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 /* Deprecated for new users, call netdev_get_by_name() instead */ 762 struct net_device *dev_get_by_name(struct net *net, const char *name) 763 { 764 struct net_device *dev; 765 766 rcu_read_lock(); 767 dev = dev_get_by_name_rcu(net, name); 768 dev_hold(dev); 769 rcu_read_unlock(); 770 return dev; 771 } 772 EXPORT_SYMBOL(dev_get_by_name); 773 774 /** 775 * netdev_get_by_name() - find a device by its name 776 * @net: the applicable net namespace 777 * @name: name to find 778 * @tracker: tracking object for the acquired reference 779 * @gfp: allocation flags for the tracker 780 * 781 * Find an interface by name. This can be called from any 782 * context and does its own locking. The returned handle has 783 * the usage count incremented and the caller must use netdev_put() to 784 * release it when it is no longer needed. %NULL is returned if no 785 * matching device is found. 786 */ 787 struct net_device *netdev_get_by_name(struct net *net, const char *name, 788 netdevice_tracker *tracker, gfp_t gfp) 789 { 790 struct net_device *dev; 791 792 dev = dev_get_by_name(net, name); 793 if (dev) 794 netdev_tracker_alloc(dev, tracker, gfp); 795 return dev; 796 } 797 EXPORT_SYMBOL(netdev_get_by_name); 798 799 /** 800 * __dev_get_by_index - find a device by its ifindex 801 * @net: the applicable net namespace 802 * @ifindex: index of device 803 * 804 * Search for an interface by index. Returns %NULL if the device 805 * is not found or a pointer to the device. The device has not 806 * had its reference counter increased so the caller must be careful 807 * about locking. The caller must hold either the RTNL semaphore 808 * or @dev_base_lock. 809 */ 810 811 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 812 { 813 struct net_device *dev; 814 struct hlist_head *head = dev_index_hash(net, ifindex); 815 816 hlist_for_each_entry(dev, head, index_hlist) 817 if (dev->ifindex == ifindex) 818 return dev; 819 820 return NULL; 821 } 822 EXPORT_SYMBOL(__dev_get_by_index); 823 824 /** 825 * dev_get_by_index_rcu - find a device by its ifindex 826 * @net: the applicable net namespace 827 * @ifindex: index of device 828 * 829 * Search for an interface by index. Returns %NULL if the device 830 * is not found or a pointer to the device. The device has not 831 * had its reference counter increased so the caller must be careful 832 * about locking. The caller must hold RCU lock. 833 */ 834 835 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 836 { 837 struct net_device *dev; 838 struct hlist_head *head = dev_index_hash(net, ifindex); 839 840 hlist_for_each_entry_rcu(dev, head, index_hlist) 841 if (dev->ifindex == ifindex) 842 return dev; 843 844 return NULL; 845 } 846 EXPORT_SYMBOL(dev_get_by_index_rcu); 847 848 /* Deprecated for new users, call netdev_get_by_index() instead */ 849 struct net_device *dev_get_by_index(struct net *net, int ifindex) 850 { 851 struct net_device *dev; 852 853 rcu_read_lock(); 854 dev = dev_get_by_index_rcu(net, ifindex); 855 dev_hold(dev); 856 rcu_read_unlock(); 857 return dev; 858 } 859 EXPORT_SYMBOL(dev_get_by_index); 860 861 /** 862 * netdev_get_by_index() - find a device by its ifindex 863 * @net: the applicable net namespace 864 * @ifindex: index of device 865 * @tracker: tracking object for the acquired reference 866 * @gfp: allocation flags for the tracker 867 * 868 * Search for an interface by index. Returns NULL if the device 869 * is not found or a pointer to the device. The device returned has 870 * had a reference added and the pointer is safe until the user calls 871 * netdev_put() to indicate they have finished with it. 872 */ 873 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 874 netdevice_tracker *tracker, gfp_t gfp) 875 { 876 struct net_device *dev; 877 878 dev = dev_get_by_index(net, ifindex); 879 if (dev) 880 netdev_tracker_alloc(dev, tracker, gfp); 881 return dev; 882 } 883 EXPORT_SYMBOL(netdev_get_by_index); 884 885 /** 886 * dev_get_by_napi_id - find a device by napi_id 887 * @napi_id: ID of the NAPI struct 888 * 889 * Search for an interface by NAPI ID. Returns %NULL if the device 890 * is not found or a pointer to the device. The device has not had 891 * its reference counter increased so the caller must be careful 892 * about locking. The caller must hold RCU lock. 893 */ 894 895 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 896 { 897 struct napi_struct *napi; 898 899 WARN_ON_ONCE(!rcu_read_lock_held()); 900 901 if (napi_id < MIN_NAPI_ID) 902 return NULL; 903 904 napi = napi_by_id(napi_id); 905 906 return napi ? napi->dev : NULL; 907 } 908 EXPORT_SYMBOL(dev_get_by_napi_id); 909 910 /** 911 * netdev_get_name - get a netdevice name, knowing its ifindex. 912 * @net: network namespace 913 * @name: a pointer to the buffer where the name will be stored. 914 * @ifindex: the ifindex of the interface to get the name from. 915 */ 916 int netdev_get_name(struct net *net, char *name, int ifindex) 917 { 918 struct net_device *dev; 919 int ret; 920 921 down_read(&devnet_rename_sem); 922 rcu_read_lock(); 923 924 dev = dev_get_by_index_rcu(net, ifindex); 925 if (!dev) { 926 ret = -ENODEV; 927 goto out; 928 } 929 930 strcpy(name, dev->name); 931 932 ret = 0; 933 out: 934 rcu_read_unlock(); 935 up_read(&devnet_rename_sem); 936 return ret; 937 } 938 939 /** 940 * dev_getbyhwaddr_rcu - find a device by its hardware address 941 * @net: the applicable net namespace 942 * @type: media type of device 943 * @ha: hardware address 944 * 945 * Search for an interface by MAC address. Returns NULL if the device 946 * is not found or a pointer to the device. 947 * The caller must hold RCU or RTNL. 948 * The returned device has not had its ref count increased 949 * and the caller must therefore be careful about locking 950 * 951 */ 952 953 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 954 const char *ha) 955 { 956 struct net_device *dev; 957 958 for_each_netdev_rcu(net, dev) 959 if (dev->type == type && 960 !memcmp(dev->dev_addr, ha, dev->addr_len)) 961 return dev; 962 963 return NULL; 964 } 965 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 966 967 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 968 { 969 struct net_device *dev, *ret = NULL; 970 971 rcu_read_lock(); 972 for_each_netdev_rcu(net, dev) 973 if (dev->type == type) { 974 dev_hold(dev); 975 ret = dev; 976 break; 977 } 978 rcu_read_unlock(); 979 return ret; 980 } 981 EXPORT_SYMBOL(dev_getfirstbyhwtype); 982 983 /** 984 * __dev_get_by_flags - find any device with given flags 985 * @net: the applicable net namespace 986 * @if_flags: IFF_* values 987 * @mask: bitmask of bits in if_flags to check 988 * 989 * Search for any interface with the given flags. Returns NULL if a device 990 * is not found or a pointer to the device. Must be called inside 991 * rtnl_lock(), and result refcount is unchanged. 992 */ 993 994 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 995 unsigned short mask) 996 { 997 struct net_device *dev, *ret; 998 999 ASSERT_RTNL(); 1000 1001 ret = NULL; 1002 for_each_netdev(net, dev) { 1003 if (((dev->flags ^ if_flags) & mask) == 0) { 1004 ret = dev; 1005 break; 1006 } 1007 } 1008 return ret; 1009 } 1010 EXPORT_SYMBOL(__dev_get_by_flags); 1011 1012 /** 1013 * dev_valid_name - check if name is okay for network device 1014 * @name: name string 1015 * 1016 * Network device names need to be valid file names to 1017 * allow sysfs to work. We also disallow any kind of 1018 * whitespace. 1019 */ 1020 bool dev_valid_name(const char *name) 1021 { 1022 if (*name == '\0') 1023 return false; 1024 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1025 return false; 1026 if (!strcmp(name, ".") || !strcmp(name, "..")) 1027 return false; 1028 1029 while (*name) { 1030 if (*name == '/' || *name == ':' || isspace(*name)) 1031 return false; 1032 name++; 1033 } 1034 return true; 1035 } 1036 EXPORT_SYMBOL(dev_valid_name); 1037 1038 /** 1039 * __dev_alloc_name - allocate a name for a device 1040 * @net: network namespace to allocate the device name in 1041 * @name: name format string 1042 * @buf: scratch buffer and result name string 1043 * 1044 * Passed a format string - eg "lt%d" it will try and find a suitable 1045 * id. It scans list of devices to build up a free map, then chooses 1046 * the first empty slot. The caller must hold the dev_base or rtnl lock 1047 * while allocating the name and adding the device in order to avoid 1048 * duplicates. 1049 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1050 * Returns the number of the unit assigned or a negative errno code. 1051 */ 1052 1053 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1054 { 1055 int i = 0; 1056 const char *p; 1057 const int max_netdevices = 8*PAGE_SIZE; 1058 unsigned long *inuse; 1059 struct net_device *d; 1060 1061 if (!dev_valid_name(name)) 1062 return -EINVAL; 1063 1064 p = strchr(name, '%'); 1065 if (p) { 1066 /* 1067 * Verify the string as this thing may have come from 1068 * the user. There must be either one "%d" and no other "%" 1069 * characters. 1070 */ 1071 if (p[1] != 'd' || strchr(p + 2, '%')) 1072 return -EINVAL; 1073 1074 /* Use one page as a bit array of possible slots */ 1075 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1076 if (!inuse) 1077 return -ENOMEM; 1078 1079 for_each_netdev(net, d) { 1080 struct netdev_name_node *name_node; 1081 list_for_each_entry(name_node, &d->name_node->list, list) { 1082 if (!sscanf(name_node->name, name, &i)) 1083 continue; 1084 if (i < 0 || i >= max_netdevices) 1085 continue; 1086 1087 /* avoid cases where sscanf is not exact inverse of printf */ 1088 snprintf(buf, IFNAMSIZ, name, i); 1089 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1090 __set_bit(i, inuse); 1091 } 1092 if (!sscanf(d->name, name, &i)) 1093 continue; 1094 if (i < 0 || i >= max_netdevices) 1095 continue; 1096 1097 /* avoid cases where sscanf is not exact inverse of printf */ 1098 snprintf(buf, IFNAMSIZ, name, i); 1099 if (!strncmp(buf, d->name, IFNAMSIZ)) 1100 __set_bit(i, inuse); 1101 } 1102 1103 i = find_first_zero_bit(inuse, max_netdevices); 1104 free_page((unsigned long) inuse); 1105 } 1106 1107 snprintf(buf, IFNAMSIZ, name, i); 1108 if (!netdev_name_in_use(net, buf)) 1109 return i; 1110 1111 /* It is possible to run out of possible slots 1112 * when the name is long and there isn't enough space left 1113 * for the digits, or if all bits are used. 1114 */ 1115 return -ENFILE; 1116 } 1117 1118 static int dev_alloc_name_ns(struct net *net, 1119 struct net_device *dev, 1120 const char *name) 1121 { 1122 char buf[IFNAMSIZ]; 1123 int ret; 1124 1125 BUG_ON(!net); 1126 ret = __dev_alloc_name(net, name, buf); 1127 if (ret >= 0) 1128 strscpy(dev->name, buf, IFNAMSIZ); 1129 return ret; 1130 } 1131 1132 /** 1133 * dev_alloc_name - allocate a name for a device 1134 * @dev: device 1135 * @name: name format string 1136 * 1137 * Passed a format string - eg "lt%d" it will try and find a suitable 1138 * id. It scans list of devices to build up a free map, then chooses 1139 * the first empty slot. The caller must hold the dev_base or rtnl lock 1140 * while allocating the name and adding the device in order to avoid 1141 * duplicates. 1142 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1143 * Returns the number of the unit assigned or a negative errno code. 1144 */ 1145 1146 int dev_alloc_name(struct net_device *dev, const char *name) 1147 { 1148 return dev_alloc_name_ns(dev_net(dev), dev, name); 1149 } 1150 EXPORT_SYMBOL(dev_alloc_name); 1151 1152 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1153 const char *name) 1154 { 1155 BUG_ON(!net); 1156 1157 if (!dev_valid_name(name)) 1158 return -EINVAL; 1159 1160 if (strchr(name, '%')) 1161 return dev_alloc_name_ns(net, dev, name); 1162 else if (netdev_name_in_use(net, name)) 1163 return -EEXIST; 1164 else if (dev->name != name) 1165 strscpy(dev->name, name, IFNAMSIZ); 1166 1167 return 0; 1168 } 1169 1170 /** 1171 * dev_change_name - change name of a device 1172 * @dev: device 1173 * @newname: name (or format string) must be at least IFNAMSIZ 1174 * 1175 * Change name of a device, can pass format strings "eth%d". 1176 * for wildcarding. 1177 */ 1178 int dev_change_name(struct net_device *dev, const char *newname) 1179 { 1180 unsigned char old_assign_type; 1181 char oldname[IFNAMSIZ]; 1182 int err = 0; 1183 int ret; 1184 struct net *net; 1185 1186 ASSERT_RTNL(); 1187 BUG_ON(!dev_net(dev)); 1188 1189 net = dev_net(dev); 1190 1191 down_write(&devnet_rename_sem); 1192 1193 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1194 up_write(&devnet_rename_sem); 1195 return 0; 1196 } 1197 1198 memcpy(oldname, dev->name, IFNAMSIZ); 1199 1200 err = dev_get_valid_name(net, dev, newname); 1201 if (err < 0) { 1202 up_write(&devnet_rename_sem); 1203 return err; 1204 } 1205 1206 if (oldname[0] && !strchr(oldname, '%')) 1207 netdev_info(dev, "renamed from %s%s\n", oldname, 1208 dev->flags & IFF_UP ? " (while UP)" : ""); 1209 1210 old_assign_type = dev->name_assign_type; 1211 dev->name_assign_type = NET_NAME_RENAMED; 1212 1213 rollback: 1214 ret = device_rename(&dev->dev, dev->name); 1215 if (ret) { 1216 memcpy(dev->name, oldname, IFNAMSIZ); 1217 dev->name_assign_type = old_assign_type; 1218 up_write(&devnet_rename_sem); 1219 return ret; 1220 } 1221 1222 up_write(&devnet_rename_sem); 1223 1224 netdev_adjacent_rename_links(dev, oldname); 1225 1226 write_lock(&dev_base_lock); 1227 netdev_name_node_del(dev->name_node); 1228 write_unlock(&dev_base_lock); 1229 1230 synchronize_rcu(); 1231 1232 write_lock(&dev_base_lock); 1233 netdev_name_node_add(net, dev->name_node); 1234 write_unlock(&dev_base_lock); 1235 1236 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1237 ret = notifier_to_errno(ret); 1238 1239 if (ret) { 1240 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1241 if (err >= 0) { 1242 err = ret; 1243 down_write(&devnet_rename_sem); 1244 memcpy(dev->name, oldname, IFNAMSIZ); 1245 memcpy(oldname, newname, IFNAMSIZ); 1246 dev->name_assign_type = old_assign_type; 1247 old_assign_type = NET_NAME_RENAMED; 1248 goto rollback; 1249 } else { 1250 netdev_err(dev, "name change rollback failed: %d\n", 1251 ret); 1252 } 1253 } 1254 1255 return err; 1256 } 1257 1258 /** 1259 * dev_set_alias - change ifalias of a device 1260 * @dev: device 1261 * @alias: name up to IFALIASZ 1262 * @len: limit of bytes to copy from info 1263 * 1264 * Set ifalias for a device, 1265 */ 1266 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1267 { 1268 struct dev_ifalias *new_alias = NULL; 1269 1270 if (len >= IFALIASZ) 1271 return -EINVAL; 1272 1273 if (len) { 1274 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1275 if (!new_alias) 1276 return -ENOMEM; 1277 1278 memcpy(new_alias->ifalias, alias, len); 1279 new_alias->ifalias[len] = 0; 1280 } 1281 1282 mutex_lock(&ifalias_mutex); 1283 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1284 mutex_is_locked(&ifalias_mutex)); 1285 mutex_unlock(&ifalias_mutex); 1286 1287 if (new_alias) 1288 kfree_rcu(new_alias, rcuhead); 1289 1290 return len; 1291 } 1292 EXPORT_SYMBOL(dev_set_alias); 1293 1294 /** 1295 * dev_get_alias - get ifalias of a device 1296 * @dev: device 1297 * @name: buffer to store name of ifalias 1298 * @len: size of buffer 1299 * 1300 * get ifalias for a device. Caller must make sure dev cannot go 1301 * away, e.g. rcu read lock or own a reference count to device. 1302 */ 1303 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1304 { 1305 const struct dev_ifalias *alias; 1306 int ret = 0; 1307 1308 rcu_read_lock(); 1309 alias = rcu_dereference(dev->ifalias); 1310 if (alias) 1311 ret = snprintf(name, len, "%s", alias->ifalias); 1312 rcu_read_unlock(); 1313 1314 return ret; 1315 } 1316 1317 /** 1318 * netdev_features_change - device changes features 1319 * @dev: device to cause notification 1320 * 1321 * Called to indicate a device has changed features. 1322 */ 1323 void netdev_features_change(struct net_device *dev) 1324 { 1325 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1326 } 1327 EXPORT_SYMBOL(netdev_features_change); 1328 1329 /** 1330 * netdev_state_change - device changes state 1331 * @dev: device to cause notification 1332 * 1333 * Called to indicate a device has changed state. This function calls 1334 * the notifier chains for netdev_chain and sends a NEWLINK message 1335 * to the routing socket. 1336 */ 1337 void netdev_state_change(struct net_device *dev) 1338 { 1339 if (dev->flags & IFF_UP) { 1340 struct netdev_notifier_change_info change_info = { 1341 .info.dev = dev, 1342 }; 1343 1344 call_netdevice_notifiers_info(NETDEV_CHANGE, 1345 &change_info.info); 1346 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1347 } 1348 } 1349 EXPORT_SYMBOL(netdev_state_change); 1350 1351 /** 1352 * __netdev_notify_peers - notify network peers about existence of @dev, 1353 * to be called when rtnl lock is already held. 1354 * @dev: network device 1355 * 1356 * Generate traffic such that interested network peers are aware of 1357 * @dev, such as by generating a gratuitous ARP. This may be used when 1358 * a device wants to inform the rest of the network about some sort of 1359 * reconfiguration such as a failover event or virtual machine 1360 * migration. 1361 */ 1362 void __netdev_notify_peers(struct net_device *dev) 1363 { 1364 ASSERT_RTNL(); 1365 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1366 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1367 } 1368 EXPORT_SYMBOL(__netdev_notify_peers); 1369 1370 /** 1371 * netdev_notify_peers - notify network peers about existence of @dev 1372 * @dev: network device 1373 * 1374 * Generate traffic such that interested network peers are aware of 1375 * @dev, such as by generating a gratuitous ARP. This may be used when 1376 * a device wants to inform the rest of the network about some sort of 1377 * reconfiguration such as a failover event or virtual machine 1378 * migration. 1379 */ 1380 void netdev_notify_peers(struct net_device *dev) 1381 { 1382 rtnl_lock(); 1383 __netdev_notify_peers(dev); 1384 rtnl_unlock(); 1385 } 1386 EXPORT_SYMBOL(netdev_notify_peers); 1387 1388 static int napi_threaded_poll(void *data); 1389 1390 static int napi_kthread_create(struct napi_struct *n) 1391 { 1392 int err = 0; 1393 1394 /* Create and wake up the kthread once to put it in 1395 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1396 * warning and work with loadavg. 1397 */ 1398 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1399 n->dev->name, n->napi_id); 1400 if (IS_ERR(n->thread)) { 1401 err = PTR_ERR(n->thread); 1402 pr_err("kthread_run failed with err %d\n", err); 1403 n->thread = NULL; 1404 } 1405 1406 return err; 1407 } 1408 1409 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1410 { 1411 const struct net_device_ops *ops = dev->netdev_ops; 1412 int ret; 1413 1414 ASSERT_RTNL(); 1415 dev_addr_check(dev); 1416 1417 if (!netif_device_present(dev)) { 1418 /* may be detached because parent is runtime-suspended */ 1419 if (dev->dev.parent) 1420 pm_runtime_resume(dev->dev.parent); 1421 if (!netif_device_present(dev)) 1422 return -ENODEV; 1423 } 1424 1425 /* Block netpoll from trying to do any rx path servicing. 1426 * If we don't do this there is a chance ndo_poll_controller 1427 * or ndo_poll may be running while we open the device 1428 */ 1429 netpoll_poll_disable(dev); 1430 1431 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1432 ret = notifier_to_errno(ret); 1433 if (ret) 1434 return ret; 1435 1436 set_bit(__LINK_STATE_START, &dev->state); 1437 1438 if (ops->ndo_validate_addr) 1439 ret = ops->ndo_validate_addr(dev); 1440 1441 if (!ret && ops->ndo_open) 1442 ret = ops->ndo_open(dev); 1443 1444 netpoll_poll_enable(dev); 1445 1446 if (ret) 1447 clear_bit(__LINK_STATE_START, &dev->state); 1448 else { 1449 dev->flags |= IFF_UP; 1450 dev_set_rx_mode(dev); 1451 dev_activate(dev); 1452 add_device_randomness(dev->dev_addr, dev->addr_len); 1453 } 1454 1455 return ret; 1456 } 1457 1458 /** 1459 * dev_open - prepare an interface for use. 1460 * @dev: device to open 1461 * @extack: netlink extended ack 1462 * 1463 * Takes a device from down to up state. The device's private open 1464 * function is invoked and then the multicast lists are loaded. Finally 1465 * the device is moved into the up state and a %NETDEV_UP message is 1466 * sent to the netdev notifier chain. 1467 * 1468 * Calling this function on an active interface is a nop. On a failure 1469 * a negative errno code is returned. 1470 */ 1471 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1472 { 1473 int ret; 1474 1475 if (dev->flags & IFF_UP) 1476 return 0; 1477 1478 ret = __dev_open(dev, extack); 1479 if (ret < 0) 1480 return ret; 1481 1482 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1483 call_netdevice_notifiers(NETDEV_UP, dev); 1484 1485 return ret; 1486 } 1487 EXPORT_SYMBOL(dev_open); 1488 1489 static void __dev_close_many(struct list_head *head) 1490 { 1491 struct net_device *dev; 1492 1493 ASSERT_RTNL(); 1494 might_sleep(); 1495 1496 list_for_each_entry(dev, head, close_list) { 1497 /* Temporarily disable netpoll until the interface is down */ 1498 netpoll_poll_disable(dev); 1499 1500 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1501 1502 clear_bit(__LINK_STATE_START, &dev->state); 1503 1504 /* Synchronize to scheduled poll. We cannot touch poll list, it 1505 * can be even on different cpu. So just clear netif_running(). 1506 * 1507 * dev->stop() will invoke napi_disable() on all of it's 1508 * napi_struct instances on this device. 1509 */ 1510 smp_mb__after_atomic(); /* Commit netif_running(). */ 1511 } 1512 1513 dev_deactivate_many(head); 1514 1515 list_for_each_entry(dev, head, close_list) { 1516 const struct net_device_ops *ops = dev->netdev_ops; 1517 1518 /* 1519 * Call the device specific close. This cannot fail. 1520 * Only if device is UP 1521 * 1522 * We allow it to be called even after a DETACH hot-plug 1523 * event. 1524 */ 1525 if (ops->ndo_stop) 1526 ops->ndo_stop(dev); 1527 1528 dev->flags &= ~IFF_UP; 1529 netpoll_poll_enable(dev); 1530 } 1531 } 1532 1533 static void __dev_close(struct net_device *dev) 1534 { 1535 LIST_HEAD(single); 1536 1537 list_add(&dev->close_list, &single); 1538 __dev_close_many(&single); 1539 list_del(&single); 1540 } 1541 1542 void dev_close_many(struct list_head *head, bool unlink) 1543 { 1544 struct net_device *dev, *tmp; 1545 1546 /* Remove the devices that don't need to be closed */ 1547 list_for_each_entry_safe(dev, tmp, head, close_list) 1548 if (!(dev->flags & IFF_UP)) 1549 list_del_init(&dev->close_list); 1550 1551 __dev_close_many(head); 1552 1553 list_for_each_entry_safe(dev, tmp, head, close_list) { 1554 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1555 call_netdevice_notifiers(NETDEV_DOWN, dev); 1556 if (unlink) 1557 list_del_init(&dev->close_list); 1558 } 1559 } 1560 EXPORT_SYMBOL(dev_close_many); 1561 1562 /** 1563 * dev_close - shutdown an interface. 1564 * @dev: device to shutdown 1565 * 1566 * This function moves an active device into down state. A 1567 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1568 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1569 * chain. 1570 */ 1571 void dev_close(struct net_device *dev) 1572 { 1573 if (dev->flags & IFF_UP) { 1574 LIST_HEAD(single); 1575 1576 list_add(&dev->close_list, &single); 1577 dev_close_many(&single, true); 1578 list_del(&single); 1579 } 1580 } 1581 EXPORT_SYMBOL(dev_close); 1582 1583 1584 /** 1585 * dev_disable_lro - disable Large Receive Offload on a device 1586 * @dev: device 1587 * 1588 * Disable Large Receive Offload (LRO) on a net device. Must be 1589 * called under RTNL. This is needed if received packets may be 1590 * forwarded to another interface. 1591 */ 1592 void dev_disable_lro(struct net_device *dev) 1593 { 1594 struct net_device *lower_dev; 1595 struct list_head *iter; 1596 1597 dev->wanted_features &= ~NETIF_F_LRO; 1598 netdev_update_features(dev); 1599 1600 if (unlikely(dev->features & NETIF_F_LRO)) 1601 netdev_WARN(dev, "failed to disable LRO!\n"); 1602 1603 netdev_for_each_lower_dev(dev, lower_dev, iter) 1604 dev_disable_lro(lower_dev); 1605 } 1606 EXPORT_SYMBOL(dev_disable_lro); 1607 1608 /** 1609 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1610 * @dev: device 1611 * 1612 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1613 * called under RTNL. This is needed if Generic XDP is installed on 1614 * the device. 1615 */ 1616 static void dev_disable_gro_hw(struct net_device *dev) 1617 { 1618 dev->wanted_features &= ~NETIF_F_GRO_HW; 1619 netdev_update_features(dev); 1620 1621 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1622 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1623 } 1624 1625 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1626 { 1627 #define N(val) \ 1628 case NETDEV_##val: \ 1629 return "NETDEV_" __stringify(val); 1630 switch (cmd) { 1631 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1632 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1633 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1634 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1635 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1636 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1637 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1638 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1639 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1640 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1641 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1642 N(XDP_FEAT_CHANGE) 1643 } 1644 #undef N 1645 return "UNKNOWN_NETDEV_EVENT"; 1646 } 1647 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1648 1649 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1650 struct net_device *dev) 1651 { 1652 struct netdev_notifier_info info = { 1653 .dev = dev, 1654 }; 1655 1656 return nb->notifier_call(nb, val, &info); 1657 } 1658 1659 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1660 struct net_device *dev) 1661 { 1662 int err; 1663 1664 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1665 err = notifier_to_errno(err); 1666 if (err) 1667 return err; 1668 1669 if (!(dev->flags & IFF_UP)) 1670 return 0; 1671 1672 call_netdevice_notifier(nb, NETDEV_UP, dev); 1673 return 0; 1674 } 1675 1676 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1677 struct net_device *dev) 1678 { 1679 if (dev->flags & IFF_UP) { 1680 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1681 dev); 1682 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1683 } 1684 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1685 } 1686 1687 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1688 struct net *net) 1689 { 1690 struct net_device *dev; 1691 int err; 1692 1693 for_each_netdev(net, dev) { 1694 err = call_netdevice_register_notifiers(nb, dev); 1695 if (err) 1696 goto rollback; 1697 } 1698 return 0; 1699 1700 rollback: 1701 for_each_netdev_continue_reverse(net, dev) 1702 call_netdevice_unregister_notifiers(nb, dev); 1703 return err; 1704 } 1705 1706 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1707 struct net *net) 1708 { 1709 struct net_device *dev; 1710 1711 for_each_netdev(net, dev) 1712 call_netdevice_unregister_notifiers(nb, dev); 1713 } 1714 1715 static int dev_boot_phase = 1; 1716 1717 /** 1718 * register_netdevice_notifier - register a network notifier block 1719 * @nb: notifier 1720 * 1721 * Register a notifier to be called when network device events occur. 1722 * The notifier passed is linked into the kernel structures and must 1723 * not be reused until it has been unregistered. A negative errno code 1724 * is returned on a failure. 1725 * 1726 * When registered all registration and up events are replayed 1727 * to the new notifier to allow device to have a race free 1728 * view of the network device list. 1729 */ 1730 1731 int register_netdevice_notifier(struct notifier_block *nb) 1732 { 1733 struct net *net; 1734 int err; 1735 1736 /* Close race with setup_net() and cleanup_net() */ 1737 down_write(&pernet_ops_rwsem); 1738 rtnl_lock(); 1739 err = raw_notifier_chain_register(&netdev_chain, nb); 1740 if (err) 1741 goto unlock; 1742 if (dev_boot_phase) 1743 goto unlock; 1744 for_each_net(net) { 1745 err = call_netdevice_register_net_notifiers(nb, net); 1746 if (err) 1747 goto rollback; 1748 } 1749 1750 unlock: 1751 rtnl_unlock(); 1752 up_write(&pernet_ops_rwsem); 1753 return err; 1754 1755 rollback: 1756 for_each_net_continue_reverse(net) 1757 call_netdevice_unregister_net_notifiers(nb, net); 1758 1759 raw_notifier_chain_unregister(&netdev_chain, nb); 1760 goto unlock; 1761 } 1762 EXPORT_SYMBOL(register_netdevice_notifier); 1763 1764 /** 1765 * unregister_netdevice_notifier - unregister a network notifier block 1766 * @nb: notifier 1767 * 1768 * Unregister a notifier previously registered by 1769 * register_netdevice_notifier(). The notifier is unlinked into the 1770 * kernel structures and may then be reused. A negative errno code 1771 * is returned on a failure. 1772 * 1773 * After unregistering unregister and down device events are synthesized 1774 * for all devices on the device list to the removed notifier to remove 1775 * the need for special case cleanup code. 1776 */ 1777 1778 int unregister_netdevice_notifier(struct notifier_block *nb) 1779 { 1780 struct net *net; 1781 int err; 1782 1783 /* Close race with setup_net() and cleanup_net() */ 1784 down_write(&pernet_ops_rwsem); 1785 rtnl_lock(); 1786 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1787 if (err) 1788 goto unlock; 1789 1790 for_each_net(net) 1791 call_netdevice_unregister_net_notifiers(nb, net); 1792 1793 unlock: 1794 rtnl_unlock(); 1795 up_write(&pernet_ops_rwsem); 1796 return err; 1797 } 1798 EXPORT_SYMBOL(unregister_netdevice_notifier); 1799 1800 static int __register_netdevice_notifier_net(struct net *net, 1801 struct notifier_block *nb, 1802 bool ignore_call_fail) 1803 { 1804 int err; 1805 1806 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1807 if (err) 1808 return err; 1809 if (dev_boot_phase) 1810 return 0; 1811 1812 err = call_netdevice_register_net_notifiers(nb, net); 1813 if (err && !ignore_call_fail) 1814 goto chain_unregister; 1815 1816 return 0; 1817 1818 chain_unregister: 1819 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1820 return err; 1821 } 1822 1823 static int __unregister_netdevice_notifier_net(struct net *net, 1824 struct notifier_block *nb) 1825 { 1826 int err; 1827 1828 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1829 if (err) 1830 return err; 1831 1832 call_netdevice_unregister_net_notifiers(nb, net); 1833 return 0; 1834 } 1835 1836 /** 1837 * register_netdevice_notifier_net - register a per-netns network notifier block 1838 * @net: network namespace 1839 * @nb: notifier 1840 * 1841 * Register a notifier to be called when network device events occur. 1842 * The notifier passed is linked into the kernel structures and must 1843 * not be reused until it has been unregistered. A negative errno code 1844 * is returned on a failure. 1845 * 1846 * When registered all registration and up events are replayed 1847 * to the new notifier to allow device to have a race free 1848 * view of the network device list. 1849 */ 1850 1851 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1852 { 1853 int err; 1854 1855 rtnl_lock(); 1856 err = __register_netdevice_notifier_net(net, nb, false); 1857 rtnl_unlock(); 1858 return err; 1859 } 1860 EXPORT_SYMBOL(register_netdevice_notifier_net); 1861 1862 /** 1863 * unregister_netdevice_notifier_net - unregister a per-netns 1864 * network notifier block 1865 * @net: network namespace 1866 * @nb: notifier 1867 * 1868 * Unregister a notifier previously registered by 1869 * register_netdevice_notifier_net(). The notifier is unlinked from the 1870 * kernel structures and may then be reused. A negative errno code 1871 * is returned on a failure. 1872 * 1873 * After unregistering unregister and down device events are synthesized 1874 * for all devices on the device list to the removed notifier to remove 1875 * the need for special case cleanup code. 1876 */ 1877 1878 int unregister_netdevice_notifier_net(struct net *net, 1879 struct notifier_block *nb) 1880 { 1881 int err; 1882 1883 rtnl_lock(); 1884 err = __unregister_netdevice_notifier_net(net, nb); 1885 rtnl_unlock(); 1886 return err; 1887 } 1888 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1889 1890 static void __move_netdevice_notifier_net(struct net *src_net, 1891 struct net *dst_net, 1892 struct notifier_block *nb) 1893 { 1894 __unregister_netdevice_notifier_net(src_net, nb); 1895 __register_netdevice_notifier_net(dst_net, nb, true); 1896 } 1897 1898 int register_netdevice_notifier_dev_net(struct net_device *dev, 1899 struct notifier_block *nb, 1900 struct netdev_net_notifier *nn) 1901 { 1902 int err; 1903 1904 rtnl_lock(); 1905 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1906 if (!err) { 1907 nn->nb = nb; 1908 list_add(&nn->list, &dev->net_notifier_list); 1909 } 1910 rtnl_unlock(); 1911 return err; 1912 } 1913 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1914 1915 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1916 struct notifier_block *nb, 1917 struct netdev_net_notifier *nn) 1918 { 1919 int err; 1920 1921 rtnl_lock(); 1922 list_del(&nn->list); 1923 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1924 rtnl_unlock(); 1925 return err; 1926 } 1927 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1928 1929 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1930 struct net *net) 1931 { 1932 struct netdev_net_notifier *nn; 1933 1934 list_for_each_entry(nn, &dev->net_notifier_list, list) 1935 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1936 } 1937 1938 /** 1939 * call_netdevice_notifiers_info - call all network notifier blocks 1940 * @val: value passed unmodified to notifier function 1941 * @info: notifier information data 1942 * 1943 * Call all network notifier blocks. Parameters and return value 1944 * are as for raw_notifier_call_chain(). 1945 */ 1946 1947 int call_netdevice_notifiers_info(unsigned long val, 1948 struct netdev_notifier_info *info) 1949 { 1950 struct net *net = dev_net(info->dev); 1951 int ret; 1952 1953 ASSERT_RTNL(); 1954 1955 /* Run per-netns notifier block chain first, then run the global one. 1956 * Hopefully, one day, the global one is going to be removed after 1957 * all notifier block registrators get converted to be per-netns. 1958 */ 1959 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1960 if (ret & NOTIFY_STOP_MASK) 1961 return ret; 1962 return raw_notifier_call_chain(&netdev_chain, val, info); 1963 } 1964 1965 /** 1966 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1967 * for and rollback on error 1968 * @val_up: value passed unmodified to notifier function 1969 * @val_down: value passed unmodified to the notifier function when 1970 * recovering from an error on @val_up 1971 * @info: notifier information data 1972 * 1973 * Call all per-netns network notifier blocks, but not notifier blocks on 1974 * the global notifier chain. Parameters and return value are as for 1975 * raw_notifier_call_chain_robust(). 1976 */ 1977 1978 static int 1979 call_netdevice_notifiers_info_robust(unsigned long val_up, 1980 unsigned long val_down, 1981 struct netdev_notifier_info *info) 1982 { 1983 struct net *net = dev_net(info->dev); 1984 1985 ASSERT_RTNL(); 1986 1987 return raw_notifier_call_chain_robust(&net->netdev_chain, 1988 val_up, val_down, info); 1989 } 1990 1991 static int call_netdevice_notifiers_extack(unsigned long val, 1992 struct net_device *dev, 1993 struct netlink_ext_ack *extack) 1994 { 1995 struct netdev_notifier_info info = { 1996 .dev = dev, 1997 .extack = extack, 1998 }; 1999 2000 return call_netdevice_notifiers_info(val, &info); 2001 } 2002 2003 /** 2004 * call_netdevice_notifiers - call all network notifier blocks 2005 * @val: value passed unmodified to notifier function 2006 * @dev: net_device pointer passed unmodified to notifier function 2007 * 2008 * Call all network notifier blocks. Parameters and return value 2009 * are as for raw_notifier_call_chain(). 2010 */ 2011 2012 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2013 { 2014 return call_netdevice_notifiers_extack(val, dev, NULL); 2015 } 2016 EXPORT_SYMBOL(call_netdevice_notifiers); 2017 2018 /** 2019 * call_netdevice_notifiers_mtu - call all network notifier blocks 2020 * @val: value passed unmodified to notifier function 2021 * @dev: net_device pointer passed unmodified to notifier function 2022 * @arg: additional u32 argument passed to the notifier function 2023 * 2024 * Call all network notifier blocks. Parameters and return value 2025 * are as for raw_notifier_call_chain(). 2026 */ 2027 static int call_netdevice_notifiers_mtu(unsigned long val, 2028 struct net_device *dev, u32 arg) 2029 { 2030 struct netdev_notifier_info_ext info = { 2031 .info.dev = dev, 2032 .ext.mtu = arg, 2033 }; 2034 2035 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2036 2037 return call_netdevice_notifiers_info(val, &info.info); 2038 } 2039 2040 #ifdef CONFIG_NET_INGRESS 2041 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2042 2043 void net_inc_ingress_queue(void) 2044 { 2045 static_branch_inc(&ingress_needed_key); 2046 } 2047 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2048 2049 void net_dec_ingress_queue(void) 2050 { 2051 static_branch_dec(&ingress_needed_key); 2052 } 2053 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2054 #endif 2055 2056 #ifdef CONFIG_NET_EGRESS 2057 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2058 2059 void net_inc_egress_queue(void) 2060 { 2061 static_branch_inc(&egress_needed_key); 2062 } 2063 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2064 2065 void net_dec_egress_queue(void) 2066 { 2067 static_branch_dec(&egress_needed_key); 2068 } 2069 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2070 #endif 2071 2072 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2073 EXPORT_SYMBOL(netstamp_needed_key); 2074 #ifdef CONFIG_JUMP_LABEL 2075 static atomic_t netstamp_needed_deferred; 2076 static atomic_t netstamp_wanted; 2077 static void netstamp_clear(struct work_struct *work) 2078 { 2079 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2080 int wanted; 2081 2082 wanted = atomic_add_return(deferred, &netstamp_wanted); 2083 if (wanted > 0) 2084 static_branch_enable(&netstamp_needed_key); 2085 else 2086 static_branch_disable(&netstamp_needed_key); 2087 } 2088 static DECLARE_WORK(netstamp_work, netstamp_clear); 2089 #endif 2090 2091 void net_enable_timestamp(void) 2092 { 2093 #ifdef CONFIG_JUMP_LABEL 2094 int wanted = atomic_read(&netstamp_wanted); 2095 2096 while (wanted > 0) { 2097 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2098 return; 2099 } 2100 atomic_inc(&netstamp_needed_deferred); 2101 schedule_work(&netstamp_work); 2102 #else 2103 static_branch_inc(&netstamp_needed_key); 2104 #endif 2105 } 2106 EXPORT_SYMBOL(net_enable_timestamp); 2107 2108 void net_disable_timestamp(void) 2109 { 2110 #ifdef CONFIG_JUMP_LABEL 2111 int wanted = atomic_read(&netstamp_wanted); 2112 2113 while (wanted > 1) { 2114 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2115 return; 2116 } 2117 atomic_dec(&netstamp_needed_deferred); 2118 schedule_work(&netstamp_work); 2119 #else 2120 static_branch_dec(&netstamp_needed_key); 2121 #endif 2122 } 2123 EXPORT_SYMBOL(net_disable_timestamp); 2124 2125 static inline void net_timestamp_set(struct sk_buff *skb) 2126 { 2127 skb->tstamp = 0; 2128 skb->mono_delivery_time = 0; 2129 if (static_branch_unlikely(&netstamp_needed_key)) 2130 skb->tstamp = ktime_get_real(); 2131 } 2132 2133 #define net_timestamp_check(COND, SKB) \ 2134 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2135 if ((COND) && !(SKB)->tstamp) \ 2136 (SKB)->tstamp = ktime_get_real(); \ 2137 } \ 2138 2139 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2140 { 2141 return __is_skb_forwardable(dev, skb, true); 2142 } 2143 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2144 2145 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2146 bool check_mtu) 2147 { 2148 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2149 2150 if (likely(!ret)) { 2151 skb->protocol = eth_type_trans(skb, dev); 2152 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2153 } 2154 2155 return ret; 2156 } 2157 2158 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2159 { 2160 return __dev_forward_skb2(dev, skb, true); 2161 } 2162 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2163 2164 /** 2165 * dev_forward_skb - loopback an skb to another netif 2166 * 2167 * @dev: destination network device 2168 * @skb: buffer to forward 2169 * 2170 * return values: 2171 * NET_RX_SUCCESS (no congestion) 2172 * NET_RX_DROP (packet was dropped, but freed) 2173 * 2174 * dev_forward_skb can be used for injecting an skb from the 2175 * start_xmit function of one device into the receive queue 2176 * of another device. 2177 * 2178 * The receiving device may be in another namespace, so 2179 * we have to clear all information in the skb that could 2180 * impact namespace isolation. 2181 */ 2182 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2183 { 2184 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2185 } 2186 EXPORT_SYMBOL_GPL(dev_forward_skb); 2187 2188 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2189 { 2190 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2191 } 2192 2193 static inline int deliver_skb(struct sk_buff *skb, 2194 struct packet_type *pt_prev, 2195 struct net_device *orig_dev) 2196 { 2197 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2198 return -ENOMEM; 2199 refcount_inc(&skb->users); 2200 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2201 } 2202 2203 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2204 struct packet_type **pt, 2205 struct net_device *orig_dev, 2206 __be16 type, 2207 struct list_head *ptype_list) 2208 { 2209 struct packet_type *ptype, *pt_prev = *pt; 2210 2211 list_for_each_entry_rcu(ptype, ptype_list, list) { 2212 if (ptype->type != type) 2213 continue; 2214 if (pt_prev) 2215 deliver_skb(skb, pt_prev, orig_dev); 2216 pt_prev = ptype; 2217 } 2218 *pt = pt_prev; 2219 } 2220 2221 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2222 { 2223 if (!ptype->af_packet_priv || !skb->sk) 2224 return false; 2225 2226 if (ptype->id_match) 2227 return ptype->id_match(ptype, skb->sk); 2228 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2229 return true; 2230 2231 return false; 2232 } 2233 2234 /** 2235 * dev_nit_active - return true if any network interface taps are in use 2236 * 2237 * @dev: network device to check for the presence of taps 2238 */ 2239 bool dev_nit_active(struct net_device *dev) 2240 { 2241 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2242 } 2243 EXPORT_SYMBOL_GPL(dev_nit_active); 2244 2245 /* 2246 * Support routine. Sends outgoing frames to any network 2247 * taps currently in use. 2248 */ 2249 2250 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2251 { 2252 struct packet_type *ptype; 2253 struct sk_buff *skb2 = NULL; 2254 struct packet_type *pt_prev = NULL; 2255 struct list_head *ptype_list = &ptype_all; 2256 2257 rcu_read_lock(); 2258 again: 2259 list_for_each_entry_rcu(ptype, ptype_list, list) { 2260 if (ptype->ignore_outgoing) 2261 continue; 2262 2263 /* Never send packets back to the socket 2264 * they originated from - MvS (miquels@drinkel.ow.org) 2265 */ 2266 if (skb_loop_sk(ptype, skb)) 2267 continue; 2268 2269 if (pt_prev) { 2270 deliver_skb(skb2, pt_prev, skb->dev); 2271 pt_prev = ptype; 2272 continue; 2273 } 2274 2275 /* need to clone skb, done only once */ 2276 skb2 = skb_clone(skb, GFP_ATOMIC); 2277 if (!skb2) 2278 goto out_unlock; 2279 2280 net_timestamp_set(skb2); 2281 2282 /* skb->nh should be correctly 2283 * set by sender, so that the second statement is 2284 * just protection against buggy protocols. 2285 */ 2286 skb_reset_mac_header(skb2); 2287 2288 if (skb_network_header(skb2) < skb2->data || 2289 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2290 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2291 ntohs(skb2->protocol), 2292 dev->name); 2293 skb_reset_network_header(skb2); 2294 } 2295 2296 skb2->transport_header = skb2->network_header; 2297 skb2->pkt_type = PACKET_OUTGOING; 2298 pt_prev = ptype; 2299 } 2300 2301 if (ptype_list == &ptype_all) { 2302 ptype_list = &dev->ptype_all; 2303 goto again; 2304 } 2305 out_unlock: 2306 if (pt_prev) { 2307 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2308 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2309 else 2310 kfree_skb(skb2); 2311 } 2312 rcu_read_unlock(); 2313 } 2314 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2315 2316 /** 2317 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2318 * @dev: Network device 2319 * @txq: number of queues available 2320 * 2321 * If real_num_tx_queues is changed the tc mappings may no longer be 2322 * valid. To resolve this verify the tc mapping remains valid and if 2323 * not NULL the mapping. With no priorities mapping to this 2324 * offset/count pair it will no longer be used. In the worst case TC0 2325 * is invalid nothing can be done so disable priority mappings. If is 2326 * expected that drivers will fix this mapping if they can before 2327 * calling netif_set_real_num_tx_queues. 2328 */ 2329 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2330 { 2331 int i; 2332 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2333 2334 /* If TC0 is invalidated disable TC mapping */ 2335 if (tc->offset + tc->count > txq) { 2336 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2337 dev->num_tc = 0; 2338 return; 2339 } 2340 2341 /* Invalidated prio to tc mappings set to TC0 */ 2342 for (i = 1; i < TC_BITMASK + 1; i++) { 2343 int q = netdev_get_prio_tc_map(dev, i); 2344 2345 tc = &dev->tc_to_txq[q]; 2346 if (tc->offset + tc->count > txq) { 2347 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", 2348 i, q); 2349 netdev_set_prio_tc_map(dev, i, 0); 2350 } 2351 } 2352 } 2353 2354 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2355 { 2356 if (dev->num_tc) { 2357 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2358 int i; 2359 2360 /* walk through the TCs and see if it falls into any of them */ 2361 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2362 if ((txq - tc->offset) < tc->count) 2363 return i; 2364 } 2365 2366 /* didn't find it, just return -1 to indicate no match */ 2367 return -1; 2368 } 2369 2370 return 0; 2371 } 2372 EXPORT_SYMBOL(netdev_txq_to_tc); 2373 2374 #ifdef CONFIG_XPS 2375 static struct static_key xps_needed __read_mostly; 2376 static struct static_key xps_rxqs_needed __read_mostly; 2377 static DEFINE_MUTEX(xps_map_mutex); 2378 #define xmap_dereference(P) \ 2379 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2380 2381 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2382 struct xps_dev_maps *old_maps, int tci, u16 index) 2383 { 2384 struct xps_map *map = NULL; 2385 int pos; 2386 2387 if (dev_maps) 2388 map = xmap_dereference(dev_maps->attr_map[tci]); 2389 if (!map) 2390 return false; 2391 2392 for (pos = map->len; pos--;) { 2393 if (map->queues[pos] != index) 2394 continue; 2395 2396 if (map->len > 1) { 2397 map->queues[pos] = map->queues[--map->len]; 2398 break; 2399 } 2400 2401 if (old_maps) 2402 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2403 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2404 kfree_rcu(map, rcu); 2405 return false; 2406 } 2407 2408 return true; 2409 } 2410 2411 static bool remove_xps_queue_cpu(struct net_device *dev, 2412 struct xps_dev_maps *dev_maps, 2413 int cpu, u16 offset, u16 count) 2414 { 2415 int num_tc = dev_maps->num_tc; 2416 bool active = false; 2417 int tci; 2418 2419 for (tci = cpu * num_tc; num_tc--; tci++) { 2420 int i, j; 2421 2422 for (i = count, j = offset; i--; j++) { 2423 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2424 break; 2425 } 2426 2427 active |= i < 0; 2428 } 2429 2430 return active; 2431 } 2432 2433 static void reset_xps_maps(struct net_device *dev, 2434 struct xps_dev_maps *dev_maps, 2435 enum xps_map_type type) 2436 { 2437 static_key_slow_dec_cpuslocked(&xps_needed); 2438 if (type == XPS_RXQS) 2439 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2440 2441 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2442 2443 kfree_rcu(dev_maps, rcu); 2444 } 2445 2446 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2447 u16 offset, u16 count) 2448 { 2449 struct xps_dev_maps *dev_maps; 2450 bool active = false; 2451 int i, j; 2452 2453 dev_maps = xmap_dereference(dev->xps_maps[type]); 2454 if (!dev_maps) 2455 return; 2456 2457 for (j = 0; j < dev_maps->nr_ids; j++) 2458 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2459 if (!active) 2460 reset_xps_maps(dev, dev_maps, type); 2461 2462 if (type == XPS_CPUS) { 2463 for (i = offset + (count - 1); count--; i--) 2464 netdev_queue_numa_node_write( 2465 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2466 } 2467 } 2468 2469 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2470 u16 count) 2471 { 2472 if (!static_key_false(&xps_needed)) 2473 return; 2474 2475 cpus_read_lock(); 2476 mutex_lock(&xps_map_mutex); 2477 2478 if (static_key_false(&xps_rxqs_needed)) 2479 clean_xps_maps(dev, XPS_RXQS, offset, count); 2480 2481 clean_xps_maps(dev, XPS_CPUS, offset, count); 2482 2483 mutex_unlock(&xps_map_mutex); 2484 cpus_read_unlock(); 2485 } 2486 2487 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2488 { 2489 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2490 } 2491 2492 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2493 u16 index, bool is_rxqs_map) 2494 { 2495 struct xps_map *new_map; 2496 int alloc_len = XPS_MIN_MAP_ALLOC; 2497 int i, pos; 2498 2499 for (pos = 0; map && pos < map->len; pos++) { 2500 if (map->queues[pos] != index) 2501 continue; 2502 return map; 2503 } 2504 2505 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2506 if (map) { 2507 if (pos < map->alloc_len) 2508 return map; 2509 2510 alloc_len = map->alloc_len * 2; 2511 } 2512 2513 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2514 * map 2515 */ 2516 if (is_rxqs_map) 2517 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2518 else 2519 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2520 cpu_to_node(attr_index)); 2521 if (!new_map) 2522 return NULL; 2523 2524 for (i = 0; i < pos; i++) 2525 new_map->queues[i] = map->queues[i]; 2526 new_map->alloc_len = alloc_len; 2527 new_map->len = pos; 2528 2529 return new_map; 2530 } 2531 2532 /* Copy xps maps at a given index */ 2533 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2534 struct xps_dev_maps *new_dev_maps, int index, 2535 int tc, bool skip_tc) 2536 { 2537 int i, tci = index * dev_maps->num_tc; 2538 struct xps_map *map; 2539 2540 /* copy maps belonging to foreign traffic classes */ 2541 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2542 if (i == tc && skip_tc) 2543 continue; 2544 2545 /* fill in the new device map from the old device map */ 2546 map = xmap_dereference(dev_maps->attr_map[tci]); 2547 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2548 } 2549 } 2550 2551 /* Must be called under cpus_read_lock */ 2552 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2553 u16 index, enum xps_map_type type) 2554 { 2555 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2556 const unsigned long *online_mask = NULL; 2557 bool active = false, copy = false; 2558 int i, j, tci, numa_node_id = -2; 2559 int maps_sz, num_tc = 1, tc = 0; 2560 struct xps_map *map, *new_map; 2561 unsigned int nr_ids; 2562 2563 WARN_ON_ONCE(index >= dev->num_tx_queues); 2564 2565 if (dev->num_tc) { 2566 /* Do not allow XPS on subordinate device directly */ 2567 num_tc = dev->num_tc; 2568 if (num_tc < 0) 2569 return -EINVAL; 2570 2571 /* If queue belongs to subordinate dev use its map */ 2572 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2573 2574 tc = netdev_txq_to_tc(dev, index); 2575 if (tc < 0) 2576 return -EINVAL; 2577 } 2578 2579 mutex_lock(&xps_map_mutex); 2580 2581 dev_maps = xmap_dereference(dev->xps_maps[type]); 2582 if (type == XPS_RXQS) { 2583 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2584 nr_ids = dev->num_rx_queues; 2585 } else { 2586 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2587 if (num_possible_cpus() > 1) 2588 online_mask = cpumask_bits(cpu_online_mask); 2589 nr_ids = nr_cpu_ids; 2590 } 2591 2592 if (maps_sz < L1_CACHE_BYTES) 2593 maps_sz = L1_CACHE_BYTES; 2594 2595 /* The old dev_maps could be larger or smaller than the one we're 2596 * setting up now, as dev->num_tc or nr_ids could have been updated in 2597 * between. We could try to be smart, but let's be safe instead and only 2598 * copy foreign traffic classes if the two map sizes match. 2599 */ 2600 if (dev_maps && 2601 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2602 copy = true; 2603 2604 /* allocate memory for queue storage */ 2605 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2606 j < nr_ids;) { 2607 if (!new_dev_maps) { 2608 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2609 if (!new_dev_maps) { 2610 mutex_unlock(&xps_map_mutex); 2611 return -ENOMEM; 2612 } 2613 2614 new_dev_maps->nr_ids = nr_ids; 2615 new_dev_maps->num_tc = num_tc; 2616 } 2617 2618 tci = j * num_tc + tc; 2619 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2620 2621 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2622 if (!map) 2623 goto error; 2624 2625 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2626 } 2627 2628 if (!new_dev_maps) 2629 goto out_no_new_maps; 2630 2631 if (!dev_maps) { 2632 /* Increment static keys at most once per type */ 2633 static_key_slow_inc_cpuslocked(&xps_needed); 2634 if (type == XPS_RXQS) 2635 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2636 } 2637 2638 for (j = 0; j < nr_ids; j++) { 2639 bool skip_tc = false; 2640 2641 tci = j * num_tc + tc; 2642 if (netif_attr_test_mask(j, mask, nr_ids) && 2643 netif_attr_test_online(j, online_mask, nr_ids)) { 2644 /* add tx-queue to CPU/rx-queue maps */ 2645 int pos = 0; 2646 2647 skip_tc = true; 2648 2649 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2650 while ((pos < map->len) && (map->queues[pos] != index)) 2651 pos++; 2652 2653 if (pos == map->len) 2654 map->queues[map->len++] = index; 2655 #ifdef CONFIG_NUMA 2656 if (type == XPS_CPUS) { 2657 if (numa_node_id == -2) 2658 numa_node_id = cpu_to_node(j); 2659 else if (numa_node_id != cpu_to_node(j)) 2660 numa_node_id = -1; 2661 } 2662 #endif 2663 } 2664 2665 if (copy) 2666 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2667 skip_tc); 2668 } 2669 2670 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2671 2672 /* Cleanup old maps */ 2673 if (!dev_maps) 2674 goto out_no_old_maps; 2675 2676 for (j = 0; j < dev_maps->nr_ids; j++) { 2677 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2678 map = xmap_dereference(dev_maps->attr_map[tci]); 2679 if (!map) 2680 continue; 2681 2682 if (copy) { 2683 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2684 if (map == new_map) 2685 continue; 2686 } 2687 2688 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2689 kfree_rcu(map, rcu); 2690 } 2691 } 2692 2693 old_dev_maps = dev_maps; 2694 2695 out_no_old_maps: 2696 dev_maps = new_dev_maps; 2697 active = true; 2698 2699 out_no_new_maps: 2700 if (type == XPS_CPUS) 2701 /* update Tx queue numa node */ 2702 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2703 (numa_node_id >= 0) ? 2704 numa_node_id : NUMA_NO_NODE); 2705 2706 if (!dev_maps) 2707 goto out_no_maps; 2708 2709 /* removes tx-queue from unused CPUs/rx-queues */ 2710 for (j = 0; j < dev_maps->nr_ids; j++) { 2711 tci = j * dev_maps->num_tc; 2712 2713 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2714 if (i == tc && 2715 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2716 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2717 continue; 2718 2719 active |= remove_xps_queue(dev_maps, 2720 copy ? old_dev_maps : NULL, 2721 tci, index); 2722 } 2723 } 2724 2725 if (old_dev_maps) 2726 kfree_rcu(old_dev_maps, rcu); 2727 2728 /* free map if not active */ 2729 if (!active) 2730 reset_xps_maps(dev, dev_maps, type); 2731 2732 out_no_maps: 2733 mutex_unlock(&xps_map_mutex); 2734 2735 return 0; 2736 error: 2737 /* remove any maps that we added */ 2738 for (j = 0; j < nr_ids; j++) { 2739 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2740 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2741 map = copy ? 2742 xmap_dereference(dev_maps->attr_map[tci]) : 2743 NULL; 2744 if (new_map && new_map != map) 2745 kfree(new_map); 2746 } 2747 } 2748 2749 mutex_unlock(&xps_map_mutex); 2750 2751 kfree(new_dev_maps); 2752 return -ENOMEM; 2753 } 2754 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2755 2756 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2757 u16 index) 2758 { 2759 int ret; 2760 2761 cpus_read_lock(); 2762 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2763 cpus_read_unlock(); 2764 2765 return ret; 2766 } 2767 EXPORT_SYMBOL(netif_set_xps_queue); 2768 2769 #endif 2770 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2771 { 2772 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2773 2774 /* Unbind any subordinate channels */ 2775 while (txq-- != &dev->_tx[0]) { 2776 if (txq->sb_dev) 2777 netdev_unbind_sb_channel(dev, txq->sb_dev); 2778 } 2779 } 2780 2781 void netdev_reset_tc(struct net_device *dev) 2782 { 2783 #ifdef CONFIG_XPS 2784 netif_reset_xps_queues_gt(dev, 0); 2785 #endif 2786 netdev_unbind_all_sb_channels(dev); 2787 2788 /* Reset TC configuration of device */ 2789 dev->num_tc = 0; 2790 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2791 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2792 } 2793 EXPORT_SYMBOL(netdev_reset_tc); 2794 2795 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2796 { 2797 if (tc >= dev->num_tc) 2798 return -EINVAL; 2799 2800 #ifdef CONFIG_XPS 2801 netif_reset_xps_queues(dev, offset, count); 2802 #endif 2803 dev->tc_to_txq[tc].count = count; 2804 dev->tc_to_txq[tc].offset = offset; 2805 return 0; 2806 } 2807 EXPORT_SYMBOL(netdev_set_tc_queue); 2808 2809 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2810 { 2811 if (num_tc > TC_MAX_QUEUE) 2812 return -EINVAL; 2813 2814 #ifdef CONFIG_XPS 2815 netif_reset_xps_queues_gt(dev, 0); 2816 #endif 2817 netdev_unbind_all_sb_channels(dev); 2818 2819 dev->num_tc = num_tc; 2820 return 0; 2821 } 2822 EXPORT_SYMBOL(netdev_set_num_tc); 2823 2824 void netdev_unbind_sb_channel(struct net_device *dev, 2825 struct net_device *sb_dev) 2826 { 2827 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2828 2829 #ifdef CONFIG_XPS 2830 netif_reset_xps_queues_gt(sb_dev, 0); 2831 #endif 2832 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2833 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2834 2835 while (txq-- != &dev->_tx[0]) { 2836 if (txq->sb_dev == sb_dev) 2837 txq->sb_dev = NULL; 2838 } 2839 } 2840 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2841 2842 int netdev_bind_sb_channel_queue(struct net_device *dev, 2843 struct net_device *sb_dev, 2844 u8 tc, u16 count, u16 offset) 2845 { 2846 /* Make certain the sb_dev and dev are already configured */ 2847 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2848 return -EINVAL; 2849 2850 /* We cannot hand out queues we don't have */ 2851 if ((offset + count) > dev->real_num_tx_queues) 2852 return -EINVAL; 2853 2854 /* Record the mapping */ 2855 sb_dev->tc_to_txq[tc].count = count; 2856 sb_dev->tc_to_txq[tc].offset = offset; 2857 2858 /* Provide a way for Tx queue to find the tc_to_txq map or 2859 * XPS map for itself. 2860 */ 2861 while (count--) 2862 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2863 2864 return 0; 2865 } 2866 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2867 2868 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2869 { 2870 /* Do not use a multiqueue device to represent a subordinate channel */ 2871 if (netif_is_multiqueue(dev)) 2872 return -ENODEV; 2873 2874 /* We allow channels 1 - 32767 to be used for subordinate channels. 2875 * Channel 0 is meant to be "native" mode and used only to represent 2876 * the main root device. We allow writing 0 to reset the device back 2877 * to normal mode after being used as a subordinate channel. 2878 */ 2879 if (channel > S16_MAX) 2880 return -EINVAL; 2881 2882 dev->num_tc = -channel; 2883 2884 return 0; 2885 } 2886 EXPORT_SYMBOL(netdev_set_sb_channel); 2887 2888 /* 2889 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2890 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2891 */ 2892 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2893 { 2894 bool disabling; 2895 int rc; 2896 2897 disabling = txq < dev->real_num_tx_queues; 2898 2899 if (txq < 1 || txq > dev->num_tx_queues) 2900 return -EINVAL; 2901 2902 if (dev->reg_state == NETREG_REGISTERED || 2903 dev->reg_state == NETREG_UNREGISTERING) { 2904 ASSERT_RTNL(); 2905 2906 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2907 txq); 2908 if (rc) 2909 return rc; 2910 2911 if (dev->num_tc) 2912 netif_setup_tc(dev, txq); 2913 2914 dev_qdisc_change_real_num_tx(dev, txq); 2915 2916 dev->real_num_tx_queues = txq; 2917 2918 if (disabling) { 2919 synchronize_net(); 2920 qdisc_reset_all_tx_gt(dev, txq); 2921 #ifdef CONFIG_XPS 2922 netif_reset_xps_queues_gt(dev, txq); 2923 #endif 2924 } 2925 } else { 2926 dev->real_num_tx_queues = txq; 2927 } 2928 2929 return 0; 2930 } 2931 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2932 2933 #ifdef CONFIG_SYSFS 2934 /** 2935 * netif_set_real_num_rx_queues - set actual number of RX queues used 2936 * @dev: Network device 2937 * @rxq: Actual number of RX queues 2938 * 2939 * This must be called either with the rtnl_lock held or before 2940 * registration of the net device. Returns 0 on success, or a 2941 * negative error code. If called before registration, it always 2942 * succeeds. 2943 */ 2944 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2945 { 2946 int rc; 2947 2948 if (rxq < 1 || rxq > dev->num_rx_queues) 2949 return -EINVAL; 2950 2951 if (dev->reg_state == NETREG_REGISTERED) { 2952 ASSERT_RTNL(); 2953 2954 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2955 rxq); 2956 if (rc) 2957 return rc; 2958 } 2959 2960 dev->real_num_rx_queues = rxq; 2961 return 0; 2962 } 2963 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2964 #endif 2965 2966 /** 2967 * netif_set_real_num_queues - set actual number of RX and TX queues used 2968 * @dev: Network device 2969 * @txq: Actual number of TX queues 2970 * @rxq: Actual number of RX queues 2971 * 2972 * Set the real number of both TX and RX queues. 2973 * Does nothing if the number of queues is already correct. 2974 */ 2975 int netif_set_real_num_queues(struct net_device *dev, 2976 unsigned int txq, unsigned int rxq) 2977 { 2978 unsigned int old_rxq = dev->real_num_rx_queues; 2979 int err; 2980 2981 if (txq < 1 || txq > dev->num_tx_queues || 2982 rxq < 1 || rxq > dev->num_rx_queues) 2983 return -EINVAL; 2984 2985 /* Start from increases, so the error path only does decreases - 2986 * decreases can't fail. 2987 */ 2988 if (rxq > dev->real_num_rx_queues) { 2989 err = netif_set_real_num_rx_queues(dev, rxq); 2990 if (err) 2991 return err; 2992 } 2993 if (txq > dev->real_num_tx_queues) { 2994 err = netif_set_real_num_tx_queues(dev, txq); 2995 if (err) 2996 goto undo_rx; 2997 } 2998 if (rxq < dev->real_num_rx_queues) 2999 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3000 if (txq < dev->real_num_tx_queues) 3001 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3002 3003 return 0; 3004 undo_rx: 3005 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3006 return err; 3007 } 3008 EXPORT_SYMBOL(netif_set_real_num_queues); 3009 3010 /** 3011 * netif_set_tso_max_size() - set the max size of TSO frames supported 3012 * @dev: netdev to update 3013 * @size: max skb->len of a TSO frame 3014 * 3015 * Set the limit on the size of TSO super-frames the device can handle. 3016 * Unless explicitly set the stack will assume the value of 3017 * %GSO_LEGACY_MAX_SIZE. 3018 */ 3019 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3020 { 3021 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3022 if (size < READ_ONCE(dev->gso_max_size)) 3023 netif_set_gso_max_size(dev, size); 3024 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3025 netif_set_gso_ipv4_max_size(dev, size); 3026 } 3027 EXPORT_SYMBOL(netif_set_tso_max_size); 3028 3029 /** 3030 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3031 * @dev: netdev to update 3032 * @segs: max number of TCP segments 3033 * 3034 * Set the limit on the number of TCP segments the device can generate from 3035 * a single TSO super-frame. 3036 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3037 */ 3038 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3039 { 3040 dev->tso_max_segs = segs; 3041 if (segs < READ_ONCE(dev->gso_max_segs)) 3042 netif_set_gso_max_segs(dev, segs); 3043 } 3044 EXPORT_SYMBOL(netif_set_tso_max_segs); 3045 3046 /** 3047 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3048 * @to: netdev to update 3049 * @from: netdev from which to copy the limits 3050 */ 3051 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3052 { 3053 netif_set_tso_max_size(to, from->tso_max_size); 3054 netif_set_tso_max_segs(to, from->tso_max_segs); 3055 } 3056 EXPORT_SYMBOL(netif_inherit_tso_max); 3057 3058 /** 3059 * netif_get_num_default_rss_queues - default number of RSS queues 3060 * 3061 * Default value is the number of physical cores if there are only 1 or 2, or 3062 * divided by 2 if there are more. 3063 */ 3064 int netif_get_num_default_rss_queues(void) 3065 { 3066 cpumask_var_t cpus; 3067 int cpu, count = 0; 3068 3069 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3070 return 1; 3071 3072 cpumask_copy(cpus, cpu_online_mask); 3073 for_each_cpu(cpu, cpus) { 3074 ++count; 3075 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3076 } 3077 free_cpumask_var(cpus); 3078 3079 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3080 } 3081 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3082 3083 static void __netif_reschedule(struct Qdisc *q) 3084 { 3085 struct softnet_data *sd; 3086 unsigned long flags; 3087 3088 local_irq_save(flags); 3089 sd = this_cpu_ptr(&softnet_data); 3090 q->next_sched = NULL; 3091 *sd->output_queue_tailp = q; 3092 sd->output_queue_tailp = &q->next_sched; 3093 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3094 local_irq_restore(flags); 3095 } 3096 3097 void __netif_schedule(struct Qdisc *q) 3098 { 3099 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3100 __netif_reschedule(q); 3101 } 3102 EXPORT_SYMBOL(__netif_schedule); 3103 3104 struct dev_kfree_skb_cb { 3105 enum skb_drop_reason reason; 3106 }; 3107 3108 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3109 { 3110 return (struct dev_kfree_skb_cb *)skb->cb; 3111 } 3112 3113 void netif_schedule_queue(struct netdev_queue *txq) 3114 { 3115 rcu_read_lock(); 3116 if (!netif_xmit_stopped(txq)) { 3117 struct Qdisc *q = rcu_dereference(txq->qdisc); 3118 3119 __netif_schedule(q); 3120 } 3121 rcu_read_unlock(); 3122 } 3123 EXPORT_SYMBOL(netif_schedule_queue); 3124 3125 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3126 { 3127 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3128 struct Qdisc *q; 3129 3130 rcu_read_lock(); 3131 q = rcu_dereference(dev_queue->qdisc); 3132 __netif_schedule(q); 3133 rcu_read_unlock(); 3134 } 3135 } 3136 EXPORT_SYMBOL(netif_tx_wake_queue); 3137 3138 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3139 { 3140 unsigned long flags; 3141 3142 if (unlikely(!skb)) 3143 return; 3144 3145 if (likely(refcount_read(&skb->users) == 1)) { 3146 smp_rmb(); 3147 refcount_set(&skb->users, 0); 3148 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3149 return; 3150 } 3151 get_kfree_skb_cb(skb)->reason = reason; 3152 local_irq_save(flags); 3153 skb->next = __this_cpu_read(softnet_data.completion_queue); 3154 __this_cpu_write(softnet_data.completion_queue, skb); 3155 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3156 local_irq_restore(flags); 3157 } 3158 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3159 3160 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3161 { 3162 if (in_hardirq() || irqs_disabled()) 3163 dev_kfree_skb_irq_reason(skb, reason); 3164 else 3165 kfree_skb_reason(skb, reason); 3166 } 3167 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3168 3169 3170 /** 3171 * netif_device_detach - mark device as removed 3172 * @dev: network device 3173 * 3174 * Mark device as removed from system and therefore no longer available. 3175 */ 3176 void netif_device_detach(struct net_device *dev) 3177 { 3178 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3179 netif_running(dev)) { 3180 netif_tx_stop_all_queues(dev); 3181 } 3182 } 3183 EXPORT_SYMBOL(netif_device_detach); 3184 3185 /** 3186 * netif_device_attach - mark device as attached 3187 * @dev: network device 3188 * 3189 * Mark device as attached from system and restart if needed. 3190 */ 3191 void netif_device_attach(struct net_device *dev) 3192 { 3193 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3194 netif_running(dev)) { 3195 netif_tx_wake_all_queues(dev); 3196 __netdev_watchdog_up(dev); 3197 } 3198 } 3199 EXPORT_SYMBOL(netif_device_attach); 3200 3201 /* 3202 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3203 * to be used as a distribution range. 3204 */ 3205 static u16 skb_tx_hash(const struct net_device *dev, 3206 const struct net_device *sb_dev, 3207 struct sk_buff *skb) 3208 { 3209 u32 hash; 3210 u16 qoffset = 0; 3211 u16 qcount = dev->real_num_tx_queues; 3212 3213 if (dev->num_tc) { 3214 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3215 3216 qoffset = sb_dev->tc_to_txq[tc].offset; 3217 qcount = sb_dev->tc_to_txq[tc].count; 3218 if (unlikely(!qcount)) { 3219 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3220 sb_dev->name, qoffset, tc); 3221 qoffset = 0; 3222 qcount = dev->real_num_tx_queues; 3223 } 3224 } 3225 3226 if (skb_rx_queue_recorded(skb)) { 3227 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3228 hash = skb_get_rx_queue(skb); 3229 if (hash >= qoffset) 3230 hash -= qoffset; 3231 while (unlikely(hash >= qcount)) 3232 hash -= qcount; 3233 return hash + qoffset; 3234 } 3235 3236 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3237 } 3238 3239 void skb_warn_bad_offload(const struct sk_buff *skb) 3240 { 3241 static const netdev_features_t null_features; 3242 struct net_device *dev = skb->dev; 3243 const char *name = ""; 3244 3245 if (!net_ratelimit()) 3246 return; 3247 3248 if (dev) { 3249 if (dev->dev.parent) 3250 name = dev_driver_string(dev->dev.parent); 3251 else 3252 name = netdev_name(dev); 3253 } 3254 skb_dump(KERN_WARNING, skb, false); 3255 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3256 name, dev ? &dev->features : &null_features, 3257 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3258 } 3259 3260 /* 3261 * Invalidate hardware checksum when packet is to be mangled, and 3262 * complete checksum manually on outgoing path. 3263 */ 3264 int skb_checksum_help(struct sk_buff *skb) 3265 { 3266 __wsum csum; 3267 int ret = 0, offset; 3268 3269 if (skb->ip_summed == CHECKSUM_COMPLETE) 3270 goto out_set_summed; 3271 3272 if (unlikely(skb_is_gso(skb))) { 3273 skb_warn_bad_offload(skb); 3274 return -EINVAL; 3275 } 3276 3277 /* Before computing a checksum, we should make sure no frag could 3278 * be modified by an external entity : checksum could be wrong. 3279 */ 3280 if (skb_has_shared_frag(skb)) { 3281 ret = __skb_linearize(skb); 3282 if (ret) 3283 goto out; 3284 } 3285 3286 offset = skb_checksum_start_offset(skb); 3287 ret = -EINVAL; 3288 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3289 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3290 goto out; 3291 } 3292 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3293 3294 offset += skb->csum_offset; 3295 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3296 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3297 goto out; 3298 } 3299 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3300 if (ret) 3301 goto out; 3302 3303 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3304 out_set_summed: 3305 skb->ip_summed = CHECKSUM_NONE; 3306 out: 3307 return ret; 3308 } 3309 EXPORT_SYMBOL(skb_checksum_help); 3310 3311 int skb_crc32c_csum_help(struct sk_buff *skb) 3312 { 3313 __le32 crc32c_csum; 3314 int ret = 0, offset, start; 3315 3316 if (skb->ip_summed != CHECKSUM_PARTIAL) 3317 goto out; 3318 3319 if (unlikely(skb_is_gso(skb))) 3320 goto out; 3321 3322 /* Before computing a checksum, we should make sure no frag could 3323 * be modified by an external entity : checksum could be wrong. 3324 */ 3325 if (unlikely(skb_has_shared_frag(skb))) { 3326 ret = __skb_linearize(skb); 3327 if (ret) 3328 goto out; 3329 } 3330 start = skb_checksum_start_offset(skb); 3331 offset = start + offsetof(struct sctphdr, checksum); 3332 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3333 ret = -EINVAL; 3334 goto out; 3335 } 3336 3337 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3338 if (ret) 3339 goto out; 3340 3341 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3342 skb->len - start, ~(__u32)0, 3343 crc32c_csum_stub)); 3344 *(__le32 *)(skb->data + offset) = crc32c_csum; 3345 skb_reset_csum_not_inet(skb); 3346 out: 3347 return ret; 3348 } 3349 3350 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3351 { 3352 __be16 type = skb->protocol; 3353 3354 /* Tunnel gso handlers can set protocol to ethernet. */ 3355 if (type == htons(ETH_P_TEB)) { 3356 struct ethhdr *eth; 3357 3358 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3359 return 0; 3360 3361 eth = (struct ethhdr *)skb->data; 3362 type = eth->h_proto; 3363 } 3364 3365 return vlan_get_protocol_and_depth(skb, type, depth); 3366 } 3367 3368 3369 /* Take action when hardware reception checksum errors are detected. */ 3370 #ifdef CONFIG_BUG 3371 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3372 { 3373 netdev_err(dev, "hw csum failure\n"); 3374 skb_dump(KERN_ERR, skb, true); 3375 dump_stack(); 3376 } 3377 3378 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3379 { 3380 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3381 } 3382 EXPORT_SYMBOL(netdev_rx_csum_fault); 3383 #endif 3384 3385 /* XXX: check that highmem exists at all on the given machine. */ 3386 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3387 { 3388 #ifdef CONFIG_HIGHMEM 3389 int i; 3390 3391 if (!(dev->features & NETIF_F_HIGHDMA)) { 3392 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3393 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3394 3395 if (PageHighMem(skb_frag_page(frag))) 3396 return 1; 3397 } 3398 } 3399 #endif 3400 return 0; 3401 } 3402 3403 /* If MPLS offload request, verify we are testing hardware MPLS features 3404 * instead of standard features for the netdev. 3405 */ 3406 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3407 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3408 netdev_features_t features, 3409 __be16 type) 3410 { 3411 if (eth_p_mpls(type)) 3412 features &= skb->dev->mpls_features; 3413 3414 return features; 3415 } 3416 #else 3417 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3418 netdev_features_t features, 3419 __be16 type) 3420 { 3421 return features; 3422 } 3423 #endif 3424 3425 static netdev_features_t harmonize_features(struct sk_buff *skb, 3426 netdev_features_t features) 3427 { 3428 __be16 type; 3429 3430 type = skb_network_protocol(skb, NULL); 3431 features = net_mpls_features(skb, features, type); 3432 3433 if (skb->ip_summed != CHECKSUM_NONE && 3434 !can_checksum_protocol(features, type)) { 3435 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3436 } 3437 if (illegal_highdma(skb->dev, skb)) 3438 features &= ~NETIF_F_SG; 3439 3440 return features; 3441 } 3442 3443 netdev_features_t passthru_features_check(struct sk_buff *skb, 3444 struct net_device *dev, 3445 netdev_features_t features) 3446 { 3447 return features; 3448 } 3449 EXPORT_SYMBOL(passthru_features_check); 3450 3451 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3452 struct net_device *dev, 3453 netdev_features_t features) 3454 { 3455 return vlan_features_check(skb, features); 3456 } 3457 3458 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3459 struct net_device *dev, 3460 netdev_features_t features) 3461 { 3462 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3463 3464 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3465 return features & ~NETIF_F_GSO_MASK; 3466 3467 if (!skb_shinfo(skb)->gso_type) { 3468 skb_warn_bad_offload(skb); 3469 return features & ~NETIF_F_GSO_MASK; 3470 } 3471 3472 /* Support for GSO partial features requires software 3473 * intervention before we can actually process the packets 3474 * so we need to strip support for any partial features now 3475 * and we can pull them back in after we have partially 3476 * segmented the frame. 3477 */ 3478 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3479 features &= ~dev->gso_partial_features; 3480 3481 /* Make sure to clear the IPv4 ID mangling feature if the 3482 * IPv4 header has the potential to be fragmented. 3483 */ 3484 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3485 struct iphdr *iph = skb->encapsulation ? 3486 inner_ip_hdr(skb) : ip_hdr(skb); 3487 3488 if (!(iph->frag_off & htons(IP_DF))) 3489 features &= ~NETIF_F_TSO_MANGLEID; 3490 } 3491 3492 return features; 3493 } 3494 3495 netdev_features_t netif_skb_features(struct sk_buff *skb) 3496 { 3497 struct net_device *dev = skb->dev; 3498 netdev_features_t features = dev->features; 3499 3500 if (skb_is_gso(skb)) 3501 features = gso_features_check(skb, dev, features); 3502 3503 /* If encapsulation offload request, verify we are testing 3504 * hardware encapsulation features instead of standard 3505 * features for the netdev 3506 */ 3507 if (skb->encapsulation) 3508 features &= dev->hw_enc_features; 3509 3510 if (skb_vlan_tagged(skb)) 3511 features = netdev_intersect_features(features, 3512 dev->vlan_features | 3513 NETIF_F_HW_VLAN_CTAG_TX | 3514 NETIF_F_HW_VLAN_STAG_TX); 3515 3516 if (dev->netdev_ops->ndo_features_check) 3517 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3518 features); 3519 else 3520 features &= dflt_features_check(skb, dev, features); 3521 3522 return harmonize_features(skb, features); 3523 } 3524 EXPORT_SYMBOL(netif_skb_features); 3525 3526 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3527 struct netdev_queue *txq, bool more) 3528 { 3529 unsigned int len; 3530 int rc; 3531 3532 if (dev_nit_active(dev)) 3533 dev_queue_xmit_nit(skb, dev); 3534 3535 len = skb->len; 3536 trace_net_dev_start_xmit(skb, dev); 3537 rc = netdev_start_xmit(skb, dev, txq, more); 3538 trace_net_dev_xmit(skb, rc, dev, len); 3539 3540 return rc; 3541 } 3542 3543 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3544 struct netdev_queue *txq, int *ret) 3545 { 3546 struct sk_buff *skb = first; 3547 int rc = NETDEV_TX_OK; 3548 3549 while (skb) { 3550 struct sk_buff *next = skb->next; 3551 3552 skb_mark_not_on_list(skb); 3553 rc = xmit_one(skb, dev, txq, next != NULL); 3554 if (unlikely(!dev_xmit_complete(rc))) { 3555 skb->next = next; 3556 goto out; 3557 } 3558 3559 skb = next; 3560 if (netif_tx_queue_stopped(txq) && skb) { 3561 rc = NETDEV_TX_BUSY; 3562 break; 3563 } 3564 } 3565 3566 out: 3567 *ret = rc; 3568 return skb; 3569 } 3570 3571 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3572 netdev_features_t features) 3573 { 3574 if (skb_vlan_tag_present(skb) && 3575 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3576 skb = __vlan_hwaccel_push_inside(skb); 3577 return skb; 3578 } 3579 3580 int skb_csum_hwoffload_help(struct sk_buff *skb, 3581 const netdev_features_t features) 3582 { 3583 if (unlikely(skb_csum_is_sctp(skb))) 3584 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3585 skb_crc32c_csum_help(skb); 3586 3587 if (features & NETIF_F_HW_CSUM) 3588 return 0; 3589 3590 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3591 switch (skb->csum_offset) { 3592 case offsetof(struct tcphdr, check): 3593 case offsetof(struct udphdr, check): 3594 return 0; 3595 } 3596 } 3597 3598 return skb_checksum_help(skb); 3599 } 3600 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3601 3602 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3603 { 3604 netdev_features_t features; 3605 3606 features = netif_skb_features(skb); 3607 skb = validate_xmit_vlan(skb, features); 3608 if (unlikely(!skb)) 3609 goto out_null; 3610 3611 skb = sk_validate_xmit_skb(skb, dev); 3612 if (unlikely(!skb)) 3613 goto out_null; 3614 3615 if (netif_needs_gso(skb, features)) { 3616 struct sk_buff *segs; 3617 3618 segs = skb_gso_segment(skb, features); 3619 if (IS_ERR(segs)) { 3620 goto out_kfree_skb; 3621 } else if (segs) { 3622 consume_skb(skb); 3623 skb = segs; 3624 } 3625 } else { 3626 if (skb_needs_linearize(skb, features) && 3627 __skb_linearize(skb)) 3628 goto out_kfree_skb; 3629 3630 /* If packet is not checksummed and device does not 3631 * support checksumming for this protocol, complete 3632 * checksumming here. 3633 */ 3634 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3635 if (skb->encapsulation) 3636 skb_set_inner_transport_header(skb, 3637 skb_checksum_start_offset(skb)); 3638 else 3639 skb_set_transport_header(skb, 3640 skb_checksum_start_offset(skb)); 3641 if (skb_csum_hwoffload_help(skb, features)) 3642 goto out_kfree_skb; 3643 } 3644 } 3645 3646 skb = validate_xmit_xfrm(skb, features, again); 3647 3648 return skb; 3649 3650 out_kfree_skb: 3651 kfree_skb(skb); 3652 out_null: 3653 dev_core_stats_tx_dropped_inc(dev); 3654 return NULL; 3655 } 3656 3657 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3658 { 3659 struct sk_buff *next, *head = NULL, *tail; 3660 3661 for (; skb != NULL; skb = next) { 3662 next = skb->next; 3663 skb_mark_not_on_list(skb); 3664 3665 /* in case skb wont be segmented, point to itself */ 3666 skb->prev = skb; 3667 3668 skb = validate_xmit_skb(skb, dev, again); 3669 if (!skb) 3670 continue; 3671 3672 if (!head) 3673 head = skb; 3674 else 3675 tail->next = skb; 3676 /* If skb was segmented, skb->prev points to 3677 * the last segment. If not, it still contains skb. 3678 */ 3679 tail = skb->prev; 3680 } 3681 return head; 3682 } 3683 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3684 3685 static void qdisc_pkt_len_init(struct sk_buff *skb) 3686 { 3687 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3688 3689 qdisc_skb_cb(skb)->pkt_len = skb->len; 3690 3691 /* To get more precise estimation of bytes sent on wire, 3692 * we add to pkt_len the headers size of all segments 3693 */ 3694 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3695 u16 gso_segs = shinfo->gso_segs; 3696 unsigned int hdr_len; 3697 3698 /* mac layer + network layer */ 3699 hdr_len = skb_transport_offset(skb); 3700 3701 /* + transport layer */ 3702 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3703 const struct tcphdr *th; 3704 struct tcphdr _tcphdr; 3705 3706 th = skb_header_pointer(skb, hdr_len, 3707 sizeof(_tcphdr), &_tcphdr); 3708 if (likely(th)) 3709 hdr_len += __tcp_hdrlen(th); 3710 } else { 3711 struct udphdr _udphdr; 3712 3713 if (skb_header_pointer(skb, hdr_len, 3714 sizeof(_udphdr), &_udphdr)) 3715 hdr_len += sizeof(struct udphdr); 3716 } 3717 3718 if (shinfo->gso_type & SKB_GSO_DODGY) 3719 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3720 shinfo->gso_size); 3721 3722 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3723 } 3724 } 3725 3726 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3727 struct sk_buff **to_free, 3728 struct netdev_queue *txq) 3729 { 3730 int rc; 3731 3732 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3733 if (rc == NET_XMIT_SUCCESS) 3734 trace_qdisc_enqueue(q, txq, skb); 3735 return rc; 3736 } 3737 3738 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3739 struct net_device *dev, 3740 struct netdev_queue *txq) 3741 { 3742 spinlock_t *root_lock = qdisc_lock(q); 3743 struct sk_buff *to_free = NULL; 3744 bool contended; 3745 int rc; 3746 3747 qdisc_calculate_pkt_len(skb, q); 3748 3749 if (q->flags & TCQ_F_NOLOCK) { 3750 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3751 qdisc_run_begin(q)) { 3752 /* Retest nolock_qdisc_is_empty() within the protection 3753 * of q->seqlock to protect from racing with requeuing. 3754 */ 3755 if (unlikely(!nolock_qdisc_is_empty(q))) { 3756 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3757 __qdisc_run(q); 3758 qdisc_run_end(q); 3759 3760 goto no_lock_out; 3761 } 3762 3763 qdisc_bstats_cpu_update(q, skb); 3764 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3765 !nolock_qdisc_is_empty(q)) 3766 __qdisc_run(q); 3767 3768 qdisc_run_end(q); 3769 return NET_XMIT_SUCCESS; 3770 } 3771 3772 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3773 qdisc_run(q); 3774 3775 no_lock_out: 3776 if (unlikely(to_free)) 3777 kfree_skb_list_reason(to_free, 3778 SKB_DROP_REASON_QDISC_DROP); 3779 return rc; 3780 } 3781 3782 /* 3783 * Heuristic to force contended enqueues to serialize on a 3784 * separate lock before trying to get qdisc main lock. 3785 * This permits qdisc->running owner to get the lock more 3786 * often and dequeue packets faster. 3787 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3788 * and then other tasks will only enqueue packets. The packets will be 3789 * sent after the qdisc owner is scheduled again. To prevent this 3790 * scenario the task always serialize on the lock. 3791 */ 3792 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3793 if (unlikely(contended)) 3794 spin_lock(&q->busylock); 3795 3796 spin_lock(root_lock); 3797 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3798 __qdisc_drop(skb, &to_free); 3799 rc = NET_XMIT_DROP; 3800 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3801 qdisc_run_begin(q)) { 3802 /* 3803 * This is a work-conserving queue; there are no old skbs 3804 * waiting to be sent out; and the qdisc is not running - 3805 * xmit the skb directly. 3806 */ 3807 3808 qdisc_bstats_update(q, skb); 3809 3810 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3811 if (unlikely(contended)) { 3812 spin_unlock(&q->busylock); 3813 contended = false; 3814 } 3815 __qdisc_run(q); 3816 } 3817 3818 qdisc_run_end(q); 3819 rc = NET_XMIT_SUCCESS; 3820 } else { 3821 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3822 if (qdisc_run_begin(q)) { 3823 if (unlikely(contended)) { 3824 spin_unlock(&q->busylock); 3825 contended = false; 3826 } 3827 __qdisc_run(q); 3828 qdisc_run_end(q); 3829 } 3830 } 3831 spin_unlock(root_lock); 3832 if (unlikely(to_free)) 3833 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3834 if (unlikely(contended)) 3835 spin_unlock(&q->busylock); 3836 return rc; 3837 } 3838 3839 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3840 static void skb_update_prio(struct sk_buff *skb) 3841 { 3842 const struct netprio_map *map; 3843 const struct sock *sk; 3844 unsigned int prioidx; 3845 3846 if (skb->priority) 3847 return; 3848 map = rcu_dereference_bh(skb->dev->priomap); 3849 if (!map) 3850 return; 3851 sk = skb_to_full_sk(skb); 3852 if (!sk) 3853 return; 3854 3855 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3856 3857 if (prioidx < map->priomap_len) 3858 skb->priority = map->priomap[prioidx]; 3859 } 3860 #else 3861 #define skb_update_prio(skb) 3862 #endif 3863 3864 /** 3865 * dev_loopback_xmit - loop back @skb 3866 * @net: network namespace this loopback is happening in 3867 * @sk: sk needed to be a netfilter okfn 3868 * @skb: buffer to transmit 3869 */ 3870 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3871 { 3872 skb_reset_mac_header(skb); 3873 __skb_pull(skb, skb_network_offset(skb)); 3874 skb->pkt_type = PACKET_LOOPBACK; 3875 if (skb->ip_summed == CHECKSUM_NONE) 3876 skb->ip_summed = CHECKSUM_UNNECESSARY; 3877 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3878 skb_dst_force(skb); 3879 netif_rx(skb); 3880 return 0; 3881 } 3882 EXPORT_SYMBOL(dev_loopback_xmit); 3883 3884 #ifdef CONFIG_NET_EGRESS 3885 static struct sk_buff * 3886 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3887 { 3888 #ifdef CONFIG_NET_CLS_ACT 3889 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3890 struct tcf_result cl_res; 3891 3892 if (!miniq) 3893 return skb; 3894 3895 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3896 tc_skb_cb(skb)->mru = 0; 3897 tc_skb_cb(skb)->post_ct = false; 3898 mini_qdisc_bstats_cpu_update(miniq, skb); 3899 3900 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3901 case TC_ACT_OK: 3902 case TC_ACT_RECLASSIFY: 3903 skb->tc_index = TC_H_MIN(cl_res.classid); 3904 break; 3905 case TC_ACT_SHOT: 3906 mini_qdisc_qstats_cpu_drop(miniq); 3907 *ret = NET_XMIT_DROP; 3908 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3909 return NULL; 3910 case TC_ACT_STOLEN: 3911 case TC_ACT_QUEUED: 3912 case TC_ACT_TRAP: 3913 *ret = NET_XMIT_SUCCESS; 3914 consume_skb(skb); 3915 return NULL; 3916 case TC_ACT_REDIRECT: 3917 /* No need to push/pop skb's mac_header here on egress! */ 3918 skb_do_redirect(skb); 3919 *ret = NET_XMIT_SUCCESS; 3920 return NULL; 3921 default: 3922 break; 3923 } 3924 #endif /* CONFIG_NET_CLS_ACT */ 3925 3926 return skb; 3927 } 3928 3929 static struct netdev_queue * 3930 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3931 { 3932 int qm = skb_get_queue_mapping(skb); 3933 3934 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3935 } 3936 3937 static bool netdev_xmit_txqueue_skipped(void) 3938 { 3939 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3940 } 3941 3942 void netdev_xmit_skip_txqueue(bool skip) 3943 { 3944 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3945 } 3946 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3947 #endif /* CONFIG_NET_EGRESS */ 3948 3949 #ifdef CONFIG_XPS 3950 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3951 struct xps_dev_maps *dev_maps, unsigned int tci) 3952 { 3953 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3954 struct xps_map *map; 3955 int queue_index = -1; 3956 3957 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3958 return queue_index; 3959 3960 tci *= dev_maps->num_tc; 3961 tci += tc; 3962 3963 map = rcu_dereference(dev_maps->attr_map[tci]); 3964 if (map) { 3965 if (map->len == 1) 3966 queue_index = map->queues[0]; 3967 else 3968 queue_index = map->queues[reciprocal_scale( 3969 skb_get_hash(skb), map->len)]; 3970 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3971 queue_index = -1; 3972 } 3973 return queue_index; 3974 } 3975 #endif 3976 3977 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3978 struct sk_buff *skb) 3979 { 3980 #ifdef CONFIG_XPS 3981 struct xps_dev_maps *dev_maps; 3982 struct sock *sk = skb->sk; 3983 int queue_index = -1; 3984 3985 if (!static_key_false(&xps_needed)) 3986 return -1; 3987 3988 rcu_read_lock(); 3989 if (!static_key_false(&xps_rxqs_needed)) 3990 goto get_cpus_map; 3991 3992 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3993 if (dev_maps) { 3994 int tci = sk_rx_queue_get(sk); 3995 3996 if (tci >= 0) 3997 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3998 tci); 3999 } 4000 4001 get_cpus_map: 4002 if (queue_index < 0) { 4003 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4004 if (dev_maps) { 4005 unsigned int tci = skb->sender_cpu - 1; 4006 4007 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4008 tci); 4009 } 4010 } 4011 rcu_read_unlock(); 4012 4013 return queue_index; 4014 #else 4015 return -1; 4016 #endif 4017 } 4018 4019 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4020 struct net_device *sb_dev) 4021 { 4022 return 0; 4023 } 4024 EXPORT_SYMBOL(dev_pick_tx_zero); 4025 4026 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4027 struct net_device *sb_dev) 4028 { 4029 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4030 } 4031 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4032 4033 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4034 struct net_device *sb_dev) 4035 { 4036 struct sock *sk = skb->sk; 4037 int queue_index = sk_tx_queue_get(sk); 4038 4039 sb_dev = sb_dev ? : dev; 4040 4041 if (queue_index < 0 || skb->ooo_okay || 4042 queue_index >= dev->real_num_tx_queues) { 4043 int new_index = get_xps_queue(dev, sb_dev, skb); 4044 4045 if (new_index < 0) 4046 new_index = skb_tx_hash(dev, sb_dev, skb); 4047 4048 if (queue_index != new_index && sk && 4049 sk_fullsock(sk) && 4050 rcu_access_pointer(sk->sk_dst_cache)) 4051 sk_tx_queue_set(sk, new_index); 4052 4053 queue_index = new_index; 4054 } 4055 4056 return queue_index; 4057 } 4058 EXPORT_SYMBOL(netdev_pick_tx); 4059 4060 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4061 struct sk_buff *skb, 4062 struct net_device *sb_dev) 4063 { 4064 int queue_index = 0; 4065 4066 #ifdef CONFIG_XPS 4067 u32 sender_cpu = skb->sender_cpu - 1; 4068 4069 if (sender_cpu >= (u32)NR_CPUS) 4070 skb->sender_cpu = raw_smp_processor_id() + 1; 4071 #endif 4072 4073 if (dev->real_num_tx_queues != 1) { 4074 const struct net_device_ops *ops = dev->netdev_ops; 4075 4076 if (ops->ndo_select_queue) 4077 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4078 else 4079 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4080 4081 queue_index = netdev_cap_txqueue(dev, queue_index); 4082 } 4083 4084 skb_set_queue_mapping(skb, queue_index); 4085 return netdev_get_tx_queue(dev, queue_index); 4086 } 4087 4088 /** 4089 * __dev_queue_xmit() - transmit a buffer 4090 * @skb: buffer to transmit 4091 * @sb_dev: suboordinate device used for L2 forwarding offload 4092 * 4093 * Queue a buffer for transmission to a network device. The caller must 4094 * have set the device and priority and built the buffer before calling 4095 * this function. The function can be called from an interrupt. 4096 * 4097 * When calling this method, interrupts MUST be enabled. This is because 4098 * the BH enable code must have IRQs enabled so that it will not deadlock. 4099 * 4100 * Regardless of the return value, the skb is consumed, so it is currently 4101 * difficult to retry a send to this method. (You can bump the ref count 4102 * before sending to hold a reference for retry if you are careful.) 4103 * 4104 * Return: 4105 * * 0 - buffer successfully transmitted 4106 * * positive qdisc return code - NET_XMIT_DROP etc. 4107 * * negative errno - other errors 4108 */ 4109 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4110 { 4111 struct net_device *dev = skb->dev; 4112 struct netdev_queue *txq = NULL; 4113 struct Qdisc *q; 4114 int rc = -ENOMEM; 4115 bool again = false; 4116 4117 skb_reset_mac_header(skb); 4118 skb_assert_len(skb); 4119 4120 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4121 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4122 4123 /* Disable soft irqs for various locks below. Also 4124 * stops preemption for RCU. 4125 */ 4126 rcu_read_lock_bh(); 4127 4128 skb_update_prio(skb); 4129 4130 qdisc_pkt_len_init(skb); 4131 #ifdef CONFIG_NET_CLS_ACT 4132 skb->tc_at_ingress = 0; 4133 #endif 4134 #ifdef CONFIG_NET_EGRESS 4135 if (static_branch_unlikely(&egress_needed_key)) { 4136 if (nf_hook_egress_active()) { 4137 skb = nf_hook_egress(skb, &rc, dev); 4138 if (!skb) 4139 goto out; 4140 } 4141 4142 netdev_xmit_skip_txqueue(false); 4143 4144 nf_skip_egress(skb, true); 4145 skb = sch_handle_egress(skb, &rc, dev); 4146 if (!skb) 4147 goto out; 4148 nf_skip_egress(skb, false); 4149 4150 if (netdev_xmit_txqueue_skipped()) 4151 txq = netdev_tx_queue_mapping(dev, skb); 4152 } 4153 #endif 4154 /* If device/qdisc don't need skb->dst, release it right now while 4155 * its hot in this cpu cache. 4156 */ 4157 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4158 skb_dst_drop(skb); 4159 else 4160 skb_dst_force(skb); 4161 4162 if (!txq) 4163 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4164 4165 q = rcu_dereference_bh(txq->qdisc); 4166 4167 trace_net_dev_queue(skb); 4168 if (q->enqueue) { 4169 rc = __dev_xmit_skb(skb, q, dev, txq); 4170 goto out; 4171 } 4172 4173 /* The device has no queue. Common case for software devices: 4174 * loopback, all the sorts of tunnels... 4175 4176 * Really, it is unlikely that netif_tx_lock protection is necessary 4177 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4178 * counters.) 4179 * However, it is possible, that they rely on protection 4180 * made by us here. 4181 4182 * Check this and shot the lock. It is not prone from deadlocks. 4183 *Either shot noqueue qdisc, it is even simpler 8) 4184 */ 4185 if (dev->flags & IFF_UP) { 4186 int cpu = smp_processor_id(); /* ok because BHs are off */ 4187 4188 /* Other cpus might concurrently change txq->xmit_lock_owner 4189 * to -1 or to their cpu id, but not to our id. 4190 */ 4191 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4192 if (dev_xmit_recursion()) 4193 goto recursion_alert; 4194 4195 skb = validate_xmit_skb(skb, dev, &again); 4196 if (!skb) 4197 goto out; 4198 4199 HARD_TX_LOCK(dev, txq, cpu); 4200 4201 if (!netif_xmit_stopped(txq)) { 4202 dev_xmit_recursion_inc(); 4203 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4204 dev_xmit_recursion_dec(); 4205 if (dev_xmit_complete(rc)) { 4206 HARD_TX_UNLOCK(dev, txq); 4207 goto out; 4208 } 4209 } 4210 HARD_TX_UNLOCK(dev, txq); 4211 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4212 dev->name); 4213 } else { 4214 /* Recursion is detected! It is possible, 4215 * unfortunately 4216 */ 4217 recursion_alert: 4218 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4219 dev->name); 4220 } 4221 } 4222 4223 rc = -ENETDOWN; 4224 rcu_read_unlock_bh(); 4225 4226 dev_core_stats_tx_dropped_inc(dev); 4227 kfree_skb_list(skb); 4228 return rc; 4229 out: 4230 rcu_read_unlock_bh(); 4231 return rc; 4232 } 4233 EXPORT_SYMBOL(__dev_queue_xmit); 4234 4235 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4236 { 4237 struct net_device *dev = skb->dev; 4238 struct sk_buff *orig_skb = skb; 4239 struct netdev_queue *txq; 4240 int ret = NETDEV_TX_BUSY; 4241 bool again = false; 4242 4243 if (unlikely(!netif_running(dev) || 4244 !netif_carrier_ok(dev))) 4245 goto drop; 4246 4247 skb = validate_xmit_skb_list(skb, dev, &again); 4248 if (skb != orig_skb) 4249 goto drop; 4250 4251 skb_set_queue_mapping(skb, queue_id); 4252 txq = skb_get_tx_queue(dev, skb); 4253 4254 local_bh_disable(); 4255 4256 dev_xmit_recursion_inc(); 4257 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4258 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4259 ret = netdev_start_xmit(skb, dev, txq, false); 4260 HARD_TX_UNLOCK(dev, txq); 4261 dev_xmit_recursion_dec(); 4262 4263 local_bh_enable(); 4264 return ret; 4265 drop: 4266 dev_core_stats_tx_dropped_inc(dev); 4267 kfree_skb_list(skb); 4268 return NET_XMIT_DROP; 4269 } 4270 EXPORT_SYMBOL(__dev_direct_xmit); 4271 4272 /************************************************************************* 4273 * Receiver routines 4274 *************************************************************************/ 4275 4276 int netdev_max_backlog __read_mostly = 1000; 4277 EXPORT_SYMBOL(netdev_max_backlog); 4278 4279 int netdev_tstamp_prequeue __read_mostly = 1; 4280 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4281 int netdev_budget __read_mostly = 300; 4282 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4283 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4284 int weight_p __read_mostly = 64; /* old backlog weight */ 4285 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4286 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4287 int dev_rx_weight __read_mostly = 64; 4288 int dev_tx_weight __read_mostly = 64; 4289 4290 /* Called with irq disabled */ 4291 static inline void ____napi_schedule(struct softnet_data *sd, 4292 struct napi_struct *napi) 4293 { 4294 struct task_struct *thread; 4295 4296 lockdep_assert_irqs_disabled(); 4297 4298 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4299 /* Paired with smp_mb__before_atomic() in 4300 * napi_enable()/dev_set_threaded(). 4301 * Use READ_ONCE() to guarantee a complete 4302 * read on napi->thread. Only call 4303 * wake_up_process() when it's not NULL. 4304 */ 4305 thread = READ_ONCE(napi->thread); 4306 if (thread) { 4307 /* Avoid doing set_bit() if the thread is in 4308 * INTERRUPTIBLE state, cause napi_thread_wait() 4309 * makes sure to proceed with napi polling 4310 * if the thread is explicitly woken from here. 4311 */ 4312 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4313 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4314 wake_up_process(thread); 4315 return; 4316 } 4317 } 4318 4319 list_add_tail(&napi->poll_list, &sd->poll_list); 4320 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4321 /* If not called from net_rx_action() 4322 * we have to raise NET_RX_SOFTIRQ. 4323 */ 4324 if (!sd->in_net_rx_action) 4325 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4326 } 4327 4328 #ifdef CONFIG_RPS 4329 4330 /* One global table that all flow-based protocols share. */ 4331 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4332 EXPORT_SYMBOL(rps_sock_flow_table); 4333 u32 rps_cpu_mask __read_mostly; 4334 EXPORT_SYMBOL(rps_cpu_mask); 4335 4336 struct static_key_false rps_needed __read_mostly; 4337 EXPORT_SYMBOL(rps_needed); 4338 struct static_key_false rfs_needed __read_mostly; 4339 EXPORT_SYMBOL(rfs_needed); 4340 4341 static struct rps_dev_flow * 4342 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4343 struct rps_dev_flow *rflow, u16 next_cpu) 4344 { 4345 if (next_cpu < nr_cpu_ids) { 4346 #ifdef CONFIG_RFS_ACCEL 4347 struct netdev_rx_queue *rxqueue; 4348 struct rps_dev_flow_table *flow_table; 4349 struct rps_dev_flow *old_rflow; 4350 u32 flow_id; 4351 u16 rxq_index; 4352 int rc; 4353 4354 /* Should we steer this flow to a different hardware queue? */ 4355 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4356 !(dev->features & NETIF_F_NTUPLE)) 4357 goto out; 4358 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4359 if (rxq_index == skb_get_rx_queue(skb)) 4360 goto out; 4361 4362 rxqueue = dev->_rx + rxq_index; 4363 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4364 if (!flow_table) 4365 goto out; 4366 flow_id = skb_get_hash(skb) & flow_table->mask; 4367 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4368 rxq_index, flow_id); 4369 if (rc < 0) 4370 goto out; 4371 old_rflow = rflow; 4372 rflow = &flow_table->flows[flow_id]; 4373 rflow->filter = rc; 4374 if (old_rflow->filter == rflow->filter) 4375 old_rflow->filter = RPS_NO_FILTER; 4376 out: 4377 #endif 4378 rflow->last_qtail = 4379 per_cpu(softnet_data, next_cpu).input_queue_head; 4380 } 4381 4382 rflow->cpu = next_cpu; 4383 return rflow; 4384 } 4385 4386 /* 4387 * get_rps_cpu is called from netif_receive_skb and returns the target 4388 * CPU from the RPS map of the receiving queue for a given skb. 4389 * rcu_read_lock must be held on entry. 4390 */ 4391 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4392 struct rps_dev_flow **rflowp) 4393 { 4394 const struct rps_sock_flow_table *sock_flow_table; 4395 struct netdev_rx_queue *rxqueue = dev->_rx; 4396 struct rps_dev_flow_table *flow_table; 4397 struct rps_map *map; 4398 int cpu = -1; 4399 u32 tcpu; 4400 u32 hash; 4401 4402 if (skb_rx_queue_recorded(skb)) { 4403 u16 index = skb_get_rx_queue(skb); 4404 4405 if (unlikely(index >= dev->real_num_rx_queues)) { 4406 WARN_ONCE(dev->real_num_rx_queues > 1, 4407 "%s received packet on queue %u, but number " 4408 "of RX queues is %u\n", 4409 dev->name, index, dev->real_num_rx_queues); 4410 goto done; 4411 } 4412 rxqueue += index; 4413 } 4414 4415 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4416 4417 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4418 map = rcu_dereference(rxqueue->rps_map); 4419 if (!flow_table && !map) 4420 goto done; 4421 4422 skb_reset_network_header(skb); 4423 hash = skb_get_hash(skb); 4424 if (!hash) 4425 goto done; 4426 4427 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4428 if (flow_table && sock_flow_table) { 4429 struct rps_dev_flow *rflow; 4430 u32 next_cpu; 4431 u32 ident; 4432 4433 /* First check into global flow table if there is a match. 4434 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4435 */ 4436 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4437 if ((ident ^ hash) & ~rps_cpu_mask) 4438 goto try_rps; 4439 4440 next_cpu = ident & rps_cpu_mask; 4441 4442 /* OK, now we know there is a match, 4443 * we can look at the local (per receive queue) flow table 4444 */ 4445 rflow = &flow_table->flows[hash & flow_table->mask]; 4446 tcpu = rflow->cpu; 4447 4448 /* 4449 * If the desired CPU (where last recvmsg was done) is 4450 * different from current CPU (one in the rx-queue flow 4451 * table entry), switch if one of the following holds: 4452 * - Current CPU is unset (>= nr_cpu_ids). 4453 * - Current CPU is offline. 4454 * - The current CPU's queue tail has advanced beyond the 4455 * last packet that was enqueued using this table entry. 4456 * This guarantees that all previous packets for the flow 4457 * have been dequeued, thus preserving in order delivery. 4458 */ 4459 if (unlikely(tcpu != next_cpu) && 4460 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4461 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4462 rflow->last_qtail)) >= 0)) { 4463 tcpu = next_cpu; 4464 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4465 } 4466 4467 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4468 *rflowp = rflow; 4469 cpu = tcpu; 4470 goto done; 4471 } 4472 } 4473 4474 try_rps: 4475 4476 if (map) { 4477 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4478 if (cpu_online(tcpu)) { 4479 cpu = tcpu; 4480 goto done; 4481 } 4482 } 4483 4484 done: 4485 return cpu; 4486 } 4487 4488 #ifdef CONFIG_RFS_ACCEL 4489 4490 /** 4491 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4492 * @dev: Device on which the filter was set 4493 * @rxq_index: RX queue index 4494 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4495 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4496 * 4497 * Drivers that implement ndo_rx_flow_steer() should periodically call 4498 * this function for each installed filter and remove the filters for 4499 * which it returns %true. 4500 */ 4501 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4502 u32 flow_id, u16 filter_id) 4503 { 4504 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4505 struct rps_dev_flow_table *flow_table; 4506 struct rps_dev_flow *rflow; 4507 bool expire = true; 4508 unsigned int cpu; 4509 4510 rcu_read_lock(); 4511 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4512 if (flow_table && flow_id <= flow_table->mask) { 4513 rflow = &flow_table->flows[flow_id]; 4514 cpu = READ_ONCE(rflow->cpu); 4515 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4516 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4517 rflow->last_qtail) < 4518 (int)(10 * flow_table->mask))) 4519 expire = false; 4520 } 4521 rcu_read_unlock(); 4522 return expire; 4523 } 4524 EXPORT_SYMBOL(rps_may_expire_flow); 4525 4526 #endif /* CONFIG_RFS_ACCEL */ 4527 4528 /* Called from hardirq (IPI) context */ 4529 static void rps_trigger_softirq(void *data) 4530 { 4531 struct softnet_data *sd = data; 4532 4533 ____napi_schedule(sd, &sd->backlog); 4534 sd->received_rps++; 4535 } 4536 4537 #endif /* CONFIG_RPS */ 4538 4539 /* Called from hardirq (IPI) context */ 4540 static void trigger_rx_softirq(void *data) 4541 { 4542 struct softnet_data *sd = data; 4543 4544 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4545 smp_store_release(&sd->defer_ipi_scheduled, 0); 4546 } 4547 4548 /* 4549 * After we queued a packet into sd->input_pkt_queue, 4550 * we need to make sure this queue is serviced soon. 4551 * 4552 * - If this is another cpu queue, link it to our rps_ipi_list, 4553 * and make sure we will process rps_ipi_list from net_rx_action(). 4554 * 4555 * - If this is our own queue, NAPI schedule our backlog. 4556 * Note that this also raises NET_RX_SOFTIRQ. 4557 */ 4558 static void napi_schedule_rps(struct softnet_data *sd) 4559 { 4560 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4561 4562 #ifdef CONFIG_RPS 4563 if (sd != mysd) { 4564 sd->rps_ipi_next = mysd->rps_ipi_list; 4565 mysd->rps_ipi_list = sd; 4566 4567 /* If not called from net_rx_action() or napi_threaded_poll() 4568 * we have to raise NET_RX_SOFTIRQ. 4569 */ 4570 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4571 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4572 return; 4573 } 4574 #endif /* CONFIG_RPS */ 4575 __napi_schedule_irqoff(&mysd->backlog); 4576 } 4577 4578 #ifdef CONFIG_NET_FLOW_LIMIT 4579 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4580 #endif 4581 4582 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4583 { 4584 #ifdef CONFIG_NET_FLOW_LIMIT 4585 struct sd_flow_limit *fl; 4586 struct softnet_data *sd; 4587 unsigned int old_flow, new_flow; 4588 4589 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4590 return false; 4591 4592 sd = this_cpu_ptr(&softnet_data); 4593 4594 rcu_read_lock(); 4595 fl = rcu_dereference(sd->flow_limit); 4596 if (fl) { 4597 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4598 old_flow = fl->history[fl->history_head]; 4599 fl->history[fl->history_head] = new_flow; 4600 4601 fl->history_head++; 4602 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4603 4604 if (likely(fl->buckets[old_flow])) 4605 fl->buckets[old_flow]--; 4606 4607 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4608 fl->count++; 4609 rcu_read_unlock(); 4610 return true; 4611 } 4612 } 4613 rcu_read_unlock(); 4614 #endif 4615 return false; 4616 } 4617 4618 /* 4619 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4620 * queue (may be a remote CPU queue). 4621 */ 4622 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4623 unsigned int *qtail) 4624 { 4625 enum skb_drop_reason reason; 4626 struct softnet_data *sd; 4627 unsigned long flags; 4628 unsigned int qlen; 4629 4630 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4631 sd = &per_cpu(softnet_data, cpu); 4632 4633 rps_lock_irqsave(sd, &flags); 4634 if (!netif_running(skb->dev)) 4635 goto drop; 4636 qlen = skb_queue_len(&sd->input_pkt_queue); 4637 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4638 if (qlen) { 4639 enqueue: 4640 __skb_queue_tail(&sd->input_pkt_queue, skb); 4641 input_queue_tail_incr_save(sd, qtail); 4642 rps_unlock_irq_restore(sd, &flags); 4643 return NET_RX_SUCCESS; 4644 } 4645 4646 /* Schedule NAPI for backlog device 4647 * We can use non atomic operation since we own the queue lock 4648 */ 4649 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4650 napi_schedule_rps(sd); 4651 goto enqueue; 4652 } 4653 reason = SKB_DROP_REASON_CPU_BACKLOG; 4654 4655 drop: 4656 sd->dropped++; 4657 rps_unlock_irq_restore(sd, &flags); 4658 4659 dev_core_stats_rx_dropped_inc(skb->dev); 4660 kfree_skb_reason(skb, reason); 4661 return NET_RX_DROP; 4662 } 4663 4664 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4665 { 4666 struct net_device *dev = skb->dev; 4667 struct netdev_rx_queue *rxqueue; 4668 4669 rxqueue = dev->_rx; 4670 4671 if (skb_rx_queue_recorded(skb)) { 4672 u16 index = skb_get_rx_queue(skb); 4673 4674 if (unlikely(index >= dev->real_num_rx_queues)) { 4675 WARN_ONCE(dev->real_num_rx_queues > 1, 4676 "%s received packet on queue %u, but number " 4677 "of RX queues is %u\n", 4678 dev->name, index, dev->real_num_rx_queues); 4679 4680 return rxqueue; /* Return first rxqueue */ 4681 } 4682 rxqueue += index; 4683 } 4684 return rxqueue; 4685 } 4686 4687 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4688 struct bpf_prog *xdp_prog) 4689 { 4690 void *orig_data, *orig_data_end, *hard_start; 4691 struct netdev_rx_queue *rxqueue; 4692 bool orig_bcast, orig_host; 4693 u32 mac_len, frame_sz; 4694 __be16 orig_eth_type; 4695 struct ethhdr *eth; 4696 u32 metalen, act; 4697 int off; 4698 4699 /* The XDP program wants to see the packet starting at the MAC 4700 * header. 4701 */ 4702 mac_len = skb->data - skb_mac_header(skb); 4703 hard_start = skb->data - skb_headroom(skb); 4704 4705 /* SKB "head" area always have tailroom for skb_shared_info */ 4706 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4707 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4708 4709 rxqueue = netif_get_rxqueue(skb); 4710 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4711 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4712 skb_headlen(skb) + mac_len, true); 4713 4714 orig_data_end = xdp->data_end; 4715 orig_data = xdp->data; 4716 eth = (struct ethhdr *)xdp->data; 4717 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4718 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4719 orig_eth_type = eth->h_proto; 4720 4721 act = bpf_prog_run_xdp(xdp_prog, xdp); 4722 4723 /* check if bpf_xdp_adjust_head was used */ 4724 off = xdp->data - orig_data; 4725 if (off) { 4726 if (off > 0) 4727 __skb_pull(skb, off); 4728 else if (off < 0) 4729 __skb_push(skb, -off); 4730 4731 skb->mac_header += off; 4732 skb_reset_network_header(skb); 4733 } 4734 4735 /* check if bpf_xdp_adjust_tail was used */ 4736 off = xdp->data_end - orig_data_end; 4737 if (off != 0) { 4738 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4739 skb->len += off; /* positive on grow, negative on shrink */ 4740 } 4741 4742 /* check if XDP changed eth hdr such SKB needs update */ 4743 eth = (struct ethhdr *)xdp->data; 4744 if ((orig_eth_type != eth->h_proto) || 4745 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4746 skb->dev->dev_addr)) || 4747 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4748 __skb_push(skb, ETH_HLEN); 4749 skb->pkt_type = PACKET_HOST; 4750 skb->protocol = eth_type_trans(skb, skb->dev); 4751 } 4752 4753 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4754 * before calling us again on redirect path. We do not call do_redirect 4755 * as we leave that up to the caller. 4756 * 4757 * Caller is responsible for managing lifetime of skb (i.e. calling 4758 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4759 */ 4760 switch (act) { 4761 case XDP_REDIRECT: 4762 case XDP_TX: 4763 __skb_push(skb, mac_len); 4764 break; 4765 case XDP_PASS: 4766 metalen = xdp->data - xdp->data_meta; 4767 if (metalen) 4768 skb_metadata_set(skb, metalen); 4769 break; 4770 } 4771 4772 return act; 4773 } 4774 4775 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4776 struct xdp_buff *xdp, 4777 struct bpf_prog *xdp_prog) 4778 { 4779 u32 act = XDP_DROP; 4780 4781 /* Reinjected packets coming from act_mirred or similar should 4782 * not get XDP generic processing. 4783 */ 4784 if (skb_is_redirected(skb)) 4785 return XDP_PASS; 4786 4787 /* XDP packets must be linear and must have sufficient headroom 4788 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4789 * native XDP provides, thus we need to do it here as well. 4790 */ 4791 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4792 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4793 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4794 int troom = skb->tail + skb->data_len - skb->end; 4795 4796 /* In case we have to go down the path and also linearize, 4797 * then lets do the pskb_expand_head() work just once here. 4798 */ 4799 if (pskb_expand_head(skb, 4800 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4801 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4802 goto do_drop; 4803 if (skb_linearize(skb)) 4804 goto do_drop; 4805 } 4806 4807 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4808 switch (act) { 4809 case XDP_REDIRECT: 4810 case XDP_TX: 4811 case XDP_PASS: 4812 break; 4813 default: 4814 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4815 fallthrough; 4816 case XDP_ABORTED: 4817 trace_xdp_exception(skb->dev, xdp_prog, act); 4818 fallthrough; 4819 case XDP_DROP: 4820 do_drop: 4821 kfree_skb(skb); 4822 break; 4823 } 4824 4825 return act; 4826 } 4827 4828 /* When doing generic XDP we have to bypass the qdisc layer and the 4829 * network taps in order to match in-driver-XDP behavior. This also means 4830 * that XDP packets are able to starve other packets going through a qdisc, 4831 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4832 * queues, so they do not have this starvation issue. 4833 */ 4834 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4835 { 4836 struct net_device *dev = skb->dev; 4837 struct netdev_queue *txq; 4838 bool free_skb = true; 4839 int cpu, rc; 4840 4841 txq = netdev_core_pick_tx(dev, skb, NULL); 4842 cpu = smp_processor_id(); 4843 HARD_TX_LOCK(dev, txq, cpu); 4844 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4845 rc = netdev_start_xmit(skb, dev, txq, 0); 4846 if (dev_xmit_complete(rc)) 4847 free_skb = false; 4848 } 4849 HARD_TX_UNLOCK(dev, txq); 4850 if (free_skb) { 4851 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4852 dev_core_stats_tx_dropped_inc(dev); 4853 kfree_skb(skb); 4854 } 4855 } 4856 4857 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4858 4859 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4860 { 4861 if (xdp_prog) { 4862 struct xdp_buff xdp; 4863 u32 act; 4864 int err; 4865 4866 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4867 if (act != XDP_PASS) { 4868 switch (act) { 4869 case XDP_REDIRECT: 4870 err = xdp_do_generic_redirect(skb->dev, skb, 4871 &xdp, xdp_prog); 4872 if (err) 4873 goto out_redir; 4874 break; 4875 case XDP_TX: 4876 generic_xdp_tx(skb, xdp_prog); 4877 break; 4878 } 4879 return XDP_DROP; 4880 } 4881 } 4882 return XDP_PASS; 4883 out_redir: 4884 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4885 return XDP_DROP; 4886 } 4887 EXPORT_SYMBOL_GPL(do_xdp_generic); 4888 4889 static int netif_rx_internal(struct sk_buff *skb) 4890 { 4891 int ret; 4892 4893 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4894 4895 trace_netif_rx(skb); 4896 4897 #ifdef CONFIG_RPS 4898 if (static_branch_unlikely(&rps_needed)) { 4899 struct rps_dev_flow voidflow, *rflow = &voidflow; 4900 int cpu; 4901 4902 rcu_read_lock(); 4903 4904 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4905 if (cpu < 0) 4906 cpu = smp_processor_id(); 4907 4908 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4909 4910 rcu_read_unlock(); 4911 } else 4912 #endif 4913 { 4914 unsigned int qtail; 4915 4916 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4917 } 4918 return ret; 4919 } 4920 4921 /** 4922 * __netif_rx - Slightly optimized version of netif_rx 4923 * @skb: buffer to post 4924 * 4925 * This behaves as netif_rx except that it does not disable bottom halves. 4926 * As a result this function may only be invoked from the interrupt context 4927 * (either hard or soft interrupt). 4928 */ 4929 int __netif_rx(struct sk_buff *skb) 4930 { 4931 int ret; 4932 4933 lockdep_assert_once(hardirq_count() | softirq_count()); 4934 4935 trace_netif_rx_entry(skb); 4936 ret = netif_rx_internal(skb); 4937 trace_netif_rx_exit(ret); 4938 return ret; 4939 } 4940 EXPORT_SYMBOL(__netif_rx); 4941 4942 /** 4943 * netif_rx - post buffer to the network code 4944 * @skb: buffer to post 4945 * 4946 * This function receives a packet from a device driver and queues it for 4947 * the upper (protocol) levels to process via the backlog NAPI device. It 4948 * always succeeds. The buffer may be dropped during processing for 4949 * congestion control or by the protocol layers. 4950 * The network buffer is passed via the backlog NAPI device. Modern NIC 4951 * driver should use NAPI and GRO. 4952 * This function can used from interrupt and from process context. The 4953 * caller from process context must not disable interrupts before invoking 4954 * this function. 4955 * 4956 * return values: 4957 * NET_RX_SUCCESS (no congestion) 4958 * NET_RX_DROP (packet was dropped) 4959 * 4960 */ 4961 int netif_rx(struct sk_buff *skb) 4962 { 4963 bool need_bh_off = !(hardirq_count() | softirq_count()); 4964 int ret; 4965 4966 if (need_bh_off) 4967 local_bh_disable(); 4968 trace_netif_rx_entry(skb); 4969 ret = netif_rx_internal(skb); 4970 trace_netif_rx_exit(ret); 4971 if (need_bh_off) 4972 local_bh_enable(); 4973 return ret; 4974 } 4975 EXPORT_SYMBOL(netif_rx); 4976 4977 static __latent_entropy void net_tx_action(struct softirq_action *h) 4978 { 4979 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4980 4981 if (sd->completion_queue) { 4982 struct sk_buff *clist; 4983 4984 local_irq_disable(); 4985 clist = sd->completion_queue; 4986 sd->completion_queue = NULL; 4987 local_irq_enable(); 4988 4989 while (clist) { 4990 struct sk_buff *skb = clist; 4991 4992 clist = clist->next; 4993 4994 WARN_ON(refcount_read(&skb->users)); 4995 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 4996 trace_consume_skb(skb, net_tx_action); 4997 else 4998 trace_kfree_skb(skb, net_tx_action, 4999 get_kfree_skb_cb(skb)->reason); 5000 5001 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5002 __kfree_skb(skb); 5003 else 5004 __napi_kfree_skb(skb, 5005 get_kfree_skb_cb(skb)->reason); 5006 } 5007 } 5008 5009 if (sd->output_queue) { 5010 struct Qdisc *head; 5011 5012 local_irq_disable(); 5013 head = sd->output_queue; 5014 sd->output_queue = NULL; 5015 sd->output_queue_tailp = &sd->output_queue; 5016 local_irq_enable(); 5017 5018 rcu_read_lock(); 5019 5020 while (head) { 5021 struct Qdisc *q = head; 5022 spinlock_t *root_lock = NULL; 5023 5024 head = head->next_sched; 5025 5026 /* We need to make sure head->next_sched is read 5027 * before clearing __QDISC_STATE_SCHED 5028 */ 5029 smp_mb__before_atomic(); 5030 5031 if (!(q->flags & TCQ_F_NOLOCK)) { 5032 root_lock = qdisc_lock(q); 5033 spin_lock(root_lock); 5034 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5035 &q->state))) { 5036 /* There is a synchronize_net() between 5037 * STATE_DEACTIVATED flag being set and 5038 * qdisc_reset()/some_qdisc_is_busy() in 5039 * dev_deactivate(), so we can safely bail out 5040 * early here to avoid data race between 5041 * qdisc_deactivate() and some_qdisc_is_busy() 5042 * for lockless qdisc. 5043 */ 5044 clear_bit(__QDISC_STATE_SCHED, &q->state); 5045 continue; 5046 } 5047 5048 clear_bit(__QDISC_STATE_SCHED, &q->state); 5049 qdisc_run(q); 5050 if (root_lock) 5051 spin_unlock(root_lock); 5052 } 5053 5054 rcu_read_unlock(); 5055 } 5056 5057 xfrm_dev_backlog(sd); 5058 } 5059 5060 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5061 /* This hook is defined here for ATM LANE */ 5062 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5063 unsigned char *addr) __read_mostly; 5064 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5065 #endif 5066 5067 static inline struct sk_buff * 5068 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5069 struct net_device *orig_dev, bool *another) 5070 { 5071 #ifdef CONFIG_NET_CLS_ACT 5072 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5073 struct tcf_result cl_res; 5074 5075 /* If there's at least one ingress present somewhere (so 5076 * we get here via enabled static key), remaining devices 5077 * that are not configured with an ingress qdisc will bail 5078 * out here. 5079 */ 5080 if (!miniq) 5081 return skb; 5082 5083 if (*pt_prev) { 5084 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5085 *pt_prev = NULL; 5086 } 5087 5088 qdisc_skb_cb(skb)->pkt_len = skb->len; 5089 tc_skb_cb(skb)->mru = 0; 5090 tc_skb_cb(skb)->post_ct = false; 5091 skb->tc_at_ingress = 1; 5092 mini_qdisc_bstats_cpu_update(miniq, skb); 5093 5094 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5095 case TC_ACT_OK: 5096 case TC_ACT_RECLASSIFY: 5097 skb->tc_index = TC_H_MIN(cl_res.classid); 5098 break; 5099 case TC_ACT_SHOT: 5100 mini_qdisc_qstats_cpu_drop(miniq); 5101 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5102 *ret = NET_RX_DROP; 5103 return NULL; 5104 case TC_ACT_STOLEN: 5105 case TC_ACT_QUEUED: 5106 case TC_ACT_TRAP: 5107 consume_skb(skb); 5108 *ret = NET_RX_SUCCESS; 5109 return NULL; 5110 case TC_ACT_REDIRECT: 5111 /* skb_mac_header check was done by cls/act_bpf, so 5112 * we can safely push the L2 header back before 5113 * redirecting to another netdev 5114 */ 5115 __skb_push(skb, skb->mac_len); 5116 if (skb_do_redirect(skb) == -EAGAIN) { 5117 __skb_pull(skb, skb->mac_len); 5118 *another = true; 5119 break; 5120 } 5121 *ret = NET_RX_SUCCESS; 5122 return NULL; 5123 case TC_ACT_CONSUMED: 5124 *ret = NET_RX_SUCCESS; 5125 return NULL; 5126 default: 5127 break; 5128 } 5129 #endif /* CONFIG_NET_CLS_ACT */ 5130 return skb; 5131 } 5132 5133 /** 5134 * netdev_is_rx_handler_busy - check if receive handler is registered 5135 * @dev: device to check 5136 * 5137 * Check if a receive handler is already registered for a given device. 5138 * Return true if there one. 5139 * 5140 * The caller must hold the rtnl_mutex. 5141 */ 5142 bool netdev_is_rx_handler_busy(struct net_device *dev) 5143 { 5144 ASSERT_RTNL(); 5145 return dev && rtnl_dereference(dev->rx_handler); 5146 } 5147 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5148 5149 /** 5150 * netdev_rx_handler_register - register receive handler 5151 * @dev: device to register a handler for 5152 * @rx_handler: receive handler to register 5153 * @rx_handler_data: data pointer that is used by rx handler 5154 * 5155 * Register a receive handler for a device. This handler will then be 5156 * called from __netif_receive_skb. A negative errno code is returned 5157 * on a failure. 5158 * 5159 * The caller must hold the rtnl_mutex. 5160 * 5161 * For a general description of rx_handler, see enum rx_handler_result. 5162 */ 5163 int netdev_rx_handler_register(struct net_device *dev, 5164 rx_handler_func_t *rx_handler, 5165 void *rx_handler_data) 5166 { 5167 if (netdev_is_rx_handler_busy(dev)) 5168 return -EBUSY; 5169 5170 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5171 return -EINVAL; 5172 5173 /* Note: rx_handler_data must be set before rx_handler */ 5174 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5175 rcu_assign_pointer(dev->rx_handler, rx_handler); 5176 5177 return 0; 5178 } 5179 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5180 5181 /** 5182 * netdev_rx_handler_unregister - unregister receive handler 5183 * @dev: device to unregister a handler from 5184 * 5185 * Unregister a receive handler from a device. 5186 * 5187 * The caller must hold the rtnl_mutex. 5188 */ 5189 void netdev_rx_handler_unregister(struct net_device *dev) 5190 { 5191 5192 ASSERT_RTNL(); 5193 RCU_INIT_POINTER(dev->rx_handler, NULL); 5194 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5195 * section has a guarantee to see a non NULL rx_handler_data 5196 * as well. 5197 */ 5198 synchronize_net(); 5199 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5200 } 5201 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5202 5203 /* 5204 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5205 * the special handling of PFMEMALLOC skbs. 5206 */ 5207 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5208 { 5209 switch (skb->protocol) { 5210 case htons(ETH_P_ARP): 5211 case htons(ETH_P_IP): 5212 case htons(ETH_P_IPV6): 5213 case htons(ETH_P_8021Q): 5214 case htons(ETH_P_8021AD): 5215 return true; 5216 default: 5217 return false; 5218 } 5219 } 5220 5221 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5222 int *ret, struct net_device *orig_dev) 5223 { 5224 if (nf_hook_ingress_active(skb)) { 5225 int ingress_retval; 5226 5227 if (*pt_prev) { 5228 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5229 *pt_prev = NULL; 5230 } 5231 5232 rcu_read_lock(); 5233 ingress_retval = nf_hook_ingress(skb); 5234 rcu_read_unlock(); 5235 return ingress_retval; 5236 } 5237 return 0; 5238 } 5239 5240 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5241 struct packet_type **ppt_prev) 5242 { 5243 struct packet_type *ptype, *pt_prev; 5244 rx_handler_func_t *rx_handler; 5245 struct sk_buff *skb = *pskb; 5246 struct net_device *orig_dev; 5247 bool deliver_exact = false; 5248 int ret = NET_RX_DROP; 5249 __be16 type; 5250 5251 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5252 5253 trace_netif_receive_skb(skb); 5254 5255 orig_dev = skb->dev; 5256 5257 skb_reset_network_header(skb); 5258 if (!skb_transport_header_was_set(skb)) 5259 skb_reset_transport_header(skb); 5260 skb_reset_mac_len(skb); 5261 5262 pt_prev = NULL; 5263 5264 another_round: 5265 skb->skb_iif = skb->dev->ifindex; 5266 5267 __this_cpu_inc(softnet_data.processed); 5268 5269 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5270 int ret2; 5271 5272 migrate_disable(); 5273 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5274 migrate_enable(); 5275 5276 if (ret2 != XDP_PASS) { 5277 ret = NET_RX_DROP; 5278 goto out; 5279 } 5280 } 5281 5282 if (eth_type_vlan(skb->protocol)) { 5283 skb = skb_vlan_untag(skb); 5284 if (unlikely(!skb)) 5285 goto out; 5286 } 5287 5288 if (skb_skip_tc_classify(skb)) 5289 goto skip_classify; 5290 5291 if (pfmemalloc) 5292 goto skip_taps; 5293 5294 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5295 if (pt_prev) 5296 ret = deliver_skb(skb, pt_prev, orig_dev); 5297 pt_prev = ptype; 5298 } 5299 5300 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5301 if (pt_prev) 5302 ret = deliver_skb(skb, pt_prev, orig_dev); 5303 pt_prev = ptype; 5304 } 5305 5306 skip_taps: 5307 #ifdef CONFIG_NET_INGRESS 5308 if (static_branch_unlikely(&ingress_needed_key)) { 5309 bool another = false; 5310 5311 nf_skip_egress(skb, true); 5312 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5313 &another); 5314 if (another) 5315 goto another_round; 5316 if (!skb) 5317 goto out; 5318 5319 nf_skip_egress(skb, false); 5320 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5321 goto out; 5322 } 5323 #endif 5324 skb_reset_redirect(skb); 5325 skip_classify: 5326 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5327 goto drop; 5328 5329 if (skb_vlan_tag_present(skb)) { 5330 if (pt_prev) { 5331 ret = deliver_skb(skb, pt_prev, orig_dev); 5332 pt_prev = NULL; 5333 } 5334 if (vlan_do_receive(&skb)) 5335 goto another_round; 5336 else if (unlikely(!skb)) 5337 goto out; 5338 } 5339 5340 rx_handler = rcu_dereference(skb->dev->rx_handler); 5341 if (rx_handler) { 5342 if (pt_prev) { 5343 ret = deliver_skb(skb, pt_prev, orig_dev); 5344 pt_prev = NULL; 5345 } 5346 switch (rx_handler(&skb)) { 5347 case RX_HANDLER_CONSUMED: 5348 ret = NET_RX_SUCCESS; 5349 goto out; 5350 case RX_HANDLER_ANOTHER: 5351 goto another_round; 5352 case RX_HANDLER_EXACT: 5353 deliver_exact = true; 5354 break; 5355 case RX_HANDLER_PASS: 5356 break; 5357 default: 5358 BUG(); 5359 } 5360 } 5361 5362 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5363 check_vlan_id: 5364 if (skb_vlan_tag_get_id(skb)) { 5365 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5366 * find vlan device. 5367 */ 5368 skb->pkt_type = PACKET_OTHERHOST; 5369 } else if (eth_type_vlan(skb->protocol)) { 5370 /* Outer header is 802.1P with vlan 0, inner header is 5371 * 802.1Q or 802.1AD and vlan_do_receive() above could 5372 * not find vlan dev for vlan id 0. 5373 */ 5374 __vlan_hwaccel_clear_tag(skb); 5375 skb = skb_vlan_untag(skb); 5376 if (unlikely(!skb)) 5377 goto out; 5378 if (vlan_do_receive(&skb)) 5379 /* After stripping off 802.1P header with vlan 0 5380 * vlan dev is found for inner header. 5381 */ 5382 goto another_round; 5383 else if (unlikely(!skb)) 5384 goto out; 5385 else 5386 /* We have stripped outer 802.1P vlan 0 header. 5387 * But could not find vlan dev. 5388 * check again for vlan id to set OTHERHOST. 5389 */ 5390 goto check_vlan_id; 5391 } 5392 /* Note: we might in the future use prio bits 5393 * and set skb->priority like in vlan_do_receive() 5394 * For the time being, just ignore Priority Code Point 5395 */ 5396 __vlan_hwaccel_clear_tag(skb); 5397 } 5398 5399 type = skb->protocol; 5400 5401 /* deliver only exact match when indicated */ 5402 if (likely(!deliver_exact)) { 5403 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5404 &ptype_base[ntohs(type) & 5405 PTYPE_HASH_MASK]); 5406 } 5407 5408 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5409 &orig_dev->ptype_specific); 5410 5411 if (unlikely(skb->dev != orig_dev)) { 5412 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5413 &skb->dev->ptype_specific); 5414 } 5415 5416 if (pt_prev) { 5417 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5418 goto drop; 5419 *ppt_prev = pt_prev; 5420 } else { 5421 drop: 5422 if (!deliver_exact) 5423 dev_core_stats_rx_dropped_inc(skb->dev); 5424 else 5425 dev_core_stats_rx_nohandler_inc(skb->dev); 5426 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5427 /* Jamal, now you will not able to escape explaining 5428 * me how you were going to use this. :-) 5429 */ 5430 ret = NET_RX_DROP; 5431 } 5432 5433 out: 5434 /* The invariant here is that if *ppt_prev is not NULL 5435 * then skb should also be non-NULL. 5436 * 5437 * Apparently *ppt_prev assignment above holds this invariant due to 5438 * skb dereferencing near it. 5439 */ 5440 *pskb = skb; 5441 return ret; 5442 } 5443 5444 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5445 { 5446 struct net_device *orig_dev = skb->dev; 5447 struct packet_type *pt_prev = NULL; 5448 int ret; 5449 5450 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5451 if (pt_prev) 5452 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5453 skb->dev, pt_prev, orig_dev); 5454 return ret; 5455 } 5456 5457 /** 5458 * netif_receive_skb_core - special purpose version of netif_receive_skb 5459 * @skb: buffer to process 5460 * 5461 * More direct receive version of netif_receive_skb(). It should 5462 * only be used by callers that have a need to skip RPS and Generic XDP. 5463 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5464 * 5465 * This function may only be called from softirq context and interrupts 5466 * should be enabled. 5467 * 5468 * Return values (usually ignored): 5469 * NET_RX_SUCCESS: no congestion 5470 * NET_RX_DROP: packet was dropped 5471 */ 5472 int netif_receive_skb_core(struct sk_buff *skb) 5473 { 5474 int ret; 5475 5476 rcu_read_lock(); 5477 ret = __netif_receive_skb_one_core(skb, false); 5478 rcu_read_unlock(); 5479 5480 return ret; 5481 } 5482 EXPORT_SYMBOL(netif_receive_skb_core); 5483 5484 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5485 struct packet_type *pt_prev, 5486 struct net_device *orig_dev) 5487 { 5488 struct sk_buff *skb, *next; 5489 5490 if (!pt_prev) 5491 return; 5492 if (list_empty(head)) 5493 return; 5494 if (pt_prev->list_func != NULL) 5495 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5496 ip_list_rcv, head, pt_prev, orig_dev); 5497 else 5498 list_for_each_entry_safe(skb, next, head, list) { 5499 skb_list_del_init(skb); 5500 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5501 } 5502 } 5503 5504 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5505 { 5506 /* Fast-path assumptions: 5507 * - There is no RX handler. 5508 * - Only one packet_type matches. 5509 * If either of these fails, we will end up doing some per-packet 5510 * processing in-line, then handling the 'last ptype' for the whole 5511 * sublist. This can't cause out-of-order delivery to any single ptype, 5512 * because the 'last ptype' must be constant across the sublist, and all 5513 * other ptypes are handled per-packet. 5514 */ 5515 /* Current (common) ptype of sublist */ 5516 struct packet_type *pt_curr = NULL; 5517 /* Current (common) orig_dev of sublist */ 5518 struct net_device *od_curr = NULL; 5519 struct list_head sublist; 5520 struct sk_buff *skb, *next; 5521 5522 INIT_LIST_HEAD(&sublist); 5523 list_for_each_entry_safe(skb, next, head, list) { 5524 struct net_device *orig_dev = skb->dev; 5525 struct packet_type *pt_prev = NULL; 5526 5527 skb_list_del_init(skb); 5528 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5529 if (!pt_prev) 5530 continue; 5531 if (pt_curr != pt_prev || od_curr != orig_dev) { 5532 /* dispatch old sublist */ 5533 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5534 /* start new sublist */ 5535 INIT_LIST_HEAD(&sublist); 5536 pt_curr = pt_prev; 5537 od_curr = orig_dev; 5538 } 5539 list_add_tail(&skb->list, &sublist); 5540 } 5541 5542 /* dispatch final sublist */ 5543 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5544 } 5545 5546 static int __netif_receive_skb(struct sk_buff *skb) 5547 { 5548 int ret; 5549 5550 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5551 unsigned int noreclaim_flag; 5552 5553 /* 5554 * PFMEMALLOC skbs are special, they should 5555 * - be delivered to SOCK_MEMALLOC sockets only 5556 * - stay away from userspace 5557 * - have bounded memory usage 5558 * 5559 * Use PF_MEMALLOC as this saves us from propagating the allocation 5560 * context down to all allocation sites. 5561 */ 5562 noreclaim_flag = memalloc_noreclaim_save(); 5563 ret = __netif_receive_skb_one_core(skb, true); 5564 memalloc_noreclaim_restore(noreclaim_flag); 5565 } else 5566 ret = __netif_receive_skb_one_core(skb, false); 5567 5568 return ret; 5569 } 5570 5571 static void __netif_receive_skb_list(struct list_head *head) 5572 { 5573 unsigned long noreclaim_flag = 0; 5574 struct sk_buff *skb, *next; 5575 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5576 5577 list_for_each_entry_safe(skb, next, head, list) { 5578 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5579 struct list_head sublist; 5580 5581 /* Handle the previous sublist */ 5582 list_cut_before(&sublist, head, &skb->list); 5583 if (!list_empty(&sublist)) 5584 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5585 pfmemalloc = !pfmemalloc; 5586 /* See comments in __netif_receive_skb */ 5587 if (pfmemalloc) 5588 noreclaim_flag = memalloc_noreclaim_save(); 5589 else 5590 memalloc_noreclaim_restore(noreclaim_flag); 5591 } 5592 } 5593 /* Handle the remaining sublist */ 5594 if (!list_empty(head)) 5595 __netif_receive_skb_list_core(head, pfmemalloc); 5596 /* Restore pflags */ 5597 if (pfmemalloc) 5598 memalloc_noreclaim_restore(noreclaim_flag); 5599 } 5600 5601 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5602 { 5603 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5604 struct bpf_prog *new = xdp->prog; 5605 int ret = 0; 5606 5607 switch (xdp->command) { 5608 case XDP_SETUP_PROG: 5609 rcu_assign_pointer(dev->xdp_prog, new); 5610 if (old) 5611 bpf_prog_put(old); 5612 5613 if (old && !new) { 5614 static_branch_dec(&generic_xdp_needed_key); 5615 } else if (new && !old) { 5616 static_branch_inc(&generic_xdp_needed_key); 5617 dev_disable_lro(dev); 5618 dev_disable_gro_hw(dev); 5619 } 5620 break; 5621 5622 default: 5623 ret = -EINVAL; 5624 break; 5625 } 5626 5627 return ret; 5628 } 5629 5630 static int netif_receive_skb_internal(struct sk_buff *skb) 5631 { 5632 int ret; 5633 5634 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5635 5636 if (skb_defer_rx_timestamp(skb)) 5637 return NET_RX_SUCCESS; 5638 5639 rcu_read_lock(); 5640 #ifdef CONFIG_RPS 5641 if (static_branch_unlikely(&rps_needed)) { 5642 struct rps_dev_flow voidflow, *rflow = &voidflow; 5643 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5644 5645 if (cpu >= 0) { 5646 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5647 rcu_read_unlock(); 5648 return ret; 5649 } 5650 } 5651 #endif 5652 ret = __netif_receive_skb(skb); 5653 rcu_read_unlock(); 5654 return ret; 5655 } 5656 5657 void netif_receive_skb_list_internal(struct list_head *head) 5658 { 5659 struct sk_buff *skb, *next; 5660 struct list_head sublist; 5661 5662 INIT_LIST_HEAD(&sublist); 5663 list_for_each_entry_safe(skb, next, head, list) { 5664 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5665 skb_list_del_init(skb); 5666 if (!skb_defer_rx_timestamp(skb)) 5667 list_add_tail(&skb->list, &sublist); 5668 } 5669 list_splice_init(&sublist, head); 5670 5671 rcu_read_lock(); 5672 #ifdef CONFIG_RPS 5673 if (static_branch_unlikely(&rps_needed)) { 5674 list_for_each_entry_safe(skb, next, head, list) { 5675 struct rps_dev_flow voidflow, *rflow = &voidflow; 5676 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5677 5678 if (cpu >= 0) { 5679 /* Will be handled, remove from list */ 5680 skb_list_del_init(skb); 5681 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5682 } 5683 } 5684 } 5685 #endif 5686 __netif_receive_skb_list(head); 5687 rcu_read_unlock(); 5688 } 5689 5690 /** 5691 * netif_receive_skb - process receive buffer from network 5692 * @skb: buffer to process 5693 * 5694 * netif_receive_skb() is the main receive data processing function. 5695 * It always succeeds. The buffer may be dropped during processing 5696 * for congestion control or by the protocol layers. 5697 * 5698 * This function may only be called from softirq context and interrupts 5699 * should be enabled. 5700 * 5701 * Return values (usually ignored): 5702 * NET_RX_SUCCESS: no congestion 5703 * NET_RX_DROP: packet was dropped 5704 */ 5705 int netif_receive_skb(struct sk_buff *skb) 5706 { 5707 int ret; 5708 5709 trace_netif_receive_skb_entry(skb); 5710 5711 ret = netif_receive_skb_internal(skb); 5712 trace_netif_receive_skb_exit(ret); 5713 5714 return ret; 5715 } 5716 EXPORT_SYMBOL(netif_receive_skb); 5717 5718 /** 5719 * netif_receive_skb_list - process many receive buffers from network 5720 * @head: list of skbs to process. 5721 * 5722 * Since return value of netif_receive_skb() is normally ignored, and 5723 * wouldn't be meaningful for a list, this function returns void. 5724 * 5725 * This function may only be called from softirq context and interrupts 5726 * should be enabled. 5727 */ 5728 void netif_receive_skb_list(struct list_head *head) 5729 { 5730 struct sk_buff *skb; 5731 5732 if (list_empty(head)) 5733 return; 5734 if (trace_netif_receive_skb_list_entry_enabled()) { 5735 list_for_each_entry(skb, head, list) 5736 trace_netif_receive_skb_list_entry(skb); 5737 } 5738 netif_receive_skb_list_internal(head); 5739 trace_netif_receive_skb_list_exit(0); 5740 } 5741 EXPORT_SYMBOL(netif_receive_skb_list); 5742 5743 static DEFINE_PER_CPU(struct work_struct, flush_works); 5744 5745 /* Network device is going away, flush any packets still pending */ 5746 static void flush_backlog(struct work_struct *work) 5747 { 5748 struct sk_buff *skb, *tmp; 5749 struct softnet_data *sd; 5750 5751 local_bh_disable(); 5752 sd = this_cpu_ptr(&softnet_data); 5753 5754 rps_lock_irq_disable(sd); 5755 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5756 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5757 __skb_unlink(skb, &sd->input_pkt_queue); 5758 dev_kfree_skb_irq(skb); 5759 input_queue_head_incr(sd); 5760 } 5761 } 5762 rps_unlock_irq_enable(sd); 5763 5764 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5765 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5766 __skb_unlink(skb, &sd->process_queue); 5767 kfree_skb(skb); 5768 input_queue_head_incr(sd); 5769 } 5770 } 5771 local_bh_enable(); 5772 } 5773 5774 static bool flush_required(int cpu) 5775 { 5776 #if IS_ENABLED(CONFIG_RPS) 5777 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5778 bool do_flush; 5779 5780 rps_lock_irq_disable(sd); 5781 5782 /* as insertion into process_queue happens with the rps lock held, 5783 * process_queue access may race only with dequeue 5784 */ 5785 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5786 !skb_queue_empty_lockless(&sd->process_queue); 5787 rps_unlock_irq_enable(sd); 5788 5789 return do_flush; 5790 #endif 5791 /* without RPS we can't safely check input_pkt_queue: during a 5792 * concurrent remote skb_queue_splice() we can detect as empty both 5793 * input_pkt_queue and process_queue even if the latter could end-up 5794 * containing a lot of packets. 5795 */ 5796 return true; 5797 } 5798 5799 static void flush_all_backlogs(void) 5800 { 5801 static cpumask_t flush_cpus; 5802 unsigned int cpu; 5803 5804 /* since we are under rtnl lock protection we can use static data 5805 * for the cpumask and avoid allocating on stack the possibly 5806 * large mask 5807 */ 5808 ASSERT_RTNL(); 5809 5810 cpus_read_lock(); 5811 5812 cpumask_clear(&flush_cpus); 5813 for_each_online_cpu(cpu) { 5814 if (flush_required(cpu)) { 5815 queue_work_on(cpu, system_highpri_wq, 5816 per_cpu_ptr(&flush_works, cpu)); 5817 cpumask_set_cpu(cpu, &flush_cpus); 5818 } 5819 } 5820 5821 /* we can have in flight packet[s] on the cpus we are not flushing, 5822 * synchronize_net() in unregister_netdevice_many() will take care of 5823 * them 5824 */ 5825 for_each_cpu(cpu, &flush_cpus) 5826 flush_work(per_cpu_ptr(&flush_works, cpu)); 5827 5828 cpus_read_unlock(); 5829 } 5830 5831 static void net_rps_send_ipi(struct softnet_data *remsd) 5832 { 5833 #ifdef CONFIG_RPS 5834 while (remsd) { 5835 struct softnet_data *next = remsd->rps_ipi_next; 5836 5837 if (cpu_online(remsd->cpu)) 5838 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5839 remsd = next; 5840 } 5841 #endif 5842 } 5843 5844 /* 5845 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5846 * Note: called with local irq disabled, but exits with local irq enabled. 5847 */ 5848 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5849 { 5850 #ifdef CONFIG_RPS 5851 struct softnet_data *remsd = sd->rps_ipi_list; 5852 5853 if (remsd) { 5854 sd->rps_ipi_list = NULL; 5855 5856 local_irq_enable(); 5857 5858 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5859 net_rps_send_ipi(remsd); 5860 } else 5861 #endif 5862 local_irq_enable(); 5863 } 5864 5865 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5866 { 5867 #ifdef CONFIG_RPS 5868 return sd->rps_ipi_list != NULL; 5869 #else 5870 return false; 5871 #endif 5872 } 5873 5874 static int process_backlog(struct napi_struct *napi, int quota) 5875 { 5876 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5877 bool again = true; 5878 int work = 0; 5879 5880 /* Check if we have pending ipi, its better to send them now, 5881 * not waiting net_rx_action() end. 5882 */ 5883 if (sd_has_rps_ipi_waiting(sd)) { 5884 local_irq_disable(); 5885 net_rps_action_and_irq_enable(sd); 5886 } 5887 5888 napi->weight = READ_ONCE(dev_rx_weight); 5889 while (again) { 5890 struct sk_buff *skb; 5891 5892 while ((skb = __skb_dequeue(&sd->process_queue))) { 5893 rcu_read_lock(); 5894 __netif_receive_skb(skb); 5895 rcu_read_unlock(); 5896 input_queue_head_incr(sd); 5897 if (++work >= quota) 5898 return work; 5899 5900 } 5901 5902 rps_lock_irq_disable(sd); 5903 if (skb_queue_empty(&sd->input_pkt_queue)) { 5904 /* 5905 * Inline a custom version of __napi_complete(). 5906 * only current cpu owns and manipulates this napi, 5907 * and NAPI_STATE_SCHED is the only possible flag set 5908 * on backlog. 5909 * We can use a plain write instead of clear_bit(), 5910 * and we dont need an smp_mb() memory barrier. 5911 */ 5912 napi->state = 0; 5913 again = false; 5914 } else { 5915 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5916 &sd->process_queue); 5917 } 5918 rps_unlock_irq_enable(sd); 5919 } 5920 5921 return work; 5922 } 5923 5924 /** 5925 * __napi_schedule - schedule for receive 5926 * @n: entry to schedule 5927 * 5928 * The entry's receive function will be scheduled to run. 5929 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5930 */ 5931 void __napi_schedule(struct napi_struct *n) 5932 { 5933 unsigned long flags; 5934 5935 local_irq_save(flags); 5936 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5937 local_irq_restore(flags); 5938 } 5939 EXPORT_SYMBOL(__napi_schedule); 5940 5941 /** 5942 * napi_schedule_prep - check if napi can be scheduled 5943 * @n: napi context 5944 * 5945 * Test if NAPI routine is already running, and if not mark 5946 * it as running. This is used as a condition variable to 5947 * insure only one NAPI poll instance runs. We also make 5948 * sure there is no pending NAPI disable. 5949 */ 5950 bool napi_schedule_prep(struct napi_struct *n) 5951 { 5952 unsigned long new, val = READ_ONCE(n->state); 5953 5954 do { 5955 if (unlikely(val & NAPIF_STATE_DISABLE)) 5956 return false; 5957 new = val | NAPIF_STATE_SCHED; 5958 5959 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5960 * This was suggested by Alexander Duyck, as compiler 5961 * emits better code than : 5962 * if (val & NAPIF_STATE_SCHED) 5963 * new |= NAPIF_STATE_MISSED; 5964 */ 5965 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5966 NAPIF_STATE_MISSED; 5967 } while (!try_cmpxchg(&n->state, &val, new)); 5968 5969 return !(val & NAPIF_STATE_SCHED); 5970 } 5971 EXPORT_SYMBOL(napi_schedule_prep); 5972 5973 /** 5974 * __napi_schedule_irqoff - schedule for receive 5975 * @n: entry to schedule 5976 * 5977 * Variant of __napi_schedule() assuming hard irqs are masked. 5978 * 5979 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5980 * because the interrupt disabled assumption might not be true 5981 * due to force-threaded interrupts and spinlock substitution. 5982 */ 5983 void __napi_schedule_irqoff(struct napi_struct *n) 5984 { 5985 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5986 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5987 else 5988 __napi_schedule(n); 5989 } 5990 EXPORT_SYMBOL(__napi_schedule_irqoff); 5991 5992 bool napi_complete_done(struct napi_struct *n, int work_done) 5993 { 5994 unsigned long flags, val, new, timeout = 0; 5995 bool ret = true; 5996 5997 /* 5998 * 1) Don't let napi dequeue from the cpu poll list 5999 * just in case its running on a different cpu. 6000 * 2) If we are busy polling, do nothing here, we have 6001 * the guarantee we will be called later. 6002 */ 6003 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6004 NAPIF_STATE_IN_BUSY_POLL))) 6005 return false; 6006 6007 if (work_done) { 6008 if (n->gro_bitmask) 6009 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6010 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6011 } 6012 if (n->defer_hard_irqs_count > 0) { 6013 n->defer_hard_irqs_count--; 6014 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6015 if (timeout) 6016 ret = false; 6017 } 6018 if (n->gro_bitmask) { 6019 /* When the NAPI instance uses a timeout and keeps postponing 6020 * it, we need to bound somehow the time packets are kept in 6021 * the GRO layer 6022 */ 6023 napi_gro_flush(n, !!timeout); 6024 } 6025 6026 gro_normal_list(n); 6027 6028 if (unlikely(!list_empty(&n->poll_list))) { 6029 /* If n->poll_list is not empty, we need to mask irqs */ 6030 local_irq_save(flags); 6031 list_del_init(&n->poll_list); 6032 local_irq_restore(flags); 6033 } 6034 WRITE_ONCE(n->list_owner, -1); 6035 6036 val = READ_ONCE(n->state); 6037 do { 6038 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6039 6040 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6041 NAPIF_STATE_SCHED_THREADED | 6042 NAPIF_STATE_PREFER_BUSY_POLL); 6043 6044 /* If STATE_MISSED was set, leave STATE_SCHED set, 6045 * because we will call napi->poll() one more time. 6046 * This C code was suggested by Alexander Duyck to help gcc. 6047 */ 6048 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6049 NAPIF_STATE_SCHED; 6050 } while (!try_cmpxchg(&n->state, &val, new)); 6051 6052 if (unlikely(val & NAPIF_STATE_MISSED)) { 6053 __napi_schedule(n); 6054 return false; 6055 } 6056 6057 if (timeout) 6058 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6059 HRTIMER_MODE_REL_PINNED); 6060 return ret; 6061 } 6062 EXPORT_SYMBOL(napi_complete_done); 6063 6064 /* must be called under rcu_read_lock(), as we dont take a reference */ 6065 static struct napi_struct *napi_by_id(unsigned int napi_id) 6066 { 6067 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6068 struct napi_struct *napi; 6069 6070 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6071 if (napi->napi_id == napi_id) 6072 return napi; 6073 6074 return NULL; 6075 } 6076 6077 #if defined(CONFIG_NET_RX_BUSY_POLL) 6078 6079 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6080 { 6081 if (!skip_schedule) { 6082 gro_normal_list(napi); 6083 __napi_schedule(napi); 6084 return; 6085 } 6086 6087 if (napi->gro_bitmask) { 6088 /* flush too old packets 6089 * If HZ < 1000, flush all packets. 6090 */ 6091 napi_gro_flush(napi, HZ >= 1000); 6092 } 6093 6094 gro_normal_list(napi); 6095 clear_bit(NAPI_STATE_SCHED, &napi->state); 6096 } 6097 6098 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6099 u16 budget) 6100 { 6101 bool skip_schedule = false; 6102 unsigned long timeout; 6103 int rc; 6104 6105 /* Busy polling means there is a high chance device driver hard irq 6106 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6107 * set in napi_schedule_prep(). 6108 * Since we are about to call napi->poll() once more, we can safely 6109 * clear NAPI_STATE_MISSED. 6110 * 6111 * Note: x86 could use a single "lock and ..." instruction 6112 * to perform these two clear_bit() 6113 */ 6114 clear_bit(NAPI_STATE_MISSED, &napi->state); 6115 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6116 6117 local_bh_disable(); 6118 6119 if (prefer_busy_poll) { 6120 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6121 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6122 if (napi->defer_hard_irqs_count && timeout) { 6123 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6124 skip_schedule = true; 6125 } 6126 } 6127 6128 /* All we really want here is to re-enable device interrupts. 6129 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6130 */ 6131 rc = napi->poll(napi, budget); 6132 /* We can't gro_normal_list() here, because napi->poll() might have 6133 * rearmed the napi (napi_complete_done()) in which case it could 6134 * already be running on another CPU. 6135 */ 6136 trace_napi_poll(napi, rc, budget); 6137 netpoll_poll_unlock(have_poll_lock); 6138 if (rc == budget) 6139 __busy_poll_stop(napi, skip_schedule); 6140 local_bh_enable(); 6141 } 6142 6143 void napi_busy_loop(unsigned int napi_id, 6144 bool (*loop_end)(void *, unsigned long), 6145 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6146 { 6147 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6148 int (*napi_poll)(struct napi_struct *napi, int budget); 6149 void *have_poll_lock = NULL; 6150 struct napi_struct *napi; 6151 6152 restart: 6153 napi_poll = NULL; 6154 6155 rcu_read_lock(); 6156 6157 napi = napi_by_id(napi_id); 6158 if (!napi) 6159 goto out; 6160 6161 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6162 preempt_disable(); 6163 for (;;) { 6164 int work = 0; 6165 6166 local_bh_disable(); 6167 if (!napi_poll) { 6168 unsigned long val = READ_ONCE(napi->state); 6169 6170 /* If multiple threads are competing for this napi, 6171 * we avoid dirtying napi->state as much as we can. 6172 */ 6173 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6174 NAPIF_STATE_IN_BUSY_POLL)) { 6175 if (prefer_busy_poll) 6176 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6177 goto count; 6178 } 6179 if (cmpxchg(&napi->state, val, 6180 val | NAPIF_STATE_IN_BUSY_POLL | 6181 NAPIF_STATE_SCHED) != val) { 6182 if (prefer_busy_poll) 6183 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6184 goto count; 6185 } 6186 have_poll_lock = netpoll_poll_lock(napi); 6187 napi_poll = napi->poll; 6188 } 6189 work = napi_poll(napi, budget); 6190 trace_napi_poll(napi, work, budget); 6191 gro_normal_list(napi); 6192 count: 6193 if (work > 0) 6194 __NET_ADD_STATS(dev_net(napi->dev), 6195 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6196 local_bh_enable(); 6197 6198 if (!loop_end || loop_end(loop_end_arg, start_time)) 6199 break; 6200 6201 if (unlikely(need_resched())) { 6202 if (napi_poll) 6203 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6204 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6205 preempt_enable(); 6206 rcu_read_unlock(); 6207 cond_resched(); 6208 if (loop_end(loop_end_arg, start_time)) 6209 return; 6210 goto restart; 6211 } 6212 cpu_relax(); 6213 } 6214 if (napi_poll) 6215 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6216 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6217 preempt_enable(); 6218 out: 6219 rcu_read_unlock(); 6220 } 6221 EXPORT_SYMBOL(napi_busy_loop); 6222 6223 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6224 6225 static void napi_hash_add(struct napi_struct *napi) 6226 { 6227 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6228 return; 6229 6230 spin_lock(&napi_hash_lock); 6231 6232 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6233 do { 6234 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6235 napi_gen_id = MIN_NAPI_ID; 6236 } while (napi_by_id(napi_gen_id)); 6237 napi->napi_id = napi_gen_id; 6238 6239 hlist_add_head_rcu(&napi->napi_hash_node, 6240 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6241 6242 spin_unlock(&napi_hash_lock); 6243 } 6244 6245 /* Warning : caller is responsible to make sure rcu grace period 6246 * is respected before freeing memory containing @napi 6247 */ 6248 static void napi_hash_del(struct napi_struct *napi) 6249 { 6250 spin_lock(&napi_hash_lock); 6251 6252 hlist_del_init_rcu(&napi->napi_hash_node); 6253 6254 spin_unlock(&napi_hash_lock); 6255 } 6256 6257 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6258 { 6259 struct napi_struct *napi; 6260 6261 napi = container_of(timer, struct napi_struct, timer); 6262 6263 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6264 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6265 */ 6266 if (!napi_disable_pending(napi) && 6267 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6268 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6269 __napi_schedule_irqoff(napi); 6270 } 6271 6272 return HRTIMER_NORESTART; 6273 } 6274 6275 static void init_gro_hash(struct napi_struct *napi) 6276 { 6277 int i; 6278 6279 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6280 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6281 napi->gro_hash[i].count = 0; 6282 } 6283 napi->gro_bitmask = 0; 6284 } 6285 6286 int dev_set_threaded(struct net_device *dev, bool threaded) 6287 { 6288 struct napi_struct *napi; 6289 int err = 0; 6290 6291 if (dev->threaded == threaded) 6292 return 0; 6293 6294 if (threaded) { 6295 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6296 if (!napi->thread) { 6297 err = napi_kthread_create(napi); 6298 if (err) { 6299 threaded = false; 6300 break; 6301 } 6302 } 6303 } 6304 } 6305 6306 dev->threaded = threaded; 6307 6308 /* Make sure kthread is created before THREADED bit 6309 * is set. 6310 */ 6311 smp_mb__before_atomic(); 6312 6313 /* Setting/unsetting threaded mode on a napi might not immediately 6314 * take effect, if the current napi instance is actively being 6315 * polled. In this case, the switch between threaded mode and 6316 * softirq mode will happen in the next round of napi_schedule(). 6317 * This should not cause hiccups/stalls to the live traffic. 6318 */ 6319 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6320 if (threaded) 6321 set_bit(NAPI_STATE_THREADED, &napi->state); 6322 else 6323 clear_bit(NAPI_STATE_THREADED, &napi->state); 6324 } 6325 6326 return err; 6327 } 6328 EXPORT_SYMBOL(dev_set_threaded); 6329 6330 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6331 int (*poll)(struct napi_struct *, int), int weight) 6332 { 6333 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6334 return; 6335 6336 INIT_LIST_HEAD(&napi->poll_list); 6337 INIT_HLIST_NODE(&napi->napi_hash_node); 6338 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6339 napi->timer.function = napi_watchdog; 6340 init_gro_hash(napi); 6341 napi->skb = NULL; 6342 INIT_LIST_HEAD(&napi->rx_list); 6343 napi->rx_count = 0; 6344 napi->poll = poll; 6345 if (weight > NAPI_POLL_WEIGHT) 6346 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6347 weight); 6348 napi->weight = weight; 6349 napi->dev = dev; 6350 #ifdef CONFIG_NETPOLL 6351 napi->poll_owner = -1; 6352 #endif 6353 napi->list_owner = -1; 6354 set_bit(NAPI_STATE_SCHED, &napi->state); 6355 set_bit(NAPI_STATE_NPSVC, &napi->state); 6356 list_add_rcu(&napi->dev_list, &dev->napi_list); 6357 napi_hash_add(napi); 6358 napi_get_frags_check(napi); 6359 /* Create kthread for this napi if dev->threaded is set. 6360 * Clear dev->threaded if kthread creation failed so that 6361 * threaded mode will not be enabled in napi_enable(). 6362 */ 6363 if (dev->threaded && napi_kthread_create(napi)) 6364 dev->threaded = 0; 6365 } 6366 EXPORT_SYMBOL(netif_napi_add_weight); 6367 6368 void napi_disable(struct napi_struct *n) 6369 { 6370 unsigned long val, new; 6371 6372 might_sleep(); 6373 set_bit(NAPI_STATE_DISABLE, &n->state); 6374 6375 val = READ_ONCE(n->state); 6376 do { 6377 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6378 usleep_range(20, 200); 6379 val = READ_ONCE(n->state); 6380 } 6381 6382 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6383 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6384 } while (!try_cmpxchg(&n->state, &val, new)); 6385 6386 hrtimer_cancel(&n->timer); 6387 6388 clear_bit(NAPI_STATE_DISABLE, &n->state); 6389 } 6390 EXPORT_SYMBOL(napi_disable); 6391 6392 /** 6393 * napi_enable - enable NAPI scheduling 6394 * @n: NAPI context 6395 * 6396 * Resume NAPI from being scheduled on this context. 6397 * Must be paired with napi_disable. 6398 */ 6399 void napi_enable(struct napi_struct *n) 6400 { 6401 unsigned long new, val = READ_ONCE(n->state); 6402 6403 do { 6404 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6405 6406 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6407 if (n->dev->threaded && n->thread) 6408 new |= NAPIF_STATE_THREADED; 6409 } while (!try_cmpxchg(&n->state, &val, new)); 6410 } 6411 EXPORT_SYMBOL(napi_enable); 6412 6413 static void flush_gro_hash(struct napi_struct *napi) 6414 { 6415 int i; 6416 6417 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6418 struct sk_buff *skb, *n; 6419 6420 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6421 kfree_skb(skb); 6422 napi->gro_hash[i].count = 0; 6423 } 6424 } 6425 6426 /* Must be called in process context */ 6427 void __netif_napi_del(struct napi_struct *napi) 6428 { 6429 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6430 return; 6431 6432 napi_hash_del(napi); 6433 list_del_rcu(&napi->dev_list); 6434 napi_free_frags(napi); 6435 6436 flush_gro_hash(napi); 6437 napi->gro_bitmask = 0; 6438 6439 if (napi->thread) { 6440 kthread_stop(napi->thread); 6441 napi->thread = NULL; 6442 } 6443 } 6444 EXPORT_SYMBOL(__netif_napi_del); 6445 6446 static int __napi_poll(struct napi_struct *n, bool *repoll) 6447 { 6448 int work, weight; 6449 6450 weight = n->weight; 6451 6452 /* This NAPI_STATE_SCHED test is for avoiding a race 6453 * with netpoll's poll_napi(). Only the entity which 6454 * obtains the lock and sees NAPI_STATE_SCHED set will 6455 * actually make the ->poll() call. Therefore we avoid 6456 * accidentally calling ->poll() when NAPI is not scheduled. 6457 */ 6458 work = 0; 6459 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6460 work = n->poll(n, weight); 6461 trace_napi_poll(n, work, weight); 6462 } 6463 6464 if (unlikely(work > weight)) 6465 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6466 n->poll, work, weight); 6467 6468 if (likely(work < weight)) 6469 return work; 6470 6471 /* Drivers must not modify the NAPI state if they 6472 * consume the entire weight. In such cases this code 6473 * still "owns" the NAPI instance and therefore can 6474 * move the instance around on the list at-will. 6475 */ 6476 if (unlikely(napi_disable_pending(n))) { 6477 napi_complete(n); 6478 return work; 6479 } 6480 6481 /* The NAPI context has more processing work, but busy-polling 6482 * is preferred. Exit early. 6483 */ 6484 if (napi_prefer_busy_poll(n)) { 6485 if (napi_complete_done(n, work)) { 6486 /* If timeout is not set, we need to make sure 6487 * that the NAPI is re-scheduled. 6488 */ 6489 napi_schedule(n); 6490 } 6491 return work; 6492 } 6493 6494 if (n->gro_bitmask) { 6495 /* flush too old packets 6496 * If HZ < 1000, flush all packets. 6497 */ 6498 napi_gro_flush(n, HZ >= 1000); 6499 } 6500 6501 gro_normal_list(n); 6502 6503 /* Some drivers may have called napi_schedule 6504 * prior to exhausting their budget. 6505 */ 6506 if (unlikely(!list_empty(&n->poll_list))) { 6507 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6508 n->dev ? n->dev->name : "backlog"); 6509 return work; 6510 } 6511 6512 *repoll = true; 6513 6514 return work; 6515 } 6516 6517 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6518 { 6519 bool do_repoll = false; 6520 void *have; 6521 int work; 6522 6523 list_del_init(&n->poll_list); 6524 6525 have = netpoll_poll_lock(n); 6526 6527 work = __napi_poll(n, &do_repoll); 6528 6529 if (do_repoll) 6530 list_add_tail(&n->poll_list, repoll); 6531 6532 netpoll_poll_unlock(have); 6533 6534 return work; 6535 } 6536 6537 static int napi_thread_wait(struct napi_struct *napi) 6538 { 6539 bool woken = false; 6540 6541 set_current_state(TASK_INTERRUPTIBLE); 6542 6543 while (!kthread_should_stop()) { 6544 /* Testing SCHED_THREADED bit here to make sure the current 6545 * kthread owns this napi and could poll on this napi. 6546 * Testing SCHED bit is not enough because SCHED bit might be 6547 * set by some other busy poll thread or by napi_disable(). 6548 */ 6549 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6550 WARN_ON(!list_empty(&napi->poll_list)); 6551 __set_current_state(TASK_RUNNING); 6552 return 0; 6553 } 6554 6555 schedule(); 6556 /* woken being true indicates this thread owns this napi. */ 6557 woken = true; 6558 set_current_state(TASK_INTERRUPTIBLE); 6559 } 6560 __set_current_state(TASK_RUNNING); 6561 6562 return -1; 6563 } 6564 6565 static void skb_defer_free_flush(struct softnet_data *sd) 6566 { 6567 struct sk_buff *skb, *next; 6568 6569 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6570 if (!READ_ONCE(sd->defer_list)) 6571 return; 6572 6573 spin_lock(&sd->defer_lock); 6574 skb = sd->defer_list; 6575 sd->defer_list = NULL; 6576 sd->defer_count = 0; 6577 spin_unlock(&sd->defer_lock); 6578 6579 while (skb != NULL) { 6580 next = skb->next; 6581 napi_consume_skb(skb, 1); 6582 skb = next; 6583 } 6584 } 6585 6586 static int napi_threaded_poll(void *data) 6587 { 6588 struct napi_struct *napi = data; 6589 struct softnet_data *sd; 6590 void *have; 6591 6592 while (!napi_thread_wait(napi)) { 6593 for (;;) { 6594 bool repoll = false; 6595 6596 local_bh_disable(); 6597 sd = this_cpu_ptr(&softnet_data); 6598 sd->in_napi_threaded_poll = true; 6599 6600 have = netpoll_poll_lock(napi); 6601 __napi_poll(napi, &repoll); 6602 netpoll_poll_unlock(have); 6603 6604 sd->in_napi_threaded_poll = false; 6605 barrier(); 6606 6607 if (sd_has_rps_ipi_waiting(sd)) { 6608 local_irq_disable(); 6609 net_rps_action_and_irq_enable(sd); 6610 } 6611 skb_defer_free_flush(sd); 6612 local_bh_enable(); 6613 6614 if (!repoll) 6615 break; 6616 6617 cond_resched(); 6618 } 6619 } 6620 return 0; 6621 } 6622 6623 static __latent_entropy void net_rx_action(struct softirq_action *h) 6624 { 6625 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6626 unsigned long time_limit = jiffies + 6627 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6628 int budget = READ_ONCE(netdev_budget); 6629 LIST_HEAD(list); 6630 LIST_HEAD(repoll); 6631 6632 start: 6633 sd->in_net_rx_action = true; 6634 local_irq_disable(); 6635 list_splice_init(&sd->poll_list, &list); 6636 local_irq_enable(); 6637 6638 for (;;) { 6639 struct napi_struct *n; 6640 6641 skb_defer_free_flush(sd); 6642 6643 if (list_empty(&list)) { 6644 if (list_empty(&repoll)) { 6645 sd->in_net_rx_action = false; 6646 barrier(); 6647 /* We need to check if ____napi_schedule() 6648 * had refilled poll_list while 6649 * sd->in_net_rx_action was true. 6650 */ 6651 if (!list_empty(&sd->poll_list)) 6652 goto start; 6653 if (!sd_has_rps_ipi_waiting(sd)) 6654 goto end; 6655 } 6656 break; 6657 } 6658 6659 n = list_first_entry(&list, struct napi_struct, poll_list); 6660 budget -= napi_poll(n, &repoll); 6661 6662 /* If softirq window is exhausted then punt. 6663 * Allow this to run for 2 jiffies since which will allow 6664 * an average latency of 1.5/HZ. 6665 */ 6666 if (unlikely(budget <= 0 || 6667 time_after_eq(jiffies, time_limit))) { 6668 sd->time_squeeze++; 6669 break; 6670 } 6671 } 6672 6673 local_irq_disable(); 6674 6675 list_splice_tail_init(&sd->poll_list, &list); 6676 list_splice_tail(&repoll, &list); 6677 list_splice(&list, &sd->poll_list); 6678 if (!list_empty(&sd->poll_list)) 6679 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6680 else 6681 sd->in_net_rx_action = false; 6682 6683 net_rps_action_and_irq_enable(sd); 6684 end:; 6685 } 6686 6687 struct netdev_adjacent { 6688 struct net_device *dev; 6689 netdevice_tracker dev_tracker; 6690 6691 /* upper master flag, there can only be one master device per list */ 6692 bool master; 6693 6694 /* lookup ignore flag */ 6695 bool ignore; 6696 6697 /* counter for the number of times this device was added to us */ 6698 u16 ref_nr; 6699 6700 /* private field for the users */ 6701 void *private; 6702 6703 struct list_head list; 6704 struct rcu_head rcu; 6705 }; 6706 6707 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6708 struct list_head *adj_list) 6709 { 6710 struct netdev_adjacent *adj; 6711 6712 list_for_each_entry(adj, adj_list, list) { 6713 if (adj->dev == adj_dev) 6714 return adj; 6715 } 6716 return NULL; 6717 } 6718 6719 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6720 struct netdev_nested_priv *priv) 6721 { 6722 struct net_device *dev = (struct net_device *)priv->data; 6723 6724 return upper_dev == dev; 6725 } 6726 6727 /** 6728 * netdev_has_upper_dev - Check if device is linked to an upper device 6729 * @dev: device 6730 * @upper_dev: upper device to check 6731 * 6732 * Find out if a device is linked to specified upper device and return true 6733 * in case it is. Note that this checks only immediate upper device, 6734 * not through a complete stack of devices. The caller must hold the RTNL lock. 6735 */ 6736 bool netdev_has_upper_dev(struct net_device *dev, 6737 struct net_device *upper_dev) 6738 { 6739 struct netdev_nested_priv priv = { 6740 .data = (void *)upper_dev, 6741 }; 6742 6743 ASSERT_RTNL(); 6744 6745 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6746 &priv); 6747 } 6748 EXPORT_SYMBOL(netdev_has_upper_dev); 6749 6750 /** 6751 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6752 * @dev: device 6753 * @upper_dev: upper device to check 6754 * 6755 * Find out if a device is linked to specified upper device and return true 6756 * in case it is. Note that this checks the entire upper device chain. 6757 * The caller must hold rcu lock. 6758 */ 6759 6760 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6761 struct net_device *upper_dev) 6762 { 6763 struct netdev_nested_priv priv = { 6764 .data = (void *)upper_dev, 6765 }; 6766 6767 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6768 &priv); 6769 } 6770 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6771 6772 /** 6773 * netdev_has_any_upper_dev - Check if device is linked to some device 6774 * @dev: device 6775 * 6776 * Find out if a device is linked to an upper device and return true in case 6777 * it is. The caller must hold the RTNL lock. 6778 */ 6779 bool netdev_has_any_upper_dev(struct net_device *dev) 6780 { 6781 ASSERT_RTNL(); 6782 6783 return !list_empty(&dev->adj_list.upper); 6784 } 6785 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6786 6787 /** 6788 * netdev_master_upper_dev_get - Get master upper device 6789 * @dev: device 6790 * 6791 * Find a master upper device and return pointer to it or NULL in case 6792 * it's not there. The caller must hold the RTNL lock. 6793 */ 6794 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6795 { 6796 struct netdev_adjacent *upper; 6797 6798 ASSERT_RTNL(); 6799 6800 if (list_empty(&dev->adj_list.upper)) 6801 return NULL; 6802 6803 upper = list_first_entry(&dev->adj_list.upper, 6804 struct netdev_adjacent, list); 6805 if (likely(upper->master)) 6806 return upper->dev; 6807 return NULL; 6808 } 6809 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6810 6811 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6812 { 6813 struct netdev_adjacent *upper; 6814 6815 ASSERT_RTNL(); 6816 6817 if (list_empty(&dev->adj_list.upper)) 6818 return NULL; 6819 6820 upper = list_first_entry(&dev->adj_list.upper, 6821 struct netdev_adjacent, list); 6822 if (likely(upper->master) && !upper->ignore) 6823 return upper->dev; 6824 return NULL; 6825 } 6826 6827 /** 6828 * netdev_has_any_lower_dev - Check if device is linked to some device 6829 * @dev: device 6830 * 6831 * Find out if a device is linked to a lower device and return true in case 6832 * it is. The caller must hold the RTNL lock. 6833 */ 6834 static bool netdev_has_any_lower_dev(struct net_device *dev) 6835 { 6836 ASSERT_RTNL(); 6837 6838 return !list_empty(&dev->adj_list.lower); 6839 } 6840 6841 void *netdev_adjacent_get_private(struct list_head *adj_list) 6842 { 6843 struct netdev_adjacent *adj; 6844 6845 adj = list_entry(adj_list, struct netdev_adjacent, list); 6846 6847 return adj->private; 6848 } 6849 EXPORT_SYMBOL(netdev_adjacent_get_private); 6850 6851 /** 6852 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6853 * @dev: device 6854 * @iter: list_head ** of the current position 6855 * 6856 * Gets the next device from the dev's upper list, starting from iter 6857 * position. The caller must hold RCU read lock. 6858 */ 6859 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6860 struct list_head **iter) 6861 { 6862 struct netdev_adjacent *upper; 6863 6864 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6865 6866 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6867 6868 if (&upper->list == &dev->adj_list.upper) 6869 return NULL; 6870 6871 *iter = &upper->list; 6872 6873 return upper->dev; 6874 } 6875 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6876 6877 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6878 struct list_head **iter, 6879 bool *ignore) 6880 { 6881 struct netdev_adjacent *upper; 6882 6883 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6884 6885 if (&upper->list == &dev->adj_list.upper) 6886 return NULL; 6887 6888 *iter = &upper->list; 6889 *ignore = upper->ignore; 6890 6891 return upper->dev; 6892 } 6893 6894 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6895 struct list_head **iter) 6896 { 6897 struct netdev_adjacent *upper; 6898 6899 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6900 6901 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6902 6903 if (&upper->list == &dev->adj_list.upper) 6904 return NULL; 6905 6906 *iter = &upper->list; 6907 6908 return upper->dev; 6909 } 6910 6911 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6912 int (*fn)(struct net_device *dev, 6913 struct netdev_nested_priv *priv), 6914 struct netdev_nested_priv *priv) 6915 { 6916 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6917 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6918 int ret, cur = 0; 6919 bool ignore; 6920 6921 now = dev; 6922 iter = &dev->adj_list.upper; 6923 6924 while (1) { 6925 if (now != dev) { 6926 ret = fn(now, priv); 6927 if (ret) 6928 return ret; 6929 } 6930 6931 next = NULL; 6932 while (1) { 6933 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6934 if (!udev) 6935 break; 6936 if (ignore) 6937 continue; 6938 6939 next = udev; 6940 niter = &udev->adj_list.upper; 6941 dev_stack[cur] = now; 6942 iter_stack[cur++] = iter; 6943 break; 6944 } 6945 6946 if (!next) { 6947 if (!cur) 6948 return 0; 6949 next = dev_stack[--cur]; 6950 niter = iter_stack[cur]; 6951 } 6952 6953 now = next; 6954 iter = niter; 6955 } 6956 6957 return 0; 6958 } 6959 6960 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6961 int (*fn)(struct net_device *dev, 6962 struct netdev_nested_priv *priv), 6963 struct netdev_nested_priv *priv) 6964 { 6965 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6966 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6967 int ret, cur = 0; 6968 6969 now = dev; 6970 iter = &dev->adj_list.upper; 6971 6972 while (1) { 6973 if (now != dev) { 6974 ret = fn(now, priv); 6975 if (ret) 6976 return ret; 6977 } 6978 6979 next = NULL; 6980 while (1) { 6981 udev = netdev_next_upper_dev_rcu(now, &iter); 6982 if (!udev) 6983 break; 6984 6985 next = udev; 6986 niter = &udev->adj_list.upper; 6987 dev_stack[cur] = now; 6988 iter_stack[cur++] = iter; 6989 break; 6990 } 6991 6992 if (!next) { 6993 if (!cur) 6994 return 0; 6995 next = dev_stack[--cur]; 6996 niter = iter_stack[cur]; 6997 } 6998 6999 now = next; 7000 iter = niter; 7001 } 7002 7003 return 0; 7004 } 7005 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7006 7007 static bool __netdev_has_upper_dev(struct net_device *dev, 7008 struct net_device *upper_dev) 7009 { 7010 struct netdev_nested_priv priv = { 7011 .flags = 0, 7012 .data = (void *)upper_dev, 7013 }; 7014 7015 ASSERT_RTNL(); 7016 7017 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7018 &priv); 7019 } 7020 7021 /** 7022 * netdev_lower_get_next_private - Get the next ->private from the 7023 * lower neighbour list 7024 * @dev: device 7025 * @iter: list_head ** of the current position 7026 * 7027 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7028 * list, starting from iter position. The caller must hold either hold the 7029 * RTNL lock or its own locking that guarantees that the neighbour lower 7030 * list will remain unchanged. 7031 */ 7032 void *netdev_lower_get_next_private(struct net_device *dev, 7033 struct list_head **iter) 7034 { 7035 struct netdev_adjacent *lower; 7036 7037 lower = list_entry(*iter, struct netdev_adjacent, list); 7038 7039 if (&lower->list == &dev->adj_list.lower) 7040 return NULL; 7041 7042 *iter = lower->list.next; 7043 7044 return lower->private; 7045 } 7046 EXPORT_SYMBOL(netdev_lower_get_next_private); 7047 7048 /** 7049 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7050 * lower neighbour list, RCU 7051 * variant 7052 * @dev: device 7053 * @iter: list_head ** of the current position 7054 * 7055 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7056 * list, starting from iter position. The caller must hold RCU read lock. 7057 */ 7058 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7059 struct list_head **iter) 7060 { 7061 struct netdev_adjacent *lower; 7062 7063 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7064 7065 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7066 7067 if (&lower->list == &dev->adj_list.lower) 7068 return NULL; 7069 7070 *iter = &lower->list; 7071 7072 return lower->private; 7073 } 7074 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7075 7076 /** 7077 * netdev_lower_get_next - Get the next device from the lower neighbour 7078 * list 7079 * @dev: device 7080 * @iter: list_head ** of the current position 7081 * 7082 * Gets the next netdev_adjacent from the dev's lower neighbour 7083 * list, starting from iter position. The caller must hold RTNL lock or 7084 * its own locking that guarantees that the neighbour lower 7085 * list will remain unchanged. 7086 */ 7087 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7088 { 7089 struct netdev_adjacent *lower; 7090 7091 lower = list_entry(*iter, struct netdev_adjacent, list); 7092 7093 if (&lower->list == &dev->adj_list.lower) 7094 return NULL; 7095 7096 *iter = lower->list.next; 7097 7098 return lower->dev; 7099 } 7100 EXPORT_SYMBOL(netdev_lower_get_next); 7101 7102 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7103 struct list_head **iter) 7104 { 7105 struct netdev_adjacent *lower; 7106 7107 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7108 7109 if (&lower->list == &dev->adj_list.lower) 7110 return NULL; 7111 7112 *iter = &lower->list; 7113 7114 return lower->dev; 7115 } 7116 7117 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7118 struct list_head **iter, 7119 bool *ignore) 7120 { 7121 struct netdev_adjacent *lower; 7122 7123 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7124 7125 if (&lower->list == &dev->adj_list.lower) 7126 return NULL; 7127 7128 *iter = &lower->list; 7129 *ignore = lower->ignore; 7130 7131 return lower->dev; 7132 } 7133 7134 int netdev_walk_all_lower_dev(struct net_device *dev, 7135 int (*fn)(struct net_device *dev, 7136 struct netdev_nested_priv *priv), 7137 struct netdev_nested_priv *priv) 7138 { 7139 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7140 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7141 int ret, cur = 0; 7142 7143 now = dev; 7144 iter = &dev->adj_list.lower; 7145 7146 while (1) { 7147 if (now != dev) { 7148 ret = fn(now, priv); 7149 if (ret) 7150 return ret; 7151 } 7152 7153 next = NULL; 7154 while (1) { 7155 ldev = netdev_next_lower_dev(now, &iter); 7156 if (!ldev) 7157 break; 7158 7159 next = ldev; 7160 niter = &ldev->adj_list.lower; 7161 dev_stack[cur] = now; 7162 iter_stack[cur++] = iter; 7163 break; 7164 } 7165 7166 if (!next) { 7167 if (!cur) 7168 return 0; 7169 next = dev_stack[--cur]; 7170 niter = iter_stack[cur]; 7171 } 7172 7173 now = next; 7174 iter = niter; 7175 } 7176 7177 return 0; 7178 } 7179 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7180 7181 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7182 int (*fn)(struct net_device *dev, 7183 struct netdev_nested_priv *priv), 7184 struct netdev_nested_priv *priv) 7185 { 7186 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7187 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7188 int ret, cur = 0; 7189 bool ignore; 7190 7191 now = dev; 7192 iter = &dev->adj_list.lower; 7193 7194 while (1) { 7195 if (now != dev) { 7196 ret = fn(now, priv); 7197 if (ret) 7198 return ret; 7199 } 7200 7201 next = NULL; 7202 while (1) { 7203 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7204 if (!ldev) 7205 break; 7206 if (ignore) 7207 continue; 7208 7209 next = ldev; 7210 niter = &ldev->adj_list.lower; 7211 dev_stack[cur] = now; 7212 iter_stack[cur++] = iter; 7213 break; 7214 } 7215 7216 if (!next) { 7217 if (!cur) 7218 return 0; 7219 next = dev_stack[--cur]; 7220 niter = iter_stack[cur]; 7221 } 7222 7223 now = next; 7224 iter = niter; 7225 } 7226 7227 return 0; 7228 } 7229 7230 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7231 struct list_head **iter) 7232 { 7233 struct netdev_adjacent *lower; 7234 7235 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7236 if (&lower->list == &dev->adj_list.lower) 7237 return NULL; 7238 7239 *iter = &lower->list; 7240 7241 return lower->dev; 7242 } 7243 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7244 7245 static u8 __netdev_upper_depth(struct net_device *dev) 7246 { 7247 struct net_device *udev; 7248 struct list_head *iter; 7249 u8 max_depth = 0; 7250 bool ignore; 7251 7252 for (iter = &dev->adj_list.upper, 7253 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7254 udev; 7255 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7256 if (ignore) 7257 continue; 7258 if (max_depth < udev->upper_level) 7259 max_depth = udev->upper_level; 7260 } 7261 7262 return max_depth; 7263 } 7264 7265 static u8 __netdev_lower_depth(struct net_device *dev) 7266 { 7267 struct net_device *ldev; 7268 struct list_head *iter; 7269 u8 max_depth = 0; 7270 bool ignore; 7271 7272 for (iter = &dev->adj_list.lower, 7273 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7274 ldev; 7275 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7276 if (ignore) 7277 continue; 7278 if (max_depth < ldev->lower_level) 7279 max_depth = ldev->lower_level; 7280 } 7281 7282 return max_depth; 7283 } 7284 7285 static int __netdev_update_upper_level(struct net_device *dev, 7286 struct netdev_nested_priv *__unused) 7287 { 7288 dev->upper_level = __netdev_upper_depth(dev) + 1; 7289 return 0; 7290 } 7291 7292 #ifdef CONFIG_LOCKDEP 7293 static LIST_HEAD(net_unlink_list); 7294 7295 static void net_unlink_todo(struct net_device *dev) 7296 { 7297 if (list_empty(&dev->unlink_list)) 7298 list_add_tail(&dev->unlink_list, &net_unlink_list); 7299 } 7300 #endif 7301 7302 static int __netdev_update_lower_level(struct net_device *dev, 7303 struct netdev_nested_priv *priv) 7304 { 7305 dev->lower_level = __netdev_lower_depth(dev) + 1; 7306 7307 #ifdef CONFIG_LOCKDEP 7308 if (!priv) 7309 return 0; 7310 7311 if (priv->flags & NESTED_SYNC_IMM) 7312 dev->nested_level = dev->lower_level - 1; 7313 if (priv->flags & NESTED_SYNC_TODO) 7314 net_unlink_todo(dev); 7315 #endif 7316 return 0; 7317 } 7318 7319 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7320 int (*fn)(struct net_device *dev, 7321 struct netdev_nested_priv *priv), 7322 struct netdev_nested_priv *priv) 7323 { 7324 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7325 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7326 int ret, cur = 0; 7327 7328 now = dev; 7329 iter = &dev->adj_list.lower; 7330 7331 while (1) { 7332 if (now != dev) { 7333 ret = fn(now, priv); 7334 if (ret) 7335 return ret; 7336 } 7337 7338 next = NULL; 7339 while (1) { 7340 ldev = netdev_next_lower_dev_rcu(now, &iter); 7341 if (!ldev) 7342 break; 7343 7344 next = ldev; 7345 niter = &ldev->adj_list.lower; 7346 dev_stack[cur] = now; 7347 iter_stack[cur++] = iter; 7348 break; 7349 } 7350 7351 if (!next) { 7352 if (!cur) 7353 return 0; 7354 next = dev_stack[--cur]; 7355 niter = iter_stack[cur]; 7356 } 7357 7358 now = next; 7359 iter = niter; 7360 } 7361 7362 return 0; 7363 } 7364 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7365 7366 /** 7367 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7368 * lower neighbour list, RCU 7369 * variant 7370 * @dev: device 7371 * 7372 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7373 * list. The caller must hold RCU read lock. 7374 */ 7375 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7376 { 7377 struct netdev_adjacent *lower; 7378 7379 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7380 struct netdev_adjacent, list); 7381 if (lower) 7382 return lower->private; 7383 return NULL; 7384 } 7385 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7386 7387 /** 7388 * netdev_master_upper_dev_get_rcu - Get master upper device 7389 * @dev: device 7390 * 7391 * Find a master upper device and return pointer to it or NULL in case 7392 * it's not there. The caller must hold the RCU read lock. 7393 */ 7394 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7395 { 7396 struct netdev_adjacent *upper; 7397 7398 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7399 struct netdev_adjacent, list); 7400 if (upper && likely(upper->master)) 7401 return upper->dev; 7402 return NULL; 7403 } 7404 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7405 7406 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7407 struct net_device *adj_dev, 7408 struct list_head *dev_list) 7409 { 7410 char linkname[IFNAMSIZ+7]; 7411 7412 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7413 "upper_%s" : "lower_%s", adj_dev->name); 7414 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7415 linkname); 7416 } 7417 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7418 char *name, 7419 struct list_head *dev_list) 7420 { 7421 char linkname[IFNAMSIZ+7]; 7422 7423 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7424 "upper_%s" : "lower_%s", name); 7425 sysfs_remove_link(&(dev->dev.kobj), linkname); 7426 } 7427 7428 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7429 struct net_device *adj_dev, 7430 struct list_head *dev_list) 7431 { 7432 return (dev_list == &dev->adj_list.upper || 7433 dev_list == &dev->adj_list.lower) && 7434 net_eq(dev_net(dev), dev_net(adj_dev)); 7435 } 7436 7437 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7438 struct net_device *adj_dev, 7439 struct list_head *dev_list, 7440 void *private, bool master) 7441 { 7442 struct netdev_adjacent *adj; 7443 int ret; 7444 7445 adj = __netdev_find_adj(adj_dev, dev_list); 7446 7447 if (adj) { 7448 adj->ref_nr += 1; 7449 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7450 dev->name, adj_dev->name, adj->ref_nr); 7451 7452 return 0; 7453 } 7454 7455 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7456 if (!adj) 7457 return -ENOMEM; 7458 7459 adj->dev = adj_dev; 7460 adj->master = master; 7461 adj->ref_nr = 1; 7462 adj->private = private; 7463 adj->ignore = false; 7464 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7465 7466 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7467 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7468 7469 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7470 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7471 if (ret) 7472 goto free_adj; 7473 } 7474 7475 /* Ensure that master link is always the first item in list. */ 7476 if (master) { 7477 ret = sysfs_create_link(&(dev->dev.kobj), 7478 &(adj_dev->dev.kobj), "master"); 7479 if (ret) 7480 goto remove_symlinks; 7481 7482 list_add_rcu(&adj->list, dev_list); 7483 } else { 7484 list_add_tail_rcu(&adj->list, dev_list); 7485 } 7486 7487 return 0; 7488 7489 remove_symlinks: 7490 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7491 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7492 free_adj: 7493 netdev_put(adj_dev, &adj->dev_tracker); 7494 kfree(adj); 7495 7496 return ret; 7497 } 7498 7499 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7500 struct net_device *adj_dev, 7501 u16 ref_nr, 7502 struct list_head *dev_list) 7503 { 7504 struct netdev_adjacent *adj; 7505 7506 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7507 dev->name, adj_dev->name, ref_nr); 7508 7509 adj = __netdev_find_adj(adj_dev, dev_list); 7510 7511 if (!adj) { 7512 pr_err("Adjacency does not exist for device %s from %s\n", 7513 dev->name, adj_dev->name); 7514 WARN_ON(1); 7515 return; 7516 } 7517 7518 if (adj->ref_nr > ref_nr) { 7519 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7520 dev->name, adj_dev->name, ref_nr, 7521 adj->ref_nr - ref_nr); 7522 adj->ref_nr -= ref_nr; 7523 return; 7524 } 7525 7526 if (adj->master) 7527 sysfs_remove_link(&(dev->dev.kobj), "master"); 7528 7529 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7530 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7531 7532 list_del_rcu(&adj->list); 7533 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7534 adj_dev->name, dev->name, adj_dev->name); 7535 netdev_put(adj_dev, &adj->dev_tracker); 7536 kfree_rcu(adj, rcu); 7537 } 7538 7539 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7540 struct net_device *upper_dev, 7541 struct list_head *up_list, 7542 struct list_head *down_list, 7543 void *private, bool master) 7544 { 7545 int ret; 7546 7547 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7548 private, master); 7549 if (ret) 7550 return ret; 7551 7552 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7553 private, false); 7554 if (ret) { 7555 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7556 return ret; 7557 } 7558 7559 return 0; 7560 } 7561 7562 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7563 struct net_device *upper_dev, 7564 u16 ref_nr, 7565 struct list_head *up_list, 7566 struct list_head *down_list) 7567 { 7568 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7569 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7570 } 7571 7572 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7573 struct net_device *upper_dev, 7574 void *private, bool master) 7575 { 7576 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7577 &dev->adj_list.upper, 7578 &upper_dev->adj_list.lower, 7579 private, master); 7580 } 7581 7582 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7583 struct net_device *upper_dev) 7584 { 7585 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7586 &dev->adj_list.upper, 7587 &upper_dev->adj_list.lower); 7588 } 7589 7590 static int __netdev_upper_dev_link(struct net_device *dev, 7591 struct net_device *upper_dev, bool master, 7592 void *upper_priv, void *upper_info, 7593 struct netdev_nested_priv *priv, 7594 struct netlink_ext_ack *extack) 7595 { 7596 struct netdev_notifier_changeupper_info changeupper_info = { 7597 .info = { 7598 .dev = dev, 7599 .extack = extack, 7600 }, 7601 .upper_dev = upper_dev, 7602 .master = master, 7603 .linking = true, 7604 .upper_info = upper_info, 7605 }; 7606 struct net_device *master_dev; 7607 int ret = 0; 7608 7609 ASSERT_RTNL(); 7610 7611 if (dev == upper_dev) 7612 return -EBUSY; 7613 7614 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7615 if (__netdev_has_upper_dev(upper_dev, dev)) 7616 return -EBUSY; 7617 7618 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7619 return -EMLINK; 7620 7621 if (!master) { 7622 if (__netdev_has_upper_dev(dev, upper_dev)) 7623 return -EEXIST; 7624 } else { 7625 master_dev = __netdev_master_upper_dev_get(dev); 7626 if (master_dev) 7627 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7628 } 7629 7630 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7631 &changeupper_info.info); 7632 ret = notifier_to_errno(ret); 7633 if (ret) 7634 return ret; 7635 7636 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7637 master); 7638 if (ret) 7639 return ret; 7640 7641 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7642 &changeupper_info.info); 7643 ret = notifier_to_errno(ret); 7644 if (ret) 7645 goto rollback; 7646 7647 __netdev_update_upper_level(dev, NULL); 7648 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7649 7650 __netdev_update_lower_level(upper_dev, priv); 7651 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7652 priv); 7653 7654 return 0; 7655 7656 rollback: 7657 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7658 7659 return ret; 7660 } 7661 7662 /** 7663 * netdev_upper_dev_link - Add a link to the upper device 7664 * @dev: device 7665 * @upper_dev: new upper device 7666 * @extack: netlink extended ack 7667 * 7668 * Adds a link to device which is upper to this one. The caller must hold 7669 * the RTNL lock. On a failure a negative errno code is returned. 7670 * On success the reference counts are adjusted and the function 7671 * returns zero. 7672 */ 7673 int netdev_upper_dev_link(struct net_device *dev, 7674 struct net_device *upper_dev, 7675 struct netlink_ext_ack *extack) 7676 { 7677 struct netdev_nested_priv priv = { 7678 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7679 .data = NULL, 7680 }; 7681 7682 return __netdev_upper_dev_link(dev, upper_dev, false, 7683 NULL, NULL, &priv, extack); 7684 } 7685 EXPORT_SYMBOL(netdev_upper_dev_link); 7686 7687 /** 7688 * netdev_master_upper_dev_link - Add a master link to the upper device 7689 * @dev: device 7690 * @upper_dev: new upper device 7691 * @upper_priv: upper device private 7692 * @upper_info: upper info to be passed down via notifier 7693 * @extack: netlink extended ack 7694 * 7695 * Adds a link to device which is upper to this one. In this case, only 7696 * one master upper device can be linked, although other non-master devices 7697 * might be linked as well. The caller must hold the RTNL lock. 7698 * On a failure a negative errno code is returned. On success the reference 7699 * counts are adjusted and the function returns zero. 7700 */ 7701 int netdev_master_upper_dev_link(struct net_device *dev, 7702 struct net_device *upper_dev, 7703 void *upper_priv, void *upper_info, 7704 struct netlink_ext_ack *extack) 7705 { 7706 struct netdev_nested_priv priv = { 7707 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7708 .data = NULL, 7709 }; 7710 7711 return __netdev_upper_dev_link(dev, upper_dev, true, 7712 upper_priv, upper_info, &priv, extack); 7713 } 7714 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7715 7716 static void __netdev_upper_dev_unlink(struct net_device *dev, 7717 struct net_device *upper_dev, 7718 struct netdev_nested_priv *priv) 7719 { 7720 struct netdev_notifier_changeupper_info changeupper_info = { 7721 .info = { 7722 .dev = dev, 7723 }, 7724 .upper_dev = upper_dev, 7725 .linking = false, 7726 }; 7727 7728 ASSERT_RTNL(); 7729 7730 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7731 7732 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7733 &changeupper_info.info); 7734 7735 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7736 7737 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7738 &changeupper_info.info); 7739 7740 __netdev_update_upper_level(dev, NULL); 7741 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7742 7743 __netdev_update_lower_level(upper_dev, priv); 7744 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7745 priv); 7746 } 7747 7748 /** 7749 * netdev_upper_dev_unlink - Removes a link to upper device 7750 * @dev: device 7751 * @upper_dev: new upper device 7752 * 7753 * Removes a link to device which is upper to this one. The caller must hold 7754 * the RTNL lock. 7755 */ 7756 void netdev_upper_dev_unlink(struct net_device *dev, 7757 struct net_device *upper_dev) 7758 { 7759 struct netdev_nested_priv priv = { 7760 .flags = NESTED_SYNC_TODO, 7761 .data = NULL, 7762 }; 7763 7764 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7765 } 7766 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7767 7768 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7769 struct net_device *lower_dev, 7770 bool val) 7771 { 7772 struct netdev_adjacent *adj; 7773 7774 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7775 if (adj) 7776 adj->ignore = val; 7777 7778 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7779 if (adj) 7780 adj->ignore = val; 7781 } 7782 7783 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7784 struct net_device *lower_dev) 7785 { 7786 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7787 } 7788 7789 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7790 struct net_device *lower_dev) 7791 { 7792 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7793 } 7794 7795 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7796 struct net_device *new_dev, 7797 struct net_device *dev, 7798 struct netlink_ext_ack *extack) 7799 { 7800 struct netdev_nested_priv priv = { 7801 .flags = 0, 7802 .data = NULL, 7803 }; 7804 int err; 7805 7806 if (!new_dev) 7807 return 0; 7808 7809 if (old_dev && new_dev != old_dev) 7810 netdev_adjacent_dev_disable(dev, old_dev); 7811 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7812 extack); 7813 if (err) { 7814 if (old_dev && new_dev != old_dev) 7815 netdev_adjacent_dev_enable(dev, old_dev); 7816 return err; 7817 } 7818 7819 return 0; 7820 } 7821 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7822 7823 void netdev_adjacent_change_commit(struct net_device *old_dev, 7824 struct net_device *new_dev, 7825 struct net_device *dev) 7826 { 7827 struct netdev_nested_priv priv = { 7828 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7829 .data = NULL, 7830 }; 7831 7832 if (!new_dev || !old_dev) 7833 return; 7834 7835 if (new_dev == old_dev) 7836 return; 7837 7838 netdev_adjacent_dev_enable(dev, old_dev); 7839 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7840 } 7841 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7842 7843 void netdev_adjacent_change_abort(struct net_device *old_dev, 7844 struct net_device *new_dev, 7845 struct net_device *dev) 7846 { 7847 struct netdev_nested_priv priv = { 7848 .flags = 0, 7849 .data = NULL, 7850 }; 7851 7852 if (!new_dev) 7853 return; 7854 7855 if (old_dev && new_dev != old_dev) 7856 netdev_adjacent_dev_enable(dev, old_dev); 7857 7858 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7859 } 7860 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7861 7862 /** 7863 * netdev_bonding_info_change - Dispatch event about slave change 7864 * @dev: device 7865 * @bonding_info: info to dispatch 7866 * 7867 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7868 * The caller must hold the RTNL lock. 7869 */ 7870 void netdev_bonding_info_change(struct net_device *dev, 7871 struct netdev_bonding_info *bonding_info) 7872 { 7873 struct netdev_notifier_bonding_info info = { 7874 .info.dev = dev, 7875 }; 7876 7877 memcpy(&info.bonding_info, bonding_info, 7878 sizeof(struct netdev_bonding_info)); 7879 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7880 &info.info); 7881 } 7882 EXPORT_SYMBOL(netdev_bonding_info_change); 7883 7884 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7885 struct netlink_ext_ack *extack) 7886 { 7887 struct netdev_notifier_offload_xstats_info info = { 7888 .info.dev = dev, 7889 .info.extack = extack, 7890 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7891 }; 7892 int err; 7893 int rc; 7894 7895 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7896 GFP_KERNEL); 7897 if (!dev->offload_xstats_l3) 7898 return -ENOMEM; 7899 7900 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7901 NETDEV_OFFLOAD_XSTATS_DISABLE, 7902 &info.info); 7903 err = notifier_to_errno(rc); 7904 if (err) 7905 goto free_stats; 7906 7907 return 0; 7908 7909 free_stats: 7910 kfree(dev->offload_xstats_l3); 7911 dev->offload_xstats_l3 = NULL; 7912 return err; 7913 } 7914 7915 int netdev_offload_xstats_enable(struct net_device *dev, 7916 enum netdev_offload_xstats_type type, 7917 struct netlink_ext_ack *extack) 7918 { 7919 ASSERT_RTNL(); 7920 7921 if (netdev_offload_xstats_enabled(dev, type)) 7922 return -EALREADY; 7923 7924 switch (type) { 7925 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7926 return netdev_offload_xstats_enable_l3(dev, extack); 7927 } 7928 7929 WARN_ON(1); 7930 return -EINVAL; 7931 } 7932 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7933 7934 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7935 { 7936 struct netdev_notifier_offload_xstats_info info = { 7937 .info.dev = dev, 7938 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7939 }; 7940 7941 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7942 &info.info); 7943 kfree(dev->offload_xstats_l3); 7944 dev->offload_xstats_l3 = NULL; 7945 } 7946 7947 int netdev_offload_xstats_disable(struct net_device *dev, 7948 enum netdev_offload_xstats_type type) 7949 { 7950 ASSERT_RTNL(); 7951 7952 if (!netdev_offload_xstats_enabled(dev, type)) 7953 return -EALREADY; 7954 7955 switch (type) { 7956 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7957 netdev_offload_xstats_disable_l3(dev); 7958 return 0; 7959 } 7960 7961 WARN_ON(1); 7962 return -EINVAL; 7963 } 7964 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7965 7966 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7967 { 7968 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7969 } 7970 7971 static struct rtnl_hw_stats64 * 7972 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7973 enum netdev_offload_xstats_type type) 7974 { 7975 switch (type) { 7976 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7977 return dev->offload_xstats_l3; 7978 } 7979 7980 WARN_ON(1); 7981 return NULL; 7982 } 7983 7984 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7985 enum netdev_offload_xstats_type type) 7986 { 7987 ASSERT_RTNL(); 7988 7989 return netdev_offload_xstats_get_ptr(dev, type); 7990 } 7991 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7992 7993 struct netdev_notifier_offload_xstats_ru { 7994 bool used; 7995 }; 7996 7997 struct netdev_notifier_offload_xstats_rd { 7998 struct rtnl_hw_stats64 stats; 7999 bool used; 8000 }; 8001 8002 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8003 const struct rtnl_hw_stats64 *src) 8004 { 8005 dest->rx_packets += src->rx_packets; 8006 dest->tx_packets += src->tx_packets; 8007 dest->rx_bytes += src->rx_bytes; 8008 dest->tx_bytes += src->tx_bytes; 8009 dest->rx_errors += src->rx_errors; 8010 dest->tx_errors += src->tx_errors; 8011 dest->rx_dropped += src->rx_dropped; 8012 dest->tx_dropped += src->tx_dropped; 8013 dest->multicast += src->multicast; 8014 } 8015 8016 static int netdev_offload_xstats_get_used(struct net_device *dev, 8017 enum netdev_offload_xstats_type type, 8018 bool *p_used, 8019 struct netlink_ext_ack *extack) 8020 { 8021 struct netdev_notifier_offload_xstats_ru report_used = {}; 8022 struct netdev_notifier_offload_xstats_info info = { 8023 .info.dev = dev, 8024 .info.extack = extack, 8025 .type = type, 8026 .report_used = &report_used, 8027 }; 8028 int rc; 8029 8030 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8031 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8032 &info.info); 8033 *p_used = report_used.used; 8034 return notifier_to_errno(rc); 8035 } 8036 8037 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8038 enum netdev_offload_xstats_type type, 8039 struct rtnl_hw_stats64 *p_stats, 8040 bool *p_used, 8041 struct netlink_ext_ack *extack) 8042 { 8043 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8044 struct netdev_notifier_offload_xstats_info info = { 8045 .info.dev = dev, 8046 .info.extack = extack, 8047 .type = type, 8048 .report_delta = &report_delta, 8049 }; 8050 struct rtnl_hw_stats64 *stats; 8051 int rc; 8052 8053 stats = netdev_offload_xstats_get_ptr(dev, type); 8054 if (WARN_ON(!stats)) 8055 return -EINVAL; 8056 8057 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8058 &info.info); 8059 8060 /* Cache whatever we got, even if there was an error, otherwise the 8061 * successful stats retrievals would get lost. 8062 */ 8063 netdev_hw_stats64_add(stats, &report_delta.stats); 8064 8065 if (p_stats) 8066 *p_stats = *stats; 8067 *p_used = report_delta.used; 8068 8069 return notifier_to_errno(rc); 8070 } 8071 8072 int netdev_offload_xstats_get(struct net_device *dev, 8073 enum netdev_offload_xstats_type type, 8074 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8075 struct netlink_ext_ack *extack) 8076 { 8077 ASSERT_RTNL(); 8078 8079 if (p_stats) 8080 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8081 p_used, extack); 8082 else 8083 return netdev_offload_xstats_get_used(dev, type, p_used, 8084 extack); 8085 } 8086 EXPORT_SYMBOL(netdev_offload_xstats_get); 8087 8088 void 8089 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8090 const struct rtnl_hw_stats64 *stats) 8091 { 8092 report_delta->used = true; 8093 netdev_hw_stats64_add(&report_delta->stats, stats); 8094 } 8095 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8096 8097 void 8098 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8099 { 8100 report_used->used = true; 8101 } 8102 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8103 8104 void netdev_offload_xstats_push_delta(struct net_device *dev, 8105 enum netdev_offload_xstats_type type, 8106 const struct rtnl_hw_stats64 *p_stats) 8107 { 8108 struct rtnl_hw_stats64 *stats; 8109 8110 ASSERT_RTNL(); 8111 8112 stats = netdev_offload_xstats_get_ptr(dev, type); 8113 if (WARN_ON(!stats)) 8114 return; 8115 8116 netdev_hw_stats64_add(stats, p_stats); 8117 } 8118 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8119 8120 /** 8121 * netdev_get_xmit_slave - Get the xmit slave of master device 8122 * @dev: device 8123 * @skb: The packet 8124 * @all_slaves: assume all the slaves are active 8125 * 8126 * The reference counters are not incremented so the caller must be 8127 * careful with locks. The caller must hold RCU lock. 8128 * %NULL is returned if no slave is found. 8129 */ 8130 8131 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8132 struct sk_buff *skb, 8133 bool all_slaves) 8134 { 8135 const struct net_device_ops *ops = dev->netdev_ops; 8136 8137 if (!ops->ndo_get_xmit_slave) 8138 return NULL; 8139 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8140 } 8141 EXPORT_SYMBOL(netdev_get_xmit_slave); 8142 8143 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8144 struct sock *sk) 8145 { 8146 const struct net_device_ops *ops = dev->netdev_ops; 8147 8148 if (!ops->ndo_sk_get_lower_dev) 8149 return NULL; 8150 return ops->ndo_sk_get_lower_dev(dev, sk); 8151 } 8152 8153 /** 8154 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8155 * @dev: device 8156 * @sk: the socket 8157 * 8158 * %NULL is returned if no lower device is found. 8159 */ 8160 8161 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8162 struct sock *sk) 8163 { 8164 struct net_device *lower; 8165 8166 lower = netdev_sk_get_lower_dev(dev, sk); 8167 while (lower) { 8168 dev = lower; 8169 lower = netdev_sk_get_lower_dev(dev, sk); 8170 } 8171 8172 return dev; 8173 } 8174 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8175 8176 static void netdev_adjacent_add_links(struct net_device *dev) 8177 { 8178 struct netdev_adjacent *iter; 8179 8180 struct net *net = dev_net(dev); 8181 8182 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8183 if (!net_eq(net, dev_net(iter->dev))) 8184 continue; 8185 netdev_adjacent_sysfs_add(iter->dev, dev, 8186 &iter->dev->adj_list.lower); 8187 netdev_adjacent_sysfs_add(dev, iter->dev, 8188 &dev->adj_list.upper); 8189 } 8190 8191 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8192 if (!net_eq(net, dev_net(iter->dev))) 8193 continue; 8194 netdev_adjacent_sysfs_add(iter->dev, dev, 8195 &iter->dev->adj_list.upper); 8196 netdev_adjacent_sysfs_add(dev, iter->dev, 8197 &dev->adj_list.lower); 8198 } 8199 } 8200 8201 static void netdev_adjacent_del_links(struct net_device *dev) 8202 { 8203 struct netdev_adjacent *iter; 8204 8205 struct net *net = dev_net(dev); 8206 8207 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8208 if (!net_eq(net, dev_net(iter->dev))) 8209 continue; 8210 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8211 &iter->dev->adj_list.lower); 8212 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8213 &dev->adj_list.upper); 8214 } 8215 8216 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8217 if (!net_eq(net, dev_net(iter->dev))) 8218 continue; 8219 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8220 &iter->dev->adj_list.upper); 8221 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8222 &dev->adj_list.lower); 8223 } 8224 } 8225 8226 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8227 { 8228 struct netdev_adjacent *iter; 8229 8230 struct net *net = dev_net(dev); 8231 8232 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8233 if (!net_eq(net, dev_net(iter->dev))) 8234 continue; 8235 netdev_adjacent_sysfs_del(iter->dev, oldname, 8236 &iter->dev->adj_list.lower); 8237 netdev_adjacent_sysfs_add(iter->dev, dev, 8238 &iter->dev->adj_list.lower); 8239 } 8240 8241 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8242 if (!net_eq(net, dev_net(iter->dev))) 8243 continue; 8244 netdev_adjacent_sysfs_del(iter->dev, oldname, 8245 &iter->dev->adj_list.upper); 8246 netdev_adjacent_sysfs_add(iter->dev, dev, 8247 &iter->dev->adj_list.upper); 8248 } 8249 } 8250 8251 void *netdev_lower_dev_get_private(struct net_device *dev, 8252 struct net_device *lower_dev) 8253 { 8254 struct netdev_adjacent *lower; 8255 8256 if (!lower_dev) 8257 return NULL; 8258 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8259 if (!lower) 8260 return NULL; 8261 8262 return lower->private; 8263 } 8264 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8265 8266 8267 /** 8268 * netdev_lower_state_changed - Dispatch event about lower device state change 8269 * @lower_dev: device 8270 * @lower_state_info: state to dispatch 8271 * 8272 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8273 * The caller must hold the RTNL lock. 8274 */ 8275 void netdev_lower_state_changed(struct net_device *lower_dev, 8276 void *lower_state_info) 8277 { 8278 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8279 .info.dev = lower_dev, 8280 }; 8281 8282 ASSERT_RTNL(); 8283 changelowerstate_info.lower_state_info = lower_state_info; 8284 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8285 &changelowerstate_info.info); 8286 } 8287 EXPORT_SYMBOL(netdev_lower_state_changed); 8288 8289 static void dev_change_rx_flags(struct net_device *dev, int flags) 8290 { 8291 const struct net_device_ops *ops = dev->netdev_ops; 8292 8293 if (ops->ndo_change_rx_flags) 8294 ops->ndo_change_rx_flags(dev, flags); 8295 } 8296 8297 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8298 { 8299 unsigned int old_flags = dev->flags; 8300 kuid_t uid; 8301 kgid_t gid; 8302 8303 ASSERT_RTNL(); 8304 8305 dev->flags |= IFF_PROMISC; 8306 dev->promiscuity += inc; 8307 if (dev->promiscuity == 0) { 8308 /* 8309 * Avoid overflow. 8310 * If inc causes overflow, untouch promisc and return error. 8311 */ 8312 if (inc < 0) 8313 dev->flags &= ~IFF_PROMISC; 8314 else { 8315 dev->promiscuity -= inc; 8316 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8317 return -EOVERFLOW; 8318 } 8319 } 8320 if (dev->flags != old_flags) { 8321 netdev_info(dev, "%s promiscuous mode\n", 8322 dev->flags & IFF_PROMISC ? "entered" : "left"); 8323 if (audit_enabled) { 8324 current_uid_gid(&uid, &gid); 8325 audit_log(audit_context(), GFP_ATOMIC, 8326 AUDIT_ANOM_PROMISCUOUS, 8327 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8328 dev->name, (dev->flags & IFF_PROMISC), 8329 (old_flags & IFF_PROMISC), 8330 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8331 from_kuid(&init_user_ns, uid), 8332 from_kgid(&init_user_ns, gid), 8333 audit_get_sessionid(current)); 8334 } 8335 8336 dev_change_rx_flags(dev, IFF_PROMISC); 8337 } 8338 if (notify) 8339 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8340 return 0; 8341 } 8342 8343 /** 8344 * dev_set_promiscuity - update promiscuity count on a device 8345 * @dev: device 8346 * @inc: modifier 8347 * 8348 * Add or remove promiscuity from a device. While the count in the device 8349 * remains above zero the interface remains promiscuous. Once it hits zero 8350 * the device reverts back to normal filtering operation. A negative inc 8351 * value is used to drop promiscuity on the device. 8352 * Return 0 if successful or a negative errno code on error. 8353 */ 8354 int dev_set_promiscuity(struct net_device *dev, int inc) 8355 { 8356 unsigned int old_flags = dev->flags; 8357 int err; 8358 8359 err = __dev_set_promiscuity(dev, inc, true); 8360 if (err < 0) 8361 return err; 8362 if (dev->flags != old_flags) 8363 dev_set_rx_mode(dev); 8364 return err; 8365 } 8366 EXPORT_SYMBOL(dev_set_promiscuity); 8367 8368 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8369 { 8370 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8371 8372 ASSERT_RTNL(); 8373 8374 dev->flags |= IFF_ALLMULTI; 8375 dev->allmulti += inc; 8376 if (dev->allmulti == 0) { 8377 /* 8378 * Avoid overflow. 8379 * If inc causes overflow, untouch allmulti and return error. 8380 */ 8381 if (inc < 0) 8382 dev->flags &= ~IFF_ALLMULTI; 8383 else { 8384 dev->allmulti -= inc; 8385 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8386 return -EOVERFLOW; 8387 } 8388 } 8389 if (dev->flags ^ old_flags) { 8390 netdev_info(dev, "%s allmulticast mode\n", 8391 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8392 dev_change_rx_flags(dev, IFF_ALLMULTI); 8393 dev_set_rx_mode(dev); 8394 if (notify) 8395 __dev_notify_flags(dev, old_flags, 8396 dev->gflags ^ old_gflags, 0, NULL); 8397 } 8398 return 0; 8399 } 8400 8401 /** 8402 * dev_set_allmulti - update allmulti count on a device 8403 * @dev: device 8404 * @inc: modifier 8405 * 8406 * Add or remove reception of all multicast frames to a device. While the 8407 * count in the device remains above zero the interface remains listening 8408 * to all interfaces. Once it hits zero the device reverts back to normal 8409 * filtering operation. A negative @inc value is used to drop the counter 8410 * when releasing a resource needing all multicasts. 8411 * Return 0 if successful or a negative errno code on error. 8412 */ 8413 8414 int dev_set_allmulti(struct net_device *dev, int inc) 8415 { 8416 return __dev_set_allmulti(dev, inc, true); 8417 } 8418 EXPORT_SYMBOL(dev_set_allmulti); 8419 8420 /* 8421 * Upload unicast and multicast address lists to device and 8422 * configure RX filtering. When the device doesn't support unicast 8423 * filtering it is put in promiscuous mode while unicast addresses 8424 * are present. 8425 */ 8426 void __dev_set_rx_mode(struct net_device *dev) 8427 { 8428 const struct net_device_ops *ops = dev->netdev_ops; 8429 8430 /* dev_open will call this function so the list will stay sane. */ 8431 if (!(dev->flags&IFF_UP)) 8432 return; 8433 8434 if (!netif_device_present(dev)) 8435 return; 8436 8437 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8438 /* Unicast addresses changes may only happen under the rtnl, 8439 * therefore calling __dev_set_promiscuity here is safe. 8440 */ 8441 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8442 __dev_set_promiscuity(dev, 1, false); 8443 dev->uc_promisc = true; 8444 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8445 __dev_set_promiscuity(dev, -1, false); 8446 dev->uc_promisc = false; 8447 } 8448 } 8449 8450 if (ops->ndo_set_rx_mode) 8451 ops->ndo_set_rx_mode(dev); 8452 } 8453 8454 void dev_set_rx_mode(struct net_device *dev) 8455 { 8456 netif_addr_lock_bh(dev); 8457 __dev_set_rx_mode(dev); 8458 netif_addr_unlock_bh(dev); 8459 } 8460 8461 /** 8462 * dev_get_flags - get flags reported to userspace 8463 * @dev: device 8464 * 8465 * Get the combination of flag bits exported through APIs to userspace. 8466 */ 8467 unsigned int dev_get_flags(const struct net_device *dev) 8468 { 8469 unsigned int flags; 8470 8471 flags = (dev->flags & ~(IFF_PROMISC | 8472 IFF_ALLMULTI | 8473 IFF_RUNNING | 8474 IFF_LOWER_UP | 8475 IFF_DORMANT)) | 8476 (dev->gflags & (IFF_PROMISC | 8477 IFF_ALLMULTI)); 8478 8479 if (netif_running(dev)) { 8480 if (netif_oper_up(dev)) 8481 flags |= IFF_RUNNING; 8482 if (netif_carrier_ok(dev)) 8483 flags |= IFF_LOWER_UP; 8484 if (netif_dormant(dev)) 8485 flags |= IFF_DORMANT; 8486 } 8487 8488 return flags; 8489 } 8490 EXPORT_SYMBOL(dev_get_flags); 8491 8492 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8493 struct netlink_ext_ack *extack) 8494 { 8495 unsigned int old_flags = dev->flags; 8496 int ret; 8497 8498 ASSERT_RTNL(); 8499 8500 /* 8501 * Set the flags on our device. 8502 */ 8503 8504 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8505 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8506 IFF_AUTOMEDIA)) | 8507 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8508 IFF_ALLMULTI)); 8509 8510 /* 8511 * Load in the correct multicast list now the flags have changed. 8512 */ 8513 8514 if ((old_flags ^ flags) & IFF_MULTICAST) 8515 dev_change_rx_flags(dev, IFF_MULTICAST); 8516 8517 dev_set_rx_mode(dev); 8518 8519 /* 8520 * Have we downed the interface. We handle IFF_UP ourselves 8521 * according to user attempts to set it, rather than blindly 8522 * setting it. 8523 */ 8524 8525 ret = 0; 8526 if ((old_flags ^ flags) & IFF_UP) { 8527 if (old_flags & IFF_UP) 8528 __dev_close(dev); 8529 else 8530 ret = __dev_open(dev, extack); 8531 } 8532 8533 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8534 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8535 unsigned int old_flags = dev->flags; 8536 8537 dev->gflags ^= IFF_PROMISC; 8538 8539 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8540 if (dev->flags != old_flags) 8541 dev_set_rx_mode(dev); 8542 } 8543 8544 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8545 * is important. Some (broken) drivers set IFF_PROMISC, when 8546 * IFF_ALLMULTI is requested not asking us and not reporting. 8547 */ 8548 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8549 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8550 8551 dev->gflags ^= IFF_ALLMULTI; 8552 __dev_set_allmulti(dev, inc, false); 8553 } 8554 8555 return ret; 8556 } 8557 8558 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8559 unsigned int gchanges, u32 portid, 8560 const struct nlmsghdr *nlh) 8561 { 8562 unsigned int changes = dev->flags ^ old_flags; 8563 8564 if (gchanges) 8565 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8566 8567 if (changes & IFF_UP) { 8568 if (dev->flags & IFF_UP) 8569 call_netdevice_notifiers(NETDEV_UP, dev); 8570 else 8571 call_netdevice_notifiers(NETDEV_DOWN, dev); 8572 } 8573 8574 if (dev->flags & IFF_UP && 8575 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8576 struct netdev_notifier_change_info change_info = { 8577 .info = { 8578 .dev = dev, 8579 }, 8580 .flags_changed = changes, 8581 }; 8582 8583 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8584 } 8585 } 8586 8587 /** 8588 * dev_change_flags - change device settings 8589 * @dev: device 8590 * @flags: device state flags 8591 * @extack: netlink extended ack 8592 * 8593 * Change settings on device based state flags. The flags are 8594 * in the userspace exported format. 8595 */ 8596 int dev_change_flags(struct net_device *dev, unsigned int flags, 8597 struct netlink_ext_ack *extack) 8598 { 8599 int ret; 8600 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8601 8602 ret = __dev_change_flags(dev, flags, extack); 8603 if (ret < 0) 8604 return ret; 8605 8606 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8607 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8608 return ret; 8609 } 8610 EXPORT_SYMBOL(dev_change_flags); 8611 8612 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8613 { 8614 const struct net_device_ops *ops = dev->netdev_ops; 8615 8616 if (ops->ndo_change_mtu) 8617 return ops->ndo_change_mtu(dev, new_mtu); 8618 8619 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8620 WRITE_ONCE(dev->mtu, new_mtu); 8621 return 0; 8622 } 8623 EXPORT_SYMBOL(__dev_set_mtu); 8624 8625 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8626 struct netlink_ext_ack *extack) 8627 { 8628 /* MTU must be positive, and in range */ 8629 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8630 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8631 return -EINVAL; 8632 } 8633 8634 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8635 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8636 return -EINVAL; 8637 } 8638 return 0; 8639 } 8640 8641 /** 8642 * dev_set_mtu_ext - Change maximum transfer unit 8643 * @dev: device 8644 * @new_mtu: new transfer unit 8645 * @extack: netlink extended ack 8646 * 8647 * Change the maximum transfer size of the network device. 8648 */ 8649 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8650 struct netlink_ext_ack *extack) 8651 { 8652 int err, orig_mtu; 8653 8654 if (new_mtu == dev->mtu) 8655 return 0; 8656 8657 err = dev_validate_mtu(dev, new_mtu, extack); 8658 if (err) 8659 return err; 8660 8661 if (!netif_device_present(dev)) 8662 return -ENODEV; 8663 8664 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8665 err = notifier_to_errno(err); 8666 if (err) 8667 return err; 8668 8669 orig_mtu = dev->mtu; 8670 err = __dev_set_mtu(dev, new_mtu); 8671 8672 if (!err) { 8673 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8674 orig_mtu); 8675 err = notifier_to_errno(err); 8676 if (err) { 8677 /* setting mtu back and notifying everyone again, 8678 * so that they have a chance to revert changes. 8679 */ 8680 __dev_set_mtu(dev, orig_mtu); 8681 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8682 new_mtu); 8683 } 8684 } 8685 return err; 8686 } 8687 8688 int dev_set_mtu(struct net_device *dev, int new_mtu) 8689 { 8690 struct netlink_ext_ack extack; 8691 int err; 8692 8693 memset(&extack, 0, sizeof(extack)); 8694 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8695 if (err && extack._msg) 8696 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8697 return err; 8698 } 8699 EXPORT_SYMBOL(dev_set_mtu); 8700 8701 /** 8702 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8703 * @dev: device 8704 * @new_len: new tx queue length 8705 */ 8706 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8707 { 8708 unsigned int orig_len = dev->tx_queue_len; 8709 int res; 8710 8711 if (new_len != (unsigned int)new_len) 8712 return -ERANGE; 8713 8714 if (new_len != orig_len) { 8715 dev->tx_queue_len = new_len; 8716 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8717 res = notifier_to_errno(res); 8718 if (res) 8719 goto err_rollback; 8720 res = dev_qdisc_change_tx_queue_len(dev); 8721 if (res) 8722 goto err_rollback; 8723 } 8724 8725 return 0; 8726 8727 err_rollback: 8728 netdev_err(dev, "refused to change device tx_queue_len\n"); 8729 dev->tx_queue_len = orig_len; 8730 return res; 8731 } 8732 8733 /** 8734 * dev_set_group - Change group this device belongs to 8735 * @dev: device 8736 * @new_group: group this device should belong to 8737 */ 8738 void dev_set_group(struct net_device *dev, int new_group) 8739 { 8740 dev->group = new_group; 8741 } 8742 8743 /** 8744 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8745 * @dev: device 8746 * @addr: new address 8747 * @extack: netlink extended ack 8748 */ 8749 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8750 struct netlink_ext_ack *extack) 8751 { 8752 struct netdev_notifier_pre_changeaddr_info info = { 8753 .info.dev = dev, 8754 .info.extack = extack, 8755 .dev_addr = addr, 8756 }; 8757 int rc; 8758 8759 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8760 return notifier_to_errno(rc); 8761 } 8762 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8763 8764 /** 8765 * dev_set_mac_address - Change Media Access Control Address 8766 * @dev: device 8767 * @sa: new address 8768 * @extack: netlink extended ack 8769 * 8770 * Change the hardware (MAC) address of the device 8771 */ 8772 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8773 struct netlink_ext_ack *extack) 8774 { 8775 const struct net_device_ops *ops = dev->netdev_ops; 8776 int err; 8777 8778 if (!ops->ndo_set_mac_address) 8779 return -EOPNOTSUPP; 8780 if (sa->sa_family != dev->type) 8781 return -EINVAL; 8782 if (!netif_device_present(dev)) 8783 return -ENODEV; 8784 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8785 if (err) 8786 return err; 8787 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8788 err = ops->ndo_set_mac_address(dev, sa); 8789 if (err) 8790 return err; 8791 } 8792 dev->addr_assign_type = NET_ADDR_SET; 8793 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8794 add_device_randomness(dev->dev_addr, dev->addr_len); 8795 return 0; 8796 } 8797 EXPORT_SYMBOL(dev_set_mac_address); 8798 8799 static DECLARE_RWSEM(dev_addr_sem); 8800 8801 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8802 struct netlink_ext_ack *extack) 8803 { 8804 int ret; 8805 8806 down_write(&dev_addr_sem); 8807 ret = dev_set_mac_address(dev, sa, extack); 8808 up_write(&dev_addr_sem); 8809 return ret; 8810 } 8811 EXPORT_SYMBOL(dev_set_mac_address_user); 8812 8813 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8814 { 8815 size_t size = sizeof(sa->sa_data_min); 8816 struct net_device *dev; 8817 int ret = 0; 8818 8819 down_read(&dev_addr_sem); 8820 rcu_read_lock(); 8821 8822 dev = dev_get_by_name_rcu(net, dev_name); 8823 if (!dev) { 8824 ret = -ENODEV; 8825 goto unlock; 8826 } 8827 if (!dev->addr_len) 8828 memset(sa->sa_data, 0, size); 8829 else 8830 memcpy(sa->sa_data, dev->dev_addr, 8831 min_t(size_t, size, dev->addr_len)); 8832 sa->sa_family = dev->type; 8833 8834 unlock: 8835 rcu_read_unlock(); 8836 up_read(&dev_addr_sem); 8837 return ret; 8838 } 8839 EXPORT_SYMBOL(dev_get_mac_address); 8840 8841 /** 8842 * dev_change_carrier - Change device carrier 8843 * @dev: device 8844 * @new_carrier: new value 8845 * 8846 * Change device carrier 8847 */ 8848 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8849 { 8850 const struct net_device_ops *ops = dev->netdev_ops; 8851 8852 if (!ops->ndo_change_carrier) 8853 return -EOPNOTSUPP; 8854 if (!netif_device_present(dev)) 8855 return -ENODEV; 8856 return ops->ndo_change_carrier(dev, new_carrier); 8857 } 8858 8859 /** 8860 * dev_get_phys_port_id - Get device physical port ID 8861 * @dev: device 8862 * @ppid: port ID 8863 * 8864 * Get device physical port ID 8865 */ 8866 int dev_get_phys_port_id(struct net_device *dev, 8867 struct netdev_phys_item_id *ppid) 8868 { 8869 const struct net_device_ops *ops = dev->netdev_ops; 8870 8871 if (!ops->ndo_get_phys_port_id) 8872 return -EOPNOTSUPP; 8873 return ops->ndo_get_phys_port_id(dev, ppid); 8874 } 8875 8876 /** 8877 * dev_get_phys_port_name - Get device physical port name 8878 * @dev: device 8879 * @name: port name 8880 * @len: limit of bytes to copy to name 8881 * 8882 * Get device physical port name 8883 */ 8884 int dev_get_phys_port_name(struct net_device *dev, 8885 char *name, size_t len) 8886 { 8887 const struct net_device_ops *ops = dev->netdev_ops; 8888 int err; 8889 8890 if (ops->ndo_get_phys_port_name) { 8891 err = ops->ndo_get_phys_port_name(dev, name, len); 8892 if (err != -EOPNOTSUPP) 8893 return err; 8894 } 8895 return devlink_compat_phys_port_name_get(dev, name, len); 8896 } 8897 8898 /** 8899 * dev_get_port_parent_id - Get the device's port parent identifier 8900 * @dev: network device 8901 * @ppid: pointer to a storage for the port's parent identifier 8902 * @recurse: allow/disallow recursion to lower devices 8903 * 8904 * Get the devices's port parent identifier 8905 */ 8906 int dev_get_port_parent_id(struct net_device *dev, 8907 struct netdev_phys_item_id *ppid, 8908 bool recurse) 8909 { 8910 const struct net_device_ops *ops = dev->netdev_ops; 8911 struct netdev_phys_item_id first = { }; 8912 struct net_device *lower_dev; 8913 struct list_head *iter; 8914 int err; 8915 8916 if (ops->ndo_get_port_parent_id) { 8917 err = ops->ndo_get_port_parent_id(dev, ppid); 8918 if (err != -EOPNOTSUPP) 8919 return err; 8920 } 8921 8922 err = devlink_compat_switch_id_get(dev, ppid); 8923 if (!recurse || err != -EOPNOTSUPP) 8924 return err; 8925 8926 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8927 err = dev_get_port_parent_id(lower_dev, ppid, true); 8928 if (err) 8929 break; 8930 if (!first.id_len) 8931 first = *ppid; 8932 else if (memcmp(&first, ppid, sizeof(*ppid))) 8933 return -EOPNOTSUPP; 8934 } 8935 8936 return err; 8937 } 8938 EXPORT_SYMBOL(dev_get_port_parent_id); 8939 8940 /** 8941 * netdev_port_same_parent_id - Indicate if two network devices have 8942 * the same port parent identifier 8943 * @a: first network device 8944 * @b: second network device 8945 */ 8946 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8947 { 8948 struct netdev_phys_item_id a_id = { }; 8949 struct netdev_phys_item_id b_id = { }; 8950 8951 if (dev_get_port_parent_id(a, &a_id, true) || 8952 dev_get_port_parent_id(b, &b_id, true)) 8953 return false; 8954 8955 return netdev_phys_item_id_same(&a_id, &b_id); 8956 } 8957 EXPORT_SYMBOL(netdev_port_same_parent_id); 8958 8959 /** 8960 * dev_change_proto_down - set carrier according to proto_down. 8961 * 8962 * @dev: device 8963 * @proto_down: new value 8964 */ 8965 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8966 { 8967 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8968 return -EOPNOTSUPP; 8969 if (!netif_device_present(dev)) 8970 return -ENODEV; 8971 if (proto_down) 8972 netif_carrier_off(dev); 8973 else 8974 netif_carrier_on(dev); 8975 dev->proto_down = proto_down; 8976 return 0; 8977 } 8978 8979 /** 8980 * dev_change_proto_down_reason - proto down reason 8981 * 8982 * @dev: device 8983 * @mask: proto down mask 8984 * @value: proto down value 8985 */ 8986 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8987 u32 value) 8988 { 8989 int b; 8990 8991 if (!mask) { 8992 dev->proto_down_reason = value; 8993 } else { 8994 for_each_set_bit(b, &mask, 32) { 8995 if (value & (1 << b)) 8996 dev->proto_down_reason |= BIT(b); 8997 else 8998 dev->proto_down_reason &= ~BIT(b); 8999 } 9000 } 9001 } 9002 9003 struct bpf_xdp_link { 9004 struct bpf_link link; 9005 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9006 int flags; 9007 }; 9008 9009 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9010 { 9011 if (flags & XDP_FLAGS_HW_MODE) 9012 return XDP_MODE_HW; 9013 if (flags & XDP_FLAGS_DRV_MODE) 9014 return XDP_MODE_DRV; 9015 if (flags & XDP_FLAGS_SKB_MODE) 9016 return XDP_MODE_SKB; 9017 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9018 } 9019 9020 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9021 { 9022 switch (mode) { 9023 case XDP_MODE_SKB: 9024 return generic_xdp_install; 9025 case XDP_MODE_DRV: 9026 case XDP_MODE_HW: 9027 return dev->netdev_ops->ndo_bpf; 9028 default: 9029 return NULL; 9030 } 9031 } 9032 9033 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9034 enum bpf_xdp_mode mode) 9035 { 9036 return dev->xdp_state[mode].link; 9037 } 9038 9039 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9040 enum bpf_xdp_mode mode) 9041 { 9042 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9043 9044 if (link) 9045 return link->link.prog; 9046 return dev->xdp_state[mode].prog; 9047 } 9048 9049 u8 dev_xdp_prog_count(struct net_device *dev) 9050 { 9051 u8 count = 0; 9052 int i; 9053 9054 for (i = 0; i < __MAX_XDP_MODE; i++) 9055 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9056 count++; 9057 return count; 9058 } 9059 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9060 9061 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9062 { 9063 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9064 9065 return prog ? prog->aux->id : 0; 9066 } 9067 9068 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9069 struct bpf_xdp_link *link) 9070 { 9071 dev->xdp_state[mode].link = link; 9072 dev->xdp_state[mode].prog = NULL; 9073 } 9074 9075 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9076 struct bpf_prog *prog) 9077 { 9078 dev->xdp_state[mode].link = NULL; 9079 dev->xdp_state[mode].prog = prog; 9080 } 9081 9082 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9083 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9084 u32 flags, struct bpf_prog *prog) 9085 { 9086 struct netdev_bpf xdp; 9087 int err; 9088 9089 memset(&xdp, 0, sizeof(xdp)); 9090 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9091 xdp.extack = extack; 9092 xdp.flags = flags; 9093 xdp.prog = prog; 9094 9095 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9096 * "moved" into driver), so they don't increment it on their own, but 9097 * they do decrement refcnt when program is detached or replaced. 9098 * Given net_device also owns link/prog, we need to bump refcnt here 9099 * to prevent drivers from underflowing it. 9100 */ 9101 if (prog) 9102 bpf_prog_inc(prog); 9103 err = bpf_op(dev, &xdp); 9104 if (err) { 9105 if (prog) 9106 bpf_prog_put(prog); 9107 return err; 9108 } 9109 9110 if (mode != XDP_MODE_HW) 9111 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9112 9113 return 0; 9114 } 9115 9116 static void dev_xdp_uninstall(struct net_device *dev) 9117 { 9118 struct bpf_xdp_link *link; 9119 struct bpf_prog *prog; 9120 enum bpf_xdp_mode mode; 9121 bpf_op_t bpf_op; 9122 9123 ASSERT_RTNL(); 9124 9125 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9126 prog = dev_xdp_prog(dev, mode); 9127 if (!prog) 9128 continue; 9129 9130 bpf_op = dev_xdp_bpf_op(dev, mode); 9131 if (!bpf_op) 9132 continue; 9133 9134 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9135 9136 /* auto-detach link from net device */ 9137 link = dev_xdp_link(dev, mode); 9138 if (link) 9139 link->dev = NULL; 9140 else 9141 bpf_prog_put(prog); 9142 9143 dev_xdp_set_link(dev, mode, NULL); 9144 } 9145 } 9146 9147 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9148 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9149 struct bpf_prog *old_prog, u32 flags) 9150 { 9151 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9152 struct bpf_prog *cur_prog; 9153 struct net_device *upper; 9154 struct list_head *iter; 9155 enum bpf_xdp_mode mode; 9156 bpf_op_t bpf_op; 9157 int err; 9158 9159 ASSERT_RTNL(); 9160 9161 /* either link or prog attachment, never both */ 9162 if (link && (new_prog || old_prog)) 9163 return -EINVAL; 9164 /* link supports only XDP mode flags */ 9165 if (link && (flags & ~XDP_FLAGS_MODES)) { 9166 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9167 return -EINVAL; 9168 } 9169 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9170 if (num_modes > 1) { 9171 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9172 return -EINVAL; 9173 } 9174 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9175 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9176 NL_SET_ERR_MSG(extack, 9177 "More than one program loaded, unset mode is ambiguous"); 9178 return -EINVAL; 9179 } 9180 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9181 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9182 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9183 return -EINVAL; 9184 } 9185 9186 mode = dev_xdp_mode(dev, flags); 9187 /* can't replace attached link */ 9188 if (dev_xdp_link(dev, mode)) { 9189 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9190 return -EBUSY; 9191 } 9192 9193 /* don't allow if an upper device already has a program */ 9194 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9195 if (dev_xdp_prog_count(upper) > 0) { 9196 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9197 return -EEXIST; 9198 } 9199 } 9200 9201 cur_prog = dev_xdp_prog(dev, mode); 9202 /* can't replace attached prog with link */ 9203 if (link && cur_prog) { 9204 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9205 return -EBUSY; 9206 } 9207 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9208 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9209 return -EEXIST; 9210 } 9211 9212 /* put effective new program into new_prog */ 9213 if (link) 9214 new_prog = link->link.prog; 9215 9216 if (new_prog) { 9217 bool offload = mode == XDP_MODE_HW; 9218 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9219 ? XDP_MODE_DRV : XDP_MODE_SKB; 9220 9221 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9222 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9223 return -EBUSY; 9224 } 9225 if (!offload && dev_xdp_prog(dev, other_mode)) { 9226 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9227 return -EEXIST; 9228 } 9229 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9230 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9231 return -EINVAL; 9232 } 9233 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9234 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9235 return -EINVAL; 9236 } 9237 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9238 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9239 return -EINVAL; 9240 } 9241 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9242 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9243 return -EINVAL; 9244 } 9245 } 9246 9247 /* don't call drivers if the effective program didn't change */ 9248 if (new_prog != cur_prog) { 9249 bpf_op = dev_xdp_bpf_op(dev, mode); 9250 if (!bpf_op) { 9251 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9252 return -EOPNOTSUPP; 9253 } 9254 9255 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9256 if (err) 9257 return err; 9258 } 9259 9260 if (link) 9261 dev_xdp_set_link(dev, mode, link); 9262 else 9263 dev_xdp_set_prog(dev, mode, new_prog); 9264 if (cur_prog) 9265 bpf_prog_put(cur_prog); 9266 9267 return 0; 9268 } 9269 9270 static int dev_xdp_attach_link(struct net_device *dev, 9271 struct netlink_ext_ack *extack, 9272 struct bpf_xdp_link *link) 9273 { 9274 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9275 } 9276 9277 static int dev_xdp_detach_link(struct net_device *dev, 9278 struct netlink_ext_ack *extack, 9279 struct bpf_xdp_link *link) 9280 { 9281 enum bpf_xdp_mode mode; 9282 bpf_op_t bpf_op; 9283 9284 ASSERT_RTNL(); 9285 9286 mode = dev_xdp_mode(dev, link->flags); 9287 if (dev_xdp_link(dev, mode) != link) 9288 return -EINVAL; 9289 9290 bpf_op = dev_xdp_bpf_op(dev, mode); 9291 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9292 dev_xdp_set_link(dev, mode, NULL); 9293 return 0; 9294 } 9295 9296 static void bpf_xdp_link_release(struct bpf_link *link) 9297 { 9298 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9299 9300 rtnl_lock(); 9301 9302 /* if racing with net_device's tear down, xdp_link->dev might be 9303 * already NULL, in which case link was already auto-detached 9304 */ 9305 if (xdp_link->dev) { 9306 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9307 xdp_link->dev = NULL; 9308 } 9309 9310 rtnl_unlock(); 9311 } 9312 9313 static int bpf_xdp_link_detach(struct bpf_link *link) 9314 { 9315 bpf_xdp_link_release(link); 9316 return 0; 9317 } 9318 9319 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9320 { 9321 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9322 9323 kfree(xdp_link); 9324 } 9325 9326 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9327 struct seq_file *seq) 9328 { 9329 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9330 u32 ifindex = 0; 9331 9332 rtnl_lock(); 9333 if (xdp_link->dev) 9334 ifindex = xdp_link->dev->ifindex; 9335 rtnl_unlock(); 9336 9337 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9338 } 9339 9340 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9341 struct bpf_link_info *info) 9342 { 9343 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9344 u32 ifindex = 0; 9345 9346 rtnl_lock(); 9347 if (xdp_link->dev) 9348 ifindex = xdp_link->dev->ifindex; 9349 rtnl_unlock(); 9350 9351 info->xdp.ifindex = ifindex; 9352 return 0; 9353 } 9354 9355 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9356 struct bpf_prog *old_prog) 9357 { 9358 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9359 enum bpf_xdp_mode mode; 9360 bpf_op_t bpf_op; 9361 int err = 0; 9362 9363 rtnl_lock(); 9364 9365 /* link might have been auto-released already, so fail */ 9366 if (!xdp_link->dev) { 9367 err = -ENOLINK; 9368 goto out_unlock; 9369 } 9370 9371 if (old_prog && link->prog != old_prog) { 9372 err = -EPERM; 9373 goto out_unlock; 9374 } 9375 old_prog = link->prog; 9376 if (old_prog->type != new_prog->type || 9377 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9378 err = -EINVAL; 9379 goto out_unlock; 9380 } 9381 9382 if (old_prog == new_prog) { 9383 /* no-op, don't disturb drivers */ 9384 bpf_prog_put(new_prog); 9385 goto out_unlock; 9386 } 9387 9388 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9389 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9390 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9391 xdp_link->flags, new_prog); 9392 if (err) 9393 goto out_unlock; 9394 9395 old_prog = xchg(&link->prog, new_prog); 9396 bpf_prog_put(old_prog); 9397 9398 out_unlock: 9399 rtnl_unlock(); 9400 return err; 9401 } 9402 9403 static const struct bpf_link_ops bpf_xdp_link_lops = { 9404 .release = bpf_xdp_link_release, 9405 .dealloc = bpf_xdp_link_dealloc, 9406 .detach = bpf_xdp_link_detach, 9407 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9408 .fill_link_info = bpf_xdp_link_fill_link_info, 9409 .update_prog = bpf_xdp_link_update, 9410 }; 9411 9412 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9413 { 9414 struct net *net = current->nsproxy->net_ns; 9415 struct bpf_link_primer link_primer; 9416 struct bpf_xdp_link *link; 9417 struct net_device *dev; 9418 int err, fd; 9419 9420 rtnl_lock(); 9421 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9422 if (!dev) { 9423 rtnl_unlock(); 9424 return -EINVAL; 9425 } 9426 9427 link = kzalloc(sizeof(*link), GFP_USER); 9428 if (!link) { 9429 err = -ENOMEM; 9430 goto unlock; 9431 } 9432 9433 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9434 link->dev = dev; 9435 link->flags = attr->link_create.flags; 9436 9437 err = bpf_link_prime(&link->link, &link_primer); 9438 if (err) { 9439 kfree(link); 9440 goto unlock; 9441 } 9442 9443 err = dev_xdp_attach_link(dev, NULL, link); 9444 rtnl_unlock(); 9445 9446 if (err) { 9447 link->dev = NULL; 9448 bpf_link_cleanup(&link_primer); 9449 goto out_put_dev; 9450 } 9451 9452 fd = bpf_link_settle(&link_primer); 9453 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9454 dev_put(dev); 9455 return fd; 9456 9457 unlock: 9458 rtnl_unlock(); 9459 9460 out_put_dev: 9461 dev_put(dev); 9462 return err; 9463 } 9464 9465 /** 9466 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9467 * @dev: device 9468 * @extack: netlink extended ack 9469 * @fd: new program fd or negative value to clear 9470 * @expected_fd: old program fd that userspace expects to replace or clear 9471 * @flags: xdp-related flags 9472 * 9473 * Set or clear a bpf program for a device 9474 */ 9475 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9476 int fd, int expected_fd, u32 flags) 9477 { 9478 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9479 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9480 int err; 9481 9482 ASSERT_RTNL(); 9483 9484 if (fd >= 0) { 9485 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9486 mode != XDP_MODE_SKB); 9487 if (IS_ERR(new_prog)) 9488 return PTR_ERR(new_prog); 9489 } 9490 9491 if (expected_fd >= 0) { 9492 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9493 mode != XDP_MODE_SKB); 9494 if (IS_ERR(old_prog)) { 9495 err = PTR_ERR(old_prog); 9496 old_prog = NULL; 9497 goto err_out; 9498 } 9499 } 9500 9501 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9502 9503 err_out: 9504 if (err && new_prog) 9505 bpf_prog_put(new_prog); 9506 if (old_prog) 9507 bpf_prog_put(old_prog); 9508 return err; 9509 } 9510 9511 /** 9512 * dev_new_index - allocate an ifindex 9513 * @net: the applicable net namespace 9514 * 9515 * Returns a suitable unique value for a new device interface 9516 * number. The caller must hold the rtnl semaphore or the 9517 * dev_base_lock to be sure it remains unique. 9518 */ 9519 static int dev_new_index(struct net *net) 9520 { 9521 int ifindex = net->ifindex; 9522 9523 for (;;) { 9524 if (++ifindex <= 0) 9525 ifindex = 1; 9526 if (!__dev_get_by_index(net, ifindex)) 9527 return net->ifindex = ifindex; 9528 } 9529 } 9530 9531 /* Delayed registration/unregisteration */ 9532 LIST_HEAD(net_todo_list); 9533 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9534 9535 static void net_set_todo(struct net_device *dev) 9536 { 9537 list_add_tail(&dev->todo_list, &net_todo_list); 9538 atomic_inc(&dev_net(dev)->dev_unreg_count); 9539 } 9540 9541 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9542 struct net_device *upper, netdev_features_t features) 9543 { 9544 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9545 netdev_features_t feature; 9546 int feature_bit; 9547 9548 for_each_netdev_feature(upper_disables, feature_bit) { 9549 feature = __NETIF_F_BIT(feature_bit); 9550 if (!(upper->wanted_features & feature) 9551 && (features & feature)) { 9552 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9553 &feature, upper->name); 9554 features &= ~feature; 9555 } 9556 } 9557 9558 return features; 9559 } 9560 9561 static void netdev_sync_lower_features(struct net_device *upper, 9562 struct net_device *lower, 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 (!(features & feature) && (lower->features & feature)) { 9571 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9572 &feature, lower->name); 9573 lower->wanted_features &= ~feature; 9574 __netdev_update_features(lower); 9575 9576 if (unlikely(lower->features & feature)) 9577 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9578 &feature, lower->name); 9579 else 9580 netdev_features_change(lower); 9581 } 9582 } 9583 } 9584 9585 static netdev_features_t netdev_fix_features(struct net_device *dev, 9586 netdev_features_t features) 9587 { 9588 /* Fix illegal checksum combinations */ 9589 if ((features & NETIF_F_HW_CSUM) && 9590 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9591 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9592 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9593 } 9594 9595 /* TSO requires that SG is present as well. */ 9596 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9597 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9598 features &= ~NETIF_F_ALL_TSO; 9599 } 9600 9601 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9602 !(features & NETIF_F_IP_CSUM)) { 9603 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9604 features &= ~NETIF_F_TSO; 9605 features &= ~NETIF_F_TSO_ECN; 9606 } 9607 9608 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9609 !(features & NETIF_F_IPV6_CSUM)) { 9610 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9611 features &= ~NETIF_F_TSO6; 9612 } 9613 9614 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9615 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9616 features &= ~NETIF_F_TSO_MANGLEID; 9617 9618 /* TSO ECN requires that TSO is present as well. */ 9619 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9620 features &= ~NETIF_F_TSO_ECN; 9621 9622 /* Software GSO depends on SG. */ 9623 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9624 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9625 features &= ~NETIF_F_GSO; 9626 } 9627 9628 /* GSO partial features require GSO partial be set */ 9629 if ((features & dev->gso_partial_features) && 9630 !(features & NETIF_F_GSO_PARTIAL)) { 9631 netdev_dbg(dev, 9632 "Dropping partially supported GSO features since no GSO partial.\n"); 9633 features &= ~dev->gso_partial_features; 9634 } 9635 9636 if (!(features & NETIF_F_RXCSUM)) { 9637 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9638 * successfully merged by hardware must also have the 9639 * checksum verified by hardware. If the user does not 9640 * want to enable RXCSUM, logically, we should disable GRO_HW. 9641 */ 9642 if (features & NETIF_F_GRO_HW) { 9643 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9644 features &= ~NETIF_F_GRO_HW; 9645 } 9646 } 9647 9648 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9649 if (features & NETIF_F_RXFCS) { 9650 if (features & NETIF_F_LRO) { 9651 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9652 features &= ~NETIF_F_LRO; 9653 } 9654 9655 if (features & NETIF_F_GRO_HW) { 9656 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9657 features &= ~NETIF_F_GRO_HW; 9658 } 9659 } 9660 9661 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9662 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9663 features &= ~NETIF_F_LRO; 9664 } 9665 9666 if (features & NETIF_F_HW_TLS_TX) { 9667 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9668 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9669 bool hw_csum = features & NETIF_F_HW_CSUM; 9670 9671 if (!ip_csum && !hw_csum) { 9672 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9673 features &= ~NETIF_F_HW_TLS_TX; 9674 } 9675 } 9676 9677 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9678 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9679 features &= ~NETIF_F_HW_TLS_RX; 9680 } 9681 9682 return features; 9683 } 9684 9685 int __netdev_update_features(struct net_device *dev) 9686 { 9687 struct net_device *upper, *lower; 9688 netdev_features_t features; 9689 struct list_head *iter; 9690 int err = -1; 9691 9692 ASSERT_RTNL(); 9693 9694 features = netdev_get_wanted_features(dev); 9695 9696 if (dev->netdev_ops->ndo_fix_features) 9697 features = dev->netdev_ops->ndo_fix_features(dev, features); 9698 9699 /* driver might be less strict about feature dependencies */ 9700 features = netdev_fix_features(dev, features); 9701 9702 /* some features can't be enabled if they're off on an upper device */ 9703 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9704 features = netdev_sync_upper_features(dev, upper, features); 9705 9706 if (dev->features == features) 9707 goto sync_lower; 9708 9709 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9710 &dev->features, &features); 9711 9712 if (dev->netdev_ops->ndo_set_features) 9713 err = dev->netdev_ops->ndo_set_features(dev, features); 9714 else 9715 err = 0; 9716 9717 if (unlikely(err < 0)) { 9718 netdev_err(dev, 9719 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9720 err, &features, &dev->features); 9721 /* return non-0 since some features might have changed and 9722 * it's better to fire a spurious notification than miss it 9723 */ 9724 return -1; 9725 } 9726 9727 sync_lower: 9728 /* some features must be disabled on lower devices when disabled 9729 * on an upper device (think: bonding master or bridge) 9730 */ 9731 netdev_for_each_lower_dev(dev, lower, iter) 9732 netdev_sync_lower_features(dev, lower, features); 9733 9734 if (!err) { 9735 netdev_features_t diff = features ^ dev->features; 9736 9737 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9738 /* udp_tunnel_{get,drop}_rx_info both need 9739 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9740 * device, or they won't do anything. 9741 * Thus we need to update dev->features 9742 * *before* calling udp_tunnel_get_rx_info, 9743 * but *after* calling udp_tunnel_drop_rx_info. 9744 */ 9745 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9746 dev->features = features; 9747 udp_tunnel_get_rx_info(dev); 9748 } else { 9749 udp_tunnel_drop_rx_info(dev); 9750 } 9751 } 9752 9753 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9754 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9755 dev->features = features; 9756 err |= vlan_get_rx_ctag_filter_info(dev); 9757 } else { 9758 vlan_drop_rx_ctag_filter_info(dev); 9759 } 9760 } 9761 9762 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9763 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9764 dev->features = features; 9765 err |= vlan_get_rx_stag_filter_info(dev); 9766 } else { 9767 vlan_drop_rx_stag_filter_info(dev); 9768 } 9769 } 9770 9771 dev->features = features; 9772 } 9773 9774 return err < 0 ? 0 : 1; 9775 } 9776 9777 /** 9778 * netdev_update_features - recalculate device features 9779 * @dev: the device to check 9780 * 9781 * Recalculate dev->features set and send notifications if it 9782 * has changed. Should be called after driver or hardware dependent 9783 * conditions might have changed that influence the features. 9784 */ 9785 void netdev_update_features(struct net_device *dev) 9786 { 9787 if (__netdev_update_features(dev)) 9788 netdev_features_change(dev); 9789 } 9790 EXPORT_SYMBOL(netdev_update_features); 9791 9792 /** 9793 * netdev_change_features - recalculate device features 9794 * @dev: the device to check 9795 * 9796 * Recalculate dev->features set and send notifications even 9797 * if they have not changed. Should be called instead of 9798 * netdev_update_features() if also dev->vlan_features might 9799 * have changed to allow the changes to be propagated to stacked 9800 * VLAN devices. 9801 */ 9802 void netdev_change_features(struct net_device *dev) 9803 { 9804 __netdev_update_features(dev); 9805 netdev_features_change(dev); 9806 } 9807 EXPORT_SYMBOL(netdev_change_features); 9808 9809 /** 9810 * netif_stacked_transfer_operstate - transfer operstate 9811 * @rootdev: the root or lower level device to transfer state from 9812 * @dev: the device to transfer operstate to 9813 * 9814 * Transfer operational state from root to device. This is normally 9815 * called when a stacking relationship exists between the root 9816 * device and the device(a leaf device). 9817 */ 9818 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9819 struct net_device *dev) 9820 { 9821 if (rootdev->operstate == IF_OPER_DORMANT) 9822 netif_dormant_on(dev); 9823 else 9824 netif_dormant_off(dev); 9825 9826 if (rootdev->operstate == IF_OPER_TESTING) 9827 netif_testing_on(dev); 9828 else 9829 netif_testing_off(dev); 9830 9831 if (netif_carrier_ok(rootdev)) 9832 netif_carrier_on(dev); 9833 else 9834 netif_carrier_off(dev); 9835 } 9836 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9837 9838 static int netif_alloc_rx_queues(struct net_device *dev) 9839 { 9840 unsigned int i, count = dev->num_rx_queues; 9841 struct netdev_rx_queue *rx; 9842 size_t sz = count * sizeof(*rx); 9843 int err = 0; 9844 9845 BUG_ON(count < 1); 9846 9847 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9848 if (!rx) 9849 return -ENOMEM; 9850 9851 dev->_rx = rx; 9852 9853 for (i = 0; i < count; i++) { 9854 rx[i].dev = dev; 9855 9856 /* XDP RX-queue setup */ 9857 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9858 if (err < 0) 9859 goto err_rxq_info; 9860 } 9861 return 0; 9862 9863 err_rxq_info: 9864 /* Rollback successful reg's and free other resources */ 9865 while (i--) 9866 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9867 kvfree(dev->_rx); 9868 dev->_rx = NULL; 9869 return err; 9870 } 9871 9872 static void netif_free_rx_queues(struct net_device *dev) 9873 { 9874 unsigned int i, count = dev->num_rx_queues; 9875 9876 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9877 if (!dev->_rx) 9878 return; 9879 9880 for (i = 0; i < count; i++) 9881 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9882 9883 kvfree(dev->_rx); 9884 } 9885 9886 static void netdev_init_one_queue(struct net_device *dev, 9887 struct netdev_queue *queue, void *_unused) 9888 { 9889 /* Initialize queue lock */ 9890 spin_lock_init(&queue->_xmit_lock); 9891 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9892 queue->xmit_lock_owner = -1; 9893 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9894 queue->dev = dev; 9895 #ifdef CONFIG_BQL 9896 dql_init(&queue->dql, HZ); 9897 #endif 9898 } 9899 9900 static void netif_free_tx_queues(struct net_device *dev) 9901 { 9902 kvfree(dev->_tx); 9903 } 9904 9905 static int netif_alloc_netdev_queues(struct net_device *dev) 9906 { 9907 unsigned int count = dev->num_tx_queues; 9908 struct netdev_queue *tx; 9909 size_t sz = count * sizeof(*tx); 9910 9911 if (count < 1 || count > 0xffff) 9912 return -EINVAL; 9913 9914 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9915 if (!tx) 9916 return -ENOMEM; 9917 9918 dev->_tx = tx; 9919 9920 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9921 spin_lock_init(&dev->tx_global_lock); 9922 9923 return 0; 9924 } 9925 9926 void netif_tx_stop_all_queues(struct net_device *dev) 9927 { 9928 unsigned int i; 9929 9930 for (i = 0; i < dev->num_tx_queues; i++) { 9931 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9932 9933 netif_tx_stop_queue(txq); 9934 } 9935 } 9936 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9937 9938 /** 9939 * register_netdevice() - register a network device 9940 * @dev: device to register 9941 * 9942 * Take a prepared network device structure and make it externally accessible. 9943 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9944 * Callers must hold the rtnl lock - you may want register_netdev() 9945 * instead of this. 9946 */ 9947 int register_netdevice(struct net_device *dev) 9948 { 9949 int ret; 9950 struct net *net = dev_net(dev); 9951 9952 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9953 NETDEV_FEATURE_COUNT); 9954 BUG_ON(dev_boot_phase); 9955 ASSERT_RTNL(); 9956 9957 might_sleep(); 9958 9959 /* When net_device's are persistent, this will be fatal. */ 9960 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9961 BUG_ON(!net); 9962 9963 ret = ethtool_check_ops(dev->ethtool_ops); 9964 if (ret) 9965 return ret; 9966 9967 spin_lock_init(&dev->addr_list_lock); 9968 netdev_set_addr_lockdep_class(dev); 9969 9970 ret = dev_get_valid_name(net, dev, dev->name); 9971 if (ret < 0) 9972 goto out; 9973 9974 ret = -ENOMEM; 9975 dev->name_node = netdev_name_node_head_alloc(dev); 9976 if (!dev->name_node) 9977 goto out; 9978 9979 /* Init, if this function is available */ 9980 if (dev->netdev_ops->ndo_init) { 9981 ret = dev->netdev_ops->ndo_init(dev); 9982 if (ret) { 9983 if (ret > 0) 9984 ret = -EIO; 9985 goto err_free_name; 9986 } 9987 } 9988 9989 if (((dev->hw_features | dev->features) & 9990 NETIF_F_HW_VLAN_CTAG_FILTER) && 9991 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9992 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9993 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9994 ret = -EINVAL; 9995 goto err_uninit; 9996 } 9997 9998 ret = -EBUSY; 9999 if (!dev->ifindex) 10000 dev->ifindex = dev_new_index(net); 10001 else if (__dev_get_by_index(net, dev->ifindex)) 10002 goto err_uninit; 10003 10004 /* Transfer changeable features to wanted_features and enable 10005 * software offloads (GSO and GRO). 10006 */ 10007 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10008 dev->features |= NETIF_F_SOFT_FEATURES; 10009 10010 if (dev->udp_tunnel_nic_info) { 10011 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10012 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10013 } 10014 10015 dev->wanted_features = dev->features & dev->hw_features; 10016 10017 if (!(dev->flags & IFF_LOOPBACK)) 10018 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10019 10020 /* If IPv4 TCP segmentation offload is supported we should also 10021 * allow the device to enable segmenting the frame with the option 10022 * of ignoring a static IP ID value. This doesn't enable the 10023 * feature itself but allows the user to enable it later. 10024 */ 10025 if (dev->hw_features & NETIF_F_TSO) 10026 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10027 if (dev->vlan_features & NETIF_F_TSO) 10028 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10029 if (dev->mpls_features & NETIF_F_TSO) 10030 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10031 if (dev->hw_enc_features & NETIF_F_TSO) 10032 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10033 10034 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10035 */ 10036 dev->vlan_features |= NETIF_F_HIGHDMA; 10037 10038 /* Make NETIF_F_SG inheritable to tunnel devices. 10039 */ 10040 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10041 10042 /* Make NETIF_F_SG inheritable to MPLS. 10043 */ 10044 dev->mpls_features |= NETIF_F_SG; 10045 10046 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10047 ret = notifier_to_errno(ret); 10048 if (ret) 10049 goto err_uninit; 10050 10051 ret = netdev_register_kobject(dev); 10052 write_lock(&dev_base_lock); 10053 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10054 write_unlock(&dev_base_lock); 10055 if (ret) 10056 goto err_uninit_notify; 10057 10058 __netdev_update_features(dev); 10059 10060 /* 10061 * Default initial state at registry is that the 10062 * device is present. 10063 */ 10064 10065 set_bit(__LINK_STATE_PRESENT, &dev->state); 10066 10067 linkwatch_init_dev(dev); 10068 10069 dev_init_scheduler(dev); 10070 10071 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10072 list_netdevice(dev); 10073 10074 add_device_randomness(dev->dev_addr, dev->addr_len); 10075 10076 /* If the device has permanent device address, driver should 10077 * set dev_addr and also addr_assign_type should be set to 10078 * NET_ADDR_PERM (default value). 10079 */ 10080 if (dev->addr_assign_type == NET_ADDR_PERM) 10081 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10082 10083 /* Notify protocols, that a new device appeared. */ 10084 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10085 ret = notifier_to_errno(ret); 10086 if (ret) { 10087 /* Expect explicit free_netdev() on failure */ 10088 dev->needs_free_netdev = false; 10089 unregister_netdevice_queue(dev, NULL); 10090 goto out; 10091 } 10092 /* 10093 * Prevent userspace races by waiting until the network 10094 * device is fully setup before sending notifications. 10095 */ 10096 if (!dev->rtnl_link_ops || 10097 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10098 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10099 10100 out: 10101 return ret; 10102 10103 err_uninit_notify: 10104 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10105 err_uninit: 10106 if (dev->netdev_ops->ndo_uninit) 10107 dev->netdev_ops->ndo_uninit(dev); 10108 if (dev->priv_destructor) 10109 dev->priv_destructor(dev); 10110 err_free_name: 10111 netdev_name_node_free(dev->name_node); 10112 goto out; 10113 } 10114 EXPORT_SYMBOL(register_netdevice); 10115 10116 /** 10117 * init_dummy_netdev - init a dummy network device for NAPI 10118 * @dev: device to init 10119 * 10120 * This takes a network device structure and initialize the minimum 10121 * amount of fields so it can be used to schedule NAPI polls without 10122 * registering a full blown interface. This is to be used by drivers 10123 * that need to tie several hardware interfaces to a single NAPI 10124 * poll scheduler due to HW limitations. 10125 */ 10126 int init_dummy_netdev(struct net_device *dev) 10127 { 10128 /* Clear everything. Note we don't initialize spinlocks 10129 * are they aren't supposed to be taken by any of the 10130 * NAPI code and this dummy netdev is supposed to be 10131 * only ever used for NAPI polls 10132 */ 10133 memset(dev, 0, sizeof(struct net_device)); 10134 10135 /* make sure we BUG if trying to hit standard 10136 * register/unregister code path 10137 */ 10138 dev->reg_state = NETREG_DUMMY; 10139 10140 /* NAPI wants this */ 10141 INIT_LIST_HEAD(&dev->napi_list); 10142 10143 /* a dummy interface is started by default */ 10144 set_bit(__LINK_STATE_PRESENT, &dev->state); 10145 set_bit(__LINK_STATE_START, &dev->state); 10146 10147 /* napi_busy_loop stats accounting wants this */ 10148 dev_net_set(dev, &init_net); 10149 10150 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10151 * because users of this 'device' dont need to change 10152 * its refcount. 10153 */ 10154 10155 return 0; 10156 } 10157 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10158 10159 10160 /** 10161 * register_netdev - register a network device 10162 * @dev: device to register 10163 * 10164 * Take a completed network device structure and add it to the kernel 10165 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10166 * chain. 0 is returned on success. A negative errno code is returned 10167 * on a failure to set up the device, or if the name is a duplicate. 10168 * 10169 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10170 * and expands the device name if you passed a format string to 10171 * alloc_netdev. 10172 */ 10173 int register_netdev(struct net_device *dev) 10174 { 10175 int err; 10176 10177 if (rtnl_lock_killable()) 10178 return -EINTR; 10179 err = register_netdevice(dev); 10180 rtnl_unlock(); 10181 return err; 10182 } 10183 EXPORT_SYMBOL(register_netdev); 10184 10185 int netdev_refcnt_read(const struct net_device *dev) 10186 { 10187 #ifdef CONFIG_PCPU_DEV_REFCNT 10188 int i, refcnt = 0; 10189 10190 for_each_possible_cpu(i) 10191 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10192 return refcnt; 10193 #else 10194 return refcount_read(&dev->dev_refcnt); 10195 #endif 10196 } 10197 EXPORT_SYMBOL(netdev_refcnt_read); 10198 10199 int netdev_unregister_timeout_secs __read_mostly = 10; 10200 10201 #define WAIT_REFS_MIN_MSECS 1 10202 #define WAIT_REFS_MAX_MSECS 250 10203 /** 10204 * netdev_wait_allrefs_any - wait until all references are gone. 10205 * @list: list of net_devices to wait on 10206 * 10207 * This is called when unregistering network devices. 10208 * 10209 * Any protocol or device that holds a reference should register 10210 * for netdevice notification, and cleanup and put back the 10211 * reference if they receive an UNREGISTER event. 10212 * We can get stuck here if buggy protocols don't correctly 10213 * call dev_put. 10214 */ 10215 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10216 { 10217 unsigned long rebroadcast_time, warning_time; 10218 struct net_device *dev; 10219 int wait = 0; 10220 10221 rebroadcast_time = warning_time = jiffies; 10222 10223 list_for_each_entry(dev, list, todo_list) 10224 if (netdev_refcnt_read(dev) == 1) 10225 return dev; 10226 10227 while (true) { 10228 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10229 rtnl_lock(); 10230 10231 /* Rebroadcast unregister notification */ 10232 list_for_each_entry(dev, list, todo_list) 10233 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10234 10235 __rtnl_unlock(); 10236 rcu_barrier(); 10237 rtnl_lock(); 10238 10239 list_for_each_entry(dev, list, todo_list) 10240 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10241 &dev->state)) { 10242 /* We must not have linkwatch events 10243 * pending on unregister. If this 10244 * happens, we simply run the queue 10245 * unscheduled, resulting in a noop 10246 * for this device. 10247 */ 10248 linkwatch_run_queue(); 10249 break; 10250 } 10251 10252 __rtnl_unlock(); 10253 10254 rebroadcast_time = jiffies; 10255 } 10256 10257 if (!wait) { 10258 rcu_barrier(); 10259 wait = WAIT_REFS_MIN_MSECS; 10260 } else { 10261 msleep(wait); 10262 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10263 } 10264 10265 list_for_each_entry(dev, list, todo_list) 10266 if (netdev_refcnt_read(dev) == 1) 10267 return dev; 10268 10269 if (time_after(jiffies, warning_time + 10270 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10271 list_for_each_entry(dev, list, todo_list) { 10272 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10273 dev->name, netdev_refcnt_read(dev)); 10274 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10275 } 10276 10277 warning_time = jiffies; 10278 } 10279 } 10280 } 10281 10282 /* The sequence is: 10283 * 10284 * rtnl_lock(); 10285 * ... 10286 * register_netdevice(x1); 10287 * register_netdevice(x2); 10288 * ... 10289 * unregister_netdevice(y1); 10290 * unregister_netdevice(y2); 10291 * ... 10292 * rtnl_unlock(); 10293 * free_netdev(y1); 10294 * free_netdev(y2); 10295 * 10296 * We are invoked by rtnl_unlock(). 10297 * This allows us to deal with problems: 10298 * 1) We can delete sysfs objects which invoke hotplug 10299 * without deadlocking with linkwatch via keventd. 10300 * 2) Since we run with the RTNL semaphore not held, we can sleep 10301 * safely in order to wait for the netdev refcnt to drop to zero. 10302 * 10303 * We must not return until all unregister events added during 10304 * the interval the lock was held have been completed. 10305 */ 10306 void netdev_run_todo(void) 10307 { 10308 struct net_device *dev, *tmp; 10309 struct list_head list; 10310 #ifdef CONFIG_LOCKDEP 10311 struct list_head unlink_list; 10312 10313 list_replace_init(&net_unlink_list, &unlink_list); 10314 10315 while (!list_empty(&unlink_list)) { 10316 struct net_device *dev = list_first_entry(&unlink_list, 10317 struct net_device, 10318 unlink_list); 10319 list_del_init(&dev->unlink_list); 10320 dev->nested_level = dev->lower_level - 1; 10321 } 10322 #endif 10323 10324 /* Snapshot list, allow later requests */ 10325 list_replace_init(&net_todo_list, &list); 10326 10327 __rtnl_unlock(); 10328 10329 /* Wait for rcu callbacks to finish before next phase */ 10330 if (!list_empty(&list)) 10331 rcu_barrier(); 10332 10333 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10334 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10335 netdev_WARN(dev, "run_todo but not unregistering\n"); 10336 list_del(&dev->todo_list); 10337 continue; 10338 } 10339 10340 write_lock(&dev_base_lock); 10341 dev->reg_state = NETREG_UNREGISTERED; 10342 write_unlock(&dev_base_lock); 10343 linkwatch_forget_dev(dev); 10344 } 10345 10346 while (!list_empty(&list)) { 10347 dev = netdev_wait_allrefs_any(&list); 10348 list_del(&dev->todo_list); 10349 10350 /* paranoia */ 10351 BUG_ON(netdev_refcnt_read(dev) != 1); 10352 BUG_ON(!list_empty(&dev->ptype_all)); 10353 BUG_ON(!list_empty(&dev->ptype_specific)); 10354 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10355 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10356 10357 if (dev->priv_destructor) 10358 dev->priv_destructor(dev); 10359 if (dev->needs_free_netdev) 10360 free_netdev(dev); 10361 10362 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10363 wake_up(&netdev_unregistering_wq); 10364 10365 /* Free network device */ 10366 kobject_put(&dev->dev.kobj); 10367 } 10368 } 10369 10370 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10371 * all the same fields in the same order as net_device_stats, with only 10372 * the type differing, but rtnl_link_stats64 may have additional fields 10373 * at the end for newer counters. 10374 */ 10375 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10376 const struct net_device_stats *netdev_stats) 10377 { 10378 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10379 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10380 u64 *dst = (u64 *)stats64; 10381 10382 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10383 for (i = 0; i < n; i++) 10384 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10385 /* zero out counters that only exist in rtnl_link_stats64 */ 10386 memset((char *)stats64 + n * sizeof(u64), 0, 10387 sizeof(*stats64) - n * sizeof(u64)); 10388 } 10389 EXPORT_SYMBOL(netdev_stats_to_stats64); 10390 10391 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10392 { 10393 struct net_device_core_stats __percpu *p; 10394 10395 p = alloc_percpu_gfp(struct net_device_core_stats, 10396 GFP_ATOMIC | __GFP_NOWARN); 10397 10398 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10399 free_percpu(p); 10400 10401 /* This READ_ONCE() pairs with the cmpxchg() above */ 10402 return READ_ONCE(dev->core_stats); 10403 } 10404 EXPORT_SYMBOL(netdev_core_stats_alloc); 10405 10406 /** 10407 * dev_get_stats - get network device statistics 10408 * @dev: device to get statistics from 10409 * @storage: place to store stats 10410 * 10411 * Get network statistics from device. Return @storage. 10412 * The device driver may provide its own method by setting 10413 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10414 * otherwise the internal statistics structure is used. 10415 */ 10416 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10417 struct rtnl_link_stats64 *storage) 10418 { 10419 const struct net_device_ops *ops = dev->netdev_ops; 10420 const struct net_device_core_stats __percpu *p; 10421 10422 if (ops->ndo_get_stats64) { 10423 memset(storage, 0, sizeof(*storage)); 10424 ops->ndo_get_stats64(dev, storage); 10425 } else if (ops->ndo_get_stats) { 10426 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10427 } else { 10428 netdev_stats_to_stats64(storage, &dev->stats); 10429 } 10430 10431 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10432 p = READ_ONCE(dev->core_stats); 10433 if (p) { 10434 const struct net_device_core_stats *core_stats; 10435 int i; 10436 10437 for_each_possible_cpu(i) { 10438 core_stats = per_cpu_ptr(p, i); 10439 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10440 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10441 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10442 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10443 } 10444 } 10445 return storage; 10446 } 10447 EXPORT_SYMBOL(dev_get_stats); 10448 10449 /** 10450 * dev_fetch_sw_netstats - get per-cpu network device statistics 10451 * @s: place to store stats 10452 * @netstats: per-cpu network stats to read from 10453 * 10454 * Read per-cpu network statistics and populate the related fields in @s. 10455 */ 10456 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10457 const struct pcpu_sw_netstats __percpu *netstats) 10458 { 10459 int cpu; 10460 10461 for_each_possible_cpu(cpu) { 10462 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10463 const struct pcpu_sw_netstats *stats; 10464 unsigned int start; 10465 10466 stats = per_cpu_ptr(netstats, cpu); 10467 do { 10468 start = u64_stats_fetch_begin(&stats->syncp); 10469 rx_packets = u64_stats_read(&stats->rx_packets); 10470 rx_bytes = u64_stats_read(&stats->rx_bytes); 10471 tx_packets = u64_stats_read(&stats->tx_packets); 10472 tx_bytes = u64_stats_read(&stats->tx_bytes); 10473 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10474 10475 s->rx_packets += rx_packets; 10476 s->rx_bytes += rx_bytes; 10477 s->tx_packets += tx_packets; 10478 s->tx_bytes += tx_bytes; 10479 } 10480 } 10481 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10482 10483 /** 10484 * dev_get_tstats64 - ndo_get_stats64 implementation 10485 * @dev: device to get statistics from 10486 * @s: place to store stats 10487 * 10488 * Populate @s from dev->stats and dev->tstats. Can be used as 10489 * ndo_get_stats64() callback. 10490 */ 10491 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10492 { 10493 netdev_stats_to_stats64(s, &dev->stats); 10494 dev_fetch_sw_netstats(s, dev->tstats); 10495 } 10496 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10497 10498 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10499 { 10500 struct netdev_queue *queue = dev_ingress_queue(dev); 10501 10502 #ifdef CONFIG_NET_CLS_ACT 10503 if (queue) 10504 return queue; 10505 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10506 if (!queue) 10507 return NULL; 10508 netdev_init_one_queue(dev, queue, NULL); 10509 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10510 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10511 rcu_assign_pointer(dev->ingress_queue, queue); 10512 #endif 10513 return queue; 10514 } 10515 10516 static const struct ethtool_ops default_ethtool_ops; 10517 10518 void netdev_set_default_ethtool_ops(struct net_device *dev, 10519 const struct ethtool_ops *ops) 10520 { 10521 if (dev->ethtool_ops == &default_ethtool_ops) 10522 dev->ethtool_ops = ops; 10523 } 10524 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10525 10526 /** 10527 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10528 * @dev: netdev to enable the IRQ coalescing on 10529 * 10530 * Sets a conservative default for SW IRQ coalescing. Users can use 10531 * sysfs attributes to override the default values. 10532 */ 10533 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10534 { 10535 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10536 10537 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10538 dev->gro_flush_timeout = 20000; 10539 dev->napi_defer_hard_irqs = 1; 10540 } 10541 } 10542 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10543 10544 void netdev_freemem(struct net_device *dev) 10545 { 10546 char *addr = (char *)dev - dev->padded; 10547 10548 kvfree(addr); 10549 } 10550 10551 /** 10552 * alloc_netdev_mqs - allocate network device 10553 * @sizeof_priv: size of private data to allocate space for 10554 * @name: device name format string 10555 * @name_assign_type: origin of device name 10556 * @setup: callback to initialize device 10557 * @txqs: the number of TX subqueues to allocate 10558 * @rxqs: the number of RX subqueues to allocate 10559 * 10560 * Allocates a struct net_device with private data area for driver use 10561 * and performs basic initialization. Also allocates subqueue structs 10562 * for each queue on the device. 10563 */ 10564 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10565 unsigned char name_assign_type, 10566 void (*setup)(struct net_device *), 10567 unsigned int txqs, unsigned int rxqs) 10568 { 10569 struct net_device *dev; 10570 unsigned int alloc_size; 10571 struct net_device *p; 10572 10573 BUG_ON(strlen(name) >= sizeof(dev->name)); 10574 10575 if (txqs < 1) { 10576 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10577 return NULL; 10578 } 10579 10580 if (rxqs < 1) { 10581 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10582 return NULL; 10583 } 10584 10585 alloc_size = sizeof(struct net_device); 10586 if (sizeof_priv) { 10587 /* ensure 32-byte alignment of private area */ 10588 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10589 alloc_size += sizeof_priv; 10590 } 10591 /* ensure 32-byte alignment of whole construct */ 10592 alloc_size += NETDEV_ALIGN - 1; 10593 10594 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10595 if (!p) 10596 return NULL; 10597 10598 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10599 dev->padded = (char *)dev - (char *)p; 10600 10601 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10602 #ifdef CONFIG_PCPU_DEV_REFCNT 10603 dev->pcpu_refcnt = alloc_percpu(int); 10604 if (!dev->pcpu_refcnt) 10605 goto free_dev; 10606 __dev_hold(dev); 10607 #else 10608 refcount_set(&dev->dev_refcnt, 1); 10609 #endif 10610 10611 if (dev_addr_init(dev)) 10612 goto free_pcpu; 10613 10614 dev_mc_init(dev); 10615 dev_uc_init(dev); 10616 10617 dev_net_set(dev, &init_net); 10618 10619 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10620 dev->gso_max_segs = GSO_MAX_SEGS; 10621 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10622 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10623 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10624 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10625 dev->tso_max_segs = TSO_MAX_SEGS; 10626 dev->upper_level = 1; 10627 dev->lower_level = 1; 10628 #ifdef CONFIG_LOCKDEP 10629 dev->nested_level = 0; 10630 INIT_LIST_HEAD(&dev->unlink_list); 10631 #endif 10632 10633 INIT_LIST_HEAD(&dev->napi_list); 10634 INIT_LIST_HEAD(&dev->unreg_list); 10635 INIT_LIST_HEAD(&dev->close_list); 10636 INIT_LIST_HEAD(&dev->link_watch_list); 10637 INIT_LIST_HEAD(&dev->adj_list.upper); 10638 INIT_LIST_HEAD(&dev->adj_list.lower); 10639 INIT_LIST_HEAD(&dev->ptype_all); 10640 INIT_LIST_HEAD(&dev->ptype_specific); 10641 INIT_LIST_HEAD(&dev->net_notifier_list); 10642 #ifdef CONFIG_NET_SCHED 10643 hash_init(dev->qdisc_hash); 10644 #endif 10645 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10646 setup(dev); 10647 10648 if (!dev->tx_queue_len) { 10649 dev->priv_flags |= IFF_NO_QUEUE; 10650 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10651 } 10652 10653 dev->num_tx_queues = txqs; 10654 dev->real_num_tx_queues = txqs; 10655 if (netif_alloc_netdev_queues(dev)) 10656 goto free_all; 10657 10658 dev->num_rx_queues = rxqs; 10659 dev->real_num_rx_queues = rxqs; 10660 if (netif_alloc_rx_queues(dev)) 10661 goto free_all; 10662 10663 strcpy(dev->name, name); 10664 dev->name_assign_type = name_assign_type; 10665 dev->group = INIT_NETDEV_GROUP; 10666 if (!dev->ethtool_ops) 10667 dev->ethtool_ops = &default_ethtool_ops; 10668 10669 nf_hook_netdev_init(dev); 10670 10671 return dev; 10672 10673 free_all: 10674 free_netdev(dev); 10675 return NULL; 10676 10677 free_pcpu: 10678 #ifdef CONFIG_PCPU_DEV_REFCNT 10679 free_percpu(dev->pcpu_refcnt); 10680 free_dev: 10681 #endif 10682 netdev_freemem(dev); 10683 return NULL; 10684 } 10685 EXPORT_SYMBOL(alloc_netdev_mqs); 10686 10687 /** 10688 * free_netdev - free network device 10689 * @dev: device 10690 * 10691 * This function does the last stage of destroying an allocated device 10692 * interface. The reference to the device object is released. If this 10693 * is the last reference then it will be freed.Must be called in process 10694 * context. 10695 */ 10696 void free_netdev(struct net_device *dev) 10697 { 10698 struct napi_struct *p, *n; 10699 10700 might_sleep(); 10701 10702 /* When called immediately after register_netdevice() failed the unwind 10703 * handling may still be dismantling the device. Handle that case by 10704 * deferring the free. 10705 */ 10706 if (dev->reg_state == NETREG_UNREGISTERING) { 10707 ASSERT_RTNL(); 10708 dev->needs_free_netdev = true; 10709 return; 10710 } 10711 10712 netif_free_tx_queues(dev); 10713 netif_free_rx_queues(dev); 10714 10715 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10716 10717 /* Flush device addresses */ 10718 dev_addr_flush(dev); 10719 10720 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10721 netif_napi_del(p); 10722 10723 ref_tracker_dir_exit(&dev->refcnt_tracker); 10724 #ifdef CONFIG_PCPU_DEV_REFCNT 10725 free_percpu(dev->pcpu_refcnt); 10726 dev->pcpu_refcnt = NULL; 10727 #endif 10728 free_percpu(dev->core_stats); 10729 dev->core_stats = NULL; 10730 free_percpu(dev->xdp_bulkq); 10731 dev->xdp_bulkq = NULL; 10732 10733 /* Compatibility with error handling in drivers */ 10734 if (dev->reg_state == NETREG_UNINITIALIZED) { 10735 netdev_freemem(dev); 10736 return; 10737 } 10738 10739 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10740 dev->reg_state = NETREG_RELEASED; 10741 10742 /* will free via device release */ 10743 put_device(&dev->dev); 10744 } 10745 EXPORT_SYMBOL(free_netdev); 10746 10747 /** 10748 * synchronize_net - Synchronize with packet receive processing 10749 * 10750 * Wait for packets currently being received to be done. 10751 * Does not block later packets from starting. 10752 */ 10753 void synchronize_net(void) 10754 { 10755 might_sleep(); 10756 if (rtnl_is_locked()) 10757 synchronize_rcu_expedited(); 10758 else 10759 synchronize_rcu(); 10760 } 10761 EXPORT_SYMBOL(synchronize_net); 10762 10763 /** 10764 * unregister_netdevice_queue - remove device from the kernel 10765 * @dev: device 10766 * @head: list 10767 * 10768 * This function shuts down a device interface and removes it 10769 * from the kernel tables. 10770 * If head not NULL, device is queued to be unregistered later. 10771 * 10772 * Callers must hold the rtnl semaphore. You may want 10773 * unregister_netdev() instead of this. 10774 */ 10775 10776 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10777 { 10778 ASSERT_RTNL(); 10779 10780 if (head) { 10781 list_move_tail(&dev->unreg_list, head); 10782 } else { 10783 LIST_HEAD(single); 10784 10785 list_add(&dev->unreg_list, &single); 10786 unregister_netdevice_many(&single); 10787 } 10788 } 10789 EXPORT_SYMBOL(unregister_netdevice_queue); 10790 10791 void unregister_netdevice_many_notify(struct list_head *head, 10792 u32 portid, const struct nlmsghdr *nlh) 10793 { 10794 struct net_device *dev, *tmp; 10795 LIST_HEAD(close_head); 10796 10797 BUG_ON(dev_boot_phase); 10798 ASSERT_RTNL(); 10799 10800 if (list_empty(head)) 10801 return; 10802 10803 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10804 /* Some devices call without registering 10805 * for initialization unwind. Remove those 10806 * devices and proceed with the remaining. 10807 */ 10808 if (dev->reg_state == NETREG_UNINITIALIZED) { 10809 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10810 dev->name, dev); 10811 10812 WARN_ON(1); 10813 list_del(&dev->unreg_list); 10814 continue; 10815 } 10816 dev->dismantle = true; 10817 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10818 } 10819 10820 /* If device is running, close it first. */ 10821 list_for_each_entry(dev, head, unreg_list) 10822 list_add_tail(&dev->close_list, &close_head); 10823 dev_close_many(&close_head, true); 10824 10825 list_for_each_entry(dev, head, unreg_list) { 10826 /* And unlink it from device chain. */ 10827 write_lock(&dev_base_lock); 10828 unlist_netdevice(dev, false); 10829 dev->reg_state = NETREG_UNREGISTERING; 10830 write_unlock(&dev_base_lock); 10831 } 10832 flush_all_backlogs(); 10833 10834 synchronize_net(); 10835 10836 list_for_each_entry(dev, head, unreg_list) { 10837 struct sk_buff *skb = NULL; 10838 10839 /* Shutdown queueing discipline. */ 10840 dev_shutdown(dev); 10841 10842 dev_xdp_uninstall(dev); 10843 bpf_dev_bound_netdev_unregister(dev); 10844 10845 netdev_offload_xstats_disable_all(dev); 10846 10847 /* Notify protocols, that we are about to destroy 10848 * this device. They should clean all the things. 10849 */ 10850 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10851 10852 if (!dev->rtnl_link_ops || 10853 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10854 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10855 GFP_KERNEL, NULL, 0, 10856 portid, nlh); 10857 10858 /* 10859 * Flush the unicast and multicast chains 10860 */ 10861 dev_uc_flush(dev); 10862 dev_mc_flush(dev); 10863 10864 netdev_name_node_alt_flush(dev); 10865 netdev_name_node_free(dev->name_node); 10866 10867 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10868 10869 if (dev->netdev_ops->ndo_uninit) 10870 dev->netdev_ops->ndo_uninit(dev); 10871 10872 if (skb) 10873 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10874 10875 /* Notifier chain MUST detach us all upper devices. */ 10876 WARN_ON(netdev_has_any_upper_dev(dev)); 10877 WARN_ON(netdev_has_any_lower_dev(dev)); 10878 10879 /* Remove entries from kobject tree */ 10880 netdev_unregister_kobject(dev); 10881 #ifdef CONFIG_XPS 10882 /* Remove XPS queueing entries */ 10883 netif_reset_xps_queues_gt(dev, 0); 10884 #endif 10885 } 10886 10887 synchronize_net(); 10888 10889 list_for_each_entry(dev, head, unreg_list) { 10890 netdev_put(dev, &dev->dev_registered_tracker); 10891 net_set_todo(dev); 10892 } 10893 10894 list_del(head); 10895 } 10896 10897 /** 10898 * unregister_netdevice_many - unregister many devices 10899 * @head: list of devices 10900 * 10901 * Note: As most callers use a stack allocated list_head, 10902 * we force a list_del() to make sure stack wont be corrupted later. 10903 */ 10904 void unregister_netdevice_many(struct list_head *head) 10905 { 10906 unregister_netdevice_many_notify(head, 0, NULL); 10907 } 10908 EXPORT_SYMBOL(unregister_netdevice_many); 10909 10910 /** 10911 * unregister_netdev - remove device from the kernel 10912 * @dev: device 10913 * 10914 * This function shuts down a device interface and removes it 10915 * from the kernel tables. 10916 * 10917 * This is just a wrapper for unregister_netdevice that takes 10918 * the rtnl semaphore. In general you want to use this and not 10919 * unregister_netdevice. 10920 */ 10921 void unregister_netdev(struct net_device *dev) 10922 { 10923 rtnl_lock(); 10924 unregister_netdevice(dev); 10925 rtnl_unlock(); 10926 } 10927 EXPORT_SYMBOL(unregister_netdev); 10928 10929 /** 10930 * __dev_change_net_namespace - move device to different nethost namespace 10931 * @dev: device 10932 * @net: network namespace 10933 * @pat: If not NULL name pattern to try if the current device name 10934 * is already taken in the destination network namespace. 10935 * @new_ifindex: If not zero, specifies device index in the target 10936 * namespace. 10937 * 10938 * This function shuts down a device interface and moves it 10939 * to a new network namespace. On success 0 is returned, on 10940 * a failure a netagive errno code is returned. 10941 * 10942 * Callers must hold the rtnl semaphore. 10943 */ 10944 10945 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10946 const char *pat, int new_ifindex) 10947 { 10948 struct net *net_old = dev_net(dev); 10949 int err, new_nsid; 10950 10951 ASSERT_RTNL(); 10952 10953 /* Don't allow namespace local devices to be moved. */ 10954 err = -EINVAL; 10955 if (dev->features & NETIF_F_NETNS_LOCAL) 10956 goto out; 10957 10958 /* Ensure the device has been registrered */ 10959 if (dev->reg_state != NETREG_REGISTERED) 10960 goto out; 10961 10962 /* Get out if there is nothing todo */ 10963 err = 0; 10964 if (net_eq(net_old, net)) 10965 goto out; 10966 10967 /* Pick the destination device name, and ensure 10968 * we can use it in the destination network namespace. 10969 */ 10970 err = -EEXIST; 10971 if (netdev_name_in_use(net, dev->name)) { 10972 /* We get here if we can't use the current device name */ 10973 if (!pat) 10974 goto out; 10975 err = dev_get_valid_name(net, dev, pat); 10976 if (err < 0) 10977 goto out; 10978 } 10979 10980 /* Check that new_ifindex isn't used yet. */ 10981 err = -EBUSY; 10982 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10983 goto out; 10984 10985 /* 10986 * And now a mini version of register_netdevice unregister_netdevice. 10987 */ 10988 10989 /* If device is running close it first. */ 10990 dev_close(dev); 10991 10992 /* And unlink it from device chain */ 10993 unlist_netdevice(dev, true); 10994 10995 synchronize_net(); 10996 10997 /* Shutdown queueing discipline. */ 10998 dev_shutdown(dev); 10999 11000 /* Notify protocols, that we are about to destroy 11001 * this device. They should clean all the things. 11002 * 11003 * Note that dev->reg_state stays at NETREG_REGISTERED. 11004 * This is wanted because this way 8021q and macvlan know 11005 * the device is just moving and can keep their slaves up. 11006 */ 11007 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11008 rcu_barrier(); 11009 11010 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11011 /* If there is an ifindex conflict assign a new one */ 11012 if (!new_ifindex) { 11013 if (__dev_get_by_index(net, dev->ifindex)) 11014 new_ifindex = dev_new_index(net); 11015 else 11016 new_ifindex = dev->ifindex; 11017 } 11018 11019 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11020 new_ifindex); 11021 11022 /* 11023 * Flush the unicast and multicast chains 11024 */ 11025 dev_uc_flush(dev); 11026 dev_mc_flush(dev); 11027 11028 /* Send a netdev-removed uevent to the old namespace */ 11029 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11030 netdev_adjacent_del_links(dev); 11031 11032 /* Move per-net netdevice notifiers that are following the netdevice */ 11033 move_netdevice_notifiers_dev_net(dev, net); 11034 11035 /* Actually switch the network namespace */ 11036 dev_net_set(dev, net); 11037 dev->ifindex = new_ifindex; 11038 11039 /* Send a netdev-add uevent to the new namespace */ 11040 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11041 netdev_adjacent_add_links(dev); 11042 11043 /* Fixup kobjects */ 11044 err = device_rename(&dev->dev, dev->name); 11045 WARN_ON(err); 11046 11047 /* Adapt owner in case owning user namespace of target network 11048 * namespace is different from the original one. 11049 */ 11050 err = netdev_change_owner(dev, net_old, net); 11051 WARN_ON(err); 11052 11053 /* Add the device back in the hashes */ 11054 list_netdevice(dev); 11055 11056 /* Notify protocols, that a new device appeared. */ 11057 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11058 11059 /* 11060 * Prevent userspace races by waiting until the network 11061 * device is fully setup before sending notifications. 11062 */ 11063 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11064 11065 synchronize_net(); 11066 err = 0; 11067 out: 11068 return err; 11069 } 11070 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11071 11072 static int dev_cpu_dead(unsigned int oldcpu) 11073 { 11074 struct sk_buff **list_skb; 11075 struct sk_buff *skb; 11076 unsigned int cpu; 11077 struct softnet_data *sd, *oldsd, *remsd = NULL; 11078 11079 local_irq_disable(); 11080 cpu = smp_processor_id(); 11081 sd = &per_cpu(softnet_data, cpu); 11082 oldsd = &per_cpu(softnet_data, oldcpu); 11083 11084 /* Find end of our completion_queue. */ 11085 list_skb = &sd->completion_queue; 11086 while (*list_skb) 11087 list_skb = &(*list_skb)->next; 11088 /* Append completion queue from offline CPU. */ 11089 *list_skb = oldsd->completion_queue; 11090 oldsd->completion_queue = NULL; 11091 11092 /* Append output queue from offline CPU. */ 11093 if (oldsd->output_queue) { 11094 *sd->output_queue_tailp = oldsd->output_queue; 11095 sd->output_queue_tailp = oldsd->output_queue_tailp; 11096 oldsd->output_queue = NULL; 11097 oldsd->output_queue_tailp = &oldsd->output_queue; 11098 } 11099 /* Append NAPI poll list from offline CPU, with one exception : 11100 * process_backlog() must be called by cpu owning percpu backlog. 11101 * We properly handle process_queue & input_pkt_queue later. 11102 */ 11103 while (!list_empty(&oldsd->poll_list)) { 11104 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11105 struct napi_struct, 11106 poll_list); 11107 11108 list_del_init(&napi->poll_list); 11109 if (napi->poll == process_backlog) 11110 napi->state = 0; 11111 else 11112 ____napi_schedule(sd, napi); 11113 } 11114 11115 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11116 local_irq_enable(); 11117 11118 #ifdef CONFIG_RPS 11119 remsd = oldsd->rps_ipi_list; 11120 oldsd->rps_ipi_list = NULL; 11121 #endif 11122 /* send out pending IPI's on offline CPU */ 11123 net_rps_send_ipi(remsd); 11124 11125 /* Process offline CPU's input_pkt_queue */ 11126 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11127 netif_rx(skb); 11128 input_queue_head_incr(oldsd); 11129 } 11130 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11131 netif_rx(skb); 11132 input_queue_head_incr(oldsd); 11133 } 11134 11135 return 0; 11136 } 11137 11138 /** 11139 * netdev_increment_features - increment feature set by one 11140 * @all: current feature set 11141 * @one: new feature set 11142 * @mask: mask feature set 11143 * 11144 * Computes a new feature set after adding a device with feature set 11145 * @one to the master device with current feature set @all. Will not 11146 * enable anything that is off in @mask. Returns the new feature set. 11147 */ 11148 netdev_features_t netdev_increment_features(netdev_features_t all, 11149 netdev_features_t one, netdev_features_t mask) 11150 { 11151 if (mask & NETIF_F_HW_CSUM) 11152 mask |= NETIF_F_CSUM_MASK; 11153 mask |= NETIF_F_VLAN_CHALLENGED; 11154 11155 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11156 all &= one | ~NETIF_F_ALL_FOR_ALL; 11157 11158 /* If one device supports hw checksumming, set for all. */ 11159 if (all & NETIF_F_HW_CSUM) 11160 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11161 11162 return all; 11163 } 11164 EXPORT_SYMBOL(netdev_increment_features); 11165 11166 static struct hlist_head * __net_init netdev_create_hash(void) 11167 { 11168 int i; 11169 struct hlist_head *hash; 11170 11171 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11172 if (hash != NULL) 11173 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11174 INIT_HLIST_HEAD(&hash[i]); 11175 11176 return hash; 11177 } 11178 11179 /* Initialize per network namespace state */ 11180 static int __net_init netdev_init(struct net *net) 11181 { 11182 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11183 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11184 11185 INIT_LIST_HEAD(&net->dev_base_head); 11186 11187 net->dev_name_head = netdev_create_hash(); 11188 if (net->dev_name_head == NULL) 11189 goto err_name; 11190 11191 net->dev_index_head = netdev_create_hash(); 11192 if (net->dev_index_head == NULL) 11193 goto err_idx; 11194 11195 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11196 11197 return 0; 11198 11199 err_idx: 11200 kfree(net->dev_name_head); 11201 err_name: 11202 return -ENOMEM; 11203 } 11204 11205 /** 11206 * netdev_drivername - network driver for the device 11207 * @dev: network device 11208 * 11209 * Determine network driver for device. 11210 */ 11211 const char *netdev_drivername(const struct net_device *dev) 11212 { 11213 const struct device_driver *driver; 11214 const struct device *parent; 11215 const char *empty = ""; 11216 11217 parent = dev->dev.parent; 11218 if (!parent) 11219 return empty; 11220 11221 driver = parent->driver; 11222 if (driver && driver->name) 11223 return driver->name; 11224 return empty; 11225 } 11226 11227 static void __netdev_printk(const char *level, const struct net_device *dev, 11228 struct va_format *vaf) 11229 { 11230 if (dev && dev->dev.parent) { 11231 dev_printk_emit(level[1] - '0', 11232 dev->dev.parent, 11233 "%s %s %s%s: %pV", 11234 dev_driver_string(dev->dev.parent), 11235 dev_name(dev->dev.parent), 11236 netdev_name(dev), netdev_reg_state(dev), 11237 vaf); 11238 } else if (dev) { 11239 printk("%s%s%s: %pV", 11240 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11241 } else { 11242 printk("%s(NULL net_device): %pV", level, vaf); 11243 } 11244 } 11245 11246 void netdev_printk(const char *level, const struct net_device *dev, 11247 const char *format, ...) 11248 { 11249 struct va_format vaf; 11250 va_list args; 11251 11252 va_start(args, format); 11253 11254 vaf.fmt = format; 11255 vaf.va = &args; 11256 11257 __netdev_printk(level, dev, &vaf); 11258 11259 va_end(args); 11260 } 11261 EXPORT_SYMBOL(netdev_printk); 11262 11263 #define define_netdev_printk_level(func, level) \ 11264 void func(const struct net_device *dev, const char *fmt, ...) \ 11265 { \ 11266 struct va_format vaf; \ 11267 va_list args; \ 11268 \ 11269 va_start(args, fmt); \ 11270 \ 11271 vaf.fmt = fmt; \ 11272 vaf.va = &args; \ 11273 \ 11274 __netdev_printk(level, dev, &vaf); \ 11275 \ 11276 va_end(args); \ 11277 } \ 11278 EXPORT_SYMBOL(func); 11279 11280 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11281 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11282 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11283 define_netdev_printk_level(netdev_err, KERN_ERR); 11284 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11285 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11286 define_netdev_printk_level(netdev_info, KERN_INFO); 11287 11288 static void __net_exit netdev_exit(struct net *net) 11289 { 11290 kfree(net->dev_name_head); 11291 kfree(net->dev_index_head); 11292 if (net != &init_net) 11293 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11294 } 11295 11296 static struct pernet_operations __net_initdata netdev_net_ops = { 11297 .init = netdev_init, 11298 .exit = netdev_exit, 11299 }; 11300 11301 static void __net_exit default_device_exit_net(struct net *net) 11302 { 11303 struct net_device *dev, *aux; 11304 /* 11305 * Push all migratable network devices back to the 11306 * initial network namespace 11307 */ 11308 ASSERT_RTNL(); 11309 for_each_netdev_safe(net, dev, aux) { 11310 int err; 11311 char fb_name[IFNAMSIZ]; 11312 11313 /* Ignore unmoveable devices (i.e. loopback) */ 11314 if (dev->features & NETIF_F_NETNS_LOCAL) 11315 continue; 11316 11317 /* Leave virtual devices for the generic cleanup */ 11318 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11319 continue; 11320 11321 /* Push remaining network devices to init_net */ 11322 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11323 if (netdev_name_in_use(&init_net, fb_name)) 11324 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11325 err = dev_change_net_namespace(dev, &init_net, fb_name); 11326 if (err) { 11327 pr_emerg("%s: failed to move %s to init_net: %d\n", 11328 __func__, dev->name, err); 11329 BUG(); 11330 } 11331 } 11332 } 11333 11334 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11335 { 11336 /* At exit all network devices most be removed from a network 11337 * namespace. Do this in the reverse order of registration. 11338 * Do this across as many network namespaces as possible to 11339 * improve batching efficiency. 11340 */ 11341 struct net_device *dev; 11342 struct net *net; 11343 LIST_HEAD(dev_kill_list); 11344 11345 rtnl_lock(); 11346 list_for_each_entry(net, net_list, exit_list) { 11347 default_device_exit_net(net); 11348 cond_resched(); 11349 } 11350 11351 list_for_each_entry(net, net_list, exit_list) { 11352 for_each_netdev_reverse(net, dev) { 11353 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11354 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11355 else 11356 unregister_netdevice_queue(dev, &dev_kill_list); 11357 } 11358 } 11359 unregister_netdevice_many(&dev_kill_list); 11360 rtnl_unlock(); 11361 } 11362 11363 static struct pernet_operations __net_initdata default_device_ops = { 11364 .exit_batch = default_device_exit_batch, 11365 }; 11366 11367 /* 11368 * Initialize the DEV module. At boot time this walks the device list and 11369 * unhooks any devices that fail to initialise (normally hardware not 11370 * present) and leaves us with a valid list of present and active devices. 11371 * 11372 */ 11373 11374 /* 11375 * This is called single threaded during boot, so no need 11376 * to take the rtnl semaphore. 11377 */ 11378 static int __init net_dev_init(void) 11379 { 11380 int i, rc = -ENOMEM; 11381 11382 BUG_ON(!dev_boot_phase); 11383 11384 if (dev_proc_init()) 11385 goto out; 11386 11387 if (netdev_kobject_init()) 11388 goto out; 11389 11390 INIT_LIST_HEAD(&ptype_all); 11391 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11392 INIT_LIST_HEAD(&ptype_base[i]); 11393 11394 if (register_pernet_subsys(&netdev_net_ops)) 11395 goto out; 11396 11397 /* 11398 * Initialise the packet receive queues. 11399 */ 11400 11401 for_each_possible_cpu(i) { 11402 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11403 struct softnet_data *sd = &per_cpu(softnet_data, i); 11404 11405 INIT_WORK(flush, flush_backlog); 11406 11407 skb_queue_head_init(&sd->input_pkt_queue); 11408 skb_queue_head_init(&sd->process_queue); 11409 #ifdef CONFIG_XFRM_OFFLOAD 11410 skb_queue_head_init(&sd->xfrm_backlog); 11411 #endif 11412 INIT_LIST_HEAD(&sd->poll_list); 11413 sd->output_queue_tailp = &sd->output_queue; 11414 #ifdef CONFIG_RPS 11415 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11416 sd->cpu = i; 11417 #endif 11418 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11419 spin_lock_init(&sd->defer_lock); 11420 11421 init_gro_hash(&sd->backlog); 11422 sd->backlog.poll = process_backlog; 11423 sd->backlog.weight = weight_p; 11424 } 11425 11426 dev_boot_phase = 0; 11427 11428 /* The loopback device is special if any other network devices 11429 * is present in a network namespace the loopback device must 11430 * be present. Since we now dynamically allocate and free the 11431 * loopback device ensure this invariant is maintained by 11432 * keeping the loopback device as the first device on the 11433 * list of network devices. Ensuring the loopback devices 11434 * is the first device that appears and the last network device 11435 * that disappears. 11436 */ 11437 if (register_pernet_device(&loopback_net_ops)) 11438 goto out; 11439 11440 if (register_pernet_device(&default_device_ops)) 11441 goto out; 11442 11443 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11444 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11445 11446 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11447 NULL, dev_cpu_dead); 11448 WARN_ON(rc < 0); 11449 rc = 0; 11450 out: 11451 return rc; 11452 } 11453 11454 subsys_initcall(net_dev_init); 11455