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 #include <linux/indirect_call_wrapper.h> 149 150 #include "net-sysfs.h" 151 152 #define MAX_GRO_SKBS 8 153 154 /* This should be increased if a protocol with a bigger head is added. */ 155 #define GRO_MAX_HEAD (MAX_HEADER + 128) 156 157 static DEFINE_SPINLOCK(ptype_lock); 158 static DEFINE_SPINLOCK(offload_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 static struct list_head offload_base __read_mostly; 162 163 static int netif_rx_internal(struct sk_buff *skb); 164 static int call_netdevice_notifiers_info(unsigned long val, 165 struct netdev_notifier_info *info); 166 static int call_netdevice_notifiers_extack(unsigned long val, 167 struct net_device *dev, 168 struct netlink_ext_ack *extack); 169 static struct napi_struct *napi_by_id(unsigned int napi_id); 170 171 /* 172 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 173 * semaphore. 174 * 175 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 176 * 177 * Writers must hold the rtnl semaphore while they loop through the 178 * dev_base_head list, and hold dev_base_lock for writing when they do the 179 * actual updates. This allows pure readers to access the list even 180 * while a writer is preparing to update it. 181 * 182 * To put it another way, dev_base_lock is held for writing only to 183 * protect against pure readers; the rtnl semaphore provides the 184 * protection against other writers. 185 * 186 * See, for example usages, register_netdevice() and 187 * unregister_netdevice(), which must be called with the rtnl 188 * semaphore held. 189 */ 190 DEFINE_RWLOCK(dev_base_lock); 191 EXPORT_SYMBOL(dev_base_lock); 192 193 static DEFINE_MUTEX(ifalias_mutex); 194 195 /* protects napi_hash addition/deletion and napi_gen_id */ 196 static DEFINE_SPINLOCK(napi_hash_lock); 197 198 static unsigned int napi_gen_id = NR_CPUS; 199 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 200 201 static seqcount_t devnet_rename_seq; 202 203 static inline void dev_base_seq_inc(struct net *net) 204 { 205 while (++net->dev_base_seq == 0) 206 ; 207 } 208 209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 210 { 211 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 212 213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 214 } 215 216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 217 { 218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 219 } 220 221 static inline void rps_lock(struct softnet_data *sd) 222 { 223 #ifdef CONFIG_RPS 224 spin_lock(&sd->input_pkt_queue.lock); 225 #endif 226 } 227 228 static inline void rps_unlock(struct softnet_data *sd) 229 { 230 #ifdef CONFIG_RPS 231 spin_unlock(&sd->input_pkt_queue.lock); 232 #endif 233 } 234 235 /* Device list insertion */ 236 static void list_netdevice(struct net_device *dev) 237 { 238 struct net *net = dev_net(dev); 239 240 ASSERT_RTNL(); 241 242 write_lock_bh(&dev_base_lock); 243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 245 hlist_add_head_rcu(&dev->index_hlist, 246 dev_index_hash(net, dev->ifindex)); 247 write_unlock_bh(&dev_base_lock); 248 249 dev_base_seq_inc(net); 250 } 251 252 /* Device list removal 253 * caller must respect a RCU grace period before freeing/reusing dev 254 */ 255 static void unlist_netdevice(struct net_device *dev) 256 { 257 ASSERT_RTNL(); 258 259 /* Unlink dev from the device chain */ 260 write_lock_bh(&dev_base_lock); 261 list_del_rcu(&dev->dev_list); 262 hlist_del_rcu(&dev->name_hlist); 263 hlist_del_rcu(&dev->index_hlist); 264 write_unlock_bh(&dev_base_lock); 265 266 dev_base_seq_inc(dev_net(dev)); 267 } 268 269 /* 270 * Our notifier list 271 */ 272 273 static RAW_NOTIFIER_HEAD(netdev_chain); 274 275 /* 276 * Device drivers call our routines to queue packets here. We empty the 277 * queue in the local softnet handler. 278 */ 279 280 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 281 EXPORT_PER_CPU_SYMBOL(softnet_data); 282 283 #ifdef CONFIG_LOCKDEP 284 /* 285 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 286 * according to dev->type 287 */ 288 static const unsigned short netdev_lock_type[] = { 289 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 290 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 291 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 292 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 293 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 294 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 295 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 296 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 297 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 298 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 299 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 300 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 301 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 302 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 303 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 304 305 static const char *const netdev_lock_name[] = { 306 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 307 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 308 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 309 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 310 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 311 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 312 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 313 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 314 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 315 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 316 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 317 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 318 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 319 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 320 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 321 322 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 323 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 324 325 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 326 { 327 int i; 328 329 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 330 if (netdev_lock_type[i] == dev_type) 331 return i; 332 /* the last key is used by default */ 333 return ARRAY_SIZE(netdev_lock_type) - 1; 334 } 335 336 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 337 unsigned short dev_type) 338 { 339 int i; 340 341 i = netdev_lock_pos(dev_type); 342 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 343 netdev_lock_name[i]); 344 } 345 346 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 347 { 348 int i; 349 350 i = netdev_lock_pos(dev->type); 351 lockdep_set_class_and_name(&dev->addr_list_lock, 352 &netdev_addr_lock_key[i], 353 netdev_lock_name[i]); 354 } 355 #else 356 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 357 unsigned short dev_type) 358 { 359 } 360 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 361 { 362 } 363 #endif 364 365 /******************************************************************************* 366 * 367 * Protocol management and registration routines 368 * 369 *******************************************************************************/ 370 371 372 /* 373 * Add a protocol ID to the list. Now that the input handler is 374 * smarter we can dispense with all the messy stuff that used to be 375 * here. 376 * 377 * BEWARE!!! Protocol handlers, mangling input packets, 378 * MUST BE last in hash buckets and checking protocol handlers 379 * MUST start from promiscuous ptype_all chain in net_bh. 380 * It is true now, do not change it. 381 * Explanation follows: if protocol handler, mangling packet, will 382 * be the first on list, it is not able to sense, that packet 383 * is cloned and should be copied-on-write, so that it will 384 * change it and subsequent readers will get broken packet. 385 * --ANK (980803) 386 */ 387 388 static inline struct list_head *ptype_head(const struct packet_type *pt) 389 { 390 if (pt->type == htons(ETH_P_ALL)) 391 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 392 else 393 return pt->dev ? &pt->dev->ptype_specific : 394 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 395 } 396 397 /** 398 * dev_add_pack - add packet handler 399 * @pt: packet type declaration 400 * 401 * Add a protocol handler to the networking stack. The passed &packet_type 402 * is linked into kernel lists and may not be freed until it has been 403 * removed from the kernel lists. 404 * 405 * This call does not sleep therefore it can not 406 * guarantee all CPU's that are in middle of receiving packets 407 * will see the new packet type (until the next received packet). 408 */ 409 410 void dev_add_pack(struct packet_type *pt) 411 { 412 struct list_head *head = ptype_head(pt); 413 414 spin_lock(&ptype_lock); 415 list_add_rcu(&pt->list, head); 416 spin_unlock(&ptype_lock); 417 } 418 EXPORT_SYMBOL(dev_add_pack); 419 420 /** 421 * __dev_remove_pack - remove packet handler 422 * @pt: packet type declaration 423 * 424 * Remove a protocol handler that was previously added to the kernel 425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 426 * from the kernel lists and can be freed or reused once this function 427 * returns. 428 * 429 * The packet type might still be in use by receivers 430 * and must not be freed until after all the CPU's have gone 431 * through a quiescent state. 432 */ 433 void __dev_remove_pack(struct packet_type *pt) 434 { 435 struct list_head *head = ptype_head(pt); 436 struct packet_type *pt1; 437 438 spin_lock(&ptype_lock); 439 440 list_for_each_entry(pt1, head, list) { 441 if (pt == pt1) { 442 list_del_rcu(&pt->list); 443 goto out; 444 } 445 } 446 447 pr_warn("dev_remove_pack: %p not found\n", pt); 448 out: 449 spin_unlock(&ptype_lock); 450 } 451 EXPORT_SYMBOL(__dev_remove_pack); 452 453 /** 454 * dev_remove_pack - remove packet handler 455 * @pt: packet type declaration 456 * 457 * Remove a protocol handler that was previously added to the kernel 458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 459 * from the kernel lists and can be freed or reused once this function 460 * returns. 461 * 462 * This call sleeps to guarantee that no CPU is looking at the packet 463 * type after return. 464 */ 465 void dev_remove_pack(struct packet_type *pt) 466 { 467 __dev_remove_pack(pt); 468 469 synchronize_net(); 470 } 471 EXPORT_SYMBOL(dev_remove_pack); 472 473 474 /** 475 * dev_add_offload - register offload handlers 476 * @po: protocol offload declaration 477 * 478 * Add protocol offload handlers to the networking stack. The passed 479 * &proto_offload is linked into kernel lists and may not be freed until 480 * it has been removed from the kernel lists. 481 * 482 * This call does not sleep therefore it can not 483 * guarantee all CPU's that are in middle of receiving packets 484 * will see the new offload handlers (until the next received packet). 485 */ 486 void dev_add_offload(struct packet_offload *po) 487 { 488 struct packet_offload *elem; 489 490 spin_lock(&offload_lock); 491 list_for_each_entry(elem, &offload_base, list) { 492 if (po->priority < elem->priority) 493 break; 494 } 495 list_add_rcu(&po->list, elem->list.prev); 496 spin_unlock(&offload_lock); 497 } 498 EXPORT_SYMBOL(dev_add_offload); 499 500 /** 501 * __dev_remove_offload - remove offload handler 502 * @po: packet offload declaration 503 * 504 * Remove a protocol offload handler that was previously added to the 505 * kernel offload handlers by dev_add_offload(). The passed &offload_type 506 * is removed from the kernel lists and can be freed or reused once this 507 * function returns. 508 * 509 * The packet type might still be in use by receivers 510 * and must not be freed until after all the CPU's have gone 511 * through a quiescent state. 512 */ 513 static void __dev_remove_offload(struct packet_offload *po) 514 { 515 struct list_head *head = &offload_base; 516 struct packet_offload *po1; 517 518 spin_lock(&offload_lock); 519 520 list_for_each_entry(po1, head, list) { 521 if (po == po1) { 522 list_del_rcu(&po->list); 523 goto out; 524 } 525 } 526 527 pr_warn("dev_remove_offload: %p not found\n", po); 528 out: 529 spin_unlock(&offload_lock); 530 } 531 532 /** 533 * dev_remove_offload - remove packet offload handler 534 * @po: packet offload declaration 535 * 536 * Remove a packet offload handler that was previously added to the kernel 537 * offload handlers by dev_add_offload(). The passed &offload_type is 538 * removed from the kernel lists and can be freed or reused once this 539 * function returns. 540 * 541 * This call sleeps to guarantee that no CPU is looking at the packet 542 * type after return. 543 */ 544 void dev_remove_offload(struct packet_offload *po) 545 { 546 __dev_remove_offload(po); 547 548 synchronize_net(); 549 } 550 EXPORT_SYMBOL(dev_remove_offload); 551 552 /****************************************************************************** 553 * 554 * Device Boot-time Settings Routines 555 * 556 ******************************************************************************/ 557 558 /* Boot time configuration table */ 559 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 560 561 /** 562 * netdev_boot_setup_add - add new setup entry 563 * @name: name of the device 564 * @map: configured settings for the device 565 * 566 * Adds new setup entry to the dev_boot_setup list. The function 567 * returns 0 on error and 1 on success. This is a generic routine to 568 * all netdevices. 569 */ 570 static int netdev_boot_setup_add(char *name, struct ifmap *map) 571 { 572 struct netdev_boot_setup *s; 573 int i; 574 575 s = dev_boot_setup; 576 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 577 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 578 memset(s[i].name, 0, sizeof(s[i].name)); 579 strlcpy(s[i].name, name, IFNAMSIZ); 580 memcpy(&s[i].map, map, sizeof(s[i].map)); 581 break; 582 } 583 } 584 585 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 586 } 587 588 /** 589 * netdev_boot_setup_check - check boot time settings 590 * @dev: the netdevice 591 * 592 * Check boot time settings for the device. 593 * The found settings are set for the device to be used 594 * later in the device probing. 595 * Returns 0 if no settings found, 1 if they are. 596 */ 597 int netdev_boot_setup_check(struct net_device *dev) 598 { 599 struct netdev_boot_setup *s = dev_boot_setup; 600 int i; 601 602 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 603 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 604 !strcmp(dev->name, s[i].name)) { 605 dev->irq = s[i].map.irq; 606 dev->base_addr = s[i].map.base_addr; 607 dev->mem_start = s[i].map.mem_start; 608 dev->mem_end = s[i].map.mem_end; 609 return 1; 610 } 611 } 612 return 0; 613 } 614 EXPORT_SYMBOL(netdev_boot_setup_check); 615 616 617 /** 618 * netdev_boot_base - get address from boot time settings 619 * @prefix: prefix for network device 620 * @unit: id for network device 621 * 622 * Check boot time settings for the base address of device. 623 * The found settings are set for the device to be used 624 * later in the device probing. 625 * Returns 0 if no settings found. 626 */ 627 unsigned long netdev_boot_base(const char *prefix, int unit) 628 { 629 const struct netdev_boot_setup *s = dev_boot_setup; 630 char name[IFNAMSIZ]; 631 int i; 632 633 sprintf(name, "%s%d", prefix, unit); 634 635 /* 636 * If device already registered then return base of 1 637 * to indicate not to probe for this interface 638 */ 639 if (__dev_get_by_name(&init_net, name)) 640 return 1; 641 642 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 643 if (!strcmp(name, s[i].name)) 644 return s[i].map.base_addr; 645 return 0; 646 } 647 648 /* 649 * Saves at boot time configured settings for any netdevice. 650 */ 651 int __init netdev_boot_setup(char *str) 652 { 653 int ints[5]; 654 struct ifmap map; 655 656 str = get_options(str, ARRAY_SIZE(ints), ints); 657 if (!str || !*str) 658 return 0; 659 660 /* Save settings */ 661 memset(&map, 0, sizeof(map)); 662 if (ints[0] > 0) 663 map.irq = ints[1]; 664 if (ints[0] > 1) 665 map.base_addr = ints[2]; 666 if (ints[0] > 2) 667 map.mem_start = ints[3]; 668 if (ints[0] > 3) 669 map.mem_end = ints[4]; 670 671 /* Add new entry to the list */ 672 return netdev_boot_setup_add(str, &map); 673 } 674 675 __setup("netdev=", netdev_boot_setup); 676 677 /******************************************************************************* 678 * 679 * Device Interface Subroutines 680 * 681 *******************************************************************************/ 682 683 /** 684 * dev_get_iflink - get 'iflink' value of a interface 685 * @dev: targeted interface 686 * 687 * Indicates the ifindex the interface is linked to. 688 * Physical interfaces have the same 'ifindex' and 'iflink' values. 689 */ 690 691 int dev_get_iflink(const struct net_device *dev) 692 { 693 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 694 return dev->netdev_ops->ndo_get_iflink(dev); 695 696 return dev->ifindex; 697 } 698 EXPORT_SYMBOL(dev_get_iflink); 699 700 /** 701 * dev_fill_metadata_dst - Retrieve tunnel egress information. 702 * @dev: targeted interface 703 * @skb: The packet. 704 * 705 * For better visibility of tunnel traffic OVS needs to retrieve 706 * egress tunnel information for a packet. Following API allows 707 * user to get this info. 708 */ 709 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 710 { 711 struct ip_tunnel_info *info; 712 713 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 714 return -EINVAL; 715 716 info = skb_tunnel_info_unclone(skb); 717 if (!info) 718 return -ENOMEM; 719 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 720 return -EINVAL; 721 722 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 723 } 724 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 725 726 /** 727 * __dev_get_by_name - find a device by its name 728 * @net: the applicable net namespace 729 * @name: name to find 730 * 731 * Find an interface by name. Must be called under RTNL semaphore 732 * or @dev_base_lock. If the name is found a pointer to the device 733 * is returned. If the name is not found then %NULL is returned. The 734 * reference counters are not incremented so the caller must be 735 * careful with locks. 736 */ 737 738 struct net_device *__dev_get_by_name(struct net *net, const char *name) 739 { 740 struct net_device *dev; 741 struct hlist_head *head = dev_name_hash(net, name); 742 743 hlist_for_each_entry(dev, head, name_hlist) 744 if (!strncmp(dev->name, name, IFNAMSIZ)) 745 return dev; 746 747 return NULL; 748 } 749 EXPORT_SYMBOL(__dev_get_by_name); 750 751 /** 752 * dev_get_by_name_rcu - find a device by its name 753 * @net: the applicable net namespace 754 * @name: name to find 755 * 756 * Find an interface by name. 757 * If the name is found a pointer to the device is returned. 758 * If the name is not found then %NULL is returned. 759 * The reference counters are not incremented so the caller must be 760 * careful with locks. The caller must hold RCU lock. 761 */ 762 763 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 764 { 765 struct net_device *dev; 766 struct hlist_head *head = dev_name_hash(net, name); 767 768 hlist_for_each_entry_rcu(dev, head, name_hlist) 769 if (!strncmp(dev->name, name, IFNAMSIZ)) 770 return dev; 771 772 return NULL; 773 } 774 EXPORT_SYMBOL(dev_get_by_name_rcu); 775 776 /** 777 * dev_get_by_name - find a device by its name 778 * @net: the applicable net namespace 779 * @name: name to find 780 * 781 * Find an interface by name. This can be called from any 782 * context and does its own locking. The returned handle has 783 * the usage count incremented and the caller must use dev_put() to 784 * release it when it is no longer needed. %NULL is returned if no 785 * matching device is found. 786 */ 787 788 struct net_device *dev_get_by_name(struct net *net, const char *name) 789 { 790 struct net_device *dev; 791 792 rcu_read_lock(); 793 dev = dev_get_by_name_rcu(net, name); 794 if (dev) 795 dev_hold(dev); 796 rcu_read_unlock(); 797 return dev; 798 } 799 EXPORT_SYMBOL(dev_get_by_name); 800 801 /** 802 * __dev_get_by_index - find a device by its ifindex 803 * @net: the applicable net namespace 804 * @ifindex: index of device 805 * 806 * Search for an interface by index. Returns %NULL if the device 807 * is not found or a pointer to the device. The device has not 808 * had its reference counter increased so the caller must be careful 809 * about locking. The caller must hold either the RTNL semaphore 810 * or @dev_base_lock. 811 */ 812 813 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 814 { 815 struct net_device *dev; 816 struct hlist_head *head = dev_index_hash(net, ifindex); 817 818 hlist_for_each_entry(dev, head, index_hlist) 819 if (dev->ifindex == ifindex) 820 return dev; 821 822 return NULL; 823 } 824 EXPORT_SYMBOL(__dev_get_by_index); 825 826 /** 827 * dev_get_by_index_rcu - find a device by its ifindex 828 * @net: the applicable net namespace 829 * @ifindex: index of device 830 * 831 * Search for an interface by index. Returns %NULL if the device 832 * is not found or a pointer to the device. The device has not 833 * had its reference counter increased so the caller must be careful 834 * about locking. The caller must hold RCU lock. 835 */ 836 837 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 838 { 839 struct net_device *dev; 840 struct hlist_head *head = dev_index_hash(net, ifindex); 841 842 hlist_for_each_entry_rcu(dev, head, index_hlist) 843 if (dev->ifindex == ifindex) 844 return dev; 845 846 return NULL; 847 } 848 EXPORT_SYMBOL(dev_get_by_index_rcu); 849 850 851 /** 852 * dev_get_by_index - find a device by its ifindex 853 * @net: the applicable net namespace 854 * @ifindex: index of device 855 * 856 * Search for an interface by index. Returns NULL if the device 857 * is not found or a pointer to the device. The device returned has 858 * had a reference added and the pointer is safe until the user calls 859 * dev_put to indicate they have finished with it. 860 */ 861 862 struct net_device *dev_get_by_index(struct net *net, int ifindex) 863 { 864 struct net_device *dev; 865 866 rcu_read_lock(); 867 dev = dev_get_by_index_rcu(net, ifindex); 868 if (dev) 869 dev_hold(dev); 870 rcu_read_unlock(); 871 return dev; 872 } 873 EXPORT_SYMBOL(dev_get_by_index); 874 875 /** 876 * dev_get_by_napi_id - find a device by napi_id 877 * @napi_id: ID of the NAPI struct 878 * 879 * Search for an interface by NAPI ID. Returns %NULL if the device 880 * is not found or a pointer to the device. The device has not had 881 * its reference counter increased so the caller must be careful 882 * about locking. The caller must hold RCU lock. 883 */ 884 885 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 886 { 887 struct napi_struct *napi; 888 889 WARN_ON_ONCE(!rcu_read_lock_held()); 890 891 if (napi_id < MIN_NAPI_ID) 892 return NULL; 893 894 napi = napi_by_id(napi_id); 895 896 return napi ? napi->dev : NULL; 897 } 898 EXPORT_SYMBOL(dev_get_by_napi_id); 899 900 /** 901 * netdev_get_name - get a netdevice name, knowing its ifindex. 902 * @net: network namespace 903 * @name: a pointer to the buffer where the name will be stored. 904 * @ifindex: the ifindex of the interface to get the name from. 905 * 906 * The use of raw_seqcount_begin() and cond_resched() before 907 * retrying is required as we want to give the writers a chance 908 * to complete when CONFIG_PREEMPT is not set. 909 */ 910 int netdev_get_name(struct net *net, char *name, int ifindex) 911 { 912 struct net_device *dev; 913 unsigned int seq; 914 915 retry: 916 seq = raw_seqcount_begin(&devnet_rename_seq); 917 rcu_read_lock(); 918 dev = dev_get_by_index_rcu(net, ifindex); 919 if (!dev) { 920 rcu_read_unlock(); 921 return -ENODEV; 922 } 923 924 strcpy(name, dev->name); 925 rcu_read_unlock(); 926 if (read_seqcount_retry(&devnet_rename_seq, seq)) { 927 cond_resched(); 928 goto retry; 929 } 930 931 return 0; 932 } 933 934 /** 935 * dev_getbyhwaddr_rcu - find a device by its hardware address 936 * @net: the applicable net namespace 937 * @type: media type of device 938 * @ha: hardware address 939 * 940 * Search for an interface by MAC address. Returns NULL if the device 941 * is not found or a pointer to the device. 942 * The caller must hold RCU or RTNL. 943 * The returned device has not had its ref count increased 944 * and the caller must therefore be careful about locking 945 * 946 */ 947 948 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 949 const char *ha) 950 { 951 struct net_device *dev; 952 953 for_each_netdev_rcu(net, dev) 954 if (dev->type == type && 955 !memcmp(dev->dev_addr, ha, dev->addr_len)) 956 return dev; 957 958 return NULL; 959 } 960 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 961 962 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) 963 { 964 struct net_device *dev; 965 966 ASSERT_RTNL(); 967 for_each_netdev(net, dev) 968 if (dev->type == type) 969 return dev; 970 971 return NULL; 972 } 973 EXPORT_SYMBOL(__dev_getfirstbyhwtype); 974 975 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 976 { 977 struct net_device *dev, *ret = NULL; 978 979 rcu_read_lock(); 980 for_each_netdev_rcu(net, dev) 981 if (dev->type == type) { 982 dev_hold(dev); 983 ret = dev; 984 break; 985 } 986 rcu_read_unlock(); 987 return ret; 988 } 989 EXPORT_SYMBOL(dev_getfirstbyhwtype); 990 991 /** 992 * __dev_get_by_flags - find any device with given flags 993 * @net: the applicable net namespace 994 * @if_flags: IFF_* values 995 * @mask: bitmask of bits in if_flags to check 996 * 997 * Search for any interface with the given flags. Returns NULL if a device 998 * is not found or a pointer to the device. Must be called inside 999 * rtnl_lock(), and result refcount is unchanged. 1000 */ 1001 1002 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1003 unsigned short mask) 1004 { 1005 struct net_device *dev, *ret; 1006 1007 ASSERT_RTNL(); 1008 1009 ret = NULL; 1010 for_each_netdev(net, dev) { 1011 if (((dev->flags ^ if_flags) & mask) == 0) { 1012 ret = dev; 1013 break; 1014 } 1015 } 1016 return ret; 1017 } 1018 EXPORT_SYMBOL(__dev_get_by_flags); 1019 1020 /** 1021 * dev_valid_name - check if name is okay for network device 1022 * @name: name string 1023 * 1024 * Network device names need to be valid file names to 1025 * to allow sysfs to work. We also disallow any kind of 1026 * whitespace. 1027 */ 1028 bool dev_valid_name(const char *name) 1029 { 1030 if (*name == '\0') 1031 return false; 1032 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1033 return false; 1034 if (!strcmp(name, ".") || !strcmp(name, "..")) 1035 return false; 1036 1037 while (*name) { 1038 if (*name == '/' || *name == ':' || isspace(*name)) 1039 return false; 1040 name++; 1041 } 1042 return true; 1043 } 1044 EXPORT_SYMBOL(dev_valid_name); 1045 1046 /** 1047 * __dev_alloc_name - allocate a name for a device 1048 * @net: network namespace to allocate the device name in 1049 * @name: name format string 1050 * @buf: scratch buffer and result name string 1051 * 1052 * Passed a format string - eg "lt%d" it will try and find a suitable 1053 * id. It scans list of devices to build up a free map, then chooses 1054 * the first empty slot. The caller must hold the dev_base or rtnl lock 1055 * while allocating the name and adding the device in order to avoid 1056 * duplicates. 1057 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1058 * Returns the number of the unit assigned or a negative errno code. 1059 */ 1060 1061 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1062 { 1063 int i = 0; 1064 const char *p; 1065 const int max_netdevices = 8*PAGE_SIZE; 1066 unsigned long *inuse; 1067 struct net_device *d; 1068 1069 if (!dev_valid_name(name)) 1070 return -EINVAL; 1071 1072 p = strchr(name, '%'); 1073 if (p) { 1074 /* 1075 * Verify the string as this thing may have come from 1076 * the user. There must be either one "%d" and no other "%" 1077 * characters. 1078 */ 1079 if (p[1] != 'd' || strchr(p + 2, '%')) 1080 return -EINVAL; 1081 1082 /* Use one page as a bit array of possible slots */ 1083 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1084 if (!inuse) 1085 return -ENOMEM; 1086 1087 for_each_netdev(net, d) { 1088 if (!sscanf(d->name, name, &i)) 1089 continue; 1090 if (i < 0 || i >= max_netdevices) 1091 continue; 1092 1093 /* avoid cases where sscanf is not exact inverse of printf */ 1094 snprintf(buf, IFNAMSIZ, name, i); 1095 if (!strncmp(buf, d->name, IFNAMSIZ)) 1096 set_bit(i, inuse); 1097 } 1098 1099 i = find_first_zero_bit(inuse, max_netdevices); 1100 free_page((unsigned long) inuse); 1101 } 1102 1103 snprintf(buf, IFNAMSIZ, name, i); 1104 if (!__dev_get_by_name(net, buf)) 1105 return i; 1106 1107 /* It is possible to run out of possible slots 1108 * when the name is long and there isn't enough space left 1109 * for the digits, or if all bits are used. 1110 */ 1111 return -ENFILE; 1112 } 1113 1114 static int dev_alloc_name_ns(struct net *net, 1115 struct net_device *dev, 1116 const char *name) 1117 { 1118 char buf[IFNAMSIZ]; 1119 int ret; 1120 1121 BUG_ON(!net); 1122 ret = __dev_alloc_name(net, name, buf); 1123 if (ret >= 0) 1124 strlcpy(dev->name, buf, IFNAMSIZ); 1125 return ret; 1126 } 1127 1128 /** 1129 * dev_alloc_name - allocate a name for a device 1130 * @dev: device 1131 * @name: name format string 1132 * 1133 * Passed a format string - eg "lt%d" it will try and find a suitable 1134 * id. It scans list of devices to build up a free map, then chooses 1135 * the first empty slot. The caller must hold the dev_base or rtnl lock 1136 * while allocating the name and adding the device in order to avoid 1137 * duplicates. 1138 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1139 * Returns the number of the unit assigned or a negative errno code. 1140 */ 1141 1142 int dev_alloc_name(struct net_device *dev, const char *name) 1143 { 1144 return dev_alloc_name_ns(dev_net(dev), dev, name); 1145 } 1146 EXPORT_SYMBOL(dev_alloc_name); 1147 1148 int dev_get_valid_name(struct net *net, struct net_device *dev, 1149 const char *name) 1150 { 1151 BUG_ON(!net); 1152 1153 if (!dev_valid_name(name)) 1154 return -EINVAL; 1155 1156 if (strchr(name, '%')) 1157 return dev_alloc_name_ns(net, dev, name); 1158 else if (__dev_get_by_name(net, name)) 1159 return -EEXIST; 1160 else if (dev->name != name) 1161 strlcpy(dev->name, name, IFNAMSIZ); 1162 1163 return 0; 1164 } 1165 EXPORT_SYMBOL(dev_get_valid_name); 1166 1167 /** 1168 * dev_change_name - change name of a device 1169 * @dev: device 1170 * @newname: name (or format string) must be at least IFNAMSIZ 1171 * 1172 * Change name of a device, can pass format strings "eth%d". 1173 * for wildcarding. 1174 */ 1175 int dev_change_name(struct net_device *dev, const char *newname) 1176 { 1177 unsigned char old_assign_type; 1178 char oldname[IFNAMSIZ]; 1179 int err = 0; 1180 int ret; 1181 struct net *net; 1182 1183 ASSERT_RTNL(); 1184 BUG_ON(!dev_net(dev)); 1185 1186 net = dev_net(dev); 1187 if (dev->flags & IFF_UP) 1188 return -EBUSY; 1189 1190 write_seqcount_begin(&devnet_rename_seq); 1191 1192 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1193 write_seqcount_end(&devnet_rename_seq); 1194 return 0; 1195 } 1196 1197 memcpy(oldname, dev->name, IFNAMSIZ); 1198 1199 err = dev_get_valid_name(net, dev, newname); 1200 if (err < 0) { 1201 write_seqcount_end(&devnet_rename_seq); 1202 return err; 1203 } 1204 1205 if (oldname[0] && !strchr(oldname, '%')) 1206 netdev_info(dev, "renamed from %s\n", oldname); 1207 1208 old_assign_type = dev->name_assign_type; 1209 dev->name_assign_type = NET_NAME_RENAMED; 1210 1211 rollback: 1212 ret = device_rename(&dev->dev, dev->name); 1213 if (ret) { 1214 memcpy(dev->name, oldname, IFNAMSIZ); 1215 dev->name_assign_type = old_assign_type; 1216 write_seqcount_end(&devnet_rename_seq); 1217 return ret; 1218 } 1219 1220 write_seqcount_end(&devnet_rename_seq); 1221 1222 netdev_adjacent_rename_links(dev, oldname); 1223 1224 write_lock_bh(&dev_base_lock); 1225 hlist_del_rcu(&dev->name_hlist); 1226 write_unlock_bh(&dev_base_lock); 1227 1228 synchronize_rcu(); 1229 1230 write_lock_bh(&dev_base_lock); 1231 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); 1232 write_unlock_bh(&dev_base_lock); 1233 1234 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1235 ret = notifier_to_errno(ret); 1236 1237 if (ret) { 1238 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1239 if (err >= 0) { 1240 err = ret; 1241 write_seqcount_begin(&devnet_rename_seq); 1242 memcpy(dev->name, oldname, IFNAMSIZ); 1243 memcpy(oldname, newname, IFNAMSIZ); 1244 dev->name_assign_type = old_assign_type; 1245 old_assign_type = NET_NAME_RENAMED; 1246 goto rollback; 1247 } else { 1248 pr_err("%s: name change rollback failed: %d\n", 1249 dev->name, ret); 1250 } 1251 } 1252 1253 return err; 1254 } 1255 1256 /** 1257 * dev_set_alias - change ifalias of a device 1258 * @dev: device 1259 * @alias: name up to IFALIASZ 1260 * @len: limit of bytes to copy from info 1261 * 1262 * Set ifalias for a device, 1263 */ 1264 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1265 { 1266 struct dev_ifalias *new_alias = NULL; 1267 1268 if (len >= IFALIASZ) 1269 return -EINVAL; 1270 1271 if (len) { 1272 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1273 if (!new_alias) 1274 return -ENOMEM; 1275 1276 memcpy(new_alias->ifalias, alias, len); 1277 new_alias->ifalias[len] = 0; 1278 } 1279 1280 mutex_lock(&ifalias_mutex); 1281 rcu_swap_protected(dev->ifalias, new_alias, 1282 mutex_is_locked(&ifalias_mutex)); 1283 mutex_unlock(&ifalias_mutex); 1284 1285 if (new_alias) 1286 kfree_rcu(new_alias, rcuhead); 1287 1288 return len; 1289 } 1290 EXPORT_SYMBOL(dev_set_alias); 1291 1292 /** 1293 * dev_get_alias - get ifalias of a device 1294 * @dev: device 1295 * @name: buffer to store name of ifalias 1296 * @len: size of buffer 1297 * 1298 * get ifalias for a device. Caller must make sure dev cannot go 1299 * away, e.g. rcu read lock or own a reference count to device. 1300 */ 1301 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1302 { 1303 const struct dev_ifalias *alias; 1304 int ret = 0; 1305 1306 rcu_read_lock(); 1307 alias = rcu_dereference(dev->ifalias); 1308 if (alias) 1309 ret = snprintf(name, len, "%s", alias->ifalias); 1310 rcu_read_unlock(); 1311 1312 return ret; 1313 } 1314 1315 /** 1316 * netdev_features_change - device changes features 1317 * @dev: device to cause notification 1318 * 1319 * Called to indicate a device has changed features. 1320 */ 1321 void netdev_features_change(struct net_device *dev) 1322 { 1323 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1324 } 1325 EXPORT_SYMBOL(netdev_features_change); 1326 1327 /** 1328 * netdev_state_change - device changes state 1329 * @dev: device to cause notification 1330 * 1331 * Called to indicate a device has changed state. This function calls 1332 * the notifier chains for netdev_chain and sends a NEWLINK message 1333 * to the routing socket. 1334 */ 1335 void netdev_state_change(struct net_device *dev) 1336 { 1337 if (dev->flags & IFF_UP) { 1338 struct netdev_notifier_change_info change_info = { 1339 .info.dev = dev, 1340 }; 1341 1342 call_netdevice_notifiers_info(NETDEV_CHANGE, 1343 &change_info.info); 1344 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1345 } 1346 } 1347 EXPORT_SYMBOL(netdev_state_change); 1348 1349 /** 1350 * netdev_notify_peers - notify network peers about existence of @dev 1351 * @dev: network device 1352 * 1353 * Generate traffic such that interested network peers are aware of 1354 * @dev, such as by generating a gratuitous ARP. This may be used when 1355 * a device wants to inform the rest of the network about some sort of 1356 * reconfiguration such as a failover event or virtual machine 1357 * migration. 1358 */ 1359 void netdev_notify_peers(struct net_device *dev) 1360 { 1361 rtnl_lock(); 1362 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1363 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1364 rtnl_unlock(); 1365 } 1366 EXPORT_SYMBOL(netdev_notify_peers); 1367 1368 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1369 { 1370 const struct net_device_ops *ops = dev->netdev_ops; 1371 int ret; 1372 1373 ASSERT_RTNL(); 1374 1375 if (!netif_device_present(dev)) 1376 return -ENODEV; 1377 1378 /* Block netpoll from trying to do any rx path servicing. 1379 * If we don't do this there is a chance ndo_poll_controller 1380 * or ndo_poll may be running while we open the device 1381 */ 1382 netpoll_poll_disable(dev); 1383 1384 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1385 ret = notifier_to_errno(ret); 1386 if (ret) 1387 return ret; 1388 1389 set_bit(__LINK_STATE_START, &dev->state); 1390 1391 if (ops->ndo_validate_addr) 1392 ret = ops->ndo_validate_addr(dev); 1393 1394 if (!ret && ops->ndo_open) 1395 ret = ops->ndo_open(dev); 1396 1397 netpoll_poll_enable(dev); 1398 1399 if (ret) 1400 clear_bit(__LINK_STATE_START, &dev->state); 1401 else { 1402 dev->flags |= IFF_UP; 1403 dev_set_rx_mode(dev); 1404 dev_activate(dev); 1405 add_device_randomness(dev->dev_addr, dev->addr_len); 1406 } 1407 1408 return ret; 1409 } 1410 1411 /** 1412 * dev_open - prepare an interface for use. 1413 * @dev: device to open 1414 * @extack: netlink extended ack 1415 * 1416 * Takes a device from down to up state. The device's private open 1417 * function is invoked and then the multicast lists are loaded. Finally 1418 * the device is moved into the up state and a %NETDEV_UP message is 1419 * sent to the netdev notifier chain. 1420 * 1421 * Calling this function on an active interface is a nop. On a failure 1422 * a negative errno code is returned. 1423 */ 1424 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1425 { 1426 int ret; 1427 1428 if (dev->flags & IFF_UP) 1429 return 0; 1430 1431 ret = __dev_open(dev, extack); 1432 if (ret < 0) 1433 return ret; 1434 1435 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1436 call_netdevice_notifiers(NETDEV_UP, dev); 1437 1438 return ret; 1439 } 1440 EXPORT_SYMBOL(dev_open); 1441 1442 static void __dev_close_many(struct list_head *head) 1443 { 1444 struct net_device *dev; 1445 1446 ASSERT_RTNL(); 1447 might_sleep(); 1448 1449 list_for_each_entry(dev, head, close_list) { 1450 /* Temporarily disable netpoll until the interface is down */ 1451 netpoll_poll_disable(dev); 1452 1453 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1454 1455 clear_bit(__LINK_STATE_START, &dev->state); 1456 1457 /* Synchronize to scheduled poll. We cannot touch poll list, it 1458 * can be even on different cpu. So just clear netif_running(). 1459 * 1460 * dev->stop() will invoke napi_disable() on all of it's 1461 * napi_struct instances on this device. 1462 */ 1463 smp_mb__after_atomic(); /* Commit netif_running(). */ 1464 } 1465 1466 dev_deactivate_many(head); 1467 1468 list_for_each_entry(dev, head, close_list) { 1469 const struct net_device_ops *ops = dev->netdev_ops; 1470 1471 /* 1472 * Call the device specific close. This cannot fail. 1473 * Only if device is UP 1474 * 1475 * We allow it to be called even after a DETACH hot-plug 1476 * event. 1477 */ 1478 if (ops->ndo_stop) 1479 ops->ndo_stop(dev); 1480 1481 dev->flags &= ~IFF_UP; 1482 netpoll_poll_enable(dev); 1483 } 1484 } 1485 1486 static void __dev_close(struct net_device *dev) 1487 { 1488 LIST_HEAD(single); 1489 1490 list_add(&dev->close_list, &single); 1491 __dev_close_many(&single); 1492 list_del(&single); 1493 } 1494 1495 void dev_close_many(struct list_head *head, bool unlink) 1496 { 1497 struct net_device *dev, *tmp; 1498 1499 /* Remove the devices that don't need to be closed */ 1500 list_for_each_entry_safe(dev, tmp, head, close_list) 1501 if (!(dev->flags & IFF_UP)) 1502 list_del_init(&dev->close_list); 1503 1504 __dev_close_many(head); 1505 1506 list_for_each_entry_safe(dev, tmp, head, close_list) { 1507 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1508 call_netdevice_notifiers(NETDEV_DOWN, dev); 1509 if (unlink) 1510 list_del_init(&dev->close_list); 1511 } 1512 } 1513 EXPORT_SYMBOL(dev_close_many); 1514 1515 /** 1516 * dev_close - shutdown an interface. 1517 * @dev: device to shutdown 1518 * 1519 * This function moves an active device into down state. A 1520 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1521 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1522 * chain. 1523 */ 1524 void dev_close(struct net_device *dev) 1525 { 1526 if (dev->flags & IFF_UP) { 1527 LIST_HEAD(single); 1528 1529 list_add(&dev->close_list, &single); 1530 dev_close_many(&single, true); 1531 list_del(&single); 1532 } 1533 } 1534 EXPORT_SYMBOL(dev_close); 1535 1536 1537 /** 1538 * dev_disable_lro - disable Large Receive Offload on a device 1539 * @dev: device 1540 * 1541 * Disable Large Receive Offload (LRO) on a net device. Must be 1542 * called under RTNL. This is needed if received packets may be 1543 * forwarded to another interface. 1544 */ 1545 void dev_disable_lro(struct net_device *dev) 1546 { 1547 struct net_device *lower_dev; 1548 struct list_head *iter; 1549 1550 dev->wanted_features &= ~NETIF_F_LRO; 1551 netdev_update_features(dev); 1552 1553 if (unlikely(dev->features & NETIF_F_LRO)) 1554 netdev_WARN(dev, "failed to disable LRO!\n"); 1555 1556 netdev_for_each_lower_dev(dev, lower_dev, iter) 1557 dev_disable_lro(lower_dev); 1558 } 1559 EXPORT_SYMBOL(dev_disable_lro); 1560 1561 /** 1562 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1563 * @dev: device 1564 * 1565 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1566 * called under RTNL. This is needed if Generic XDP is installed on 1567 * the device. 1568 */ 1569 static void dev_disable_gro_hw(struct net_device *dev) 1570 { 1571 dev->wanted_features &= ~NETIF_F_GRO_HW; 1572 netdev_update_features(dev); 1573 1574 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1575 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1576 } 1577 1578 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1579 { 1580 #define N(val) \ 1581 case NETDEV_##val: \ 1582 return "NETDEV_" __stringify(val); 1583 switch (cmd) { 1584 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1585 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1586 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1587 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1588 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1589 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1590 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1591 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1592 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1593 N(PRE_CHANGEADDR) 1594 } 1595 #undef N 1596 return "UNKNOWN_NETDEV_EVENT"; 1597 } 1598 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1599 1600 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1601 struct net_device *dev) 1602 { 1603 struct netdev_notifier_info info = { 1604 .dev = dev, 1605 }; 1606 1607 return nb->notifier_call(nb, val, &info); 1608 } 1609 1610 static int dev_boot_phase = 1; 1611 1612 /** 1613 * register_netdevice_notifier - register a network notifier block 1614 * @nb: notifier 1615 * 1616 * Register a notifier to be called when network device events occur. 1617 * The notifier passed is linked into the kernel structures and must 1618 * not be reused until it has been unregistered. A negative errno code 1619 * is returned on a failure. 1620 * 1621 * When registered all registration and up events are replayed 1622 * to the new notifier to allow device to have a race free 1623 * view of the network device list. 1624 */ 1625 1626 int register_netdevice_notifier(struct notifier_block *nb) 1627 { 1628 struct net_device *dev; 1629 struct net_device *last; 1630 struct net *net; 1631 int err; 1632 1633 /* Close race with setup_net() and cleanup_net() */ 1634 down_write(&pernet_ops_rwsem); 1635 rtnl_lock(); 1636 err = raw_notifier_chain_register(&netdev_chain, nb); 1637 if (err) 1638 goto unlock; 1639 if (dev_boot_phase) 1640 goto unlock; 1641 for_each_net(net) { 1642 for_each_netdev(net, dev) { 1643 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1644 err = notifier_to_errno(err); 1645 if (err) 1646 goto rollback; 1647 1648 if (!(dev->flags & IFF_UP)) 1649 continue; 1650 1651 call_netdevice_notifier(nb, NETDEV_UP, dev); 1652 } 1653 } 1654 1655 unlock: 1656 rtnl_unlock(); 1657 up_write(&pernet_ops_rwsem); 1658 return err; 1659 1660 rollback: 1661 last = dev; 1662 for_each_net(net) { 1663 for_each_netdev(net, dev) { 1664 if (dev == last) 1665 goto outroll; 1666 1667 if (dev->flags & IFF_UP) { 1668 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1669 dev); 1670 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1671 } 1672 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1673 } 1674 } 1675 1676 outroll: 1677 raw_notifier_chain_unregister(&netdev_chain, nb); 1678 goto unlock; 1679 } 1680 EXPORT_SYMBOL(register_netdevice_notifier); 1681 1682 /** 1683 * unregister_netdevice_notifier - unregister a network notifier block 1684 * @nb: notifier 1685 * 1686 * Unregister a notifier previously registered by 1687 * register_netdevice_notifier(). The notifier is unlinked into the 1688 * kernel structures and may then be reused. A negative errno code 1689 * is returned on a failure. 1690 * 1691 * After unregistering unregister and down device events are synthesized 1692 * for all devices on the device list to the removed notifier to remove 1693 * the need for special case cleanup code. 1694 */ 1695 1696 int unregister_netdevice_notifier(struct notifier_block *nb) 1697 { 1698 struct net_device *dev; 1699 struct net *net; 1700 int err; 1701 1702 /* Close race with setup_net() and cleanup_net() */ 1703 down_write(&pernet_ops_rwsem); 1704 rtnl_lock(); 1705 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1706 if (err) 1707 goto unlock; 1708 1709 for_each_net(net) { 1710 for_each_netdev(net, dev) { 1711 if (dev->flags & IFF_UP) { 1712 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1713 dev); 1714 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1715 } 1716 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1717 } 1718 } 1719 unlock: 1720 rtnl_unlock(); 1721 up_write(&pernet_ops_rwsem); 1722 return err; 1723 } 1724 EXPORT_SYMBOL(unregister_netdevice_notifier); 1725 1726 /** 1727 * call_netdevice_notifiers_info - call all network notifier blocks 1728 * @val: value passed unmodified to notifier function 1729 * @info: notifier information data 1730 * 1731 * Call all network notifier blocks. Parameters and return value 1732 * are as for raw_notifier_call_chain(). 1733 */ 1734 1735 static int call_netdevice_notifiers_info(unsigned long val, 1736 struct netdev_notifier_info *info) 1737 { 1738 ASSERT_RTNL(); 1739 return raw_notifier_call_chain(&netdev_chain, val, info); 1740 } 1741 1742 static int call_netdevice_notifiers_extack(unsigned long val, 1743 struct net_device *dev, 1744 struct netlink_ext_ack *extack) 1745 { 1746 struct netdev_notifier_info info = { 1747 .dev = dev, 1748 .extack = extack, 1749 }; 1750 1751 return call_netdevice_notifiers_info(val, &info); 1752 } 1753 1754 /** 1755 * call_netdevice_notifiers - call all network notifier blocks 1756 * @val: value passed unmodified to notifier function 1757 * @dev: net_device pointer passed unmodified to notifier function 1758 * 1759 * Call all network notifier blocks. Parameters and return value 1760 * are as for raw_notifier_call_chain(). 1761 */ 1762 1763 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1764 { 1765 return call_netdevice_notifiers_extack(val, dev, NULL); 1766 } 1767 EXPORT_SYMBOL(call_netdevice_notifiers); 1768 1769 /** 1770 * call_netdevice_notifiers_mtu - call all network notifier blocks 1771 * @val: value passed unmodified to notifier function 1772 * @dev: net_device pointer passed unmodified to notifier function 1773 * @arg: additional u32 argument passed to the notifier function 1774 * 1775 * Call all network notifier blocks. Parameters and return value 1776 * are as for raw_notifier_call_chain(). 1777 */ 1778 static int call_netdevice_notifiers_mtu(unsigned long val, 1779 struct net_device *dev, u32 arg) 1780 { 1781 struct netdev_notifier_info_ext info = { 1782 .info.dev = dev, 1783 .ext.mtu = arg, 1784 }; 1785 1786 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 1787 1788 return call_netdevice_notifiers_info(val, &info.info); 1789 } 1790 1791 #ifdef CONFIG_NET_INGRESS 1792 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 1793 1794 void net_inc_ingress_queue(void) 1795 { 1796 static_branch_inc(&ingress_needed_key); 1797 } 1798 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1799 1800 void net_dec_ingress_queue(void) 1801 { 1802 static_branch_dec(&ingress_needed_key); 1803 } 1804 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1805 #endif 1806 1807 #ifdef CONFIG_NET_EGRESS 1808 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 1809 1810 void net_inc_egress_queue(void) 1811 { 1812 static_branch_inc(&egress_needed_key); 1813 } 1814 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 1815 1816 void net_dec_egress_queue(void) 1817 { 1818 static_branch_dec(&egress_needed_key); 1819 } 1820 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 1821 #endif 1822 1823 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 1824 #ifdef CONFIG_JUMP_LABEL 1825 static atomic_t netstamp_needed_deferred; 1826 static atomic_t netstamp_wanted; 1827 static void netstamp_clear(struct work_struct *work) 1828 { 1829 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 1830 int wanted; 1831 1832 wanted = atomic_add_return(deferred, &netstamp_wanted); 1833 if (wanted > 0) 1834 static_branch_enable(&netstamp_needed_key); 1835 else 1836 static_branch_disable(&netstamp_needed_key); 1837 } 1838 static DECLARE_WORK(netstamp_work, netstamp_clear); 1839 #endif 1840 1841 void net_enable_timestamp(void) 1842 { 1843 #ifdef CONFIG_JUMP_LABEL 1844 int wanted; 1845 1846 while (1) { 1847 wanted = atomic_read(&netstamp_wanted); 1848 if (wanted <= 0) 1849 break; 1850 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 1851 return; 1852 } 1853 atomic_inc(&netstamp_needed_deferred); 1854 schedule_work(&netstamp_work); 1855 #else 1856 static_branch_inc(&netstamp_needed_key); 1857 #endif 1858 } 1859 EXPORT_SYMBOL(net_enable_timestamp); 1860 1861 void net_disable_timestamp(void) 1862 { 1863 #ifdef CONFIG_JUMP_LABEL 1864 int wanted; 1865 1866 while (1) { 1867 wanted = atomic_read(&netstamp_wanted); 1868 if (wanted <= 1) 1869 break; 1870 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 1871 return; 1872 } 1873 atomic_dec(&netstamp_needed_deferred); 1874 schedule_work(&netstamp_work); 1875 #else 1876 static_branch_dec(&netstamp_needed_key); 1877 #endif 1878 } 1879 EXPORT_SYMBOL(net_disable_timestamp); 1880 1881 static inline void net_timestamp_set(struct sk_buff *skb) 1882 { 1883 skb->tstamp = 0; 1884 if (static_branch_unlikely(&netstamp_needed_key)) 1885 __net_timestamp(skb); 1886 } 1887 1888 #define net_timestamp_check(COND, SKB) \ 1889 if (static_branch_unlikely(&netstamp_needed_key)) { \ 1890 if ((COND) && !(SKB)->tstamp) \ 1891 __net_timestamp(SKB); \ 1892 } \ 1893 1894 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 1895 { 1896 unsigned int len; 1897 1898 if (!(dev->flags & IFF_UP)) 1899 return false; 1900 1901 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 1902 if (skb->len <= len) 1903 return true; 1904 1905 /* if TSO is enabled, we don't care about the length as the packet 1906 * could be forwarded without being segmented before 1907 */ 1908 if (skb_is_gso(skb)) 1909 return true; 1910 1911 return false; 1912 } 1913 EXPORT_SYMBOL_GPL(is_skb_forwardable); 1914 1915 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1916 { 1917 int ret = ____dev_forward_skb(dev, skb); 1918 1919 if (likely(!ret)) { 1920 skb->protocol = eth_type_trans(skb, dev); 1921 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1922 } 1923 1924 return ret; 1925 } 1926 EXPORT_SYMBOL_GPL(__dev_forward_skb); 1927 1928 /** 1929 * dev_forward_skb - loopback an skb to another netif 1930 * 1931 * @dev: destination network device 1932 * @skb: buffer to forward 1933 * 1934 * return values: 1935 * NET_RX_SUCCESS (no congestion) 1936 * NET_RX_DROP (packet was dropped, but freed) 1937 * 1938 * dev_forward_skb can be used for injecting an skb from the 1939 * start_xmit function of one device into the receive queue 1940 * of another device. 1941 * 1942 * The receiving device may be in another namespace, so 1943 * we have to clear all information in the skb that could 1944 * impact namespace isolation. 1945 */ 1946 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 1947 { 1948 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 1949 } 1950 EXPORT_SYMBOL_GPL(dev_forward_skb); 1951 1952 static inline int deliver_skb(struct sk_buff *skb, 1953 struct packet_type *pt_prev, 1954 struct net_device *orig_dev) 1955 { 1956 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 1957 return -ENOMEM; 1958 refcount_inc(&skb->users); 1959 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 1960 } 1961 1962 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 1963 struct packet_type **pt, 1964 struct net_device *orig_dev, 1965 __be16 type, 1966 struct list_head *ptype_list) 1967 { 1968 struct packet_type *ptype, *pt_prev = *pt; 1969 1970 list_for_each_entry_rcu(ptype, ptype_list, list) { 1971 if (ptype->type != type) 1972 continue; 1973 if (pt_prev) 1974 deliver_skb(skb, pt_prev, orig_dev); 1975 pt_prev = ptype; 1976 } 1977 *pt = pt_prev; 1978 } 1979 1980 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 1981 { 1982 if (!ptype->af_packet_priv || !skb->sk) 1983 return false; 1984 1985 if (ptype->id_match) 1986 return ptype->id_match(ptype, skb->sk); 1987 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 1988 return true; 1989 1990 return false; 1991 } 1992 1993 /** 1994 * dev_nit_active - return true if any network interface taps are in use 1995 * 1996 * @dev: network device to check for the presence of taps 1997 */ 1998 bool dev_nit_active(struct net_device *dev) 1999 { 2000 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2001 } 2002 EXPORT_SYMBOL_GPL(dev_nit_active); 2003 2004 /* 2005 * Support routine. Sends outgoing frames to any network 2006 * taps currently in use. 2007 */ 2008 2009 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2010 { 2011 struct packet_type *ptype; 2012 struct sk_buff *skb2 = NULL; 2013 struct packet_type *pt_prev = NULL; 2014 struct list_head *ptype_list = &ptype_all; 2015 2016 rcu_read_lock(); 2017 again: 2018 list_for_each_entry_rcu(ptype, ptype_list, list) { 2019 if (ptype->ignore_outgoing) 2020 continue; 2021 2022 /* Never send packets back to the socket 2023 * they originated from - MvS (miquels@drinkel.ow.org) 2024 */ 2025 if (skb_loop_sk(ptype, skb)) 2026 continue; 2027 2028 if (pt_prev) { 2029 deliver_skb(skb2, pt_prev, skb->dev); 2030 pt_prev = ptype; 2031 continue; 2032 } 2033 2034 /* need to clone skb, done only once */ 2035 skb2 = skb_clone(skb, GFP_ATOMIC); 2036 if (!skb2) 2037 goto out_unlock; 2038 2039 net_timestamp_set(skb2); 2040 2041 /* skb->nh should be correctly 2042 * set by sender, so that the second statement is 2043 * just protection against buggy protocols. 2044 */ 2045 skb_reset_mac_header(skb2); 2046 2047 if (skb_network_header(skb2) < skb2->data || 2048 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2049 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2050 ntohs(skb2->protocol), 2051 dev->name); 2052 skb_reset_network_header(skb2); 2053 } 2054 2055 skb2->transport_header = skb2->network_header; 2056 skb2->pkt_type = PACKET_OUTGOING; 2057 pt_prev = ptype; 2058 } 2059 2060 if (ptype_list == &ptype_all) { 2061 ptype_list = &dev->ptype_all; 2062 goto again; 2063 } 2064 out_unlock: 2065 if (pt_prev) { 2066 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2067 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2068 else 2069 kfree_skb(skb2); 2070 } 2071 rcu_read_unlock(); 2072 } 2073 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2074 2075 /** 2076 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2077 * @dev: Network device 2078 * @txq: number of queues available 2079 * 2080 * If real_num_tx_queues is changed the tc mappings may no longer be 2081 * valid. To resolve this verify the tc mapping remains valid and if 2082 * not NULL the mapping. With no priorities mapping to this 2083 * offset/count pair it will no longer be used. In the worst case TC0 2084 * is invalid nothing can be done so disable priority mappings. If is 2085 * expected that drivers will fix this mapping if they can before 2086 * calling netif_set_real_num_tx_queues. 2087 */ 2088 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2089 { 2090 int i; 2091 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2092 2093 /* If TC0 is invalidated disable TC mapping */ 2094 if (tc->offset + tc->count > txq) { 2095 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2096 dev->num_tc = 0; 2097 return; 2098 } 2099 2100 /* Invalidated prio to tc mappings set to TC0 */ 2101 for (i = 1; i < TC_BITMASK + 1; i++) { 2102 int q = netdev_get_prio_tc_map(dev, i); 2103 2104 tc = &dev->tc_to_txq[q]; 2105 if (tc->offset + tc->count > txq) { 2106 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2107 i, q); 2108 netdev_set_prio_tc_map(dev, i, 0); 2109 } 2110 } 2111 } 2112 2113 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2114 { 2115 if (dev->num_tc) { 2116 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2117 int i; 2118 2119 /* walk through the TCs and see if it falls into any of them */ 2120 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2121 if ((txq - tc->offset) < tc->count) 2122 return i; 2123 } 2124 2125 /* didn't find it, just return -1 to indicate no match */ 2126 return -1; 2127 } 2128 2129 return 0; 2130 } 2131 EXPORT_SYMBOL(netdev_txq_to_tc); 2132 2133 #ifdef CONFIG_XPS 2134 struct static_key xps_needed __read_mostly; 2135 EXPORT_SYMBOL(xps_needed); 2136 struct static_key xps_rxqs_needed __read_mostly; 2137 EXPORT_SYMBOL(xps_rxqs_needed); 2138 static DEFINE_MUTEX(xps_map_mutex); 2139 #define xmap_dereference(P) \ 2140 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2141 2142 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2143 int tci, u16 index) 2144 { 2145 struct xps_map *map = NULL; 2146 int pos; 2147 2148 if (dev_maps) 2149 map = xmap_dereference(dev_maps->attr_map[tci]); 2150 if (!map) 2151 return false; 2152 2153 for (pos = map->len; pos--;) { 2154 if (map->queues[pos] != index) 2155 continue; 2156 2157 if (map->len > 1) { 2158 map->queues[pos] = map->queues[--map->len]; 2159 break; 2160 } 2161 2162 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2163 kfree_rcu(map, rcu); 2164 return false; 2165 } 2166 2167 return true; 2168 } 2169 2170 static bool remove_xps_queue_cpu(struct net_device *dev, 2171 struct xps_dev_maps *dev_maps, 2172 int cpu, u16 offset, u16 count) 2173 { 2174 int num_tc = dev->num_tc ? : 1; 2175 bool active = false; 2176 int tci; 2177 2178 for (tci = cpu * num_tc; num_tc--; tci++) { 2179 int i, j; 2180 2181 for (i = count, j = offset; i--; j++) { 2182 if (!remove_xps_queue(dev_maps, tci, j)) 2183 break; 2184 } 2185 2186 active |= i < 0; 2187 } 2188 2189 return active; 2190 } 2191 2192 static void reset_xps_maps(struct net_device *dev, 2193 struct xps_dev_maps *dev_maps, 2194 bool is_rxqs_map) 2195 { 2196 if (is_rxqs_map) { 2197 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2198 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); 2199 } else { 2200 RCU_INIT_POINTER(dev->xps_cpus_map, NULL); 2201 } 2202 static_key_slow_dec_cpuslocked(&xps_needed); 2203 kfree_rcu(dev_maps, rcu); 2204 } 2205 2206 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, 2207 struct xps_dev_maps *dev_maps, unsigned int nr_ids, 2208 u16 offset, u16 count, bool is_rxqs_map) 2209 { 2210 bool active = false; 2211 int i, j; 2212 2213 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), 2214 j < nr_ids;) 2215 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, 2216 count); 2217 if (!active) 2218 reset_xps_maps(dev, dev_maps, is_rxqs_map); 2219 2220 if (!is_rxqs_map) { 2221 for (i = offset + (count - 1); count--; i--) { 2222 netdev_queue_numa_node_write( 2223 netdev_get_tx_queue(dev, i), 2224 NUMA_NO_NODE); 2225 } 2226 } 2227 } 2228 2229 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2230 u16 count) 2231 { 2232 const unsigned long *possible_mask = NULL; 2233 struct xps_dev_maps *dev_maps; 2234 unsigned int nr_ids; 2235 2236 if (!static_key_false(&xps_needed)) 2237 return; 2238 2239 cpus_read_lock(); 2240 mutex_lock(&xps_map_mutex); 2241 2242 if (static_key_false(&xps_rxqs_needed)) { 2243 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2244 if (dev_maps) { 2245 nr_ids = dev->num_rx_queues; 2246 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, 2247 offset, count, true); 2248 } 2249 } 2250 2251 dev_maps = xmap_dereference(dev->xps_cpus_map); 2252 if (!dev_maps) 2253 goto out_no_maps; 2254 2255 if (num_possible_cpus() > 1) 2256 possible_mask = cpumask_bits(cpu_possible_mask); 2257 nr_ids = nr_cpu_ids; 2258 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, 2259 false); 2260 2261 out_no_maps: 2262 mutex_unlock(&xps_map_mutex); 2263 cpus_read_unlock(); 2264 } 2265 2266 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2267 { 2268 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2269 } 2270 2271 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2272 u16 index, bool is_rxqs_map) 2273 { 2274 struct xps_map *new_map; 2275 int alloc_len = XPS_MIN_MAP_ALLOC; 2276 int i, pos; 2277 2278 for (pos = 0; map && pos < map->len; pos++) { 2279 if (map->queues[pos] != index) 2280 continue; 2281 return map; 2282 } 2283 2284 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2285 if (map) { 2286 if (pos < map->alloc_len) 2287 return map; 2288 2289 alloc_len = map->alloc_len * 2; 2290 } 2291 2292 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2293 * map 2294 */ 2295 if (is_rxqs_map) 2296 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2297 else 2298 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2299 cpu_to_node(attr_index)); 2300 if (!new_map) 2301 return NULL; 2302 2303 for (i = 0; i < pos; i++) 2304 new_map->queues[i] = map->queues[i]; 2305 new_map->alloc_len = alloc_len; 2306 new_map->len = pos; 2307 2308 return new_map; 2309 } 2310 2311 /* Must be called under cpus_read_lock */ 2312 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2313 u16 index, bool is_rxqs_map) 2314 { 2315 const unsigned long *online_mask = NULL, *possible_mask = NULL; 2316 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2317 int i, j, tci, numa_node_id = -2; 2318 int maps_sz, num_tc = 1, tc = 0; 2319 struct xps_map *map, *new_map; 2320 bool active = false; 2321 unsigned int nr_ids; 2322 2323 if (dev->num_tc) { 2324 /* Do not allow XPS on subordinate device directly */ 2325 num_tc = dev->num_tc; 2326 if (num_tc < 0) 2327 return -EINVAL; 2328 2329 /* If queue belongs to subordinate dev use its map */ 2330 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2331 2332 tc = netdev_txq_to_tc(dev, index); 2333 if (tc < 0) 2334 return -EINVAL; 2335 } 2336 2337 mutex_lock(&xps_map_mutex); 2338 if (is_rxqs_map) { 2339 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2340 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2341 nr_ids = dev->num_rx_queues; 2342 } else { 2343 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2344 if (num_possible_cpus() > 1) { 2345 online_mask = cpumask_bits(cpu_online_mask); 2346 possible_mask = cpumask_bits(cpu_possible_mask); 2347 } 2348 dev_maps = xmap_dereference(dev->xps_cpus_map); 2349 nr_ids = nr_cpu_ids; 2350 } 2351 2352 if (maps_sz < L1_CACHE_BYTES) 2353 maps_sz = L1_CACHE_BYTES; 2354 2355 /* allocate memory for queue storage */ 2356 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2357 j < nr_ids;) { 2358 if (!new_dev_maps) 2359 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2360 if (!new_dev_maps) { 2361 mutex_unlock(&xps_map_mutex); 2362 return -ENOMEM; 2363 } 2364 2365 tci = j * num_tc + tc; 2366 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : 2367 NULL; 2368 2369 map = expand_xps_map(map, j, index, is_rxqs_map); 2370 if (!map) 2371 goto error; 2372 2373 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2374 } 2375 2376 if (!new_dev_maps) 2377 goto out_no_new_maps; 2378 2379 if (!dev_maps) { 2380 /* Increment static keys at most once per type */ 2381 static_key_slow_inc_cpuslocked(&xps_needed); 2382 if (is_rxqs_map) 2383 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2384 } 2385 2386 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2387 j < nr_ids;) { 2388 /* copy maps belonging to foreign traffic classes */ 2389 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { 2390 /* fill in the new device map from the old device map */ 2391 map = xmap_dereference(dev_maps->attr_map[tci]); 2392 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2393 } 2394 2395 /* We need to explicitly update tci as prevous loop 2396 * could break out early if dev_maps is NULL. 2397 */ 2398 tci = j * num_tc + tc; 2399 2400 if (netif_attr_test_mask(j, mask, nr_ids) && 2401 netif_attr_test_online(j, online_mask, nr_ids)) { 2402 /* add tx-queue to CPU/rx-queue maps */ 2403 int pos = 0; 2404 2405 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2406 while ((pos < map->len) && (map->queues[pos] != index)) 2407 pos++; 2408 2409 if (pos == map->len) 2410 map->queues[map->len++] = index; 2411 #ifdef CONFIG_NUMA 2412 if (!is_rxqs_map) { 2413 if (numa_node_id == -2) 2414 numa_node_id = cpu_to_node(j); 2415 else if (numa_node_id != cpu_to_node(j)) 2416 numa_node_id = -1; 2417 } 2418 #endif 2419 } else if (dev_maps) { 2420 /* fill in the new device map from the old device map */ 2421 map = xmap_dereference(dev_maps->attr_map[tci]); 2422 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2423 } 2424 2425 /* copy maps belonging to foreign traffic classes */ 2426 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { 2427 /* fill in the new device map from the old device map */ 2428 map = xmap_dereference(dev_maps->attr_map[tci]); 2429 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2430 } 2431 } 2432 2433 if (is_rxqs_map) 2434 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); 2435 else 2436 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); 2437 2438 /* Cleanup old maps */ 2439 if (!dev_maps) 2440 goto out_no_old_maps; 2441 2442 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2443 j < nr_ids;) { 2444 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2445 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2446 map = xmap_dereference(dev_maps->attr_map[tci]); 2447 if (map && map != new_map) 2448 kfree_rcu(map, rcu); 2449 } 2450 } 2451 2452 kfree_rcu(dev_maps, rcu); 2453 2454 out_no_old_maps: 2455 dev_maps = new_dev_maps; 2456 active = true; 2457 2458 out_no_new_maps: 2459 if (!is_rxqs_map) { 2460 /* update Tx queue numa node */ 2461 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2462 (numa_node_id >= 0) ? 2463 numa_node_id : NUMA_NO_NODE); 2464 } 2465 2466 if (!dev_maps) 2467 goto out_no_maps; 2468 2469 /* removes tx-queue from unused CPUs/rx-queues */ 2470 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2471 j < nr_ids;) { 2472 for (i = tc, tci = j * num_tc; i--; tci++) 2473 active |= remove_xps_queue(dev_maps, tci, index); 2474 if (!netif_attr_test_mask(j, mask, nr_ids) || 2475 !netif_attr_test_online(j, online_mask, nr_ids)) 2476 active |= remove_xps_queue(dev_maps, tci, index); 2477 for (i = num_tc - tc, tci++; --i; tci++) 2478 active |= remove_xps_queue(dev_maps, tci, index); 2479 } 2480 2481 /* free map if not active */ 2482 if (!active) 2483 reset_xps_maps(dev, dev_maps, is_rxqs_map); 2484 2485 out_no_maps: 2486 mutex_unlock(&xps_map_mutex); 2487 2488 return 0; 2489 error: 2490 /* remove any maps that we added */ 2491 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2492 j < nr_ids;) { 2493 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2494 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2495 map = dev_maps ? 2496 xmap_dereference(dev_maps->attr_map[tci]) : 2497 NULL; 2498 if (new_map && new_map != map) 2499 kfree(new_map); 2500 } 2501 } 2502 2503 mutex_unlock(&xps_map_mutex); 2504 2505 kfree(new_dev_maps); 2506 return -ENOMEM; 2507 } 2508 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2509 2510 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2511 u16 index) 2512 { 2513 int ret; 2514 2515 cpus_read_lock(); 2516 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); 2517 cpus_read_unlock(); 2518 2519 return ret; 2520 } 2521 EXPORT_SYMBOL(netif_set_xps_queue); 2522 2523 #endif 2524 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2525 { 2526 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2527 2528 /* Unbind any subordinate channels */ 2529 while (txq-- != &dev->_tx[0]) { 2530 if (txq->sb_dev) 2531 netdev_unbind_sb_channel(dev, txq->sb_dev); 2532 } 2533 } 2534 2535 void netdev_reset_tc(struct net_device *dev) 2536 { 2537 #ifdef CONFIG_XPS 2538 netif_reset_xps_queues_gt(dev, 0); 2539 #endif 2540 netdev_unbind_all_sb_channels(dev); 2541 2542 /* Reset TC configuration of device */ 2543 dev->num_tc = 0; 2544 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2545 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2546 } 2547 EXPORT_SYMBOL(netdev_reset_tc); 2548 2549 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2550 { 2551 if (tc >= dev->num_tc) 2552 return -EINVAL; 2553 2554 #ifdef CONFIG_XPS 2555 netif_reset_xps_queues(dev, offset, count); 2556 #endif 2557 dev->tc_to_txq[tc].count = count; 2558 dev->tc_to_txq[tc].offset = offset; 2559 return 0; 2560 } 2561 EXPORT_SYMBOL(netdev_set_tc_queue); 2562 2563 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2564 { 2565 if (num_tc > TC_MAX_QUEUE) 2566 return -EINVAL; 2567 2568 #ifdef CONFIG_XPS 2569 netif_reset_xps_queues_gt(dev, 0); 2570 #endif 2571 netdev_unbind_all_sb_channels(dev); 2572 2573 dev->num_tc = num_tc; 2574 return 0; 2575 } 2576 EXPORT_SYMBOL(netdev_set_num_tc); 2577 2578 void netdev_unbind_sb_channel(struct net_device *dev, 2579 struct net_device *sb_dev) 2580 { 2581 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2582 2583 #ifdef CONFIG_XPS 2584 netif_reset_xps_queues_gt(sb_dev, 0); 2585 #endif 2586 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2587 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2588 2589 while (txq-- != &dev->_tx[0]) { 2590 if (txq->sb_dev == sb_dev) 2591 txq->sb_dev = NULL; 2592 } 2593 } 2594 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2595 2596 int netdev_bind_sb_channel_queue(struct net_device *dev, 2597 struct net_device *sb_dev, 2598 u8 tc, u16 count, u16 offset) 2599 { 2600 /* Make certain the sb_dev and dev are already configured */ 2601 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2602 return -EINVAL; 2603 2604 /* We cannot hand out queues we don't have */ 2605 if ((offset + count) > dev->real_num_tx_queues) 2606 return -EINVAL; 2607 2608 /* Record the mapping */ 2609 sb_dev->tc_to_txq[tc].count = count; 2610 sb_dev->tc_to_txq[tc].offset = offset; 2611 2612 /* Provide a way for Tx queue to find the tc_to_txq map or 2613 * XPS map for itself. 2614 */ 2615 while (count--) 2616 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2617 2618 return 0; 2619 } 2620 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2621 2622 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2623 { 2624 /* Do not use a multiqueue device to represent a subordinate channel */ 2625 if (netif_is_multiqueue(dev)) 2626 return -ENODEV; 2627 2628 /* We allow channels 1 - 32767 to be used for subordinate channels. 2629 * Channel 0 is meant to be "native" mode and used only to represent 2630 * the main root device. We allow writing 0 to reset the device back 2631 * to normal mode after being used as a subordinate channel. 2632 */ 2633 if (channel > S16_MAX) 2634 return -EINVAL; 2635 2636 dev->num_tc = -channel; 2637 2638 return 0; 2639 } 2640 EXPORT_SYMBOL(netdev_set_sb_channel); 2641 2642 /* 2643 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2644 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2645 */ 2646 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2647 { 2648 bool disabling; 2649 int rc; 2650 2651 disabling = txq < dev->real_num_tx_queues; 2652 2653 if (txq < 1 || txq > dev->num_tx_queues) 2654 return -EINVAL; 2655 2656 if (dev->reg_state == NETREG_REGISTERED || 2657 dev->reg_state == NETREG_UNREGISTERING) { 2658 ASSERT_RTNL(); 2659 2660 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2661 txq); 2662 if (rc) 2663 return rc; 2664 2665 if (dev->num_tc) 2666 netif_setup_tc(dev, txq); 2667 2668 dev->real_num_tx_queues = txq; 2669 2670 if (disabling) { 2671 synchronize_net(); 2672 qdisc_reset_all_tx_gt(dev, txq); 2673 #ifdef CONFIG_XPS 2674 netif_reset_xps_queues_gt(dev, txq); 2675 #endif 2676 } 2677 } else { 2678 dev->real_num_tx_queues = txq; 2679 } 2680 2681 return 0; 2682 } 2683 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2684 2685 #ifdef CONFIG_SYSFS 2686 /** 2687 * netif_set_real_num_rx_queues - set actual number of RX queues used 2688 * @dev: Network device 2689 * @rxq: Actual number of RX queues 2690 * 2691 * This must be called either with the rtnl_lock held or before 2692 * registration of the net device. Returns 0 on success, or a 2693 * negative error code. If called before registration, it always 2694 * succeeds. 2695 */ 2696 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2697 { 2698 int rc; 2699 2700 if (rxq < 1 || rxq > dev->num_rx_queues) 2701 return -EINVAL; 2702 2703 if (dev->reg_state == NETREG_REGISTERED) { 2704 ASSERT_RTNL(); 2705 2706 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2707 rxq); 2708 if (rc) 2709 return rc; 2710 } 2711 2712 dev->real_num_rx_queues = rxq; 2713 return 0; 2714 } 2715 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2716 #endif 2717 2718 /** 2719 * netif_get_num_default_rss_queues - default number of RSS queues 2720 * 2721 * This routine should set an upper limit on the number of RSS queues 2722 * used by default by multiqueue devices. 2723 */ 2724 int netif_get_num_default_rss_queues(void) 2725 { 2726 return is_kdump_kernel() ? 2727 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2728 } 2729 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2730 2731 static void __netif_reschedule(struct Qdisc *q) 2732 { 2733 struct softnet_data *sd; 2734 unsigned long flags; 2735 2736 local_irq_save(flags); 2737 sd = this_cpu_ptr(&softnet_data); 2738 q->next_sched = NULL; 2739 *sd->output_queue_tailp = q; 2740 sd->output_queue_tailp = &q->next_sched; 2741 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2742 local_irq_restore(flags); 2743 } 2744 2745 void __netif_schedule(struct Qdisc *q) 2746 { 2747 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2748 __netif_reschedule(q); 2749 } 2750 EXPORT_SYMBOL(__netif_schedule); 2751 2752 struct dev_kfree_skb_cb { 2753 enum skb_free_reason reason; 2754 }; 2755 2756 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2757 { 2758 return (struct dev_kfree_skb_cb *)skb->cb; 2759 } 2760 2761 void netif_schedule_queue(struct netdev_queue *txq) 2762 { 2763 rcu_read_lock(); 2764 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { 2765 struct Qdisc *q = rcu_dereference(txq->qdisc); 2766 2767 __netif_schedule(q); 2768 } 2769 rcu_read_unlock(); 2770 } 2771 EXPORT_SYMBOL(netif_schedule_queue); 2772 2773 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2774 { 2775 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 2776 struct Qdisc *q; 2777 2778 rcu_read_lock(); 2779 q = rcu_dereference(dev_queue->qdisc); 2780 __netif_schedule(q); 2781 rcu_read_unlock(); 2782 } 2783 } 2784 EXPORT_SYMBOL(netif_tx_wake_queue); 2785 2786 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 2787 { 2788 unsigned long flags; 2789 2790 if (unlikely(!skb)) 2791 return; 2792 2793 if (likely(refcount_read(&skb->users) == 1)) { 2794 smp_rmb(); 2795 refcount_set(&skb->users, 0); 2796 } else if (likely(!refcount_dec_and_test(&skb->users))) { 2797 return; 2798 } 2799 get_kfree_skb_cb(skb)->reason = reason; 2800 local_irq_save(flags); 2801 skb->next = __this_cpu_read(softnet_data.completion_queue); 2802 __this_cpu_write(softnet_data.completion_queue, skb); 2803 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2804 local_irq_restore(flags); 2805 } 2806 EXPORT_SYMBOL(__dev_kfree_skb_irq); 2807 2808 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 2809 { 2810 if (in_irq() || irqs_disabled()) 2811 __dev_kfree_skb_irq(skb, reason); 2812 else 2813 dev_kfree_skb(skb); 2814 } 2815 EXPORT_SYMBOL(__dev_kfree_skb_any); 2816 2817 2818 /** 2819 * netif_device_detach - mark device as removed 2820 * @dev: network device 2821 * 2822 * Mark device as removed from system and therefore no longer available. 2823 */ 2824 void netif_device_detach(struct net_device *dev) 2825 { 2826 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 2827 netif_running(dev)) { 2828 netif_tx_stop_all_queues(dev); 2829 } 2830 } 2831 EXPORT_SYMBOL(netif_device_detach); 2832 2833 /** 2834 * netif_device_attach - mark device as attached 2835 * @dev: network device 2836 * 2837 * Mark device as attached from system and restart if needed. 2838 */ 2839 void netif_device_attach(struct net_device *dev) 2840 { 2841 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 2842 netif_running(dev)) { 2843 netif_tx_wake_all_queues(dev); 2844 __netdev_watchdog_up(dev); 2845 } 2846 } 2847 EXPORT_SYMBOL(netif_device_attach); 2848 2849 /* 2850 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 2851 * to be used as a distribution range. 2852 */ 2853 static u16 skb_tx_hash(const struct net_device *dev, 2854 const struct net_device *sb_dev, 2855 struct sk_buff *skb) 2856 { 2857 u32 hash; 2858 u16 qoffset = 0; 2859 u16 qcount = dev->real_num_tx_queues; 2860 2861 if (dev->num_tc) { 2862 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 2863 2864 qoffset = sb_dev->tc_to_txq[tc].offset; 2865 qcount = sb_dev->tc_to_txq[tc].count; 2866 } 2867 2868 if (skb_rx_queue_recorded(skb)) { 2869 hash = skb_get_rx_queue(skb); 2870 while (unlikely(hash >= qcount)) 2871 hash -= qcount; 2872 return hash + qoffset; 2873 } 2874 2875 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 2876 } 2877 2878 static void skb_warn_bad_offload(const struct sk_buff *skb) 2879 { 2880 static const netdev_features_t null_features; 2881 struct net_device *dev = skb->dev; 2882 const char *name = ""; 2883 2884 if (!net_ratelimit()) 2885 return; 2886 2887 if (dev) { 2888 if (dev->dev.parent) 2889 name = dev_driver_string(dev->dev.parent); 2890 else 2891 name = netdev_name(dev); 2892 } 2893 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " 2894 "gso_type=%d ip_summed=%d\n", 2895 name, dev ? &dev->features : &null_features, 2896 skb->sk ? &skb->sk->sk_route_caps : &null_features, 2897 skb->len, skb->data_len, skb_shinfo(skb)->gso_size, 2898 skb_shinfo(skb)->gso_type, skb->ip_summed); 2899 } 2900 2901 /* 2902 * Invalidate hardware checksum when packet is to be mangled, and 2903 * complete checksum manually on outgoing path. 2904 */ 2905 int skb_checksum_help(struct sk_buff *skb) 2906 { 2907 __wsum csum; 2908 int ret = 0, offset; 2909 2910 if (skb->ip_summed == CHECKSUM_COMPLETE) 2911 goto out_set_summed; 2912 2913 if (unlikely(skb_shinfo(skb)->gso_size)) { 2914 skb_warn_bad_offload(skb); 2915 return -EINVAL; 2916 } 2917 2918 /* Before computing a checksum, we should make sure no frag could 2919 * be modified by an external entity : checksum could be wrong. 2920 */ 2921 if (skb_has_shared_frag(skb)) { 2922 ret = __skb_linearize(skb); 2923 if (ret) 2924 goto out; 2925 } 2926 2927 offset = skb_checksum_start_offset(skb); 2928 BUG_ON(offset >= skb_headlen(skb)); 2929 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2930 2931 offset += skb->csum_offset; 2932 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 2933 2934 if (skb_cloned(skb) && 2935 !skb_clone_writable(skb, offset + sizeof(__sum16))) { 2936 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2937 if (ret) 2938 goto out; 2939 } 2940 2941 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 2942 out_set_summed: 2943 skb->ip_summed = CHECKSUM_NONE; 2944 out: 2945 return ret; 2946 } 2947 EXPORT_SYMBOL(skb_checksum_help); 2948 2949 int skb_crc32c_csum_help(struct sk_buff *skb) 2950 { 2951 __le32 crc32c_csum; 2952 int ret = 0, offset, start; 2953 2954 if (skb->ip_summed != CHECKSUM_PARTIAL) 2955 goto out; 2956 2957 if (unlikely(skb_is_gso(skb))) 2958 goto out; 2959 2960 /* Before computing a checksum, we should make sure no frag could 2961 * be modified by an external entity : checksum could be wrong. 2962 */ 2963 if (unlikely(skb_has_shared_frag(skb))) { 2964 ret = __skb_linearize(skb); 2965 if (ret) 2966 goto out; 2967 } 2968 start = skb_checksum_start_offset(skb); 2969 offset = start + offsetof(struct sctphdr, checksum); 2970 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 2971 ret = -EINVAL; 2972 goto out; 2973 } 2974 if (skb_cloned(skb) && 2975 !skb_clone_writable(skb, offset + sizeof(__le32))) { 2976 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2977 if (ret) 2978 goto out; 2979 } 2980 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 2981 skb->len - start, ~(__u32)0, 2982 crc32c_csum_stub)); 2983 *(__le32 *)(skb->data + offset) = crc32c_csum; 2984 skb->ip_summed = CHECKSUM_NONE; 2985 skb->csum_not_inet = 0; 2986 out: 2987 return ret; 2988 } 2989 2990 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 2991 { 2992 __be16 type = skb->protocol; 2993 2994 /* Tunnel gso handlers can set protocol to ethernet. */ 2995 if (type == htons(ETH_P_TEB)) { 2996 struct ethhdr *eth; 2997 2998 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 2999 return 0; 3000 3001 eth = (struct ethhdr *)skb->data; 3002 type = eth->h_proto; 3003 } 3004 3005 return __vlan_get_protocol(skb, type, depth); 3006 } 3007 3008 /** 3009 * skb_mac_gso_segment - mac layer segmentation handler. 3010 * @skb: buffer to segment 3011 * @features: features for the output path (see dev->features) 3012 */ 3013 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3014 netdev_features_t features) 3015 { 3016 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3017 struct packet_offload *ptype; 3018 int vlan_depth = skb->mac_len; 3019 __be16 type = skb_network_protocol(skb, &vlan_depth); 3020 3021 if (unlikely(!type)) 3022 return ERR_PTR(-EINVAL); 3023 3024 __skb_pull(skb, vlan_depth); 3025 3026 rcu_read_lock(); 3027 list_for_each_entry_rcu(ptype, &offload_base, list) { 3028 if (ptype->type == type && ptype->callbacks.gso_segment) { 3029 segs = ptype->callbacks.gso_segment(skb, features); 3030 break; 3031 } 3032 } 3033 rcu_read_unlock(); 3034 3035 __skb_push(skb, skb->data - skb_mac_header(skb)); 3036 3037 return segs; 3038 } 3039 EXPORT_SYMBOL(skb_mac_gso_segment); 3040 3041 3042 /* openvswitch calls this on rx path, so we need a different check. 3043 */ 3044 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3045 { 3046 if (tx_path) 3047 return skb->ip_summed != CHECKSUM_PARTIAL && 3048 skb->ip_summed != CHECKSUM_UNNECESSARY; 3049 3050 return skb->ip_summed == CHECKSUM_NONE; 3051 } 3052 3053 /** 3054 * __skb_gso_segment - Perform segmentation on skb. 3055 * @skb: buffer to segment 3056 * @features: features for the output path (see dev->features) 3057 * @tx_path: whether it is called in TX path 3058 * 3059 * This function segments the given skb and returns a list of segments. 3060 * 3061 * It may return NULL if the skb requires no segmentation. This is 3062 * only possible when GSO is used for verifying header integrity. 3063 * 3064 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. 3065 */ 3066 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3067 netdev_features_t features, bool tx_path) 3068 { 3069 struct sk_buff *segs; 3070 3071 if (unlikely(skb_needs_check(skb, tx_path))) { 3072 int err; 3073 3074 /* We're going to init ->check field in TCP or UDP header */ 3075 err = skb_cow_head(skb, 0); 3076 if (err < 0) 3077 return ERR_PTR(err); 3078 } 3079 3080 /* Only report GSO partial support if it will enable us to 3081 * support segmentation on this frame without needing additional 3082 * work. 3083 */ 3084 if (features & NETIF_F_GSO_PARTIAL) { 3085 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3086 struct net_device *dev = skb->dev; 3087 3088 partial_features |= dev->features & dev->gso_partial_features; 3089 if (!skb_gso_ok(skb, features | partial_features)) 3090 features &= ~NETIF_F_GSO_PARTIAL; 3091 } 3092 3093 BUILD_BUG_ON(SKB_SGO_CB_OFFSET + 3094 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3095 3096 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3097 SKB_GSO_CB(skb)->encap_level = 0; 3098 3099 skb_reset_mac_header(skb); 3100 skb_reset_mac_len(skb); 3101 3102 segs = skb_mac_gso_segment(skb, features); 3103 3104 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3105 skb_warn_bad_offload(skb); 3106 3107 return segs; 3108 } 3109 EXPORT_SYMBOL(__skb_gso_segment); 3110 3111 /* Take action when hardware reception checksum errors are detected. */ 3112 #ifdef CONFIG_BUG 3113 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3114 { 3115 if (net_ratelimit()) { 3116 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3117 if (dev) 3118 pr_err("dev features: %pNF\n", &dev->features); 3119 pr_err("skb len=%u data_len=%u pkt_type=%u gso_size=%u gso_type=%u nr_frags=%u ip_summed=%u csum=%x csum_complete_sw=%d csum_valid=%d csum_level=%u\n", 3120 skb->len, skb->data_len, skb->pkt_type, 3121 skb_shinfo(skb)->gso_size, skb_shinfo(skb)->gso_type, 3122 skb_shinfo(skb)->nr_frags, skb->ip_summed, skb->csum, 3123 skb->csum_complete_sw, skb->csum_valid, skb->csum_level); 3124 dump_stack(); 3125 } 3126 } 3127 EXPORT_SYMBOL(netdev_rx_csum_fault); 3128 #endif 3129 3130 /* XXX: check that highmem exists at all on the given machine. */ 3131 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3132 { 3133 #ifdef CONFIG_HIGHMEM 3134 int i; 3135 3136 if (!(dev->features & NETIF_F_HIGHDMA)) { 3137 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3138 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3139 3140 if (PageHighMem(skb_frag_page(frag))) 3141 return 1; 3142 } 3143 } 3144 #endif 3145 return 0; 3146 } 3147 3148 /* If MPLS offload request, verify we are testing hardware MPLS features 3149 * instead of standard features for the netdev. 3150 */ 3151 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3152 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3153 netdev_features_t features, 3154 __be16 type) 3155 { 3156 if (eth_p_mpls(type)) 3157 features &= skb->dev->mpls_features; 3158 3159 return features; 3160 } 3161 #else 3162 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3163 netdev_features_t features, 3164 __be16 type) 3165 { 3166 return features; 3167 } 3168 #endif 3169 3170 static netdev_features_t harmonize_features(struct sk_buff *skb, 3171 netdev_features_t features) 3172 { 3173 int tmp; 3174 __be16 type; 3175 3176 type = skb_network_protocol(skb, &tmp); 3177 features = net_mpls_features(skb, features, type); 3178 3179 if (skb->ip_summed != CHECKSUM_NONE && 3180 !can_checksum_protocol(features, type)) { 3181 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3182 } 3183 if (illegal_highdma(skb->dev, skb)) 3184 features &= ~NETIF_F_SG; 3185 3186 return features; 3187 } 3188 3189 netdev_features_t passthru_features_check(struct sk_buff *skb, 3190 struct net_device *dev, 3191 netdev_features_t features) 3192 { 3193 return features; 3194 } 3195 EXPORT_SYMBOL(passthru_features_check); 3196 3197 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3198 struct net_device *dev, 3199 netdev_features_t features) 3200 { 3201 return vlan_features_check(skb, features); 3202 } 3203 3204 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3205 struct net_device *dev, 3206 netdev_features_t features) 3207 { 3208 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3209 3210 if (gso_segs > dev->gso_max_segs) 3211 return features & ~NETIF_F_GSO_MASK; 3212 3213 /* Support for GSO partial features requires software 3214 * intervention before we can actually process the packets 3215 * so we need to strip support for any partial features now 3216 * and we can pull them back in after we have partially 3217 * segmented the frame. 3218 */ 3219 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3220 features &= ~dev->gso_partial_features; 3221 3222 /* Make sure to clear the IPv4 ID mangling feature if the 3223 * IPv4 header has the potential to be fragmented. 3224 */ 3225 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3226 struct iphdr *iph = skb->encapsulation ? 3227 inner_ip_hdr(skb) : ip_hdr(skb); 3228 3229 if (!(iph->frag_off & htons(IP_DF))) 3230 features &= ~NETIF_F_TSO_MANGLEID; 3231 } 3232 3233 return features; 3234 } 3235 3236 netdev_features_t netif_skb_features(struct sk_buff *skb) 3237 { 3238 struct net_device *dev = skb->dev; 3239 netdev_features_t features = dev->features; 3240 3241 if (skb_is_gso(skb)) 3242 features = gso_features_check(skb, dev, features); 3243 3244 /* If encapsulation offload request, verify we are testing 3245 * hardware encapsulation features instead of standard 3246 * features for the netdev 3247 */ 3248 if (skb->encapsulation) 3249 features &= dev->hw_enc_features; 3250 3251 if (skb_vlan_tagged(skb)) 3252 features = netdev_intersect_features(features, 3253 dev->vlan_features | 3254 NETIF_F_HW_VLAN_CTAG_TX | 3255 NETIF_F_HW_VLAN_STAG_TX); 3256 3257 if (dev->netdev_ops->ndo_features_check) 3258 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3259 features); 3260 else 3261 features &= dflt_features_check(skb, dev, features); 3262 3263 return harmonize_features(skb, features); 3264 } 3265 EXPORT_SYMBOL(netif_skb_features); 3266 3267 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3268 struct netdev_queue *txq, bool more) 3269 { 3270 unsigned int len; 3271 int rc; 3272 3273 if (dev_nit_active(dev)) 3274 dev_queue_xmit_nit(skb, dev); 3275 3276 len = skb->len; 3277 trace_net_dev_start_xmit(skb, dev); 3278 rc = netdev_start_xmit(skb, dev, txq, more); 3279 trace_net_dev_xmit(skb, rc, dev, len); 3280 3281 return rc; 3282 } 3283 3284 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3285 struct netdev_queue *txq, int *ret) 3286 { 3287 struct sk_buff *skb = first; 3288 int rc = NETDEV_TX_OK; 3289 3290 while (skb) { 3291 struct sk_buff *next = skb->next; 3292 3293 skb_mark_not_on_list(skb); 3294 rc = xmit_one(skb, dev, txq, next != NULL); 3295 if (unlikely(!dev_xmit_complete(rc))) { 3296 skb->next = next; 3297 goto out; 3298 } 3299 3300 skb = next; 3301 if (netif_tx_queue_stopped(txq) && skb) { 3302 rc = NETDEV_TX_BUSY; 3303 break; 3304 } 3305 } 3306 3307 out: 3308 *ret = rc; 3309 return skb; 3310 } 3311 3312 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3313 netdev_features_t features) 3314 { 3315 if (skb_vlan_tag_present(skb) && 3316 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3317 skb = __vlan_hwaccel_push_inside(skb); 3318 return skb; 3319 } 3320 3321 int skb_csum_hwoffload_help(struct sk_buff *skb, 3322 const netdev_features_t features) 3323 { 3324 if (unlikely(skb->csum_not_inet)) 3325 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3326 skb_crc32c_csum_help(skb); 3327 3328 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); 3329 } 3330 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3331 3332 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3333 { 3334 netdev_features_t features; 3335 3336 features = netif_skb_features(skb); 3337 skb = validate_xmit_vlan(skb, features); 3338 if (unlikely(!skb)) 3339 goto out_null; 3340 3341 skb = sk_validate_xmit_skb(skb, dev); 3342 if (unlikely(!skb)) 3343 goto out_null; 3344 3345 if (netif_needs_gso(skb, features)) { 3346 struct sk_buff *segs; 3347 3348 segs = skb_gso_segment(skb, features); 3349 if (IS_ERR(segs)) { 3350 goto out_kfree_skb; 3351 } else if (segs) { 3352 consume_skb(skb); 3353 skb = segs; 3354 } 3355 } else { 3356 if (skb_needs_linearize(skb, features) && 3357 __skb_linearize(skb)) 3358 goto out_kfree_skb; 3359 3360 /* If packet is not checksummed and device does not 3361 * support checksumming for this protocol, complete 3362 * checksumming here. 3363 */ 3364 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3365 if (skb->encapsulation) 3366 skb_set_inner_transport_header(skb, 3367 skb_checksum_start_offset(skb)); 3368 else 3369 skb_set_transport_header(skb, 3370 skb_checksum_start_offset(skb)); 3371 if (skb_csum_hwoffload_help(skb, features)) 3372 goto out_kfree_skb; 3373 } 3374 } 3375 3376 skb = validate_xmit_xfrm(skb, features, again); 3377 3378 return skb; 3379 3380 out_kfree_skb: 3381 kfree_skb(skb); 3382 out_null: 3383 atomic_long_inc(&dev->tx_dropped); 3384 return NULL; 3385 } 3386 3387 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3388 { 3389 struct sk_buff *next, *head = NULL, *tail; 3390 3391 for (; skb != NULL; skb = next) { 3392 next = skb->next; 3393 skb_mark_not_on_list(skb); 3394 3395 /* in case skb wont be segmented, point to itself */ 3396 skb->prev = skb; 3397 3398 skb = validate_xmit_skb(skb, dev, again); 3399 if (!skb) 3400 continue; 3401 3402 if (!head) 3403 head = skb; 3404 else 3405 tail->next = skb; 3406 /* If skb was segmented, skb->prev points to 3407 * the last segment. If not, it still contains skb. 3408 */ 3409 tail = skb->prev; 3410 } 3411 return head; 3412 } 3413 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3414 3415 static void qdisc_pkt_len_init(struct sk_buff *skb) 3416 { 3417 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3418 3419 qdisc_skb_cb(skb)->pkt_len = skb->len; 3420 3421 /* To get more precise estimation of bytes sent on wire, 3422 * we add to pkt_len the headers size of all segments 3423 */ 3424 if (shinfo->gso_size) { 3425 unsigned int hdr_len; 3426 u16 gso_segs = shinfo->gso_segs; 3427 3428 /* mac layer + network layer */ 3429 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3430 3431 /* + transport layer */ 3432 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3433 const struct tcphdr *th; 3434 struct tcphdr _tcphdr; 3435 3436 th = skb_header_pointer(skb, skb_transport_offset(skb), 3437 sizeof(_tcphdr), &_tcphdr); 3438 if (likely(th)) 3439 hdr_len += __tcp_hdrlen(th); 3440 } else { 3441 struct udphdr _udphdr; 3442 3443 if (skb_header_pointer(skb, skb_transport_offset(skb), 3444 sizeof(_udphdr), &_udphdr)) 3445 hdr_len += sizeof(struct udphdr); 3446 } 3447 3448 if (shinfo->gso_type & SKB_GSO_DODGY) 3449 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3450 shinfo->gso_size); 3451 3452 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3453 } 3454 } 3455 3456 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3457 struct net_device *dev, 3458 struct netdev_queue *txq) 3459 { 3460 spinlock_t *root_lock = qdisc_lock(q); 3461 struct sk_buff *to_free = NULL; 3462 bool contended; 3463 int rc; 3464 3465 qdisc_calculate_pkt_len(skb, q); 3466 3467 if (q->flags & TCQ_F_NOLOCK) { 3468 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3469 __qdisc_drop(skb, &to_free); 3470 rc = NET_XMIT_DROP; 3471 } else { 3472 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3473 qdisc_run(q); 3474 } 3475 3476 if (unlikely(to_free)) 3477 kfree_skb_list(to_free); 3478 return rc; 3479 } 3480 3481 /* 3482 * Heuristic to force contended enqueues to serialize on a 3483 * separate lock before trying to get qdisc main lock. 3484 * This permits qdisc->running owner to get the lock more 3485 * often and dequeue packets faster. 3486 */ 3487 contended = qdisc_is_running(q); 3488 if (unlikely(contended)) 3489 spin_lock(&q->busylock); 3490 3491 spin_lock(root_lock); 3492 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3493 __qdisc_drop(skb, &to_free); 3494 rc = NET_XMIT_DROP; 3495 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3496 qdisc_run_begin(q)) { 3497 /* 3498 * This is a work-conserving queue; there are no old skbs 3499 * waiting to be sent out; and the qdisc is not running - 3500 * xmit the skb directly. 3501 */ 3502 3503 qdisc_bstats_update(q, skb); 3504 3505 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3506 if (unlikely(contended)) { 3507 spin_unlock(&q->busylock); 3508 contended = false; 3509 } 3510 __qdisc_run(q); 3511 } 3512 3513 qdisc_run_end(q); 3514 rc = NET_XMIT_SUCCESS; 3515 } else { 3516 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3517 if (qdisc_run_begin(q)) { 3518 if (unlikely(contended)) { 3519 spin_unlock(&q->busylock); 3520 contended = false; 3521 } 3522 __qdisc_run(q); 3523 qdisc_run_end(q); 3524 } 3525 } 3526 spin_unlock(root_lock); 3527 if (unlikely(to_free)) 3528 kfree_skb_list(to_free); 3529 if (unlikely(contended)) 3530 spin_unlock(&q->busylock); 3531 return rc; 3532 } 3533 3534 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3535 static void skb_update_prio(struct sk_buff *skb) 3536 { 3537 const struct netprio_map *map; 3538 const struct sock *sk; 3539 unsigned int prioidx; 3540 3541 if (skb->priority) 3542 return; 3543 map = rcu_dereference_bh(skb->dev->priomap); 3544 if (!map) 3545 return; 3546 sk = skb_to_full_sk(skb); 3547 if (!sk) 3548 return; 3549 3550 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3551 3552 if (prioidx < map->priomap_len) 3553 skb->priority = map->priomap[prioidx]; 3554 } 3555 #else 3556 #define skb_update_prio(skb) 3557 #endif 3558 3559 DEFINE_PER_CPU(int, xmit_recursion); 3560 EXPORT_SYMBOL(xmit_recursion); 3561 3562 /** 3563 * dev_loopback_xmit - loop back @skb 3564 * @net: network namespace this loopback is happening in 3565 * @sk: sk needed to be a netfilter okfn 3566 * @skb: buffer to transmit 3567 */ 3568 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3569 { 3570 skb_reset_mac_header(skb); 3571 __skb_pull(skb, skb_network_offset(skb)); 3572 skb->pkt_type = PACKET_LOOPBACK; 3573 skb->ip_summed = CHECKSUM_UNNECESSARY; 3574 WARN_ON(!skb_dst(skb)); 3575 skb_dst_force(skb); 3576 netif_rx_ni(skb); 3577 return 0; 3578 } 3579 EXPORT_SYMBOL(dev_loopback_xmit); 3580 3581 #ifdef CONFIG_NET_EGRESS 3582 static struct sk_buff * 3583 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3584 { 3585 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3586 struct tcf_result cl_res; 3587 3588 if (!miniq) 3589 return skb; 3590 3591 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3592 mini_qdisc_bstats_cpu_update(miniq, skb); 3593 3594 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3595 case TC_ACT_OK: 3596 case TC_ACT_RECLASSIFY: 3597 skb->tc_index = TC_H_MIN(cl_res.classid); 3598 break; 3599 case TC_ACT_SHOT: 3600 mini_qdisc_qstats_cpu_drop(miniq); 3601 *ret = NET_XMIT_DROP; 3602 kfree_skb(skb); 3603 return NULL; 3604 case TC_ACT_STOLEN: 3605 case TC_ACT_QUEUED: 3606 case TC_ACT_TRAP: 3607 *ret = NET_XMIT_SUCCESS; 3608 consume_skb(skb); 3609 return NULL; 3610 case TC_ACT_REDIRECT: 3611 /* No need to push/pop skb's mac_header here on egress! */ 3612 skb_do_redirect(skb); 3613 *ret = NET_XMIT_SUCCESS; 3614 return NULL; 3615 default: 3616 break; 3617 } 3618 3619 return skb; 3620 } 3621 #endif /* CONFIG_NET_EGRESS */ 3622 3623 #ifdef CONFIG_XPS 3624 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3625 struct xps_dev_maps *dev_maps, unsigned int tci) 3626 { 3627 struct xps_map *map; 3628 int queue_index = -1; 3629 3630 if (dev->num_tc) { 3631 tci *= dev->num_tc; 3632 tci += netdev_get_prio_tc_map(dev, skb->priority); 3633 } 3634 3635 map = rcu_dereference(dev_maps->attr_map[tci]); 3636 if (map) { 3637 if (map->len == 1) 3638 queue_index = map->queues[0]; 3639 else 3640 queue_index = map->queues[reciprocal_scale( 3641 skb_get_hash(skb), map->len)]; 3642 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3643 queue_index = -1; 3644 } 3645 return queue_index; 3646 } 3647 #endif 3648 3649 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3650 struct sk_buff *skb) 3651 { 3652 #ifdef CONFIG_XPS 3653 struct xps_dev_maps *dev_maps; 3654 struct sock *sk = skb->sk; 3655 int queue_index = -1; 3656 3657 if (!static_key_false(&xps_needed)) 3658 return -1; 3659 3660 rcu_read_lock(); 3661 if (!static_key_false(&xps_rxqs_needed)) 3662 goto get_cpus_map; 3663 3664 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); 3665 if (dev_maps) { 3666 int tci = sk_rx_queue_get(sk); 3667 3668 if (tci >= 0 && tci < dev->num_rx_queues) 3669 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3670 tci); 3671 } 3672 3673 get_cpus_map: 3674 if (queue_index < 0) { 3675 dev_maps = rcu_dereference(sb_dev->xps_cpus_map); 3676 if (dev_maps) { 3677 unsigned int tci = skb->sender_cpu - 1; 3678 3679 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3680 tci); 3681 } 3682 } 3683 rcu_read_unlock(); 3684 3685 return queue_index; 3686 #else 3687 return -1; 3688 #endif 3689 } 3690 3691 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3692 struct net_device *sb_dev, 3693 select_queue_fallback_t fallback) 3694 { 3695 return 0; 3696 } 3697 EXPORT_SYMBOL(dev_pick_tx_zero); 3698 3699 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 3700 struct net_device *sb_dev, 3701 select_queue_fallback_t fallback) 3702 { 3703 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 3704 } 3705 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 3706 3707 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 3708 struct net_device *sb_dev) 3709 { 3710 struct sock *sk = skb->sk; 3711 int queue_index = sk_tx_queue_get(sk); 3712 3713 sb_dev = sb_dev ? : dev; 3714 3715 if (queue_index < 0 || skb->ooo_okay || 3716 queue_index >= dev->real_num_tx_queues) { 3717 int new_index = get_xps_queue(dev, sb_dev, skb); 3718 3719 if (new_index < 0) 3720 new_index = skb_tx_hash(dev, sb_dev, skb); 3721 3722 if (queue_index != new_index && sk && 3723 sk_fullsock(sk) && 3724 rcu_access_pointer(sk->sk_dst_cache)) 3725 sk_tx_queue_set(sk, new_index); 3726 3727 queue_index = new_index; 3728 } 3729 3730 return queue_index; 3731 } 3732 3733 struct netdev_queue *netdev_pick_tx(struct net_device *dev, 3734 struct sk_buff *skb, 3735 struct net_device *sb_dev) 3736 { 3737 int queue_index = 0; 3738 3739 #ifdef CONFIG_XPS 3740 u32 sender_cpu = skb->sender_cpu - 1; 3741 3742 if (sender_cpu >= (u32)NR_CPUS) 3743 skb->sender_cpu = raw_smp_processor_id() + 1; 3744 #endif 3745 3746 if (dev->real_num_tx_queues != 1) { 3747 const struct net_device_ops *ops = dev->netdev_ops; 3748 3749 if (ops->ndo_select_queue) 3750 queue_index = ops->ndo_select_queue(dev, skb, sb_dev, 3751 __netdev_pick_tx); 3752 else 3753 queue_index = __netdev_pick_tx(dev, skb, sb_dev); 3754 3755 queue_index = netdev_cap_txqueue(dev, queue_index); 3756 } 3757 3758 skb_set_queue_mapping(skb, queue_index); 3759 return netdev_get_tx_queue(dev, queue_index); 3760 } 3761 3762 /** 3763 * __dev_queue_xmit - transmit a buffer 3764 * @skb: buffer to transmit 3765 * @sb_dev: suboordinate device used for L2 forwarding offload 3766 * 3767 * Queue a buffer for transmission to a network device. The caller must 3768 * have set the device and priority and built the buffer before calling 3769 * this function. The function can be called from an interrupt. 3770 * 3771 * A negative errno code is returned on a failure. A success does not 3772 * guarantee the frame will be transmitted as it may be dropped due 3773 * to congestion or traffic shaping. 3774 * 3775 * ----------------------------------------------------------------------------------- 3776 * I notice this method can also return errors from the queue disciplines, 3777 * including NET_XMIT_DROP, which is a positive value. So, errors can also 3778 * be positive. 3779 * 3780 * Regardless of the return value, the skb is consumed, so it is currently 3781 * difficult to retry a send to this method. (You can bump the ref count 3782 * before sending to hold a reference for retry if you are careful.) 3783 * 3784 * When calling this method, interrupts MUST be enabled. This is because 3785 * the BH enable code must have IRQs enabled so that it will not deadlock. 3786 * --BLG 3787 */ 3788 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 3789 { 3790 struct net_device *dev = skb->dev; 3791 struct netdev_queue *txq; 3792 struct Qdisc *q; 3793 int rc = -ENOMEM; 3794 bool again = false; 3795 3796 skb_reset_mac_header(skb); 3797 3798 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 3799 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 3800 3801 /* Disable soft irqs for various locks below. Also 3802 * stops preemption for RCU. 3803 */ 3804 rcu_read_lock_bh(); 3805 3806 skb_update_prio(skb); 3807 3808 qdisc_pkt_len_init(skb); 3809 #ifdef CONFIG_NET_CLS_ACT 3810 skb->tc_at_ingress = 0; 3811 # ifdef CONFIG_NET_EGRESS 3812 if (static_branch_unlikely(&egress_needed_key)) { 3813 skb = sch_handle_egress(skb, &rc, dev); 3814 if (!skb) 3815 goto out; 3816 } 3817 # endif 3818 #endif 3819 /* If device/qdisc don't need skb->dst, release it right now while 3820 * its hot in this cpu cache. 3821 */ 3822 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 3823 skb_dst_drop(skb); 3824 else 3825 skb_dst_force(skb); 3826 3827 txq = netdev_pick_tx(dev, skb, sb_dev); 3828 q = rcu_dereference_bh(txq->qdisc); 3829 3830 trace_net_dev_queue(skb); 3831 if (q->enqueue) { 3832 rc = __dev_xmit_skb(skb, q, dev, txq); 3833 goto out; 3834 } 3835 3836 /* The device has no queue. Common case for software devices: 3837 * loopback, all the sorts of tunnels... 3838 3839 * Really, it is unlikely that netif_tx_lock protection is necessary 3840 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 3841 * counters.) 3842 * However, it is possible, that they rely on protection 3843 * made by us here. 3844 3845 * Check this and shot the lock. It is not prone from deadlocks. 3846 *Either shot noqueue qdisc, it is even simpler 8) 3847 */ 3848 if (dev->flags & IFF_UP) { 3849 int cpu = smp_processor_id(); /* ok because BHs are off */ 3850 3851 if (txq->xmit_lock_owner != cpu) { 3852 if (unlikely(__this_cpu_read(xmit_recursion) > 3853 XMIT_RECURSION_LIMIT)) 3854 goto recursion_alert; 3855 3856 skb = validate_xmit_skb(skb, dev, &again); 3857 if (!skb) 3858 goto out; 3859 3860 HARD_TX_LOCK(dev, txq, cpu); 3861 3862 if (!netif_xmit_stopped(txq)) { 3863 __this_cpu_inc(xmit_recursion); 3864 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 3865 __this_cpu_dec(xmit_recursion); 3866 if (dev_xmit_complete(rc)) { 3867 HARD_TX_UNLOCK(dev, txq); 3868 goto out; 3869 } 3870 } 3871 HARD_TX_UNLOCK(dev, txq); 3872 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 3873 dev->name); 3874 } else { 3875 /* Recursion is detected! It is possible, 3876 * unfortunately 3877 */ 3878 recursion_alert: 3879 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 3880 dev->name); 3881 } 3882 } 3883 3884 rc = -ENETDOWN; 3885 rcu_read_unlock_bh(); 3886 3887 atomic_long_inc(&dev->tx_dropped); 3888 kfree_skb_list(skb); 3889 return rc; 3890 out: 3891 rcu_read_unlock_bh(); 3892 return rc; 3893 } 3894 3895 int dev_queue_xmit(struct sk_buff *skb) 3896 { 3897 return __dev_queue_xmit(skb, NULL); 3898 } 3899 EXPORT_SYMBOL(dev_queue_xmit); 3900 3901 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 3902 { 3903 return __dev_queue_xmit(skb, sb_dev); 3904 } 3905 EXPORT_SYMBOL(dev_queue_xmit_accel); 3906 3907 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 3908 { 3909 struct net_device *dev = skb->dev; 3910 struct sk_buff *orig_skb = skb; 3911 struct netdev_queue *txq; 3912 int ret = NETDEV_TX_BUSY; 3913 bool again = false; 3914 3915 if (unlikely(!netif_running(dev) || 3916 !netif_carrier_ok(dev))) 3917 goto drop; 3918 3919 skb = validate_xmit_skb_list(skb, dev, &again); 3920 if (skb != orig_skb) 3921 goto drop; 3922 3923 skb_set_queue_mapping(skb, queue_id); 3924 txq = skb_get_tx_queue(dev, skb); 3925 3926 local_bh_disable(); 3927 3928 HARD_TX_LOCK(dev, txq, smp_processor_id()); 3929 if (!netif_xmit_frozen_or_drv_stopped(txq)) 3930 ret = netdev_start_xmit(skb, dev, txq, false); 3931 HARD_TX_UNLOCK(dev, txq); 3932 3933 local_bh_enable(); 3934 3935 if (!dev_xmit_complete(ret)) 3936 kfree_skb(skb); 3937 3938 return ret; 3939 drop: 3940 atomic_long_inc(&dev->tx_dropped); 3941 kfree_skb_list(skb); 3942 return NET_XMIT_DROP; 3943 } 3944 EXPORT_SYMBOL(dev_direct_xmit); 3945 3946 /************************************************************************* 3947 * Receiver routines 3948 *************************************************************************/ 3949 3950 int netdev_max_backlog __read_mostly = 1000; 3951 EXPORT_SYMBOL(netdev_max_backlog); 3952 3953 int netdev_tstamp_prequeue __read_mostly = 1; 3954 int netdev_budget __read_mostly = 300; 3955 unsigned int __read_mostly netdev_budget_usecs = 2000; 3956 int weight_p __read_mostly = 64; /* old backlog weight */ 3957 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 3958 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 3959 int dev_rx_weight __read_mostly = 64; 3960 int dev_tx_weight __read_mostly = 64; 3961 3962 /* Called with irq disabled */ 3963 static inline void ____napi_schedule(struct softnet_data *sd, 3964 struct napi_struct *napi) 3965 { 3966 list_add_tail(&napi->poll_list, &sd->poll_list); 3967 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 3968 } 3969 3970 #ifdef CONFIG_RPS 3971 3972 /* One global table that all flow-based protocols share. */ 3973 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 3974 EXPORT_SYMBOL(rps_sock_flow_table); 3975 u32 rps_cpu_mask __read_mostly; 3976 EXPORT_SYMBOL(rps_cpu_mask); 3977 3978 struct static_key rps_needed __read_mostly; 3979 EXPORT_SYMBOL(rps_needed); 3980 struct static_key rfs_needed __read_mostly; 3981 EXPORT_SYMBOL(rfs_needed); 3982 3983 static struct rps_dev_flow * 3984 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 3985 struct rps_dev_flow *rflow, u16 next_cpu) 3986 { 3987 if (next_cpu < nr_cpu_ids) { 3988 #ifdef CONFIG_RFS_ACCEL 3989 struct netdev_rx_queue *rxqueue; 3990 struct rps_dev_flow_table *flow_table; 3991 struct rps_dev_flow *old_rflow; 3992 u32 flow_id; 3993 u16 rxq_index; 3994 int rc; 3995 3996 /* Should we steer this flow to a different hardware queue? */ 3997 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 3998 !(dev->features & NETIF_F_NTUPLE)) 3999 goto out; 4000 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4001 if (rxq_index == skb_get_rx_queue(skb)) 4002 goto out; 4003 4004 rxqueue = dev->_rx + rxq_index; 4005 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4006 if (!flow_table) 4007 goto out; 4008 flow_id = skb_get_hash(skb) & flow_table->mask; 4009 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4010 rxq_index, flow_id); 4011 if (rc < 0) 4012 goto out; 4013 old_rflow = rflow; 4014 rflow = &flow_table->flows[flow_id]; 4015 rflow->filter = rc; 4016 if (old_rflow->filter == rflow->filter) 4017 old_rflow->filter = RPS_NO_FILTER; 4018 out: 4019 #endif 4020 rflow->last_qtail = 4021 per_cpu(softnet_data, next_cpu).input_queue_head; 4022 } 4023 4024 rflow->cpu = next_cpu; 4025 return rflow; 4026 } 4027 4028 /* 4029 * get_rps_cpu is called from netif_receive_skb and returns the target 4030 * CPU from the RPS map of the receiving queue for a given skb. 4031 * rcu_read_lock must be held on entry. 4032 */ 4033 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4034 struct rps_dev_flow **rflowp) 4035 { 4036 const struct rps_sock_flow_table *sock_flow_table; 4037 struct netdev_rx_queue *rxqueue = dev->_rx; 4038 struct rps_dev_flow_table *flow_table; 4039 struct rps_map *map; 4040 int cpu = -1; 4041 u32 tcpu; 4042 u32 hash; 4043 4044 if (skb_rx_queue_recorded(skb)) { 4045 u16 index = skb_get_rx_queue(skb); 4046 4047 if (unlikely(index >= dev->real_num_rx_queues)) { 4048 WARN_ONCE(dev->real_num_rx_queues > 1, 4049 "%s received packet on queue %u, but number " 4050 "of RX queues is %u\n", 4051 dev->name, index, dev->real_num_rx_queues); 4052 goto done; 4053 } 4054 rxqueue += index; 4055 } 4056 4057 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4058 4059 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4060 map = rcu_dereference(rxqueue->rps_map); 4061 if (!flow_table && !map) 4062 goto done; 4063 4064 skb_reset_network_header(skb); 4065 hash = skb_get_hash(skb); 4066 if (!hash) 4067 goto done; 4068 4069 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4070 if (flow_table && sock_flow_table) { 4071 struct rps_dev_flow *rflow; 4072 u32 next_cpu; 4073 u32 ident; 4074 4075 /* First check into global flow table if there is a match */ 4076 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4077 if ((ident ^ hash) & ~rps_cpu_mask) 4078 goto try_rps; 4079 4080 next_cpu = ident & rps_cpu_mask; 4081 4082 /* OK, now we know there is a match, 4083 * we can look at the local (per receive queue) flow table 4084 */ 4085 rflow = &flow_table->flows[hash & flow_table->mask]; 4086 tcpu = rflow->cpu; 4087 4088 /* 4089 * If the desired CPU (where last recvmsg was done) is 4090 * different from current CPU (one in the rx-queue flow 4091 * table entry), switch if one of the following holds: 4092 * - Current CPU is unset (>= nr_cpu_ids). 4093 * - Current CPU is offline. 4094 * - The current CPU's queue tail has advanced beyond the 4095 * last packet that was enqueued using this table entry. 4096 * This guarantees that all previous packets for the flow 4097 * have been dequeued, thus preserving in order delivery. 4098 */ 4099 if (unlikely(tcpu != next_cpu) && 4100 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4101 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4102 rflow->last_qtail)) >= 0)) { 4103 tcpu = next_cpu; 4104 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4105 } 4106 4107 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4108 *rflowp = rflow; 4109 cpu = tcpu; 4110 goto done; 4111 } 4112 } 4113 4114 try_rps: 4115 4116 if (map) { 4117 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4118 if (cpu_online(tcpu)) { 4119 cpu = tcpu; 4120 goto done; 4121 } 4122 } 4123 4124 done: 4125 return cpu; 4126 } 4127 4128 #ifdef CONFIG_RFS_ACCEL 4129 4130 /** 4131 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4132 * @dev: Device on which the filter was set 4133 * @rxq_index: RX queue index 4134 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4135 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4136 * 4137 * Drivers that implement ndo_rx_flow_steer() should periodically call 4138 * this function for each installed filter and remove the filters for 4139 * which it returns %true. 4140 */ 4141 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4142 u32 flow_id, u16 filter_id) 4143 { 4144 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4145 struct rps_dev_flow_table *flow_table; 4146 struct rps_dev_flow *rflow; 4147 bool expire = true; 4148 unsigned int cpu; 4149 4150 rcu_read_lock(); 4151 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4152 if (flow_table && flow_id <= flow_table->mask) { 4153 rflow = &flow_table->flows[flow_id]; 4154 cpu = READ_ONCE(rflow->cpu); 4155 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4156 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4157 rflow->last_qtail) < 4158 (int)(10 * flow_table->mask))) 4159 expire = false; 4160 } 4161 rcu_read_unlock(); 4162 return expire; 4163 } 4164 EXPORT_SYMBOL(rps_may_expire_flow); 4165 4166 #endif /* CONFIG_RFS_ACCEL */ 4167 4168 /* Called from hardirq (IPI) context */ 4169 static void rps_trigger_softirq(void *data) 4170 { 4171 struct softnet_data *sd = data; 4172 4173 ____napi_schedule(sd, &sd->backlog); 4174 sd->received_rps++; 4175 } 4176 4177 #endif /* CONFIG_RPS */ 4178 4179 /* 4180 * Check if this softnet_data structure is another cpu one 4181 * If yes, queue it to our IPI list and return 1 4182 * If no, return 0 4183 */ 4184 static int rps_ipi_queued(struct softnet_data *sd) 4185 { 4186 #ifdef CONFIG_RPS 4187 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4188 4189 if (sd != mysd) { 4190 sd->rps_ipi_next = mysd->rps_ipi_list; 4191 mysd->rps_ipi_list = sd; 4192 4193 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4194 return 1; 4195 } 4196 #endif /* CONFIG_RPS */ 4197 return 0; 4198 } 4199 4200 #ifdef CONFIG_NET_FLOW_LIMIT 4201 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4202 #endif 4203 4204 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4205 { 4206 #ifdef CONFIG_NET_FLOW_LIMIT 4207 struct sd_flow_limit *fl; 4208 struct softnet_data *sd; 4209 unsigned int old_flow, new_flow; 4210 4211 if (qlen < (netdev_max_backlog >> 1)) 4212 return false; 4213 4214 sd = this_cpu_ptr(&softnet_data); 4215 4216 rcu_read_lock(); 4217 fl = rcu_dereference(sd->flow_limit); 4218 if (fl) { 4219 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4220 old_flow = fl->history[fl->history_head]; 4221 fl->history[fl->history_head] = new_flow; 4222 4223 fl->history_head++; 4224 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4225 4226 if (likely(fl->buckets[old_flow])) 4227 fl->buckets[old_flow]--; 4228 4229 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4230 fl->count++; 4231 rcu_read_unlock(); 4232 return true; 4233 } 4234 } 4235 rcu_read_unlock(); 4236 #endif 4237 return false; 4238 } 4239 4240 /* 4241 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4242 * queue (may be a remote CPU queue). 4243 */ 4244 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4245 unsigned int *qtail) 4246 { 4247 struct softnet_data *sd; 4248 unsigned long flags; 4249 unsigned int qlen; 4250 4251 sd = &per_cpu(softnet_data, cpu); 4252 4253 local_irq_save(flags); 4254 4255 rps_lock(sd); 4256 if (!netif_running(skb->dev)) 4257 goto drop; 4258 qlen = skb_queue_len(&sd->input_pkt_queue); 4259 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4260 if (qlen) { 4261 enqueue: 4262 __skb_queue_tail(&sd->input_pkt_queue, skb); 4263 input_queue_tail_incr_save(sd, qtail); 4264 rps_unlock(sd); 4265 local_irq_restore(flags); 4266 return NET_RX_SUCCESS; 4267 } 4268 4269 /* Schedule NAPI for backlog device 4270 * We can use non atomic operation since we own the queue lock 4271 */ 4272 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4273 if (!rps_ipi_queued(sd)) 4274 ____napi_schedule(sd, &sd->backlog); 4275 } 4276 goto enqueue; 4277 } 4278 4279 drop: 4280 sd->dropped++; 4281 rps_unlock(sd); 4282 4283 local_irq_restore(flags); 4284 4285 atomic_long_inc(&skb->dev->rx_dropped); 4286 kfree_skb(skb); 4287 return NET_RX_DROP; 4288 } 4289 4290 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4291 { 4292 struct net_device *dev = skb->dev; 4293 struct netdev_rx_queue *rxqueue; 4294 4295 rxqueue = dev->_rx; 4296 4297 if (skb_rx_queue_recorded(skb)) { 4298 u16 index = skb_get_rx_queue(skb); 4299 4300 if (unlikely(index >= dev->real_num_rx_queues)) { 4301 WARN_ONCE(dev->real_num_rx_queues > 1, 4302 "%s received packet on queue %u, but number " 4303 "of RX queues is %u\n", 4304 dev->name, index, dev->real_num_rx_queues); 4305 4306 return rxqueue; /* Return first rxqueue */ 4307 } 4308 rxqueue += index; 4309 } 4310 return rxqueue; 4311 } 4312 4313 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4314 struct xdp_buff *xdp, 4315 struct bpf_prog *xdp_prog) 4316 { 4317 struct netdev_rx_queue *rxqueue; 4318 void *orig_data, *orig_data_end; 4319 u32 metalen, act = XDP_DROP; 4320 __be16 orig_eth_type; 4321 struct ethhdr *eth; 4322 bool orig_bcast; 4323 int hlen, off; 4324 u32 mac_len; 4325 4326 /* Reinjected packets coming from act_mirred or similar should 4327 * not get XDP generic processing. 4328 */ 4329 if (skb_cloned(skb) || skb_is_tc_redirected(skb)) 4330 return XDP_PASS; 4331 4332 /* XDP packets must be linear and must have sufficient headroom 4333 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4334 * native XDP provides, thus we need to do it here as well. 4335 */ 4336 if (skb_is_nonlinear(skb) || 4337 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4338 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4339 int troom = skb->tail + skb->data_len - skb->end; 4340 4341 /* In case we have to go down the path and also linearize, 4342 * then lets do the pskb_expand_head() work just once here. 4343 */ 4344 if (pskb_expand_head(skb, 4345 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4346 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4347 goto do_drop; 4348 if (skb_linearize(skb)) 4349 goto do_drop; 4350 } 4351 4352 /* The XDP program wants to see the packet starting at the MAC 4353 * header. 4354 */ 4355 mac_len = skb->data - skb_mac_header(skb); 4356 hlen = skb_headlen(skb) + mac_len; 4357 xdp->data = skb->data - mac_len; 4358 xdp->data_meta = xdp->data; 4359 xdp->data_end = xdp->data + hlen; 4360 xdp->data_hard_start = skb->data - skb_headroom(skb); 4361 orig_data_end = xdp->data_end; 4362 orig_data = xdp->data; 4363 eth = (struct ethhdr *)xdp->data; 4364 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4365 orig_eth_type = eth->h_proto; 4366 4367 rxqueue = netif_get_rxqueue(skb); 4368 xdp->rxq = &rxqueue->xdp_rxq; 4369 4370 act = bpf_prog_run_xdp(xdp_prog, xdp); 4371 4372 off = xdp->data - orig_data; 4373 if (off > 0) 4374 __skb_pull(skb, off); 4375 else if (off < 0) 4376 __skb_push(skb, -off); 4377 skb->mac_header += off; 4378 4379 /* check if bpf_xdp_adjust_tail was used. it can only "shrink" 4380 * pckt. 4381 */ 4382 off = orig_data_end - xdp->data_end; 4383 if (off != 0) { 4384 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4385 skb->len -= off; 4386 4387 } 4388 4389 /* check if XDP changed eth hdr such SKB needs update */ 4390 eth = (struct ethhdr *)xdp->data; 4391 if ((orig_eth_type != eth->h_proto) || 4392 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4393 __skb_push(skb, ETH_HLEN); 4394 skb->protocol = eth_type_trans(skb, skb->dev); 4395 } 4396 4397 switch (act) { 4398 case XDP_REDIRECT: 4399 case XDP_TX: 4400 __skb_push(skb, mac_len); 4401 break; 4402 case XDP_PASS: 4403 metalen = xdp->data - xdp->data_meta; 4404 if (metalen) 4405 skb_metadata_set(skb, metalen); 4406 break; 4407 default: 4408 bpf_warn_invalid_xdp_action(act); 4409 /* fall through */ 4410 case XDP_ABORTED: 4411 trace_xdp_exception(skb->dev, xdp_prog, act); 4412 /* fall through */ 4413 case XDP_DROP: 4414 do_drop: 4415 kfree_skb(skb); 4416 break; 4417 } 4418 4419 return act; 4420 } 4421 4422 /* When doing generic XDP we have to bypass the qdisc layer and the 4423 * network taps in order to match in-driver-XDP behavior. 4424 */ 4425 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4426 { 4427 struct net_device *dev = skb->dev; 4428 struct netdev_queue *txq; 4429 bool free_skb = true; 4430 int cpu, rc; 4431 4432 txq = netdev_pick_tx(dev, skb, NULL); 4433 cpu = smp_processor_id(); 4434 HARD_TX_LOCK(dev, txq, cpu); 4435 if (!netif_xmit_stopped(txq)) { 4436 rc = netdev_start_xmit(skb, dev, txq, 0); 4437 if (dev_xmit_complete(rc)) 4438 free_skb = false; 4439 } 4440 HARD_TX_UNLOCK(dev, txq); 4441 if (free_skb) { 4442 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4443 kfree_skb(skb); 4444 } 4445 } 4446 EXPORT_SYMBOL_GPL(generic_xdp_tx); 4447 4448 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4449 4450 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4451 { 4452 if (xdp_prog) { 4453 struct xdp_buff xdp; 4454 u32 act; 4455 int err; 4456 4457 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4458 if (act != XDP_PASS) { 4459 switch (act) { 4460 case XDP_REDIRECT: 4461 err = xdp_do_generic_redirect(skb->dev, skb, 4462 &xdp, xdp_prog); 4463 if (err) 4464 goto out_redir; 4465 break; 4466 case XDP_TX: 4467 generic_xdp_tx(skb, xdp_prog); 4468 break; 4469 } 4470 return XDP_DROP; 4471 } 4472 } 4473 return XDP_PASS; 4474 out_redir: 4475 kfree_skb(skb); 4476 return XDP_DROP; 4477 } 4478 EXPORT_SYMBOL_GPL(do_xdp_generic); 4479 4480 static int netif_rx_internal(struct sk_buff *skb) 4481 { 4482 int ret; 4483 4484 net_timestamp_check(netdev_tstamp_prequeue, skb); 4485 4486 trace_netif_rx(skb); 4487 4488 if (static_branch_unlikely(&generic_xdp_needed_key)) { 4489 int ret; 4490 4491 preempt_disable(); 4492 rcu_read_lock(); 4493 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 4494 rcu_read_unlock(); 4495 preempt_enable(); 4496 4497 /* Consider XDP consuming the packet a success from 4498 * the netdev point of view we do not want to count 4499 * this as an error. 4500 */ 4501 if (ret != XDP_PASS) 4502 return NET_RX_SUCCESS; 4503 } 4504 4505 #ifdef CONFIG_RPS 4506 if (static_key_false(&rps_needed)) { 4507 struct rps_dev_flow voidflow, *rflow = &voidflow; 4508 int cpu; 4509 4510 preempt_disable(); 4511 rcu_read_lock(); 4512 4513 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4514 if (cpu < 0) 4515 cpu = smp_processor_id(); 4516 4517 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4518 4519 rcu_read_unlock(); 4520 preempt_enable(); 4521 } else 4522 #endif 4523 { 4524 unsigned int qtail; 4525 4526 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4527 put_cpu(); 4528 } 4529 return ret; 4530 } 4531 4532 /** 4533 * netif_rx - post buffer to the network code 4534 * @skb: buffer to post 4535 * 4536 * This function receives a packet from a device driver and queues it for 4537 * the upper (protocol) levels to process. It always succeeds. The buffer 4538 * may be dropped during processing for congestion control or by the 4539 * protocol layers. 4540 * 4541 * return values: 4542 * NET_RX_SUCCESS (no congestion) 4543 * NET_RX_DROP (packet was dropped) 4544 * 4545 */ 4546 4547 int netif_rx(struct sk_buff *skb) 4548 { 4549 int ret; 4550 4551 trace_netif_rx_entry(skb); 4552 4553 ret = netif_rx_internal(skb); 4554 trace_netif_rx_exit(ret); 4555 4556 return ret; 4557 } 4558 EXPORT_SYMBOL(netif_rx); 4559 4560 int netif_rx_ni(struct sk_buff *skb) 4561 { 4562 int err; 4563 4564 trace_netif_rx_ni_entry(skb); 4565 4566 preempt_disable(); 4567 err = netif_rx_internal(skb); 4568 if (local_softirq_pending()) 4569 do_softirq(); 4570 preempt_enable(); 4571 trace_netif_rx_ni_exit(err); 4572 4573 return err; 4574 } 4575 EXPORT_SYMBOL(netif_rx_ni); 4576 4577 static __latent_entropy void net_tx_action(struct softirq_action *h) 4578 { 4579 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4580 4581 if (sd->completion_queue) { 4582 struct sk_buff *clist; 4583 4584 local_irq_disable(); 4585 clist = sd->completion_queue; 4586 sd->completion_queue = NULL; 4587 local_irq_enable(); 4588 4589 while (clist) { 4590 struct sk_buff *skb = clist; 4591 4592 clist = clist->next; 4593 4594 WARN_ON(refcount_read(&skb->users)); 4595 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4596 trace_consume_skb(skb); 4597 else 4598 trace_kfree_skb(skb, net_tx_action); 4599 4600 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4601 __kfree_skb(skb); 4602 else 4603 __kfree_skb_defer(skb); 4604 } 4605 4606 __kfree_skb_flush(); 4607 } 4608 4609 if (sd->output_queue) { 4610 struct Qdisc *head; 4611 4612 local_irq_disable(); 4613 head = sd->output_queue; 4614 sd->output_queue = NULL; 4615 sd->output_queue_tailp = &sd->output_queue; 4616 local_irq_enable(); 4617 4618 while (head) { 4619 struct Qdisc *q = head; 4620 spinlock_t *root_lock = NULL; 4621 4622 head = head->next_sched; 4623 4624 if (!(q->flags & TCQ_F_NOLOCK)) { 4625 root_lock = qdisc_lock(q); 4626 spin_lock(root_lock); 4627 } 4628 /* We need to make sure head->next_sched is read 4629 * before clearing __QDISC_STATE_SCHED 4630 */ 4631 smp_mb__before_atomic(); 4632 clear_bit(__QDISC_STATE_SCHED, &q->state); 4633 qdisc_run(q); 4634 if (root_lock) 4635 spin_unlock(root_lock); 4636 } 4637 } 4638 4639 xfrm_dev_backlog(sd); 4640 } 4641 4642 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4643 /* This hook is defined here for ATM LANE */ 4644 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4645 unsigned char *addr) __read_mostly; 4646 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4647 #endif 4648 4649 static inline struct sk_buff * 4650 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4651 struct net_device *orig_dev) 4652 { 4653 #ifdef CONFIG_NET_CLS_ACT 4654 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4655 struct tcf_result cl_res; 4656 4657 /* If there's at least one ingress present somewhere (so 4658 * we get here via enabled static key), remaining devices 4659 * that are not configured with an ingress qdisc will bail 4660 * out here. 4661 */ 4662 if (!miniq) 4663 return skb; 4664 4665 if (*pt_prev) { 4666 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4667 *pt_prev = NULL; 4668 } 4669 4670 qdisc_skb_cb(skb)->pkt_len = skb->len; 4671 skb->tc_at_ingress = 1; 4672 mini_qdisc_bstats_cpu_update(miniq, skb); 4673 4674 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 4675 case TC_ACT_OK: 4676 case TC_ACT_RECLASSIFY: 4677 skb->tc_index = TC_H_MIN(cl_res.classid); 4678 break; 4679 case TC_ACT_SHOT: 4680 mini_qdisc_qstats_cpu_drop(miniq); 4681 kfree_skb(skb); 4682 return NULL; 4683 case TC_ACT_STOLEN: 4684 case TC_ACT_QUEUED: 4685 case TC_ACT_TRAP: 4686 consume_skb(skb); 4687 return NULL; 4688 case TC_ACT_REDIRECT: 4689 /* skb_mac_header check was done by cls/act_bpf, so 4690 * we can safely push the L2 header back before 4691 * redirecting to another netdev 4692 */ 4693 __skb_push(skb, skb->mac_len); 4694 skb_do_redirect(skb); 4695 return NULL; 4696 case TC_ACT_REINSERT: 4697 /* this does not scrub the packet, and updates stats on error */ 4698 skb_tc_reinsert(skb, &cl_res); 4699 return NULL; 4700 default: 4701 break; 4702 } 4703 #endif /* CONFIG_NET_CLS_ACT */ 4704 return skb; 4705 } 4706 4707 /** 4708 * netdev_is_rx_handler_busy - check if receive handler is registered 4709 * @dev: device to check 4710 * 4711 * Check if a receive handler is already registered for a given device. 4712 * Return true if there one. 4713 * 4714 * The caller must hold the rtnl_mutex. 4715 */ 4716 bool netdev_is_rx_handler_busy(struct net_device *dev) 4717 { 4718 ASSERT_RTNL(); 4719 return dev && rtnl_dereference(dev->rx_handler); 4720 } 4721 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 4722 4723 /** 4724 * netdev_rx_handler_register - register receive handler 4725 * @dev: device to register a handler for 4726 * @rx_handler: receive handler to register 4727 * @rx_handler_data: data pointer that is used by rx handler 4728 * 4729 * Register a receive handler for a device. This handler will then be 4730 * called from __netif_receive_skb. A negative errno code is returned 4731 * on a failure. 4732 * 4733 * The caller must hold the rtnl_mutex. 4734 * 4735 * For a general description of rx_handler, see enum rx_handler_result. 4736 */ 4737 int netdev_rx_handler_register(struct net_device *dev, 4738 rx_handler_func_t *rx_handler, 4739 void *rx_handler_data) 4740 { 4741 if (netdev_is_rx_handler_busy(dev)) 4742 return -EBUSY; 4743 4744 if (dev->priv_flags & IFF_NO_RX_HANDLER) 4745 return -EINVAL; 4746 4747 /* Note: rx_handler_data must be set before rx_handler */ 4748 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 4749 rcu_assign_pointer(dev->rx_handler, rx_handler); 4750 4751 return 0; 4752 } 4753 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 4754 4755 /** 4756 * netdev_rx_handler_unregister - unregister receive handler 4757 * @dev: device to unregister a handler from 4758 * 4759 * Unregister a receive handler from a device. 4760 * 4761 * The caller must hold the rtnl_mutex. 4762 */ 4763 void netdev_rx_handler_unregister(struct net_device *dev) 4764 { 4765 4766 ASSERT_RTNL(); 4767 RCU_INIT_POINTER(dev->rx_handler, NULL); 4768 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 4769 * section has a guarantee to see a non NULL rx_handler_data 4770 * as well. 4771 */ 4772 synchronize_net(); 4773 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 4774 } 4775 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 4776 4777 /* 4778 * Limit the use of PFMEMALLOC reserves to those protocols that implement 4779 * the special handling of PFMEMALLOC skbs. 4780 */ 4781 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 4782 { 4783 switch (skb->protocol) { 4784 case htons(ETH_P_ARP): 4785 case htons(ETH_P_IP): 4786 case htons(ETH_P_IPV6): 4787 case htons(ETH_P_8021Q): 4788 case htons(ETH_P_8021AD): 4789 return true; 4790 default: 4791 return false; 4792 } 4793 } 4794 4795 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 4796 int *ret, struct net_device *orig_dev) 4797 { 4798 #ifdef CONFIG_NETFILTER_INGRESS 4799 if (nf_hook_ingress_active(skb)) { 4800 int ingress_retval; 4801 4802 if (*pt_prev) { 4803 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4804 *pt_prev = NULL; 4805 } 4806 4807 rcu_read_lock(); 4808 ingress_retval = nf_hook_ingress(skb); 4809 rcu_read_unlock(); 4810 return ingress_retval; 4811 } 4812 #endif /* CONFIG_NETFILTER_INGRESS */ 4813 return 0; 4814 } 4815 4816 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, 4817 struct packet_type **ppt_prev) 4818 { 4819 struct packet_type *ptype, *pt_prev; 4820 rx_handler_func_t *rx_handler; 4821 struct net_device *orig_dev; 4822 bool deliver_exact = false; 4823 int ret = NET_RX_DROP; 4824 __be16 type; 4825 4826 net_timestamp_check(!netdev_tstamp_prequeue, skb); 4827 4828 trace_netif_receive_skb(skb); 4829 4830 orig_dev = skb->dev; 4831 4832 skb_reset_network_header(skb); 4833 if (!skb_transport_header_was_set(skb)) 4834 skb_reset_transport_header(skb); 4835 skb_reset_mac_len(skb); 4836 4837 pt_prev = NULL; 4838 4839 another_round: 4840 skb->skb_iif = skb->dev->ifindex; 4841 4842 __this_cpu_inc(softnet_data.processed); 4843 4844 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 4845 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 4846 skb = skb_vlan_untag(skb); 4847 if (unlikely(!skb)) 4848 goto out; 4849 } 4850 4851 if (skb_skip_tc_classify(skb)) 4852 goto skip_classify; 4853 4854 if (pfmemalloc) 4855 goto skip_taps; 4856 4857 list_for_each_entry_rcu(ptype, &ptype_all, list) { 4858 if (pt_prev) 4859 ret = deliver_skb(skb, pt_prev, orig_dev); 4860 pt_prev = ptype; 4861 } 4862 4863 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 4864 if (pt_prev) 4865 ret = deliver_skb(skb, pt_prev, orig_dev); 4866 pt_prev = ptype; 4867 } 4868 4869 skip_taps: 4870 #ifdef CONFIG_NET_INGRESS 4871 if (static_branch_unlikely(&ingress_needed_key)) { 4872 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); 4873 if (!skb) 4874 goto out; 4875 4876 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 4877 goto out; 4878 } 4879 #endif 4880 skb_reset_tc(skb); 4881 skip_classify: 4882 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 4883 goto drop; 4884 4885 if (skb_vlan_tag_present(skb)) { 4886 if (pt_prev) { 4887 ret = deliver_skb(skb, pt_prev, orig_dev); 4888 pt_prev = NULL; 4889 } 4890 if (vlan_do_receive(&skb)) 4891 goto another_round; 4892 else if (unlikely(!skb)) 4893 goto out; 4894 } 4895 4896 rx_handler = rcu_dereference(skb->dev->rx_handler); 4897 if (rx_handler) { 4898 if (pt_prev) { 4899 ret = deliver_skb(skb, pt_prev, orig_dev); 4900 pt_prev = NULL; 4901 } 4902 switch (rx_handler(&skb)) { 4903 case RX_HANDLER_CONSUMED: 4904 ret = NET_RX_SUCCESS; 4905 goto out; 4906 case RX_HANDLER_ANOTHER: 4907 goto another_round; 4908 case RX_HANDLER_EXACT: 4909 deliver_exact = true; 4910 case RX_HANDLER_PASS: 4911 break; 4912 default: 4913 BUG(); 4914 } 4915 } 4916 4917 if (unlikely(skb_vlan_tag_present(skb))) { 4918 if (skb_vlan_tag_get_id(skb)) 4919 skb->pkt_type = PACKET_OTHERHOST; 4920 /* Note: we might in the future use prio bits 4921 * and set skb->priority like in vlan_do_receive() 4922 * For the time being, just ignore Priority Code Point 4923 */ 4924 __vlan_hwaccel_clear_tag(skb); 4925 } 4926 4927 type = skb->protocol; 4928 4929 /* deliver only exact match when indicated */ 4930 if (likely(!deliver_exact)) { 4931 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4932 &ptype_base[ntohs(type) & 4933 PTYPE_HASH_MASK]); 4934 } 4935 4936 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4937 &orig_dev->ptype_specific); 4938 4939 if (unlikely(skb->dev != orig_dev)) { 4940 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 4941 &skb->dev->ptype_specific); 4942 } 4943 4944 if (pt_prev) { 4945 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 4946 goto drop; 4947 *ppt_prev = pt_prev; 4948 } else { 4949 drop: 4950 if (!deliver_exact) 4951 atomic_long_inc(&skb->dev->rx_dropped); 4952 else 4953 atomic_long_inc(&skb->dev->rx_nohandler); 4954 kfree_skb(skb); 4955 /* Jamal, now you will not able to escape explaining 4956 * me how you were going to use this. :-) 4957 */ 4958 ret = NET_RX_DROP; 4959 } 4960 4961 out: 4962 return ret; 4963 } 4964 4965 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 4966 { 4967 struct net_device *orig_dev = skb->dev; 4968 struct packet_type *pt_prev = NULL; 4969 int ret; 4970 4971 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); 4972 if (pt_prev) 4973 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 4974 return ret; 4975 } 4976 4977 /** 4978 * netif_receive_skb_core - special purpose version of netif_receive_skb 4979 * @skb: buffer to process 4980 * 4981 * More direct receive version of netif_receive_skb(). It should 4982 * only be used by callers that have a need to skip RPS and Generic XDP. 4983 * Caller must also take care of handling if (page_is_)pfmemalloc. 4984 * 4985 * This function may only be called from softirq context and interrupts 4986 * should be enabled. 4987 * 4988 * Return values (usually ignored): 4989 * NET_RX_SUCCESS: no congestion 4990 * NET_RX_DROP: packet was dropped 4991 */ 4992 int netif_receive_skb_core(struct sk_buff *skb) 4993 { 4994 int ret; 4995 4996 rcu_read_lock(); 4997 ret = __netif_receive_skb_one_core(skb, false); 4998 rcu_read_unlock(); 4999 5000 return ret; 5001 } 5002 EXPORT_SYMBOL(netif_receive_skb_core); 5003 5004 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5005 struct packet_type *pt_prev, 5006 struct net_device *orig_dev) 5007 { 5008 struct sk_buff *skb, *next; 5009 5010 if (!pt_prev) 5011 return; 5012 if (list_empty(head)) 5013 return; 5014 if (pt_prev->list_func != NULL) 5015 pt_prev->list_func(head, pt_prev, orig_dev); 5016 else 5017 list_for_each_entry_safe(skb, next, head, list) 5018 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5019 } 5020 5021 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5022 { 5023 /* Fast-path assumptions: 5024 * - There is no RX handler. 5025 * - Only one packet_type matches. 5026 * If either of these fails, we will end up doing some per-packet 5027 * processing in-line, then handling the 'last ptype' for the whole 5028 * sublist. This can't cause out-of-order delivery to any single ptype, 5029 * because the 'last ptype' must be constant across the sublist, and all 5030 * other ptypes are handled per-packet. 5031 */ 5032 /* Current (common) ptype of sublist */ 5033 struct packet_type *pt_curr = NULL; 5034 /* Current (common) orig_dev of sublist */ 5035 struct net_device *od_curr = NULL; 5036 struct list_head sublist; 5037 struct sk_buff *skb, *next; 5038 5039 INIT_LIST_HEAD(&sublist); 5040 list_for_each_entry_safe(skb, next, head, list) { 5041 struct net_device *orig_dev = skb->dev; 5042 struct packet_type *pt_prev = NULL; 5043 5044 skb_list_del_init(skb); 5045 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); 5046 if (!pt_prev) 5047 continue; 5048 if (pt_curr != pt_prev || od_curr != orig_dev) { 5049 /* dispatch old sublist */ 5050 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5051 /* start new sublist */ 5052 INIT_LIST_HEAD(&sublist); 5053 pt_curr = pt_prev; 5054 od_curr = orig_dev; 5055 } 5056 list_add_tail(&skb->list, &sublist); 5057 } 5058 5059 /* dispatch final sublist */ 5060 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5061 } 5062 5063 static int __netif_receive_skb(struct sk_buff *skb) 5064 { 5065 int ret; 5066 5067 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5068 unsigned int noreclaim_flag; 5069 5070 /* 5071 * PFMEMALLOC skbs are special, they should 5072 * - be delivered to SOCK_MEMALLOC sockets only 5073 * - stay away from userspace 5074 * - have bounded memory usage 5075 * 5076 * Use PF_MEMALLOC as this saves us from propagating the allocation 5077 * context down to all allocation sites. 5078 */ 5079 noreclaim_flag = memalloc_noreclaim_save(); 5080 ret = __netif_receive_skb_one_core(skb, true); 5081 memalloc_noreclaim_restore(noreclaim_flag); 5082 } else 5083 ret = __netif_receive_skb_one_core(skb, false); 5084 5085 return ret; 5086 } 5087 5088 static void __netif_receive_skb_list(struct list_head *head) 5089 { 5090 unsigned long noreclaim_flag = 0; 5091 struct sk_buff *skb, *next; 5092 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5093 5094 list_for_each_entry_safe(skb, next, head, list) { 5095 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5096 struct list_head sublist; 5097 5098 /* Handle the previous sublist */ 5099 list_cut_before(&sublist, head, &skb->list); 5100 if (!list_empty(&sublist)) 5101 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5102 pfmemalloc = !pfmemalloc; 5103 /* See comments in __netif_receive_skb */ 5104 if (pfmemalloc) 5105 noreclaim_flag = memalloc_noreclaim_save(); 5106 else 5107 memalloc_noreclaim_restore(noreclaim_flag); 5108 } 5109 } 5110 /* Handle the remaining sublist */ 5111 if (!list_empty(head)) 5112 __netif_receive_skb_list_core(head, pfmemalloc); 5113 /* Restore pflags */ 5114 if (pfmemalloc) 5115 memalloc_noreclaim_restore(noreclaim_flag); 5116 } 5117 5118 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5119 { 5120 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5121 struct bpf_prog *new = xdp->prog; 5122 int ret = 0; 5123 5124 switch (xdp->command) { 5125 case XDP_SETUP_PROG: 5126 rcu_assign_pointer(dev->xdp_prog, new); 5127 if (old) 5128 bpf_prog_put(old); 5129 5130 if (old && !new) { 5131 static_branch_dec(&generic_xdp_needed_key); 5132 } else if (new && !old) { 5133 static_branch_inc(&generic_xdp_needed_key); 5134 dev_disable_lro(dev); 5135 dev_disable_gro_hw(dev); 5136 } 5137 break; 5138 5139 case XDP_QUERY_PROG: 5140 xdp->prog_id = old ? old->aux->id : 0; 5141 break; 5142 5143 default: 5144 ret = -EINVAL; 5145 break; 5146 } 5147 5148 return ret; 5149 } 5150 5151 static int netif_receive_skb_internal(struct sk_buff *skb) 5152 { 5153 int ret; 5154 5155 net_timestamp_check(netdev_tstamp_prequeue, skb); 5156 5157 if (skb_defer_rx_timestamp(skb)) 5158 return NET_RX_SUCCESS; 5159 5160 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5161 int ret; 5162 5163 preempt_disable(); 5164 rcu_read_lock(); 5165 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5166 rcu_read_unlock(); 5167 preempt_enable(); 5168 5169 if (ret != XDP_PASS) 5170 return NET_RX_DROP; 5171 } 5172 5173 rcu_read_lock(); 5174 #ifdef CONFIG_RPS 5175 if (static_key_false(&rps_needed)) { 5176 struct rps_dev_flow voidflow, *rflow = &voidflow; 5177 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5178 5179 if (cpu >= 0) { 5180 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5181 rcu_read_unlock(); 5182 return ret; 5183 } 5184 } 5185 #endif 5186 ret = __netif_receive_skb(skb); 5187 rcu_read_unlock(); 5188 return ret; 5189 } 5190 5191 static void netif_receive_skb_list_internal(struct list_head *head) 5192 { 5193 struct bpf_prog *xdp_prog = NULL; 5194 struct sk_buff *skb, *next; 5195 struct list_head sublist; 5196 5197 INIT_LIST_HEAD(&sublist); 5198 list_for_each_entry_safe(skb, next, head, list) { 5199 net_timestamp_check(netdev_tstamp_prequeue, skb); 5200 skb_list_del_init(skb); 5201 if (!skb_defer_rx_timestamp(skb)) 5202 list_add_tail(&skb->list, &sublist); 5203 } 5204 list_splice_init(&sublist, head); 5205 5206 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5207 preempt_disable(); 5208 rcu_read_lock(); 5209 list_for_each_entry_safe(skb, next, head, list) { 5210 xdp_prog = rcu_dereference(skb->dev->xdp_prog); 5211 skb_list_del_init(skb); 5212 if (do_xdp_generic(xdp_prog, skb) == XDP_PASS) 5213 list_add_tail(&skb->list, &sublist); 5214 } 5215 rcu_read_unlock(); 5216 preempt_enable(); 5217 /* Put passed packets back on main list */ 5218 list_splice_init(&sublist, head); 5219 } 5220 5221 rcu_read_lock(); 5222 #ifdef CONFIG_RPS 5223 if (static_key_false(&rps_needed)) { 5224 list_for_each_entry_safe(skb, next, head, list) { 5225 struct rps_dev_flow voidflow, *rflow = &voidflow; 5226 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5227 5228 if (cpu >= 0) { 5229 /* Will be handled, remove from list */ 5230 skb_list_del_init(skb); 5231 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5232 } 5233 } 5234 } 5235 #endif 5236 __netif_receive_skb_list(head); 5237 rcu_read_unlock(); 5238 } 5239 5240 /** 5241 * netif_receive_skb - process receive buffer from network 5242 * @skb: buffer to process 5243 * 5244 * netif_receive_skb() is the main receive data processing function. 5245 * It always succeeds. The buffer may be dropped during processing 5246 * for congestion control or by the protocol layers. 5247 * 5248 * This function may only be called from softirq context and interrupts 5249 * should be enabled. 5250 * 5251 * Return values (usually ignored): 5252 * NET_RX_SUCCESS: no congestion 5253 * NET_RX_DROP: packet was dropped 5254 */ 5255 int netif_receive_skb(struct sk_buff *skb) 5256 { 5257 int ret; 5258 5259 trace_netif_receive_skb_entry(skb); 5260 5261 ret = netif_receive_skb_internal(skb); 5262 trace_netif_receive_skb_exit(ret); 5263 5264 return ret; 5265 } 5266 EXPORT_SYMBOL(netif_receive_skb); 5267 5268 /** 5269 * netif_receive_skb_list - process many receive buffers from network 5270 * @head: list of skbs to process. 5271 * 5272 * Since return value of netif_receive_skb() is normally ignored, and 5273 * wouldn't be meaningful for a list, this function returns void. 5274 * 5275 * This function may only be called from softirq context and interrupts 5276 * should be enabled. 5277 */ 5278 void netif_receive_skb_list(struct list_head *head) 5279 { 5280 struct sk_buff *skb; 5281 5282 if (list_empty(head)) 5283 return; 5284 if (trace_netif_receive_skb_list_entry_enabled()) { 5285 list_for_each_entry(skb, head, list) 5286 trace_netif_receive_skb_list_entry(skb); 5287 } 5288 netif_receive_skb_list_internal(head); 5289 trace_netif_receive_skb_list_exit(0); 5290 } 5291 EXPORT_SYMBOL(netif_receive_skb_list); 5292 5293 DEFINE_PER_CPU(struct work_struct, flush_works); 5294 5295 /* Network device is going away, flush any packets still pending */ 5296 static void flush_backlog(struct work_struct *work) 5297 { 5298 struct sk_buff *skb, *tmp; 5299 struct softnet_data *sd; 5300 5301 local_bh_disable(); 5302 sd = this_cpu_ptr(&softnet_data); 5303 5304 local_irq_disable(); 5305 rps_lock(sd); 5306 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5307 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5308 __skb_unlink(skb, &sd->input_pkt_queue); 5309 kfree_skb(skb); 5310 input_queue_head_incr(sd); 5311 } 5312 } 5313 rps_unlock(sd); 5314 local_irq_enable(); 5315 5316 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5317 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5318 __skb_unlink(skb, &sd->process_queue); 5319 kfree_skb(skb); 5320 input_queue_head_incr(sd); 5321 } 5322 } 5323 local_bh_enable(); 5324 } 5325 5326 static void flush_all_backlogs(void) 5327 { 5328 unsigned int cpu; 5329 5330 get_online_cpus(); 5331 5332 for_each_online_cpu(cpu) 5333 queue_work_on(cpu, system_highpri_wq, 5334 per_cpu_ptr(&flush_works, cpu)); 5335 5336 for_each_online_cpu(cpu) 5337 flush_work(per_cpu_ptr(&flush_works, cpu)); 5338 5339 put_online_cpus(); 5340 } 5341 5342 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int)); 5343 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int)); 5344 static int napi_gro_complete(struct sk_buff *skb) 5345 { 5346 struct packet_offload *ptype; 5347 __be16 type = skb->protocol; 5348 struct list_head *head = &offload_base; 5349 int err = -ENOENT; 5350 5351 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5352 5353 if (NAPI_GRO_CB(skb)->count == 1) { 5354 skb_shinfo(skb)->gso_size = 0; 5355 goto out; 5356 } 5357 5358 rcu_read_lock(); 5359 list_for_each_entry_rcu(ptype, head, list) { 5360 if (ptype->type != type || !ptype->callbacks.gro_complete) 5361 continue; 5362 5363 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5364 ipv6_gro_complete, inet_gro_complete, 5365 skb, 0); 5366 break; 5367 } 5368 rcu_read_unlock(); 5369 5370 if (err) { 5371 WARN_ON(&ptype->list == head); 5372 kfree_skb(skb); 5373 return NET_RX_SUCCESS; 5374 } 5375 5376 out: 5377 return netif_receive_skb_internal(skb); 5378 } 5379 5380 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5381 bool flush_old) 5382 { 5383 struct list_head *head = &napi->gro_hash[index].list; 5384 struct sk_buff *skb, *p; 5385 5386 list_for_each_entry_safe_reverse(skb, p, head, list) { 5387 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5388 return; 5389 skb_list_del_init(skb); 5390 napi_gro_complete(skb); 5391 napi->gro_hash[index].count--; 5392 } 5393 5394 if (!napi->gro_hash[index].count) 5395 __clear_bit(index, &napi->gro_bitmask); 5396 } 5397 5398 /* napi->gro_hash[].list contains packets ordered by age. 5399 * youngest packets at the head of it. 5400 * Complete skbs in reverse order to reduce latencies. 5401 */ 5402 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5403 { 5404 unsigned long bitmask = napi->gro_bitmask; 5405 unsigned int i, base = ~0U; 5406 5407 while ((i = ffs(bitmask)) != 0) { 5408 bitmask >>= i; 5409 base += i; 5410 __napi_gro_flush_chain(napi, base, flush_old); 5411 } 5412 } 5413 EXPORT_SYMBOL(napi_gro_flush); 5414 5415 static struct list_head *gro_list_prepare(struct napi_struct *napi, 5416 struct sk_buff *skb) 5417 { 5418 unsigned int maclen = skb->dev->hard_header_len; 5419 u32 hash = skb_get_hash_raw(skb); 5420 struct list_head *head; 5421 struct sk_buff *p; 5422 5423 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; 5424 list_for_each_entry(p, head, list) { 5425 unsigned long diffs; 5426 5427 NAPI_GRO_CB(p)->flush = 0; 5428 5429 if (hash != skb_get_hash_raw(p)) { 5430 NAPI_GRO_CB(p)->same_flow = 0; 5431 continue; 5432 } 5433 5434 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5435 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5436 if (skb_vlan_tag_present(p)) 5437 diffs |= p->vlan_tci ^ skb->vlan_tci; 5438 diffs |= skb_metadata_dst_cmp(p, skb); 5439 diffs |= skb_metadata_differs(p, skb); 5440 if (maclen == ETH_HLEN) 5441 diffs |= compare_ether_header(skb_mac_header(p), 5442 skb_mac_header(skb)); 5443 else if (!diffs) 5444 diffs = memcmp(skb_mac_header(p), 5445 skb_mac_header(skb), 5446 maclen); 5447 NAPI_GRO_CB(p)->same_flow = !diffs; 5448 } 5449 5450 return head; 5451 } 5452 5453 static void skb_gro_reset_offset(struct sk_buff *skb) 5454 { 5455 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5456 const skb_frag_t *frag0 = &pinfo->frags[0]; 5457 5458 NAPI_GRO_CB(skb)->data_offset = 0; 5459 NAPI_GRO_CB(skb)->frag0 = NULL; 5460 NAPI_GRO_CB(skb)->frag0_len = 0; 5461 5462 if (skb_mac_header(skb) == skb_tail_pointer(skb) && 5463 pinfo->nr_frags && 5464 !PageHighMem(skb_frag_page(frag0))) { 5465 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5466 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5467 skb_frag_size(frag0), 5468 skb->end - skb->tail); 5469 } 5470 } 5471 5472 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5473 { 5474 struct skb_shared_info *pinfo = skb_shinfo(skb); 5475 5476 BUG_ON(skb->end - skb->tail < grow); 5477 5478 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5479 5480 skb->data_len -= grow; 5481 skb->tail += grow; 5482 5483 pinfo->frags[0].page_offset += grow; 5484 skb_frag_size_sub(&pinfo->frags[0], grow); 5485 5486 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 5487 skb_frag_unref(skb, 0); 5488 memmove(pinfo->frags, pinfo->frags + 1, 5489 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 5490 } 5491 } 5492 5493 static void gro_flush_oldest(struct list_head *head) 5494 { 5495 struct sk_buff *oldest; 5496 5497 oldest = list_last_entry(head, struct sk_buff, list); 5498 5499 /* We are called with head length >= MAX_GRO_SKBS, so this is 5500 * impossible. 5501 */ 5502 if (WARN_ON_ONCE(!oldest)) 5503 return; 5504 5505 /* Do not adjust napi->gro_hash[].count, caller is adding a new 5506 * SKB to the chain. 5507 */ 5508 skb_list_del_init(oldest); 5509 napi_gro_complete(oldest); 5510 } 5511 5512 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *, 5513 struct sk_buff *)); 5514 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *, 5515 struct sk_buff *)); 5516 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5517 { 5518 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 5519 struct list_head *head = &offload_base; 5520 struct packet_offload *ptype; 5521 __be16 type = skb->protocol; 5522 struct list_head *gro_head; 5523 struct sk_buff *pp = NULL; 5524 enum gro_result ret; 5525 int same_flow; 5526 int grow; 5527 5528 if (netif_elide_gro(skb->dev)) 5529 goto normal; 5530 5531 gro_head = gro_list_prepare(napi, skb); 5532 5533 rcu_read_lock(); 5534 list_for_each_entry_rcu(ptype, head, list) { 5535 if (ptype->type != type || !ptype->callbacks.gro_receive) 5536 continue; 5537 5538 skb_set_network_header(skb, skb_gro_offset(skb)); 5539 skb_reset_mac_len(skb); 5540 NAPI_GRO_CB(skb)->same_flow = 0; 5541 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 5542 NAPI_GRO_CB(skb)->free = 0; 5543 NAPI_GRO_CB(skb)->encap_mark = 0; 5544 NAPI_GRO_CB(skb)->recursion_counter = 0; 5545 NAPI_GRO_CB(skb)->is_fou = 0; 5546 NAPI_GRO_CB(skb)->is_atomic = 1; 5547 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 5548 5549 /* Setup for GRO checksum validation */ 5550 switch (skb->ip_summed) { 5551 case CHECKSUM_COMPLETE: 5552 NAPI_GRO_CB(skb)->csum = skb->csum; 5553 NAPI_GRO_CB(skb)->csum_valid = 1; 5554 NAPI_GRO_CB(skb)->csum_cnt = 0; 5555 break; 5556 case CHECKSUM_UNNECESSARY: 5557 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 5558 NAPI_GRO_CB(skb)->csum_valid = 0; 5559 break; 5560 default: 5561 NAPI_GRO_CB(skb)->csum_cnt = 0; 5562 NAPI_GRO_CB(skb)->csum_valid = 0; 5563 } 5564 5565 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 5566 ipv6_gro_receive, inet_gro_receive, 5567 gro_head, skb); 5568 break; 5569 } 5570 rcu_read_unlock(); 5571 5572 if (&ptype->list == head) 5573 goto normal; 5574 5575 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { 5576 ret = GRO_CONSUMED; 5577 goto ok; 5578 } 5579 5580 same_flow = NAPI_GRO_CB(skb)->same_flow; 5581 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 5582 5583 if (pp) { 5584 skb_list_del_init(pp); 5585 napi_gro_complete(pp); 5586 napi->gro_hash[hash].count--; 5587 } 5588 5589 if (same_flow) 5590 goto ok; 5591 5592 if (NAPI_GRO_CB(skb)->flush) 5593 goto normal; 5594 5595 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { 5596 gro_flush_oldest(gro_head); 5597 } else { 5598 napi->gro_hash[hash].count++; 5599 } 5600 NAPI_GRO_CB(skb)->count = 1; 5601 NAPI_GRO_CB(skb)->age = jiffies; 5602 NAPI_GRO_CB(skb)->last = skb; 5603 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 5604 list_add(&skb->list, gro_head); 5605 ret = GRO_HELD; 5606 5607 pull: 5608 grow = skb_gro_offset(skb) - skb_headlen(skb); 5609 if (grow > 0) 5610 gro_pull_from_frag0(skb, grow); 5611 ok: 5612 if (napi->gro_hash[hash].count) { 5613 if (!test_bit(hash, &napi->gro_bitmask)) 5614 __set_bit(hash, &napi->gro_bitmask); 5615 } else if (test_bit(hash, &napi->gro_bitmask)) { 5616 __clear_bit(hash, &napi->gro_bitmask); 5617 } 5618 5619 return ret; 5620 5621 normal: 5622 ret = GRO_NORMAL; 5623 goto pull; 5624 } 5625 5626 struct packet_offload *gro_find_receive_by_type(__be16 type) 5627 { 5628 struct list_head *offload_head = &offload_base; 5629 struct packet_offload *ptype; 5630 5631 list_for_each_entry_rcu(ptype, offload_head, list) { 5632 if (ptype->type != type || !ptype->callbacks.gro_receive) 5633 continue; 5634 return ptype; 5635 } 5636 return NULL; 5637 } 5638 EXPORT_SYMBOL(gro_find_receive_by_type); 5639 5640 struct packet_offload *gro_find_complete_by_type(__be16 type) 5641 { 5642 struct list_head *offload_head = &offload_base; 5643 struct packet_offload *ptype; 5644 5645 list_for_each_entry_rcu(ptype, offload_head, list) { 5646 if (ptype->type != type || !ptype->callbacks.gro_complete) 5647 continue; 5648 return ptype; 5649 } 5650 return NULL; 5651 } 5652 EXPORT_SYMBOL(gro_find_complete_by_type); 5653 5654 static void napi_skb_free_stolen_head(struct sk_buff *skb) 5655 { 5656 skb_dst_drop(skb); 5657 secpath_reset(skb); 5658 kmem_cache_free(skbuff_head_cache, skb); 5659 } 5660 5661 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) 5662 { 5663 switch (ret) { 5664 case GRO_NORMAL: 5665 if (netif_receive_skb_internal(skb)) 5666 ret = GRO_DROP; 5667 break; 5668 5669 case GRO_DROP: 5670 kfree_skb(skb); 5671 break; 5672 5673 case GRO_MERGED_FREE: 5674 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 5675 napi_skb_free_stolen_head(skb); 5676 else 5677 __kfree_skb(skb); 5678 break; 5679 5680 case GRO_HELD: 5681 case GRO_MERGED: 5682 case GRO_CONSUMED: 5683 break; 5684 } 5685 5686 return ret; 5687 } 5688 5689 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5690 { 5691 gro_result_t ret; 5692 5693 skb_mark_napi_id(skb, napi); 5694 trace_napi_gro_receive_entry(skb); 5695 5696 skb_gro_reset_offset(skb); 5697 5698 ret = napi_skb_finish(dev_gro_receive(napi, skb), skb); 5699 trace_napi_gro_receive_exit(ret); 5700 5701 return ret; 5702 } 5703 EXPORT_SYMBOL(napi_gro_receive); 5704 5705 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 5706 { 5707 if (unlikely(skb->pfmemalloc)) { 5708 consume_skb(skb); 5709 return; 5710 } 5711 __skb_pull(skb, skb_headlen(skb)); 5712 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 5713 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 5714 __vlan_hwaccel_clear_tag(skb); 5715 skb->dev = napi->dev; 5716 skb->skb_iif = 0; 5717 5718 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 5719 skb->pkt_type = PACKET_HOST; 5720 5721 skb->encapsulation = 0; 5722 skb_shinfo(skb)->gso_type = 0; 5723 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5724 secpath_reset(skb); 5725 5726 napi->skb = skb; 5727 } 5728 5729 struct sk_buff *napi_get_frags(struct napi_struct *napi) 5730 { 5731 struct sk_buff *skb = napi->skb; 5732 5733 if (!skb) { 5734 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 5735 if (skb) { 5736 napi->skb = skb; 5737 skb_mark_napi_id(skb, napi); 5738 } 5739 } 5740 return skb; 5741 } 5742 EXPORT_SYMBOL(napi_get_frags); 5743 5744 static gro_result_t napi_frags_finish(struct napi_struct *napi, 5745 struct sk_buff *skb, 5746 gro_result_t ret) 5747 { 5748 switch (ret) { 5749 case GRO_NORMAL: 5750 case GRO_HELD: 5751 __skb_push(skb, ETH_HLEN); 5752 skb->protocol = eth_type_trans(skb, skb->dev); 5753 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) 5754 ret = GRO_DROP; 5755 break; 5756 5757 case GRO_DROP: 5758 napi_reuse_skb(napi, skb); 5759 break; 5760 5761 case GRO_MERGED_FREE: 5762 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 5763 napi_skb_free_stolen_head(skb); 5764 else 5765 napi_reuse_skb(napi, skb); 5766 break; 5767 5768 case GRO_MERGED: 5769 case GRO_CONSUMED: 5770 break; 5771 } 5772 5773 return ret; 5774 } 5775 5776 /* Upper GRO stack assumes network header starts at gro_offset=0 5777 * Drivers could call both napi_gro_frags() and napi_gro_receive() 5778 * We copy ethernet header into skb->data to have a common layout. 5779 */ 5780 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 5781 { 5782 struct sk_buff *skb = napi->skb; 5783 const struct ethhdr *eth; 5784 unsigned int hlen = sizeof(*eth); 5785 5786 napi->skb = NULL; 5787 5788 skb_reset_mac_header(skb); 5789 skb_gro_reset_offset(skb); 5790 5791 eth = skb_gro_header_fast(skb, 0); 5792 if (unlikely(skb_gro_header_hard(skb, hlen))) { 5793 eth = skb_gro_header_slow(skb, hlen, 0); 5794 if (unlikely(!eth)) { 5795 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 5796 __func__, napi->dev->name); 5797 napi_reuse_skb(napi, skb); 5798 return NULL; 5799 } 5800 } else { 5801 gro_pull_from_frag0(skb, hlen); 5802 NAPI_GRO_CB(skb)->frag0 += hlen; 5803 NAPI_GRO_CB(skb)->frag0_len -= hlen; 5804 } 5805 __skb_pull(skb, hlen); 5806 5807 /* 5808 * This works because the only protocols we care about don't require 5809 * special handling. 5810 * We'll fix it up properly in napi_frags_finish() 5811 */ 5812 skb->protocol = eth->h_proto; 5813 5814 return skb; 5815 } 5816 5817 gro_result_t napi_gro_frags(struct napi_struct *napi) 5818 { 5819 gro_result_t ret; 5820 struct sk_buff *skb = napi_frags_skb(napi); 5821 5822 if (!skb) 5823 return GRO_DROP; 5824 5825 trace_napi_gro_frags_entry(skb); 5826 5827 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 5828 trace_napi_gro_frags_exit(ret); 5829 5830 return ret; 5831 } 5832 EXPORT_SYMBOL(napi_gro_frags); 5833 5834 /* Compute the checksum from gro_offset and return the folded value 5835 * after adding in any pseudo checksum. 5836 */ 5837 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 5838 { 5839 __wsum wsum; 5840 __sum16 sum; 5841 5842 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 5843 5844 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 5845 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 5846 /* See comments in __skb_checksum_complete(). */ 5847 if (likely(!sum)) { 5848 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 5849 !skb->csum_complete_sw) 5850 netdev_rx_csum_fault(skb->dev, skb); 5851 } 5852 5853 NAPI_GRO_CB(skb)->csum = wsum; 5854 NAPI_GRO_CB(skb)->csum_valid = 1; 5855 5856 return sum; 5857 } 5858 EXPORT_SYMBOL(__skb_gro_checksum_complete); 5859 5860 static void net_rps_send_ipi(struct softnet_data *remsd) 5861 { 5862 #ifdef CONFIG_RPS 5863 while (remsd) { 5864 struct softnet_data *next = remsd->rps_ipi_next; 5865 5866 if (cpu_online(remsd->cpu)) 5867 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5868 remsd = next; 5869 } 5870 #endif 5871 } 5872 5873 /* 5874 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5875 * Note: called with local irq disabled, but exits with local irq enabled. 5876 */ 5877 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5878 { 5879 #ifdef CONFIG_RPS 5880 struct softnet_data *remsd = sd->rps_ipi_list; 5881 5882 if (remsd) { 5883 sd->rps_ipi_list = NULL; 5884 5885 local_irq_enable(); 5886 5887 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5888 net_rps_send_ipi(remsd); 5889 } else 5890 #endif 5891 local_irq_enable(); 5892 } 5893 5894 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5895 { 5896 #ifdef CONFIG_RPS 5897 return sd->rps_ipi_list != NULL; 5898 #else 5899 return false; 5900 #endif 5901 } 5902 5903 static int process_backlog(struct napi_struct *napi, int quota) 5904 { 5905 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5906 bool again = true; 5907 int work = 0; 5908 5909 /* Check if we have pending ipi, its better to send them now, 5910 * not waiting net_rx_action() end. 5911 */ 5912 if (sd_has_rps_ipi_waiting(sd)) { 5913 local_irq_disable(); 5914 net_rps_action_and_irq_enable(sd); 5915 } 5916 5917 napi->weight = dev_rx_weight; 5918 while (again) { 5919 struct sk_buff *skb; 5920 5921 while ((skb = __skb_dequeue(&sd->process_queue))) { 5922 rcu_read_lock(); 5923 __netif_receive_skb(skb); 5924 rcu_read_unlock(); 5925 input_queue_head_incr(sd); 5926 if (++work >= quota) 5927 return work; 5928 5929 } 5930 5931 local_irq_disable(); 5932 rps_lock(sd); 5933 if (skb_queue_empty(&sd->input_pkt_queue)) { 5934 /* 5935 * Inline a custom version of __napi_complete(). 5936 * only current cpu owns and manipulates this napi, 5937 * and NAPI_STATE_SCHED is the only possible flag set 5938 * on backlog. 5939 * We can use a plain write instead of clear_bit(), 5940 * and we dont need an smp_mb() memory barrier. 5941 */ 5942 napi->state = 0; 5943 again = false; 5944 } else { 5945 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5946 &sd->process_queue); 5947 } 5948 rps_unlock(sd); 5949 local_irq_enable(); 5950 } 5951 5952 return work; 5953 } 5954 5955 /** 5956 * __napi_schedule - schedule for receive 5957 * @n: entry to schedule 5958 * 5959 * The entry's receive function will be scheduled to run. 5960 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5961 */ 5962 void __napi_schedule(struct napi_struct *n) 5963 { 5964 unsigned long flags; 5965 5966 local_irq_save(flags); 5967 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5968 local_irq_restore(flags); 5969 } 5970 EXPORT_SYMBOL(__napi_schedule); 5971 5972 /** 5973 * napi_schedule_prep - check if napi can be scheduled 5974 * @n: napi context 5975 * 5976 * Test if NAPI routine is already running, and if not mark 5977 * it as running. This is used as a condition variable 5978 * insure only one NAPI poll instance runs. We also make 5979 * sure there is no pending NAPI disable. 5980 */ 5981 bool napi_schedule_prep(struct napi_struct *n) 5982 { 5983 unsigned long val, new; 5984 5985 do { 5986 val = READ_ONCE(n->state); 5987 if (unlikely(val & NAPIF_STATE_DISABLE)) 5988 return false; 5989 new = val | NAPIF_STATE_SCHED; 5990 5991 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5992 * This was suggested by Alexander Duyck, as compiler 5993 * emits better code than : 5994 * if (val & NAPIF_STATE_SCHED) 5995 * new |= NAPIF_STATE_MISSED; 5996 */ 5997 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5998 NAPIF_STATE_MISSED; 5999 } while (cmpxchg(&n->state, val, new) != val); 6000 6001 return !(val & NAPIF_STATE_SCHED); 6002 } 6003 EXPORT_SYMBOL(napi_schedule_prep); 6004 6005 /** 6006 * __napi_schedule_irqoff - schedule for receive 6007 * @n: entry to schedule 6008 * 6009 * Variant of __napi_schedule() assuming hard irqs are masked 6010 */ 6011 void __napi_schedule_irqoff(struct napi_struct *n) 6012 { 6013 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6014 } 6015 EXPORT_SYMBOL(__napi_schedule_irqoff); 6016 6017 bool napi_complete_done(struct napi_struct *n, int work_done) 6018 { 6019 unsigned long flags, val, new; 6020 6021 /* 6022 * 1) Don't let napi dequeue from the cpu poll list 6023 * just in case its running on a different cpu. 6024 * 2) If we are busy polling, do nothing here, we have 6025 * the guarantee we will be called later. 6026 */ 6027 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6028 NAPIF_STATE_IN_BUSY_POLL))) 6029 return false; 6030 6031 if (n->gro_bitmask) { 6032 unsigned long timeout = 0; 6033 6034 if (work_done) 6035 timeout = n->dev->gro_flush_timeout; 6036 6037 /* When the NAPI instance uses a timeout and keeps postponing 6038 * it, we need to bound somehow the time packets are kept in 6039 * the GRO layer 6040 */ 6041 napi_gro_flush(n, !!timeout); 6042 if (timeout) 6043 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6044 HRTIMER_MODE_REL_PINNED); 6045 } 6046 if (unlikely(!list_empty(&n->poll_list))) { 6047 /* If n->poll_list is not empty, we need to mask irqs */ 6048 local_irq_save(flags); 6049 list_del_init(&n->poll_list); 6050 local_irq_restore(flags); 6051 } 6052 6053 do { 6054 val = READ_ONCE(n->state); 6055 6056 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6057 6058 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); 6059 6060 /* If STATE_MISSED was set, leave STATE_SCHED set, 6061 * because we will call napi->poll() one more time. 6062 * This C code was suggested by Alexander Duyck to help gcc. 6063 */ 6064 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6065 NAPIF_STATE_SCHED; 6066 } while (cmpxchg(&n->state, val, new) != val); 6067 6068 if (unlikely(val & NAPIF_STATE_MISSED)) { 6069 __napi_schedule(n); 6070 return false; 6071 } 6072 6073 return true; 6074 } 6075 EXPORT_SYMBOL(napi_complete_done); 6076 6077 /* must be called under rcu_read_lock(), as we dont take a reference */ 6078 static struct napi_struct *napi_by_id(unsigned int napi_id) 6079 { 6080 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6081 struct napi_struct *napi; 6082 6083 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6084 if (napi->napi_id == napi_id) 6085 return napi; 6086 6087 return NULL; 6088 } 6089 6090 #if defined(CONFIG_NET_RX_BUSY_POLL) 6091 6092 #define BUSY_POLL_BUDGET 8 6093 6094 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) 6095 { 6096 int rc; 6097 6098 /* Busy polling means there is a high chance device driver hard irq 6099 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6100 * set in napi_schedule_prep(). 6101 * Since we are about to call napi->poll() once more, we can safely 6102 * clear NAPI_STATE_MISSED. 6103 * 6104 * Note: x86 could use a single "lock and ..." instruction 6105 * to perform these two clear_bit() 6106 */ 6107 clear_bit(NAPI_STATE_MISSED, &napi->state); 6108 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6109 6110 local_bh_disable(); 6111 6112 /* All we really want here is to re-enable device interrupts. 6113 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6114 */ 6115 rc = napi->poll(napi, BUSY_POLL_BUDGET); 6116 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); 6117 netpoll_poll_unlock(have_poll_lock); 6118 if (rc == BUSY_POLL_BUDGET) 6119 __napi_schedule(napi); 6120 local_bh_enable(); 6121 } 6122 6123 void napi_busy_loop(unsigned int napi_id, 6124 bool (*loop_end)(void *, unsigned long), 6125 void *loop_end_arg) 6126 { 6127 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6128 int (*napi_poll)(struct napi_struct *napi, int budget); 6129 void *have_poll_lock = NULL; 6130 struct napi_struct *napi; 6131 6132 restart: 6133 napi_poll = NULL; 6134 6135 rcu_read_lock(); 6136 6137 napi = napi_by_id(napi_id); 6138 if (!napi) 6139 goto out; 6140 6141 preempt_disable(); 6142 for (;;) { 6143 int work = 0; 6144 6145 local_bh_disable(); 6146 if (!napi_poll) { 6147 unsigned long val = READ_ONCE(napi->state); 6148 6149 /* If multiple threads are competing for this napi, 6150 * we avoid dirtying napi->state as much as we can. 6151 */ 6152 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6153 NAPIF_STATE_IN_BUSY_POLL)) 6154 goto count; 6155 if (cmpxchg(&napi->state, val, 6156 val | NAPIF_STATE_IN_BUSY_POLL | 6157 NAPIF_STATE_SCHED) != val) 6158 goto count; 6159 have_poll_lock = netpoll_poll_lock(napi); 6160 napi_poll = napi->poll; 6161 } 6162 work = napi_poll(napi, BUSY_POLL_BUDGET); 6163 trace_napi_poll(napi, work, BUSY_POLL_BUDGET); 6164 count: 6165 if (work > 0) 6166 __NET_ADD_STATS(dev_net(napi->dev), 6167 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6168 local_bh_enable(); 6169 6170 if (!loop_end || loop_end(loop_end_arg, start_time)) 6171 break; 6172 6173 if (unlikely(need_resched())) { 6174 if (napi_poll) 6175 busy_poll_stop(napi, have_poll_lock); 6176 preempt_enable(); 6177 rcu_read_unlock(); 6178 cond_resched(); 6179 if (loop_end(loop_end_arg, start_time)) 6180 return; 6181 goto restart; 6182 } 6183 cpu_relax(); 6184 } 6185 if (napi_poll) 6186 busy_poll_stop(napi, have_poll_lock); 6187 preempt_enable(); 6188 out: 6189 rcu_read_unlock(); 6190 } 6191 EXPORT_SYMBOL(napi_busy_loop); 6192 6193 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6194 6195 static void napi_hash_add(struct napi_struct *napi) 6196 { 6197 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || 6198 test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) 6199 return; 6200 6201 spin_lock(&napi_hash_lock); 6202 6203 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6204 do { 6205 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6206 napi_gen_id = MIN_NAPI_ID; 6207 } while (napi_by_id(napi_gen_id)); 6208 napi->napi_id = napi_gen_id; 6209 6210 hlist_add_head_rcu(&napi->napi_hash_node, 6211 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6212 6213 spin_unlock(&napi_hash_lock); 6214 } 6215 6216 /* Warning : caller is responsible to make sure rcu grace period 6217 * is respected before freeing memory containing @napi 6218 */ 6219 bool napi_hash_del(struct napi_struct *napi) 6220 { 6221 bool rcu_sync_needed = false; 6222 6223 spin_lock(&napi_hash_lock); 6224 6225 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { 6226 rcu_sync_needed = true; 6227 hlist_del_rcu(&napi->napi_hash_node); 6228 } 6229 spin_unlock(&napi_hash_lock); 6230 return rcu_sync_needed; 6231 } 6232 EXPORT_SYMBOL_GPL(napi_hash_del); 6233 6234 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6235 { 6236 struct napi_struct *napi; 6237 6238 napi = container_of(timer, struct napi_struct, timer); 6239 6240 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6241 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6242 */ 6243 if (napi->gro_bitmask && !napi_disable_pending(napi) && 6244 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) 6245 __napi_schedule_irqoff(napi); 6246 6247 return HRTIMER_NORESTART; 6248 } 6249 6250 static void init_gro_hash(struct napi_struct *napi) 6251 { 6252 int i; 6253 6254 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6255 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6256 napi->gro_hash[i].count = 0; 6257 } 6258 napi->gro_bitmask = 0; 6259 } 6260 6261 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6262 int (*poll)(struct napi_struct *, int), int weight) 6263 { 6264 INIT_LIST_HEAD(&napi->poll_list); 6265 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6266 napi->timer.function = napi_watchdog; 6267 init_gro_hash(napi); 6268 napi->skb = NULL; 6269 napi->poll = poll; 6270 if (weight > NAPI_POLL_WEIGHT) 6271 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6272 weight); 6273 napi->weight = weight; 6274 list_add(&napi->dev_list, &dev->napi_list); 6275 napi->dev = dev; 6276 #ifdef CONFIG_NETPOLL 6277 napi->poll_owner = -1; 6278 #endif 6279 set_bit(NAPI_STATE_SCHED, &napi->state); 6280 napi_hash_add(napi); 6281 } 6282 EXPORT_SYMBOL(netif_napi_add); 6283 6284 void napi_disable(struct napi_struct *n) 6285 { 6286 might_sleep(); 6287 set_bit(NAPI_STATE_DISABLE, &n->state); 6288 6289 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6290 msleep(1); 6291 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6292 msleep(1); 6293 6294 hrtimer_cancel(&n->timer); 6295 6296 clear_bit(NAPI_STATE_DISABLE, &n->state); 6297 } 6298 EXPORT_SYMBOL(napi_disable); 6299 6300 static void flush_gro_hash(struct napi_struct *napi) 6301 { 6302 int i; 6303 6304 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6305 struct sk_buff *skb, *n; 6306 6307 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6308 kfree_skb(skb); 6309 napi->gro_hash[i].count = 0; 6310 } 6311 } 6312 6313 /* Must be called in process context */ 6314 void netif_napi_del(struct napi_struct *napi) 6315 { 6316 might_sleep(); 6317 if (napi_hash_del(napi)) 6318 synchronize_net(); 6319 list_del_init(&napi->dev_list); 6320 napi_free_frags(napi); 6321 6322 flush_gro_hash(napi); 6323 napi->gro_bitmask = 0; 6324 } 6325 EXPORT_SYMBOL(netif_napi_del); 6326 6327 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6328 { 6329 void *have; 6330 int work, weight; 6331 6332 list_del_init(&n->poll_list); 6333 6334 have = netpoll_poll_lock(n); 6335 6336 weight = n->weight; 6337 6338 /* This NAPI_STATE_SCHED test is for avoiding a race 6339 * with netpoll's poll_napi(). Only the entity which 6340 * obtains the lock and sees NAPI_STATE_SCHED set will 6341 * actually make the ->poll() call. Therefore we avoid 6342 * accidentally calling ->poll() when NAPI is not scheduled. 6343 */ 6344 work = 0; 6345 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6346 work = n->poll(n, weight); 6347 trace_napi_poll(n, work, weight); 6348 } 6349 6350 WARN_ON_ONCE(work > weight); 6351 6352 if (likely(work < weight)) 6353 goto out_unlock; 6354 6355 /* Drivers must not modify the NAPI state if they 6356 * consume the entire weight. In such cases this code 6357 * still "owns" the NAPI instance and therefore can 6358 * move the instance around on the list at-will. 6359 */ 6360 if (unlikely(napi_disable_pending(n))) { 6361 napi_complete(n); 6362 goto out_unlock; 6363 } 6364 6365 if (n->gro_bitmask) { 6366 /* flush too old packets 6367 * If HZ < 1000, flush all packets. 6368 */ 6369 napi_gro_flush(n, HZ >= 1000); 6370 } 6371 6372 /* Some drivers may have called napi_schedule 6373 * prior to exhausting their budget. 6374 */ 6375 if (unlikely(!list_empty(&n->poll_list))) { 6376 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6377 n->dev ? n->dev->name : "backlog"); 6378 goto out_unlock; 6379 } 6380 6381 list_add_tail(&n->poll_list, repoll); 6382 6383 out_unlock: 6384 netpoll_poll_unlock(have); 6385 6386 return work; 6387 } 6388 6389 static __latent_entropy void net_rx_action(struct softirq_action *h) 6390 { 6391 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6392 unsigned long time_limit = jiffies + 6393 usecs_to_jiffies(netdev_budget_usecs); 6394 int budget = netdev_budget; 6395 LIST_HEAD(list); 6396 LIST_HEAD(repoll); 6397 6398 local_irq_disable(); 6399 list_splice_init(&sd->poll_list, &list); 6400 local_irq_enable(); 6401 6402 for (;;) { 6403 struct napi_struct *n; 6404 6405 if (list_empty(&list)) { 6406 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6407 goto out; 6408 break; 6409 } 6410 6411 n = list_first_entry(&list, struct napi_struct, poll_list); 6412 budget -= napi_poll(n, &repoll); 6413 6414 /* If softirq window is exhausted then punt. 6415 * Allow this to run for 2 jiffies since which will allow 6416 * an average latency of 1.5/HZ. 6417 */ 6418 if (unlikely(budget <= 0 || 6419 time_after_eq(jiffies, time_limit))) { 6420 sd->time_squeeze++; 6421 break; 6422 } 6423 } 6424 6425 local_irq_disable(); 6426 6427 list_splice_tail_init(&sd->poll_list, &list); 6428 list_splice_tail(&repoll, &list); 6429 list_splice(&list, &sd->poll_list); 6430 if (!list_empty(&sd->poll_list)) 6431 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6432 6433 net_rps_action_and_irq_enable(sd); 6434 out: 6435 __kfree_skb_flush(); 6436 } 6437 6438 struct netdev_adjacent { 6439 struct net_device *dev; 6440 6441 /* upper master flag, there can only be one master device per list */ 6442 bool master; 6443 6444 /* counter for the number of times this device was added to us */ 6445 u16 ref_nr; 6446 6447 /* private field for the users */ 6448 void *private; 6449 6450 struct list_head list; 6451 struct rcu_head rcu; 6452 }; 6453 6454 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6455 struct list_head *adj_list) 6456 { 6457 struct netdev_adjacent *adj; 6458 6459 list_for_each_entry(adj, adj_list, list) { 6460 if (adj->dev == adj_dev) 6461 return adj; 6462 } 6463 return NULL; 6464 } 6465 6466 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) 6467 { 6468 struct net_device *dev = data; 6469 6470 return upper_dev == dev; 6471 } 6472 6473 /** 6474 * netdev_has_upper_dev - Check if device is linked to an upper device 6475 * @dev: device 6476 * @upper_dev: upper device to check 6477 * 6478 * Find out if a device is linked to specified upper device and return true 6479 * in case it is. Note that this checks only immediate upper device, 6480 * not through a complete stack of devices. The caller must hold the RTNL lock. 6481 */ 6482 bool netdev_has_upper_dev(struct net_device *dev, 6483 struct net_device *upper_dev) 6484 { 6485 ASSERT_RTNL(); 6486 6487 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 6488 upper_dev); 6489 } 6490 EXPORT_SYMBOL(netdev_has_upper_dev); 6491 6492 /** 6493 * netdev_has_upper_dev_all - Check if device is linked to an upper device 6494 * @dev: device 6495 * @upper_dev: upper device to check 6496 * 6497 * Find out if a device is linked to specified upper device and return true 6498 * in case it is. Note that this checks the entire upper device chain. 6499 * The caller must hold rcu lock. 6500 */ 6501 6502 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6503 struct net_device *upper_dev) 6504 { 6505 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, 6506 upper_dev); 6507 } 6508 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6509 6510 /** 6511 * netdev_has_any_upper_dev - Check if device is linked to some device 6512 * @dev: device 6513 * 6514 * Find out if a device is linked to an upper device and return true in case 6515 * it is. The caller must hold the RTNL lock. 6516 */ 6517 bool netdev_has_any_upper_dev(struct net_device *dev) 6518 { 6519 ASSERT_RTNL(); 6520 6521 return !list_empty(&dev->adj_list.upper); 6522 } 6523 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6524 6525 /** 6526 * netdev_master_upper_dev_get - Get master upper device 6527 * @dev: device 6528 * 6529 * Find a master upper device and return pointer to it or NULL in case 6530 * it's not there. The caller must hold the RTNL lock. 6531 */ 6532 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6533 { 6534 struct netdev_adjacent *upper; 6535 6536 ASSERT_RTNL(); 6537 6538 if (list_empty(&dev->adj_list.upper)) 6539 return NULL; 6540 6541 upper = list_first_entry(&dev->adj_list.upper, 6542 struct netdev_adjacent, list); 6543 if (likely(upper->master)) 6544 return upper->dev; 6545 return NULL; 6546 } 6547 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6548 6549 /** 6550 * netdev_has_any_lower_dev - Check if device is linked to some device 6551 * @dev: device 6552 * 6553 * Find out if a device is linked to a lower device and return true in case 6554 * it is. The caller must hold the RTNL lock. 6555 */ 6556 static bool netdev_has_any_lower_dev(struct net_device *dev) 6557 { 6558 ASSERT_RTNL(); 6559 6560 return !list_empty(&dev->adj_list.lower); 6561 } 6562 6563 void *netdev_adjacent_get_private(struct list_head *adj_list) 6564 { 6565 struct netdev_adjacent *adj; 6566 6567 adj = list_entry(adj_list, struct netdev_adjacent, list); 6568 6569 return adj->private; 6570 } 6571 EXPORT_SYMBOL(netdev_adjacent_get_private); 6572 6573 /** 6574 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6575 * @dev: device 6576 * @iter: list_head ** of the current position 6577 * 6578 * Gets the next device from the dev's upper list, starting from iter 6579 * position. The caller must hold RCU read lock. 6580 */ 6581 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6582 struct list_head **iter) 6583 { 6584 struct netdev_adjacent *upper; 6585 6586 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6587 6588 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6589 6590 if (&upper->list == &dev->adj_list.upper) 6591 return NULL; 6592 6593 *iter = &upper->list; 6594 6595 return upper->dev; 6596 } 6597 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6598 6599 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6600 struct list_head **iter) 6601 { 6602 struct netdev_adjacent *upper; 6603 6604 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6605 6606 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6607 6608 if (&upper->list == &dev->adj_list.upper) 6609 return NULL; 6610 6611 *iter = &upper->list; 6612 6613 return upper->dev; 6614 } 6615 6616 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6617 int (*fn)(struct net_device *dev, 6618 void *data), 6619 void *data) 6620 { 6621 struct net_device *udev; 6622 struct list_head *iter; 6623 int ret; 6624 6625 for (iter = &dev->adj_list.upper, 6626 udev = netdev_next_upper_dev_rcu(dev, &iter); 6627 udev; 6628 udev = netdev_next_upper_dev_rcu(dev, &iter)) { 6629 /* first is the upper device itself */ 6630 ret = fn(udev, data); 6631 if (ret) 6632 return ret; 6633 6634 /* then look at all of its upper devices */ 6635 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); 6636 if (ret) 6637 return ret; 6638 } 6639 6640 return 0; 6641 } 6642 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6643 6644 /** 6645 * netdev_lower_get_next_private - Get the next ->private from the 6646 * lower neighbour list 6647 * @dev: device 6648 * @iter: list_head ** of the current position 6649 * 6650 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6651 * list, starting from iter position. The caller must hold either hold the 6652 * RTNL lock or its own locking that guarantees that the neighbour lower 6653 * list will remain unchanged. 6654 */ 6655 void *netdev_lower_get_next_private(struct net_device *dev, 6656 struct list_head **iter) 6657 { 6658 struct netdev_adjacent *lower; 6659 6660 lower = list_entry(*iter, struct netdev_adjacent, list); 6661 6662 if (&lower->list == &dev->adj_list.lower) 6663 return NULL; 6664 6665 *iter = lower->list.next; 6666 6667 return lower->private; 6668 } 6669 EXPORT_SYMBOL(netdev_lower_get_next_private); 6670 6671 /** 6672 * netdev_lower_get_next_private_rcu - Get the next ->private from the 6673 * lower neighbour list, RCU 6674 * variant 6675 * @dev: device 6676 * @iter: list_head ** of the current position 6677 * 6678 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6679 * list, starting from iter position. The caller must hold RCU read lock. 6680 */ 6681 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 6682 struct list_head **iter) 6683 { 6684 struct netdev_adjacent *lower; 6685 6686 WARN_ON_ONCE(!rcu_read_lock_held()); 6687 6688 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6689 6690 if (&lower->list == &dev->adj_list.lower) 6691 return NULL; 6692 6693 *iter = &lower->list; 6694 6695 return lower->private; 6696 } 6697 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 6698 6699 /** 6700 * netdev_lower_get_next - Get the next device from the lower neighbour 6701 * list 6702 * @dev: device 6703 * @iter: list_head ** of the current position 6704 * 6705 * Gets the next netdev_adjacent from the dev's lower neighbour 6706 * list, starting from iter position. The caller must hold RTNL lock or 6707 * its own locking that guarantees that the neighbour lower 6708 * list will remain unchanged. 6709 */ 6710 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 6711 { 6712 struct netdev_adjacent *lower; 6713 6714 lower = list_entry(*iter, struct netdev_adjacent, list); 6715 6716 if (&lower->list == &dev->adj_list.lower) 6717 return NULL; 6718 6719 *iter = lower->list.next; 6720 6721 return lower->dev; 6722 } 6723 EXPORT_SYMBOL(netdev_lower_get_next); 6724 6725 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 6726 struct list_head **iter) 6727 { 6728 struct netdev_adjacent *lower; 6729 6730 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 6731 6732 if (&lower->list == &dev->adj_list.lower) 6733 return NULL; 6734 6735 *iter = &lower->list; 6736 6737 return lower->dev; 6738 } 6739 6740 int netdev_walk_all_lower_dev(struct net_device *dev, 6741 int (*fn)(struct net_device *dev, 6742 void *data), 6743 void *data) 6744 { 6745 struct net_device *ldev; 6746 struct list_head *iter; 6747 int ret; 6748 6749 for (iter = &dev->adj_list.lower, 6750 ldev = netdev_next_lower_dev(dev, &iter); 6751 ldev; 6752 ldev = netdev_next_lower_dev(dev, &iter)) { 6753 /* first is the lower device itself */ 6754 ret = fn(ldev, data); 6755 if (ret) 6756 return ret; 6757 6758 /* then look at all of its lower devices */ 6759 ret = netdev_walk_all_lower_dev(ldev, fn, data); 6760 if (ret) 6761 return ret; 6762 } 6763 6764 return 0; 6765 } 6766 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 6767 6768 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 6769 struct list_head **iter) 6770 { 6771 struct netdev_adjacent *lower; 6772 6773 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6774 if (&lower->list == &dev->adj_list.lower) 6775 return NULL; 6776 6777 *iter = &lower->list; 6778 6779 return lower->dev; 6780 } 6781 6782 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 6783 int (*fn)(struct net_device *dev, 6784 void *data), 6785 void *data) 6786 { 6787 struct net_device *ldev; 6788 struct list_head *iter; 6789 int ret; 6790 6791 for (iter = &dev->adj_list.lower, 6792 ldev = netdev_next_lower_dev_rcu(dev, &iter); 6793 ldev; 6794 ldev = netdev_next_lower_dev_rcu(dev, &iter)) { 6795 /* first is the lower device itself */ 6796 ret = fn(ldev, data); 6797 if (ret) 6798 return ret; 6799 6800 /* then look at all of its lower devices */ 6801 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); 6802 if (ret) 6803 return ret; 6804 } 6805 6806 return 0; 6807 } 6808 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 6809 6810 /** 6811 * netdev_lower_get_first_private_rcu - Get the first ->private from the 6812 * lower neighbour list, RCU 6813 * variant 6814 * @dev: device 6815 * 6816 * Gets the first netdev_adjacent->private from the dev's lower neighbour 6817 * list. The caller must hold RCU read lock. 6818 */ 6819 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 6820 { 6821 struct netdev_adjacent *lower; 6822 6823 lower = list_first_or_null_rcu(&dev->adj_list.lower, 6824 struct netdev_adjacent, list); 6825 if (lower) 6826 return lower->private; 6827 return NULL; 6828 } 6829 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 6830 6831 /** 6832 * netdev_master_upper_dev_get_rcu - Get master upper device 6833 * @dev: device 6834 * 6835 * Find a master upper device and return pointer to it or NULL in case 6836 * it's not there. The caller must hold the RCU read lock. 6837 */ 6838 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 6839 { 6840 struct netdev_adjacent *upper; 6841 6842 upper = list_first_or_null_rcu(&dev->adj_list.upper, 6843 struct netdev_adjacent, list); 6844 if (upper && likely(upper->master)) 6845 return upper->dev; 6846 return NULL; 6847 } 6848 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 6849 6850 static int netdev_adjacent_sysfs_add(struct net_device *dev, 6851 struct net_device *adj_dev, 6852 struct list_head *dev_list) 6853 { 6854 char linkname[IFNAMSIZ+7]; 6855 6856 sprintf(linkname, dev_list == &dev->adj_list.upper ? 6857 "upper_%s" : "lower_%s", adj_dev->name); 6858 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 6859 linkname); 6860 } 6861 static void netdev_adjacent_sysfs_del(struct net_device *dev, 6862 char *name, 6863 struct list_head *dev_list) 6864 { 6865 char linkname[IFNAMSIZ+7]; 6866 6867 sprintf(linkname, dev_list == &dev->adj_list.upper ? 6868 "upper_%s" : "lower_%s", name); 6869 sysfs_remove_link(&(dev->dev.kobj), linkname); 6870 } 6871 6872 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 6873 struct net_device *adj_dev, 6874 struct list_head *dev_list) 6875 { 6876 return (dev_list == &dev->adj_list.upper || 6877 dev_list == &dev->adj_list.lower) && 6878 net_eq(dev_net(dev), dev_net(adj_dev)); 6879 } 6880 6881 static int __netdev_adjacent_dev_insert(struct net_device *dev, 6882 struct net_device *adj_dev, 6883 struct list_head *dev_list, 6884 void *private, bool master) 6885 { 6886 struct netdev_adjacent *adj; 6887 int ret; 6888 6889 adj = __netdev_find_adj(adj_dev, dev_list); 6890 6891 if (adj) { 6892 adj->ref_nr += 1; 6893 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 6894 dev->name, adj_dev->name, adj->ref_nr); 6895 6896 return 0; 6897 } 6898 6899 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 6900 if (!adj) 6901 return -ENOMEM; 6902 6903 adj->dev = adj_dev; 6904 adj->master = master; 6905 adj->ref_nr = 1; 6906 adj->private = private; 6907 dev_hold(adj_dev); 6908 6909 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 6910 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 6911 6912 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 6913 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 6914 if (ret) 6915 goto free_adj; 6916 } 6917 6918 /* Ensure that master link is always the first item in list. */ 6919 if (master) { 6920 ret = sysfs_create_link(&(dev->dev.kobj), 6921 &(adj_dev->dev.kobj), "master"); 6922 if (ret) 6923 goto remove_symlinks; 6924 6925 list_add_rcu(&adj->list, dev_list); 6926 } else { 6927 list_add_tail_rcu(&adj->list, dev_list); 6928 } 6929 6930 return 0; 6931 6932 remove_symlinks: 6933 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 6934 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 6935 free_adj: 6936 kfree(adj); 6937 dev_put(adj_dev); 6938 6939 return ret; 6940 } 6941 6942 static void __netdev_adjacent_dev_remove(struct net_device *dev, 6943 struct net_device *adj_dev, 6944 u16 ref_nr, 6945 struct list_head *dev_list) 6946 { 6947 struct netdev_adjacent *adj; 6948 6949 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 6950 dev->name, adj_dev->name, ref_nr); 6951 6952 adj = __netdev_find_adj(adj_dev, dev_list); 6953 6954 if (!adj) { 6955 pr_err("Adjacency does not exist for device %s from %s\n", 6956 dev->name, adj_dev->name); 6957 WARN_ON(1); 6958 return; 6959 } 6960 6961 if (adj->ref_nr > ref_nr) { 6962 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 6963 dev->name, adj_dev->name, ref_nr, 6964 adj->ref_nr - ref_nr); 6965 adj->ref_nr -= ref_nr; 6966 return; 6967 } 6968 6969 if (adj->master) 6970 sysfs_remove_link(&(dev->dev.kobj), "master"); 6971 6972 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 6973 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 6974 6975 list_del_rcu(&adj->list); 6976 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 6977 adj_dev->name, dev->name, adj_dev->name); 6978 dev_put(adj_dev); 6979 kfree_rcu(adj, rcu); 6980 } 6981 6982 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 6983 struct net_device *upper_dev, 6984 struct list_head *up_list, 6985 struct list_head *down_list, 6986 void *private, bool master) 6987 { 6988 int ret; 6989 6990 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 6991 private, master); 6992 if (ret) 6993 return ret; 6994 6995 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 6996 private, false); 6997 if (ret) { 6998 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 6999 return ret; 7000 } 7001 7002 return 0; 7003 } 7004 7005 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7006 struct net_device *upper_dev, 7007 u16 ref_nr, 7008 struct list_head *up_list, 7009 struct list_head *down_list) 7010 { 7011 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7012 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7013 } 7014 7015 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7016 struct net_device *upper_dev, 7017 void *private, bool master) 7018 { 7019 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7020 &dev->adj_list.upper, 7021 &upper_dev->adj_list.lower, 7022 private, master); 7023 } 7024 7025 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7026 struct net_device *upper_dev) 7027 { 7028 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7029 &dev->adj_list.upper, 7030 &upper_dev->adj_list.lower); 7031 } 7032 7033 static int __netdev_upper_dev_link(struct net_device *dev, 7034 struct net_device *upper_dev, bool master, 7035 void *upper_priv, void *upper_info, 7036 struct netlink_ext_ack *extack) 7037 { 7038 struct netdev_notifier_changeupper_info changeupper_info = { 7039 .info = { 7040 .dev = dev, 7041 .extack = extack, 7042 }, 7043 .upper_dev = upper_dev, 7044 .master = master, 7045 .linking = true, 7046 .upper_info = upper_info, 7047 }; 7048 struct net_device *master_dev; 7049 int ret = 0; 7050 7051 ASSERT_RTNL(); 7052 7053 if (dev == upper_dev) 7054 return -EBUSY; 7055 7056 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7057 if (netdev_has_upper_dev(upper_dev, dev)) 7058 return -EBUSY; 7059 7060 if (!master) { 7061 if (netdev_has_upper_dev(dev, upper_dev)) 7062 return -EEXIST; 7063 } else { 7064 master_dev = netdev_master_upper_dev_get(dev); 7065 if (master_dev) 7066 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7067 } 7068 7069 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7070 &changeupper_info.info); 7071 ret = notifier_to_errno(ret); 7072 if (ret) 7073 return ret; 7074 7075 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7076 master); 7077 if (ret) 7078 return ret; 7079 7080 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7081 &changeupper_info.info); 7082 ret = notifier_to_errno(ret); 7083 if (ret) 7084 goto rollback; 7085 7086 return 0; 7087 7088 rollback: 7089 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7090 7091 return ret; 7092 } 7093 7094 /** 7095 * netdev_upper_dev_link - Add a link to the upper device 7096 * @dev: device 7097 * @upper_dev: new upper device 7098 * @extack: netlink extended ack 7099 * 7100 * Adds a link to device which is upper to this one. The caller must hold 7101 * the RTNL lock. On a failure a negative errno code is returned. 7102 * On success the reference counts are adjusted and the function 7103 * returns zero. 7104 */ 7105 int netdev_upper_dev_link(struct net_device *dev, 7106 struct net_device *upper_dev, 7107 struct netlink_ext_ack *extack) 7108 { 7109 return __netdev_upper_dev_link(dev, upper_dev, false, 7110 NULL, NULL, extack); 7111 } 7112 EXPORT_SYMBOL(netdev_upper_dev_link); 7113 7114 /** 7115 * netdev_master_upper_dev_link - Add a master link to the upper device 7116 * @dev: device 7117 * @upper_dev: new upper device 7118 * @upper_priv: upper device private 7119 * @upper_info: upper info to be passed down via notifier 7120 * @extack: netlink extended ack 7121 * 7122 * Adds a link to device which is upper to this one. In this case, only 7123 * one master upper device can be linked, although other non-master devices 7124 * might be linked as well. The caller must hold the RTNL lock. 7125 * On a failure a negative errno code is returned. On success the reference 7126 * counts are adjusted and the function returns zero. 7127 */ 7128 int netdev_master_upper_dev_link(struct net_device *dev, 7129 struct net_device *upper_dev, 7130 void *upper_priv, void *upper_info, 7131 struct netlink_ext_ack *extack) 7132 { 7133 return __netdev_upper_dev_link(dev, upper_dev, true, 7134 upper_priv, upper_info, extack); 7135 } 7136 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7137 7138 /** 7139 * netdev_upper_dev_unlink - Removes a link to upper device 7140 * @dev: device 7141 * @upper_dev: new upper device 7142 * 7143 * Removes a link to device which is upper to this one. The caller must hold 7144 * the RTNL lock. 7145 */ 7146 void netdev_upper_dev_unlink(struct net_device *dev, 7147 struct net_device *upper_dev) 7148 { 7149 struct netdev_notifier_changeupper_info changeupper_info = { 7150 .info = { 7151 .dev = dev, 7152 }, 7153 .upper_dev = upper_dev, 7154 .linking = false, 7155 }; 7156 7157 ASSERT_RTNL(); 7158 7159 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7160 7161 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7162 &changeupper_info.info); 7163 7164 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7165 7166 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7167 &changeupper_info.info); 7168 } 7169 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7170 7171 /** 7172 * netdev_bonding_info_change - Dispatch event about slave change 7173 * @dev: device 7174 * @bonding_info: info to dispatch 7175 * 7176 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7177 * The caller must hold the RTNL lock. 7178 */ 7179 void netdev_bonding_info_change(struct net_device *dev, 7180 struct netdev_bonding_info *bonding_info) 7181 { 7182 struct netdev_notifier_bonding_info info = { 7183 .info.dev = dev, 7184 }; 7185 7186 memcpy(&info.bonding_info, bonding_info, 7187 sizeof(struct netdev_bonding_info)); 7188 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7189 &info.info); 7190 } 7191 EXPORT_SYMBOL(netdev_bonding_info_change); 7192 7193 static void netdev_adjacent_add_links(struct net_device *dev) 7194 { 7195 struct netdev_adjacent *iter; 7196 7197 struct net *net = dev_net(dev); 7198 7199 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7200 if (!net_eq(net, dev_net(iter->dev))) 7201 continue; 7202 netdev_adjacent_sysfs_add(iter->dev, dev, 7203 &iter->dev->adj_list.lower); 7204 netdev_adjacent_sysfs_add(dev, iter->dev, 7205 &dev->adj_list.upper); 7206 } 7207 7208 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7209 if (!net_eq(net, dev_net(iter->dev))) 7210 continue; 7211 netdev_adjacent_sysfs_add(iter->dev, dev, 7212 &iter->dev->adj_list.upper); 7213 netdev_adjacent_sysfs_add(dev, iter->dev, 7214 &dev->adj_list.lower); 7215 } 7216 } 7217 7218 static void netdev_adjacent_del_links(struct net_device *dev) 7219 { 7220 struct netdev_adjacent *iter; 7221 7222 struct net *net = dev_net(dev); 7223 7224 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7225 if (!net_eq(net, dev_net(iter->dev))) 7226 continue; 7227 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7228 &iter->dev->adj_list.lower); 7229 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7230 &dev->adj_list.upper); 7231 } 7232 7233 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7234 if (!net_eq(net, dev_net(iter->dev))) 7235 continue; 7236 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7237 &iter->dev->adj_list.upper); 7238 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7239 &dev->adj_list.lower); 7240 } 7241 } 7242 7243 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 7244 { 7245 struct netdev_adjacent *iter; 7246 7247 struct net *net = dev_net(dev); 7248 7249 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7250 if (!net_eq(net, dev_net(iter->dev))) 7251 continue; 7252 netdev_adjacent_sysfs_del(iter->dev, oldname, 7253 &iter->dev->adj_list.lower); 7254 netdev_adjacent_sysfs_add(iter->dev, dev, 7255 &iter->dev->adj_list.lower); 7256 } 7257 7258 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7259 if (!net_eq(net, dev_net(iter->dev))) 7260 continue; 7261 netdev_adjacent_sysfs_del(iter->dev, oldname, 7262 &iter->dev->adj_list.upper); 7263 netdev_adjacent_sysfs_add(iter->dev, dev, 7264 &iter->dev->adj_list.upper); 7265 } 7266 } 7267 7268 void *netdev_lower_dev_get_private(struct net_device *dev, 7269 struct net_device *lower_dev) 7270 { 7271 struct netdev_adjacent *lower; 7272 7273 if (!lower_dev) 7274 return NULL; 7275 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 7276 if (!lower) 7277 return NULL; 7278 7279 return lower->private; 7280 } 7281 EXPORT_SYMBOL(netdev_lower_dev_get_private); 7282 7283 7284 int dev_get_nest_level(struct net_device *dev) 7285 { 7286 struct net_device *lower = NULL; 7287 struct list_head *iter; 7288 int max_nest = -1; 7289 int nest; 7290 7291 ASSERT_RTNL(); 7292 7293 netdev_for_each_lower_dev(dev, lower, iter) { 7294 nest = dev_get_nest_level(lower); 7295 if (max_nest < nest) 7296 max_nest = nest; 7297 } 7298 7299 return max_nest + 1; 7300 } 7301 EXPORT_SYMBOL(dev_get_nest_level); 7302 7303 /** 7304 * netdev_lower_change - Dispatch event about lower device state change 7305 * @lower_dev: device 7306 * @lower_state_info: state to dispatch 7307 * 7308 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 7309 * The caller must hold the RTNL lock. 7310 */ 7311 void netdev_lower_state_changed(struct net_device *lower_dev, 7312 void *lower_state_info) 7313 { 7314 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 7315 .info.dev = lower_dev, 7316 }; 7317 7318 ASSERT_RTNL(); 7319 changelowerstate_info.lower_state_info = lower_state_info; 7320 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 7321 &changelowerstate_info.info); 7322 } 7323 EXPORT_SYMBOL(netdev_lower_state_changed); 7324 7325 static void dev_change_rx_flags(struct net_device *dev, int flags) 7326 { 7327 const struct net_device_ops *ops = dev->netdev_ops; 7328 7329 if (ops->ndo_change_rx_flags) 7330 ops->ndo_change_rx_flags(dev, flags); 7331 } 7332 7333 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 7334 { 7335 unsigned int old_flags = dev->flags; 7336 kuid_t uid; 7337 kgid_t gid; 7338 7339 ASSERT_RTNL(); 7340 7341 dev->flags |= IFF_PROMISC; 7342 dev->promiscuity += inc; 7343 if (dev->promiscuity == 0) { 7344 /* 7345 * Avoid overflow. 7346 * If inc causes overflow, untouch promisc and return error. 7347 */ 7348 if (inc < 0) 7349 dev->flags &= ~IFF_PROMISC; 7350 else { 7351 dev->promiscuity -= inc; 7352 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 7353 dev->name); 7354 return -EOVERFLOW; 7355 } 7356 } 7357 if (dev->flags != old_flags) { 7358 pr_info("device %s %s promiscuous mode\n", 7359 dev->name, 7360 dev->flags & IFF_PROMISC ? "entered" : "left"); 7361 if (audit_enabled) { 7362 current_uid_gid(&uid, &gid); 7363 audit_log(audit_context(), GFP_ATOMIC, 7364 AUDIT_ANOM_PROMISCUOUS, 7365 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 7366 dev->name, (dev->flags & IFF_PROMISC), 7367 (old_flags & IFF_PROMISC), 7368 from_kuid(&init_user_ns, audit_get_loginuid(current)), 7369 from_kuid(&init_user_ns, uid), 7370 from_kgid(&init_user_ns, gid), 7371 audit_get_sessionid(current)); 7372 } 7373 7374 dev_change_rx_flags(dev, IFF_PROMISC); 7375 } 7376 if (notify) 7377 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 7378 return 0; 7379 } 7380 7381 /** 7382 * dev_set_promiscuity - update promiscuity count on a device 7383 * @dev: device 7384 * @inc: modifier 7385 * 7386 * Add or remove promiscuity from a device. While the count in the device 7387 * remains above zero the interface remains promiscuous. Once it hits zero 7388 * the device reverts back to normal filtering operation. A negative inc 7389 * value is used to drop promiscuity on the device. 7390 * Return 0 if successful or a negative errno code on error. 7391 */ 7392 int dev_set_promiscuity(struct net_device *dev, int inc) 7393 { 7394 unsigned int old_flags = dev->flags; 7395 int err; 7396 7397 err = __dev_set_promiscuity(dev, inc, true); 7398 if (err < 0) 7399 return err; 7400 if (dev->flags != old_flags) 7401 dev_set_rx_mode(dev); 7402 return err; 7403 } 7404 EXPORT_SYMBOL(dev_set_promiscuity); 7405 7406 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 7407 { 7408 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 7409 7410 ASSERT_RTNL(); 7411 7412 dev->flags |= IFF_ALLMULTI; 7413 dev->allmulti += inc; 7414 if (dev->allmulti == 0) { 7415 /* 7416 * Avoid overflow. 7417 * If inc causes overflow, untouch allmulti and return error. 7418 */ 7419 if (inc < 0) 7420 dev->flags &= ~IFF_ALLMULTI; 7421 else { 7422 dev->allmulti -= inc; 7423 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 7424 dev->name); 7425 return -EOVERFLOW; 7426 } 7427 } 7428 if (dev->flags ^ old_flags) { 7429 dev_change_rx_flags(dev, IFF_ALLMULTI); 7430 dev_set_rx_mode(dev); 7431 if (notify) 7432 __dev_notify_flags(dev, old_flags, 7433 dev->gflags ^ old_gflags); 7434 } 7435 return 0; 7436 } 7437 7438 /** 7439 * dev_set_allmulti - update allmulti count on a device 7440 * @dev: device 7441 * @inc: modifier 7442 * 7443 * Add or remove reception of all multicast frames to a device. While the 7444 * count in the device remains above zero the interface remains listening 7445 * to all interfaces. Once it hits zero the device reverts back to normal 7446 * filtering operation. A negative @inc value is used to drop the counter 7447 * when releasing a resource needing all multicasts. 7448 * Return 0 if successful or a negative errno code on error. 7449 */ 7450 7451 int dev_set_allmulti(struct net_device *dev, int inc) 7452 { 7453 return __dev_set_allmulti(dev, inc, true); 7454 } 7455 EXPORT_SYMBOL(dev_set_allmulti); 7456 7457 /* 7458 * Upload unicast and multicast address lists to device and 7459 * configure RX filtering. When the device doesn't support unicast 7460 * filtering it is put in promiscuous mode while unicast addresses 7461 * are present. 7462 */ 7463 void __dev_set_rx_mode(struct net_device *dev) 7464 { 7465 const struct net_device_ops *ops = dev->netdev_ops; 7466 7467 /* dev_open will call this function so the list will stay sane. */ 7468 if (!(dev->flags&IFF_UP)) 7469 return; 7470 7471 if (!netif_device_present(dev)) 7472 return; 7473 7474 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 7475 /* Unicast addresses changes may only happen under the rtnl, 7476 * therefore calling __dev_set_promiscuity here is safe. 7477 */ 7478 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 7479 __dev_set_promiscuity(dev, 1, false); 7480 dev->uc_promisc = true; 7481 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 7482 __dev_set_promiscuity(dev, -1, false); 7483 dev->uc_promisc = false; 7484 } 7485 } 7486 7487 if (ops->ndo_set_rx_mode) 7488 ops->ndo_set_rx_mode(dev); 7489 } 7490 7491 void dev_set_rx_mode(struct net_device *dev) 7492 { 7493 netif_addr_lock_bh(dev); 7494 __dev_set_rx_mode(dev); 7495 netif_addr_unlock_bh(dev); 7496 } 7497 7498 /** 7499 * dev_get_flags - get flags reported to userspace 7500 * @dev: device 7501 * 7502 * Get the combination of flag bits exported through APIs to userspace. 7503 */ 7504 unsigned int dev_get_flags(const struct net_device *dev) 7505 { 7506 unsigned int flags; 7507 7508 flags = (dev->flags & ~(IFF_PROMISC | 7509 IFF_ALLMULTI | 7510 IFF_RUNNING | 7511 IFF_LOWER_UP | 7512 IFF_DORMANT)) | 7513 (dev->gflags & (IFF_PROMISC | 7514 IFF_ALLMULTI)); 7515 7516 if (netif_running(dev)) { 7517 if (netif_oper_up(dev)) 7518 flags |= IFF_RUNNING; 7519 if (netif_carrier_ok(dev)) 7520 flags |= IFF_LOWER_UP; 7521 if (netif_dormant(dev)) 7522 flags |= IFF_DORMANT; 7523 } 7524 7525 return flags; 7526 } 7527 EXPORT_SYMBOL(dev_get_flags); 7528 7529 int __dev_change_flags(struct net_device *dev, unsigned int flags, 7530 struct netlink_ext_ack *extack) 7531 { 7532 unsigned int old_flags = dev->flags; 7533 int ret; 7534 7535 ASSERT_RTNL(); 7536 7537 /* 7538 * Set the flags on our device. 7539 */ 7540 7541 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 7542 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 7543 IFF_AUTOMEDIA)) | 7544 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 7545 IFF_ALLMULTI)); 7546 7547 /* 7548 * Load in the correct multicast list now the flags have changed. 7549 */ 7550 7551 if ((old_flags ^ flags) & IFF_MULTICAST) 7552 dev_change_rx_flags(dev, IFF_MULTICAST); 7553 7554 dev_set_rx_mode(dev); 7555 7556 /* 7557 * Have we downed the interface. We handle IFF_UP ourselves 7558 * according to user attempts to set it, rather than blindly 7559 * setting it. 7560 */ 7561 7562 ret = 0; 7563 if ((old_flags ^ flags) & IFF_UP) { 7564 if (old_flags & IFF_UP) 7565 __dev_close(dev); 7566 else 7567 ret = __dev_open(dev, extack); 7568 } 7569 7570 if ((flags ^ dev->gflags) & IFF_PROMISC) { 7571 int inc = (flags & IFF_PROMISC) ? 1 : -1; 7572 unsigned int old_flags = dev->flags; 7573 7574 dev->gflags ^= IFF_PROMISC; 7575 7576 if (__dev_set_promiscuity(dev, inc, false) >= 0) 7577 if (dev->flags != old_flags) 7578 dev_set_rx_mode(dev); 7579 } 7580 7581 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 7582 * is important. Some (broken) drivers set IFF_PROMISC, when 7583 * IFF_ALLMULTI is requested not asking us and not reporting. 7584 */ 7585 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 7586 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 7587 7588 dev->gflags ^= IFF_ALLMULTI; 7589 __dev_set_allmulti(dev, inc, false); 7590 } 7591 7592 return ret; 7593 } 7594 7595 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 7596 unsigned int gchanges) 7597 { 7598 unsigned int changes = dev->flags ^ old_flags; 7599 7600 if (gchanges) 7601 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 7602 7603 if (changes & IFF_UP) { 7604 if (dev->flags & IFF_UP) 7605 call_netdevice_notifiers(NETDEV_UP, dev); 7606 else 7607 call_netdevice_notifiers(NETDEV_DOWN, dev); 7608 } 7609 7610 if (dev->flags & IFF_UP && 7611 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 7612 struct netdev_notifier_change_info change_info = { 7613 .info = { 7614 .dev = dev, 7615 }, 7616 .flags_changed = changes, 7617 }; 7618 7619 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 7620 } 7621 } 7622 7623 /** 7624 * dev_change_flags - change device settings 7625 * @dev: device 7626 * @flags: device state flags 7627 * @extack: netlink extended ack 7628 * 7629 * Change settings on device based state flags. The flags are 7630 * in the userspace exported format. 7631 */ 7632 int dev_change_flags(struct net_device *dev, unsigned int flags, 7633 struct netlink_ext_ack *extack) 7634 { 7635 int ret; 7636 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 7637 7638 ret = __dev_change_flags(dev, flags, extack); 7639 if (ret < 0) 7640 return ret; 7641 7642 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 7643 __dev_notify_flags(dev, old_flags, changes); 7644 return ret; 7645 } 7646 EXPORT_SYMBOL(dev_change_flags); 7647 7648 int __dev_set_mtu(struct net_device *dev, int new_mtu) 7649 { 7650 const struct net_device_ops *ops = dev->netdev_ops; 7651 7652 if (ops->ndo_change_mtu) 7653 return ops->ndo_change_mtu(dev, new_mtu); 7654 7655 dev->mtu = new_mtu; 7656 return 0; 7657 } 7658 EXPORT_SYMBOL(__dev_set_mtu); 7659 7660 /** 7661 * dev_set_mtu_ext - Change maximum transfer unit 7662 * @dev: device 7663 * @new_mtu: new transfer unit 7664 * @extack: netlink extended ack 7665 * 7666 * Change the maximum transfer size of the network device. 7667 */ 7668 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 7669 struct netlink_ext_ack *extack) 7670 { 7671 int err, orig_mtu; 7672 7673 if (new_mtu == dev->mtu) 7674 return 0; 7675 7676 /* MTU must be positive, and in range */ 7677 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 7678 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 7679 return -EINVAL; 7680 } 7681 7682 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 7683 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 7684 return -EINVAL; 7685 } 7686 7687 if (!netif_device_present(dev)) 7688 return -ENODEV; 7689 7690 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 7691 err = notifier_to_errno(err); 7692 if (err) 7693 return err; 7694 7695 orig_mtu = dev->mtu; 7696 err = __dev_set_mtu(dev, new_mtu); 7697 7698 if (!err) { 7699 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 7700 orig_mtu); 7701 err = notifier_to_errno(err); 7702 if (err) { 7703 /* setting mtu back and notifying everyone again, 7704 * so that they have a chance to revert changes. 7705 */ 7706 __dev_set_mtu(dev, orig_mtu); 7707 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 7708 new_mtu); 7709 } 7710 } 7711 return err; 7712 } 7713 7714 int dev_set_mtu(struct net_device *dev, int new_mtu) 7715 { 7716 struct netlink_ext_ack extack; 7717 int err; 7718 7719 memset(&extack, 0, sizeof(extack)); 7720 err = dev_set_mtu_ext(dev, new_mtu, &extack); 7721 if (err && extack._msg) 7722 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 7723 return err; 7724 } 7725 EXPORT_SYMBOL(dev_set_mtu); 7726 7727 /** 7728 * dev_change_tx_queue_len - Change TX queue length of a netdevice 7729 * @dev: device 7730 * @new_len: new tx queue length 7731 */ 7732 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 7733 { 7734 unsigned int orig_len = dev->tx_queue_len; 7735 int res; 7736 7737 if (new_len != (unsigned int)new_len) 7738 return -ERANGE; 7739 7740 if (new_len != orig_len) { 7741 dev->tx_queue_len = new_len; 7742 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 7743 res = notifier_to_errno(res); 7744 if (res) 7745 goto err_rollback; 7746 res = dev_qdisc_change_tx_queue_len(dev); 7747 if (res) 7748 goto err_rollback; 7749 } 7750 7751 return 0; 7752 7753 err_rollback: 7754 netdev_err(dev, "refused to change device tx_queue_len\n"); 7755 dev->tx_queue_len = orig_len; 7756 return res; 7757 } 7758 7759 /** 7760 * dev_set_group - Change group this device belongs to 7761 * @dev: device 7762 * @new_group: group this device should belong to 7763 */ 7764 void dev_set_group(struct net_device *dev, int new_group) 7765 { 7766 dev->group = new_group; 7767 } 7768 EXPORT_SYMBOL(dev_set_group); 7769 7770 /** 7771 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 7772 * @dev: device 7773 * @addr: new address 7774 * @extack: netlink extended ack 7775 */ 7776 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 7777 struct netlink_ext_ack *extack) 7778 { 7779 struct netdev_notifier_pre_changeaddr_info info = { 7780 .info.dev = dev, 7781 .info.extack = extack, 7782 .dev_addr = addr, 7783 }; 7784 int rc; 7785 7786 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 7787 return notifier_to_errno(rc); 7788 } 7789 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 7790 7791 /** 7792 * dev_set_mac_address - Change Media Access Control Address 7793 * @dev: device 7794 * @sa: new address 7795 * @extack: netlink extended ack 7796 * 7797 * Change the hardware (MAC) address of the device 7798 */ 7799 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 7800 struct netlink_ext_ack *extack) 7801 { 7802 const struct net_device_ops *ops = dev->netdev_ops; 7803 int err; 7804 7805 if (!ops->ndo_set_mac_address) 7806 return -EOPNOTSUPP; 7807 if (sa->sa_family != dev->type) 7808 return -EINVAL; 7809 if (!netif_device_present(dev)) 7810 return -ENODEV; 7811 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 7812 if (err) 7813 return err; 7814 err = ops->ndo_set_mac_address(dev, sa); 7815 if (err) 7816 return err; 7817 dev->addr_assign_type = NET_ADDR_SET; 7818 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 7819 add_device_randomness(dev->dev_addr, dev->addr_len); 7820 return 0; 7821 } 7822 EXPORT_SYMBOL(dev_set_mac_address); 7823 7824 /** 7825 * dev_change_carrier - Change device carrier 7826 * @dev: device 7827 * @new_carrier: new value 7828 * 7829 * Change device carrier 7830 */ 7831 int dev_change_carrier(struct net_device *dev, bool new_carrier) 7832 { 7833 const struct net_device_ops *ops = dev->netdev_ops; 7834 7835 if (!ops->ndo_change_carrier) 7836 return -EOPNOTSUPP; 7837 if (!netif_device_present(dev)) 7838 return -ENODEV; 7839 return ops->ndo_change_carrier(dev, new_carrier); 7840 } 7841 EXPORT_SYMBOL(dev_change_carrier); 7842 7843 /** 7844 * dev_get_phys_port_id - Get device physical port ID 7845 * @dev: device 7846 * @ppid: port ID 7847 * 7848 * Get device physical port ID 7849 */ 7850 int dev_get_phys_port_id(struct net_device *dev, 7851 struct netdev_phys_item_id *ppid) 7852 { 7853 const struct net_device_ops *ops = dev->netdev_ops; 7854 7855 if (!ops->ndo_get_phys_port_id) 7856 return -EOPNOTSUPP; 7857 return ops->ndo_get_phys_port_id(dev, ppid); 7858 } 7859 EXPORT_SYMBOL(dev_get_phys_port_id); 7860 7861 /** 7862 * dev_get_phys_port_name - Get device physical port name 7863 * @dev: device 7864 * @name: port name 7865 * @len: limit of bytes to copy to name 7866 * 7867 * Get device physical port name 7868 */ 7869 int dev_get_phys_port_name(struct net_device *dev, 7870 char *name, size_t len) 7871 { 7872 const struct net_device_ops *ops = dev->netdev_ops; 7873 7874 if (!ops->ndo_get_phys_port_name) 7875 return -EOPNOTSUPP; 7876 return ops->ndo_get_phys_port_name(dev, name, len); 7877 } 7878 EXPORT_SYMBOL(dev_get_phys_port_name); 7879 7880 /** 7881 * dev_change_proto_down - update protocol port state information 7882 * @dev: device 7883 * @proto_down: new value 7884 * 7885 * This info can be used by switch drivers to set the phys state of the 7886 * port. 7887 */ 7888 int dev_change_proto_down(struct net_device *dev, bool proto_down) 7889 { 7890 const struct net_device_ops *ops = dev->netdev_ops; 7891 7892 if (!ops->ndo_change_proto_down) 7893 return -EOPNOTSUPP; 7894 if (!netif_device_present(dev)) 7895 return -ENODEV; 7896 return ops->ndo_change_proto_down(dev, proto_down); 7897 } 7898 EXPORT_SYMBOL(dev_change_proto_down); 7899 7900 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op, 7901 enum bpf_netdev_command cmd) 7902 { 7903 struct netdev_bpf xdp; 7904 7905 if (!bpf_op) 7906 return 0; 7907 7908 memset(&xdp, 0, sizeof(xdp)); 7909 xdp.command = cmd; 7910 7911 /* Query must always succeed. */ 7912 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG); 7913 7914 return xdp.prog_id; 7915 } 7916 7917 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, 7918 struct netlink_ext_ack *extack, u32 flags, 7919 struct bpf_prog *prog) 7920 { 7921 struct netdev_bpf xdp; 7922 7923 memset(&xdp, 0, sizeof(xdp)); 7924 if (flags & XDP_FLAGS_HW_MODE) 7925 xdp.command = XDP_SETUP_PROG_HW; 7926 else 7927 xdp.command = XDP_SETUP_PROG; 7928 xdp.extack = extack; 7929 xdp.flags = flags; 7930 xdp.prog = prog; 7931 7932 return bpf_op(dev, &xdp); 7933 } 7934 7935 static void dev_xdp_uninstall(struct net_device *dev) 7936 { 7937 struct netdev_bpf xdp; 7938 bpf_op_t ndo_bpf; 7939 7940 /* Remove generic XDP */ 7941 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL)); 7942 7943 /* Remove from the driver */ 7944 ndo_bpf = dev->netdev_ops->ndo_bpf; 7945 if (!ndo_bpf) 7946 return; 7947 7948 memset(&xdp, 0, sizeof(xdp)); 7949 xdp.command = XDP_QUERY_PROG; 7950 WARN_ON(ndo_bpf(dev, &xdp)); 7951 if (xdp.prog_id) 7952 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, 7953 NULL)); 7954 7955 /* Remove HW offload */ 7956 memset(&xdp, 0, sizeof(xdp)); 7957 xdp.command = XDP_QUERY_PROG_HW; 7958 if (!ndo_bpf(dev, &xdp) && xdp.prog_id) 7959 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, 7960 NULL)); 7961 } 7962 7963 /** 7964 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 7965 * @dev: device 7966 * @extack: netlink extended ack 7967 * @fd: new program fd or negative value to clear 7968 * @flags: xdp-related flags 7969 * 7970 * Set or clear a bpf program for a device 7971 */ 7972 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 7973 int fd, u32 flags) 7974 { 7975 const struct net_device_ops *ops = dev->netdev_ops; 7976 enum bpf_netdev_command query; 7977 struct bpf_prog *prog = NULL; 7978 bpf_op_t bpf_op, bpf_chk; 7979 int err; 7980 7981 ASSERT_RTNL(); 7982 7983 query = flags & XDP_FLAGS_HW_MODE ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG; 7984 7985 bpf_op = bpf_chk = ops->ndo_bpf; 7986 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) 7987 return -EOPNOTSUPP; 7988 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) 7989 bpf_op = generic_xdp_install; 7990 if (bpf_op == bpf_chk) 7991 bpf_chk = generic_xdp_install; 7992 7993 if (fd >= 0) { 7994 if (__dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG) || 7995 __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG_HW)) 7996 return -EEXIST; 7997 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && 7998 __dev_xdp_query(dev, bpf_op, query)) 7999 return -EBUSY; 8000 8001 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 8002 bpf_op == ops->ndo_bpf); 8003 if (IS_ERR(prog)) 8004 return PTR_ERR(prog); 8005 8006 if (!(flags & XDP_FLAGS_HW_MODE) && 8007 bpf_prog_is_dev_bound(prog->aux)) { 8008 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported"); 8009 bpf_prog_put(prog); 8010 return -EINVAL; 8011 } 8012 } 8013 8014 err = dev_xdp_install(dev, bpf_op, extack, flags, prog); 8015 if (err < 0 && prog) 8016 bpf_prog_put(prog); 8017 8018 return err; 8019 } 8020 8021 /** 8022 * dev_new_index - allocate an ifindex 8023 * @net: the applicable net namespace 8024 * 8025 * Returns a suitable unique value for a new device interface 8026 * number. The caller must hold the rtnl semaphore or the 8027 * dev_base_lock to be sure it remains unique. 8028 */ 8029 static int dev_new_index(struct net *net) 8030 { 8031 int ifindex = net->ifindex; 8032 8033 for (;;) { 8034 if (++ifindex <= 0) 8035 ifindex = 1; 8036 if (!__dev_get_by_index(net, ifindex)) 8037 return net->ifindex = ifindex; 8038 } 8039 } 8040 8041 /* Delayed registration/unregisteration */ 8042 static LIST_HEAD(net_todo_list); 8043 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 8044 8045 static void net_set_todo(struct net_device *dev) 8046 { 8047 list_add_tail(&dev->todo_list, &net_todo_list); 8048 dev_net(dev)->dev_unreg_count++; 8049 } 8050 8051 static void rollback_registered_many(struct list_head *head) 8052 { 8053 struct net_device *dev, *tmp; 8054 LIST_HEAD(close_head); 8055 8056 BUG_ON(dev_boot_phase); 8057 ASSERT_RTNL(); 8058 8059 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 8060 /* Some devices call without registering 8061 * for initialization unwind. Remove those 8062 * devices and proceed with the remaining. 8063 */ 8064 if (dev->reg_state == NETREG_UNINITIALIZED) { 8065 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 8066 dev->name, dev); 8067 8068 WARN_ON(1); 8069 list_del(&dev->unreg_list); 8070 continue; 8071 } 8072 dev->dismantle = true; 8073 BUG_ON(dev->reg_state != NETREG_REGISTERED); 8074 } 8075 8076 /* If device is running, close it first. */ 8077 list_for_each_entry(dev, head, unreg_list) 8078 list_add_tail(&dev->close_list, &close_head); 8079 dev_close_many(&close_head, true); 8080 8081 list_for_each_entry(dev, head, unreg_list) { 8082 /* And unlink it from device chain. */ 8083 unlist_netdevice(dev); 8084 8085 dev->reg_state = NETREG_UNREGISTERING; 8086 } 8087 flush_all_backlogs(); 8088 8089 synchronize_net(); 8090 8091 list_for_each_entry(dev, head, unreg_list) { 8092 struct sk_buff *skb = NULL; 8093 8094 /* Shutdown queueing discipline. */ 8095 dev_shutdown(dev); 8096 8097 dev_xdp_uninstall(dev); 8098 8099 /* Notify protocols, that we are about to destroy 8100 * this device. They should clean all the things. 8101 */ 8102 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 8103 8104 if (!dev->rtnl_link_ops || 8105 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 8106 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 8107 GFP_KERNEL, NULL, 0); 8108 8109 /* 8110 * Flush the unicast and multicast chains 8111 */ 8112 dev_uc_flush(dev); 8113 dev_mc_flush(dev); 8114 8115 if (dev->netdev_ops->ndo_uninit) 8116 dev->netdev_ops->ndo_uninit(dev); 8117 8118 if (skb) 8119 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 8120 8121 /* Notifier chain MUST detach us all upper devices. */ 8122 WARN_ON(netdev_has_any_upper_dev(dev)); 8123 WARN_ON(netdev_has_any_lower_dev(dev)); 8124 8125 /* Remove entries from kobject tree */ 8126 netdev_unregister_kobject(dev); 8127 #ifdef CONFIG_XPS 8128 /* Remove XPS queueing entries */ 8129 netif_reset_xps_queues_gt(dev, 0); 8130 #endif 8131 } 8132 8133 synchronize_net(); 8134 8135 list_for_each_entry(dev, head, unreg_list) 8136 dev_put(dev); 8137 } 8138 8139 static void rollback_registered(struct net_device *dev) 8140 { 8141 LIST_HEAD(single); 8142 8143 list_add(&dev->unreg_list, &single); 8144 rollback_registered_many(&single); 8145 list_del(&single); 8146 } 8147 8148 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 8149 struct net_device *upper, netdev_features_t features) 8150 { 8151 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 8152 netdev_features_t feature; 8153 int feature_bit; 8154 8155 for_each_netdev_feature(&upper_disables, feature_bit) { 8156 feature = __NETIF_F_BIT(feature_bit); 8157 if (!(upper->wanted_features & feature) 8158 && (features & feature)) { 8159 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 8160 &feature, upper->name); 8161 features &= ~feature; 8162 } 8163 } 8164 8165 return features; 8166 } 8167 8168 static void netdev_sync_lower_features(struct net_device *upper, 8169 struct net_device *lower, netdev_features_t features) 8170 { 8171 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 8172 netdev_features_t feature; 8173 int feature_bit; 8174 8175 for_each_netdev_feature(&upper_disables, feature_bit) { 8176 feature = __NETIF_F_BIT(feature_bit); 8177 if (!(features & feature) && (lower->features & feature)) { 8178 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 8179 &feature, lower->name); 8180 lower->wanted_features &= ~feature; 8181 netdev_update_features(lower); 8182 8183 if (unlikely(lower->features & feature)) 8184 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 8185 &feature, lower->name); 8186 } 8187 } 8188 } 8189 8190 static netdev_features_t netdev_fix_features(struct net_device *dev, 8191 netdev_features_t features) 8192 { 8193 /* Fix illegal checksum combinations */ 8194 if ((features & NETIF_F_HW_CSUM) && 8195 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 8196 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 8197 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 8198 } 8199 8200 /* TSO requires that SG is present as well. */ 8201 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 8202 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 8203 features &= ~NETIF_F_ALL_TSO; 8204 } 8205 8206 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 8207 !(features & NETIF_F_IP_CSUM)) { 8208 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 8209 features &= ~NETIF_F_TSO; 8210 features &= ~NETIF_F_TSO_ECN; 8211 } 8212 8213 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 8214 !(features & NETIF_F_IPV6_CSUM)) { 8215 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 8216 features &= ~NETIF_F_TSO6; 8217 } 8218 8219 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 8220 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 8221 features &= ~NETIF_F_TSO_MANGLEID; 8222 8223 /* TSO ECN requires that TSO is present as well. */ 8224 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 8225 features &= ~NETIF_F_TSO_ECN; 8226 8227 /* Software GSO depends on SG. */ 8228 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 8229 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 8230 features &= ~NETIF_F_GSO; 8231 } 8232 8233 /* GSO partial features require GSO partial be set */ 8234 if ((features & dev->gso_partial_features) && 8235 !(features & NETIF_F_GSO_PARTIAL)) { 8236 netdev_dbg(dev, 8237 "Dropping partially supported GSO features since no GSO partial.\n"); 8238 features &= ~dev->gso_partial_features; 8239 } 8240 8241 if (!(features & NETIF_F_RXCSUM)) { 8242 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 8243 * successfully merged by hardware must also have the 8244 * checksum verified by hardware. If the user does not 8245 * want to enable RXCSUM, logically, we should disable GRO_HW. 8246 */ 8247 if (features & NETIF_F_GRO_HW) { 8248 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 8249 features &= ~NETIF_F_GRO_HW; 8250 } 8251 } 8252 8253 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 8254 if (features & NETIF_F_RXFCS) { 8255 if (features & NETIF_F_LRO) { 8256 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 8257 features &= ~NETIF_F_LRO; 8258 } 8259 8260 if (features & NETIF_F_GRO_HW) { 8261 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 8262 features &= ~NETIF_F_GRO_HW; 8263 } 8264 } 8265 8266 return features; 8267 } 8268 8269 int __netdev_update_features(struct net_device *dev) 8270 { 8271 struct net_device *upper, *lower; 8272 netdev_features_t features; 8273 struct list_head *iter; 8274 int err = -1; 8275 8276 ASSERT_RTNL(); 8277 8278 features = netdev_get_wanted_features(dev); 8279 8280 if (dev->netdev_ops->ndo_fix_features) 8281 features = dev->netdev_ops->ndo_fix_features(dev, features); 8282 8283 /* driver might be less strict about feature dependencies */ 8284 features = netdev_fix_features(dev, features); 8285 8286 /* some features can't be enabled if they're off an an upper device */ 8287 netdev_for_each_upper_dev_rcu(dev, upper, iter) 8288 features = netdev_sync_upper_features(dev, upper, features); 8289 8290 if (dev->features == features) 8291 goto sync_lower; 8292 8293 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 8294 &dev->features, &features); 8295 8296 if (dev->netdev_ops->ndo_set_features) 8297 err = dev->netdev_ops->ndo_set_features(dev, features); 8298 else 8299 err = 0; 8300 8301 if (unlikely(err < 0)) { 8302 netdev_err(dev, 8303 "set_features() failed (%d); wanted %pNF, left %pNF\n", 8304 err, &features, &dev->features); 8305 /* return non-0 since some features might have changed and 8306 * it's better to fire a spurious notification than miss it 8307 */ 8308 return -1; 8309 } 8310 8311 sync_lower: 8312 /* some features must be disabled on lower devices when disabled 8313 * on an upper device (think: bonding master or bridge) 8314 */ 8315 netdev_for_each_lower_dev(dev, lower, iter) 8316 netdev_sync_lower_features(dev, lower, features); 8317 8318 if (!err) { 8319 netdev_features_t diff = features ^ dev->features; 8320 8321 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 8322 /* udp_tunnel_{get,drop}_rx_info both need 8323 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 8324 * device, or they won't do anything. 8325 * Thus we need to update dev->features 8326 * *before* calling udp_tunnel_get_rx_info, 8327 * but *after* calling udp_tunnel_drop_rx_info. 8328 */ 8329 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 8330 dev->features = features; 8331 udp_tunnel_get_rx_info(dev); 8332 } else { 8333 udp_tunnel_drop_rx_info(dev); 8334 } 8335 } 8336 8337 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 8338 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 8339 dev->features = features; 8340 err |= vlan_get_rx_ctag_filter_info(dev); 8341 } else { 8342 vlan_drop_rx_ctag_filter_info(dev); 8343 } 8344 } 8345 8346 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 8347 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 8348 dev->features = features; 8349 err |= vlan_get_rx_stag_filter_info(dev); 8350 } else { 8351 vlan_drop_rx_stag_filter_info(dev); 8352 } 8353 } 8354 8355 dev->features = features; 8356 } 8357 8358 return err < 0 ? 0 : 1; 8359 } 8360 8361 /** 8362 * netdev_update_features - recalculate device features 8363 * @dev: the device to check 8364 * 8365 * Recalculate dev->features set and send notifications if it 8366 * has changed. Should be called after driver or hardware dependent 8367 * conditions might have changed that influence the features. 8368 */ 8369 void netdev_update_features(struct net_device *dev) 8370 { 8371 if (__netdev_update_features(dev)) 8372 netdev_features_change(dev); 8373 } 8374 EXPORT_SYMBOL(netdev_update_features); 8375 8376 /** 8377 * netdev_change_features - recalculate device features 8378 * @dev: the device to check 8379 * 8380 * Recalculate dev->features set and send notifications even 8381 * if they have not changed. Should be called instead of 8382 * netdev_update_features() if also dev->vlan_features might 8383 * have changed to allow the changes to be propagated to stacked 8384 * VLAN devices. 8385 */ 8386 void netdev_change_features(struct net_device *dev) 8387 { 8388 __netdev_update_features(dev); 8389 netdev_features_change(dev); 8390 } 8391 EXPORT_SYMBOL(netdev_change_features); 8392 8393 /** 8394 * netif_stacked_transfer_operstate - transfer operstate 8395 * @rootdev: the root or lower level device to transfer state from 8396 * @dev: the device to transfer operstate to 8397 * 8398 * Transfer operational state from root to device. This is normally 8399 * called when a stacking relationship exists between the root 8400 * device and the device(a leaf device). 8401 */ 8402 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 8403 struct net_device *dev) 8404 { 8405 if (rootdev->operstate == IF_OPER_DORMANT) 8406 netif_dormant_on(dev); 8407 else 8408 netif_dormant_off(dev); 8409 8410 if (netif_carrier_ok(rootdev)) 8411 netif_carrier_on(dev); 8412 else 8413 netif_carrier_off(dev); 8414 } 8415 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 8416 8417 static int netif_alloc_rx_queues(struct net_device *dev) 8418 { 8419 unsigned int i, count = dev->num_rx_queues; 8420 struct netdev_rx_queue *rx; 8421 size_t sz = count * sizeof(*rx); 8422 int err = 0; 8423 8424 BUG_ON(count < 1); 8425 8426 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 8427 if (!rx) 8428 return -ENOMEM; 8429 8430 dev->_rx = rx; 8431 8432 for (i = 0; i < count; i++) { 8433 rx[i].dev = dev; 8434 8435 /* XDP RX-queue setup */ 8436 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i); 8437 if (err < 0) 8438 goto err_rxq_info; 8439 } 8440 return 0; 8441 8442 err_rxq_info: 8443 /* Rollback successful reg's and free other resources */ 8444 while (i--) 8445 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 8446 kvfree(dev->_rx); 8447 dev->_rx = NULL; 8448 return err; 8449 } 8450 8451 static void netif_free_rx_queues(struct net_device *dev) 8452 { 8453 unsigned int i, count = dev->num_rx_queues; 8454 8455 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 8456 if (!dev->_rx) 8457 return; 8458 8459 for (i = 0; i < count; i++) 8460 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 8461 8462 kvfree(dev->_rx); 8463 } 8464 8465 static void netdev_init_one_queue(struct net_device *dev, 8466 struct netdev_queue *queue, void *_unused) 8467 { 8468 /* Initialize queue lock */ 8469 spin_lock_init(&queue->_xmit_lock); 8470 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 8471 queue->xmit_lock_owner = -1; 8472 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 8473 queue->dev = dev; 8474 #ifdef CONFIG_BQL 8475 dql_init(&queue->dql, HZ); 8476 #endif 8477 } 8478 8479 static void netif_free_tx_queues(struct net_device *dev) 8480 { 8481 kvfree(dev->_tx); 8482 } 8483 8484 static int netif_alloc_netdev_queues(struct net_device *dev) 8485 { 8486 unsigned int count = dev->num_tx_queues; 8487 struct netdev_queue *tx; 8488 size_t sz = count * sizeof(*tx); 8489 8490 if (count < 1 || count > 0xffff) 8491 return -EINVAL; 8492 8493 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 8494 if (!tx) 8495 return -ENOMEM; 8496 8497 dev->_tx = tx; 8498 8499 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 8500 spin_lock_init(&dev->tx_global_lock); 8501 8502 return 0; 8503 } 8504 8505 void netif_tx_stop_all_queues(struct net_device *dev) 8506 { 8507 unsigned int i; 8508 8509 for (i = 0; i < dev->num_tx_queues; i++) { 8510 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 8511 8512 netif_tx_stop_queue(txq); 8513 } 8514 } 8515 EXPORT_SYMBOL(netif_tx_stop_all_queues); 8516 8517 /** 8518 * register_netdevice - register a network device 8519 * @dev: device to register 8520 * 8521 * Take a completed network device structure and add it to the kernel 8522 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 8523 * chain. 0 is returned on success. A negative errno code is returned 8524 * on a failure to set up the device, or if the name is a duplicate. 8525 * 8526 * Callers must hold the rtnl semaphore. You may want 8527 * register_netdev() instead of this. 8528 * 8529 * BUGS: 8530 * The locking appears insufficient to guarantee two parallel registers 8531 * will not get the same name. 8532 */ 8533 8534 int register_netdevice(struct net_device *dev) 8535 { 8536 int ret; 8537 struct net *net = dev_net(dev); 8538 8539 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 8540 NETDEV_FEATURE_COUNT); 8541 BUG_ON(dev_boot_phase); 8542 ASSERT_RTNL(); 8543 8544 might_sleep(); 8545 8546 /* When net_device's are persistent, this will be fatal. */ 8547 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 8548 BUG_ON(!net); 8549 8550 spin_lock_init(&dev->addr_list_lock); 8551 netdev_set_addr_lockdep_class(dev); 8552 8553 ret = dev_get_valid_name(net, dev, dev->name); 8554 if (ret < 0) 8555 goto out; 8556 8557 /* Init, if this function is available */ 8558 if (dev->netdev_ops->ndo_init) { 8559 ret = dev->netdev_ops->ndo_init(dev); 8560 if (ret) { 8561 if (ret > 0) 8562 ret = -EIO; 8563 goto out; 8564 } 8565 } 8566 8567 if (((dev->hw_features | dev->features) & 8568 NETIF_F_HW_VLAN_CTAG_FILTER) && 8569 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 8570 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 8571 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 8572 ret = -EINVAL; 8573 goto err_uninit; 8574 } 8575 8576 ret = -EBUSY; 8577 if (!dev->ifindex) 8578 dev->ifindex = dev_new_index(net); 8579 else if (__dev_get_by_index(net, dev->ifindex)) 8580 goto err_uninit; 8581 8582 /* Transfer changeable features to wanted_features and enable 8583 * software offloads (GSO and GRO). 8584 */ 8585 dev->hw_features |= NETIF_F_SOFT_FEATURES; 8586 dev->features |= NETIF_F_SOFT_FEATURES; 8587 8588 if (dev->netdev_ops->ndo_udp_tunnel_add) { 8589 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 8590 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 8591 } 8592 8593 dev->wanted_features = dev->features & dev->hw_features; 8594 8595 if (!(dev->flags & IFF_LOOPBACK)) 8596 dev->hw_features |= NETIF_F_NOCACHE_COPY; 8597 8598 /* If IPv4 TCP segmentation offload is supported we should also 8599 * allow the device to enable segmenting the frame with the option 8600 * of ignoring a static IP ID value. This doesn't enable the 8601 * feature itself but allows the user to enable it later. 8602 */ 8603 if (dev->hw_features & NETIF_F_TSO) 8604 dev->hw_features |= NETIF_F_TSO_MANGLEID; 8605 if (dev->vlan_features & NETIF_F_TSO) 8606 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 8607 if (dev->mpls_features & NETIF_F_TSO) 8608 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 8609 if (dev->hw_enc_features & NETIF_F_TSO) 8610 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 8611 8612 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 8613 */ 8614 dev->vlan_features |= NETIF_F_HIGHDMA; 8615 8616 /* Make NETIF_F_SG inheritable to tunnel devices. 8617 */ 8618 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 8619 8620 /* Make NETIF_F_SG inheritable to MPLS. 8621 */ 8622 dev->mpls_features |= NETIF_F_SG; 8623 8624 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 8625 ret = notifier_to_errno(ret); 8626 if (ret) 8627 goto err_uninit; 8628 8629 ret = netdev_register_kobject(dev); 8630 if (ret) 8631 goto err_uninit; 8632 dev->reg_state = NETREG_REGISTERED; 8633 8634 __netdev_update_features(dev); 8635 8636 /* 8637 * Default initial state at registry is that the 8638 * device is present. 8639 */ 8640 8641 set_bit(__LINK_STATE_PRESENT, &dev->state); 8642 8643 linkwatch_init_dev(dev); 8644 8645 dev_init_scheduler(dev); 8646 dev_hold(dev); 8647 list_netdevice(dev); 8648 add_device_randomness(dev->dev_addr, dev->addr_len); 8649 8650 /* If the device has permanent device address, driver should 8651 * set dev_addr and also addr_assign_type should be set to 8652 * NET_ADDR_PERM (default value). 8653 */ 8654 if (dev->addr_assign_type == NET_ADDR_PERM) 8655 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 8656 8657 /* Notify protocols, that a new device appeared. */ 8658 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 8659 ret = notifier_to_errno(ret); 8660 if (ret) { 8661 rollback_registered(dev); 8662 dev->reg_state = NETREG_UNREGISTERED; 8663 } 8664 /* 8665 * Prevent userspace races by waiting until the network 8666 * device is fully setup before sending notifications. 8667 */ 8668 if (!dev->rtnl_link_ops || 8669 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 8670 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 8671 8672 out: 8673 return ret; 8674 8675 err_uninit: 8676 if (dev->netdev_ops->ndo_uninit) 8677 dev->netdev_ops->ndo_uninit(dev); 8678 if (dev->priv_destructor) 8679 dev->priv_destructor(dev); 8680 goto out; 8681 } 8682 EXPORT_SYMBOL(register_netdevice); 8683 8684 /** 8685 * init_dummy_netdev - init a dummy network device for NAPI 8686 * @dev: device to init 8687 * 8688 * This takes a network device structure and initialize the minimum 8689 * amount of fields so it can be used to schedule NAPI polls without 8690 * registering a full blown interface. This is to be used by drivers 8691 * that need to tie several hardware interfaces to a single NAPI 8692 * poll scheduler due to HW limitations. 8693 */ 8694 int init_dummy_netdev(struct net_device *dev) 8695 { 8696 /* Clear everything. Note we don't initialize spinlocks 8697 * are they aren't supposed to be taken by any of the 8698 * NAPI code and this dummy netdev is supposed to be 8699 * only ever used for NAPI polls 8700 */ 8701 memset(dev, 0, sizeof(struct net_device)); 8702 8703 /* make sure we BUG if trying to hit standard 8704 * register/unregister code path 8705 */ 8706 dev->reg_state = NETREG_DUMMY; 8707 8708 /* NAPI wants this */ 8709 INIT_LIST_HEAD(&dev->napi_list); 8710 8711 /* a dummy interface is started by default */ 8712 set_bit(__LINK_STATE_PRESENT, &dev->state); 8713 set_bit(__LINK_STATE_START, &dev->state); 8714 8715 /* Note : We dont allocate pcpu_refcnt for dummy devices, 8716 * because users of this 'device' dont need to change 8717 * its refcount. 8718 */ 8719 8720 return 0; 8721 } 8722 EXPORT_SYMBOL_GPL(init_dummy_netdev); 8723 8724 8725 /** 8726 * register_netdev - register a network device 8727 * @dev: device to register 8728 * 8729 * Take a completed network device structure and add it to the kernel 8730 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 8731 * chain. 0 is returned on success. A negative errno code is returned 8732 * on a failure to set up the device, or if the name is a duplicate. 8733 * 8734 * This is a wrapper around register_netdevice that takes the rtnl semaphore 8735 * and expands the device name if you passed a format string to 8736 * alloc_netdev. 8737 */ 8738 int register_netdev(struct net_device *dev) 8739 { 8740 int err; 8741 8742 if (rtnl_lock_killable()) 8743 return -EINTR; 8744 err = register_netdevice(dev); 8745 rtnl_unlock(); 8746 return err; 8747 } 8748 EXPORT_SYMBOL(register_netdev); 8749 8750 int netdev_refcnt_read(const struct net_device *dev) 8751 { 8752 int i, refcnt = 0; 8753 8754 for_each_possible_cpu(i) 8755 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 8756 return refcnt; 8757 } 8758 EXPORT_SYMBOL(netdev_refcnt_read); 8759 8760 /** 8761 * netdev_wait_allrefs - wait until all references are gone. 8762 * @dev: target net_device 8763 * 8764 * This is called when unregistering network devices. 8765 * 8766 * Any protocol or device that holds a reference should register 8767 * for netdevice notification, and cleanup and put back the 8768 * reference if they receive an UNREGISTER event. 8769 * We can get stuck here if buggy protocols don't correctly 8770 * call dev_put. 8771 */ 8772 static void netdev_wait_allrefs(struct net_device *dev) 8773 { 8774 unsigned long rebroadcast_time, warning_time; 8775 int refcnt; 8776 8777 linkwatch_forget_dev(dev); 8778 8779 rebroadcast_time = warning_time = jiffies; 8780 refcnt = netdev_refcnt_read(dev); 8781 8782 while (refcnt != 0) { 8783 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 8784 rtnl_lock(); 8785 8786 /* Rebroadcast unregister notification */ 8787 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 8788 8789 __rtnl_unlock(); 8790 rcu_barrier(); 8791 rtnl_lock(); 8792 8793 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 8794 &dev->state)) { 8795 /* We must not have linkwatch events 8796 * pending on unregister. If this 8797 * happens, we simply run the queue 8798 * unscheduled, resulting in a noop 8799 * for this device. 8800 */ 8801 linkwatch_run_queue(); 8802 } 8803 8804 __rtnl_unlock(); 8805 8806 rebroadcast_time = jiffies; 8807 } 8808 8809 msleep(250); 8810 8811 refcnt = netdev_refcnt_read(dev); 8812 8813 if (time_after(jiffies, warning_time + 10 * HZ)) { 8814 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 8815 dev->name, refcnt); 8816 warning_time = jiffies; 8817 } 8818 } 8819 } 8820 8821 /* The sequence is: 8822 * 8823 * rtnl_lock(); 8824 * ... 8825 * register_netdevice(x1); 8826 * register_netdevice(x2); 8827 * ... 8828 * unregister_netdevice(y1); 8829 * unregister_netdevice(y2); 8830 * ... 8831 * rtnl_unlock(); 8832 * free_netdev(y1); 8833 * free_netdev(y2); 8834 * 8835 * We are invoked by rtnl_unlock(). 8836 * This allows us to deal with problems: 8837 * 1) We can delete sysfs objects which invoke hotplug 8838 * without deadlocking with linkwatch via keventd. 8839 * 2) Since we run with the RTNL semaphore not held, we can sleep 8840 * safely in order to wait for the netdev refcnt to drop to zero. 8841 * 8842 * We must not return until all unregister events added during 8843 * the interval the lock was held have been completed. 8844 */ 8845 void netdev_run_todo(void) 8846 { 8847 struct list_head list; 8848 8849 /* Snapshot list, allow later requests */ 8850 list_replace_init(&net_todo_list, &list); 8851 8852 __rtnl_unlock(); 8853 8854 8855 /* Wait for rcu callbacks to finish before next phase */ 8856 if (!list_empty(&list)) 8857 rcu_barrier(); 8858 8859 while (!list_empty(&list)) { 8860 struct net_device *dev 8861 = list_first_entry(&list, struct net_device, todo_list); 8862 list_del(&dev->todo_list); 8863 8864 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 8865 pr_err("network todo '%s' but state %d\n", 8866 dev->name, dev->reg_state); 8867 dump_stack(); 8868 continue; 8869 } 8870 8871 dev->reg_state = NETREG_UNREGISTERED; 8872 8873 netdev_wait_allrefs(dev); 8874 8875 /* paranoia */ 8876 BUG_ON(netdev_refcnt_read(dev)); 8877 BUG_ON(!list_empty(&dev->ptype_all)); 8878 BUG_ON(!list_empty(&dev->ptype_specific)); 8879 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 8880 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 8881 #if IS_ENABLED(CONFIG_DECNET) 8882 WARN_ON(dev->dn_ptr); 8883 #endif 8884 if (dev->priv_destructor) 8885 dev->priv_destructor(dev); 8886 if (dev->needs_free_netdev) 8887 free_netdev(dev); 8888 8889 /* Report a network device has been unregistered */ 8890 rtnl_lock(); 8891 dev_net(dev)->dev_unreg_count--; 8892 __rtnl_unlock(); 8893 wake_up(&netdev_unregistering_wq); 8894 8895 /* Free network device */ 8896 kobject_put(&dev->dev.kobj); 8897 } 8898 } 8899 8900 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 8901 * all the same fields in the same order as net_device_stats, with only 8902 * the type differing, but rtnl_link_stats64 may have additional fields 8903 * at the end for newer counters. 8904 */ 8905 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 8906 const struct net_device_stats *netdev_stats) 8907 { 8908 #if BITS_PER_LONG == 64 8909 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 8910 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 8911 /* zero out counters that only exist in rtnl_link_stats64 */ 8912 memset((char *)stats64 + sizeof(*netdev_stats), 0, 8913 sizeof(*stats64) - sizeof(*netdev_stats)); 8914 #else 8915 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 8916 const unsigned long *src = (const unsigned long *)netdev_stats; 8917 u64 *dst = (u64 *)stats64; 8918 8919 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 8920 for (i = 0; i < n; i++) 8921 dst[i] = src[i]; 8922 /* zero out counters that only exist in rtnl_link_stats64 */ 8923 memset((char *)stats64 + n * sizeof(u64), 0, 8924 sizeof(*stats64) - n * sizeof(u64)); 8925 #endif 8926 } 8927 EXPORT_SYMBOL(netdev_stats_to_stats64); 8928 8929 /** 8930 * dev_get_stats - get network device statistics 8931 * @dev: device to get statistics from 8932 * @storage: place to store stats 8933 * 8934 * Get network statistics from device. Return @storage. 8935 * The device driver may provide its own method by setting 8936 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 8937 * otherwise the internal statistics structure is used. 8938 */ 8939 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 8940 struct rtnl_link_stats64 *storage) 8941 { 8942 const struct net_device_ops *ops = dev->netdev_ops; 8943 8944 if (ops->ndo_get_stats64) { 8945 memset(storage, 0, sizeof(*storage)); 8946 ops->ndo_get_stats64(dev, storage); 8947 } else if (ops->ndo_get_stats) { 8948 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 8949 } else { 8950 netdev_stats_to_stats64(storage, &dev->stats); 8951 } 8952 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 8953 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 8954 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 8955 return storage; 8956 } 8957 EXPORT_SYMBOL(dev_get_stats); 8958 8959 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 8960 { 8961 struct netdev_queue *queue = dev_ingress_queue(dev); 8962 8963 #ifdef CONFIG_NET_CLS_ACT 8964 if (queue) 8965 return queue; 8966 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 8967 if (!queue) 8968 return NULL; 8969 netdev_init_one_queue(dev, queue, NULL); 8970 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 8971 queue->qdisc_sleeping = &noop_qdisc; 8972 rcu_assign_pointer(dev->ingress_queue, queue); 8973 #endif 8974 return queue; 8975 } 8976 8977 static const struct ethtool_ops default_ethtool_ops; 8978 8979 void netdev_set_default_ethtool_ops(struct net_device *dev, 8980 const struct ethtool_ops *ops) 8981 { 8982 if (dev->ethtool_ops == &default_ethtool_ops) 8983 dev->ethtool_ops = ops; 8984 } 8985 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 8986 8987 void netdev_freemem(struct net_device *dev) 8988 { 8989 char *addr = (char *)dev - dev->padded; 8990 8991 kvfree(addr); 8992 } 8993 8994 /** 8995 * alloc_netdev_mqs - allocate network device 8996 * @sizeof_priv: size of private data to allocate space for 8997 * @name: device name format string 8998 * @name_assign_type: origin of device name 8999 * @setup: callback to initialize device 9000 * @txqs: the number of TX subqueues to allocate 9001 * @rxqs: the number of RX subqueues to allocate 9002 * 9003 * Allocates a struct net_device with private data area for driver use 9004 * and performs basic initialization. Also allocates subqueue structs 9005 * for each queue on the device. 9006 */ 9007 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 9008 unsigned char name_assign_type, 9009 void (*setup)(struct net_device *), 9010 unsigned int txqs, unsigned int rxqs) 9011 { 9012 struct net_device *dev; 9013 unsigned int alloc_size; 9014 struct net_device *p; 9015 9016 BUG_ON(strlen(name) >= sizeof(dev->name)); 9017 9018 if (txqs < 1) { 9019 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 9020 return NULL; 9021 } 9022 9023 if (rxqs < 1) { 9024 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 9025 return NULL; 9026 } 9027 9028 alloc_size = sizeof(struct net_device); 9029 if (sizeof_priv) { 9030 /* ensure 32-byte alignment of private area */ 9031 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 9032 alloc_size += sizeof_priv; 9033 } 9034 /* ensure 32-byte alignment of whole construct */ 9035 alloc_size += NETDEV_ALIGN - 1; 9036 9037 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 9038 if (!p) 9039 return NULL; 9040 9041 dev = PTR_ALIGN(p, NETDEV_ALIGN); 9042 dev->padded = (char *)dev - (char *)p; 9043 9044 dev->pcpu_refcnt = alloc_percpu(int); 9045 if (!dev->pcpu_refcnt) 9046 goto free_dev; 9047 9048 if (dev_addr_init(dev)) 9049 goto free_pcpu; 9050 9051 dev_mc_init(dev); 9052 dev_uc_init(dev); 9053 9054 dev_net_set(dev, &init_net); 9055 9056 dev->gso_max_size = GSO_MAX_SIZE; 9057 dev->gso_max_segs = GSO_MAX_SEGS; 9058 9059 INIT_LIST_HEAD(&dev->napi_list); 9060 INIT_LIST_HEAD(&dev->unreg_list); 9061 INIT_LIST_HEAD(&dev->close_list); 9062 INIT_LIST_HEAD(&dev->link_watch_list); 9063 INIT_LIST_HEAD(&dev->adj_list.upper); 9064 INIT_LIST_HEAD(&dev->adj_list.lower); 9065 INIT_LIST_HEAD(&dev->ptype_all); 9066 INIT_LIST_HEAD(&dev->ptype_specific); 9067 #ifdef CONFIG_NET_SCHED 9068 hash_init(dev->qdisc_hash); 9069 #endif 9070 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 9071 setup(dev); 9072 9073 if (!dev->tx_queue_len) { 9074 dev->priv_flags |= IFF_NO_QUEUE; 9075 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 9076 } 9077 9078 dev->num_tx_queues = txqs; 9079 dev->real_num_tx_queues = txqs; 9080 if (netif_alloc_netdev_queues(dev)) 9081 goto free_all; 9082 9083 dev->num_rx_queues = rxqs; 9084 dev->real_num_rx_queues = rxqs; 9085 if (netif_alloc_rx_queues(dev)) 9086 goto free_all; 9087 9088 strcpy(dev->name, name); 9089 dev->name_assign_type = name_assign_type; 9090 dev->group = INIT_NETDEV_GROUP; 9091 if (!dev->ethtool_ops) 9092 dev->ethtool_ops = &default_ethtool_ops; 9093 9094 nf_hook_ingress_init(dev); 9095 9096 return dev; 9097 9098 free_all: 9099 free_netdev(dev); 9100 return NULL; 9101 9102 free_pcpu: 9103 free_percpu(dev->pcpu_refcnt); 9104 free_dev: 9105 netdev_freemem(dev); 9106 return NULL; 9107 } 9108 EXPORT_SYMBOL(alloc_netdev_mqs); 9109 9110 /** 9111 * free_netdev - free network device 9112 * @dev: device 9113 * 9114 * This function does the last stage of destroying an allocated device 9115 * interface. The reference to the device object is released. If this 9116 * is the last reference then it will be freed.Must be called in process 9117 * context. 9118 */ 9119 void free_netdev(struct net_device *dev) 9120 { 9121 struct napi_struct *p, *n; 9122 9123 might_sleep(); 9124 netif_free_tx_queues(dev); 9125 netif_free_rx_queues(dev); 9126 9127 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 9128 9129 /* Flush device addresses */ 9130 dev_addr_flush(dev); 9131 9132 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 9133 netif_napi_del(p); 9134 9135 free_percpu(dev->pcpu_refcnt); 9136 dev->pcpu_refcnt = NULL; 9137 9138 /* Compatibility with error handling in drivers */ 9139 if (dev->reg_state == NETREG_UNINITIALIZED) { 9140 netdev_freemem(dev); 9141 return; 9142 } 9143 9144 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 9145 dev->reg_state = NETREG_RELEASED; 9146 9147 /* will free via device release */ 9148 put_device(&dev->dev); 9149 } 9150 EXPORT_SYMBOL(free_netdev); 9151 9152 /** 9153 * synchronize_net - Synchronize with packet receive processing 9154 * 9155 * Wait for packets currently being received to be done. 9156 * Does not block later packets from starting. 9157 */ 9158 void synchronize_net(void) 9159 { 9160 might_sleep(); 9161 if (rtnl_is_locked()) 9162 synchronize_rcu_expedited(); 9163 else 9164 synchronize_rcu(); 9165 } 9166 EXPORT_SYMBOL(synchronize_net); 9167 9168 /** 9169 * unregister_netdevice_queue - remove device from the kernel 9170 * @dev: device 9171 * @head: list 9172 * 9173 * This function shuts down a device interface and removes it 9174 * from the kernel tables. 9175 * If head not NULL, device is queued to be unregistered later. 9176 * 9177 * Callers must hold the rtnl semaphore. You may want 9178 * unregister_netdev() instead of this. 9179 */ 9180 9181 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 9182 { 9183 ASSERT_RTNL(); 9184 9185 if (head) { 9186 list_move_tail(&dev->unreg_list, head); 9187 } else { 9188 rollback_registered(dev); 9189 /* Finish processing unregister after unlock */ 9190 net_set_todo(dev); 9191 } 9192 } 9193 EXPORT_SYMBOL(unregister_netdevice_queue); 9194 9195 /** 9196 * unregister_netdevice_many - unregister many devices 9197 * @head: list of devices 9198 * 9199 * Note: As most callers use a stack allocated list_head, 9200 * we force a list_del() to make sure stack wont be corrupted later. 9201 */ 9202 void unregister_netdevice_many(struct list_head *head) 9203 { 9204 struct net_device *dev; 9205 9206 if (!list_empty(head)) { 9207 rollback_registered_many(head); 9208 list_for_each_entry(dev, head, unreg_list) 9209 net_set_todo(dev); 9210 list_del(head); 9211 } 9212 } 9213 EXPORT_SYMBOL(unregister_netdevice_many); 9214 9215 /** 9216 * unregister_netdev - remove device from the kernel 9217 * @dev: device 9218 * 9219 * This function shuts down a device interface and removes it 9220 * from the kernel tables. 9221 * 9222 * This is just a wrapper for unregister_netdevice that takes 9223 * the rtnl semaphore. In general you want to use this and not 9224 * unregister_netdevice. 9225 */ 9226 void unregister_netdev(struct net_device *dev) 9227 { 9228 rtnl_lock(); 9229 unregister_netdevice(dev); 9230 rtnl_unlock(); 9231 } 9232 EXPORT_SYMBOL(unregister_netdev); 9233 9234 /** 9235 * dev_change_net_namespace - move device to different nethost namespace 9236 * @dev: device 9237 * @net: network namespace 9238 * @pat: If not NULL name pattern to try if the current device name 9239 * is already taken in the destination network namespace. 9240 * 9241 * This function shuts down a device interface and moves it 9242 * to a new network namespace. On success 0 is returned, on 9243 * a failure a netagive errno code is returned. 9244 * 9245 * Callers must hold the rtnl semaphore. 9246 */ 9247 9248 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 9249 { 9250 int err, new_nsid, new_ifindex; 9251 9252 ASSERT_RTNL(); 9253 9254 /* Don't allow namespace local devices to be moved. */ 9255 err = -EINVAL; 9256 if (dev->features & NETIF_F_NETNS_LOCAL) 9257 goto out; 9258 9259 /* Ensure the device has been registrered */ 9260 if (dev->reg_state != NETREG_REGISTERED) 9261 goto out; 9262 9263 /* Get out if there is nothing todo */ 9264 err = 0; 9265 if (net_eq(dev_net(dev), net)) 9266 goto out; 9267 9268 /* Pick the destination device name, and ensure 9269 * we can use it in the destination network namespace. 9270 */ 9271 err = -EEXIST; 9272 if (__dev_get_by_name(net, dev->name)) { 9273 /* We get here if we can't use the current device name */ 9274 if (!pat) 9275 goto out; 9276 err = dev_get_valid_name(net, dev, pat); 9277 if (err < 0) 9278 goto out; 9279 } 9280 9281 /* 9282 * And now a mini version of register_netdevice unregister_netdevice. 9283 */ 9284 9285 /* If device is running close it first. */ 9286 dev_close(dev); 9287 9288 /* And unlink it from device chain */ 9289 unlist_netdevice(dev); 9290 9291 synchronize_net(); 9292 9293 /* Shutdown queueing discipline. */ 9294 dev_shutdown(dev); 9295 9296 /* Notify protocols, that we are about to destroy 9297 * this device. They should clean all the things. 9298 * 9299 * Note that dev->reg_state stays at NETREG_REGISTERED. 9300 * This is wanted because this way 8021q and macvlan know 9301 * the device is just moving and can keep their slaves up. 9302 */ 9303 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 9304 rcu_barrier(); 9305 9306 new_nsid = peernet2id_alloc(dev_net(dev), net); 9307 /* If there is an ifindex conflict assign a new one */ 9308 if (__dev_get_by_index(net, dev->ifindex)) 9309 new_ifindex = dev_new_index(net); 9310 else 9311 new_ifindex = dev->ifindex; 9312 9313 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 9314 new_ifindex); 9315 9316 /* 9317 * Flush the unicast and multicast chains 9318 */ 9319 dev_uc_flush(dev); 9320 dev_mc_flush(dev); 9321 9322 /* Send a netdev-removed uevent to the old namespace */ 9323 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 9324 netdev_adjacent_del_links(dev); 9325 9326 /* Actually switch the network namespace */ 9327 dev_net_set(dev, net); 9328 dev->ifindex = new_ifindex; 9329 9330 /* Send a netdev-add uevent to the new namespace */ 9331 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 9332 netdev_adjacent_add_links(dev); 9333 9334 /* Fixup kobjects */ 9335 err = device_rename(&dev->dev, dev->name); 9336 WARN_ON(err); 9337 9338 /* Add the device back in the hashes */ 9339 list_netdevice(dev); 9340 9341 /* Notify protocols, that a new device appeared. */ 9342 call_netdevice_notifiers(NETDEV_REGISTER, dev); 9343 9344 /* 9345 * Prevent userspace races by waiting until the network 9346 * device is fully setup before sending notifications. 9347 */ 9348 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 9349 9350 synchronize_net(); 9351 err = 0; 9352 out: 9353 return err; 9354 } 9355 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 9356 9357 static int dev_cpu_dead(unsigned int oldcpu) 9358 { 9359 struct sk_buff **list_skb; 9360 struct sk_buff *skb; 9361 unsigned int cpu; 9362 struct softnet_data *sd, *oldsd, *remsd = NULL; 9363 9364 local_irq_disable(); 9365 cpu = smp_processor_id(); 9366 sd = &per_cpu(softnet_data, cpu); 9367 oldsd = &per_cpu(softnet_data, oldcpu); 9368 9369 /* Find end of our completion_queue. */ 9370 list_skb = &sd->completion_queue; 9371 while (*list_skb) 9372 list_skb = &(*list_skb)->next; 9373 /* Append completion queue from offline CPU. */ 9374 *list_skb = oldsd->completion_queue; 9375 oldsd->completion_queue = NULL; 9376 9377 /* Append output queue from offline CPU. */ 9378 if (oldsd->output_queue) { 9379 *sd->output_queue_tailp = oldsd->output_queue; 9380 sd->output_queue_tailp = oldsd->output_queue_tailp; 9381 oldsd->output_queue = NULL; 9382 oldsd->output_queue_tailp = &oldsd->output_queue; 9383 } 9384 /* Append NAPI poll list from offline CPU, with one exception : 9385 * process_backlog() must be called by cpu owning percpu backlog. 9386 * We properly handle process_queue & input_pkt_queue later. 9387 */ 9388 while (!list_empty(&oldsd->poll_list)) { 9389 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 9390 struct napi_struct, 9391 poll_list); 9392 9393 list_del_init(&napi->poll_list); 9394 if (napi->poll == process_backlog) 9395 napi->state = 0; 9396 else 9397 ____napi_schedule(sd, napi); 9398 } 9399 9400 raise_softirq_irqoff(NET_TX_SOFTIRQ); 9401 local_irq_enable(); 9402 9403 #ifdef CONFIG_RPS 9404 remsd = oldsd->rps_ipi_list; 9405 oldsd->rps_ipi_list = NULL; 9406 #endif 9407 /* send out pending IPI's on offline CPU */ 9408 net_rps_send_ipi(remsd); 9409 9410 /* Process offline CPU's input_pkt_queue */ 9411 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 9412 netif_rx_ni(skb); 9413 input_queue_head_incr(oldsd); 9414 } 9415 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 9416 netif_rx_ni(skb); 9417 input_queue_head_incr(oldsd); 9418 } 9419 9420 return 0; 9421 } 9422 9423 /** 9424 * netdev_increment_features - increment feature set by one 9425 * @all: current feature set 9426 * @one: new feature set 9427 * @mask: mask feature set 9428 * 9429 * Computes a new feature set after adding a device with feature set 9430 * @one to the master device with current feature set @all. Will not 9431 * enable anything that is off in @mask. Returns the new feature set. 9432 */ 9433 netdev_features_t netdev_increment_features(netdev_features_t all, 9434 netdev_features_t one, netdev_features_t mask) 9435 { 9436 if (mask & NETIF_F_HW_CSUM) 9437 mask |= NETIF_F_CSUM_MASK; 9438 mask |= NETIF_F_VLAN_CHALLENGED; 9439 9440 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 9441 all &= one | ~NETIF_F_ALL_FOR_ALL; 9442 9443 /* If one device supports hw checksumming, set for all. */ 9444 if (all & NETIF_F_HW_CSUM) 9445 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 9446 9447 return all; 9448 } 9449 EXPORT_SYMBOL(netdev_increment_features); 9450 9451 static struct hlist_head * __net_init netdev_create_hash(void) 9452 { 9453 int i; 9454 struct hlist_head *hash; 9455 9456 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 9457 if (hash != NULL) 9458 for (i = 0; i < NETDEV_HASHENTRIES; i++) 9459 INIT_HLIST_HEAD(&hash[i]); 9460 9461 return hash; 9462 } 9463 9464 /* Initialize per network namespace state */ 9465 static int __net_init netdev_init(struct net *net) 9466 { 9467 BUILD_BUG_ON(GRO_HASH_BUCKETS > 9468 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask)); 9469 9470 if (net != &init_net) 9471 INIT_LIST_HEAD(&net->dev_base_head); 9472 9473 net->dev_name_head = netdev_create_hash(); 9474 if (net->dev_name_head == NULL) 9475 goto err_name; 9476 9477 net->dev_index_head = netdev_create_hash(); 9478 if (net->dev_index_head == NULL) 9479 goto err_idx; 9480 9481 return 0; 9482 9483 err_idx: 9484 kfree(net->dev_name_head); 9485 err_name: 9486 return -ENOMEM; 9487 } 9488 9489 /** 9490 * netdev_drivername - network driver for the device 9491 * @dev: network device 9492 * 9493 * Determine network driver for device. 9494 */ 9495 const char *netdev_drivername(const struct net_device *dev) 9496 { 9497 const struct device_driver *driver; 9498 const struct device *parent; 9499 const char *empty = ""; 9500 9501 parent = dev->dev.parent; 9502 if (!parent) 9503 return empty; 9504 9505 driver = parent->driver; 9506 if (driver && driver->name) 9507 return driver->name; 9508 return empty; 9509 } 9510 9511 static void __netdev_printk(const char *level, const struct net_device *dev, 9512 struct va_format *vaf) 9513 { 9514 if (dev && dev->dev.parent) { 9515 dev_printk_emit(level[1] - '0', 9516 dev->dev.parent, 9517 "%s %s %s%s: %pV", 9518 dev_driver_string(dev->dev.parent), 9519 dev_name(dev->dev.parent), 9520 netdev_name(dev), netdev_reg_state(dev), 9521 vaf); 9522 } else if (dev) { 9523 printk("%s%s%s: %pV", 9524 level, netdev_name(dev), netdev_reg_state(dev), vaf); 9525 } else { 9526 printk("%s(NULL net_device): %pV", level, vaf); 9527 } 9528 } 9529 9530 void netdev_printk(const char *level, const struct net_device *dev, 9531 const char *format, ...) 9532 { 9533 struct va_format vaf; 9534 va_list args; 9535 9536 va_start(args, format); 9537 9538 vaf.fmt = format; 9539 vaf.va = &args; 9540 9541 __netdev_printk(level, dev, &vaf); 9542 9543 va_end(args); 9544 } 9545 EXPORT_SYMBOL(netdev_printk); 9546 9547 #define define_netdev_printk_level(func, level) \ 9548 void func(const struct net_device *dev, const char *fmt, ...) \ 9549 { \ 9550 struct va_format vaf; \ 9551 va_list args; \ 9552 \ 9553 va_start(args, fmt); \ 9554 \ 9555 vaf.fmt = fmt; \ 9556 vaf.va = &args; \ 9557 \ 9558 __netdev_printk(level, dev, &vaf); \ 9559 \ 9560 va_end(args); \ 9561 } \ 9562 EXPORT_SYMBOL(func); 9563 9564 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 9565 define_netdev_printk_level(netdev_alert, KERN_ALERT); 9566 define_netdev_printk_level(netdev_crit, KERN_CRIT); 9567 define_netdev_printk_level(netdev_err, KERN_ERR); 9568 define_netdev_printk_level(netdev_warn, KERN_WARNING); 9569 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 9570 define_netdev_printk_level(netdev_info, KERN_INFO); 9571 9572 static void __net_exit netdev_exit(struct net *net) 9573 { 9574 kfree(net->dev_name_head); 9575 kfree(net->dev_index_head); 9576 if (net != &init_net) 9577 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 9578 } 9579 9580 static struct pernet_operations __net_initdata netdev_net_ops = { 9581 .init = netdev_init, 9582 .exit = netdev_exit, 9583 }; 9584 9585 static void __net_exit default_device_exit(struct net *net) 9586 { 9587 struct net_device *dev, *aux; 9588 /* 9589 * Push all migratable network devices back to the 9590 * initial network namespace 9591 */ 9592 rtnl_lock(); 9593 for_each_netdev_safe(net, dev, aux) { 9594 int err; 9595 char fb_name[IFNAMSIZ]; 9596 9597 /* Ignore unmoveable devices (i.e. loopback) */ 9598 if (dev->features & NETIF_F_NETNS_LOCAL) 9599 continue; 9600 9601 /* Leave virtual devices for the generic cleanup */ 9602 if (dev->rtnl_link_ops) 9603 continue; 9604 9605 /* Push remaining network devices to init_net */ 9606 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 9607 err = dev_change_net_namespace(dev, &init_net, fb_name); 9608 if (err) { 9609 pr_emerg("%s: failed to move %s to init_net: %d\n", 9610 __func__, dev->name, err); 9611 BUG(); 9612 } 9613 } 9614 rtnl_unlock(); 9615 } 9616 9617 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 9618 { 9619 /* Return with the rtnl_lock held when there are no network 9620 * devices unregistering in any network namespace in net_list. 9621 */ 9622 struct net *net; 9623 bool unregistering; 9624 DEFINE_WAIT_FUNC(wait, woken_wake_function); 9625 9626 add_wait_queue(&netdev_unregistering_wq, &wait); 9627 for (;;) { 9628 unregistering = false; 9629 rtnl_lock(); 9630 list_for_each_entry(net, net_list, exit_list) { 9631 if (net->dev_unreg_count > 0) { 9632 unregistering = true; 9633 break; 9634 } 9635 } 9636 if (!unregistering) 9637 break; 9638 __rtnl_unlock(); 9639 9640 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 9641 } 9642 remove_wait_queue(&netdev_unregistering_wq, &wait); 9643 } 9644 9645 static void __net_exit default_device_exit_batch(struct list_head *net_list) 9646 { 9647 /* At exit all network devices most be removed from a network 9648 * namespace. Do this in the reverse order of registration. 9649 * Do this across as many network namespaces as possible to 9650 * improve batching efficiency. 9651 */ 9652 struct net_device *dev; 9653 struct net *net; 9654 LIST_HEAD(dev_kill_list); 9655 9656 /* To prevent network device cleanup code from dereferencing 9657 * loopback devices or network devices that have been freed 9658 * wait here for all pending unregistrations to complete, 9659 * before unregistring the loopback device and allowing the 9660 * network namespace be freed. 9661 * 9662 * The netdev todo list containing all network devices 9663 * unregistrations that happen in default_device_exit_batch 9664 * will run in the rtnl_unlock() at the end of 9665 * default_device_exit_batch. 9666 */ 9667 rtnl_lock_unregistering(net_list); 9668 list_for_each_entry(net, net_list, exit_list) { 9669 for_each_netdev_reverse(net, dev) { 9670 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 9671 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 9672 else 9673 unregister_netdevice_queue(dev, &dev_kill_list); 9674 } 9675 } 9676 unregister_netdevice_many(&dev_kill_list); 9677 rtnl_unlock(); 9678 } 9679 9680 static struct pernet_operations __net_initdata default_device_ops = { 9681 .exit = default_device_exit, 9682 .exit_batch = default_device_exit_batch, 9683 }; 9684 9685 /* 9686 * Initialize the DEV module. At boot time this walks the device list and 9687 * unhooks any devices that fail to initialise (normally hardware not 9688 * present) and leaves us with a valid list of present and active devices. 9689 * 9690 */ 9691 9692 /* 9693 * This is called single threaded during boot, so no need 9694 * to take the rtnl semaphore. 9695 */ 9696 static int __init net_dev_init(void) 9697 { 9698 int i, rc = -ENOMEM; 9699 9700 BUG_ON(!dev_boot_phase); 9701 9702 if (dev_proc_init()) 9703 goto out; 9704 9705 if (netdev_kobject_init()) 9706 goto out; 9707 9708 INIT_LIST_HEAD(&ptype_all); 9709 for (i = 0; i < PTYPE_HASH_SIZE; i++) 9710 INIT_LIST_HEAD(&ptype_base[i]); 9711 9712 INIT_LIST_HEAD(&offload_base); 9713 9714 if (register_pernet_subsys(&netdev_net_ops)) 9715 goto out; 9716 9717 /* 9718 * Initialise the packet receive queues. 9719 */ 9720 9721 for_each_possible_cpu(i) { 9722 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 9723 struct softnet_data *sd = &per_cpu(softnet_data, i); 9724 9725 INIT_WORK(flush, flush_backlog); 9726 9727 skb_queue_head_init(&sd->input_pkt_queue); 9728 skb_queue_head_init(&sd->process_queue); 9729 #ifdef CONFIG_XFRM_OFFLOAD 9730 skb_queue_head_init(&sd->xfrm_backlog); 9731 #endif 9732 INIT_LIST_HEAD(&sd->poll_list); 9733 sd->output_queue_tailp = &sd->output_queue; 9734 #ifdef CONFIG_RPS 9735 sd->csd.func = rps_trigger_softirq; 9736 sd->csd.info = sd; 9737 sd->cpu = i; 9738 #endif 9739 9740 init_gro_hash(&sd->backlog); 9741 sd->backlog.poll = process_backlog; 9742 sd->backlog.weight = weight_p; 9743 } 9744 9745 dev_boot_phase = 0; 9746 9747 /* The loopback device is special if any other network devices 9748 * is present in a network namespace the loopback device must 9749 * be present. Since we now dynamically allocate and free the 9750 * loopback device ensure this invariant is maintained by 9751 * keeping the loopback device as the first device on the 9752 * list of network devices. Ensuring the loopback devices 9753 * is the first device that appears and the last network device 9754 * that disappears. 9755 */ 9756 if (register_pernet_device(&loopback_net_ops)) 9757 goto out; 9758 9759 if (register_pernet_device(&default_device_ops)) 9760 goto out; 9761 9762 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 9763 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 9764 9765 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 9766 NULL, dev_cpu_dead); 9767 WARN_ON(rc < 0); 9768 rc = 0; 9769 out: 9770 return rc; 9771 } 9772 9773 subsys_initcall(net_dev_init); 9774