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_extack(unsigned long val, 164 struct net_device *dev, 165 struct netlink_ext_ack *extack); 166 static struct napi_struct *napi_by_id(unsigned int napi_id); 167 168 /* 169 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 170 * semaphore. 171 * 172 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 173 * 174 * Writers must hold the rtnl semaphore while they loop through the 175 * dev_base_head list, and hold dev_base_lock for writing when they do the 176 * actual updates. This allows pure readers to access the list even 177 * while a writer is preparing to update it. 178 * 179 * To put it another way, dev_base_lock is held for writing only to 180 * protect against pure readers; the rtnl semaphore provides the 181 * protection against other writers. 182 * 183 * See, for example usages, register_netdevice() and 184 * unregister_netdevice(), which must be called with the rtnl 185 * semaphore held. 186 */ 187 DEFINE_RWLOCK(dev_base_lock); 188 EXPORT_SYMBOL(dev_base_lock); 189 190 static DEFINE_MUTEX(ifalias_mutex); 191 192 /* protects napi_hash addition/deletion and napi_gen_id */ 193 static DEFINE_SPINLOCK(napi_hash_lock); 194 195 static unsigned int napi_gen_id = NR_CPUS; 196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 197 198 static DECLARE_RWSEM(devnet_rename_sem); 199 200 static inline void dev_base_seq_inc(struct net *net) 201 { 202 while (++net->dev_base_seq == 0) 203 ; 204 } 205 206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 207 { 208 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 209 210 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 211 } 212 213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 214 { 215 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 216 } 217 218 static inline void rps_lock_irqsave(struct softnet_data *sd, 219 unsigned long *flags) 220 { 221 if (IS_ENABLED(CONFIG_RPS)) 222 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 223 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 224 local_irq_save(*flags); 225 } 226 227 static inline void rps_lock_irq_disable(struct softnet_data *sd) 228 { 229 if (IS_ENABLED(CONFIG_RPS)) 230 spin_lock_irq(&sd->input_pkt_queue.lock); 231 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 232 local_irq_disable(); 233 } 234 235 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 236 unsigned long *flags) 237 { 238 if (IS_ENABLED(CONFIG_RPS)) 239 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 240 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 241 local_irq_restore(*flags); 242 } 243 244 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 245 { 246 if (IS_ENABLED(CONFIG_RPS)) 247 spin_unlock_irq(&sd->input_pkt_queue.lock); 248 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 249 local_irq_enable(); 250 } 251 252 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 253 const char *name) 254 { 255 struct netdev_name_node *name_node; 256 257 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 258 if (!name_node) 259 return NULL; 260 INIT_HLIST_NODE(&name_node->hlist); 261 name_node->dev = dev; 262 name_node->name = name; 263 return name_node; 264 } 265 266 static struct netdev_name_node * 267 netdev_name_node_head_alloc(struct net_device *dev) 268 { 269 struct netdev_name_node *name_node; 270 271 name_node = netdev_name_node_alloc(dev, dev->name); 272 if (!name_node) 273 return NULL; 274 INIT_LIST_HEAD(&name_node->list); 275 return name_node; 276 } 277 278 static void netdev_name_node_free(struct netdev_name_node *name_node) 279 { 280 kfree(name_node); 281 } 282 283 static void netdev_name_node_add(struct net *net, 284 struct netdev_name_node *name_node) 285 { 286 hlist_add_head_rcu(&name_node->hlist, 287 dev_name_hash(net, name_node->name)); 288 } 289 290 static void netdev_name_node_del(struct netdev_name_node *name_node) 291 { 292 hlist_del_rcu(&name_node->hlist); 293 } 294 295 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 296 const char *name) 297 { 298 struct hlist_head *head = dev_name_hash(net, name); 299 struct netdev_name_node *name_node; 300 301 hlist_for_each_entry(name_node, head, hlist) 302 if (!strcmp(name_node->name, name)) 303 return name_node; 304 return NULL; 305 } 306 307 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 308 const char *name) 309 { 310 struct hlist_head *head = dev_name_hash(net, name); 311 struct netdev_name_node *name_node; 312 313 hlist_for_each_entry_rcu(name_node, head, hlist) 314 if (!strcmp(name_node->name, name)) 315 return name_node; 316 return NULL; 317 } 318 319 bool netdev_name_in_use(struct net *net, const char *name) 320 { 321 return netdev_name_node_lookup(net, name); 322 } 323 EXPORT_SYMBOL(netdev_name_in_use); 324 325 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 326 { 327 struct netdev_name_node *name_node; 328 struct net *net = dev_net(dev); 329 330 name_node = netdev_name_node_lookup(net, name); 331 if (name_node) 332 return -EEXIST; 333 name_node = netdev_name_node_alloc(dev, name); 334 if (!name_node) 335 return -ENOMEM; 336 netdev_name_node_add(net, name_node); 337 /* The node that holds dev->name acts as a head of per-device list. */ 338 list_add_tail(&name_node->list, &dev->name_node->list); 339 340 return 0; 341 } 342 343 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 344 { 345 list_del(&name_node->list); 346 netdev_name_node_del(name_node); 347 kfree(name_node->name); 348 netdev_name_node_free(name_node); 349 } 350 351 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 352 { 353 struct netdev_name_node *name_node; 354 struct net *net = dev_net(dev); 355 356 name_node = netdev_name_node_lookup(net, name); 357 if (!name_node) 358 return -ENOENT; 359 /* lookup might have found our primary name or a name belonging 360 * to another device. 361 */ 362 if (name_node == dev->name_node || name_node->dev != dev) 363 return -EINVAL; 364 365 __netdev_name_node_alt_destroy(name_node); 366 367 return 0; 368 } 369 370 static void netdev_name_node_alt_flush(struct net_device *dev) 371 { 372 struct netdev_name_node *name_node, *tmp; 373 374 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 375 __netdev_name_node_alt_destroy(name_node); 376 } 377 378 /* Device list insertion */ 379 static void list_netdevice(struct net_device *dev) 380 { 381 struct net *net = dev_net(dev); 382 383 ASSERT_RTNL(); 384 385 write_lock(&dev_base_lock); 386 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 387 netdev_name_node_add(net, dev->name_node); 388 hlist_add_head_rcu(&dev->index_hlist, 389 dev_index_hash(net, dev->ifindex)); 390 write_unlock(&dev_base_lock); 391 392 dev_base_seq_inc(net); 393 } 394 395 /* Device list removal 396 * caller must respect a RCU grace period before freeing/reusing dev 397 */ 398 static void unlist_netdevice(struct net_device *dev, bool lock) 399 { 400 ASSERT_RTNL(); 401 402 /* Unlink dev from the device chain */ 403 if (lock) 404 write_lock(&dev_base_lock); 405 list_del_rcu(&dev->dev_list); 406 netdev_name_node_del(dev->name_node); 407 hlist_del_rcu(&dev->index_hlist); 408 if (lock) 409 write_unlock(&dev_base_lock); 410 411 dev_base_seq_inc(dev_net(dev)); 412 } 413 414 /* 415 * Our notifier list 416 */ 417 418 static RAW_NOTIFIER_HEAD(netdev_chain); 419 420 /* 421 * Device drivers call our routines to queue packets here. We empty the 422 * queue in the local softnet handler. 423 */ 424 425 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 426 EXPORT_PER_CPU_SYMBOL(softnet_data); 427 428 #ifdef CONFIG_LOCKDEP 429 /* 430 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 431 * according to dev->type 432 */ 433 static const unsigned short netdev_lock_type[] = { 434 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 435 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 436 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 437 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 438 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 439 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 440 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 441 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 442 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 443 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 444 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 445 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 446 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 447 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 448 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 449 450 static const char *const netdev_lock_name[] = { 451 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 452 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 453 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 454 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 455 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 456 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 457 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 458 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 459 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 460 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 461 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 462 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 463 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 464 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 465 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 466 467 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 468 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 469 470 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 471 { 472 int i; 473 474 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 475 if (netdev_lock_type[i] == dev_type) 476 return i; 477 /* the last key is used by default */ 478 return ARRAY_SIZE(netdev_lock_type) - 1; 479 } 480 481 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 482 unsigned short dev_type) 483 { 484 int i; 485 486 i = netdev_lock_pos(dev_type); 487 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 488 netdev_lock_name[i]); 489 } 490 491 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 492 { 493 int i; 494 495 i = netdev_lock_pos(dev->type); 496 lockdep_set_class_and_name(&dev->addr_list_lock, 497 &netdev_addr_lock_key[i], 498 netdev_lock_name[i]); 499 } 500 #else 501 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 502 unsigned short dev_type) 503 { 504 } 505 506 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 507 { 508 } 509 #endif 510 511 /******************************************************************************* 512 * 513 * Protocol management and registration routines 514 * 515 *******************************************************************************/ 516 517 518 /* 519 * Add a protocol ID to the list. Now that the input handler is 520 * smarter we can dispense with all the messy stuff that used to be 521 * here. 522 * 523 * BEWARE!!! Protocol handlers, mangling input packets, 524 * MUST BE last in hash buckets and checking protocol handlers 525 * MUST start from promiscuous ptype_all chain in net_bh. 526 * It is true now, do not change it. 527 * Explanation follows: if protocol handler, mangling packet, will 528 * be the first on list, it is not able to sense, that packet 529 * is cloned and should be copied-on-write, so that it will 530 * change it and subsequent readers will get broken packet. 531 * --ANK (980803) 532 */ 533 534 static inline struct list_head *ptype_head(const struct packet_type *pt) 535 { 536 if (pt->type == htons(ETH_P_ALL)) 537 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 538 else 539 return pt->dev ? &pt->dev->ptype_specific : 540 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 541 } 542 543 /** 544 * dev_add_pack - add packet handler 545 * @pt: packet type declaration 546 * 547 * Add a protocol handler to the networking stack. The passed &packet_type 548 * is linked into kernel lists and may not be freed until it has been 549 * removed from the kernel lists. 550 * 551 * This call does not sleep therefore it can not 552 * guarantee all CPU's that are in middle of receiving packets 553 * will see the new packet type (until the next received packet). 554 */ 555 556 void dev_add_pack(struct packet_type *pt) 557 { 558 struct list_head *head = ptype_head(pt); 559 560 spin_lock(&ptype_lock); 561 list_add_rcu(&pt->list, head); 562 spin_unlock(&ptype_lock); 563 } 564 EXPORT_SYMBOL(dev_add_pack); 565 566 /** 567 * __dev_remove_pack - remove packet handler 568 * @pt: packet type declaration 569 * 570 * Remove a protocol handler that was previously added to the kernel 571 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 572 * from the kernel lists and can be freed or reused once this function 573 * returns. 574 * 575 * The packet type might still be in use by receivers 576 * and must not be freed until after all the CPU's have gone 577 * through a quiescent state. 578 */ 579 void __dev_remove_pack(struct packet_type *pt) 580 { 581 struct list_head *head = ptype_head(pt); 582 struct packet_type *pt1; 583 584 spin_lock(&ptype_lock); 585 586 list_for_each_entry(pt1, head, list) { 587 if (pt == pt1) { 588 list_del_rcu(&pt->list); 589 goto out; 590 } 591 } 592 593 pr_warn("dev_remove_pack: %p not found\n", pt); 594 out: 595 spin_unlock(&ptype_lock); 596 } 597 EXPORT_SYMBOL(__dev_remove_pack); 598 599 /** 600 * dev_remove_pack - remove packet handler 601 * @pt: packet type declaration 602 * 603 * Remove a protocol handler that was previously added to the kernel 604 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 605 * from the kernel lists and can be freed or reused once this function 606 * returns. 607 * 608 * This call sleeps to guarantee that no CPU is looking at the packet 609 * type after return. 610 */ 611 void dev_remove_pack(struct packet_type *pt) 612 { 613 __dev_remove_pack(pt); 614 615 synchronize_net(); 616 } 617 EXPORT_SYMBOL(dev_remove_pack); 618 619 620 /******************************************************************************* 621 * 622 * Device Interface Subroutines 623 * 624 *******************************************************************************/ 625 626 /** 627 * dev_get_iflink - get 'iflink' value of a interface 628 * @dev: targeted interface 629 * 630 * Indicates the ifindex the interface is linked to. 631 * Physical interfaces have the same 'ifindex' and 'iflink' values. 632 */ 633 634 int dev_get_iflink(const struct net_device *dev) 635 { 636 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 637 return dev->netdev_ops->ndo_get_iflink(dev); 638 639 return dev->ifindex; 640 } 641 EXPORT_SYMBOL(dev_get_iflink); 642 643 /** 644 * dev_fill_metadata_dst - Retrieve tunnel egress information. 645 * @dev: targeted interface 646 * @skb: The packet. 647 * 648 * For better visibility of tunnel traffic OVS needs to retrieve 649 * egress tunnel information for a packet. Following API allows 650 * user to get this info. 651 */ 652 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 653 { 654 struct ip_tunnel_info *info; 655 656 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 657 return -EINVAL; 658 659 info = skb_tunnel_info_unclone(skb); 660 if (!info) 661 return -ENOMEM; 662 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 663 return -EINVAL; 664 665 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 666 } 667 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 668 669 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 670 { 671 int k = stack->num_paths++; 672 673 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 674 return NULL; 675 676 return &stack->path[k]; 677 } 678 679 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 680 struct net_device_path_stack *stack) 681 { 682 const struct net_device *last_dev; 683 struct net_device_path_ctx ctx = { 684 .dev = dev, 685 }; 686 struct net_device_path *path; 687 int ret = 0; 688 689 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 690 stack->num_paths = 0; 691 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 692 last_dev = ctx.dev; 693 path = dev_fwd_path(stack); 694 if (!path) 695 return -1; 696 697 memset(path, 0, sizeof(struct net_device_path)); 698 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 699 if (ret < 0) 700 return -1; 701 702 if (WARN_ON_ONCE(last_dev == ctx.dev)) 703 return -1; 704 } 705 706 if (!ctx.dev) 707 return ret; 708 709 path = dev_fwd_path(stack); 710 if (!path) 711 return -1; 712 path->type = DEV_PATH_ETHERNET; 713 path->dev = ctx.dev; 714 715 return ret; 716 } 717 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 718 719 /** 720 * __dev_get_by_name - find a device by its name 721 * @net: the applicable net namespace 722 * @name: name to find 723 * 724 * Find an interface by name. Must be called under RTNL semaphore 725 * or @dev_base_lock. If the name is found a pointer to the device 726 * is returned. If the name is not found then %NULL is returned. The 727 * reference counters are not incremented so the caller must be 728 * careful with locks. 729 */ 730 731 struct net_device *__dev_get_by_name(struct net *net, const char *name) 732 { 733 struct netdev_name_node *node_name; 734 735 node_name = netdev_name_node_lookup(net, name); 736 return node_name ? node_name->dev : NULL; 737 } 738 EXPORT_SYMBOL(__dev_get_by_name); 739 740 /** 741 * dev_get_by_name_rcu - find a device by its name 742 * @net: the applicable net namespace 743 * @name: name to find 744 * 745 * Find an interface by name. 746 * If the name is found a pointer to the device is returned. 747 * If the name is not found then %NULL is returned. 748 * The reference counters are not incremented so the caller must be 749 * careful with locks. The caller must hold RCU lock. 750 */ 751 752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 753 { 754 struct netdev_name_node *node_name; 755 756 node_name = netdev_name_node_lookup_rcu(net, name); 757 return node_name ? node_name->dev : NULL; 758 } 759 EXPORT_SYMBOL(dev_get_by_name_rcu); 760 761 /** 762 * dev_get_by_name - find a device by its name 763 * @net: the applicable net namespace 764 * @name: name to find 765 * 766 * Find an interface by name. This can be called from any 767 * context and does its own locking. The returned handle has 768 * the usage count incremented and the caller must use dev_put() to 769 * release it when it is no longer needed. %NULL is returned if no 770 * matching device is found. 771 */ 772 773 struct net_device *dev_get_by_name(struct net *net, const char *name) 774 { 775 struct net_device *dev; 776 777 rcu_read_lock(); 778 dev = dev_get_by_name_rcu(net, name); 779 dev_hold(dev); 780 rcu_read_unlock(); 781 return dev; 782 } 783 EXPORT_SYMBOL(dev_get_by_name); 784 785 /** 786 * __dev_get_by_index - find a device by its ifindex 787 * @net: the applicable net namespace 788 * @ifindex: index of device 789 * 790 * Search for an interface by index. Returns %NULL if the device 791 * is not found or a pointer to the device. The device has not 792 * had its reference counter increased so the caller must be careful 793 * about locking. The caller must hold either the RTNL semaphore 794 * or @dev_base_lock. 795 */ 796 797 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 798 { 799 struct net_device *dev; 800 struct hlist_head *head = dev_index_hash(net, ifindex); 801 802 hlist_for_each_entry(dev, head, index_hlist) 803 if (dev->ifindex == ifindex) 804 return dev; 805 806 return NULL; 807 } 808 EXPORT_SYMBOL(__dev_get_by_index); 809 810 /** 811 * dev_get_by_index_rcu - find a device by its ifindex 812 * @net: the applicable net namespace 813 * @ifindex: index of device 814 * 815 * Search for an interface by index. Returns %NULL if the device 816 * is not found or a pointer to the device. The device has not 817 * had its reference counter increased so the caller must be careful 818 * about locking. The caller must hold RCU lock. 819 */ 820 821 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 822 { 823 struct net_device *dev; 824 struct hlist_head *head = dev_index_hash(net, ifindex); 825 826 hlist_for_each_entry_rcu(dev, head, index_hlist) 827 if (dev->ifindex == ifindex) 828 return dev; 829 830 return NULL; 831 } 832 EXPORT_SYMBOL(dev_get_by_index_rcu); 833 834 835 /** 836 * dev_get_by_index - find a device by its ifindex 837 * @net: the applicable net namespace 838 * @ifindex: index of device 839 * 840 * Search for an interface by index. Returns NULL if the device 841 * is not found or a pointer to the device. The device returned has 842 * had a reference added and the pointer is safe until the user calls 843 * dev_put to indicate they have finished with it. 844 */ 845 846 struct net_device *dev_get_by_index(struct net *net, int ifindex) 847 { 848 struct net_device *dev; 849 850 rcu_read_lock(); 851 dev = dev_get_by_index_rcu(net, ifindex); 852 dev_hold(dev); 853 rcu_read_unlock(); 854 return dev; 855 } 856 EXPORT_SYMBOL(dev_get_by_index); 857 858 /** 859 * dev_get_by_napi_id - find a device by napi_id 860 * @napi_id: ID of the NAPI struct 861 * 862 * Search for an interface by NAPI ID. Returns %NULL if the device 863 * is not found or a pointer to the device. The device has not had 864 * its reference counter increased so the caller must be careful 865 * about locking. The caller must hold RCU lock. 866 */ 867 868 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 869 { 870 struct napi_struct *napi; 871 872 WARN_ON_ONCE(!rcu_read_lock_held()); 873 874 if (napi_id < MIN_NAPI_ID) 875 return NULL; 876 877 napi = napi_by_id(napi_id); 878 879 return napi ? napi->dev : NULL; 880 } 881 EXPORT_SYMBOL(dev_get_by_napi_id); 882 883 /** 884 * netdev_get_name - get a netdevice name, knowing its ifindex. 885 * @net: network namespace 886 * @name: a pointer to the buffer where the name will be stored. 887 * @ifindex: the ifindex of the interface to get the name from. 888 */ 889 int netdev_get_name(struct net *net, char *name, int ifindex) 890 { 891 struct net_device *dev; 892 int ret; 893 894 down_read(&devnet_rename_sem); 895 rcu_read_lock(); 896 897 dev = dev_get_by_index_rcu(net, ifindex); 898 if (!dev) { 899 ret = -ENODEV; 900 goto out; 901 } 902 903 strcpy(name, dev->name); 904 905 ret = 0; 906 out: 907 rcu_read_unlock(); 908 up_read(&devnet_rename_sem); 909 return ret; 910 } 911 912 /** 913 * dev_getbyhwaddr_rcu - find a device by its hardware address 914 * @net: the applicable net namespace 915 * @type: media type of device 916 * @ha: hardware address 917 * 918 * Search for an interface by MAC address. Returns NULL if the device 919 * is not found or a pointer to the device. 920 * The caller must hold RCU or RTNL. 921 * The returned device has not had its ref count increased 922 * and the caller must therefore be careful about locking 923 * 924 */ 925 926 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 927 const char *ha) 928 { 929 struct net_device *dev; 930 931 for_each_netdev_rcu(net, dev) 932 if (dev->type == type && 933 !memcmp(dev->dev_addr, ha, dev->addr_len)) 934 return dev; 935 936 return NULL; 937 } 938 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 939 940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 941 { 942 struct net_device *dev, *ret = NULL; 943 944 rcu_read_lock(); 945 for_each_netdev_rcu(net, dev) 946 if (dev->type == type) { 947 dev_hold(dev); 948 ret = dev; 949 break; 950 } 951 rcu_read_unlock(); 952 return ret; 953 } 954 EXPORT_SYMBOL(dev_getfirstbyhwtype); 955 956 /** 957 * __dev_get_by_flags - find any device with given flags 958 * @net: the applicable net namespace 959 * @if_flags: IFF_* values 960 * @mask: bitmask of bits in if_flags to check 961 * 962 * Search for any interface with the given flags. Returns NULL if a device 963 * is not found or a pointer to the device. Must be called inside 964 * rtnl_lock(), and result refcount is unchanged. 965 */ 966 967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 968 unsigned short mask) 969 { 970 struct net_device *dev, *ret; 971 972 ASSERT_RTNL(); 973 974 ret = NULL; 975 for_each_netdev(net, dev) { 976 if (((dev->flags ^ if_flags) & mask) == 0) { 977 ret = dev; 978 break; 979 } 980 } 981 return ret; 982 } 983 EXPORT_SYMBOL(__dev_get_by_flags); 984 985 /** 986 * dev_valid_name - check if name is okay for network device 987 * @name: name string 988 * 989 * Network device names need to be valid file names to 990 * allow sysfs to work. We also disallow any kind of 991 * whitespace. 992 */ 993 bool dev_valid_name(const char *name) 994 { 995 if (*name == '\0') 996 return false; 997 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 998 return false; 999 if (!strcmp(name, ".") || !strcmp(name, "..")) 1000 return false; 1001 1002 while (*name) { 1003 if (*name == '/' || *name == ':' || isspace(*name)) 1004 return false; 1005 name++; 1006 } 1007 return true; 1008 } 1009 EXPORT_SYMBOL(dev_valid_name); 1010 1011 /** 1012 * __dev_alloc_name - allocate a name for a device 1013 * @net: network namespace to allocate the device name in 1014 * @name: name format string 1015 * @buf: scratch buffer and result name string 1016 * 1017 * Passed a format string - eg "lt%d" it will try and find a suitable 1018 * id. It scans list of devices to build up a free map, then chooses 1019 * the first empty slot. The caller must hold the dev_base or rtnl lock 1020 * while allocating the name and adding the device in order to avoid 1021 * duplicates. 1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1023 * Returns the number of the unit assigned or a negative errno code. 1024 */ 1025 1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1027 { 1028 int i = 0; 1029 const char *p; 1030 const int max_netdevices = 8*PAGE_SIZE; 1031 unsigned long *inuse; 1032 struct net_device *d; 1033 1034 if (!dev_valid_name(name)) 1035 return -EINVAL; 1036 1037 p = strchr(name, '%'); 1038 if (p) { 1039 /* 1040 * Verify the string as this thing may have come from 1041 * the user. There must be either one "%d" and no other "%" 1042 * characters. 1043 */ 1044 if (p[1] != 'd' || strchr(p + 2, '%')) 1045 return -EINVAL; 1046 1047 /* Use one page as a bit array of possible slots */ 1048 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1049 if (!inuse) 1050 return -ENOMEM; 1051 1052 for_each_netdev(net, d) { 1053 struct netdev_name_node *name_node; 1054 list_for_each_entry(name_node, &d->name_node->list, list) { 1055 if (!sscanf(name_node->name, name, &i)) 1056 continue; 1057 if (i < 0 || i >= max_netdevices) 1058 continue; 1059 1060 /* avoid cases where sscanf is not exact inverse of printf */ 1061 snprintf(buf, IFNAMSIZ, name, i); 1062 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1063 __set_bit(i, inuse); 1064 } 1065 if (!sscanf(d->name, name, &i)) 1066 continue; 1067 if (i < 0 || i >= max_netdevices) 1068 continue; 1069 1070 /* avoid cases where sscanf is not exact inverse of printf */ 1071 snprintf(buf, IFNAMSIZ, name, i); 1072 if (!strncmp(buf, d->name, IFNAMSIZ)) 1073 __set_bit(i, inuse); 1074 } 1075 1076 i = find_first_zero_bit(inuse, max_netdevices); 1077 free_page((unsigned long) inuse); 1078 } 1079 1080 snprintf(buf, IFNAMSIZ, name, i); 1081 if (!netdev_name_in_use(net, buf)) 1082 return i; 1083 1084 /* It is possible to run out of possible slots 1085 * when the name is long and there isn't enough space left 1086 * for the digits, or if all bits are used. 1087 */ 1088 return -ENFILE; 1089 } 1090 1091 static int dev_alloc_name_ns(struct net *net, 1092 struct net_device *dev, 1093 const char *name) 1094 { 1095 char buf[IFNAMSIZ]; 1096 int ret; 1097 1098 BUG_ON(!net); 1099 ret = __dev_alloc_name(net, name, buf); 1100 if (ret >= 0) 1101 strscpy(dev->name, buf, IFNAMSIZ); 1102 return ret; 1103 } 1104 1105 /** 1106 * dev_alloc_name - allocate a name for a device 1107 * @dev: device 1108 * @name: name format string 1109 * 1110 * Passed a format string - eg "lt%d" it will try and find a suitable 1111 * id. It scans list of devices to build up a free map, then chooses 1112 * the first empty slot. The caller must hold the dev_base or rtnl lock 1113 * while allocating the name and adding the device in order to avoid 1114 * duplicates. 1115 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1116 * Returns the number of the unit assigned or a negative errno code. 1117 */ 1118 1119 int dev_alloc_name(struct net_device *dev, const char *name) 1120 { 1121 return dev_alloc_name_ns(dev_net(dev), dev, name); 1122 } 1123 EXPORT_SYMBOL(dev_alloc_name); 1124 1125 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1126 const char *name) 1127 { 1128 BUG_ON(!net); 1129 1130 if (!dev_valid_name(name)) 1131 return -EINVAL; 1132 1133 if (strchr(name, '%')) 1134 return dev_alloc_name_ns(net, dev, name); 1135 else if (netdev_name_in_use(net, name)) 1136 return -EEXIST; 1137 else if (dev->name != name) 1138 strscpy(dev->name, name, IFNAMSIZ); 1139 1140 return 0; 1141 } 1142 1143 /** 1144 * dev_change_name - change name of a device 1145 * @dev: device 1146 * @newname: name (or format string) must be at least IFNAMSIZ 1147 * 1148 * Change name of a device, can pass format strings "eth%d". 1149 * for wildcarding. 1150 */ 1151 int dev_change_name(struct net_device *dev, const char *newname) 1152 { 1153 unsigned char old_assign_type; 1154 char oldname[IFNAMSIZ]; 1155 int err = 0; 1156 int ret; 1157 struct net *net; 1158 1159 ASSERT_RTNL(); 1160 BUG_ON(!dev_net(dev)); 1161 1162 net = dev_net(dev); 1163 1164 down_write(&devnet_rename_sem); 1165 1166 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1167 up_write(&devnet_rename_sem); 1168 return 0; 1169 } 1170 1171 memcpy(oldname, dev->name, IFNAMSIZ); 1172 1173 err = dev_get_valid_name(net, dev, newname); 1174 if (err < 0) { 1175 up_write(&devnet_rename_sem); 1176 return err; 1177 } 1178 1179 if (oldname[0] && !strchr(oldname, '%')) 1180 netdev_info(dev, "renamed from %s%s\n", oldname, 1181 dev->flags & IFF_UP ? " (while UP)" : ""); 1182 1183 old_assign_type = dev->name_assign_type; 1184 dev->name_assign_type = NET_NAME_RENAMED; 1185 1186 rollback: 1187 ret = device_rename(&dev->dev, dev->name); 1188 if (ret) { 1189 memcpy(dev->name, oldname, IFNAMSIZ); 1190 dev->name_assign_type = old_assign_type; 1191 up_write(&devnet_rename_sem); 1192 return ret; 1193 } 1194 1195 up_write(&devnet_rename_sem); 1196 1197 netdev_adjacent_rename_links(dev, oldname); 1198 1199 write_lock(&dev_base_lock); 1200 netdev_name_node_del(dev->name_node); 1201 write_unlock(&dev_base_lock); 1202 1203 synchronize_rcu(); 1204 1205 write_lock(&dev_base_lock); 1206 netdev_name_node_add(net, dev->name_node); 1207 write_unlock(&dev_base_lock); 1208 1209 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1210 ret = notifier_to_errno(ret); 1211 1212 if (ret) { 1213 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1214 if (err >= 0) { 1215 err = ret; 1216 down_write(&devnet_rename_sem); 1217 memcpy(dev->name, oldname, IFNAMSIZ); 1218 memcpy(oldname, newname, IFNAMSIZ); 1219 dev->name_assign_type = old_assign_type; 1220 old_assign_type = NET_NAME_RENAMED; 1221 goto rollback; 1222 } else { 1223 netdev_err(dev, "name change rollback failed: %d\n", 1224 ret); 1225 } 1226 } 1227 1228 return err; 1229 } 1230 1231 /** 1232 * dev_set_alias - change ifalias of a device 1233 * @dev: device 1234 * @alias: name up to IFALIASZ 1235 * @len: limit of bytes to copy from info 1236 * 1237 * Set ifalias for a device, 1238 */ 1239 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1240 { 1241 struct dev_ifalias *new_alias = NULL; 1242 1243 if (len >= IFALIASZ) 1244 return -EINVAL; 1245 1246 if (len) { 1247 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1248 if (!new_alias) 1249 return -ENOMEM; 1250 1251 memcpy(new_alias->ifalias, alias, len); 1252 new_alias->ifalias[len] = 0; 1253 } 1254 1255 mutex_lock(&ifalias_mutex); 1256 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1257 mutex_is_locked(&ifalias_mutex)); 1258 mutex_unlock(&ifalias_mutex); 1259 1260 if (new_alias) 1261 kfree_rcu(new_alias, rcuhead); 1262 1263 return len; 1264 } 1265 EXPORT_SYMBOL(dev_set_alias); 1266 1267 /** 1268 * dev_get_alias - get ifalias of a device 1269 * @dev: device 1270 * @name: buffer to store name of ifalias 1271 * @len: size of buffer 1272 * 1273 * get ifalias for a device. Caller must make sure dev cannot go 1274 * away, e.g. rcu read lock or own a reference count to device. 1275 */ 1276 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1277 { 1278 const struct dev_ifalias *alias; 1279 int ret = 0; 1280 1281 rcu_read_lock(); 1282 alias = rcu_dereference(dev->ifalias); 1283 if (alias) 1284 ret = snprintf(name, len, "%s", alias->ifalias); 1285 rcu_read_unlock(); 1286 1287 return ret; 1288 } 1289 1290 /** 1291 * netdev_features_change - device changes features 1292 * @dev: device to cause notification 1293 * 1294 * Called to indicate a device has changed features. 1295 */ 1296 void netdev_features_change(struct net_device *dev) 1297 { 1298 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1299 } 1300 EXPORT_SYMBOL(netdev_features_change); 1301 1302 /** 1303 * netdev_state_change - device changes state 1304 * @dev: device to cause notification 1305 * 1306 * Called to indicate a device has changed state. This function calls 1307 * the notifier chains for netdev_chain and sends a NEWLINK message 1308 * to the routing socket. 1309 */ 1310 void netdev_state_change(struct net_device *dev) 1311 { 1312 if (dev->flags & IFF_UP) { 1313 struct netdev_notifier_change_info change_info = { 1314 .info.dev = dev, 1315 }; 1316 1317 call_netdevice_notifiers_info(NETDEV_CHANGE, 1318 &change_info.info); 1319 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1320 } 1321 } 1322 EXPORT_SYMBOL(netdev_state_change); 1323 1324 /** 1325 * __netdev_notify_peers - notify network peers about existence of @dev, 1326 * to be called when rtnl lock is already held. 1327 * @dev: network device 1328 * 1329 * Generate traffic such that interested network peers are aware of 1330 * @dev, such as by generating a gratuitous ARP. This may be used when 1331 * a device wants to inform the rest of the network about some sort of 1332 * reconfiguration such as a failover event or virtual machine 1333 * migration. 1334 */ 1335 void __netdev_notify_peers(struct net_device *dev) 1336 { 1337 ASSERT_RTNL(); 1338 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1339 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1340 } 1341 EXPORT_SYMBOL(__netdev_notify_peers); 1342 1343 /** 1344 * netdev_notify_peers - notify network peers about existence of @dev 1345 * @dev: network device 1346 * 1347 * Generate traffic such that interested network peers are aware of 1348 * @dev, such as by generating a gratuitous ARP. This may be used when 1349 * a device wants to inform the rest of the network about some sort of 1350 * reconfiguration such as a failover event or virtual machine 1351 * migration. 1352 */ 1353 void netdev_notify_peers(struct net_device *dev) 1354 { 1355 rtnl_lock(); 1356 __netdev_notify_peers(dev); 1357 rtnl_unlock(); 1358 } 1359 EXPORT_SYMBOL(netdev_notify_peers); 1360 1361 static int napi_threaded_poll(void *data); 1362 1363 static int napi_kthread_create(struct napi_struct *n) 1364 { 1365 int err = 0; 1366 1367 /* Create and wake up the kthread once to put it in 1368 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1369 * warning and work with loadavg. 1370 */ 1371 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1372 n->dev->name, n->napi_id); 1373 if (IS_ERR(n->thread)) { 1374 err = PTR_ERR(n->thread); 1375 pr_err("kthread_run failed with err %d\n", err); 1376 n->thread = NULL; 1377 } 1378 1379 return err; 1380 } 1381 1382 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1383 { 1384 const struct net_device_ops *ops = dev->netdev_ops; 1385 int ret; 1386 1387 ASSERT_RTNL(); 1388 dev_addr_check(dev); 1389 1390 if (!netif_device_present(dev)) { 1391 /* may be detached because parent is runtime-suspended */ 1392 if (dev->dev.parent) 1393 pm_runtime_resume(dev->dev.parent); 1394 if (!netif_device_present(dev)) 1395 return -ENODEV; 1396 } 1397 1398 /* Block netpoll from trying to do any rx path servicing. 1399 * If we don't do this there is a chance ndo_poll_controller 1400 * or ndo_poll may be running while we open the device 1401 */ 1402 netpoll_poll_disable(dev); 1403 1404 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1405 ret = notifier_to_errno(ret); 1406 if (ret) 1407 return ret; 1408 1409 set_bit(__LINK_STATE_START, &dev->state); 1410 1411 if (ops->ndo_validate_addr) 1412 ret = ops->ndo_validate_addr(dev); 1413 1414 if (!ret && ops->ndo_open) 1415 ret = ops->ndo_open(dev); 1416 1417 netpoll_poll_enable(dev); 1418 1419 if (ret) 1420 clear_bit(__LINK_STATE_START, &dev->state); 1421 else { 1422 dev->flags |= IFF_UP; 1423 dev_set_rx_mode(dev); 1424 dev_activate(dev); 1425 add_device_randomness(dev->dev_addr, dev->addr_len); 1426 } 1427 1428 return ret; 1429 } 1430 1431 /** 1432 * dev_open - prepare an interface for use. 1433 * @dev: device to open 1434 * @extack: netlink extended ack 1435 * 1436 * Takes a device from down to up state. The device's private open 1437 * function is invoked and then the multicast lists are loaded. Finally 1438 * the device is moved into the up state and a %NETDEV_UP message is 1439 * sent to the netdev notifier chain. 1440 * 1441 * Calling this function on an active interface is a nop. On a failure 1442 * a negative errno code is returned. 1443 */ 1444 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1445 { 1446 int ret; 1447 1448 if (dev->flags & IFF_UP) 1449 return 0; 1450 1451 ret = __dev_open(dev, extack); 1452 if (ret < 0) 1453 return ret; 1454 1455 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1456 call_netdevice_notifiers(NETDEV_UP, dev); 1457 1458 return ret; 1459 } 1460 EXPORT_SYMBOL(dev_open); 1461 1462 static void __dev_close_many(struct list_head *head) 1463 { 1464 struct net_device *dev; 1465 1466 ASSERT_RTNL(); 1467 might_sleep(); 1468 1469 list_for_each_entry(dev, head, close_list) { 1470 /* Temporarily disable netpoll until the interface is down */ 1471 netpoll_poll_disable(dev); 1472 1473 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1474 1475 clear_bit(__LINK_STATE_START, &dev->state); 1476 1477 /* Synchronize to scheduled poll. We cannot touch poll list, it 1478 * can be even on different cpu. So just clear netif_running(). 1479 * 1480 * dev->stop() will invoke napi_disable() on all of it's 1481 * napi_struct instances on this device. 1482 */ 1483 smp_mb__after_atomic(); /* Commit netif_running(). */ 1484 } 1485 1486 dev_deactivate_many(head); 1487 1488 list_for_each_entry(dev, head, close_list) { 1489 const struct net_device_ops *ops = dev->netdev_ops; 1490 1491 /* 1492 * Call the device specific close. This cannot fail. 1493 * Only if device is UP 1494 * 1495 * We allow it to be called even after a DETACH hot-plug 1496 * event. 1497 */ 1498 if (ops->ndo_stop) 1499 ops->ndo_stop(dev); 1500 1501 dev->flags &= ~IFF_UP; 1502 netpoll_poll_enable(dev); 1503 } 1504 } 1505 1506 static void __dev_close(struct net_device *dev) 1507 { 1508 LIST_HEAD(single); 1509 1510 list_add(&dev->close_list, &single); 1511 __dev_close_many(&single); 1512 list_del(&single); 1513 } 1514 1515 void dev_close_many(struct list_head *head, bool unlink) 1516 { 1517 struct net_device *dev, *tmp; 1518 1519 /* Remove the devices that don't need to be closed */ 1520 list_for_each_entry_safe(dev, tmp, head, close_list) 1521 if (!(dev->flags & IFF_UP)) 1522 list_del_init(&dev->close_list); 1523 1524 __dev_close_many(head); 1525 1526 list_for_each_entry_safe(dev, tmp, head, close_list) { 1527 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1528 call_netdevice_notifiers(NETDEV_DOWN, dev); 1529 if (unlink) 1530 list_del_init(&dev->close_list); 1531 } 1532 } 1533 EXPORT_SYMBOL(dev_close_many); 1534 1535 /** 1536 * dev_close - shutdown an interface. 1537 * @dev: device to shutdown 1538 * 1539 * This function moves an active device into down state. A 1540 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1541 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1542 * chain. 1543 */ 1544 void dev_close(struct net_device *dev) 1545 { 1546 if (dev->flags & IFF_UP) { 1547 LIST_HEAD(single); 1548 1549 list_add(&dev->close_list, &single); 1550 dev_close_many(&single, true); 1551 list_del(&single); 1552 } 1553 } 1554 EXPORT_SYMBOL(dev_close); 1555 1556 1557 /** 1558 * dev_disable_lro - disable Large Receive Offload on a device 1559 * @dev: device 1560 * 1561 * Disable Large Receive Offload (LRO) on a net device. Must be 1562 * called under RTNL. This is needed if received packets may be 1563 * forwarded to another interface. 1564 */ 1565 void dev_disable_lro(struct net_device *dev) 1566 { 1567 struct net_device *lower_dev; 1568 struct list_head *iter; 1569 1570 dev->wanted_features &= ~NETIF_F_LRO; 1571 netdev_update_features(dev); 1572 1573 if (unlikely(dev->features & NETIF_F_LRO)) 1574 netdev_WARN(dev, "failed to disable LRO!\n"); 1575 1576 netdev_for_each_lower_dev(dev, lower_dev, iter) 1577 dev_disable_lro(lower_dev); 1578 } 1579 EXPORT_SYMBOL(dev_disable_lro); 1580 1581 /** 1582 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1583 * @dev: device 1584 * 1585 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1586 * called under RTNL. This is needed if Generic XDP is installed on 1587 * the device. 1588 */ 1589 static void dev_disable_gro_hw(struct net_device *dev) 1590 { 1591 dev->wanted_features &= ~NETIF_F_GRO_HW; 1592 netdev_update_features(dev); 1593 1594 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1595 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1596 } 1597 1598 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1599 { 1600 #define N(val) \ 1601 case NETDEV_##val: \ 1602 return "NETDEV_" __stringify(val); 1603 switch (cmd) { 1604 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1605 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1606 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1607 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1608 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1609 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1610 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1611 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1612 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1613 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1614 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1615 N(XDP_FEAT_CHANGE) 1616 } 1617 #undef N 1618 return "UNKNOWN_NETDEV_EVENT"; 1619 } 1620 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1621 1622 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1623 struct net_device *dev) 1624 { 1625 struct netdev_notifier_info info = { 1626 .dev = dev, 1627 }; 1628 1629 return nb->notifier_call(nb, val, &info); 1630 } 1631 1632 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1633 struct net_device *dev) 1634 { 1635 int err; 1636 1637 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1638 err = notifier_to_errno(err); 1639 if (err) 1640 return err; 1641 1642 if (!(dev->flags & IFF_UP)) 1643 return 0; 1644 1645 call_netdevice_notifier(nb, NETDEV_UP, dev); 1646 return 0; 1647 } 1648 1649 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1650 struct net_device *dev) 1651 { 1652 if (dev->flags & IFF_UP) { 1653 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1654 dev); 1655 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1656 } 1657 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1658 } 1659 1660 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1661 struct net *net) 1662 { 1663 struct net_device *dev; 1664 int err; 1665 1666 for_each_netdev(net, dev) { 1667 err = call_netdevice_register_notifiers(nb, dev); 1668 if (err) 1669 goto rollback; 1670 } 1671 return 0; 1672 1673 rollback: 1674 for_each_netdev_continue_reverse(net, dev) 1675 call_netdevice_unregister_notifiers(nb, dev); 1676 return err; 1677 } 1678 1679 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1680 struct net *net) 1681 { 1682 struct net_device *dev; 1683 1684 for_each_netdev(net, dev) 1685 call_netdevice_unregister_notifiers(nb, dev); 1686 } 1687 1688 static int dev_boot_phase = 1; 1689 1690 /** 1691 * register_netdevice_notifier - register a network notifier block 1692 * @nb: notifier 1693 * 1694 * Register a notifier to be called when network device events occur. 1695 * The notifier passed is linked into the kernel structures and must 1696 * not be reused until it has been unregistered. A negative errno code 1697 * is returned on a failure. 1698 * 1699 * When registered all registration and up events are replayed 1700 * to the new notifier to allow device to have a race free 1701 * view of the network device list. 1702 */ 1703 1704 int register_netdevice_notifier(struct notifier_block *nb) 1705 { 1706 struct net *net; 1707 int err; 1708 1709 /* Close race with setup_net() and cleanup_net() */ 1710 down_write(&pernet_ops_rwsem); 1711 rtnl_lock(); 1712 err = raw_notifier_chain_register(&netdev_chain, nb); 1713 if (err) 1714 goto unlock; 1715 if (dev_boot_phase) 1716 goto unlock; 1717 for_each_net(net) { 1718 err = call_netdevice_register_net_notifiers(nb, net); 1719 if (err) 1720 goto rollback; 1721 } 1722 1723 unlock: 1724 rtnl_unlock(); 1725 up_write(&pernet_ops_rwsem); 1726 return err; 1727 1728 rollback: 1729 for_each_net_continue_reverse(net) 1730 call_netdevice_unregister_net_notifiers(nb, net); 1731 1732 raw_notifier_chain_unregister(&netdev_chain, nb); 1733 goto unlock; 1734 } 1735 EXPORT_SYMBOL(register_netdevice_notifier); 1736 1737 /** 1738 * unregister_netdevice_notifier - unregister a network notifier block 1739 * @nb: notifier 1740 * 1741 * Unregister a notifier previously registered by 1742 * register_netdevice_notifier(). The notifier is unlinked into the 1743 * kernel structures and may then be reused. A negative errno code 1744 * is returned on a failure. 1745 * 1746 * After unregistering unregister and down device events are synthesized 1747 * for all devices on the device list to the removed notifier to remove 1748 * the need for special case cleanup code. 1749 */ 1750 1751 int unregister_netdevice_notifier(struct notifier_block *nb) 1752 { 1753 struct net *net; 1754 int err; 1755 1756 /* Close race with setup_net() and cleanup_net() */ 1757 down_write(&pernet_ops_rwsem); 1758 rtnl_lock(); 1759 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1760 if (err) 1761 goto unlock; 1762 1763 for_each_net(net) 1764 call_netdevice_unregister_net_notifiers(nb, net); 1765 1766 unlock: 1767 rtnl_unlock(); 1768 up_write(&pernet_ops_rwsem); 1769 return err; 1770 } 1771 EXPORT_SYMBOL(unregister_netdevice_notifier); 1772 1773 static int __register_netdevice_notifier_net(struct net *net, 1774 struct notifier_block *nb, 1775 bool ignore_call_fail) 1776 { 1777 int err; 1778 1779 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1780 if (err) 1781 return err; 1782 if (dev_boot_phase) 1783 return 0; 1784 1785 err = call_netdevice_register_net_notifiers(nb, net); 1786 if (err && !ignore_call_fail) 1787 goto chain_unregister; 1788 1789 return 0; 1790 1791 chain_unregister: 1792 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1793 return err; 1794 } 1795 1796 static int __unregister_netdevice_notifier_net(struct net *net, 1797 struct notifier_block *nb) 1798 { 1799 int err; 1800 1801 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1802 if (err) 1803 return err; 1804 1805 call_netdevice_unregister_net_notifiers(nb, net); 1806 return 0; 1807 } 1808 1809 /** 1810 * register_netdevice_notifier_net - register a per-netns network notifier block 1811 * @net: network namespace 1812 * @nb: notifier 1813 * 1814 * Register a notifier to be called when network device events occur. 1815 * The notifier passed is linked into the kernel structures and must 1816 * not be reused until it has been unregistered. A negative errno code 1817 * is returned on a failure. 1818 * 1819 * When registered all registration and up events are replayed 1820 * to the new notifier to allow device to have a race free 1821 * view of the network device list. 1822 */ 1823 1824 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1825 { 1826 int err; 1827 1828 rtnl_lock(); 1829 err = __register_netdevice_notifier_net(net, nb, false); 1830 rtnl_unlock(); 1831 return err; 1832 } 1833 EXPORT_SYMBOL(register_netdevice_notifier_net); 1834 1835 /** 1836 * unregister_netdevice_notifier_net - unregister a per-netns 1837 * network notifier block 1838 * @net: network namespace 1839 * @nb: notifier 1840 * 1841 * Unregister a notifier previously registered by 1842 * register_netdevice_notifier_net(). The notifier is unlinked from the 1843 * kernel structures and may then be reused. A negative errno code 1844 * is returned on a failure. 1845 * 1846 * After unregistering unregister and down device events are synthesized 1847 * for all devices on the device list to the removed notifier to remove 1848 * the need for special case cleanup code. 1849 */ 1850 1851 int unregister_netdevice_notifier_net(struct net *net, 1852 struct notifier_block *nb) 1853 { 1854 int err; 1855 1856 rtnl_lock(); 1857 err = __unregister_netdevice_notifier_net(net, nb); 1858 rtnl_unlock(); 1859 return err; 1860 } 1861 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1862 1863 static void __move_netdevice_notifier_net(struct net *src_net, 1864 struct net *dst_net, 1865 struct notifier_block *nb) 1866 { 1867 __unregister_netdevice_notifier_net(src_net, nb); 1868 __register_netdevice_notifier_net(dst_net, nb, true); 1869 } 1870 1871 int register_netdevice_notifier_dev_net(struct net_device *dev, 1872 struct notifier_block *nb, 1873 struct netdev_net_notifier *nn) 1874 { 1875 int err; 1876 1877 rtnl_lock(); 1878 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1879 if (!err) { 1880 nn->nb = nb; 1881 list_add(&nn->list, &dev->net_notifier_list); 1882 } 1883 rtnl_unlock(); 1884 return err; 1885 } 1886 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1887 1888 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1889 struct notifier_block *nb, 1890 struct netdev_net_notifier *nn) 1891 { 1892 int err; 1893 1894 rtnl_lock(); 1895 list_del(&nn->list); 1896 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1897 rtnl_unlock(); 1898 return err; 1899 } 1900 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1901 1902 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1903 struct net *net) 1904 { 1905 struct netdev_net_notifier *nn; 1906 1907 list_for_each_entry(nn, &dev->net_notifier_list, list) 1908 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1909 } 1910 1911 /** 1912 * call_netdevice_notifiers_info - call all network notifier blocks 1913 * @val: value passed unmodified to notifier function 1914 * @info: notifier information data 1915 * 1916 * Call all network notifier blocks. Parameters and return value 1917 * are as for raw_notifier_call_chain(). 1918 */ 1919 1920 int call_netdevice_notifiers_info(unsigned long val, 1921 struct netdev_notifier_info *info) 1922 { 1923 struct net *net = dev_net(info->dev); 1924 int ret; 1925 1926 ASSERT_RTNL(); 1927 1928 /* Run per-netns notifier block chain first, then run the global one. 1929 * Hopefully, one day, the global one is going to be removed after 1930 * all notifier block registrators get converted to be per-netns. 1931 */ 1932 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1933 if (ret & NOTIFY_STOP_MASK) 1934 return ret; 1935 return raw_notifier_call_chain(&netdev_chain, val, info); 1936 } 1937 1938 /** 1939 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1940 * for and rollback on error 1941 * @val_up: value passed unmodified to notifier function 1942 * @val_down: value passed unmodified to the notifier function when 1943 * recovering from an error on @val_up 1944 * @info: notifier information data 1945 * 1946 * Call all per-netns network notifier blocks, but not notifier blocks on 1947 * the global notifier chain. Parameters and return value are as for 1948 * raw_notifier_call_chain_robust(). 1949 */ 1950 1951 static int 1952 call_netdevice_notifiers_info_robust(unsigned long val_up, 1953 unsigned long val_down, 1954 struct netdev_notifier_info *info) 1955 { 1956 struct net *net = dev_net(info->dev); 1957 1958 ASSERT_RTNL(); 1959 1960 return raw_notifier_call_chain_robust(&net->netdev_chain, 1961 val_up, val_down, info); 1962 } 1963 1964 static int call_netdevice_notifiers_extack(unsigned long val, 1965 struct net_device *dev, 1966 struct netlink_ext_ack *extack) 1967 { 1968 struct netdev_notifier_info info = { 1969 .dev = dev, 1970 .extack = extack, 1971 }; 1972 1973 return call_netdevice_notifiers_info(val, &info); 1974 } 1975 1976 /** 1977 * call_netdevice_notifiers - call all network notifier blocks 1978 * @val: value passed unmodified to notifier function 1979 * @dev: net_device pointer passed unmodified to notifier function 1980 * 1981 * Call all network notifier blocks. Parameters and return value 1982 * are as for raw_notifier_call_chain(). 1983 */ 1984 1985 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1986 { 1987 return call_netdevice_notifiers_extack(val, dev, NULL); 1988 } 1989 EXPORT_SYMBOL(call_netdevice_notifiers); 1990 1991 /** 1992 * call_netdevice_notifiers_mtu - call all network notifier blocks 1993 * @val: value passed unmodified to notifier function 1994 * @dev: net_device pointer passed unmodified to notifier function 1995 * @arg: additional u32 argument passed to the notifier function 1996 * 1997 * Call all network notifier blocks. Parameters and return value 1998 * are as for raw_notifier_call_chain(). 1999 */ 2000 static int call_netdevice_notifiers_mtu(unsigned long val, 2001 struct net_device *dev, u32 arg) 2002 { 2003 struct netdev_notifier_info_ext info = { 2004 .info.dev = dev, 2005 .ext.mtu = arg, 2006 }; 2007 2008 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2009 2010 return call_netdevice_notifiers_info(val, &info.info); 2011 } 2012 2013 #ifdef CONFIG_NET_INGRESS 2014 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2015 2016 void net_inc_ingress_queue(void) 2017 { 2018 static_branch_inc(&ingress_needed_key); 2019 } 2020 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2021 2022 void net_dec_ingress_queue(void) 2023 { 2024 static_branch_dec(&ingress_needed_key); 2025 } 2026 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2027 #endif 2028 2029 #ifdef CONFIG_NET_EGRESS 2030 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2031 2032 void net_inc_egress_queue(void) 2033 { 2034 static_branch_inc(&egress_needed_key); 2035 } 2036 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2037 2038 void net_dec_egress_queue(void) 2039 { 2040 static_branch_dec(&egress_needed_key); 2041 } 2042 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2043 #endif 2044 2045 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2046 EXPORT_SYMBOL(netstamp_needed_key); 2047 #ifdef CONFIG_JUMP_LABEL 2048 static atomic_t netstamp_needed_deferred; 2049 static atomic_t netstamp_wanted; 2050 static void netstamp_clear(struct work_struct *work) 2051 { 2052 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2053 int wanted; 2054 2055 wanted = atomic_add_return(deferred, &netstamp_wanted); 2056 if (wanted > 0) 2057 static_branch_enable(&netstamp_needed_key); 2058 else 2059 static_branch_disable(&netstamp_needed_key); 2060 } 2061 static DECLARE_WORK(netstamp_work, netstamp_clear); 2062 #endif 2063 2064 void net_enable_timestamp(void) 2065 { 2066 #ifdef CONFIG_JUMP_LABEL 2067 int wanted = atomic_read(&netstamp_wanted); 2068 2069 while (wanted > 0) { 2070 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2071 return; 2072 } 2073 atomic_inc(&netstamp_needed_deferred); 2074 schedule_work(&netstamp_work); 2075 #else 2076 static_branch_inc(&netstamp_needed_key); 2077 #endif 2078 } 2079 EXPORT_SYMBOL(net_enable_timestamp); 2080 2081 void net_disable_timestamp(void) 2082 { 2083 #ifdef CONFIG_JUMP_LABEL 2084 int wanted = atomic_read(&netstamp_wanted); 2085 2086 while (wanted > 1) { 2087 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2088 return; 2089 } 2090 atomic_dec(&netstamp_needed_deferred); 2091 schedule_work(&netstamp_work); 2092 #else 2093 static_branch_dec(&netstamp_needed_key); 2094 #endif 2095 } 2096 EXPORT_SYMBOL(net_disable_timestamp); 2097 2098 static inline void net_timestamp_set(struct sk_buff *skb) 2099 { 2100 skb->tstamp = 0; 2101 skb->mono_delivery_time = 0; 2102 if (static_branch_unlikely(&netstamp_needed_key)) 2103 skb->tstamp = ktime_get_real(); 2104 } 2105 2106 #define net_timestamp_check(COND, SKB) \ 2107 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2108 if ((COND) && !(SKB)->tstamp) \ 2109 (SKB)->tstamp = ktime_get_real(); \ 2110 } \ 2111 2112 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2113 { 2114 return __is_skb_forwardable(dev, skb, true); 2115 } 2116 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2117 2118 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2119 bool check_mtu) 2120 { 2121 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2122 2123 if (likely(!ret)) { 2124 skb->protocol = eth_type_trans(skb, dev); 2125 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2126 } 2127 2128 return ret; 2129 } 2130 2131 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2132 { 2133 return __dev_forward_skb2(dev, skb, true); 2134 } 2135 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2136 2137 /** 2138 * dev_forward_skb - loopback an skb to another netif 2139 * 2140 * @dev: destination network device 2141 * @skb: buffer to forward 2142 * 2143 * return values: 2144 * NET_RX_SUCCESS (no congestion) 2145 * NET_RX_DROP (packet was dropped, but freed) 2146 * 2147 * dev_forward_skb can be used for injecting an skb from the 2148 * start_xmit function of one device into the receive queue 2149 * of another device. 2150 * 2151 * The receiving device may be in another namespace, so 2152 * we have to clear all information in the skb that could 2153 * impact namespace isolation. 2154 */ 2155 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2156 { 2157 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2158 } 2159 EXPORT_SYMBOL_GPL(dev_forward_skb); 2160 2161 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2162 { 2163 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2164 } 2165 2166 static inline int deliver_skb(struct sk_buff *skb, 2167 struct packet_type *pt_prev, 2168 struct net_device *orig_dev) 2169 { 2170 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2171 return -ENOMEM; 2172 refcount_inc(&skb->users); 2173 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2174 } 2175 2176 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2177 struct packet_type **pt, 2178 struct net_device *orig_dev, 2179 __be16 type, 2180 struct list_head *ptype_list) 2181 { 2182 struct packet_type *ptype, *pt_prev = *pt; 2183 2184 list_for_each_entry_rcu(ptype, ptype_list, list) { 2185 if (ptype->type != type) 2186 continue; 2187 if (pt_prev) 2188 deliver_skb(skb, pt_prev, orig_dev); 2189 pt_prev = ptype; 2190 } 2191 *pt = pt_prev; 2192 } 2193 2194 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2195 { 2196 if (!ptype->af_packet_priv || !skb->sk) 2197 return false; 2198 2199 if (ptype->id_match) 2200 return ptype->id_match(ptype, skb->sk); 2201 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2202 return true; 2203 2204 return false; 2205 } 2206 2207 /** 2208 * dev_nit_active - return true if any network interface taps are in use 2209 * 2210 * @dev: network device to check for the presence of taps 2211 */ 2212 bool dev_nit_active(struct net_device *dev) 2213 { 2214 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2215 } 2216 EXPORT_SYMBOL_GPL(dev_nit_active); 2217 2218 /* 2219 * Support routine. Sends outgoing frames to any network 2220 * taps currently in use. 2221 */ 2222 2223 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2224 { 2225 struct packet_type *ptype; 2226 struct sk_buff *skb2 = NULL; 2227 struct packet_type *pt_prev = NULL; 2228 struct list_head *ptype_list = &ptype_all; 2229 2230 rcu_read_lock(); 2231 again: 2232 list_for_each_entry_rcu(ptype, ptype_list, list) { 2233 if (ptype->ignore_outgoing) 2234 continue; 2235 2236 /* Never send packets back to the socket 2237 * they originated from - MvS (miquels@drinkel.ow.org) 2238 */ 2239 if (skb_loop_sk(ptype, skb)) 2240 continue; 2241 2242 if (pt_prev) { 2243 deliver_skb(skb2, pt_prev, skb->dev); 2244 pt_prev = ptype; 2245 continue; 2246 } 2247 2248 /* need to clone skb, done only once */ 2249 skb2 = skb_clone(skb, GFP_ATOMIC); 2250 if (!skb2) 2251 goto out_unlock; 2252 2253 net_timestamp_set(skb2); 2254 2255 /* skb->nh should be correctly 2256 * set by sender, so that the second statement is 2257 * just protection against buggy protocols. 2258 */ 2259 skb_reset_mac_header(skb2); 2260 2261 if (skb_network_header(skb2) < skb2->data || 2262 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2263 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2264 ntohs(skb2->protocol), 2265 dev->name); 2266 skb_reset_network_header(skb2); 2267 } 2268 2269 skb2->transport_header = skb2->network_header; 2270 skb2->pkt_type = PACKET_OUTGOING; 2271 pt_prev = ptype; 2272 } 2273 2274 if (ptype_list == &ptype_all) { 2275 ptype_list = &dev->ptype_all; 2276 goto again; 2277 } 2278 out_unlock: 2279 if (pt_prev) { 2280 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2281 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2282 else 2283 kfree_skb(skb2); 2284 } 2285 rcu_read_unlock(); 2286 } 2287 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2288 2289 /** 2290 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2291 * @dev: Network device 2292 * @txq: number of queues available 2293 * 2294 * If real_num_tx_queues is changed the tc mappings may no longer be 2295 * valid. To resolve this verify the tc mapping remains valid and if 2296 * not NULL the mapping. With no priorities mapping to this 2297 * offset/count pair it will no longer be used. In the worst case TC0 2298 * is invalid nothing can be done so disable priority mappings. If is 2299 * expected that drivers will fix this mapping if they can before 2300 * calling netif_set_real_num_tx_queues. 2301 */ 2302 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2303 { 2304 int i; 2305 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2306 2307 /* If TC0 is invalidated disable TC mapping */ 2308 if (tc->offset + tc->count > txq) { 2309 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2310 dev->num_tc = 0; 2311 return; 2312 } 2313 2314 /* Invalidated prio to tc mappings set to TC0 */ 2315 for (i = 1; i < TC_BITMASK + 1; i++) { 2316 int q = netdev_get_prio_tc_map(dev, i); 2317 2318 tc = &dev->tc_to_txq[q]; 2319 if (tc->offset + tc->count > txq) { 2320 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", 2321 i, q); 2322 netdev_set_prio_tc_map(dev, i, 0); 2323 } 2324 } 2325 } 2326 2327 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2328 { 2329 if (dev->num_tc) { 2330 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2331 int i; 2332 2333 /* walk through the TCs and see if it falls into any of them */ 2334 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2335 if ((txq - tc->offset) < tc->count) 2336 return i; 2337 } 2338 2339 /* didn't find it, just return -1 to indicate no match */ 2340 return -1; 2341 } 2342 2343 return 0; 2344 } 2345 EXPORT_SYMBOL(netdev_txq_to_tc); 2346 2347 #ifdef CONFIG_XPS 2348 static struct static_key xps_needed __read_mostly; 2349 static struct static_key xps_rxqs_needed __read_mostly; 2350 static DEFINE_MUTEX(xps_map_mutex); 2351 #define xmap_dereference(P) \ 2352 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2353 2354 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2355 struct xps_dev_maps *old_maps, int tci, u16 index) 2356 { 2357 struct xps_map *map = NULL; 2358 int pos; 2359 2360 if (dev_maps) 2361 map = xmap_dereference(dev_maps->attr_map[tci]); 2362 if (!map) 2363 return false; 2364 2365 for (pos = map->len; pos--;) { 2366 if (map->queues[pos] != index) 2367 continue; 2368 2369 if (map->len > 1) { 2370 map->queues[pos] = map->queues[--map->len]; 2371 break; 2372 } 2373 2374 if (old_maps) 2375 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2376 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2377 kfree_rcu(map, rcu); 2378 return false; 2379 } 2380 2381 return true; 2382 } 2383 2384 static bool remove_xps_queue_cpu(struct net_device *dev, 2385 struct xps_dev_maps *dev_maps, 2386 int cpu, u16 offset, u16 count) 2387 { 2388 int num_tc = dev_maps->num_tc; 2389 bool active = false; 2390 int tci; 2391 2392 for (tci = cpu * num_tc; num_tc--; tci++) { 2393 int i, j; 2394 2395 for (i = count, j = offset; i--; j++) { 2396 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2397 break; 2398 } 2399 2400 active |= i < 0; 2401 } 2402 2403 return active; 2404 } 2405 2406 static void reset_xps_maps(struct net_device *dev, 2407 struct xps_dev_maps *dev_maps, 2408 enum xps_map_type type) 2409 { 2410 static_key_slow_dec_cpuslocked(&xps_needed); 2411 if (type == XPS_RXQS) 2412 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2413 2414 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2415 2416 kfree_rcu(dev_maps, rcu); 2417 } 2418 2419 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2420 u16 offset, u16 count) 2421 { 2422 struct xps_dev_maps *dev_maps; 2423 bool active = false; 2424 int i, j; 2425 2426 dev_maps = xmap_dereference(dev->xps_maps[type]); 2427 if (!dev_maps) 2428 return; 2429 2430 for (j = 0; j < dev_maps->nr_ids; j++) 2431 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2432 if (!active) 2433 reset_xps_maps(dev, dev_maps, type); 2434 2435 if (type == XPS_CPUS) { 2436 for (i = offset + (count - 1); count--; i--) 2437 netdev_queue_numa_node_write( 2438 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2439 } 2440 } 2441 2442 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2443 u16 count) 2444 { 2445 if (!static_key_false(&xps_needed)) 2446 return; 2447 2448 cpus_read_lock(); 2449 mutex_lock(&xps_map_mutex); 2450 2451 if (static_key_false(&xps_rxqs_needed)) 2452 clean_xps_maps(dev, XPS_RXQS, offset, count); 2453 2454 clean_xps_maps(dev, XPS_CPUS, offset, count); 2455 2456 mutex_unlock(&xps_map_mutex); 2457 cpus_read_unlock(); 2458 } 2459 2460 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2461 { 2462 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2463 } 2464 2465 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2466 u16 index, bool is_rxqs_map) 2467 { 2468 struct xps_map *new_map; 2469 int alloc_len = XPS_MIN_MAP_ALLOC; 2470 int i, pos; 2471 2472 for (pos = 0; map && pos < map->len; pos++) { 2473 if (map->queues[pos] != index) 2474 continue; 2475 return map; 2476 } 2477 2478 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2479 if (map) { 2480 if (pos < map->alloc_len) 2481 return map; 2482 2483 alloc_len = map->alloc_len * 2; 2484 } 2485 2486 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2487 * map 2488 */ 2489 if (is_rxqs_map) 2490 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2491 else 2492 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2493 cpu_to_node(attr_index)); 2494 if (!new_map) 2495 return NULL; 2496 2497 for (i = 0; i < pos; i++) 2498 new_map->queues[i] = map->queues[i]; 2499 new_map->alloc_len = alloc_len; 2500 new_map->len = pos; 2501 2502 return new_map; 2503 } 2504 2505 /* Copy xps maps at a given index */ 2506 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2507 struct xps_dev_maps *new_dev_maps, int index, 2508 int tc, bool skip_tc) 2509 { 2510 int i, tci = index * dev_maps->num_tc; 2511 struct xps_map *map; 2512 2513 /* copy maps belonging to foreign traffic classes */ 2514 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2515 if (i == tc && skip_tc) 2516 continue; 2517 2518 /* fill in the new device map from the old device map */ 2519 map = xmap_dereference(dev_maps->attr_map[tci]); 2520 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2521 } 2522 } 2523 2524 /* Must be called under cpus_read_lock */ 2525 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2526 u16 index, enum xps_map_type type) 2527 { 2528 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2529 const unsigned long *online_mask = NULL; 2530 bool active = false, copy = false; 2531 int i, j, tci, numa_node_id = -2; 2532 int maps_sz, num_tc = 1, tc = 0; 2533 struct xps_map *map, *new_map; 2534 unsigned int nr_ids; 2535 2536 WARN_ON_ONCE(index >= dev->num_tx_queues); 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_drop_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_reason(struct sk_buff *skb, enum skb_drop_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_reason); 3132 3133 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3134 { 3135 if (in_hardirq() || irqs_disabled()) 3136 dev_kfree_skb_irq_reason(skb, reason); 3137 else 3138 kfree_skb_reason(skb, reason); 3139 } 3140 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3141 3142 3143 /** 3144 * netif_device_detach - mark device as removed 3145 * @dev: network device 3146 * 3147 * Mark device as removed from system and therefore no longer available. 3148 */ 3149 void netif_device_detach(struct net_device *dev) 3150 { 3151 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3152 netif_running(dev)) { 3153 netif_tx_stop_all_queues(dev); 3154 } 3155 } 3156 EXPORT_SYMBOL(netif_device_detach); 3157 3158 /** 3159 * netif_device_attach - mark device as attached 3160 * @dev: network device 3161 * 3162 * Mark device as attached from system and restart if needed. 3163 */ 3164 void netif_device_attach(struct net_device *dev) 3165 { 3166 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3167 netif_running(dev)) { 3168 netif_tx_wake_all_queues(dev); 3169 __netdev_watchdog_up(dev); 3170 } 3171 } 3172 EXPORT_SYMBOL(netif_device_attach); 3173 3174 /* 3175 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3176 * to be used as a distribution range. 3177 */ 3178 static u16 skb_tx_hash(const struct net_device *dev, 3179 const struct net_device *sb_dev, 3180 struct sk_buff *skb) 3181 { 3182 u32 hash; 3183 u16 qoffset = 0; 3184 u16 qcount = dev->real_num_tx_queues; 3185 3186 if (dev->num_tc) { 3187 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3188 3189 qoffset = sb_dev->tc_to_txq[tc].offset; 3190 qcount = sb_dev->tc_to_txq[tc].count; 3191 if (unlikely(!qcount)) { 3192 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3193 sb_dev->name, qoffset, tc); 3194 qoffset = 0; 3195 qcount = dev->real_num_tx_queues; 3196 } 3197 } 3198 3199 if (skb_rx_queue_recorded(skb)) { 3200 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3201 hash = skb_get_rx_queue(skb); 3202 if (hash >= qoffset) 3203 hash -= qoffset; 3204 while (unlikely(hash >= qcount)) 3205 hash -= qcount; 3206 return hash + qoffset; 3207 } 3208 3209 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3210 } 3211 3212 static void skb_warn_bad_offload(const struct sk_buff *skb) 3213 { 3214 static const netdev_features_t null_features; 3215 struct net_device *dev = skb->dev; 3216 const char *name = ""; 3217 3218 if (!net_ratelimit()) 3219 return; 3220 3221 if (dev) { 3222 if (dev->dev.parent) 3223 name = dev_driver_string(dev->dev.parent); 3224 else 3225 name = netdev_name(dev); 3226 } 3227 skb_dump(KERN_WARNING, skb, false); 3228 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3229 name, dev ? &dev->features : &null_features, 3230 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3231 } 3232 3233 /* 3234 * Invalidate hardware checksum when packet is to be mangled, and 3235 * complete checksum manually on outgoing path. 3236 */ 3237 int skb_checksum_help(struct sk_buff *skb) 3238 { 3239 __wsum csum; 3240 int ret = 0, offset; 3241 3242 if (skb->ip_summed == CHECKSUM_COMPLETE) 3243 goto out_set_summed; 3244 3245 if (unlikely(skb_is_gso(skb))) { 3246 skb_warn_bad_offload(skb); 3247 return -EINVAL; 3248 } 3249 3250 /* Before computing a checksum, we should make sure no frag could 3251 * be modified by an external entity : checksum could be wrong. 3252 */ 3253 if (skb_has_shared_frag(skb)) { 3254 ret = __skb_linearize(skb); 3255 if (ret) 3256 goto out; 3257 } 3258 3259 offset = skb_checksum_start_offset(skb); 3260 ret = -EINVAL; 3261 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3262 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3263 goto out; 3264 } 3265 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3266 3267 offset += skb->csum_offset; 3268 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3269 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3270 goto out; 3271 } 3272 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3273 if (ret) 3274 goto out; 3275 3276 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3277 out_set_summed: 3278 skb->ip_summed = CHECKSUM_NONE; 3279 out: 3280 return ret; 3281 } 3282 EXPORT_SYMBOL(skb_checksum_help); 3283 3284 int skb_crc32c_csum_help(struct sk_buff *skb) 3285 { 3286 __le32 crc32c_csum; 3287 int ret = 0, offset, start; 3288 3289 if (skb->ip_summed != CHECKSUM_PARTIAL) 3290 goto out; 3291 3292 if (unlikely(skb_is_gso(skb))) 3293 goto out; 3294 3295 /* Before computing a checksum, we should make sure no frag could 3296 * be modified by an external entity : checksum could be wrong. 3297 */ 3298 if (unlikely(skb_has_shared_frag(skb))) { 3299 ret = __skb_linearize(skb); 3300 if (ret) 3301 goto out; 3302 } 3303 start = skb_checksum_start_offset(skb); 3304 offset = start + offsetof(struct sctphdr, checksum); 3305 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3306 ret = -EINVAL; 3307 goto out; 3308 } 3309 3310 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3311 if (ret) 3312 goto out; 3313 3314 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3315 skb->len - start, ~(__u32)0, 3316 crc32c_csum_stub)); 3317 *(__le32 *)(skb->data + offset) = crc32c_csum; 3318 skb->ip_summed = CHECKSUM_NONE; 3319 skb->csum_not_inet = 0; 3320 out: 3321 return ret; 3322 } 3323 3324 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3325 { 3326 __be16 type = skb->protocol; 3327 3328 /* Tunnel gso handlers can set protocol to ethernet. */ 3329 if (type == htons(ETH_P_TEB)) { 3330 struct ethhdr *eth; 3331 3332 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3333 return 0; 3334 3335 eth = (struct ethhdr *)skb->data; 3336 type = eth->h_proto; 3337 } 3338 3339 return __vlan_get_protocol(skb, type, depth); 3340 } 3341 3342 /* openvswitch calls this on rx path, so we need a different check. 3343 */ 3344 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3345 { 3346 if (tx_path) 3347 return skb->ip_summed != CHECKSUM_PARTIAL && 3348 skb->ip_summed != CHECKSUM_UNNECESSARY; 3349 3350 return skb->ip_summed == CHECKSUM_NONE; 3351 } 3352 3353 /** 3354 * __skb_gso_segment - Perform segmentation on skb. 3355 * @skb: buffer to segment 3356 * @features: features for the output path (see dev->features) 3357 * @tx_path: whether it is called in TX path 3358 * 3359 * This function segments the given skb and returns a list of segments. 3360 * 3361 * It may return NULL if the skb requires no segmentation. This is 3362 * only possible when GSO is used for verifying header integrity. 3363 * 3364 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3365 */ 3366 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3367 netdev_features_t features, bool tx_path) 3368 { 3369 struct sk_buff *segs; 3370 3371 if (unlikely(skb_needs_check(skb, tx_path))) { 3372 int err; 3373 3374 /* We're going to init ->check field in TCP or UDP header */ 3375 err = skb_cow_head(skb, 0); 3376 if (err < 0) 3377 return ERR_PTR(err); 3378 } 3379 3380 /* Only report GSO partial support if it will enable us to 3381 * support segmentation on this frame without needing additional 3382 * work. 3383 */ 3384 if (features & NETIF_F_GSO_PARTIAL) { 3385 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3386 struct net_device *dev = skb->dev; 3387 3388 partial_features |= dev->features & dev->gso_partial_features; 3389 if (!skb_gso_ok(skb, features | partial_features)) 3390 features &= ~NETIF_F_GSO_PARTIAL; 3391 } 3392 3393 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3394 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3395 3396 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3397 SKB_GSO_CB(skb)->encap_level = 0; 3398 3399 skb_reset_mac_header(skb); 3400 skb_reset_mac_len(skb); 3401 3402 segs = skb_mac_gso_segment(skb, features); 3403 3404 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3405 skb_warn_bad_offload(skb); 3406 3407 return segs; 3408 } 3409 EXPORT_SYMBOL(__skb_gso_segment); 3410 3411 /* Take action when hardware reception checksum errors are detected. */ 3412 #ifdef CONFIG_BUG 3413 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3414 { 3415 netdev_err(dev, "hw csum failure\n"); 3416 skb_dump(KERN_ERR, skb, true); 3417 dump_stack(); 3418 } 3419 3420 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3421 { 3422 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3423 } 3424 EXPORT_SYMBOL(netdev_rx_csum_fault); 3425 #endif 3426 3427 /* XXX: check that highmem exists at all on the given machine. */ 3428 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3429 { 3430 #ifdef CONFIG_HIGHMEM 3431 int i; 3432 3433 if (!(dev->features & NETIF_F_HIGHDMA)) { 3434 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3435 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3436 3437 if (PageHighMem(skb_frag_page(frag))) 3438 return 1; 3439 } 3440 } 3441 #endif 3442 return 0; 3443 } 3444 3445 /* If MPLS offload request, verify we are testing hardware MPLS features 3446 * instead of standard features for the netdev. 3447 */ 3448 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3449 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3450 netdev_features_t features, 3451 __be16 type) 3452 { 3453 if (eth_p_mpls(type)) 3454 features &= skb->dev->mpls_features; 3455 3456 return features; 3457 } 3458 #else 3459 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3460 netdev_features_t features, 3461 __be16 type) 3462 { 3463 return features; 3464 } 3465 #endif 3466 3467 static netdev_features_t harmonize_features(struct sk_buff *skb, 3468 netdev_features_t features) 3469 { 3470 __be16 type; 3471 3472 type = skb_network_protocol(skb, NULL); 3473 features = net_mpls_features(skb, features, type); 3474 3475 if (skb->ip_summed != CHECKSUM_NONE && 3476 !can_checksum_protocol(features, type)) { 3477 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3478 } 3479 if (illegal_highdma(skb->dev, skb)) 3480 features &= ~NETIF_F_SG; 3481 3482 return features; 3483 } 3484 3485 netdev_features_t passthru_features_check(struct sk_buff *skb, 3486 struct net_device *dev, 3487 netdev_features_t features) 3488 { 3489 return features; 3490 } 3491 EXPORT_SYMBOL(passthru_features_check); 3492 3493 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3494 struct net_device *dev, 3495 netdev_features_t features) 3496 { 3497 return vlan_features_check(skb, features); 3498 } 3499 3500 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3501 struct net_device *dev, 3502 netdev_features_t features) 3503 { 3504 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3505 3506 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3507 return features & ~NETIF_F_GSO_MASK; 3508 3509 if (!skb_shinfo(skb)->gso_type) { 3510 skb_warn_bad_offload(skb); 3511 return features & ~NETIF_F_GSO_MASK; 3512 } 3513 3514 /* Support for GSO partial features requires software 3515 * intervention before we can actually process the packets 3516 * so we need to strip support for any partial features now 3517 * and we can pull them back in after we have partially 3518 * segmented the frame. 3519 */ 3520 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3521 features &= ~dev->gso_partial_features; 3522 3523 /* Make sure to clear the IPv4 ID mangling feature if the 3524 * IPv4 header has the potential to be fragmented. 3525 */ 3526 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3527 struct iphdr *iph = skb->encapsulation ? 3528 inner_ip_hdr(skb) : ip_hdr(skb); 3529 3530 if (!(iph->frag_off & htons(IP_DF))) 3531 features &= ~NETIF_F_TSO_MANGLEID; 3532 } 3533 3534 return features; 3535 } 3536 3537 netdev_features_t netif_skb_features(struct sk_buff *skb) 3538 { 3539 struct net_device *dev = skb->dev; 3540 netdev_features_t features = dev->features; 3541 3542 if (skb_is_gso(skb)) 3543 features = gso_features_check(skb, dev, features); 3544 3545 /* If encapsulation offload request, verify we are testing 3546 * hardware encapsulation features instead of standard 3547 * features for the netdev 3548 */ 3549 if (skb->encapsulation) 3550 features &= dev->hw_enc_features; 3551 3552 if (skb_vlan_tagged(skb)) 3553 features = netdev_intersect_features(features, 3554 dev->vlan_features | 3555 NETIF_F_HW_VLAN_CTAG_TX | 3556 NETIF_F_HW_VLAN_STAG_TX); 3557 3558 if (dev->netdev_ops->ndo_features_check) 3559 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3560 features); 3561 else 3562 features &= dflt_features_check(skb, dev, features); 3563 3564 return harmonize_features(skb, features); 3565 } 3566 EXPORT_SYMBOL(netif_skb_features); 3567 3568 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3569 struct netdev_queue *txq, bool more) 3570 { 3571 unsigned int len; 3572 int rc; 3573 3574 if (dev_nit_active(dev)) 3575 dev_queue_xmit_nit(skb, dev); 3576 3577 len = skb->len; 3578 trace_net_dev_start_xmit(skb, dev); 3579 rc = netdev_start_xmit(skb, dev, txq, more); 3580 trace_net_dev_xmit(skb, rc, dev, len); 3581 3582 return rc; 3583 } 3584 3585 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3586 struct netdev_queue *txq, int *ret) 3587 { 3588 struct sk_buff *skb = first; 3589 int rc = NETDEV_TX_OK; 3590 3591 while (skb) { 3592 struct sk_buff *next = skb->next; 3593 3594 skb_mark_not_on_list(skb); 3595 rc = xmit_one(skb, dev, txq, next != NULL); 3596 if (unlikely(!dev_xmit_complete(rc))) { 3597 skb->next = next; 3598 goto out; 3599 } 3600 3601 skb = next; 3602 if (netif_tx_queue_stopped(txq) && skb) { 3603 rc = NETDEV_TX_BUSY; 3604 break; 3605 } 3606 } 3607 3608 out: 3609 *ret = rc; 3610 return skb; 3611 } 3612 3613 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3614 netdev_features_t features) 3615 { 3616 if (skb_vlan_tag_present(skb) && 3617 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3618 skb = __vlan_hwaccel_push_inside(skb); 3619 return skb; 3620 } 3621 3622 int skb_csum_hwoffload_help(struct sk_buff *skb, 3623 const netdev_features_t features) 3624 { 3625 if (unlikely(skb_csum_is_sctp(skb))) 3626 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3627 skb_crc32c_csum_help(skb); 3628 3629 if (features & NETIF_F_HW_CSUM) 3630 return 0; 3631 3632 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3633 switch (skb->csum_offset) { 3634 case offsetof(struct tcphdr, check): 3635 case offsetof(struct udphdr, check): 3636 return 0; 3637 } 3638 } 3639 3640 return skb_checksum_help(skb); 3641 } 3642 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3643 3644 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3645 { 3646 netdev_features_t features; 3647 3648 features = netif_skb_features(skb); 3649 skb = validate_xmit_vlan(skb, features); 3650 if (unlikely(!skb)) 3651 goto out_null; 3652 3653 skb = sk_validate_xmit_skb(skb, dev); 3654 if (unlikely(!skb)) 3655 goto out_null; 3656 3657 if (netif_needs_gso(skb, features)) { 3658 struct sk_buff *segs; 3659 3660 segs = skb_gso_segment(skb, features); 3661 if (IS_ERR(segs)) { 3662 goto out_kfree_skb; 3663 } else if (segs) { 3664 consume_skb(skb); 3665 skb = segs; 3666 } 3667 } else { 3668 if (skb_needs_linearize(skb, features) && 3669 __skb_linearize(skb)) 3670 goto out_kfree_skb; 3671 3672 /* If packet is not checksummed and device does not 3673 * support checksumming for this protocol, complete 3674 * checksumming here. 3675 */ 3676 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3677 if (skb->encapsulation) 3678 skb_set_inner_transport_header(skb, 3679 skb_checksum_start_offset(skb)); 3680 else 3681 skb_set_transport_header(skb, 3682 skb_checksum_start_offset(skb)); 3683 if (skb_csum_hwoffload_help(skb, features)) 3684 goto out_kfree_skb; 3685 } 3686 } 3687 3688 skb = validate_xmit_xfrm(skb, features, again); 3689 3690 return skb; 3691 3692 out_kfree_skb: 3693 kfree_skb(skb); 3694 out_null: 3695 dev_core_stats_tx_dropped_inc(dev); 3696 return NULL; 3697 } 3698 3699 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3700 { 3701 struct sk_buff *next, *head = NULL, *tail; 3702 3703 for (; skb != NULL; skb = next) { 3704 next = skb->next; 3705 skb_mark_not_on_list(skb); 3706 3707 /* in case skb wont be segmented, point to itself */ 3708 skb->prev = skb; 3709 3710 skb = validate_xmit_skb(skb, dev, again); 3711 if (!skb) 3712 continue; 3713 3714 if (!head) 3715 head = skb; 3716 else 3717 tail->next = skb; 3718 /* If skb was segmented, skb->prev points to 3719 * the last segment. If not, it still contains skb. 3720 */ 3721 tail = skb->prev; 3722 } 3723 return head; 3724 } 3725 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3726 3727 static void qdisc_pkt_len_init(struct sk_buff *skb) 3728 { 3729 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3730 3731 qdisc_skb_cb(skb)->pkt_len = skb->len; 3732 3733 /* To get more precise estimation of bytes sent on wire, 3734 * we add to pkt_len the headers size of all segments 3735 */ 3736 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3737 u16 gso_segs = shinfo->gso_segs; 3738 unsigned int hdr_len; 3739 3740 /* mac layer + network layer */ 3741 hdr_len = skb_transport_offset(skb); 3742 3743 /* + transport layer */ 3744 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3745 const struct tcphdr *th; 3746 struct tcphdr _tcphdr; 3747 3748 th = skb_header_pointer(skb, hdr_len, 3749 sizeof(_tcphdr), &_tcphdr); 3750 if (likely(th)) 3751 hdr_len += __tcp_hdrlen(th); 3752 } else { 3753 struct udphdr _udphdr; 3754 3755 if (skb_header_pointer(skb, hdr_len, 3756 sizeof(_udphdr), &_udphdr)) 3757 hdr_len += sizeof(struct udphdr); 3758 } 3759 3760 if (shinfo->gso_type & SKB_GSO_DODGY) 3761 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3762 shinfo->gso_size); 3763 3764 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3765 } 3766 } 3767 3768 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3769 struct sk_buff **to_free, 3770 struct netdev_queue *txq) 3771 { 3772 int rc; 3773 3774 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3775 if (rc == NET_XMIT_SUCCESS) 3776 trace_qdisc_enqueue(q, txq, skb); 3777 return rc; 3778 } 3779 3780 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3781 struct net_device *dev, 3782 struct netdev_queue *txq) 3783 { 3784 spinlock_t *root_lock = qdisc_lock(q); 3785 struct sk_buff *to_free = NULL; 3786 bool contended; 3787 int rc; 3788 3789 qdisc_calculate_pkt_len(skb, q); 3790 3791 if (q->flags & TCQ_F_NOLOCK) { 3792 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3793 qdisc_run_begin(q)) { 3794 /* Retest nolock_qdisc_is_empty() within the protection 3795 * of q->seqlock to protect from racing with requeuing. 3796 */ 3797 if (unlikely(!nolock_qdisc_is_empty(q))) { 3798 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3799 __qdisc_run(q); 3800 qdisc_run_end(q); 3801 3802 goto no_lock_out; 3803 } 3804 3805 qdisc_bstats_cpu_update(q, skb); 3806 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3807 !nolock_qdisc_is_empty(q)) 3808 __qdisc_run(q); 3809 3810 qdisc_run_end(q); 3811 return NET_XMIT_SUCCESS; 3812 } 3813 3814 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3815 qdisc_run(q); 3816 3817 no_lock_out: 3818 if (unlikely(to_free)) 3819 kfree_skb_list_reason(to_free, 3820 SKB_DROP_REASON_QDISC_DROP); 3821 return rc; 3822 } 3823 3824 /* 3825 * Heuristic to force contended enqueues to serialize on a 3826 * separate lock before trying to get qdisc main lock. 3827 * This permits qdisc->running owner to get the lock more 3828 * often and dequeue packets faster. 3829 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3830 * and then other tasks will only enqueue packets. The packets will be 3831 * sent after the qdisc owner is scheduled again. To prevent this 3832 * scenario the task always serialize on the lock. 3833 */ 3834 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3835 if (unlikely(contended)) 3836 spin_lock(&q->busylock); 3837 3838 spin_lock(root_lock); 3839 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3840 __qdisc_drop(skb, &to_free); 3841 rc = NET_XMIT_DROP; 3842 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3843 qdisc_run_begin(q)) { 3844 /* 3845 * This is a work-conserving queue; there are no old skbs 3846 * waiting to be sent out; and the qdisc is not running - 3847 * xmit the skb directly. 3848 */ 3849 3850 qdisc_bstats_update(q, skb); 3851 3852 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3853 if (unlikely(contended)) { 3854 spin_unlock(&q->busylock); 3855 contended = false; 3856 } 3857 __qdisc_run(q); 3858 } 3859 3860 qdisc_run_end(q); 3861 rc = NET_XMIT_SUCCESS; 3862 } else { 3863 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3864 if (qdisc_run_begin(q)) { 3865 if (unlikely(contended)) { 3866 spin_unlock(&q->busylock); 3867 contended = false; 3868 } 3869 __qdisc_run(q); 3870 qdisc_run_end(q); 3871 } 3872 } 3873 spin_unlock(root_lock); 3874 if (unlikely(to_free)) 3875 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3876 if (unlikely(contended)) 3877 spin_unlock(&q->busylock); 3878 return rc; 3879 } 3880 3881 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3882 static void skb_update_prio(struct sk_buff *skb) 3883 { 3884 const struct netprio_map *map; 3885 const struct sock *sk; 3886 unsigned int prioidx; 3887 3888 if (skb->priority) 3889 return; 3890 map = rcu_dereference_bh(skb->dev->priomap); 3891 if (!map) 3892 return; 3893 sk = skb_to_full_sk(skb); 3894 if (!sk) 3895 return; 3896 3897 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3898 3899 if (prioidx < map->priomap_len) 3900 skb->priority = map->priomap[prioidx]; 3901 } 3902 #else 3903 #define skb_update_prio(skb) 3904 #endif 3905 3906 /** 3907 * dev_loopback_xmit - loop back @skb 3908 * @net: network namespace this loopback is happening in 3909 * @sk: sk needed to be a netfilter okfn 3910 * @skb: buffer to transmit 3911 */ 3912 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3913 { 3914 skb_reset_mac_header(skb); 3915 __skb_pull(skb, skb_network_offset(skb)); 3916 skb->pkt_type = PACKET_LOOPBACK; 3917 if (skb->ip_summed == CHECKSUM_NONE) 3918 skb->ip_summed = CHECKSUM_UNNECESSARY; 3919 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3920 skb_dst_force(skb); 3921 netif_rx(skb); 3922 return 0; 3923 } 3924 EXPORT_SYMBOL(dev_loopback_xmit); 3925 3926 #ifdef CONFIG_NET_EGRESS 3927 static struct sk_buff * 3928 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3929 { 3930 #ifdef CONFIG_NET_CLS_ACT 3931 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3932 struct tcf_result cl_res; 3933 3934 if (!miniq) 3935 return skb; 3936 3937 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3938 tc_skb_cb(skb)->mru = 0; 3939 tc_skb_cb(skb)->post_ct = false; 3940 mini_qdisc_bstats_cpu_update(miniq, skb); 3941 3942 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3943 case TC_ACT_OK: 3944 case TC_ACT_RECLASSIFY: 3945 skb->tc_index = TC_H_MIN(cl_res.classid); 3946 break; 3947 case TC_ACT_SHOT: 3948 mini_qdisc_qstats_cpu_drop(miniq); 3949 *ret = NET_XMIT_DROP; 3950 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3951 return NULL; 3952 case TC_ACT_STOLEN: 3953 case TC_ACT_QUEUED: 3954 case TC_ACT_TRAP: 3955 *ret = NET_XMIT_SUCCESS; 3956 consume_skb(skb); 3957 return NULL; 3958 case TC_ACT_REDIRECT: 3959 /* No need to push/pop skb's mac_header here on egress! */ 3960 skb_do_redirect(skb); 3961 *ret = NET_XMIT_SUCCESS; 3962 return NULL; 3963 default: 3964 break; 3965 } 3966 #endif /* CONFIG_NET_CLS_ACT */ 3967 3968 return skb; 3969 } 3970 3971 static struct netdev_queue * 3972 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3973 { 3974 int qm = skb_get_queue_mapping(skb); 3975 3976 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3977 } 3978 3979 static bool netdev_xmit_txqueue_skipped(void) 3980 { 3981 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3982 } 3983 3984 void netdev_xmit_skip_txqueue(bool skip) 3985 { 3986 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3987 } 3988 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3989 #endif /* CONFIG_NET_EGRESS */ 3990 3991 #ifdef CONFIG_XPS 3992 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3993 struct xps_dev_maps *dev_maps, unsigned int tci) 3994 { 3995 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3996 struct xps_map *map; 3997 int queue_index = -1; 3998 3999 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4000 return queue_index; 4001 4002 tci *= dev_maps->num_tc; 4003 tci += tc; 4004 4005 map = rcu_dereference(dev_maps->attr_map[tci]); 4006 if (map) { 4007 if (map->len == 1) 4008 queue_index = map->queues[0]; 4009 else 4010 queue_index = map->queues[reciprocal_scale( 4011 skb_get_hash(skb), map->len)]; 4012 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4013 queue_index = -1; 4014 } 4015 return queue_index; 4016 } 4017 #endif 4018 4019 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4020 struct sk_buff *skb) 4021 { 4022 #ifdef CONFIG_XPS 4023 struct xps_dev_maps *dev_maps; 4024 struct sock *sk = skb->sk; 4025 int queue_index = -1; 4026 4027 if (!static_key_false(&xps_needed)) 4028 return -1; 4029 4030 rcu_read_lock(); 4031 if (!static_key_false(&xps_rxqs_needed)) 4032 goto get_cpus_map; 4033 4034 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4035 if (dev_maps) { 4036 int tci = sk_rx_queue_get(sk); 4037 4038 if (tci >= 0) 4039 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4040 tci); 4041 } 4042 4043 get_cpus_map: 4044 if (queue_index < 0) { 4045 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4046 if (dev_maps) { 4047 unsigned int tci = skb->sender_cpu - 1; 4048 4049 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4050 tci); 4051 } 4052 } 4053 rcu_read_unlock(); 4054 4055 return queue_index; 4056 #else 4057 return -1; 4058 #endif 4059 } 4060 4061 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4062 struct net_device *sb_dev) 4063 { 4064 return 0; 4065 } 4066 EXPORT_SYMBOL(dev_pick_tx_zero); 4067 4068 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4069 struct net_device *sb_dev) 4070 { 4071 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4072 } 4073 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4074 4075 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4076 struct net_device *sb_dev) 4077 { 4078 struct sock *sk = skb->sk; 4079 int queue_index = sk_tx_queue_get(sk); 4080 4081 sb_dev = sb_dev ? : dev; 4082 4083 if (queue_index < 0 || skb->ooo_okay || 4084 queue_index >= dev->real_num_tx_queues) { 4085 int new_index = get_xps_queue(dev, sb_dev, skb); 4086 4087 if (new_index < 0) 4088 new_index = skb_tx_hash(dev, sb_dev, skb); 4089 4090 if (queue_index != new_index && sk && 4091 sk_fullsock(sk) && 4092 rcu_access_pointer(sk->sk_dst_cache)) 4093 sk_tx_queue_set(sk, new_index); 4094 4095 queue_index = new_index; 4096 } 4097 4098 return queue_index; 4099 } 4100 EXPORT_SYMBOL(netdev_pick_tx); 4101 4102 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4103 struct sk_buff *skb, 4104 struct net_device *sb_dev) 4105 { 4106 int queue_index = 0; 4107 4108 #ifdef CONFIG_XPS 4109 u32 sender_cpu = skb->sender_cpu - 1; 4110 4111 if (sender_cpu >= (u32)NR_CPUS) 4112 skb->sender_cpu = raw_smp_processor_id() + 1; 4113 #endif 4114 4115 if (dev->real_num_tx_queues != 1) { 4116 const struct net_device_ops *ops = dev->netdev_ops; 4117 4118 if (ops->ndo_select_queue) 4119 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4120 else 4121 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4122 4123 queue_index = netdev_cap_txqueue(dev, queue_index); 4124 } 4125 4126 skb_set_queue_mapping(skb, queue_index); 4127 return netdev_get_tx_queue(dev, queue_index); 4128 } 4129 4130 /** 4131 * __dev_queue_xmit() - transmit a buffer 4132 * @skb: buffer to transmit 4133 * @sb_dev: suboordinate device used for L2 forwarding offload 4134 * 4135 * Queue a buffer for transmission to a network device. The caller must 4136 * have set the device and priority and built the buffer before calling 4137 * this function. The function can be called from an interrupt. 4138 * 4139 * When calling this method, interrupts MUST be enabled. This is because 4140 * the BH enable code must have IRQs enabled so that it will not deadlock. 4141 * 4142 * Regardless of the return value, the skb is consumed, so it is currently 4143 * difficult to retry a send to this method. (You can bump the ref count 4144 * before sending to hold a reference for retry if you are careful.) 4145 * 4146 * Return: 4147 * * 0 - buffer successfully transmitted 4148 * * positive qdisc return code - NET_XMIT_DROP etc. 4149 * * negative errno - other errors 4150 */ 4151 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4152 { 4153 struct net_device *dev = skb->dev; 4154 struct netdev_queue *txq = NULL; 4155 struct Qdisc *q; 4156 int rc = -ENOMEM; 4157 bool again = false; 4158 4159 skb_reset_mac_header(skb); 4160 skb_assert_len(skb); 4161 4162 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4163 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4164 4165 /* Disable soft irqs for various locks below. Also 4166 * stops preemption for RCU. 4167 */ 4168 rcu_read_lock_bh(); 4169 4170 skb_update_prio(skb); 4171 4172 qdisc_pkt_len_init(skb); 4173 #ifdef CONFIG_NET_CLS_ACT 4174 skb->tc_at_ingress = 0; 4175 #endif 4176 #ifdef CONFIG_NET_EGRESS 4177 if (static_branch_unlikely(&egress_needed_key)) { 4178 if (nf_hook_egress_active()) { 4179 skb = nf_hook_egress(skb, &rc, dev); 4180 if (!skb) 4181 goto out; 4182 } 4183 4184 netdev_xmit_skip_txqueue(false); 4185 4186 nf_skip_egress(skb, true); 4187 skb = sch_handle_egress(skb, &rc, dev); 4188 if (!skb) 4189 goto out; 4190 nf_skip_egress(skb, false); 4191 4192 if (netdev_xmit_txqueue_skipped()) 4193 txq = netdev_tx_queue_mapping(dev, skb); 4194 } 4195 #endif 4196 /* If device/qdisc don't need skb->dst, release it right now while 4197 * its hot in this cpu cache. 4198 */ 4199 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4200 skb_dst_drop(skb); 4201 else 4202 skb_dst_force(skb); 4203 4204 if (!txq) 4205 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4206 4207 q = rcu_dereference_bh(txq->qdisc); 4208 4209 trace_net_dev_queue(skb); 4210 if (q->enqueue) { 4211 rc = __dev_xmit_skb(skb, q, dev, txq); 4212 goto out; 4213 } 4214 4215 /* The device has no queue. Common case for software devices: 4216 * loopback, all the sorts of tunnels... 4217 4218 * Really, it is unlikely that netif_tx_lock protection is necessary 4219 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4220 * counters.) 4221 * However, it is possible, that they rely on protection 4222 * made by us here. 4223 4224 * Check this and shot the lock. It is not prone from deadlocks. 4225 *Either shot noqueue qdisc, it is even simpler 8) 4226 */ 4227 if (dev->flags & IFF_UP) { 4228 int cpu = smp_processor_id(); /* ok because BHs are off */ 4229 4230 /* Other cpus might concurrently change txq->xmit_lock_owner 4231 * to -1 or to their cpu id, but not to our id. 4232 */ 4233 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4234 if (dev_xmit_recursion()) 4235 goto recursion_alert; 4236 4237 skb = validate_xmit_skb(skb, dev, &again); 4238 if (!skb) 4239 goto out; 4240 4241 HARD_TX_LOCK(dev, txq, cpu); 4242 4243 if (!netif_xmit_stopped(txq)) { 4244 dev_xmit_recursion_inc(); 4245 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4246 dev_xmit_recursion_dec(); 4247 if (dev_xmit_complete(rc)) { 4248 HARD_TX_UNLOCK(dev, txq); 4249 goto out; 4250 } 4251 } 4252 HARD_TX_UNLOCK(dev, txq); 4253 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4254 dev->name); 4255 } else { 4256 /* Recursion is detected! It is possible, 4257 * unfortunately 4258 */ 4259 recursion_alert: 4260 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4261 dev->name); 4262 } 4263 } 4264 4265 rc = -ENETDOWN; 4266 rcu_read_unlock_bh(); 4267 4268 dev_core_stats_tx_dropped_inc(dev); 4269 kfree_skb_list(skb); 4270 return rc; 4271 out: 4272 rcu_read_unlock_bh(); 4273 return rc; 4274 } 4275 EXPORT_SYMBOL(__dev_queue_xmit); 4276 4277 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4278 { 4279 struct net_device *dev = skb->dev; 4280 struct sk_buff *orig_skb = skb; 4281 struct netdev_queue *txq; 4282 int ret = NETDEV_TX_BUSY; 4283 bool again = false; 4284 4285 if (unlikely(!netif_running(dev) || 4286 !netif_carrier_ok(dev))) 4287 goto drop; 4288 4289 skb = validate_xmit_skb_list(skb, dev, &again); 4290 if (skb != orig_skb) 4291 goto drop; 4292 4293 skb_set_queue_mapping(skb, queue_id); 4294 txq = skb_get_tx_queue(dev, skb); 4295 4296 local_bh_disable(); 4297 4298 dev_xmit_recursion_inc(); 4299 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4300 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4301 ret = netdev_start_xmit(skb, dev, txq, false); 4302 HARD_TX_UNLOCK(dev, txq); 4303 dev_xmit_recursion_dec(); 4304 4305 local_bh_enable(); 4306 return ret; 4307 drop: 4308 dev_core_stats_tx_dropped_inc(dev); 4309 kfree_skb_list(skb); 4310 return NET_XMIT_DROP; 4311 } 4312 EXPORT_SYMBOL(__dev_direct_xmit); 4313 4314 /************************************************************************* 4315 * Receiver routines 4316 *************************************************************************/ 4317 4318 int netdev_max_backlog __read_mostly = 1000; 4319 EXPORT_SYMBOL(netdev_max_backlog); 4320 4321 int netdev_tstamp_prequeue __read_mostly = 1; 4322 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4323 int netdev_budget __read_mostly = 300; 4324 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4325 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4326 int weight_p __read_mostly = 64; /* old backlog weight */ 4327 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4328 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4329 int dev_rx_weight __read_mostly = 64; 4330 int dev_tx_weight __read_mostly = 64; 4331 4332 /* Called with irq disabled */ 4333 static inline void ____napi_schedule(struct softnet_data *sd, 4334 struct napi_struct *napi) 4335 { 4336 struct task_struct *thread; 4337 4338 lockdep_assert_irqs_disabled(); 4339 4340 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4341 /* Paired with smp_mb__before_atomic() in 4342 * napi_enable()/dev_set_threaded(). 4343 * Use READ_ONCE() to guarantee a complete 4344 * read on napi->thread. Only call 4345 * wake_up_process() when it's not NULL. 4346 */ 4347 thread = READ_ONCE(napi->thread); 4348 if (thread) { 4349 /* Avoid doing set_bit() if the thread is in 4350 * INTERRUPTIBLE state, cause napi_thread_wait() 4351 * makes sure to proceed with napi polling 4352 * if the thread is explicitly woken from here. 4353 */ 4354 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4355 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4356 wake_up_process(thread); 4357 return; 4358 } 4359 } 4360 4361 list_add_tail(&napi->poll_list, &sd->poll_list); 4362 /* If not called from net_rx_action() 4363 * we have to raise NET_RX_SOFTIRQ. 4364 */ 4365 if (!sd->in_net_rx_action) 4366 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4367 } 4368 4369 #ifdef CONFIG_RPS 4370 4371 /* One global table that all flow-based protocols share. */ 4372 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4373 EXPORT_SYMBOL(rps_sock_flow_table); 4374 u32 rps_cpu_mask __read_mostly; 4375 EXPORT_SYMBOL(rps_cpu_mask); 4376 4377 struct static_key_false rps_needed __read_mostly; 4378 EXPORT_SYMBOL(rps_needed); 4379 struct static_key_false rfs_needed __read_mostly; 4380 EXPORT_SYMBOL(rfs_needed); 4381 4382 static struct rps_dev_flow * 4383 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4384 struct rps_dev_flow *rflow, u16 next_cpu) 4385 { 4386 if (next_cpu < nr_cpu_ids) { 4387 #ifdef CONFIG_RFS_ACCEL 4388 struct netdev_rx_queue *rxqueue; 4389 struct rps_dev_flow_table *flow_table; 4390 struct rps_dev_flow *old_rflow; 4391 u32 flow_id; 4392 u16 rxq_index; 4393 int rc; 4394 4395 /* Should we steer this flow to a different hardware queue? */ 4396 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4397 !(dev->features & NETIF_F_NTUPLE)) 4398 goto out; 4399 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4400 if (rxq_index == skb_get_rx_queue(skb)) 4401 goto out; 4402 4403 rxqueue = dev->_rx + rxq_index; 4404 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4405 if (!flow_table) 4406 goto out; 4407 flow_id = skb_get_hash(skb) & flow_table->mask; 4408 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4409 rxq_index, flow_id); 4410 if (rc < 0) 4411 goto out; 4412 old_rflow = rflow; 4413 rflow = &flow_table->flows[flow_id]; 4414 rflow->filter = rc; 4415 if (old_rflow->filter == rflow->filter) 4416 old_rflow->filter = RPS_NO_FILTER; 4417 out: 4418 #endif 4419 rflow->last_qtail = 4420 per_cpu(softnet_data, next_cpu).input_queue_head; 4421 } 4422 4423 rflow->cpu = next_cpu; 4424 return rflow; 4425 } 4426 4427 /* 4428 * get_rps_cpu is called from netif_receive_skb and returns the target 4429 * CPU from the RPS map of the receiving queue for a given skb. 4430 * rcu_read_lock must be held on entry. 4431 */ 4432 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4433 struct rps_dev_flow **rflowp) 4434 { 4435 const struct rps_sock_flow_table *sock_flow_table; 4436 struct netdev_rx_queue *rxqueue = dev->_rx; 4437 struct rps_dev_flow_table *flow_table; 4438 struct rps_map *map; 4439 int cpu = -1; 4440 u32 tcpu; 4441 u32 hash; 4442 4443 if (skb_rx_queue_recorded(skb)) { 4444 u16 index = skb_get_rx_queue(skb); 4445 4446 if (unlikely(index >= dev->real_num_rx_queues)) { 4447 WARN_ONCE(dev->real_num_rx_queues > 1, 4448 "%s received packet on queue %u, but number " 4449 "of RX queues is %u\n", 4450 dev->name, index, dev->real_num_rx_queues); 4451 goto done; 4452 } 4453 rxqueue += index; 4454 } 4455 4456 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4457 4458 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4459 map = rcu_dereference(rxqueue->rps_map); 4460 if (!flow_table && !map) 4461 goto done; 4462 4463 skb_reset_network_header(skb); 4464 hash = skb_get_hash(skb); 4465 if (!hash) 4466 goto done; 4467 4468 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4469 if (flow_table && sock_flow_table) { 4470 struct rps_dev_flow *rflow; 4471 u32 next_cpu; 4472 u32 ident; 4473 4474 /* First check into global flow table if there is a match */ 4475 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4476 if ((ident ^ hash) & ~rps_cpu_mask) 4477 goto try_rps; 4478 4479 next_cpu = ident & rps_cpu_mask; 4480 4481 /* OK, now we know there is a match, 4482 * we can look at the local (per receive queue) flow table 4483 */ 4484 rflow = &flow_table->flows[hash & flow_table->mask]; 4485 tcpu = rflow->cpu; 4486 4487 /* 4488 * If the desired CPU (where last recvmsg was done) is 4489 * different from current CPU (one in the rx-queue flow 4490 * table entry), switch if one of the following holds: 4491 * - Current CPU is unset (>= nr_cpu_ids). 4492 * - Current CPU is offline. 4493 * - The current CPU's queue tail has advanced beyond the 4494 * last packet that was enqueued using this table entry. 4495 * This guarantees that all previous packets for the flow 4496 * have been dequeued, thus preserving in order delivery. 4497 */ 4498 if (unlikely(tcpu != next_cpu) && 4499 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4500 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4501 rflow->last_qtail)) >= 0)) { 4502 tcpu = next_cpu; 4503 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4504 } 4505 4506 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4507 *rflowp = rflow; 4508 cpu = tcpu; 4509 goto done; 4510 } 4511 } 4512 4513 try_rps: 4514 4515 if (map) { 4516 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4517 if (cpu_online(tcpu)) { 4518 cpu = tcpu; 4519 goto done; 4520 } 4521 } 4522 4523 done: 4524 return cpu; 4525 } 4526 4527 #ifdef CONFIG_RFS_ACCEL 4528 4529 /** 4530 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4531 * @dev: Device on which the filter was set 4532 * @rxq_index: RX queue index 4533 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4534 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4535 * 4536 * Drivers that implement ndo_rx_flow_steer() should periodically call 4537 * this function for each installed filter and remove the filters for 4538 * which it returns %true. 4539 */ 4540 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4541 u32 flow_id, u16 filter_id) 4542 { 4543 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4544 struct rps_dev_flow_table *flow_table; 4545 struct rps_dev_flow *rflow; 4546 bool expire = true; 4547 unsigned int cpu; 4548 4549 rcu_read_lock(); 4550 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4551 if (flow_table && flow_id <= flow_table->mask) { 4552 rflow = &flow_table->flows[flow_id]; 4553 cpu = READ_ONCE(rflow->cpu); 4554 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4555 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4556 rflow->last_qtail) < 4557 (int)(10 * flow_table->mask))) 4558 expire = false; 4559 } 4560 rcu_read_unlock(); 4561 return expire; 4562 } 4563 EXPORT_SYMBOL(rps_may_expire_flow); 4564 4565 #endif /* CONFIG_RFS_ACCEL */ 4566 4567 /* Called from hardirq (IPI) context */ 4568 static void rps_trigger_softirq(void *data) 4569 { 4570 struct softnet_data *sd = data; 4571 4572 ____napi_schedule(sd, &sd->backlog); 4573 sd->received_rps++; 4574 } 4575 4576 #endif /* CONFIG_RPS */ 4577 4578 /* Called from hardirq (IPI) context */ 4579 static void trigger_rx_softirq(void *data) 4580 { 4581 struct softnet_data *sd = data; 4582 4583 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4584 smp_store_release(&sd->defer_ipi_scheduled, 0); 4585 } 4586 4587 /* 4588 * After we queued a packet into sd->input_pkt_queue, 4589 * we need to make sure this queue is serviced soon. 4590 * 4591 * - If this is another cpu queue, link it to our rps_ipi_list, 4592 * and make sure we will process rps_ipi_list from net_rx_action(). 4593 * 4594 * - If this is our own queue, NAPI schedule our backlog. 4595 * Note that this also raises NET_RX_SOFTIRQ. 4596 */ 4597 static void napi_schedule_rps(struct softnet_data *sd) 4598 { 4599 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4600 4601 #ifdef CONFIG_RPS 4602 if (sd != mysd) { 4603 sd->rps_ipi_next = mysd->rps_ipi_list; 4604 mysd->rps_ipi_list = sd; 4605 4606 /* If not called from net_rx_action() 4607 * we have to raise NET_RX_SOFTIRQ. 4608 */ 4609 if (!mysd->in_net_rx_action) 4610 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4611 return; 4612 } 4613 #endif /* CONFIG_RPS */ 4614 __napi_schedule_irqoff(&mysd->backlog); 4615 } 4616 4617 #ifdef CONFIG_NET_FLOW_LIMIT 4618 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4619 #endif 4620 4621 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4622 { 4623 #ifdef CONFIG_NET_FLOW_LIMIT 4624 struct sd_flow_limit *fl; 4625 struct softnet_data *sd; 4626 unsigned int old_flow, new_flow; 4627 4628 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4629 return false; 4630 4631 sd = this_cpu_ptr(&softnet_data); 4632 4633 rcu_read_lock(); 4634 fl = rcu_dereference(sd->flow_limit); 4635 if (fl) { 4636 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4637 old_flow = fl->history[fl->history_head]; 4638 fl->history[fl->history_head] = new_flow; 4639 4640 fl->history_head++; 4641 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4642 4643 if (likely(fl->buckets[old_flow])) 4644 fl->buckets[old_flow]--; 4645 4646 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4647 fl->count++; 4648 rcu_read_unlock(); 4649 return true; 4650 } 4651 } 4652 rcu_read_unlock(); 4653 #endif 4654 return false; 4655 } 4656 4657 /* 4658 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4659 * queue (may be a remote CPU queue). 4660 */ 4661 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4662 unsigned int *qtail) 4663 { 4664 enum skb_drop_reason reason; 4665 struct softnet_data *sd; 4666 unsigned long flags; 4667 unsigned int qlen; 4668 4669 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4670 sd = &per_cpu(softnet_data, cpu); 4671 4672 rps_lock_irqsave(sd, &flags); 4673 if (!netif_running(skb->dev)) 4674 goto drop; 4675 qlen = skb_queue_len(&sd->input_pkt_queue); 4676 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4677 if (qlen) { 4678 enqueue: 4679 __skb_queue_tail(&sd->input_pkt_queue, skb); 4680 input_queue_tail_incr_save(sd, qtail); 4681 rps_unlock_irq_restore(sd, &flags); 4682 return NET_RX_SUCCESS; 4683 } 4684 4685 /* Schedule NAPI for backlog device 4686 * We can use non atomic operation since we own the queue lock 4687 */ 4688 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4689 napi_schedule_rps(sd); 4690 goto enqueue; 4691 } 4692 reason = SKB_DROP_REASON_CPU_BACKLOG; 4693 4694 drop: 4695 sd->dropped++; 4696 rps_unlock_irq_restore(sd, &flags); 4697 4698 dev_core_stats_rx_dropped_inc(skb->dev); 4699 kfree_skb_reason(skb, reason); 4700 return NET_RX_DROP; 4701 } 4702 4703 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4704 { 4705 struct net_device *dev = skb->dev; 4706 struct netdev_rx_queue *rxqueue; 4707 4708 rxqueue = dev->_rx; 4709 4710 if (skb_rx_queue_recorded(skb)) { 4711 u16 index = skb_get_rx_queue(skb); 4712 4713 if (unlikely(index >= dev->real_num_rx_queues)) { 4714 WARN_ONCE(dev->real_num_rx_queues > 1, 4715 "%s received packet on queue %u, but number " 4716 "of RX queues is %u\n", 4717 dev->name, index, dev->real_num_rx_queues); 4718 4719 return rxqueue; /* Return first rxqueue */ 4720 } 4721 rxqueue += index; 4722 } 4723 return rxqueue; 4724 } 4725 4726 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4727 struct bpf_prog *xdp_prog) 4728 { 4729 void *orig_data, *orig_data_end, *hard_start; 4730 struct netdev_rx_queue *rxqueue; 4731 bool orig_bcast, orig_host; 4732 u32 mac_len, frame_sz; 4733 __be16 orig_eth_type; 4734 struct ethhdr *eth; 4735 u32 metalen, act; 4736 int off; 4737 4738 /* The XDP program wants to see the packet starting at the MAC 4739 * header. 4740 */ 4741 mac_len = skb->data - skb_mac_header(skb); 4742 hard_start = skb->data - skb_headroom(skb); 4743 4744 /* SKB "head" area always have tailroom for skb_shared_info */ 4745 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4746 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4747 4748 rxqueue = netif_get_rxqueue(skb); 4749 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4750 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4751 skb_headlen(skb) + mac_len, true); 4752 4753 orig_data_end = xdp->data_end; 4754 orig_data = xdp->data; 4755 eth = (struct ethhdr *)xdp->data; 4756 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4757 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4758 orig_eth_type = eth->h_proto; 4759 4760 act = bpf_prog_run_xdp(xdp_prog, xdp); 4761 4762 /* check if bpf_xdp_adjust_head was used */ 4763 off = xdp->data - orig_data; 4764 if (off) { 4765 if (off > 0) 4766 __skb_pull(skb, off); 4767 else if (off < 0) 4768 __skb_push(skb, -off); 4769 4770 skb->mac_header += off; 4771 skb_reset_network_header(skb); 4772 } 4773 4774 /* check if bpf_xdp_adjust_tail was used */ 4775 off = xdp->data_end - orig_data_end; 4776 if (off != 0) { 4777 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4778 skb->len += off; /* positive on grow, negative on shrink */ 4779 } 4780 4781 /* check if XDP changed eth hdr such SKB needs update */ 4782 eth = (struct ethhdr *)xdp->data; 4783 if ((orig_eth_type != eth->h_proto) || 4784 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4785 skb->dev->dev_addr)) || 4786 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4787 __skb_push(skb, ETH_HLEN); 4788 skb->pkt_type = PACKET_HOST; 4789 skb->protocol = eth_type_trans(skb, skb->dev); 4790 } 4791 4792 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4793 * before calling us again on redirect path. We do not call do_redirect 4794 * as we leave that up to the caller. 4795 * 4796 * Caller is responsible for managing lifetime of skb (i.e. calling 4797 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4798 */ 4799 switch (act) { 4800 case XDP_REDIRECT: 4801 case XDP_TX: 4802 __skb_push(skb, mac_len); 4803 break; 4804 case XDP_PASS: 4805 metalen = xdp->data - xdp->data_meta; 4806 if (metalen) 4807 skb_metadata_set(skb, metalen); 4808 break; 4809 } 4810 4811 return act; 4812 } 4813 4814 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4815 struct xdp_buff *xdp, 4816 struct bpf_prog *xdp_prog) 4817 { 4818 u32 act = XDP_DROP; 4819 4820 /* Reinjected packets coming from act_mirred or similar should 4821 * not get XDP generic processing. 4822 */ 4823 if (skb_is_redirected(skb)) 4824 return XDP_PASS; 4825 4826 /* XDP packets must be linear and must have sufficient headroom 4827 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4828 * native XDP provides, thus we need to do it here as well. 4829 */ 4830 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4831 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4832 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4833 int troom = skb->tail + skb->data_len - skb->end; 4834 4835 /* In case we have to go down the path and also linearize, 4836 * then lets do the pskb_expand_head() work just once here. 4837 */ 4838 if (pskb_expand_head(skb, 4839 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4840 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4841 goto do_drop; 4842 if (skb_linearize(skb)) 4843 goto do_drop; 4844 } 4845 4846 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4847 switch (act) { 4848 case XDP_REDIRECT: 4849 case XDP_TX: 4850 case XDP_PASS: 4851 break; 4852 default: 4853 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4854 fallthrough; 4855 case XDP_ABORTED: 4856 trace_xdp_exception(skb->dev, xdp_prog, act); 4857 fallthrough; 4858 case XDP_DROP: 4859 do_drop: 4860 kfree_skb(skb); 4861 break; 4862 } 4863 4864 return act; 4865 } 4866 4867 /* When doing generic XDP we have to bypass the qdisc layer and the 4868 * network taps in order to match in-driver-XDP behavior. This also means 4869 * that XDP packets are able to starve other packets going through a qdisc, 4870 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4871 * queues, so they do not have this starvation issue. 4872 */ 4873 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4874 { 4875 struct net_device *dev = skb->dev; 4876 struct netdev_queue *txq; 4877 bool free_skb = true; 4878 int cpu, rc; 4879 4880 txq = netdev_core_pick_tx(dev, skb, NULL); 4881 cpu = smp_processor_id(); 4882 HARD_TX_LOCK(dev, txq, cpu); 4883 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4884 rc = netdev_start_xmit(skb, dev, txq, 0); 4885 if (dev_xmit_complete(rc)) 4886 free_skb = false; 4887 } 4888 HARD_TX_UNLOCK(dev, txq); 4889 if (free_skb) { 4890 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4891 dev_core_stats_tx_dropped_inc(dev); 4892 kfree_skb(skb); 4893 } 4894 } 4895 4896 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4897 4898 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4899 { 4900 if (xdp_prog) { 4901 struct xdp_buff xdp; 4902 u32 act; 4903 int err; 4904 4905 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4906 if (act != XDP_PASS) { 4907 switch (act) { 4908 case XDP_REDIRECT: 4909 err = xdp_do_generic_redirect(skb->dev, skb, 4910 &xdp, xdp_prog); 4911 if (err) 4912 goto out_redir; 4913 break; 4914 case XDP_TX: 4915 generic_xdp_tx(skb, xdp_prog); 4916 break; 4917 } 4918 return XDP_DROP; 4919 } 4920 } 4921 return XDP_PASS; 4922 out_redir: 4923 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4924 return XDP_DROP; 4925 } 4926 EXPORT_SYMBOL_GPL(do_xdp_generic); 4927 4928 static int netif_rx_internal(struct sk_buff *skb) 4929 { 4930 int ret; 4931 4932 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4933 4934 trace_netif_rx(skb); 4935 4936 #ifdef CONFIG_RPS 4937 if (static_branch_unlikely(&rps_needed)) { 4938 struct rps_dev_flow voidflow, *rflow = &voidflow; 4939 int cpu; 4940 4941 rcu_read_lock(); 4942 4943 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4944 if (cpu < 0) 4945 cpu = smp_processor_id(); 4946 4947 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4948 4949 rcu_read_unlock(); 4950 } else 4951 #endif 4952 { 4953 unsigned int qtail; 4954 4955 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4956 } 4957 return ret; 4958 } 4959 4960 /** 4961 * __netif_rx - Slightly optimized version of netif_rx 4962 * @skb: buffer to post 4963 * 4964 * This behaves as netif_rx except that it does not disable bottom halves. 4965 * As a result this function may only be invoked from the interrupt context 4966 * (either hard or soft interrupt). 4967 */ 4968 int __netif_rx(struct sk_buff *skb) 4969 { 4970 int ret; 4971 4972 lockdep_assert_once(hardirq_count() | softirq_count()); 4973 4974 trace_netif_rx_entry(skb); 4975 ret = netif_rx_internal(skb); 4976 trace_netif_rx_exit(ret); 4977 return ret; 4978 } 4979 EXPORT_SYMBOL(__netif_rx); 4980 4981 /** 4982 * netif_rx - post buffer to the network code 4983 * @skb: buffer to post 4984 * 4985 * This function receives a packet from a device driver and queues it for 4986 * the upper (protocol) levels to process via the backlog NAPI device. It 4987 * always succeeds. The buffer may be dropped during processing for 4988 * congestion control or by the protocol layers. 4989 * The network buffer is passed via the backlog NAPI device. Modern NIC 4990 * driver should use NAPI and GRO. 4991 * This function can used from interrupt and from process context. The 4992 * caller from process context must not disable interrupts before invoking 4993 * this function. 4994 * 4995 * return values: 4996 * NET_RX_SUCCESS (no congestion) 4997 * NET_RX_DROP (packet was dropped) 4998 * 4999 */ 5000 int netif_rx(struct sk_buff *skb) 5001 { 5002 bool need_bh_off = !(hardirq_count() | softirq_count()); 5003 int ret; 5004 5005 if (need_bh_off) 5006 local_bh_disable(); 5007 trace_netif_rx_entry(skb); 5008 ret = netif_rx_internal(skb); 5009 trace_netif_rx_exit(ret); 5010 if (need_bh_off) 5011 local_bh_enable(); 5012 return ret; 5013 } 5014 EXPORT_SYMBOL(netif_rx); 5015 5016 static __latent_entropy void net_tx_action(struct softirq_action *h) 5017 { 5018 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5019 5020 if (sd->completion_queue) { 5021 struct sk_buff *clist; 5022 5023 local_irq_disable(); 5024 clist = sd->completion_queue; 5025 sd->completion_queue = NULL; 5026 local_irq_enable(); 5027 5028 while (clist) { 5029 struct sk_buff *skb = clist; 5030 5031 clist = clist->next; 5032 5033 WARN_ON(refcount_read(&skb->users)); 5034 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5035 trace_consume_skb(skb, net_tx_action); 5036 else 5037 trace_kfree_skb(skb, net_tx_action, 5038 get_kfree_skb_cb(skb)->reason); 5039 5040 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5041 __kfree_skb(skb); 5042 else 5043 __kfree_skb_defer(skb); 5044 } 5045 } 5046 5047 if (sd->output_queue) { 5048 struct Qdisc *head; 5049 5050 local_irq_disable(); 5051 head = sd->output_queue; 5052 sd->output_queue = NULL; 5053 sd->output_queue_tailp = &sd->output_queue; 5054 local_irq_enable(); 5055 5056 rcu_read_lock(); 5057 5058 while (head) { 5059 struct Qdisc *q = head; 5060 spinlock_t *root_lock = NULL; 5061 5062 head = head->next_sched; 5063 5064 /* We need to make sure head->next_sched is read 5065 * before clearing __QDISC_STATE_SCHED 5066 */ 5067 smp_mb__before_atomic(); 5068 5069 if (!(q->flags & TCQ_F_NOLOCK)) { 5070 root_lock = qdisc_lock(q); 5071 spin_lock(root_lock); 5072 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5073 &q->state))) { 5074 /* There is a synchronize_net() between 5075 * STATE_DEACTIVATED flag being set and 5076 * qdisc_reset()/some_qdisc_is_busy() in 5077 * dev_deactivate(), so we can safely bail out 5078 * early here to avoid data race between 5079 * qdisc_deactivate() and some_qdisc_is_busy() 5080 * for lockless qdisc. 5081 */ 5082 clear_bit(__QDISC_STATE_SCHED, &q->state); 5083 continue; 5084 } 5085 5086 clear_bit(__QDISC_STATE_SCHED, &q->state); 5087 qdisc_run(q); 5088 if (root_lock) 5089 spin_unlock(root_lock); 5090 } 5091 5092 rcu_read_unlock(); 5093 } 5094 5095 xfrm_dev_backlog(sd); 5096 } 5097 5098 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5099 /* This hook is defined here for ATM LANE */ 5100 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5101 unsigned char *addr) __read_mostly; 5102 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5103 #endif 5104 5105 static inline struct sk_buff * 5106 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5107 struct net_device *orig_dev, bool *another) 5108 { 5109 #ifdef CONFIG_NET_CLS_ACT 5110 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5111 struct tcf_result cl_res; 5112 5113 /* If there's at least one ingress present somewhere (so 5114 * we get here via enabled static key), remaining devices 5115 * that are not configured with an ingress qdisc will bail 5116 * out here. 5117 */ 5118 if (!miniq) 5119 return skb; 5120 5121 if (*pt_prev) { 5122 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5123 *pt_prev = NULL; 5124 } 5125 5126 qdisc_skb_cb(skb)->pkt_len = skb->len; 5127 tc_skb_cb(skb)->mru = 0; 5128 tc_skb_cb(skb)->post_ct = false; 5129 skb->tc_at_ingress = 1; 5130 mini_qdisc_bstats_cpu_update(miniq, skb); 5131 5132 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5133 case TC_ACT_OK: 5134 case TC_ACT_RECLASSIFY: 5135 skb->tc_index = TC_H_MIN(cl_res.classid); 5136 break; 5137 case TC_ACT_SHOT: 5138 mini_qdisc_qstats_cpu_drop(miniq); 5139 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5140 *ret = NET_RX_DROP; 5141 return NULL; 5142 case TC_ACT_STOLEN: 5143 case TC_ACT_QUEUED: 5144 case TC_ACT_TRAP: 5145 consume_skb(skb); 5146 *ret = NET_RX_SUCCESS; 5147 return NULL; 5148 case TC_ACT_REDIRECT: 5149 /* skb_mac_header check was done by cls/act_bpf, so 5150 * we can safely push the L2 header back before 5151 * redirecting to another netdev 5152 */ 5153 __skb_push(skb, skb->mac_len); 5154 if (skb_do_redirect(skb) == -EAGAIN) { 5155 __skb_pull(skb, skb->mac_len); 5156 *another = true; 5157 break; 5158 } 5159 *ret = NET_RX_SUCCESS; 5160 return NULL; 5161 case TC_ACT_CONSUMED: 5162 *ret = NET_RX_SUCCESS; 5163 return NULL; 5164 default: 5165 break; 5166 } 5167 #endif /* CONFIG_NET_CLS_ACT */ 5168 return skb; 5169 } 5170 5171 /** 5172 * netdev_is_rx_handler_busy - check if receive handler is registered 5173 * @dev: device to check 5174 * 5175 * Check if a receive handler is already registered for a given device. 5176 * Return true if there one. 5177 * 5178 * The caller must hold the rtnl_mutex. 5179 */ 5180 bool netdev_is_rx_handler_busy(struct net_device *dev) 5181 { 5182 ASSERT_RTNL(); 5183 return dev && rtnl_dereference(dev->rx_handler); 5184 } 5185 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5186 5187 /** 5188 * netdev_rx_handler_register - register receive handler 5189 * @dev: device to register a handler for 5190 * @rx_handler: receive handler to register 5191 * @rx_handler_data: data pointer that is used by rx handler 5192 * 5193 * Register a receive handler for a device. This handler will then be 5194 * called from __netif_receive_skb. A negative errno code is returned 5195 * on a failure. 5196 * 5197 * The caller must hold the rtnl_mutex. 5198 * 5199 * For a general description of rx_handler, see enum rx_handler_result. 5200 */ 5201 int netdev_rx_handler_register(struct net_device *dev, 5202 rx_handler_func_t *rx_handler, 5203 void *rx_handler_data) 5204 { 5205 if (netdev_is_rx_handler_busy(dev)) 5206 return -EBUSY; 5207 5208 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5209 return -EINVAL; 5210 5211 /* Note: rx_handler_data must be set before rx_handler */ 5212 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5213 rcu_assign_pointer(dev->rx_handler, rx_handler); 5214 5215 return 0; 5216 } 5217 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5218 5219 /** 5220 * netdev_rx_handler_unregister - unregister receive handler 5221 * @dev: device to unregister a handler from 5222 * 5223 * Unregister a receive handler from a device. 5224 * 5225 * The caller must hold the rtnl_mutex. 5226 */ 5227 void netdev_rx_handler_unregister(struct net_device *dev) 5228 { 5229 5230 ASSERT_RTNL(); 5231 RCU_INIT_POINTER(dev->rx_handler, NULL); 5232 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5233 * section has a guarantee to see a non NULL rx_handler_data 5234 * as well. 5235 */ 5236 synchronize_net(); 5237 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5238 } 5239 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5240 5241 /* 5242 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5243 * the special handling of PFMEMALLOC skbs. 5244 */ 5245 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5246 { 5247 switch (skb->protocol) { 5248 case htons(ETH_P_ARP): 5249 case htons(ETH_P_IP): 5250 case htons(ETH_P_IPV6): 5251 case htons(ETH_P_8021Q): 5252 case htons(ETH_P_8021AD): 5253 return true; 5254 default: 5255 return false; 5256 } 5257 } 5258 5259 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5260 int *ret, struct net_device *orig_dev) 5261 { 5262 if (nf_hook_ingress_active(skb)) { 5263 int ingress_retval; 5264 5265 if (*pt_prev) { 5266 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5267 *pt_prev = NULL; 5268 } 5269 5270 rcu_read_lock(); 5271 ingress_retval = nf_hook_ingress(skb); 5272 rcu_read_unlock(); 5273 return ingress_retval; 5274 } 5275 return 0; 5276 } 5277 5278 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5279 struct packet_type **ppt_prev) 5280 { 5281 struct packet_type *ptype, *pt_prev; 5282 rx_handler_func_t *rx_handler; 5283 struct sk_buff *skb = *pskb; 5284 struct net_device *orig_dev; 5285 bool deliver_exact = false; 5286 int ret = NET_RX_DROP; 5287 __be16 type; 5288 5289 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5290 5291 trace_netif_receive_skb(skb); 5292 5293 orig_dev = skb->dev; 5294 5295 skb_reset_network_header(skb); 5296 if (!skb_transport_header_was_set(skb)) 5297 skb_reset_transport_header(skb); 5298 skb_reset_mac_len(skb); 5299 5300 pt_prev = NULL; 5301 5302 another_round: 5303 skb->skb_iif = skb->dev->ifindex; 5304 5305 __this_cpu_inc(softnet_data.processed); 5306 5307 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5308 int ret2; 5309 5310 migrate_disable(); 5311 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5312 migrate_enable(); 5313 5314 if (ret2 != XDP_PASS) { 5315 ret = NET_RX_DROP; 5316 goto out; 5317 } 5318 } 5319 5320 if (eth_type_vlan(skb->protocol)) { 5321 skb = skb_vlan_untag(skb); 5322 if (unlikely(!skb)) 5323 goto out; 5324 } 5325 5326 if (skb_skip_tc_classify(skb)) 5327 goto skip_classify; 5328 5329 if (pfmemalloc) 5330 goto skip_taps; 5331 5332 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5333 if (pt_prev) 5334 ret = deliver_skb(skb, pt_prev, orig_dev); 5335 pt_prev = ptype; 5336 } 5337 5338 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5339 if (pt_prev) 5340 ret = deliver_skb(skb, pt_prev, orig_dev); 5341 pt_prev = ptype; 5342 } 5343 5344 skip_taps: 5345 #ifdef CONFIG_NET_INGRESS 5346 if (static_branch_unlikely(&ingress_needed_key)) { 5347 bool another = false; 5348 5349 nf_skip_egress(skb, true); 5350 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5351 &another); 5352 if (another) 5353 goto another_round; 5354 if (!skb) 5355 goto out; 5356 5357 nf_skip_egress(skb, false); 5358 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5359 goto out; 5360 } 5361 #endif 5362 skb_reset_redirect(skb); 5363 skip_classify: 5364 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5365 goto drop; 5366 5367 if (skb_vlan_tag_present(skb)) { 5368 if (pt_prev) { 5369 ret = deliver_skb(skb, pt_prev, orig_dev); 5370 pt_prev = NULL; 5371 } 5372 if (vlan_do_receive(&skb)) 5373 goto another_round; 5374 else if (unlikely(!skb)) 5375 goto out; 5376 } 5377 5378 rx_handler = rcu_dereference(skb->dev->rx_handler); 5379 if (rx_handler) { 5380 if (pt_prev) { 5381 ret = deliver_skb(skb, pt_prev, orig_dev); 5382 pt_prev = NULL; 5383 } 5384 switch (rx_handler(&skb)) { 5385 case RX_HANDLER_CONSUMED: 5386 ret = NET_RX_SUCCESS; 5387 goto out; 5388 case RX_HANDLER_ANOTHER: 5389 goto another_round; 5390 case RX_HANDLER_EXACT: 5391 deliver_exact = true; 5392 break; 5393 case RX_HANDLER_PASS: 5394 break; 5395 default: 5396 BUG(); 5397 } 5398 } 5399 5400 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5401 check_vlan_id: 5402 if (skb_vlan_tag_get_id(skb)) { 5403 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5404 * find vlan device. 5405 */ 5406 skb->pkt_type = PACKET_OTHERHOST; 5407 } else if (eth_type_vlan(skb->protocol)) { 5408 /* Outer header is 802.1P with vlan 0, inner header is 5409 * 802.1Q or 802.1AD and vlan_do_receive() above could 5410 * not find vlan dev for vlan id 0. 5411 */ 5412 __vlan_hwaccel_clear_tag(skb); 5413 skb = skb_vlan_untag(skb); 5414 if (unlikely(!skb)) 5415 goto out; 5416 if (vlan_do_receive(&skb)) 5417 /* After stripping off 802.1P header with vlan 0 5418 * vlan dev is found for inner header. 5419 */ 5420 goto another_round; 5421 else if (unlikely(!skb)) 5422 goto out; 5423 else 5424 /* We have stripped outer 802.1P vlan 0 header. 5425 * But could not find vlan dev. 5426 * check again for vlan id to set OTHERHOST. 5427 */ 5428 goto check_vlan_id; 5429 } 5430 /* Note: we might in the future use prio bits 5431 * and set skb->priority like in vlan_do_receive() 5432 * For the time being, just ignore Priority Code Point 5433 */ 5434 __vlan_hwaccel_clear_tag(skb); 5435 } 5436 5437 type = skb->protocol; 5438 5439 /* deliver only exact match when indicated */ 5440 if (likely(!deliver_exact)) { 5441 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5442 &ptype_base[ntohs(type) & 5443 PTYPE_HASH_MASK]); 5444 } 5445 5446 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5447 &orig_dev->ptype_specific); 5448 5449 if (unlikely(skb->dev != orig_dev)) { 5450 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5451 &skb->dev->ptype_specific); 5452 } 5453 5454 if (pt_prev) { 5455 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5456 goto drop; 5457 *ppt_prev = pt_prev; 5458 } else { 5459 drop: 5460 if (!deliver_exact) 5461 dev_core_stats_rx_dropped_inc(skb->dev); 5462 else 5463 dev_core_stats_rx_nohandler_inc(skb->dev); 5464 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5465 /* Jamal, now you will not able to escape explaining 5466 * me how you were going to use this. :-) 5467 */ 5468 ret = NET_RX_DROP; 5469 } 5470 5471 out: 5472 /* The invariant here is that if *ppt_prev is not NULL 5473 * then skb should also be non-NULL. 5474 * 5475 * Apparently *ppt_prev assignment above holds this invariant due to 5476 * skb dereferencing near it. 5477 */ 5478 *pskb = skb; 5479 return ret; 5480 } 5481 5482 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5483 { 5484 struct net_device *orig_dev = skb->dev; 5485 struct packet_type *pt_prev = NULL; 5486 int ret; 5487 5488 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5489 if (pt_prev) 5490 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5491 skb->dev, pt_prev, orig_dev); 5492 return ret; 5493 } 5494 5495 /** 5496 * netif_receive_skb_core - special purpose version of netif_receive_skb 5497 * @skb: buffer to process 5498 * 5499 * More direct receive version of netif_receive_skb(). It should 5500 * only be used by callers that have a need to skip RPS and Generic XDP. 5501 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5502 * 5503 * This function may only be called from softirq context and interrupts 5504 * should be enabled. 5505 * 5506 * Return values (usually ignored): 5507 * NET_RX_SUCCESS: no congestion 5508 * NET_RX_DROP: packet was dropped 5509 */ 5510 int netif_receive_skb_core(struct sk_buff *skb) 5511 { 5512 int ret; 5513 5514 rcu_read_lock(); 5515 ret = __netif_receive_skb_one_core(skb, false); 5516 rcu_read_unlock(); 5517 5518 return ret; 5519 } 5520 EXPORT_SYMBOL(netif_receive_skb_core); 5521 5522 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5523 struct packet_type *pt_prev, 5524 struct net_device *orig_dev) 5525 { 5526 struct sk_buff *skb, *next; 5527 5528 if (!pt_prev) 5529 return; 5530 if (list_empty(head)) 5531 return; 5532 if (pt_prev->list_func != NULL) 5533 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5534 ip_list_rcv, head, pt_prev, orig_dev); 5535 else 5536 list_for_each_entry_safe(skb, next, head, list) { 5537 skb_list_del_init(skb); 5538 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5539 } 5540 } 5541 5542 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5543 { 5544 /* Fast-path assumptions: 5545 * - There is no RX handler. 5546 * - Only one packet_type matches. 5547 * If either of these fails, we will end up doing some per-packet 5548 * processing in-line, then handling the 'last ptype' for the whole 5549 * sublist. This can't cause out-of-order delivery to any single ptype, 5550 * because the 'last ptype' must be constant across the sublist, and all 5551 * other ptypes are handled per-packet. 5552 */ 5553 /* Current (common) ptype of sublist */ 5554 struct packet_type *pt_curr = NULL; 5555 /* Current (common) orig_dev of sublist */ 5556 struct net_device *od_curr = NULL; 5557 struct list_head sublist; 5558 struct sk_buff *skb, *next; 5559 5560 INIT_LIST_HEAD(&sublist); 5561 list_for_each_entry_safe(skb, next, head, list) { 5562 struct net_device *orig_dev = skb->dev; 5563 struct packet_type *pt_prev = NULL; 5564 5565 skb_list_del_init(skb); 5566 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5567 if (!pt_prev) 5568 continue; 5569 if (pt_curr != pt_prev || od_curr != orig_dev) { 5570 /* dispatch old sublist */ 5571 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5572 /* start new sublist */ 5573 INIT_LIST_HEAD(&sublist); 5574 pt_curr = pt_prev; 5575 od_curr = orig_dev; 5576 } 5577 list_add_tail(&skb->list, &sublist); 5578 } 5579 5580 /* dispatch final sublist */ 5581 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5582 } 5583 5584 static int __netif_receive_skb(struct sk_buff *skb) 5585 { 5586 int ret; 5587 5588 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5589 unsigned int noreclaim_flag; 5590 5591 /* 5592 * PFMEMALLOC skbs are special, they should 5593 * - be delivered to SOCK_MEMALLOC sockets only 5594 * - stay away from userspace 5595 * - have bounded memory usage 5596 * 5597 * Use PF_MEMALLOC as this saves us from propagating the allocation 5598 * context down to all allocation sites. 5599 */ 5600 noreclaim_flag = memalloc_noreclaim_save(); 5601 ret = __netif_receive_skb_one_core(skb, true); 5602 memalloc_noreclaim_restore(noreclaim_flag); 5603 } else 5604 ret = __netif_receive_skb_one_core(skb, false); 5605 5606 return ret; 5607 } 5608 5609 static void __netif_receive_skb_list(struct list_head *head) 5610 { 5611 unsigned long noreclaim_flag = 0; 5612 struct sk_buff *skb, *next; 5613 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5614 5615 list_for_each_entry_safe(skb, next, head, list) { 5616 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5617 struct list_head sublist; 5618 5619 /* Handle the previous sublist */ 5620 list_cut_before(&sublist, head, &skb->list); 5621 if (!list_empty(&sublist)) 5622 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5623 pfmemalloc = !pfmemalloc; 5624 /* See comments in __netif_receive_skb */ 5625 if (pfmemalloc) 5626 noreclaim_flag = memalloc_noreclaim_save(); 5627 else 5628 memalloc_noreclaim_restore(noreclaim_flag); 5629 } 5630 } 5631 /* Handle the remaining sublist */ 5632 if (!list_empty(head)) 5633 __netif_receive_skb_list_core(head, pfmemalloc); 5634 /* Restore pflags */ 5635 if (pfmemalloc) 5636 memalloc_noreclaim_restore(noreclaim_flag); 5637 } 5638 5639 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5640 { 5641 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5642 struct bpf_prog *new = xdp->prog; 5643 int ret = 0; 5644 5645 switch (xdp->command) { 5646 case XDP_SETUP_PROG: 5647 rcu_assign_pointer(dev->xdp_prog, new); 5648 if (old) 5649 bpf_prog_put(old); 5650 5651 if (old && !new) { 5652 static_branch_dec(&generic_xdp_needed_key); 5653 } else if (new && !old) { 5654 static_branch_inc(&generic_xdp_needed_key); 5655 dev_disable_lro(dev); 5656 dev_disable_gro_hw(dev); 5657 } 5658 break; 5659 5660 default: 5661 ret = -EINVAL; 5662 break; 5663 } 5664 5665 return ret; 5666 } 5667 5668 static int netif_receive_skb_internal(struct sk_buff *skb) 5669 { 5670 int ret; 5671 5672 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5673 5674 if (skb_defer_rx_timestamp(skb)) 5675 return NET_RX_SUCCESS; 5676 5677 rcu_read_lock(); 5678 #ifdef CONFIG_RPS 5679 if (static_branch_unlikely(&rps_needed)) { 5680 struct rps_dev_flow voidflow, *rflow = &voidflow; 5681 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5682 5683 if (cpu >= 0) { 5684 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5685 rcu_read_unlock(); 5686 return ret; 5687 } 5688 } 5689 #endif 5690 ret = __netif_receive_skb(skb); 5691 rcu_read_unlock(); 5692 return ret; 5693 } 5694 5695 void netif_receive_skb_list_internal(struct list_head *head) 5696 { 5697 struct sk_buff *skb, *next; 5698 struct list_head sublist; 5699 5700 INIT_LIST_HEAD(&sublist); 5701 list_for_each_entry_safe(skb, next, head, list) { 5702 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5703 skb_list_del_init(skb); 5704 if (!skb_defer_rx_timestamp(skb)) 5705 list_add_tail(&skb->list, &sublist); 5706 } 5707 list_splice_init(&sublist, head); 5708 5709 rcu_read_lock(); 5710 #ifdef CONFIG_RPS 5711 if (static_branch_unlikely(&rps_needed)) { 5712 list_for_each_entry_safe(skb, next, head, list) { 5713 struct rps_dev_flow voidflow, *rflow = &voidflow; 5714 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5715 5716 if (cpu >= 0) { 5717 /* Will be handled, remove from list */ 5718 skb_list_del_init(skb); 5719 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5720 } 5721 } 5722 } 5723 #endif 5724 __netif_receive_skb_list(head); 5725 rcu_read_unlock(); 5726 } 5727 5728 /** 5729 * netif_receive_skb - process receive buffer from network 5730 * @skb: buffer to process 5731 * 5732 * netif_receive_skb() is the main receive data processing function. 5733 * It always succeeds. The buffer may be dropped during processing 5734 * for congestion control or by the protocol layers. 5735 * 5736 * This function may only be called from softirq context and interrupts 5737 * should be enabled. 5738 * 5739 * Return values (usually ignored): 5740 * NET_RX_SUCCESS: no congestion 5741 * NET_RX_DROP: packet was dropped 5742 */ 5743 int netif_receive_skb(struct sk_buff *skb) 5744 { 5745 int ret; 5746 5747 trace_netif_receive_skb_entry(skb); 5748 5749 ret = netif_receive_skb_internal(skb); 5750 trace_netif_receive_skb_exit(ret); 5751 5752 return ret; 5753 } 5754 EXPORT_SYMBOL(netif_receive_skb); 5755 5756 /** 5757 * netif_receive_skb_list - process many receive buffers from network 5758 * @head: list of skbs to process. 5759 * 5760 * Since return value of netif_receive_skb() is normally ignored, and 5761 * wouldn't be meaningful for a list, this function returns void. 5762 * 5763 * This function may only be called from softirq context and interrupts 5764 * should be enabled. 5765 */ 5766 void netif_receive_skb_list(struct list_head *head) 5767 { 5768 struct sk_buff *skb; 5769 5770 if (list_empty(head)) 5771 return; 5772 if (trace_netif_receive_skb_list_entry_enabled()) { 5773 list_for_each_entry(skb, head, list) 5774 trace_netif_receive_skb_list_entry(skb); 5775 } 5776 netif_receive_skb_list_internal(head); 5777 trace_netif_receive_skb_list_exit(0); 5778 } 5779 EXPORT_SYMBOL(netif_receive_skb_list); 5780 5781 static DEFINE_PER_CPU(struct work_struct, flush_works); 5782 5783 /* Network device is going away, flush any packets still pending */ 5784 static void flush_backlog(struct work_struct *work) 5785 { 5786 struct sk_buff *skb, *tmp; 5787 struct softnet_data *sd; 5788 5789 local_bh_disable(); 5790 sd = this_cpu_ptr(&softnet_data); 5791 5792 rps_lock_irq_disable(sd); 5793 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5794 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5795 __skb_unlink(skb, &sd->input_pkt_queue); 5796 dev_kfree_skb_irq(skb); 5797 input_queue_head_incr(sd); 5798 } 5799 } 5800 rps_unlock_irq_enable(sd); 5801 5802 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5803 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5804 __skb_unlink(skb, &sd->process_queue); 5805 kfree_skb(skb); 5806 input_queue_head_incr(sd); 5807 } 5808 } 5809 local_bh_enable(); 5810 } 5811 5812 static bool flush_required(int cpu) 5813 { 5814 #if IS_ENABLED(CONFIG_RPS) 5815 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5816 bool do_flush; 5817 5818 rps_lock_irq_disable(sd); 5819 5820 /* as insertion into process_queue happens with the rps lock held, 5821 * process_queue access may race only with dequeue 5822 */ 5823 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5824 !skb_queue_empty_lockless(&sd->process_queue); 5825 rps_unlock_irq_enable(sd); 5826 5827 return do_flush; 5828 #endif 5829 /* without RPS we can't safely check input_pkt_queue: during a 5830 * concurrent remote skb_queue_splice() we can detect as empty both 5831 * input_pkt_queue and process_queue even if the latter could end-up 5832 * containing a lot of packets. 5833 */ 5834 return true; 5835 } 5836 5837 static void flush_all_backlogs(void) 5838 { 5839 static cpumask_t flush_cpus; 5840 unsigned int cpu; 5841 5842 /* since we are under rtnl lock protection we can use static data 5843 * for the cpumask and avoid allocating on stack the possibly 5844 * large mask 5845 */ 5846 ASSERT_RTNL(); 5847 5848 cpus_read_lock(); 5849 5850 cpumask_clear(&flush_cpus); 5851 for_each_online_cpu(cpu) { 5852 if (flush_required(cpu)) { 5853 queue_work_on(cpu, system_highpri_wq, 5854 per_cpu_ptr(&flush_works, cpu)); 5855 cpumask_set_cpu(cpu, &flush_cpus); 5856 } 5857 } 5858 5859 /* we can have in flight packet[s] on the cpus we are not flushing, 5860 * synchronize_net() in unregister_netdevice_many() will take care of 5861 * them 5862 */ 5863 for_each_cpu(cpu, &flush_cpus) 5864 flush_work(per_cpu_ptr(&flush_works, cpu)); 5865 5866 cpus_read_unlock(); 5867 } 5868 5869 static void net_rps_send_ipi(struct softnet_data *remsd) 5870 { 5871 #ifdef CONFIG_RPS 5872 while (remsd) { 5873 struct softnet_data *next = remsd->rps_ipi_next; 5874 5875 if (cpu_online(remsd->cpu)) 5876 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5877 remsd = next; 5878 } 5879 #endif 5880 } 5881 5882 /* 5883 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5884 * Note: called with local irq disabled, but exits with local irq enabled. 5885 */ 5886 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5887 { 5888 #ifdef CONFIG_RPS 5889 struct softnet_data *remsd = sd->rps_ipi_list; 5890 5891 if (remsd) { 5892 sd->rps_ipi_list = NULL; 5893 5894 local_irq_enable(); 5895 5896 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5897 net_rps_send_ipi(remsd); 5898 } else 5899 #endif 5900 local_irq_enable(); 5901 } 5902 5903 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5904 { 5905 #ifdef CONFIG_RPS 5906 return sd->rps_ipi_list != NULL; 5907 #else 5908 return false; 5909 #endif 5910 } 5911 5912 static int process_backlog(struct napi_struct *napi, int quota) 5913 { 5914 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5915 bool again = true; 5916 int work = 0; 5917 5918 /* Check if we have pending ipi, its better to send them now, 5919 * not waiting net_rx_action() end. 5920 */ 5921 if (sd_has_rps_ipi_waiting(sd)) { 5922 local_irq_disable(); 5923 net_rps_action_and_irq_enable(sd); 5924 } 5925 5926 napi->weight = READ_ONCE(dev_rx_weight); 5927 while (again) { 5928 struct sk_buff *skb; 5929 5930 while ((skb = __skb_dequeue(&sd->process_queue))) { 5931 rcu_read_lock(); 5932 __netif_receive_skb(skb); 5933 rcu_read_unlock(); 5934 input_queue_head_incr(sd); 5935 if (++work >= quota) 5936 return work; 5937 5938 } 5939 5940 rps_lock_irq_disable(sd); 5941 if (skb_queue_empty(&sd->input_pkt_queue)) { 5942 /* 5943 * Inline a custom version of __napi_complete(). 5944 * only current cpu owns and manipulates this napi, 5945 * and NAPI_STATE_SCHED is the only possible flag set 5946 * on backlog. 5947 * We can use a plain write instead of clear_bit(), 5948 * and we dont need an smp_mb() memory barrier. 5949 */ 5950 napi->state = 0; 5951 again = false; 5952 } else { 5953 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5954 &sd->process_queue); 5955 } 5956 rps_unlock_irq_enable(sd); 5957 } 5958 5959 return work; 5960 } 5961 5962 /** 5963 * __napi_schedule - schedule for receive 5964 * @n: entry to schedule 5965 * 5966 * The entry's receive function will be scheduled to run. 5967 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5968 */ 5969 void __napi_schedule(struct napi_struct *n) 5970 { 5971 unsigned long flags; 5972 5973 local_irq_save(flags); 5974 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5975 local_irq_restore(flags); 5976 } 5977 EXPORT_SYMBOL(__napi_schedule); 5978 5979 /** 5980 * napi_schedule_prep - check if napi can be scheduled 5981 * @n: napi context 5982 * 5983 * Test if NAPI routine is already running, and if not mark 5984 * it as running. This is used as a condition variable to 5985 * insure only one NAPI poll instance runs. We also make 5986 * sure there is no pending NAPI disable. 5987 */ 5988 bool napi_schedule_prep(struct napi_struct *n) 5989 { 5990 unsigned long new, val = READ_ONCE(n->state); 5991 5992 do { 5993 if (unlikely(val & NAPIF_STATE_DISABLE)) 5994 return false; 5995 new = val | NAPIF_STATE_SCHED; 5996 5997 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5998 * This was suggested by Alexander Duyck, as compiler 5999 * emits better code than : 6000 * if (val & NAPIF_STATE_SCHED) 6001 * new |= NAPIF_STATE_MISSED; 6002 */ 6003 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6004 NAPIF_STATE_MISSED; 6005 } while (!try_cmpxchg(&n->state, &val, new)); 6006 6007 return !(val & NAPIF_STATE_SCHED); 6008 } 6009 EXPORT_SYMBOL(napi_schedule_prep); 6010 6011 /** 6012 * __napi_schedule_irqoff - schedule for receive 6013 * @n: entry to schedule 6014 * 6015 * Variant of __napi_schedule() assuming hard irqs are masked. 6016 * 6017 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6018 * because the interrupt disabled assumption might not be true 6019 * due to force-threaded interrupts and spinlock substitution. 6020 */ 6021 void __napi_schedule_irqoff(struct napi_struct *n) 6022 { 6023 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6024 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6025 else 6026 __napi_schedule(n); 6027 } 6028 EXPORT_SYMBOL(__napi_schedule_irqoff); 6029 6030 bool napi_complete_done(struct napi_struct *n, int work_done) 6031 { 6032 unsigned long flags, val, new, timeout = 0; 6033 bool ret = true; 6034 6035 /* 6036 * 1) Don't let napi dequeue from the cpu poll list 6037 * just in case its running on a different cpu. 6038 * 2) If we are busy polling, do nothing here, we have 6039 * the guarantee we will be called later. 6040 */ 6041 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6042 NAPIF_STATE_IN_BUSY_POLL))) 6043 return false; 6044 6045 if (work_done) { 6046 if (n->gro_bitmask) 6047 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6048 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6049 } 6050 if (n->defer_hard_irqs_count > 0) { 6051 n->defer_hard_irqs_count--; 6052 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6053 if (timeout) 6054 ret = false; 6055 } 6056 if (n->gro_bitmask) { 6057 /* When the NAPI instance uses a timeout and keeps postponing 6058 * it, we need to bound somehow the time packets are kept in 6059 * the GRO layer 6060 */ 6061 napi_gro_flush(n, !!timeout); 6062 } 6063 6064 gro_normal_list(n); 6065 6066 if (unlikely(!list_empty(&n->poll_list))) { 6067 /* If n->poll_list is not empty, we need to mask irqs */ 6068 local_irq_save(flags); 6069 list_del_init(&n->poll_list); 6070 local_irq_restore(flags); 6071 } 6072 6073 val = READ_ONCE(n->state); 6074 do { 6075 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6076 6077 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6078 NAPIF_STATE_SCHED_THREADED | 6079 NAPIF_STATE_PREFER_BUSY_POLL); 6080 6081 /* If STATE_MISSED was set, leave STATE_SCHED set, 6082 * because we will call napi->poll() one more time. 6083 * This C code was suggested by Alexander Duyck to help gcc. 6084 */ 6085 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6086 NAPIF_STATE_SCHED; 6087 } while (!try_cmpxchg(&n->state, &val, new)); 6088 6089 if (unlikely(val & NAPIF_STATE_MISSED)) { 6090 __napi_schedule(n); 6091 return false; 6092 } 6093 6094 if (timeout) 6095 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6096 HRTIMER_MODE_REL_PINNED); 6097 return ret; 6098 } 6099 EXPORT_SYMBOL(napi_complete_done); 6100 6101 /* must be called under rcu_read_lock(), as we dont take a reference */ 6102 static struct napi_struct *napi_by_id(unsigned int napi_id) 6103 { 6104 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6105 struct napi_struct *napi; 6106 6107 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6108 if (napi->napi_id == napi_id) 6109 return napi; 6110 6111 return NULL; 6112 } 6113 6114 #if defined(CONFIG_NET_RX_BUSY_POLL) 6115 6116 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6117 { 6118 if (!skip_schedule) { 6119 gro_normal_list(napi); 6120 __napi_schedule(napi); 6121 return; 6122 } 6123 6124 if (napi->gro_bitmask) { 6125 /* flush too old packets 6126 * If HZ < 1000, flush all packets. 6127 */ 6128 napi_gro_flush(napi, HZ >= 1000); 6129 } 6130 6131 gro_normal_list(napi); 6132 clear_bit(NAPI_STATE_SCHED, &napi->state); 6133 } 6134 6135 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6136 u16 budget) 6137 { 6138 bool skip_schedule = false; 6139 unsigned long timeout; 6140 int rc; 6141 6142 /* Busy polling means there is a high chance device driver hard irq 6143 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6144 * set in napi_schedule_prep(). 6145 * Since we are about to call napi->poll() once more, we can safely 6146 * clear NAPI_STATE_MISSED. 6147 * 6148 * Note: x86 could use a single "lock and ..." instruction 6149 * to perform these two clear_bit() 6150 */ 6151 clear_bit(NAPI_STATE_MISSED, &napi->state); 6152 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6153 6154 local_bh_disable(); 6155 6156 if (prefer_busy_poll) { 6157 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6158 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6159 if (napi->defer_hard_irqs_count && timeout) { 6160 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6161 skip_schedule = true; 6162 } 6163 } 6164 6165 /* All we really want here is to re-enable device interrupts. 6166 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6167 */ 6168 rc = napi->poll(napi, budget); 6169 /* We can't gro_normal_list() here, because napi->poll() might have 6170 * rearmed the napi (napi_complete_done()) in which case it could 6171 * already be running on another CPU. 6172 */ 6173 trace_napi_poll(napi, rc, budget); 6174 netpoll_poll_unlock(have_poll_lock); 6175 if (rc == budget) 6176 __busy_poll_stop(napi, skip_schedule); 6177 local_bh_enable(); 6178 } 6179 6180 void napi_busy_loop(unsigned int napi_id, 6181 bool (*loop_end)(void *, unsigned long), 6182 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6183 { 6184 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6185 int (*napi_poll)(struct napi_struct *napi, int budget); 6186 void *have_poll_lock = NULL; 6187 struct napi_struct *napi; 6188 6189 restart: 6190 napi_poll = NULL; 6191 6192 rcu_read_lock(); 6193 6194 napi = napi_by_id(napi_id); 6195 if (!napi) 6196 goto out; 6197 6198 preempt_disable(); 6199 for (;;) { 6200 int work = 0; 6201 6202 local_bh_disable(); 6203 if (!napi_poll) { 6204 unsigned long val = READ_ONCE(napi->state); 6205 6206 /* If multiple threads are competing for this napi, 6207 * we avoid dirtying napi->state as much as we can. 6208 */ 6209 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6210 NAPIF_STATE_IN_BUSY_POLL)) { 6211 if (prefer_busy_poll) 6212 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6213 goto count; 6214 } 6215 if (cmpxchg(&napi->state, val, 6216 val | NAPIF_STATE_IN_BUSY_POLL | 6217 NAPIF_STATE_SCHED) != val) { 6218 if (prefer_busy_poll) 6219 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6220 goto count; 6221 } 6222 have_poll_lock = netpoll_poll_lock(napi); 6223 napi_poll = napi->poll; 6224 } 6225 work = napi_poll(napi, budget); 6226 trace_napi_poll(napi, work, budget); 6227 gro_normal_list(napi); 6228 count: 6229 if (work > 0) 6230 __NET_ADD_STATS(dev_net(napi->dev), 6231 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6232 local_bh_enable(); 6233 6234 if (!loop_end || loop_end(loop_end_arg, start_time)) 6235 break; 6236 6237 if (unlikely(need_resched())) { 6238 if (napi_poll) 6239 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6240 preempt_enable(); 6241 rcu_read_unlock(); 6242 cond_resched(); 6243 if (loop_end(loop_end_arg, start_time)) 6244 return; 6245 goto restart; 6246 } 6247 cpu_relax(); 6248 } 6249 if (napi_poll) 6250 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6251 preempt_enable(); 6252 out: 6253 rcu_read_unlock(); 6254 } 6255 EXPORT_SYMBOL(napi_busy_loop); 6256 6257 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6258 6259 static void napi_hash_add(struct napi_struct *napi) 6260 { 6261 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6262 return; 6263 6264 spin_lock(&napi_hash_lock); 6265 6266 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6267 do { 6268 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6269 napi_gen_id = MIN_NAPI_ID; 6270 } while (napi_by_id(napi_gen_id)); 6271 napi->napi_id = napi_gen_id; 6272 6273 hlist_add_head_rcu(&napi->napi_hash_node, 6274 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6275 6276 spin_unlock(&napi_hash_lock); 6277 } 6278 6279 /* Warning : caller is responsible to make sure rcu grace period 6280 * is respected before freeing memory containing @napi 6281 */ 6282 static void napi_hash_del(struct napi_struct *napi) 6283 { 6284 spin_lock(&napi_hash_lock); 6285 6286 hlist_del_init_rcu(&napi->napi_hash_node); 6287 6288 spin_unlock(&napi_hash_lock); 6289 } 6290 6291 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6292 { 6293 struct napi_struct *napi; 6294 6295 napi = container_of(timer, struct napi_struct, timer); 6296 6297 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6298 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6299 */ 6300 if (!napi_disable_pending(napi) && 6301 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6302 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6303 __napi_schedule_irqoff(napi); 6304 } 6305 6306 return HRTIMER_NORESTART; 6307 } 6308 6309 static void init_gro_hash(struct napi_struct *napi) 6310 { 6311 int i; 6312 6313 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6314 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6315 napi->gro_hash[i].count = 0; 6316 } 6317 napi->gro_bitmask = 0; 6318 } 6319 6320 int dev_set_threaded(struct net_device *dev, bool threaded) 6321 { 6322 struct napi_struct *napi; 6323 int err = 0; 6324 6325 if (dev->threaded == threaded) 6326 return 0; 6327 6328 if (threaded) { 6329 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6330 if (!napi->thread) { 6331 err = napi_kthread_create(napi); 6332 if (err) { 6333 threaded = false; 6334 break; 6335 } 6336 } 6337 } 6338 } 6339 6340 dev->threaded = threaded; 6341 6342 /* Make sure kthread is created before THREADED bit 6343 * is set. 6344 */ 6345 smp_mb__before_atomic(); 6346 6347 /* Setting/unsetting threaded mode on a napi might not immediately 6348 * take effect, if the current napi instance is actively being 6349 * polled. In this case, the switch between threaded mode and 6350 * softirq mode will happen in the next round of napi_schedule(). 6351 * This should not cause hiccups/stalls to the live traffic. 6352 */ 6353 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6354 if (threaded) 6355 set_bit(NAPI_STATE_THREADED, &napi->state); 6356 else 6357 clear_bit(NAPI_STATE_THREADED, &napi->state); 6358 } 6359 6360 return err; 6361 } 6362 EXPORT_SYMBOL(dev_set_threaded); 6363 6364 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6365 int (*poll)(struct napi_struct *, int), int weight) 6366 { 6367 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6368 return; 6369 6370 INIT_LIST_HEAD(&napi->poll_list); 6371 INIT_HLIST_NODE(&napi->napi_hash_node); 6372 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6373 napi->timer.function = napi_watchdog; 6374 init_gro_hash(napi); 6375 napi->skb = NULL; 6376 INIT_LIST_HEAD(&napi->rx_list); 6377 napi->rx_count = 0; 6378 napi->poll = poll; 6379 if (weight > NAPI_POLL_WEIGHT) 6380 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6381 weight); 6382 napi->weight = weight; 6383 napi->dev = dev; 6384 #ifdef CONFIG_NETPOLL 6385 napi->poll_owner = -1; 6386 #endif 6387 set_bit(NAPI_STATE_SCHED, &napi->state); 6388 set_bit(NAPI_STATE_NPSVC, &napi->state); 6389 list_add_rcu(&napi->dev_list, &dev->napi_list); 6390 napi_hash_add(napi); 6391 napi_get_frags_check(napi); 6392 /* Create kthread for this napi if dev->threaded is set. 6393 * Clear dev->threaded if kthread creation failed so that 6394 * threaded mode will not be enabled in napi_enable(). 6395 */ 6396 if (dev->threaded && napi_kthread_create(napi)) 6397 dev->threaded = 0; 6398 } 6399 EXPORT_SYMBOL(netif_napi_add_weight); 6400 6401 void napi_disable(struct napi_struct *n) 6402 { 6403 unsigned long val, new; 6404 6405 might_sleep(); 6406 set_bit(NAPI_STATE_DISABLE, &n->state); 6407 6408 val = READ_ONCE(n->state); 6409 do { 6410 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6411 usleep_range(20, 200); 6412 val = READ_ONCE(n->state); 6413 } 6414 6415 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6416 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6417 } while (!try_cmpxchg(&n->state, &val, new)); 6418 6419 hrtimer_cancel(&n->timer); 6420 6421 clear_bit(NAPI_STATE_DISABLE, &n->state); 6422 } 6423 EXPORT_SYMBOL(napi_disable); 6424 6425 /** 6426 * napi_enable - enable NAPI scheduling 6427 * @n: NAPI context 6428 * 6429 * Resume NAPI from being scheduled on this context. 6430 * Must be paired with napi_disable. 6431 */ 6432 void napi_enable(struct napi_struct *n) 6433 { 6434 unsigned long new, val = READ_ONCE(n->state); 6435 6436 do { 6437 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6438 6439 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6440 if (n->dev->threaded && n->thread) 6441 new |= NAPIF_STATE_THREADED; 6442 } while (!try_cmpxchg(&n->state, &val, new)); 6443 } 6444 EXPORT_SYMBOL(napi_enable); 6445 6446 static void flush_gro_hash(struct napi_struct *napi) 6447 { 6448 int i; 6449 6450 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6451 struct sk_buff *skb, *n; 6452 6453 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6454 kfree_skb(skb); 6455 napi->gro_hash[i].count = 0; 6456 } 6457 } 6458 6459 /* Must be called in process context */ 6460 void __netif_napi_del(struct napi_struct *napi) 6461 { 6462 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6463 return; 6464 6465 napi_hash_del(napi); 6466 list_del_rcu(&napi->dev_list); 6467 napi_free_frags(napi); 6468 6469 flush_gro_hash(napi); 6470 napi->gro_bitmask = 0; 6471 6472 if (napi->thread) { 6473 kthread_stop(napi->thread); 6474 napi->thread = NULL; 6475 } 6476 } 6477 EXPORT_SYMBOL(__netif_napi_del); 6478 6479 static int __napi_poll(struct napi_struct *n, bool *repoll) 6480 { 6481 int work, weight; 6482 6483 weight = n->weight; 6484 6485 /* This NAPI_STATE_SCHED test is for avoiding a race 6486 * with netpoll's poll_napi(). Only the entity which 6487 * obtains the lock and sees NAPI_STATE_SCHED set will 6488 * actually make the ->poll() call. Therefore we avoid 6489 * accidentally calling ->poll() when NAPI is not scheduled. 6490 */ 6491 work = 0; 6492 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6493 work = n->poll(n, weight); 6494 trace_napi_poll(n, work, weight); 6495 } 6496 6497 if (unlikely(work > weight)) 6498 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6499 n->poll, work, weight); 6500 6501 if (likely(work < weight)) 6502 return work; 6503 6504 /* Drivers must not modify the NAPI state if they 6505 * consume the entire weight. In such cases this code 6506 * still "owns" the NAPI instance and therefore can 6507 * move the instance around on the list at-will. 6508 */ 6509 if (unlikely(napi_disable_pending(n))) { 6510 napi_complete(n); 6511 return work; 6512 } 6513 6514 /* The NAPI context has more processing work, but busy-polling 6515 * is preferred. Exit early. 6516 */ 6517 if (napi_prefer_busy_poll(n)) { 6518 if (napi_complete_done(n, work)) { 6519 /* If timeout is not set, we need to make sure 6520 * that the NAPI is re-scheduled. 6521 */ 6522 napi_schedule(n); 6523 } 6524 return work; 6525 } 6526 6527 if (n->gro_bitmask) { 6528 /* flush too old packets 6529 * If HZ < 1000, flush all packets. 6530 */ 6531 napi_gro_flush(n, HZ >= 1000); 6532 } 6533 6534 gro_normal_list(n); 6535 6536 /* Some drivers may have called napi_schedule 6537 * prior to exhausting their budget. 6538 */ 6539 if (unlikely(!list_empty(&n->poll_list))) { 6540 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6541 n->dev ? n->dev->name : "backlog"); 6542 return work; 6543 } 6544 6545 *repoll = true; 6546 6547 return work; 6548 } 6549 6550 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6551 { 6552 bool do_repoll = false; 6553 void *have; 6554 int work; 6555 6556 list_del_init(&n->poll_list); 6557 6558 have = netpoll_poll_lock(n); 6559 6560 work = __napi_poll(n, &do_repoll); 6561 6562 if (do_repoll) 6563 list_add_tail(&n->poll_list, repoll); 6564 6565 netpoll_poll_unlock(have); 6566 6567 return work; 6568 } 6569 6570 static int napi_thread_wait(struct napi_struct *napi) 6571 { 6572 bool woken = false; 6573 6574 set_current_state(TASK_INTERRUPTIBLE); 6575 6576 while (!kthread_should_stop()) { 6577 /* Testing SCHED_THREADED bit here to make sure the current 6578 * kthread owns this napi and could poll on this napi. 6579 * Testing SCHED bit is not enough because SCHED bit might be 6580 * set by some other busy poll thread or by napi_disable(). 6581 */ 6582 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6583 WARN_ON(!list_empty(&napi->poll_list)); 6584 __set_current_state(TASK_RUNNING); 6585 return 0; 6586 } 6587 6588 schedule(); 6589 /* woken being true indicates this thread owns this napi. */ 6590 woken = true; 6591 set_current_state(TASK_INTERRUPTIBLE); 6592 } 6593 __set_current_state(TASK_RUNNING); 6594 6595 return -1; 6596 } 6597 6598 static int napi_threaded_poll(void *data) 6599 { 6600 struct napi_struct *napi = data; 6601 void *have; 6602 6603 while (!napi_thread_wait(napi)) { 6604 for (;;) { 6605 bool repoll = false; 6606 6607 local_bh_disable(); 6608 6609 have = netpoll_poll_lock(napi); 6610 __napi_poll(napi, &repoll); 6611 netpoll_poll_unlock(have); 6612 6613 local_bh_enable(); 6614 6615 if (!repoll) 6616 break; 6617 6618 cond_resched(); 6619 } 6620 } 6621 return 0; 6622 } 6623 6624 static void skb_defer_free_flush(struct softnet_data *sd) 6625 { 6626 struct sk_buff *skb, *next; 6627 6628 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6629 if (!READ_ONCE(sd->defer_list)) 6630 return; 6631 6632 spin_lock_irq(&sd->defer_lock); 6633 skb = sd->defer_list; 6634 sd->defer_list = NULL; 6635 sd->defer_count = 0; 6636 spin_unlock_irq(&sd->defer_lock); 6637 6638 while (skb != NULL) { 6639 next = skb->next; 6640 napi_consume_skb(skb, 1); 6641 skb = next; 6642 } 6643 } 6644 6645 static __latent_entropy void net_rx_action(struct softirq_action *h) 6646 { 6647 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6648 unsigned long time_limit = jiffies + 6649 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6650 int budget = READ_ONCE(netdev_budget); 6651 LIST_HEAD(list); 6652 LIST_HEAD(repoll); 6653 6654 start: 6655 sd->in_net_rx_action = true; 6656 local_irq_disable(); 6657 list_splice_init(&sd->poll_list, &list); 6658 local_irq_enable(); 6659 6660 for (;;) { 6661 struct napi_struct *n; 6662 6663 skb_defer_free_flush(sd); 6664 6665 if (list_empty(&list)) { 6666 if (list_empty(&repoll)) { 6667 sd->in_net_rx_action = false; 6668 barrier(); 6669 /* We need to check if ____napi_schedule() 6670 * had refilled poll_list while 6671 * sd->in_net_rx_action was true. 6672 */ 6673 if (!list_empty(&sd->poll_list)) 6674 goto start; 6675 if (!sd_has_rps_ipi_waiting(sd)) 6676 goto end; 6677 } 6678 break; 6679 } 6680 6681 n = list_first_entry(&list, struct napi_struct, poll_list); 6682 budget -= napi_poll(n, &repoll); 6683 6684 /* If softirq window is exhausted then punt. 6685 * Allow this to run for 2 jiffies since which will allow 6686 * an average latency of 1.5/HZ. 6687 */ 6688 if (unlikely(budget <= 0 || 6689 time_after_eq(jiffies, time_limit))) { 6690 sd->time_squeeze++; 6691 break; 6692 } 6693 } 6694 6695 local_irq_disable(); 6696 6697 list_splice_tail_init(&sd->poll_list, &list); 6698 list_splice_tail(&repoll, &list); 6699 list_splice(&list, &sd->poll_list); 6700 if (!list_empty(&sd->poll_list)) 6701 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6702 else 6703 sd->in_net_rx_action = false; 6704 6705 net_rps_action_and_irq_enable(sd); 6706 end:; 6707 } 6708 6709 struct netdev_adjacent { 6710 struct net_device *dev; 6711 netdevice_tracker dev_tracker; 6712 6713 /* upper master flag, there can only be one master device per list */ 6714 bool master; 6715 6716 /* lookup ignore flag */ 6717 bool ignore; 6718 6719 /* counter for the number of times this device was added to us */ 6720 u16 ref_nr; 6721 6722 /* private field for the users */ 6723 void *private; 6724 6725 struct list_head list; 6726 struct rcu_head rcu; 6727 }; 6728 6729 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6730 struct list_head *adj_list) 6731 { 6732 struct netdev_adjacent *adj; 6733 6734 list_for_each_entry(adj, adj_list, list) { 6735 if (adj->dev == adj_dev) 6736 return adj; 6737 } 6738 return NULL; 6739 } 6740 6741 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6742 struct netdev_nested_priv *priv) 6743 { 6744 struct net_device *dev = (struct net_device *)priv->data; 6745 6746 return upper_dev == dev; 6747 } 6748 6749 /** 6750 * netdev_has_upper_dev - Check if device is linked to an upper device 6751 * @dev: device 6752 * @upper_dev: upper device to check 6753 * 6754 * Find out if a device is linked to specified upper device and return true 6755 * in case it is. Note that this checks only immediate upper device, 6756 * not through a complete stack of devices. The caller must hold the RTNL lock. 6757 */ 6758 bool netdev_has_upper_dev(struct net_device *dev, 6759 struct net_device *upper_dev) 6760 { 6761 struct netdev_nested_priv priv = { 6762 .data = (void *)upper_dev, 6763 }; 6764 6765 ASSERT_RTNL(); 6766 6767 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6768 &priv); 6769 } 6770 EXPORT_SYMBOL(netdev_has_upper_dev); 6771 6772 /** 6773 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6774 * @dev: device 6775 * @upper_dev: upper device to check 6776 * 6777 * Find out if a device is linked to specified upper device and return true 6778 * in case it is. Note that this checks the entire upper device chain. 6779 * The caller must hold rcu lock. 6780 */ 6781 6782 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6783 struct net_device *upper_dev) 6784 { 6785 struct netdev_nested_priv priv = { 6786 .data = (void *)upper_dev, 6787 }; 6788 6789 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6790 &priv); 6791 } 6792 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6793 6794 /** 6795 * netdev_has_any_upper_dev - Check if device is linked to some device 6796 * @dev: device 6797 * 6798 * Find out if a device is linked to an upper device and return true in case 6799 * it is. The caller must hold the RTNL lock. 6800 */ 6801 bool netdev_has_any_upper_dev(struct net_device *dev) 6802 { 6803 ASSERT_RTNL(); 6804 6805 return !list_empty(&dev->adj_list.upper); 6806 } 6807 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6808 6809 /** 6810 * netdev_master_upper_dev_get - Get master upper device 6811 * @dev: device 6812 * 6813 * Find a master upper device and return pointer to it or NULL in case 6814 * it's not there. The caller must hold the RTNL lock. 6815 */ 6816 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6817 { 6818 struct netdev_adjacent *upper; 6819 6820 ASSERT_RTNL(); 6821 6822 if (list_empty(&dev->adj_list.upper)) 6823 return NULL; 6824 6825 upper = list_first_entry(&dev->adj_list.upper, 6826 struct netdev_adjacent, list); 6827 if (likely(upper->master)) 6828 return upper->dev; 6829 return NULL; 6830 } 6831 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6832 6833 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6834 { 6835 struct netdev_adjacent *upper; 6836 6837 ASSERT_RTNL(); 6838 6839 if (list_empty(&dev->adj_list.upper)) 6840 return NULL; 6841 6842 upper = list_first_entry(&dev->adj_list.upper, 6843 struct netdev_adjacent, list); 6844 if (likely(upper->master) && !upper->ignore) 6845 return upper->dev; 6846 return NULL; 6847 } 6848 6849 /** 6850 * netdev_has_any_lower_dev - Check if device is linked to some device 6851 * @dev: device 6852 * 6853 * Find out if a device is linked to a lower device and return true in case 6854 * it is. The caller must hold the RTNL lock. 6855 */ 6856 static bool netdev_has_any_lower_dev(struct net_device *dev) 6857 { 6858 ASSERT_RTNL(); 6859 6860 return !list_empty(&dev->adj_list.lower); 6861 } 6862 6863 void *netdev_adjacent_get_private(struct list_head *adj_list) 6864 { 6865 struct netdev_adjacent *adj; 6866 6867 adj = list_entry(adj_list, struct netdev_adjacent, list); 6868 6869 return adj->private; 6870 } 6871 EXPORT_SYMBOL(netdev_adjacent_get_private); 6872 6873 /** 6874 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6875 * @dev: device 6876 * @iter: list_head ** of the current position 6877 * 6878 * Gets the next device from the dev's upper list, starting from iter 6879 * position. The caller must hold RCU read lock. 6880 */ 6881 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6882 struct list_head **iter) 6883 { 6884 struct netdev_adjacent *upper; 6885 6886 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6887 6888 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6889 6890 if (&upper->list == &dev->adj_list.upper) 6891 return NULL; 6892 6893 *iter = &upper->list; 6894 6895 return upper->dev; 6896 } 6897 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6898 6899 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6900 struct list_head **iter, 6901 bool *ignore) 6902 { 6903 struct netdev_adjacent *upper; 6904 6905 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6906 6907 if (&upper->list == &dev->adj_list.upper) 6908 return NULL; 6909 6910 *iter = &upper->list; 6911 *ignore = upper->ignore; 6912 6913 return upper->dev; 6914 } 6915 6916 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6917 struct list_head **iter) 6918 { 6919 struct netdev_adjacent *upper; 6920 6921 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6922 6923 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6924 6925 if (&upper->list == &dev->adj_list.upper) 6926 return NULL; 6927 6928 *iter = &upper->list; 6929 6930 return upper->dev; 6931 } 6932 6933 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6934 int (*fn)(struct net_device *dev, 6935 struct netdev_nested_priv *priv), 6936 struct netdev_nested_priv *priv) 6937 { 6938 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6939 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6940 int ret, cur = 0; 6941 bool ignore; 6942 6943 now = dev; 6944 iter = &dev->adj_list.upper; 6945 6946 while (1) { 6947 if (now != dev) { 6948 ret = fn(now, priv); 6949 if (ret) 6950 return ret; 6951 } 6952 6953 next = NULL; 6954 while (1) { 6955 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6956 if (!udev) 6957 break; 6958 if (ignore) 6959 continue; 6960 6961 next = udev; 6962 niter = &udev->adj_list.upper; 6963 dev_stack[cur] = now; 6964 iter_stack[cur++] = iter; 6965 break; 6966 } 6967 6968 if (!next) { 6969 if (!cur) 6970 return 0; 6971 next = dev_stack[--cur]; 6972 niter = iter_stack[cur]; 6973 } 6974 6975 now = next; 6976 iter = niter; 6977 } 6978 6979 return 0; 6980 } 6981 6982 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6983 int (*fn)(struct net_device *dev, 6984 struct netdev_nested_priv *priv), 6985 struct netdev_nested_priv *priv) 6986 { 6987 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6988 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6989 int ret, cur = 0; 6990 6991 now = dev; 6992 iter = &dev->adj_list.upper; 6993 6994 while (1) { 6995 if (now != dev) { 6996 ret = fn(now, priv); 6997 if (ret) 6998 return ret; 6999 } 7000 7001 next = NULL; 7002 while (1) { 7003 udev = netdev_next_upper_dev_rcu(now, &iter); 7004 if (!udev) 7005 break; 7006 7007 next = udev; 7008 niter = &udev->adj_list.upper; 7009 dev_stack[cur] = now; 7010 iter_stack[cur++] = iter; 7011 break; 7012 } 7013 7014 if (!next) { 7015 if (!cur) 7016 return 0; 7017 next = dev_stack[--cur]; 7018 niter = iter_stack[cur]; 7019 } 7020 7021 now = next; 7022 iter = niter; 7023 } 7024 7025 return 0; 7026 } 7027 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7028 7029 static bool __netdev_has_upper_dev(struct net_device *dev, 7030 struct net_device *upper_dev) 7031 { 7032 struct netdev_nested_priv priv = { 7033 .flags = 0, 7034 .data = (void *)upper_dev, 7035 }; 7036 7037 ASSERT_RTNL(); 7038 7039 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7040 &priv); 7041 } 7042 7043 /** 7044 * netdev_lower_get_next_private - Get the next ->private from the 7045 * lower neighbour list 7046 * @dev: device 7047 * @iter: list_head ** of the current position 7048 * 7049 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7050 * list, starting from iter position. The caller must hold either hold the 7051 * RTNL lock or its own locking that guarantees that the neighbour lower 7052 * list will remain unchanged. 7053 */ 7054 void *netdev_lower_get_next_private(struct net_device *dev, 7055 struct list_head **iter) 7056 { 7057 struct netdev_adjacent *lower; 7058 7059 lower = list_entry(*iter, struct netdev_adjacent, list); 7060 7061 if (&lower->list == &dev->adj_list.lower) 7062 return NULL; 7063 7064 *iter = lower->list.next; 7065 7066 return lower->private; 7067 } 7068 EXPORT_SYMBOL(netdev_lower_get_next_private); 7069 7070 /** 7071 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7072 * lower neighbour list, RCU 7073 * variant 7074 * @dev: device 7075 * @iter: list_head ** of the current position 7076 * 7077 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7078 * list, starting from iter position. The caller must hold RCU read lock. 7079 */ 7080 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7081 struct list_head **iter) 7082 { 7083 struct netdev_adjacent *lower; 7084 7085 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7086 7087 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7088 7089 if (&lower->list == &dev->adj_list.lower) 7090 return NULL; 7091 7092 *iter = &lower->list; 7093 7094 return lower->private; 7095 } 7096 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7097 7098 /** 7099 * netdev_lower_get_next - Get the next device from the lower neighbour 7100 * list 7101 * @dev: device 7102 * @iter: list_head ** of the current position 7103 * 7104 * Gets the next netdev_adjacent from the dev's lower neighbour 7105 * list, starting from iter position. The caller must hold RTNL lock or 7106 * its own locking that guarantees that the neighbour lower 7107 * list will remain unchanged. 7108 */ 7109 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7110 { 7111 struct netdev_adjacent *lower; 7112 7113 lower = list_entry(*iter, struct netdev_adjacent, list); 7114 7115 if (&lower->list == &dev->adj_list.lower) 7116 return NULL; 7117 7118 *iter = lower->list.next; 7119 7120 return lower->dev; 7121 } 7122 EXPORT_SYMBOL(netdev_lower_get_next); 7123 7124 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7125 struct list_head **iter) 7126 { 7127 struct netdev_adjacent *lower; 7128 7129 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7130 7131 if (&lower->list == &dev->adj_list.lower) 7132 return NULL; 7133 7134 *iter = &lower->list; 7135 7136 return lower->dev; 7137 } 7138 7139 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7140 struct list_head **iter, 7141 bool *ignore) 7142 { 7143 struct netdev_adjacent *lower; 7144 7145 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7146 7147 if (&lower->list == &dev->adj_list.lower) 7148 return NULL; 7149 7150 *iter = &lower->list; 7151 *ignore = lower->ignore; 7152 7153 return lower->dev; 7154 } 7155 7156 int netdev_walk_all_lower_dev(struct net_device *dev, 7157 int (*fn)(struct net_device *dev, 7158 struct netdev_nested_priv *priv), 7159 struct netdev_nested_priv *priv) 7160 { 7161 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7162 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7163 int ret, cur = 0; 7164 7165 now = dev; 7166 iter = &dev->adj_list.lower; 7167 7168 while (1) { 7169 if (now != dev) { 7170 ret = fn(now, priv); 7171 if (ret) 7172 return ret; 7173 } 7174 7175 next = NULL; 7176 while (1) { 7177 ldev = netdev_next_lower_dev(now, &iter); 7178 if (!ldev) 7179 break; 7180 7181 next = ldev; 7182 niter = &ldev->adj_list.lower; 7183 dev_stack[cur] = now; 7184 iter_stack[cur++] = iter; 7185 break; 7186 } 7187 7188 if (!next) { 7189 if (!cur) 7190 return 0; 7191 next = dev_stack[--cur]; 7192 niter = iter_stack[cur]; 7193 } 7194 7195 now = next; 7196 iter = niter; 7197 } 7198 7199 return 0; 7200 } 7201 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7202 7203 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7204 int (*fn)(struct net_device *dev, 7205 struct netdev_nested_priv *priv), 7206 struct netdev_nested_priv *priv) 7207 { 7208 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7209 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7210 int ret, cur = 0; 7211 bool ignore; 7212 7213 now = dev; 7214 iter = &dev->adj_list.lower; 7215 7216 while (1) { 7217 if (now != dev) { 7218 ret = fn(now, priv); 7219 if (ret) 7220 return ret; 7221 } 7222 7223 next = NULL; 7224 while (1) { 7225 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7226 if (!ldev) 7227 break; 7228 if (ignore) 7229 continue; 7230 7231 next = ldev; 7232 niter = &ldev->adj_list.lower; 7233 dev_stack[cur] = now; 7234 iter_stack[cur++] = iter; 7235 break; 7236 } 7237 7238 if (!next) { 7239 if (!cur) 7240 return 0; 7241 next = dev_stack[--cur]; 7242 niter = iter_stack[cur]; 7243 } 7244 7245 now = next; 7246 iter = niter; 7247 } 7248 7249 return 0; 7250 } 7251 7252 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7253 struct list_head **iter) 7254 { 7255 struct netdev_adjacent *lower; 7256 7257 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7258 if (&lower->list == &dev->adj_list.lower) 7259 return NULL; 7260 7261 *iter = &lower->list; 7262 7263 return lower->dev; 7264 } 7265 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7266 7267 static u8 __netdev_upper_depth(struct net_device *dev) 7268 { 7269 struct net_device *udev; 7270 struct list_head *iter; 7271 u8 max_depth = 0; 7272 bool ignore; 7273 7274 for (iter = &dev->adj_list.upper, 7275 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7276 udev; 7277 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7278 if (ignore) 7279 continue; 7280 if (max_depth < udev->upper_level) 7281 max_depth = udev->upper_level; 7282 } 7283 7284 return max_depth; 7285 } 7286 7287 static u8 __netdev_lower_depth(struct net_device *dev) 7288 { 7289 struct net_device *ldev; 7290 struct list_head *iter; 7291 u8 max_depth = 0; 7292 bool ignore; 7293 7294 for (iter = &dev->adj_list.lower, 7295 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7296 ldev; 7297 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7298 if (ignore) 7299 continue; 7300 if (max_depth < ldev->lower_level) 7301 max_depth = ldev->lower_level; 7302 } 7303 7304 return max_depth; 7305 } 7306 7307 static int __netdev_update_upper_level(struct net_device *dev, 7308 struct netdev_nested_priv *__unused) 7309 { 7310 dev->upper_level = __netdev_upper_depth(dev) + 1; 7311 return 0; 7312 } 7313 7314 #ifdef CONFIG_LOCKDEP 7315 static LIST_HEAD(net_unlink_list); 7316 7317 static void net_unlink_todo(struct net_device *dev) 7318 { 7319 if (list_empty(&dev->unlink_list)) 7320 list_add_tail(&dev->unlink_list, &net_unlink_list); 7321 } 7322 #endif 7323 7324 static int __netdev_update_lower_level(struct net_device *dev, 7325 struct netdev_nested_priv *priv) 7326 { 7327 dev->lower_level = __netdev_lower_depth(dev) + 1; 7328 7329 #ifdef CONFIG_LOCKDEP 7330 if (!priv) 7331 return 0; 7332 7333 if (priv->flags & NESTED_SYNC_IMM) 7334 dev->nested_level = dev->lower_level - 1; 7335 if (priv->flags & NESTED_SYNC_TODO) 7336 net_unlink_todo(dev); 7337 #endif 7338 return 0; 7339 } 7340 7341 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7342 int (*fn)(struct net_device *dev, 7343 struct netdev_nested_priv *priv), 7344 struct netdev_nested_priv *priv) 7345 { 7346 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7347 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7348 int ret, cur = 0; 7349 7350 now = dev; 7351 iter = &dev->adj_list.lower; 7352 7353 while (1) { 7354 if (now != dev) { 7355 ret = fn(now, priv); 7356 if (ret) 7357 return ret; 7358 } 7359 7360 next = NULL; 7361 while (1) { 7362 ldev = netdev_next_lower_dev_rcu(now, &iter); 7363 if (!ldev) 7364 break; 7365 7366 next = ldev; 7367 niter = &ldev->adj_list.lower; 7368 dev_stack[cur] = now; 7369 iter_stack[cur++] = iter; 7370 break; 7371 } 7372 7373 if (!next) { 7374 if (!cur) 7375 return 0; 7376 next = dev_stack[--cur]; 7377 niter = iter_stack[cur]; 7378 } 7379 7380 now = next; 7381 iter = niter; 7382 } 7383 7384 return 0; 7385 } 7386 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7387 7388 /** 7389 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7390 * lower neighbour list, RCU 7391 * variant 7392 * @dev: device 7393 * 7394 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7395 * list. The caller must hold RCU read lock. 7396 */ 7397 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7398 { 7399 struct netdev_adjacent *lower; 7400 7401 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7402 struct netdev_adjacent, list); 7403 if (lower) 7404 return lower->private; 7405 return NULL; 7406 } 7407 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7408 7409 /** 7410 * netdev_master_upper_dev_get_rcu - Get master upper device 7411 * @dev: device 7412 * 7413 * Find a master upper device and return pointer to it or NULL in case 7414 * it's not there. The caller must hold the RCU read lock. 7415 */ 7416 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7417 { 7418 struct netdev_adjacent *upper; 7419 7420 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7421 struct netdev_adjacent, list); 7422 if (upper && likely(upper->master)) 7423 return upper->dev; 7424 return NULL; 7425 } 7426 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7427 7428 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7429 struct net_device *adj_dev, 7430 struct list_head *dev_list) 7431 { 7432 char linkname[IFNAMSIZ+7]; 7433 7434 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7435 "upper_%s" : "lower_%s", adj_dev->name); 7436 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7437 linkname); 7438 } 7439 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7440 char *name, 7441 struct list_head *dev_list) 7442 { 7443 char linkname[IFNAMSIZ+7]; 7444 7445 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7446 "upper_%s" : "lower_%s", name); 7447 sysfs_remove_link(&(dev->dev.kobj), linkname); 7448 } 7449 7450 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7451 struct net_device *adj_dev, 7452 struct list_head *dev_list) 7453 { 7454 return (dev_list == &dev->adj_list.upper || 7455 dev_list == &dev->adj_list.lower) && 7456 net_eq(dev_net(dev), dev_net(adj_dev)); 7457 } 7458 7459 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7460 struct net_device *adj_dev, 7461 struct list_head *dev_list, 7462 void *private, bool master) 7463 { 7464 struct netdev_adjacent *adj; 7465 int ret; 7466 7467 adj = __netdev_find_adj(adj_dev, dev_list); 7468 7469 if (adj) { 7470 adj->ref_nr += 1; 7471 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7472 dev->name, adj_dev->name, adj->ref_nr); 7473 7474 return 0; 7475 } 7476 7477 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7478 if (!adj) 7479 return -ENOMEM; 7480 7481 adj->dev = adj_dev; 7482 adj->master = master; 7483 adj->ref_nr = 1; 7484 adj->private = private; 7485 adj->ignore = false; 7486 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7487 7488 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7489 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7490 7491 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7492 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7493 if (ret) 7494 goto free_adj; 7495 } 7496 7497 /* Ensure that master link is always the first item in list. */ 7498 if (master) { 7499 ret = sysfs_create_link(&(dev->dev.kobj), 7500 &(adj_dev->dev.kobj), "master"); 7501 if (ret) 7502 goto remove_symlinks; 7503 7504 list_add_rcu(&adj->list, dev_list); 7505 } else { 7506 list_add_tail_rcu(&adj->list, dev_list); 7507 } 7508 7509 return 0; 7510 7511 remove_symlinks: 7512 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7513 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7514 free_adj: 7515 netdev_put(adj_dev, &adj->dev_tracker); 7516 kfree(adj); 7517 7518 return ret; 7519 } 7520 7521 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7522 struct net_device *adj_dev, 7523 u16 ref_nr, 7524 struct list_head *dev_list) 7525 { 7526 struct netdev_adjacent *adj; 7527 7528 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7529 dev->name, adj_dev->name, ref_nr); 7530 7531 adj = __netdev_find_adj(adj_dev, dev_list); 7532 7533 if (!adj) { 7534 pr_err("Adjacency does not exist for device %s from %s\n", 7535 dev->name, adj_dev->name); 7536 WARN_ON(1); 7537 return; 7538 } 7539 7540 if (adj->ref_nr > ref_nr) { 7541 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7542 dev->name, adj_dev->name, ref_nr, 7543 adj->ref_nr - ref_nr); 7544 adj->ref_nr -= ref_nr; 7545 return; 7546 } 7547 7548 if (adj->master) 7549 sysfs_remove_link(&(dev->dev.kobj), "master"); 7550 7551 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7552 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7553 7554 list_del_rcu(&adj->list); 7555 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7556 adj_dev->name, dev->name, adj_dev->name); 7557 netdev_put(adj_dev, &adj->dev_tracker); 7558 kfree_rcu(adj, rcu); 7559 } 7560 7561 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7562 struct net_device *upper_dev, 7563 struct list_head *up_list, 7564 struct list_head *down_list, 7565 void *private, bool master) 7566 { 7567 int ret; 7568 7569 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7570 private, master); 7571 if (ret) 7572 return ret; 7573 7574 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7575 private, false); 7576 if (ret) { 7577 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7578 return ret; 7579 } 7580 7581 return 0; 7582 } 7583 7584 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7585 struct net_device *upper_dev, 7586 u16 ref_nr, 7587 struct list_head *up_list, 7588 struct list_head *down_list) 7589 { 7590 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7591 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7592 } 7593 7594 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7595 struct net_device *upper_dev, 7596 void *private, bool master) 7597 { 7598 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7599 &dev->adj_list.upper, 7600 &upper_dev->adj_list.lower, 7601 private, master); 7602 } 7603 7604 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7605 struct net_device *upper_dev) 7606 { 7607 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7608 &dev->adj_list.upper, 7609 &upper_dev->adj_list.lower); 7610 } 7611 7612 static int __netdev_upper_dev_link(struct net_device *dev, 7613 struct net_device *upper_dev, bool master, 7614 void *upper_priv, void *upper_info, 7615 struct netdev_nested_priv *priv, 7616 struct netlink_ext_ack *extack) 7617 { 7618 struct netdev_notifier_changeupper_info changeupper_info = { 7619 .info = { 7620 .dev = dev, 7621 .extack = extack, 7622 }, 7623 .upper_dev = upper_dev, 7624 .master = master, 7625 .linking = true, 7626 .upper_info = upper_info, 7627 }; 7628 struct net_device *master_dev; 7629 int ret = 0; 7630 7631 ASSERT_RTNL(); 7632 7633 if (dev == upper_dev) 7634 return -EBUSY; 7635 7636 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7637 if (__netdev_has_upper_dev(upper_dev, dev)) 7638 return -EBUSY; 7639 7640 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7641 return -EMLINK; 7642 7643 if (!master) { 7644 if (__netdev_has_upper_dev(dev, upper_dev)) 7645 return -EEXIST; 7646 } else { 7647 master_dev = __netdev_master_upper_dev_get(dev); 7648 if (master_dev) 7649 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7650 } 7651 7652 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7653 &changeupper_info.info); 7654 ret = notifier_to_errno(ret); 7655 if (ret) 7656 return ret; 7657 7658 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7659 master); 7660 if (ret) 7661 return ret; 7662 7663 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7664 &changeupper_info.info); 7665 ret = notifier_to_errno(ret); 7666 if (ret) 7667 goto rollback; 7668 7669 __netdev_update_upper_level(dev, NULL); 7670 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7671 7672 __netdev_update_lower_level(upper_dev, priv); 7673 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7674 priv); 7675 7676 return 0; 7677 7678 rollback: 7679 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7680 7681 return ret; 7682 } 7683 7684 /** 7685 * netdev_upper_dev_link - Add a link to the upper device 7686 * @dev: device 7687 * @upper_dev: new upper device 7688 * @extack: netlink extended ack 7689 * 7690 * Adds a link to device which is upper to this one. The caller must hold 7691 * the RTNL lock. On a failure a negative errno code is returned. 7692 * On success the reference counts are adjusted and the function 7693 * returns zero. 7694 */ 7695 int netdev_upper_dev_link(struct net_device *dev, 7696 struct net_device *upper_dev, 7697 struct netlink_ext_ack *extack) 7698 { 7699 struct netdev_nested_priv priv = { 7700 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7701 .data = NULL, 7702 }; 7703 7704 return __netdev_upper_dev_link(dev, upper_dev, false, 7705 NULL, NULL, &priv, extack); 7706 } 7707 EXPORT_SYMBOL(netdev_upper_dev_link); 7708 7709 /** 7710 * netdev_master_upper_dev_link - Add a master link to the upper device 7711 * @dev: device 7712 * @upper_dev: new upper device 7713 * @upper_priv: upper device private 7714 * @upper_info: upper info to be passed down via notifier 7715 * @extack: netlink extended ack 7716 * 7717 * Adds a link to device which is upper to this one. In this case, only 7718 * one master upper device can be linked, although other non-master devices 7719 * might be linked as well. The caller must hold the RTNL lock. 7720 * On a failure a negative errno code is returned. On success the reference 7721 * counts are adjusted and the function returns zero. 7722 */ 7723 int netdev_master_upper_dev_link(struct net_device *dev, 7724 struct net_device *upper_dev, 7725 void *upper_priv, void *upper_info, 7726 struct netlink_ext_ack *extack) 7727 { 7728 struct netdev_nested_priv priv = { 7729 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7730 .data = NULL, 7731 }; 7732 7733 return __netdev_upper_dev_link(dev, upper_dev, true, 7734 upper_priv, upper_info, &priv, extack); 7735 } 7736 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7737 7738 static void __netdev_upper_dev_unlink(struct net_device *dev, 7739 struct net_device *upper_dev, 7740 struct netdev_nested_priv *priv) 7741 { 7742 struct netdev_notifier_changeupper_info changeupper_info = { 7743 .info = { 7744 .dev = dev, 7745 }, 7746 .upper_dev = upper_dev, 7747 .linking = false, 7748 }; 7749 7750 ASSERT_RTNL(); 7751 7752 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7753 7754 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7755 &changeupper_info.info); 7756 7757 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7758 7759 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7760 &changeupper_info.info); 7761 7762 __netdev_update_upper_level(dev, NULL); 7763 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7764 7765 __netdev_update_lower_level(upper_dev, priv); 7766 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7767 priv); 7768 } 7769 7770 /** 7771 * netdev_upper_dev_unlink - Removes a link to upper device 7772 * @dev: device 7773 * @upper_dev: new upper device 7774 * 7775 * Removes a link to device which is upper to this one. The caller must hold 7776 * the RTNL lock. 7777 */ 7778 void netdev_upper_dev_unlink(struct net_device *dev, 7779 struct net_device *upper_dev) 7780 { 7781 struct netdev_nested_priv priv = { 7782 .flags = NESTED_SYNC_TODO, 7783 .data = NULL, 7784 }; 7785 7786 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7787 } 7788 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7789 7790 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7791 struct net_device *lower_dev, 7792 bool val) 7793 { 7794 struct netdev_adjacent *adj; 7795 7796 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7797 if (adj) 7798 adj->ignore = val; 7799 7800 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7801 if (adj) 7802 adj->ignore = val; 7803 } 7804 7805 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7806 struct net_device *lower_dev) 7807 { 7808 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7809 } 7810 7811 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7812 struct net_device *lower_dev) 7813 { 7814 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7815 } 7816 7817 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7818 struct net_device *new_dev, 7819 struct net_device *dev, 7820 struct netlink_ext_ack *extack) 7821 { 7822 struct netdev_nested_priv priv = { 7823 .flags = 0, 7824 .data = NULL, 7825 }; 7826 int err; 7827 7828 if (!new_dev) 7829 return 0; 7830 7831 if (old_dev && new_dev != old_dev) 7832 netdev_adjacent_dev_disable(dev, old_dev); 7833 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7834 extack); 7835 if (err) { 7836 if (old_dev && new_dev != old_dev) 7837 netdev_adjacent_dev_enable(dev, old_dev); 7838 return err; 7839 } 7840 7841 return 0; 7842 } 7843 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7844 7845 void netdev_adjacent_change_commit(struct net_device *old_dev, 7846 struct net_device *new_dev, 7847 struct net_device *dev) 7848 { 7849 struct netdev_nested_priv priv = { 7850 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7851 .data = NULL, 7852 }; 7853 7854 if (!new_dev || !old_dev) 7855 return; 7856 7857 if (new_dev == old_dev) 7858 return; 7859 7860 netdev_adjacent_dev_enable(dev, old_dev); 7861 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7862 } 7863 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7864 7865 void netdev_adjacent_change_abort(struct net_device *old_dev, 7866 struct net_device *new_dev, 7867 struct net_device *dev) 7868 { 7869 struct netdev_nested_priv priv = { 7870 .flags = 0, 7871 .data = NULL, 7872 }; 7873 7874 if (!new_dev) 7875 return; 7876 7877 if (old_dev && new_dev != old_dev) 7878 netdev_adjacent_dev_enable(dev, old_dev); 7879 7880 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7881 } 7882 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7883 7884 /** 7885 * netdev_bonding_info_change - Dispatch event about slave change 7886 * @dev: device 7887 * @bonding_info: info to dispatch 7888 * 7889 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7890 * The caller must hold the RTNL lock. 7891 */ 7892 void netdev_bonding_info_change(struct net_device *dev, 7893 struct netdev_bonding_info *bonding_info) 7894 { 7895 struct netdev_notifier_bonding_info info = { 7896 .info.dev = dev, 7897 }; 7898 7899 memcpy(&info.bonding_info, bonding_info, 7900 sizeof(struct netdev_bonding_info)); 7901 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7902 &info.info); 7903 } 7904 EXPORT_SYMBOL(netdev_bonding_info_change); 7905 7906 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7907 struct netlink_ext_ack *extack) 7908 { 7909 struct netdev_notifier_offload_xstats_info info = { 7910 .info.dev = dev, 7911 .info.extack = extack, 7912 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7913 }; 7914 int err; 7915 int rc; 7916 7917 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7918 GFP_KERNEL); 7919 if (!dev->offload_xstats_l3) 7920 return -ENOMEM; 7921 7922 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7923 NETDEV_OFFLOAD_XSTATS_DISABLE, 7924 &info.info); 7925 err = notifier_to_errno(rc); 7926 if (err) 7927 goto free_stats; 7928 7929 return 0; 7930 7931 free_stats: 7932 kfree(dev->offload_xstats_l3); 7933 dev->offload_xstats_l3 = NULL; 7934 return err; 7935 } 7936 7937 int netdev_offload_xstats_enable(struct net_device *dev, 7938 enum netdev_offload_xstats_type type, 7939 struct netlink_ext_ack *extack) 7940 { 7941 ASSERT_RTNL(); 7942 7943 if (netdev_offload_xstats_enabled(dev, type)) 7944 return -EALREADY; 7945 7946 switch (type) { 7947 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7948 return netdev_offload_xstats_enable_l3(dev, extack); 7949 } 7950 7951 WARN_ON(1); 7952 return -EINVAL; 7953 } 7954 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7955 7956 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7957 { 7958 struct netdev_notifier_offload_xstats_info info = { 7959 .info.dev = dev, 7960 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7961 }; 7962 7963 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7964 &info.info); 7965 kfree(dev->offload_xstats_l3); 7966 dev->offload_xstats_l3 = NULL; 7967 } 7968 7969 int netdev_offload_xstats_disable(struct net_device *dev, 7970 enum netdev_offload_xstats_type type) 7971 { 7972 ASSERT_RTNL(); 7973 7974 if (!netdev_offload_xstats_enabled(dev, type)) 7975 return -EALREADY; 7976 7977 switch (type) { 7978 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7979 netdev_offload_xstats_disable_l3(dev); 7980 return 0; 7981 } 7982 7983 WARN_ON(1); 7984 return -EINVAL; 7985 } 7986 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7987 7988 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7989 { 7990 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7991 } 7992 7993 static struct rtnl_hw_stats64 * 7994 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7995 enum netdev_offload_xstats_type type) 7996 { 7997 switch (type) { 7998 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7999 return dev->offload_xstats_l3; 8000 } 8001 8002 WARN_ON(1); 8003 return NULL; 8004 } 8005 8006 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8007 enum netdev_offload_xstats_type type) 8008 { 8009 ASSERT_RTNL(); 8010 8011 return netdev_offload_xstats_get_ptr(dev, type); 8012 } 8013 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8014 8015 struct netdev_notifier_offload_xstats_ru { 8016 bool used; 8017 }; 8018 8019 struct netdev_notifier_offload_xstats_rd { 8020 struct rtnl_hw_stats64 stats; 8021 bool used; 8022 }; 8023 8024 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8025 const struct rtnl_hw_stats64 *src) 8026 { 8027 dest->rx_packets += src->rx_packets; 8028 dest->tx_packets += src->tx_packets; 8029 dest->rx_bytes += src->rx_bytes; 8030 dest->tx_bytes += src->tx_bytes; 8031 dest->rx_errors += src->rx_errors; 8032 dest->tx_errors += src->tx_errors; 8033 dest->rx_dropped += src->rx_dropped; 8034 dest->tx_dropped += src->tx_dropped; 8035 dest->multicast += src->multicast; 8036 } 8037 8038 static int netdev_offload_xstats_get_used(struct net_device *dev, 8039 enum netdev_offload_xstats_type type, 8040 bool *p_used, 8041 struct netlink_ext_ack *extack) 8042 { 8043 struct netdev_notifier_offload_xstats_ru report_used = {}; 8044 struct netdev_notifier_offload_xstats_info info = { 8045 .info.dev = dev, 8046 .info.extack = extack, 8047 .type = type, 8048 .report_used = &report_used, 8049 }; 8050 int rc; 8051 8052 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8053 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8054 &info.info); 8055 *p_used = report_used.used; 8056 return notifier_to_errno(rc); 8057 } 8058 8059 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8060 enum netdev_offload_xstats_type type, 8061 struct rtnl_hw_stats64 *p_stats, 8062 bool *p_used, 8063 struct netlink_ext_ack *extack) 8064 { 8065 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8066 struct netdev_notifier_offload_xstats_info info = { 8067 .info.dev = dev, 8068 .info.extack = extack, 8069 .type = type, 8070 .report_delta = &report_delta, 8071 }; 8072 struct rtnl_hw_stats64 *stats; 8073 int rc; 8074 8075 stats = netdev_offload_xstats_get_ptr(dev, type); 8076 if (WARN_ON(!stats)) 8077 return -EINVAL; 8078 8079 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8080 &info.info); 8081 8082 /* Cache whatever we got, even if there was an error, otherwise the 8083 * successful stats retrievals would get lost. 8084 */ 8085 netdev_hw_stats64_add(stats, &report_delta.stats); 8086 8087 if (p_stats) 8088 *p_stats = *stats; 8089 *p_used = report_delta.used; 8090 8091 return notifier_to_errno(rc); 8092 } 8093 8094 int netdev_offload_xstats_get(struct net_device *dev, 8095 enum netdev_offload_xstats_type type, 8096 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8097 struct netlink_ext_ack *extack) 8098 { 8099 ASSERT_RTNL(); 8100 8101 if (p_stats) 8102 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8103 p_used, extack); 8104 else 8105 return netdev_offload_xstats_get_used(dev, type, p_used, 8106 extack); 8107 } 8108 EXPORT_SYMBOL(netdev_offload_xstats_get); 8109 8110 void 8111 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8112 const struct rtnl_hw_stats64 *stats) 8113 { 8114 report_delta->used = true; 8115 netdev_hw_stats64_add(&report_delta->stats, stats); 8116 } 8117 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8118 8119 void 8120 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8121 { 8122 report_used->used = true; 8123 } 8124 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8125 8126 void netdev_offload_xstats_push_delta(struct net_device *dev, 8127 enum netdev_offload_xstats_type type, 8128 const struct rtnl_hw_stats64 *p_stats) 8129 { 8130 struct rtnl_hw_stats64 *stats; 8131 8132 ASSERT_RTNL(); 8133 8134 stats = netdev_offload_xstats_get_ptr(dev, type); 8135 if (WARN_ON(!stats)) 8136 return; 8137 8138 netdev_hw_stats64_add(stats, p_stats); 8139 } 8140 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8141 8142 /** 8143 * netdev_get_xmit_slave - Get the xmit slave of master device 8144 * @dev: device 8145 * @skb: The packet 8146 * @all_slaves: assume all the slaves are active 8147 * 8148 * The reference counters are not incremented so the caller must be 8149 * careful with locks. The caller must hold RCU lock. 8150 * %NULL is returned if no slave is found. 8151 */ 8152 8153 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8154 struct sk_buff *skb, 8155 bool all_slaves) 8156 { 8157 const struct net_device_ops *ops = dev->netdev_ops; 8158 8159 if (!ops->ndo_get_xmit_slave) 8160 return NULL; 8161 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8162 } 8163 EXPORT_SYMBOL(netdev_get_xmit_slave); 8164 8165 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8166 struct sock *sk) 8167 { 8168 const struct net_device_ops *ops = dev->netdev_ops; 8169 8170 if (!ops->ndo_sk_get_lower_dev) 8171 return NULL; 8172 return ops->ndo_sk_get_lower_dev(dev, sk); 8173 } 8174 8175 /** 8176 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8177 * @dev: device 8178 * @sk: the socket 8179 * 8180 * %NULL is returned if no lower device is found. 8181 */ 8182 8183 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8184 struct sock *sk) 8185 { 8186 struct net_device *lower; 8187 8188 lower = netdev_sk_get_lower_dev(dev, sk); 8189 while (lower) { 8190 dev = lower; 8191 lower = netdev_sk_get_lower_dev(dev, sk); 8192 } 8193 8194 return dev; 8195 } 8196 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8197 8198 static void netdev_adjacent_add_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_add(iter->dev, dev, 8208 &iter->dev->adj_list.lower); 8209 netdev_adjacent_sysfs_add(dev, iter->dev, 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_add(iter->dev, dev, 8217 &iter->dev->adj_list.upper); 8218 netdev_adjacent_sysfs_add(dev, iter->dev, 8219 &dev->adj_list.lower); 8220 } 8221 } 8222 8223 static void netdev_adjacent_del_links(struct net_device *dev) 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, dev->name, 8233 &iter->dev->adj_list.lower); 8234 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8235 &dev->adj_list.upper); 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, dev->name, 8242 &iter->dev->adj_list.upper); 8243 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8244 &dev->adj_list.lower); 8245 } 8246 } 8247 8248 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8249 { 8250 struct netdev_adjacent *iter; 8251 8252 struct net *net = dev_net(dev); 8253 8254 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8255 if (!net_eq(net, dev_net(iter->dev))) 8256 continue; 8257 netdev_adjacent_sysfs_del(iter->dev, oldname, 8258 &iter->dev->adj_list.lower); 8259 netdev_adjacent_sysfs_add(iter->dev, dev, 8260 &iter->dev->adj_list.lower); 8261 } 8262 8263 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8264 if (!net_eq(net, dev_net(iter->dev))) 8265 continue; 8266 netdev_adjacent_sysfs_del(iter->dev, oldname, 8267 &iter->dev->adj_list.upper); 8268 netdev_adjacent_sysfs_add(iter->dev, dev, 8269 &iter->dev->adj_list.upper); 8270 } 8271 } 8272 8273 void *netdev_lower_dev_get_private(struct net_device *dev, 8274 struct net_device *lower_dev) 8275 { 8276 struct netdev_adjacent *lower; 8277 8278 if (!lower_dev) 8279 return NULL; 8280 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8281 if (!lower) 8282 return NULL; 8283 8284 return lower->private; 8285 } 8286 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8287 8288 8289 /** 8290 * netdev_lower_state_changed - Dispatch event about lower device state change 8291 * @lower_dev: device 8292 * @lower_state_info: state to dispatch 8293 * 8294 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8295 * The caller must hold the RTNL lock. 8296 */ 8297 void netdev_lower_state_changed(struct net_device *lower_dev, 8298 void *lower_state_info) 8299 { 8300 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8301 .info.dev = lower_dev, 8302 }; 8303 8304 ASSERT_RTNL(); 8305 changelowerstate_info.lower_state_info = lower_state_info; 8306 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8307 &changelowerstate_info.info); 8308 } 8309 EXPORT_SYMBOL(netdev_lower_state_changed); 8310 8311 static void dev_change_rx_flags(struct net_device *dev, int flags) 8312 { 8313 const struct net_device_ops *ops = dev->netdev_ops; 8314 8315 if (ops->ndo_change_rx_flags) 8316 ops->ndo_change_rx_flags(dev, flags); 8317 } 8318 8319 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8320 { 8321 unsigned int old_flags = dev->flags; 8322 kuid_t uid; 8323 kgid_t gid; 8324 8325 ASSERT_RTNL(); 8326 8327 dev->flags |= IFF_PROMISC; 8328 dev->promiscuity += inc; 8329 if (dev->promiscuity == 0) { 8330 /* 8331 * Avoid overflow. 8332 * If inc causes overflow, untouch promisc and return error. 8333 */ 8334 if (inc < 0) 8335 dev->flags &= ~IFF_PROMISC; 8336 else { 8337 dev->promiscuity -= inc; 8338 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8339 return -EOVERFLOW; 8340 } 8341 } 8342 if (dev->flags != old_flags) { 8343 netdev_info(dev, "%s promiscuous mode\n", 8344 dev->flags & IFF_PROMISC ? "entered" : "left"); 8345 if (audit_enabled) { 8346 current_uid_gid(&uid, &gid); 8347 audit_log(audit_context(), GFP_ATOMIC, 8348 AUDIT_ANOM_PROMISCUOUS, 8349 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8350 dev->name, (dev->flags & IFF_PROMISC), 8351 (old_flags & IFF_PROMISC), 8352 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8353 from_kuid(&init_user_ns, uid), 8354 from_kgid(&init_user_ns, gid), 8355 audit_get_sessionid(current)); 8356 } 8357 8358 dev_change_rx_flags(dev, IFF_PROMISC); 8359 } 8360 if (notify) 8361 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8362 return 0; 8363 } 8364 8365 /** 8366 * dev_set_promiscuity - update promiscuity count on a device 8367 * @dev: device 8368 * @inc: modifier 8369 * 8370 * Add or remove promiscuity from a device. While the count in the device 8371 * remains above zero the interface remains promiscuous. Once it hits zero 8372 * the device reverts back to normal filtering operation. A negative inc 8373 * value is used to drop promiscuity on the device. 8374 * Return 0 if successful or a negative errno code on error. 8375 */ 8376 int dev_set_promiscuity(struct net_device *dev, int inc) 8377 { 8378 unsigned int old_flags = dev->flags; 8379 int err; 8380 8381 err = __dev_set_promiscuity(dev, inc, true); 8382 if (err < 0) 8383 return err; 8384 if (dev->flags != old_flags) 8385 dev_set_rx_mode(dev); 8386 return err; 8387 } 8388 EXPORT_SYMBOL(dev_set_promiscuity); 8389 8390 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8391 { 8392 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8393 8394 ASSERT_RTNL(); 8395 8396 dev->flags |= IFF_ALLMULTI; 8397 dev->allmulti += inc; 8398 if (dev->allmulti == 0) { 8399 /* 8400 * Avoid overflow. 8401 * If inc causes overflow, untouch allmulti and return error. 8402 */ 8403 if (inc < 0) 8404 dev->flags &= ~IFF_ALLMULTI; 8405 else { 8406 dev->allmulti -= inc; 8407 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8408 return -EOVERFLOW; 8409 } 8410 } 8411 if (dev->flags ^ old_flags) { 8412 netdev_info(dev, "%s allmulticast mode\n", 8413 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8414 dev_change_rx_flags(dev, IFF_ALLMULTI); 8415 dev_set_rx_mode(dev); 8416 if (notify) 8417 __dev_notify_flags(dev, old_flags, 8418 dev->gflags ^ old_gflags, 0, NULL); 8419 } 8420 return 0; 8421 } 8422 8423 /** 8424 * dev_set_allmulti - update allmulti count on a device 8425 * @dev: device 8426 * @inc: modifier 8427 * 8428 * Add or remove reception of all multicast frames to a device. While the 8429 * count in the device remains above zero the interface remains listening 8430 * to all interfaces. Once it hits zero the device reverts back to normal 8431 * filtering operation. A negative @inc value is used to drop the counter 8432 * when releasing a resource needing all multicasts. 8433 * Return 0 if successful or a negative errno code on error. 8434 */ 8435 8436 int dev_set_allmulti(struct net_device *dev, int inc) 8437 { 8438 return __dev_set_allmulti(dev, inc, true); 8439 } 8440 EXPORT_SYMBOL(dev_set_allmulti); 8441 8442 /* 8443 * Upload unicast and multicast address lists to device and 8444 * configure RX filtering. When the device doesn't support unicast 8445 * filtering it is put in promiscuous mode while unicast addresses 8446 * are present. 8447 */ 8448 void __dev_set_rx_mode(struct net_device *dev) 8449 { 8450 const struct net_device_ops *ops = dev->netdev_ops; 8451 8452 /* dev_open will call this function so the list will stay sane. */ 8453 if (!(dev->flags&IFF_UP)) 8454 return; 8455 8456 if (!netif_device_present(dev)) 8457 return; 8458 8459 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8460 /* Unicast addresses changes may only happen under the rtnl, 8461 * therefore calling __dev_set_promiscuity here is safe. 8462 */ 8463 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8464 __dev_set_promiscuity(dev, 1, false); 8465 dev->uc_promisc = true; 8466 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8467 __dev_set_promiscuity(dev, -1, false); 8468 dev->uc_promisc = false; 8469 } 8470 } 8471 8472 if (ops->ndo_set_rx_mode) 8473 ops->ndo_set_rx_mode(dev); 8474 } 8475 8476 void dev_set_rx_mode(struct net_device *dev) 8477 { 8478 netif_addr_lock_bh(dev); 8479 __dev_set_rx_mode(dev); 8480 netif_addr_unlock_bh(dev); 8481 } 8482 8483 /** 8484 * dev_get_flags - get flags reported to userspace 8485 * @dev: device 8486 * 8487 * Get the combination of flag bits exported through APIs to userspace. 8488 */ 8489 unsigned int dev_get_flags(const struct net_device *dev) 8490 { 8491 unsigned int flags; 8492 8493 flags = (dev->flags & ~(IFF_PROMISC | 8494 IFF_ALLMULTI | 8495 IFF_RUNNING | 8496 IFF_LOWER_UP | 8497 IFF_DORMANT)) | 8498 (dev->gflags & (IFF_PROMISC | 8499 IFF_ALLMULTI)); 8500 8501 if (netif_running(dev)) { 8502 if (netif_oper_up(dev)) 8503 flags |= IFF_RUNNING; 8504 if (netif_carrier_ok(dev)) 8505 flags |= IFF_LOWER_UP; 8506 if (netif_dormant(dev)) 8507 flags |= IFF_DORMANT; 8508 } 8509 8510 return flags; 8511 } 8512 EXPORT_SYMBOL(dev_get_flags); 8513 8514 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8515 struct netlink_ext_ack *extack) 8516 { 8517 unsigned int old_flags = dev->flags; 8518 int ret; 8519 8520 ASSERT_RTNL(); 8521 8522 /* 8523 * Set the flags on our device. 8524 */ 8525 8526 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8527 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8528 IFF_AUTOMEDIA)) | 8529 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8530 IFF_ALLMULTI)); 8531 8532 /* 8533 * Load in the correct multicast list now the flags have changed. 8534 */ 8535 8536 if ((old_flags ^ flags) & IFF_MULTICAST) 8537 dev_change_rx_flags(dev, IFF_MULTICAST); 8538 8539 dev_set_rx_mode(dev); 8540 8541 /* 8542 * Have we downed the interface. We handle IFF_UP ourselves 8543 * according to user attempts to set it, rather than blindly 8544 * setting it. 8545 */ 8546 8547 ret = 0; 8548 if ((old_flags ^ flags) & IFF_UP) { 8549 if (old_flags & IFF_UP) 8550 __dev_close(dev); 8551 else 8552 ret = __dev_open(dev, extack); 8553 } 8554 8555 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8556 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8557 unsigned int old_flags = dev->flags; 8558 8559 dev->gflags ^= IFF_PROMISC; 8560 8561 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8562 if (dev->flags != old_flags) 8563 dev_set_rx_mode(dev); 8564 } 8565 8566 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8567 * is important. Some (broken) drivers set IFF_PROMISC, when 8568 * IFF_ALLMULTI is requested not asking us and not reporting. 8569 */ 8570 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8571 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8572 8573 dev->gflags ^= IFF_ALLMULTI; 8574 __dev_set_allmulti(dev, inc, false); 8575 } 8576 8577 return ret; 8578 } 8579 8580 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8581 unsigned int gchanges, u32 portid, 8582 const struct nlmsghdr *nlh) 8583 { 8584 unsigned int changes = dev->flags ^ old_flags; 8585 8586 if (gchanges) 8587 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8588 8589 if (changes & IFF_UP) { 8590 if (dev->flags & IFF_UP) 8591 call_netdevice_notifiers(NETDEV_UP, dev); 8592 else 8593 call_netdevice_notifiers(NETDEV_DOWN, dev); 8594 } 8595 8596 if (dev->flags & IFF_UP && 8597 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8598 struct netdev_notifier_change_info change_info = { 8599 .info = { 8600 .dev = dev, 8601 }, 8602 .flags_changed = changes, 8603 }; 8604 8605 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8606 } 8607 } 8608 8609 /** 8610 * dev_change_flags - change device settings 8611 * @dev: device 8612 * @flags: device state flags 8613 * @extack: netlink extended ack 8614 * 8615 * Change settings on device based state flags. The flags are 8616 * in the userspace exported format. 8617 */ 8618 int dev_change_flags(struct net_device *dev, unsigned int flags, 8619 struct netlink_ext_ack *extack) 8620 { 8621 int ret; 8622 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8623 8624 ret = __dev_change_flags(dev, flags, extack); 8625 if (ret < 0) 8626 return ret; 8627 8628 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8629 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8630 return ret; 8631 } 8632 EXPORT_SYMBOL(dev_change_flags); 8633 8634 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8635 { 8636 const struct net_device_ops *ops = dev->netdev_ops; 8637 8638 if (ops->ndo_change_mtu) 8639 return ops->ndo_change_mtu(dev, new_mtu); 8640 8641 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8642 WRITE_ONCE(dev->mtu, new_mtu); 8643 return 0; 8644 } 8645 EXPORT_SYMBOL(__dev_set_mtu); 8646 8647 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8648 struct netlink_ext_ack *extack) 8649 { 8650 /* MTU must be positive, and in range */ 8651 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8652 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8653 return -EINVAL; 8654 } 8655 8656 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8657 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8658 return -EINVAL; 8659 } 8660 return 0; 8661 } 8662 8663 /** 8664 * dev_set_mtu_ext - Change maximum transfer unit 8665 * @dev: device 8666 * @new_mtu: new transfer unit 8667 * @extack: netlink extended ack 8668 * 8669 * Change the maximum transfer size of the network device. 8670 */ 8671 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8672 struct netlink_ext_ack *extack) 8673 { 8674 int err, orig_mtu; 8675 8676 if (new_mtu == dev->mtu) 8677 return 0; 8678 8679 err = dev_validate_mtu(dev, new_mtu, extack); 8680 if (err) 8681 return err; 8682 8683 if (!netif_device_present(dev)) 8684 return -ENODEV; 8685 8686 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8687 err = notifier_to_errno(err); 8688 if (err) 8689 return err; 8690 8691 orig_mtu = dev->mtu; 8692 err = __dev_set_mtu(dev, new_mtu); 8693 8694 if (!err) { 8695 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8696 orig_mtu); 8697 err = notifier_to_errno(err); 8698 if (err) { 8699 /* setting mtu back and notifying everyone again, 8700 * so that they have a chance to revert changes. 8701 */ 8702 __dev_set_mtu(dev, orig_mtu); 8703 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8704 new_mtu); 8705 } 8706 } 8707 return err; 8708 } 8709 8710 int dev_set_mtu(struct net_device *dev, int new_mtu) 8711 { 8712 struct netlink_ext_ack extack; 8713 int err; 8714 8715 memset(&extack, 0, sizeof(extack)); 8716 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8717 if (err && extack._msg) 8718 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8719 return err; 8720 } 8721 EXPORT_SYMBOL(dev_set_mtu); 8722 8723 /** 8724 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8725 * @dev: device 8726 * @new_len: new tx queue length 8727 */ 8728 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8729 { 8730 unsigned int orig_len = dev->tx_queue_len; 8731 int res; 8732 8733 if (new_len != (unsigned int)new_len) 8734 return -ERANGE; 8735 8736 if (new_len != orig_len) { 8737 dev->tx_queue_len = new_len; 8738 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8739 res = notifier_to_errno(res); 8740 if (res) 8741 goto err_rollback; 8742 res = dev_qdisc_change_tx_queue_len(dev); 8743 if (res) 8744 goto err_rollback; 8745 } 8746 8747 return 0; 8748 8749 err_rollback: 8750 netdev_err(dev, "refused to change device tx_queue_len\n"); 8751 dev->tx_queue_len = orig_len; 8752 return res; 8753 } 8754 8755 /** 8756 * dev_set_group - Change group this device belongs to 8757 * @dev: device 8758 * @new_group: group this device should belong to 8759 */ 8760 void dev_set_group(struct net_device *dev, int new_group) 8761 { 8762 dev->group = new_group; 8763 } 8764 8765 /** 8766 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8767 * @dev: device 8768 * @addr: new address 8769 * @extack: netlink extended ack 8770 */ 8771 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8772 struct netlink_ext_ack *extack) 8773 { 8774 struct netdev_notifier_pre_changeaddr_info info = { 8775 .info.dev = dev, 8776 .info.extack = extack, 8777 .dev_addr = addr, 8778 }; 8779 int rc; 8780 8781 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8782 return notifier_to_errno(rc); 8783 } 8784 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8785 8786 /** 8787 * dev_set_mac_address - Change Media Access Control Address 8788 * @dev: device 8789 * @sa: new address 8790 * @extack: netlink extended ack 8791 * 8792 * Change the hardware (MAC) address of the device 8793 */ 8794 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8795 struct netlink_ext_ack *extack) 8796 { 8797 const struct net_device_ops *ops = dev->netdev_ops; 8798 int err; 8799 8800 if (!ops->ndo_set_mac_address) 8801 return -EOPNOTSUPP; 8802 if (sa->sa_family != dev->type) 8803 return -EINVAL; 8804 if (!netif_device_present(dev)) 8805 return -ENODEV; 8806 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8807 if (err) 8808 return err; 8809 err = ops->ndo_set_mac_address(dev, sa); 8810 if (err) 8811 return err; 8812 dev->addr_assign_type = NET_ADDR_SET; 8813 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8814 add_device_randomness(dev->dev_addr, dev->addr_len); 8815 return 0; 8816 } 8817 EXPORT_SYMBOL(dev_set_mac_address); 8818 8819 static DECLARE_RWSEM(dev_addr_sem); 8820 8821 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8822 struct netlink_ext_ack *extack) 8823 { 8824 int ret; 8825 8826 down_write(&dev_addr_sem); 8827 ret = dev_set_mac_address(dev, sa, extack); 8828 up_write(&dev_addr_sem); 8829 return ret; 8830 } 8831 EXPORT_SYMBOL(dev_set_mac_address_user); 8832 8833 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8834 { 8835 size_t size = sizeof(sa->sa_data_min); 8836 struct net_device *dev; 8837 int ret = 0; 8838 8839 down_read(&dev_addr_sem); 8840 rcu_read_lock(); 8841 8842 dev = dev_get_by_name_rcu(net, dev_name); 8843 if (!dev) { 8844 ret = -ENODEV; 8845 goto unlock; 8846 } 8847 if (!dev->addr_len) 8848 memset(sa->sa_data, 0, size); 8849 else 8850 memcpy(sa->sa_data, dev->dev_addr, 8851 min_t(size_t, size, dev->addr_len)); 8852 sa->sa_family = dev->type; 8853 8854 unlock: 8855 rcu_read_unlock(); 8856 up_read(&dev_addr_sem); 8857 return ret; 8858 } 8859 EXPORT_SYMBOL(dev_get_mac_address); 8860 8861 /** 8862 * dev_change_carrier - Change device carrier 8863 * @dev: device 8864 * @new_carrier: new value 8865 * 8866 * Change device carrier 8867 */ 8868 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8869 { 8870 const struct net_device_ops *ops = dev->netdev_ops; 8871 8872 if (!ops->ndo_change_carrier) 8873 return -EOPNOTSUPP; 8874 if (!netif_device_present(dev)) 8875 return -ENODEV; 8876 return ops->ndo_change_carrier(dev, new_carrier); 8877 } 8878 8879 /** 8880 * dev_get_phys_port_id - Get device physical port ID 8881 * @dev: device 8882 * @ppid: port ID 8883 * 8884 * Get device physical port ID 8885 */ 8886 int dev_get_phys_port_id(struct net_device *dev, 8887 struct netdev_phys_item_id *ppid) 8888 { 8889 const struct net_device_ops *ops = dev->netdev_ops; 8890 8891 if (!ops->ndo_get_phys_port_id) 8892 return -EOPNOTSUPP; 8893 return ops->ndo_get_phys_port_id(dev, ppid); 8894 } 8895 8896 /** 8897 * dev_get_phys_port_name - Get device physical port name 8898 * @dev: device 8899 * @name: port name 8900 * @len: limit of bytes to copy to name 8901 * 8902 * Get device physical port name 8903 */ 8904 int dev_get_phys_port_name(struct net_device *dev, 8905 char *name, size_t len) 8906 { 8907 const struct net_device_ops *ops = dev->netdev_ops; 8908 int err; 8909 8910 if (ops->ndo_get_phys_port_name) { 8911 err = ops->ndo_get_phys_port_name(dev, name, len); 8912 if (err != -EOPNOTSUPP) 8913 return err; 8914 } 8915 return devlink_compat_phys_port_name_get(dev, name, len); 8916 } 8917 8918 /** 8919 * dev_get_port_parent_id - Get the device's port parent identifier 8920 * @dev: network device 8921 * @ppid: pointer to a storage for the port's parent identifier 8922 * @recurse: allow/disallow recursion to lower devices 8923 * 8924 * Get the devices's port parent identifier 8925 */ 8926 int dev_get_port_parent_id(struct net_device *dev, 8927 struct netdev_phys_item_id *ppid, 8928 bool recurse) 8929 { 8930 const struct net_device_ops *ops = dev->netdev_ops; 8931 struct netdev_phys_item_id first = { }; 8932 struct net_device *lower_dev; 8933 struct list_head *iter; 8934 int err; 8935 8936 if (ops->ndo_get_port_parent_id) { 8937 err = ops->ndo_get_port_parent_id(dev, ppid); 8938 if (err != -EOPNOTSUPP) 8939 return err; 8940 } 8941 8942 err = devlink_compat_switch_id_get(dev, ppid); 8943 if (!recurse || err != -EOPNOTSUPP) 8944 return err; 8945 8946 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8947 err = dev_get_port_parent_id(lower_dev, ppid, true); 8948 if (err) 8949 break; 8950 if (!first.id_len) 8951 first = *ppid; 8952 else if (memcmp(&first, ppid, sizeof(*ppid))) 8953 return -EOPNOTSUPP; 8954 } 8955 8956 return err; 8957 } 8958 EXPORT_SYMBOL(dev_get_port_parent_id); 8959 8960 /** 8961 * netdev_port_same_parent_id - Indicate if two network devices have 8962 * the same port parent identifier 8963 * @a: first network device 8964 * @b: second network device 8965 */ 8966 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8967 { 8968 struct netdev_phys_item_id a_id = { }; 8969 struct netdev_phys_item_id b_id = { }; 8970 8971 if (dev_get_port_parent_id(a, &a_id, true) || 8972 dev_get_port_parent_id(b, &b_id, true)) 8973 return false; 8974 8975 return netdev_phys_item_id_same(&a_id, &b_id); 8976 } 8977 EXPORT_SYMBOL(netdev_port_same_parent_id); 8978 8979 /** 8980 * dev_change_proto_down - set carrier according to proto_down. 8981 * 8982 * @dev: device 8983 * @proto_down: new value 8984 */ 8985 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8986 { 8987 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8988 return -EOPNOTSUPP; 8989 if (!netif_device_present(dev)) 8990 return -ENODEV; 8991 if (proto_down) 8992 netif_carrier_off(dev); 8993 else 8994 netif_carrier_on(dev); 8995 dev->proto_down = proto_down; 8996 return 0; 8997 } 8998 8999 /** 9000 * dev_change_proto_down_reason - proto down reason 9001 * 9002 * @dev: device 9003 * @mask: proto down mask 9004 * @value: proto down value 9005 */ 9006 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9007 u32 value) 9008 { 9009 int b; 9010 9011 if (!mask) { 9012 dev->proto_down_reason = value; 9013 } else { 9014 for_each_set_bit(b, &mask, 32) { 9015 if (value & (1 << b)) 9016 dev->proto_down_reason |= BIT(b); 9017 else 9018 dev->proto_down_reason &= ~BIT(b); 9019 } 9020 } 9021 } 9022 9023 struct bpf_xdp_link { 9024 struct bpf_link link; 9025 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9026 int flags; 9027 }; 9028 9029 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9030 { 9031 if (flags & XDP_FLAGS_HW_MODE) 9032 return XDP_MODE_HW; 9033 if (flags & XDP_FLAGS_DRV_MODE) 9034 return XDP_MODE_DRV; 9035 if (flags & XDP_FLAGS_SKB_MODE) 9036 return XDP_MODE_SKB; 9037 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9038 } 9039 9040 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9041 { 9042 switch (mode) { 9043 case XDP_MODE_SKB: 9044 return generic_xdp_install; 9045 case XDP_MODE_DRV: 9046 case XDP_MODE_HW: 9047 return dev->netdev_ops->ndo_bpf; 9048 default: 9049 return NULL; 9050 } 9051 } 9052 9053 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9054 enum bpf_xdp_mode mode) 9055 { 9056 return dev->xdp_state[mode].link; 9057 } 9058 9059 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9060 enum bpf_xdp_mode mode) 9061 { 9062 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9063 9064 if (link) 9065 return link->link.prog; 9066 return dev->xdp_state[mode].prog; 9067 } 9068 9069 u8 dev_xdp_prog_count(struct net_device *dev) 9070 { 9071 u8 count = 0; 9072 int i; 9073 9074 for (i = 0; i < __MAX_XDP_MODE; i++) 9075 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9076 count++; 9077 return count; 9078 } 9079 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9080 9081 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9082 { 9083 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9084 9085 return prog ? prog->aux->id : 0; 9086 } 9087 9088 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9089 struct bpf_xdp_link *link) 9090 { 9091 dev->xdp_state[mode].link = link; 9092 dev->xdp_state[mode].prog = NULL; 9093 } 9094 9095 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9096 struct bpf_prog *prog) 9097 { 9098 dev->xdp_state[mode].link = NULL; 9099 dev->xdp_state[mode].prog = prog; 9100 } 9101 9102 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9103 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9104 u32 flags, struct bpf_prog *prog) 9105 { 9106 struct netdev_bpf xdp; 9107 int err; 9108 9109 memset(&xdp, 0, sizeof(xdp)); 9110 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9111 xdp.extack = extack; 9112 xdp.flags = flags; 9113 xdp.prog = prog; 9114 9115 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9116 * "moved" into driver), so they don't increment it on their own, but 9117 * they do decrement refcnt when program is detached or replaced. 9118 * Given net_device also owns link/prog, we need to bump refcnt here 9119 * to prevent drivers from underflowing it. 9120 */ 9121 if (prog) 9122 bpf_prog_inc(prog); 9123 err = bpf_op(dev, &xdp); 9124 if (err) { 9125 if (prog) 9126 bpf_prog_put(prog); 9127 return err; 9128 } 9129 9130 if (mode != XDP_MODE_HW) 9131 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9132 9133 return 0; 9134 } 9135 9136 static void dev_xdp_uninstall(struct net_device *dev) 9137 { 9138 struct bpf_xdp_link *link; 9139 struct bpf_prog *prog; 9140 enum bpf_xdp_mode mode; 9141 bpf_op_t bpf_op; 9142 9143 ASSERT_RTNL(); 9144 9145 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9146 prog = dev_xdp_prog(dev, mode); 9147 if (!prog) 9148 continue; 9149 9150 bpf_op = dev_xdp_bpf_op(dev, mode); 9151 if (!bpf_op) 9152 continue; 9153 9154 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9155 9156 /* auto-detach link from net device */ 9157 link = dev_xdp_link(dev, mode); 9158 if (link) 9159 link->dev = NULL; 9160 else 9161 bpf_prog_put(prog); 9162 9163 dev_xdp_set_link(dev, mode, NULL); 9164 } 9165 } 9166 9167 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9168 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9169 struct bpf_prog *old_prog, u32 flags) 9170 { 9171 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9172 struct bpf_prog *cur_prog; 9173 struct net_device *upper; 9174 struct list_head *iter; 9175 enum bpf_xdp_mode mode; 9176 bpf_op_t bpf_op; 9177 int err; 9178 9179 ASSERT_RTNL(); 9180 9181 /* either link or prog attachment, never both */ 9182 if (link && (new_prog || old_prog)) 9183 return -EINVAL; 9184 /* link supports only XDP mode flags */ 9185 if (link && (flags & ~XDP_FLAGS_MODES)) { 9186 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9187 return -EINVAL; 9188 } 9189 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9190 if (num_modes > 1) { 9191 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9192 return -EINVAL; 9193 } 9194 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9195 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9196 NL_SET_ERR_MSG(extack, 9197 "More than one program loaded, unset mode is ambiguous"); 9198 return -EINVAL; 9199 } 9200 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9201 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9202 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9203 return -EINVAL; 9204 } 9205 9206 mode = dev_xdp_mode(dev, flags); 9207 /* can't replace attached link */ 9208 if (dev_xdp_link(dev, mode)) { 9209 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9210 return -EBUSY; 9211 } 9212 9213 /* don't allow if an upper device already has a program */ 9214 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9215 if (dev_xdp_prog_count(upper) > 0) { 9216 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9217 return -EEXIST; 9218 } 9219 } 9220 9221 cur_prog = dev_xdp_prog(dev, mode); 9222 /* can't replace attached prog with link */ 9223 if (link && cur_prog) { 9224 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9225 return -EBUSY; 9226 } 9227 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9228 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9229 return -EEXIST; 9230 } 9231 9232 /* put effective new program into new_prog */ 9233 if (link) 9234 new_prog = link->link.prog; 9235 9236 if (new_prog) { 9237 bool offload = mode == XDP_MODE_HW; 9238 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9239 ? XDP_MODE_DRV : XDP_MODE_SKB; 9240 9241 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9242 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9243 return -EBUSY; 9244 } 9245 if (!offload && dev_xdp_prog(dev, other_mode)) { 9246 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9247 return -EEXIST; 9248 } 9249 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9250 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9251 return -EINVAL; 9252 } 9253 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9254 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9255 return -EINVAL; 9256 } 9257 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9258 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9259 return -EINVAL; 9260 } 9261 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9262 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9263 return -EINVAL; 9264 } 9265 } 9266 9267 /* don't call drivers if the effective program didn't change */ 9268 if (new_prog != cur_prog) { 9269 bpf_op = dev_xdp_bpf_op(dev, mode); 9270 if (!bpf_op) { 9271 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9272 return -EOPNOTSUPP; 9273 } 9274 9275 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9276 if (err) 9277 return err; 9278 } 9279 9280 if (link) 9281 dev_xdp_set_link(dev, mode, link); 9282 else 9283 dev_xdp_set_prog(dev, mode, new_prog); 9284 if (cur_prog) 9285 bpf_prog_put(cur_prog); 9286 9287 return 0; 9288 } 9289 9290 static int dev_xdp_attach_link(struct net_device *dev, 9291 struct netlink_ext_ack *extack, 9292 struct bpf_xdp_link *link) 9293 { 9294 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9295 } 9296 9297 static int dev_xdp_detach_link(struct net_device *dev, 9298 struct netlink_ext_ack *extack, 9299 struct bpf_xdp_link *link) 9300 { 9301 enum bpf_xdp_mode mode; 9302 bpf_op_t bpf_op; 9303 9304 ASSERT_RTNL(); 9305 9306 mode = dev_xdp_mode(dev, link->flags); 9307 if (dev_xdp_link(dev, mode) != link) 9308 return -EINVAL; 9309 9310 bpf_op = dev_xdp_bpf_op(dev, mode); 9311 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9312 dev_xdp_set_link(dev, mode, NULL); 9313 return 0; 9314 } 9315 9316 static void bpf_xdp_link_release(struct bpf_link *link) 9317 { 9318 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9319 9320 rtnl_lock(); 9321 9322 /* if racing with net_device's tear down, xdp_link->dev might be 9323 * already NULL, in which case link was already auto-detached 9324 */ 9325 if (xdp_link->dev) { 9326 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9327 xdp_link->dev = NULL; 9328 } 9329 9330 rtnl_unlock(); 9331 } 9332 9333 static int bpf_xdp_link_detach(struct bpf_link *link) 9334 { 9335 bpf_xdp_link_release(link); 9336 return 0; 9337 } 9338 9339 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9340 { 9341 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9342 9343 kfree(xdp_link); 9344 } 9345 9346 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9347 struct seq_file *seq) 9348 { 9349 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9350 u32 ifindex = 0; 9351 9352 rtnl_lock(); 9353 if (xdp_link->dev) 9354 ifindex = xdp_link->dev->ifindex; 9355 rtnl_unlock(); 9356 9357 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9358 } 9359 9360 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9361 struct bpf_link_info *info) 9362 { 9363 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9364 u32 ifindex = 0; 9365 9366 rtnl_lock(); 9367 if (xdp_link->dev) 9368 ifindex = xdp_link->dev->ifindex; 9369 rtnl_unlock(); 9370 9371 info->xdp.ifindex = ifindex; 9372 return 0; 9373 } 9374 9375 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9376 struct bpf_prog *old_prog) 9377 { 9378 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9379 enum bpf_xdp_mode mode; 9380 bpf_op_t bpf_op; 9381 int err = 0; 9382 9383 rtnl_lock(); 9384 9385 /* link might have been auto-released already, so fail */ 9386 if (!xdp_link->dev) { 9387 err = -ENOLINK; 9388 goto out_unlock; 9389 } 9390 9391 if (old_prog && link->prog != old_prog) { 9392 err = -EPERM; 9393 goto out_unlock; 9394 } 9395 old_prog = link->prog; 9396 if (old_prog->type != new_prog->type || 9397 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9398 err = -EINVAL; 9399 goto out_unlock; 9400 } 9401 9402 if (old_prog == new_prog) { 9403 /* no-op, don't disturb drivers */ 9404 bpf_prog_put(new_prog); 9405 goto out_unlock; 9406 } 9407 9408 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9409 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9410 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9411 xdp_link->flags, new_prog); 9412 if (err) 9413 goto out_unlock; 9414 9415 old_prog = xchg(&link->prog, new_prog); 9416 bpf_prog_put(old_prog); 9417 9418 out_unlock: 9419 rtnl_unlock(); 9420 return err; 9421 } 9422 9423 static const struct bpf_link_ops bpf_xdp_link_lops = { 9424 .release = bpf_xdp_link_release, 9425 .dealloc = bpf_xdp_link_dealloc, 9426 .detach = bpf_xdp_link_detach, 9427 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9428 .fill_link_info = bpf_xdp_link_fill_link_info, 9429 .update_prog = bpf_xdp_link_update, 9430 }; 9431 9432 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9433 { 9434 struct net *net = current->nsproxy->net_ns; 9435 struct bpf_link_primer link_primer; 9436 struct bpf_xdp_link *link; 9437 struct net_device *dev; 9438 int err, fd; 9439 9440 rtnl_lock(); 9441 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9442 if (!dev) { 9443 rtnl_unlock(); 9444 return -EINVAL; 9445 } 9446 9447 link = kzalloc(sizeof(*link), GFP_USER); 9448 if (!link) { 9449 err = -ENOMEM; 9450 goto unlock; 9451 } 9452 9453 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9454 link->dev = dev; 9455 link->flags = attr->link_create.flags; 9456 9457 err = bpf_link_prime(&link->link, &link_primer); 9458 if (err) { 9459 kfree(link); 9460 goto unlock; 9461 } 9462 9463 err = dev_xdp_attach_link(dev, NULL, link); 9464 rtnl_unlock(); 9465 9466 if (err) { 9467 link->dev = NULL; 9468 bpf_link_cleanup(&link_primer); 9469 goto out_put_dev; 9470 } 9471 9472 fd = bpf_link_settle(&link_primer); 9473 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9474 dev_put(dev); 9475 return fd; 9476 9477 unlock: 9478 rtnl_unlock(); 9479 9480 out_put_dev: 9481 dev_put(dev); 9482 return err; 9483 } 9484 9485 /** 9486 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9487 * @dev: device 9488 * @extack: netlink extended ack 9489 * @fd: new program fd or negative value to clear 9490 * @expected_fd: old program fd that userspace expects to replace or clear 9491 * @flags: xdp-related flags 9492 * 9493 * Set or clear a bpf program for a device 9494 */ 9495 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9496 int fd, int expected_fd, u32 flags) 9497 { 9498 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9499 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9500 int err; 9501 9502 ASSERT_RTNL(); 9503 9504 if (fd >= 0) { 9505 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9506 mode != XDP_MODE_SKB); 9507 if (IS_ERR(new_prog)) 9508 return PTR_ERR(new_prog); 9509 } 9510 9511 if (expected_fd >= 0) { 9512 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9513 mode != XDP_MODE_SKB); 9514 if (IS_ERR(old_prog)) { 9515 err = PTR_ERR(old_prog); 9516 old_prog = NULL; 9517 goto err_out; 9518 } 9519 } 9520 9521 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9522 9523 err_out: 9524 if (err && new_prog) 9525 bpf_prog_put(new_prog); 9526 if (old_prog) 9527 bpf_prog_put(old_prog); 9528 return err; 9529 } 9530 9531 /** 9532 * dev_new_index - allocate an ifindex 9533 * @net: the applicable net namespace 9534 * 9535 * Returns a suitable unique value for a new device interface 9536 * number. The caller must hold the rtnl semaphore or the 9537 * dev_base_lock to be sure it remains unique. 9538 */ 9539 static int dev_new_index(struct net *net) 9540 { 9541 int ifindex = net->ifindex; 9542 9543 for (;;) { 9544 if (++ifindex <= 0) 9545 ifindex = 1; 9546 if (!__dev_get_by_index(net, ifindex)) 9547 return net->ifindex = ifindex; 9548 } 9549 } 9550 9551 /* Delayed registration/unregisteration */ 9552 LIST_HEAD(net_todo_list); 9553 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9554 9555 static void net_set_todo(struct net_device *dev) 9556 { 9557 list_add_tail(&dev->todo_list, &net_todo_list); 9558 atomic_inc(&dev_net(dev)->dev_unreg_count); 9559 } 9560 9561 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9562 struct net_device *upper, netdev_features_t features) 9563 { 9564 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9565 netdev_features_t feature; 9566 int feature_bit; 9567 9568 for_each_netdev_feature(upper_disables, feature_bit) { 9569 feature = __NETIF_F_BIT(feature_bit); 9570 if (!(upper->wanted_features & feature) 9571 && (features & feature)) { 9572 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9573 &feature, upper->name); 9574 features &= ~feature; 9575 } 9576 } 9577 9578 return features; 9579 } 9580 9581 static void netdev_sync_lower_features(struct net_device *upper, 9582 struct net_device *lower, netdev_features_t features) 9583 { 9584 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9585 netdev_features_t feature; 9586 int feature_bit; 9587 9588 for_each_netdev_feature(upper_disables, feature_bit) { 9589 feature = __NETIF_F_BIT(feature_bit); 9590 if (!(features & feature) && (lower->features & feature)) { 9591 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9592 &feature, lower->name); 9593 lower->wanted_features &= ~feature; 9594 __netdev_update_features(lower); 9595 9596 if (unlikely(lower->features & feature)) 9597 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9598 &feature, lower->name); 9599 else 9600 netdev_features_change(lower); 9601 } 9602 } 9603 } 9604 9605 static netdev_features_t netdev_fix_features(struct net_device *dev, 9606 netdev_features_t features) 9607 { 9608 /* Fix illegal checksum combinations */ 9609 if ((features & NETIF_F_HW_CSUM) && 9610 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9611 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9612 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9613 } 9614 9615 /* TSO requires that SG is present as well. */ 9616 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9617 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9618 features &= ~NETIF_F_ALL_TSO; 9619 } 9620 9621 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9622 !(features & NETIF_F_IP_CSUM)) { 9623 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9624 features &= ~NETIF_F_TSO; 9625 features &= ~NETIF_F_TSO_ECN; 9626 } 9627 9628 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9629 !(features & NETIF_F_IPV6_CSUM)) { 9630 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9631 features &= ~NETIF_F_TSO6; 9632 } 9633 9634 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9635 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9636 features &= ~NETIF_F_TSO_MANGLEID; 9637 9638 /* TSO ECN requires that TSO is present as well. */ 9639 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9640 features &= ~NETIF_F_TSO_ECN; 9641 9642 /* Software GSO depends on SG. */ 9643 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9644 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9645 features &= ~NETIF_F_GSO; 9646 } 9647 9648 /* GSO partial features require GSO partial be set */ 9649 if ((features & dev->gso_partial_features) && 9650 !(features & NETIF_F_GSO_PARTIAL)) { 9651 netdev_dbg(dev, 9652 "Dropping partially supported GSO features since no GSO partial.\n"); 9653 features &= ~dev->gso_partial_features; 9654 } 9655 9656 if (!(features & NETIF_F_RXCSUM)) { 9657 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9658 * successfully merged by hardware must also have the 9659 * checksum verified by hardware. If the user does not 9660 * want to enable RXCSUM, logically, we should disable GRO_HW. 9661 */ 9662 if (features & NETIF_F_GRO_HW) { 9663 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9664 features &= ~NETIF_F_GRO_HW; 9665 } 9666 } 9667 9668 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9669 if (features & NETIF_F_RXFCS) { 9670 if (features & NETIF_F_LRO) { 9671 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9672 features &= ~NETIF_F_LRO; 9673 } 9674 9675 if (features & NETIF_F_GRO_HW) { 9676 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9677 features &= ~NETIF_F_GRO_HW; 9678 } 9679 } 9680 9681 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9682 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9683 features &= ~NETIF_F_LRO; 9684 } 9685 9686 if (features & NETIF_F_HW_TLS_TX) { 9687 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9688 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9689 bool hw_csum = features & NETIF_F_HW_CSUM; 9690 9691 if (!ip_csum && !hw_csum) { 9692 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9693 features &= ~NETIF_F_HW_TLS_TX; 9694 } 9695 } 9696 9697 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9698 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9699 features &= ~NETIF_F_HW_TLS_RX; 9700 } 9701 9702 return features; 9703 } 9704 9705 int __netdev_update_features(struct net_device *dev) 9706 { 9707 struct net_device *upper, *lower; 9708 netdev_features_t features; 9709 struct list_head *iter; 9710 int err = -1; 9711 9712 ASSERT_RTNL(); 9713 9714 features = netdev_get_wanted_features(dev); 9715 9716 if (dev->netdev_ops->ndo_fix_features) 9717 features = dev->netdev_ops->ndo_fix_features(dev, features); 9718 9719 /* driver might be less strict about feature dependencies */ 9720 features = netdev_fix_features(dev, features); 9721 9722 /* some features can't be enabled if they're off on an upper device */ 9723 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9724 features = netdev_sync_upper_features(dev, upper, features); 9725 9726 if (dev->features == features) 9727 goto sync_lower; 9728 9729 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9730 &dev->features, &features); 9731 9732 if (dev->netdev_ops->ndo_set_features) 9733 err = dev->netdev_ops->ndo_set_features(dev, features); 9734 else 9735 err = 0; 9736 9737 if (unlikely(err < 0)) { 9738 netdev_err(dev, 9739 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9740 err, &features, &dev->features); 9741 /* return non-0 since some features might have changed and 9742 * it's better to fire a spurious notification than miss it 9743 */ 9744 return -1; 9745 } 9746 9747 sync_lower: 9748 /* some features must be disabled on lower devices when disabled 9749 * on an upper device (think: bonding master or bridge) 9750 */ 9751 netdev_for_each_lower_dev(dev, lower, iter) 9752 netdev_sync_lower_features(dev, lower, features); 9753 9754 if (!err) { 9755 netdev_features_t diff = features ^ dev->features; 9756 9757 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9758 /* udp_tunnel_{get,drop}_rx_info both need 9759 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9760 * device, or they won't do anything. 9761 * Thus we need to update dev->features 9762 * *before* calling udp_tunnel_get_rx_info, 9763 * but *after* calling udp_tunnel_drop_rx_info. 9764 */ 9765 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9766 dev->features = features; 9767 udp_tunnel_get_rx_info(dev); 9768 } else { 9769 udp_tunnel_drop_rx_info(dev); 9770 } 9771 } 9772 9773 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9774 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9775 dev->features = features; 9776 err |= vlan_get_rx_ctag_filter_info(dev); 9777 } else { 9778 vlan_drop_rx_ctag_filter_info(dev); 9779 } 9780 } 9781 9782 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9783 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9784 dev->features = features; 9785 err |= vlan_get_rx_stag_filter_info(dev); 9786 } else { 9787 vlan_drop_rx_stag_filter_info(dev); 9788 } 9789 } 9790 9791 dev->features = features; 9792 } 9793 9794 return err < 0 ? 0 : 1; 9795 } 9796 9797 /** 9798 * netdev_update_features - recalculate device features 9799 * @dev: the device to check 9800 * 9801 * Recalculate dev->features set and send notifications if it 9802 * has changed. Should be called after driver or hardware dependent 9803 * conditions might have changed that influence the features. 9804 */ 9805 void netdev_update_features(struct net_device *dev) 9806 { 9807 if (__netdev_update_features(dev)) 9808 netdev_features_change(dev); 9809 } 9810 EXPORT_SYMBOL(netdev_update_features); 9811 9812 /** 9813 * netdev_change_features - recalculate device features 9814 * @dev: the device to check 9815 * 9816 * Recalculate dev->features set and send notifications even 9817 * if they have not changed. Should be called instead of 9818 * netdev_update_features() if also dev->vlan_features might 9819 * have changed to allow the changes to be propagated to stacked 9820 * VLAN devices. 9821 */ 9822 void netdev_change_features(struct net_device *dev) 9823 { 9824 __netdev_update_features(dev); 9825 netdev_features_change(dev); 9826 } 9827 EXPORT_SYMBOL(netdev_change_features); 9828 9829 /** 9830 * netif_stacked_transfer_operstate - transfer operstate 9831 * @rootdev: the root or lower level device to transfer state from 9832 * @dev: the device to transfer operstate to 9833 * 9834 * Transfer operational state from root to device. This is normally 9835 * called when a stacking relationship exists between the root 9836 * device and the device(a leaf device). 9837 */ 9838 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9839 struct net_device *dev) 9840 { 9841 if (rootdev->operstate == IF_OPER_DORMANT) 9842 netif_dormant_on(dev); 9843 else 9844 netif_dormant_off(dev); 9845 9846 if (rootdev->operstate == IF_OPER_TESTING) 9847 netif_testing_on(dev); 9848 else 9849 netif_testing_off(dev); 9850 9851 if (netif_carrier_ok(rootdev)) 9852 netif_carrier_on(dev); 9853 else 9854 netif_carrier_off(dev); 9855 } 9856 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9857 9858 static int netif_alloc_rx_queues(struct net_device *dev) 9859 { 9860 unsigned int i, count = dev->num_rx_queues; 9861 struct netdev_rx_queue *rx; 9862 size_t sz = count * sizeof(*rx); 9863 int err = 0; 9864 9865 BUG_ON(count < 1); 9866 9867 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9868 if (!rx) 9869 return -ENOMEM; 9870 9871 dev->_rx = rx; 9872 9873 for (i = 0; i < count; i++) { 9874 rx[i].dev = dev; 9875 9876 /* XDP RX-queue setup */ 9877 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9878 if (err < 0) 9879 goto err_rxq_info; 9880 } 9881 return 0; 9882 9883 err_rxq_info: 9884 /* Rollback successful reg's and free other resources */ 9885 while (i--) 9886 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9887 kvfree(dev->_rx); 9888 dev->_rx = NULL; 9889 return err; 9890 } 9891 9892 static void netif_free_rx_queues(struct net_device *dev) 9893 { 9894 unsigned int i, count = dev->num_rx_queues; 9895 9896 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9897 if (!dev->_rx) 9898 return; 9899 9900 for (i = 0; i < count; i++) 9901 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9902 9903 kvfree(dev->_rx); 9904 } 9905 9906 static void netdev_init_one_queue(struct net_device *dev, 9907 struct netdev_queue *queue, void *_unused) 9908 { 9909 /* Initialize queue lock */ 9910 spin_lock_init(&queue->_xmit_lock); 9911 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9912 queue->xmit_lock_owner = -1; 9913 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9914 queue->dev = dev; 9915 #ifdef CONFIG_BQL 9916 dql_init(&queue->dql, HZ); 9917 #endif 9918 } 9919 9920 static void netif_free_tx_queues(struct net_device *dev) 9921 { 9922 kvfree(dev->_tx); 9923 } 9924 9925 static int netif_alloc_netdev_queues(struct net_device *dev) 9926 { 9927 unsigned int count = dev->num_tx_queues; 9928 struct netdev_queue *tx; 9929 size_t sz = count * sizeof(*tx); 9930 9931 if (count < 1 || count > 0xffff) 9932 return -EINVAL; 9933 9934 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9935 if (!tx) 9936 return -ENOMEM; 9937 9938 dev->_tx = tx; 9939 9940 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9941 spin_lock_init(&dev->tx_global_lock); 9942 9943 return 0; 9944 } 9945 9946 void netif_tx_stop_all_queues(struct net_device *dev) 9947 { 9948 unsigned int i; 9949 9950 for (i = 0; i < dev->num_tx_queues; i++) { 9951 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9952 9953 netif_tx_stop_queue(txq); 9954 } 9955 } 9956 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9957 9958 /** 9959 * register_netdevice() - register a network device 9960 * @dev: device to register 9961 * 9962 * Take a prepared network device structure and make it externally accessible. 9963 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9964 * Callers must hold the rtnl lock - you may want register_netdev() 9965 * instead of this. 9966 */ 9967 int register_netdevice(struct net_device *dev) 9968 { 9969 int ret; 9970 struct net *net = dev_net(dev); 9971 9972 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9973 NETDEV_FEATURE_COUNT); 9974 BUG_ON(dev_boot_phase); 9975 ASSERT_RTNL(); 9976 9977 might_sleep(); 9978 9979 /* When net_device's are persistent, this will be fatal. */ 9980 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9981 BUG_ON(!net); 9982 9983 ret = ethtool_check_ops(dev->ethtool_ops); 9984 if (ret) 9985 return ret; 9986 9987 spin_lock_init(&dev->addr_list_lock); 9988 netdev_set_addr_lockdep_class(dev); 9989 9990 ret = dev_get_valid_name(net, dev, dev->name); 9991 if (ret < 0) 9992 goto out; 9993 9994 ret = -ENOMEM; 9995 dev->name_node = netdev_name_node_head_alloc(dev); 9996 if (!dev->name_node) 9997 goto out; 9998 9999 /* Init, if this function is available */ 10000 if (dev->netdev_ops->ndo_init) { 10001 ret = dev->netdev_ops->ndo_init(dev); 10002 if (ret) { 10003 if (ret > 0) 10004 ret = -EIO; 10005 goto err_free_name; 10006 } 10007 } 10008 10009 if (((dev->hw_features | dev->features) & 10010 NETIF_F_HW_VLAN_CTAG_FILTER) && 10011 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10012 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10013 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10014 ret = -EINVAL; 10015 goto err_uninit; 10016 } 10017 10018 ret = -EBUSY; 10019 if (!dev->ifindex) 10020 dev->ifindex = dev_new_index(net); 10021 else if (__dev_get_by_index(net, dev->ifindex)) 10022 goto err_uninit; 10023 10024 /* Transfer changeable features to wanted_features and enable 10025 * software offloads (GSO and GRO). 10026 */ 10027 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10028 dev->features |= NETIF_F_SOFT_FEATURES; 10029 10030 if (dev->udp_tunnel_nic_info) { 10031 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10032 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10033 } 10034 10035 dev->wanted_features = dev->features & dev->hw_features; 10036 10037 if (!(dev->flags & IFF_LOOPBACK)) 10038 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10039 10040 /* If IPv4 TCP segmentation offload is supported we should also 10041 * allow the device to enable segmenting the frame with the option 10042 * of ignoring a static IP ID value. This doesn't enable the 10043 * feature itself but allows the user to enable it later. 10044 */ 10045 if (dev->hw_features & NETIF_F_TSO) 10046 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10047 if (dev->vlan_features & NETIF_F_TSO) 10048 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10049 if (dev->mpls_features & NETIF_F_TSO) 10050 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10051 if (dev->hw_enc_features & NETIF_F_TSO) 10052 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10053 10054 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10055 */ 10056 dev->vlan_features |= NETIF_F_HIGHDMA; 10057 10058 /* Make NETIF_F_SG inheritable to tunnel devices. 10059 */ 10060 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10061 10062 /* Make NETIF_F_SG inheritable to MPLS. 10063 */ 10064 dev->mpls_features |= NETIF_F_SG; 10065 10066 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10067 ret = notifier_to_errno(ret); 10068 if (ret) 10069 goto err_uninit; 10070 10071 ret = netdev_register_kobject(dev); 10072 write_lock(&dev_base_lock); 10073 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10074 write_unlock(&dev_base_lock); 10075 if (ret) 10076 goto err_uninit_notify; 10077 10078 __netdev_update_features(dev); 10079 10080 /* 10081 * Default initial state at registry is that the 10082 * device is present. 10083 */ 10084 10085 set_bit(__LINK_STATE_PRESENT, &dev->state); 10086 10087 linkwatch_init_dev(dev); 10088 10089 dev_init_scheduler(dev); 10090 10091 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10092 list_netdevice(dev); 10093 10094 add_device_randomness(dev->dev_addr, dev->addr_len); 10095 10096 /* If the device has permanent device address, driver should 10097 * set dev_addr and also addr_assign_type should be set to 10098 * NET_ADDR_PERM (default value). 10099 */ 10100 if (dev->addr_assign_type == NET_ADDR_PERM) 10101 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10102 10103 /* Notify protocols, that a new device appeared. */ 10104 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10105 ret = notifier_to_errno(ret); 10106 if (ret) { 10107 /* Expect explicit free_netdev() on failure */ 10108 dev->needs_free_netdev = false; 10109 unregister_netdevice_queue(dev, NULL); 10110 goto out; 10111 } 10112 /* 10113 * Prevent userspace races by waiting until the network 10114 * device is fully setup before sending notifications. 10115 */ 10116 if (!dev->rtnl_link_ops || 10117 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10118 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10119 10120 out: 10121 return ret; 10122 10123 err_uninit_notify: 10124 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10125 err_uninit: 10126 if (dev->netdev_ops->ndo_uninit) 10127 dev->netdev_ops->ndo_uninit(dev); 10128 if (dev->priv_destructor) 10129 dev->priv_destructor(dev); 10130 err_free_name: 10131 netdev_name_node_free(dev->name_node); 10132 goto out; 10133 } 10134 EXPORT_SYMBOL(register_netdevice); 10135 10136 /** 10137 * init_dummy_netdev - init a dummy network device for NAPI 10138 * @dev: device to init 10139 * 10140 * This takes a network device structure and initialize the minimum 10141 * amount of fields so it can be used to schedule NAPI polls without 10142 * registering a full blown interface. This is to be used by drivers 10143 * that need to tie several hardware interfaces to a single NAPI 10144 * poll scheduler due to HW limitations. 10145 */ 10146 int init_dummy_netdev(struct net_device *dev) 10147 { 10148 /* Clear everything. Note we don't initialize spinlocks 10149 * are they aren't supposed to be taken by any of the 10150 * NAPI code and this dummy netdev is supposed to be 10151 * only ever used for NAPI polls 10152 */ 10153 memset(dev, 0, sizeof(struct net_device)); 10154 10155 /* make sure we BUG if trying to hit standard 10156 * register/unregister code path 10157 */ 10158 dev->reg_state = NETREG_DUMMY; 10159 10160 /* NAPI wants this */ 10161 INIT_LIST_HEAD(&dev->napi_list); 10162 10163 /* a dummy interface is started by default */ 10164 set_bit(__LINK_STATE_PRESENT, &dev->state); 10165 set_bit(__LINK_STATE_START, &dev->state); 10166 10167 /* napi_busy_loop stats accounting wants this */ 10168 dev_net_set(dev, &init_net); 10169 10170 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10171 * because users of this 'device' dont need to change 10172 * its refcount. 10173 */ 10174 10175 return 0; 10176 } 10177 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10178 10179 10180 /** 10181 * register_netdev - register a network device 10182 * @dev: device to register 10183 * 10184 * Take a completed network device structure and add it to the kernel 10185 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10186 * chain. 0 is returned on success. A negative errno code is returned 10187 * on a failure to set up the device, or if the name is a duplicate. 10188 * 10189 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10190 * and expands the device name if you passed a format string to 10191 * alloc_netdev. 10192 */ 10193 int register_netdev(struct net_device *dev) 10194 { 10195 int err; 10196 10197 if (rtnl_lock_killable()) 10198 return -EINTR; 10199 err = register_netdevice(dev); 10200 rtnl_unlock(); 10201 return err; 10202 } 10203 EXPORT_SYMBOL(register_netdev); 10204 10205 int netdev_refcnt_read(const struct net_device *dev) 10206 { 10207 #ifdef CONFIG_PCPU_DEV_REFCNT 10208 int i, refcnt = 0; 10209 10210 for_each_possible_cpu(i) 10211 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10212 return refcnt; 10213 #else 10214 return refcount_read(&dev->dev_refcnt); 10215 #endif 10216 } 10217 EXPORT_SYMBOL(netdev_refcnt_read); 10218 10219 int netdev_unregister_timeout_secs __read_mostly = 10; 10220 10221 #define WAIT_REFS_MIN_MSECS 1 10222 #define WAIT_REFS_MAX_MSECS 250 10223 /** 10224 * netdev_wait_allrefs_any - wait until all references are gone. 10225 * @list: list of net_devices to wait on 10226 * 10227 * This is called when unregistering network devices. 10228 * 10229 * Any protocol or device that holds a reference should register 10230 * for netdevice notification, and cleanup and put back the 10231 * reference if they receive an UNREGISTER event. 10232 * We can get stuck here if buggy protocols don't correctly 10233 * call dev_put. 10234 */ 10235 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10236 { 10237 unsigned long rebroadcast_time, warning_time; 10238 struct net_device *dev; 10239 int wait = 0; 10240 10241 rebroadcast_time = warning_time = jiffies; 10242 10243 list_for_each_entry(dev, list, todo_list) 10244 if (netdev_refcnt_read(dev) == 1) 10245 return dev; 10246 10247 while (true) { 10248 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10249 rtnl_lock(); 10250 10251 /* Rebroadcast unregister notification */ 10252 list_for_each_entry(dev, list, todo_list) 10253 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10254 10255 __rtnl_unlock(); 10256 rcu_barrier(); 10257 rtnl_lock(); 10258 10259 list_for_each_entry(dev, list, todo_list) 10260 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10261 &dev->state)) { 10262 /* We must not have linkwatch events 10263 * pending on unregister. If this 10264 * happens, we simply run the queue 10265 * unscheduled, resulting in a noop 10266 * for this device. 10267 */ 10268 linkwatch_run_queue(); 10269 break; 10270 } 10271 10272 __rtnl_unlock(); 10273 10274 rebroadcast_time = jiffies; 10275 } 10276 10277 if (!wait) { 10278 rcu_barrier(); 10279 wait = WAIT_REFS_MIN_MSECS; 10280 } else { 10281 msleep(wait); 10282 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10283 } 10284 10285 list_for_each_entry(dev, list, todo_list) 10286 if (netdev_refcnt_read(dev) == 1) 10287 return dev; 10288 10289 if (time_after(jiffies, warning_time + 10290 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10291 list_for_each_entry(dev, list, todo_list) { 10292 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10293 dev->name, netdev_refcnt_read(dev)); 10294 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10295 } 10296 10297 warning_time = jiffies; 10298 } 10299 } 10300 } 10301 10302 /* The sequence is: 10303 * 10304 * rtnl_lock(); 10305 * ... 10306 * register_netdevice(x1); 10307 * register_netdevice(x2); 10308 * ... 10309 * unregister_netdevice(y1); 10310 * unregister_netdevice(y2); 10311 * ... 10312 * rtnl_unlock(); 10313 * free_netdev(y1); 10314 * free_netdev(y2); 10315 * 10316 * We are invoked by rtnl_unlock(). 10317 * This allows us to deal with problems: 10318 * 1) We can delete sysfs objects which invoke hotplug 10319 * without deadlocking with linkwatch via keventd. 10320 * 2) Since we run with the RTNL semaphore not held, we can sleep 10321 * safely in order to wait for the netdev refcnt to drop to zero. 10322 * 10323 * We must not return until all unregister events added during 10324 * the interval the lock was held have been completed. 10325 */ 10326 void netdev_run_todo(void) 10327 { 10328 struct net_device *dev, *tmp; 10329 struct list_head list; 10330 #ifdef CONFIG_LOCKDEP 10331 struct list_head unlink_list; 10332 10333 list_replace_init(&net_unlink_list, &unlink_list); 10334 10335 while (!list_empty(&unlink_list)) { 10336 struct net_device *dev = list_first_entry(&unlink_list, 10337 struct net_device, 10338 unlink_list); 10339 list_del_init(&dev->unlink_list); 10340 dev->nested_level = dev->lower_level - 1; 10341 } 10342 #endif 10343 10344 /* Snapshot list, allow later requests */ 10345 list_replace_init(&net_todo_list, &list); 10346 10347 __rtnl_unlock(); 10348 10349 /* Wait for rcu callbacks to finish before next phase */ 10350 if (!list_empty(&list)) 10351 rcu_barrier(); 10352 10353 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10354 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10355 netdev_WARN(dev, "run_todo but not unregistering\n"); 10356 list_del(&dev->todo_list); 10357 continue; 10358 } 10359 10360 write_lock(&dev_base_lock); 10361 dev->reg_state = NETREG_UNREGISTERED; 10362 write_unlock(&dev_base_lock); 10363 linkwatch_forget_dev(dev); 10364 } 10365 10366 while (!list_empty(&list)) { 10367 dev = netdev_wait_allrefs_any(&list); 10368 list_del(&dev->todo_list); 10369 10370 /* paranoia */ 10371 BUG_ON(netdev_refcnt_read(dev) != 1); 10372 BUG_ON(!list_empty(&dev->ptype_all)); 10373 BUG_ON(!list_empty(&dev->ptype_specific)); 10374 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10375 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10376 10377 if (dev->priv_destructor) 10378 dev->priv_destructor(dev); 10379 if (dev->needs_free_netdev) 10380 free_netdev(dev); 10381 10382 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10383 wake_up(&netdev_unregistering_wq); 10384 10385 /* Free network device */ 10386 kobject_put(&dev->dev.kobj); 10387 } 10388 } 10389 10390 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10391 * all the same fields in the same order as net_device_stats, with only 10392 * the type differing, but rtnl_link_stats64 may have additional fields 10393 * at the end for newer counters. 10394 */ 10395 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10396 const struct net_device_stats *netdev_stats) 10397 { 10398 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10399 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10400 u64 *dst = (u64 *)stats64; 10401 10402 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10403 for (i = 0; i < n; i++) 10404 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10405 /* zero out counters that only exist in rtnl_link_stats64 */ 10406 memset((char *)stats64 + n * sizeof(u64), 0, 10407 sizeof(*stats64) - n * sizeof(u64)); 10408 } 10409 EXPORT_SYMBOL(netdev_stats_to_stats64); 10410 10411 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10412 { 10413 struct net_device_core_stats __percpu *p; 10414 10415 p = alloc_percpu_gfp(struct net_device_core_stats, 10416 GFP_ATOMIC | __GFP_NOWARN); 10417 10418 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10419 free_percpu(p); 10420 10421 /* This READ_ONCE() pairs with the cmpxchg() above */ 10422 return READ_ONCE(dev->core_stats); 10423 } 10424 EXPORT_SYMBOL(netdev_core_stats_alloc); 10425 10426 /** 10427 * dev_get_stats - get network device statistics 10428 * @dev: device to get statistics from 10429 * @storage: place to store stats 10430 * 10431 * Get network statistics from device. Return @storage. 10432 * The device driver may provide its own method by setting 10433 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10434 * otherwise the internal statistics structure is used. 10435 */ 10436 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10437 struct rtnl_link_stats64 *storage) 10438 { 10439 const struct net_device_ops *ops = dev->netdev_ops; 10440 const struct net_device_core_stats __percpu *p; 10441 10442 if (ops->ndo_get_stats64) { 10443 memset(storage, 0, sizeof(*storage)); 10444 ops->ndo_get_stats64(dev, storage); 10445 } else if (ops->ndo_get_stats) { 10446 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10447 } else { 10448 netdev_stats_to_stats64(storage, &dev->stats); 10449 } 10450 10451 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10452 p = READ_ONCE(dev->core_stats); 10453 if (p) { 10454 const struct net_device_core_stats *core_stats; 10455 int i; 10456 10457 for_each_possible_cpu(i) { 10458 core_stats = per_cpu_ptr(p, i); 10459 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10460 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10461 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10462 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10463 } 10464 } 10465 return storage; 10466 } 10467 EXPORT_SYMBOL(dev_get_stats); 10468 10469 /** 10470 * dev_fetch_sw_netstats - get per-cpu network device statistics 10471 * @s: place to store stats 10472 * @netstats: per-cpu network stats to read from 10473 * 10474 * Read per-cpu network statistics and populate the related fields in @s. 10475 */ 10476 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10477 const struct pcpu_sw_netstats __percpu *netstats) 10478 { 10479 int cpu; 10480 10481 for_each_possible_cpu(cpu) { 10482 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10483 const struct pcpu_sw_netstats *stats; 10484 unsigned int start; 10485 10486 stats = per_cpu_ptr(netstats, cpu); 10487 do { 10488 start = u64_stats_fetch_begin(&stats->syncp); 10489 rx_packets = u64_stats_read(&stats->rx_packets); 10490 rx_bytes = u64_stats_read(&stats->rx_bytes); 10491 tx_packets = u64_stats_read(&stats->tx_packets); 10492 tx_bytes = u64_stats_read(&stats->tx_bytes); 10493 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10494 10495 s->rx_packets += rx_packets; 10496 s->rx_bytes += rx_bytes; 10497 s->tx_packets += tx_packets; 10498 s->tx_bytes += tx_bytes; 10499 } 10500 } 10501 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10502 10503 /** 10504 * dev_get_tstats64 - ndo_get_stats64 implementation 10505 * @dev: device to get statistics from 10506 * @s: place to store stats 10507 * 10508 * Populate @s from dev->stats and dev->tstats. Can be used as 10509 * ndo_get_stats64() callback. 10510 */ 10511 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10512 { 10513 netdev_stats_to_stats64(s, &dev->stats); 10514 dev_fetch_sw_netstats(s, dev->tstats); 10515 } 10516 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10517 10518 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10519 { 10520 struct netdev_queue *queue = dev_ingress_queue(dev); 10521 10522 #ifdef CONFIG_NET_CLS_ACT 10523 if (queue) 10524 return queue; 10525 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10526 if (!queue) 10527 return NULL; 10528 netdev_init_one_queue(dev, queue, NULL); 10529 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10530 queue->qdisc_sleeping = &noop_qdisc; 10531 rcu_assign_pointer(dev->ingress_queue, queue); 10532 #endif 10533 return queue; 10534 } 10535 10536 static const struct ethtool_ops default_ethtool_ops; 10537 10538 void netdev_set_default_ethtool_ops(struct net_device *dev, 10539 const struct ethtool_ops *ops) 10540 { 10541 if (dev->ethtool_ops == &default_ethtool_ops) 10542 dev->ethtool_ops = ops; 10543 } 10544 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10545 10546 /** 10547 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10548 * @dev: netdev to enable the IRQ coalescing on 10549 * 10550 * Sets a conservative default for SW IRQ coalescing. Users can use 10551 * sysfs attributes to override the default values. 10552 */ 10553 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10554 { 10555 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10556 10557 dev->gro_flush_timeout = 20000; 10558 dev->napi_defer_hard_irqs = 1; 10559 } 10560 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10561 10562 void netdev_freemem(struct net_device *dev) 10563 { 10564 char *addr = (char *)dev - dev->padded; 10565 10566 kvfree(addr); 10567 } 10568 10569 /** 10570 * alloc_netdev_mqs - allocate network device 10571 * @sizeof_priv: size of private data to allocate space for 10572 * @name: device name format string 10573 * @name_assign_type: origin of device name 10574 * @setup: callback to initialize device 10575 * @txqs: the number of TX subqueues to allocate 10576 * @rxqs: the number of RX subqueues to allocate 10577 * 10578 * Allocates a struct net_device with private data area for driver use 10579 * and performs basic initialization. Also allocates subqueue structs 10580 * for each queue on the device. 10581 */ 10582 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10583 unsigned char name_assign_type, 10584 void (*setup)(struct net_device *), 10585 unsigned int txqs, unsigned int rxqs) 10586 { 10587 struct net_device *dev; 10588 unsigned int alloc_size; 10589 struct net_device *p; 10590 10591 BUG_ON(strlen(name) >= sizeof(dev->name)); 10592 10593 if (txqs < 1) { 10594 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10595 return NULL; 10596 } 10597 10598 if (rxqs < 1) { 10599 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10600 return NULL; 10601 } 10602 10603 alloc_size = sizeof(struct net_device); 10604 if (sizeof_priv) { 10605 /* ensure 32-byte alignment of private area */ 10606 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10607 alloc_size += sizeof_priv; 10608 } 10609 /* ensure 32-byte alignment of whole construct */ 10610 alloc_size += NETDEV_ALIGN - 1; 10611 10612 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10613 if (!p) 10614 return NULL; 10615 10616 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10617 dev->padded = (char *)dev - (char *)p; 10618 10619 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10620 #ifdef CONFIG_PCPU_DEV_REFCNT 10621 dev->pcpu_refcnt = alloc_percpu(int); 10622 if (!dev->pcpu_refcnt) 10623 goto free_dev; 10624 __dev_hold(dev); 10625 #else 10626 refcount_set(&dev->dev_refcnt, 1); 10627 #endif 10628 10629 if (dev_addr_init(dev)) 10630 goto free_pcpu; 10631 10632 dev_mc_init(dev); 10633 dev_uc_init(dev); 10634 10635 dev_net_set(dev, &init_net); 10636 10637 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10638 dev->gso_max_segs = GSO_MAX_SEGS; 10639 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10640 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10641 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10642 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10643 dev->tso_max_segs = TSO_MAX_SEGS; 10644 dev->upper_level = 1; 10645 dev->lower_level = 1; 10646 #ifdef CONFIG_LOCKDEP 10647 dev->nested_level = 0; 10648 INIT_LIST_HEAD(&dev->unlink_list); 10649 #endif 10650 10651 INIT_LIST_HEAD(&dev->napi_list); 10652 INIT_LIST_HEAD(&dev->unreg_list); 10653 INIT_LIST_HEAD(&dev->close_list); 10654 INIT_LIST_HEAD(&dev->link_watch_list); 10655 INIT_LIST_HEAD(&dev->adj_list.upper); 10656 INIT_LIST_HEAD(&dev->adj_list.lower); 10657 INIT_LIST_HEAD(&dev->ptype_all); 10658 INIT_LIST_HEAD(&dev->ptype_specific); 10659 INIT_LIST_HEAD(&dev->net_notifier_list); 10660 #ifdef CONFIG_NET_SCHED 10661 hash_init(dev->qdisc_hash); 10662 #endif 10663 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10664 setup(dev); 10665 10666 if (!dev->tx_queue_len) { 10667 dev->priv_flags |= IFF_NO_QUEUE; 10668 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10669 } 10670 10671 dev->num_tx_queues = txqs; 10672 dev->real_num_tx_queues = txqs; 10673 if (netif_alloc_netdev_queues(dev)) 10674 goto free_all; 10675 10676 dev->num_rx_queues = rxqs; 10677 dev->real_num_rx_queues = rxqs; 10678 if (netif_alloc_rx_queues(dev)) 10679 goto free_all; 10680 10681 strcpy(dev->name, name); 10682 dev->name_assign_type = name_assign_type; 10683 dev->group = INIT_NETDEV_GROUP; 10684 if (!dev->ethtool_ops) 10685 dev->ethtool_ops = &default_ethtool_ops; 10686 10687 nf_hook_netdev_init(dev); 10688 10689 return dev; 10690 10691 free_all: 10692 free_netdev(dev); 10693 return NULL; 10694 10695 free_pcpu: 10696 #ifdef CONFIG_PCPU_DEV_REFCNT 10697 free_percpu(dev->pcpu_refcnt); 10698 free_dev: 10699 #endif 10700 netdev_freemem(dev); 10701 return NULL; 10702 } 10703 EXPORT_SYMBOL(alloc_netdev_mqs); 10704 10705 /** 10706 * free_netdev - free network device 10707 * @dev: device 10708 * 10709 * This function does the last stage of destroying an allocated device 10710 * interface. The reference to the device object is released. If this 10711 * is the last reference then it will be freed.Must be called in process 10712 * context. 10713 */ 10714 void free_netdev(struct net_device *dev) 10715 { 10716 struct napi_struct *p, *n; 10717 10718 might_sleep(); 10719 10720 /* When called immediately after register_netdevice() failed the unwind 10721 * handling may still be dismantling the device. Handle that case by 10722 * deferring the free. 10723 */ 10724 if (dev->reg_state == NETREG_UNREGISTERING) { 10725 ASSERT_RTNL(); 10726 dev->needs_free_netdev = true; 10727 return; 10728 } 10729 10730 netif_free_tx_queues(dev); 10731 netif_free_rx_queues(dev); 10732 10733 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10734 10735 /* Flush device addresses */ 10736 dev_addr_flush(dev); 10737 10738 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10739 netif_napi_del(p); 10740 10741 ref_tracker_dir_exit(&dev->refcnt_tracker); 10742 #ifdef CONFIG_PCPU_DEV_REFCNT 10743 free_percpu(dev->pcpu_refcnt); 10744 dev->pcpu_refcnt = NULL; 10745 #endif 10746 free_percpu(dev->core_stats); 10747 dev->core_stats = NULL; 10748 free_percpu(dev->xdp_bulkq); 10749 dev->xdp_bulkq = NULL; 10750 10751 /* Compatibility with error handling in drivers */ 10752 if (dev->reg_state == NETREG_UNINITIALIZED) { 10753 netdev_freemem(dev); 10754 return; 10755 } 10756 10757 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10758 dev->reg_state = NETREG_RELEASED; 10759 10760 /* will free via device release */ 10761 put_device(&dev->dev); 10762 } 10763 EXPORT_SYMBOL(free_netdev); 10764 10765 /** 10766 * synchronize_net - Synchronize with packet receive processing 10767 * 10768 * Wait for packets currently being received to be done. 10769 * Does not block later packets from starting. 10770 */ 10771 void synchronize_net(void) 10772 { 10773 might_sleep(); 10774 if (rtnl_is_locked()) 10775 synchronize_rcu_expedited(); 10776 else 10777 synchronize_rcu(); 10778 } 10779 EXPORT_SYMBOL(synchronize_net); 10780 10781 /** 10782 * unregister_netdevice_queue - remove device from the kernel 10783 * @dev: device 10784 * @head: list 10785 * 10786 * This function shuts down a device interface and removes it 10787 * from the kernel tables. 10788 * If head not NULL, device is queued to be unregistered later. 10789 * 10790 * Callers must hold the rtnl semaphore. You may want 10791 * unregister_netdev() instead of this. 10792 */ 10793 10794 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10795 { 10796 ASSERT_RTNL(); 10797 10798 if (head) { 10799 list_move_tail(&dev->unreg_list, head); 10800 } else { 10801 LIST_HEAD(single); 10802 10803 list_add(&dev->unreg_list, &single); 10804 unregister_netdevice_many(&single); 10805 } 10806 } 10807 EXPORT_SYMBOL(unregister_netdevice_queue); 10808 10809 void unregister_netdevice_many_notify(struct list_head *head, 10810 u32 portid, const struct nlmsghdr *nlh) 10811 { 10812 struct net_device *dev, *tmp; 10813 LIST_HEAD(close_head); 10814 10815 BUG_ON(dev_boot_phase); 10816 ASSERT_RTNL(); 10817 10818 if (list_empty(head)) 10819 return; 10820 10821 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10822 /* Some devices call without registering 10823 * for initialization unwind. Remove those 10824 * devices and proceed with the remaining. 10825 */ 10826 if (dev->reg_state == NETREG_UNINITIALIZED) { 10827 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10828 dev->name, dev); 10829 10830 WARN_ON(1); 10831 list_del(&dev->unreg_list); 10832 continue; 10833 } 10834 dev->dismantle = true; 10835 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10836 } 10837 10838 /* If device is running, close it first. */ 10839 list_for_each_entry(dev, head, unreg_list) 10840 list_add_tail(&dev->close_list, &close_head); 10841 dev_close_many(&close_head, true); 10842 10843 list_for_each_entry(dev, head, unreg_list) { 10844 /* And unlink it from device chain. */ 10845 write_lock(&dev_base_lock); 10846 unlist_netdevice(dev, false); 10847 dev->reg_state = NETREG_UNREGISTERING; 10848 write_unlock(&dev_base_lock); 10849 } 10850 flush_all_backlogs(); 10851 10852 synchronize_net(); 10853 10854 list_for_each_entry(dev, head, unreg_list) { 10855 struct sk_buff *skb = NULL; 10856 10857 /* Shutdown queueing discipline. */ 10858 dev_shutdown(dev); 10859 10860 dev_xdp_uninstall(dev); 10861 bpf_dev_bound_netdev_unregister(dev); 10862 10863 netdev_offload_xstats_disable_all(dev); 10864 10865 /* Notify protocols, that we are about to destroy 10866 * this device. They should clean all the things. 10867 */ 10868 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10869 10870 if (!dev->rtnl_link_ops || 10871 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10872 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10873 GFP_KERNEL, NULL, 0, 10874 portid, nlh); 10875 10876 /* 10877 * Flush the unicast and multicast chains 10878 */ 10879 dev_uc_flush(dev); 10880 dev_mc_flush(dev); 10881 10882 netdev_name_node_alt_flush(dev); 10883 netdev_name_node_free(dev->name_node); 10884 10885 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10886 10887 if (dev->netdev_ops->ndo_uninit) 10888 dev->netdev_ops->ndo_uninit(dev); 10889 10890 if (skb) 10891 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10892 10893 /* Notifier chain MUST detach us all upper devices. */ 10894 WARN_ON(netdev_has_any_upper_dev(dev)); 10895 WARN_ON(netdev_has_any_lower_dev(dev)); 10896 10897 /* Remove entries from kobject tree */ 10898 netdev_unregister_kobject(dev); 10899 #ifdef CONFIG_XPS 10900 /* Remove XPS queueing entries */ 10901 netif_reset_xps_queues_gt(dev, 0); 10902 #endif 10903 } 10904 10905 synchronize_net(); 10906 10907 list_for_each_entry(dev, head, unreg_list) { 10908 netdev_put(dev, &dev->dev_registered_tracker); 10909 net_set_todo(dev); 10910 } 10911 10912 list_del(head); 10913 } 10914 10915 /** 10916 * unregister_netdevice_many - unregister many devices 10917 * @head: list of devices 10918 * 10919 * Note: As most callers use a stack allocated list_head, 10920 * we force a list_del() to make sure stack wont be corrupted later. 10921 */ 10922 void unregister_netdevice_many(struct list_head *head) 10923 { 10924 unregister_netdevice_many_notify(head, 0, NULL); 10925 } 10926 EXPORT_SYMBOL(unregister_netdevice_many); 10927 10928 /** 10929 * unregister_netdev - remove device from the kernel 10930 * @dev: device 10931 * 10932 * This function shuts down a device interface and removes it 10933 * from the kernel tables. 10934 * 10935 * This is just a wrapper for unregister_netdevice that takes 10936 * the rtnl semaphore. In general you want to use this and not 10937 * unregister_netdevice. 10938 */ 10939 void unregister_netdev(struct net_device *dev) 10940 { 10941 rtnl_lock(); 10942 unregister_netdevice(dev); 10943 rtnl_unlock(); 10944 } 10945 EXPORT_SYMBOL(unregister_netdev); 10946 10947 /** 10948 * __dev_change_net_namespace - move device to different nethost namespace 10949 * @dev: device 10950 * @net: network namespace 10951 * @pat: If not NULL name pattern to try if the current device name 10952 * is already taken in the destination network namespace. 10953 * @new_ifindex: If not zero, specifies device index in the target 10954 * namespace. 10955 * 10956 * This function shuts down a device interface and moves it 10957 * to a new network namespace. On success 0 is returned, on 10958 * a failure a netagive errno code is returned. 10959 * 10960 * Callers must hold the rtnl semaphore. 10961 */ 10962 10963 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10964 const char *pat, int new_ifindex) 10965 { 10966 struct net *net_old = dev_net(dev); 10967 int err, new_nsid; 10968 10969 ASSERT_RTNL(); 10970 10971 /* Don't allow namespace local devices to be moved. */ 10972 err = -EINVAL; 10973 if (dev->features & NETIF_F_NETNS_LOCAL) 10974 goto out; 10975 10976 /* Ensure the device has been registrered */ 10977 if (dev->reg_state != NETREG_REGISTERED) 10978 goto out; 10979 10980 /* Get out if there is nothing todo */ 10981 err = 0; 10982 if (net_eq(net_old, net)) 10983 goto out; 10984 10985 /* Pick the destination device name, and ensure 10986 * we can use it in the destination network namespace. 10987 */ 10988 err = -EEXIST; 10989 if (netdev_name_in_use(net, dev->name)) { 10990 /* We get here if we can't use the current device name */ 10991 if (!pat) 10992 goto out; 10993 err = dev_get_valid_name(net, dev, pat); 10994 if (err < 0) 10995 goto out; 10996 } 10997 10998 /* Check that new_ifindex isn't used yet. */ 10999 err = -EBUSY; 11000 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11001 goto out; 11002 11003 /* 11004 * And now a mini version of register_netdevice unregister_netdevice. 11005 */ 11006 11007 /* If device is running close it first. */ 11008 dev_close(dev); 11009 11010 /* And unlink it from device chain */ 11011 unlist_netdevice(dev, true); 11012 11013 synchronize_net(); 11014 11015 /* Shutdown queueing discipline. */ 11016 dev_shutdown(dev); 11017 11018 /* Notify protocols, that we are about to destroy 11019 * this device. They should clean all the things. 11020 * 11021 * Note that dev->reg_state stays at NETREG_REGISTERED. 11022 * This is wanted because this way 8021q and macvlan know 11023 * the device is just moving and can keep their slaves up. 11024 */ 11025 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11026 rcu_barrier(); 11027 11028 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11029 /* If there is an ifindex conflict assign a new one */ 11030 if (!new_ifindex) { 11031 if (__dev_get_by_index(net, dev->ifindex)) 11032 new_ifindex = dev_new_index(net); 11033 else 11034 new_ifindex = dev->ifindex; 11035 } 11036 11037 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11038 new_ifindex); 11039 11040 /* 11041 * Flush the unicast and multicast chains 11042 */ 11043 dev_uc_flush(dev); 11044 dev_mc_flush(dev); 11045 11046 /* Send a netdev-removed uevent to the old namespace */ 11047 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11048 netdev_adjacent_del_links(dev); 11049 11050 /* Move per-net netdevice notifiers that are following the netdevice */ 11051 move_netdevice_notifiers_dev_net(dev, net); 11052 11053 /* Actually switch the network namespace */ 11054 dev_net_set(dev, net); 11055 dev->ifindex = new_ifindex; 11056 11057 /* Send a netdev-add uevent to the new namespace */ 11058 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11059 netdev_adjacent_add_links(dev); 11060 11061 /* Fixup kobjects */ 11062 err = device_rename(&dev->dev, dev->name); 11063 WARN_ON(err); 11064 11065 /* Adapt owner in case owning user namespace of target network 11066 * namespace is different from the original one. 11067 */ 11068 err = netdev_change_owner(dev, net_old, net); 11069 WARN_ON(err); 11070 11071 /* Add the device back in the hashes */ 11072 list_netdevice(dev); 11073 11074 /* Notify protocols, that a new device appeared. */ 11075 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11076 11077 /* 11078 * Prevent userspace races by waiting until the network 11079 * device is fully setup before sending notifications. 11080 */ 11081 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11082 11083 synchronize_net(); 11084 err = 0; 11085 out: 11086 return err; 11087 } 11088 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11089 11090 static int dev_cpu_dead(unsigned int oldcpu) 11091 { 11092 struct sk_buff **list_skb; 11093 struct sk_buff *skb; 11094 unsigned int cpu; 11095 struct softnet_data *sd, *oldsd, *remsd = NULL; 11096 11097 local_irq_disable(); 11098 cpu = smp_processor_id(); 11099 sd = &per_cpu(softnet_data, cpu); 11100 oldsd = &per_cpu(softnet_data, oldcpu); 11101 11102 /* Find end of our completion_queue. */ 11103 list_skb = &sd->completion_queue; 11104 while (*list_skb) 11105 list_skb = &(*list_skb)->next; 11106 /* Append completion queue from offline CPU. */ 11107 *list_skb = oldsd->completion_queue; 11108 oldsd->completion_queue = NULL; 11109 11110 /* Append output queue from offline CPU. */ 11111 if (oldsd->output_queue) { 11112 *sd->output_queue_tailp = oldsd->output_queue; 11113 sd->output_queue_tailp = oldsd->output_queue_tailp; 11114 oldsd->output_queue = NULL; 11115 oldsd->output_queue_tailp = &oldsd->output_queue; 11116 } 11117 /* Append NAPI poll list from offline CPU, with one exception : 11118 * process_backlog() must be called by cpu owning percpu backlog. 11119 * We properly handle process_queue & input_pkt_queue later. 11120 */ 11121 while (!list_empty(&oldsd->poll_list)) { 11122 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11123 struct napi_struct, 11124 poll_list); 11125 11126 list_del_init(&napi->poll_list); 11127 if (napi->poll == process_backlog) 11128 napi->state = 0; 11129 else 11130 ____napi_schedule(sd, napi); 11131 } 11132 11133 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11134 local_irq_enable(); 11135 11136 #ifdef CONFIG_RPS 11137 remsd = oldsd->rps_ipi_list; 11138 oldsd->rps_ipi_list = NULL; 11139 #endif 11140 /* send out pending IPI's on offline CPU */ 11141 net_rps_send_ipi(remsd); 11142 11143 /* Process offline CPU's input_pkt_queue */ 11144 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11145 netif_rx(skb); 11146 input_queue_head_incr(oldsd); 11147 } 11148 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11149 netif_rx(skb); 11150 input_queue_head_incr(oldsd); 11151 } 11152 11153 return 0; 11154 } 11155 11156 /** 11157 * netdev_increment_features - increment feature set by one 11158 * @all: current feature set 11159 * @one: new feature set 11160 * @mask: mask feature set 11161 * 11162 * Computes a new feature set after adding a device with feature set 11163 * @one to the master device with current feature set @all. Will not 11164 * enable anything that is off in @mask. Returns the new feature set. 11165 */ 11166 netdev_features_t netdev_increment_features(netdev_features_t all, 11167 netdev_features_t one, netdev_features_t mask) 11168 { 11169 if (mask & NETIF_F_HW_CSUM) 11170 mask |= NETIF_F_CSUM_MASK; 11171 mask |= NETIF_F_VLAN_CHALLENGED; 11172 11173 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11174 all &= one | ~NETIF_F_ALL_FOR_ALL; 11175 11176 /* If one device supports hw checksumming, set for all. */ 11177 if (all & NETIF_F_HW_CSUM) 11178 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11179 11180 return all; 11181 } 11182 EXPORT_SYMBOL(netdev_increment_features); 11183 11184 static struct hlist_head * __net_init netdev_create_hash(void) 11185 { 11186 int i; 11187 struct hlist_head *hash; 11188 11189 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11190 if (hash != NULL) 11191 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11192 INIT_HLIST_HEAD(&hash[i]); 11193 11194 return hash; 11195 } 11196 11197 /* Initialize per network namespace state */ 11198 static int __net_init netdev_init(struct net *net) 11199 { 11200 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11201 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11202 11203 INIT_LIST_HEAD(&net->dev_base_head); 11204 11205 net->dev_name_head = netdev_create_hash(); 11206 if (net->dev_name_head == NULL) 11207 goto err_name; 11208 11209 net->dev_index_head = netdev_create_hash(); 11210 if (net->dev_index_head == NULL) 11211 goto err_idx; 11212 11213 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11214 11215 return 0; 11216 11217 err_idx: 11218 kfree(net->dev_name_head); 11219 err_name: 11220 return -ENOMEM; 11221 } 11222 11223 /** 11224 * netdev_drivername - network driver for the device 11225 * @dev: network device 11226 * 11227 * Determine network driver for device. 11228 */ 11229 const char *netdev_drivername(const struct net_device *dev) 11230 { 11231 const struct device_driver *driver; 11232 const struct device *parent; 11233 const char *empty = ""; 11234 11235 parent = dev->dev.parent; 11236 if (!parent) 11237 return empty; 11238 11239 driver = parent->driver; 11240 if (driver && driver->name) 11241 return driver->name; 11242 return empty; 11243 } 11244 11245 static void __netdev_printk(const char *level, const struct net_device *dev, 11246 struct va_format *vaf) 11247 { 11248 if (dev && dev->dev.parent) { 11249 dev_printk_emit(level[1] - '0', 11250 dev->dev.parent, 11251 "%s %s %s%s: %pV", 11252 dev_driver_string(dev->dev.parent), 11253 dev_name(dev->dev.parent), 11254 netdev_name(dev), netdev_reg_state(dev), 11255 vaf); 11256 } else if (dev) { 11257 printk("%s%s%s: %pV", 11258 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11259 } else { 11260 printk("%s(NULL net_device): %pV", level, vaf); 11261 } 11262 } 11263 11264 void netdev_printk(const char *level, const struct net_device *dev, 11265 const char *format, ...) 11266 { 11267 struct va_format vaf; 11268 va_list args; 11269 11270 va_start(args, format); 11271 11272 vaf.fmt = format; 11273 vaf.va = &args; 11274 11275 __netdev_printk(level, dev, &vaf); 11276 11277 va_end(args); 11278 } 11279 EXPORT_SYMBOL(netdev_printk); 11280 11281 #define define_netdev_printk_level(func, level) \ 11282 void func(const struct net_device *dev, const char *fmt, ...) \ 11283 { \ 11284 struct va_format vaf; \ 11285 va_list args; \ 11286 \ 11287 va_start(args, fmt); \ 11288 \ 11289 vaf.fmt = fmt; \ 11290 vaf.va = &args; \ 11291 \ 11292 __netdev_printk(level, dev, &vaf); \ 11293 \ 11294 va_end(args); \ 11295 } \ 11296 EXPORT_SYMBOL(func); 11297 11298 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11299 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11300 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11301 define_netdev_printk_level(netdev_err, KERN_ERR); 11302 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11303 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11304 define_netdev_printk_level(netdev_info, KERN_INFO); 11305 11306 static void __net_exit netdev_exit(struct net *net) 11307 { 11308 kfree(net->dev_name_head); 11309 kfree(net->dev_index_head); 11310 if (net != &init_net) 11311 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11312 } 11313 11314 static struct pernet_operations __net_initdata netdev_net_ops = { 11315 .init = netdev_init, 11316 .exit = netdev_exit, 11317 }; 11318 11319 static void __net_exit default_device_exit_net(struct net *net) 11320 { 11321 struct net_device *dev, *aux; 11322 /* 11323 * Push all migratable network devices back to the 11324 * initial network namespace 11325 */ 11326 ASSERT_RTNL(); 11327 for_each_netdev_safe(net, dev, aux) { 11328 int err; 11329 char fb_name[IFNAMSIZ]; 11330 11331 /* Ignore unmoveable devices (i.e. loopback) */ 11332 if (dev->features & NETIF_F_NETNS_LOCAL) 11333 continue; 11334 11335 /* Leave virtual devices for the generic cleanup */ 11336 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11337 continue; 11338 11339 /* Push remaining network devices to init_net */ 11340 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11341 if (netdev_name_in_use(&init_net, fb_name)) 11342 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11343 err = dev_change_net_namespace(dev, &init_net, fb_name); 11344 if (err) { 11345 pr_emerg("%s: failed to move %s to init_net: %d\n", 11346 __func__, dev->name, err); 11347 BUG(); 11348 } 11349 } 11350 } 11351 11352 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11353 { 11354 /* At exit all network devices most be removed from a network 11355 * namespace. Do this in the reverse order of registration. 11356 * Do this across as many network namespaces as possible to 11357 * improve batching efficiency. 11358 */ 11359 struct net_device *dev; 11360 struct net *net; 11361 LIST_HEAD(dev_kill_list); 11362 11363 rtnl_lock(); 11364 list_for_each_entry(net, net_list, exit_list) { 11365 default_device_exit_net(net); 11366 cond_resched(); 11367 } 11368 11369 list_for_each_entry(net, net_list, exit_list) { 11370 for_each_netdev_reverse(net, dev) { 11371 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11372 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11373 else 11374 unregister_netdevice_queue(dev, &dev_kill_list); 11375 } 11376 } 11377 unregister_netdevice_many(&dev_kill_list); 11378 rtnl_unlock(); 11379 } 11380 11381 static struct pernet_operations __net_initdata default_device_ops = { 11382 .exit_batch = default_device_exit_batch, 11383 }; 11384 11385 /* 11386 * Initialize the DEV module. At boot time this walks the device list and 11387 * unhooks any devices that fail to initialise (normally hardware not 11388 * present) and leaves us with a valid list of present and active devices. 11389 * 11390 */ 11391 11392 /* 11393 * This is called single threaded during boot, so no need 11394 * to take the rtnl semaphore. 11395 */ 11396 static int __init net_dev_init(void) 11397 { 11398 int i, rc = -ENOMEM; 11399 11400 BUG_ON(!dev_boot_phase); 11401 11402 if (dev_proc_init()) 11403 goto out; 11404 11405 if (netdev_kobject_init()) 11406 goto out; 11407 11408 INIT_LIST_HEAD(&ptype_all); 11409 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11410 INIT_LIST_HEAD(&ptype_base[i]); 11411 11412 if (register_pernet_subsys(&netdev_net_ops)) 11413 goto out; 11414 11415 /* 11416 * Initialise the packet receive queues. 11417 */ 11418 11419 for_each_possible_cpu(i) { 11420 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11421 struct softnet_data *sd = &per_cpu(softnet_data, i); 11422 11423 INIT_WORK(flush, flush_backlog); 11424 11425 skb_queue_head_init(&sd->input_pkt_queue); 11426 skb_queue_head_init(&sd->process_queue); 11427 #ifdef CONFIG_XFRM_OFFLOAD 11428 skb_queue_head_init(&sd->xfrm_backlog); 11429 #endif 11430 INIT_LIST_HEAD(&sd->poll_list); 11431 sd->output_queue_tailp = &sd->output_queue; 11432 #ifdef CONFIG_RPS 11433 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11434 sd->cpu = i; 11435 #endif 11436 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11437 spin_lock_init(&sd->defer_lock); 11438 11439 init_gro_hash(&sd->backlog); 11440 sd->backlog.poll = process_backlog; 11441 sd->backlog.weight = weight_p; 11442 } 11443 11444 dev_boot_phase = 0; 11445 11446 /* The loopback device is special if any other network devices 11447 * is present in a network namespace the loopback device must 11448 * be present. Since we now dynamically allocate and free the 11449 * loopback device ensure this invariant is maintained by 11450 * keeping the loopback device as the first device on the 11451 * list of network devices. Ensuring the loopback devices 11452 * is the first device that appears and the last network device 11453 * that disappears. 11454 */ 11455 if (register_pernet_device(&loopback_net_ops)) 11456 goto out; 11457 11458 if (register_pernet_device(&default_device_ops)) 11459 goto out; 11460 11461 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11462 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11463 11464 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11465 NULL, dev_cpu_dead); 11466 WARN_ON(rc < 0); 11467 rc = 0; 11468 out: 11469 return rc; 11470 } 11471 11472 subsys_initcall(net_dev_init); 11473