRO_HW. 8097 */ 8098 if (features & NETIF_F_GRO_HW) { 8099 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 8100 features &= ~NETIF_F_GRO_HW; 8101 } 8102 } 8103 8104 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 8105 if (features & NETIF_F_RXFCS) { 8106 if (features & NETIF_F_LRO) { 8107 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 8108 features &= ~NETIF_F_LRO; 8109 } 8110 8111 if (features & NETIF_F_GRO_HW) { 8112 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 8113 features &= ~NETIF_F_GRO_HW; 8114 } 8115 } 8116 8117 return features; 8118 } 8119 8120 int __netdev_update_features(struct net_device *dev) 8121 { 8122 struct net_device *upper, *lower; 8123 netdev_features_t features; 8124 struct list_head *iter; 8125 int err = -1; 8126 8127 ASSERT_RTNL(); 8128 8129 features = netdev_get_wanted_features(dev); 8130 8131 if (dev->netdev_ops->ndo_fix_features) 8132 features = dev->netdev_ops->ndo_fix_features(dev, features); 8133 8134 /* driver might be less strict about feature dependencies */ 8135 features = netdev_fix_features(dev, features); 8136 8137 /* some features can't be enabled if they're off an an upper device */ 8138 netdev_for_each_upper_dev_rcu(dev, upper, iter) 8139 features = netdev_sync_upper_features(dev, upper, features); 8140 8141 if (dev->features == features) 8142 goto sync_lower; 8143 8144 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 8145 &dev->features, &features); 8146 8147 if (dev->netdev_ops->ndo_set_features) 8148 err = dev->netdev_ops->ndo_set_features(dev, features); 8149 else 8150 err = 0; 8151 8152 if (unlikely(err < 0)) { 8153 netdev_err(dev, 8154 "set_features() failed (%d); wanted %pNF, left %pNF\n", 8155 err, &features, &dev->features); 8156 /* return non-0 since some features might have changed and 8157 * it's better to fire a spurious notification than miss it 8158 */ 8159 return -1; 8160 } 8161 8162 sync_lower: 8163 /* some features must be disabled on lower devices when disabled 8164 * on an upper device (think: bonding master or bridge) 8165 */ 8166 netdev_for_each_lower_dev(dev, lower, iter) 8167 netdev_sync_lower_features(dev, lower, features); 8168 8169 if (!err) { 8170 netdev_features_t diff = features ^ dev->features; 8171 8172 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 8173 /* udp_tunnel_{get,drop}_rx_info both need 8174 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 8175 * device, or they won't do anything. 8176 * Thus we need to update dev->features 8177 * *before* calling udp_tunnel_get_rx_info, 8178 * but *after* calling udp_tunnel_drop_rx_info. 8179 */ 8180 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 8181 dev->features = features; 8182 udp_tunnel_get_rx_info(dev); 8183 } else { 8184 udp_tunnel_drop_rx_info(dev); 8185 } 8186 } 8187 8188 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 8189 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 8190 dev->features = features; 8191 err |= vlan_get_rx_ctag_filter_info(dev); 8192 } else { 8193 vlan_drop_rx_ctag_filter_info(dev); 8194 } 8195 } 8196 8197 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 8198 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 8199 dev->features = features; 8200 err |= vlan_get_rx_stag_filter_info(dev); 8201 } else { 8202 vlan_drop_rx_stag_filter_info(dev); 8203 } 8204 } 8205 8206 dev->features = features; 8207 } 8208 8209 return err < 0 ? 0 : 1; 8210 } 8211 8212 /** 8213 * netdev_update_features - recalculate device features 8214 * @dev: the device to check 8215 * 8216 * Recalculate dev->features set and send notifications if it 8217 * has changed. Should be called after driver or hardware dependent 8218 * conditions might have changed that influence the features. 8219 */ 8220 void netdev_update_features(struct net_device *dev) 8221 { 8222 if (__netdev_update_features(dev)) 8223 netdev_features_change(dev); 8224 } 8225 EXPORT_SYMBOL(netdev_update_features); 8226 8227 /** 8228 * netdev_change_features - recalculate device features 8229 * @dev: the device to check 8230 * 8231 * Recalculate dev->features set and send notifications even 8232 * if they have not changed. Should be called instead of 8233 * netdev_update_features() if also dev->vlan_features might 8234 * have changed to allow the changes to be propagated to stacked 8235 * VLAN devices. 8236 */ 8237 void netdev_change_features(struct net_device *dev) 8238 { 8239 __netdev_update_features(dev); 8240 netdev_features_change(dev); 8241 } 8242 EXPORT_SYMBOL(netdev_change_features); 8243 8244 /** 8245 * netif_stacked_transfer_operstate - transfer operstate 8246 * @rootdev: the root or lower level device to transfer state from 8247 * @dev: the device to transfer operstate to 8248 * 8249 * Transfer operational state from root to device. This is normally 8250 * called when a stacking relationship exists between the root 8251 * device and the device(a leaf device). 8252 */ 8253 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 8254 struct net_device *dev) 8255 { 8256 if (rootdev->operstate == IF_OPER_DORMANT) 8257 netif_dormant_on(dev); 8258 else 8259 netif_dormant_off(dev); 8260 8261 if (netif_carrier_ok(rootdev)) 8262 netif_carrier_on(dev); 8263 else 8264 netif_carrier_off(dev); 8265 } 8266 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 8267 8268 static int netif_alloc_rx_queues(struct net_device *dev) 8269 { 8270 unsigned int i, count = dev->num_rx_queues; 8271 struct netdev_rx_queue *rx; 8272 size_t sz = count * sizeof(*rx); 8273 int err = 0; 8274 8275 BUG_ON(count < 1); 8276 8277 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 8278 if (!rx) 8279 return -ENOMEM; 8280 8281 dev->_rx = rx; 8282 8283 for (i = 0; i < count; i++) { 8284 rx[i].dev = dev; 8285 8286 /* XDP RX-queue setup */ 8287 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i); 8288 if (err < 0) 8289 goto err_rxq_info; 8290 } 8291 return 0; 8292 8293 err_rxq_info: 8294 /* Rollback successful reg's and free other resources */ 8295 while (i--) 8296 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 8297 kvfree(dev->_rx); 8298 dev->_rx = NULL; 8299 return err; 8300 } 8301 8302 static void netif_free_rx_queues(struct net_device *dev) 8303 { 8304 unsigned int i, count = dev->num_rx_queues; 8305 8306 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 8307 if (!dev->_rx) 8308 return; 8309 8310 for (i = 0; i < count; i++) 8311 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 8312 8313 kvfree(dev->_rx); 8314 } 8315 8316 static void netdev_init_one_queue(struct net_device *dev, 8317 struct netdev_queue *queue, void *_unused) 8318 { 8319 /* Initialize queue lock */ 8320 spin_lock_init(&queue->_xmit_lock); 8321 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 8322 queue->xmit_lock_owner = -1; 8323 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 8324 queue->dev = dev; 8325 #ifdef CONFIG_BQL 8326 dql_init(&queue->dql, HZ); 8327 #endif 8328 } 8329 8330 static void netif_free_tx_queues(struct net_device *dev) 8331 { 8332 kvfree(dev->_tx); 8333 } 8334 8335 static int netif_alloc_netdev_queues(struct net_device *dev) 8336 { 8337 unsigned int count = dev->num_tx_queues; 8338 struct netdev_queue *tx; 8339 size_t sz = count * sizeof(*tx); 8340 8341 if (count < 1 || count > 0xffff) 8342 return -EINVAL; 8343 8344 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 8345 if (!tx) 8346 return -ENOMEM; 8347 8348 dev->_tx = tx; 8349 8350 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 8351 spin_lock_init(&dev->tx_global_lock); 8352 8353 return 0; 8354 } 8355 8356 void netif_tx_stop_all_queues(struct net_device *dev) 8357 { 8358 unsigned int i; 8359 8360 for (i = 0; i < dev->num_tx_queues; i++) { 8361 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 8362 8363 netif_tx_stop_queue(txq); 8364 } 8365 } 8366 EXPORT_SYMBOL(netif_tx_stop_all_queues); 8367 8368 /** 8369 * register_netdevice - register a network device 8370 * @dev: device to register 8371 * 8372 * Take a completed network device structure and add it to the kernel 8373 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 8374 * chain. 0 is returned on success. A negative errno code is returned 8375 * on a failure to set up the device, or if the name is a duplicate. 8376 * 8377 * Callers must hold the rtnl semaphore. You may want 8378 * register_netdev() instead of this. 8379 * 8380 * BUGS: 8381 * The locking appears insufficient to guarantee two parallel registers 8382 * will not get the same name. 8383 */ 8384 8385 int register_netdevice(struct net_device *dev) 8386 { 8387 int ret; 8388 struct net *net = dev_net(dev); 8389 8390 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 8391 NETDEV_FEATURE_COUNT); 8392 BUG_ON(dev_boot_phase); 8393 ASSERT_RTNL(); 8394 8395 might_sleep(); 8396 8397 /* When net_device's are persistent, this will be fatal. */ 8398 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 8399 BUG_ON(!net); 8400 8401 spin_lock_init(&dev->addr_list_lock); 8402 netdev_set_addr_lockdep_class(dev); 8403 8404 ret = dev_get_valid_name(net, dev, dev->name); 8405 if (ret < 0) 8406 goto out; 8407 8408 /* Init, if this function is available */ 8409 if (dev->netdev_ops->ndo_init) { 8410 ret = dev->netdev_ops->ndo_init(dev); 8411 if (ret) { 8412 if (ret > 0) 8413 ret = -EIO; 8414 goto out; 8415 } 8416 } 8417 8418 if (((dev->hw_features | dev->features) & 8419 NETIF_F_HW_VLAN_CTAG_FILTER) && 8420 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 8421 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 8422 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 8423 ret = -EINVAL; 8424 goto err_uninit; 8425 } 8426 8427 ret = -EBUSY; 8428 if (!dev->ifindex) 8429 dev->ifindex = dev_new_index(net); 8430 else if (__dev_get_by_index(net, dev->ifindex)) 8431 goto err_uninit; 8432 8433 /* Transfer changeable features to wanted_features and enable 8434 * software offloads (GSO and GRO). 8435 */ 8436 dev->hw_features |= NETIF_F_SOFT_FEATURES; 8437 dev->features |= NETIF_F_SOFT_FEATURES; 8438 8439 if (dev->netdev_ops->ndo_udp_tunnel_add) { 8440 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 8441 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 8442 } 8443 8444 dev->wanted_features = dev->features & dev->hw_features; 8445 8446 if (!(dev->flags & IFF_LOOPBACK)) 8447 dev->hw_features |= NETIF_F_NOCACHE_COPY; 8448 8449 /* If IPv4 TCP segmentation offload is supported we should also 8450 * allow the device to enable segmenting the frame with the option 8451 * of ignoring a static IP ID value. This doesn't enable the 8452 * feature itself but allows the user to enable it later. 8453 */ 8454 if (dev->hw_features & NETIF_F_TSO) 8455 dev->hw_features |= NETIF_F_TSO_MANGLEID; 8456 if (dev->vlan_features & NETIF_F_TSO) 8457 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 8458 if (dev->mpls_features & NETIF_F_TSO) 8459 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 8460 if (dev->hw_enc_features & NETIF_F_TSO) 8461 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 8462 8463 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 8464 */ 8465 dev->vlan_features |= NETIF_F_HIGHDMA; 8466 8467 /* Make NETIF_F_SG inheritable to tunnel devices. 8468 */ 8469 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 8470 8471 /* Make NETIF_F_SG inheritable to MPLS. 8472 */ 8473 dev->mpls_features |= NETIF_F_SG; 8474 8475 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 8476 ret = notifier_to_errno(ret); 8477 if (ret) 8478 goto err_uninit; 8479 8480 ret = netdev_register_kobject(dev); 8481 if (ret) 8482 goto err_uninit; 8483 dev->reg_state = NETREG_REGISTERED; 8484 8485 __netdev_update_features(dev); 8486 8487 /* 8488 * Default initial state at registry is that the 8489 * device is present. 8490 */ 8491 8492 set_bit(__LINK_STATE_PRESENT, &dev->state); 8493 8494 linkwatch_init_dev(dev); 8495 8496 dev_init_scheduler(dev); 8497 dev_hold(dev); 8498 list_netdevice(dev); 8499 add_device_randomness(dev->dev_addr, dev->addr_len); 8500 8501 /* If the device has permanent device address, driver should 8502 * set dev_addr and also addr_assign_type should be set to 8503 * NET_ADDR_PERM (default value). 8504 */ 8505 if (dev->addr_assign_type == NET_ADDR_PERM) 8506 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 8507 8508 /* Notify protocols, that a new device appeared. */ 8509 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 8510 ret = notifier_to_errno(ret); 8511 if (ret) { 8512 rollback_registered(dev); 8513 dev->reg_state = NETREG_UNREGISTERED; 8514 } 8515 /* 8516 * Prevent userspace races by waiting until the network 8517 * device is fully setup before sending notifications. 8518 */ 8519 if (!dev->rtnl_link_ops || 8520 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 8521 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 8522 8523 out: 8524 return ret; 8525 8526 err_uninit: 8527 if (dev->netdev_ops->ndo_uninit) 8528 dev->netdev_ops->ndo_uninit(dev); 8529 if (dev->priv_destructor) 8530 dev->priv_destructor(dev); 8531 goto out; 8532 } 8533 EXPORT_SYMBOL(register_netdevice); 8534 8535 /** 8536 * init_dummy_netdev - init a dummy network device for NAPI 8537 * @dev: device to init 8538 * 8539 * This takes a network device structure and initialize the minimum 8540 * amount of fields so it can be used to schedule NAPI polls without 8541 * registering a full blown interface. This is to be used by drivers 8542 * that need to tie several hardware interfaces to a single NAPI 8543 * poll scheduler due to HW limitations. 8544 */ 8545 int init_dummy_netdev(struct net_device *dev) 8546 { 8547 /* Clear everything. Note we don't initialize spinlocks 8548 * are they aren't supposed to be taken by any of the 8549 * NAPI code and this dummy netdev is supposed to be 8550 * only ever used for NAPI polls 8551 */ 8552 memset(dev, 0, sizeof(struct net_device)); 8553 8554 /* make sure we BUG if trying to hit standard 8555 * register/unregister code path 8556 */ 8557 dev->reg_state = NETREG_DUMMY; 8558 8559 /* NAPI wants this */ 8560 INIT_LIST_HEAD(&dev->napi_list); 8561 8562 /* a dummy interface is started by default */ 8563 set_bit(__LINK_STATE_PRESENT, &dev->state); 8564 set_bit(__LINK_STATE_START, &dev->state); 8565 8566 /* Note : We dont allocate pcpu_refcnt for dummy devices, 8567 * because users of this 'device' dont need to change 8568 * its refcount. 8569 */ 8570 8571 return 0; 8572 } 8573 EXPORT_SYMBOL_GPL(init_dummy_netdev); 8574 8575 8576 /** 8577 * register_netdev - register a network device 8578 * @dev: device to register 8579 * 8580 * Take a completed network device structure and add it to the kernel 8581 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 8582 * chain. 0 is returned on success. A negative errno code is returned 8583 * on a failure to set up the device, or if the name is a duplicate. 8584 * 8585 * This is a wrapper around register_netdevice that takes the rtnl semaphore 8586 * and expands the device name if you passed a format string to 8587 * alloc_netdev. 8588 */ 8589 int register_netdev(struct net_device *dev) 8590 { 8591 int err; 8592 8593 if (rtnl_lock_killable()) 8594 return -EINTR; 8595 err = register_netdevice(dev); 8596 rtnl_unlock(); 8597 return err; 8598 } 8599 EXPORT_SYMBOL(register_netdev); 8600 8601 int netdev_refcnt_read(const struct net_device *dev) 8602 { 8603 int i, refcnt = 0; 8604 8605 for_each_possible_cpu(i) 8606 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 8607 return refcnt; 8608 } 8609 EXPORT_SYMBOL(netdev_refcnt_read); 8610 8611 /** 8612 * netdev_wait_allrefs - wait until all references are gone. 8613 * @dev: target net_device 8614 * 8615 * This is called when unregistering network devices. 8616 * 8617 * Any protocol or device that holds a reference should register 8618 * for netdevice notification, and cleanup and put back the 8619 * reference if they receive an UNREGISTER event. 8620 * We can get stuck here if buggy protocols don't correctly 8621 * call dev_put. 8622 */ 8623 static void netdev_wait_allrefs(struct net_device *dev) 8624 { 8625 unsigned long rebroadcast_time, warning_time; 8626 int refcnt; 8627 8628 linkwatch_forget_dev(dev); 8629 8630 rebroadcast_time = warning_time = jiffies; 8631 refcnt = netdev_refcnt_read(dev); 8632 8633 while (refcnt != 0) { 8634 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 8635 rtnl_lock(); 8636 8637 /* Rebroadcast unregister notification */ 8638 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 8639 8640 __rtnl_unlock(); 8641 rcu_barrier(); 8642 rtnl_lock(); 8643 8644 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 8645 &dev->state)) { 8646 /* We must not have linkwatch events 8647 * pending on unregister. If this 8648 * happens, we simply run the queue 8649 * unscheduled, resulting in a noop 8650 * for this device. 8651 */ 8652 linkwatch_run_queue(); 8653 } 8654 8655 __rtnl_unlock(); 8656 8657 rebroadcast_time = jiffies; 8658 } 8659 8660 msleep(250); 8661 8662 refcnt = netdev_refcnt_read(dev); 8663 8664 if (time_after(jiffies, warning_time + 10 * HZ)) { 8665 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 8666 dev->name, refcnt); 8667 warning_time = jiffies; 8668 } 8669 } 8670 } 8671 8672 /* The sequence is: 8673 * 8674 * rtnl_lock(); 8675 * ... 8676 * register_netdevice(x1); 8677 * register_netdevice(x2); 8678 * ... 8679 * unregister_netdevice(y1); 8680 * unregister_netdevice(y2); 8681 * ... 8682 * rtnl_unlock(); 8683 * free_netdev(y1); 8684 * free_netdev(y2); 8685 * 8686 * We are invoked by rtnl_unlock(). 8687 * This allows us to deal with problems: 8688 * 1) We can delete sysfs objects which invoke hotplug 8689 * without deadlocking with linkwatch via keventd. 8690 * 2) Since we run with the RTNL semaphore not held, we can sleep 8691 * safely in order to wait for the netdev refcnt to drop to zero. 8692 * 8693 * We must not return until all unregister events added during 8694 * the interval the lock was held have been completed. 8695 */ 8696 void netdev_run_todo(void) 8697 { 8698 struct list_head list; 8699 8700 /* Snapshot list, allow later requests */ 8701 list_replace_init(&net_todo_list, &list); 8702 8703 __rtnl_unlock(); 8704 8705 8706 /* Wait for rcu callbacks to finish before next phase */ 8707 if (!list_empty(&list)) 8708 rcu_barrier(); 8709 8710 while (!list_empty(&list)) { 8711 struct net_device *dev 8712 = list_first_entry(&list, struct net_device, todo_list); 8713 list_del(&dev->todo_list); 8714 8715 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 8716 pr_err("network todo '%s' but state %d\n", 8717 dev->name, dev->reg_state); 8718 dump_stack(); 8719 continue; 8720 } 8721 8722 dev->reg_state = NETREG_UNREGISTERED; 8723 8724 netdev_wait_allrefs(dev); 8725 8726 /* paranoia */ 8727 BUG_ON(netdev_refcnt_read(dev)); 8728 BUG_ON(!list_empty(&dev->ptype_all)); 8729 BUG_ON(!list_empty(&dev->ptype_specific)); 8730 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 8731 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 8732 #if IS_ENABLED(CONFIG_DECNET) 8733 WARN_ON(dev->dn_ptr); 8734 #endif 8735 if (dev->priv_destructor) 8736 dev->priv_destructor(dev); 8737 if (dev->needs_free_netdev) 8738 free_netdev(dev); 8739 8740 /* Report a network device has been unregistered */ 8741 rtnl_lock(); 8742 dev_net(dev)->dev_unreg_count--; 8743 __rtnl_unlock(); 8744 wake_up(&netdev_unregistering_wq); 8745 8746 /* Free network device */ 8747 kobject_put(&dev->dev.kobj); 8748 } 8749 } 8750 8751 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 8752 * all the same fields in the same order as net_device_stats, with only 8753 * the type differing, but rtnl_link_stats64 may have additional fields 8754 * at the end for newer counters. 8755 */ 8756 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 8757 const struct net_device_stats *netdev_stats) 8758 { 8759 #if BITS_PER_LONG == 64 8760 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 8761 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 8762 /* zero out counters that only exist in rtnl_link_stats64 */ 8763 memset((char *)stats64 + sizeof(*netdev_stats), 0, 8764 sizeof(*stats64) - sizeof(*netdev_stats)); 8765 #else 8766 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 8767 const unsigned long *src = (const unsigned long *)netdev_stats; 8768 u64 *dst = (u64 *)stats64; 8769 8770 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 8771 for (i = 0; i < n; i++) 8772 dst[i] = src[i]; 8773 /* zero out counters that only exist in rtnl_link_stats64 */ 8774 memset((char *)stats64 + n * sizeof(u64), 0, 8775 sizeof(*stats64) - n * sizeof(u64)); 8776 #endif 8777 } 8778 EXPORT_SYMBOL(netdev_stats_to_stats64); 8779 8780 /** 8781 * dev_get_stats - get network device statistics 8782 * @dev: device to get statistics from 8783 * @storage: place to store stats 8784 * 8785 * Get network statistics from device. Return @storage. 8786 * The device driver may provide its own method by setting 8787 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 8788 * otherwise the internal statistics structure is used. 8789 */ 8790 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 8791 struct rtnl_link_stats64 *storage) 8792 { 8793 const struct net_device_ops *ops = dev->netdev_ops; 8794 8795 if (ops->ndo_get_stats64) { 8796 memset(storage, 0, sizeof(*storage)); 8797 ops->ndo_get_stats64(dev, storage); 8798 } else if (ops->ndo_get_stats) { 8799 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 8800 } else { 8801 netdev_stats_to_stats64(storage, &dev->stats); 8802 } 8803 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 8804 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 8805 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 8806 return storage; 8807 } 8808 EXPORT_SYMBOL(dev_get_stats); 8809 8810 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 8811 { 8812 struct netdev_queue *queue = dev_ingress_queue(dev); 8813 8814 #ifdef CONFIG_NET_CLS_ACT 8815 if (queue) 8816 return queue; 8817 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 8818 if (!queue) 8819 return NULL; 8820 netdev_init_one_queue(dev, queue, NULL); 8821 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 8822 queue->qdisc_sleeping = &noop_qdisc; 8823 rcu_assign_pointer(dev->ingress_queue, queue); 8824 #endif 8825 return queue; 8826 } 8827 8828 static const struct ethtool_ops default_ethtool_ops; 8829 8830 void netdev_set_default_ethtool_ops(struct net_device *dev, 8831 const struct ethtool_ops *ops) 8832 { 8833 if (dev->ethtool_ops == &default_ethtool_ops) 8834 dev->ethtool_ops = ops; 8835 } 8836 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 8837 8838 void netdev_freemem(struct net_device *dev) 8839 { 8840 char *addr = (char *)dev - dev->padded; 8841 8842 kvfree(addr); 8843 } 8844 8845 /** 8846 * alloc_netdev_mqs - allocate network device 8847 * @sizeof_priv: size of private data to allocate space for 8848 * @name: device name format string 8849 * @name_assign_type: origin of device name 8850 * @setup: callback to initialize device 8851 * @txqs: the number of TX subqueues to allocate 8852 * @rxqs: the number of RX subqueues to allocate 8853 * 8854 * Allocates a struct net_device with private data area for driver use 8855 * and performs basic initialization. Also allocates subqueue structs 8856 * for each queue on the device. 8857 */ 8858 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 8859 unsigned char name_assign_type, 8860 void (*setup)(struct net_device *), 8861 unsigned int txqs, unsigned int rxqs) 8862 { 8863 struct net_device *dev; 8864 unsigned int alloc_size; 8865 struct net_device *p; 8866 8867 BUG_ON(strlen(name) >= sizeof(dev->name)); 8868 8869 if (txqs < 1) { 8870 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 8871 return NULL; 8872 } 8873 8874 if (rxqs < 1) { 8875 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 8876 return NULL; 8877 } 8878 8879 alloc_size = sizeof(struct net_device); 8880 if (sizeof_priv) { 8881 /* ensure 32-byte alignment of private area */ 8882 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 8883 alloc_size += sizeof_priv; 8884 } 8885 /* ensure 32-byte alignment of whole construct */ 8886 alloc_size += NETDEV_ALIGN - 1; 8887 8888 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 8889 if (!p) 8890 return NULL; 8891 8892 dev = PTR_ALIGN(p, NETDEV_ALIGN); 8893 dev->padded = (char *)dev - (char *)p; 8894 8895 dev->pcpu_refcnt = alloc_percpu(int); 8896 if (!dev->pcpu_refcnt) 8897 goto free_dev; 8898 8899 if (dev_addr_init(dev)) 8900 goto free_pcpu; 8901 8902 dev_mc_init(dev); 8903 dev_uc_init(dev); 8904 8905 dev_net_set(dev, &init_net); 8906 8907 dev->gso_max_size = GSO_MAX_SIZE; 8908 dev->gso_max_segs = GSO_MAX_SEGS; 8909 8910 INIT_LIST_HEAD(&dev->napi_list); 8911 INIT_LIST_HEAD(&dev->unreg_list); 8912 INIT_LIST_HEAD(&dev->close_list); 8913 INIT_LIST_HEAD(&dev->link_watch_list); 8914 INIT_LIST_HEAD(&dev->adj_list.upper); 8915 INIT_LIST_HEAD(&dev->adj_list.lower); 8916 INIT_LIST_HEAD(&dev->ptype_all); 8917 INIT_LIST_HEAD(&dev->ptype_specific); 8918 #ifdef CONFIG_NET_SCHED 8919 hash_init(dev->qdisc_hash); 8920 #endif 8921 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 8922 setup(dev); 8923 8924 if (!dev->tx_queue_len) { 8925 dev->priv_flags |= IFF_NO_QUEUE; 8926 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 8927 } 8928 8929 dev->num_tx_queues = txqs; 8930 dev->real_num_tx_queues = txqs; 8931 if (netif_alloc_netdev_queues(dev)) 8932 goto free_all; 8933 8934 dev->num_rx_queues = rxqs; 8935 dev->real_num_rx_queues = rxqs; 8936 if (netif_alloc_rx_queues(dev)) 8937 goto free_all; 8938 8939 strcpy(dev->name, name); 8940 dev->name_assign_type = name_assign_type; 8941 dev->group = INIT_NETDEV_GROUP; 8942 if (!dev->ethtool_ops) 8943 dev->ethtool_ops = &default_ethtool_ops; 8944 8945 nf_hook_ingress_init(dev); 8946 8947 return dev; 8948 8949 free_all: 8950 free_netdev(dev); 8951 return NULL; 8952 8953 free_pcpu: 8954 free_percpu(dev->pcpu_refcnt); 8955 free_dev: 8956 netdev_freemem(dev); 8957 return NULL; 8958 } 8959 EXPORT_SYMBOL(alloc_netdev_mqs); 8960 8961 /** 8962 * free_netdev - free network device 8963 * @dev: device 8964 * 8965 * This function does the last stage of destroying an allocated device 8966 * interface. The reference to the device object is released. If this 8967 * is the last reference then it will be freed.Must be called in process 8968 * context. 8969 */ 8970 void free_netdev(struct net_device *dev) 8971 { 8972 struct napi_struct *p, *n; 8973 8974 might_sleep(); 8975 netif_free_tx_queues(dev); 8976 netif_free_rx_queues(dev); 8977 8978 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 8979 8980 /* Flush device addresses */ 8981 dev_addr_flush(dev); 8982 8983 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 8984 netif_napi_del(p); 8985 8986 free_percpu(dev->pcpu_refcnt); 8987 dev->pcpu_refcnt = NULL; 8988 8989 /* Compatibility with error handling in drivers */ 8990 if (dev->reg_state == NETREG_UNINITIALIZED) { 8991 netdev_freemem(dev); 8992 return; 8993 } 8994 8995 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 8996 dev->reg_state = NETREG_RELEASED; 8997 8998 /* will free via device release */ 8999 put_device(&dev->dev); 9000 } 9001 EXPORT_SYMBOL(free_netdev); 9002 9003 /** 9004 * synchronize_net - Synchronize with packet receive processing 9005 * 9006 * Wait for packets currently being received to be done. 9007 * Does not block later packets from starting. 9008 */ 9009 void synchronize_net(void) 9010 { 9011 might_sleep(); 9012 if (rtnl_is_locked()) 9013 synchronize_rcu_expedited(); 9014 else 9015 synchronize_rcu(); 9016 } 9017 EXPORT_SYMBOL(synchronize_net); 9018 9019 /** 9020 * unregister_netdevice_queue - remove device from the kernel 9021 * @dev: device 9022 * @head: list 9023 * 9024 * This function shuts down a device interface and removes it 9025 * from the kernel tables. 9026 * If head not NULL, device is queued to be unregistered later. 9027 * 9028 * Callers must hold the rtnl semaphore. You may want 9029 * unregister_netdev() instead of this. 9030 */ 9031 9032 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 9033 { 9034 ASSERT_RTNL(); 9035 9036 if (head) { 9037 list_move_tail(&dev->unreg_list, head); 9038 } else { 9039 rollback_registered(dev); 9040 /* Finish processing unregister after unlock */ 9041 net_set_todo(dev); 9042 } 9043 } 9044 EXPORT_SYMBOL(unregister_netdevice_queue); 9045 9046 /** 9047 * unregister_netdevice_many - unregister many devices 9048 * @head: list of devices 9049 * 9050 * Note: As most callers use a stack allocated list_head, 9051 * we force a list_del() to make sure stack wont be corrupted later. 9052 */ 9053 void unregister_netdevice_many(struct list_head *head) 9054 { 9055 struct net_device *dev; 9056 9057 if (!list_empty(head)) { 9058 rollback_registered_many(head); 9059 list_for_each_entry(dev, head, unreg_list) 9060 net_set_todo(dev); 9061 list_del(head); 9062 } 9063 } 9064 EXPORT_SYMBOL(unregister_netdevice_many); 9065 9066 /** 9067 * unregister_netdev - remove device from the kernel 9068 * @dev: device 9069 * 9070 * This function shuts down a device interface and removes it 9071 * from the kernel tables. 9072 * 9073 * This is just a wrapper for unregister_netdevice that takes 9074 * the rtnl semaphore. In general you want to use this and not 9075 * unregister_netdevice. 9076 */ 9077 void unregister_netdev(struct net_device *dev) 9078 { 9079 rtnl_lock(); 9080 unregister_netdevice(dev); 9081 rtnl_unlock(); 9082 } 9083 EXPORT_SYMBOL(unregister_netdev); 9084 9085 /** 9086 * dev_change_net_namespace - move device to different nethost namespace 9087 * @dev: device 9088 * @net: network namespace 9089 * @pat: If not NULL name pattern to try if the current device name 9090 * is already taken in the destination network namespace. 9091 * 9092 * This function shuts down a device interface and moves it 9093 * to a new network namespace. On success 0 is returned, on 9094 * a failure a netagive errno code is returned. 9095 * 9096 * Callers must hold the rtnl semaphore. 9097 */ 9098 9099 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 9100 { 9101 int err, new_nsid, new_ifindex; 9102 9103 ASSERT_RTNL(); 9104 9105 /* Don't allow namespace local devices to be moved. */ 9106 err = -EINVAL; 9107 if (dev->features & NETIF_F_NETNS_LOCAL) 9108 goto out; 9109 9110 /* Ensure the device has been registrered */ 9111 if (dev->reg_state != NETREG_REGISTERED) 9112 goto out; 9113 9114 /* Get out if there is nothing todo */ 9115 err = 0; 9116 if (net_eq(dev_net(dev), net)) 9117 goto out; 9118 9119 /* Pick the destination device name, and ensure 9120 * we can use it in the destination network namespace. 9121 */ 9122 err = -EEXIST; 9123 if (__dev_get_by_name(net, dev->name)) { 9124 /* We get here if we can't use the current device name */ 9125 if (!pat) 9126 goto out; 9127 err = dev_get_valid_name(net, dev, pat); 9128 if (err < 0) 9129 goto out; 9130 } 9131 9132 /* 9133 * And now a mini version of register_netdevice unregister_netdevice. 9134 */ 9135 9136 /* If device is running close it first. */ 9137 dev_close(dev); 9138 9139 /* And unlink it from device chain */ 9140 unlist_netdevice(dev); 9141 9142 synchronize_net(); 9143 9144 /* Shutdown queueing discipline. */ 9145 dev_shutdown(dev); 9146 9147 /* Notify protocols, that we are about to destroy 9148 * this device. They should clean all the things. 9149 * 9150 * Note that dev->reg_state stays at NETREG_REGISTERED. 9151 * This is wanted because this way 8021q and macvlan know 9152 * the device is just moving and can keep their slaves up. 9153 */ 9154 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 9155 rcu_barrier(); 9156 9157 new_nsid = peernet2id_alloc(dev_net(dev), net); 9158 /* If there is an ifindex conflict assign a new one */ 9159 if (__dev_get_by_index(net, dev->ifindex)) 9160 new_ifindex = dev_new_index(net); 9161 else 9162 new_ifindex = dev->ifindex; 9163 9164 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 9165 new_ifindex); 9166 9167 /* 9168 * Flush the unicast and multicast chains 9169 */ 9170 dev_uc_flush(dev); 9171 dev_mc_flush(dev); 9172 9173 /* Send a netdev-removed uevent to the old namespace */ 9174 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 9175 netdev_adjacent_del_links(dev); 9176 9177 /* Actually switch the network namespace */ 9178 dev_net_set(dev, net); 9179 dev->ifindex = new_ifindex; 9180 9181 /* Send a netdev-add uevent to the new namespace */ 9182 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 9183 netdev_adjacent_add_links(dev); 9184 9185 /* Fixup kobjects */ 9186 err = device_rename(&dev->dev, dev->name); 9187 WARN_ON(err); 9188 9189 /* Add the device back in the hashes */ 9190 list_netdevice(dev); 9191 9192 /* Notify protocols, that a new device appeared. */ 9193 call_netdevice_notifiers(NETDEV_REGISTER, dev); 9194 9195 /* 9196 * Prevent userspace races by waiting until the network 9197 * device is fully setup before sending notifications. 9198 */ 9199 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 9200 9201 synchronize_net(); 9202 err = 0; 9203 out: 9204 return err; 9205 } 9206 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 9207 9208 static int dev_cpu_dead(unsigned int oldcpu) 9209 { 9210 struct sk_buff **list_skb; 9211 struct sk_buff *skb; 9212 unsigned int cpu; 9213 struct softnet_data *sd, *oldsd, *remsd = NULL; 9214 9215 local_irq_disable(); 9216 cpu = smp_processor_id(); 9217 sd = &per_cpu(softnet_data, cpu); 9218 oldsd = &per_cpu(softnet_data, oldcpu); 9219 9220 /* Find end of our completion_queue. */ 9221 list_skb = &sd->completion_queue; 9222 while (*list_skb) 9223 list_skb = &(*list_skb)->next; 9224 /* Append completion queue from offline CPU. */ 9225 *list_skb = oldsd->completion_queue; 9226 oldsd->completion_queue = NULL; 9227 9228 /* Append output queue from offline CPU. */ 9229 if (oldsd->output_queue) { 9230 *sd->output_queue_tailp = oldsd->output_queue; 9231 sd->output_queue_tailp = oldsd->output_queue_tailp; 9232 oldsd->output_queue = NULL; 9233 oldsd->output_queue_tailp = &oldsd->output_queue; 9234 } 9235 /* Append NAPI poll list from offline CPU, with one exception : 9236 * process_backlog() must be called by cpu owning percpu backlog. 9237 * We properly handle process_queue & input_pkt_queue later. 9238 */ 9239 while (!list_empty(&oldsd->poll_list)) { 9240 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 9241 struct napi_struct, 9242 poll_list); 9243 9244 list_del_init(&napi->poll_list); 9245 if (napi->poll == process_backlog) 9246 napi->state = 0; 9247 else 9248 ____napi_schedule(sd, napi); 9249 } 9250 9251 raise_softirq_irqoff(NET_TX_SOFTIRQ); 9252 local_irq_enable(); 9253 9254 #ifdef CONFIG_RPS 9255 remsd = oldsd->rps_ipi_list; 9256 oldsd->rps_ipi_list = NULL; 9257 #endif 9258 /* send out pending IPI's on offline CPU */ 9259 net_rps_send_ipi(remsd); 9260 9261 /* Process offline CPU's input_pkt_queue */ 9262 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 9263 netif_rx_ni(skb); 9264 input_queue_head_incr(oldsd); 9265 } 9266 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 9267 netif_rx_ni(skb); 9268 input_queue_head_incr(oldsd); 9269 } 9270 9271 return 0; 9272 } 9273 9274 /** 9275 * netdev_increment_features - increment feature set by one 9276 * @all: current feature set 9277 * @one: new feature set 9278 * @mask: mask feature set 9279 * 9280 * Computes a new feature set after adding a device with feature set 9281 * @one to the master device with current feature set @all. Will not 9282 * enable anything that is off in @mask. Returns the new feature set. 9283 */ 9284 netdev_features_t netdev_increment_features(netdev_features_t all, 9285 netdev_features_t one, netdev_features_t mask) 9286 { 9287 if (mask & NETIF_F_HW_CSUM) 9288 mask |= NETIF_F_CSUM_MASK; 9289 mask |= NETIF_F_VLAN_CHALLENGED; 9290 9291 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 9292 all &= one | ~NETIF_F_ALL_FOR_ALL; 9293 9294 /* If one device supports hw checksumming, set for all. */ 9295 if (all & NETIF_F_HW_CSUM) 9296 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 9297 9298 return all; 9299 } 9300 EXPORT_SYMBOL(netdev_increment_features); 9301 9302 static struct hlist_head * __net_init netdev_create_hash(void) 9303 { 9304 int i; 9305 struct hlist_head *hash; 9306 9307 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 9308 if (hash != NULL) 9309 for (i = 0; i < NETDEV_HASHENTRIES; i++) 9310 INIT_HLIST_HEAD(&hash[i]); 9311 9312 return hash; 9313 } 9314 9315 /* Initialize per network namespace state */ 9316 static int __net_init netdev_init(struct net *net) 9317 { 9318 BUILD_BUG_ON(GRO_HASH_BUCKETS > 9319 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask)); 9320 9321 if (net != &init_net) 9322 INIT_LIST_HEAD(&net->dev_base_head); 9323 9324 net->dev_name_head = netdev_create_hash(); 9325 if (net->dev_name_head == NULL) 9326 goto err_name; 9327 9328 net->dev_index_head = netdev_create_hash(); 9329 if (net->dev_index_head == NULL) 9330 goto err_idx; 9331 9332 return 0; 9333 9334 err_idx: 9335 kfree(net->dev_name_head); 9336 err_name: 9337 return -ENOMEM; 9338 } 9339 9340 /** 9341 * netdev_drivername - network driver for the device 9342 * @dev: network device 9343 * 9344 * Determine network driver for device. 9345 */ 9346 const char *netdev_drivername(const struct net_device *dev) 9347 { 9348 const struct device_driver *driver; 9349 const struct device *parent; 9350 const char *empty = ""; 9351 9352 parent = dev->dev.parent; 9353 if (!parent) 9354 return empty; 9355 9356 driver = parent->driver; 9357 if (driver && driver->name) 9358 return driver->name; 9359 return empty; 9360 } 9361 9362 static void __netdev_printk(const char *level, const struct net_device *dev, 9363 struct va_format *vaf) 9364 { 9365 if (dev && dev->dev.parent) { 9366 dev_printk_emit(level[1] - '0', 9367 dev->dev.parent, 9368 "%s %s %s%s: %pV", 9369 dev_driver_string(dev->dev.parent), 9370 dev_name(dev->dev.parent), 9371 netdev_name(dev), netdev_reg_state(dev), 9372 vaf); 9373 } else if (dev) { 9374 printk("%s%s%s: %pV", 9375 level, netdev_name(dev), netdev_reg_state(dev), vaf); 9376 } else { 9377 printk("%s(NULL net_device): %pV", level, vaf); 9378 } 9379 } 9380 9381 void netdev_printk(const char *level, const struct net_device *dev, 9382 const char *format, ...) 9383 { 9384 struct va_format vaf; 9385 va_list args; 9386 9387 va_start(args, format); 9388 9389 vaf.fmt = format; 9390 vaf.va = &args; 9391 9392 __netdev_printk(level, dev, &vaf); 9393 9394 va_end(args); 9395 } 9396 EXPORT_SYMBOL(netdev_printk); 9397 9398 #define define_netdev_printk_level(func, level) \ 9399 void func(const struct net_device *dev, const char *fmt, ...) \ 9400 { \ 9401 struct va_format vaf; \ 9402 va_list args; \ 9403 \ 9404 va_start(args, fmt); \ 9405 \ 9406 vaf.fmt = fmt; \ 9407 vaf.va = &args; \ 9408 \ 9409 __netdev_printk(level, dev, &vaf); \ 9410 \ 9411 va_end(args); \ 9412 } \ 9413 EXPORT_SYMBOL(func); 9414 9415 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 9416 define_netdev_printk_level(netdev_alert, KERN_ALERT); 9417 define_netdev_printk_level(netdev_crit, KERN_CRIT); 9418 define_netdev_printk_level(netdev_err, KERN_ERR); 9419 define_netdev_printk_level(netdev_warn, KERN_WARNING); 9420 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 9421 define_netdev_printk_level(netdev_info, KERN_INFO); 9422 9423 static void __net_exit netdev_exit(struct net *net) 9424 { 9425 kfree(net->dev_name_head); 9426 kfree(net->dev_index_head); 9427 if (net != &init_net) 9428 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 9429 } 9430 9431 static struct pernet_operations __net_initdata netdev_net_ops = { 9432 .init = netdev_init, 9433 .exit = netdev_exit, 9434 }; 9435 9436 static void __net_exit default_device_exit(struct net *net) 9437 { 9438 struct net_device *dev, *aux; 9439 /* 9440 * Push all migratable network devices back to the 9441 * initial network namespace 9442 */ 9443 rtnl_lock(); 9444 for_each_netdev_safe(net, dev, aux) { 9445 int err; 9446 char fb_name[IFNAMSIZ]; 9447 9448 /* Ignore unmoveable devices (i.e. loopback) */ 9449 if (dev->features & NETIF_F_NETNS_LOCAL) 9450 continue; 9451 9452 /* Leave virtual devices for the generic cleanup */ 9453 if (dev->rtnl_link_ops) 9454 continue; 9455 9456 /* Push remaining network devices to init_net */ 9457 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 9458 err = dev_change_net_namespace(dev, &init_net, fb_name); 9459 if (err) { 9460 pr_emerg("%s: failed to move %s to init_net: %d\n", 9461 __func__, dev->name, err); 9462 BUG(); 9463 } 9464 } 9465 rtnl_unlock(); 9466 } 9467 9468 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 9469 { 9470 /* Return with the rtnl_lock held when there are no network 9471 * devices unregistering in any network namespace in net_list. 9472 */ 9473 struct net *net; 9474 bool unregistering; 9475 DEFINE_WAIT_FUNC(wait, woken_wake_function); 9476 9477 add_wait_queue(&netdev_unregistering_wq, &wait); 9478 for (;;) { 9479 unregistering = false; 9480 rtnl_lock(); 9481 list_for_each_entry(net, net_list, exit_list) { 9482 if (net->dev_unreg_count > 0) { 9483 unregistering = true; 9484 break; 9485 } 9486 } 9487 if (!unregistering) 9488 break; 9489 __rtnl_unlock(); 9490 9491 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 9492 } 9493 remove_wait_queue(&netdev_unregistering_wq, &wait); 9494 } 9495 9496 static void __net_exit default_device_exit_batch(struct list_head *net_list) 9497 { 9498 /* At exit all network devices most be removed from a network 9499 * namespace. Do this in the reverse order of registration. 9500 * Do this across as many network namespaces as possible to 9501 * improve batching efficiency. 9502 */ 9503 struct net_device *dev; 9504 struct net *net; 9505 LIST_HEAD(dev_kill_list); 9506 9507 /* To prevent network device cleanup code from dereferencing 9508 * loopback devices or network devices that have been freed 9509 * wait here for all pending unregistrations to complete, 9510 * before unregistring the loopback device and allowing the 9511 * network namespace be freed. 9512 * 9513 * The netdev todo list containing all network devices 9514 * unregistrations that happen in default_device_exit_batch 9515 * will run in the rtnl_unlock() at the end of 9516 * default_device_exit_batch. 9517 */ 9518 rtnl_lock_unregistering(net_list); 9519 list_for_each_entry(net, net_list, exit_list) { 9520 for_each_netdev_reverse(net, dev) { 9521 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 9522 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 9523 else 9524 unregister_netdevice_queue(dev, &dev_kill_list); 9525 } 9526 } 9527 unregister_netdevice_many(&dev_kill_list); 9528 rtnl_unlock(); 9529 } 9530 9531 static struct pernet_operations __net_initdata default_device_ops = { 9532 .exit = default_device_exit, 9533 .exit_batch = default_device_exit_batch, 9534 }; 9535 9536 /* 9537 * Initialize the DEV module. At boot time this walks the device list and 9538 * unhooks any devices that fail to initialise (normally hardware not 9539 * present) and leaves us with a valid list of present and active devices. 9540 * 9541 */ 9542 9543 /* 9544 * This is called single threaded during boot, so no need 9545 * to take the rtnl semaphore. 9546 */ 9547 static int __init net_dev_init(void) 9548 { 9549 int i, rc = -ENOMEM; 9550 9551 BUG_ON(!dev_boot_phase); 9552 9553 if (dev_proc_init()) 9554 goto out; 9555 9556 if (netdev_kobject_init()) 9557 goto out; 9558 9559 INIT_LIST_HEAD(&ptype_all); 9560 for (i = 0; i < PTYPE_HASH_SIZE; i++) 9561 INIT_LIST_HEAD(&ptype_base[i]); 9562 9563 INIT_LIST_HEAD(&offload_base); 9564 9565 if (register_pernet_subsys(&netdev_net_ops)) 9566 goto out; 9567 9568 /* 9569 * Initialise the packet receive queues. 9570 */ 9571 9572 for_each_possible_cpu(i) { 9573 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 9574 struct softnet_data *sd = &per_cpu(softnet_data, i); 9575 9576 INIT_WORK(flush, flush_backlog); 9577 9578 skb_queue_head_init(&sd->input_pkt_queue); 9579 skb_queue_head_init(&sd->process_queue); 9580 #ifdef CONFIG_XFRM_OFFLOAD 9581 skb_queue_head_init(&sd->xfrm_backlog); 9582 #endif 9583 INIT_LIST_HEAD(&sd->poll_list); 9584 sd->output_queue_tailp = &sd->output_queue; 9585 #ifdef CONFIG_RPS 9586 sd->csd.func = rps_trigger_softirq; 9587 sd->csd.info = sd; 9588 sd->cpu = i; 9589 #endif 9590 9591 init_gro_hash(&sd->backlog); 9592 sd->backlog.poll = process_backlog; 9593 sd->backlog.weight = weight_p; 9594 } 9595 9596 dev_boot_phase = 0; 9597 9598 /* The loopback device is special if any other network devices 9599 * is present in a network namespace the loopback device must 9600 * be present. Since we now dynamically allocate and free the 9601 * loopback device ensure this invariant is maintained by 9602 * keeping the loopback device as the first device on the 9603 * list of network devices. Ensuring the loopback devices 9604 * is the first device that appears and the last network device 9605 * that disappears. 9606 */ 9607 if (register_pernet_device(&loopback_net_ops)) 9608 goto out; 9609 9610 if (register_pernet_device(&default_device_ops)) 9611 goto out; 9612 9613 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 9614 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 9615 9616 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 9617 NULL, dev_cpu_dead); 9618 WARN_ON(rc < 0); 9619 rc = 0; 9620 out: 9621 return rc; 9622 } 9623 9624 subsys_initcall(net_dev_init); 9625