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 if (unlikely(reason == SKB_REASON_DROPPED)) 3138 kfree_skb(skb); 3139 else 3140 consume_skb(skb); 3141 } 3142 EXPORT_SYMBOL(__dev_kfree_skb_any); 3143 3144 3145 /** 3146 * netif_device_detach - mark device as removed 3147 * @dev: network device 3148 * 3149 * Mark device as removed from system and therefore no longer available. 3150 */ 3151 void netif_device_detach(struct net_device *dev) 3152 { 3153 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3154 netif_running(dev)) { 3155 netif_tx_stop_all_queues(dev); 3156 } 3157 } 3158 EXPORT_SYMBOL(netif_device_detach); 3159 3160 /** 3161 * netif_device_attach - mark device as attached 3162 * @dev: network device 3163 * 3164 * Mark device as attached from system and restart if needed. 3165 */ 3166 void netif_device_attach(struct net_device *dev) 3167 { 3168 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3169 netif_running(dev)) { 3170 netif_tx_wake_all_queues(dev); 3171 __netdev_watchdog_up(dev); 3172 } 3173 } 3174 EXPORT_SYMBOL(netif_device_attach); 3175 3176 /* 3177 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3178 * to be used as a distribution range. 3179 */ 3180 static u16 skb_tx_hash(const struct net_device *dev, 3181 const struct net_device *sb_dev, 3182 struct sk_buff *skb) 3183 { 3184 u32 hash; 3185 u16 qoffset = 0; 3186 u16 qcount = dev->real_num_tx_queues; 3187 3188 if (dev->num_tc) { 3189 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3190 3191 qoffset = sb_dev->tc_to_txq[tc].offset; 3192 qcount = sb_dev->tc_to_txq[tc].count; 3193 if (unlikely(!qcount)) { 3194 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3195 sb_dev->name, qoffset, tc); 3196 qoffset = 0; 3197 qcount = dev->real_num_tx_queues; 3198 } 3199 } 3200 3201 if (skb_rx_queue_recorded(skb)) { 3202 hash = skb_get_rx_queue(skb); 3203 if (hash >= qoffset) 3204 hash -= qoffset; 3205 while (unlikely(hash >= qcount)) 3206 hash -= qcount; 3207 return hash + qoffset; 3208 } 3209 3210 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3211 } 3212 3213 static void skb_warn_bad_offload(const struct sk_buff *skb) 3214 { 3215 static const netdev_features_t null_features; 3216 struct net_device *dev = skb->dev; 3217 const char *name = ""; 3218 3219 if (!net_ratelimit()) 3220 return; 3221 3222 if (dev) { 3223 if (dev->dev.parent) 3224 name = dev_driver_string(dev->dev.parent); 3225 else 3226 name = netdev_name(dev); 3227 } 3228 skb_dump(KERN_WARNING, skb, false); 3229 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3230 name, dev ? &dev->features : &null_features, 3231 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3232 } 3233 3234 /* 3235 * Invalidate hardware checksum when packet is to be mangled, and 3236 * complete checksum manually on outgoing path. 3237 */ 3238 int skb_checksum_help(struct sk_buff *skb) 3239 { 3240 __wsum csum; 3241 int ret = 0, offset; 3242 3243 if (skb->ip_summed == CHECKSUM_COMPLETE) 3244 goto out_set_summed; 3245 3246 if (unlikely(skb_is_gso(skb))) { 3247 skb_warn_bad_offload(skb); 3248 return -EINVAL; 3249 } 3250 3251 /* Before computing a checksum, we should make sure no frag could 3252 * be modified by an external entity : checksum could be wrong. 3253 */ 3254 if (skb_has_shared_frag(skb)) { 3255 ret = __skb_linearize(skb); 3256 if (ret) 3257 goto out; 3258 } 3259 3260 offset = skb_checksum_start_offset(skb); 3261 ret = -EINVAL; 3262 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3263 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3264 goto out; 3265 } 3266 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3267 3268 offset += skb->csum_offset; 3269 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3270 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3271 goto out; 3272 } 3273 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3274 if (ret) 3275 goto out; 3276 3277 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3278 out_set_summed: 3279 skb->ip_summed = CHECKSUM_NONE; 3280 out: 3281 return ret; 3282 } 3283 EXPORT_SYMBOL(skb_checksum_help); 3284 3285 int skb_crc32c_csum_help(struct sk_buff *skb) 3286 { 3287 __le32 crc32c_csum; 3288 int ret = 0, offset, start; 3289 3290 if (skb->ip_summed != CHECKSUM_PARTIAL) 3291 goto out; 3292 3293 if (unlikely(skb_is_gso(skb))) 3294 goto out; 3295 3296 /* Before computing a checksum, we should make sure no frag could 3297 * be modified by an external entity : checksum could be wrong. 3298 */ 3299 if (unlikely(skb_has_shared_frag(skb))) { 3300 ret = __skb_linearize(skb); 3301 if (ret) 3302 goto out; 3303 } 3304 start = skb_checksum_start_offset(skb); 3305 offset = start + offsetof(struct sctphdr, checksum); 3306 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3307 ret = -EINVAL; 3308 goto out; 3309 } 3310 3311 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3312 if (ret) 3313 goto out; 3314 3315 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3316 skb->len - start, ~(__u32)0, 3317 crc32c_csum_stub)); 3318 *(__le32 *)(skb->data + offset) = crc32c_csum; 3319 skb->ip_summed = CHECKSUM_NONE; 3320 skb->csum_not_inet = 0; 3321 out: 3322 return ret; 3323 } 3324 3325 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3326 { 3327 __be16 type = skb->protocol; 3328 3329 /* Tunnel gso handlers can set protocol to ethernet. */ 3330 if (type == htons(ETH_P_TEB)) { 3331 struct ethhdr *eth; 3332 3333 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3334 return 0; 3335 3336 eth = (struct ethhdr *)skb->data; 3337 type = eth->h_proto; 3338 } 3339 3340 return __vlan_get_protocol(skb, type, depth); 3341 } 3342 3343 /* openvswitch calls this on rx path, so we need a different check. 3344 */ 3345 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3346 { 3347 if (tx_path) 3348 return skb->ip_summed != CHECKSUM_PARTIAL && 3349 skb->ip_summed != CHECKSUM_UNNECESSARY; 3350 3351 return skb->ip_summed == CHECKSUM_NONE; 3352 } 3353 3354 /** 3355 * __skb_gso_segment - Perform segmentation on skb. 3356 * @skb: buffer to segment 3357 * @features: features for the output path (see dev->features) 3358 * @tx_path: whether it is called in TX path 3359 * 3360 * This function segments the given skb and returns a list of segments. 3361 * 3362 * It may return NULL if the skb requires no segmentation. This is 3363 * only possible when GSO is used for verifying header integrity. 3364 * 3365 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3366 */ 3367 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3368 netdev_features_t features, bool tx_path) 3369 { 3370 struct sk_buff *segs; 3371 3372 if (unlikely(skb_needs_check(skb, tx_path))) { 3373 int err; 3374 3375 /* We're going to init ->check field in TCP or UDP header */ 3376 err = skb_cow_head(skb, 0); 3377 if (err < 0) 3378 return ERR_PTR(err); 3379 } 3380 3381 /* Only report GSO partial support if it will enable us to 3382 * support segmentation on this frame without needing additional 3383 * work. 3384 */ 3385 if (features & NETIF_F_GSO_PARTIAL) { 3386 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3387 struct net_device *dev = skb->dev; 3388 3389 partial_features |= dev->features & dev->gso_partial_features; 3390 if (!skb_gso_ok(skb, features | partial_features)) 3391 features &= ~NETIF_F_GSO_PARTIAL; 3392 } 3393 3394 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3395 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3396 3397 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3398 SKB_GSO_CB(skb)->encap_level = 0; 3399 3400 skb_reset_mac_header(skb); 3401 skb_reset_mac_len(skb); 3402 3403 segs = skb_mac_gso_segment(skb, features); 3404 3405 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3406 skb_warn_bad_offload(skb); 3407 3408 return segs; 3409 } 3410 EXPORT_SYMBOL(__skb_gso_segment); 3411 3412 /* Take action when hardware reception checksum errors are detected. */ 3413 #ifdef CONFIG_BUG 3414 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3415 { 3416 netdev_err(dev, "hw csum failure\n"); 3417 skb_dump(KERN_ERR, skb, true); 3418 dump_stack(); 3419 } 3420 3421 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3422 { 3423 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3424 } 3425 EXPORT_SYMBOL(netdev_rx_csum_fault); 3426 #endif 3427 3428 /* XXX: check that highmem exists at all on the given machine. */ 3429 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3430 { 3431 #ifdef CONFIG_HIGHMEM 3432 int i; 3433 3434 if (!(dev->features & NETIF_F_HIGHDMA)) { 3435 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3436 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3437 3438 if (PageHighMem(skb_frag_page(frag))) 3439 return 1; 3440 } 3441 } 3442 #endif 3443 return 0; 3444 } 3445 3446 /* If MPLS offload request, verify we are testing hardware MPLS features 3447 * instead of standard features for the netdev. 3448 */ 3449 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3450 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3451 netdev_features_t features, 3452 __be16 type) 3453 { 3454 if (eth_p_mpls(type)) 3455 features &= skb->dev->mpls_features; 3456 3457 return features; 3458 } 3459 #else 3460 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3461 netdev_features_t features, 3462 __be16 type) 3463 { 3464 return features; 3465 } 3466 #endif 3467 3468 static netdev_features_t harmonize_features(struct sk_buff *skb, 3469 netdev_features_t features) 3470 { 3471 __be16 type; 3472 3473 type = skb_network_protocol(skb, NULL); 3474 features = net_mpls_features(skb, features, type); 3475 3476 if (skb->ip_summed != CHECKSUM_NONE && 3477 !can_checksum_protocol(features, type)) { 3478 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3479 } 3480 if (illegal_highdma(skb->dev, skb)) 3481 features &= ~NETIF_F_SG; 3482 3483 return features; 3484 } 3485 3486 netdev_features_t passthru_features_check(struct sk_buff *skb, 3487 struct net_device *dev, 3488 netdev_features_t features) 3489 { 3490 return features; 3491 } 3492 EXPORT_SYMBOL(passthru_features_check); 3493 3494 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3495 struct net_device *dev, 3496 netdev_features_t features) 3497 { 3498 return vlan_features_check(skb, features); 3499 } 3500 3501 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3502 struct net_device *dev, 3503 netdev_features_t features) 3504 { 3505 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3506 3507 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3508 return features & ~NETIF_F_GSO_MASK; 3509 3510 if (!skb_shinfo(skb)->gso_type) { 3511 skb_warn_bad_offload(skb); 3512 return features & ~NETIF_F_GSO_MASK; 3513 } 3514 3515 /* Support for GSO partial features requires software 3516 * intervention before we can actually process the packets 3517 * so we need to strip support for any partial features now 3518 * and we can pull them back in after we have partially 3519 * segmented the frame. 3520 */ 3521 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3522 features &= ~dev->gso_partial_features; 3523 3524 /* Make sure to clear the IPv4 ID mangling feature if the 3525 * IPv4 header has the potential to be fragmented. 3526 */ 3527 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3528 struct iphdr *iph = skb->encapsulation ? 3529 inner_ip_hdr(skb) : ip_hdr(skb); 3530 3531 if (!(iph->frag_off & htons(IP_DF))) 3532 features &= ~NETIF_F_TSO_MANGLEID; 3533 } 3534 3535 return features; 3536 } 3537 3538 netdev_features_t netif_skb_features(struct sk_buff *skb) 3539 { 3540 struct net_device *dev = skb->dev; 3541 netdev_features_t features = dev->features; 3542 3543 if (skb_is_gso(skb)) 3544 features = gso_features_check(skb, dev, features); 3545 3546 /* If encapsulation offload request, verify we are testing 3547 * hardware encapsulation features instead of standard 3548 * features for the netdev 3549 */ 3550 if (skb->encapsulation) 3551 features &= dev->hw_enc_features; 3552 3553 if (skb_vlan_tagged(skb)) 3554 features = netdev_intersect_features(features, 3555 dev->vlan_features | 3556 NETIF_F_HW_VLAN_CTAG_TX | 3557 NETIF_F_HW_VLAN_STAG_TX); 3558 3559 if (dev->netdev_ops->ndo_features_check) 3560 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3561 features); 3562 else 3563 features &= dflt_features_check(skb, dev, features); 3564 3565 return harmonize_features(skb, features); 3566 } 3567 EXPORT_SYMBOL(netif_skb_features); 3568 3569 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3570 struct netdev_queue *txq, bool more) 3571 { 3572 unsigned int len; 3573 int rc; 3574 3575 if (dev_nit_active(dev)) 3576 dev_queue_xmit_nit(skb, dev); 3577 3578 len = skb->len; 3579 trace_net_dev_start_xmit(skb, dev); 3580 rc = netdev_start_xmit(skb, dev, txq, more); 3581 trace_net_dev_xmit(skb, rc, dev, len); 3582 3583 return rc; 3584 } 3585 3586 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3587 struct netdev_queue *txq, int *ret) 3588 { 3589 struct sk_buff *skb = first; 3590 int rc = NETDEV_TX_OK; 3591 3592 while (skb) { 3593 struct sk_buff *next = skb->next; 3594 3595 skb_mark_not_on_list(skb); 3596 rc = xmit_one(skb, dev, txq, next != NULL); 3597 if (unlikely(!dev_xmit_complete(rc))) { 3598 skb->next = next; 3599 goto out; 3600 } 3601 3602 skb = next; 3603 if (netif_tx_queue_stopped(txq) && skb) { 3604 rc = NETDEV_TX_BUSY; 3605 break; 3606 } 3607 } 3608 3609 out: 3610 *ret = rc; 3611 return skb; 3612 } 3613 3614 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3615 netdev_features_t features) 3616 { 3617 if (skb_vlan_tag_present(skb) && 3618 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3619 skb = __vlan_hwaccel_push_inside(skb); 3620 return skb; 3621 } 3622 3623 int skb_csum_hwoffload_help(struct sk_buff *skb, 3624 const netdev_features_t features) 3625 { 3626 if (unlikely(skb_csum_is_sctp(skb))) 3627 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3628 skb_crc32c_csum_help(skb); 3629 3630 if (features & NETIF_F_HW_CSUM) 3631 return 0; 3632 3633 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3634 switch (skb->csum_offset) { 3635 case offsetof(struct tcphdr, check): 3636 case offsetof(struct udphdr, check): 3637 return 0; 3638 } 3639 } 3640 3641 return skb_checksum_help(skb); 3642 } 3643 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3644 3645 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3646 { 3647 netdev_features_t features; 3648 3649 features = netif_skb_features(skb); 3650 skb = validate_xmit_vlan(skb, features); 3651 if (unlikely(!skb)) 3652 goto out_null; 3653 3654 skb = sk_validate_xmit_skb(skb, dev); 3655 if (unlikely(!skb)) 3656 goto out_null; 3657 3658 if (netif_needs_gso(skb, features)) { 3659 struct sk_buff *segs; 3660 3661 segs = skb_gso_segment(skb, features); 3662 if (IS_ERR(segs)) { 3663 goto out_kfree_skb; 3664 } else if (segs) { 3665 consume_skb(skb); 3666 skb = segs; 3667 } 3668 } else { 3669 if (skb_needs_linearize(skb, features) && 3670 __skb_linearize(skb)) 3671 goto out_kfree_skb; 3672 3673 /* If packet is not checksummed and device does not 3674 * support checksumming for this protocol, complete 3675 * checksumming here. 3676 */ 3677 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3678 if (skb->encapsulation) 3679 skb_set_inner_transport_header(skb, 3680 skb_checksum_start_offset(skb)); 3681 else 3682 skb_set_transport_header(skb, 3683 skb_checksum_start_offset(skb)); 3684 if (skb_csum_hwoffload_help(skb, features)) 3685 goto out_kfree_skb; 3686 } 3687 } 3688 3689 skb = validate_xmit_xfrm(skb, features, again); 3690 3691 return skb; 3692 3693 out_kfree_skb: 3694 kfree_skb(skb); 3695 out_null: 3696 dev_core_stats_tx_dropped_inc(dev); 3697 return NULL; 3698 } 3699 3700 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3701 { 3702 struct sk_buff *next, *head = NULL, *tail; 3703 3704 for (; skb != NULL; skb = next) { 3705 next = skb->next; 3706 skb_mark_not_on_list(skb); 3707 3708 /* in case skb wont be segmented, point to itself */ 3709 skb->prev = skb; 3710 3711 skb = validate_xmit_skb(skb, dev, again); 3712 if (!skb) 3713 continue; 3714 3715 if (!head) 3716 head = skb; 3717 else 3718 tail->next = skb; 3719 /* If skb was segmented, skb->prev points to 3720 * the last segment. If not, it still contains skb. 3721 */ 3722 tail = skb->prev; 3723 } 3724 return head; 3725 } 3726 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3727 3728 static void qdisc_pkt_len_init(struct sk_buff *skb) 3729 { 3730 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3731 3732 qdisc_skb_cb(skb)->pkt_len = skb->len; 3733 3734 /* To get more precise estimation of bytes sent on wire, 3735 * we add to pkt_len the headers size of all segments 3736 */ 3737 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3738 unsigned int hdr_len; 3739 u16 gso_segs = shinfo->gso_segs; 3740 3741 /* mac layer + network layer */ 3742 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3743 3744 /* + transport layer */ 3745 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3746 const struct tcphdr *th; 3747 struct tcphdr _tcphdr; 3748 3749 th = skb_header_pointer(skb, skb_transport_offset(skb), 3750 sizeof(_tcphdr), &_tcphdr); 3751 if (likely(th)) 3752 hdr_len += __tcp_hdrlen(th); 3753 } else { 3754 struct udphdr _udphdr; 3755 3756 if (skb_header_pointer(skb, skb_transport_offset(skb), 3757 sizeof(_udphdr), &_udphdr)) 3758 hdr_len += sizeof(struct udphdr); 3759 } 3760 3761 if (shinfo->gso_type & SKB_GSO_DODGY) 3762 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3763 shinfo->gso_size); 3764 3765 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3766 } 3767 } 3768 3769 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3770 struct sk_buff **to_free, 3771 struct netdev_queue *txq) 3772 { 3773 int rc; 3774 3775 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3776 if (rc == NET_XMIT_SUCCESS) 3777 trace_qdisc_enqueue(q, txq, skb); 3778 return rc; 3779 } 3780 3781 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3782 struct net_device *dev, 3783 struct netdev_queue *txq) 3784 { 3785 spinlock_t *root_lock = qdisc_lock(q); 3786 struct sk_buff *to_free = NULL; 3787 bool contended; 3788 int rc; 3789 3790 qdisc_calculate_pkt_len(skb, q); 3791 3792 if (q->flags & TCQ_F_NOLOCK) { 3793 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3794 qdisc_run_begin(q)) { 3795 /* Retest nolock_qdisc_is_empty() within the protection 3796 * of q->seqlock to protect from racing with requeuing. 3797 */ 3798 if (unlikely(!nolock_qdisc_is_empty(q))) { 3799 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3800 __qdisc_run(q); 3801 qdisc_run_end(q); 3802 3803 goto no_lock_out; 3804 } 3805 3806 qdisc_bstats_cpu_update(q, skb); 3807 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3808 !nolock_qdisc_is_empty(q)) 3809 __qdisc_run(q); 3810 3811 qdisc_run_end(q); 3812 return NET_XMIT_SUCCESS; 3813 } 3814 3815 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3816 qdisc_run(q); 3817 3818 no_lock_out: 3819 if (unlikely(to_free)) 3820 kfree_skb_list_reason(to_free, 3821 SKB_DROP_REASON_QDISC_DROP); 3822 return rc; 3823 } 3824 3825 /* 3826 * Heuristic to force contended enqueues to serialize on a 3827 * separate lock before trying to get qdisc main lock. 3828 * This permits qdisc->running owner to get the lock more 3829 * often and dequeue packets faster. 3830 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3831 * and then other tasks will only enqueue packets. The packets will be 3832 * sent after the qdisc owner is scheduled again. To prevent this 3833 * scenario the task always serialize on the lock. 3834 */ 3835 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3836 if (unlikely(contended)) 3837 spin_lock(&q->busylock); 3838 3839 spin_lock(root_lock); 3840 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3841 __qdisc_drop(skb, &to_free); 3842 rc = NET_XMIT_DROP; 3843 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3844 qdisc_run_begin(q)) { 3845 /* 3846 * This is a work-conserving queue; there are no old skbs 3847 * waiting to be sent out; and the qdisc is not running - 3848 * xmit the skb directly. 3849 */ 3850 3851 qdisc_bstats_update(q, skb); 3852 3853 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3854 if (unlikely(contended)) { 3855 spin_unlock(&q->busylock); 3856 contended = false; 3857 } 3858 __qdisc_run(q); 3859 } 3860 3861 qdisc_run_end(q); 3862 rc = NET_XMIT_SUCCESS; 3863 } else { 3864 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3865 if (qdisc_run_begin(q)) { 3866 if (unlikely(contended)) { 3867 spin_unlock(&q->busylock); 3868 contended = false; 3869 } 3870 __qdisc_run(q); 3871 qdisc_run_end(q); 3872 } 3873 } 3874 spin_unlock(root_lock); 3875 if (unlikely(to_free)) 3876 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3877 if (unlikely(contended)) 3878 spin_unlock(&q->busylock); 3879 return rc; 3880 } 3881 3882 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3883 static void skb_update_prio(struct sk_buff *skb) 3884 { 3885 const struct netprio_map *map; 3886 const struct sock *sk; 3887 unsigned int prioidx; 3888 3889 if (skb->priority) 3890 return; 3891 map = rcu_dereference_bh(skb->dev->priomap); 3892 if (!map) 3893 return; 3894 sk = skb_to_full_sk(skb); 3895 if (!sk) 3896 return; 3897 3898 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3899 3900 if (prioidx < map->priomap_len) 3901 skb->priority = map->priomap[prioidx]; 3902 } 3903 #else 3904 #define skb_update_prio(skb) 3905 #endif 3906 3907 /** 3908 * dev_loopback_xmit - loop back @skb 3909 * @net: network namespace this loopback is happening in 3910 * @sk: sk needed to be a netfilter okfn 3911 * @skb: buffer to transmit 3912 */ 3913 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3914 { 3915 skb_reset_mac_header(skb); 3916 __skb_pull(skb, skb_network_offset(skb)); 3917 skb->pkt_type = PACKET_LOOPBACK; 3918 if (skb->ip_summed == CHECKSUM_NONE) 3919 skb->ip_summed = CHECKSUM_UNNECESSARY; 3920 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3921 skb_dst_force(skb); 3922 netif_rx(skb); 3923 return 0; 3924 } 3925 EXPORT_SYMBOL(dev_loopback_xmit); 3926 3927 #ifdef CONFIG_NET_EGRESS 3928 static struct sk_buff * 3929 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3930 { 3931 #ifdef CONFIG_NET_CLS_ACT 3932 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3933 struct tcf_result cl_res; 3934 3935 if (!miniq) 3936 return skb; 3937 3938 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3939 tc_skb_cb(skb)->mru = 0; 3940 tc_skb_cb(skb)->post_ct = false; 3941 mini_qdisc_bstats_cpu_update(miniq, skb); 3942 3943 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3944 case TC_ACT_OK: 3945 case TC_ACT_RECLASSIFY: 3946 skb->tc_index = TC_H_MIN(cl_res.classid); 3947 break; 3948 case TC_ACT_SHOT: 3949 mini_qdisc_qstats_cpu_drop(miniq); 3950 *ret = NET_XMIT_DROP; 3951 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3952 return NULL; 3953 case TC_ACT_STOLEN: 3954 case TC_ACT_QUEUED: 3955 case TC_ACT_TRAP: 3956 *ret = NET_XMIT_SUCCESS; 3957 consume_skb(skb); 3958 return NULL; 3959 case TC_ACT_REDIRECT: 3960 /* No need to push/pop skb's mac_header here on egress! */ 3961 skb_do_redirect(skb); 3962 *ret = NET_XMIT_SUCCESS; 3963 return NULL; 3964 default: 3965 break; 3966 } 3967 #endif /* CONFIG_NET_CLS_ACT */ 3968 3969 return skb; 3970 } 3971 3972 static struct netdev_queue * 3973 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3974 { 3975 int qm = skb_get_queue_mapping(skb); 3976 3977 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3978 } 3979 3980 static bool netdev_xmit_txqueue_skipped(void) 3981 { 3982 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3983 } 3984 3985 void netdev_xmit_skip_txqueue(bool skip) 3986 { 3987 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3988 } 3989 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3990 #endif /* CONFIG_NET_EGRESS */ 3991 3992 #ifdef CONFIG_XPS 3993 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3994 struct xps_dev_maps *dev_maps, unsigned int tci) 3995 { 3996 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3997 struct xps_map *map; 3998 int queue_index = -1; 3999 4000 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4001 return queue_index; 4002 4003 tci *= dev_maps->num_tc; 4004 tci += tc; 4005 4006 map = rcu_dereference(dev_maps->attr_map[tci]); 4007 if (map) { 4008 if (map->len == 1) 4009 queue_index = map->queues[0]; 4010 else 4011 queue_index = map->queues[reciprocal_scale( 4012 skb_get_hash(skb), map->len)]; 4013 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4014 queue_index = -1; 4015 } 4016 return queue_index; 4017 } 4018 #endif 4019 4020 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4021 struct sk_buff *skb) 4022 { 4023 #ifdef CONFIG_XPS 4024 struct xps_dev_maps *dev_maps; 4025 struct sock *sk = skb->sk; 4026 int queue_index = -1; 4027 4028 if (!static_key_false(&xps_needed)) 4029 return -1; 4030 4031 rcu_read_lock(); 4032 if (!static_key_false(&xps_rxqs_needed)) 4033 goto get_cpus_map; 4034 4035 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4036 if (dev_maps) { 4037 int tci = sk_rx_queue_get(sk); 4038 4039 if (tci >= 0) 4040 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4041 tci); 4042 } 4043 4044 get_cpus_map: 4045 if (queue_index < 0) { 4046 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4047 if (dev_maps) { 4048 unsigned int tci = skb->sender_cpu - 1; 4049 4050 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4051 tci); 4052 } 4053 } 4054 rcu_read_unlock(); 4055 4056 return queue_index; 4057 #else 4058 return -1; 4059 #endif 4060 } 4061 4062 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4063 struct net_device *sb_dev) 4064 { 4065 return 0; 4066 } 4067 EXPORT_SYMBOL(dev_pick_tx_zero); 4068 4069 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4070 struct net_device *sb_dev) 4071 { 4072 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4073 } 4074 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4075 4076 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4077 struct net_device *sb_dev) 4078 { 4079 struct sock *sk = skb->sk; 4080 int queue_index = sk_tx_queue_get(sk); 4081 4082 sb_dev = sb_dev ? : dev; 4083 4084 if (queue_index < 0 || skb->ooo_okay || 4085 queue_index >= dev->real_num_tx_queues) { 4086 int new_index = get_xps_queue(dev, sb_dev, skb); 4087 4088 if (new_index < 0) 4089 new_index = skb_tx_hash(dev, sb_dev, skb); 4090 4091 if (queue_index != new_index && sk && 4092 sk_fullsock(sk) && 4093 rcu_access_pointer(sk->sk_dst_cache)) 4094 sk_tx_queue_set(sk, new_index); 4095 4096 queue_index = new_index; 4097 } 4098 4099 return queue_index; 4100 } 4101 EXPORT_SYMBOL(netdev_pick_tx); 4102 4103 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4104 struct sk_buff *skb, 4105 struct net_device *sb_dev) 4106 { 4107 int queue_index = 0; 4108 4109 #ifdef CONFIG_XPS 4110 u32 sender_cpu = skb->sender_cpu - 1; 4111 4112 if (sender_cpu >= (u32)NR_CPUS) 4113 skb->sender_cpu = raw_smp_processor_id() + 1; 4114 #endif 4115 4116 if (dev->real_num_tx_queues != 1) { 4117 const struct net_device_ops *ops = dev->netdev_ops; 4118 4119 if (ops->ndo_select_queue) 4120 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4121 else 4122 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4123 4124 queue_index = netdev_cap_txqueue(dev, queue_index); 4125 } 4126 4127 skb_set_queue_mapping(skb, queue_index); 4128 return netdev_get_tx_queue(dev, queue_index); 4129 } 4130 4131 /** 4132 * __dev_queue_xmit() - transmit a buffer 4133 * @skb: buffer to transmit 4134 * @sb_dev: suboordinate device used for L2 forwarding offload 4135 * 4136 * Queue a buffer for transmission to a network device. The caller must 4137 * have set the device and priority and built the buffer before calling 4138 * this function. The function can be called from an interrupt. 4139 * 4140 * When calling this method, interrupts MUST be enabled. This is because 4141 * the BH enable code must have IRQs enabled so that it will not deadlock. 4142 * 4143 * Regardless of the return value, the skb is consumed, so it is currently 4144 * difficult to retry a send to this method. (You can bump the ref count 4145 * before sending to hold a reference for retry if you are careful.) 4146 * 4147 * Return: 4148 * * 0 - buffer successfully transmitted 4149 * * positive qdisc return code - NET_XMIT_DROP etc. 4150 * * negative errno - other errors 4151 */ 4152 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4153 { 4154 struct net_device *dev = skb->dev; 4155 struct netdev_queue *txq = NULL; 4156 struct Qdisc *q; 4157 int rc = -ENOMEM; 4158 bool again = false; 4159 4160 skb_reset_mac_header(skb); 4161 skb_assert_len(skb); 4162 4163 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4164 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4165 4166 /* Disable soft irqs for various locks below. Also 4167 * stops preemption for RCU. 4168 */ 4169 rcu_read_lock_bh(); 4170 4171 skb_update_prio(skb); 4172 4173 qdisc_pkt_len_init(skb); 4174 #ifdef CONFIG_NET_CLS_ACT 4175 skb->tc_at_ingress = 0; 4176 #endif 4177 #ifdef CONFIG_NET_EGRESS 4178 if (static_branch_unlikely(&egress_needed_key)) { 4179 if (nf_hook_egress_active()) { 4180 skb = nf_hook_egress(skb, &rc, dev); 4181 if (!skb) 4182 goto out; 4183 } 4184 4185 netdev_xmit_skip_txqueue(false); 4186 4187 nf_skip_egress(skb, true); 4188 skb = sch_handle_egress(skb, &rc, dev); 4189 if (!skb) 4190 goto out; 4191 nf_skip_egress(skb, false); 4192 4193 if (netdev_xmit_txqueue_skipped()) 4194 txq = netdev_tx_queue_mapping(dev, skb); 4195 } 4196 #endif 4197 /* If device/qdisc don't need skb->dst, release it right now while 4198 * its hot in this cpu cache. 4199 */ 4200 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4201 skb_dst_drop(skb); 4202 else 4203 skb_dst_force(skb); 4204 4205 if (!txq) 4206 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4207 4208 q = rcu_dereference_bh(txq->qdisc); 4209 4210 trace_net_dev_queue(skb); 4211 if (q->enqueue) { 4212 rc = __dev_xmit_skb(skb, q, dev, txq); 4213 goto out; 4214 } 4215 4216 /* The device has no queue. Common case for software devices: 4217 * loopback, all the sorts of tunnels... 4218 4219 * Really, it is unlikely that netif_tx_lock protection is necessary 4220 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4221 * counters.) 4222 * However, it is possible, that they rely on protection 4223 * made by us here. 4224 4225 * Check this and shot the lock. It is not prone from deadlocks. 4226 *Either shot noqueue qdisc, it is even simpler 8) 4227 */ 4228 if (dev->flags & IFF_UP) { 4229 int cpu = smp_processor_id(); /* ok because BHs are off */ 4230 4231 /* Other cpus might concurrently change txq->xmit_lock_owner 4232 * to -1 or to their cpu id, but not to our id. 4233 */ 4234 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4235 if (dev_xmit_recursion()) 4236 goto recursion_alert; 4237 4238 skb = validate_xmit_skb(skb, dev, &again); 4239 if (!skb) 4240 goto out; 4241 4242 HARD_TX_LOCK(dev, txq, cpu); 4243 4244 if (!netif_xmit_stopped(txq)) { 4245 dev_xmit_recursion_inc(); 4246 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4247 dev_xmit_recursion_dec(); 4248 if (dev_xmit_complete(rc)) { 4249 HARD_TX_UNLOCK(dev, txq); 4250 goto out; 4251 } 4252 } 4253 HARD_TX_UNLOCK(dev, txq); 4254 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4255 dev->name); 4256 } else { 4257 /* Recursion is detected! It is possible, 4258 * unfortunately 4259 */ 4260 recursion_alert: 4261 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4262 dev->name); 4263 } 4264 } 4265 4266 rc = -ENETDOWN; 4267 rcu_read_unlock_bh(); 4268 4269 dev_core_stats_tx_dropped_inc(dev); 4270 kfree_skb_list(skb); 4271 return rc; 4272 out: 4273 rcu_read_unlock_bh(); 4274 return rc; 4275 } 4276 EXPORT_SYMBOL(__dev_queue_xmit); 4277 4278 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4279 { 4280 struct net_device *dev = skb->dev; 4281 struct sk_buff *orig_skb = skb; 4282 struct netdev_queue *txq; 4283 int ret = NETDEV_TX_BUSY; 4284 bool again = false; 4285 4286 if (unlikely(!netif_running(dev) || 4287 !netif_carrier_ok(dev))) 4288 goto drop; 4289 4290 skb = validate_xmit_skb_list(skb, dev, &again); 4291 if (skb != orig_skb) 4292 goto drop; 4293 4294 skb_set_queue_mapping(skb, queue_id); 4295 txq = skb_get_tx_queue(dev, skb); 4296 4297 local_bh_disable(); 4298 4299 dev_xmit_recursion_inc(); 4300 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4301 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4302 ret = netdev_start_xmit(skb, dev, txq, false); 4303 HARD_TX_UNLOCK(dev, txq); 4304 dev_xmit_recursion_dec(); 4305 4306 local_bh_enable(); 4307 return ret; 4308 drop: 4309 dev_core_stats_tx_dropped_inc(dev); 4310 kfree_skb_list(skb); 4311 return NET_XMIT_DROP; 4312 } 4313 EXPORT_SYMBOL(__dev_direct_xmit); 4314 4315 /************************************************************************* 4316 * Receiver routines 4317 *************************************************************************/ 4318 4319 int netdev_max_backlog __read_mostly = 1000; 4320 EXPORT_SYMBOL(netdev_max_backlog); 4321 4322 int netdev_tstamp_prequeue __read_mostly = 1; 4323 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4324 int netdev_budget __read_mostly = 300; 4325 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4326 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4327 int weight_p __read_mostly = 64; /* old backlog weight */ 4328 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4329 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4330 int dev_rx_weight __read_mostly = 64; 4331 int dev_tx_weight __read_mostly = 64; 4332 4333 /* Called with irq disabled */ 4334 static inline void ____napi_schedule(struct softnet_data *sd, 4335 struct napi_struct *napi) 4336 { 4337 struct task_struct *thread; 4338 4339 lockdep_assert_irqs_disabled(); 4340 4341 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4342 /* Paired with smp_mb__before_atomic() in 4343 * napi_enable()/dev_set_threaded(). 4344 * Use READ_ONCE() to guarantee a complete 4345 * read on napi->thread. Only call 4346 * wake_up_process() when it's not NULL. 4347 */ 4348 thread = READ_ONCE(napi->thread); 4349 if (thread) { 4350 /* Avoid doing set_bit() if the thread is in 4351 * INTERRUPTIBLE state, cause napi_thread_wait() 4352 * makes sure to proceed with napi polling 4353 * if the thread is explicitly woken from here. 4354 */ 4355 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4356 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4357 wake_up_process(thread); 4358 return; 4359 } 4360 } 4361 4362 list_add_tail(&napi->poll_list, &sd->poll_list); 4363 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4364 } 4365 4366 #ifdef CONFIG_RPS 4367 4368 /* One global table that all flow-based protocols share. */ 4369 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4370 EXPORT_SYMBOL(rps_sock_flow_table); 4371 u32 rps_cpu_mask __read_mostly; 4372 EXPORT_SYMBOL(rps_cpu_mask); 4373 4374 struct static_key_false rps_needed __read_mostly; 4375 EXPORT_SYMBOL(rps_needed); 4376 struct static_key_false rfs_needed __read_mostly; 4377 EXPORT_SYMBOL(rfs_needed); 4378 4379 static struct rps_dev_flow * 4380 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4381 struct rps_dev_flow *rflow, u16 next_cpu) 4382 { 4383 if (next_cpu < nr_cpu_ids) { 4384 #ifdef CONFIG_RFS_ACCEL 4385 struct netdev_rx_queue *rxqueue; 4386 struct rps_dev_flow_table *flow_table; 4387 struct rps_dev_flow *old_rflow; 4388 u32 flow_id; 4389 u16 rxq_index; 4390 int rc; 4391 4392 /* Should we steer this flow to a different hardware queue? */ 4393 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4394 !(dev->features & NETIF_F_NTUPLE)) 4395 goto out; 4396 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4397 if (rxq_index == skb_get_rx_queue(skb)) 4398 goto out; 4399 4400 rxqueue = dev->_rx + rxq_index; 4401 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4402 if (!flow_table) 4403 goto out; 4404 flow_id = skb_get_hash(skb) & flow_table->mask; 4405 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4406 rxq_index, flow_id); 4407 if (rc < 0) 4408 goto out; 4409 old_rflow = rflow; 4410 rflow = &flow_table->flows[flow_id]; 4411 rflow->filter = rc; 4412 if (old_rflow->filter == rflow->filter) 4413 old_rflow->filter = RPS_NO_FILTER; 4414 out: 4415 #endif 4416 rflow->last_qtail = 4417 per_cpu(softnet_data, next_cpu).input_queue_head; 4418 } 4419 4420 rflow->cpu = next_cpu; 4421 return rflow; 4422 } 4423 4424 /* 4425 * get_rps_cpu is called from netif_receive_skb and returns the target 4426 * CPU from the RPS map of the receiving queue for a given skb. 4427 * rcu_read_lock must be held on entry. 4428 */ 4429 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4430 struct rps_dev_flow **rflowp) 4431 { 4432 const struct rps_sock_flow_table *sock_flow_table; 4433 struct netdev_rx_queue *rxqueue = dev->_rx; 4434 struct rps_dev_flow_table *flow_table; 4435 struct rps_map *map; 4436 int cpu = -1; 4437 u32 tcpu; 4438 u32 hash; 4439 4440 if (skb_rx_queue_recorded(skb)) { 4441 u16 index = skb_get_rx_queue(skb); 4442 4443 if (unlikely(index >= dev->real_num_rx_queues)) { 4444 WARN_ONCE(dev->real_num_rx_queues > 1, 4445 "%s received packet on queue %u, but number " 4446 "of RX queues is %u\n", 4447 dev->name, index, dev->real_num_rx_queues); 4448 goto done; 4449 } 4450 rxqueue += index; 4451 } 4452 4453 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4454 4455 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4456 map = rcu_dereference(rxqueue->rps_map); 4457 if (!flow_table && !map) 4458 goto done; 4459 4460 skb_reset_network_header(skb); 4461 hash = skb_get_hash(skb); 4462 if (!hash) 4463 goto done; 4464 4465 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4466 if (flow_table && sock_flow_table) { 4467 struct rps_dev_flow *rflow; 4468 u32 next_cpu; 4469 u32 ident; 4470 4471 /* First check into global flow table if there is a match */ 4472 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4473 if ((ident ^ hash) & ~rps_cpu_mask) 4474 goto try_rps; 4475 4476 next_cpu = ident & rps_cpu_mask; 4477 4478 /* OK, now we know there is a match, 4479 * we can look at the local (per receive queue) flow table 4480 */ 4481 rflow = &flow_table->flows[hash & flow_table->mask]; 4482 tcpu = rflow->cpu; 4483 4484 /* 4485 * If the desired CPU (where last recvmsg was done) is 4486 * different from current CPU (one in the rx-queue flow 4487 * table entry), switch if one of the following holds: 4488 * - Current CPU is unset (>= nr_cpu_ids). 4489 * - Current CPU is offline. 4490 * - The current CPU's queue tail has advanced beyond the 4491 * last packet that was enqueued using this table entry. 4492 * This guarantees that all previous packets for the flow 4493 * have been dequeued, thus preserving in order delivery. 4494 */ 4495 if (unlikely(tcpu != next_cpu) && 4496 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4497 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4498 rflow->last_qtail)) >= 0)) { 4499 tcpu = next_cpu; 4500 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4501 } 4502 4503 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4504 *rflowp = rflow; 4505 cpu = tcpu; 4506 goto done; 4507 } 4508 } 4509 4510 try_rps: 4511 4512 if (map) { 4513 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4514 if (cpu_online(tcpu)) { 4515 cpu = tcpu; 4516 goto done; 4517 } 4518 } 4519 4520 done: 4521 return cpu; 4522 } 4523 4524 #ifdef CONFIG_RFS_ACCEL 4525 4526 /** 4527 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4528 * @dev: Device on which the filter was set 4529 * @rxq_index: RX queue index 4530 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4531 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4532 * 4533 * Drivers that implement ndo_rx_flow_steer() should periodically call 4534 * this function for each installed filter and remove the filters for 4535 * which it returns %true. 4536 */ 4537 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4538 u32 flow_id, u16 filter_id) 4539 { 4540 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4541 struct rps_dev_flow_table *flow_table; 4542 struct rps_dev_flow *rflow; 4543 bool expire = true; 4544 unsigned int cpu; 4545 4546 rcu_read_lock(); 4547 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4548 if (flow_table && flow_id <= flow_table->mask) { 4549 rflow = &flow_table->flows[flow_id]; 4550 cpu = READ_ONCE(rflow->cpu); 4551 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4552 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4553 rflow->last_qtail) < 4554 (int)(10 * flow_table->mask))) 4555 expire = false; 4556 } 4557 rcu_read_unlock(); 4558 return expire; 4559 } 4560 EXPORT_SYMBOL(rps_may_expire_flow); 4561 4562 #endif /* CONFIG_RFS_ACCEL */ 4563 4564 /* Called from hardirq (IPI) context */ 4565 static void rps_trigger_softirq(void *data) 4566 { 4567 struct softnet_data *sd = data; 4568 4569 ____napi_schedule(sd, &sd->backlog); 4570 sd->received_rps++; 4571 } 4572 4573 #endif /* CONFIG_RPS */ 4574 4575 /* Called from hardirq (IPI) context */ 4576 static void trigger_rx_softirq(void *data) 4577 { 4578 struct softnet_data *sd = data; 4579 4580 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4581 smp_store_release(&sd->defer_ipi_scheduled, 0); 4582 } 4583 4584 /* 4585 * Check if this softnet_data structure is another cpu one 4586 * If yes, queue it to our IPI list and return 1 4587 * If no, return 0 4588 */ 4589 static int napi_schedule_rps(struct softnet_data *sd) 4590 { 4591 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4592 4593 #ifdef CONFIG_RPS 4594 if (sd != mysd) { 4595 sd->rps_ipi_next = mysd->rps_ipi_list; 4596 mysd->rps_ipi_list = sd; 4597 4598 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4599 return 1; 4600 } 4601 #endif /* CONFIG_RPS */ 4602 __napi_schedule_irqoff(&mysd->backlog); 4603 return 0; 4604 } 4605 4606 #ifdef CONFIG_NET_FLOW_LIMIT 4607 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4608 #endif 4609 4610 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4611 { 4612 #ifdef CONFIG_NET_FLOW_LIMIT 4613 struct sd_flow_limit *fl; 4614 struct softnet_data *sd; 4615 unsigned int old_flow, new_flow; 4616 4617 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4618 return false; 4619 4620 sd = this_cpu_ptr(&softnet_data); 4621 4622 rcu_read_lock(); 4623 fl = rcu_dereference(sd->flow_limit); 4624 if (fl) { 4625 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4626 old_flow = fl->history[fl->history_head]; 4627 fl->history[fl->history_head] = new_flow; 4628 4629 fl->history_head++; 4630 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4631 4632 if (likely(fl->buckets[old_flow])) 4633 fl->buckets[old_flow]--; 4634 4635 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4636 fl->count++; 4637 rcu_read_unlock(); 4638 return true; 4639 } 4640 } 4641 rcu_read_unlock(); 4642 #endif 4643 return false; 4644 } 4645 4646 /* 4647 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4648 * queue (may be a remote CPU queue). 4649 */ 4650 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4651 unsigned int *qtail) 4652 { 4653 enum skb_drop_reason reason; 4654 struct softnet_data *sd; 4655 unsigned long flags; 4656 unsigned int qlen; 4657 4658 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4659 sd = &per_cpu(softnet_data, cpu); 4660 4661 rps_lock_irqsave(sd, &flags); 4662 if (!netif_running(skb->dev)) 4663 goto drop; 4664 qlen = skb_queue_len(&sd->input_pkt_queue); 4665 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4666 if (qlen) { 4667 enqueue: 4668 __skb_queue_tail(&sd->input_pkt_queue, skb); 4669 input_queue_tail_incr_save(sd, qtail); 4670 rps_unlock_irq_restore(sd, &flags); 4671 return NET_RX_SUCCESS; 4672 } 4673 4674 /* Schedule NAPI for backlog device 4675 * We can use non atomic operation since we own the queue lock 4676 */ 4677 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4678 napi_schedule_rps(sd); 4679 goto enqueue; 4680 } 4681 reason = SKB_DROP_REASON_CPU_BACKLOG; 4682 4683 drop: 4684 sd->dropped++; 4685 rps_unlock_irq_restore(sd, &flags); 4686 4687 dev_core_stats_rx_dropped_inc(skb->dev); 4688 kfree_skb_reason(skb, reason); 4689 return NET_RX_DROP; 4690 } 4691 4692 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4693 { 4694 struct net_device *dev = skb->dev; 4695 struct netdev_rx_queue *rxqueue; 4696 4697 rxqueue = dev->_rx; 4698 4699 if (skb_rx_queue_recorded(skb)) { 4700 u16 index = skb_get_rx_queue(skb); 4701 4702 if (unlikely(index >= dev->real_num_rx_queues)) { 4703 WARN_ONCE(dev->real_num_rx_queues > 1, 4704 "%s received packet on queue %u, but number " 4705 "of RX queues is %u\n", 4706 dev->name, index, dev->real_num_rx_queues); 4707 4708 return rxqueue; /* Return first rxqueue */ 4709 } 4710 rxqueue += index; 4711 } 4712 return rxqueue; 4713 } 4714 4715 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4716 struct bpf_prog *xdp_prog) 4717 { 4718 void *orig_data, *orig_data_end, *hard_start; 4719 struct netdev_rx_queue *rxqueue; 4720 bool orig_bcast, orig_host; 4721 u32 mac_len, frame_sz; 4722 __be16 orig_eth_type; 4723 struct ethhdr *eth; 4724 u32 metalen, act; 4725 int off; 4726 4727 /* The XDP program wants to see the packet starting at the MAC 4728 * header. 4729 */ 4730 mac_len = skb->data - skb_mac_header(skb); 4731 hard_start = skb->data - skb_headroom(skb); 4732 4733 /* SKB "head" area always have tailroom for skb_shared_info */ 4734 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4735 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4736 4737 rxqueue = netif_get_rxqueue(skb); 4738 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4739 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4740 skb_headlen(skb) + mac_len, true); 4741 4742 orig_data_end = xdp->data_end; 4743 orig_data = xdp->data; 4744 eth = (struct ethhdr *)xdp->data; 4745 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4746 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4747 orig_eth_type = eth->h_proto; 4748 4749 act = bpf_prog_run_xdp(xdp_prog, xdp); 4750 4751 /* check if bpf_xdp_adjust_head was used */ 4752 off = xdp->data - orig_data; 4753 if (off) { 4754 if (off > 0) 4755 __skb_pull(skb, off); 4756 else if (off < 0) 4757 __skb_push(skb, -off); 4758 4759 skb->mac_header += off; 4760 skb_reset_network_header(skb); 4761 } 4762 4763 /* check if bpf_xdp_adjust_tail was used */ 4764 off = xdp->data_end - orig_data_end; 4765 if (off != 0) { 4766 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4767 skb->len += off; /* positive on grow, negative on shrink */ 4768 } 4769 4770 /* check if XDP changed eth hdr such SKB needs update */ 4771 eth = (struct ethhdr *)xdp->data; 4772 if ((orig_eth_type != eth->h_proto) || 4773 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4774 skb->dev->dev_addr)) || 4775 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4776 __skb_push(skb, ETH_HLEN); 4777 skb->pkt_type = PACKET_HOST; 4778 skb->protocol = eth_type_trans(skb, skb->dev); 4779 } 4780 4781 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4782 * before calling us again on redirect path. We do not call do_redirect 4783 * as we leave that up to the caller. 4784 * 4785 * Caller is responsible for managing lifetime of skb (i.e. calling 4786 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4787 */ 4788 switch (act) { 4789 case XDP_REDIRECT: 4790 case XDP_TX: 4791 __skb_push(skb, mac_len); 4792 break; 4793 case XDP_PASS: 4794 metalen = xdp->data - xdp->data_meta; 4795 if (metalen) 4796 skb_metadata_set(skb, metalen); 4797 break; 4798 } 4799 4800 return act; 4801 } 4802 4803 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4804 struct xdp_buff *xdp, 4805 struct bpf_prog *xdp_prog) 4806 { 4807 u32 act = XDP_DROP; 4808 4809 /* Reinjected packets coming from act_mirred or similar should 4810 * not get XDP generic processing. 4811 */ 4812 if (skb_is_redirected(skb)) 4813 return XDP_PASS; 4814 4815 /* XDP packets must be linear and must have sufficient headroom 4816 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4817 * native XDP provides, thus we need to do it here as well. 4818 */ 4819 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4820 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4821 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4822 int troom = skb->tail + skb->data_len - skb->end; 4823 4824 /* In case we have to go down the path and also linearize, 4825 * then lets do the pskb_expand_head() work just once here. 4826 */ 4827 if (pskb_expand_head(skb, 4828 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4829 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4830 goto do_drop; 4831 if (skb_linearize(skb)) 4832 goto do_drop; 4833 } 4834 4835 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4836 switch (act) { 4837 case XDP_REDIRECT: 4838 case XDP_TX: 4839 case XDP_PASS: 4840 break; 4841 default: 4842 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4843 fallthrough; 4844 case XDP_ABORTED: 4845 trace_xdp_exception(skb->dev, xdp_prog, act); 4846 fallthrough; 4847 case XDP_DROP: 4848 do_drop: 4849 kfree_skb(skb); 4850 break; 4851 } 4852 4853 return act; 4854 } 4855 4856 /* When doing generic XDP we have to bypass the qdisc layer and the 4857 * network taps in order to match in-driver-XDP behavior. This also means 4858 * that XDP packets are able to starve other packets going through a qdisc, 4859 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4860 * queues, so they do not have this starvation issue. 4861 */ 4862 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4863 { 4864 struct net_device *dev = skb->dev; 4865 struct netdev_queue *txq; 4866 bool free_skb = true; 4867 int cpu, rc; 4868 4869 txq = netdev_core_pick_tx(dev, skb, NULL); 4870 cpu = smp_processor_id(); 4871 HARD_TX_LOCK(dev, txq, cpu); 4872 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4873 rc = netdev_start_xmit(skb, dev, txq, 0); 4874 if (dev_xmit_complete(rc)) 4875 free_skb = false; 4876 } 4877 HARD_TX_UNLOCK(dev, txq); 4878 if (free_skb) { 4879 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4880 dev_core_stats_tx_dropped_inc(dev); 4881 kfree_skb(skb); 4882 } 4883 } 4884 4885 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4886 4887 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4888 { 4889 if (xdp_prog) { 4890 struct xdp_buff xdp; 4891 u32 act; 4892 int err; 4893 4894 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4895 if (act != XDP_PASS) { 4896 switch (act) { 4897 case XDP_REDIRECT: 4898 err = xdp_do_generic_redirect(skb->dev, skb, 4899 &xdp, xdp_prog); 4900 if (err) 4901 goto out_redir; 4902 break; 4903 case XDP_TX: 4904 generic_xdp_tx(skb, xdp_prog); 4905 break; 4906 } 4907 return XDP_DROP; 4908 } 4909 } 4910 return XDP_PASS; 4911 out_redir: 4912 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4913 return XDP_DROP; 4914 } 4915 EXPORT_SYMBOL_GPL(do_xdp_generic); 4916 4917 static int netif_rx_internal(struct sk_buff *skb) 4918 { 4919 int ret; 4920 4921 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4922 4923 trace_netif_rx(skb); 4924 4925 #ifdef CONFIG_RPS 4926 if (static_branch_unlikely(&rps_needed)) { 4927 struct rps_dev_flow voidflow, *rflow = &voidflow; 4928 int cpu; 4929 4930 rcu_read_lock(); 4931 4932 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4933 if (cpu < 0) 4934 cpu = smp_processor_id(); 4935 4936 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4937 4938 rcu_read_unlock(); 4939 } else 4940 #endif 4941 { 4942 unsigned int qtail; 4943 4944 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4945 } 4946 return ret; 4947 } 4948 4949 /** 4950 * __netif_rx - Slightly optimized version of netif_rx 4951 * @skb: buffer to post 4952 * 4953 * This behaves as netif_rx except that it does not disable bottom halves. 4954 * As a result this function may only be invoked from the interrupt context 4955 * (either hard or soft interrupt). 4956 */ 4957 int __netif_rx(struct sk_buff *skb) 4958 { 4959 int ret; 4960 4961 lockdep_assert_once(hardirq_count() | softirq_count()); 4962 4963 trace_netif_rx_entry(skb); 4964 ret = netif_rx_internal(skb); 4965 trace_netif_rx_exit(ret); 4966 return ret; 4967 } 4968 EXPORT_SYMBOL(__netif_rx); 4969 4970 /** 4971 * netif_rx - post buffer to the network code 4972 * @skb: buffer to post 4973 * 4974 * This function receives a packet from a device driver and queues it for 4975 * the upper (protocol) levels to process via the backlog NAPI device. It 4976 * always succeeds. The buffer may be dropped during processing for 4977 * congestion control or by the protocol layers. 4978 * The network buffer is passed via the backlog NAPI device. Modern NIC 4979 * driver should use NAPI and GRO. 4980 * This function can used from interrupt and from process context. The 4981 * caller from process context must not disable interrupts before invoking 4982 * this function. 4983 * 4984 * return values: 4985 * NET_RX_SUCCESS (no congestion) 4986 * NET_RX_DROP (packet was dropped) 4987 * 4988 */ 4989 int netif_rx(struct sk_buff *skb) 4990 { 4991 bool need_bh_off = !(hardirq_count() | softirq_count()); 4992 int ret; 4993 4994 if (need_bh_off) 4995 local_bh_disable(); 4996 trace_netif_rx_entry(skb); 4997 ret = netif_rx_internal(skb); 4998 trace_netif_rx_exit(ret); 4999 if (need_bh_off) 5000 local_bh_enable(); 5001 return ret; 5002 } 5003 EXPORT_SYMBOL(netif_rx); 5004 5005 static __latent_entropy void net_tx_action(struct softirq_action *h) 5006 { 5007 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5008 5009 if (sd->completion_queue) { 5010 struct sk_buff *clist; 5011 5012 local_irq_disable(); 5013 clist = sd->completion_queue; 5014 sd->completion_queue = NULL; 5015 local_irq_enable(); 5016 5017 while (clist) { 5018 struct sk_buff *skb = clist; 5019 5020 clist = clist->next; 5021 5022 WARN_ON(refcount_read(&skb->users)); 5023 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5024 trace_consume_skb(skb, net_tx_action); 5025 else 5026 trace_kfree_skb(skb, net_tx_action, 5027 SKB_DROP_REASON_NOT_SPECIFIED); 5028 5029 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5030 __kfree_skb(skb); 5031 else 5032 __kfree_skb_defer(skb); 5033 } 5034 } 5035 5036 if (sd->output_queue) { 5037 struct Qdisc *head; 5038 5039 local_irq_disable(); 5040 head = sd->output_queue; 5041 sd->output_queue = NULL; 5042 sd->output_queue_tailp = &sd->output_queue; 5043 local_irq_enable(); 5044 5045 rcu_read_lock(); 5046 5047 while (head) { 5048 struct Qdisc *q = head; 5049 spinlock_t *root_lock = NULL; 5050 5051 head = head->next_sched; 5052 5053 /* We need to make sure head->next_sched is read 5054 * before clearing __QDISC_STATE_SCHED 5055 */ 5056 smp_mb__before_atomic(); 5057 5058 if (!(q->flags & TCQ_F_NOLOCK)) { 5059 root_lock = qdisc_lock(q); 5060 spin_lock(root_lock); 5061 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5062 &q->state))) { 5063 /* There is a synchronize_net() between 5064 * STATE_DEACTIVATED flag being set and 5065 * qdisc_reset()/some_qdisc_is_busy() in 5066 * dev_deactivate(), so we can safely bail out 5067 * early here to avoid data race between 5068 * qdisc_deactivate() and some_qdisc_is_busy() 5069 * for lockless qdisc. 5070 */ 5071 clear_bit(__QDISC_STATE_SCHED, &q->state); 5072 continue; 5073 } 5074 5075 clear_bit(__QDISC_STATE_SCHED, &q->state); 5076 qdisc_run(q); 5077 if (root_lock) 5078 spin_unlock(root_lock); 5079 } 5080 5081 rcu_read_unlock(); 5082 } 5083 5084 xfrm_dev_backlog(sd); 5085 } 5086 5087 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5088 /* This hook is defined here for ATM LANE */ 5089 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5090 unsigned char *addr) __read_mostly; 5091 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5092 #endif 5093 5094 static inline struct sk_buff * 5095 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5096 struct net_device *orig_dev, bool *another) 5097 { 5098 #ifdef CONFIG_NET_CLS_ACT 5099 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5100 struct tcf_result cl_res; 5101 5102 /* If there's at least one ingress present somewhere (so 5103 * we get here via enabled static key), remaining devices 5104 * that are not configured with an ingress qdisc will bail 5105 * out here. 5106 */ 5107 if (!miniq) 5108 return skb; 5109 5110 if (*pt_prev) { 5111 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5112 *pt_prev = NULL; 5113 } 5114 5115 qdisc_skb_cb(skb)->pkt_len = skb->len; 5116 tc_skb_cb(skb)->mru = 0; 5117 tc_skb_cb(skb)->post_ct = false; 5118 skb->tc_at_ingress = 1; 5119 mini_qdisc_bstats_cpu_update(miniq, skb); 5120 5121 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5122 case TC_ACT_OK: 5123 case TC_ACT_RECLASSIFY: 5124 skb->tc_index = TC_H_MIN(cl_res.classid); 5125 break; 5126 case TC_ACT_SHOT: 5127 mini_qdisc_qstats_cpu_drop(miniq); 5128 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5129 *ret = NET_RX_DROP; 5130 return NULL; 5131 case TC_ACT_STOLEN: 5132 case TC_ACT_QUEUED: 5133 case TC_ACT_TRAP: 5134 consume_skb(skb); 5135 *ret = NET_RX_SUCCESS; 5136 return NULL; 5137 case TC_ACT_REDIRECT: 5138 /* skb_mac_header check was done by cls/act_bpf, so 5139 * we can safely push the L2 header back before 5140 * redirecting to another netdev 5141 */ 5142 __skb_push(skb, skb->mac_len); 5143 if (skb_do_redirect(skb) == -EAGAIN) { 5144 __skb_pull(skb, skb->mac_len); 5145 *another = true; 5146 break; 5147 } 5148 *ret = NET_RX_SUCCESS; 5149 return NULL; 5150 case TC_ACT_CONSUMED: 5151 *ret = NET_RX_SUCCESS; 5152 return NULL; 5153 default: 5154 break; 5155 } 5156 #endif /* CONFIG_NET_CLS_ACT */ 5157 return skb; 5158 } 5159 5160 /** 5161 * netdev_is_rx_handler_busy - check if receive handler is registered 5162 * @dev: device to check 5163 * 5164 * Check if a receive handler is already registered for a given device. 5165 * Return true if there one. 5166 * 5167 * The caller must hold the rtnl_mutex. 5168 */ 5169 bool netdev_is_rx_handler_busy(struct net_device *dev) 5170 { 5171 ASSERT_RTNL(); 5172 return dev && rtnl_dereference(dev->rx_handler); 5173 } 5174 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5175 5176 /** 5177 * netdev_rx_handler_register - register receive handler 5178 * @dev: device to register a handler for 5179 * @rx_handler: receive handler to register 5180 * @rx_handler_data: data pointer that is used by rx handler 5181 * 5182 * Register a receive handler for a device. This handler will then be 5183 * called from __netif_receive_skb. A negative errno code is returned 5184 * on a failure. 5185 * 5186 * The caller must hold the rtnl_mutex. 5187 * 5188 * For a general description of rx_handler, see enum rx_handler_result. 5189 */ 5190 int netdev_rx_handler_register(struct net_device *dev, 5191 rx_handler_func_t *rx_handler, 5192 void *rx_handler_data) 5193 { 5194 if (netdev_is_rx_handler_busy(dev)) 5195 return -EBUSY; 5196 5197 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5198 return -EINVAL; 5199 5200 /* Note: rx_handler_data must be set before rx_handler */ 5201 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5202 rcu_assign_pointer(dev->rx_handler, rx_handler); 5203 5204 return 0; 5205 } 5206 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5207 5208 /** 5209 * netdev_rx_handler_unregister - unregister receive handler 5210 * @dev: device to unregister a handler from 5211 * 5212 * Unregister a receive handler from a device. 5213 * 5214 * The caller must hold the rtnl_mutex. 5215 */ 5216 void netdev_rx_handler_unregister(struct net_device *dev) 5217 { 5218 5219 ASSERT_RTNL(); 5220 RCU_INIT_POINTER(dev->rx_handler, NULL); 5221 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5222 * section has a guarantee to see a non NULL rx_handler_data 5223 * as well. 5224 */ 5225 synchronize_net(); 5226 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5227 } 5228 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5229 5230 /* 5231 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5232 * the special handling of PFMEMALLOC skbs. 5233 */ 5234 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5235 { 5236 switch (skb->protocol) { 5237 case htons(ETH_P_ARP): 5238 case htons(ETH_P_IP): 5239 case htons(ETH_P_IPV6): 5240 case htons(ETH_P_8021Q): 5241 case htons(ETH_P_8021AD): 5242 return true; 5243 default: 5244 return false; 5245 } 5246 } 5247 5248 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5249 int *ret, struct net_device *orig_dev) 5250 { 5251 if (nf_hook_ingress_active(skb)) { 5252 int ingress_retval; 5253 5254 if (*pt_prev) { 5255 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5256 *pt_prev = NULL; 5257 } 5258 5259 rcu_read_lock(); 5260 ingress_retval = nf_hook_ingress(skb); 5261 rcu_read_unlock(); 5262 return ingress_retval; 5263 } 5264 return 0; 5265 } 5266 5267 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5268 struct packet_type **ppt_prev) 5269 { 5270 struct packet_type *ptype, *pt_prev; 5271 rx_handler_func_t *rx_handler; 5272 struct sk_buff *skb = *pskb; 5273 struct net_device *orig_dev; 5274 bool deliver_exact = false; 5275 int ret = NET_RX_DROP; 5276 __be16 type; 5277 5278 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5279 5280 trace_netif_receive_skb(skb); 5281 5282 orig_dev = skb->dev; 5283 5284 skb_reset_network_header(skb); 5285 if (!skb_transport_header_was_set(skb)) 5286 skb_reset_transport_header(skb); 5287 skb_reset_mac_len(skb); 5288 5289 pt_prev = NULL; 5290 5291 another_round: 5292 skb->skb_iif = skb->dev->ifindex; 5293 5294 __this_cpu_inc(softnet_data.processed); 5295 5296 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5297 int ret2; 5298 5299 migrate_disable(); 5300 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5301 migrate_enable(); 5302 5303 if (ret2 != XDP_PASS) { 5304 ret = NET_RX_DROP; 5305 goto out; 5306 } 5307 } 5308 5309 if (eth_type_vlan(skb->protocol)) { 5310 skb = skb_vlan_untag(skb); 5311 if (unlikely(!skb)) 5312 goto out; 5313 } 5314 5315 if (skb_skip_tc_classify(skb)) 5316 goto skip_classify; 5317 5318 if (pfmemalloc) 5319 goto skip_taps; 5320 5321 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5322 if (pt_prev) 5323 ret = deliver_skb(skb, pt_prev, orig_dev); 5324 pt_prev = ptype; 5325 } 5326 5327 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5328 if (pt_prev) 5329 ret = deliver_skb(skb, pt_prev, orig_dev); 5330 pt_prev = ptype; 5331 } 5332 5333 skip_taps: 5334 #ifdef CONFIG_NET_INGRESS 5335 if (static_branch_unlikely(&ingress_needed_key)) { 5336 bool another = false; 5337 5338 nf_skip_egress(skb, true); 5339 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5340 &another); 5341 if (another) 5342 goto another_round; 5343 if (!skb) 5344 goto out; 5345 5346 nf_skip_egress(skb, false); 5347 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5348 goto out; 5349 } 5350 #endif 5351 skb_reset_redirect(skb); 5352 skip_classify: 5353 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5354 goto drop; 5355 5356 if (skb_vlan_tag_present(skb)) { 5357 if (pt_prev) { 5358 ret = deliver_skb(skb, pt_prev, orig_dev); 5359 pt_prev = NULL; 5360 } 5361 if (vlan_do_receive(&skb)) 5362 goto another_round; 5363 else if (unlikely(!skb)) 5364 goto out; 5365 } 5366 5367 rx_handler = rcu_dereference(skb->dev->rx_handler); 5368 if (rx_handler) { 5369 if (pt_prev) { 5370 ret = deliver_skb(skb, pt_prev, orig_dev); 5371 pt_prev = NULL; 5372 } 5373 switch (rx_handler(&skb)) { 5374 case RX_HANDLER_CONSUMED: 5375 ret = NET_RX_SUCCESS; 5376 goto out; 5377 case RX_HANDLER_ANOTHER: 5378 goto another_round; 5379 case RX_HANDLER_EXACT: 5380 deliver_exact = true; 5381 break; 5382 case RX_HANDLER_PASS: 5383 break; 5384 default: 5385 BUG(); 5386 } 5387 } 5388 5389 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5390 check_vlan_id: 5391 if (skb_vlan_tag_get_id(skb)) { 5392 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5393 * find vlan device. 5394 */ 5395 skb->pkt_type = PACKET_OTHERHOST; 5396 } else if (eth_type_vlan(skb->protocol)) { 5397 /* Outer header is 802.1P with vlan 0, inner header is 5398 * 802.1Q or 802.1AD and vlan_do_receive() above could 5399 * not find vlan dev for vlan id 0. 5400 */ 5401 __vlan_hwaccel_clear_tag(skb); 5402 skb = skb_vlan_untag(skb); 5403 if (unlikely(!skb)) 5404 goto out; 5405 if (vlan_do_receive(&skb)) 5406 /* After stripping off 802.1P header with vlan 0 5407 * vlan dev is found for inner header. 5408 */ 5409 goto another_round; 5410 else if (unlikely(!skb)) 5411 goto out; 5412 else 5413 /* We have stripped outer 802.1P vlan 0 header. 5414 * But could not find vlan dev. 5415 * check again for vlan id to set OTHERHOST. 5416 */ 5417 goto check_vlan_id; 5418 } 5419 /* Note: we might in the future use prio bits 5420 * and set skb->priority like in vlan_do_receive() 5421 * For the time being, just ignore Priority Code Point 5422 */ 5423 __vlan_hwaccel_clear_tag(skb); 5424 } 5425 5426 type = skb->protocol; 5427 5428 /* deliver only exact match when indicated */ 5429 if (likely(!deliver_exact)) { 5430 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5431 &ptype_base[ntohs(type) & 5432 PTYPE_HASH_MASK]); 5433 } 5434 5435 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5436 &orig_dev->ptype_specific); 5437 5438 if (unlikely(skb->dev != orig_dev)) { 5439 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5440 &skb->dev->ptype_specific); 5441 } 5442 5443 if (pt_prev) { 5444 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5445 goto drop; 5446 *ppt_prev = pt_prev; 5447 } else { 5448 drop: 5449 if (!deliver_exact) 5450 dev_core_stats_rx_dropped_inc(skb->dev); 5451 else 5452 dev_core_stats_rx_nohandler_inc(skb->dev); 5453 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5454 /* Jamal, now you will not able to escape explaining 5455 * me how you were going to use this. :-) 5456 */ 5457 ret = NET_RX_DROP; 5458 } 5459 5460 out: 5461 /* The invariant here is that if *ppt_prev is not NULL 5462 * then skb should also be non-NULL. 5463 * 5464 * Apparently *ppt_prev assignment above holds this invariant due to 5465 * skb dereferencing near it. 5466 */ 5467 *pskb = skb; 5468 return ret; 5469 } 5470 5471 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5472 { 5473 struct net_device *orig_dev = skb->dev; 5474 struct packet_type *pt_prev = NULL; 5475 int ret; 5476 5477 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5478 if (pt_prev) 5479 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5480 skb->dev, pt_prev, orig_dev); 5481 return ret; 5482 } 5483 5484 /** 5485 * netif_receive_skb_core - special purpose version of netif_receive_skb 5486 * @skb: buffer to process 5487 * 5488 * More direct receive version of netif_receive_skb(). It should 5489 * only be used by callers that have a need to skip RPS and Generic XDP. 5490 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5491 * 5492 * This function may only be called from softirq context and interrupts 5493 * should be enabled. 5494 * 5495 * Return values (usually ignored): 5496 * NET_RX_SUCCESS: no congestion 5497 * NET_RX_DROP: packet was dropped 5498 */ 5499 int netif_receive_skb_core(struct sk_buff *skb) 5500 { 5501 int ret; 5502 5503 rcu_read_lock(); 5504 ret = __netif_receive_skb_one_core(skb, false); 5505 rcu_read_unlock(); 5506 5507 return ret; 5508 } 5509 EXPORT_SYMBOL(netif_receive_skb_core); 5510 5511 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5512 struct packet_type *pt_prev, 5513 struct net_device *orig_dev) 5514 { 5515 struct sk_buff *skb, *next; 5516 5517 if (!pt_prev) 5518 return; 5519 if (list_empty(head)) 5520 return; 5521 if (pt_prev->list_func != NULL) 5522 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5523 ip_list_rcv, head, pt_prev, orig_dev); 5524 else 5525 list_for_each_entry_safe(skb, next, head, list) { 5526 skb_list_del_init(skb); 5527 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5528 } 5529 } 5530 5531 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5532 { 5533 /* Fast-path assumptions: 5534 * - There is no RX handler. 5535 * - Only one packet_type matches. 5536 * If either of these fails, we will end up doing some per-packet 5537 * processing in-line, then handling the 'last ptype' for the whole 5538 * sublist. This can't cause out-of-order delivery to any single ptype, 5539 * because the 'last ptype' must be constant across the sublist, and all 5540 * other ptypes are handled per-packet. 5541 */ 5542 /* Current (common) ptype of sublist */ 5543 struct packet_type *pt_curr = NULL; 5544 /* Current (common) orig_dev of sublist */ 5545 struct net_device *od_curr = NULL; 5546 struct list_head sublist; 5547 struct sk_buff *skb, *next; 5548 5549 INIT_LIST_HEAD(&sublist); 5550 list_for_each_entry_safe(skb, next, head, list) { 5551 struct net_device *orig_dev = skb->dev; 5552 struct packet_type *pt_prev = NULL; 5553 5554 skb_list_del_init(skb); 5555 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5556 if (!pt_prev) 5557 continue; 5558 if (pt_curr != pt_prev || od_curr != orig_dev) { 5559 /* dispatch old sublist */ 5560 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5561 /* start new sublist */ 5562 INIT_LIST_HEAD(&sublist); 5563 pt_curr = pt_prev; 5564 od_curr = orig_dev; 5565 } 5566 list_add_tail(&skb->list, &sublist); 5567 } 5568 5569 /* dispatch final sublist */ 5570 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5571 } 5572 5573 static int __netif_receive_skb(struct sk_buff *skb) 5574 { 5575 int ret; 5576 5577 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5578 unsigned int noreclaim_flag; 5579 5580 /* 5581 * PFMEMALLOC skbs are special, they should 5582 * - be delivered to SOCK_MEMALLOC sockets only 5583 * - stay away from userspace 5584 * - have bounded memory usage 5585 * 5586 * Use PF_MEMALLOC as this saves us from propagating the allocation 5587 * context down to all allocation sites. 5588 */ 5589 noreclaim_flag = memalloc_noreclaim_save(); 5590 ret = __netif_receive_skb_one_core(skb, true); 5591 memalloc_noreclaim_restore(noreclaim_flag); 5592 } else 5593 ret = __netif_receive_skb_one_core(skb, false); 5594 5595 return ret; 5596 } 5597 5598 static void __netif_receive_skb_list(struct list_head *head) 5599 { 5600 unsigned long noreclaim_flag = 0; 5601 struct sk_buff *skb, *next; 5602 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5603 5604 list_for_each_entry_safe(skb, next, head, list) { 5605 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5606 struct list_head sublist; 5607 5608 /* Handle the previous sublist */ 5609 list_cut_before(&sublist, head, &skb->list); 5610 if (!list_empty(&sublist)) 5611 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5612 pfmemalloc = !pfmemalloc; 5613 /* See comments in __netif_receive_skb */ 5614 if (pfmemalloc) 5615 noreclaim_flag = memalloc_noreclaim_save(); 5616 else 5617 memalloc_noreclaim_restore(noreclaim_flag); 5618 } 5619 } 5620 /* Handle the remaining sublist */ 5621 if (!list_empty(head)) 5622 __netif_receive_skb_list_core(head, pfmemalloc); 5623 /* Restore pflags */ 5624 if (pfmemalloc) 5625 memalloc_noreclaim_restore(noreclaim_flag); 5626 } 5627 5628 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5629 { 5630 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5631 struct bpf_prog *new = xdp->prog; 5632 int ret = 0; 5633 5634 switch (xdp->command) { 5635 case XDP_SETUP_PROG: 5636 rcu_assign_pointer(dev->xdp_prog, new); 5637 if (old) 5638 bpf_prog_put(old); 5639 5640 if (old && !new) { 5641 static_branch_dec(&generic_xdp_needed_key); 5642 } else if (new && !old) { 5643 static_branch_inc(&generic_xdp_needed_key); 5644 dev_disable_lro(dev); 5645 dev_disable_gro_hw(dev); 5646 } 5647 break; 5648 5649 default: 5650 ret = -EINVAL; 5651 break; 5652 } 5653 5654 return ret; 5655 } 5656 5657 static int netif_receive_skb_internal(struct sk_buff *skb) 5658 { 5659 int ret; 5660 5661 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5662 5663 if (skb_defer_rx_timestamp(skb)) 5664 return NET_RX_SUCCESS; 5665 5666 rcu_read_lock(); 5667 #ifdef CONFIG_RPS 5668 if (static_branch_unlikely(&rps_needed)) { 5669 struct rps_dev_flow voidflow, *rflow = &voidflow; 5670 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5671 5672 if (cpu >= 0) { 5673 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5674 rcu_read_unlock(); 5675 return ret; 5676 } 5677 } 5678 #endif 5679 ret = __netif_receive_skb(skb); 5680 rcu_read_unlock(); 5681 return ret; 5682 } 5683 5684 void netif_receive_skb_list_internal(struct list_head *head) 5685 { 5686 struct sk_buff *skb, *next; 5687 struct list_head sublist; 5688 5689 INIT_LIST_HEAD(&sublist); 5690 list_for_each_entry_safe(skb, next, head, list) { 5691 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5692 skb_list_del_init(skb); 5693 if (!skb_defer_rx_timestamp(skb)) 5694 list_add_tail(&skb->list, &sublist); 5695 } 5696 list_splice_init(&sublist, head); 5697 5698 rcu_read_lock(); 5699 #ifdef CONFIG_RPS 5700 if (static_branch_unlikely(&rps_needed)) { 5701 list_for_each_entry_safe(skb, next, head, list) { 5702 struct rps_dev_flow voidflow, *rflow = &voidflow; 5703 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5704 5705 if (cpu >= 0) { 5706 /* Will be handled, remove from list */ 5707 skb_list_del_init(skb); 5708 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5709 } 5710 } 5711 } 5712 #endif 5713 __netif_receive_skb_list(head); 5714 rcu_read_unlock(); 5715 } 5716 5717 /** 5718 * netif_receive_skb - process receive buffer from network 5719 * @skb: buffer to process 5720 * 5721 * netif_receive_skb() is the main receive data processing function. 5722 * It always succeeds. The buffer may be dropped during processing 5723 * for congestion control or by the protocol layers. 5724 * 5725 * This function may only be called from softirq context and interrupts 5726 * should be enabled. 5727 * 5728 * Return values (usually ignored): 5729 * NET_RX_SUCCESS: no congestion 5730 * NET_RX_DROP: packet was dropped 5731 */ 5732 int netif_receive_skb(struct sk_buff *skb) 5733 { 5734 int ret; 5735 5736 trace_netif_receive_skb_entry(skb); 5737 5738 ret = netif_receive_skb_internal(skb); 5739 trace_netif_receive_skb_exit(ret); 5740 5741 return ret; 5742 } 5743 EXPORT_SYMBOL(netif_receive_skb); 5744 5745 /** 5746 * netif_receive_skb_list - process many receive buffers from network 5747 * @head: list of skbs to process. 5748 * 5749 * Since return value of netif_receive_skb() is normally ignored, and 5750 * wouldn't be meaningful for a list, this function returns void. 5751 * 5752 * This function may only be called from softirq context and interrupts 5753 * should be enabled. 5754 */ 5755 void netif_receive_skb_list(struct list_head *head) 5756 { 5757 struct sk_buff *skb; 5758 5759 if (list_empty(head)) 5760 return; 5761 if (trace_netif_receive_skb_list_entry_enabled()) { 5762 list_for_each_entry(skb, head, list) 5763 trace_netif_receive_skb_list_entry(skb); 5764 } 5765 netif_receive_skb_list_internal(head); 5766 trace_netif_receive_skb_list_exit(0); 5767 } 5768 EXPORT_SYMBOL(netif_receive_skb_list); 5769 5770 static DEFINE_PER_CPU(struct work_struct, flush_works); 5771 5772 /* Network device is going away, flush any packets still pending */ 5773 static void flush_backlog(struct work_struct *work) 5774 { 5775 struct sk_buff *skb, *tmp; 5776 struct softnet_data *sd; 5777 5778 local_bh_disable(); 5779 sd = this_cpu_ptr(&softnet_data); 5780 5781 rps_lock_irq_disable(sd); 5782 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5783 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5784 __skb_unlink(skb, &sd->input_pkt_queue); 5785 dev_kfree_skb_irq(skb); 5786 input_queue_head_incr(sd); 5787 } 5788 } 5789 rps_unlock_irq_enable(sd); 5790 5791 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5792 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5793 __skb_unlink(skb, &sd->process_queue); 5794 kfree_skb(skb); 5795 input_queue_head_incr(sd); 5796 } 5797 } 5798 local_bh_enable(); 5799 } 5800 5801 static bool flush_required(int cpu) 5802 { 5803 #if IS_ENABLED(CONFIG_RPS) 5804 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5805 bool do_flush; 5806 5807 rps_lock_irq_disable(sd); 5808 5809 /* as insertion into process_queue happens with the rps lock held, 5810 * process_queue access may race only with dequeue 5811 */ 5812 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5813 !skb_queue_empty_lockless(&sd->process_queue); 5814 rps_unlock_irq_enable(sd); 5815 5816 return do_flush; 5817 #endif 5818 /* without RPS we can't safely check input_pkt_queue: during a 5819 * concurrent remote skb_queue_splice() we can detect as empty both 5820 * input_pkt_queue and process_queue even if the latter could end-up 5821 * containing a lot of packets. 5822 */ 5823 return true; 5824 } 5825 5826 static void flush_all_backlogs(void) 5827 { 5828 static cpumask_t flush_cpus; 5829 unsigned int cpu; 5830 5831 /* since we are under rtnl lock protection we can use static data 5832 * for the cpumask and avoid allocating on stack the possibly 5833 * large mask 5834 */ 5835 ASSERT_RTNL(); 5836 5837 cpus_read_lock(); 5838 5839 cpumask_clear(&flush_cpus); 5840 for_each_online_cpu(cpu) { 5841 if (flush_required(cpu)) { 5842 queue_work_on(cpu, system_highpri_wq, 5843 per_cpu_ptr(&flush_works, cpu)); 5844 cpumask_set_cpu(cpu, &flush_cpus); 5845 } 5846 } 5847 5848 /* we can have in flight packet[s] on the cpus we are not flushing, 5849 * synchronize_net() in unregister_netdevice_many() will take care of 5850 * them 5851 */ 5852 for_each_cpu(cpu, &flush_cpus) 5853 flush_work(per_cpu_ptr(&flush_works, cpu)); 5854 5855 cpus_read_unlock(); 5856 } 5857 5858 static void net_rps_send_ipi(struct softnet_data *remsd) 5859 { 5860 #ifdef CONFIG_RPS 5861 while (remsd) { 5862 struct softnet_data *next = remsd->rps_ipi_next; 5863 5864 if (cpu_online(remsd->cpu)) 5865 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5866 remsd = next; 5867 } 5868 #endif 5869 } 5870 5871 /* 5872 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5873 * Note: called with local irq disabled, but exits with local irq enabled. 5874 */ 5875 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5876 { 5877 #ifdef CONFIG_RPS 5878 struct softnet_data *remsd = sd->rps_ipi_list; 5879 5880 if (remsd) { 5881 sd->rps_ipi_list = NULL; 5882 5883 local_irq_enable(); 5884 5885 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5886 net_rps_send_ipi(remsd); 5887 } else 5888 #endif 5889 local_irq_enable(); 5890 } 5891 5892 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5893 { 5894 #ifdef CONFIG_RPS 5895 return sd->rps_ipi_list != NULL; 5896 #else 5897 return false; 5898 #endif 5899 } 5900 5901 static int process_backlog(struct napi_struct *napi, int quota) 5902 { 5903 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5904 bool again = true; 5905 int work = 0; 5906 5907 /* Check if we have pending ipi, its better to send them now, 5908 * not waiting net_rx_action() end. 5909 */ 5910 if (sd_has_rps_ipi_waiting(sd)) { 5911 local_irq_disable(); 5912 net_rps_action_and_irq_enable(sd); 5913 } 5914 5915 napi->weight = READ_ONCE(dev_rx_weight); 5916 while (again) { 5917 struct sk_buff *skb; 5918 5919 while ((skb = __skb_dequeue(&sd->process_queue))) { 5920 rcu_read_lock(); 5921 __netif_receive_skb(skb); 5922 rcu_read_unlock(); 5923 input_queue_head_incr(sd); 5924 if (++work >= quota) 5925 return work; 5926 5927 } 5928 5929 rps_lock_irq_disable(sd); 5930 if (skb_queue_empty(&sd->input_pkt_queue)) { 5931 /* 5932 * Inline a custom version of __napi_complete(). 5933 * only current cpu owns and manipulates this napi, 5934 * and NAPI_STATE_SCHED is the only possible flag set 5935 * on backlog. 5936 * We can use a plain write instead of clear_bit(), 5937 * and we dont need an smp_mb() memory barrier. 5938 */ 5939 napi->state = 0; 5940 again = false; 5941 } else { 5942 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5943 &sd->process_queue); 5944 } 5945 rps_unlock_irq_enable(sd); 5946 } 5947 5948 return work; 5949 } 5950 5951 /** 5952 * __napi_schedule - schedule for receive 5953 * @n: entry to schedule 5954 * 5955 * The entry's receive function will be scheduled to run. 5956 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5957 */ 5958 void __napi_schedule(struct napi_struct *n) 5959 { 5960 unsigned long flags; 5961 5962 local_irq_save(flags); 5963 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5964 local_irq_restore(flags); 5965 } 5966 EXPORT_SYMBOL(__napi_schedule); 5967 5968 /** 5969 * napi_schedule_prep - check if napi can be scheduled 5970 * @n: napi context 5971 * 5972 * Test if NAPI routine is already running, and if not mark 5973 * it as running. This is used as a condition variable to 5974 * insure only one NAPI poll instance runs. We also make 5975 * sure there is no pending NAPI disable. 5976 */ 5977 bool napi_schedule_prep(struct napi_struct *n) 5978 { 5979 unsigned long new, val = READ_ONCE(n->state); 5980 5981 do { 5982 if (unlikely(val & NAPIF_STATE_DISABLE)) 5983 return false; 5984 new = val | NAPIF_STATE_SCHED; 5985 5986 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5987 * This was suggested by Alexander Duyck, as compiler 5988 * emits better code than : 5989 * if (val & NAPIF_STATE_SCHED) 5990 * new |= NAPIF_STATE_MISSED; 5991 */ 5992 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5993 NAPIF_STATE_MISSED; 5994 } while (!try_cmpxchg(&n->state, &val, new)); 5995 5996 return !(val & NAPIF_STATE_SCHED); 5997 } 5998 EXPORT_SYMBOL(napi_schedule_prep); 5999 6000 /** 6001 * __napi_schedule_irqoff - schedule for receive 6002 * @n: entry to schedule 6003 * 6004 * Variant of __napi_schedule() assuming hard irqs are masked. 6005 * 6006 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6007 * because the interrupt disabled assumption might not be true 6008 * due to force-threaded interrupts and spinlock substitution. 6009 */ 6010 void __napi_schedule_irqoff(struct napi_struct *n) 6011 { 6012 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6013 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6014 else 6015 __napi_schedule(n); 6016 } 6017 EXPORT_SYMBOL(__napi_schedule_irqoff); 6018 6019 bool napi_complete_done(struct napi_struct *n, int work_done) 6020 { 6021 unsigned long flags, val, new, timeout = 0; 6022 bool ret = true; 6023 6024 /* 6025 * 1) Don't let napi dequeue from the cpu poll list 6026 * just in case its running on a different cpu. 6027 * 2) If we are busy polling, do nothing here, we have 6028 * the guarantee we will be called later. 6029 */ 6030 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6031 NAPIF_STATE_IN_BUSY_POLL))) 6032 return false; 6033 6034 if (work_done) { 6035 if (n->gro_bitmask) 6036 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6037 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6038 } 6039 if (n->defer_hard_irqs_count > 0) { 6040 n->defer_hard_irqs_count--; 6041 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6042 if (timeout) 6043 ret = false; 6044 } 6045 if (n->gro_bitmask) { 6046 /* When the NAPI instance uses a timeout and keeps postponing 6047 * it, we need to bound somehow the time packets are kept in 6048 * the GRO layer 6049 */ 6050 napi_gro_flush(n, !!timeout); 6051 } 6052 6053 gro_normal_list(n); 6054 6055 if (unlikely(!list_empty(&n->poll_list))) { 6056 /* If n->poll_list is not empty, we need to mask irqs */ 6057 local_irq_save(flags); 6058 list_del_init(&n->poll_list); 6059 local_irq_restore(flags); 6060 } 6061 6062 val = READ_ONCE(n->state); 6063 do { 6064 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6065 6066 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6067 NAPIF_STATE_SCHED_THREADED | 6068 NAPIF_STATE_PREFER_BUSY_POLL); 6069 6070 /* If STATE_MISSED was set, leave STATE_SCHED set, 6071 * because we will call napi->poll() one more time. 6072 * This C code was suggested by Alexander Duyck to help gcc. 6073 */ 6074 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6075 NAPIF_STATE_SCHED; 6076 } while (!try_cmpxchg(&n->state, &val, new)); 6077 6078 if (unlikely(val & NAPIF_STATE_MISSED)) { 6079 __napi_schedule(n); 6080 return false; 6081 } 6082 6083 if (timeout) 6084 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6085 HRTIMER_MODE_REL_PINNED); 6086 return ret; 6087 } 6088 EXPORT_SYMBOL(napi_complete_done); 6089 6090 /* must be called under rcu_read_lock(), as we dont take a reference */ 6091 static struct napi_struct *napi_by_id(unsigned int napi_id) 6092 { 6093 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6094 struct napi_struct *napi; 6095 6096 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6097 if (napi->napi_id == napi_id) 6098 return napi; 6099 6100 return NULL; 6101 } 6102 6103 #if defined(CONFIG_NET_RX_BUSY_POLL) 6104 6105 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6106 { 6107 if (!skip_schedule) { 6108 gro_normal_list(napi); 6109 __napi_schedule(napi); 6110 return; 6111 } 6112 6113 if (napi->gro_bitmask) { 6114 /* flush too old packets 6115 * If HZ < 1000, flush all packets. 6116 */ 6117 napi_gro_flush(napi, HZ >= 1000); 6118 } 6119 6120 gro_normal_list(napi); 6121 clear_bit(NAPI_STATE_SCHED, &napi->state); 6122 } 6123 6124 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6125 u16 budget) 6126 { 6127 bool skip_schedule = false; 6128 unsigned long timeout; 6129 int rc; 6130 6131 /* Busy polling means there is a high chance device driver hard irq 6132 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6133 * set in napi_schedule_prep(). 6134 * Since we are about to call napi->poll() once more, we can safely 6135 * clear NAPI_STATE_MISSED. 6136 * 6137 * Note: x86 could use a single "lock and ..." instruction 6138 * to perform these two clear_bit() 6139 */ 6140 clear_bit(NAPI_STATE_MISSED, &napi->state); 6141 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6142 6143 local_bh_disable(); 6144 6145 if (prefer_busy_poll) { 6146 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6147 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6148 if (napi->defer_hard_irqs_count && timeout) { 6149 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6150 skip_schedule = true; 6151 } 6152 } 6153 6154 /* All we really want here is to re-enable device interrupts. 6155 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6156 */ 6157 rc = napi->poll(napi, budget); 6158 /* We can't gro_normal_list() here, because napi->poll() might have 6159 * rearmed the napi (napi_complete_done()) in which case it could 6160 * already be running on another CPU. 6161 */ 6162 trace_napi_poll(napi, rc, budget); 6163 netpoll_poll_unlock(have_poll_lock); 6164 if (rc == budget) 6165 __busy_poll_stop(napi, skip_schedule); 6166 local_bh_enable(); 6167 } 6168 6169 void napi_busy_loop(unsigned int napi_id, 6170 bool (*loop_end)(void *, unsigned long), 6171 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6172 { 6173 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6174 int (*napi_poll)(struct napi_struct *napi, int budget); 6175 void *have_poll_lock = NULL; 6176 struct napi_struct *napi; 6177 6178 restart: 6179 napi_poll = NULL; 6180 6181 rcu_read_lock(); 6182 6183 napi = napi_by_id(napi_id); 6184 if (!napi) 6185 goto out; 6186 6187 preempt_disable(); 6188 for (;;) { 6189 int work = 0; 6190 6191 local_bh_disable(); 6192 if (!napi_poll) { 6193 unsigned long val = READ_ONCE(napi->state); 6194 6195 /* If multiple threads are competing for this napi, 6196 * we avoid dirtying napi->state as much as we can. 6197 */ 6198 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6199 NAPIF_STATE_IN_BUSY_POLL)) { 6200 if (prefer_busy_poll) 6201 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6202 goto count; 6203 } 6204 if (cmpxchg(&napi->state, val, 6205 val | NAPIF_STATE_IN_BUSY_POLL | 6206 NAPIF_STATE_SCHED) != val) { 6207 if (prefer_busy_poll) 6208 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6209 goto count; 6210 } 6211 have_poll_lock = netpoll_poll_lock(napi); 6212 napi_poll = napi->poll; 6213 } 6214 work = napi_poll(napi, budget); 6215 trace_napi_poll(napi, work, budget); 6216 gro_normal_list(napi); 6217 count: 6218 if (work > 0) 6219 __NET_ADD_STATS(dev_net(napi->dev), 6220 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6221 local_bh_enable(); 6222 6223 if (!loop_end || loop_end(loop_end_arg, start_time)) 6224 break; 6225 6226 if (unlikely(need_resched())) { 6227 if (napi_poll) 6228 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6229 preempt_enable(); 6230 rcu_read_unlock(); 6231 cond_resched(); 6232 if (loop_end(loop_end_arg, start_time)) 6233 return; 6234 goto restart; 6235 } 6236 cpu_relax(); 6237 } 6238 if (napi_poll) 6239 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6240 preempt_enable(); 6241 out: 6242 rcu_read_unlock(); 6243 } 6244 EXPORT_SYMBOL(napi_busy_loop); 6245 6246 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6247 6248 static void napi_hash_add(struct napi_struct *napi) 6249 { 6250 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6251 return; 6252 6253 spin_lock(&napi_hash_lock); 6254 6255 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6256 do { 6257 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6258 napi_gen_id = MIN_NAPI_ID; 6259 } while (napi_by_id(napi_gen_id)); 6260 napi->napi_id = napi_gen_id; 6261 6262 hlist_add_head_rcu(&napi->napi_hash_node, 6263 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6264 6265 spin_unlock(&napi_hash_lock); 6266 } 6267 6268 /* Warning : caller is responsible to make sure rcu grace period 6269 * is respected before freeing memory containing @napi 6270 */ 6271 static void napi_hash_del(struct napi_struct *napi) 6272 { 6273 spin_lock(&napi_hash_lock); 6274 6275 hlist_del_init_rcu(&napi->napi_hash_node); 6276 6277 spin_unlock(&napi_hash_lock); 6278 } 6279 6280 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6281 { 6282 struct napi_struct *napi; 6283 6284 napi = container_of(timer, struct napi_struct, timer); 6285 6286 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6287 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6288 */ 6289 if (!napi_disable_pending(napi) && 6290 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6291 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6292 __napi_schedule_irqoff(napi); 6293 } 6294 6295 return HRTIMER_NORESTART; 6296 } 6297 6298 static void init_gro_hash(struct napi_struct *napi) 6299 { 6300 int i; 6301 6302 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6303 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6304 napi->gro_hash[i].count = 0; 6305 } 6306 napi->gro_bitmask = 0; 6307 } 6308 6309 int dev_set_threaded(struct net_device *dev, bool threaded) 6310 { 6311 struct napi_struct *napi; 6312 int err = 0; 6313 6314 if (dev->threaded == threaded) 6315 return 0; 6316 6317 if (threaded) { 6318 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6319 if (!napi->thread) { 6320 err = napi_kthread_create(napi); 6321 if (err) { 6322 threaded = false; 6323 break; 6324 } 6325 } 6326 } 6327 } 6328 6329 dev->threaded = threaded; 6330 6331 /* Make sure kthread is created before THREADED bit 6332 * is set. 6333 */ 6334 smp_mb__before_atomic(); 6335 6336 /* Setting/unsetting threaded mode on a napi might not immediately 6337 * take effect, if the current napi instance is actively being 6338 * polled. In this case, the switch between threaded mode and 6339 * softirq mode will happen in the next round of napi_schedule(). 6340 * This should not cause hiccups/stalls to the live traffic. 6341 */ 6342 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6343 if (threaded) 6344 set_bit(NAPI_STATE_THREADED, &napi->state); 6345 else 6346 clear_bit(NAPI_STATE_THREADED, &napi->state); 6347 } 6348 6349 return err; 6350 } 6351 EXPORT_SYMBOL(dev_set_threaded); 6352 6353 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6354 int (*poll)(struct napi_struct *, int), int weight) 6355 { 6356 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6357 return; 6358 6359 INIT_LIST_HEAD(&napi->poll_list); 6360 INIT_HLIST_NODE(&napi->napi_hash_node); 6361 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6362 napi->timer.function = napi_watchdog; 6363 init_gro_hash(napi); 6364 napi->skb = NULL; 6365 INIT_LIST_HEAD(&napi->rx_list); 6366 napi->rx_count = 0; 6367 napi->poll = poll; 6368 if (weight > NAPI_POLL_WEIGHT) 6369 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6370 weight); 6371 napi->weight = weight; 6372 napi->dev = dev; 6373 #ifdef CONFIG_NETPOLL 6374 napi->poll_owner = -1; 6375 #endif 6376 set_bit(NAPI_STATE_SCHED, &napi->state); 6377 set_bit(NAPI_STATE_NPSVC, &napi->state); 6378 list_add_rcu(&napi->dev_list, &dev->napi_list); 6379 napi_hash_add(napi); 6380 napi_get_frags_check(napi); 6381 /* Create kthread for this napi if dev->threaded is set. 6382 * Clear dev->threaded if kthread creation failed so that 6383 * threaded mode will not be enabled in napi_enable(). 6384 */ 6385 if (dev->threaded && napi_kthread_create(napi)) 6386 dev->threaded = 0; 6387 } 6388 EXPORT_SYMBOL(netif_napi_add_weight); 6389 6390 void napi_disable(struct napi_struct *n) 6391 { 6392 unsigned long val, new; 6393 6394 might_sleep(); 6395 set_bit(NAPI_STATE_DISABLE, &n->state); 6396 6397 val = READ_ONCE(n->state); 6398 do { 6399 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6400 usleep_range(20, 200); 6401 val = READ_ONCE(n->state); 6402 } 6403 6404 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6405 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6406 } while (!try_cmpxchg(&n->state, &val, new)); 6407 6408 hrtimer_cancel(&n->timer); 6409 6410 clear_bit(NAPI_STATE_DISABLE, &n->state); 6411 } 6412 EXPORT_SYMBOL(napi_disable); 6413 6414 /** 6415 * napi_enable - enable NAPI scheduling 6416 * @n: NAPI context 6417 * 6418 * Resume NAPI from being scheduled on this context. 6419 * Must be paired with napi_disable. 6420 */ 6421 void napi_enable(struct napi_struct *n) 6422 { 6423 unsigned long new, val = READ_ONCE(n->state); 6424 6425 do { 6426 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6427 6428 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6429 if (n->dev->threaded && n->thread) 6430 new |= NAPIF_STATE_THREADED; 6431 } while (!try_cmpxchg(&n->state, &val, new)); 6432 } 6433 EXPORT_SYMBOL(napi_enable); 6434 6435 static void flush_gro_hash(struct napi_struct *napi) 6436 { 6437 int i; 6438 6439 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6440 struct sk_buff *skb, *n; 6441 6442 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6443 kfree_skb(skb); 6444 napi->gro_hash[i].count = 0; 6445 } 6446 } 6447 6448 /* Must be called in process context */ 6449 void __netif_napi_del(struct napi_struct *napi) 6450 { 6451 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6452 return; 6453 6454 napi_hash_del(napi); 6455 list_del_rcu(&napi->dev_list); 6456 napi_free_frags(napi); 6457 6458 flush_gro_hash(napi); 6459 napi->gro_bitmask = 0; 6460 6461 if (napi->thread) { 6462 kthread_stop(napi->thread); 6463 napi->thread = NULL; 6464 } 6465 } 6466 EXPORT_SYMBOL(__netif_napi_del); 6467 6468 static int __napi_poll(struct napi_struct *n, bool *repoll) 6469 { 6470 int work, weight; 6471 6472 weight = n->weight; 6473 6474 /* This NAPI_STATE_SCHED test is for avoiding a race 6475 * with netpoll's poll_napi(). Only the entity which 6476 * obtains the lock and sees NAPI_STATE_SCHED set will 6477 * actually make the ->poll() call. Therefore we avoid 6478 * accidentally calling ->poll() when NAPI is not scheduled. 6479 */ 6480 work = 0; 6481 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6482 work = n->poll(n, weight); 6483 trace_napi_poll(n, work, weight); 6484 } 6485 6486 if (unlikely(work > weight)) 6487 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6488 n->poll, work, weight); 6489 6490 if (likely(work < weight)) 6491 return work; 6492 6493 /* Drivers must not modify the NAPI state if they 6494 * consume the entire weight. In such cases this code 6495 * still "owns" the NAPI instance and therefore can 6496 * move the instance around on the list at-will. 6497 */ 6498 if (unlikely(napi_disable_pending(n))) { 6499 napi_complete(n); 6500 return work; 6501 } 6502 6503 /* The NAPI context has more processing work, but busy-polling 6504 * is preferred. Exit early. 6505 */ 6506 if (napi_prefer_busy_poll(n)) { 6507 if (napi_complete_done(n, work)) { 6508 /* If timeout is not set, we need to make sure 6509 * that the NAPI is re-scheduled. 6510 */ 6511 napi_schedule(n); 6512 } 6513 return work; 6514 } 6515 6516 if (n->gro_bitmask) { 6517 /* flush too old packets 6518 * If HZ < 1000, flush all packets. 6519 */ 6520 napi_gro_flush(n, HZ >= 1000); 6521 } 6522 6523 gro_normal_list(n); 6524 6525 /* Some drivers may have called napi_schedule 6526 * prior to exhausting their budget. 6527 */ 6528 if (unlikely(!list_empty(&n->poll_list))) { 6529 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6530 n->dev ? n->dev->name : "backlog"); 6531 return work; 6532 } 6533 6534 *repoll = true; 6535 6536 return work; 6537 } 6538 6539 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6540 { 6541 bool do_repoll = false; 6542 void *have; 6543 int work; 6544 6545 list_del_init(&n->poll_list); 6546 6547 have = netpoll_poll_lock(n); 6548 6549 work = __napi_poll(n, &do_repoll); 6550 6551 if (do_repoll) 6552 list_add_tail(&n->poll_list, repoll); 6553 6554 netpoll_poll_unlock(have); 6555 6556 return work; 6557 } 6558 6559 static int napi_thread_wait(struct napi_struct *napi) 6560 { 6561 bool woken = false; 6562 6563 set_current_state(TASK_INTERRUPTIBLE); 6564 6565 while (!kthread_should_stop()) { 6566 /* Testing SCHED_THREADED bit here to make sure the current 6567 * kthread owns this napi and could poll on this napi. 6568 * Testing SCHED bit is not enough because SCHED bit might be 6569 * set by some other busy poll thread or by napi_disable(). 6570 */ 6571 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6572 WARN_ON(!list_empty(&napi->poll_list)); 6573 __set_current_state(TASK_RUNNING); 6574 return 0; 6575 } 6576 6577 schedule(); 6578 /* woken being true indicates this thread owns this napi. */ 6579 woken = true; 6580 set_current_state(TASK_INTERRUPTIBLE); 6581 } 6582 __set_current_state(TASK_RUNNING); 6583 6584 return -1; 6585 } 6586 6587 static int napi_threaded_poll(void *data) 6588 { 6589 struct napi_struct *napi = data; 6590 void *have; 6591 6592 while (!napi_thread_wait(napi)) { 6593 for (;;) { 6594 bool repoll = false; 6595 6596 local_bh_disable(); 6597 6598 have = netpoll_poll_lock(napi); 6599 __napi_poll(napi, &repoll); 6600 netpoll_poll_unlock(have); 6601 6602 local_bh_enable(); 6603 6604 if (!repoll) 6605 break; 6606 6607 cond_resched(); 6608 } 6609 } 6610 return 0; 6611 } 6612 6613 static void skb_defer_free_flush(struct softnet_data *sd) 6614 { 6615 struct sk_buff *skb, *next; 6616 6617 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6618 if (!READ_ONCE(sd->defer_list)) 6619 return; 6620 6621 spin_lock_irq(&sd->defer_lock); 6622 skb = sd->defer_list; 6623 sd->defer_list = NULL; 6624 sd->defer_count = 0; 6625 spin_unlock_irq(&sd->defer_lock); 6626 6627 while (skb != NULL) { 6628 next = skb->next; 6629 napi_consume_skb(skb, 1); 6630 skb = next; 6631 } 6632 } 6633 6634 static __latent_entropy void net_rx_action(struct softirq_action *h) 6635 { 6636 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6637 unsigned long time_limit = jiffies + 6638 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6639 int budget = READ_ONCE(netdev_budget); 6640 LIST_HEAD(list); 6641 LIST_HEAD(repoll); 6642 6643 local_irq_disable(); 6644 list_splice_init(&sd->poll_list, &list); 6645 local_irq_enable(); 6646 6647 for (;;) { 6648 struct napi_struct *n; 6649 6650 skb_defer_free_flush(sd); 6651 6652 if (list_empty(&list)) { 6653 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6654 goto end; 6655 break; 6656 } 6657 6658 n = list_first_entry(&list, struct napi_struct, poll_list); 6659 budget -= napi_poll(n, &repoll); 6660 6661 /* If softirq window is exhausted then punt. 6662 * Allow this to run for 2 jiffies since which will allow 6663 * an average latency of 1.5/HZ. 6664 */ 6665 if (unlikely(budget <= 0 || 6666 time_after_eq(jiffies, time_limit))) { 6667 sd->time_squeeze++; 6668 break; 6669 } 6670 } 6671 6672 local_irq_disable(); 6673 6674 list_splice_tail_init(&sd->poll_list, &list); 6675 list_splice_tail(&repoll, &list); 6676 list_splice(&list, &sd->poll_list); 6677 if (!list_empty(&sd->poll_list)) 6678 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6679 6680 net_rps_action_and_irq_enable(sd); 6681 end:; 6682 } 6683 6684 struct netdev_adjacent { 6685 struct net_device *dev; 6686 netdevice_tracker dev_tracker; 6687 6688 /* upper master flag, there can only be one master device per list */ 6689 bool master; 6690 6691 /* lookup ignore flag */ 6692 bool ignore; 6693 6694 /* counter for the number of times this device was added to us */ 6695 u16 ref_nr; 6696 6697 /* private field for the users */ 6698 void *private; 6699 6700 struct list_head list; 6701 struct rcu_head rcu; 6702 }; 6703 6704 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6705 struct list_head *adj_list) 6706 { 6707 struct netdev_adjacent *adj; 6708 6709 list_for_each_entry(adj, adj_list, list) { 6710 if (adj->dev == adj_dev) 6711 return adj; 6712 } 6713 return NULL; 6714 } 6715 6716 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6717 struct netdev_nested_priv *priv) 6718 { 6719 struct net_device *dev = (struct net_device *)priv->data; 6720 6721 return upper_dev == dev; 6722 } 6723 6724 /** 6725 * netdev_has_upper_dev - Check if device is linked to an upper device 6726 * @dev: device 6727 * @upper_dev: upper device to check 6728 * 6729 * Find out if a device is linked to specified upper device and return true 6730 * in case it is. Note that this checks only immediate upper device, 6731 * not through a complete stack of devices. The caller must hold the RTNL lock. 6732 */ 6733 bool netdev_has_upper_dev(struct net_device *dev, 6734 struct net_device *upper_dev) 6735 { 6736 struct netdev_nested_priv priv = { 6737 .data = (void *)upper_dev, 6738 }; 6739 6740 ASSERT_RTNL(); 6741 6742 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6743 &priv); 6744 } 6745 EXPORT_SYMBOL(netdev_has_upper_dev); 6746 6747 /** 6748 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6749 * @dev: device 6750 * @upper_dev: upper device to check 6751 * 6752 * Find out if a device is linked to specified upper device and return true 6753 * in case it is. Note that this checks the entire upper device chain. 6754 * The caller must hold rcu lock. 6755 */ 6756 6757 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6758 struct net_device *upper_dev) 6759 { 6760 struct netdev_nested_priv priv = { 6761 .data = (void *)upper_dev, 6762 }; 6763 6764 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6765 &priv); 6766 } 6767 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6768 6769 /** 6770 * netdev_has_any_upper_dev - Check if device is linked to some device 6771 * @dev: device 6772 * 6773 * Find out if a device is linked to an upper device and return true in case 6774 * it is. The caller must hold the RTNL lock. 6775 */ 6776 bool netdev_has_any_upper_dev(struct net_device *dev) 6777 { 6778 ASSERT_RTNL(); 6779 6780 return !list_empty(&dev->adj_list.upper); 6781 } 6782 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6783 6784 /** 6785 * netdev_master_upper_dev_get - Get master upper device 6786 * @dev: device 6787 * 6788 * Find a master upper device and return pointer to it or NULL in case 6789 * it's not there. The caller must hold the RTNL lock. 6790 */ 6791 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6792 { 6793 struct netdev_adjacent *upper; 6794 6795 ASSERT_RTNL(); 6796 6797 if (list_empty(&dev->adj_list.upper)) 6798 return NULL; 6799 6800 upper = list_first_entry(&dev->adj_list.upper, 6801 struct netdev_adjacent, list); 6802 if (likely(upper->master)) 6803 return upper->dev; 6804 return NULL; 6805 } 6806 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6807 6808 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6809 { 6810 struct netdev_adjacent *upper; 6811 6812 ASSERT_RTNL(); 6813 6814 if (list_empty(&dev->adj_list.upper)) 6815 return NULL; 6816 6817 upper = list_first_entry(&dev->adj_list.upper, 6818 struct netdev_adjacent, list); 6819 if (likely(upper->master) && !upper->ignore) 6820 return upper->dev; 6821 return NULL; 6822 } 6823 6824 /** 6825 * netdev_has_any_lower_dev - Check if device is linked to some device 6826 * @dev: device 6827 * 6828 * Find out if a device is linked to a lower device and return true in case 6829 * it is. The caller must hold the RTNL lock. 6830 */ 6831 static bool netdev_has_any_lower_dev(struct net_device *dev) 6832 { 6833 ASSERT_RTNL(); 6834 6835 return !list_empty(&dev->adj_list.lower); 6836 } 6837 6838 void *netdev_adjacent_get_private(struct list_head *adj_list) 6839 { 6840 struct netdev_adjacent *adj; 6841 6842 adj = list_entry(adj_list, struct netdev_adjacent, list); 6843 6844 return adj->private; 6845 } 6846 EXPORT_SYMBOL(netdev_adjacent_get_private); 6847 6848 /** 6849 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6850 * @dev: device 6851 * @iter: list_head ** of the current position 6852 * 6853 * Gets the next device from the dev's upper list, starting from iter 6854 * position. The caller must hold RCU read lock. 6855 */ 6856 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6857 struct list_head **iter) 6858 { 6859 struct netdev_adjacent *upper; 6860 6861 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6862 6863 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6864 6865 if (&upper->list == &dev->adj_list.upper) 6866 return NULL; 6867 6868 *iter = &upper->list; 6869 6870 return upper->dev; 6871 } 6872 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6873 6874 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6875 struct list_head **iter, 6876 bool *ignore) 6877 { 6878 struct netdev_adjacent *upper; 6879 6880 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6881 6882 if (&upper->list == &dev->adj_list.upper) 6883 return NULL; 6884 6885 *iter = &upper->list; 6886 *ignore = upper->ignore; 6887 6888 return upper->dev; 6889 } 6890 6891 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6892 struct list_head **iter) 6893 { 6894 struct netdev_adjacent *upper; 6895 6896 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6897 6898 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6899 6900 if (&upper->list == &dev->adj_list.upper) 6901 return NULL; 6902 6903 *iter = &upper->list; 6904 6905 return upper->dev; 6906 } 6907 6908 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6909 int (*fn)(struct net_device *dev, 6910 struct netdev_nested_priv *priv), 6911 struct netdev_nested_priv *priv) 6912 { 6913 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6914 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6915 int ret, cur = 0; 6916 bool ignore; 6917 6918 now = dev; 6919 iter = &dev->adj_list.upper; 6920 6921 while (1) { 6922 if (now != dev) { 6923 ret = fn(now, priv); 6924 if (ret) 6925 return ret; 6926 } 6927 6928 next = NULL; 6929 while (1) { 6930 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6931 if (!udev) 6932 break; 6933 if (ignore) 6934 continue; 6935 6936 next = udev; 6937 niter = &udev->adj_list.upper; 6938 dev_stack[cur] = now; 6939 iter_stack[cur++] = iter; 6940 break; 6941 } 6942 6943 if (!next) { 6944 if (!cur) 6945 return 0; 6946 next = dev_stack[--cur]; 6947 niter = iter_stack[cur]; 6948 } 6949 6950 now = next; 6951 iter = niter; 6952 } 6953 6954 return 0; 6955 } 6956 6957 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6958 int (*fn)(struct net_device *dev, 6959 struct netdev_nested_priv *priv), 6960 struct netdev_nested_priv *priv) 6961 { 6962 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6963 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6964 int ret, cur = 0; 6965 6966 now = dev; 6967 iter = &dev->adj_list.upper; 6968 6969 while (1) { 6970 if (now != dev) { 6971 ret = fn(now, priv); 6972 if (ret) 6973 return ret; 6974 } 6975 6976 next = NULL; 6977 while (1) { 6978 udev = netdev_next_upper_dev_rcu(now, &iter); 6979 if (!udev) 6980 break; 6981 6982 next = udev; 6983 niter = &udev->adj_list.upper; 6984 dev_stack[cur] = now; 6985 iter_stack[cur++] = iter; 6986 break; 6987 } 6988 6989 if (!next) { 6990 if (!cur) 6991 return 0; 6992 next = dev_stack[--cur]; 6993 niter = iter_stack[cur]; 6994 } 6995 6996 now = next; 6997 iter = niter; 6998 } 6999 7000 return 0; 7001 } 7002 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7003 7004 static bool __netdev_has_upper_dev(struct net_device *dev, 7005 struct net_device *upper_dev) 7006 { 7007 struct netdev_nested_priv priv = { 7008 .flags = 0, 7009 .data = (void *)upper_dev, 7010 }; 7011 7012 ASSERT_RTNL(); 7013 7014 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7015 &priv); 7016 } 7017 7018 /** 7019 * netdev_lower_get_next_private - Get the next ->private from the 7020 * lower neighbour list 7021 * @dev: device 7022 * @iter: list_head ** of the current position 7023 * 7024 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7025 * list, starting from iter position. The caller must hold either hold the 7026 * RTNL lock or its own locking that guarantees that the neighbour lower 7027 * list will remain unchanged. 7028 */ 7029 void *netdev_lower_get_next_private(struct net_device *dev, 7030 struct list_head **iter) 7031 { 7032 struct netdev_adjacent *lower; 7033 7034 lower = list_entry(*iter, struct netdev_adjacent, list); 7035 7036 if (&lower->list == &dev->adj_list.lower) 7037 return NULL; 7038 7039 *iter = lower->list.next; 7040 7041 return lower->private; 7042 } 7043 EXPORT_SYMBOL(netdev_lower_get_next_private); 7044 7045 /** 7046 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7047 * lower neighbour list, RCU 7048 * variant 7049 * @dev: device 7050 * @iter: list_head ** of the current position 7051 * 7052 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7053 * list, starting from iter position. The caller must hold RCU read lock. 7054 */ 7055 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7056 struct list_head **iter) 7057 { 7058 struct netdev_adjacent *lower; 7059 7060 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7061 7062 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7063 7064 if (&lower->list == &dev->adj_list.lower) 7065 return NULL; 7066 7067 *iter = &lower->list; 7068 7069 return lower->private; 7070 } 7071 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7072 7073 /** 7074 * netdev_lower_get_next - Get the next device from the lower neighbour 7075 * list 7076 * @dev: device 7077 * @iter: list_head ** of the current position 7078 * 7079 * Gets the next netdev_adjacent from the dev's lower neighbour 7080 * list, starting from iter position. The caller must hold RTNL lock or 7081 * its own locking that guarantees that the neighbour lower 7082 * list will remain unchanged. 7083 */ 7084 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7085 { 7086 struct netdev_adjacent *lower; 7087 7088 lower = list_entry(*iter, struct netdev_adjacent, list); 7089 7090 if (&lower->list == &dev->adj_list.lower) 7091 return NULL; 7092 7093 *iter = lower->list.next; 7094 7095 return lower->dev; 7096 } 7097 EXPORT_SYMBOL(netdev_lower_get_next); 7098 7099 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7100 struct list_head **iter) 7101 { 7102 struct netdev_adjacent *lower; 7103 7104 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7105 7106 if (&lower->list == &dev->adj_list.lower) 7107 return NULL; 7108 7109 *iter = &lower->list; 7110 7111 return lower->dev; 7112 } 7113 7114 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7115 struct list_head **iter, 7116 bool *ignore) 7117 { 7118 struct netdev_adjacent *lower; 7119 7120 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7121 7122 if (&lower->list == &dev->adj_list.lower) 7123 return NULL; 7124 7125 *iter = &lower->list; 7126 *ignore = lower->ignore; 7127 7128 return lower->dev; 7129 } 7130 7131 int netdev_walk_all_lower_dev(struct net_device *dev, 7132 int (*fn)(struct net_device *dev, 7133 struct netdev_nested_priv *priv), 7134 struct netdev_nested_priv *priv) 7135 { 7136 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7137 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7138 int ret, cur = 0; 7139 7140 now = dev; 7141 iter = &dev->adj_list.lower; 7142 7143 while (1) { 7144 if (now != dev) { 7145 ret = fn(now, priv); 7146 if (ret) 7147 return ret; 7148 } 7149 7150 next = NULL; 7151 while (1) { 7152 ldev = netdev_next_lower_dev(now, &iter); 7153 if (!ldev) 7154 break; 7155 7156 next = ldev; 7157 niter = &ldev->adj_list.lower; 7158 dev_stack[cur] = now; 7159 iter_stack[cur++] = iter; 7160 break; 7161 } 7162 7163 if (!next) { 7164 if (!cur) 7165 return 0; 7166 next = dev_stack[--cur]; 7167 niter = iter_stack[cur]; 7168 } 7169 7170 now = next; 7171 iter = niter; 7172 } 7173 7174 return 0; 7175 } 7176 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7177 7178 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7179 int (*fn)(struct net_device *dev, 7180 struct netdev_nested_priv *priv), 7181 struct netdev_nested_priv *priv) 7182 { 7183 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7184 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7185 int ret, cur = 0; 7186 bool ignore; 7187 7188 now = dev; 7189 iter = &dev->adj_list.lower; 7190 7191 while (1) { 7192 if (now != dev) { 7193 ret = fn(now, priv); 7194 if (ret) 7195 return ret; 7196 } 7197 7198 next = NULL; 7199 while (1) { 7200 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7201 if (!ldev) 7202 break; 7203 if (ignore) 7204 continue; 7205 7206 next = ldev; 7207 niter = &ldev->adj_list.lower; 7208 dev_stack[cur] = now; 7209 iter_stack[cur++] = iter; 7210 break; 7211 } 7212 7213 if (!next) { 7214 if (!cur) 7215 return 0; 7216 next = dev_stack[--cur]; 7217 niter = iter_stack[cur]; 7218 } 7219 7220 now = next; 7221 iter = niter; 7222 } 7223 7224 return 0; 7225 } 7226 7227 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7228 struct list_head **iter) 7229 { 7230 struct netdev_adjacent *lower; 7231 7232 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7233 if (&lower->list == &dev->adj_list.lower) 7234 return NULL; 7235 7236 *iter = &lower->list; 7237 7238 return lower->dev; 7239 } 7240 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7241 7242 static u8 __netdev_upper_depth(struct net_device *dev) 7243 { 7244 struct net_device *udev; 7245 struct list_head *iter; 7246 u8 max_depth = 0; 7247 bool ignore; 7248 7249 for (iter = &dev->adj_list.upper, 7250 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7251 udev; 7252 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7253 if (ignore) 7254 continue; 7255 if (max_depth < udev->upper_level) 7256 max_depth = udev->upper_level; 7257 } 7258 7259 return max_depth; 7260 } 7261 7262 static u8 __netdev_lower_depth(struct net_device *dev) 7263 { 7264 struct net_device *ldev; 7265 struct list_head *iter; 7266 u8 max_depth = 0; 7267 bool ignore; 7268 7269 for (iter = &dev->adj_list.lower, 7270 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7271 ldev; 7272 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7273 if (ignore) 7274 continue; 7275 if (max_depth < ldev->lower_level) 7276 max_depth = ldev->lower_level; 7277 } 7278 7279 return max_depth; 7280 } 7281 7282 static int __netdev_update_upper_level(struct net_device *dev, 7283 struct netdev_nested_priv *__unused) 7284 { 7285 dev->upper_level = __netdev_upper_depth(dev) + 1; 7286 return 0; 7287 } 7288 7289 #ifdef CONFIG_LOCKDEP 7290 static LIST_HEAD(net_unlink_list); 7291 7292 static void net_unlink_todo(struct net_device *dev) 7293 { 7294 if (list_empty(&dev->unlink_list)) 7295 list_add_tail(&dev->unlink_list, &net_unlink_list); 7296 } 7297 #endif 7298 7299 static int __netdev_update_lower_level(struct net_device *dev, 7300 struct netdev_nested_priv *priv) 7301 { 7302 dev->lower_level = __netdev_lower_depth(dev) + 1; 7303 7304 #ifdef CONFIG_LOCKDEP 7305 if (!priv) 7306 return 0; 7307 7308 if (priv->flags & NESTED_SYNC_IMM) 7309 dev->nested_level = dev->lower_level - 1; 7310 if (priv->flags & NESTED_SYNC_TODO) 7311 net_unlink_todo(dev); 7312 #endif 7313 return 0; 7314 } 7315 7316 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7317 int (*fn)(struct net_device *dev, 7318 struct netdev_nested_priv *priv), 7319 struct netdev_nested_priv *priv) 7320 { 7321 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7322 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7323 int ret, cur = 0; 7324 7325 now = dev; 7326 iter = &dev->adj_list.lower; 7327 7328 while (1) { 7329 if (now != dev) { 7330 ret = fn(now, priv); 7331 if (ret) 7332 return ret; 7333 } 7334 7335 next = NULL; 7336 while (1) { 7337 ldev = netdev_next_lower_dev_rcu(now, &iter); 7338 if (!ldev) 7339 break; 7340 7341 next = ldev; 7342 niter = &ldev->adj_list.lower; 7343 dev_stack[cur] = now; 7344 iter_stack[cur++] = iter; 7345 break; 7346 } 7347 7348 if (!next) { 7349 if (!cur) 7350 return 0; 7351 next = dev_stack[--cur]; 7352 niter = iter_stack[cur]; 7353 } 7354 7355 now = next; 7356 iter = niter; 7357 } 7358 7359 return 0; 7360 } 7361 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7362 7363 /** 7364 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7365 * lower neighbour list, RCU 7366 * variant 7367 * @dev: device 7368 * 7369 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7370 * list. The caller must hold RCU read lock. 7371 */ 7372 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7373 { 7374 struct netdev_adjacent *lower; 7375 7376 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7377 struct netdev_adjacent, list); 7378 if (lower) 7379 return lower->private; 7380 return NULL; 7381 } 7382 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7383 7384 /** 7385 * netdev_master_upper_dev_get_rcu - Get master upper device 7386 * @dev: device 7387 * 7388 * Find a master upper device and return pointer to it or NULL in case 7389 * it's not there. The caller must hold the RCU read lock. 7390 */ 7391 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7392 { 7393 struct netdev_adjacent *upper; 7394 7395 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7396 struct netdev_adjacent, list); 7397 if (upper && likely(upper->master)) 7398 return upper->dev; 7399 return NULL; 7400 } 7401 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7402 7403 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7404 struct net_device *adj_dev, 7405 struct list_head *dev_list) 7406 { 7407 char linkname[IFNAMSIZ+7]; 7408 7409 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7410 "upper_%s" : "lower_%s", adj_dev->name); 7411 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7412 linkname); 7413 } 7414 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7415 char *name, 7416 struct list_head *dev_list) 7417 { 7418 char linkname[IFNAMSIZ+7]; 7419 7420 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7421 "upper_%s" : "lower_%s", name); 7422 sysfs_remove_link(&(dev->dev.kobj), linkname); 7423 } 7424 7425 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7426 struct net_device *adj_dev, 7427 struct list_head *dev_list) 7428 { 7429 return (dev_list == &dev->adj_list.upper || 7430 dev_list == &dev->adj_list.lower) && 7431 net_eq(dev_net(dev), dev_net(adj_dev)); 7432 } 7433 7434 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7435 struct net_device *adj_dev, 7436 struct list_head *dev_list, 7437 void *private, bool master) 7438 { 7439 struct netdev_adjacent *adj; 7440 int ret; 7441 7442 adj = __netdev_find_adj(adj_dev, dev_list); 7443 7444 if (adj) { 7445 adj->ref_nr += 1; 7446 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7447 dev->name, adj_dev->name, adj->ref_nr); 7448 7449 return 0; 7450 } 7451 7452 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7453 if (!adj) 7454 return -ENOMEM; 7455 7456 adj->dev = adj_dev; 7457 adj->master = master; 7458 adj->ref_nr = 1; 7459 adj->private = private; 7460 adj->ignore = false; 7461 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7462 7463 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7464 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7465 7466 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7467 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7468 if (ret) 7469 goto free_adj; 7470 } 7471 7472 /* Ensure that master link is always the first item in list. */ 7473 if (master) { 7474 ret = sysfs_create_link(&(dev->dev.kobj), 7475 &(adj_dev->dev.kobj), "master"); 7476 if (ret) 7477 goto remove_symlinks; 7478 7479 list_add_rcu(&adj->list, dev_list); 7480 } else { 7481 list_add_tail_rcu(&adj->list, dev_list); 7482 } 7483 7484 return 0; 7485 7486 remove_symlinks: 7487 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7488 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7489 free_adj: 7490 netdev_put(adj_dev, &adj->dev_tracker); 7491 kfree(adj); 7492 7493 return ret; 7494 } 7495 7496 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7497 struct net_device *adj_dev, 7498 u16 ref_nr, 7499 struct list_head *dev_list) 7500 { 7501 struct netdev_adjacent *adj; 7502 7503 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7504 dev->name, adj_dev->name, ref_nr); 7505 7506 adj = __netdev_find_adj(adj_dev, dev_list); 7507 7508 if (!adj) { 7509 pr_err("Adjacency does not exist for device %s from %s\n", 7510 dev->name, adj_dev->name); 7511 WARN_ON(1); 7512 return; 7513 } 7514 7515 if (adj->ref_nr > ref_nr) { 7516 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7517 dev->name, adj_dev->name, ref_nr, 7518 adj->ref_nr - ref_nr); 7519 adj->ref_nr -= ref_nr; 7520 return; 7521 } 7522 7523 if (adj->master) 7524 sysfs_remove_link(&(dev->dev.kobj), "master"); 7525 7526 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7527 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7528 7529 list_del_rcu(&adj->list); 7530 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7531 adj_dev->name, dev->name, adj_dev->name); 7532 netdev_put(adj_dev, &adj->dev_tracker); 7533 kfree_rcu(adj, rcu); 7534 } 7535 7536 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7537 struct net_device *upper_dev, 7538 struct list_head *up_list, 7539 struct list_head *down_list, 7540 void *private, bool master) 7541 { 7542 int ret; 7543 7544 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7545 private, master); 7546 if (ret) 7547 return ret; 7548 7549 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7550 private, false); 7551 if (ret) { 7552 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7553 return ret; 7554 } 7555 7556 return 0; 7557 } 7558 7559 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7560 struct net_device *upper_dev, 7561 u16 ref_nr, 7562 struct list_head *up_list, 7563 struct list_head *down_list) 7564 { 7565 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7566 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7567 } 7568 7569 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7570 struct net_device *upper_dev, 7571 void *private, bool master) 7572 { 7573 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7574 &dev->adj_list.upper, 7575 &upper_dev->adj_list.lower, 7576 private, master); 7577 } 7578 7579 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7580 struct net_device *upper_dev) 7581 { 7582 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7583 &dev->adj_list.upper, 7584 &upper_dev->adj_list.lower); 7585 } 7586 7587 static int __netdev_upper_dev_link(struct net_device *dev, 7588 struct net_device *upper_dev, bool master, 7589 void *upper_priv, void *upper_info, 7590 struct netdev_nested_priv *priv, 7591 struct netlink_ext_ack *extack) 7592 { 7593 struct netdev_notifier_changeupper_info changeupper_info = { 7594 .info = { 7595 .dev = dev, 7596 .extack = extack, 7597 }, 7598 .upper_dev = upper_dev, 7599 .master = master, 7600 .linking = true, 7601 .upper_info = upper_info, 7602 }; 7603 struct net_device *master_dev; 7604 int ret = 0; 7605 7606 ASSERT_RTNL(); 7607 7608 if (dev == upper_dev) 7609 return -EBUSY; 7610 7611 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7612 if (__netdev_has_upper_dev(upper_dev, dev)) 7613 return -EBUSY; 7614 7615 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7616 return -EMLINK; 7617 7618 if (!master) { 7619 if (__netdev_has_upper_dev(dev, upper_dev)) 7620 return -EEXIST; 7621 } else { 7622 master_dev = __netdev_master_upper_dev_get(dev); 7623 if (master_dev) 7624 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7625 } 7626 7627 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7628 &changeupper_info.info); 7629 ret = notifier_to_errno(ret); 7630 if (ret) 7631 return ret; 7632 7633 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7634 master); 7635 if (ret) 7636 return ret; 7637 7638 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7639 &changeupper_info.info); 7640 ret = notifier_to_errno(ret); 7641 if (ret) 7642 goto rollback; 7643 7644 __netdev_update_upper_level(dev, NULL); 7645 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7646 7647 __netdev_update_lower_level(upper_dev, priv); 7648 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7649 priv); 7650 7651 return 0; 7652 7653 rollback: 7654 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7655 7656 return ret; 7657 } 7658 7659 /** 7660 * netdev_upper_dev_link - Add a link to the upper device 7661 * @dev: device 7662 * @upper_dev: new upper device 7663 * @extack: netlink extended ack 7664 * 7665 * Adds a link to device which is upper to this one. The caller must hold 7666 * the RTNL lock. On a failure a negative errno code is returned. 7667 * On success the reference counts are adjusted and the function 7668 * returns zero. 7669 */ 7670 int netdev_upper_dev_link(struct net_device *dev, 7671 struct net_device *upper_dev, 7672 struct netlink_ext_ack *extack) 7673 { 7674 struct netdev_nested_priv priv = { 7675 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7676 .data = NULL, 7677 }; 7678 7679 return __netdev_upper_dev_link(dev, upper_dev, false, 7680 NULL, NULL, &priv, extack); 7681 } 7682 EXPORT_SYMBOL(netdev_upper_dev_link); 7683 7684 /** 7685 * netdev_master_upper_dev_link - Add a master link to the upper device 7686 * @dev: device 7687 * @upper_dev: new upper device 7688 * @upper_priv: upper device private 7689 * @upper_info: upper info to be passed down via notifier 7690 * @extack: netlink extended ack 7691 * 7692 * Adds a link to device which is upper to this one. In this case, only 7693 * one master upper device can be linked, although other non-master devices 7694 * might be linked as well. The caller must hold the RTNL lock. 7695 * On a failure a negative errno code is returned. On success the reference 7696 * counts are adjusted and the function returns zero. 7697 */ 7698 int netdev_master_upper_dev_link(struct net_device *dev, 7699 struct net_device *upper_dev, 7700 void *upper_priv, void *upper_info, 7701 struct netlink_ext_ack *extack) 7702 { 7703 struct netdev_nested_priv priv = { 7704 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7705 .data = NULL, 7706 }; 7707 7708 return __netdev_upper_dev_link(dev, upper_dev, true, 7709 upper_priv, upper_info, &priv, extack); 7710 } 7711 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7712 7713 static void __netdev_upper_dev_unlink(struct net_device *dev, 7714 struct net_device *upper_dev, 7715 struct netdev_nested_priv *priv) 7716 { 7717 struct netdev_notifier_changeupper_info changeupper_info = { 7718 .info = { 7719 .dev = dev, 7720 }, 7721 .upper_dev = upper_dev, 7722 .linking = false, 7723 }; 7724 7725 ASSERT_RTNL(); 7726 7727 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7728 7729 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7730 &changeupper_info.info); 7731 7732 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7733 7734 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7735 &changeupper_info.info); 7736 7737 __netdev_update_upper_level(dev, NULL); 7738 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7739 7740 __netdev_update_lower_level(upper_dev, priv); 7741 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7742 priv); 7743 } 7744 7745 /** 7746 * netdev_upper_dev_unlink - Removes a link to upper device 7747 * @dev: device 7748 * @upper_dev: new upper device 7749 * 7750 * Removes a link to device which is upper to this one. The caller must hold 7751 * the RTNL lock. 7752 */ 7753 void netdev_upper_dev_unlink(struct net_device *dev, 7754 struct net_device *upper_dev) 7755 { 7756 struct netdev_nested_priv priv = { 7757 .flags = NESTED_SYNC_TODO, 7758 .data = NULL, 7759 }; 7760 7761 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7762 } 7763 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7764 7765 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7766 struct net_device *lower_dev, 7767 bool val) 7768 { 7769 struct netdev_adjacent *adj; 7770 7771 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7772 if (adj) 7773 adj->ignore = val; 7774 7775 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7776 if (adj) 7777 adj->ignore = val; 7778 } 7779 7780 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7781 struct net_device *lower_dev) 7782 { 7783 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7784 } 7785 7786 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7787 struct net_device *lower_dev) 7788 { 7789 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7790 } 7791 7792 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7793 struct net_device *new_dev, 7794 struct net_device *dev, 7795 struct netlink_ext_ack *extack) 7796 { 7797 struct netdev_nested_priv priv = { 7798 .flags = 0, 7799 .data = NULL, 7800 }; 7801 int err; 7802 7803 if (!new_dev) 7804 return 0; 7805 7806 if (old_dev && new_dev != old_dev) 7807 netdev_adjacent_dev_disable(dev, old_dev); 7808 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7809 extack); 7810 if (err) { 7811 if (old_dev && new_dev != old_dev) 7812 netdev_adjacent_dev_enable(dev, old_dev); 7813 return err; 7814 } 7815 7816 return 0; 7817 } 7818 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7819 7820 void netdev_adjacent_change_commit(struct net_device *old_dev, 7821 struct net_device *new_dev, 7822 struct net_device *dev) 7823 { 7824 struct netdev_nested_priv priv = { 7825 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7826 .data = NULL, 7827 }; 7828 7829 if (!new_dev || !old_dev) 7830 return; 7831 7832 if (new_dev == old_dev) 7833 return; 7834 7835 netdev_adjacent_dev_enable(dev, old_dev); 7836 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7837 } 7838 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7839 7840 void netdev_adjacent_change_abort(struct net_device *old_dev, 7841 struct net_device *new_dev, 7842 struct net_device *dev) 7843 { 7844 struct netdev_nested_priv priv = { 7845 .flags = 0, 7846 .data = NULL, 7847 }; 7848 7849 if (!new_dev) 7850 return; 7851 7852 if (old_dev && new_dev != old_dev) 7853 netdev_adjacent_dev_enable(dev, old_dev); 7854 7855 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7856 } 7857 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7858 7859 /** 7860 * netdev_bonding_info_change - Dispatch event about slave change 7861 * @dev: device 7862 * @bonding_info: info to dispatch 7863 * 7864 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7865 * The caller must hold the RTNL lock. 7866 */ 7867 void netdev_bonding_info_change(struct net_device *dev, 7868 struct netdev_bonding_info *bonding_info) 7869 { 7870 struct netdev_notifier_bonding_info info = { 7871 .info.dev = dev, 7872 }; 7873 7874 memcpy(&info.bonding_info, bonding_info, 7875 sizeof(struct netdev_bonding_info)); 7876 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7877 &info.info); 7878 } 7879 EXPORT_SYMBOL(netdev_bonding_info_change); 7880 7881 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7882 struct netlink_ext_ack *extack) 7883 { 7884 struct netdev_notifier_offload_xstats_info info = { 7885 .info.dev = dev, 7886 .info.extack = extack, 7887 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7888 }; 7889 int err; 7890 int rc; 7891 7892 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7893 GFP_KERNEL); 7894 if (!dev->offload_xstats_l3) 7895 return -ENOMEM; 7896 7897 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7898 NETDEV_OFFLOAD_XSTATS_DISABLE, 7899 &info.info); 7900 err = notifier_to_errno(rc); 7901 if (err) 7902 goto free_stats; 7903 7904 return 0; 7905 7906 free_stats: 7907 kfree(dev->offload_xstats_l3); 7908 dev->offload_xstats_l3 = NULL; 7909 return err; 7910 } 7911 7912 int netdev_offload_xstats_enable(struct net_device *dev, 7913 enum netdev_offload_xstats_type type, 7914 struct netlink_ext_ack *extack) 7915 { 7916 ASSERT_RTNL(); 7917 7918 if (netdev_offload_xstats_enabled(dev, type)) 7919 return -EALREADY; 7920 7921 switch (type) { 7922 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7923 return netdev_offload_xstats_enable_l3(dev, extack); 7924 } 7925 7926 WARN_ON(1); 7927 return -EINVAL; 7928 } 7929 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7930 7931 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7932 { 7933 struct netdev_notifier_offload_xstats_info info = { 7934 .info.dev = dev, 7935 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7936 }; 7937 7938 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7939 &info.info); 7940 kfree(dev->offload_xstats_l3); 7941 dev->offload_xstats_l3 = NULL; 7942 } 7943 7944 int netdev_offload_xstats_disable(struct net_device *dev, 7945 enum netdev_offload_xstats_type type) 7946 { 7947 ASSERT_RTNL(); 7948 7949 if (!netdev_offload_xstats_enabled(dev, type)) 7950 return -EALREADY; 7951 7952 switch (type) { 7953 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7954 netdev_offload_xstats_disable_l3(dev); 7955 return 0; 7956 } 7957 7958 WARN_ON(1); 7959 return -EINVAL; 7960 } 7961 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7962 7963 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7964 { 7965 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7966 } 7967 7968 static struct rtnl_hw_stats64 * 7969 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7970 enum netdev_offload_xstats_type type) 7971 { 7972 switch (type) { 7973 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7974 return dev->offload_xstats_l3; 7975 } 7976 7977 WARN_ON(1); 7978 return NULL; 7979 } 7980 7981 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7982 enum netdev_offload_xstats_type type) 7983 { 7984 ASSERT_RTNL(); 7985 7986 return netdev_offload_xstats_get_ptr(dev, type); 7987 } 7988 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7989 7990 struct netdev_notifier_offload_xstats_ru { 7991 bool used; 7992 }; 7993 7994 struct netdev_notifier_offload_xstats_rd { 7995 struct rtnl_hw_stats64 stats; 7996 bool used; 7997 }; 7998 7999 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8000 const struct rtnl_hw_stats64 *src) 8001 { 8002 dest->rx_packets += src->rx_packets; 8003 dest->tx_packets += src->tx_packets; 8004 dest->rx_bytes += src->rx_bytes; 8005 dest->tx_bytes += src->tx_bytes; 8006 dest->rx_errors += src->rx_errors; 8007 dest->tx_errors += src->tx_errors; 8008 dest->rx_dropped += src->rx_dropped; 8009 dest->tx_dropped += src->tx_dropped; 8010 dest->multicast += src->multicast; 8011 } 8012 8013 static int netdev_offload_xstats_get_used(struct net_device *dev, 8014 enum netdev_offload_xstats_type type, 8015 bool *p_used, 8016 struct netlink_ext_ack *extack) 8017 { 8018 struct netdev_notifier_offload_xstats_ru report_used = {}; 8019 struct netdev_notifier_offload_xstats_info info = { 8020 .info.dev = dev, 8021 .info.extack = extack, 8022 .type = type, 8023 .report_used = &report_used, 8024 }; 8025 int rc; 8026 8027 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8028 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8029 &info.info); 8030 *p_used = report_used.used; 8031 return notifier_to_errno(rc); 8032 } 8033 8034 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8035 enum netdev_offload_xstats_type type, 8036 struct rtnl_hw_stats64 *p_stats, 8037 bool *p_used, 8038 struct netlink_ext_ack *extack) 8039 { 8040 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8041 struct netdev_notifier_offload_xstats_info info = { 8042 .info.dev = dev, 8043 .info.extack = extack, 8044 .type = type, 8045 .report_delta = &report_delta, 8046 }; 8047 struct rtnl_hw_stats64 *stats; 8048 int rc; 8049 8050 stats = netdev_offload_xstats_get_ptr(dev, type); 8051 if (WARN_ON(!stats)) 8052 return -EINVAL; 8053 8054 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8055 &info.info); 8056 8057 /* Cache whatever we got, even if there was an error, otherwise the 8058 * successful stats retrievals would get lost. 8059 */ 8060 netdev_hw_stats64_add(stats, &report_delta.stats); 8061 8062 if (p_stats) 8063 *p_stats = *stats; 8064 *p_used = report_delta.used; 8065 8066 return notifier_to_errno(rc); 8067 } 8068 8069 int netdev_offload_xstats_get(struct net_device *dev, 8070 enum netdev_offload_xstats_type type, 8071 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8072 struct netlink_ext_ack *extack) 8073 { 8074 ASSERT_RTNL(); 8075 8076 if (p_stats) 8077 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8078 p_used, extack); 8079 else 8080 return netdev_offload_xstats_get_used(dev, type, p_used, 8081 extack); 8082 } 8083 EXPORT_SYMBOL(netdev_offload_xstats_get); 8084 8085 void 8086 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8087 const struct rtnl_hw_stats64 *stats) 8088 { 8089 report_delta->used = true; 8090 netdev_hw_stats64_add(&report_delta->stats, stats); 8091 } 8092 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8093 8094 void 8095 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8096 { 8097 report_used->used = true; 8098 } 8099 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8100 8101 void netdev_offload_xstats_push_delta(struct net_device *dev, 8102 enum netdev_offload_xstats_type type, 8103 const struct rtnl_hw_stats64 *p_stats) 8104 { 8105 struct rtnl_hw_stats64 *stats; 8106 8107 ASSERT_RTNL(); 8108 8109 stats = netdev_offload_xstats_get_ptr(dev, type); 8110 if (WARN_ON(!stats)) 8111 return; 8112 8113 netdev_hw_stats64_add(stats, p_stats); 8114 } 8115 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8116 8117 /** 8118 * netdev_get_xmit_slave - Get the xmit slave of master device 8119 * @dev: device 8120 * @skb: The packet 8121 * @all_slaves: assume all the slaves are active 8122 * 8123 * The reference counters are not incremented so the caller must be 8124 * careful with locks. The caller must hold RCU lock. 8125 * %NULL is returned if no slave is found. 8126 */ 8127 8128 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8129 struct sk_buff *skb, 8130 bool all_slaves) 8131 { 8132 const struct net_device_ops *ops = dev->netdev_ops; 8133 8134 if (!ops->ndo_get_xmit_slave) 8135 return NULL; 8136 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8137 } 8138 EXPORT_SYMBOL(netdev_get_xmit_slave); 8139 8140 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8141 struct sock *sk) 8142 { 8143 const struct net_device_ops *ops = dev->netdev_ops; 8144 8145 if (!ops->ndo_sk_get_lower_dev) 8146 return NULL; 8147 return ops->ndo_sk_get_lower_dev(dev, sk); 8148 } 8149 8150 /** 8151 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8152 * @dev: device 8153 * @sk: the socket 8154 * 8155 * %NULL is returned if no lower device is found. 8156 */ 8157 8158 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8159 struct sock *sk) 8160 { 8161 struct net_device *lower; 8162 8163 lower = netdev_sk_get_lower_dev(dev, sk); 8164 while (lower) { 8165 dev = lower; 8166 lower = netdev_sk_get_lower_dev(dev, sk); 8167 } 8168 8169 return dev; 8170 } 8171 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8172 8173 static void netdev_adjacent_add_links(struct net_device *dev) 8174 { 8175 struct netdev_adjacent *iter; 8176 8177 struct net *net = dev_net(dev); 8178 8179 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8180 if (!net_eq(net, dev_net(iter->dev))) 8181 continue; 8182 netdev_adjacent_sysfs_add(iter->dev, dev, 8183 &iter->dev->adj_list.lower); 8184 netdev_adjacent_sysfs_add(dev, iter->dev, 8185 &dev->adj_list.upper); 8186 } 8187 8188 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8189 if (!net_eq(net, dev_net(iter->dev))) 8190 continue; 8191 netdev_adjacent_sysfs_add(iter->dev, dev, 8192 &iter->dev->adj_list.upper); 8193 netdev_adjacent_sysfs_add(dev, iter->dev, 8194 &dev->adj_list.lower); 8195 } 8196 } 8197 8198 static void netdev_adjacent_del_links(struct net_device *dev) 8199 { 8200 struct netdev_adjacent *iter; 8201 8202 struct net *net = dev_net(dev); 8203 8204 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8205 if (!net_eq(net, dev_net(iter->dev))) 8206 continue; 8207 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8208 &iter->dev->adj_list.lower); 8209 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8210 &dev->adj_list.upper); 8211 } 8212 8213 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8214 if (!net_eq(net, dev_net(iter->dev))) 8215 continue; 8216 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8217 &iter->dev->adj_list.upper); 8218 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8219 &dev->adj_list.lower); 8220 } 8221 } 8222 8223 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8224 { 8225 struct netdev_adjacent *iter; 8226 8227 struct net *net = dev_net(dev); 8228 8229 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8230 if (!net_eq(net, dev_net(iter->dev))) 8231 continue; 8232 netdev_adjacent_sysfs_del(iter->dev, oldname, 8233 &iter->dev->adj_list.lower); 8234 netdev_adjacent_sysfs_add(iter->dev, dev, 8235 &iter->dev->adj_list.lower); 8236 } 8237 8238 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8239 if (!net_eq(net, dev_net(iter->dev))) 8240 continue; 8241 netdev_adjacent_sysfs_del(iter->dev, oldname, 8242 &iter->dev->adj_list.upper); 8243 netdev_adjacent_sysfs_add(iter->dev, dev, 8244 &iter->dev->adj_list.upper); 8245 } 8246 } 8247 8248 void *netdev_lower_dev_get_private(struct net_device *dev, 8249 struct net_device *lower_dev) 8250 { 8251 struct netdev_adjacent *lower; 8252 8253 if (!lower_dev) 8254 return NULL; 8255 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8256 if (!lower) 8257 return NULL; 8258 8259 return lower->private; 8260 } 8261 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8262 8263 8264 /** 8265 * netdev_lower_state_changed - Dispatch event about lower device state change 8266 * @lower_dev: device 8267 * @lower_state_info: state to dispatch 8268 * 8269 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8270 * The caller must hold the RTNL lock. 8271 */ 8272 void netdev_lower_state_changed(struct net_device *lower_dev, 8273 void *lower_state_info) 8274 { 8275 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8276 .info.dev = lower_dev, 8277 }; 8278 8279 ASSERT_RTNL(); 8280 changelowerstate_info.lower_state_info = lower_state_info; 8281 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8282 &changelowerstate_info.info); 8283 } 8284 EXPORT_SYMBOL(netdev_lower_state_changed); 8285 8286 static void dev_change_rx_flags(struct net_device *dev, int flags) 8287 { 8288 const struct net_device_ops *ops = dev->netdev_ops; 8289 8290 if (ops->ndo_change_rx_flags) 8291 ops->ndo_change_rx_flags(dev, flags); 8292 } 8293 8294 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8295 { 8296 unsigned int old_flags = dev->flags; 8297 kuid_t uid; 8298 kgid_t gid; 8299 8300 ASSERT_RTNL(); 8301 8302 dev->flags |= IFF_PROMISC; 8303 dev->promiscuity += inc; 8304 if (dev->promiscuity == 0) { 8305 /* 8306 * Avoid overflow. 8307 * If inc causes overflow, untouch promisc and return error. 8308 */ 8309 if (inc < 0) 8310 dev->flags &= ~IFF_PROMISC; 8311 else { 8312 dev->promiscuity -= inc; 8313 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8314 return -EOVERFLOW; 8315 } 8316 } 8317 if (dev->flags != old_flags) { 8318 netdev_info(dev, "%s promiscuous mode\n", 8319 dev->flags & IFF_PROMISC ? "entered" : "left"); 8320 if (audit_enabled) { 8321 current_uid_gid(&uid, &gid); 8322 audit_log(audit_context(), GFP_ATOMIC, 8323 AUDIT_ANOM_PROMISCUOUS, 8324 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8325 dev->name, (dev->flags & IFF_PROMISC), 8326 (old_flags & IFF_PROMISC), 8327 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8328 from_kuid(&init_user_ns, uid), 8329 from_kgid(&init_user_ns, gid), 8330 audit_get_sessionid(current)); 8331 } 8332 8333 dev_change_rx_flags(dev, IFF_PROMISC); 8334 } 8335 if (notify) 8336 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8337 return 0; 8338 } 8339 8340 /** 8341 * dev_set_promiscuity - update promiscuity count on a device 8342 * @dev: device 8343 * @inc: modifier 8344 * 8345 * Add or remove promiscuity from a device. While the count in the device 8346 * remains above zero the interface remains promiscuous. Once it hits zero 8347 * the device reverts back to normal filtering operation. A negative inc 8348 * value is used to drop promiscuity on the device. 8349 * Return 0 if successful or a negative errno code on error. 8350 */ 8351 int dev_set_promiscuity(struct net_device *dev, int inc) 8352 { 8353 unsigned int old_flags = dev->flags; 8354 int err; 8355 8356 err = __dev_set_promiscuity(dev, inc, true); 8357 if (err < 0) 8358 return err; 8359 if (dev->flags != old_flags) 8360 dev_set_rx_mode(dev); 8361 return err; 8362 } 8363 EXPORT_SYMBOL(dev_set_promiscuity); 8364 8365 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8366 { 8367 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8368 8369 ASSERT_RTNL(); 8370 8371 dev->flags |= IFF_ALLMULTI; 8372 dev->allmulti += inc; 8373 if (dev->allmulti == 0) { 8374 /* 8375 * Avoid overflow. 8376 * If inc causes overflow, untouch allmulti and return error. 8377 */ 8378 if (inc < 0) 8379 dev->flags &= ~IFF_ALLMULTI; 8380 else { 8381 dev->allmulti -= inc; 8382 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8383 return -EOVERFLOW; 8384 } 8385 } 8386 if (dev->flags ^ old_flags) { 8387 netdev_info(dev, "%s allmulticast mode\n", 8388 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8389 dev_change_rx_flags(dev, IFF_ALLMULTI); 8390 dev_set_rx_mode(dev); 8391 if (notify) 8392 __dev_notify_flags(dev, old_flags, 8393 dev->gflags ^ old_gflags, 0, NULL); 8394 } 8395 return 0; 8396 } 8397 8398 /** 8399 * dev_set_allmulti - update allmulti count on a device 8400 * @dev: device 8401 * @inc: modifier 8402 * 8403 * Add or remove reception of all multicast frames to a device. While the 8404 * count in the device remains above zero the interface remains listening 8405 * to all interfaces. Once it hits zero the device reverts back to normal 8406 * filtering operation. A negative @inc value is used to drop the counter 8407 * when releasing a resource needing all multicasts. 8408 * Return 0 if successful or a negative errno code on error. 8409 */ 8410 8411 int dev_set_allmulti(struct net_device *dev, int inc) 8412 { 8413 return __dev_set_allmulti(dev, inc, true); 8414 } 8415 EXPORT_SYMBOL(dev_set_allmulti); 8416 8417 /* 8418 * Upload unicast and multicast address lists to device and 8419 * configure RX filtering. When the device doesn't support unicast 8420 * filtering it is put in promiscuous mode while unicast addresses 8421 * are present. 8422 */ 8423 void __dev_set_rx_mode(struct net_device *dev) 8424 { 8425 const struct net_device_ops *ops = dev->netdev_ops; 8426 8427 /* dev_open will call this function so the list will stay sane. */ 8428 if (!(dev->flags&IFF_UP)) 8429 return; 8430 8431 if (!netif_device_present(dev)) 8432 return; 8433 8434 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8435 /* Unicast addresses changes may only happen under the rtnl, 8436 * therefore calling __dev_set_promiscuity here is safe. 8437 */ 8438 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8439 __dev_set_promiscuity(dev, 1, false); 8440 dev->uc_promisc = true; 8441 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8442 __dev_set_promiscuity(dev, -1, false); 8443 dev->uc_promisc = false; 8444 } 8445 } 8446 8447 if (ops->ndo_set_rx_mode) 8448 ops->ndo_set_rx_mode(dev); 8449 } 8450 8451 void dev_set_rx_mode(struct net_device *dev) 8452 { 8453 netif_addr_lock_bh(dev); 8454 __dev_set_rx_mode(dev); 8455 netif_addr_unlock_bh(dev); 8456 } 8457 8458 /** 8459 * dev_get_flags - get flags reported to userspace 8460 * @dev: device 8461 * 8462 * Get the combination of flag bits exported through APIs to userspace. 8463 */ 8464 unsigned int dev_get_flags(const struct net_device *dev) 8465 { 8466 unsigned int flags; 8467 8468 flags = (dev->flags & ~(IFF_PROMISC | 8469 IFF_ALLMULTI | 8470 IFF_RUNNING | 8471 IFF_LOWER_UP | 8472 IFF_DORMANT)) | 8473 (dev->gflags & (IFF_PROMISC | 8474 IFF_ALLMULTI)); 8475 8476 if (netif_running(dev)) { 8477 if (netif_oper_up(dev)) 8478 flags |= IFF_RUNNING; 8479 if (netif_carrier_ok(dev)) 8480 flags |= IFF_LOWER_UP; 8481 if (netif_dormant(dev)) 8482 flags |= IFF_DORMANT; 8483 } 8484 8485 return flags; 8486 } 8487 EXPORT_SYMBOL(dev_get_flags); 8488 8489 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8490 struct netlink_ext_ack *extack) 8491 { 8492 unsigned int old_flags = dev->flags; 8493 int ret; 8494 8495 ASSERT_RTNL(); 8496 8497 /* 8498 * Set the flags on our device. 8499 */ 8500 8501 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8502 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8503 IFF_AUTOMEDIA)) | 8504 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8505 IFF_ALLMULTI)); 8506 8507 /* 8508 * Load in the correct multicast list now the flags have changed. 8509 */ 8510 8511 if ((old_flags ^ flags) & IFF_MULTICAST) 8512 dev_change_rx_flags(dev, IFF_MULTICAST); 8513 8514 dev_set_rx_mode(dev); 8515 8516 /* 8517 * Have we downed the interface. We handle IFF_UP ourselves 8518 * according to user attempts to set it, rather than blindly 8519 * setting it. 8520 */ 8521 8522 ret = 0; 8523 if ((old_flags ^ flags) & IFF_UP) { 8524 if (old_flags & IFF_UP) 8525 __dev_close(dev); 8526 else 8527 ret = __dev_open(dev, extack); 8528 } 8529 8530 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8531 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8532 unsigned int old_flags = dev->flags; 8533 8534 dev->gflags ^= IFF_PROMISC; 8535 8536 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8537 if (dev->flags != old_flags) 8538 dev_set_rx_mode(dev); 8539 } 8540 8541 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8542 * is important. Some (broken) drivers set IFF_PROMISC, when 8543 * IFF_ALLMULTI is requested not asking us and not reporting. 8544 */ 8545 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8546 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8547 8548 dev->gflags ^= IFF_ALLMULTI; 8549 __dev_set_allmulti(dev, inc, false); 8550 } 8551 8552 return ret; 8553 } 8554 8555 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8556 unsigned int gchanges, u32 portid, 8557 const struct nlmsghdr *nlh) 8558 { 8559 unsigned int changes = dev->flags ^ old_flags; 8560 8561 if (gchanges) 8562 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8563 8564 if (changes & IFF_UP) { 8565 if (dev->flags & IFF_UP) 8566 call_netdevice_notifiers(NETDEV_UP, dev); 8567 else 8568 call_netdevice_notifiers(NETDEV_DOWN, dev); 8569 } 8570 8571 if (dev->flags & IFF_UP && 8572 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8573 struct netdev_notifier_change_info change_info = { 8574 .info = { 8575 .dev = dev, 8576 }, 8577 .flags_changed = changes, 8578 }; 8579 8580 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8581 } 8582 } 8583 8584 /** 8585 * dev_change_flags - change device settings 8586 * @dev: device 8587 * @flags: device state flags 8588 * @extack: netlink extended ack 8589 * 8590 * Change settings on device based state flags. The flags are 8591 * in the userspace exported format. 8592 */ 8593 int dev_change_flags(struct net_device *dev, unsigned int flags, 8594 struct netlink_ext_ack *extack) 8595 { 8596 int ret; 8597 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8598 8599 ret = __dev_change_flags(dev, flags, extack); 8600 if (ret < 0) 8601 return ret; 8602 8603 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8604 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8605 return ret; 8606 } 8607 EXPORT_SYMBOL(dev_change_flags); 8608 8609 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8610 { 8611 const struct net_device_ops *ops = dev->netdev_ops; 8612 8613 if (ops->ndo_change_mtu) 8614 return ops->ndo_change_mtu(dev, new_mtu); 8615 8616 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8617 WRITE_ONCE(dev->mtu, new_mtu); 8618 return 0; 8619 } 8620 EXPORT_SYMBOL(__dev_set_mtu); 8621 8622 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8623 struct netlink_ext_ack *extack) 8624 { 8625 /* MTU must be positive, and in range */ 8626 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8627 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8628 return -EINVAL; 8629 } 8630 8631 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8632 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8633 return -EINVAL; 8634 } 8635 return 0; 8636 } 8637 8638 /** 8639 * dev_set_mtu_ext - Change maximum transfer unit 8640 * @dev: device 8641 * @new_mtu: new transfer unit 8642 * @extack: netlink extended ack 8643 * 8644 * Change the maximum transfer size of the network device. 8645 */ 8646 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8647 struct netlink_ext_ack *extack) 8648 { 8649 int err, orig_mtu; 8650 8651 if (new_mtu == dev->mtu) 8652 return 0; 8653 8654 err = dev_validate_mtu(dev, new_mtu, extack); 8655 if (err) 8656 return err; 8657 8658 if (!netif_device_present(dev)) 8659 return -ENODEV; 8660 8661 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8662 err = notifier_to_errno(err); 8663 if (err) 8664 return err; 8665 8666 orig_mtu = dev->mtu; 8667 err = __dev_set_mtu(dev, new_mtu); 8668 8669 if (!err) { 8670 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8671 orig_mtu); 8672 err = notifier_to_errno(err); 8673 if (err) { 8674 /* setting mtu back and notifying everyone again, 8675 * so that they have a chance to revert changes. 8676 */ 8677 __dev_set_mtu(dev, orig_mtu); 8678 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8679 new_mtu); 8680 } 8681 } 8682 return err; 8683 } 8684 8685 int dev_set_mtu(struct net_device *dev, int new_mtu) 8686 { 8687 struct netlink_ext_ack extack; 8688 int err; 8689 8690 memset(&extack, 0, sizeof(extack)); 8691 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8692 if (err && extack._msg) 8693 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8694 return err; 8695 } 8696 EXPORT_SYMBOL(dev_set_mtu); 8697 8698 /** 8699 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8700 * @dev: device 8701 * @new_len: new tx queue length 8702 */ 8703 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8704 { 8705 unsigned int orig_len = dev->tx_queue_len; 8706 int res; 8707 8708 if (new_len != (unsigned int)new_len) 8709 return -ERANGE; 8710 8711 if (new_len != orig_len) { 8712 dev->tx_queue_len = new_len; 8713 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8714 res = notifier_to_errno(res); 8715 if (res) 8716 goto err_rollback; 8717 res = dev_qdisc_change_tx_queue_len(dev); 8718 if (res) 8719 goto err_rollback; 8720 } 8721 8722 return 0; 8723 8724 err_rollback: 8725 netdev_err(dev, "refused to change device tx_queue_len\n"); 8726 dev->tx_queue_len = orig_len; 8727 return res; 8728 } 8729 8730 /** 8731 * dev_set_group - Change group this device belongs to 8732 * @dev: device 8733 * @new_group: group this device should belong to 8734 */ 8735 void dev_set_group(struct net_device *dev, int new_group) 8736 { 8737 dev->group = new_group; 8738 } 8739 8740 /** 8741 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8742 * @dev: device 8743 * @addr: new address 8744 * @extack: netlink extended ack 8745 */ 8746 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8747 struct netlink_ext_ack *extack) 8748 { 8749 struct netdev_notifier_pre_changeaddr_info info = { 8750 .info.dev = dev, 8751 .info.extack = extack, 8752 .dev_addr = addr, 8753 }; 8754 int rc; 8755 8756 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8757 return notifier_to_errno(rc); 8758 } 8759 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8760 8761 /** 8762 * dev_set_mac_address - Change Media Access Control Address 8763 * @dev: device 8764 * @sa: new address 8765 * @extack: netlink extended ack 8766 * 8767 * Change the hardware (MAC) address of the device 8768 */ 8769 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8770 struct netlink_ext_ack *extack) 8771 { 8772 const struct net_device_ops *ops = dev->netdev_ops; 8773 int err; 8774 8775 if (!ops->ndo_set_mac_address) 8776 return -EOPNOTSUPP; 8777 if (sa->sa_family != dev->type) 8778 return -EINVAL; 8779 if (!netif_device_present(dev)) 8780 return -ENODEV; 8781 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8782 if (err) 8783 return err; 8784 err = ops->ndo_set_mac_address(dev, sa); 8785 if (err) 8786 return err; 8787 dev->addr_assign_type = NET_ADDR_SET; 8788 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8789 add_device_randomness(dev->dev_addr, dev->addr_len); 8790 return 0; 8791 } 8792 EXPORT_SYMBOL(dev_set_mac_address); 8793 8794 static DECLARE_RWSEM(dev_addr_sem); 8795 8796 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8797 struct netlink_ext_ack *extack) 8798 { 8799 int ret; 8800 8801 down_write(&dev_addr_sem); 8802 ret = dev_set_mac_address(dev, sa, extack); 8803 up_write(&dev_addr_sem); 8804 return ret; 8805 } 8806 EXPORT_SYMBOL(dev_set_mac_address_user); 8807 8808 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8809 { 8810 size_t size = sizeof(sa->sa_data_min); 8811 struct net_device *dev; 8812 int ret = 0; 8813 8814 down_read(&dev_addr_sem); 8815 rcu_read_lock(); 8816 8817 dev = dev_get_by_name_rcu(net, dev_name); 8818 if (!dev) { 8819 ret = -ENODEV; 8820 goto unlock; 8821 } 8822 if (!dev->addr_len) 8823 memset(sa->sa_data, 0, size); 8824 else 8825 memcpy(sa->sa_data, dev->dev_addr, 8826 min_t(size_t, size, dev->addr_len)); 8827 sa->sa_family = dev->type; 8828 8829 unlock: 8830 rcu_read_unlock(); 8831 up_read(&dev_addr_sem); 8832 return ret; 8833 } 8834 EXPORT_SYMBOL(dev_get_mac_address); 8835 8836 /** 8837 * dev_change_carrier - Change device carrier 8838 * @dev: device 8839 * @new_carrier: new value 8840 * 8841 * Change device carrier 8842 */ 8843 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8844 { 8845 const struct net_device_ops *ops = dev->netdev_ops; 8846 8847 if (!ops->ndo_change_carrier) 8848 return -EOPNOTSUPP; 8849 if (!netif_device_present(dev)) 8850 return -ENODEV; 8851 return ops->ndo_change_carrier(dev, new_carrier); 8852 } 8853 8854 /** 8855 * dev_get_phys_port_id - Get device physical port ID 8856 * @dev: device 8857 * @ppid: port ID 8858 * 8859 * Get device physical port ID 8860 */ 8861 int dev_get_phys_port_id(struct net_device *dev, 8862 struct netdev_phys_item_id *ppid) 8863 { 8864 const struct net_device_ops *ops = dev->netdev_ops; 8865 8866 if (!ops->ndo_get_phys_port_id) 8867 return -EOPNOTSUPP; 8868 return ops->ndo_get_phys_port_id(dev, ppid); 8869 } 8870 8871 /** 8872 * dev_get_phys_port_name - Get device physical port name 8873 * @dev: device 8874 * @name: port name 8875 * @len: limit of bytes to copy to name 8876 * 8877 * Get device physical port name 8878 */ 8879 int dev_get_phys_port_name(struct net_device *dev, 8880 char *name, size_t len) 8881 { 8882 const struct net_device_ops *ops = dev->netdev_ops; 8883 int err; 8884 8885 if (ops->ndo_get_phys_port_name) { 8886 err = ops->ndo_get_phys_port_name(dev, name, len); 8887 if (err != -EOPNOTSUPP) 8888 return err; 8889 } 8890 return devlink_compat_phys_port_name_get(dev, name, len); 8891 } 8892 8893 /** 8894 * dev_get_port_parent_id - Get the device's port parent identifier 8895 * @dev: network device 8896 * @ppid: pointer to a storage for the port's parent identifier 8897 * @recurse: allow/disallow recursion to lower devices 8898 * 8899 * Get the devices's port parent identifier 8900 */ 8901 int dev_get_port_parent_id(struct net_device *dev, 8902 struct netdev_phys_item_id *ppid, 8903 bool recurse) 8904 { 8905 const struct net_device_ops *ops = dev->netdev_ops; 8906 struct netdev_phys_item_id first = { }; 8907 struct net_device *lower_dev; 8908 struct list_head *iter; 8909 int err; 8910 8911 if (ops->ndo_get_port_parent_id) { 8912 err = ops->ndo_get_port_parent_id(dev, ppid); 8913 if (err != -EOPNOTSUPP) 8914 return err; 8915 } 8916 8917 err = devlink_compat_switch_id_get(dev, ppid); 8918 if (!recurse || err != -EOPNOTSUPP) 8919 return err; 8920 8921 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8922 err = dev_get_port_parent_id(lower_dev, ppid, true); 8923 if (err) 8924 break; 8925 if (!first.id_len) 8926 first = *ppid; 8927 else if (memcmp(&first, ppid, sizeof(*ppid))) 8928 return -EOPNOTSUPP; 8929 } 8930 8931 return err; 8932 } 8933 EXPORT_SYMBOL(dev_get_port_parent_id); 8934 8935 /** 8936 * netdev_port_same_parent_id - Indicate if two network devices have 8937 * the same port parent identifier 8938 * @a: first network device 8939 * @b: second network device 8940 */ 8941 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8942 { 8943 struct netdev_phys_item_id a_id = { }; 8944 struct netdev_phys_item_id b_id = { }; 8945 8946 if (dev_get_port_parent_id(a, &a_id, true) || 8947 dev_get_port_parent_id(b, &b_id, true)) 8948 return false; 8949 8950 return netdev_phys_item_id_same(&a_id, &b_id); 8951 } 8952 EXPORT_SYMBOL(netdev_port_same_parent_id); 8953 8954 /** 8955 * dev_change_proto_down - set carrier according to proto_down. 8956 * 8957 * @dev: device 8958 * @proto_down: new value 8959 */ 8960 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8961 { 8962 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8963 return -EOPNOTSUPP; 8964 if (!netif_device_present(dev)) 8965 return -ENODEV; 8966 if (proto_down) 8967 netif_carrier_off(dev); 8968 else 8969 netif_carrier_on(dev); 8970 dev->proto_down = proto_down; 8971 return 0; 8972 } 8973 8974 /** 8975 * dev_change_proto_down_reason - proto down reason 8976 * 8977 * @dev: device 8978 * @mask: proto down mask 8979 * @value: proto down value 8980 */ 8981 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8982 u32 value) 8983 { 8984 int b; 8985 8986 if (!mask) { 8987 dev->proto_down_reason = value; 8988 } else { 8989 for_each_set_bit(b, &mask, 32) { 8990 if (value & (1 << b)) 8991 dev->proto_down_reason |= BIT(b); 8992 else 8993 dev->proto_down_reason &= ~BIT(b); 8994 } 8995 } 8996 } 8997 8998 struct bpf_xdp_link { 8999 struct bpf_link link; 9000 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9001 int flags; 9002 }; 9003 9004 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9005 { 9006 if (flags & XDP_FLAGS_HW_MODE) 9007 return XDP_MODE_HW; 9008 if (flags & XDP_FLAGS_DRV_MODE) 9009 return XDP_MODE_DRV; 9010 if (flags & XDP_FLAGS_SKB_MODE) 9011 return XDP_MODE_SKB; 9012 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9013 } 9014 9015 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9016 { 9017 switch (mode) { 9018 case XDP_MODE_SKB: 9019 return generic_xdp_install; 9020 case XDP_MODE_DRV: 9021 case XDP_MODE_HW: 9022 return dev->netdev_ops->ndo_bpf; 9023 default: 9024 return NULL; 9025 } 9026 } 9027 9028 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9029 enum bpf_xdp_mode mode) 9030 { 9031 return dev->xdp_state[mode].link; 9032 } 9033 9034 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9035 enum bpf_xdp_mode mode) 9036 { 9037 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9038 9039 if (link) 9040 return link->link.prog; 9041 return dev->xdp_state[mode].prog; 9042 } 9043 9044 u8 dev_xdp_prog_count(struct net_device *dev) 9045 { 9046 u8 count = 0; 9047 int i; 9048 9049 for (i = 0; i < __MAX_XDP_MODE; i++) 9050 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9051 count++; 9052 return count; 9053 } 9054 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9055 9056 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9057 { 9058 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9059 9060 return prog ? prog->aux->id : 0; 9061 } 9062 9063 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9064 struct bpf_xdp_link *link) 9065 { 9066 dev->xdp_state[mode].link = link; 9067 dev->xdp_state[mode].prog = NULL; 9068 } 9069 9070 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9071 struct bpf_prog *prog) 9072 { 9073 dev->xdp_state[mode].link = NULL; 9074 dev->xdp_state[mode].prog = prog; 9075 } 9076 9077 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9078 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9079 u32 flags, struct bpf_prog *prog) 9080 { 9081 struct netdev_bpf xdp; 9082 int err; 9083 9084 memset(&xdp, 0, sizeof(xdp)); 9085 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9086 xdp.extack = extack; 9087 xdp.flags = flags; 9088 xdp.prog = prog; 9089 9090 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9091 * "moved" into driver), so they don't increment it on their own, but 9092 * they do decrement refcnt when program is detached or replaced. 9093 * Given net_device also owns link/prog, we need to bump refcnt here 9094 * to prevent drivers from underflowing it. 9095 */ 9096 if (prog) 9097 bpf_prog_inc(prog); 9098 err = bpf_op(dev, &xdp); 9099 if (err) { 9100 if (prog) 9101 bpf_prog_put(prog); 9102 return err; 9103 } 9104 9105 if (mode != XDP_MODE_HW) 9106 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9107 9108 return 0; 9109 } 9110 9111 static void dev_xdp_uninstall(struct net_device *dev) 9112 { 9113 struct bpf_xdp_link *link; 9114 struct bpf_prog *prog; 9115 enum bpf_xdp_mode mode; 9116 bpf_op_t bpf_op; 9117 9118 ASSERT_RTNL(); 9119 9120 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9121 prog = dev_xdp_prog(dev, mode); 9122 if (!prog) 9123 continue; 9124 9125 bpf_op = dev_xdp_bpf_op(dev, mode); 9126 if (!bpf_op) 9127 continue; 9128 9129 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9130 9131 /* auto-detach link from net device */ 9132 link = dev_xdp_link(dev, mode); 9133 if (link) 9134 link->dev = NULL; 9135 else 9136 bpf_prog_put(prog); 9137 9138 dev_xdp_set_link(dev, mode, NULL); 9139 } 9140 } 9141 9142 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9143 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9144 struct bpf_prog *old_prog, u32 flags) 9145 { 9146 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9147 struct bpf_prog *cur_prog; 9148 struct net_device *upper; 9149 struct list_head *iter; 9150 enum bpf_xdp_mode mode; 9151 bpf_op_t bpf_op; 9152 int err; 9153 9154 ASSERT_RTNL(); 9155 9156 /* either link or prog attachment, never both */ 9157 if (link && (new_prog || old_prog)) 9158 return -EINVAL; 9159 /* link supports only XDP mode flags */ 9160 if (link && (flags & ~XDP_FLAGS_MODES)) { 9161 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9162 return -EINVAL; 9163 } 9164 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9165 if (num_modes > 1) { 9166 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9167 return -EINVAL; 9168 } 9169 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9170 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9171 NL_SET_ERR_MSG(extack, 9172 "More than one program loaded, unset mode is ambiguous"); 9173 return -EINVAL; 9174 } 9175 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9176 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9177 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9178 return -EINVAL; 9179 } 9180 9181 mode = dev_xdp_mode(dev, flags); 9182 /* can't replace attached link */ 9183 if (dev_xdp_link(dev, mode)) { 9184 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9185 return -EBUSY; 9186 } 9187 9188 /* don't allow if an upper device already has a program */ 9189 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9190 if (dev_xdp_prog_count(upper) > 0) { 9191 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9192 return -EEXIST; 9193 } 9194 } 9195 9196 cur_prog = dev_xdp_prog(dev, mode); 9197 /* can't replace attached prog with link */ 9198 if (link && cur_prog) { 9199 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9200 return -EBUSY; 9201 } 9202 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9203 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9204 return -EEXIST; 9205 } 9206 9207 /* put effective new program into new_prog */ 9208 if (link) 9209 new_prog = link->link.prog; 9210 9211 if (new_prog) { 9212 bool offload = mode == XDP_MODE_HW; 9213 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9214 ? XDP_MODE_DRV : XDP_MODE_SKB; 9215 9216 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9217 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9218 return -EBUSY; 9219 } 9220 if (!offload && dev_xdp_prog(dev, other_mode)) { 9221 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9222 return -EEXIST; 9223 } 9224 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9225 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9226 return -EINVAL; 9227 } 9228 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9229 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9230 return -EINVAL; 9231 } 9232 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9233 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9234 return -EINVAL; 9235 } 9236 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9237 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9238 return -EINVAL; 9239 } 9240 } 9241 9242 /* don't call drivers if the effective program didn't change */ 9243 if (new_prog != cur_prog) { 9244 bpf_op = dev_xdp_bpf_op(dev, mode); 9245 if (!bpf_op) { 9246 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9247 return -EOPNOTSUPP; 9248 } 9249 9250 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9251 if (err) 9252 return err; 9253 } 9254 9255 if (link) 9256 dev_xdp_set_link(dev, mode, link); 9257 else 9258 dev_xdp_set_prog(dev, mode, new_prog); 9259 if (cur_prog) 9260 bpf_prog_put(cur_prog); 9261 9262 return 0; 9263 } 9264 9265 static int dev_xdp_attach_link(struct net_device *dev, 9266 struct netlink_ext_ack *extack, 9267 struct bpf_xdp_link *link) 9268 { 9269 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9270 } 9271 9272 static int dev_xdp_detach_link(struct net_device *dev, 9273 struct netlink_ext_ack *extack, 9274 struct bpf_xdp_link *link) 9275 { 9276 enum bpf_xdp_mode mode; 9277 bpf_op_t bpf_op; 9278 9279 ASSERT_RTNL(); 9280 9281 mode = dev_xdp_mode(dev, link->flags); 9282 if (dev_xdp_link(dev, mode) != link) 9283 return -EINVAL; 9284 9285 bpf_op = dev_xdp_bpf_op(dev, mode); 9286 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9287 dev_xdp_set_link(dev, mode, NULL); 9288 return 0; 9289 } 9290 9291 static void bpf_xdp_link_release(struct bpf_link *link) 9292 { 9293 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9294 9295 rtnl_lock(); 9296 9297 /* if racing with net_device's tear down, xdp_link->dev might be 9298 * already NULL, in which case link was already auto-detached 9299 */ 9300 if (xdp_link->dev) { 9301 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9302 xdp_link->dev = NULL; 9303 } 9304 9305 rtnl_unlock(); 9306 } 9307 9308 static int bpf_xdp_link_detach(struct bpf_link *link) 9309 { 9310 bpf_xdp_link_release(link); 9311 return 0; 9312 } 9313 9314 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9315 { 9316 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9317 9318 kfree(xdp_link); 9319 } 9320 9321 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9322 struct seq_file *seq) 9323 { 9324 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9325 u32 ifindex = 0; 9326 9327 rtnl_lock(); 9328 if (xdp_link->dev) 9329 ifindex = xdp_link->dev->ifindex; 9330 rtnl_unlock(); 9331 9332 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9333 } 9334 9335 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9336 struct bpf_link_info *info) 9337 { 9338 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9339 u32 ifindex = 0; 9340 9341 rtnl_lock(); 9342 if (xdp_link->dev) 9343 ifindex = xdp_link->dev->ifindex; 9344 rtnl_unlock(); 9345 9346 info->xdp.ifindex = ifindex; 9347 return 0; 9348 } 9349 9350 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9351 struct bpf_prog *old_prog) 9352 { 9353 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9354 enum bpf_xdp_mode mode; 9355 bpf_op_t bpf_op; 9356 int err = 0; 9357 9358 rtnl_lock(); 9359 9360 /* link might have been auto-released already, so fail */ 9361 if (!xdp_link->dev) { 9362 err = -ENOLINK; 9363 goto out_unlock; 9364 } 9365 9366 if (old_prog && link->prog != old_prog) { 9367 err = -EPERM; 9368 goto out_unlock; 9369 } 9370 old_prog = link->prog; 9371 if (old_prog->type != new_prog->type || 9372 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9373 err = -EINVAL; 9374 goto out_unlock; 9375 } 9376 9377 if (old_prog == new_prog) { 9378 /* no-op, don't disturb drivers */ 9379 bpf_prog_put(new_prog); 9380 goto out_unlock; 9381 } 9382 9383 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9384 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9385 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9386 xdp_link->flags, new_prog); 9387 if (err) 9388 goto out_unlock; 9389 9390 old_prog = xchg(&link->prog, new_prog); 9391 bpf_prog_put(old_prog); 9392 9393 out_unlock: 9394 rtnl_unlock(); 9395 return err; 9396 } 9397 9398 static const struct bpf_link_ops bpf_xdp_link_lops = { 9399 .release = bpf_xdp_link_release, 9400 .dealloc = bpf_xdp_link_dealloc, 9401 .detach = bpf_xdp_link_detach, 9402 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9403 .fill_link_info = bpf_xdp_link_fill_link_info, 9404 .update_prog = bpf_xdp_link_update, 9405 }; 9406 9407 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9408 { 9409 struct net *net = current->nsproxy->net_ns; 9410 struct bpf_link_primer link_primer; 9411 struct bpf_xdp_link *link; 9412 struct net_device *dev; 9413 int err, fd; 9414 9415 rtnl_lock(); 9416 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9417 if (!dev) { 9418 rtnl_unlock(); 9419 return -EINVAL; 9420 } 9421 9422 link = kzalloc(sizeof(*link), GFP_USER); 9423 if (!link) { 9424 err = -ENOMEM; 9425 goto unlock; 9426 } 9427 9428 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9429 link->dev = dev; 9430 link->flags = attr->link_create.flags; 9431 9432 err = bpf_link_prime(&link->link, &link_primer); 9433 if (err) { 9434 kfree(link); 9435 goto unlock; 9436 } 9437 9438 err = dev_xdp_attach_link(dev, NULL, link); 9439 rtnl_unlock(); 9440 9441 if (err) { 9442 link->dev = NULL; 9443 bpf_link_cleanup(&link_primer); 9444 goto out_put_dev; 9445 } 9446 9447 fd = bpf_link_settle(&link_primer); 9448 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9449 dev_put(dev); 9450 return fd; 9451 9452 unlock: 9453 rtnl_unlock(); 9454 9455 out_put_dev: 9456 dev_put(dev); 9457 return err; 9458 } 9459 9460 /** 9461 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9462 * @dev: device 9463 * @extack: netlink extended ack 9464 * @fd: new program fd or negative value to clear 9465 * @expected_fd: old program fd that userspace expects to replace or clear 9466 * @flags: xdp-related flags 9467 * 9468 * Set or clear a bpf program for a device 9469 */ 9470 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9471 int fd, int expected_fd, u32 flags) 9472 { 9473 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9474 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9475 int err; 9476 9477 ASSERT_RTNL(); 9478 9479 if (fd >= 0) { 9480 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9481 mode != XDP_MODE_SKB); 9482 if (IS_ERR(new_prog)) 9483 return PTR_ERR(new_prog); 9484 } 9485 9486 if (expected_fd >= 0) { 9487 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9488 mode != XDP_MODE_SKB); 9489 if (IS_ERR(old_prog)) { 9490 err = PTR_ERR(old_prog); 9491 old_prog = NULL; 9492 goto err_out; 9493 } 9494 } 9495 9496 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9497 9498 err_out: 9499 if (err && new_prog) 9500 bpf_prog_put(new_prog); 9501 if (old_prog) 9502 bpf_prog_put(old_prog); 9503 return err; 9504 } 9505 9506 /** 9507 * dev_new_index - allocate an ifindex 9508 * @net: the applicable net namespace 9509 * 9510 * Returns a suitable unique value for a new device interface 9511 * number. The caller must hold the rtnl semaphore or the 9512 * dev_base_lock to be sure it remains unique. 9513 */ 9514 static int dev_new_index(struct net *net) 9515 { 9516 int ifindex = net->ifindex; 9517 9518 for (;;) { 9519 if (++ifindex <= 0) 9520 ifindex = 1; 9521 if (!__dev_get_by_index(net, ifindex)) 9522 return net->ifindex = ifindex; 9523 } 9524 } 9525 9526 /* Delayed registration/unregisteration */ 9527 LIST_HEAD(net_todo_list); 9528 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9529 9530 static void net_set_todo(struct net_device *dev) 9531 { 9532 list_add_tail(&dev->todo_list, &net_todo_list); 9533 atomic_inc(&dev_net(dev)->dev_unreg_count); 9534 } 9535 9536 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9537 struct net_device *upper, netdev_features_t features) 9538 { 9539 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9540 netdev_features_t feature; 9541 int feature_bit; 9542 9543 for_each_netdev_feature(upper_disables, feature_bit) { 9544 feature = __NETIF_F_BIT(feature_bit); 9545 if (!(upper->wanted_features & feature) 9546 && (features & feature)) { 9547 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9548 &feature, upper->name); 9549 features &= ~feature; 9550 } 9551 } 9552 9553 return features; 9554 } 9555 9556 static void netdev_sync_lower_features(struct net_device *upper, 9557 struct net_device *lower, netdev_features_t features) 9558 { 9559 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9560 netdev_features_t feature; 9561 int feature_bit; 9562 9563 for_each_netdev_feature(upper_disables, feature_bit) { 9564 feature = __NETIF_F_BIT(feature_bit); 9565 if (!(features & feature) && (lower->features & feature)) { 9566 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9567 &feature, lower->name); 9568 lower->wanted_features &= ~feature; 9569 __netdev_update_features(lower); 9570 9571 if (unlikely(lower->features & feature)) 9572 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9573 &feature, lower->name); 9574 else 9575 netdev_features_change(lower); 9576 } 9577 } 9578 } 9579 9580 static netdev_features_t netdev_fix_features(struct net_device *dev, 9581 netdev_features_t features) 9582 { 9583 /* Fix illegal checksum combinations */ 9584 if ((features & NETIF_F_HW_CSUM) && 9585 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9586 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9587 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9588 } 9589 9590 /* TSO requires that SG is present as well. */ 9591 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9592 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9593 features &= ~NETIF_F_ALL_TSO; 9594 } 9595 9596 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9597 !(features & NETIF_F_IP_CSUM)) { 9598 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9599 features &= ~NETIF_F_TSO; 9600 features &= ~NETIF_F_TSO_ECN; 9601 } 9602 9603 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9604 !(features & NETIF_F_IPV6_CSUM)) { 9605 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9606 features &= ~NETIF_F_TSO6; 9607 } 9608 9609 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9610 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9611 features &= ~NETIF_F_TSO_MANGLEID; 9612 9613 /* TSO ECN requires that TSO is present as well. */ 9614 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9615 features &= ~NETIF_F_TSO_ECN; 9616 9617 /* Software GSO depends on SG. */ 9618 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9619 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9620 features &= ~NETIF_F_GSO; 9621 } 9622 9623 /* GSO partial features require GSO partial be set */ 9624 if ((features & dev->gso_partial_features) && 9625 !(features & NETIF_F_GSO_PARTIAL)) { 9626 netdev_dbg(dev, 9627 "Dropping partially supported GSO features since no GSO partial.\n"); 9628 features &= ~dev->gso_partial_features; 9629 } 9630 9631 if (!(features & NETIF_F_RXCSUM)) { 9632 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9633 * successfully merged by hardware must also have the 9634 * checksum verified by hardware. If the user does not 9635 * want to enable RXCSUM, logically, we should disable GRO_HW. 9636 */ 9637 if (features & NETIF_F_GRO_HW) { 9638 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9639 features &= ~NETIF_F_GRO_HW; 9640 } 9641 } 9642 9643 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9644 if (features & NETIF_F_RXFCS) { 9645 if (features & NETIF_F_LRO) { 9646 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9647 features &= ~NETIF_F_LRO; 9648 } 9649 9650 if (features & NETIF_F_GRO_HW) { 9651 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9652 features &= ~NETIF_F_GRO_HW; 9653 } 9654 } 9655 9656 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9657 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9658 features &= ~NETIF_F_LRO; 9659 } 9660 9661 if (features & NETIF_F_HW_TLS_TX) { 9662 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9663 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9664 bool hw_csum = features & NETIF_F_HW_CSUM; 9665 9666 if (!ip_csum && !hw_csum) { 9667 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9668 features &= ~NETIF_F_HW_TLS_TX; 9669 } 9670 } 9671 9672 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9673 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9674 features &= ~NETIF_F_HW_TLS_RX; 9675 } 9676 9677 return features; 9678 } 9679 9680 int __netdev_update_features(struct net_device *dev) 9681 { 9682 struct net_device *upper, *lower; 9683 netdev_features_t features; 9684 struct list_head *iter; 9685 int err = -1; 9686 9687 ASSERT_RTNL(); 9688 9689 features = netdev_get_wanted_features(dev); 9690 9691 if (dev->netdev_ops->ndo_fix_features) 9692 features = dev->netdev_ops->ndo_fix_features(dev, features); 9693 9694 /* driver might be less strict about feature dependencies */ 9695 features = netdev_fix_features(dev, features); 9696 9697 /* some features can't be enabled if they're off on an upper device */ 9698 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9699 features = netdev_sync_upper_features(dev, upper, features); 9700 9701 if (dev->features == features) 9702 goto sync_lower; 9703 9704 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9705 &dev->features, &features); 9706 9707 if (dev->netdev_ops->ndo_set_features) 9708 err = dev->netdev_ops->ndo_set_features(dev, features); 9709 else 9710 err = 0; 9711 9712 if (unlikely(err < 0)) { 9713 netdev_err(dev, 9714 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9715 err, &features, &dev->features); 9716 /* return non-0 since some features might have changed and 9717 * it's better to fire a spurious notification than miss it 9718 */ 9719 return -1; 9720 } 9721 9722 sync_lower: 9723 /* some features must be disabled on lower devices when disabled 9724 * on an upper device (think: bonding master or bridge) 9725 */ 9726 netdev_for_each_lower_dev(dev, lower, iter) 9727 netdev_sync_lower_features(dev, lower, features); 9728 9729 if (!err) { 9730 netdev_features_t diff = features ^ dev->features; 9731 9732 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9733 /* udp_tunnel_{get,drop}_rx_info both need 9734 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9735 * device, or they won't do anything. 9736 * Thus we need to update dev->features 9737 * *before* calling udp_tunnel_get_rx_info, 9738 * but *after* calling udp_tunnel_drop_rx_info. 9739 */ 9740 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9741 dev->features = features; 9742 udp_tunnel_get_rx_info(dev); 9743 } else { 9744 udp_tunnel_drop_rx_info(dev); 9745 } 9746 } 9747 9748 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9749 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9750 dev->features = features; 9751 err |= vlan_get_rx_ctag_filter_info(dev); 9752 } else { 9753 vlan_drop_rx_ctag_filter_info(dev); 9754 } 9755 } 9756 9757 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9758 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9759 dev->features = features; 9760 err |= vlan_get_rx_stag_filter_info(dev); 9761 } else { 9762 vlan_drop_rx_stag_filter_info(dev); 9763 } 9764 } 9765 9766 dev->features = features; 9767 } 9768 9769 return err < 0 ? 0 : 1; 9770 } 9771 9772 /** 9773 * netdev_update_features - recalculate device features 9774 * @dev: the device to check 9775 * 9776 * Recalculate dev->features set and send notifications if it 9777 * has changed. Should be called after driver or hardware dependent 9778 * conditions might have changed that influence the features. 9779 */ 9780 void netdev_update_features(struct net_device *dev) 9781 { 9782 if (__netdev_update_features(dev)) 9783 netdev_features_change(dev); 9784 } 9785 EXPORT_SYMBOL(netdev_update_features); 9786 9787 /** 9788 * netdev_change_features - recalculate device features 9789 * @dev: the device to check 9790 * 9791 * Recalculate dev->features set and send notifications even 9792 * if they have not changed. Should be called instead of 9793 * netdev_update_features() if also dev->vlan_features might 9794 * have changed to allow the changes to be propagated to stacked 9795 * VLAN devices. 9796 */ 9797 void netdev_change_features(struct net_device *dev) 9798 { 9799 __netdev_update_features(dev); 9800 netdev_features_change(dev); 9801 } 9802 EXPORT_SYMBOL(netdev_change_features); 9803 9804 /** 9805 * netif_stacked_transfer_operstate - transfer operstate 9806 * @rootdev: the root or lower level device to transfer state from 9807 * @dev: the device to transfer operstate to 9808 * 9809 * Transfer operational state from root to device. This is normally 9810 * called when a stacking relationship exists between the root 9811 * device and the device(a leaf device). 9812 */ 9813 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9814 struct net_device *dev) 9815 { 9816 if (rootdev->operstate == IF_OPER_DORMANT) 9817 netif_dormant_on(dev); 9818 else 9819 netif_dormant_off(dev); 9820 9821 if (rootdev->operstate == IF_OPER_TESTING) 9822 netif_testing_on(dev); 9823 else 9824 netif_testing_off(dev); 9825 9826 if (netif_carrier_ok(rootdev)) 9827 netif_carrier_on(dev); 9828 else 9829 netif_carrier_off(dev); 9830 } 9831 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9832 9833 static int netif_alloc_rx_queues(struct net_device *dev) 9834 { 9835 unsigned int i, count = dev->num_rx_queues; 9836 struct netdev_rx_queue *rx; 9837 size_t sz = count * sizeof(*rx); 9838 int err = 0; 9839 9840 BUG_ON(count < 1); 9841 9842 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9843 if (!rx) 9844 return -ENOMEM; 9845 9846 dev->_rx = rx; 9847 9848 for (i = 0; i < count; i++) { 9849 rx[i].dev = dev; 9850 9851 /* XDP RX-queue setup */ 9852 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9853 if (err < 0) 9854 goto err_rxq_info; 9855 } 9856 return 0; 9857 9858 err_rxq_info: 9859 /* Rollback successful reg's and free other resources */ 9860 while (i--) 9861 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9862 kvfree(dev->_rx); 9863 dev->_rx = NULL; 9864 return err; 9865 } 9866 9867 static void netif_free_rx_queues(struct net_device *dev) 9868 { 9869 unsigned int i, count = dev->num_rx_queues; 9870 9871 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9872 if (!dev->_rx) 9873 return; 9874 9875 for (i = 0; i < count; i++) 9876 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9877 9878 kvfree(dev->_rx); 9879 } 9880 9881 static void netdev_init_one_queue(struct net_device *dev, 9882 struct netdev_queue *queue, void *_unused) 9883 { 9884 /* Initialize queue lock */ 9885 spin_lock_init(&queue->_xmit_lock); 9886 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9887 queue->xmit_lock_owner = -1; 9888 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9889 queue->dev = dev; 9890 #ifdef CONFIG_BQL 9891 dql_init(&queue->dql, HZ); 9892 #endif 9893 } 9894 9895 static void netif_free_tx_queues(struct net_device *dev) 9896 { 9897 kvfree(dev->_tx); 9898 } 9899 9900 static int netif_alloc_netdev_queues(struct net_device *dev) 9901 { 9902 unsigned int count = dev->num_tx_queues; 9903 struct netdev_queue *tx; 9904 size_t sz = count * sizeof(*tx); 9905 9906 if (count < 1 || count > 0xffff) 9907 return -EINVAL; 9908 9909 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9910 if (!tx) 9911 return -ENOMEM; 9912 9913 dev->_tx = tx; 9914 9915 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9916 spin_lock_init(&dev->tx_global_lock); 9917 9918 return 0; 9919 } 9920 9921 void netif_tx_stop_all_queues(struct net_device *dev) 9922 { 9923 unsigned int i; 9924 9925 for (i = 0; i < dev->num_tx_queues; i++) { 9926 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9927 9928 netif_tx_stop_queue(txq); 9929 } 9930 } 9931 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9932 9933 /** 9934 * register_netdevice() - register a network device 9935 * @dev: device to register 9936 * 9937 * Take a prepared network device structure and make it externally accessible. 9938 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9939 * Callers must hold the rtnl lock - you may want register_netdev() 9940 * instead of this. 9941 */ 9942 int register_netdevice(struct net_device *dev) 9943 { 9944 int ret; 9945 struct net *net = dev_net(dev); 9946 9947 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9948 NETDEV_FEATURE_COUNT); 9949 BUG_ON(dev_boot_phase); 9950 ASSERT_RTNL(); 9951 9952 might_sleep(); 9953 9954 /* When net_device's are persistent, this will be fatal. */ 9955 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9956 BUG_ON(!net); 9957 9958 ret = ethtool_check_ops(dev->ethtool_ops); 9959 if (ret) 9960 return ret; 9961 9962 spin_lock_init(&dev->addr_list_lock); 9963 netdev_set_addr_lockdep_class(dev); 9964 9965 ret = dev_get_valid_name(net, dev, dev->name); 9966 if (ret < 0) 9967 goto out; 9968 9969 ret = -ENOMEM; 9970 dev->name_node = netdev_name_node_head_alloc(dev); 9971 if (!dev->name_node) 9972 goto out; 9973 9974 /* Init, if this function is available */ 9975 if (dev->netdev_ops->ndo_init) { 9976 ret = dev->netdev_ops->ndo_init(dev); 9977 if (ret) { 9978 if (ret > 0) 9979 ret = -EIO; 9980 goto err_free_name; 9981 } 9982 } 9983 9984 if (((dev->hw_features | dev->features) & 9985 NETIF_F_HW_VLAN_CTAG_FILTER) && 9986 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9987 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9988 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9989 ret = -EINVAL; 9990 goto err_uninit; 9991 } 9992 9993 ret = -EBUSY; 9994 if (!dev->ifindex) 9995 dev->ifindex = dev_new_index(net); 9996 else if (__dev_get_by_index(net, dev->ifindex)) 9997 goto err_uninit; 9998 9999 /* Transfer changeable features to wanted_features and enable 10000 * software offloads (GSO and GRO). 10001 */ 10002 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10003 dev->features |= NETIF_F_SOFT_FEATURES; 10004 10005 if (dev->udp_tunnel_nic_info) { 10006 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10007 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10008 } 10009 10010 dev->wanted_features = dev->features & dev->hw_features; 10011 10012 if (!(dev->flags & IFF_LOOPBACK)) 10013 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10014 10015 /* If IPv4 TCP segmentation offload is supported we should also 10016 * allow the device to enable segmenting the frame with the option 10017 * of ignoring a static IP ID value. This doesn't enable the 10018 * feature itself but allows the user to enable it later. 10019 */ 10020 if (dev->hw_features & NETIF_F_TSO) 10021 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10022 if (dev->vlan_features & NETIF_F_TSO) 10023 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10024 if (dev->mpls_features & NETIF_F_TSO) 10025 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10026 if (dev->hw_enc_features & NETIF_F_TSO) 10027 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10028 10029 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10030 */ 10031 dev->vlan_features |= NETIF_F_HIGHDMA; 10032 10033 /* Make NETIF_F_SG inheritable to tunnel devices. 10034 */ 10035 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10036 10037 /* Make NETIF_F_SG inheritable to MPLS. 10038 */ 10039 dev->mpls_features |= NETIF_F_SG; 10040 10041 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10042 ret = notifier_to_errno(ret); 10043 if (ret) 10044 goto err_uninit; 10045 10046 ret = netdev_register_kobject(dev); 10047 write_lock(&dev_base_lock); 10048 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10049 write_unlock(&dev_base_lock); 10050 if (ret) 10051 goto err_uninit_notify; 10052 10053 __netdev_update_features(dev); 10054 10055 /* 10056 * Default initial state at registry is that the 10057 * device is present. 10058 */ 10059 10060 set_bit(__LINK_STATE_PRESENT, &dev->state); 10061 10062 linkwatch_init_dev(dev); 10063 10064 dev_init_scheduler(dev); 10065 10066 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10067 list_netdevice(dev); 10068 10069 add_device_randomness(dev->dev_addr, dev->addr_len); 10070 10071 /* If the device has permanent device address, driver should 10072 * set dev_addr and also addr_assign_type should be set to 10073 * NET_ADDR_PERM (default value). 10074 */ 10075 if (dev->addr_assign_type == NET_ADDR_PERM) 10076 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10077 10078 /* Notify protocols, that a new device appeared. */ 10079 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10080 ret = notifier_to_errno(ret); 10081 if (ret) { 10082 /* Expect explicit free_netdev() on failure */ 10083 dev->needs_free_netdev = false; 10084 unregister_netdevice_queue(dev, NULL); 10085 goto out; 10086 } 10087 /* 10088 * Prevent userspace races by waiting until the network 10089 * device is fully setup before sending notifications. 10090 */ 10091 if (!dev->rtnl_link_ops || 10092 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10093 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10094 10095 out: 10096 return ret; 10097 10098 err_uninit_notify: 10099 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10100 err_uninit: 10101 if (dev->netdev_ops->ndo_uninit) 10102 dev->netdev_ops->ndo_uninit(dev); 10103 if (dev->priv_destructor) 10104 dev->priv_destructor(dev); 10105 err_free_name: 10106 netdev_name_node_free(dev->name_node); 10107 goto out; 10108 } 10109 EXPORT_SYMBOL(register_netdevice); 10110 10111 /** 10112 * init_dummy_netdev - init a dummy network device for NAPI 10113 * @dev: device to init 10114 * 10115 * This takes a network device structure and initialize the minimum 10116 * amount of fields so it can be used to schedule NAPI polls without 10117 * registering a full blown interface. This is to be used by drivers 10118 * that need to tie several hardware interfaces to a single NAPI 10119 * poll scheduler due to HW limitations. 10120 */ 10121 int init_dummy_netdev(struct net_device *dev) 10122 { 10123 /* Clear everything. Note we don't initialize spinlocks 10124 * are they aren't supposed to be taken by any of the 10125 * NAPI code and this dummy netdev is supposed to be 10126 * only ever used for NAPI polls 10127 */ 10128 memset(dev, 0, sizeof(struct net_device)); 10129 10130 /* make sure we BUG if trying to hit standard 10131 * register/unregister code path 10132 */ 10133 dev->reg_state = NETREG_DUMMY; 10134 10135 /* NAPI wants this */ 10136 INIT_LIST_HEAD(&dev->napi_list); 10137 10138 /* a dummy interface is started by default */ 10139 set_bit(__LINK_STATE_PRESENT, &dev->state); 10140 set_bit(__LINK_STATE_START, &dev->state); 10141 10142 /* napi_busy_loop stats accounting wants this */ 10143 dev_net_set(dev, &init_net); 10144 10145 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10146 * because users of this 'device' dont need to change 10147 * its refcount. 10148 */ 10149 10150 return 0; 10151 } 10152 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10153 10154 10155 /** 10156 * register_netdev - register a network device 10157 * @dev: device to register 10158 * 10159 * Take a completed network device structure and add it to the kernel 10160 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10161 * chain. 0 is returned on success. A negative errno code is returned 10162 * on a failure to set up the device, or if the name is a duplicate. 10163 * 10164 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10165 * and expands the device name if you passed a format string to 10166 * alloc_netdev. 10167 */ 10168 int register_netdev(struct net_device *dev) 10169 { 10170 int err; 10171 10172 if (rtnl_lock_killable()) 10173 return -EINTR; 10174 err = register_netdevice(dev); 10175 rtnl_unlock(); 10176 return err; 10177 } 10178 EXPORT_SYMBOL(register_netdev); 10179 10180 int netdev_refcnt_read(const struct net_device *dev) 10181 { 10182 #ifdef CONFIG_PCPU_DEV_REFCNT 10183 int i, refcnt = 0; 10184 10185 for_each_possible_cpu(i) 10186 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10187 return refcnt; 10188 #else 10189 return refcount_read(&dev->dev_refcnt); 10190 #endif 10191 } 10192 EXPORT_SYMBOL(netdev_refcnt_read); 10193 10194 int netdev_unregister_timeout_secs __read_mostly = 10; 10195 10196 #define WAIT_REFS_MIN_MSECS 1 10197 #define WAIT_REFS_MAX_MSECS 250 10198 /** 10199 * netdev_wait_allrefs_any - wait until all references are gone. 10200 * @list: list of net_devices to wait on 10201 * 10202 * This is called when unregistering network devices. 10203 * 10204 * Any protocol or device that holds a reference should register 10205 * for netdevice notification, and cleanup and put back the 10206 * reference if they receive an UNREGISTER event. 10207 * We can get stuck here if buggy protocols don't correctly 10208 * call dev_put. 10209 */ 10210 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10211 { 10212 unsigned long rebroadcast_time, warning_time; 10213 struct net_device *dev; 10214 int wait = 0; 10215 10216 rebroadcast_time = warning_time = jiffies; 10217 10218 list_for_each_entry(dev, list, todo_list) 10219 if (netdev_refcnt_read(dev) == 1) 10220 return dev; 10221 10222 while (true) { 10223 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10224 rtnl_lock(); 10225 10226 /* Rebroadcast unregister notification */ 10227 list_for_each_entry(dev, list, todo_list) 10228 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10229 10230 __rtnl_unlock(); 10231 rcu_barrier(); 10232 rtnl_lock(); 10233 10234 list_for_each_entry(dev, list, todo_list) 10235 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10236 &dev->state)) { 10237 /* We must not have linkwatch events 10238 * pending on unregister. If this 10239 * happens, we simply run the queue 10240 * unscheduled, resulting in a noop 10241 * for this device. 10242 */ 10243 linkwatch_run_queue(); 10244 break; 10245 } 10246 10247 __rtnl_unlock(); 10248 10249 rebroadcast_time = jiffies; 10250 } 10251 10252 if (!wait) { 10253 rcu_barrier(); 10254 wait = WAIT_REFS_MIN_MSECS; 10255 } else { 10256 msleep(wait); 10257 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10258 } 10259 10260 list_for_each_entry(dev, list, todo_list) 10261 if (netdev_refcnt_read(dev) == 1) 10262 return dev; 10263 10264 if (time_after(jiffies, warning_time + 10265 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10266 list_for_each_entry(dev, list, todo_list) { 10267 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10268 dev->name, netdev_refcnt_read(dev)); 10269 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10270 } 10271 10272 warning_time = jiffies; 10273 } 10274 } 10275 } 10276 10277 /* The sequence is: 10278 * 10279 * rtnl_lock(); 10280 * ... 10281 * register_netdevice(x1); 10282 * register_netdevice(x2); 10283 * ... 10284 * unregister_netdevice(y1); 10285 * unregister_netdevice(y2); 10286 * ... 10287 * rtnl_unlock(); 10288 * free_netdev(y1); 10289 * free_netdev(y2); 10290 * 10291 * We are invoked by rtnl_unlock(). 10292 * This allows us to deal with problems: 10293 * 1) We can delete sysfs objects which invoke hotplug 10294 * without deadlocking with linkwatch via keventd. 10295 * 2) Since we run with the RTNL semaphore not held, we can sleep 10296 * safely in order to wait for the netdev refcnt to drop to zero. 10297 * 10298 * We must not return until all unregister events added during 10299 * the interval the lock was held have been completed. 10300 */ 10301 void netdev_run_todo(void) 10302 { 10303 struct net_device *dev, *tmp; 10304 struct list_head list; 10305 #ifdef CONFIG_LOCKDEP 10306 struct list_head unlink_list; 10307 10308 list_replace_init(&net_unlink_list, &unlink_list); 10309 10310 while (!list_empty(&unlink_list)) { 10311 struct net_device *dev = list_first_entry(&unlink_list, 10312 struct net_device, 10313 unlink_list); 10314 list_del_init(&dev->unlink_list); 10315 dev->nested_level = dev->lower_level - 1; 10316 } 10317 #endif 10318 10319 /* Snapshot list, allow later requests */ 10320 list_replace_init(&net_todo_list, &list); 10321 10322 __rtnl_unlock(); 10323 10324 /* Wait for rcu callbacks to finish before next phase */ 10325 if (!list_empty(&list)) 10326 rcu_barrier(); 10327 10328 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10329 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10330 netdev_WARN(dev, "run_todo but not unregistering\n"); 10331 list_del(&dev->todo_list); 10332 continue; 10333 } 10334 10335 write_lock(&dev_base_lock); 10336 dev->reg_state = NETREG_UNREGISTERED; 10337 write_unlock(&dev_base_lock); 10338 linkwatch_forget_dev(dev); 10339 } 10340 10341 while (!list_empty(&list)) { 10342 dev = netdev_wait_allrefs_any(&list); 10343 list_del(&dev->todo_list); 10344 10345 /* paranoia */ 10346 BUG_ON(netdev_refcnt_read(dev) != 1); 10347 BUG_ON(!list_empty(&dev->ptype_all)); 10348 BUG_ON(!list_empty(&dev->ptype_specific)); 10349 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10350 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10351 10352 if (dev->priv_destructor) 10353 dev->priv_destructor(dev); 10354 if (dev->needs_free_netdev) 10355 free_netdev(dev); 10356 10357 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10358 wake_up(&netdev_unregistering_wq); 10359 10360 /* Free network device */ 10361 kobject_put(&dev->dev.kobj); 10362 } 10363 } 10364 10365 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10366 * all the same fields in the same order as net_device_stats, with only 10367 * the type differing, but rtnl_link_stats64 may have additional fields 10368 * at the end for newer counters. 10369 */ 10370 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10371 const struct net_device_stats *netdev_stats) 10372 { 10373 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10374 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10375 u64 *dst = (u64 *)stats64; 10376 10377 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10378 for (i = 0; i < n; i++) 10379 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10380 /* zero out counters that only exist in rtnl_link_stats64 */ 10381 memset((char *)stats64 + n * sizeof(u64), 0, 10382 sizeof(*stats64) - n * sizeof(u64)); 10383 } 10384 EXPORT_SYMBOL(netdev_stats_to_stats64); 10385 10386 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10387 { 10388 struct net_device_core_stats __percpu *p; 10389 10390 p = alloc_percpu_gfp(struct net_device_core_stats, 10391 GFP_ATOMIC | __GFP_NOWARN); 10392 10393 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10394 free_percpu(p); 10395 10396 /* This READ_ONCE() pairs with the cmpxchg() above */ 10397 return READ_ONCE(dev->core_stats); 10398 } 10399 EXPORT_SYMBOL(netdev_core_stats_alloc); 10400 10401 /** 10402 * dev_get_stats - get network device statistics 10403 * @dev: device to get statistics from 10404 * @storage: place to store stats 10405 * 10406 * Get network statistics from device. Return @storage. 10407 * The device driver may provide its own method by setting 10408 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10409 * otherwise the internal statistics structure is used. 10410 */ 10411 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10412 struct rtnl_link_stats64 *storage) 10413 { 10414 const struct net_device_ops *ops = dev->netdev_ops; 10415 const struct net_device_core_stats __percpu *p; 10416 10417 if (ops->ndo_get_stats64) { 10418 memset(storage, 0, sizeof(*storage)); 10419 ops->ndo_get_stats64(dev, storage); 10420 } else if (ops->ndo_get_stats) { 10421 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10422 } else { 10423 netdev_stats_to_stats64(storage, &dev->stats); 10424 } 10425 10426 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10427 p = READ_ONCE(dev->core_stats); 10428 if (p) { 10429 const struct net_device_core_stats *core_stats; 10430 int i; 10431 10432 for_each_possible_cpu(i) { 10433 core_stats = per_cpu_ptr(p, i); 10434 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10435 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10436 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10437 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10438 } 10439 } 10440 return storage; 10441 } 10442 EXPORT_SYMBOL(dev_get_stats); 10443 10444 /** 10445 * dev_fetch_sw_netstats - get per-cpu network device statistics 10446 * @s: place to store stats 10447 * @netstats: per-cpu network stats to read from 10448 * 10449 * Read per-cpu network statistics and populate the related fields in @s. 10450 */ 10451 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10452 const struct pcpu_sw_netstats __percpu *netstats) 10453 { 10454 int cpu; 10455 10456 for_each_possible_cpu(cpu) { 10457 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10458 const struct pcpu_sw_netstats *stats; 10459 unsigned int start; 10460 10461 stats = per_cpu_ptr(netstats, cpu); 10462 do { 10463 start = u64_stats_fetch_begin(&stats->syncp); 10464 rx_packets = u64_stats_read(&stats->rx_packets); 10465 rx_bytes = u64_stats_read(&stats->rx_bytes); 10466 tx_packets = u64_stats_read(&stats->tx_packets); 10467 tx_bytes = u64_stats_read(&stats->tx_bytes); 10468 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10469 10470 s->rx_packets += rx_packets; 10471 s->rx_bytes += rx_bytes; 10472 s->tx_packets += tx_packets; 10473 s->tx_bytes += tx_bytes; 10474 } 10475 } 10476 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10477 10478 /** 10479 * dev_get_tstats64 - ndo_get_stats64 implementation 10480 * @dev: device to get statistics from 10481 * @s: place to store stats 10482 * 10483 * Populate @s from dev->stats and dev->tstats. Can be used as 10484 * ndo_get_stats64() callback. 10485 */ 10486 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10487 { 10488 netdev_stats_to_stats64(s, &dev->stats); 10489 dev_fetch_sw_netstats(s, dev->tstats); 10490 } 10491 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10492 10493 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10494 { 10495 struct netdev_queue *queue = dev_ingress_queue(dev); 10496 10497 #ifdef CONFIG_NET_CLS_ACT 10498 if (queue) 10499 return queue; 10500 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10501 if (!queue) 10502 return NULL; 10503 netdev_init_one_queue(dev, queue, NULL); 10504 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10505 queue->qdisc_sleeping = &noop_qdisc; 10506 rcu_assign_pointer(dev->ingress_queue, queue); 10507 #endif 10508 return queue; 10509 } 10510 10511 static const struct ethtool_ops default_ethtool_ops; 10512 10513 void netdev_set_default_ethtool_ops(struct net_device *dev, 10514 const struct ethtool_ops *ops) 10515 { 10516 if (dev->ethtool_ops == &default_ethtool_ops) 10517 dev->ethtool_ops = ops; 10518 } 10519 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10520 10521 /** 10522 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10523 * @dev: netdev to enable the IRQ coalescing on 10524 * 10525 * Sets a conservative default for SW IRQ coalescing. Users can use 10526 * sysfs attributes to override the default values. 10527 */ 10528 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10529 { 10530 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10531 10532 dev->gro_flush_timeout = 20000; 10533 dev->napi_defer_hard_irqs = 1; 10534 } 10535 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10536 10537 void netdev_freemem(struct net_device *dev) 10538 { 10539 char *addr = (char *)dev - dev->padded; 10540 10541 kvfree(addr); 10542 } 10543 10544 /** 10545 * alloc_netdev_mqs - allocate network device 10546 * @sizeof_priv: size of private data to allocate space for 10547 * @name: device name format string 10548 * @name_assign_type: origin of device name 10549 * @setup: callback to initialize device 10550 * @txqs: the number of TX subqueues to allocate 10551 * @rxqs: the number of RX subqueues to allocate 10552 * 10553 * Allocates a struct net_device with private data area for driver use 10554 * and performs basic initialization. Also allocates subqueue structs 10555 * for each queue on the device. 10556 */ 10557 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10558 unsigned char name_assign_type, 10559 void (*setup)(struct net_device *), 10560 unsigned int txqs, unsigned int rxqs) 10561 { 10562 struct net_device *dev; 10563 unsigned int alloc_size; 10564 struct net_device *p; 10565 10566 BUG_ON(strlen(name) >= sizeof(dev->name)); 10567 10568 if (txqs < 1) { 10569 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10570 return NULL; 10571 } 10572 10573 if (rxqs < 1) { 10574 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10575 return NULL; 10576 } 10577 10578 alloc_size = sizeof(struct net_device); 10579 if (sizeof_priv) { 10580 /* ensure 32-byte alignment of private area */ 10581 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10582 alloc_size += sizeof_priv; 10583 } 10584 /* ensure 32-byte alignment of whole construct */ 10585 alloc_size += NETDEV_ALIGN - 1; 10586 10587 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10588 if (!p) 10589 return NULL; 10590 10591 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10592 dev->padded = (char *)dev - (char *)p; 10593 10594 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10595 #ifdef CONFIG_PCPU_DEV_REFCNT 10596 dev->pcpu_refcnt = alloc_percpu(int); 10597 if (!dev->pcpu_refcnt) 10598 goto free_dev; 10599 __dev_hold(dev); 10600 #else 10601 refcount_set(&dev->dev_refcnt, 1); 10602 #endif 10603 10604 if (dev_addr_init(dev)) 10605 goto free_pcpu; 10606 10607 dev_mc_init(dev); 10608 dev_uc_init(dev); 10609 10610 dev_net_set(dev, &init_net); 10611 10612 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10613 dev->gso_max_segs = GSO_MAX_SEGS; 10614 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10615 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10616 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10617 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10618 dev->tso_max_segs = TSO_MAX_SEGS; 10619 dev->upper_level = 1; 10620 dev->lower_level = 1; 10621 #ifdef CONFIG_LOCKDEP 10622 dev->nested_level = 0; 10623 INIT_LIST_HEAD(&dev->unlink_list); 10624 #endif 10625 10626 INIT_LIST_HEAD(&dev->napi_list); 10627 INIT_LIST_HEAD(&dev->unreg_list); 10628 INIT_LIST_HEAD(&dev->close_list); 10629 INIT_LIST_HEAD(&dev->link_watch_list); 10630 INIT_LIST_HEAD(&dev->adj_list.upper); 10631 INIT_LIST_HEAD(&dev->adj_list.lower); 10632 INIT_LIST_HEAD(&dev->ptype_all); 10633 INIT_LIST_HEAD(&dev->ptype_specific); 10634 INIT_LIST_HEAD(&dev->net_notifier_list); 10635 #ifdef CONFIG_NET_SCHED 10636 hash_init(dev->qdisc_hash); 10637 #endif 10638 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10639 setup(dev); 10640 10641 if (!dev->tx_queue_len) { 10642 dev->priv_flags |= IFF_NO_QUEUE; 10643 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10644 } 10645 10646 dev->num_tx_queues = txqs; 10647 dev->real_num_tx_queues = txqs; 10648 if (netif_alloc_netdev_queues(dev)) 10649 goto free_all; 10650 10651 dev->num_rx_queues = rxqs; 10652 dev->real_num_rx_queues = rxqs; 10653 if (netif_alloc_rx_queues(dev)) 10654 goto free_all; 10655 10656 strcpy(dev->name, name); 10657 dev->name_assign_type = name_assign_type; 10658 dev->group = INIT_NETDEV_GROUP; 10659 if (!dev->ethtool_ops) 10660 dev->ethtool_ops = &default_ethtool_ops; 10661 10662 nf_hook_netdev_init(dev); 10663 10664 return dev; 10665 10666 free_all: 10667 free_netdev(dev); 10668 return NULL; 10669 10670 free_pcpu: 10671 #ifdef CONFIG_PCPU_DEV_REFCNT 10672 free_percpu(dev->pcpu_refcnt); 10673 free_dev: 10674 #endif 10675 netdev_freemem(dev); 10676 return NULL; 10677 } 10678 EXPORT_SYMBOL(alloc_netdev_mqs); 10679 10680 /** 10681 * free_netdev - free network device 10682 * @dev: device 10683 * 10684 * This function does the last stage of destroying an allocated device 10685 * interface. The reference to the device object is released. If this 10686 * is the last reference then it will be freed.Must be called in process 10687 * context. 10688 */ 10689 void free_netdev(struct net_device *dev) 10690 { 10691 struct napi_struct *p, *n; 10692 10693 might_sleep(); 10694 10695 /* When called immediately after register_netdevice() failed the unwind 10696 * handling may still be dismantling the device. Handle that case by 10697 * deferring the free. 10698 */ 10699 if (dev->reg_state == NETREG_UNREGISTERING) { 10700 ASSERT_RTNL(); 10701 dev->needs_free_netdev = true; 10702 return; 10703 } 10704 10705 netif_free_tx_queues(dev); 10706 netif_free_rx_queues(dev); 10707 10708 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10709 10710 /* Flush device addresses */ 10711 dev_addr_flush(dev); 10712 10713 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10714 netif_napi_del(p); 10715 10716 ref_tracker_dir_exit(&dev->refcnt_tracker); 10717 #ifdef CONFIG_PCPU_DEV_REFCNT 10718 free_percpu(dev->pcpu_refcnt); 10719 dev->pcpu_refcnt = NULL; 10720 #endif 10721 free_percpu(dev->core_stats); 10722 dev->core_stats = NULL; 10723 free_percpu(dev->xdp_bulkq); 10724 dev->xdp_bulkq = NULL; 10725 10726 /* Compatibility with error handling in drivers */ 10727 if (dev->reg_state == NETREG_UNINITIALIZED) { 10728 netdev_freemem(dev); 10729 return; 10730 } 10731 10732 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10733 dev->reg_state = NETREG_RELEASED; 10734 10735 /* will free via device release */ 10736 put_device(&dev->dev); 10737 } 10738 EXPORT_SYMBOL(free_netdev); 10739 10740 /** 10741 * synchronize_net - Synchronize with packet receive processing 10742 * 10743 * Wait for packets currently being received to be done. 10744 * Does not block later packets from starting. 10745 */ 10746 void synchronize_net(void) 10747 { 10748 might_sleep(); 10749 if (rtnl_is_locked()) 10750 synchronize_rcu_expedited(); 10751 else 10752 synchronize_rcu(); 10753 } 10754 EXPORT_SYMBOL(synchronize_net); 10755 10756 /** 10757 * unregister_netdevice_queue - remove device from the kernel 10758 * @dev: device 10759 * @head: list 10760 * 10761 * This function shuts down a device interface and removes it 10762 * from the kernel tables. 10763 * If head not NULL, device is queued to be unregistered later. 10764 * 10765 * Callers must hold the rtnl semaphore. You may want 10766 * unregister_netdev() instead of this. 10767 */ 10768 10769 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10770 { 10771 ASSERT_RTNL(); 10772 10773 if (head) { 10774 list_move_tail(&dev->unreg_list, head); 10775 } else { 10776 LIST_HEAD(single); 10777 10778 list_add(&dev->unreg_list, &single); 10779 unregister_netdevice_many(&single); 10780 } 10781 } 10782 EXPORT_SYMBOL(unregister_netdevice_queue); 10783 10784 void unregister_netdevice_many_notify(struct list_head *head, 10785 u32 portid, const struct nlmsghdr *nlh) 10786 { 10787 struct net_device *dev, *tmp; 10788 LIST_HEAD(close_head); 10789 10790 BUG_ON(dev_boot_phase); 10791 ASSERT_RTNL(); 10792 10793 if (list_empty(head)) 10794 return; 10795 10796 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10797 /* Some devices call without registering 10798 * for initialization unwind. Remove those 10799 * devices and proceed with the remaining. 10800 */ 10801 if (dev->reg_state == NETREG_UNINITIALIZED) { 10802 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10803 dev->name, dev); 10804 10805 WARN_ON(1); 10806 list_del(&dev->unreg_list); 10807 continue; 10808 } 10809 dev->dismantle = true; 10810 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10811 } 10812 10813 /* If device is running, close it first. */ 10814 list_for_each_entry(dev, head, unreg_list) 10815 list_add_tail(&dev->close_list, &close_head); 10816 dev_close_many(&close_head, true); 10817 10818 list_for_each_entry(dev, head, unreg_list) { 10819 /* And unlink it from device chain. */ 10820 write_lock(&dev_base_lock); 10821 unlist_netdevice(dev, false); 10822 dev->reg_state = NETREG_UNREGISTERING; 10823 write_unlock(&dev_base_lock); 10824 } 10825 flush_all_backlogs(); 10826 10827 synchronize_net(); 10828 10829 list_for_each_entry(dev, head, unreg_list) { 10830 struct sk_buff *skb = NULL; 10831 10832 /* Shutdown queueing discipline. */ 10833 dev_shutdown(dev); 10834 10835 dev_xdp_uninstall(dev); 10836 bpf_dev_bound_netdev_unregister(dev); 10837 10838 netdev_offload_xstats_disable_all(dev); 10839 10840 /* Notify protocols, that we are about to destroy 10841 * this device. They should clean all the things. 10842 */ 10843 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10844 10845 if (!dev->rtnl_link_ops || 10846 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10847 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10848 GFP_KERNEL, NULL, 0, 10849 portid, nlmsg_seq(nlh)); 10850 10851 /* 10852 * Flush the unicast and multicast chains 10853 */ 10854 dev_uc_flush(dev); 10855 dev_mc_flush(dev); 10856 10857 netdev_name_node_alt_flush(dev); 10858 netdev_name_node_free(dev->name_node); 10859 10860 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10861 10862 if (dev->netdev_ops->ndo_uninit) 10863 dev->netdev_ops->ndo_uninit(dev); 10864 10865 if (skb) 10866 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10867 10868 /* Notifier chain MUST detach us all upper devices. */ 10869 WARN_ON(netdev_has_any_upper_dev(dev)); 10870 WARN_ON(netdev_has_any_lower_dev(dev)); 10871 10872 /* Remove entries from kobject tree */ 10873 netdev_unregister_kobject(dev); 10874 #ifdef CONFIG_XPS 10875 /* Remove XPS queueing entries */ 10876 netif_reset_xps_queues_gt(dev, 0); 10877 #endif 10878 } 10879 10880 synchronize_net(); 10881 10882 list_for_each_entry(dev, head, unreg_list) { 10883 netdev_put(dev, &dev->dev_registered_tracker); 10884 net_set_todo(dev); 10885 } 10886 10887 list_del(head); 10888 } 10889 10890 /** 10891 * unregister_netdevice_many - unregister many devices 10892 * @head: list of devices 10893 * 10894 * Note: As most callers use a stack allocated list_head, 10895 * we force a list_del() to make sure stack wont be corrupted later. 10896 */ 10897 void unregister_netdevice_many(struct list_head *head) 10898 { 10899 unregister_netdevice_many_notify(head, 0, NULL); 10900 } 10901 EXPORT_SYMBOL(unregister_netdevice_many); 10902 10903 /** 10904 * unregister_netdev - remove device from the kernel 10905 * @dev: device 10906 * 10907 * This function shuts down a device interface and removes it 10908 * from the kernel tables. 10909 * 10910 * This is just a wrapper for unregister_netdevice that takes 10911 * the rtnl semaphore. In general you want to use this and not 10912 * unregister_netdevice. 10913 */ 10914 void unregister_netdev(struct net_device *dev) 10915 { 10916 rtnl_lock(); 10917 unregister_netdevice(dev); 10918 rtnl_unlock(); 10919 } 10920 EXPORT_SYMBOL(unregister_netdev); 10921 10922 /** 10923 * __dev_change_net_namespace - move device to different nethost namespace 10924 * @dev: device 10925 * @net: network namespace 10926 * @pat: If not NULL name pattern to try if the current device name 10927 * is already taken in the destination network namespace. 10928 * @new_ifindex: If not zero, specifies device index in the target 10929 * namespace. 10930 * 10931 * This function shuts down a device interface and moves it 10932 * to a new network namespace. On success 0 is returned, on 10933 * a failure a netagive errno code is returned. 10934 * 10935 * Callers must hold the rtnl semaphore. 10936 */ 10937 10938 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10939 const char *pat, int new_ifindex) 10940 { 10941 struct net *net_old = dev_net(dev); 10942 int err, new_nsid; 10943 10944 ASSERT_RTNL(); 10945 10946 /* Don't allow namespace local devices to be moved. */ 10947 err = -EINVAL; 10948 if (dev->features & NETIF_F_NETNS_LOCAL) 10949 goto out; 10950 10951 /* Ensure the device has been registrered */ 10952 if (dev->reg_state != NETREG_REGISTERED) 10953 goto out; 10954 10955 /* Get out if there is nothing todo */ 10956 err = 0; 10957 if (net_eq(net_old, net)) 10958 goto out; 10959 10960 /* Pick the destination device name, and ensure 10961 * we can use it in the destination network namespace. 10962 */ 10963 err = -EEXIST; 10964 if (netdev_name_in_use(net, dev->name)) { 10965 /* We get here if we can't use the current device name */ 10966 if (!pat) 10967 goto out; 10968 err = dev_get_valid_name(net, dev, pat); 10969 if (err < 0) 10970 goto out; 10971 } 10972 10973 /* Check that new_ifindex isn't used yet. */ 10974 err = -EBUSY; 10975 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10976 goto out; 10977 10978 /* 10979 * And now a mini version of register_netdevice unregister_netdevice. 10980 */ 10981 10982 /* If device is running close it first. */ 10983 dev_close(dev); 10984 10985 /* And unlink it from device chain */ 10986 unlist_netdevice(dev, true); 10987 10988 synchronize_net(); 10989 10990 /* Shutdown queueing discipline. */ 10991 dev_shutdown(dev); 10992 10993 /* Notify protocols, that we are about to destroy 10994 * this device. They should clean all the things. 10995 * 10996 * Note that dev->reg_state stays at NETREG_REGISTERED. 10997 * This is wanted because this way 8021q and macvlan know 10998 * the device is just moving and can keep their slaves up. 10999 */ 11000 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11001 rcu_barrier(); 11002 11003 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11004 /* If there is an ifindex conflict assign a new one */ 11005 if (!new_ifindex) { 11006 if (__dev_get_by_index(net, dev->ifindex)) 11007 new_ifindex = dev_new_index(net); 11008 else 11009 new_ifindex = dev->ifindex; 11010 } 11011 11012 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11013 new_ifindex); 11014 11015 /* 11016 * Flush the unicast and multicast chains 11017 */ 11018 dev_uc_flush(dev); 11019 dev_mc_flush(dev); 11020 11021 /* Send a netdev-removed uevent to the old namespace */ 11022 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11023 netdev_adjacent_del_links(dev); 11024 11025 /* Move per-net netdevice notifiers that are following the netdevice */ 11026 move_netdevice_notifiers_dev_net(dev, net); 11027 11028 /* Actually switch the network namespace */ 11029 dev_net_set(dev, net); 11030 dev->ifindex = new_ifindex; 11031 11032 /* Send a netdev-add uevent to the new namespace */ 11033 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11034 netdev_adjacent_add_links(dev); 11035 11036 /* Fixup kobjects */ 11037 err = device_rename(&dev->dev, dev->name); 11038 WARN_ON(err); 11039 11040 /* Adapt owner in case owning user namespace of target network 11041 * namespace is different from the original one. 11042 */ 11043 err = netdev_change_owner(dev, net_old, net); 11044 WARN_ON(err); 11045 11046 /* Add the device back in the hashes */ 11047 list_netdevice(dev); 11048 11049 /* Notify protocols, that a new device appeared. */ 11050 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11051 11052 /* 11053 * Prevent userspace races by waiting until the network 11054 * device is fully setup before sending notifications. 11055 */ 11056 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11057 11058 synchronize_net(); 11059 err = 0; 11060 out: 11061 return err; 11062 } 11063 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11064 11065 static int dev_cpu_dead(unsigned int oldcpu) 11066 { 11067 struct sk_buff **list_skb; 11068 struct sk_buff *skb; 11069 unsigned int cpu; 11070 struct softnet_data *sd, *oldsd, *remsd = NULL; 11071 11072 local_irq_disable(); 11073 cpu = smp_processor_id(); 11074 sd = &per_cpu(softnet_data, cpu); 11075 oldsd = &per_cpu(softnet_data, oldcpu); 11076 11077 /* Find end of our completion_queue. */ 11078 list_skb = &sd->completion_queue; 11079 while (*list_skb) 11080 list_skb = &(*list_skb)->next; 11081 /* Append completion queue from offline CPU. */ 11082 *list_skb = oldsd->completion_queue; 11083 oldsd->completion_queue = NULL; 11084 11085 /* Append output queue from offline CPU. */ 11086 if (oldsd->output_queue) { 11087 *sd->output_queue_tailp = oldsd->output_queue; 11088 sd->output_queue_tailp = oldsd->output_queue_tailp; 11089 oldsd->output_queue = NULL; 11090 oldsd->output_queue_tailp = &oldsd->output_queue; 11091 } 11092 /* Append NAPI poll list from offline CPU, with one exception : 11093 * process_backlog() must be called by cpu owning percpu backlog. 11094 * We properly handle process_queue & input_pkt_queue later. 11095 */ 11096 while (!list_empty(&oldsd->poll_list)) { 11097 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11098 struct napi_struct, 11099 poll_list); 11100 11101 list_del_init(&napi->poll_list); 11102 if (napi->poll == process_backlog) 11103 napi->state = 0; 11104 else 11105 ____napi_schedule(sd, napi); 11106 } 11107 11108 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11109 local_irq_enable(); 11110 11111 #ifdef CONFIG_RPS 11112 remsd = oldsd->rps_ipi_list; 11113 oldsd->rps_ipi_list = NULL; 11114 #endif 11115 /* send out pending IPI's on offline CPU */ 11116 net_rps_send_ipi(remsd); 11117 11118 /* Process offline CPU's input_pkt_queue */ 11119 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11120 netif_rx(skb); 11121 input_queue_head_incr(oldsd); 11122 } 11123 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11124 netif_rx(skb); 11125 input_queue_head_incr(oldsd); 11126 } 11127 11128 return 0; 11129 } 11130 11131 /** 11132 * netdev_increment_features - increment feature set by one 11133 * @all: current feature set 11134 * @one: new feature set 11135 * @mask: mask feature set 11136 * 11137 * Computes a new feature set after adding a device with feature set 11138 * @one to the master device with current feature set @all. Will not 11139 * enable anything that is off in @mask. Returns the new feature set. 11140 */ 11141 netdev_features_t netdev_increment_features(netdev_features_t all, 11142 netdev_features_t one, netdev_features_t mask) 11143 { 11144 if (mask & NETIF_F_HW_CSUM) 11145 mask |= NETIF_F_CSUM_MASK; 11146 mask |= NETIF_F_VLAN_CHALLENGED; 11147 11148 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11149 all &= one | ~NETIF_F_ALL_FOR_ALL; 11150 11151 /* If one device supports hw checksumming, set for all. */ 11152 if (all & NETIF_F_HW_CSUM) 11153 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11154 11155 return all; 11156 } 11157 EXPORT_SYMBOL(netdev_increment_features); 11158 11159 static struct hlist_head * __net_init netdev_create_hash(void) 11160 { 11161 int i; 11162 struct hlist_head *hash; 11163 11164 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11165 if (hash != NULL) 11166 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11167 INIT_HLIST_HEAD(&hash[i]); 11168 11169 return hash; 11170 } 11171 11172 /* Initialize per network namespace state */ 11173 static int __net_init netdev_init(struct net *net) 11174 { 11175 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11176 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11177 11178 INIT_LIST_HEAD(&net->dev_base_head); 11179 11180 net->dev_name_head = netdev_create_hash(); 11181 if (net->dev_name_head == NULL) 11182 goto err_name; 11183 11184 net->dev_index_head = netdev_create_hash(); 11185 if (net->dev_index_head == NULL) 11186 goto err_idx; 11187 11188 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11189 11190 return 0; 11191 11192 err_idx: 11193 kfree(net->dev_name_head); 11194 err_name: 11195 return -ENOMEM; 11196 } 11197 11198 /** 11199 * netdev_drivername - network driver for the device 11200 * @dev: network device 11201 * 11202 * Determine network driver for device. 11203 */ 11204 const char *netdev_drivername(const struct net_device *dev) 11205 { 11206 const struct device_driver *driver; 11207 const struct device *parent; 11208 const char *empty = ""; 11209 11210 parent = dev->dev.parent; 11211 if (!parent) 11212 return empty; 11213 11214 driver = parent->driver; 11215 if (driver && driver->name) 11216 return driver->name; 11217 return empty; 11218 } 11219 11220 static void __netdev_printk(const char *level, const struct net_device *dev, 11221 struct va_format *vaf) 11222 { 11223 if (dev && dev->dev.parent) { 11224 dev_printk_emit(level[1] - '0', 11225 dev->dev.parent, 11226 "%s %s %s%s: %pV", 11227 dev_driver_string(dev->dev.parent), 11228 dev_name(dev->dev.parent), 11229 netdev_name(dev), netdev_reg_state(dev), 11230 vaf); 11231 } else if (dev) { 11232 printk("%s%s%s: %pV", 11233 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11234 } else { 11235 printk("%s(NULL net_device): %pV", level, vaf); 11236 } 11237 } 11238 11239 void netdev_printk(const char *level, const struct net_device *dev, 11240 const char *format, ...) 11241 { 11242 struct va_format vaf; 11243 va_list args; 11244 11245 va_start(args, format); 11246 11247 vaf.fmt = format; 11248 vaf.va = &args; 11249 11250 __netdev_printk(level, dev, &vaf); 11251 11252 va_end(args); 11253 } 11254 EXPORT_SYMBOL(netdev_printk); 11255 11256 #define define_netdev_printk_level(func, level) \ 11257 void func(const struct net_device *dev, const char *fmt, ...) \ 11258 { \ 11259 struct va_format vaf; \ 11260 va_list args; \ 11261 \ 11262 va_start(args, fmt); \ 11263 \ 11264 vaf.fmt = fmt; \ 11265 vaf.va = &args; \ 11266 \ 11267 __netdev_printk(level, dev, &vaf); \ 11268 \ 11269 va_end(args); \ 11270 } \ 11271 EXPORT_SYMBOL(func); 11272 11273 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11274 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11275 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11276 define_netdev_printk_level(netdev_err, KERN_ERR); 11277 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11278 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11279 define_netdev_printk_level(netdev_info, KERN_INFO); 11280 11281 static void __net_exit netdev_exit(struct net *net) 11282 { 11283 kfree(net->dev_name_head); 11284 kfree(net->dev_index_head); 11285 if (net != &init_net) 11286 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11287 } 11288 11289 static struct pernet_operations __net_initdata netdev_net_ops = { 11290 .init = netdev_init, 11291 .exit = netdev_exit, 11292 }; 11293 11294 static void __net_exit default_device_exit_net(struct net *net) 11295 { 11296 struct net_device *dev, *aux; 11297 /* 11298 * Push all migratable network devices back to the 11299 * initial network namespace 11300 */ 11301 ASSERT_RTNL(); 11302 for_each_netdev_safe(net, dev, aux) { 11303 int err; 11304 char fb_name[IFNAMSIZ]; 11305 11306 /* Ignore unmoveable devices (i.e. loopback) */ 11307 if (dev->features & NETIF_F_NETNS_LOCAL) 11308 continue; 11309 11310 /* Leave virtual devices for the generic cleanup */ 11311 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11312 continue; 11313 11314 /* Push remaining network devices to init_net */ 11315 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11316 if (netdev_name_in_use(&init_net, fb_name)) 11317 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11318 err = dev_change_net_namespace(dev, &init_net, fb_name); 11319 if (err) { 11320 pr_emerg("%s: failed to move %s to init_net: %d\n", 11321 __func__, dev->name, err); 11322 BUG(); 11323 } 11324 } 11325 } 11326 11327 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11328 { 11329 /* At exit all network devices most be removed from a network 11330 * namespace. Do this in the reverse order of registration. 11331 * Do this across as many network namespaces as possible to 11332 * improve batching efficiency. 11333 */ 11334 struct net_device *dev; 11335 struct net *net; 11336 LIST_HEAD(dev_kill_list); 11337 11338 rtnl_lock(); 11339 list_for_each_entry(net, net_list, exit_list) { 11340 default_device_exit_net(net); 11341 cond_resched(); 11342 } 11343 11344 list_for_each_entry(net, net_list, exit_list) { 11345 for_each_netdev_reverse(net, dev) { 11346 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11347 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11348 else 11349 unregister_netdevice_queue(dev, &dev_kill_list); 11350 } 11351 } 11352 unregister_netdevice_many(&dev_kill_list); 11353 rtnl_unlock(); 11354 } 11355 11356 static struct pernet_operations __net_initdata default_device_ops = { 11357 .exit_batch = default_device_exit_batch, 11358 }; 11359 11360 /* 11361 * Initialize the DEV module. At boot time this walks the device list and 11362 * unhooks any devices that fail to initialise (normally hardware not 11363 * present) and leaves us with a valid list of present and active devices. 11364 * 11365 */ 11366 11367 /* 11368 * This is called single threaded during boot, so no need 11369 * to take the rtnl semaphore. 11370 */ 11371 static int __init net_dev_init(void) 11372 { 11373 int i, rc = -ENOMEM; 11374 11375 BUG_ON(!dev_boot_phase); 11376 11377 if (dev_proc_init()) 11378 goto out; 11379 11380 if (netdev_kobject_init()) 11381 goto out; 11382 11383 INIT_LIST_HEAD(&ptype_all); 11384 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11385 INIT_LIST_HEAD(&ptype_base[i]); 11386 11387 if (register_pernet_subsys(&netdev_net_ops)) 11388 goto out; 11389 11390 /* 11391 * Initialise the packet receive queues. 11392 */ 11393 11394 for_each_possible_cpu(i) { 11395 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11396 struct softnet_data *sd = &per_cpu(softnet_data, i); 11397 11398 INIT_WORK(flush, flush_backlog); 11399 11400 skb_queue_head_init(&sd->input_pkt_queue); 11401 skb_queue_head_init(&sd->process_queue); 11402 #ifdef CONFIG_XFRM_OFFLOAD 11403 skb_queue_head_init(&sd->xfrm_backlog); 11404 #endif 11405 INIT_LIST_HEAD(&sd->poll_list); 11406 sd->output_queue_tailp = &sd->output_queue; 11407 #ifdef CONFIG_RPS 11408 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11409 sd->cpu = i; 11410 #endif 11411 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11412 spin_lock_init(&sd->defer_lock); 11413 11414 init_gro_hash(&sd->backlog); 11415 sd->backlog.poll = process_backlog; 11416 sd->backlog.weight = weight_p; 11417 } 11418 11419 dev_boot_phase = 0; 11420 11421 /* The loopback device is special if any other network devices 11422 * is present in a network namespace the loopback device must 11423 * be present. Since we now dynamically allocate and free the 11424 * loopback device ensure this invariant is maintained by 11425 * keeping the loopback device as the first device on the 11426 * list of network devices. Ensuring the loopback devices 11427 * is the first device that appears and the last network device 11428 * that disappears. 11429 */ 11430 if (register_pernet_device(&loopback_net_ops)) 11431 goto out; 11432 11433 if (register_pernet_device(&default_device_ops)) 11434 goto out; 11435 11436 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11437 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11438 11439 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11440 NULL, dev_cpu_dead); 11441 WARN_ON(rc < 0); 11442 rc = 0; 11443 out: 11444 return rc; 11445 } 11446 11447 subsys_initcall(net_dev_init); 11448