1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.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 <trace/events/qdisc.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_netdev.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 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "dev.h" 155 #include "net-sysfs.h" 156 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_info(unsigned long val, 164 struct netdev_notifier_info *info); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 netdev_name_node_del(name_node); 349 kfree(name_node->name); 350 netdev_name_node_free(name_node); 351 } 352 353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 354 { 355 struct netdev_name_node *name_node; 356 struct net *net = dev_net(dev); 357 358 name_node = netdev_name_node_lookup(net, name); 359 if (!name_node) 360 return -ENOENT; 361 /* lookup might have found our primary name or a name belonging 362 * to another device. 363 */ 364 if (name_node == dev->name_node || name_node->dev != dev) 365 return -EINVAL; 366 367 __netdev_name_node_alt_destroy(name_node); 368 369 return 0; 370 } 371 372 static void netdev_name_node_alt_flush(struct net_device *dev) 373 { 374 struct netdev_name_node *name_node, *tmp; 375 376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 377 __netdev_name_node_alt_destroy(name_node); 378 } 379 380 /* Device list insertion */ 381 static void list_netdevice(struct net_device *dev) 382 { 383 struct net *net = dev_net(dev); 384 385 ASSERT_RTNL(); 386 387 write_lock(&dev_base_lock); 388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 389 netdev_name_node_add(net, dev->name_node); 390 hlist_add_head_rcu(&dev->index_hlist, 391 dev_index_hash(net, dev->ifindex)); 392 write_unlock(&dev_base_lock); 393 394 dev_base_seq_inc(net); 395 } 396 397 /* Device list removal 398 * caller must respect a RCU grace period before freeing/reusing dev 399 */ 400 static void unlist_netdevice(struct net_device *dev, bool lock) 401 { 402 ASSERT_RTNL(); 403 404 /* Unlink dev from the device chain */ 405 if (lock) 406 write_lock(&dev_base_lock); 407 list_del_rcu(&dev->dev_list); 408 netdev_name_node_del(dev->name_node); 409 hlist_del_rcu(&dev->index_hlist); 410 if (lock) 411 write_unlock(&dev_base_lock); 412 413 dev_base_seq_inc(dev_net(dev)); 414 } 415 416 /* 417 * Our notifier list 418 */ 419 420 static RAW_NOTIFIER_HEAD(netdev_chain); 421 422 /* 423 * Device drivers call our routines to queue packets here. We empty the 424 * queue in the local softnet handler. 425 */ 426 427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 428 EXPORT_PER_CPU_SYMBOL(softnet_data); 429 430 #ifdef CONFIG_LOCKDEP 431 /* 432 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 433 * according to dev->type 434 */ 435 static const unsigned short netdev_lock_type[] = { 436 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 437 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 438 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 439 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 440 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 441 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 442 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 443 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 444 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 445 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 446 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 447 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 448 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 449 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 450 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 451 452 static const char *const netdev_lock_name[] = { 453 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 454 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 455 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 456 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 457 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 458 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 459 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 460 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 461 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 462 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 463 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 464 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 465 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 466 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 467 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 468 469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 471 472 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 473 { 474 int i; 475 476 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 477 if (netdev_lock_type[i] == dev_type) 478 return i; 479 /* the last key is used by default */ 480 return ARRAY_SIZE(netdev_lock_type) - 1; 481 } 482 483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 484 unsigned short dev_type) 485 { 486 int i; 487 488 i = netdev_lock_pos(dev_type); 489 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 490 netdev_lock_name[i]); 491 } 492 493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 494 { 495 int i; 496 497 i = netdev_lock_pos(dev->type); 498 lockdep_set_class_and_name(&dev->addr_list_lock, 499 &netdev_addr_lock_key[i], 500 netdev_lock_name[i]); 501 } 502 #else 503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 504 unsigned short dev_type) 505 { 506 } 507 508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 509 { 510 } 511 #endif 512 513 /******************************************************************************* 514 * 515 * Protocol management and registration routines 516 * 517 *******************************************************************************/ 518 519 520 /* 521 * Add a protocol ID to the list. Now that the input handler is 522 * smarter we can dispense with all the messy stuff that used to be 523 * here. 524 * 525 * BEWARE!!! Protocol handlers, mangling input packets, 526 * MUST BE last in hash buckets and checking protocol handlers 527 * MUST start from promiscuous ptype_all chain in net_bh. 528 * It is true now, do not change it. 529 * Explanation follows: if protocol handler, mangling packet, will 530 * be the first on list, it is not able to sense, that packet 531 * is cloned and should be copied-on-write, so that it will 532 * change it and subsequent readers will get broken packet. 533 * --ANK (980803) 534 */ 535 536 static inline struct list_head *ptype_head(const struct packet_type *pt) 537 { 538 if (pt->type == htons(ETH_P_ALL)) 539 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 540 else 541 return pt->dev ? &pt->dev->ptype_specific : 542 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 543 } 544 545 /** 546 * dev_add_pack - add packet handler 547 * @pt: packet type declaration 548 * 549 * Add a protocol handler to the networking stack. The passed &packet_type 550 * is linked into kernel lists and may not be freed until it has been 551 * removed from the kernel lists. 552 * 553 * This call does not sleep therefore it can not 554 * guarantee all CPU's that are in middle of receiving packets 555 * will see the new packet type (until the next received packet). 556 */ 557 558 void dev_add_pack(struct packet_type *pt) 559 { 560 struct list_head *head = ptype_head(pt); 561 562 spin_lock(&ptype_lock); 563 list_add_rcu(&pt->list, head); 564 spin_unlock(&ptype_lock); 565 } 566 EXPORT_SYMBOL(dev_add_pack); 567 568 /** 569 * __dev_remove_pack - remove packet handler 570 * @pt: packet type declaration 571 * 572 * Remove a protocol handler that was previously added to the kernel 573 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 574 * from the kernel lists and can be freed or reused once this function 575 * returns. 576 * 577 * The packet type might still be in use by receivers 578 * and must not be freed until after all the CPU's have gone 579 * through a quiescent state. 580 */ 581 void __dev_remove_pack(struct packet_type *pt) 582 { 583 struct list_head *head = ptype_head(pt); 584 struct packet_type *pt1; 585 586 spin_lock(&ptype_lock); 587 588 list_for_each_entry(pt1, head, list) { 589 if (pt == pt1) { 590 list_del_rcu(&pt->list); 591 goto out; 592 } 593 } 594 595 pr_warn("dev_remove_pack: %p not found\n", pt); 596 out: 597 spin_unlock(&ptype_lock); 598 } 599 EXPORT_SYMBOL(__dev_remove_pack); 600 601 /** 602 * dev_remove_pack - remove packet handler 603 * @pt: packet type declaration 604 * 605 * Remove a protocol handler that was previously added to the kernel 606 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 607 * from the kernel lists and can be freed or reused once this function 608 * returns. 609 * 610 * This call sleeps to guarantee that no CPU is looking at the packet 611 * type after return. 612 */ 613 void dev_remove_pack(struct packet_type *pt) 614 { 615 __dev_remove_pack(pt); 616 617 synchronize_net(); 618 } 619 EXPORT_SYMBOL(dev_remove_pack); 620 621 622 /******************************************************************************* 623 * 624 * Device Interface Subroutines 625 * 626 *******************************************************************************/ 627 628 /** 629 * dev_get_iflink - get 'iflink' value of a interface 630 * @dev: targeted interface 631 * 632 * Indicates the ifindex the interface is linked to. 633 * Physical interfaces have the same 'ifindex' and 'iflink' values. 634 */ 635 636 int dev_get_iflink(const struct net_device *dev) 637 { 638 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 639 return dev->netdev_ops->ndo_get_iflink(dev); 640 641 return dev->ifindex; 642 } 643 EXPORT_SYMBOL(dev_get_iflink); 644 645 /** 646 * dev_fill_metadata_dst - Retrieve tunnel egress information. 647 * @dev: targeted interface 648 * @skb: The packet. 649 * 650 * For better visibility of tunnel traffic OVS needs to retrieve 651 * egress tunnel information for a packet. Following API allows 652 * user to get this info. 653 */ 654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 655 { 656 struct ip_tunnel_info *info; 657 658 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 659 return -EINVAL; 660 661 info = skb_tunnel_info_unclone(skb); 662 if (!info) 663 return -ENOMEM; 664 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 665 return -EINVAL; 666 667 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 668 } 669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 670 671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 672 { 673 int k = stack->num_paths++; 674 675 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 676 return NULL; 677 678 return &stack->path[k]; 679 } 680 681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 682 struct net_device_path_stack *stack) 683 { 684 const struct net_device *last_dev; 685 struct net_device_path_ctx ctx = { 686 .dev = dev, 687 }; 688 struct net_device_path *path; 689 int ret = 0; 690 691 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 692 stack->num_paths = 0; 693 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 694 last_dev = ctx.dev; 695 path = dev_fwd_path(stack); 696 if (!path) 697 return -1; 698 699 memset(path, 0, sizeof(struct net_device_path)); 700 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 701 if (ret < 0) 702 return -1; 703 704 if (WARN_ON_ONCE(last_dev == ctx.dev)) 705 return -1; 706 } 707 708 if (!ctx.dev) 709 return ret; 710 711 path = dev_fwd_path(stack); 712 if (!path) 713 return -1; 714 path->type = DEV_PATH_ETHERNET; 715 path->dev = ctx.dev; 716 717 return ret; 718 } 719 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 720 721 /** 722 * __dev_get_by_name - find a device by its name 723 * @net: the applicable net namespace 724 * @name: name to find 725 * 726 * Find an interface by name. Must be called under RTNL semaphore 727 * or @dev_base_lock. If the name is found a pointer to the device 728 * is returned. If the name is not found then %NULL is returned. The 729 * reference counters are not incremented so the caller must be 730 * careful with locks. 731 */ 732 733 struct net_device *__dev_get_by_name(struct net *net, const char *name) 734 { 735 struct netdev_name_node *node_name; 736 737 node_name = netdev_name_node_lookup(net, name); 738 return node_name ? node_name->dev : NULL; 739 } 740 EXPORT_SYMBOL(__dev_get_by_name); 741 742 /** 743 * dev_get_by_name_rcu - find a device by its name 744 * @net: the applicable net namespace 745 * @name: name to find 746 * 747 * Find an interface by name. 748 * If the name is found a pointer to the device is returned. 749 * If the name is not found then %NULL is returned. 750 * The reference counters are not incremented so the caller must be 751 * careful with locks. The caller must hold RCU lock. 752 */ 753 754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 755 { 756 struct netdev_name_node *node_name; 757 758 node_name = netdev_name_node_lookup_rcu(net, name); 759 return node_name ? node_name->dev : NULL; 760 } 761 EXPORT_SYMBOL(dev_get_by_name_rcu); 762 763 /** 764 * dev_get_by_name - find a device by its name 765 * @net: the applicable net namespace 766 * @name: name to find 767 * 768 * Find an interface by name. This can be called from any 769 * context and does its own locking. The returned handle has 770 * the usage count incremented and the caller must use dev_put() to 771 * release it when it is no longer needed. %NULL is returned if no 772 * matching device is found. 773 */ 774 775 struct net_device *dev_get_by_name(struct net *net, const char *name) 776 { 777 struct net_device *dev; 778 779 rcu_read_lock(); 780 dev = dev_get_by_name_rcu(net, name); 781 dev_hold(dev); 782 rcu_read_unlock(); 783 return dev; 784 } 785 EXPORT_SYMBOL(dev_get_by_name); 786 787 /** 788 * __dev_get_by_index - find a device by its ifindex 789 * @net: the applicable net namespace 790 * @ifindex: index of device 791 * 792 * Search for an interface by index. Returns %NULL if the device 793 * is not found or a pointer to the device. The device has not 794 * had its reference counter increased so the caller must be careful 795 * about locking. The caller must hold either the RTNL semaphore 796 * or @dev_base_lock. 797 */ 798 799 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 800 { 801 struct net_device *dev; 802 struct hlist_head *head = dev_index_hash(net, ifindex); 803 804 hlist_for_each_entry(dev, head, index_hlist) 805 if (dev->ifindex == ifindex) 806 return dev; 807 808 return NULL; 809 } 810 EXPORT_SYMBOL(__dev_get_by_index); 811 812 /** 813 * dev_get_by_index_rcu - find a device by its ifindex 814 * @net: the applicable net namespace 815 * @ifindex: index of device 816 * 817 * Search for an interface by index. Returns %NULL if the device 818 * is not found or a pointer to the device. The device has not 819 * had its reference counter increased so the caller must be careful 820 * about locking. The caller must hold RCU lock. 821 */ 822 823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 824 { 825 struct net_device *dev; 826 struct hlist_head *head = dev_index_hash(net, ifindex); 827 828 hlist_for_each_entry_rcu(dev, head, index_hlist) 829 if (dev->ifindex == ifindex) 830 return dev; 831 832 return NULL; 833 } 834 EXPORT_SYMBOL(dev_get_by_index_rcu); 835 836 837 /** 838 * dev_get_by_index - find a device by its ifindex 839 * @net: the applicable net namespace 840 * @ifindex: index of device 841 * 842 * Search for an interface by index. Returns NULL if the device 843 * is not found or a pointer to the device. The device returned has 844 * had a reference added and the pointer is safe until the user calls 845 * dev_put to indicate they have finished with it. 846 */ 847 848 struct net_device *dev_get_by_index(struct net *net, int ifindex) 849 { 850 struct net_device *dev; 851 852 rcu_read_lock(); 853 dev = dev_get_by_index_rcu(net, ifindex); 854 dev_hold(dev); 855 rcu_read_unlock(); 856 return dev; 857 } 858 EXPORT_SYMBOL(dev_get_by_index); 859 860 /** 861 * dev_get_by_napi_id - find a device by napi_id 862 * @napi_id: ID of the NAPI struct 863 * 864 * Search for an interface by NAPI ID. Returns %NULL if the device 865 * is not found or a pointer to the device. The device has not had 866 * its reference counter increased so the caller must be careful 867 * about locking. The caller must hold RCU lock. 868 */ 869 870 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 871 { 872 struct napi_struct *napi; 873 874 WARN_ON_ONCE(!rcu_read_lock_held()); 875 876 if (napi_id < MIN_NAPI_ID) 877 return NULL; 878 879 napi = napi_by_id(napi_id); 880 881 return napi ? napi->dev : NULL; 882 } 883 EXPORT_SYMBOL(dev_get_by_napi_id); 884 885 /** 886 * netdev_get_name - get a netdevice name, knowing its ifindex. 887 * @net: network namespace 888 * @name: a pointer to the buffer where the name will be stored. 889 * @ifindex: the ifindex of the interface to get the name from. 890 */ 891 int netdev_get_name(struct net *net, char *name, int ifindex) 892 { 893 struct net_device *dev; 894 int ret; 895 896 down_read(&devnet_rename_sem); 897 rcu_read_lock(); 898 899 dev = dev_get_by_index_rcu(net, ifindex); 900 if (!dev) { 901 ret = -ENODEV; 902 goto out; 903 } 904 905 strcpy(name, dev->name); 906 907 ret = 0; 908 out: 909 rcu_read_unlock(); 910 up_read(&devnet_rename_sem); 911 return ret; 912 } 913 914 /** 915 * dev_getbyhwaddr_rcu - find a device by its hardware address 916 * @net: the applicable net namespace 917 * @type: media type of device 918 * @ha: hardware address 919 * 920 * Search for an interface by MAC address. Returns NULL if the device 921 * is not found or a pointer to the device. 922 * The caller must hold RCU or RTNL. 923 * The returned device has not had its ref count increased 924 * and the caller must therefore be careful about locking 925 * 926 */ 927 928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 929 const char *ha) 930 { 931 struct net_device *dev; 932 933 for_each_netdev_rcu(net, dev) 934 if (dev->type == type && 935 !memcmp(dev->dev_addr, ha, dev->addr_len)) 936 return dev; 937 938 return NULL; 939 } 940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 941 942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 943 { 944 struct net_device *dev, *ret = NULL; 945 946 rcu_read_lock(); 947 for_each_netdev_rcu(net, dev) 948 if (dev->type == type) { 949 dev_hold(dev); 950 ret = dev; 951 break; 952 } 953 rcu_read_unlock(); 954 return ret; 955 } 956 EXPORT_SYMBOL(dev_getfirstbyhwtype); 957 958 /** 959 * __dev_get_by_flags - find any device with given flags 960 * @net: the applicable net namespace 961 * @if_flags: IFF_* values 962 * @mask: bitmask of bits in if_flags to check 963 * 964 * Search for any interface with the given flags. Returns NULL if a device 965 * is not found or a pointer to the device. Must be called inside 966 * rtnl_lock(), and result refcount is unchanged. 967 */ 968 969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 970 unsigned short mask) 971 { 972 struct net_device *dev, *ret; 973 974 ASSERT_RTNL(); 975 976 ret = NULL; 977 for_each_netdev(net, dev) { 978 if (((dev->flags ^ if_flags) & mask) == 0) { 979 ret = dev; 980 break; 981 } 982 } 983 return ret; 984 } 985 EXPORT_SYMBOL(__dev_get_by_flags); 986 987 /** 988 * dev_valid_name - check if name is okay for network device 989 * @name: name string 990 * 991 * Network device names need to be valid file names to 992 * allow sysfs to work. We also disallow any kind of 993 * whitespace. 994 */ 995 bool dev_valid_name(const char *name) 996 { 997 if (*name == '\0') 998 return false; 999 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1000 return false; 1001 if (!strcmp(name, ".") || !strcmp(name, "..")) 1002 return false; 1003 1004 while (*name) { 1005 if (*name == '/' || *name == ':' || isspace(*name)) 1006 return false; 1007 name++; 1008 } 1009 return true; 1010 } 1011 EXPORT_SYMBOL(dev_valid_name); 1012 1013 /** 1014 * __dev_alloc_name - allocate a name for a device 1015 * @net: network namespace to allocate the device name in 1016 * @name: name format string 1017 * @buf: scratch buffer and result name string 1018 * 1019 * Passed a format string - eg "lt%d" it will try and find a suitable 1020 * id. It scans list of devices to build up a free map, then chooses 1021 * the first empty slot. The caller must hold the dev_base or rtnl lock 1022 * while allocating the name and adding the device in order to avoid 1023 * duplicates. 1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1025 * Returns the number of the unit assigned or a negative errno code. 1026 */ 1027 1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1029 { 1030 int i = 0; 1031 const char *p; 1032 const int max_netdevices = 8*PAGE_SIZE; 1033 unsigned long *inuse; 1034 struct net_device *d; 1035 1036 if (!dev_valid_name(name)) 1037 return -EINVAL; 1038 1039 p = strchr(name, '%'); 1040 if (p) { 1041 /* 1042 * Verify the string as this thing may have come from 1043 * the user. There must be either one "%d" and no other "%" 1044 * characters. 1045 */ 1046 if (p[1] != 'd' || strchr(p + 2, '%')) 1047 return -EINVAL; 1048 1049 /* Use one page as a bit array of possible slots */ 1050 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1051 if (!inuse) 1052 return -ENOMEM; 1053 1054 for_each_netdev(net, d) { 1055 struct netdev_name_node *name_node; 1056 list_for_each_entry(name_node, &d->name_node->list, list) { 1057 if (!sscanf(name_node->name, name, &i)) 1058 continue; 1059 if (i < 0 || i >= max_netdevices) 1060 continue; 1061 1062 /* avoid cases where sscanf is not exact inverse of printf */ 1063 snprintf(buf, IFNAMSIZ, name, i); 1064 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1065 __set_bit(i, inuse); 1066 } 1067 if (!sscanf(d->name, name, &i)) 1068 continue; 1069 if (i < 0 || i >= max_netdevices) 1070 continue; 1071 1072 /* avoid cases where sscanf is not exact inverse of printf */ 1073 snprintf(buf, IFNAMSIZ, name, i); 1074 if (!strncmp(buf, d->name, IFNAMSIZ)) 1075 __set_bit(i, inuse); 1076 } 1077 1078 i = find_first_zero_bit(inuse, max_netdevices); 1079 free_page((unsigned long) inuse); 1080 } 1081 1082 snprintf(buf, IFNAMSIZ, name, i); 1083 if (!netdev_name_in_use(net, buf)) 1084 return i; 1085 1086 /* It is possible to run out of possible slots 1087 * when the name is long and there isn't enough space left 1088 * for the digits, or if all bits are used. 1089 */ 1090 return -ENFILE; 1091 } 1092 1093 static int dev_alloc_name_ns(struct net *net, 1094 struct net_device *dev, 1095 const char *name) 1096 { 1097 char buf[IFNAMSIZ]; 1098 int ret; 1099 1100 BUG_ON(!net); 1101 ret = __dev_alloc_name(net, name, buf); 1102 if (ret >= 0) 1103 strscpy(dev->name, buf, IFNAMSIZ); 1104 return ret; 1105 } 1106 1107 /** 1108 * dev_alloc_name - allocate a name for a device 1109 * @dev: device 1110 * @name: name format string 1111 * 1112 * Passed a format string - eg "lt%d" it will try and find a suitable 1113 * id. It scans list of devices to build up a free map, then chooses 1114 * the first empty slot. The caller must hold the dev_base or rtnl lock 1115 * while allocating the name and adding the device in order to avoid 1116 * duplicates. 1117 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1118 * Returns the number of the unit assigned or a negative errno code. 1119 */ 1120 1121 int dev_alloc_name(struct net_device *dev, const char *name) 1122 { 1123 return dev_alloc_name_ns(dev_net(dev), dev, name); 1124 } 1125 EXPORT_SYMBOL(dev_alloc_name); 1126 1127 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1128 const char *name) 1129 { 1130 BUG_ON(!net); 1131 1132 if (!dev_valid_name(name)) 1133 return -EINVAL; 1134 1135 if (strchr(name, '%')) 1136 return dev_alloc_name_ns(net, dev, name); 1137 else if (netdev_name_in_use(net, name)) 1138 return -EEXIST; 1139 else if (dev->name != name) 1140 strscpy(dev->name, name, IFNAMSIZ); 1141 1142 return 0; 1143 } 1144 1145 /** 1146 * dev_change_name - change name of a device 1147 * @dev: device 1148 * @newname: name (or format string) must be at least IFNAMSIZ 1149 * 1150 * Change name of a device, can pass format strings "eth%d". 1151 * for wildcarding. 1152 */ 1153 int dev_change_name(struct net_device *dev, const char *newname) 1154 { 1155 unsigned char old_assign_type; 1156 char oldname[IFNAMSIZ]; 1157 int err = 0; 1158 int ret; 1159 struct net *net; 1160 1161 ASSERT_RTNL(); 1162 BUG_ON(!dev_net(dev)); 1163 1164 net = dev_net(dev); 1165 1166 down_write(&devnet_rename_sem); 1167 1168 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1169 up_write(&devnet_rename_sem); 1170 return 0; 1171 } 1172 1173 memcpy(oldname, dev->name, IFNAMSIZ); 1174 1175 err = dev_get_valid_name(net, dev, newname); 1176 if (err < 0) { 1177 up_write(&devnet_rename_sem); 1178 return err; 1179 } 1180 1181 if (oldname[0] && !strchr(oldname, '%')) 1182 netdev_info(dev, "renamed from %s%s\n", oldname, 1183 dev->flags & IFF_UP ? " (while UP)" : ""); 1184 1185 old_assign_type = dev->name_assign_type; 1186 dev->name_assign_type = NET_NAME_RENAMED; 1187 1188 rollback: 1189 ret = device_rename(&dev->dev, dev->name); 1190 if (ret) { 1191 memcpy(dev->name, oldname, IFNAMSIZ); 1192 dev->name_assign_type = old_assign_type; 1193 up_write(&devnet_rename_sem); 1194 return ret; 1195 } 1196 1197 up_write(&devnet_rename_sem); 1198 1199 netdev_adjacent_rename_links(dev, oldname); 1200 1201 write_lock(&dev_base_lock); 1202 netdev_name_node_del(dev->name_node); 1203 write_unlock(&dev_base_lock); 1204 1205 synchronize_rcu(); 1206 1207 write_lock(&dev_base_lock); 1208 netdev_name_node_add(net, dev->name_node); 1209 write_unlock(&dev_base_lock); 1210 1211 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1212 ret = notifier_to_errno(ret); 1213 1214 if (ret) { 1215 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1216 if (err >= 0) { 1217 err = ret; 1218 down_write(&devnet_rename_sem); 1219 memcpy(dev->name, oldname, IFNAMSIZ); 1220 memcpy(oldname, newname, IFNAMSIZ); 1221 dev->name_assign_type = old_assign_type; 1222 old_assign_type = NET_NAME_RENAMED; 1223 goto rollback; 1224 } else { 1225 netdev_err(dev, "name change rollback failed: %d\n", 1226 ret); 1227 } 1228 } 1229 1230 return err; 1231 } 1232 1233 /** 1234 * dev_set_alias - change ifalias of a device 1235 * @dev: device 1236 * @alias: name up to IFALIASZ 1237 * @len: limit of bytes to copy from info 1238 * 1239 * Set ifalias for a device, 1240 */ 1241 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1242 { 1243 struct dev_ifalias *new_alias = NULL; 1244 1245 if (len >= IFALIASZ) 1246 return -EINVAL; 1247 1248 if (len) { 1249 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1250 if (!new_alias) 1251 return -ENOMEM; 1252 1253 memcpy(new_alias->ifalias, alias, len); 1254 new_alias->ifalias[len] = 0; 1255 } 1256 1257 mutex_lock(&ifalias_mutex); 1258 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1259 mutex_is_locked(&ifalias_mutex)); 1260 mutex_unlock(&ifalias_mutex); 1261 1262 if (new_alias) 1263 kfree_rcu(new_alias, rcuhead); 1264 1265 return len; 1266 } 1267 EXPORT_SYMBOL(dev_set_alias); 1268 1269 /** 1270 * dev_get_alias - get ifalias of a device 1271 * @dev: device 1272 * @name: buffer to store name of ifalias 1273 * @len: size of buffer 1274 * 1275 * get ifalias for a device. Caller must make sure dev cannot go 1276 * away, e.g. rcu read lock or own a reference count to device. 1277 */ 1278 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1279 { 1280 const struct dev_ifalias *alias; 1281 int ret = 0; 1282 1283 rcu_read_lock(); 1284 alias = rcu_dereference(dev->ifalias); 1285 if (alias) 1286 ret = snprintf(name, len, "%s", alias->ifalias); 1287 rcu_read_unlock(); 1288 1289 return ret; 1290 } 1291 1292 /** 1293 * netdev_features_change - device changes features 1294 * @dev: device to cause notification 1295 * 1296 * Called to indicate a device has changed features. 1297 */ 1298 void netdev_features_change(struct net_device *dev) 1299 { 1300 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1301 } 1302 EXPORT_SYMBOL(netdev_features_change); 1303 1304 /** 1305 * netdev_state_change - device changes state 1306 * @dev: device to cause notification 1307 * 1308 * Called to indicate a device has changed state. This function calls 1309 * the notifier chains for netdev_chain and sends a NEWLINK message 1310 * to the routing socket. 1311 */ 1312 void netdev_state_change(struct net_device *dev) 1313 { 1314 if (dev->flags & IFF_UP) { 1315 struct netdev_notifier_change_info change_info = { 1316 .info.dev = dev, 1317 }; 1318 1319 call_netdevice_notifiers_info(NETDEV_CHANGE, 1320 &change_info.info); 1321 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1322 } 1323 } 1324 EXPORT_SYMBOL(netdev_state_change); 1325 1326 /** 1327 * __netdev_notify_peers - notify network peers about existence of @dev, 1328 * to be called when rtnl lock is already held. 1329 * @dev: network device 1330 * 1331 * Generate traffic such that interested network peers are aware of 1332 * @dev, such as by generating a gratuitous ARP. This may be used when 1333 * a device wants to inform the rest of the network about some sort of 1334 * reconfiguration such as a failover event or virtual machine 1335 * migration. 1336 */ 1337 void __netdev_notify_peers(struct net_device *dev) 1338 { 1339 ASSERT_RTNL(); 1340 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1341 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1342 } 1343 EXPORT_SYMBOL(__netdev_notify_peers); 1344 1345 /** 1346 * netdev_notify_peers - notify network peers about existence of @dev 1347 * @dev: network device 1348 * 1349 * Generate traffic such that interested network peers are aware of 1350 * @dev, such as by generating a gratuitous ARP. This may be used when 1351 * a device wants to inform the rest of the network about some sort of 1352 * reconfiguration such as a failover event or virtual machine 1353 * migration. 1354 */ 1355 void netdev_notify_peers(struct net_device *dev) 1356 { 1357 rtnl_lock(); 1358 __netdev_notify_peers(dev); 1359 rtnl_unlock(); 1360 } 1361 EXPORT_SYMBOL(netdev_notify_peers); 1362 1363 static int napi_threaded_poll(void *data); 1364 1365 static int napi_kthread_create(struct napi_struct *n) 1366 { 1367 int err = 0; 1368 1369 /* Create and wake up the kthread once to put it in 1370 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1371 * warning and work with loadavg. 1372 */ 1373 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1374 n->dev->name, n->napi_id); 1375 if (IS_ERR(n->thread)) { 1376 err = PTR_ERR(n->thread); 1377 pr_err("kthread_run failed with err %d\n", err); 1378 n->thread = NULL; 1379 } 1380 1381 return err; 1382 } 1383 1384 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1385 { 1386 const struct net_device_ops *ops = dev->netdev_ops; 1387 int ret; 1388 1389 ASSERT_RTNL(); 1390 dev_addr_check(dev); 1391 1392 if (!netif_device_present(dev)) { 1393 /* may be detached because parent is runtime-suspended */ 1394 if (dev->dev.parent) 1395 pm_runtime_resume(dev->dev.parent); 1396 if (!netif_device_present(dev)) 1397 return -ENODEV; 1398 } 1399 1400 /* Block netpoll from trying to do any rx path servicing. 1401 * If we don't do this there is a chance ndo_poll_controller 1402 * or ndo_poll may be running while we open the device 1403 */ 1404 netpoll_poll_disable(dev); 1405 1406 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1407 ret = notifier_to_errno(ret); 1408 if (ret) 1409 return ret; 1410 1411 set_bit(__LINK_STATE_START, &dev->state); 1412 1413 if (ops->ndo_validate_addr) 1414 ret = ops->ndo_validate_addr(dev); 1415 1416 if (!ret && ops->ndo_open) 1417 ret = ops->ndo_open(dev); 1418 1419 netpoll_poll_enable(dev); 1420 1421 if (ret) 1422 clear_bit(__LINK_STATE_START, &dev->state); 1423 else { 1424 dev->flags |= IFF_UP; 1425 dev_set_rx_mode(dev); 1426 dev_activate(dev); 1427 add_device_randomness(dev->dev_addr, dev->addr_len); 1428 } 1429 1430 return ret; 1431 } 1432 1433 /** 1434 * dev_open - prepare an interface for use. 1435 * @dev: device to open 1436 * @extack: netlink extended ack 1437 * 1438 * Takes a device from down to up state. The device's private open 1439 * function is invoked and then the multicast lists are loaded. Finally 1440 * the device is moved into the up state and a %NETDEV_UP message is 1441 * sent to the netdev notifier chain. 1442 * 1443 * Calling this function on an active interface is a nop. On a failure 1444 * a negative errno code is returned. 1445 */ 1446 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1447 { 1448 int ret; 1449 1450 if (dev->flags & IFF_UP) 1451 return 0; 1452 1453 ret = __dev_open(dev, extack); 1454 if (ret < 0) 1455 return ret; 1456 1457 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1458 call_netdevice_notifiers(NETDEV_UP, dev); 1459 1460 return ret; 1461 } 1462 EXPORT_SYMBOL(dev_open); 1463 1464 static void __dev_close_many(struct list_head *head) 1465 { 1466 struct net_device *dev; 1467 1468 ASSERT_RTNL(); 1469 might_sleep(); 1470 1471 list_for_each_entry(dev, head, close_list) { 1472 /* Temporarily disable netpoll until the interface is down */ 1473 netpoll_poll_disable(dev); 1474 1475 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1476 1477 clear_bit(__LINK_STATE_START, &dev->state); 1478 1479 /* Synchronize to scheduled poll. We cannot touch poll list, it 1480 * can be even on different cpu. So just clear netif_running(). 1481 * 1482 * dev->stop() will invoke napi_disable() on all of it's 1483 * napi_struct instances on this device. 1484 */ 1485 smp_mb__after_atomic(); /* Commit netif_running(). */ 1486 } 1487 1488 dev_deactivate_many(head); 1489 1490 list_for_each_entry(dev, head, close_list) { 1491 const struct net_device_ops *ops = dev->netdev_ops; 1492 1493 /* 1494 * Call the device specific close. This cannot fail. 1495 * Only if device is UP 1496 * 1497 * We allow it to be called even after a DETACH hot-plug 1498 * event. 1499 */ 1500 if (ops->ndo_stop) 1501 ops->ndo_stop(dev); 1502 1503 dev->flags &= ~IFF_UP; 1504 netpoll_poll_enable(dev); 1505 } 1506 } 1507 1508 static void __dev_close(struct net_device *dev) 1509 { 1510 LIST_HEAD(single); 1511 1512 list_add(&dev->close_list, &single); 1513 __dev_close_many(&single); 1514 list_del(&single); 1515 } 1516 1517 void dev_close_many(struct list_head *head, bool unlink) 1518 { 1519 struct net_device *dev, *tmp; 1520 1521 /* Remove the devices that don't need to be closed */ 1522 list_for_each_entry_safe(dev, tmp, head, close_list) 1523 if (!(dev->flags & IFF_UP)) 1524 list_del_init(&dev->close_list); 1525 1526 __dev_close_many(head); 1527 1528 list_for_each_entry_safe(dev, tmp, head, close_list) { 1529 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1530 call_netdevice_notifiers(NETDEV_DOWN, dev); 1531 if (unlink) 1532 list_del_init(&dev->close_list); 1533 } 1534 } 1535 EXPORT_SYMBOL(dev_close_many); 1536 1537 /** 1538 * dev_close - shutdown an interface. 1539 * @dev: device to shutdown 1540 * 1541 * This function moves an active device into down state. A 1542 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1543 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1544 * chain. 1545 */ 1546 void dev_close(struct net_device *dev) 1547 { 1548 if (dev->flags & IFF_UP) { 1549 LIST_HEAD(single); 1550 1551 list_add(&dev->close_list, &single); 1552 dev_close_many(&single, true); 1553 list_del(&single); 1554 } 1555 } 1556 EXPORT_SYMBOL(dev_close); 1557 1558 1559 /** 1560 * dev_disable_lro - disable Large Receive Offload on a device 1561 * @dev: device 1562 * 1563 * Disable Large Receive Offload (LRO) on a net device. Must be 1564 * called under RTNL. This is needed if received packets may be 1565 * forwarded to another interface. 1566 */ 1567 void dev_disable_lro(struct net_device *dev) 1568 { 1569 struct net_device *lower_dev; 1570 struct list_head *iter; 1571 1572 dev->wanted_features &= ~NETIF_F_LRO; 1573 netdev_update_features(dev); 1574 1575 if (unlikely(dev->features & NETIF_F_LRO)) 1576 netdev_WARN(dev, "failed to disable LRO!\n"); 1577 1578 netdev_for_each_lower_dev(dev, lower_dev, iter) 1579 dev_disable_lro(lower_dev); 1580 } 1581 EXPORT_SYMBOL(dev_disable_lro); 1582 1583 /** 1584 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1585 * @dev: device 1586 * 1587 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1588 * called under RTNL. This is needed if Generic XDP is installed on 1589 * the device. 1590 */ 1591 static void dev_disable_gro_hw(struct net_device *dev) 1592 { 1593 dev->wanted_features &= ~NETIF_F_GRO_HW; 1594 netdev_update_features(dev); 1595 1596 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1597 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1598 } 1599 1600 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1601 { 1602 #define N(val) \ 1603 case NETDEV_##val: \ 1604 return "NETDEV_" __stringify(val); 1605 switch (cmd) { 1606 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1607 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1608 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1609 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1610 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1611 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1612 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1613 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1614 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1615 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1616 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1617 N(XDP_FEAT_CHANGE) 1618 } 1619 #undef N 1620 return "UNKNOWN_NETDEV_EVENT"; 1621 } 1622 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1623 1624 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1625 struct net_device *dev) 1626 { 1627 struct netdev_notifier_info info = { 1628 .dev = dev, 1629 }; 1630 1631 return nb->notifier_call(nb, val, &info); 1632 } 1633 1634 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1635 struct net_device *dev) 1636 { 1637 int err; 1638 1639 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1640 err = notifier_to_errno(err); 1641 if (err) 1642 return err; 1643 1644 if (!(dev->flags & IFF_UP)) 1645 return 0; 1646 1647 call_netdevice_notifier(nb, NETDEV_UP, dev); 1648 return 0; 1649 } 1650 1651 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1652 struct net_device *dev) 1653 { 1654 if (dev->flags & IFF_UP) { 1655 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1656 dev); 1657 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1658 } 1659 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1660 } 1661 1662 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1663 struct net *net) 1664 { 1665 struct net_device *dev; 1666 int err; 1667 1668 for_each_netdev(net, dev) { 1669 err = call_netdevice_register_notifiers(nb, dev); 1670 if (err) 1671 goto rollback; 1672 } 1673 return 0; 1674 1675 rollback: 1676 for_each_netdev_continue_reverse(net, dev) 1677 call_netdevice_unregister_notifiers(nb, dev); 1678 return err; 1679 } 1680 1681 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1682 struct net *net) 1683 { 1684 struct net_device *dev; 1685 1686 for_each_netdev(net, dev) 1687 call_netdevice_unregister_notifiers(nb, dev); 1688 } 1689 1690 static int dev_boot_phase = 1; 1691 1692 /** 1693 * register_netdevice_notifier - register a network notifier block 1694 * @nb: notifier 1695 * 1696 * Register a notifier to be called when network device events occur. 1697 * The notifier passed is linked into the kernel structures and must 1698 * not be reused until it has been unregistered. A negative errno code 1699 * is returned on a failure. 1700 * 1701 * When registered all registration and up events are replayed 1702 * to the new notifier to allow device to have a race free 1703 * view of the network device list. 1704 */ 1705 1706 int register_netdevice_notifier(struct notifier_block *nb) 1707 { 1708 struct net *net; 1709 int err; 1710 1711 /* Close race with setup_net() and cleanup_net() */ 1712 down_write(&pernet_ops_rwsem); 1713 rtnl_lock(); 1714 err = raw_notifier_chain_register(&netdev_chain, nb); 1715 if (err) 1716 goto unlock; 1717 if (dev_boot_phase) 1718 goto unlock; 1719 for_each_net(net) { 1720 err = call_netdevice_register_net_notifiers(nb, net); 1721 if (err) 1722 goto rollback; 1723 } 1724 1725 unlock: 1726 rtnl_unlock(); 1727 up_write(&pernet_ops_rwsem); 1728 return err; 1729 1730 rollback: 1731 for_each_net_continue_reverse(net) 1732 call_netdevice_unregister_net_notifiers(nb, net); 1733 1734 raw_notifier_chain_unregister(&netdev_chain, nb); 1735 goto unlock; 1736 } 1737 EXPORT_SYMBOL(register_netdevice_notifier); 1738 1739 /** 1740 * unregister_netdevice_notifier - unregister a network notifier block 1741 * @nb: notifier 1742 * 1743 * Unregister a notifier previously registered by 1744 * register_netdevice_notifier(). The notifier is unlinked into the 1745 * kernel structures and may then be reused. A negative errno code 1746 * is returned on a failure. 1747 * 1748 * After unregistering unregister and down device events are synthesized 1749 * for all devices on the device list to the removed notifier to remove 1750 * the need for special case cleanup code. 1751 */ 1752 1753 int unregister_netdevice_notifier(struct notifier_block *nb) 1754 { 1755 struct net *net; 1756 int err; 1757 1758 /* Close race with setup_net() and cleanup_net() */ 1759 down_write(&pernet_ops_rwsem); 1760 rtnl_lock(); 1761 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1762 if (err) 1763 goto unlock; 1764 1765 for_each_net(net) 1766 call_netdevice_unregister_net_notifiers(nb, net); 1767 1768 unlock: 1769 rtnl_unlock(); 1770 up_write(&pernet_ops_rwsem); 1771 return err; 1772 } 1773 EXPORT_SYMBOL(unregister_netdevice_notifier); 1774 1775 static int __register_netdevice_notifier_net(struct net *net, 1776 struct notifier_block *nb, 1777 bool ignore_call_fail) 1778 { 1779 int err; 1780 1781 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1782 if (err) 1783 return err; 1784 if (dev_boot_phase) 1785 return 0; 1786 1787 err = call_netdevice_register_net_notifiers(nb, net); 1788 if (err && !ignore_call_fail) 1789 goto chain_unregister; 1790 1791 return 0; 1792 1793 chain_unregister: 1794 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1795 return err; 1796 } 1797 1798 static int __unregister_netdevice_notifier_net(struct net *net, 1799 struct notifier_block *nb) 1800 { 1801 int err; 1802 1803 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1804 if (err) 1805 return err; 1806 1807 call_netdevice_unregister_net_notifiers(nb, net); 1808 return 0; 1809 } 1810 1811 /** 1812 * register_netdevice_notifier_net - register a per-netns network notifier block 1813 * @net: network namespace 1814 * @nb: notifier 1815 * 1816 * Register a notifier to be called when network device events occur. 1817 * The notifier passed is linked into the kernel structures and must 1818 * not be reused until it has been unregistered. A negative errno code 1819 * is returned on a failure. 1820 * 1821 * When registered all registration and up events are replayed 1822 * to the new notifier to allow device to have a race free 1823 * view of the network device list. 1824 */ 1825 1826 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1827 { 1828 int err; 1829 1830 rtnl_lock(); 1831 err = __register_netdevice_notifier_net(net, nb, false); 1832 rtnl_unlock(); 1833 return err; 1834 } 1835 EXPORT_SYMBOL(register_netdevice_notifier_net); 1836 1837 /** 1838 * unregister_netdevice_notifier_net - unregister a per-netns 1839 * network notifier block 1840 * @net: network namespace 1841 * @nb: notifier 1842 * 1843 * Unregister a notifier previously registered by 1844 * register_netdevice_notifier_net(). The notifier is unlinked from the 1845 * kernel structures and may then be reused. A negative errno code 1846 * is returned on a failure. 1847 * 1848 * After unregistering unregister and down device events are synthesized 1849 * for all devices on the device list to the removed notifier to remove 1850 * the need for special case cleanup code. 1851 */ 1852 1853 int unregister_netdevice_notifier_net(struct net *net, 1854 struct notifier_block *nb) 1855 { 1856 int err; 1857 1858 rtnl_lock(); 1859 err = __unregister_netdevice_notifier_net(net, nb); 1860 rtnl_unlock(); 1861 return err; 1862 } 1863 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1864 1865 static void __move_netdevice_notifier_net(struct net *src_net, 1866 struct net *dst_net, 1867 struct notifier_block *nb) 1868 { 1869 __unregister_netdevice_notifier_net(src_net, nb); 1870 __register_netdevice_notifier_net(dst_net, nb, true); 1871 } 1872 1873 int register_netdevice_notifier_dev_net(struct net_device *dev, 1874 struct notifier_block *nb, 1875 struct netdev_net_notifier *nn) 1876 { 1877 int err; 1878 1879 rtnl_lock(); 1880 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1881 if (!err) { 1882 nn->nb = nb; 1883 list_add(&nn->list, &dev->net_notifier_list); 1884 } 1885 rtnl_unlock(); 1886 return err; 1887 } 1888 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1889 1890 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1891 struct notifier_block *nb, 1892 struct netdev_net_notifier *nn) 1893 { 1894 int err; 1895 1896 rtnl_lock(); 1897 list_del(&nn->list); 1898 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1899 rtnl_unlock(); 1900 return err; 1901 } 1902 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1903 1904 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1905 struct net *net) 1906 { 1907 struct netdev_net_notifier *nn; 1908 1909 list_for_each_entry(nn, &dev->net_notifier_list, list) 1910 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1911 } 1912 1913 /** 1914 * call_netdevice_notifiers_info - call all network notifier blocks 1915 * @val: value passed unmodified to notifier function 1916 * @info: notifier information data 1917 * 1918 * Call all network notifier blocks. Parameters and return value 1919 * are as for raw_notifier_call_chain(). 1920 */ 1921 1922 static int call_netdevice_notifiers_info(unsigned long val, 1923 struct netdev_notifier_info *info) 1924 { 1925 struct net *net = dev_net(info->dev); 1926 int ret; 1927 1928 ASSERT_RTNL(); 1929 1930 /* Run per-netns notifier block chain first, then run the global one. 1931 * Hopefully, one day, the global one is going to be removed after 1932 * all notifier block registrators get converted to be per-netns. 1933 */ 1934 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1935 if (ret & NOTIFY_STOP_MASK) 1936 return ret; 1937 return raw_notifier_call_chain(&netdev_chain, val, info); 1938 } 1939 1940 /** 1941 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1942 * for and rollback on error 1943 * @val_up: value passed unmodified to notifier function 1944 * @val_down: value passed unmodified to the notifier function when 1945 * recovering from an error on @val_up 1946 * @info: notifier information data 1947 * 1948 * Call all per-netns network notifier blocks, but not notifier blocks on 1949 * the global notifier chain. Parameters and return value are as for 1950 * raw_notifier_call_chain_robust(). 1951 */ 1952 1953 static int 1954 call_netdevice_notifiers_info_robust(unsigned long val_up, 1955 unsigned long val_down, 1956 struct netdev_notifier_info *info) 1957 { 1958 struct net *net = dev_net(info->dev); 1959 1960 ASSERT_RTNL(); 1961 1962 return raw_notifier_call_chain_robust(&net->netdev_chain, 1963 val_up, val_down, info); 1964 } 1965 1966 static int call_netdevice_notifiers_extack(unsigned long val, 1967 struct net_device *dev, 1968 struct netlink_ext_ack *extack) 1969 { 1970 struct netdev_notifier_info info = { 1971 .dev = dev, 1972 .extack = extack, 1973 }; 1974 1975 return call_netdevice_notifiers_info(val, &info); 1976 } 1977 1978 /** 1979 * call_netdevice_notifiers - call all network notifier blocks 1980 * @val: value passed unmodified to notifier function 1981 * @dev: net_device pointer passed unmodified to notifier function 1982 * 1983 * Call all network notifier blocks. Parameters and return value 1984 * are as for raw_notifier_call_chain(). 1985 */ 1986 1987 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1988 { 1989 return call_netdevice_notifiers_extack(val, dev, NULL); 1990 } 1991 EXPORT_SYMBOL(call_netdevice_notifiers); 1992 1993 /** 1994 * call_netdevice_notifiers_mtu - call all network notifier blocks 1995 * @val: value passed unmodified to notifier function 1996 * @dev: net_device pointer passed unmodified to notifier function 1997 * @arg: additional u32 argument passed to the notifier function 1998 * 1999 * Call all network notifier blocks. Parameters and return value 2000 * are as for raw_notifier_call_chain(). 2001 */ 2002 static int call_netdevice_notifiers_mtu(unsigned long val, 2003 struct net_device *dev, u32 arg) 2004 { 2005 struct netdev_notifier_info_ext info = { 2006 .info.dev = dev, 2007 .ext.mtu = arg, 2008 }; 2009 2010 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2011 2012 return call_netdevice_notifiers_info(val, &info.info); 2013 } 2014 2015 #ifdef CONFIG_NET_INGRESS 2016 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2017 2018 void net_inc_ingress_queue(void) 2019 { 2020 static_branch_inc(&ingress_needed_key); 2021 } 2022 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2023 2024 void net_dec_ingress_queue(void) 2025 { 2026 static_branch_dec(&ingress_needed_key); 2027 } 2028 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2029 #endif 2030 2031 #ifdef CONFIG_NET_EGRESS 2032 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2033 2034 void net_inc_egress_queue(void) 2035 { 2036 static_branch_inc(&egress_needed_key); 2037 } 2038 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2039 2040 void net_dec_egress_queue(void) 2041 { 2042 static_branch_dec(&egress_needed_key); 2043 } 2044 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2045 #endif 2046 2047 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2048 EXPORT_SYMBOL(netstamp_needed_key); 2049 #ifdef CONFIG_JUMP_LABEL 2050 static atomic_t netstamp_needed_deferred; 2051 static atomic_t netstamp_wanted; 2052 static void netstamp_clear(struct work_struct *work) 2053 { 2054 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2055 int wanted; 2056 2057 wanted = atomic_add_return(deferred, &netstamp_wanted); 2058 if (wanted > 0) 2059 static_branch_enable(&netstamp_needed_key); 2060 else 2061 static_branch_disable(&netstamp_needed_key); 2062 } 2063 static DECLARE_WORK(netstamp_work, netstamp_clear); 2064 #endif 2065 2066 void net_enable_timestamp(void) 2067 { 2068 #ifdef CONFIG_JUMP_LABEL 2069 int wanted = atomic_read(&netstamp_wanted); 2070 2071 while (wanted > 0) { 2072 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2073 return; 2074 } 2075 atomic_inc(&netstamp_needed_deferred); 2076 schedule_work(&netstamp_work); 2077 #else 2078 static_branch_inc(&netstamp_needed_key); 2079 #endif 2080 } 2081 EXPORT_SYMBOL(net_enable_timestamp); 2082 2083 void net_disable_timestamp(void) 2084 { 2085 #ifdef CONFIG_JUMP_LABEL 2086 int wanted = atomic_read(&netstamp_wanted); 2087 2088 while (wanted > 1) { 2089 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2090 return; 2091 } 2092 atomic_dec(&netstamp_needed_deferred); 2093 schedule_work(&netstamp_work); 2094 #else 2095 static_branch_dec(&netstamp_needed_key); 2096 #endif 2097 } 2098 EXPORT_SYMBOL(net_disable_timestamp); 2099 2100 static inline void net_timestamp_set(struct sk_buff *skb) 2101 { 2102 skb->tstamp = 0; 2103 skb->mono_delivery_time = 0; 2104 if (static_branch_unlikely(&netstamp_needed_key)) 2105 skb->tstamp = ktime_get_real(); 2106 } 2107 2108 #define net_timestamp_check(COND, SKB) \ 2109 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2110 if ((COND) && !(SKB)->tstamp) \ 2111 (SKB)->tstamp = ktime_get_real(); \ 2112 } \ 2113 2114 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2115 { 2116 return __is_skb_forwardable(dev, skb, true); 2117 } 2118 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2119 2120 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2121 bool check_mtu) 2122 { 2123 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2124 2125 if (likely(!ret)) { 2126 skb->protocol = eth_type_trans(skb, dev); 2127 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2128 } 2129 2130 return ret; 2131 } 2132 2133 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2134 { 2135 return __dev_forward_skb2(dev, skb, true); 2136 } 2137 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2138 2139 /** 2140 * dev_forward_skb - loopback an skb to another netif 2141 * 2142 * @dev: destination network device 2143 * @skb: buffer to forward 2144 * 2145 * return values: 2146 * NET_RX_SUCCESS (no congestion) 2147 * NET_RX_DROP (packet was dropped, but freed) 2148 * 2149 * dev_forward_skb can be used for injecting an skb from the 2150 * start_xmit function of one device into the receive queue 2151 * of another device. 2152 * 2153 * The receiving device may be in another namespace, so 2154 * we have to clear all information in the skb that could 2155 * impact namespace isolation. 2156 */ 2157 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2158 { 2159 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2160 } 2161 EXPORT_SYMBOL_GPL(dev_forward_skb); 2162 2163 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2164 { 2165 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2166 } 2167 2168 static inline int deliver_skb(struct sk_buff *skb, 2169 struct packet_type *pt_prev, 2170 struct net_device *orig_dev) 2171 { 2172 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2173 return -ENOMEM; 2174 refcount_inc(&skb->users); 2175 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2176 } 2177 2178 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2179 struct packet_type **pt, 2180 struct net_device *orig_dev, 2181 __be16 type, 2182 struct list_head *ptype_list) 2183 { 2184 struct packet_type *ptype, *pt_prev = *pt; 2185 2186 list_for_each_entry_rcu(ptype, ptype_list, list) { 2187 if (ptype->type != type) 2188 continue; 2189 if (pt_prev) 2190 deliver_skb(skb, pt_prev, orig_dev); 2191 pt_prev = ptype; 2192 } 2193 *pt = pt_prev; 2194 } 2195 2196 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2197 { 2198 if (!ptype->af_packet_priv || !skb->sk) 2199 return false; 2200 2201 if (ptype->id_match) 2202 return ptype->id_match(ptype, skb->sk); 2203 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2204 return true; 2205 2206 return false; 2207 } 2208 2209 /** 2210 * dev_nit_active - return true if any network interface taps are in use 2211 * 2212 * @dev: network device to check for the presence of taps 2213 */ 2214 bool dev_nit_active(struct net_device *dev) 2215 { 2216 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2217 } 2218 EXPORT_SYMBOL_GPL(dev_nit_active); 2219 2220 /* 2221 * Support routine. Sends outgoing frames to any network 2222 * taps currently in use. 2223 */ 2224 2225 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2226 { 2227 struct packet_type *ptype; 2228 struct sk_buff *skb2 = NULL; 2229 struct packet_type *pt_prev = NULL; 2230 struct list_head *ptype_list = &ptype_all; 2231 2232 rcu_read_lock(); 2233 again: 2234 list_for_each_entry_rcu(ptype, ptype_list, list) { 2235 if (ptype->ignore_outgoing) 2236 continue; 2237 2238 /* Never send packets back to the socket 2239 * they originated from - MvS (miquels@drinkel.ow.org) 2240 */ 2241 if (skb_loop_sk(ptype, skb)) 2242 continue; 2243 2244 if (pt_prev) { 2245 deliver_skb(skb2, pt_prev, skb->dev); 2246 pt_prev = ptype; 2247 continue; 2248 } 2249 2250 /* need to clone skb, done only once */ 2251 skb2 = skb_clone(skb, GFP_ATOMIC); 2252 if (!skb2) 2253 goto out_unlock; 2254 2255 net_timestamp_set(skb2); 2256 2257 /* skb->nh should be correctly 2258 * set by sender, so that the second statement is 2259 * just protection against buggy protocols. 2260 */ 2261 skb_reset_mac_header(skb2); 2262 2263 if (skb_network_header(skb2) < skb2->data || 2264 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2265 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2266 ntohs(skb2->protocol), 2267 dev->name); 2268 skb_reset_network_header(skb2); 2269 } 2270 2271 skb2->transport_header = skb2->network_header; 2272 skb2->pkt_type = PACKET_OUTGOING; 2273 pt_prev = ptype; 2274 } 2275 2276 if (ptype_list == &ptype_all) { 2277 ptype_list = &dev->ptype_all; 2278 goto again; 2279 } 2280 out_unlock: 2281 if (pt_prev) { 2282 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2283 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2284 else 2285 kfree_skb(skb2); 2286 } 2287 rcu_read_unlock(); 2288 } 2289 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2290 2291 /** 2292 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2293 * @dev: Network device 2294 * @txq: number of queues available 2295 * 2296 * If real_num_tx_queues is changed the tc mappings may no longer be 2297 * valid. To resolve this verify the tc mapping remains valid and if 2298 * not NULL the mapping. With no priorities mapping to this 2299 * offset/count pair it will no longer be used. In the worst case TC0 2300 * is invalid nothing can be done so disable priority mappings. If is 2301 * expected that drivers will fix this mapping if they can before 2302 * calling netif_set_real_num_tx_queues. 2303 */ 2304 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2305 { 2306 int i; 2307 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2308 2309 /* If TC0 is invalidated disable TC mapping */ 2310 if (tc->offset + tc->count > txq) { 2311 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2312 dev->num_tc = 0; 2313 return; 2314 } 2315 2316 /* Invalidated prio to tc mappings set to TC0 */ 2317 for (i = 1; i < TC_BITMASK + 1; i++) { 2318 int q = netdev_get_prio_tc_map(dev, i); 2319 2320 tc = &dev->tc_to_txq[q]; 2321 if (tc->offset + tc->count > txq) { 2322 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2323 i, q); 2324 netdev_set_prio_tc_map(dev, i, 0); 2325 } 2326 } 2327 } 2328 2329 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2330 { 2331 if (dev->num_tc) { 2332 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2333 int i; 2334 2335 /* walk through the TCs and see if it falls into any of them */ 2336 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2337 if ((txq - tc->offset) < tc->count) 2338 return i; 2339 } 2340 2341 /* didn't find it, just return -1 to indicate no match */ 2342 return -1; 2343 } 2344 2345 return 0; 2346 } 2347 EXPORT_SYMBOL(netdev_txq_to_tc); 2348 2349 #ifdef CONFIG_XPS 2350 static struct static_key xps_needed __read_mostly; 2351 static struct static_key xps_rxqs_needed __read_mostly; 2352 static DEFINE_MUTEX(xps_map_mutex); 2353 #define xmap_dereference(P) \ 2354 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2355 2356 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2357 struct xps_dev_maps *old_maps, int tci, u16 index) 2358 { 2359 struct xps_map *map = NULL; 2360 int pos; 2361 2362 if (dev_maps) 2363 map = xmap_dereference(dev_maps->attr_map[tci]); 2364 if (!map) 2365 return false; 2366 2367 for (pos = map->len; pos--;) { 2368 if (map->queues[pos] != index) 2369 continue; 2370 2371 if (map->len > 1) { 2372 map->queues[pos] = map->queues[--map->len]; 2373 break; 2374 } 2375 2376 if (old_maps) 2377 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2378 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2379 kfree_rcu(map, rcu); 2380 return false; 2381 } 2382 2383 return true; 2384 } 2385 2386 static bool remove_xps_queue_cpu(struct net_device *dev, 2387 struct xps_dev_maps *dev_maps, 2388 int cpu, u16 offset, u16 count) 2389 { 2390 int num_tc = dev_maps->num_tc; 2391 bool active = false; 2392 int tci; 2393 2394 for (tci = cpu * num_tc; num_tc--; tci++) { 2395 int i, j; 2396 2397 for (i = count, j = offset; i--; j++) { 2398 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2399 break; 2400 } 2401 2402 active |= i < 0; 2403 } 2404 2405 return active; 2406 } 2407 2408 static void reset_xps_maps(struct net_device *dev, 2409 struct xps_dev_maps *dev_maps, 2410 enum xps_map_type type) 2411 { 2412 static_key_slow_dec_cpuslocked(&xps_needed); 2413 if (type == XPS_RXQS) 2414 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2415 2416 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2417 2418 kfree_rcu(dev_maps, rcu); 2419 } 2420 2421 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2422 u16 offset, u16 count) 2423 { 2424 struct xps_dev_maps *dev_maps; 2425 bool active = false; 2426 int i, j; 2427 2428 dev_maps = xmap_dereference(dev->xps_maps[type]); 2429 if (!dev_maps) 2430 return; 2431 2432 for (j = 0; j < dev_maps->nr_ids; j++) 2433 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2434 if (!active) 2435 reset_xps_maps(dev, dev_maps, type); 2436 2437 if (type == XPS_CPUS) { 2438 for (i = offset + (count - 1); count--; i--) 2439 netdev_queue_numa_node_write( 2440 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2441 } 2442 } 2443 2444 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2445 u16 count) 2446 { 2447 if (!static_key_false(&xps_needed)) 2448 return; 2449 2450 cpus_read_lock(); 2451 mutex_lock(&xps_map_mutex); 2452 2453 if (static_key_false(&xps_rxqs_needed)) 2454 clean_xps_maps(dev, XPS_RXQS, offset, count); 2455 2456 clean_xps_maps(dev, XPS_CPUS, offset, count); 2457 2458 mutex_unlock(&xps_map_mutex); 2459 cpus_read_unlock(); 2460 } 2461 2462 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2463 { 2464 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2465 } 2466 2467 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2468 u16 index, bool is_rxqs_map) 2469 { 2470 struct xps_map *new_map; 2471 int alloc_len = XPS_MIN_MAP_ALLOC; 2472 int i, pos; 2473 2474 for (pos = 0; map && pos < map->len; pos++) { 2475 if (map->queues[pos] != index) 2476 continue; 2477 return map; 2478 } 2479 2480 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2481 if (map) { 2482 if (pos < map->alloc_len) 2483 return map; 2484 2485 alloc_len = map->alloc_len * 2; 2486 } 2487 2488 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2489 * map 2490 */ 2491 if (is_rxqs_map) 2492 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2493 else 2494 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2495 cpu_to_node(attr_index)); 2496 if (!new_map) 2497 return NULL; 2498 2499 for (i = 0; i < pos; i++) 2500 new_map->queues[i] = map->queues[i]; 2501 new_map->alloc_len = alloc_len; 2502 new_map->len = pos; 2503 2504 return new_map; 2505 } 2506 2507 /* Copy xps maps at a given index */ 2508 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2509 struct xps_dev_maps *new_dev_maps, int index, 2510 int tc, bool skip_tc) 2511 { 2512 int i, tci = index * dev_maps->num_tc; 2513 struct xps_map *map; 2514 2515 /* copy maps belonging to foreign traffic classes */ 2516 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2517 if (i == tc && skip_tc) 2518 continue; 2519 2520 /* fill in the new device map from the old device map */ 2521 map = xmap_dereference(dev_maps->attr_map[tci]); 2522 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2523 } 2524 } 2525 2526 /* Must be called under cpus_read_lock */ 2527 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2528 u16 index, enum xps_map_type type) 2529 { 2530 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2531 const unsigned long *online_mask = NULL; 2532 bool active = false, copy = false; 2533 int i, j, tci, numa_node_id = -2; 2534 int maps_sz, num_tc = 1, tc = 0; 2535 struct xps_map *map, *new_map; 2536 unsigned int nr_ids; 2537 2538 if (dev->num_tc) { 2539 /* Do not allow XPS on subordinate device directly */ 2540 num_tc = dev->num_tc; 2541 if (num_tc < 0) 2542 return -EINVAL; 2543 2544 /* If queue belongs to subordinate dev use its map */ 2545 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2546 2547 tc = netdev_txq_to_tc(dev, index); 2548 if (tc < 0) 2549 return -EINVAL; 2550 } 2551 2552 mutex_lock(&xps_map_mutex); 2553 2554 dev_maps = xmap_dereference(dev->xps_maps[type]); 2555 if (type == XPS_RXQS) { 2556 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2557 nr_ids = dev->num_rx_queues; 2558 } else { 2559 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2560 if (num_possible_cpus() > 1) 2561 online_mask = cpumask_bits(cpu_online_mask); 2562 nr_ids = nr_cpu_ids; 2563 } 2564 2565 if (maps_sz < L1_CACHE_BYTES) 2566 maps_sz = L1_CACHE_BYTES; 2567 2568 /* The old dev_maps could be larger or smaller than the one we're 2569 * setting up now, as dev->num_tc or nr_ids could have been updated in 2570 * between. We could try to be smart, but let's be safe instead and only 2571 * copy foreign traffic classes if the two map sizes match. 2572 */ 2573 if (dev_maps && 2574 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2575 copy = true; 2576 2577 /* allocate memory for queue storage */ 2578 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2579 j < nr_ids;) { 2580 if (!new_dev_maps) { 2581 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2582 if (!new_dev_maps) { 2583 mutex_unlock(&xps_map_mutex); 2584 return -ENOMEM; 2585 } 2586 2587 new_dev_maps->nr_ids = nr_ids; 2588 new_dev_maps->num_tc = num_tc; 2589 } 2590 2591 tci = j * num_tc + tc; 2592 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2593 2594 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2595 if (!map) 2596 goto error; 2597 2598 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2599 } 2600 2601 if (!new_dev_maps) 2602 goto out_no_new_maps; 2603 2604 if (!dev_maps) { 2605 /* Increment static keys at most once per type */ 2606 static_key_slow_inc_cpuslocked(&xps_needed); 2607 if (type == XPS_RXQS) 2608 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2609 } 2610 2611 for (j = 0; j < nr_ids; j++) { 2612 bool skip_tc = false; 2613 2614 tci = j * num_tc + tc; 2615 if (netif_attr_test_mask(j, mask, nr_ids) && 2616 netif_attr_test_online(j, online_mask, nr_ids)) { 2617 /* add tx-queue to CPU/rx-queue maps */ 2618 int pos = 0; 2619 2620 skip_tc = true; 2621 2622 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2623 while ((pos < map->len) && (map->queues[pos] != index)) 2624 pos++; 2625 2626 if (pos == map->len) 2627 map->queues[map->len++] = index; 2628 #ifdef CONFIG_NUMA 2629 if (type == XPS_CPUS) { 2630 if (numa_node_id == -2) 2631 numa_node_id = cpu_to_node(j); 2632 else if (numa_node_id != cpu_to_node(j)) 2633 numa_node_id = -1; 2634 } 2635 #endif 2636 } 2637 2638 if (copy) 2639 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2640 skip_tc); 2641 } 2642 2643 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2644 2645 /* Cleanup old maps */ 2646 if (!dev_maps) 2647 goto out_no_old_maps; 2648 2649 for (j = 0; j < dev_maps->nr_ids; j++) { 2650 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2651 map = xmap_dereference(dev_maps->attr_map[tci]); 2652 if (!map) 2653 continue; 2654 2655 if (copy) { 2656 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2657 if (map == new_map) 2658 continue; 2659 } 2660 2661 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2662 kfree_rcu(map, rcu); 2663 } 2664 } 2665 2666 old_dev_maps = dev_maps; 2667 2668 out_no_old_maps: 2669 dev_maps = new_dev_maps; 2670 active = true; 2671 2672 out_no_new_maps: 2673 if (type == XPS_CPUS) 2674 /* update Tx queue numa node */ 2675 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2676 (numa_node_id >= 0) ? 2677 numa_node_id : NUMA_NO_NODE); 2678 2679 if (!dev_maps) 2680 goto out_no_maps; 2681 2682 /* removes tx-queue from unused CPUs/rx-queues */ 2683 for (j = 0; j < dev_maps->nr_ids; j++) { 2684 tci = j * dev_maps->num_tc; 2685 2686 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2687 if (i == tc && 2688 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2689 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2690 continue; 2691 2692 active |= remove_xps_queue(dev_maps, 2693 copy ? old_dev_maps : NULL, 2694 tci, index); 2695 } 2696 } 2697 2698 if (old_dev_maps) 2699 kfree_rcu(old_dev_maps, rcu); 2700 2701 /* free map if not active */ 2702 if (!active) 2703 reset_xps_maps(dev, dev_maps, type); 2704 2705 out_no_maps: 2706 mutex_unlock(&xps_map_mutex); 2707 2708 return 0; 2709 error: 2710 /* remove any maps that we added */ 2711 for (j = 0; j < nr_ids; j++) { 2712 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2713 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2714 map = copy ? 2715 xmap_dereference(dev_maps->attr_map[tci]) : 2716 NULL; 2717 if (new_map && new_map != map) 2718 kfree(new_map); 2719 } 2720 } 2721 2722 mutex_unlock(&xps_map_mutex); 2723 2724 kfree(new_dev_maps); 2725 return -ENOMEM; 2726 } 2727 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2728 2729 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2730 u16 index) 2731 { 2732 int ret; 2733 2734 cpus_read_lock(); 2735 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2736 cpus_read_unlock(); 2737 2738 return ret; 2739 } 2740 EXPORT_SYMBOL(netif_set_xps_queue); 2741 2742 #endif 2743 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2744 { 2745 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2746 2747 /* Unbind any subordinate channels */ 2748 while (txq-- != &dev->_tx[0]) { 2749 if (txq->sb_dev) 2750 netdev_unbind_sb_channel(dev, txq->sb_dev); 2751 } 2752 } 2753 2754 void netdev_reset_tc(struct net_device *dev) 2755 { 2756 #ifdef CONFIG_XPS 2757 netif_reset_xps_queues_gt(dev, 0); 2758 #endif 2759 netdev_unbind_all_sb_channels(dev); 2760 2761 /* Reset TC configuration of device */ 2762 dev->num_tc = 0; 2763 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2764 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2765 } 2766 EXPORT_SYMBOL(netdev_reset_tc); 2767 2768 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2769 { 2770 if (tc >= dev->num_tc) 2771 return -EINVAL; 2772 2773 #ifdef CONFIG_XPS 2774 netif_reset_xps_queues(dev, offset, count); 2775 #endif 2776 dev->tc_to_txq[tc].count = count; 2777 dev->tc_to_txq[tc].offset = offset; 2778 return 0; 2779 } 2780 EXPORT_SYMBOL(netdev_set_tc_queue); 2781 2782 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2783 { 2784 if (num_tc > TC_MAX_QUEUE) 2785 return -EINVAL; 2786 2787 #ifdef CONFIG_XPS 2788 netif_reset_xps_queues_gt(dev, 0); 2789 #endif 2790 netdev_unbind_all_sb_channels(dev); 2791 2792 dev->num_tc = num_tc; 2793 return 0; 2794 } 2795 EXPORT_SYMBOL(netdev_set_num_tc); 2796 2797 void netdev_unbind_sb_channel(struct net_device *dev, 2798 struct net_device *sb_dev) 2799 { 2800 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2801 2802 #ifdef CONFIG_XPS 2803 netif_reset_xps_queues_gt(sb_dev, 0); 2804 #endif 2805 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2806 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2807 2808 while (txq-- != &dev->_tx[0]) { 2809 if (txq->sb_dev == sb_dev) 2810 txq->sb_dev = NULL; 2811 } 2812 } 2813 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2814 2815 int netdev_bind_sb_channel_queue(struct net_device *dev, 2816 struct net_device *sb_dev, 2817 u8 tc, u16 count, u16 offset) 2818 { 2819 /* Make certain the sb_dev and dev are already configured */ 2820 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2821 return -EINVAL; 2822 2823 /* We cannot hand out queues we don't have */ 2824 if ((offset + count) > dev->real_num_tx_queues) 2825 return -EINVAL; 2826 2827 /* Record the mapping */ 2828 sb_dev->tc_to_txq[tc].count = count; 2829 sb_dev->tc_to_txq[tc].offset = offset; 2830 2831 /* Provide a way for Tx queue to find the tc_to_txq map or 2832 * XPS map for itself. 2833 */ 2834 while (count--) 2835 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2836 2837 return 0; 2838 } 2839 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2840 2841 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2842 { 2843 /* Do not use a multiqueue device to represent a subordinate channel */ 2844 if (netif_is_multiqueue(dev)) 2845 return -ENODEV; 2846 2847 /* We allow channels 1 - 32767 to be used for subordinate channels. 2848 * Channel 0 is meant to be "native" mode and used only to represent 2849 * the main root device. We allow writing 0 to reset the device back 2850 * to normal mode after being used as a subordinate channel. 2851 */ 2852 if (channel > S16_MAX) 2853 return -EINVAL; 2854 2855 dev->num_tc = -channel; 2856 2857 return 0; 2858 } 2859 EXPORT_SYMBOL(netdev_set_sb_channel); 2860 2861 /* 2862 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2863 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2864 */ 2865 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2866 { 2867 bool disabling; 2868 int rc; 2869 2870 disabling = txq < dev->real_num_tx_queues; 2871 2872 if (txq < 1 || txq > dev->num_tx_queues) 2873 return -EINVAL; 2874 2875 if (dev->reg_state == NETREG_REGISTERED || 2876 dev->reg_state == NETREG_UNREGISTERING) { 2877 ASSERT_RTNL(); 2878 2879 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2880 txq); 2881 if (rc) 2882 return rc; 2883 2884 if (dev->num_tc) 2885 netif_setup_tc(dev, txq); 2886 2887 dev_qdisc_change_real_num_tx(dev, txq); 2888 2889 dev->real_num_tx_queues = txq; 2890 2891 if (disabling) { 2892 synchronize_net(); 2893 qdisc_reset_all_tx_gt(dev, txq); 2894 #ifdef CONFIG_XPS 2895 netif_reset_xps_queues_gt(dev, txq); 2896 #endif 2897 } 2898 } else { 2899 dev->real_num_tx_queues = txq; 2900 } 2901 2902 return 0; 2903 } 2904 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2905 2906 #ifdef CONFIG_SYSFS 2907 /** 2908 * netif_set_real_num_rx_queues - set actual number of RX queues used 2909 * @dev: Network device 2910 * @rxq: Actual number of RX queues 2911 * 2912 * This must be called either with the rtnl_lock held or before 2913 * registration of the net device. Returns 0 on success, or a 2914 * negative error code. If called before registration, it always 2915 * succeeds. 2916 */ 2917 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2918 { 2919 int rc; 2920 2921 if (rxq < 1 || rxq > dev->num_rx_queues) 2922 return -EINVAL; 2923 2924 if (dev->reg_state == NETREG_REGISTERED) { 2925 ASSERT_RTNL(); 2926 2927 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2928 rxq); 2929 if (rc) 2930 return rc; 2931 } 2932 2933 dev->real_num_rx_queues = rxq; 2934 return 0; 2935 } 2936 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2937 #endif 2938 2939 /** 2940 * netif_set_real_num_queues - set actual number of RX and TX queues used 2941 * @dev: Network device 2942 * @txq: Actual number of TX queues 2943 * @rxq: Actual number of RX queues 2944 * 2945 * Set the real number of both TX and RX queues. 2946 * Does nothing if the number of queues is already correct. 2947 */ 2948 int netif_set_real_num_queues(struct net_device *dev, 2949 unsigned int txq, unsigned int rxq) 2950 { 2951 unsigned int old_rxq = dev->real_num_rx_queues; 2952 int err; 2953 2954 if (txq < 1 || txq > dev->num_tx_queues || 2955 rxq < 1 || rxq > dev->num_rx_queues) 2956 return -EINVAL; 2957 2958 /* Start from increases, so the error path only does decreases - 2959 * decreases can't fail. 2960 */ 2961 if (rxq > dev->real_num_rx_queues) { 2962 err = netif_set_real_num_rx_queues(dev, rxq); 2963 if (err) 2964 return err; 2965 } 2966 if (txq > dev->real_num_tx_queues) { 2967 err = netif_set_real_num_tx_queues(dev, txq); 2968 if (err) 2969 goto undo_rx; 2970 } 2971 if (rxq < dev->real_num_rx_queues) 2972 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2973 if (txq < dev->real_num_tx_queues) 2974 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2975 2976 return 0; 2977 undo_rx: 2978 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2979 return err; 2980 } 2981 EXPORT_SYMBOL(netif_set_real_num_queues); 2982 2983 /** 2984 * netif_set_tso_max_size() - set the max size of TSO frames supported 2985 * @dev: netdev to update 2986 * @size: max skb->len of a TSO frame 2987 * 2988 * Set the limit on the size of TSO super-frames the device can handle. 2989 * Unless explicitly set the stack will assume the value of 2990 * %GSO_LEGACY_MAX_SIZE. 2991 */ 2992 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 2993 { 2994 dev->tso_max_size = min(GSO_MAX_SIZE, size); 2995 if (size < READ_ONCE(dev->gso_max_size)) 2996 netif_set_gso_max_size(dev, size); 2997 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 2998 netif_set_gso_ipv4_max_size(dev, size); 2999 } 3000 EXPORT_SYMBOL(netif_set_tso_max_size); 3001 3002 /** 3003 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3004 * @dev: netdev to update 3005 * @segs: max number of TCP segments 3006 * 3007 * Set the limit on the number of TCP segments the device can generate from 3008 * a single TSO super-frame. 3009 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3010 */ 3011 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3012 { 3013 dev->tso_max_segs = segs; 3014 if (segs < READ_ONCE(dev->gso_max_segs)) 3015 netif_set_gso_max_segs(dev, segs); 3016 } 3017 EXPORT_SYMBOL(netif_set_tso_max_segs); 3018 3019 /** 3020 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3021 * @to: netdev to update 3022 * @from: netdev from which to copy the limits 3023 */ 3024 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3025 { 3026 netif_set_tso_max_size(to, from->tso_max_size); 3027 netif_set_tso_max_segs(to, from->tso_max_segs); 3028 } 3029 EXPORT_SYMBOL(netif_inherit_tso_max); 3030 3031 /** 3032 * netif_get_num_default_rss_queues - default number of RSS queues 3033 * 3034 * Default value is the number of physical cores if there are only 1 or 2, or 3035 * divided by 2 if there are more. 3036 */ 3037 int netif_get_num_default_rss_queues(void) 3038 { 3039 cpumask_var_t cpus; 3040 int cpu, count = 0; 3041 3042 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3043 return 1; 3044 3045 cpumask_copy(cpus, cpu_online_mask); 3046 for_each_cpu(cpu, cpus) { 3047 ++count; 3048 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3049 } 3050 free_cpumask_var(cpus); 3051 3052 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3053 } 3054 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3055 3056 static void __netif_reschedule(struct Qdisc *q) 3057 { 3058 struct softnet_data *sd; 3059 unsigned long flags; 3060 3061 local_irq_save(flags); 3062 sd = this_cpu_ptr(&softnet_data); 3063 q->next_sched = NULL; 3064 *sd->output_queue_tailp = q; 3065 sd->output_queue_tailp = &q->next_sched; 3066 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3067 local_irq_restore(flags); 3068 } 3069 3070 void __netif_schedule(struct Qdisc *q) 3071 { 3072 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3073 __netif_reschedule(q); 3074 } 3075 EXPORT_SYMBOL(__netif_schedule); 3076 3077 struct dev_kfree_skb_cb { 3078 enum skb_free_reason reason; 3079 }; 3080 3081 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3082 { 3083 return (struct dev_kfree_skb_cb *)skb->cb; 3084 } 3085 3086 void netif_schedule_queue(struct netdev_queue *txq) 3087 { 3088 rcu_read_lock(); 3089 if (!netif_xmit_stopped(txq)) { 3090 struct Qdisc *q = rcu_dereference(txq->qdisc); 3091 3092 __netif_schedule(q); 3093 } 3094 rcu_read_unlock(); 3095 } 3096 EXPORT_SYMBOL(netif_schedule_queue); 3097 3098 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3099 { 3100 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3101 struct Qdisc *q; 3102 3103 rcu_read_lock(); 3104 q = rcu_dereference(dev_queue->qdisc); 3105 __netif_schedule(q); 3106 rcu_read_unlock(); 3107 } 3108 } 3109 EXPORT_SYMBOL(netif_tx_wake_queue); 3110 3111 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3112 { 3113 unsigned long flags; 3114 3115 if (unlikely(!skb)) 3116 return; 3117 3118 if (likely(refcount_read(&skb->users) == 1)) { 3119 smp_rmb(); 3120 refcount_set(&skb->users, 0); 3121 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3122 return; 3123 } 3124 get_kfree_skb_cb(skb)->reason = reason; 3125 local_irq_save(flags); 3126 skb->next = __this_cpu_read(softnet_data.completion_queue); 3127 __this_cpu_write(softnet_data.completion_queue, skb); 3128 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3129 local_irq_restore(flags); 3130 } 3131 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3132 3133 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3134 { 3135 if (in_hardirq() || irqs_disabled()) 3136 __dev_kfree_skb_irq(skb, reason); 3137 else 3138 dev_kfree_skb(skb); 3139 } 3140 EXPORT_SYMBOL(__dev_kfree_skb_any); 3141 3142 3143 /** 3144 * netif_device_detach - mark device as removed 3145 * @dev: network device 3146 * 3147 * Mark device as removed from system and therefore no longer available. 3148 */ 3149 void netif_device_detach(struct net_device *dev) 3150 { 3151 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3152 netif_running(dev)) { 3153 netif_tx_stop_all_queues(dev); 3154 } 3155 } 3156 EXPORT_SYMBOL(netif_device_detach); 3157 3158 /** 3159 * netif_device_attach - mark device as attached 3160 * @dev: network device 3161 * 3162 * Mark device as attached from system and restart if needed. 3163 */ 3164 void netif_device_attach(struct net_device *dev) 3165 { 3166 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3167 netif_running(dev)) { 3168 netif_tx_wake_all_queues(dev); 3169 __netdev_watchdog_up(dev); 3170 } 3171 } 3172 EXPORT_SYMBOL(netif_device_attach); 3173 3174 /* 3175 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3176 * to be used as a distribution range. 3177 */ 3178 static u16 skb_tx_hash(const struct net_device *dev, 3179 const struct net_device *sb_dev, 3180 struct sk_buff *skb) 3181 { 3182 u32 hash; 3183 u16 qoffset = 0; 3184 u16 qcount = dev->real_num_tx_queues; 3185 3186 if (dev->num_tc) { 3187 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3188 3189 qoffset = sb_dev->tc_to_txq[tc].offset; 3190 qcount = sb_dev->tc_to_txq[tc].count; 3191 if (unlikely(!qcount)) { 3192 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3193 sb_dev->name, qoffset, tc); 3194 qoffset = 0; 3195 qcount = dev->real_num_tx_queues; 3196 } 3197 } 3198 3199 if (skb_rx_queue_recorded(skb)) { 3200 hash = skb_get_rx_queue(skb); 3201 if (hash >= qoffset) 3202 hash -= qoffset; 3203 while (unlikely(hash >= qcount)) 3204 hash -= qcount; 3205 return hash + qoffset; 3206 } 3207 3208 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3209 } 3210 3211 static void skb_warn_bad_offload(const struct sk_buff *skb) 3212 { 3213 static const netdev_features_t null_features; 3214 struct net_device *dev = skb->dev; 3215 const char *name = ""; 3216 3217 if (!net_ratelimit()) 3218 return; 3219 3220 if (dev) { 3221 if (dev->dev.parent) 3222 name = dev_driver_string(dev->dev.parent); 3223 else 3224 name = netdev_name(dev); 3225 } 3226 skb_dump(KERN_WARNING, skb, false); 3227 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3228 name, dev ? &dev->features : &null_features, 3229 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3230 } 3231 3232 /* 3233 * Invalidate hardware checksum when packet is to be mangled, and 3234 * complete checksum manually on outgoing path. 3235 */ 3236 int skb_checksum_help(struct sk_buff *skb) 3237 { 3238 __wsum csum; 3239 int ret = 0, offset; 3240 3241 if (skb->ip_summed == CHECKSUM_COMPLETE) 3242 goto out_set_summed; 3243 3244 if (unlikely(skb_is_gso(skb))) { 3245 skb_warn_bad_offload(skb); 3246 return -EINVAL; 3247 } 3248 3249 /* Before computing a checksum, we should make sure no frag could 3250 * be modified by an external entity : checksum could be wrong. 3251 */ 3252 if (skb_has_shared_frag(skb)) { 3253 ret = __skb_linearize(skb); 3254 if (ret) 3255 goto out; 3256 } 3257 3258 offset = skb_checksum_start_offset(skb); 3259 ret = -EINVAL; 3260 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3261 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3262 goto out; 3263 } 3264 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3265 3266 offset += skb->csum_offset; 3267 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3268 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3269 goto out; 3270 } 3271 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3272 if (ret) 3273 goto out; 3274 3275 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3276 out_set_summed: 3277 skb->ip_summed = CHECKSUM_NONE; 3278 out: 3279 return ret; 3280 } 3281 EXPORT_SYMBOL(skb_checksum_help); 3282 3283 int skb_crc32c_csum_help(struct sk_buff *skb) 3284 { 3285 __le32 crc32c_csum; 3286 int ret = 0, offset, start; 3287 3288 if (skb->ip_summed != CHECKSUM_PARTIAL) 3289 goto out; 3290 3291 if (unlikely(skb_is_gso(skb))) 3292 goto out; 3293 3294 /* Before computing a checksum, we should make sure no frag could 3295 * be modified by an external entity : checksum could be wrong. 3296 */ 3297 if (unlikely(skb_has_shared_frag(skb))) { 3298 ret = __skb_linearize(skb); 3299 if (ret) 3300 goto out; 3301 } 3302 start = skb_checksum_start_offset(skb); 3303 offset = start + offsetof(struct sctphdr, checksum); 3304 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3305 ret = -EINVAL; 3306 goto out; 3307 } 3308 3309 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3310 if (ret) 3311 goto out; 3312 3313 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3314 skb->len - start, ~(__u32)0, 3315 crc32c_csum_stub)); 3316 *(__le32 *)(skb->data + offset) = crc32c_csum; 3317 skb->ip_summed = CHECKSUM_NONE; 3318 skb->csum_not_inet = 0; 3319 out: 3320 return ret; 3321 } 3322 3323 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3324 { 3325 __be16 type = skb->protocol; 3326 3327 /* Tunnel gso handlers can set protocol to ethernet. */ 3328 if (type == htons(ETH_P_TEB)) { 3329 struct ethhdr *eth; 3330 3331 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3332 return 0; 3333 3334 eth = (struct ethhdr *)skb->data; 3335 type = eth->h_proto; 3336 } 3337 3338 return __vlan_get_protocol(skb, type, depth); 3339 } 3340 3341 /* openvswitch calls this on rx path, so we need a different check. 3342 */ 3343 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3344 { 3345 if (tx_path) 3346 return skb->ip_summed != CHECKSUM_PARTIAL && 3347 skb->ip_summed != CHECKSUM_UNNECESSARY; 3348 3349 return skb->ip_summed == CHECKSUM_NONE; 3350 } 3351 3352 /** 3353 * __skb_gso_segment - Perform segmentation on skb. 3354 * @skb: buffer to segment 3355 * @features: features for the output path (see dev->features) 3356 * @tx_path: whether it is called in TX path 3357 * 3358 * This function segments the given skb and returns a list of segments. 3359 * 3360 * It may return NULL if the skb requires no segmentation. This is 3361 * only possible when GSO is used for verifying header integrity. 3362 * 3363 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3364 */ 3365 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3366 netdev_features_t features, bool tx_path) 3367 { 3368 struct sk_buff *segs; 3369 3370 if (unlikely(skb_needs_check(skb, tx_path))) { 3371 int err; 3372 3373 /* We're going to init ->check field in TCP or UDP header */ 3374 err = skb_cow_head(skb, 0); 3375 if (err < 0) 3376 return ERR_PTR(err); 3377 } 3378 3379 /* Only report GSO partial support if it will enable us to 3380 * support segmentation on this frame without needing additional 3381 * work. 3382 */ 3383 if (features & NETIF_F_GSO_PARTIAL) { 3384 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3385 struct net_device *dev = skb->dev; 3386 3387 partial_features |= dev->features & dev->gso_partial_features; 3388 if (!skb_gso_ok(skb, features | partial_features)) 3389 features &= ~NETIF_F_GSO_PARTIAL; 3390 } 3391 3392 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3393 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3394 3395 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3396 SKB_GSO_CB(skb)->encap_level = 0; 3397 3398 skb_reset_mac_header(skb); 3399 skb_reset_mac_len(skb); 3400 3401 segs = skb_mac_gso_segment(skb, features); 3402 3403 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3404 skb_warn_bad_offload(skb); 3405 3406 return segs; 3407 } 3408 EXPORT_SYMBOL(__skb_gso_segment); 3409 3410 /* Take action when hardware reception checksum errors are detected. */ 3411 #ifdef CONFIG_BUG 3412 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3413 { 3414 netdev_err(dev, "hw csum failure\n"); 3415 skb_dump(KERN_ERR, skb, true); 3416 dump_stack(); 3417 } 3418 3419 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3420 { 3421 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3422 } 3423 EXPORT_SYMBOL(netdev_rx_csum_fault); 3424 #endif 3425 3426 /* XXX: check that highmem exists at all on the given machine. */ 3427 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3428 { 3429 #ifdef CONFIG_HIGHMEM 3430 int i; 3431 3432 if (!(dev->features & NETIF_F_HIGHDMA)) { 3433 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3434 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3435 3436 if (PageHighMem(skb_frag_page(frag))) 3437 return 1; 3438 } 3439 } 3440 #endif 3441 return 0; 3442 } 3443 3444 /* If MPLS offload request, verify we are testing hardware MPLS features 3445 * instead of standard features for the netdev. 3446 */ 3447 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3448 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3449 netdev_features_t features, 3450 __be16 type) 3451 { 3452 if (eth_p_mpls(type)) 3453 features &= skb->dev->mpls_features; 3454 3455 return features; 3456 } 3457 #else 3458 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3459 netdev_features_t features, 3460 __be16 type) 3461 { 3462 return features; 3463 } 3464 #endif 3465 3466 static netdev_features_t harmonize_features(struct sk_buff *skb, 3467 netdev_features_t features) 3468 { 3469 __be16 type; 3470 3471 type = skb_network_protocol(skb, NULL); 3472 features = net_mpls_features(skb, features, type); 3473 3474 if (skb->ip_summed != CHECKSUM_NONE && 3475 !can_checksum_protocol(features, type)) { 3476 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3477 } 3478 if (illegal_highdma(skb->dev, skb)) 3479 features &= ~NETIF_F_SG; 3480 3481 return features; 3482 } 3483 3484 netdev_features_t passthru_features_check(struct sk_buff *skb, 3485 struct net_device *dev, 3486 netdev_features_t features) 3487 { 3488 return features; 3489 } 3490 EXPORT_SYMBOL(passthru_features_check); 3491 3492 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3493 struct net_device *dev, 3494 netdev_features_t features) 3495 { 3496 return vlan_features_check(skb, features); 3497 } 3498 3499 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3500 struct net_device *dev, 3501 netdev_features_t features) 3502 { 3503 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3504 3505 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3506 return features & ~NETIF_F_GSO_MASK; 3507 3508 if (!skb_shinfo(skb)->gso_type) { 3509 skb_warn_bad_offload(skb); 3510 return features & ~NETIF_F_GSO_MASK; 3511 } 3512 3513 /* Support for GSO partial features requires software 3514 * intervention before we can actually process the packets 3515 * so we need to strip support for any partial features now 3516 * and we can pull them back in after we have partially 3517 * segmented the frame. 3518 */ 3519 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3520 features &= ~dev->gso_partial_features; 3521 3522 /* Make sure to clear the IPv4 ID mangling feature if the 3523 * IPv4 header has the potential to be fragmented. 3524 */ 3525 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3526 struct iphdr *iph = skb->encapsulation ? 3527 inner_ip_hdr(skb) : ip_hdr(skb); 3528 3529 if (!(iph->frag_off & htons(IP_DF))) 3530 features &= ~NETIF_F_TSO_MANGLEID; 3531 } 3532 3533 return features; 3534 } 3535 3536 netdev_features_t netif_skb_features(struct sk_buff *skb) 3537 { 3538 struct net_device *dev = skb->dev; 3539 netdev_features_t features = dev->features; 3540 3541 if (skb_is_gso(skb)) 3542 features = gso_features_check(skb, dev, features); 3543 3544 /* If encapsulation offload request, verify we are testing 3545 * hardware encapsulation features instead of standard 3546 * features for the netdev 3547 */ 3548 if (skb->encapsulation) 3549 features &= dev->hw_enc_features; 3550 3551 if (skb_vlan_tagged(skb)) 3552 features = netdev_intersect_features(features, 3553 dev->vlan_features | 3554 NETIF_F_HW_VLAN_CTAG_TX | 3555 NETIF_F_HW_VLAN_STAG_TX); 3556 3557 if (dev->netdev_ops->ndo_features_check) 3558 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3559 features); 3560 else 3561 features &= dflt_features_check(skb, dev, features); 3562 3563 return harmonize_features(skb, features); 3564 } 3565 EXPORT_SYMBOL(netif_skb_features); 3566 3567 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3568 struct netdev_queue *txq, bool more) 3569 { 3570 unsigned int len; 3571 int rc; 3572 3573 if (dev_nit_active(dev)) 3574 dev_queue_xmit_nit(skb, dev); 3575 3576 len = skb->len; 3577 trace_net_dev_start_xmit(skb, dev); 3578 rc = netdev_start_xmit(skb, dev, txq, more); 3579 trace_net_dev_xmit(skb, rc, dev, len); 3580 3581 return rc; 3582 } 3583 3584 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3585 struct netdev_queue *txq, int *ret) 3586 { 3587 struct sk_buff *skb = first; 3588 int rc = NETDEV_TX_OK; 3589 3590 while (skb) { 3591 struct sk_buff *next = skb->next; 3592 3593 skb_mark_not_on_list(skb); 3594 rc = xmit_one(skb, dev, txq, next != NULL); 3595 if (unlikely(!dev_xmit_complete(rc))) { 3596 skb->next = next; 3597 goto out; 3598 } 3599 3600 skb = next; 3601 if (netif_tx_queue_stopped(txq) && skb) { 3602 rc = NETDEV_TX_BUSY; 3603 break; 3604 } 3605 } 3606 3607 out: 3608 *ret = rc; 3609 return skb; 3610 } 3611 3612 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3613 netdev_features_t features) 3614 { 3615 if (skb_vlan_tag_present(skb) && 3616 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3617 skb = __vlan_hwaccel_push_inside(skb); 3618 return skb; 3619 } 3620 3621 int skb_csum_hwoffload_help(struct sk_buff *skb, 3622 const netdev_features_t features) 3623 { 3624 if (unlikely(skb_csum_is_sctp(skb))) 3625 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3626 skb_crc32c_csum_help(skb); 3627 3628 if (features & NETIF_F_HW_CSUM) 3629 return 0; 3630 3631 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3632 switch (skb->csum_offset) { 3633 case offsetof(struct tcphdr, check): 3634 case offsetof(struct udphdr, check): 3635 return 0; 3636 } 3637 } 3638 3639 return skb_checksum_help(skb); 3640 } 3641 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3642 3643 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3644 { 3645 netdev_features_t features; 3646 3647 features = netif_skb_features(skb); 3648 skb = validate_xmit_vlan(skb, features); 3649 if (unlikely(!skb)) 3650 goto out_null; 3651 3652 skb = sk_validate_xmit_skb(skb, dev); 3653 if (unlikely(!skb)) 3654 goto out_null; 3655 3656 if (netif_needs_gso(skb, features)) { 3657 struct sk_buff *segs; 3658 3659 segs = skb_gso_segment(skb, features); 3660 if (IS_ERR(segs)) { 3661 goto out_kfree_skb; 3662 } else if (segs) { 3663 consume_skb(skb); 3664 skb = segs; 3665 } 3666 } else { 3667 if (skb_needs_linearize(skb, features) && 3668 __skb_linearize(skb)) 3669 goto out_kfree_skb; 3670 3671 /* If packet is not checksummed and device does not 3672 * support checksumming for this protocol, complete 3673 * checksumming here. 3674 */ 3675 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3676 if (skb->encapsulation) 3677 skb_set_inner_transport_header(skb, 3678 skb_checksum_start_offset(skb)); 3679 else 3680 skb_set_transport_header(skb, 3681 skb_checksum_start_offset(skb)); 3682 if (skb_csum_hwoffload_help(skb, features)) 3683 goto out_kfree_skb; 3684 } 3685 } 3686 3687 skb = validate_xmit_xfrm(skb, features, again); 3688 3689 return skb; 3690 3691 out_kfree_skb: 3692 kfree_skb(skb); 3693 out_null: 3694 dev_core_stats_tx_dropped_inc(dev); 3695 return NULL; 3696 } 3697 3698 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3699 { 3700 struct sk_buff *next, *head = NULL, *tail; 3701 3702 for (; skb != NULL; skb = next) { 3703 next = skb->next; 3704 skb_mark_not_on_list(skb); 3705 3706 /* in case skb wont be segmented, point to itself */ 3707 skb->prev = skb; 3708 3709 skb = validate_xmit_skb(skb, dev, again); 3710 if (!skb) 3711 continue; 3712 3713 if (!head) 3714 head = skb; 3715 else 3716 tail->next = skb; 3717 /* If skb was segmented, skb->prev points to 3718 * the last segment. If not, it still contains skb. 3719 */ 3720 tail = skb->prev; 3721 } 3722 return head; 3723 } 3724 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3725 3726 static void qdisc_pkt_len_init(struct sk_buff *skb) 3727 { 3728 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3729 3730 qdisc_skb_cb(skb)->pkt_len = skb->len; 3731 3732 /* To get more precise estimation of bytes sent on wire, 3733 * we add to pkt_len the headers size of all segments 3734 */ 3735 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3736 unsigned int hdr_len; 3737 u16 gso_segs = shinfo->gso_segs; 3738 3739 /* mac layer + network layer */ 3740 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3741 3742 /* + transport layer */ 3743 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3744 const struct tcphdr *th; 3745 struct tcphdr _tcphdr; 3746 3747 th = skb_header_pointer(skb, skb_transport_offset(skb), 3748 sizeof(_tcphdr), &_tcphdr); 3749 if (likely(th)) 3750 hdr_len += __tcp_hdrlen(th); 3751 } else { 3752 struct udphdr _udphdr; 3753 3754 if (skb_header_pointer(skb, skb_transport_offset(skb), 3755 sizeof(_udphdr), &_udphdr)) 3756 hdr_len += sizeof(struct udphdr); 3757 } 3758 3759 if (shinfo->gso_type & SKB_GSO_DODGY) 3760 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3761 shinfo->gso_size); 3762 3763 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3764 } 3765 } 3766 3767 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3768 struct sk_buff **to_free, 3769 struct netdev_queue *txq) 3770 { 3771 int rc; 3772 3773 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3774 if (rc == NET_XMIT_SUCCESS) 3775 trace_qdisc_enqueue(q, txq, skb); 3776 return rc; 3777 } 3778 3779 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3780 struct net_device *dev, 3781 struct netdev_queue *txq) 3782 { 3783 spinlock_t *root_lock = qdisc_lock(q); 3784 struct sk_buff *to_free = NULL; 3785 bool contended; 3786 int rc; 3787 3788 qdisc_calculate_pkt_len(skb, q); 3789 3790 if (q->flags & TCQ_F_NOLOCK) { 3791 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3792 qdisc_run_begin(q)) { 3793 /* Retest nolock_qdisc_is_empty() within the protection 3794 * of q->seqlock to protect from racing with requeuing. 3795 */ 3796 if (unlikely(!nolock_qdisc_is_empty(q))) { 3797 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3798 __qdisc_run(q); 3799 qdisc_run_end(q); 3800 3801 goto no_lock_out; 3802 } 3803 3804 qdisc_bstats_cpu_update(q, skb); 3805 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3806 !nolock_qdisc_is_empty(q)) 3807 __qdisc_run(q); 3808 3809 qdisc_run_end(q); 3810 return NET_XMIT_SUCCESS; 3811 } 3812 3813 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3814 qdisc_run(q); 3815 3816 no_lock_out: 3817 if (unlikely(to_free)) 3818 kfree_skb_list_reason(to_free, 3819 SKB_DROP_REASON_QDISC_DROP); 3820 return rc; 3821 } 3822 3823 /* 3824 * Heuristic to force contended enqueues to serialize on a 3825 * separate lock before trying to get qdisc main lock. 3826 * This permits qdisc->running owner to get the lock more 3827 * often and dequeue packets faster. 3828 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3829 * and then other tasks will only enqueue packets. The packets will be 3830 * sent after the qdisc owner is scheduled again. To prevent this 3831 * scenario the task always serialize on the lock. 3832 */ 3833 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3834 if (unlikely(contended)) 3835 spin_lock(&q->busylock); 3836 3837 spin_lock(root_lock); 3838 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3839 __qdisc_drop(skb, &to_free); 3840 rc = NET_XMIT_DROP; 3841 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3842 qdisc_run_begin(q)) { 3843 /* 3844 * This is a work-conserving queue; there are no old skbs 3845 * waiting to be sent out; and the qdisc is not running - 3846 * xmit the skb directly. 3847 */ 3848 3849 qdisc_bstats_update(q, skb); 3850 3851 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3852 if (unlikely(contended)) { 3853 spin_unlock(&q->busylock); 3854 contended = false; 3855 } 3856 __qdisc_run(q); 3857 } 3858 3859 qdisc_run_end(q); 3860 rc = NET_XMIT_SUCCESS; 3861 } else { 3862 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3863 if (qdisc_run_begin(q)) { 3864 if (unlikely(contended)) { 3865 spin_unlock(&q->busylock); 3866 contended = false; 3867 } 3868 __qdisc_run(q); 3869 qdisc_run_end(q); 3870 } 3871 } 3872 spin_unlock(root_lock); 3873 if (unlikely(to_free)) 3874 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3875 if (unlikely(contended)) 3876 spin_unlock(&q->busylock); 3877 return rc; 3878 } 3879 3880 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3881 static void skb_update_prio(struct sk_buff *skb) 3882 { 3883 const struct netprio_map *map; 3884 const struct sock *sk; 3885 unsigned int prioidx; 3886 3887 if (skb->priority) 3888 return; 3889 map = rcu_dereference_bh(skb->dev->priomap); 3890 if (!map) 3891 return; 3892 sk = skb_to_full_sk(skb); 3893 if (!sk) 3894 return; 3895 3896 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3897 3898 if (prioidx < map->priomap_len) 3899 skb->priority = map->priomap[prioidx]; 3900 } 3901 #else 3902 #define skb_update_prio(skb) 3903 #endif 3904 3905 /** 3906 * dev_loopback_xmit - loop back @skb 3907 * @net: network namespace this loopback is happening in 3908 * @sk: sk needed to be a netfilter okfn 3909 * @skb: buffer to transmit 3910 */ 3911 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3912 { 3913 skb_reset_mac_header(skb); 3914 __skb_pull(skb, skb_network_offset(skb)); 3915 skb->pkt_type = PACKET_LOOPBACK; 3916 if (skb->ip_summed == CHECKSUM_NONE) 3917 skb->ip_summed = CHECKSUM_UNNECESSARY; 3918 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3919 skb_dst_force(skb); 3920 netif_rx(skb); 3921 return 0; 3922 } 3923 EXPORT_SYMBOL(dev_loopback_xmit); 3924 3925 #ifdef CONFIG_NET_EGRESS 3926 static struct sk_buff * 3927 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3928 { 3929 #ifdef CONFIG_NET_CLS_ACT 3930 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3931 struct tcf_result cl_res; 3932 3933 if (!miniq) 3934 return skb; 3935 3936 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3937 tc_skb_cb(skb)->mru = 0; 3938 tc_skb_cb(skb)->post_ct = false; 3939 mini_qdisc_bstats_cpu_update(miniq, skb); 3940 3941 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3942 case TC_ACT_OK: 3943 case TC_ACT_RECLASSIFY: 3944 skb->tc_index = TC_H_MIN(cl_res.classid); 3945 break; 3946 case TC_ACT_SHOT: 3947 mini_qdisc_qstats_cpu_drop(miniq); 3948 *ret = NET_XMIT_DROP; 3949 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3950 return NULL; 3951 case TC_ACT_STOLEN: 3952 case TC_ACT_QUEUED: 3953 case TC_ACT_TRAP: 3954 *ret = NET_XMIT_SUCCESS; 3955 consume_skb(skb); 3956 return NULL; 3957 case TC_ACT_REDIRECT: 3958 /* No need to push/pop skb's mac_header here on egress! */ 3959 skb_do_redirect(skb); 3960 *ret = NET_XMIT_SUCCESS; 3961 return NULL; 3962 default: 3963 break; 3964 } 3965 #endif /* CONFIG_NET_CLS_ACT */ 3966 3967 return skb; 3968 } 3969 3970 static struct netdev_queue * 3971 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3972 { 3973 int qm = skb_get_queue_mapping(skb); 3974 3975 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3976 } 3977 3978 static bool netdev_xmit_txqueue_skipped(void) 3979 { 3980 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3981 } 3982 3983 void netdev_xmit_skip_txqueue(bool skip) 3984 { 3985 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3986 } 3987 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3988 #endif /* CONFIG_NET_EGRESS */ 3989 3990 #ifdef CONFIG_XPS 3991 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3992 struct xps_dev_maps *dev_maps, unsigned int tci) 3993 { 3994 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3995 struct xps_map *map; 3996 int queue_index = -1; 3997 3998 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3999 return queue_index; 4000 4001 tci *= dev_maps->num_tc; 4002 tci += tc; 4003 4004 map = rcu_dereference(dev_maps->attr_map[tci]); 4005 if (map) { 4006 if (map->len == 1) 4007 queue_index = map->queues[0]; 4008 else 4009 queue_index = map->queues[reciprocal_scale( 4010 skb_get_hash(skb), map->len)]; 4011 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4012 queue_index = -1; 4013 } 4014 return queue_index; 4015 } 4016 #endif 4017 4018 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4019 struct sk_buff *skb) 4020 { 4021 #ifdef CONFIG_XPS 4022 struct xps_dev_maps *dev_maps; 4023 struct sock *sk = skb->sk; 4024 int queue_index = -1; 4025 4026 if (!static_key_false(&xps_needed)) 4027 return -1; 4028 4029 rcu_read_lock(); 4030 if (!static_key_false(&xps_rxqs_needed)) 4031 goto get_cpus_map; 4032 4033 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4034 if (dev_maps) { 4035 int tci = sk_rx_queue_get(sk); 4036 4037 if (tci >= 0) 4038 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4039 tci); 4040 } 4041 4042 get_cpus_map: 4043 if (queue_index < 0) { 4044 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4045 if (dev_maps) { 4046 unsigned int tci = skb->sender_cpu - 1; 4047 4048 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4049 tci); 4050 } 4051 } 4052 rcu_read_unlock(); 4053 4054 return queue_index; 4055 #else 4056 return -1; 4057 #endif 4058 } 4059 4060 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4061 struct net_device *sb_dev) 4062 { 4063 return 0; 4064 } 4065 EXPORT_SYMBOL(dev_pick_tx_zero); 4066 4067 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4068 struct net_device *sb_dev) 4069 { 4070 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4071 } 4072 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4073 4074 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4075 struct net_device *sb_dev) 4076 { 4077 struct sock *sk = skb->sk; 4078 int queue_index = sk_tx_queue_get(sk); 4079 4080 sb_dev = sb_dev ? : dev; 4081 4082 if (queue_index < 0 || skb->ooo_okay || 4083 queue_index >= dev->real_num_tx_queues) { 4084 int new_index = get_xps_queue(dev, sb_dev, skb); 4085 4086 if (new_index < 0) 4087 new_index = skb_tx_hash(dev, sb_dev, skb); 4088 4089 if (queue_index != new_index && sk && 4090 sk_fullsock(sk) && 4091 rcu_access_pointer(sk->sk_dst_cache)) 4092 sk_tx_queue_set(sk, new_index); 4093 4094 queue_index = new_index; 4095 } 4096 4097 return queue_index; 4098 } 4099 EXPORT_SYMBOL(netdev_pick_tx); 4100 4101 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4102 struct sk_buff *skb, 4103 struct net_device *sb_dev) 4104 { 4105 int queue_index = 0; 4106 4107 #ifdef CONFIG_XPS 4108 u32 sender_cpu = skb->sender_cpu - 1; 4109 4110 if (sender_cpu >= (u32)NR_CPUS) 4111 skb->sender_cpu = raw_smp_processor_id() + 1; 4112 #endif 4113 4114 if (dev->real_num_tx_queues != 1) { 4115 const struct net_device_ops *ops = dev->netdev_ops; 4116 4117 if (ops->ndo_select_queue) 4118 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4119 else 4120 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4121 4122 queue_index = netdev_cap_txqueue(dev, queue_index); 4123 } 4124 4125 skb_set_queue_mapping(skb, queue_index); 4126 return netdev_get_tx_queue(dev, queue_index); 4127 } 4128 4129 /** 4130 * __dev_queue_xmit() - transmit a buffer 4131 * @skb: buffer to transmit 4132 * @sb_dev: suboordinate device used for L2 forwarding offload 4133 * 4134 * Queue a buffer for transmission to a network device. The caller must 4135 * have set the device and priority and built the buffer before calling 4136 * this function. The function can be called from an interrupt. 4137 * 4138 * When calling this method, interrupts MUST be enabled. This is because 4139 * the BH enable code must have IRQs enabled so that it will not deadlock. 4140 * 4141 * Regardless of the return value, the skb is consumed, so it is currently 4142 * difficult to retry a send to this method. (You can bump the ref count 4143 * before sending to hold a reference for retry if you are careful.) 4144 * 4145 * Return: 4146 * * 0 - buffer successfully transmitted 4147 * * positive qdisc return code - NET_XMIT_DROP etc. 4148 * * negative errno - other errors 4149 */ 4150 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4151 { 4152 struct net_device *dev = skb->dev; 4153 struct netdev_queue *txq = NULL; 4154 struct Qdisc *q; 4155 int rc = -ENOMEM; 4156 bool again = false; 4157 4158 skb_reset_mac_header(skb); 4159 skb_assert_len(skb); 4160 4161 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4162 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4163 4164 /* Disable soft irqs for various locks below. Also 4165 * stops preemption for RCU. 4166 */ 4167 rcu_read_lock_bh(); 4168 4169 skb_update_prio(skb); 4170 4171 qdisc_pkt_len_init(skb); 4172 #ifdef CONFIG_NET_CLS_ACT 4173 skb->tc_at_ingress = 0; 4174 #endif 4175 #ifdef CONFIG_NET_EGRESS 4176 if (static_branch_unlikely(&egress_needed_key)) { 4177 if (nf_hook_egress_active()) { 4178 skb = nf_hook_egress(skb, &rc, dev); 4179 if (!skb) 4180 goto out; 4181 } 4182 4183 netdev_xmit_skip_txqueue(false); 4184 4185 nf_skip_egress(skb, true); 4186 skb = sch_handle_egress(skb, &rc, dev); 4187 if (!skb) 4188 goto out; 4189 nf_skip_egress(skb, false); 4190 4191 if (netdev_xmit_txqueue_skipped()) 4192 txq = netdev_tx_queue_mapping(dev, skb); 4193 } 4194 #endif 4195 /* If device/qdisc don't need skb->dst, release it right now while 4196 * its hot in this cpu cache. 4197 */ 4198 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4199 skb_dst_drop(skb); 4200 else 4201 skb_dst_force(skb); 4202 4203 if (!txq) 4204 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4205 4206 q = rcu_dereference_bh(txq->qdisc); 4207 4208 trace_net_dev_queue(skb); 4209 if (q->enqueue) { 4210 rc = __dev_xmit_skb(skb, q, dev, txq); 4211 goto out; 4212 } 4213 4214 /* The device has no queue. Common case for software devices: 4215 * loopback, all the sorts of tunnels... 4216 4217 * Really, it is unlikely that netif_tx_lock protection is necessary 4218 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4219 * counters.) 4220 * However, it is possible, that they rely on protection 4221 * made by us here. 4222 4223 * Check this and shot the lock. It is not prone from deadlocks. 4224 *Either shot noqueue qdisc, it is even simpler 8) 4225 */ 4226 if (dev->flags & IFF_UP) { 4227 int cpu = smp_processor_id(); /* ok because BHs are off */ 4228 4229 /* Other cpus might concurrently change txq->xmit_lock_owner 4230 * to -1 or to their cpu id, but not to our id. 4231 */ 4232 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4233 if (dev_xmit_recursion()) 4234 goto recursion_alert; 4235 4236 skb = validate_xmit_skb(skb, dev, &again); 4237 if (!skb) 4238 goto out; 4239 4240 HARD_TX_LOCK(dev, txq, cpu); 4241 4242 if (!netif_xmit_stopped(txq)) { 4243 dev_xmit_recursion_inc(); 4244 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4245 dev_xmit_recursion_dec(); 4246 if (dev_xmit_complete(rc)) { 4247 HARD_TX_UNLOCK(dev, txq); 4248 goto out; 4249 } 4250 } 4251 HARD_TX_UNLOCK(dev, txq); 4252 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4253 dev->name); 4254 } else { 4255 /* Recursion is detected! It is possible, 4256 * unfortunately 4257 */ 4258 recursion_alert: 4259 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4260 dev->name); 4261 } 4262 } 4263 4264 rc = -ENETDOWN; 4265 rcu_read_unlock_bh(); 4266 4267 dev_core_stats_tx_dropped_inc(dev); 4268 kfree_skb_list(skb); 4269 return rc; 4270 out: 4271 rcu_read_unlock_bh(); 4272 return rc; 4273 } 4274 EXPORT_SYMBOL(__dev_queue_xmit); 4275 4276 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4277 { 4278 struct net_device *dev = skb->dev; 4279 struct sk_buff *orig_skb = skb; 4280 struct netdev_queue *txq; 4281 int ret = NETDEV_TX_BUSY; 4282 bool again = false; 4283 4284 if (unlikely(!netif_running(dev) || 4285 !netif_carrier_ok(dev))) 4286 goto drop; 4287 4288 skb = validate_xmit_skb_list(skb, dev, &again); 4289 if (skb != orig_skb) 4290 goto drop; 4291 4292 skb_set_queue_mapping(skb, queue_id); 4293 txq = skb_get_tx_queue(dev, skb); 4294 4295 local_bh_disable(); 4296 4297 dev_xmit_recursion_inc(); 4298 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4299 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4300 ret = netdev_start_xmit(skb, dev, txq, false); 4301 HARD_TX_UNLOCK(dev, txq); 4302 dev_xmit_recursion_dec(); 4303 4304 local_bh_enable(); 4305 return ret; 4306 drop: 4307 dev_core_stats_tx_dropped_inc(dev); 4308 kfree_skb_list(skb); 4309 return NET_XMIT_DROP; 4310 } 4311 EXPORT_SYMBOL(__dev_direct_xmit); 4312 4313 /************************************************************************* 4314 * Receiver routines 4315 *************************************************************************/ 4316 4317 int netdev_max_backlog __read_mostly = 1000; 4318 EXPORT_SYMBOL(netdev_max_backlog); 4319 4320 int netdev_tstamp_prequeue __read_mostly = 1; 4321 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4322 int netdev_budget __read_mostly = 300; 4323 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4324 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4325 int weight_p __read_mostly = 64; /* old backlog weight */ 4326 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4327 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4328 int dev_rx_weight __read_mostly = 64; 4329 int dev_tx_weight __read_mostly = 64; 4330 4331 /* Called with irq disabled */ 4332 static inline void ____napi_schedule(struct softnet_data *sd, 4333 struct napi_struct *napi) 4334 { 4335 struct task_struct *thread; 4336 4337 lockdep_assert_irqs_disabled(); 4338 4339 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4340 /* Paired with smp_mb__before_atomic() in 4341 * napi_enable()/dev_set_threaded(). 4342 * Use READ_ONCE() to guarantee a complete 4343 * read on napi->thread. Only call 4344 * wake_up_process() when it's not NULL. 4345 */ 4346 thread = READ_ONCE(napi->thread); 4347 if (thread) { 4348 /* Avoid doing set_bit() if the thread is in 4349 * INTERRUPTIBLE state, cause napi_thread_wait() 4350 * makes sure to proceed with napi polling 4351 * if the thread is explicitly woken from here. 4352 */ 4353 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4354 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4355 wake_up_process(thread); 4356 return; 4357 } 4358 } 4359 4360 list_add_tail(&napi->poll_list, &sd->poll_list); 4361 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4362 } 4363 4364 #ifdef CONFIG_RPS 4365 4366 /* One global table that all flow-based protocols share. */ 4367 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4368 EXPORT_SYMBOL(rps_sock_flow_table); 4369 u32 rps_cpu_mask __read_mostly; 4370 EXPORT_SYMBOL(rps_cpu_mask); 4371 4372 struct static_key_false rps_needed __read_mostly; 4373 EXPORT_SYMBOL(rps_needed); 4374 struct static_key_false rfs_needed __read_mostly; 4375 EXPORT_SYMBOL(rfs_needed); 4376 4377 static struct rps_dev_flow * 4378 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4379 struct rps_dev_flow *rflow, u16 next_cpu) 4380 { 4381 if (next_cpu < nr_cpu_ids) { 4382 #ifdef CONFIG_RFS_ACCEL 4383 struct netdev_rx_queue *rxqueue; 4384 struct rps_dev_flow_table *flow_table; 4385 struct rps_dev_flow *old_rflow; 4386 u32 flow_id; 4387 u16 rxq_index; 4388 int rc; 4389 4390 /* Should we steer this flow to a different hardware queue? */ 4391 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4392 !(dev->features & NETIF_F_NTUPLE)) 4393 goto out; 4394 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4395 if (rxq_index == skb_get_rx_queue(skb)) 4396 goto out; 4397 4398 rxqueue = dev->_rx + rxq_index; 4399 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4400 if (!flow_table) 4401 goto out; 4402 flow_id = skb_get_hash(skb) & flow_table->mask; 4403 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4404 rxq_index, flow_id); 4405 if (rc < 0) 4406 goto out; 4407 old_rflow = rflow; 4408 rflow = &flow_table->flows[flow_id]; 4409 rflow->filter = rc; 4410 if (old_rflow->filter == rflow->filter) 4411 old_rflow->filter = RPS_NO_FILTER; 4412 out: 4413 #endif 4414 rflow->last_qtail = 4415 per_cpu(softnet_data, next_cpu).input_queue_head; 4416 } 4417 4418 rflow->cpu = next_cpu; 4419 return rflow; 4420 } 4421 4422 /* 4423 * get_rps_cpu is called from netif_receive_skb and returns the target 4424 * CPU from the RPS map of the receiving queue for a given skb. 4425 * rcu_read_lock must be held on entry. 4426 */ 4427 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4428 struct rps_dev_flow **rflowp) 4429 { 4430 const struct rps_sock_flow_table *sock_flow_table; 4431 struct netdev_rx_queue *rxqueue = dev->_rx; 4432 struct rps_dev_flow_table *flow_table; 4433 struct rps_map *map; 4434 int cpu = -1; 4435 u32 tcpu; 4436 u32 hash; 4437 4438 if (skb_rx_queue_recorded(skb)) { 4439 u16 index = skb_get_rx_queue(skb); 4440 4441 if (unlikely(index >= dev->real_num_rx_queues)) { 4442 WARN_ONCE(dev->real_num_rx_queues > 1, 4443 "%s received packet on queue %u, but number " 4444 "of RX queues is %u\n", 4445 dev->name, index, dev->real_num_rx_queues); 4446 goto done; 4447 } 4448 rxqueue += index; 4449 } 4450 4451 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4452 4453 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4454 map = rcu_dereference(rxqueue->rps_map); 4455 if (!flow_table && !map) 4456 goto done; 4457 4458 skb_reset_network_header(skb); 4459 hash = skb_get_hash(skb); 4460 if (!hash) 4461 goto done; 4462 4463 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4464 if (flow_table && sock_flow_table) { 4465 struct rps_dev_flow *rflow; 4466 u32 next_cpu; 4467 u32 ident; 4468 4469 /* First check into global flow table if there is a match */ 4470 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4471 if ((ident ^ hash) & ~rps_cpu_mask) 4472 goto try_rps; 4473 4474 next_cpu = ident & rps_cpu_mask; 4475 4476 /* OK, now we know there is a match, 4477 * we can look at the local (per receive queue) flow table 4478 */ 4479 rflow = &flow_table->flows[hash & flow_table->mask]; 4480 tcpu = rflow->cpu; 4481 4482 /* 4483 * If the desired CPU (where last recvmsg was done) is 4484 * different from current CPU (one in the rx-queue flow 4485 * table entry), switch if one of the following holds: 4486 * - Current CPU is unset (>= nr_cpu_ids). 4487 * - Current CPU is offline. 4488 * - The current CPU's queue tail has advanced beyond the 4489 * last packet that was enqueued using this table entry. 4490 * This guarantees that all previous packets for the flow 4491 * have been dequeued, thus preserving in order delivery. 4492 */ 4493 if (unlikely(tcpu != next_cpu) && 4494 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4495 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4496 rflow->last_qtail)) >= 0)) { 4497 tcpu = next_cpu; 4498 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4499 } 4500 4501 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4502 *rflowp = rflow; 4503 cpu = tcpu; 4504 goto done; 4505 } 4506 } 4507 4508 try_rps: 4509 4510 if (map) { 4511 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4512 if (cpu_online(tcpu)) { 4513 cpu = tcpu; 4514 goto done; 4515 } 4516 } 4517 4518 done: 4519 return cpu; 4520 } 4521 4522 #ifdef CONFIG_RFS_ACCEL 4523 4524 /** 4525 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4526 * @dev: Device on which the filter was set 4527 * @rxq_index: RX queue index 4528 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4529 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4530 * 4531 * Drivers that implement ndo_rx_flow_steer() should periodically call 4532 * this function for each installed filter and remove the filters for 4533 * which it returns %true. 4534 */ 4535 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4536 u32 flow_id, u16 filter_id) 4537 { 4538 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4539 struct rps_dev_flow_table *flow_table; 4540 struct rps_dev_flow *rflow; 4541 bool expire = true; 4542 unsigned int cpu; 4543 4544 rcu_read_lock(); 4545 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4546 if (flow_table && flow_id <= flow_table->mask) { 4547 rflow = &flow_table->flows[flow_id]; 4548 cpu = READ_ONCE(rflow->cpu); 4549 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4550 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4551 rflow->last_qtail) < 4552 (int)(10 * flow_table->mask))) 4553 expire = false; 4554 } 4555 rcu_read_unlock(); 4556 return expire; 4557 } 4558 EXPORT_SYMBOL(rps_may_expire_flow); 4559 4560 #endif /* CONFIG_RFS_ACCEL */ 4561 4562 /* Called from hardirq (IPI) context */ 4563 static void rps_trigger_softirq(void *data) 4564 { 4565 struct softnet_data *sd = data; 4566 4567 ____napi_schedule(sd, &sd->backlog); 4568 sd->received_rps++; 4569 } 4570 4571 #endif /* CONFIG_RPS */ 4572 4573 /* Called from hardirq (IPI) context */ 4574 static void trigger_rx_softirq(void *data) 4575 { 4576 struct softnet_data *sd = data; 4577 4578 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4579 smp_store_release(&sd->defer_ipi_scheduled, 0); 4580 } 4581 4582 /* 4583 * Check if this softnet_data structure is another cpu one 4584 * If yes, queue it to our IPI list and return 1 4585 * If no, return 0 4586 */ 4587 static int napi_schedule_rps(struct softnet_data *sd) 4588 { 4589 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4590 4591 #ifdef CONFIG_RPS 4592 if (sd != mysd) { 4593 sd->rps_ipi_next = mysd->rps_ipi_list; 4594 mysd->rps_ipi_list = sd; 4595 4596 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4597 return 1; 4598 } 4599 #endif /* CONFIG_RPS */ 4600 __napi_schedule_irqoff(&mysd->backlog); 4601 return 0; 4602 } 4603 4604 #ifdef CONFIG_NET_FLOW_LIMIT 4605 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4606 #endif 4607 4608 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4609 { 4610 #ifdef CONFIG_NET_FLOW_LIMIT 4611 struct sd_flow_limit *fl; 4612 struct softnet_data *sd; 4613 unsigned int old_flow, new_flow; 4614 4615 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4616 return false; 4617 4618 sd = this_cpu_ptr(&softnet_data); 4619 4620 rcu_read_lock(); 4621 fl = rcu_dereference(sd->flow_limit); 4622 if (fl) { 4623 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4624 old_flow = fl->history[fl->history_head]; 4625 fl->history[fl->history_head] = new_flow; 4626 4627 fl->history_head++; 4628 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4629 4630 if (likely(fl->buckets[old_flow])) 4631 fl->buckets[old_flow]--; 4632 4633 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4634 fl->count++; 4635 rcu_read_unlock(); 4636 return true; 4637 } 4638 } 4639 rcu_read_unlock(); 4640 #endif 4641 return false; 4642 } 4643 4644 /* 4645 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4646 * queue (may be a remote CPU queue). 4647 */ 4648 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4649 unsigned int *qtail) 4650 { 4651 enum skb_drop_reason reason; 4652 struct softnet_data *sd; 4653 unsigned long flags; 4654 unsigned int qlen; 4655 4656 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4657 sd = &per_cpu(softnet_data, cpu); 4658 4659 rps_lock_irqsave(sd, &flags); 4660 if (!netif_running(skb->dev)) 4661 goto drop; 4662 qlen = skb_queue_len(&sd->input_pkt_queue); 4663 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4664 if (qlen) { 4665 enqueue: 4666 __skb_queue_tail(&sd->input_pkt_queue, skb); 4667 input_queue_tail_incr_save(sd, qtail); 4668 rps_unlock_irq_restore(sd, &flags); 4669 return NET_RX_SUCCESS; 4670 } 4671 4672 /* Schedule NAPI for backlog device 4673 * We can use non atomic operation since we own the queue lock 4674 */ 4675 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4676 napi_schedule_rps(sd); 4677 goto enqueue; 4678 } 4679 reason = SKB_DROP_REASON_CPU_BACKLOG; 4680 4681 drop: 4682 sd->dropped++; 4683 rps_unlock_irq_restore(sd, &flags); 4684 4685 dev_core_stats_rx_dropped_inc(skb->dev); 4686 kfree_skb_reason(skb, reason); 4687 return NET_RX_DROP; 4688 } 4689 4690 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4691 { 4692 struct net_device *dev = skb->dev; 4693 struct netdev_rx_queue *rxqueue; 4694 4695 rxqueue = dev->_rx; 4696 4697 if (skb_rx_queue_recorded(skb)) { 4698 u16 index = skb_get_rx_queue(skb); 4699 4700 if (unlikely(index >= dev->real_num_rx_queues)) { 4701 WARN_ONCE(dev->real_num_rx_queues > 1, 4702 "%s received packet on queue %u, but number " 4703 "of RX queues is %u\n", 4704 dev->name, index, dev->real_num_rx_queues); 4705 4706 return rxqueue; /* Return first rxqueue */ 4707 } 4708 rxqueue += index; 4709 } 4710 return rxqueue; 4711 } 4712 4713 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4714 struct bpf_prog *xdp_prog) 4715 { 4716 void *orig_data, *orig_data_end, *hard_start; 4717 struct netdev_rx_queue *rxqueue; 4718 bool orig_bcast, orig_host; 4719 u32 mac_len, frame_sz; 4720 __be16 orig_eth_type; 4721 struct ethhdr *eth; 4722 u32 metalen, act; 4723 int off; 4724 4725 /* The XDP program wants to see the packet starting at the MAC 4726 * header. 4727 */ 4728 mac_len = skb->data - skb_mac_header(skb); 4729 hard_start = skb->data - skb_headroom(skb); 4730 4731 /* SKB "head" area always have tailroom for skb_shared_info */ 4732 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4733 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4734 4735 rxqueue = netif_get_rxqueue(skb); 4736 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4737 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4738 skb_headlen(skb) + mac_len, true); 4739 4740 orig_data_end = xdp->data_end; 4741 orig_data = xdp->data; 4742 eth = (struct ethhdr *)xdp->data; 4743 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4744 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4745 orig_eth_type = eth->h_proto; 4746 4747 act = bpf_prog_run_xdp(xdp_prog, xdp); 4748 4749 /* check if bpf_xdp_adjust_head was used */ 4750 off = xdp->data - orig_data; 4751 if (off) { 4752 if (off > 0) 4753 __skb_pull(skb, off); 4754 else if (off < 0) 4755 __skb_push(skb, -off); 4756 4757 skb->mac_header += off; 4758 skb_reset_network_header(skb); 4759 } 4760 4761 /* check if bpf_xdp_adjust_tail was used */ 4762 off = xdp->data_end - orig_data_end; 4763 if (off != 0) { 4764 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4765 skb->len += off; /* positive on grow, negative on shrink */ 4766 } 4767 4768 /* check if XDP changed eth hdr such SKB needs update */ 4769 eth = (struct ethhdr *)xdp->data; 4770 if ((orig_eth_type != eth->h_proto) || 4771 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4772 skb->dev->dev_addr)) || 4773 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4774 __skb_push(skb, ETH_HLEN); 4775 skb->pkt_type = PACKET_HOST; 4776 skb->protocol = eth_type_trans(skb, skb->dev); 4777 } 4778 4779 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4780 * before calling us again on redirect path. We do not call do_redirect 4781 * as we leave that up to the caller. 4782 * 4783 * Caller is responsible for managing lifetime of skb (i.e. calling 4784 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4785 */ 4786 switch (act) { 4787 case XDP_REDIRECT: 4788 case XDP_TX: 4789 __skb_push(skb, mac_len); 4790 break; 4791 case XDP_PASS: 4792 metalen = xdp->data - xdp->data_meta; 4793 if (metalen) 4794 skb_metadata_set(skb, metalen); 4795 break; 4796 } 4797 4798 return act; 4799 } 4800 4801 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4802 struct xdp_buff *xdp, 4803 struct bpf_prog *xdp_prog) 4804 { 4805 u32 act = XDP_DROP; 4806 4807 /* Reinjected packets coming from act_mirred or similar should 4808 * not get XDP generic processing. 4809 */ 4810 if (skb_is_redirected(skb)) 4811 return XDP_PASS; 4812 4813 /* XDP packets must be linear and must have sufficient headroom 4814 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4815 * native XDP provides, thus we need to do it here as well. 4816 */ 4817 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4818 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4819 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4820 int troom = skb->tail + skb->data_len - skb->end; 4821 4822 /* In case we have to go down the path and also linearize, 4823 * then lets do the pskb_expand_head() work just once here. 4824 */ 4825 if (pskb_expand_head(skb, 4826 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4827 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4828 goto do_drop; 4829 if (skb_linearize(skb)) 4830 goto do_drop; 4831 } 4832 4833 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4834 switch (act) { 4835 case XDP_REDIRECT: 4836 case XDP_TX: 4837 case XDP_PASS: 4838 break; 4839 default: 4840 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4841 fallthrough; 4842 case XDP_ABORTED: 4843 trace_xdp_exception(skb->dev, xdp_prog, act); 4844 fallthrough; 4845 case XDP_DROP: 4846 do_drop: 4847 kfree_skb(skb); 4848 break; 4849 } 4850 4851 return act; 4852 } 4853 4854 /* When doing generic XDP we have to bypass the qdisc layer and the 4855 * network taps in order to match in-driver-XDP behavior. This also means 4856 * that XDP packets are able to starve other packets going through a qdisc, 4857 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4858 * queues, so they do not have this starvation issue. 4859 */ 4860 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4861 { 4862 struct net_device *dev = skb->dev; 4863 struct netdev_queue *txq; 4864 bool free_skb = true; 4865 int cpu, rc; 4866 4867 txq = netdev_core_pick_tx(dev, skb, NULL); 4868 cpu = smp_processor_id(); 4869 HARD_TX_LOCK(dev, txq, cpu); 4870 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4871 rc = netdev_start_xmit(skb, dev, txq, 0); 4872 if (dev_xmit_complete(rc)) 4873 free_skb = false; 4874 } 4875 HARD_TX_UNLOCK(dev, txq); 4876 if (free_skb) { 4877 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4878 dev_core_stats_tx_dropped_inc(dev); 4879 kfree_skb(skb); 4880 } 4881 } 4882 4883 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4884 4885 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4886 { 4887 if (xdp_prog) { 4888 struct xdp_buff xdp; 4889 u32 act; 4890 int err; 4891 4892 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4893 if (act != XDP_PASS) { 4894 switch (act) { 4895 case XDP_REDIRECT: 4896 err = xdp_do_generic_redirect(skb->dev, skb, 4897 &xdp, xdp_prog); 4898 if (err) 4899 goto out_redir; 4900 break; 4901 case XDP_TX: 4902 generic_xdp_tx(skb, xdp_prog); 4903 break; 4904 } 4905 return XDP_DROP; 4906 } 4907 } 4908 return XDP_PASS; 4909 out_redir: 4910 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4911 return XDP_DROP; 4912 } 4913 EXPORT_SYMBOL_GPL(do_xdp_generic); 4914 4915 static int netif_rx_internal(struct sk_buff *skb) 4916 { 4917 int ret; 4918 4919 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4920 4921 trace_netif_rx(skb); 4922 4923 #ifdef CONFIG_RPS 4924 if (static_branch_unlikely(&rps_needed)) { 4925 struct rps_dev_flow voidflow, *rflow = &voidflow; 4926 int cpu; 4927 4928 rcu_read_lock(); 4929 4930 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4931 if (cpu < 0) 4932 cpu = smp_processor_id(); 4933 4934 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4935 4936 rcu_read_unlock(); 4937 } else 4938 #endif 4939 { 4940 unsigned int qtail; 4941 4942 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4943 } 4944 return ret; 4945 } 4946 4947 /** 4948 * __netif_rx - Slightly optimized version of netif_rx 4949 * @skb: buffer to post 4950 * 4951 * This behaves as netif_rx except that it does not disable bottom halves. 4952 * As a result this function may only be invoked from the interrupt context 4953 * (either hard or soft interrupt). 4954 */ 4955 int __netif_rx(struct sk_buff *skb) 4956 { 4957 int ret; 4958 4959 lockdep_assert_once(hardirq_count() | softirq_count()); 4960 4961 trace_netif_rx_entry(skb); 4962 ret = netif_rx_internal(skb); 4963 trace_netif_rx_exit(ret); 4964 return ret; 4965 } 4966 EXPORT_SYMBOL(__netif_rx); 4967 4968 /** 4969 * netif_rx - post buffer to the network code 4970 * @skb: buffer to post 4971 * 4972 * This function receives a packet from a device driver and queues it for 4973 * the upper (protocol) levels to process via the backlog NAPI device. It 4974 * always succeeds. The buffer may be dropped during processing for 4975 * congestion control or by the protocol layers. 4976 * The network buffer is passed via the backlog NAPI device. Modern NIC 4977 * driver should use NAPI and GRO. 4978 * This function can used from interrupt and from process context. The 4979 * caller from process context must not disable interrupts before invoking 4980 * this function. 4981 * 4982 * return values: 4983 * NET_RX_SUCCESS (no congestion) 4984 * NET_RX_DROP (packet was dropped) 4985 * 4986 */ 4987 int netif_rx(struct sk_buff *skb) 4988 { 4989 bool need_bh_off = !(hardirq_count() | softirq_count()); 4990 int ret; 4991 4992 if (need_bh_off) 4993 local_bh_disable(); 4994 trace_netif_rx_entry(skb); 4995 ret = netif_rx_internal(skb); 4996 trace_netif_rx_exit(ret); 4997 if (need_bh_off) 4998 local_bh_enable(); 4999 return ret; 5000 } 5001 EXPORT_SYMBOL(netif_rx); 5002 5003 static __latent_entropy void net_tx_action(struct softirq_action *h) 5004 { 5005 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5006 5007 if (sd->completion_queue) { 5008 struct sk_buff *clist; 5009 5010 local_irq_disable(); 5011 clist = sd->completion_queue; 5012 sd->completion_queue = NULL; 5013 local_irq_enable(); 5014 5015 while (clist) { 5016 struct sk_buff *skb = clist; 5017 5018 clist = clist->next; 5019 5020 WARN_ON(refcount_read(&skb->users)); 5021 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5022 trace_consume_skb(skb); 5023 else 5024 trace_kfree_skb(skb, net_tx_action, 5025 SKB_DROP_REASON_NOT_SPECIFIED); 5026 5027 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5028 __kfree_skb(skb); 5029 else 5030 __kfree_skb_defer(skb); 5031 } 5032 } 5033 5034 if (sd->output_queue) { 5035 struct Qdisc *head; 5036 5037 local_irq_disable(); 5038 head = sd->output_queue; 5039 sd->output_queue = NULL; 5040 sd->output_queue_tailp = &sd->output_queue; 5041 local_irq_enable(); 5042 5043 rcu_read_lock(); 5044 5045 while (head) { 5046 struct Qdisc *q = head; 5047 spinlock_t *root_lock = NULL; 5048 5049 head = head->next_sched; 5050 5051 /* We need to make sure head->next_sched is read 5052 * before clearing __QDISC_STATE_SCHED 5053 */ 5054 smp_mb__before_atomic(); 5055 5056 if (!(q->flags & TCQ_F_NOLOCK)) { 5057 root_lock = qdisc_lock(q); 5058 spin_lock(root_lock); 5059 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5060 &q->state))) { 5061 /* There is a synchronize_net() between 5062 * STATE_DEACTIVATED flag being set and 5063 * qdisc_reset()/some_qdisc_is_busy() in 5064 * dev_deactivate(), so we can safely bail out 5065 * early here to avoid data race between 5066 * qdisc_deactivate() and some_qdisc_is_busy() 5067 * for lockless qdisc. 5068 */ 5069 clear_bit(__QDISC_STATE_SCHED, &q->state); 5070 continue; 5071 } 5072 5073 clear_bit(__QDISC_STATE_SCHED, &q->state); 5074 qdisc_run(q); 5075 if (root_lock) 5076 spin_unlock(root_lock); 5077 } 5078 5079 rcu_read_unlock(); 5080 } 5081 5082 xfrm_dev_backlog(sd); 5083 } 5084 5085 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5086 /* This hook is defined here for ATM LANE */ 5087 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5088 unsigned char *addr) __read_mostly; 5089 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5090 #endif 5091 5092 static inline struct sk_buff * 5093 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5094 struct net_device *orig_dev, bool *another) 5095 { 5096 #ifdef CONFIG_NET_CLS_ACT 5097 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5098 struct tcf_result cl_res; 5099 5100 /* If there's at least one ingress present somewhere (so 5101 * we get here via enabled static key), remaining devices 5102 * that are not configured with an ingress qdisc will bail 5103 * out here. 5104 */ 5105 if (!miniq) 5106 return skb; 5107 5108 if (*pt_prev) { 5109 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5110 *pt_prev = NULL; 5111 } 5112 5113 qdisc_skb_cb(skb)->pkt_len = skb->len; 5114 tc_skb_cb(skb)->mru = 0; 5115 tc_skb_cb(skb)->post_ct = false; 5116 skb->tc_at_ingress = 1; 5117 mini_qdisc_bstats_cpu_update(miniq, skb); 5118 5119 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5120 case TC_ACT_OK: 5121 case TC_ACT_RECLASSIFY: 5122 skb->tc_index = TC_H_MIN(cl_res.classid); 5123 break; 5124 case TC_ACT_SHOT: 5125 mini_qdisc_qstats_cpu_drop(miniq); 5126 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5127 *ret = NET_RX_DROP; 5128 return NULL; 5129 case TC_ACT_STOLEN: 5130 case TC_ACT_QUEUED: 5131 case TC_ACT_TRAP: 5132 consume_skb(skb); 5133 *ret = NET_RX_SUCCESS; 5134 return NULL; 5135 case TC_ACT_REDIRECT: 5136 /* skb_mac_header check was done by cls/act_bpf, so 5137 * we can safely push the L2 header back before 5138 * redirecting to another netdev 5139 */ 5140 __skb_push(skb, skb->mac_len); 5141 if (skb_do_redirect(skb) == -EAGAIN) { 5142 __skb_pull(skb, skb->mac_len); 5143 *another = true; 5144 break; 5145 } 5146 *ret = NET_RX_SUCCESS; 5147 return NULL; 5148 case TC_ACT_CONSUMED: 5149 *ret = NET_RX_SUCCESS; 5150 return NULL; 5151 default: 5152 break; 5153 } 5154 #endif /* CONFIG_NET_CLS_ACT */ 5155 return skb; 5156 } 5157 5158 /** 5159 * netdev_is_rx_handler_busy - check if receive handler is registered 5160 * @dev: device to check 5161 * 5162 * Check if a receive handler is already registered for a given device. 5163 * Return true if there one. 5164 * 5165 * The caller must hold the rtnl_mutex. 5166 */ 5167 bool netdev_is_rx_handler_busy(struct net_device *dev) 5168 { 5169 ASSERT_RTNL(); 5170 return dev && rtnl_dereference(dev->rx_handler); 5171 } 5172 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5173 5174 /** 5175 * netdev_rx_handler_register - register receive handler 5176 * @dev: device to register a handler for 5177 * @rx_handler: receive handler to register 5178 * @rx_handler_data: data pointer that is used by rx handler 5179 * 5180 * Register a receive handler for a device. This handler will then be 5181 * called from __netif_receive_skb. A negative errno code is returned 5182 * on a failure. 5183 * 5184 * The caller must hold the rtnl_mutex. 5185 * 5186 * For a general description of rx_handler, see enum rx_handler_result. 5187 */ 5188 int netdev_rx_handler_register(struct net_device *dev, 5189 rx_handler_func_t *rx_handler, 5190 void *rx_handler_data) 5191 { 5192 if (netdev_is_rx_handler_busy(dev)) 5193 return -EBUSY; 5194 5195 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5196 return -EINVAL; 5197 5198 /* Note: rx_handler_data must be set before rx_handler */ 5199 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5200 rcu_assign_pointer(dev->rx_handler, rx_handler); 5201 5202 return 0; 5203 } 5204 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5205 5206 /** 5207 * netdev_rx_handler_unregister - unregister receive handler 5208 * @dev: device to unregister a handler from 5209 * 5210 * Unregister a receive handler from a device. 5211 * 5212 * The caller must hold the rtnl_mutex. 5213 */ 5214 void netdev_rx_handler_unregister(struct net_device *dev) 5215 { 5216 5217 ASSERT_RTNL(); 5218 RCU_INIT_POINTER(dev->rx_handler, NULL); 5219 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5220 * section has a guarantee to see a non NULL rx_handler_data 5221 * as well. 5222 */ 5223 synchronize_net(); 5224 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5225 } 5226 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5227 5228 /* 5229 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5230 * the special handling of PFMEMALLOC skbs. 5231 */ 5232 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5233 { 5234 switch (skb->protocol) { 5235 case htons(ETH_P_ARP): 5236 case htons(ETH_P_IP): 5237 case htons(ETH_P_IPV6): 5238 case htons(ETH_P_8021Q): 5239 case htons(ETH_P_8021AD): 5240 return true; 5241 default: 5242 return false; 5243 } 5244 } 5245 5246 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5247 int *ret, struct net_device *orig_dev) 5248 { 5249 if (nf_hook_ingress_active(skb)) { 5250 int ingress_retval; 5251 5252 if (*pt_prev) { 5253 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5254 *pt_prev = NULL; 5255 } 5256 5257 rcu_read_lock(); 5258 ingress_retval = nf_hook_ingress(skb); 5259 rcu_read_unlock(); 5260 return ingress_retval; 5261 } 5262 return 0; 5263 } 5264 5265 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5266 struct packet_type **ppt_prev) 5267 { 5268 struct packet_type *ptype, *pt_prev; 5269 rx_handler_func_t *rx_handler; 5270 struct sk_buff *skb = *pskb; 5271 struct net_device *orig_dev; 5272 bool deliver_exact = false; 5273 int ret = NET_RX_DROP; 5274 __be16 type; 5275 5276 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5277 5278 trace_netif_receive_skb(skb); 5279 5280 orig_dev = skb->dev; 5281 5282 skb_reset_network_header(skb); 5283 if (!skb_transport_header_was_set(skb)) 5284 skb_reset_transport_header(skb); 5285 skb_reset_mac_len(skb); 5286 5287 pt_prev = NULL; 5288 5289 another_round: 5290 skb->skb_iif = skb->dev->ifindex; 5291 5292 __this_cpu_inc(softnet_data.processed); 5293 5294 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5295 int ret2; 5296 5297 migrate_disable(); 5298 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5299 migrate_enable(); 5300 5301 if (ret2 != XDP_PASS) { 5302 ret = NET_RX_DROP; 5303 goto out; 5304 } 5305 } 5306 5307 if (eth_type_vlan(skb->protocol)) { 5308 skb = skb_vlan_untag(skb); 5309 if (unlikely(!skb)) 5310 goto out; 5311 } 5312 5313 if (skb_skip_tc_classify(skb)) 5314 goto skip_classify; 5315 5316 if (pfmemalloc) 5317 goto skip_taps; 5318 5319 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5320 if (pt_prev) 5321 ret = deliver_skb(skb, pt_prev, orig_dev); 5322 pt_prev = ptype; 5323 } 5324 5325 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5326 if (pt_prev) 5327 ret = deliver_skb(skb, pt_prev, orig_dev); 5328 pt_prev = ptype; 5329 } 5330 5331 skip_taps: 5332 #ifdef CONFIG_NET_INGRESS 5333 if (static_branch_unlikely(&ingress_needed_key)) { 5334 bool another = false; 5335 5336 nf_skip_egress(skb, true); 5337 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5338 &another); 5339 if (another) 5340 goto another_round; 5341 if (!skb) 5342 goto out; 5343 5344 nf_skip_egress(skb, false); 5345 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5346 goto out; 5347 } 5348 #endif 5349 skb_reset_redirect(skb); 5350 skip_classify: 5351 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5352 goto drop; 5353 5354 if (skb_vlan_tag_present(skb)) { 5355 if (pt_prev) { 5356 ret = deliver_skb(skb, pt_prev, orig_dev); 5357 pt_prev = NULL; 5358 } 5359 if (vlan_do_receive(&skb)) 5360 goto another_round; 5361 else if (unlikely(!skb)) 5362 goto out; 5363 } 5364 5365 rx_handler = rcu_dereference(skb->dev->rx_handler); 5366 if (rx_handler) { 5367 if (pt_prev) { 5368 ret = deliver_skb(skb, pt_prev, orig_dev); 5369 pt_prev = NULL; 5370 } 5371 switch (rx_handler(&skb)) { 5372 case RX_HANDLER_CONSUMED: 5373 ret = NET_RX_SUCCESS; 5374 goto out; 5375 case RX_HANDLER_ANOTHER: 5376 goto another_round; 5377 case RX_HANDLER_EXACT: 5378 deliver_exact = true; 5379 break; 5380 case RX_HANDLER_PASS: 5381 break; 5382 default: 5383 BUG(); 5384 } 5385 } 5386 5387 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5388 check_vlan_id: 5389 if (skb_vlan_tag_get_id(skb)) { 5390 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5391 * find vlan device. 5392 */ 5393 skb->pkt_type = PACKET_OTHERHOST; 5394 } else if (eth_type_vlan(skb->protocol)) { 5395 /* Outer header is 802.1P with vlan 0, inner header is 5396 * 802.1Q or 802.1AD and vlan_do_receive() above could 5397 * not find vlan dev for vlan id 0. 5398 */ 5399 __vlan_hwaccel_clear_tag(skb); 5400 skb = skb_vlan_untag(skb); 5401 if (unlikely(!skb)) 5402 goto out; 5403 if (vlan_do_receive(&skb)) 5404 /* After stripping off 802.1P header with vlan 0 5405 * vlan dev is found for inner header. 5406 */ 5407 goto another_round; 5408 else if (unlikely(!skb)) 5409 goto out; 5410 else 5411 /* We have stripped outer 802.1P vlan 0 header. 5412 * But could not find vlan dev. 5413 * check again for vlan id to set OTHERHOST. 5414 */ 5415 goto check_vlan_id; 5416 } 5417 /* Note: we might in the future use prio bits 5418 * and set skb->priority like in vlan_do_receive() 5419 * For the time being, just ignore Priority Code Point 5420 */ 5421 __vlan_hwaccel_clear_tag(skb); 5422 } 5423 5424 type = skb->protocol; 5425 5426 /* deliver only exact match when indicated */ 5427 if (likely(!deliver_exact)) { 5428 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5429 &ptype_base[ntohs(type) & 5430 PTYPE_HASH_MASK]); 5431 } 5432 5433 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5434 &orig_dev->ptype_specific); 5435 5436 if (unlikely(skb->dev != orig_dev)) { 5437 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5438 &skb->dev->ptype_specific); 5439 } 5440 5441 if (pt_prev) { 5442 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5443 goto drop; 5444 *ppt_prev = pt_prev; 5445 } else { 5446 drop: 5447 if (!deliver_exact) 5448 dev_core_stats_rx_dropped_inc(skb->dev); 5449 else 5450 dev_core_stats_rx_nohandler_inc(skb->dev); 5451 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5452 /* Jamal, now you will not able to escape explaining 5453 * me how you were going to use this. :-) 5454 */ 5455 ret = NET_RX_DROP; 5456 } 5457 5458 out: 5459 /* The invariant here is that if *ppt_prev is not NULL 5460 * then skb should also be non-NULL. 5461 * 5462 * Apparently *ppt_prev assignment above holds this invariant due to 5463 * skb dereferencing near it. 5464 */ 5465 *pskb = skb; 5466 return ret; 5467 } 5468 5469 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5470 { 5471 struct net_device *orig_dev = skb->dev; 5472 struct packet_type *pt_prev = NULL; 5473 int ret; 5474 5475 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5476 if (pt_prev) 5477 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5478 skb->dev, pt_prev, orig_dev); 5479 return ret; 5480 } 5481 5482 /** 5483 * netif_receive_skb_core - special purpose version of netif_receive_skb 5484 * @skb: buffer to process 5485 * 5486 * More direct receive version of netif_receive_skb(). It should 5487 * only be used by callers that have a need to skip RPS and Generic XDP. 5488 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5489 * 5490 * This function may only be called from softirq context and interrupts 5491 * should be enabled. 5492 * 5493 * Return values (usually ignored): 5494 * NET_RX_SUCCESS: no congestion 5495 * NET_RX_DROP: packet was dropped 5496 */ 5497 int netif_receive_skb_core(struct sk_buff *skb) 5498 { 5499 int ret; 5500 5501 rcu_read_lock(); 5502 ret = __netif_receive_skb_one_core(skb, false); 5503 rcu_read_unlock(); 5504 5505 return ret; 5506 } 5507 EXPORT_SYMBOL(netif_receive_skb_core); 5508 5509 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5510 struct packet_type *pt_prev, 5511 struct net_device *orig_dev) 5512 { 5513 struct sk_buff *skb, *next; 5514 5515 if (!pt_prev) 5516 return; 5517 if (list_empty(head)) 5518 return; 5519 if (pt_prev->list_func != NULL) 5520 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5521 ip_list_rcv, head, pt_prev, orig_dev); 5522 else 5523 list_for_each_entry_safe(skb, next, head, list) { 5524 skb_list_del_init(skb); 5525 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5526 } 5527 } 5528 5529 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5530 { 5531 /* Fast-path assumptions: 5532 * - There is no RX handler. 5533 * - Only one packet_type matches. 5534 * If either of these fails, we will end up doing some per-packet 5535 * processing in-line, then handling the 'last ptype' for the whole 5536 * sublist. This can't cause out-of-order delivery to any single ptype, 5537 * because the 'last ptype' must be constant across the sublist, and all 5538 * other ptypes are handled per-packet. 5539 */ 5540 /* Current (common) ptype of sublist */ 5541 struct packet_type *pt_curr = NULL; 5542 /* Current (common) orig_dev of sublist */ 5543 struct net_device *od_curr = NULL; 5544 struct list_head sublist; 5545 struct sk_buff *skb, *next; 5546 5547 INIT_LIST_HEAD(&sublist); 5548 list_for_each_entry_safe(skb, next, head, list) { 5549 struct net_device *orig_dev = skb->dev; 5550 struct packet_type *pt_prev = NULL; 5551 5552 skb_list_del_init(skb); 5553 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5554 if (!pt_prev) 5555 continue; 5556 if (pt_curr != pt_prev || od_curr != orig_dev) { 5557 /* dispatch old sublist */ 5558 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5559 /* start new sublist */ 5560 INIT_LIST_HEAD(&sublist); 5561 pt_curr = pt_prev; 5562 od_curr = orig_dev; 5563 } 5564 list_add_tail(&skb->list, &sublist); 5565 } 5566 5567 /* dispatch final sublist */ 5568 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5569 } 5570 5571 static int __netif_receive_skb(struct sk_buff *skb) 5572 { 5573 int ret; 5574 5575 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5576 unsigned int noreclaim_flag; 5577 5578 /* 5579 * PFMEMALLOC skbs are special, they should 5580 * - be delivered to SOCK_MEMALLOC sockets only 5581 * - stay away from userspace 5582 * - have bounded memory usage 5583 * 5584 * Use PF_MEMALLOC as this saves us from propagating the allocation 5585 * context down to all allocation sites. 5586 */ 5587 noreclaim_flag = memalloc_noreclaim_save(); 5588 ret = __netif_receive_skb_one_core(skb, true); 5589 memalloc_noreclaim_restore(noreclaim_flag); 5590 } else 5591 ret = __netif_receive_skb_one_core(skb, false); 5592 5593 return ret; 5594 } 5595 5596 static void __netif_receive_skb_list(struct list_head *head) 5597 { 5598 unsigned long noreclaim_flag = 0; 5599 struct sk_buff *skb, *next; 5600 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5601 5602 list_for_each_entry_safe(skb, next, head, list) { 5603 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5604 struct list_head sublist; 5605 5606 /* Handle the previous sublist */ 5607 list_cut_before(&sublist, head, &skb->list); 5608 if (!list_empty(&sublist)) 5609 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5610 pfmemalloc = !pfmemalloc; 5611 /* See comments in __netif_receive_skb */ 5612 if (pfmemalloc) 5613 noreclaim_flag = memalloc_noreclaim_save(); 5614 else 5615 memalloc_noreclaim_restore(noreclaim_flag); 5616 } 5617 } 5618 /* Handle the remaining sublist */ 5619 if (!list_empty(head)) 5620 __netif_receive_skb_list_core(head, pfmemalloc); 5621 /* Restore pflags */ 5622 if (pfmemalloc) 5623 memalloc_noreclaim_restore(noreclaim_flag); 5624 } 5625 5626 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5627 { 5628 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5629 struct bpf_prog *new = xdp->prog; 5630 int ret = 0; 5631 5632 switch (xdp->command) { 5633 case XDP_SETUP_PROG: 5634 rcu_assign_pointer(dev->xdp_prog, new); 5635 if (old) 5636 bpf_prog_put(old); 5637 5638 if (old && !new) { 5639 static_branch_dec(&generic_xdp_needed_key); 5640 } else if (new && !old) { 5641 static_branch_inc(&generic_xdp_needed_key); 5642 dev_disable_lro(dev); 5643 dev_disable_gro_hw(dev); 5644 } 5645 break; 5646 5647 default: 5648 ret = -EINVAL; 5649 break; 5650 } 5651 5652 return ret; 5653 } 5654 5655 static int netif_receive_skb_internal(struct sk_buff *skb) 5656 { 5657 int ret; 5658 5659 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5660 5661 if (skb_defer_rx_timestamp(skb)) 5662 return NET_RX_SUCCESS; 5663 5664 rcu_read_lock(); 5665 #ifdef CONFIG_RPS 5666 if (static_branch_unlikely(&rps_needed)) { 5667 struct rps_dev_flow voidflow, *rflow = &voidflow; 5668 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5669 5670 if (cpu >= 0) { 5671 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5672 rcu_read_unlock(); 5673 return ret; 5674 } 5675 } 5676 #endif 5677 ret = __netif_receive_skb(skb); 5678 rcu_read_unlock(); 5679 return ret; 5680 } 5681 5682 void netif_receive_skb_list_internal(struct list_head *head) 5683 { 5684 struct sk_buff *skb, *next; 5685 struct list_head sublist; 5686 5687 INIT_LIST_HEAD(&sublist); 5688 list_for_each_entry_safe(skb, next, head, list) { 5689 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5690 skb_list_del_init(skb); 5691 if (!skb_defer_rx_timestamp(skb)) 5692 list_add_tail(&skb->list, &sublist); 5693 } 5694 list_splice_init(&sublist, head); 5695 5696 rcu_read_lock(); 5697 #ifdef CONFIG_RPS 5698 if (static_branch_unlikely(&rps_needed)) { 5699 list_for_each_entry_safe(skb, next, head, list) { 5700 struct rps_dev_flow voidflow, *rflow = &voidflow; 5701 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5702 5703 if (cpu >= 0) { 5704 /* Will be handled, remove from list */ 5705 skb_list_del_init(skb); 5706 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5707 } 5708 } 5709 } 5710 #endif 5711 __netif_receive_skb_list(head); 5712 rcu_read_unlock(); 5713 } 5714 5715 /** 5716 * netif_receive_skb - process receive buffer from network 5717 * @skb: buffer to process 5718 * 5719 * netif_receive_skb() is the main receive data processing function. 5720 * It always succeeds. The buffer may be dropped during processing 5721 * for congestion control or by the protocol layers. 5722 * 5723 * This function may only be called from softirq context and interrupts 5724 * should be enabled. 5725 * 5726 * Return values (usually ignored): 5727 * NET_RX_SUCCESS: no congestion 5728 * NET_RX_DROP: packet was dropped 5729 */ 5730 int netif_receive_skb(struct sk_buff *skb) 5731 { 5732 int ret; 5733 5734 trace_netif_receive_skb_entry(skb); 5735 5736 ret = netif_receive_skb_internal(skb); 5737 trace_netif_receive_skb_exit(ret); 5738 5739 return ret; 5740 } 5741 EXPORT_SYMBOL(netif_receive_skb); 5742 5743 /** 5744 * netif_receive_skb_list - process many receive buffers from network 5745 * @head: list of skbs to process. 5746 * 5747 * Since return value of netif_receive_skb() is normally ignored, and 5748 * wouldn't be meaningful for a list, this function returns void. 5749 * 5750 * This function may only be called from softirq context and interrupts 5751 * should be enabled. 5752 */ 5753 void netif_receive_skb_list(struct list_head *head) 5754 { 5755 struct sk_buff *skb; 5756 5757 if (list_empty(head)) 5758 return; 5759 if (trace_netif_receive_skb_list_entry_enabled()) { 5760 list_for_each_entry(skb, head, list) 5761 trace_netif_receive_skb_list_entry(skb); 5762 } 5763 netif_receive_skb_list_internal(head); 5764 trace_netif_receive_skb_list_exit(0); 5765 } 5766 EXPORT_SYMBOL(netif_receive_skb_list); 5767 5768 static DEFINE_PER_CPU(struct work_struct, flush_works); 5769 5770 /* Network device is going away, flush any packets still pending */ 5771 static void flush_backlog(struct work_struct *work) 5772 { 5773 struct sk_buff *skb, *tmp; 5774 struct softnet_data *sd; 5775 5776 local_bh_disable(); 5777 sd = this_cpu_ptr(&softnet_data); 5778 5779 rps_lock_irq_disable(sd); 5780 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5781 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5782 __skb_unlink(skb, &sd->input_pkt_queue); 5783 dev_kfree_skb_irq(skb); 5784 input_queue_head_incr(sd); 5785 } 5786 } 5787 rps_unlock_irq_enable(sd); 5788 5789 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5790 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5791 __skb_unlink(skb, &sd->process_queue); 5792 kfree_skb(skb); 5793 input_queue_head_incr(sd); 5794 } 5795 } 5796 local_bh_enable(); 5797 } 5798 5799 static bool flush_required(int cpu) 5800 { 5801 #if IS_ENABLED(CONFIG_RPS) 5802 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5803 bool do_flush; 5804 5805 rps_lock_irq_disable(sd); 5806 5807 /* as insertion into process_queue happens with the rps lock held, 5808 * process_queue access may race only with dequeue 5809 */ 5810 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5811 !skb_queue_empty_lockless(&sd->process_queue); 5812 rps_unlock_irq_enable(sd); 5813 5814 return do_flush; 5815 #endif 5816 /* without RPS we can't safely check input_pkt_queue: during a 5817 * concurrent remote skb_queue_splice() we can detect as empty both 5818 * input_pkt_queue and process_queue even if the latter could end-up 5819 * containing a lot of packets. 5820 */ 5821 return true; 5822 } 5823 5824 static void flush_all_backlogs(void) 5825 { 5826 static cpumask_t flush_cpus; 5827 unsigned int cpu; 5828 5829 /* since we are under rtnl lock protection we can use static data 5830 * for the cpumask and avoid allocating on stack the possibly 5831 * large mask 5832 */ 5833 ASSERT_RTNL(); 5834 5835 cpus_read_lock(); 5836 5837 cpumask_clear(&flush_cpus); 5838 for_each_online_cpu(cpu) { 5839 if (flush_required(cpu)) { 5840 queue_work_on(cpu, system_highpri_wq, 5841 per_cpu_ptr(&flush_works, cpu)); 5842 cpumask_set_cpu(cpu, &flush_cpus); 5843 } 5844 } 5845 5846 /* we can have in flight packet[s] on the cpus we are not flushing, 5847 * synchronize_net() in unregister_netdevice_many() will take care of 5848 * them 5849 */ 5850 for_each_cpu(cpu, &flush_cpus) 5851 flush_work(per_cpu_ptr(&flush_works, cpu)); 5852 5853 cpus_read_unlock(); 5854 } 5855 5856 static void net_rps_send_ipi(struct softnet_data *remsd) 5857 { 5858 #ifdef CONFIG_RPS 5859 while (remsd) { 5860 struct softnet_data *next = remsd->rps_ipi_next; 5861 5862 if (cpu_online(remsd->cpu)) 5863 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5864 remsd = next; 5865 } 5866 #endif 5867 } 5868 5869 /* 5870 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5871 * Note: called with local irq disabled, but exits with local irq enabled. 5872 */ 5873 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5874 { 5875 #ifdef CONFIG_RPS 5876 struct softnet_data *remsd = sd->rps_ipi_list; 5877 5878 if (remsd) { 5879 sd->rps_ipi_list = NULL; 5880 5881 local_irq_enable(); 5882 5883 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5884 net_rps_send_ipi(remsd); 5885 } else 5886 #endif 5887 local_irq_enable(); 5888 } 5889 5890 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5891 { 5892 #ifdef CONFIG_RPS 5893 return sd->rps_ipi_list != NULL; 5894 #else 5895 return false; 5896 #endif 5897 } 5898 5899 static int process_backlog(struct napi_struct *napi, int quota) 5900 { 5901 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5902 bool again = true; 5903 int work = 0; 5904 5905 /* Check if we have pending ipi, its better to send them now, 5906 * not waiting net_rx_action() end. 5907 */ 5908 if (sd_has_rps_ipi_waiting(sd)) { 5909 local_irq_disable(); 5910 net_rps_action_and_irq_enable(sd); 5911 } 5912 5913 napi->weight = READ_ONCE(dev_rx_weight); 5914 while (again) { 5915 struct sk_buff *skb; 5916 5917 while ((skb = __skb_dequeue(&sd->process_queue))) { 5918 rcu_read_lock(); 5919 __netif_receive_skb(skb); 5920 rcu_read_unlock(); 5921 input_queue_head_incr(sd); 5922 if (++work >= quota) 5923 return work; 5924 5925 } 5926 5927 rps_lock_irq_disable(sd); 5928 if (skb_queue_empty(&sd->input_pkt_queue)) { 5929 /* 5930 * Inline a custom version of __napi_complete(). 5931 * only current cpu owns and manipulates this napi, 5932 * and NAPI_STATE_SCHED is the only possible flag set 5933 * on backlog. 5934 * We can use a plain write instead of clear_bit(), 5935 * and we dont need an smp_mb() memory barrier. 5936 */ 5937 napi->state = 0; 5938 again = false; 5939 } else { 5940 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5941 &sd->process_queue); 5942 } 5943 rps_unlock_irq_enable(sd); 5944 } 5945 5946 return work; 5947 } 5948 5949 /** 5950 * __napi_schedule - schedule for receive 5951 * @n: entry to schedule 5952 * 5953 * The entry's receive function will be scheduled to run. 5954 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5955 */ 5956 void __napi_schedule(struct napi_struct *n) 5957 { 5958 unsigned long flags; 5959 5960 local_irq_save(flags); 5961 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5962 local_irq_restore(flags); 5963 } 5964 EXPORT_SYMBOL(__napi_schedule); 5965 5966 /** 5967 * napi_schedule_prep - check if napi can be scheduled 5968 * @n: napi context 5969 * 5970 * Test if NAPI routine is already running, and if not mark 5971 * it as running. This is used as a condition variable to 5972 * insure only one NAPI poll instance runs. We also make 5973 * sure there is no pending NAPI disable. 5974 */ 5975 bool napi_schedule_prep(struct napi_struct *n) 5976 { 5977 unsigned long new, val = READ_ONCE(n->state); 5978 5979 do { 5980 if (unlikely(val & NAPIF_STATE_DISABLE)) 5981 return false; 5982 new = val | NAPIF_STATE_SCHED; 5983 5984 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5985 * This was suggested by Alexander Duyck, as compiler 5986 * emits better code than : 5987 * if (val & NAPIF_STATE_SCHED) 5988 * new |= NAPIF_STATE_MISSED; 5989 */ 5990 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5991 NAPIF_STATE_MISSED; 5992 } while (!try_cmpxchg(&n->state, &val, new)); 5993 5994 return !(val & NAPIF_STATE_SCHED); 5995 } 5996 EXPORT_SYMBOL(napi_schedule_prep); 5997 5998 /** 5999 * __napi_schedule_irqoff - schedule for receive 6000 * @n: entry to schedule 6001 * 6002 * Variant of __napi_schedule() assuming hard irqs are masked. 6003 * 6004 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6005 * because the interrupt disabled assumption might not be true 6006 * due to force-threaded interrupts and spinlock substitution. 6007 */ 6008 void __napi_schedule_irqoff(struct napi_struct *n) 6009 { 6010 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6011 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6012 else 6013 __napi_schedule(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, timeout = 0; 6020 bool ret = true; 6021 6022 /* 6023 * 1) Don't let napi dequeue from the cpu poll list 6024 * just in case its running on a different cpu. 6025 * 2) If we are busy polling, do nothing here, we have 6026 * the guarantee we will be called later. 6027 */ 6028 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6029 NAPIF_STATE_IN_BUSY_POLL))) 6030 return false; 6031 6032 if (work_done) { 6033 if (n->gro_bitmask) 6034 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6035 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6036 } 6037 if (n->defer_hard_irqs_count > 0) { 6038 n->defer_hard_irqs_count--; 6039 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6040 if (timeout) 6041 ret = false; 6042 } 6043 if (n->gro_bitmask) { 6044 /* When the NAPI instance uses a timeout and keeps postponing 6045 * it, we need to bound somehow the time packets are kept in 6046 * the GRO layer 6047 */ 6048 napi_gro_flush(n, !!timeout); 6049 } 6050 6051 gro_normal_list(n); 6052 6053 if (unlikely(!list_empty(&n->poll_list))) { 6054 /* If n->poll_list is not empty, we need to mask irqs */ 6055 local_irq_save(flags); 6056 list_del_init(&n->poll_list); 6057 local_irq_restore(flags); 6058 } 6059 6060 val = READ_ONCE(n->state); 6061 do { 6062 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6063 6064 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6065 NAPIF_STATE_SCHED_THREADED | 6066 NAPIF_STATE_PREFER_BUSY_POLL); 6067 6068 /* If STATE_MISSED was set, leave STATE_SCHED set, 6069 * because we will call napi->poll() one more time. 6070 * This C code was suggested by Alexander Duyck to help gcc. 6071 */ 6072 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6073 NAPIF_STATE_SCHED; 6074 } while (!try_cmpxchg(&n->state, &val, new)); 6075 6076 if (unlikely(val & NAPIF_STATE_MISSED)) { 6077 __napi_schedule(n); 6078 return false; 6079 } 6080 6081 if (timeout) 6082 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6083 HRTIMER_MODE_REL_PINNED); 6084 return ret; 6085 } 6086 EXPORT_SYMBOL(napi_complete_done); 6087 6088 /* must be called under rcu_read_lock(), as we dont take a reference */ 6089 static struct napi_struct *napi_by_id(unsigned int napi_id) 6090 { 6091 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6092 struct napi_struct *napi; 6093 6094 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6095 if (napi->napi_id == napi_id) 6096 return napi; 6097 6098 return NULL; 6099 } 6100 6101 #if defined(CONFIG_NET_RX_BUSY_POLL) 6102 6103 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6104 { 6105 if (!skip_schedule) { 6106 gro_normal_list(napi); 6107 __napi_schedule(napi); 6108 return; 6109 } 6110 6111 if (napi->gro_bitmask) { 6112 /* flush too old packets 6113 * If HZ < 1000, flush all packets. 6114 */ 6115 napi_gro_flush(napi, HZ >= 1000); 6116 } 6117 6118 gro_normal_list(napi); 6119 clear_bit(NAPI_STATE_SCHED, &napi->state); 6120 } 6121 6122 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6123 u16 budget) 6124 { 6125 bool skip_schedule = false; 6126 unsigned long timeout; 6127 int rc; 6128 6129 /* Busy polling means there is a high chance device driver hard irq 6130 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6131 * set in napi_schedule_prep(). 6132 * Since we are about to call napi->poll() once more, we can safely 6133 * clear NAPI_STATE_MISSED. 6134 * 6135 * Note: x86 could use a single "lock and ..." instruction 6136 * to perform these two clear_bit() 6137 */ 6138 clear_bit(NAPI_STATE_MISSED, &napi->state); 6139 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6140 6141 local_bh_disable(); 6142 6143 if (prefer_busy_poll) { 6144 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6145 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6146 if (napi->defer_hard_irqs_count && timeout) { 6147 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6148 skip_schedule = true; 6149 } 6150 } 6151 6152 /* All we really want here is to re-enable device interrupts. 6153 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6154 */ 6155 rc = napi->poll(napi, budget); 6156 /* We can't gro_normal_list() here, because napi->poll() might have 6157 * rearmed the napi (napi_complete_done()) in which case it could 6158 * already be running on another CPU. 6159 */ 6160 trace_napi_poll(napi, rc, budget); 6161 netpoll_poll_unlock(have_poll_lock); 6162 if (rc == budget) 6163 __busy_poll_stop(napi, skip_schedule); 6164 local_bh_enable(); 6165 } 6166 6167 void napi_busy_loop(unsigned int napi_id, 6168 bool (*loop_end)(void *, unsigned long), 6169 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6170 { 6171 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6172 int (*napi_poll)(struct napi_struct *napi, int budget); 6173 void *have_poll_lock = NULL; 6174 struct napi_struct *napi; 6175 6176 restart: 6177 napi_poll = NULL; 6178 6179 rcu_read_lock(); 6180 6181 napi = napi_by_id(napi_id); 6182 if (!napi) 6183 goto out; 6184 6185 preempt_disable(); 6186 for (;;) { 6187 int work = 0; 6188 6189 local_bh_disable(); 6190 if (!napi_poll) { 6191 unsigned long val = READ_ONCE(napi->state); 6192 6193 /* If multiple threads are competing for this napi, 6194 * we avoid dirtying napi->state as much as we can. 6195 */ 6196 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6197 NAPIF_STATE_IN_BUSY_POLL)) { 6198 if (prefer_busy_poll) 6199 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6200 goto count; 6201 } 6202 if (cmpxchg(&napi->state, val, 6203 val | NAPIF_STATE_IN_BUSY_POLL | 6204 NAPIF_STATE_SCHED) != val) { 6205 if (prefer_busy_poll) 6206 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6207 goto count; 6208 } 6209 have_poll_lock = netpoll_poll_lock(napi); 6210 napi_poll = napi->poll; 6211 } 6212 work = napi_poll(napi, budget); 6213 trace_napi_poll(napi, work, budget); 6214 gro_normal_list(napi); 6215 count: 6216 if (work > 0) 6217 __NET_ADD_STATS(dev_net(napi->dev), 6218 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6219 local_bh_enable(); 6220 6221 if (!loop_end || loop_end(loop_end_arg, start_time)) 6222 break; 6223 6224 if (unlikely(need_resched())) { 6225 if (napi_poll) 6226 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6227 preempt_enable(); 6228 rcu_read_unlock(); 6229 cond_resched(); 6230 if (loop_end(loop_end_arg, start_time)) 6231 return; 6232 goto restart; 6233 } 6234 cpu_relax(); 6235 } 6236 if (napi_poll) 6237 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6238 preempt_enable(); 6239 out: 6240 rcu_read_unlock(); 6241 } 6242 EXPORT_SYMBOL(napi_busy_loop); 6243 6244 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6245 6246 static void napi_hash_add(struct napi_struct *napi) 6247 { 6248 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6249 return; 6250 6251 spin_lock(&napi_hash_lock); 6252 6253 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6254 do { 6255 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6256 napi_gen_id = MIN_NAPI_ID; 6257 } while (napi_by_id(napi_gen_id)); 6258 napi->napi_id = napi_gen_id; 6259 6260 hlist_add_head_rcu(&napi->napi_hash_node, 6261 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6262 6263 spin_unlock(&napi_hash_lock); 6264 } 6265 6266 /* Warning : caller is responsible to make sure rcu grace period 6267 * is respected before freeing memory containing @napi 6268 */ 6269 static void napi_hash_del(struct napi_struct *napi) 6270 { 6271 spin_lock(&napi_hash_lock); 6272 6273 hlist_del_init_rcu(&napi->napi_hash_node); 6274 6275 spin_unlock(&napi_hash_lock); 6276 } 6277 6278 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6279 { 6280 struct napi_struct *napi; 6281 6282 napi = container_of(timer, struct napi_struct, timer); 6283 6284 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6285 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6286 */ 6287 if (!napi_disable_pending(napi) && 6288 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6289 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6290 __napi_schedule_irqoff(napi); 6291 } 6292 6293 return HRTIMER_NORESTART; 6294 } 6295 6296 static void init_gro_hash(struct napi_struct *napi) 6297 { 6298 int i; 6299 6300 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6301 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6302 napi->gro_hash[i].count = 0; 6303 } 6304 napi->gro_bitmask = 0; 6305 } 6306 6307 int dev_set_threaded(struct net_device *dev, bool threaded) 6308 { 6309 struct napi_struct *napi; 6310 int err = 0; 6311 6312 if (dev->threaded == threaded) 6313 return 0; 6314 6315 if (threaded) { 6316 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6317 if (!napi->thread) { 6318 err = napi_kthread_create(napi); 6319 if (err) { 6320 threaded = false; 6321 break; 6322 } 6323 } 6324 } 6325 } 6326 6327 dev->threaded = threaded; 6328 6329 /* Make sure kthread is created before THREADED bit 6330 * is set. 6331 */ 6332 smp_mb__before_atomic(); 6333 6334 /* Setting/unsetting threaded mode on a napi might not immediately 6335 * take effect, if the current napi instance is actively being 6336 * polled. In this case, the switch between threaded mode and 6337 * softirq mode will happen in the next round of napi_schedule(). 6338 * This should not cause hiccups/stalls to the live traffic. 6339 */ 6340 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6341 if (threaded) 6342 set_bit(NAPI_STATE_THREADED, &napi->state); 6343 else 6344 clear_bit(NAPI_STATE_THREADED, &napi->state); 6345 } 6346 6347 return err; 6348 } 6349 EXPORT_SYMBOL(dev_set_threaded); 6350 6351 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6352 int (*poll)(struct napi_struct *, int), int weight) 6353 { 6354 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6355 return; 6356 6357 INIT_LIST_HEAD(&napi->poll_list); 6358 INIT_HLIST_NODE(&napi->napi_hash_node); 6359 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6360 napi->timer.function = napi_watchdog; 6361 init_gro_hash(napi); 6362 napi->skb = NULL; 6363 INIT_LIST_HEAD(&napi->rx_list); 6364 napi->rx_count = 0; 6365 napi->poll = poll; 6366 if (weight > NAPI_POLL_WEIGHT) 6367 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6368 weight); 6369 napi->weight = weight; 6370 napi->dev = dev; 6371 #ifdef CONFIG_NETPOLL 6372 napi->poll_owner = -1; 6373 #endif 6374 set_bit(NAPI_STATE_SCHED, &napi->state); 6375 set_bit(NAPI_STATE_NPSVC, &napi->state); 6376 list_add_rcu(&napi->dev_list, &dev->napi_list); 6377 napi_hash_add(napi); 6378 napi_get_frags_check(napi); 6379 /* Create kthread for this napi if dev->threaded is set. 6380 * Clear dev->threaded if kthread creation failed so that 6381 * threaded mode will not be enabled in napi_enable(). 6382 */ 6383 if (dev->threaded && napi_kthread_create(napi)) 6384 dev->threaded = 0; 6385 } 6386 EXPORT_SYMBOL(netif_napi_add_weight); 6387 6388 void napi_disable(struct napi_struct *n) 6389 { 6390 unsigned long val, new; 6391 6392 might_sleep(); 6393 set_bit(NAPI_STATE_DISABLE, &n->state); 6394 6395 val = READ_ONCE(n->state); 6396 do { 6397 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6398 usleep_range(20, 200); 6399 val = READ_ONCE(n->state); 6400 } 6401 6402 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6403 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6404 } while (!try_cmpxchg(&n->state, &val, new)); 6405 6406 hrtimer_cancel(&n->timer); 6407 6408 clear_bit(NAPI_STATE_DISABLE, &n->state); 6409 } 6410 EXPORT_SYMBOL(napi_disable); 6411 6412 /** 6413 * napi_enable - enable NAPI scheduling 6414 * @n: NAPI context 6415 * 6416 * Resume NAPI from being scheduled on this context. 6417 * Must be paired with napi_disable. 6418 */ 6419 void napi_enable(struct napi_struct *n) 6420 { 6421 unsigned long new, val = READ_ONCE(n->state); 6422 6423 do { 6424 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6425 6426 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6427 if (n->dev->threaded && n->thread) 6428 new |= NAPIF_STATE_THREADED; 6429 } while (!try_cmpxchg(&n->state, &val, new)); 6430 } 6431 EXPORT_SYMBOL(napi_enable); 6432 6433 static void flush_gro_hash(struct napi_struct *napi) 6434 { 6435 int i; 6436 6437 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6438 struct sk_buff *skb, *n; 6439 6440 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6441 kfree_skb(skb); 6442 napi->gro_hash[i].count = 0; 6443 } 6444 } 6445 6446 /* Must be called in process context */ 6447 void __netif_napi_del(struct napi_struct *napi) 6448 { 6449 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6450 return; 6451 6452 napi_hash_del(napi); 6453 list_del_rcu(&napi->dev_list); 6454 napi_free_frags(napi); 6455 6456 flush_gro_hash(napi); 6457 napi->gro_bitmask = 0; 6458 6459 if (napi->thread) { 6460 kthread_stop(napi->thread); 6461 napi->thread = NULL; 6462 } 6463 } 6464 EXPORT_SYMBOL(__netif_napi_del); 6465 6466 static int __napi_poll(struct napi_struct *n, bool *repoll) 6467 { 6468 int work, weight; 6469 6470 weight = n->weight; 6471 6472 /* This NAPI_STATE_SCHED test is for avoiding a race 6473 * with netpoll's poll_napi(). Only the entity which 6474 * obtains the lock and sees NAPI_STATE_SCHED set will 6475 * actually make the ->poll() call. Therefore we avoid 6476 * accidentally calling ->poll() when NAPI is not scheduled. 6477 */ 6478 work = 0; 6479 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6480 work = n->poll(n, weight); 6481 trace_napi_poll(n, work, weight); 6482 } 6483 6484 if (unlikely(work > weight)) 6485 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6486 n->poll, work, weight); 6487 6488 if (likely(work < weight)) 6489 return work; 6490 6491 /* Drivers must not modify the NAPI state if they 6492 * consume the entire weight. In such cases this code 6493 * still "owns" the NAPI instance and therefore can 6494 * move the instance around on the list at-will. 6495 */ 6496 if (unlikely(napi_disable_pending(n))) { 6497 napi_complete(n); 6498 return work; 6499 } 6500 6501 /* The NAPI context has more processing work, but busy-polling 6502 * is preferred. Exit early. 6503 */ 6504 if (napi_prefer_busy_poll(n)) { 6505 if (napi_complete_done(n, work)) { 6506 /* If timeout is not set, we need to make sure 6507 * that the NAPI is re-scheduled. 6508 */ 6509 napi_schedule(n); 6510 } 6511 return work; 6512 } 6513 6514 if (n->gro_bitmask) { 6515 /* flush too old packets 6516 * If HZ < 1000, flush all packets. 6517 */ 6518 napi_gro_flush(n, HZ >= 1000); 6519 } 6520 6521 gro_normal_list(n); 6522 6523 /* Some drivers may have called napi_schedule 6524 * prior to exhausting their budget. 6525 */ 6526 if (unlikely(!list_empty(&n->poll_list))) { 6527 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6528 n->dev ? n->dev->name : "backlog"); 6529 return work; 6530 } 6531 6532 *repoll = true; 6533 6534 return work; 6535 } 6536 6537 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6538 { 6539 bool do_repoll = false; 6540 void *have; 6541 int work; 6542 6543 list_del_init(&n->poll_list); 6544 6545 have = netpoll_poll_lock(n); 6546 6547 work = __napi_poll(n, &do_repoll); 6548 6549 if (do_repoll) 6550 list_add_tail(&n->poll_list, repoll); 6551 6552 netpoll_poll_unlock(have); 6553 6554 return work; 6555 } 6556 6557 static int napi_thread_wait(struct napi_struct *napi) 6558 { 6559 bool woken = false; 6560 6561 set_current_state(TASK_INTERRUPTIBLE); 6562 6563 while (!kthread_should_stop()) { 6564 /* Testing SCHED_THREADED bit here to make sure the current 6565 * kthread owns this napi and could poll on this napi. 6566 * Testing SCHED bit is not enough because SCHED bit might be 6567 * set by some other busy poll thread or by napi_disable(). 6568 */ 6569 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6570 WARN_ON(!list_empty(&napi->poll_list)); 6571 __set_current_state(TASK_RUNNING); 6572 return 0; 6573 } 6574 6575 schedule(); 6576 /* woken being true indicates this thread owns this napi. */ 6577 woken = true; 6578 set_current_state(TASK_INTERRUPTIBLE); 6579 } 6580 __set_current_state(TASK_RUNNING); 6581 6582 return -1; 6583 } 6584 6585 static int napi_threaded_poll(void *data) 6586 { 6587 struct napi_struct *napi = data; 6588 void *have; 6589 6590 while (!napi_thread_wait(napi)) { 6591 for (;;) { 6592 bool repoll = false; 6593 6594 local_bh_disable(); 6595 6596 have = netpoll_poll_lock(napi); 6597 __napi_poll(napi, &repoll); 6598 netpoll_poll_unlock(have); 6599 6600 local_bh_enable(); 6601 6602 if (!repoll) 6603 break; 6604 6605 cond_resched(); 6606 } 6607 } 6608 return 0; 6609 } 6610 6611 static void skb_defer_free_flush(struct softnet_data *sd) 6612 { 6613 struct sk_buff *skb, *next; 6614 6615 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6616 if (!READ_ONCE(sd->defer_list)) 6617 return; 6618 6619 spin_lock_irq(&sd->defer_lock); 6620 skb = sd->defer_list; 6621 sd->defer_list = NULL; 6622 sd->defer_count = 0; 6623 spin_unlock_irq(&sd->defer_lock); 6624 6625 while (skb != NULL) { 6626 next = skb->next; 6627 napi_consume_skb(skb, 1); 6628 skb = next; 6629 } 6630 } 6631 6632 static __latent_entropy void net_rx_action(struct softirq_action *h) 6633 { 6634 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6635 unsigned long time_limit = jiffies + 6636 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6637 int budget = READ_ONCE(netdev_budget); 6638 LIST_HEAD(list); 6639 LIST_HEAD(repoll); 6640 6641 local_irq_disable(); 6642 list_splice_init(&sd->poll_list, &list); 6643 local_irq_enable(); 6644 6645 for (;;) { 6646 struct napi_struct *n; 6647 6648 skb_defer_free_flush(sd); 6649 6650 if (list_empty(&list)) { 6651 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6652 goto end; 6653 break; 6654 } 6655 6656 n = list_first_entry(&list, struct napi_struct, poll_list); 6657 budget -= napi_poll(n, &repoll); 6658 6659 /* If softirq window is exhausted then punt. 6660 * Allow this to run for 2 jiffies since which will allow 6661 * an average latency of 1.5/HZ. 6662 */ 6663 if (unlikely(budget <= 0 || 6664 time_after_eq(jiffies, time_limit))) { 6665 sd->time_squeeze++; 6666 break; 6667 } 6668 } 6669 6670 local_irq_disable(); 6671 6672 list_splice_tail_init(&sd->poll_list, &list); 6673 list_splice_tail(&repoll, &list); 6674 list_splice(&list, &sd->poll_list); 6675 if (!list_empty(&sd->poll_list)) 6676 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6677 6678 net_rps_action_and_irq_enable(sd); 6679 end:; 6680 } 6681 6682 struct netdev_adjacent { 6683 struct net_device *dev; 6684 netdevice_tracker dev_tracker; 6685 6686 /* upper master flag, there can only be one master device per list */ 6687 bool master; 6688 6689 /* lookup ignore flag */ 6690 bool ignore; 6691 6692 /* counter for the number of times this device was added to us */ 6693 u16 ref_nr; 6694 6695 /* private field for the users */ 6696 void *private; 6697 6698 struct list_head list; 6699 struct rcu_head rcu; 6700 }; 6701 6702 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6703 struct list_head *adj_list) 6704 { 6705 struct netdev_adjacent *adj; 6706 6707 list_for_each_entry(adj, adj_list, list) { 6708 if (adj->dev == adj_dev) 6709 return adj; 6710 } 6711 return NULL; 6712 } 6713 6714 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6715 struct netdev_nested_priv *priv) 6716 { 6717 struct net_device *dev = (struct net_device *)priv->data; 6718 6719 return upper_dev == dev; 6720 } 6721 6722 /** 6723 * netdev_has_upper_dev - Check if device is linked to an upper device 6724 * @dev: device 6725 * @upper_dev: upper device to check 6726 * 6727 * Find out if a device is linked to specified upper device and return true 6728 * in case it is. Note that this checks only immediate upper device, 6729 * not through a complete stack of devices. The caller must hold the RTNL lock. 6730 */ 6731 bool netdev_has_upper_dev(struct net_device *dev, 6732 struct net_device *upper_dev) 6733 { 6734 struct netdev_nested_priv priv = { 6735 .data = (void *)upper_dev, 6736 }; 6737 6738 ASSERT_RTNL(); 6739 6740 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6741 &priv); 6742 } 6743 EXPORT_SYMBOL(netdev_has_upper_dev); 6744 6745 /** 6746 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6747 * @dev: device 6748 * @upper_dev: upper device to check 6749 * 6750 * Find out if a device is linked to specified upper device and return true 6751 * in case it is. Note that this checks the entire upper device chain. 6752 * The caller must hold rcu lock. 6753 */ 6754 6755 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6756 struct net_device *upper_dev) 6757 { 6758 struct netdev_nested_priv priv = { 6759 .data = (void *)upper_dev, 6760 }; 6761 6762 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6763 &priv); 6764 } 6765 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6766 6767 /** 6768 * netdev_has_any_upper_dev - Check if device is linked to some device 6769 * @dev: device 6770 * 6771 * Find out if a device is linked to an upper device and return true in case 6772 * it is. The caller must hold the RTNL lock. 6773 */ 6774 bool netdev_has_any_upper_dev(struct net_device *dev) 6775 { 6776 ASSERT_RTNL(); 6777 6778 return !list_empty(&dev->adj_list.upper); 6779 } 6780 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6781 6782 /** 6783 * netdev_master_upper_dev_get - Get master upper device 6784 * @dev: device 6785 * 6786 * Find a master upper device and return pointer to it or NULL in case 6787 * it's not there. The caller must hold the RTNL lock. 6788 */ 6789 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6790 { 6791 struct netdev_adjacent *upper; 6792 6793 ASSERT_RTNL(); 6794 6795 if (list_empty(&dev->adj_list.upper)) 6796 return NULL; 6797 6798 upper = list_first_entry(&dev->adj_list.upper, 6799 struct netdev_adjacent, list); 6800 if (likely(upper->master)) 6801 return upper->dev; 6802 return NULL; 6803 } 6804 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6805 6806 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6807 { 6808 struct netdev_adjacent *upper; 6809 6810 ASSERT_RTNL(); 6811 6812 if (list_empty(&dev->adj_list.upper)) 6813 return NULL; 6814 6815 upper = list_first_entry(&dev->adj_list.upper, 6816 struct netdev_adjacent, list); 6817 if (likely(upper->master) && !upper->ignore) 6818 return upper->dev; 6819 return NULL; 6820 } 6821 6822 /** 6823 * netdev_has_any_lower_dev - Check if device is linked to some device 6824 * @dev: device 6825 * 6826 * Find out if a device is linked to a lower device and return true in case 6827 * it is. The caller must hold the RTNL lock. 6828 */ 6829 static bool netdev_has_any_lower_dev(struct net_device *dev) 6830 { 6831 ASSERT_RTNL(); 6832 6833 return !list_empty(&dev->adj_list.lower); 6834 } 6835 6836 void *netdev_adjacent_get_private(struct list_head *adj_list) 6837 { 6838 struct netdev_adjacent *adj; 6839 6840 adj = list_entry(adj_list, struct netdev_adjacent, list); 6841 6842 return adj->private; 6843 } 6844 EXPORT_SYMBOL(netdev_adjacent_get_private); 6845 6846 /** 6847 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6848 * @dev: device 6849 * @iter: list_head ** of the current position 6850 * 6851 * Gets the next device from the dev's upper list, starting from iter 6852 * position. The caller must hold RCU read lock. 6853 */ 6854 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6855 struct list_head **iter) 6856 { 6857 struct netdev_adjacent *upper; 6858 6859 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6860 6861 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6862 6863 if (&upper->list == &dev->adj_list.upper) 6864 return NULL; 6865 6866 *iter = &upper->list; 6867 6868 return upper->dev; 6869 } 6870 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6871 6872 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6873 struct list_head **iter, 6874 bool *ignore) 6875 { 6876 struct netdev_adjacent *upper; 6877 6878 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6879 6880 if (&upper->list == &dev->adj_list.upper) 6881 return NULL; 6882 6883 *iter = &upper->list; 6884 *ignore = upper->ignore; 6885 6886 return upper->dev; 6887 } 6888 6889 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6890 struct list_head **iter) 6891 { 6892 struct netdev_adjacent *upper; 6893 6894 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6895 6896 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6897 6898 if (&upper->list == &dev->adj_list.upper) 6899 return NULL; 6900 6901 *iter = &upper->list; 6902 6903 return upper->dev; 6904 } 6905 6906 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6907 int (*fn)(struct net_device *dev, 6908 struct netdev_nested_priv *priv), 6909 struct netdev_nested_priv *priv) 6910 { 6911 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6912 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6913 int ret, cur = 0; 6914 bool ignore; 6915 6916 now = dev; 6917 iter = &dev->adj_list.upper; 6918 6919 while (1) { 6920 if (now != dev) { 6921 ret = fn(now, priv); 6922 if (ret) 6923 return ret; 6924 } 6925 6926 next = NULL; 6927 while (1) { 6928 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6929 if (!udev) 6930 break; 6931 if (ignore) 6932 continue; 6933 6934 next = udev; 6935 niter = &udev->adj_list.upper; 6936 dev_stack[cur] = now; 6937 iter_stack[cur++] = iter; 6938 break; 6939 } 6940 6941 if (!next) { 6942 if (!cur) 6943 return 0; 6944 next = dev_stack[--cur]; 6945 niter = iter_stack[cur]; 6946 } 6947 6948 now = next; 6949 iter = niter; 6950 } 6951 6952 return 0; 6953 } 6954 6955 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6956 int (*fn)(struct net_device *dev, 6957 struct netdev_nested_priv *priv), 6958 struct netdev_nested_priv *priv) 6959 { 6960 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6961 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6962 int ret, cur = 0; 6963 6964 now = dev; 6965 iter = &dev->adj_list.upper; 6966 6967 while (1) { 6968 if (now != dev) { 6969 ret = fn(now, priv); 6970 if (ret) 6971 return ret; 6972 } 6973 6974 next = NULL; 6975 while (1) { 6976 udev = netdev_next_upper_dev_rcu(now, &iter); 6977 if (!udev) 6978 break; 6979 6980 next = udev; 6981 niter = &udev->adj_list.upper; 6982 dev_stack[cur] = now; 6983 iter_stack[cur++] = iter; 6984 break; 6985 } 6986 6987 if (!next) { 6988 if (!cur) 6989 return 0; 6990 next = dev_stack[--cur]; 6991 niter = iter_stack[cur]; 6992 } 6993 6994 now = next; 6995 iter = niter; 6996 } 6997 6998 return 0; 6999 } 7000 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7001 7002 static bool __netdev_has_upper_dev(struct net_device *dev, 7003 struct net_device *upper_dev) 7004 { 7005 struct netdev_nested_priv priv = { 7006 .flags = 0, 7007 .data = (void *)upper_dev, 7008 }; 7009 7010 ASSERT_RTNL(); 7011 7012 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7013 &priv); 7014 } 7015 7016 /** 7017 * netdev_lower_get_next_private - Get the next ->private from the 7018 * lower neighbour list 7019 * @dev: device 7020 * @iter: list_head ** of the current position 7021 * 7022 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7023 * list, starting from iter position. The caller must hold either hold the 7024 * RTNL lock or its own locking that guarantees that the neighbour lower 7025 * list will remain unchanged. 7026 */ 7027 void *netdev_lower_get_next_private(struct net_device *dev, 7028 struct list_head **iter) 7029 { 7030 struct netdev_adjacent *lower; 7031 7032 lower = list_entry(*iter, struct netdev_adjacent, list); 7033 7034 if (&lower->list == &dev->adj_list.lower) 7035 return NULL; 7036 7037 *iter = lower->list.next; 7038 7039 return lower->private; 7040 } 7041 EXPORT_SYMBOL(netdev_lower_get_next_private); 7042 7043 /** 7044 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7045 * lower neighbour list, RCU 7046 * variant 7047 * @dev: device 7048 * @iter: list_head ** of the current position 7049 * 7050 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7051 * list, starting from iter position. The caller must hold RCU read lock. 7052 */ 7053 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7054 struct list_head **iter) 7055 { 7056 struct netdev_adjacent *lower; 7057 7058 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7059 7060 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7061 7062 if (&lower->list == &dev->adj_list.lower) 7063 return NULL; 7064 7065 *iter = &lower->list; 7066 7067 return lower->private; 7068 } 7069 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7070 7071 /** 7072 * netdev_lower_get_next - Get the next device from the lower neighbour 7073 * list 7074 * @dev: device 7075 * @iter: list_head ** of the current position 7076 * 7077 * Gets the next netdev_adjacent from the dev's lower neighbour 7078 * list, starting from iter position. The caller must hold RTNL lock or 7079 * its own locking that guarantees that the neighbour lower 7080 * list will remain unchanged. 7081 */ 7082 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7083 { 7084 struct netdev_adjacent *lower; 7085 7086 lower = list_entry(*iter, struct netdev_adjacent, list); 7087 7088 if (&lower->list == &dev->adj_list.lower) 7089 return NULL; 7090 7091 *iter = lower->list.next; 7092 7093 return lower->dev; 7094 } 7095 EXPORT_SYMBOL(netdev_lower_get_next); 7096 7097 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7098 struct list_head **iter) 7099 { 7100 struct netdev_adjacent *lower; 7101 7102 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7103 7104 if (&lower->list == &dev->adj_list.lower) 7105 return NULL; 7106 7107 *iter = &lower->list; 7108 7109 return lower->dev; 7110 } 7111 7112 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7113 struct list_head **iter, 7114 bool *ignore) 7115 { 7116 struct netdev_adjacent *lower; 7117 7118 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7119 7120 if (&lower->list == &dev->adj_list.lower) 7121 return NULL; 7122 7123 *iter = &lower->list; 7124 *ignore = lower->ignore; 7125 7126 return lower->dev; 7127 } 7128 7129 int netdev_walk_all_lower_dev(struct net_device *dev, 7130 int (*fn)(struct net_device *dev, 7131 struct netdev_nested_priv *priv), 7132 struct netdev_nested_priv *priv) 7133 { 7134 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7135 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7136 int ret, cur = 0; 7137 7138 now = dev; 7139 iter = &dev->adj_list.lower; 7140 7141 while (1) { 7142 if (now != dev) { 7143 ret = fn(now, priv); 7144 if (ret) 7145 return ret; 7146 } 7147 7148 next = NULL; 7149 while (1) { 7150 ldev = netdev_next_lower_dev(now, &iter); 7151 if (!ldev) 7152 break; 7153 7154 next = ldev; 7155 niter = &ldev->adj_list.lower; 7156 dev_stack[cur] = now; 7157 iter_stack[cur++] = iter; 7158 break; 7159 } 7160 7161 if (!next) { 7162 if (!cur) 7163 return 0; 7164 next = dev_stack[--cur]; 7165 niter = iter_stack[cur]; 7166 } 7167 7168 now = next; 7169 iter = niter; 7170 } 7171 7172 return 0; 7173 } 7174 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7175 7176 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7177 int (*fn)(struct net_device *dev, 7178 struct netdev_nested_priv *priv), 7179 struct netdev_nested_priv *priv) 7180 { 7181 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7182 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7183 int ret, cur = 0; 7184 bool ignore; 7185 7186 now = dev; 7187 iter = &dev->adj_list.lower; 7188 7189 while (1) { 7190 if (now != dev) { 7191 ret = fn(now, priv); 7192 if (ret) 7193 return ret; 7194 } 7195 7196 next = NULL; 7197 while (1) { 7198 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7199 if (!ldev) 7200 break; 7201 if (ignore) 7202 continue; 7203 7204 next = ldev; 7205 niter = &ldev->adj_list.lower; 7206 dev_stack[cur] = now; 7207 iter_stack[cur++] = iter; 7208 break; 7209 } 7210 7211 if (!next) { 7212 if (!cur) 7213 return 0; 7214 next = dev_stack[--cur]; 7215 niter = iter_stack[cur]; 7216 } 7217 7218 now = next; 7219 iter = niter; 7220 } 7221 7222 return 0; 7223 } 7224 7225 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7226 struct list_head **iter) 7227 { 7228 struct netdev_adjacent *lower; 7229 7230 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7231 if (&lower->list == &dev->adj_list.lower) 7232 return NULL; 7233 7234 *iter = &lower->list; 7235 7236 return lower->dev; 7237 } 7238 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7239 7240 static u8 __netdev_upper_depth(struct net_device *dev) 7241 { 7242 struct net_device *udev; 7243 struct list_head *iter; 7244 u8 max_depth = 0; 7245 bool ignore; 7246 7247 for (iter = &dev->adj_list.upper, 7248 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7249 udev; 7250 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7251 if (ignore) 7252 continue; 7253 if (max_depth < udev->upper_level) 7254 max_depth = udev->upper_level; 7255 } 7256 7257 return max_depth; 7258 } 7259 7260 static u8 __netdev_lower_depth(struct net_device *dev) 7261 { 7262 struct net_device *ldev; 7263 struct list_head *iter; 7264 u8 max_depth = 0; 7265 bool ignore; 7266 7267 for (iter = &dev->adj_list.lower, 7268 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7269 ldev; 7270 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7271 if (ignore) 7272 continue; 7273 if (max_depth < ldev->lower_level) 7274 max_depth = ldev->lower_level; 7275 } 7276 7277 return max_depth; 7278 } 7279 7280 static int __netdev_update_upper_level(struct net_device *dev, 7281 struct netdev_nested_priv *__unused) 7282 { 7283 dev->upper_level = __netdev_upper_depth(dev) + 1; 7284 return 0; 7285 } 7286 7287 #ifdef CONFIG_LOCKDEP 7288 static LIST_HEAD(net_unlink_list); 7289 7290 static void net_unlink_todo(struct net_device *dev) 7291 { 7292 if (list_empty(&dev->unlink_list)) 7293 list_add_tail(&dev->unlink_list, &net_unlink_list); 7294 } 7295 #endif 7296 7297 static int __netdev_update_lower_level(struct net_device *dev, 7298 struct netdev_nested_priv *priv) 7299 { 7300 dev->lower_level = __netdev_lower_depth(dev) + 1; 7301 7302 #ifdef CONFIG_LOCKDEP 7303 if (!priv) 7304 return 0; 7305 7306 if (priv->flags & NESTED_SYNC_IMM) 7307 dev->nested_level = dev->lower_level - 1; 7308 if (priv->flags & NESTED_SYNC_TODO) 7309 net_unlink_todo(dev); 7310 #endif 7311 return 0; 7312 } 7313 7314 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7315 int (*fn)(struct net_device *dev, 7316 struct netdev_nested_priv *priv), 7317 struct netdev_nested_priv *priv) 7318 { 7319 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7320 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7321 int ret, cur = 0; 7322 7323 now = dev; 7324 iter = &dev->adj_list.lower; 7325 7326 while (1) { 7327 if (now != dev) { 7328 ret = fn(now, priv); 7329 if (ret) 7330 return ret; 7331 } 7332 7333 next = NULL; 7334 while (1) { 7335 ldev = netdev_next_lower_dev_rcu(now, &iter); 7336 if (!ldev) 7337 break; 7338 7339 next = ldev; 7340 niter = &ldev->adj_list.lower; 7341 dev_stack[cur] = now; 7342 iter_stack[cur++] = iter; 7343 break; 7344 } 7345 7346 if (!next) { 7347 if (!cur) 7348 return 0; 7349 next = dev_stack[--cur]; 7350 niter = iter_stack[cur]; 7351 } 7352 7353 now = next; 7354 iter = niter; 7355 } 7356 7357 return 0; 7358 } 7359 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7360 7361 /** 7362 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7363 * lower neighbour list, RCU 7364 * variant 7365 * @dev: device 7366 * 7367 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7368 * list. The caller must hold RCU read lock. 7369 */ 7370 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7371 { 7372 struct netdev_adjacent *lower; 7373 7374 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7375 struct netdev_adjacent, list); 7376 if (lower) 7377 return lower->private; 7378 return NULL; 7379 } 7380 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7381 7382 /** 7383 * netdev_master_upper_dev_get_rcu - Get master upper device 7384 * @dev: device 7385 * 7386 * Find a master upper device and return pointer to it or NULL in case 7387 * it's not there. The caller must hold the RCU read lock. 7388 */ 7389 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7390 { 7391 struct netdev_adjacent *upper; 7392 7393 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7394 struct netdev_adjacent, list); 7395 if (upper && likely(upper->master)) 7396 return upper->dev; 7397 return NULL; 7398 } 7399 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7400 7401 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7402 struct net_device *adj_dev, 7403 struct list_head *dev_list) 7404 { 7405 char linkname[IFNAMSIZ+7]; 7406 7407 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7408 "upper_%s" : "lower_%s", adj_dev->name); 7409 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7410 linkname); 7411 } 7412 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7413 char *name, 7414 struct list_head *dev_list) 7415 { 7416 char linkname[IFNAMSIZ+7]; 7417 7418 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7419 "upper_%s" : "lower_%s", name); 7420 sysfs_remove_link(&(dev->dev.kobj), linkname); 7421 } 7422 7423 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7424 struct net_device *adj_dev, 7425 struct list_head *dev_list) 7426 { 7427 return (dev_list == &dev->adj_list.upper || 7428 dev_list == &dev->adj_list.lower) && 7429 net_eq(dev_net(dev), dev_net(adj_dev)); 7430 } 7431 7432 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7433 struct net_device *adj_dev, 7434 struct list_head *dev_list, 7435 void *private, bool master) 7436 { 7437 struct netdev_adjacent *adj; 7438 int ret; 7439 7440 adj = __netdev_find_adj(adj_dev, dev_list); 7441 7442 if (adj) { 7443 adj->ref_nr += 1; 7444 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7445 dev->name, adj_dev->name, adj->ref_nr); 7446 7447 return 0; 7448 } 7449 7450 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7451 if (!adj) 7452 return -ENOMEM; 7453 7454 adj->dev = adj_dev; 7455 adj->master = master; 7456 adj->ref_nr = 1; 7457 adj->private = private; 7458 adj->ignore = false; 7459 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7460 7461 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7462 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7463 7464 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7465 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7466 if (ret) 7467 goto free_adj; 7468 } 7469 7470 /* Ensure that master link is always the first item in list. */ 7471 if (master) { 7472 ret = sysfs_create_link(&(dev->dev.kobj), 7473 &(adj_dev->dev.kobj), "master"); 7474 if (ret) 7475 goto remove_symlinks; 7476 7477 list_add_rcu(&adj->list, dev_list); 7478 } else { 7479 list_add_tail_rcu(&adj->list, dev_list); 7480 } 7481 7482 return 0; 7483 7484 remove_symlinks: 7485 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7486 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7487 free_adj: 7488 netdev_put(adj_dev, &adj->dev_tracker); 7489 kfree(adj); 7490 7491 return ret; 7492 } 7493 7494 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7495 struct net_device *adj_dev, 7496 u16 ref_nr, 7497 struct list_head *dev_list) 7498 { 7499 struct netdev_adjacent *adj; 7500 7501 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7502 dev->name, adj_dev->name, ref_nr); 7503 7504 adj = __netdev_find_adj(adj_dev, dev_list); 7505 7506 if (!adj) { 7507 pr_err("Adjacency does not exist for device %s from %s\n", 7508 dev->name, adj_dev->name); 7509 WARN_ON(1); 7510 return; 7511 } 7512 7513 if (adj->ref_nr > ref_nr) { 7514 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7515 dev->name, adj_dev->name, ref_nr, 7516 adj->ref_nr - ref_nr); 7517 adj->ref_nr -= ref_nr; 7518 return; 7519 } 7520 7521 if (adj->master) 7522 sysfs_remove_link(&(dev->dev.kobj), "master"); 7523 7524 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7525 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7526 7527 list_del_rcu(&adj->list); 7528 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7529 adj_dev->name, dev->name, adj_dev->name); 7530 netdev_put(adj_dev, &adj->dev_tracker); 7531 kfree_rcu(adj, rcu); 7532 } 7533 7534 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7535 struct net_device *upper_dev, 7536 struct list_head *up_list, 7537 struct list_head *down_list, 7538 void *private, bool master) 7539 { 7540 int ret; 7541 7542 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7543 private, master); 7544 if (ret) 7545 return ret; 7546 7547 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7548 private, false); 7549 if (ret) { 7550 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7551 return ret; 7552 } 7553 7554 return 0; 7555 } 7556 7557 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7558 struct net_device *upper_dev, 7559 u16 ref_nr, 7560 struct list_head *up_list, 7561 struct list_head *down_list) 7562 { 7563 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7564 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7565 } 7566 7567 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7568 struct net_device *upper_dev, 7569 void *private, bool master) 7570 { 7571 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7572 &dev->adj_list.upper, 7573 &upper_dev->adj_list.lower, 7574 private, master); 7575 } 7576 7577 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7578 struct net_device *upper_dev) 7579 { 7580 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7581 &dev->adj_list.upper, 7582 &upper_dev->adj_list.lower); 7583 } 7584 7585 static int __netdev_upper_dev_link(struct net_device *dev, 7586 struct net_device *upper_dev, bool master, 7587 void *upper_priv, void *upper_info, 7588 struct netdev_nested_priv *priv, 7589 struct netlink_ext_ack *extack) 7590 { 7591 struct netdev_notifier_changeupper_info changeupper_info = { 7592 .info = { 7593 .dev = dev, 7594 .extack = extack, 7595 }, 7596 .upper_dev = upper_dev, 7597 .master = master, 7598 .linking = true, 7599 .upper_info = upper_info, 7600 }; 7601 struct net_device *master_dev; 7602 int ret = 0; 7603 7604 ASSERT_RTNL(); 7605 7606 if (dev == upper_dev) 7607 return -EBUSY; 7608 7609 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7610 if (__netdev_has_upper_dev(upper_dev, dev)) 7611 return -EBUSY; 7612 7613 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7614 return -EMLINK; 7615 7616 if (!master) { 7617 if (__netdev_has_upper_dev(dev, upper_dev)) 7618 return -EEXIST; 7619 } else { 7620 master_dev = __netdev_master_upper_dev_get(dev); 7621 if (master_dev) 7622 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7623 } 7624 7625 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7626 &changeupper_info.info); 7627 ret = notifier_to_errno(ret); 7628 if (ret) 7629 return ret; 7630 7631 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7632 master); 7633 if (ret) 7634 return ret; 7635 7636 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7637 &changeupper_info.info); 7638 ret = notifier_to_errno(ret); 7639 if (ret) 7640 goto rollback; 7641 7642 __netdev_update_upper_level(dev, NULL); 7643 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7644 7645 __netdev_update_lower_level(upper_dev, priv); 7646 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7647 priv); 7648 7649 return 0; 7650 7651 rollback: 7652 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7653 7654 return ret; 7655 } 7656 7657 /** 7658 * netdev_upper_dev_link - Add a link to the upper device 7659 * @dev: device 7660 * @upper_dev: new upper device 7661 * @extack: netlink extended ack 7662 * 7663 * Adds a link to device which is upper to this one. The caller must hold 7664 * the RTNL lock. On a failure a negative errno code is returned. 7665 * On success the reference counts are adjusted and the function 7666 * returns zero. 7667 */ 7668 int netdev_upper_dev_link(struct net_device *dev, 7669 struct net_device *upper_dev, 7670 struct netlink_ext_ack *extack) 7671 { 7672 struct netdev_nested_priv priv = { 7673 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7674 .data = NULL, 7675 }; 7676 7677 return __netdev_upper_dev_link(dev, upper_dev, false, 7678 NULL, NULL, &priv, extack); 7679 } 7680 EXPORT_SYMBOL(netdev_upper_dev_link); 7681 7682 /** 7683 * netdev_master_upper_dev_link - Add a master link to the upper device 7684 * @dev: device 7685 * @upper_dev: new upper device 7686 * @upper_priv: upper device private 7687 * @upper_info: upper info to be passed down via notifier 7688 * @extack: netlink extended ack 7689 * 7690 * Adds a link to device which is upper to this one. In this case, only 7691 * one master upper device can be linked, although other non-master devices 7692 * might be linked as well. The caller must hold the RTNL lock. 7693 * On a failure a negative errno code is returned. On success the reference 7694 * counts are adjusted and the function returns zero. 7695 */ 7696 int netdev_master_upper_dev_link(struct net_device *dev, 7697 struct net_device *upper_dev, 7698 void *upper_priv, void *upper_info, 7699 struct netlink_ext_ack *extack) 7700 { 7701 struct netdev_nested_priv priv = { 7702 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7703 .data = NULL, 7704 }; 7705 7706 return __netdev_upper_dev_link(dev, upper_dev, true, 7707 upper_priv, upper_info, &priv, extack); 7708 } 7709 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7710 7711 static void __netdev_upper_dev_unlink(struct net_device *dev, 7712 struct net_device *upper_dev, 7713 struct netdev_nested_priv *priv) 7714 { 7715 struct netdev_notifier_changeupper_info changeupper_info = { 7716 .info = { 7717 .dev = dev, 7718 }, 7719 .upper_dev = upper_dev, 7720 .linking = false, 7721 }; 7722 7723 ASSERT_RTNL(); 7724 7725 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7726 7727 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7728 &changeupper_info.info); 7729 7730 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7731 7732 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7733 &changeupper_info.info); 7734 7735 __netdev_update_upper_level(dev, NULL); 7736 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7737 7738 __netdev_update_lower_level(upper_dev, priv); 7739 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7740 priv); 7741 } 7742 7743 /** 7744 * netdev_upper_dev_unlink - Removes a link to upper device 7745 * @dev: device 7746 * @upper_dev: new upper device 7747 * 7748 * Removes a link to device which is upper to this one. The caller must hold 7749 * the RTNL lock. 7750 */ 7751 void netdev_upper_dev_unlink(struct net_device *dev, 7752 struct net_device *upper_dev) 7753 { 7754 struct netdev_nested_priv priv = { 7755 .flags = NESTED_SYNC_TODO, 7756 .data = NULL, 7757 }; 7758 7759 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7760 } 7761 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7762 7763 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7764 struct net_device *lower_dev, 7765 bool val) 7766 { 7767 struct netdev_adjacent *adj; 7768 7769 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7770 if (adj) 7771 adj->ignore = val; 7772 7773 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7774 if (adj) 7775 adj->ignore = val; 7776 } 7777 7778 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7779 struct net_device *lower_dev) 7780 { 7781 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7782 } 7783 7784 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7785 struct net_device *lower_dev) 7786 { 7787 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7788 } 7789 7790 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7791 struct net_device *new_dev, 7792 struct net_device *dev, 7793 struct netlink_ext_ack *extack) 7794 { 7795 struct netdev_nested_priv priv = { 7796 .flags = 0, 7797 .data = NULL, 7798 }; 7799 int err; 7800 7801 if (!new_dev) 7802 return 0; 7803 7804 if (old_dev && new_dev != old_dev) 7805 netdev_adjacent_dev_disable(dev, old_dev); 7806 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7807 extack); 7808 if (err) { 7809 if (old_dev && new_dev != old_dev) 7810 netdev_adjacent_dev_enable(dev, old_dev); 7811 return err; 7812 } 7813 7814 return 0; 7815 } 7816 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7817 7818 void netdev_adjacent_change_commit(struct net_device *old_dev, 7819 struct net_device *new_dev, 7820 struct net_device *dev) 7821 { 7822 struct netdev_nested_priv priv = { 7823 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7824 .data = NULL, 7825 }; 7826 7827 if (!new_dev || !old_dev) 7828 return; 7829 7830 if (new_dev == old_dev) 7831 return; 7832 7833 netdev_adjacent_dev_enable(dev, old_dev); 7834 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7835 } 7836 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7837 7838 void netdev_adjacent_change_abort(struct net_device *old_dev, 7839 struct net_device *new_dev, 7840 struct net_device *dev) 7841 { 7842 struct netdev_nested_priv priv = { 7843 .flags = 0, 7844 .data = NULL, 7845 }; 7846 7847 if (!new_dev) 7848 return; 7849 7850 if (old_dev && new_dev != old_dev) 7851 netdev_adjacent_dev_enable(dev, old_dev); 7852 7853 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7854 } 7855 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7856 7857 /** 7858 * netdev_bonding_info_change - Dispatch event about slave change 7859 * @dev: device 7860 * @bonding_info: info to dispatch 7861 * 7862 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7863 * The caller must hold the RTNL lock. 7864 */ 7865 void netdev_bonding_info_change(struct net_device *dev, 7866 struct netdev_bonding_info *bonding_info) 7867 { 7868 struct netdev_notifier_bonding_info info = { 7869 .info.dev = dev, 7870 }; 7871 7872 memcpy(&info.bonding_info, bonding_info, 7873 sizeof(struct netdev_bonding_info)); 7874 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7875 &info.info); 7876 } 7877 EXPORT_SYMBOL(netdev_bonding_info_change); 7878 7879 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7880 struct netlink_ext_ack *extack) 7881 { 7882 struct netdev_notifier_offload_xstats_info info = { 7883 .info.dev = dev, 7884 .info.extack = extack, 7885 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7886 }; 7887 int err; 7888 int rc; 7889 7890 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7891 GFP_KERNEL); 7892 if (!dev->offload_xstats_l3) 7893 return -ENOMEM; 7894 7895 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7896 NETDEV_OFFLOAD_XSTATS_DISABLE, 7897 &info.info); 7898 err = notifier_to_errno(rc); 7899 if (err) 7900 goto free_stats; 7901 7902 return 0; 7903 7904 free_stats: 7905 kfree(dev->offload_xstats_l3); 7906 dev->offload_xstats_l3 = NULL; 7907 return err; 7908 } 7909 7910 int netdev_offload_xstats_enable(struct net_device *dev, 7911 enum netdev_offload_xstats_type type, 7912 struct netlink_ext_ack *extack) 7913 { 7914 ASSERT_RTNL(); 7915 7916 if (netdev_offload_xstats_enabled(dev, type)) 7917 return -EALREADY; 7918 7919 switch (type) { 7920 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7921 return netdev_offload_xstats_enable_l3(dev, extack); 7922 } 7923 7924 WARN_ON(1); 7925 return -EINVAL; 7926 } 7927 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7928 7929 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7930 { 7931 struct netdev_notifier_offload_xstats_info info = { 7932 .info.dev = dev, 7933 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7934 }; 7935 7936 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7937 &info.info); 7938 kfree(dev->offload_xstats_l3); 7939 dev->offload_xstats_l3 = NULL; 7940 } 7941 7942 int netdev_offload_xstats_disable(struct net_device *dev, 7943 enum netdev_offload_xstats_type type) 7944 { 7945 ASSERT_RTNL(); 7946 7947 if (!netdev_offload_xstats_enabled(dev, type)) 7948 return -EALREADY; 7949 7950 switch (type) { 7951 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7952 netdev_offload_xstats_disable_l3(dev); 7953 return 0; 7954 } 7955 7956 WARN_ON(1); 7957 return -EINVAL; 7958 } 7959 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7960 7961 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7962 { 7963 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7964 } 7965 7966 static struct rtnl_hw_stats64 * 7967 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7968 enum netdev_offload_xstats_type type) 7969 { 7970 switch (type) { 7971 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7972 return dev->offload_xstats_l3; 7973 } 7974 7975 WARN_ON(1); 7976 return NULL; 7977 } 7978 7979 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7980 enum netdev_offload_xstats_type type) 7981 { 7982 ASSERT_RTNL(); 7983 7984 return netdev_offload_xstats_get_ptr(dev, type); 7985 } 7986 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7987 7988 struct netdev_notifier_offload_xstats_ru { 7989 bool used; 7990 }; 7991 7992 struct netdev_notifier_offload_xstats_rd { 7993 struct rtnl_hw_stats64 stats; 7994 bool used; 7995 }; 7996 7997 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 7998 const struct rtnl_hw_stats64 *src) 7999 { 8000 dest->rx_packets += src->rx_packets; 8001 dest->tx_packets += src->tx_packets; 8002 dest->rx_bytes += src->rx_bytes; 8003 dest->tx_bytes += src->tx_bytes; 8004 dest->rx_errors += src->rx_errors; 8005 dest->tx_errors += src->tx_errors; 8006 dest->rx_dropped += src->rx_dropped; 8007 dest->tx_dropped += src->tx_dropped; 8008 dest->multicast += src->multicast; 8009 } 8010 8011 static int netdev_offload_xstats_get_used(struct net_device *dev, 8012 enum netdev_offload_xstats_type type, 8013 bool *p_used, 8014 struct netlink_ext_ack *extack) 8015 { 8016 struct netdev_notifier_offload_xstats_ru report_used = {}; 8017 struct netdev_notifier_offload_xstats_info info = { 8018 .info.dev = dev, 8019 .info.extack = extack, 8020 .type = type, 8021 .report_used = &report_used, 8022 }; 8023 int rc; 8024 8025 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8026 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8027 &info.info); 8028 *p_used = report_used.used; 8029 return notifier_to_errno(rc); 8030 } 8031 8032 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8033 enum netdev_offload_xstats_type type, 8034 struct rtnl_hw_stats64 *p_stats, 8035 bool *p_used, 8036 struct netlink_ext_ack *extack) 8037 { 8038 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8039 struct netdev_notifier_offload_xstats_info info = { 8040 .info.dev = dev, 8041 .info.extack = extack, 8042 .type = type, 8043 .report_delta = &report_delta, 8044 }; 8045 struct rtnl_hw_stats64 *stats; 8046 int rc; 8047 8048 stats = netdev_offload_xstats_get_ptr(dev, type); 8049 if (WARN_ON(!stats)) 8050 return -EINVAL; 8051 8052 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8053 &info.info); 8054 8055 /* Cache whatever we got, even if there was an error, otherwise the 8056 * successful stats retrievals would get lost. 8057 */ 8058 netdev_hw_stats64_add(stats, &report_delta.stats); 8059 8060 if (p_stats) 8061 *p_stats = *stats; 8062 *p_used = report_delta.used; 8063 8064 return notifier_to_errno(rc); 8065 } 8066 8067 int netdev_offload_xstats_get(struct net_device *dev, 8068 enum netdev_offload_xstats_type type, 8069 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8070 struct netlink_ext_ack *extack) 8071 { 8072 ASSERT_RTNL(); 8073 8074 if (p_stats) 8075 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8076 p_used, extack); 8077 else 8078 return netdev_offload_xstats_get_used(dev, type, p_used, 8079 extack); 8080 } 8081 EXPORT_SYMBOL(netdev_offload_xstats_get); 8082 8083 void 8084 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8085 const struct rtnl_hw_stats64 *stats) 8086 { 8087 report_delta->used = true; 8088 netdev_hw_stats64_add(&report_delta->stats, stats); 8089 } 8090 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8091 8092 void 8093 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8094 { 8095 report_used->used = true; 8096 } 8097 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8098 8099 void netdev_offload_xstats_push_delta(struct net_device *dev, 8100 enum netdev_offload_xstats_type type, 8101 const struct rtnl_hw_stats64 *p_stats) 8102 { 8103 struct rtnl_hw_stats64 *stats; 8104 8105 ASSERT_RTNL(); 8106 8107 stats = netdev_offload_xstats_get_ptr(dev, type); 8108 if (WARN_ON(!stats)) 8109 return; 8110 8111 netdev_hw_stats64_add(stats, p_stats); 8112 } 8113 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8114 8115 /** 8116 * netdev_get_xmit_slave - Get the xmit slave of master device 8117 * @dev: device 8118 * @skb: The packet 8119 * @all_slaves: assume all the slaves are active 8120 * 8121 * The reference counters are not incremented so the caller must be 8122 * careful with locks. The caller must hold RCU lock. 8123 * %NULL is returned if no slave is found. 8124 */ 8125 8126 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8127 struct sk_buff *skb, 8128 bool all_slaves) 8129 { 8130 const struct net_device_ops *ops = dev->netdev_ops; 8131 8132 if (!ops->ndo_get_xmit_slave) 8133 return NULL; 8134 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8135 } 8136 EXPORT_SYMBOL(netdev_get_xmit_slave); 8137 8138 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8139 struct sock *sk) 8140 { 8141 const struct net_device_ops *ops = dev->netdev_ops; 8142 8143 if (!ops->ndo_sk_get_lower_dev) 8144 return NULL; 8145 return ops->ndo_sk_get_lower_dev(dev, sk); 8146 } 8147 8148 /** 8149 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8150 * @dev: device 8151 * @sk: the socket 8152 * 8153 * %NULL is returned if no lower device is found. 8154 */ 8155 8156 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8157 struct sock *sk) 8158 { 8159 struct net_device *lower; 8160 8161 lower = netdev_sk_get_lower_dev(dev, sk); 8162 while (lower) { 8163 dev = lower; 8164 lower = netdev_sk_get_lower_dev(dev, sk); 8165 } 8166 8167 return dev; 8168 } 8169 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8170 8171 static void netdev_adjacent_add_links(struct net_device *dev) 8172 { 8173 struct netdev_adjacent *iter; 8174 8175 struct net *net = dev_net(dev); 8176 8177 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8178 if (!net_eq(net, dev_net(iter->dev))) 8179 continue; 8180 netdev_adjacent_sysfs_add(iter->dev, dev, 8181 &iter->dev->adj_list.lower); 8182 netdev_adjacent_sysfs_add(dev, iter->dev, 8183 &dev->adj_list.upper); 8184 } 8185 8186 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8187 if (!net_eq(net, dev_net(iter->dev))) 8188 continue; 8189 netdev_adjacent_sysfs_add(iter->dev, dev, 8190 &iter->dev->adj_list.upper); 8191 netdev_adjacent_sysfs_add(dev, iter->dev, 8192 &dev->adj_list.lower); 8193 } 8194 } 8195 8196 static void netdev_adjacent_del_links(struct net_device *dev) 8197 { 8198 struct netdev_adjacent *iter; 8199 8200 struct net *net = dev_net(dev); 8201 8202 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8203 if (!net_eq(net, dev_net(iter->dev))) 8204 continue; 8205 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8206 &iter->dev->adj_list.lower); 8207 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8208 &dev->adj_list.upper); 8209 } 8210 8211 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8212 if (!net_eq(net, dev_net(iter->dev))) 8213 continue; 8214 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8215 &iter->dev->adj_list.upper); 8216 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8217 &dev->adj_list.lower); 8218 } 8219 } 8220 8221 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8222 { 8223 struct netdev_adjacent *iter; 8224 8225 struct net *net = dev_net(dev); 8226 8227 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8228 if (!net_eq(net, dev_net(iter->dev))) 8229 continue; 8230 netdev_adjacent_sysfs_del(iter->dev, oldname, 8231 &iter->dev->adj_list.lower); 8232 netdev_adjacent_sysfs_add(iter->dev, dev, 8233 &iter->dev->adj_list.lower); 8234 } 8235 8236 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8237 if (!net_eq(net, dev_net(iter->dev))) 8238 continue; 8239 netdev_adjacent_sysfs_del(iter->dev, oldname, 8240 &iter->dev->adj_list.upper); 8241 netdev_adjacent_sysfs_add(iter->dev, dev, 8242 &iter->dev->adj_list.upper); 8243 } 8244 } 8245 8246 void *netdev_lower_dev_get_private(struct net_device *dev, 8247 struct net_device *lower_dev) 8248 { 8249 struct netdev_adjacent *lower; 8250 8251 if (!lower_dev) 8252 return NULL; 8253 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8254 if (!lower) 8255 return NULL; 8256 8257 return lower->private; 8258 } 8259 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8260 8261 8262 /** 8263 * netdev_lower_state_changed - Dispatch event about lower device state change 8264 * @lower_dev: device 8265 * @lower_state_info: state to dispatch 8266 * 8267 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8268 * The caller must hold the RTNL lock. 8269 */ 8270 void netdev_lower_state_changed(struct net_device *lower_dev, 8271 void *lower_state_info) 8272 { 8273 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8274 .info.dev = lower_dev, 8275 }; 8276 8277 ASSERT_RTNL(); 8278 changelowerstate_info.lower_state_info = lower_state_info; 8279 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8280 &changelowerstate_info.info); 8281 } 8282 EXPORT_SYMBOL(netdev_lower_state_changed); 8283 8284 static void dev_change_rx_flags(struct net_device *dev, int flags) 8285 { 8286 const struct net_device_ops *ops = dev->netdev_ops; 8287 8288 if (ops->ndo_change_rx_flags) 8289 ops->ndo_change_rx_flags(dev, flags); 8290 } 8291 8292 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8293 { 8294 unsigned int old_flags = dev->flags; 8295 kuid_t uid; 8296 kgid_t gid; 8297 8298 ASSERT_RTNL(); 8299 8300 dev->flags |= IFF_PROMISC; 8301 dev->promiscuity += inc; 8302 if (dev->promiscuity == 0) { 8303 /* 8304 * Avoid overflow. 8305 * If inc causes overflow, untouch promisc and return error. 8306 */ 8307 if (inc < 0) 8308 dev->flags &= ~IFF_PROMISC; 8309 else { 8310 dev->promiscuity -= inc; 8311 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8312 return -EOVERFLOW; 8313 } 8314 } 8315 if (dev->flags != old_flags) { 8316 netdev_info(dev, "%s promiscuous mode\n", 8317 dev->flags & IFF_PROMISC ? "entered" : "left"); 8318 if (audit_enabled) { 8319 current_uid_gid(&uid, &gid); 8320 audit_log(audit_context(), GFP_ATOMIC, 8321 AUDIT_ANOM_PROMISCUOUS, 8322 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8323 dev->name, (dev->flags & IFF_PROMISC), 8324 (old_flags & IFF_PROMISC), 8325 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8326 from_kuid(&init_user_ns, uid), 8327 from_kgid(&init_user_ns, gid), 8328 audit_get_sessionid(current)); 8329 } 8330 8331 dev_change_rx_flags(dev, IFF_PROMISC); 8332 } 8333 if (notify) 8334 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8335 return 0; 8336 } 8337 8338 /** 8339 * dev_set_promiscuity - update promiscuity count on a device 8340 * @dev: device 8341 * @inc: modifier 8342 * 8343 * Add or remove promiscuity from a device. While the count in the device 8344 * remains above zero the interface remains promiscuous. Once it hits zero 8345 * the device reverts back to normal filtering operation. A negative inc 8346 * value is used to drop promiscuity on the device. 8347 * Return 0 if successful or a negative errno code on error. 8348 */ 8349 int dev_set_promiscuity(struct net_device *dev, int inc) 8350 { 8351 unsigned int old_flags = dev->flags; 8352 int err; 8353 8354 err = __dev_set_promiscuity(dev, inc, true); 8355 if (err < 0) 8356 return err; 8357 if (dev->flags != old_flags) 8358 dev_set_rx_mode(dev); 8359 return err; 8360 } 8361 EXPORT_SYMBOL(dev_set_promiscuity); 8362 8363 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8364 { 8365 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8366 8367 ASSERT_RTNL(); 8368 8369 dev->flags |= IFF_ALLMULTI; 8370 dev->allmulti += inc; 8371 if (dev->allmulti == 0) { 8372 /* 8373 * Avoid overflow. 8374 * If inc causes overflow, untouch allmulti and return error. 8375 */ 8376 if (inc < 0) 8377 dev->flags &= ~IFF_ALLMULTI; 8378 else { 8379 dev->allmulti -= inc; 8380 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8381 return -EOVERFLOW; 8382 } 8383 } 8384 if (dev->flags ^ old_flags) { 8385 netdev_info(dev, "%s allmulticast mode\n", 8386 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8387 dev_change_rx_flags(dev, IFF_ALLMULTI); 8388 dev_set_rx_mode(dev); 8389 if (notify) 8390 __dev_notify_flags(dev, old_flags, 8391 dev->gflags ^ old_gflags, 0, NULL); 8392 } 8393 return 0; 8394 } 8395 8396 /** 8397 * dev_set_allmulti - update allmulti count on a device 8398 * @dev: device 8399 * @inc: modifier 8400 * 8401 * Add or remove reception of all multicast frames to a device. While the 8402 * count in the device remains above zero the interface remains listening 8403 * to all interfaces. Once it hits zero the device reverts back to normal 8404 * filtering operation. A negative @inc value is used to drop the counter 8405 * when releasing a resource needing all multicasts. 8406 * Return 0 if successful or a negative errno code on error. 8407 */ 8408 8409 int dev_set_allmulti(struct net_device *dev, int inc) 8410 { 8411 return __dev_set_allmulti(dev, inc, true); 8412 } 8413 EXPORT_SYMBOL(dev_set_allmulti); 8414 8415 /* 8416 * Upload unicast and multicast address lists to device and 8417 * configure RX filtering. When the device doesn't support unicast 8418 * filtering it is put in promiscuous mode while unicast addresses 8419 * are present. 8420 */ 8421 void __dev_set_rx_mode(struct net_device *dev) 8422 { 8423 const struct net_device_ops *ops = dev->netdev_ops; 8424 8425 /* dev_open will call this function so the list will stay sane. */ 8426 if (!(dev->flags&IFF_UP)) 8427 return; 8428 8429 if (!netif_device_present(dev)) 8430 return; 8431 8432 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8433 /* Unicast addresses changes may only happen under the rtnl, 8434 * therefore calling __dev_set_promiscuity here is safe. 8435 */ 8436 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8437 __dev_set_promiscuity(dev, 1, false); 8438 dev->uc_promisc = true; 8439 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8440 __dev_set_promiscuity(dev, -1, false); 8441 dev->uc_promisc = false; 8442 } 8443 } 8444 8445 if (ops->ndo_set_rx_mode) 8446 ops->ndo_set_rx_mode(dev); 8447 } 8448 8449 void dev_set_rx_mode(struct net_device *dev) 8450 { 8451 netif_addr_lock_bh(dev); 8452 __dev_set_rx_mode(dev); 8453 netif_addr_unlock_bh(dev); 8454 } 8455 8456 /** 8457 * dev_get_flags - get flags reported to userspace 8458 * @dev: device 8459 * 8460 * Get the combination of flag bits exported through APIs to userspace. 8461 */ 8462 unsigned int dev_get_flags(const struct net_device *dev) 8463 { 8464 unsigned int flags; 8465 8466 flags = (dev->flags & ~(IFF_PROMISC | 8467 IFF_ALLMULTI | 8468 IFF_RUNNING | 8469 IFF_LOWER_UP | 8470 IFF_DORMANT)) | 8471 (dev->gflags & (IFF_PROMISC | 8472 IFF_ALLMULTI)); 8473 8474 if (netif_running(dev)) { 8475 if (netif_oper_up(dev)) 8476 flags |= IFF_RUNNING; 8477 if (netif_carrier_ok(dev)) 8478 flags |= IFF_LOWER_UP; 8479 if (netif_dormant(dev)) 8480 flags |= IFF_DORMANT; 8481 } 8482 8483 return flags; 8484 } 8485 EXPORT_SYMBOL(dev_get_flags); 8486 8487 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8488 struct netlink_ext_ack *extack) 8489 { 8490 unsigned int old_flags = dev->flags; 8491 int ret; 8492 8493 ASSERT_RTNL(); 8494 8495 /* 8496 * Set the flags on our device. 8497 */ 8498 8499 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8500 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8501 IFF_AUTOMEDIA)) | 8502 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8503 IFF_ALLMULTI)); 8504 8505 /* 8506 * Load in the correct multicast list now the flags have changed. 8507 */ 8508 8509 if ((old_flags ^ flags) & IFF_MULTICAST) 8510 dev_change_rx_flags(dev, IFF_MULTICAST); 8511 8512 dev_set_rx_mode(dev); 8513 8514 /* 8515 * Have we downed the interface. We handle IFF_UP ourselves 8516 * according to user attempts to set it, rather than blindly 8517 * setting it. 8518 */ 8519 8520 ret = 0; 8521 if ((old_flags ^ flags) & IFF_UP) { 8522 if (old_flags & IFF_UP) 8523 __dev_close(dev); 8524 else 8525 ret = __dev_open(dev, extack); 8526 } 8527 8528 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8529 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8530 unsigned int old_flags = dev->flags; 8531 8532 dev->gflags ^= IFF_PROMISC; 8533 8534 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8535 if (dev->flags != old_flags) 8536 dev_set_rx_mode(dev); 8537 } 8538 8539 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8540 * is important. Some (broken) drivers set IFF_PROMISC, when 8541 * IFF_ALLMULTI is requested not asking us and not reporting. 8542 */ 8543 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8544 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8545 8546 dev->gflags ^= IFF_ALLMULTI; 8547 __dev_set_allmulti(dev, inc, false); 8548 } 8549 8550 return ret; 8551 } 8552 8553 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8554 unsigned int gchanges, u32 portid, 8555 const struct nlmsghdr *nlh) 8556 { 8557 unsigned int changes = dev->flags ^ old_flags; 8558 8559 if (gchanges) 8560 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8561 8562 if (changes & IFF_UP) { 8563 if (dev->flags & IFF_UP) 8564 call_netdevice_notifiers(NETDEV_UP, dev); 8565 else 8566 call_netdevice_notifiers(NETDEV_DOWN, dev); 8567 } 8568 8569 if (dev->flags & IFF_UP && 8570 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8571 struct netdev_notifier_change_info change_info = { 8572 .info = { 8573 .dev = dev, 8574 }, 8575 .flags_changed = changes, 8576 }; 8577 8578 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8579 } 8580 } 8581 8582 /** 8583 * dev_change_flags - change device settings 8584 * @dev: device 8585 * @flags: device state flags 8586 * @extack: netlink extended ack 8587 * 8588 * Change settings on device based state flags. The flags are 8589 * in the userspace exported format. 8590 */ 8591 int dev_change_flags(struct net_device *dev, unsigned int flags, 8592 struct netlink_ext_ack *extack) 8593 { 8594 int ret; 8595 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8596 8597 ret = __dev_change_flags(dev, flags, extack); 8598 if (ret < 0) 8599 return ret; 8600 8601 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8602 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8603 return ret; 8604 } 8605 EXPORT_SYMBOL(dev_change_flags); 8606 8607 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8608 { 8609 const struct net_device_ops *ops = dev->netdev_ops; 8610 8611 if (ops->ndo_change_mtu) 8612 return ops->ndo_change_mtu(dev, new_mtu); 8613 8614 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8615 WRITE_ONCE(dev->mtu, new_mtu); 8616 return 0; 8617 } 8618 EXPORT_SYMBOL(__dev_set_mtu); 8619 8620 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8621 struct netlink_ext_ack *extack) 8622 { 8623 /* MTU must be positive, and in range */ 8624 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8625 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8626 return -EINVAL; 8627 } 8628 8629 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8630 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8631 return -EINVAL; 8632 } 8633 return 0; 8634 } 8635 8636 /** 8637 * dev_set_mtu_ext - Change maximum transfer unit 8638 * @dev: device 8639 * @new_mtu: new transfer unit 8640 * @extack: netlink extended ack 8641 * 8642 * Change the maximum transfer size of the network device. 8643 */ 8644 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8645 struct netlink_ext_ack *extack) 8646 { 8647 int err, orig_mtu; 8648 8649 if (new_mtu == dev->mtu) 8650 return 0; 8651 8652 err = dev_validate_mtu(dev, new_mtu, extack); 8653 if (err) 8654 return err; 8655 8656 if (!netif_device_present(dev)) 8657 return -ENODEV; 8658 8659 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8660 err = notifier_to_errno(err); 8661 if (err) 8662 return err; 8663 8664 orig_mtu = dev->mtu; 8665 err = __dev_set_mtu(dev, new_mtu); 8666 8667 if (!err) { 8668 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8669 orig_mtu); 8670 err = notifier_to_errno(err); 8671 if (err) { 8672 /* setting mtu back and notifying everyone again, 8673 * so that they have a chance to revert changes. 8674 */ 8675 __dev_set_mtu(dev, orig_mtu); 8676 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8677 new_mtu); 8678 } 8679 } 8680 return err; 8681 } 8682 8683 int dev_set_mtu(struct net_device *dev, int new_mtu) 8684 { 8685 struct netlink_ext_ack extack; 8686 int err; 8687 8688 memset(&extack, 0, sizeof(extack)); 8689 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8690 if (err && extack._msg) 8691 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8692 return err; 8693 } 8694 EXPORT_SYMBOL(dev_set_mtu); 8695 8696 /** 8697 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8698 * @dev: device 8699 * @new_len: new tx queue length 8700 */ 8701 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8702 { 8703 unsigned int orig_len = dev->tx_queue_len; 8704 int res; 8705 8706 if (new_len != (unsigned int)new_len) 8707 return -ERANGE; 8708 8709 if (new_len != orig_len) { 8710 dev->tx_queue_len = new_len; 8711 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8712 res = notifier_to_errno(res); 8713 if (res) 8714 goto err_rollback; 8715 res = dev_qdisc_change_tx_queue_len(dev); 8716 if (res) 8717 goto err_rollback; 8718 } 8719 8720 return 0; 8721 8722 err_rollback: 8723 netdev_err(dev, "refused to change device tx_queue_len\n"); 8724 dev->tx_queue_len = orig_len; 8725 return res; 8726 } 8727 8728 /** 8729 * dev_set_group - Change group this device belongs to 8730 * @dev: device 8731 * @new_group: group this device should belong to 8732 */ 8733 void dev_set_group(struct net_device *dev, int new_group) 8734 { 8735 dev->group = new_group; 8736 } 8737 8738 /** 8739 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8740 * @dev: device 8741 * @addr: new address 8742 * @extack: netlink extended ack 8743 */ 8744 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8745 struct netlink_ext_ack *extack) 8746 { 8747 struct netdev_notifier_pre_changeaddr_info info = { 8748 .info.dev = dev, 8749 .info.extack = extack, 8750 .dev_addr = addr, 8751 }; 8752 int rc; 8753 8754 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8755 return notifier_to_errno(rc); 8756 } 8757 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8758 8759 /** 8760 * dev_set_mac_address - Change Media Access Control Address 8761 * @dev: device 8762 * @sa: new address 8763 * @extack: netlink extended ack 8764 * 8765 * Change the hardware (MAC) address of the device 8766 */ 8767 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8768 struct netlink_ext_ack *extack) 8769 { 8770 const struct net_device_ops *ops = dev->netdev_ops; 8771 int err; 8772 8773 if (!ops->ndo_set_mac_address) 8774 return -EOPNOTSUPP; 8775 if (sa->sa_family != dev->type) 8776 return -EINVAL; 8777 if (!netif_device_present(dev)) 8778 return -ENODEV; 8779 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8780 if (err) 8781 return err; 8782 err = ops->ndo_set_mac_address(dev, sa); 8783 if (err) 8784 return err; 8785 dev->addr_assign_type = NET_ADDR_SET; 8786 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8787 add_device_randomness(dev->dev_addr, dev->addr_len); 8788 return 0; 8789 } 8790 EXPORT_SYMBOL(dev_set_mac_address); 8791 8792 static DECLARE_RWSEM(dev_addr_sem); 8793 8794 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8795 struct netlink_ext_ack *extack) 8796 { 8797 int ret; 8798 8799 down_write(&dev_addr_sem); 8800 ret = dev_set_mac_address(dev, sa, extack); 8801 up_write(&dev_addr_sem); 8802 return ret; 8803 } 8804 EXPORT_SYMBOL(dev_set_mac_address_user); 8805 8806 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8807 { 8808 size_t size = sizeof(sa->sa_data_min); 8809 struct net_device *dev; 8810 int ret = 0; 8811 8812 down_read(&dev_addr_sem); 8813 rcu_read_lock(); 8814 8815 dev = dev_get_by_name_rcu(net, dev_name); 8816 if (!dev) { 8817 ret = -ENODEV; 8818 goto unlock; 8819 } 8820 if (!dev->addr_len) 8821 memset(sa->sa_data, 0, size); 8822 else 8823 memcpy(sa->sa_data, dev->dev_addr, 8824 min_t(size_t, size, dev->addr_len)); 8825 sa->sa_family = dev->type; 8826 8827 unlock: 8828 rcu_read_unlock(); 8829 up_read(&dev_addr_sem); 8830 return ret; 8831 } 8832 EXPORT_SYMBOL(dev_get_mac_address); 8833 8834 /** 8835 * dev_change_carrier - Change device carrier 8836 * @dev: device 8837 * @new_carrier: new value 8838 * 8839 * Change device carrier 8840 */ 8841 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8842 { 8843 const struct net_device_ops *ops = dev->netdev_ops; 8844 8845 if (!ops->ndo_change_carrier) 8846 return -EOPNOTSUPP; 8847 if (!netif_device_present(dev)) 8848 return -ENODEV; 8849 return ops->ndo_change_carrier(dev, new_carrier); 8850 } 8851 8852 /** 8853 * dev_get_phys_port_id - Get device physical port ID 8854 * @dev: device 8855 * @ppid: port ID 8856 * 8857 * Get device physical port ID 8858 */ 8859 int dev_get_phys_port_id(struct net_device *dev, 8860 struct netdev_phys_item_id *ppid) 8861 { 8862 const struct net_device_ops *ops = dev->netdev_ops; 8863 8864 if (!ops->ndo_get_phys_port_id) 8865 return -EOPNOTSUPP; 8866 return ops->ndo_get_phys_port_id(dev, ppid); 8867 } 8868 8869 /** 8870 * dev_get_phys_port_name - Get device physical port name 8871 * @dev: device 8872 * @name: port name 8873 * @len: limit of bytes to copy to name 8874 * 8875 * Get device physical port name 8876 */ 8877 int dev_get_phys_port_name(struct net_device *dev, 8878 char *name, size_t len) 8879 { 8880 const struct net_device_ops *ops = dev->netdev_ops; 8881 int err; 8882 8883 if (ops->ndo_get_phys_port_name) { 8884 err = ops->ndo_get_phys_port_name(dev, name, len); 8885 if (err != -EOPNOTSUPP) 8886 return err; 8887 } 8888 return devlink_compat_phys_port_name_get(dev, name, len); 8889 } 8890 8891 /** 8892 * dev_get_port_parent_id - Get the device's port parent identifier 8893 * @dev: network device 8894 * @ppid: pointer to a storage for the port's parent identifier 8895 * @recurse: allow/disallow recursion to lower devices 8896 * 8897 * Get the devices's port parent identifier 8898 */ 8899 int dev_get_port_parent_id(struct net_device *dev, 8900 struct netdev_phys_item_id *ppid, 8901 bool recurse) 8902 { 8903 const struct net_device_ops *ops = dev->netdev_ops; 8904 struct netdev_phys_item_id first = { }; 8905 struct net_device *lower_dev; 8906 struct list_head *iter; 8907 int err; 8908 8909 if (ops->ndo_get_port_parent_id) { 8910 err = ops->ndo_get_port_parent_id(dev, ppid); 8911 if (err != -EOPNOTSUPP) 8912 return err; 8913 } 8914 8915 err = devlink_compat_switch_id_get(dev, ppid); 8916 if (!recurse || err != -EOPNOTSUPP) 8917 return err; 8918 8919 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8920 err = dev_get_port_parent_id(lower_dev, ppid, true); 8921 if (err) 8922 break; 8923 if (!first.id_len) 8924 first = *ppid; 8925 else if (memcmp(&first, ppid, sizeof(*ppid))) 8926 return -EOPNOTSUPP; 8927 } 8928 8929 return err; 8930 } 8931 EXPORT_SYMBOL(dev_get_port_parent_id); 8932 8933 /** 8934 * netdev_port_same_parent_id - Indicate if two network devices have 8935 * the same port parent identifier 8936 * @a: first network device 8937 * @b: second network device 8938 */ 8939 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8940 { 8941 struct netdev_phys_item_id a_id = { }; 8942 struct netdev_phys_item_id b_id = { }; 8943 8944 if (dev_get_port_parent_id(a, &a_id, true) || 8945 dev_get_port_parent_id(b, &b_id, true)) 8946 return false; 8947 8948 return netdev_phys_item_id_same(&a_id, &b_id); 8949 } 8950 EXPORT_SYMBOL(netdev_port_same_parent_id); 8951 8952 /** 8953 * dev_change_proto_down - set carrier according to proto_down. 8954 * 8955 * @dev: device 8956 * @proto_down: new value 8957 */ 8958 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8959 { 8960 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8961 return -EOPNOTSUPP; 8962 if (!netif_device_present(dev)) 8963 return -ENODEV; 8964 if (proto_down) 8965 netif_carrier_off(dev); 8966 else 8967 netif_carrier_on(dev); 8968 dev->proto_down = proto_down; 8969 return 0; 8970 } 8971 8972 /** 8973 * dev_change_proto_down_reason - proto down reason 8974 * 8975 * @dev: device 8976 * @mask: proto down mask 8977 * @value: proto down value 8978 */ 8979 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8980 u32 value) 8981 { 8982 int b; 8983 8984 if (!mask) { 8985 dev->proto_down_reason = value; 8986 } else { 8987 for_each_set_bit(b, &mask, 32) { 8988 if (value & (1 << b)) 8989 dev->proto_down_reason |= BIT(b); 8990 else 8991 dev->proto_down_reason &= ~BIT(b); 8992 } 8993 } 8994 } 8995 8996 struct bpf_xdp_link { 8997 struct bpf_link link; 8998 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8999 int flags; 9000 }; 9001 9002 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9003 { 9004 if (flags & XDP_FLAGS_HW_MODE) 9005 return XDP_MODE_HW; 9006 if (flags & XDP_FLAGS_DRV_MODE) 9007 return XDP_MODE_DRV; 9008 if (flags & XDP_FLAGS_SKB_MODE) 9009 return XDP_MODE_SKB; 9010 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9011 } 9012 9013 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9014 { 9015 switch (mode) { 9016 case XDP_MODE_SKB: 9017 return generic_xdp_install; 9018 case XDP_MODE_DRV: 9019 case XDP_MODE_HW: 9020 return dev->netdev_ops->ndo_bpf; 9021 default: 9022 return NULL; 9023 } 9024 } 9025 9026 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9027 enum bpf_xdp_mode mode) 9028 { 9029 return dev->xdp_state[mode].link; 9030 } 9031 9032 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9033 enum bpf_xdp_mode mode) 9034 { 9035 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9036 9037 if (link) 9038 return link->link.prog; 9039 return dev->xdp_state[mode].prog; 9040 } 9041 9042 u8 dev_xdp_prog_count(struct net_device *dev) 9043 { 9044 u8 count = 0; 9045 int i; 9046 9047 for (i = 0; i < __MAX_XDP_MODE; i++) 9048 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9049 count++; 9050 return count; 9051 } 9052 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9053 9054 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9055 { 9056 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9057 9058 return prog ? prog->aux->id : 0; 9059 } 9060 9061 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9062 struct bpf_xdp_link *link) 9063 { 9064 dev->xdp_state[mode].link = link; 9065 dev->xdp_state[mode].prog = NULL; 9066 } 9067 9068 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9069 struct bpf_prog *prog) 9070 { 9071 dev->xdp_state[mode].link = NULL; 9072 dev->xdp_state[mode].prog = prog; 9073 } 9074 9075 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9076 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9077 u32 flags, struct bpf_prog *prog) 9078 { 9079 struct netdev_bpf xdp; 9080 int err; 9081 9082 memset(&xdp, 0, sizeof(xdp)); 9083 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9084 xdp.extack = extack; 9085 xdp.flags = flags; 9086 xdp.prog = prog; 9087 9088 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9089 * "moved" into driver), so they don't increment it on their own, but 9090 * they do decrement refcnt when program is detached or replaced. 9091 * Given net_device also owns link/prog, we need to bump refcnt here 9092 * to prevent drivers from underflowing it. 9093 */ 9094 if (prog) 9095 bpf_prog_inc(prog); 9096 err = bpf_op(dev, &xdp); 9097 if (err) { 9098 if (prog) 9099 bpf_prog_put(prog); 9100 return err; 9101 } 9102 9103 if (mode != XDP_MODE_HW) 9104 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9105 9106 return 0; 9107 } 9108 9109 static void dev_xdp_uninstall(struct net_device *dev) 9110 { 9111 struct bpf_xdp_link *link; 9112 struct bpf_prog *prog; 9113 enum bpf_xdp_mode mode; 9114 bpf_op_t bpf_op; 9115 9116 ASSERT_RTNL(); 9117 9118 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9119 prog = dev_xdp_prog(dev, mode); 9120 if (!prog) 9121 continue; 9122 9123 bpf_op = dev_xdp_bpf_op(dev, mode); 9124 if (!bpf_op) 9125 continue; 9126 9127 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9128 9129 /* auto-detach link from net device */ 9130 link = dev_xdp_link(dev, mode); 9131 if (link) 9132 link->dev = NULL; 9133 else 9134 bpf_prog_put(prog); 9135 9136 dev_xdp_set_link(dev, mode, NULL); 9137 } 9138 } 9139 9140 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9141 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9142 struct bpf_prog *old_prog, u32 flags) 9143 { 9144 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9145 struct bpf_prog *cur_prog; 9146 struct net_device *upper; 9147 struct list_head *iter; 9148 enum bpf_xdp_mode mode; 9149 bpf_op_t bpf_op; 9150 int err; 9151 9152 ASSERT_RTNL(); 9153 9154 /* either link or prog attachment, never both */ 9155 if (link && (new_prog || old_prog)) 9156 return -EINVAL; 9157 /* link supports only XDP mode flags */ 9158 if (link && (flags & ~XDP_FLAGS_MODES)) { 9159 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9160 return -EINVAL; 9161 } 9162 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9163 if (num_modes > 1) { 9164 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9165 return -EINVAL; 9166 } 9167 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9168 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9169 NL_SET_ERR_MSG(extack, 9170 "More than one program loaded, unset mode is ambiguous"); 9171 return -EINVAL; 9172 } 9173 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9174 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9175 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9176 return -EINVAL; 9177 } 9178 9179 mode = dev_xdp_mode(dev, flags); 9180 /* can't replace attached link */ 9181 if (dev_xdp_link(dev, mode)) { 9182 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9183 return -EBUSY; 9184 } 9185 9186 /* don't allow if an upper device already has a program */ 9187 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9188 if (dev_xdp_prog_count(upper) > 0) { 9189 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9190 return -EEXIST; 9191 } 9192 } 9193 9194 cur_prog = dev_xdp_prog(dev, mode); 9195 /* can't replace attached prog with link */ 9196 if (link && cur_prog) { 9197 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9198 return -EBUSY; 9199 } 9200 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9201 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9202 return -EEXIST; 9203 } 9204 9205 /* put effective new program into new_prog */ 9206 if (link) 9207 new_prog = link->link.prog; 9208 9209 if (new_prog) { 9210 bool offload = mode == XDP_MODE_HW; 9211 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9212 ? XDP_MODE_DRV : XDP_MODE_SKB; 9213 9214 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9215 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9216 return -EBUSY; 9217 } 9218 if (!offload && dev_xdp_prog(dev, other_mode)) { 9219 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9220 return -EEXIST; 9221 } 9222 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9223 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9224 return -EINVAL; 9225 } 9226 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9227 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9228 return -EINVAL; 9229 } 9230 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9231 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9232 return -EINVAL; 9233 } 9234 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9235 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9236 return -EINVAL; 9237 } 9238 } 9239 9240 /* don't call drivers if the effective program didn't change */ 9241 if (new_prog != cur_prog) { 9242 bpf_op = dev_xdp_bpf_op(dev, mode); 9243 if (!bpf_op) { 9244 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9245 return -EOPNOTSUPP; 9246 } 9247 9248 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9249 if (err) 9250 return err; 9251 } 9252 9253 if (link) 9254 dev_xdp_set_link(dev, mode, link); 9255 else 9256 dev_xdp_set_prog(dev, mode, new_prog); 9257 if (cur_prog) 9258 bpf_prog_put(cur_prog); 9259 9260 return 0; 9261 } 9262 9263 static int dev_xdp_attach_link(struct net_device *dev, 9264 struct netlink_ext_ack *extack, 9265 struct bpf_xdp_link *link) 9266 { 9267 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9268 } 9269 9270 static int dev_xdp_detach_link(struct net_device *dev, 9271 struct netlink_ext_ack *extack, 9272 struct bpf_xdp_link *link) 9273 { 9274 enum bpf_xdp_mode mode; 9275 bpf_op_t bpf_op; 9276 9277 ASSERT_RTNL(); 9278 9279 mode = dev_xdp_mode(dev, link->flags); 9280 if (dev_xdp_link(dev, mode) != link) 9281 return -EINVAL; 9282 9283 bpf_op = dev_xdp_bpf_op(dev, mode); 9284 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9285 dev_xdp_set_link(dev, mode, NULL); 9286 return 0; 9287 } 9288 9289 static void bpf_xdp_link_release(struct bpf_link *link) 9290 { 9291 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9292 9293 rtnl_lock(); 9294 9295 /* if racing with net_device's tear down, xdp_link->dev might be 9296 * already NULL, in which case link was already auto-detached 9297 */ 9298 if (xdp_link->dev) { 9299 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9300 xdp_link->dev = NULL; 9301 } 9302 9303 rtnl_unlock(); 9304 } 9305 9306 static int bpf_xdp_link_detach(struct bpf_link *link) 9307 { 9308 bpf_xdp_link_release(link); 9309 return 0; 9310 } 9311 9312 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9313 { 9314 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9315 9316 kfree(xdp_link); 9317 } 9318 9319 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9320 struct seq_file *seq) 9321 { 9322 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9323 u32 ifindex = 0; 9324 9325 rtnl_lock(); 9326 if (xdp_link->dev) 9327 ifindex = xdp_link->dev->ifindex; 9328 rtnl_unlock(); 9329 9330 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9331 } 9332 9333 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9334 struct bpf_link_info *info) 9335 { 9336 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9337 u32 ifindex = 0; 9338 9339 rtnl_lock(); 9340 if (xdp_link->dev) 9341 ifindex = xdp_link->dev->ifindex; 9342 rtnl_unlock(); 9343 9344 info->xdp.ifindex = ifindex; 9345 return 0; 9346 } 9347 9348 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9349 struct bpf_prog *old_prog) 9350 { 9351 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9352 enum bpf_xdp_mode mode; 9353 bpf_op_t bpf_op; 9354 int err = 0; 9355 9356 rtnl_lock(); 9357 9358 /* link might have been auto-released already, so fail */ 9359 if (!xdp_link->dev) { 9360 err = -ENOLINK; 9361 goto out_unlock; 9362 } 9363 9364 if (old_prog && link->prog != old_prog) { 9365 err = -EPERM; 9366 goto out_unlock; 9367 } 9368 old_prog = link->prog; 9369 if (old_prog->type != new_prog->type || 9370 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9371 err = -EINVAL; 9372 goto out_unlock; 9373 } 9374 9375 if (old_prog == new_prog) { 9376 /* no-op, don't disturb drivers */ 9377 bpf_prog_put(new_prog); 9378 goto out_unlock; 9379 } 9380 9381 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9382 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9383 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9384 xdp_link->flags, new_prog); 9385 if (err) 9386 goto out_unlock; 9387 9388 old_prog = xchg(&link->prog, new_prog); 9389 bpf_prog_put(old_prog); 9390 9391 out_unlock: 9392 rtnl_unlock(); 9393 return err; 9394 } 9395 9396 static const struct bpf_link_ops bpf_xdp_link_lops = { 9397 .release = bpf_xdp_link_release, 9398 .dealloc = bpf_xdp_link_dealloc, 9399 .detach = bpf_xdp_link_detach, 9400 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9401 .fill_link_info = bpf_xdp_link_fill_link_info, 9402 .update_prog = bpf_xdp_link_update, 9403 }; 9404 9405 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9406 { 9407 struct net *net = current->nsproxy->net_ns; 9408 struct bpf_link_primer link_primer; 9409 struct bpf_xdp_link *link; 9410 struct net_device *dev; 9411 int err, fd; 9412 9413 rtnl_lock(); 9414 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9415 if (!dev) { 9416 rtnl_unlock(); 9417 return -EINVAL; 9418 } 9419 9420 link = kzalloc(sizeof(*link), GFP_USER); 9421 if (!link) { 9422 err = -ENOMEM; 9423 goto unlock; 9424 } 9425 9426 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9427 link->dev = dev; 9428 link->flags = attr->link_create.flags; 9429 9430 err = bpf_link_prime(&link->link, &link_primer); 9431 if (err) { 9432 kfree(link); 9433 goto unlock; 9434 } 9435 9436 err = dev_xdp_attach_link(dev, NULL, link); 9437 rtnl_unlock(); 9438 9439 if (err) { 9440 link->dev = NULL; 9441 bpf_link_cleanup(&link_primer); 9442 goto out_put_dev; 9443 } 9444 9445 fd = bpf_link_settle(&link_primer); 9446 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9447 dev_put(dev); 9448 return fd; 9449 9450 unlock: 9451 rtnl_unlock(); 9452 9453 out_put_dev: 9454 dev_put(dev); 9455 return err; 9456 } 9457 9458 /** 9459 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9460 * @dev: device 9461 * @extack: netlink extended ack 9462 * @fd: new program fd or negative value to clear 9463 * @expected_fd: old program fd that userspace expects to replace or clear 9464 * @flags: xdp-related flags 9465 * 9466 * Set or clear a bpf program for a device 9467 */ 9468 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9469 int fd, int expected_fd, u32 flags) 9470 { 9471 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9472 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9473 int err; 9474 9475 ASSERT_RTNL(); 9476 9477 if (fd >= 0) { 9478 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9479 mode != XDP_MODE_SKB); 9480 if (IS_ERR(new_prog)) 9481 return PTR_ERR(new_prog); 9482 } 9483 9484 if (expected_fd >= 0) { 9485 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9486 mode != XDP_MODE_SKB); 9487 if (IS_ERR(old_prog)) { 9488 err = PTR_ERR(old_prog); 9489 old_prog = NULL; 9490 goto err_out; 9491 } 9492 } 9493 9494 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9495 9496 err_out: 9497 if (err && new_prog) 9498 bpf_prog_put(new_prog); 9499 if (old_prog) 9500 bpf_prog_put(old_prog); 9501 return err; 9502 } 9503 9504 /** 9505 * dev_new_index - allocate an ifindex 9506 * @net: the applicable net namespace 9507 * 9508 * Returns a suitable unique value for a new device interface 9509 * number. The caller must hold the rtnl semaphore or the 9510 * dev_base_lock to be sure it remains unique. 9511 */ 9512 static int dev_new_index(struct net *net) 9513 { 9514 int ifindex = net->ifindex; 9515 9516 for (;;) { 9517 if (++ifindex <= 0) 9518 ifindex = 1; 9519 if (!__dev_get_by_index(net, ifindex)) 9520 return net->ifindex = ifindex; 9521 } 9522 } 9523 9524 /* Delayed registration/unregisteration */ 9525 LIST_HEAD(net_todo_list); 9526 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9527 9528 static void net_set_todo(struct net_device *dev) 9529 { 9530 list_add_tail(&dev->todo_list, &net_todo_list); 9531 atomic_inc(&dev_net(dev)->dev_unreg_count); 9532 } 9533 9534 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9535 struct net_device *upper, netdev_features_t features) 9536 { 9537 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9538 netdev_features_t feature; 9539 int feature_bit; 9540 9541 for_each_netdev_feature(upper_disables, feature_bit) { 9542 feature = __NETIF_F_BIT(feature_bit); 9543 if (!(upper->wanted_features & feature) 9544 && (features & feature)) { 9545 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9546 &feature, upper->name); 9547 features &= ~feature; 9548 } 9549 } 9550 9551 return features; 9552 } 9553 9554 static void netdev_sync_lower_features(struct net_device *upper, 9555 struct net_device *lower, netdev_features_t features) 9556 { 9557 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9558 netdev_features_t feature; 9559 int feature_bit; 9560 9561 for_each_netdev_feature(upper_disables, feature_bit) { 9562 feature = __NETIF_F_BIT(feature_bit); 9563 if (!(features & feature) && (lower->features & feature)) { 9564 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9565 &feature, lower->name); 9566 lower->wanted_features &= ~feature; 9567 __netdev_update_features(lower); 9568 9569 if (unlikely(lower->features & feature)) 9570 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9571 &feature, lower->name); 9572 else 9573 netdev_features_change(lower); 9574 } 9575 } 9576 } 9577 9578 static netdev_features_t netdev_fix_features(struct net_device *dev, 9579 netdev_features_t features) 9580 { 9581 /* Fix illegal checksum combinations */ 9582 if ((features & NETIF_F_HW_CSUM) && 9583 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9584 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9585 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9586 } 9587 9588 /* TSO requires that SG is present as well. */ 9589 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9590 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9591 features &= ~NETIF_F_ALL_TSO; 9592 } 9593 9594 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9595 !(features & NETIF_F_IP_CSUM)) { 9596 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9597 features &= ~NETIF_F_TSO; 9598 features &= ~NETIF_F_TSO_ECN; 9599 } 9600 9601 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9602 !(features & NETIF_F_IPV6_CSUM)) { 9603 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9604 features &= ~NETIF_F_TSO6; 9605 } 9606 9607 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9608 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9609 features &= ~NETIF_F_TSO_MANGLEID; 9610 9611 /* TSO ECN requires that TSO is present as well. */ 9612 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9613 features &= ~NETIF_F_TSO_ECN; 9614 9615 /* Software GSO depends on SG. */ 9616 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9617 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9618 features &= ~NETIF_F_GSO; 9619 } 9620 9621 /* GSO partial features require GSO partial be set */ 9622 if ((features & dev->gso_partial_features) && 9623 !(features & NETIF_F_GSO_PARTIAL)) { 9624 netdev_dbg(dev, 9625 "Dropping partially supported GSO features since no GSO partial.\n"); 9626 features &= ~dev->gso_partial_features; 9627 } 9628 9629 if (!(features & NETIF_F_RXCSUM)) { 9630 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9631 * successfully merged by hardware must also have the 9632 * checksum verified by hardware. If the user does not 9633 * want to enable RXCSUM, logically, we should disable GRO_HW. 9634 */ 9635 if (features & NETIF_F_GRO_HW) { 9636 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9637 features &= ~NETIF_F_GRO_HW; 9638 } 9639 } 9640 9641 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9642 if (features & NETIF_F_RXFCS) { 9643 if (features & NETIF_F_LRO) { 9644 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9645 features &= ~NETIF_F_LRO; 9646 } 9647 9648 if (features & NETIF_F_GRO_HW) { 9649 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9650 features &= ~NETIF_F_GRO_HW; 9651 } 9652 } 9653 9654 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9655 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9656 features &= ~NETIF_F_LRO; 9657 } 9658 9659 if (features & NETIF_F_HW_TLS_TX) { 9660 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9661 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9662 bool hw_csum = features & NETIF_F_HW_CSUM; 9663 9664 if (!ip_csum && !hw_csum) { 9665 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9666 features &= ~NETIF_F_HW_TLS_TX; 9667 } 9668 } 9669 9670 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9671 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9672 features &= ~NETIF_F_HW_TLS_RX; 9673 } 9674 9675 return features; 9676 } 9677 9678 int __netdev_update_features(struct net_device *dev) 9679 { 9680 struct net_device *upper, *lower; 9681 netdev_features_t features; 9682 struct list_head *iter; 9683 int err = -1; 9684 9685 ASSERT_RTNL(); 9686 9687 features = netdev_get_wanted_features(dev); 9688 9689 if (dev->netdev_ops->ndo_fix_features) 9690 features = dev->netdev_ops->ndo_fix_features(dev, features); 9691 9692 /* driver might be less strict about feature dependencies */ 9693 features = netdev_fix_features(dev, features); 9694 9695 /* some features can't be enabled if they're off on an upper device */ 9696 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9697 features = netdev_sync_upper_features(dev, upper, features); 9698 9699 if (dev->features == features) 9700 goto sync_lower; 9701 9702 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9703 &dev->features, &features); 9704 9705 if (dev->netdev_ops->ndo_set_features) 9706 err = dev->netdev_ops->ndo_set_features(dev, features); 9707 else 9708 err = 0; 9709 9710 if (unlikely(err < 0)) { 9711 netdev_err(dev, 9712 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9713 err, &features, &dev->features); 9714 /* return non-0 since some features might have changed and 9715 * it's better to fire a spurious notification than miss it 9716 */ 9717 return -1; 9718 } 9719 9720 sync_lower: 9721 /* some features must be disabled on lower devices when disabled 9722 * on an upper device (think: bonding master or bridge) 9723 */ 9724 netdev_for_each_lower_dev(dev, lower, iter) 9725 netdev_sync_lower_features(dev, lower, features); 9726 9727 if (!err) { 9728 netdev_features_t diff = features ^ dev->features; 9729 9730 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9731 /* udp_tunnel_{get,drop}_rx_info both need 9732 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9733 * device, or they won't do anything. 9734 * Thus we need to update dev->features 9735 * *before* calling udp_tunnel_get_rx_info, 9736 * but *after* calling udp_tunnel_drop_rx_info. 9737 */ 9738 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9739 dev->features = features; 9740 udp_tunnel_get_rx_info(dev); 9741 } else { 9742 udp_tunnel_drop_rx_info(dev); 9743 } 9744 } 9745 9746 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9747 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9748 dev->features = features; 9749 err |= vlan_get_rx_ctag_filter_info(dev); 9750 } else { 9751 vlan_drop_rx_ctag_filter_info(dev); 9752 } 9753 } 9754 9755 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9756 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9757 dev->features = features; 9758 err |= vlan_get_rx_stag_filter_info(dev); 9759 } else { 9760 vlan_drop_rx_stag_filter_info(dev); 9761 } 9762 } 9763 9764 dev->features = features; 9765 } 9766 9767 return err < 0 ? 0 : 1; 9768 } 9769 9770 /** 9771 * netdev_update_features - recalculate device features 9772 * @dev: the device to check 9773 * 9774 * Recalculate dev->features set and send notifications if it 9775 * has changed. Should be called after driver or hardware dependent 9776 * conditions might have changed that influence the features. 9777 */ 9778 void netdev_update_features(struct net_device *dev) 9779 { 9780 if (__netdev_update_features(dev)) 9781 netdev_features_change(dev); 9782 } 9783 EXPORT_SYMBOL(netdev_update_features); 9784 9785 /** 9786 * netdev_change_features - recalculate device features 9787 * @dev: the device to check 9788 * 9789 * Recalculate dev->features set and send notifications even 9790 * if they have not changed. Should be called instead of 9791 * netdev_update_features() if also dev->vlan_features might 9792 * have changed to allow the changes to be propagated to stacked 9793 * VLAN devices. 9794 */ 9795 void netdev_change_features(struct net_device *dev) 9796 { 9797 __netdev_update_features(dev); 9798 netdev_features_change(dev); 9799 } 9800 EXPORT_SYMBOL(netdev_change_features); 9801 9802 /** 9803 * netif_stacked_transfer_operstate - transfer operstate 9804 * @rootdev: the root or lower level device to transfer state from 9805 * @dev: the device to transfer operstate to 9806 * 9807 * Transfer operational state from root to device. This is normally 9808 * called when a stacking relationship exists between the root 9809 * device and the device(a leaf device). 9810 */ 9811 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9812 struct net_device *dev) 9813 { 9814 if (rootdev->operstate == IF_OPER_DORMANT) 9815 netif_dormant_on(dev); 9816 else 9817 netif_dormant_off(dev); 9818 9819 if (rootdev->operstate == IF_OPER_TESTING) 9820 netif_testing_on(dev); 9821 else 9822 netif_testing_off(dev); 9823 9824 if (netif_carrier_ok(rootdev)) 9825 netif_carrier_on(dev); 9826 else 9827 netif_carrier_off(dev); 9828 } 9829 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9830 9831 static int netif_alloc_rx_queues(struct net_device *dev) 9832 { 9833 unsigned int i, count = dev->num_rx_queues; 9834 struct netdev_rx_queue *rx; 9835 size_t sz = count * sizeof(*rx); 9836 int err = 0; 9837 9838 BUG_ON(count < 1); 9839 9840 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9841 if (!rx) 9842 return -ENOMEM; 9843 9844 dev->_rx = rx; 9845 9846 for (i = 0; i < count; i++) { 9847 rx[i].dev = dev; 9848 9849 /* XDP RX-queue setup */ 9850 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9851 if (err < 0) 9852 goto err_rxq_info; 9853 } 9854 return 0; 9855 9856 err_rxq_info: 9857 /* Rollback successful reg's and free other resources */ 9858 while (i--) 9859 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9860 kvfree(dev->_rx); 9861 dev->_rx = NULL; 9862 return err; 9863 } 9864 9865 static void netif_free_rx_queues(struct net_device *dev) 9866 { 9867 unsigned int i, count = dev->num_rx_queues; 9868 9869 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9870 if (!dev->_rx) 9871 return; 9872 9873 for (i = 0; i < count; i++) 9874 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9875 9876 kvfree(dev->_rx); 9877 } 9878 9879 static void netdev_init_one_queue(struct net_device *dev, 9880 struct netdev_queue *queue, void *_unused) 9881 { 9882 /* Initialize queue lock */ 9883 spin_lock_init(&queue->_xmit_lock); 9884 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9885 queue->xmit_lock_owner = -1; 9886 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9887 queue->dev = dev; 9888 #ifdef CONFIG_BQL 9889 dql_init(&queue->dql, HZ); 9890 #endif 9891 } 9892 9893 static void netif_free_tx_queues(struct net_device *dev) 9894 { 9895 kvfree(dev->_tx); 9896 } 9897 9898 static int netif_alloc_netdev_queues(struct net_device *dev) 9899 { 9900 unsigned int count = dev->num_tx_queues; 9901 struct netdev_queue *tx; 9902 size_t sz = count * sizeof(*tx); 9903 9904 if (count < 1 || count > 0xffff) 9905 return -EINVAL; 9906 9907 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9908 if (!tx) 9909 return -ENOMEM; 9910 9911 dev->_tx = tx; 9912 9913 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9914 spin_lock_init(&dev->tx_global_lock); 9915 9916 return 0; 9917 } 9918 9919 void netif_tx_stop_all_queues(struct net_device *dev) 9920 { 9921 unsigned int i; 9922 9923 for (i = 0; i < dev->num_tx_queues; i++) { 9924 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9925 9926 netif_tx_stop_queue(txq); 9927 } 9928 } 9929 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9930 9931 /** 9932 * register_netdevice() - register a network device 9933 * @dev: device to register 9934 * 9935 * Take a prepared network device structure and make it externally accessible. 9936 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9937 * Callers must hold the rtnl lock - you may want register_netdev() 9938 * instead of this. 9939 */ 9940 int register_netdevice(struct net_device *dev) 9941 { 9942 int ret; 9943 struct net *net = dev_net(dev); 9944 9945 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9946 NETDEV_FEATURE_COUNT); 9947 BUG_ON(dev_boot_phase); 9948 ASSERT_RTNL(); 9949 9950 might_sleep(); 9951 9952 /* When net_device's are persistent, this will be fatal. */ 9953 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9954 BUG_ON(!net); 9955 9956 ret = ethtool_check_ops(dev->ethtool_ops); 9957 if (ret) 9958 return ret; 9959 9960 spin_lock_init(&dev->addr_list_lock); 9961 netdev_set_addr_lockdep_class(dev); 9962 9963 ret = dev_get_valid_name(net, dev, dev->name); 9964 if (ret < 0) 9965 goto out; 9966 9967 ret = -ENOMEM; 9968 dev->name_node = netdev_name_node_head_alloc(dev); 9969 if (!dev->name_node) 9970 goto out; 9971 9972 /* Init, if this function is available */ 9973 if (dev->netdev_ops->ndo_init) { 9974 ret = dev->netdev_ops->ndo_init(dev); 9975 if (ret) { 9976 if (ret > 0) 9977 ret = -EIO; 9978 goto err_free_name; 9979 } 9980 } 9981 9982 if (((dev->hw_features | dev->features) & 9983 NETIF_F_HW_VLAN_CTAG_FILTER) && 9984 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9985 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9986 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9987 ret = -EINVAL; 9988 goto err_uninit; 9989 } 9990 9991 ret = -EBUSY; 9992 if (!dev->ifindex) 9993 dev->ifindex = dev_new_index(net); 9994 else if (__dev_get_by_index(net, dev->ifindex)) 9995 goto err_uninit; 9996 9997 /* Transfer changeable features to wanted_features and enable 9998 * software offloads (GSO and GRO). 9999 */ 10000 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10001 dev->features |= NETIF_F_SOFT_FEATURES; 10002 10003 if (dev->udp_tunnel_nic_info) { 10004 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10005 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10006 } 10007 10008 dev->wanted_features = dev->features & dev->hw_features; 10009 10010 if (!(dev->flags & IFF_LOOPBACK)) 10011 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10012 10013 /* If IPv4 TCP segmentation offload is supported we should also 10014 * allow the device to enable segmenting the frame with the option 10015 * of ignoring a static IP ID value. This doesn't enable the 10016 * feature itself but allows the user to enable it later. 10017 */ 10018 if (dev->hw_features & NETIF_F_TSO) 10019 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10020 if (dev->vlan_features & NETIF_F_TSO) 10021 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10022 if (dev->mpls_features & NETIF_F_TSO) 10023 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10024 if (dev->hw_enc_features & NETIF_F_TSO) 10025 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10026 10027 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10028 */ 10029 dev->vlan_features |= NETIF_F_HIGHDMA; 10030 10031 /* Make NETIF_F_SG inheritable to tunnel devices. 10032 */ 10033 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10034 10035 /* Make NETIF_F_SG inheritable to MPLS. 10036 */ 10037 dev->mpls_features |= NETIF_F_SG; 10038 10039 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10040 ret = notifier_to_errno(ret); 10041 if (ret) 10042 goto err_uninit; 10043 10044 ret = netdev_register_kobject(dev); 10045 write_lock(&dev_base_lock); 10046 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10047 write_unlock(&dev_base_lock); 10048 if (ret) 10049 goto err_uninit_notify; 10050 10051 __netdev_update_features(dev); 10052 10053 /* 10054 * Default initial state at registry is that the 10055 * device is present. 10056 */ 10057 10058 set_bit(__LINK_STATE_PRESENT, &dev->state); 10059 10060 linkwatch_init_dev(dev); 10061 10062 dev_init_scheduler(dev); 10063 10064 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10065 list_netdevice(dev); 10066 10067 add_device_randomness(dev->dev_addr, dev->addr_len); 10068 10069 /* If the device has permanent device address, driver should 10070 * set dev_addr and also addr_assign_type should be set to 10071 * NET_ADDR_PERM (default value). 10072 */ 10073 if (dev->addr_assign_type == NET_ADDR_PERM) 10074 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10075 10076 /* Notify protocols, that a new device appeared. */ 10077 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10078 ret = notifier_to_errno(ret); 10079 if (ret) { 10080 /* Expect explicit free_netdev() on failure */ 10081 dev->needs_free_netdev = false; 10082 unregister_netdevice_queue(dev, NULL); 10083 goto out; 10084 } 10085 /* 10086 * Prevent userspace races by waiting until the network 10087 * device is fully setup before sending notifications. 10088 */ 10089 if (!dev->rtnl_link_ops || 10090 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10091 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10092 10093 out: 10094 return ret; 10095 10096 err_uninit_notify: 10097 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10098 err_uninit: 10099 if (dev->netdev_ops->ndo_uninit) 10100 dev->netdev_ops->ndo_uninit(dev); 10101 if (dev->priv_destructor) 10102 dev->priv_destructor(dev); 10103 err_free_name: 10104 netdev_name_node_free(dev->name_node); 10105 goto out; 10106 } 10107 EXPORT_SYMBOL(register_netdevice); 10108 10109 /** 10110 * init_dummy_netdev - init a dummy network device for NAPI 10111 * @dev: device to init 10112 * 10113 * This takes a network device structure and initialize the minimum 10114 * amount of fields so it can be used to schedule NAPI polls without 10115 * registering a full blown interface. This is to be used by drivers 10116 * that need to tie several hardware interfaces to a single NAPI 10117 * poll scheduler due to HW limitations. 10118 */ 10119 int init_dummy_netdev(struct net_device *dev) 10120 { 10121 /* Clear everything. Note we don't initialize spinlocks 10122 * are they aren't supposed to be taken by any of the 10123 * NAPI code and this dummy netdev is supposed to be 10124 * only ever used for NAPI polls 10125 */ 10126 memset(dev, 0, sizeof(struct net_device)); 10127 10128 /* make sure we BUG if trying to hit standard 10129 * register/unregister code path 10130 */ 10131 dev->reg_state = NETREG_DUMMY; 10132 10133 /* NAPI wants this */ 10134 INIT_LIST_HEAD(&dev->napi_list); 10135 10136 /* a dummy interface is started by default */ 10137 set_bit(__LINK_STATE_PRESENT, &dev->state); 10138 set_bit(__LINK_STATE_START, &dev->state); 10139 10140 /* napi_busy_loop stats accounting wants this */ 10141 dev_net_set(dev, &init_net); 10142 10143 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10144 * because users of this 'device' dont need to change 10145 * its refcount. 10146 */ 10147 10148 return 0; 10149 } 10150 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10151 10152 10153 /** 10154 * register_netdev - register a network device 10155 * @dev: device to register 10156 * 10157 * Take a completed network device structure and add it to the kernel 10158 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10159 * chain. 0 is returned on success. A negative errno code is returned 10160 * on a failure to set up the device, or if the name is a duplicate. 10161 * 10162 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10163 * and expands the device name if you passed a format string to 10164 * alloc_netdev. 10165 */ 10166 int register_netdev(struct net_device *dev) 10167 { 10168 int err; 10169 10170 if (rtnl_lock_killable()) 10171 return -EINTR; 10172 err = register_netdevice(dev); 10173 rtnl_unlock(); 10174 return err; 10175 } 10176 EXPORT_SYMBOL(register_netdev); 10177 10178 int netdev_refcnt_read(const struct net_device *dev) 10179 { 10180 #ifdef CONFIG_PCPU_DEV_REFCNT 10181 int i, refcnt = 0; 10182 10183 for_each_possible_cpu(i) 10184 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10185 return refcnt; 10186 #else 10187 return refcount_read(&dev->dev_refcnt); 10188 #endif 10189 } 10190 EXPORT_SYMBOL(netdev_refcnt_read); 10191 10192 int netdev_unregister_timeout_secs __read_mostly = 10; 10193 10194 #define WAIT_REFS_MIN_MSECS 1 10195 #define WAIT_REFS_MAX_MSECS 250 10196 /** 10197 * netdev_wait_allrefs_any - wait until all references are gone. 10198 * @list: list of net_devices to wait on 10199 * 10200 * This is called when unregistering network devices. 10201 * 10202 * Any protocol or device that holds a reference should register 10203 * for netdevice notification, and cleanup and put back the 10204 * reference if they receive an UNREGISTER event. 10205 * We can get stuck here if buggy protocols don't correctly 10206 * call dev_put. 10207 */ 10208 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10209 { 10210 unsigned long rebroadcast_time, warning_time; 10211 struct net_device *dev; 10212 int wait = 0; 10213 10214 rebroadcast_time = warning_time = jiffies; 10215 10216 list_for_each_entry(dev, list, todo_list) 10217 if (netdev_refcnt_read(dev) == 1) 10218 return dev; 10219 10220 while (true) { 10221 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10222 rtnl_lock(); 10223 10224 /* Rebroadcast unregister notification */ 10225 list_for_each_entry(dev, list, todo_list) 10226 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10227 10228 __rtnl_unlock(); 10229 rcu_barrier(); 10230 rtnl_lock(); 10231 10232 list_for_each_entry(dev, list, todo_list) 10233 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10234 &dev->state)) { 10235 /* We must not have linkwatch events 10236 * pending on unregister. If this 10237 * happens, we simply run the queue 10238 * unscheduled, resulting in a noop 10239 * for this device. 10240 */ 10241 linkwatch_run_queue(); 10242 break; 10243 } 10244 10245 __rtnl_unlock(); 10246 10247 rebroadcast_time = jiffies; 10248 } 10249 10250 if (!wait) { 10251 rcu_barrier(); 10252 wait = WAIT_REFS_MIN_MSECS; 10253 } else { 10254 msleep(wait); 10255 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10256 } 10257 10258 list_for_each_entry(dev, list, todo_list) 10259 if (netdev_refcnt_read(dev) == 1) 10260 return dev; 10261 10262 if (time_after(jiffies, warning_time + 10263 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10264 list_for_each_entry(dev, list, todo_list) { 10265 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10266 dev->name, netdev_refcnt_read(dev)); 10267 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10268 } 10269 10270 warning_time = jiffies; 10271 } 10272 } 10273 } 10274 10275 /* The sequence is: 10276 * 10277 * rtnl_lock(); 10278 * ... 10279 * register_netdevice(x1); 10280 * register_netdevice(x2); 10281 * ... 10282 * unregister_netdevice(y1); 10283 * unregister_netdevice(y2); 10284 * ... 10285 * rtnl_unlock(); 10286 * free_netdev(y1); 10287 * free_netdev(y2); 10288 * 10289 * We are invoked by rtnl_unlock(). 10290 * This allows us to deal with problems: 10291 * 1) We can delete sysfs objects which invoke hotplug 10292 * without deadlocking with linkwatch via keventd. 10293 * 2) Since we run with the RTNL semaphore not held, we can sleep 10294 * safely in order to wait for the netdev refcnt to drop to zero. 10295 * 10296 * We must not return until all unregister events added during 10297 * the interval the lock was held have been completed. 10298 */ 10299 void netdev_run_todo(void) 10300 { 10301 struct net_device *dev, *tmp; 10302 struct list_head list; 10303 #ifdef CONFIG_LOCKDEP 10304 struct list_head unlink_list; 10305 10306 list_replace_init(&net_unlink_list, &unlink_list); 10307 10308 while (!list_empty(&unlink_list)) { 10309 struct net_device *dev = list_first_entry(&unlink_list, 10310 struct net_device, 10311 unlink_list); 10312 list_del_init(&dev->unlink_list); 10313 dev->nested_level = dev->lower_level - 1; 10314 } 10315 #endif 10316 10317 /* Snapshot list, allow later requests */ 10318 list_replace_init(&net_todo_list, &list); 10319 10320 __rtnl_unlock(); 10321 10322 /* Wait for rcu callbacks to finish before next phase */ 10323 if (!list_empty(&list)) 10324 rcu_barrier(); 10325 10326 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10327 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10328 netdev_WARN(dev, "run_todo but not unregistering\n"); 10329 list_del(&dev->todo_list); 10330 continue; 10331 } 10332 10333 write_lock(&dev_base_lock); 10334 dev->reg_state = NETREG_UNREGISTERED; 10335 write_unlock(&dev_base_lock); 10336 linkwatch_forget_dev(dev); 10337 } 10338 10339 while (!list_empty(&list)) { 10340 dev = netdev_wait_allrefs_any(&list); 10341 list_del(&dev->todo_list); 10342 10343 /* paranoia */ 10344 BUG_ON(netdev_refcnt_read(dev) != 1); 10345 BUG_ON(!list_empty(&dev->ptype_all)); 10346 BUG_ON(!list_empty(&dev->ptype_specific)); 10347 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10348 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10349 10350 if (dev->priv_destructor) 10351 dev->priv_destructor(dev); 10352 if (dev->needs_free_netdev) 10353 free_netdev(dev); 10354 10355 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10356 wake_up(&netdev_unregistering_wq); 10357 10358 /* Free network device */ 10359 kobject_put(&dev->dev.kobj); 10360 } 10361 } 10362 10363 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10364 * all the same fields in the same order as net_device_stats, with only 10365 * the type differing, but rtnl_link_stats64 may have additional fields 10366 * at the end for newer counters. 10367 */ 10368 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10369 const struct net_device_stats *netdev_stats) 10370 { 10371 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10372 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10373 u64 *dst = (u64 *)stats64; 10374 10375 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10376 for (i = 0; i < n; i++) 10377 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10378 /* zero out counters that only exist in rtnl_link_stats64 */ 10379 memset((char *)stats64 + n * sizeof(u64), 0, 10380 sizeof(*stats64) - n * sizeof(u64)); 10381 } 10382 EXPORT_SYMBOL(netdev_stats_to_stats64); 10383 10384 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10385 { 10386 struct net_device_core_stats __percpu *p; 10387 10388 p = alloc_percpu_gfp(struct net_device_core_stats, 10389 GFP_ATOMIC | __GFP_NOWARN); 10390 10391 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10392 free_percpu(p); 10393 10394 /* This READ_ONCE() pairs with the cmpxchg() above */ 10395 return READ_ONCE(dev->core_stats); 10396 } 10397 EXPORT_SYMBOL(netdev_core_stats_alloc); 10398 10399 /** 10400 * dev_get_stats - get network device statistics 10401 * @dev: device to get statistics from 10402 * @storage: place to store stats 10403 * 10404 * Get network statistics from device. Return @storage. 10405 * The device driver may provide its own method by setting 10406 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10407 * otherwise the internal statistics structure is used. 10408 */ 10409 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10410 struct rtnl_link_stats64 *storage) 10411 { 10412 const struct net_device_ops *ops = dev->netdev_ops; 10413 const struct net_device_core_stats __percpu *p; 10414 10415 if (ops->ndo_get_stats64) { 10416 memset(storage, 0, sizeof(*storage)); 10417 ops->ndo_get_stats64(dev, storage); 10418 } else if (ops->ndo_get_stats) { 10419 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10420 } else { 10421 netdev_stats_to_stats64(storage, &dev->stats); 10422 } 10423 10424 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10425 p = READ_ONCE(dev->core_stats); 10426 if (p) { 10427 const struct net_device_core_stats *core_stats; 10428 int i; 10429 10430 for_each_possible_cpu(i) { 10431 core_stats = per_cpu_ptr(p, i); 10432 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10433 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10434 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10435 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10436 } 10437 } 10438 return storage; 10439 } 10440 EXPORT_SYMBOL(dev_get_stats); 10441 10442 /** 10443 * dev_fetch_sw_netstats - get per-cpu network device statistics 10444 * @s: place to store stats 10445 * @netstats: per-cpu network stats to read from 10446 * 10447 * Read per-cpu network statistics and populate the related fields in @s. 10448 */ 10449 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10450 const struct pcpu_sw_netstats __percpu *netstats) 10451 { 10452 int cpu; 10453 10454 for_each_possible_cpu(cpu) { 10455 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10456 const struct pcpu_sw_netstats *stats; 10457 unsigned int start; 10458 10459 stats = per_cpu_ptr(netstats, cpu); 10460 do { 10461 start = u64_stats_fetch_begin(&stats->syncp); 10462 rx_packets = u64_stats_read(&stats->rx_packets); 10463 rx_bytes = u64_stats_read(&stats->rx_bytes); 10464 tx_packets = u64_stats_read(&stats->tx_packets); 10465 tx_bytes = u64_stats_read(&stats->tx_bytes); 10466 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10467 10468 s->rx_packets += rx_packets; 10469 s->rx_bytes += rx_bytes; 10470 s->tx_packets += tx_packets; 10471 s->tx_bytes += tx_bytes; 10472 } 10473 } 10474 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10475 10476 /** 10477 * dev_get_tstats64 - ndo_get_stats64 implementation 10478 * @dev: device to get statistics from 10479 * @s: place to store stats 10480 * 10481 * Populate @s from dev->stats and dev->tstats. Can be used as 10482 * ndo_get_stats64() callback. 10483 */ 10484 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10485 { 10486 netdev_stats_to_stats64(s, &dev->stats); 10487 dev_fetch_sw_netstats(s, dev->tstats); 10488 } 10489 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10490 10491 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10492 { 10493 struct netdev_queue *queue = dev_ingress_queue(dev); 10494 10495 #ifdef CONFIG_NET_CLS_ACT 10496 if (queue) 10497 return queue; 10498 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10499 if (!queue) 10500 return NULL; 10501 netdev_init_one_queue(dev, queue, NULL); 10502 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10503 queue->qdisc_sleeping = &noop_qdisc; 10504 rcu_assign_pointer(dev->ingress_queue, queue); 10505 #endif 10506 return queue; 10507 } 10508 10509 static const struct ethtool_ops default_ethtool_ops; 10510 10511 void netdev_set_default_ethtool_ops(struct net_device *dev, 10512 const struct ethtool_ops *ops) 10513 { 10514 if (dev->ethtool_ops == &default_ethtool_ops) 10515 dev->ethtool_ops = ops; 10516 } 10517 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10518 10519 /** 10520 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10521 * @dev: netdev to enable the IRQ coalescing on 10522 * 10523 * Sets a conservative default for SW IRQ coalescing. Users can use 10524 * sysfs attributes to override the default values. 10525 */ 10526 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10527 { 10528 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10529 10530 dev->gro_flush_timeout = 20000; 10531 dev->napi_defer_hard_irqs = 1; 10532 } 10533 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10534 10535 void netdev_freemem(struct net_device *dev) 10536 { 10537 char *addr = (char *)dev - dev->padded; 10538 10539 kvfree(addr); 10540 } 10541 10542 /** 10543 * alloc_netdev_mqs - allocate network device 10544 * @sizeof_priv: size of private data to allocate space for 10545 * @name: device name format string 10546 * @name_assign_type: origin of device name 10547 * @setup: callback to initialize device 10548 * @txqs: the number of TX subqueues to allocate 10549 * @rxqs: the number of RX subqueues to allocate 10550 * 10551 * Allocates a struct net_device with private data area for driver use 10552 * and performs basic initialization. Also allocates subqueue structs 10553 * for each queue on the device. 10554 */ 10555 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10556 unsigned char name_assign_type, 10557 void (*setup)(struct net_device *), 10558 unsigned int txqs, unsigned int rxqs) 10559 { 10560 struct net_device *dev; 10561 unsigned int alloc_size; 10562 struct net_device *p; 10563 10564 BUG_ON(strlen(name) >= sizeof(dev->name)); 10565 10566 if (txqs < 1) { 10567 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10568 return NULL; 10569 } 10570 10571 if (rxqs < 1) { 10572 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10573 return NULL; 10574 } 10575 10576 alloc_size = sizeof(struct net_device); 10577 if (sizeof_priv) { 10578 /* ensure 32-byte alignment of private area */ 10579 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10580 alloc_size += sizeof_priv; 10581 } 10582 /* ensure 32-byte alignment of whole construct */ 10583 alloc_size += NETDEV_ALIGN - 1; 10584 10585 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10586 if (!p) 10587 return NULL; 10588 10589 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10590 dev->padded = (char *)dev - (char *)p; 10591 10592 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10593 #ifdef CONFIG_PCPU_DEV_REFCNT 10594 dev->pcpu_refcnt = alloc_percpu(int); 10595 if (!dev->pcpu_refcnt) 10596 goto free_dev; 10597 __dev_hold(dev); 10598 #else 10599 refcount_set(&dev->dev_refcnt, 1); 10600 #endif 10601 10602 if (dev_addr_init(dev)) 10603 goto free_pcpu; 10604 10605 dev_mc_init(dev); 10606 dev_uc_init(dev); 10607 10608 dev_net_set(dev, &init_net); 10609 10610 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10611 dev->gso_max_segs = GSO_MAX_SEGS; 10612 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10613 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10614 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10615 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10616 dev->tso_max_segs = TSO_MAX_SEGS; 10617 dev->upper_level = 1; 10618 dev->lower_level = 1; 10619 #ifdef CONFIG_LOCKDEP 10620 dev->nested_level = 0; 10621 INIT_LIST_HEAD(&dev->unlink_list); 10622 #endif 10623 10624 INIT_LIST_HEAD(&dev->napi_list); 10625 INIT_LIST_HEAD(&dev->unreg_list); 10626 INIT_LIST_HEAD(&dev->close_list); 10627 INIT_LIST_HEAD(&dev->link_watch_list); 10628 INIT_LIST_HEAD(&dev->adj_list.upper); 10629 INIT_LIST_HEAD(&dev->adj_list.lower); 10630 INIT_LIST_HEAD(&dev->ptype_all); 10631 INIT_LIST_HEAD(&dev->ptype_specific); 10632 INIT_LIST_HEAD(&dev->net_notifier_list); 10633 #ifdef CONFIG_NET_SCHED 10634 hash_init(dev->qdisc_hash); 10635 #endif 10636 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10637 setup(dev); 10638 10639 if (!dev->tx_queue_len) { 10640 dev->priv_flags |= IFF_NO_QUEUE; 10641 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10642 } 10643 10644 dev->num_tx_queues = txqs; 10645 dev->real_num_tx_queues = txqs; 10646 if (netif_alloc_netdev_queues(dev)) 10647 goto free_all; 10648 10649 dev->num_rx_queues = rxqs; 10650 dev->real_num_rx_queues = rxqs; 10651 if (netif_alloc_rx_queues(dev)) 10652 goto free_all; 10653 10654 strcpy(dev->name, name); 10655 dev->name_assign_type = name_assign_type; 10656 dev->group = INIT_NETDEV_GROUP; 10657 if (!dev->ethtool_ops) 10658 dev->ethtool_ops = &default_ethtool_ops; 10659 10660 nf_hook_netdev_init(dev); 10661 10662 return dev; 10663 10664 free_all: 10665 free_netdev(dev); 10666 return NULL; 10667 10668 free_pcpu: 10669 #ifdef CONFIG_PCPU_DEV_REFCNT 10670 free_percpu(dev->pcpu_refcnt); 10671 free_dev: 10672 #endif 10673 netdev_freemem(dev); 10674 return NULL; 10675 } 10676 EXPORT_SYMBOL(alloc_netdev_mqs); 10677 10678 /** 10679 * free_netdev - free network device 10680 * @dev: device 10681 * 10682 * This function does the last stage of destroying an allocated device 10683 * interface. The reference to the device object is released. If this 10684 * is the last reference then it will be freed.Must be called in process 10685 * context. 10686 */ 10687 void free_netdev(struct net_device *dev) 10688 { 10689 struct napi_struct *p, *n; 10690 10691 might_sleep(); 10692 10693 /* When called immediately after register_netdevice() failed the unwind 10694 * handling may still be dismantling the device. Handle that case by 10695 * deferring the free. 10696 */ 10697 if (dev->reg_state == NETREG_UNREGISTERING) { 10698 ASSERT_RTNL(); 10699 dev->needs_free_netdev = true; 10700 return; 10701 } 10702 10703 netif_free_tx_queues(dev); 10704 netif_free_rx_queues(dev); 10705 10706 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10707 10708 /* Flush device addresses */ 10709 dev_addr_flush(dev); 10710 10711 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10712 netif_napi_del(p); 10713 10714 ref_tracker_dir_exit(&dev->refcnt_tracker); 10715 #ifdef CONFIG_PCPU_DEV_REFCNT 10716 free_percpu(dev->pcpu_refcnt); 10717 dev->pcpu_refcnt = NULL; 10718 #endif 10719 free_percpu(dev->core_stats); 10720 dev->core_stats = NULL; 10721 free_percpu(dev->xdp_bulkq); 10722 dev->xdp_bulkq = NULL; 10723 10724 /* Compatibility with error handling in drivers */ 10725 if (dev->reg_state == NETREG_UNINITIALIZED) { 10726 netdev_freemem(dev); 10727 return; 10728 } 10729 10730 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10731 dev->reg_state = NETREG_RELEASED; 10732 10733 /* will free via device release */ 10734 put_device(&dev->dev); 10735 } 10736 EXPORT_SYMBOL(free_netdev); 10737 10738 /** 10739 * synchronize_net - Synchronize with packet receive processing 10740 * 10741 * Wait for packets currently being received to be done. 10742 * Does not block later packets from starting. 10743 */ 10744 void synchronize_net(void) 10745 { 10746 might_sleep(); 10747 if (rtnl_is_locked()) 10748 synchronize_rcu_expedited(); 10749 else 10750 synchronize_rcu(); 10751 } 10752 EXPORT_SYMBOL(synchronize_net); 10753 10754 /** 10755 * unregister_netdevice_queue - remove device from the kernel 10756 * @dev: device 10757 * @head: list 10758 * 10759 * This function shuts down a device interface and removes it 10760 * from the kernel tables. 10761 * If head not NULL, device is queued to be unregistered later. 10762 * 10763 * Callers must hold the rtnl semaphore. You may want 10764 * unregister_netdev() instead of this. 10765 */ 10766 10767 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10768 { 10769 ASSERT_RTNL(); 10770 10771 if (head) { 10772 list_move_tail(&dev->unreg_list, head); 10773 } else { 10774 LIST_HEAD(single); 10775 10776 list_add(&dev->unreg_list, &single); 10777 unregister_netdevice_many(&single); 10778 } 10779 } 10780 EXPORT_SYMBOL(unregister_netdevice_queue); 10781 10782 void unregister_netdevice_many_notify(struct list_head *head, 10783 u32 portid, const struct nlmsghdr *nlh) 10784 { 10785 struct net_device *dev, *tmp; 10786 LIST_HEAD(close_head); 10787 10788 BUG_ON(dev_boot_phase); 10789 ASSERT_RTNL(); 10790 10791 if (list_empty(head)) 10792 return; 10793 10794 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10795 /* Some devices call without registering 10796 * for initialization unwind. Remove those 10797 * devices and proceed with the remaining. 10798 */ 10799 if (dev->reg_state == NETREG_UNINITIALIZED) { 10800 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10801 dev->name, dev); 10802 10803 WARN_ON(1); 10804 list_del(&dev->unreg_list); 10805 continue; 10806 } 10807 dev->dismantle = true; 10808 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10809 } 10810 10811 /* If device is running, close it first. */ 10812 list_for_each_entry(dev, head, unreg_list) 10813 list_add_tail(&dev->close_list, &close_head); 10814 dev_close_many(&close_head, true); 10815 10816 list_for_each_entry(dev, head, unreg_list) { 10817 /* And unlink it from device chain. */ 10818 write_lock(&dev_base_lock); 10819 unlist_netdevice(dev, false); 10820 dev->reg_state = NETREG_UNREGISTERING; 10821 write_unlock(&dev_base_lock); 10822 } 10823 flush_all_backlogs(); 10824 10825 synchronize_net(); 10826 10827 list_for_each_entry(dev, head, unreg_list) { 10828 struct sk_buff *skb = NULL; 10829 10830 /* Shutdown queueing discipline. */ 10831 dev_shutdown(dev); 10832 10833 dev_xdp_uninstall(dev); 10834 bpf_dev_bound_netdev_unregister(dev); 10835 10836 netdev_offload_xstats_disable_all(dev); 10837 10838 /* Notify protocols, that we are about to destroy 10839 * this device. They should clean all the things. 10840 */ 10841 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10842 10843 if (!dev->rtnl_link_ops || 10844 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10845 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10846 GFP_KERNEL, NULL, 0, 10847 portid, nlmsg_seq(nlh)); 10848 10849 /* 10850 * Flush the unicast and multicast chains 10851 */ 10852 dev_uc_flush(dev); 10853 dev_mc_flush(dev); 10854 10855 netdev_name_node_alt_flush(dev); 10856 netdev_name_node_free(dev->name_node); 10857 10858 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10859 10860 if (dev->netdev_ops->ndo_uninit) 10861 dev->netdev_ops->ndo_uninit(dev); 10862 10863 if (skb) 10864 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10865 10866 /* Notifier chain MUST detach us all upper devices. */ 10867 WARN_ON(netdev_has_any_upper_dev(dev)); 10868 WARN_ON(netdev_has_any_lower_dev(dev)); 10869 10870 /* Remove entries from kobject tree */ 10871 netdev_unregister_kobject(dev); 10872 #ifdef CONFIG_XPS 10873 /* Remove XPS queueing entries */ 10874 netif_reset_xps_queues_gt(dev, 0); 10875 #endif 10876 } 10877 10878 synchronize_net(); 10879 10880 list_for_each_entry(dev, head, unreg_list) { 10881 netdev_put(dev, &dev->dev_registered_tracker); 10882 net_set_todo(dev); 10883 } 10884 10885 list_del(head); 10886 } 10887 10888 /** 10889 * unregister_netdevice_many - unregister many devices 10890 * @head: list of devices 10891 * 10892 * Note: As most callers use a stack allocated list_head, 10893 * we force a list_del() to make sure stack wont be corrupted later. 10894 */ 10895 void unregister_netdevice_many(struct list_head *head) 10896 { 10897 unregister_netdevice_many_notify(head, 0, NULL); 10898 } 10899 EXPORT_SYMBOL(unregister_netdevice_many); 10900 10901 /** 10902 * unregister_netdev - remove device from the kernel 10903 * @dev: device 10904 * 10905 * This function shuts down a device interface and removes it 10906 * from the kernel tables. 10907 * 10908 * This is just a wrapper for unregister_netdevice that takes 10909 * the rtnl semaphore. In general you want to use this and not 10910 * unregister_netdevice. 10911 */ 10912 void unregister_netdev(struct net_device *dev) 10913 { 10914 rtnl_lock(); 10915 unregister_netdevice(dev); 10916 rtnl_unlock(); 10917 } 10918 EXPORT_SYMBOL(unregister_netdev); 10919 10920 /** 10921 * __dev_change_net_namespace - move device to different nethost namespace 10922 * @dev: device 10923 * @net: network namespace 10924 * @pat: If not NULL name pattern to try if the current device name 10925 * is already taken in the destination network namespace. 10926 * @new_ifindex: If not zero, specifies device index in the target 10927 * namespace. 10928 * 10929 * This function shuts down a device interface and moves it 10930 * to a new network namespace. On success 0 is returned, on 10931 * a failure a netagive errno code is returned. 10932 * 10933 * Callers must hold the rtnl semaphore. 10934 */ 10935 10936 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10937 const char *pat, int new_ifindex) 10938 { 10939 struct net *net_old = dev_net(dev); 10940 int err, new_nsid; 10941 10942 ASSERT_RTNL(); 10943 10944 /* Don't allow namespace local devices to be moved. */ 10945 err = -EINVAL; 10946 if (dev->features & NETIF_F_NETNS_LOCAL) 10947 goto out; 10948 10949 /* Ensure the device has been registrered */ 10950 if (dev->reg_state != NETREG_REGISTERED) 10951 goto out; 10952 10953 /* Get out if there is nothing todo */ 10954 err = 0; 10955 if (net_eq(net_old, net)) 10956 goto out; 10957 10958 /* Pick the destination device name, and ensure 10959 * we can use it in the destination network namespace. 10960 */ 10961 err = -EEXIST; 10962 if (netdev_name_in_use(net, dev->name)) { 10963 /* We get here if we can't use the current device name */ 10964 if (!pat) 10965 goto out; 10966 err = dev_get_valid_name(net, dev, pat); 10967 if (err < 0) 10968 goto out; 10969 } 10970 10971 /* Check that new_ifindex isn't used yet. */ 10972 err = -EBUSY; 10973 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10974 goto out; 10975 10976 /* 10977 * And now a mini version of register_netdevice unregister_netdevice. 10978 */ 10979 10980 /* If device is running close it first. */ 10981 dev_close(dev); 10982 10983 /* And unlink it from device chain */ 10984 unlist_netdevice(dev, true); 10985 10986 synchronize_net(); 10987 10988 /* Shutdown queueing discipline. */ 10989 dev_shutdown(dev); 10990 10991 /* Notify protocols, that we are about to destroy 10992 * this device. They should clean all the things. 10993 * 10994 * Note that dev->reg_state stays at NETREG_REGISTERED. 10995 * This is wanted because this way 8021q and macvlan know 10996 * the device is just moving and can keep their slaves up. 10997 */ 10998 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10999 rcu_barrier(); 11000 11001 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11002 /* If there is an ifindex conflict assign a new one */ 11003 if (!new_ifindex) { 11004 if (__dev_get_by_index(net, dev->ifindex)) 11005 new_ifindex = dev_new_index(net); 11006 else 11007 new_ifindex = dev->ifindex; 11008 } 11009 11010 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11011 new_ifindex); 11012 11013 /* 11014 * Flush the unicast and multicast chains 11015 */ 11016 dev_uc_flush(dev); 11017 dev_mc_flush(dev); 11018 11019 /* Send a netdev-removed uevent to the old namespace */ 11020 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11021 netdev_adjacent_del_links(dev); 11022 11023 /* Move per-net netdevice notifiers that are following the netdevice */ 11024 move_netdevice_notifiers_dev_net(dev, net); 11025 11026 /* Actually switch the network namespace */ 11027 dev_net_set(dev, net); 11028 dev->ifindex = new_ifindex; 11029 11030 /* Send a netdev-add uevent to the new namespace */ 11031 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11032 netdev_adjacent_add_links(dev); 11033 11034 /* Fixup kobjects */ 11035 err = device_rename(&dev->dev, dev->name); 11036 WARN_ON(err); 11037 11038 /* Adapt owner in case owning user namespace of target network 11039 * namespace is different from the original one. 11040 */ 11041 err = netdev_change_owner(dev, net_old, net); 11042 WARN_ON(err); 11043 11044 /* Add the device back in the hashes */ 11045 list_netdevice(dev); 11046 11047 /* Notify protocols, that a new device appeared. */ 11048 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11049 11050 /* 11051 * Prevent userspace races by waiting until the network 11052 * device is fully setup before sending notifications. 11053 */ 11054 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11055 11056 synchronize_net(); 11057 err = 0; 11058 out: 11059 return err; 11060 } 11061 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11062 11063 static int dev_cpu_dead(unsigned int oldcpu) 11064 { 11065 struct sk_buff **list_skb; 11066 struct sk_buff *skb; 11067 unsigned int cpu; 11068 struct softnet_data *sd, *oldsd, *remsd = NULL; 11069 11070 local_irq_disable(); 11071 cpu = smp_processor_id(); 11072 sd = &per_cpu(softnet_data, cpu); 11073 oldsd = &per_cpu(softnet_data, oldcpu); 11074 11075 /* Find end of our completion_queue. */ 11076 list_skb = &sd->completion_queue; 11077 while (*list_skb) 11078 list_skb = &(*list_skb)->next; 11079 /* Append completion queue from offline CPU. */ 11080 *list_skb = oldsd->completion_queue; 11081 oldsd->completion_queue = NULL; 11082 11083 /* Append output queue from offline CPU. */ 11084 if (oldsd->output_queue) { 11085 *sd->output_queue_tailp = oldsd->output_queue; 11086 sd->output_queue_tailp = oldsd->output_queue_tailp; 11087 oldsd->output_queue = NULL; 11088 oldsd->output_queue_tailp = &oldsd->output_queue; 11089 } 11090 /* Append NAPI poll list from offline CPU, with one exception : 11091 * process_backlog() must be called by cpu owning percpu backlog. 11092 * We properly handle process_queue & input_pkt_queue later. 11093 */ 11094 while (!list_empty(&oldsd->poll_list)) { 11095 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11096 struct napi_struct, 11097 poll_list); 11098 11099 list_del_init(&napi->poll_list); 11100 if (napi->poll == process_backlog) 11101 napi->state = 0; 11102 else 11103 ____napi_schedule(sd, napi); 11104 } 11105 11106 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11107 local_irq_enable(); 11108 11109 #ifdef CONFIG_RPS 11110 remsd = oldsd->rps_ipi_list; 11111 oldsd->rps_ipi_list = NULL; 11112 #endif 11113 /* send out pending IPI's on offline CPU */ 11114 net_rps_send_ipi(remsd); 11115 11116 /* Process offline CPU's input_pkt_queue */ 11117 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11118 netif_rx(skb); 11119 input_queue_head_incr(oldsd); 11120 } 11121 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11122 netif_rx(skb); 11123 input_queue_head_incr(oldsd); 11124 } 11125 11126 return 0; 11127 } 11128 11129 /** 11130 * netdev_increment_features - increment feature set by one 11131 * @all: current feature set 11132 * @one: new feature set 11133 * @mask: mask feature set 11134 * 11135 * Computes a new feature set after adding a device with feature set 11136 * @one to the master device with current feature set @all. Will not 11137 * enable anything that is off in @mask. Returns the new feature set. 11138 */ 11139 netdev_features_t netdev_increment_features(netdev_features_t all, 11140 netdev_features_t one, netdev_features_t mask) 11141 { 11142 if (mask & NETIF_F_HW_CSUM) 11143 mask |= NETIF_F_CSUM_MASK; 11144 mask |= NETIF_F_VLAN_CHALLENGED; 11145 11146 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11147 all &= one | ~NETIF_F_ALL_FOR_ALL; 11148 11149 /* If one device supports hw checksumming, set for all. */ 11150 if (all & NETIF_F_HW_CSUM) 11151 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11152 11153 return all; 11154 } 11155 EXPORT_SYMBOL(netdev_increment_features); 11156 11157 static struct hlist_head * __net_init netdev_create_hash(void) 11158 { 11159 int i; 11160 struct hlist_head *hash; 11161 11162 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11163 if (hash != NULL) 11164 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11165 INIT_HLIST_HEAD(&hash[i]); 11166 11167 return hash; 11168 } 11169 11170 /* Initialize per network namespace state */ 11171 static int __net_init netdev_init(struct net *net) 11172 { 11173 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11174 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11175 11176 INIT_LIST_HEAD(&net->dev_base_head); 11177 11178 net->dev_name_head = netdev_create_hash(); 11179 if (net->dev_name_head == NULL) 11180 goto err_name; 11181 11182 net->dev_index_head = netdev_create_hash(); 11183 if (net->dev_index_head == NULL) 11184 goto err_idx; 11185 11186 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11187 11188 return 0; 11189 11190 err_idx: 11191 kfree(net->dev_name_head); 11192 err_name: 11193 return -ENOMEM; 11194 } 11195 11196 /** 11197 * netdev_drivername - network driver for the device 11198 * @dev: network device 11199 * 11200 * Determine network driver for device. 11201 */ 11202 const char *netdev_drivername(const struct net_device *dev) 11203 { 11204 const struct device_driver *driver; 11205 const struct device *parent; 11206 const char *empty = ""; 11207 11208 parent = dev->dev.parent; 11209 if (!parent) 11210 return empty; 11211 11212 driver = parent->driver; 11213 if (driver && driver->name) 11214 return driver->name; 11215 return empty; 11216 } 11217 11218 static void __netdev_printk(const char *level, const struct net_device *dev, 11219 struct va_format *vaf) 11220 { 11221 if (dev && dev->dev.parent) { 11222 dev_printk_emit(level[1] - '0', 11223 dev->dev.parent, 11224 "%s %s %s%s: %pV", 11225 dev_driver_string(dev->dev.parent), 11226 dev_name(dev->dev.parent), 11227 netdev_name(dev), netdev_reg_state(dev), 11228 vaf); 11229 } else if (dev) { 11230 printk("%s%s%s: %pV", 11231 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11232 } else { 11233 printk("%s(NULL net_device): %pV", level, vaf); 11234 } 11235 } 11236 11237 void netdev_printk(const char *level, const struct net_device *dev, 11238 const char *format, ...) 11239 { 11240 struct va_format vaf; 11241 va_list args; 11242 11243 va_start(args, format); 11244 11245 vaf.fmt = format; 11246 vaf.va = &args; 11247 11248 __netdev_printk(level, dev, &vaf); 11249 11250 va_end(args); 11251 } 11252 EXPORT_SYMBOL(netdev_printk); 11253 11254 #define define_netdev_printk_level(func, level) \ 11255 void func(const struct net_device *dev, const char *fmt, ...) \ 11256 { \ 11257 struct va_format vaf; \ 11258 va_list args; \ 11259 \ 11260 va_start(args, fmt); \ 11261 \ 11262 vaf.fmt = fmt; \ 11263 vaf.va = &args; \ 11264 \ 11265 __netdev_printk(level, dev, &vaf); \ 11266 \ 11267 va_end(args); \ 11268 } \ 11269 EXPORT_SYMBOL(func); 11270 11271 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11272 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11273 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11274 define_netdev_printk_level(netdev_err, KERN_ERR); 11275 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11276 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11277 define_netdev_printk_level(netdev_info, KERN_INFO); 11278 11279 static void __net_exit netdev_exit(struct net *net) 11280 { 11281 kfree(net->dev_name_head); 11282 kfree(net->dev_index_head); 11283 if (net != &init_net) 11284 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11285 } 11286 11287 static struct pernet_operations __net_initdata netdev_net_ops = { 11288 .init = netdev_init, 11289 .exit = netdev_exit, 11290 }; 11291 11292 static void __net_exit default_device_exit_net(struct net *net) 11293 { 11294 struct net_device *dev, *aux; 11295 /* 11296 * Push all migratable network devices back to the 11297 * initial network namespace 11298 */ 11299 ASSERT_RTNL(); 11300 for_each_netdev_safe(net, dev, aux) { 11301 int err; 11302 char fb_name[IFNAMSIZ]; 11303 11304 /* Ignore unmoveable devices (i.e. loopback) */ 11305 if (dev->features & NETIF_F_NETNS_LOCAL) 11306 continue; 11307 11308 /* Leave virtual devices for the generic cleanup */ 11309 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11310 continue; 11311 11312 /* Push remaining network devices to init_net */ 11313 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11314 if (netdev_name_in_use(&init_net, fb_name)) 11315 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11316 err = dev_change_net_namespace(dev, &init_net, fb_name); 11317 if (err) { 11318 pr_emerg("%s: failed to move %s to init_net: %d\n", 11319 __func__, dev->name, err); 11320 BUG(); 11321 } 11322 } 11323 } 11324 11325 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11326 { 11327 /* At exit all network devices most be removed from a network 11328 * namespace. Do this in the reverse order of registration. 11329 * Do this across as many network namespaces as possible to 11330 * improve batching efficiency. 11331 */ 11332 struct net_device *dev; 11333 struct net *net; 11334 LIST_HEAD(dev_kill_list); 11335 11336 rtnl_lock(); 11337 list_for_each_entry(net, net_list, exit_list) { 11338 default_device_exit_net(net); 11339 cond_resched(); 11340 } 11341 11342 list_for_each_entry(net, net_list, exit_list) { 11343 for_each_netdev_reverse(net, dev) { 11344 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11345 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11346 else 11347 unregister_netdevice_queue(dev, &dev_kill_list); 11348 } 11349 } 11350 unregister_netdevice_many(&dev_kill_list); 11351 rtnl_unlock(); 11352 } 11353 11354 static struct pernet_operations __net_initdata default_device_ops = { 11355 .exit_batch = default_device_exit_batch, 11356 }; 11357 11358 /* 11359 * Initialize the DEV module. At boot time this walks the device list and 11360 * unhooks any devices that fail to initialise (normally hardware not 11361 * present) and leaves us with a valid list of present and active devices. 11362 * 11363 */ 11364 11365 /* 11366 * This is called single threaded during boot, so no need 11367 * to take the rtnl semaphore. 11368 */ 11369 static int __init net_dev_init(void) 11370 { 11371 int i, rc = -ENOMEM; 11372 11373 BUG_ON(!dev_boot_phase); 11374 11375 if (dev_proc_init()) 11376 goto out; 11377 11378 if (netdev_kobject_init()) 11379 goto out; 11380 11381 INIT_LIST_HEAD(&ptype_all); 11382 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11383 INIT_LIST_HEAD(&ptype_base[i]); 11384 11385 if (register_pernet_subsys(&netdev_net_ops)) 11386 goto out; 11387 11388 /* 11389 * Initialise the packet receive queues. 11390 */ 11391 11392 for_each_possible_cpu(i) { 11393 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11394 struct softnet_data *sd = &per_cpu(softnet_data, i); 11395 11396 INIT_WORK(flush, flush_backlog); 11397 11398 skb_queue_head_init(&sd->input_pkt_queue); 11399 skb_queue_head_init(&sd->process_queue); 11400 #ifdef CONFIG_XFRM_OFFLOAD 11401 skb_queue_head_init(&sd->xfrm_backlog); 11402 #endif 11403 INIT_LIST_HEAD(&sd->poll_list); 11404 sd->output_queue_tailp = &sd->output_queue; 11405 #ifdef CONFIG_RPS 11406 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11407 sd->cpu = i; 11408 #endif 11409 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11410 spin_lock_init(&sd->defer_lock); 11411 11412 init_gro_hash(&sd->backlog); 11413 sd->backlog.poll = process_backlog; 11414 sd->backlog.weight = weight_p; 11415 } 11416 11417 dev_boot_phase = 0; 11418 11419 /* The loopback device is special if any other network devices 11420 * is present in a network namespace the loopback device must 11421 * be present. Since we now dynamically allocate and free the 11422 * loopback device ensure this invariant is maintained by 11423 * keeping the loopback device as the first device on the 11424 * list of network devices. Ensuring the loopback devices 11425 * is the first device that appears and the last network device 11426 * that disappears. 11427 */ 11428 if (register_pernet_device(&loopback_net_ops)) 11429 goto out; 11430 11431 if (register_pernet_device(&default_device_ops)) 11432 goto out; 11433 11434 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11435 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11436 11437 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11438 NULL, dev_cpu_dead); 11439 WARN_ON(rc < 0); 11440 rc = 0; 11441 out: 11442 return rc; 11443 } 11444 11445 subsys_initcall(net_dev_init); 11446