1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "dev.h" 155 #include "net-sysfs.h" 156 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_info(unsigned long val, 164 struct netdev_notifier_info *info); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 netdev_name_node_del(name_node); 349 kfree(name_node->name); 350 netdev_name_node_free(name_node); 351 } 352 353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 354 { 355 struct netdev_name_node *name_node; 356 struct net *net = dev_net(dev); 357 358 name_node = netdev_name_node_lookup(net, name); 359 if (!name_node) 360 return -ENOENT; 361 /* lookup might have found our primary name or a name belonging 362 * to another device. 363 */ 364 if (name_node == dev->name_node || name_node->dev != dev) 365 return -EINVAL; 366 367 __netdev_name_node_alt_destroy(name_node); 368 369 return 0; 370 } 371 372 static void netdev_name_node_alt_flush(struct net_device *dev) 373 { 374 struct netdev_name_node *name_node, *tmp; 375 376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 377 __netdev_name_node_alt_destroy(name_node); 378 } 379 380 /* Device list insertion */ 381 static void list_netdevice(struct net_device *dev) 382 { 383 struct net *net = dev_net(dev); 384 385 ASSERT_RTNL(); 386 387 write_lock(&dev_base_lock); 388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 389 netdev_name_node_add(net, dev->name_node); 390 hlist_add_head_rcu(&dev->index_hlist, 391 dev_index_hash(net, dev->ifindex)); 392 write_unlock(&dev_base_lock); 393 394 dev_base_seq_inc(net); 395 } 396 397 /* Device list removal 398 * caller must respect a RCU grace period before freeing/reusing dev 399 */ 400 static void unlist_netdevice(struct net_device *dev, bool lock) 401 { 402 ASSERT_RTNL(); 403 404 /* Unlink dev from the device chain */ 405 if (lock) 406 write_lock(&dev_base_lock); 407 list_del_rcu(&dev->dev_list); 408 netdev_name_node_del(dev->name_node); 409 hlist_del_rcu(&dev->index_hlist); 410 if (lock) 411 write_unlock(&dev_base_lock); 412 413 dev_base_seq_inc(dev_net(dev)); 414 } 415 416 /* 417 * Our notifier list 418 */ 419 420 static RAW_NOTIFIER_HEAD(netdev_chain); 421 422 /* 423 * Device drivers call our routines to queue packets here. We empty the 424 * queue in the local softnet handler. 425 */ 426 427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 428 EXPORT_PER_CPU_SYMBOL(softnet_data); 429 430 #ifdef CONFIG_LOCKDEP 431 /* 432 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 433 * according to dev->type 434 */ 435 static const unsigned short netdev_lock_type[] = { 436 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 437 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 438 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 439 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 440 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 441 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 442 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 443 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 444 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 445 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 446 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 447 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 448 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 449 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 450 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 451 452 static const char *const netdev_lock_name[] = { 453 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 454 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 455 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 456 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 457 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 458 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 459 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 460 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 461 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 462 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 463 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 464 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 465 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 466 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 467 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 468 469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 471 472 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 473 { 474 int i; 475 476 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 477 if (netdev_lock_type[i] == dev_type) 478 return i; 479 /* the last key is used by default */ 480 return ARRAY_SIZE(netdev_lock_type) - 1; 481 } 482 483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 484 unsigned short dev_type) 485 { 486 int i; 487 488 i = netdev_lock_pos(dev_type); 489 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 490 netdev_lock_name[i]); 491 } 492 493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 494 { 495 int i; 496 497 i = netdev_lock_pos(dev->type); 498 lockdep_set_class_and_name(&dev->addr_list_lock, 499 &netdev_addr_lock_key[i], 500 netdev_lock_name[i]); 501 } 502 #else 503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 504 unsigned short dev_type) 505 { 506 } 507 508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 509 { 510 } 511 #endif 512 513 /******************************************************************************* 514 * 515 * Protocol management and registration routines 516 * 517 *******************************************************************************/ 518 519 520 /* 521 * Add a protocol ID to the list. Now that the input handler is 522 * smarter we can dispense with all the messy stuff that used to be 523 * here. 524 * 525 * BEWARE!!! Protocol handlers, mangling input packets, 526 * MUST BE last in hash buckets and checking protocol handlers 527 * MUST start from promiscuous ptype_all chain in net_bh. 528 * It is true now, do not change it. 529 * Explanation follows: if protocol handler, mangling packet, will 530 * be the first on list, it is not able to sense, that packet 531 * is cloned and should be copied-on-write, so that it will 532 * change it and subsequent readers will get broken packet. 533 * --ANK (980803) 534 */ 535 536 static inline struct list_head *ptype_head(const struct packet_type *pt) 537 { 538 if (pt->type == htons(ETH_P_ALL)) 539 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 540 else 541 return pt->dev ? &pt->dev->ptype_specific : 542 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 543 } 544 545 /** 546 * dev_add_pack - add packet handler 547 * @pt: packet type declaration 548 * 549 * Add a protocol handler to the networking stack. The passed &packet_type 550 * is linked into kernel lists and may not be freed until it has been 551 * removed from the kernel lists. 552 * 553 * This call does not sleep therefore it can not 554 * guarantee all CPU's that are in middle of receiving packets 555 * will see the new packet type (until the next received packet). 556 */ 557 558 void dev_add_pack(struct packet_type *pt) 559 { 560 struct list_head *head = ptype_head(pt); 561 562 spin_lock(&ptype_lock); 563 list_add_rcu(&pt->list, head); 564 spin_unlock(&ptype_lock); 565 } 566 EXPORT_SYMBOL(dev_add_pack); 567 568 /** 569 * __dev_remove_pack - remove packet handler 570 * @pt: packet type declaration 571 * 572 * Remove a protocol handler that was previously added to the kernel 573 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 574 * from the kernel lists and can be freed or reused once this function 575 * returns. 576 * 577 * The packet type might still be in use by receivers 578 * and must not be freed until after all the CPU's have gone 579 * through a quiescent state. 580 */ 581 void __dev_remove_pack(struct packet_type *pt) 582 { 583 struct list_head *head = ptype_head(pt); 584 struct packet_type *pt1; 585 586 spin_lock(&ptype_lock); 587 588 list_for_each_entry(pt1, head, list) { 589 if (pt == pt1) { 590 list_del_rcu(&pt->list); 591 goto out; 592 } 593 } 594 595 pr_warn("dev_remove_pack: %p not found\n", pt); 596 out: 597 spin_unlock(&ptype_lock); 598 } 599 EXPORT_SYMBOL(__dev_remove_pack); 600 601 /** 602 * dev_remove_pack - remove packet handler 603 * @pt: packet type declaration 604 * 605 * Remove a protocol handler that was previously added to the kernel 606 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 607 * from the kernel lists and can be freed or reused once this function 608 * returns. 609 * 610 * This call sleeps to guarantee that no CPU is looking at the packet 611 * type after return. 612 */ 613 void dev_remove_pack(struct packet_type *pt) 614 { 615 __dev_remove_pack(pt); 616 617 synchronize_net(); 618 } 619 EXPORT_SYMBOL(dev_remove_pack); 620 621 622 /******************************************************************************* 623 * 624 * Device Interface Subroutines 625 * 626 *******************************************************************************/ 627 628 /** 629 * dev_get_iflink - get 'iflink' value of a interface 630 * @dev: targeted interface 631 * 632 * Indicates the ifindex the interface is linked to. 633 * Physical interfaces have the same 'ifindex' and 'iflink' values. 634 */ 635 636 int dev_get_iflink(const struct net_device *dev) 637 { 638 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 639 return dev->netdev_ops->ndo_get_iflink(dev); 640 641 return dev->ifindex; 642 } 643 EXPORT_SYMBOL(dev_get_iflink); 644 645 /** 646 * dev_fill_metadata_dst - Retrieve tunnel egress information. 647 * @dev: targeted interface 648 * @skb: The packet. 649 * 650 * For better visibility of tunnel traffic OVS needs to retrieve 651 * egress tunnel information for a packet. Following API allows 652 * user to get this info. 653 */ 654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 655 { 656 struct ip_tunnel_info *info; 657 658 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 659 return -EINVAL; 660 661 info = skb_tunnel_info_unclone(skb); 662 if (!info) 663 return -ENOMEM; 664 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 665 return -EINVAL; 666 667 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 668 } 669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 670 671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 672 { 673 int k = stack->num_paths++; 674 675 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 676 return NULL; 677 678 return &stack->path[k]; 679 } 680 681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 682 struct net_device_path_stack *stack) 683 { 684 const struct net_device *last_dev; 685 struct net_device_path_ctx ctx = { 686 .dev = dev, 687 }; 688 struct net_device_path *path; 689 int ret = 0; 690 691 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 692 stack->num_paths = 0; 693 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 694 last_dev = ctx.dev; 695 path = dev_fwd_path(stack); 696 if (!path) 697 return -1; 698 699 memset(path, 0, sizeof(struct net_device_path)); 700 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 701 if (ret < 0) 702 return -1; 703 704 if (WARN_ON_ONCE(last_dev == ctx.dev)) 705 return -1; 706 } 707 708 if (!ctx.dev) 709 return ret; 710 711 path = dev_fwd_path(stack); 712 if (!path) 713 return -1; 714 path->type = DEV_PATH_ETHERNET; 715 path->dev = ctx.dev; 716 717 return ret; 718 } 719 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 720 721 /** 722 * __dev_get_by_name - find a device by its name 723 * @net: the applicable net namespace 724 * @name: name to find 725 * 726 * Find an interface by name. Must be called under RTNL semaphore 727 * or @dev_base_lock. If the name is found a pointer to the device 728 * is returned. If the name is not found then %NULL is returned. The 729 * reference counters are not incremented so the caller must be 730 * careful with locks. 731 */ 732 733 struct net_device *__dev_get_by_name(struct net *net, const char *name) 734 { 735 struct netdev_name_node *node_name; 736 737 node_name = netdev_name_node_lookup(net, name); 738 return node_name ? node_name->dev : NULL; 739 } 740 EXPORT_SYMBOL(__dev_get_by_name); 741 742 /** 743 * dev_get_by_name_rcu - find a device by its name 744 * @net: the applicable net namespace 745 * @name: name to find 746 * 747 * Find an interface by name. 748 * If the name is found a pointer to the device is returned. 749 * If the name is not found then %NULL is returned. 750 * The reference counters are not incremented so the caller must be 751 * careful with locks. The caller must hold RCU lock. 752 */ 753 754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 755 { 756 struct netdev_name_node *node_name; 757 758 node_name = netdev_name_node_lookup_rcu(net, name); 759 return node_name ? node_name->dev : NULL; 760 } 761 EXPORT_SYMBOL(dev_get_by_name_rcu); 762 763 /** 764 * dev_get_by_name - find a device by its name 765 * @net: the applicable net namespace 766 * @name: name to find 767 * 768 * Find an interface by name. This can be called from any 769 * context and does its own locking. The returned handle has 770 * the usage count incremented and the caller must use dev_put() to 771 * release it when it is no longer needed. %NULL is returned if no 772 * matching device is found. 773 */ 774 775 struct net_device *dev_get_by_name(struct net *net, const char *name) 776 { 777 struct net_device *dev; 778 779 rcu_read_lock(); 780 dev = dev_get_by_name_rcu(net, name); 781 dev_hold(dev); 782 rcu_read_unlock(); 783 return dev; 784 } 785 EXPORT_SYMBOL(dev_get_by_name); 786 787 /** 788 * __dev_get_by_index - find a device by its ifindex 789 * @net: the applicable net namespace 790 * @ifindex: index of device 791 * 792 * Search for an interface by index. Returns %NULL if the device 793 * is not found or a pointer to the device. The device has not 794 * had its reference counter increased so the caller must be careful 795 * about locking. The caller must hold either the RTNL semaphore 796 * or @dev_base_lock. 797 */ 798 799 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 800 { 801 struct net_device *dev; 802 struct hlist_head *head = dev_index_hash(net, ifindex); 803 804 hlist_for_each_entry(dev, head, index_hlist) 805 if (dev->ifindex == ifindex) 806 return dev; 807 808 return NULL; 809 } 810 EXPORT_SYMBOL(__dev_get_by_index); 811 812 /** 813 * dev_get_by_index_rcu - find a device by its ifindex 814 * @net: the applicable net namespace 815 * @ifindex: index of device 816 * 817 * Search for an interface by index. Returns %NULL if the device 818 * is not found or a pointer to the device. The device has not 819 * had its reference counter increased so the caller must be careful 820 * about locking. The caller must hold RCU lock. 821 */ 822 823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 824 { 825 struct net_device *dev; 826 struct hlist_head *head = dev_index_hash(net, ifindex); 827 828 hlist_for_each_entry_rcu(dev, head, index_hlist) 829 if (dev->ifindex == ifindex) 830 return dev; 831 832 return NULL; 833 } 834 EXPORT_SYMBOL(dev_get_by_index_rcu); 835 836 837 /** 838 * dev_get_by_index - find a device by its ifindex 839 * @net: the applicable net namespace 840 * @ifindex: index of device 841 * 842 * Search for an interface by index. Returns NULL if the device 843 * is not found or a pointer to the device. The device returned has 844 * had a reference added and the pointer is safe until the user calls 845 * dev_put to indicate they have finished with it. 846 */ 847 848 struct net_device *dev_get_by_index(struct net *net, int ifindex) 849 { 850 struct net_device *dev; 851 852 rcu_read_lock(); 853 dev = dev_get_by_index_rcu(net, ifindex); 854 dev_hold(dev); 855 rcu_read_unlock(); 856 return dev; 857 } 858 EXPORT_SYMBOL(dev_get_by_index); 859 860 /** 861 * dev_get_by_napi_id - find a device by napi_id 862 * @napi_id: ID of the NAPI struct 863 * 864 * Search for an interface by NAPI ID. Returns %NULL if the device 865 * is not found or a pointer to the device. The device has not had 866 * its reference counter increased so the caller must be careful 867 * about locking. The caller must hold RCU lock. 868 */ 869 870 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 871 { 872 struct napi_struct *napi; 873 874 WARN_ON_ONCE(!rcu_read_lock_held()); 875 876 if (napi_id < MIN_NAPI_ID) 877 return NULL; 878 879 napi = napi_by_id(napi_id); 880 881 return napi ? napi->dev : NULL; 882 } 883 EXPORT_SYMBOL(dev_get_by_napi_id); 884 885 /** 886 * netdev_get_name - get a netdevice name, knowing its ifindex. 887 * @net: network namespace 888 * @name: a pointer to the buffer where the name will be stored. 889 * @ifindex: the ifindex of the interface to get the name from. 890 */ 891 int netdev_get_name(struct net *net, char *name, int ifindex) 892 { 893 struct net_device *dev; 894 int ret; 895 896 down_read(&devnet_rename_sem); 897 rcu_read_lock(); 898 899 dev = dev_get_by_index_rcu(net, ifindex); 900 if (!dev) { 901 ret = -ENODEV; 902 goto out; 903 } 904 905 strcpy(name, dev->name); 906 907 ret = 0; 908 out: 909 rcu_read_unlock(); 910 up_read(&devnet_rename_sem); 911 return ret; 912 } 913 914 /** 915 * dev_getbyhwaddr_rcu - find a device by its hardware address 916 * @net: the applicable net namespace 917 * @type: media type of device 918 * @ha: hardware address 919 * 920 * Search for an interface by MAC address. Returns NULL if the device 921 * is not found or a pointer to the device. 922 * The caller must hold RCU or RTNL. 923 * The returned device has not had its ref count increased 924 * and the caller must therefore be careful about locking 925 * 926 */ 927 928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 929 const char *ha) 930 { 931 struct net_device *dev; 932 933 for_each_netdev_rcu(net, dev) 934 if (dev->type == type && 935 !memcmp(dev->dev_addr, ha, dev->addr_len)) 936 return dev; 937 938 return NULL; 939 } 940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 941 942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 943 { 944 struct net_device *dev, *ret = NULL; 945 946 rcu_read_lock(); 947 for_each_netdev_rcu(net, dev) 948 if (dev->type == type) { 949 dev_hold(dev); 950 ret = dev; 951 break; 952 } 953 rcu_read_unlock(); 954 return ret; 955 } 956 EXPORT_SYMBOL(dev_getfirstbyhwtype); 957 958 /** 959 * __dev_get_by_flags - find any device with given flags 960 * @net: the applicable net namespace 961 * @if_flags: IFF_* values 962 * @mask: bitmask of bits in if_flags to check 963 * 964 * Search for any interface with the given flags. Returns NULL if a device 965 * is not found or a pointer to the device. Must be called inside 966 * rtnl_lock(), and result refcount is unchanged. 967 */ 968 969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 970 unsigned short mask) 971 { 972 struct net_device *dev, *ret; 973 974 ASSERT_RTNL(); 975 976 ret = NULL; 977 for_each_netdev(net, dev) { 978 if (((dev->flags ^ if_flags) & mask) == 0) { 979 ret = dev; 980 break; 981 } 982 } 983 return ret; 984 } 985 EXPORT_SYMBOL(__dev_get_by_flags); 986 987 /** 988 * dev_valid_name - check if name is okay for network device 989 * @name: name string 990 * 991 * Network device names need to be valid file names to 992 * allow sysfs to work. We also disallow any kind of 993 * whitespace. 994 */ 995 bool dev_valid_name(const char *name) 996 { 997 if (*name == '\0') 998 return false; 999 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1000 return false; 1001 if (!strcmp(name, ".") || !strcmp(name, "..")) 1002 return false; 1003 1004 while (*name) { 1005 if (*name == '/' || *name == ':' || isspace(*name)) 1006 return false; 1007 name++; 1008 } 1009 return true; 1010 } 1011 EXPORT_SYMBOL(dev_valid_name); 1012 1013 /** 1014 * __dev_alloc_name - allocate a name for a device 1015 * @net: network namespace to allocate the device name in 1016 * @name: name format string 1017 * @buf: scratch buffer and result name string 1018 * 1019 * Passed a format string - eg "lt%d" it will try and find a suitable 1020 * id. It scans list of devices to build up a free map, then chooses 1021 * the first empty slot. The caller must hold the dev_base or rtnl lock 1022 * while allocating the name and adding the device in order to avoid 1023 * duplicates. 1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1025 * Returns the number of the unit assigned or a negative errno code. 1026 */ 1027 1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1029 { 1030 int i = 0; 1031 const char *p; 1032 const int max_netdevices = 8*PAGE_SIZE; 1033 unsigned long *inuse; 1034 struct net_device *d; 1035 1036 if (!dev_valid_name(name)) 1037 return -EINVAL; 1038 1039 p = strchr(name, '%'); 1040 if (p) { 1041 /* 1042 * Verify the string as this thing may have come from 1043 * the user. There must be either one "%d" and no other "%" 1044 * characters. 1045 */ 1046 if (p[1] != 'd' || strchr(p + 2, '%')) 1047 return -EINVAL; 1048 1049 /* Use one page as a bit array of possible slots */ 1050 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1051 if (!inuse) 1052 return -ENOMEM; 1053 1054 for_each_netdev(net, d) { 1055 struct netdev_name_node *name_node; 1056 list_for_each_entry(name_node, &d->name_node->list, list) { 1057 if (!sscanf(name_node->name, name, &i)) 1058 continue; 1059 if (i < 0 || i >= max_netdevices) 1060 continue; 1061 1062 /* avoid cases where sscanf is not exact inverse of printf */ 1063 snprintf(buf, IFNAMSIZ, name, i); 1064 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1065 __set_bit(i, inuse); 1066 } 1067 if (!sscanf(d->name, name, &i)) 1068 continue; 1069 if (i < 0 || i >= max_netdevices) 1070 continue; 1071 1072 /* avoid cases where sscanf is not exact inverse of printf */ 1073 snprintf(buf, IFNAMSIZ, name, i); 1074 if (!strncmp(buf, d->name, IFNAMSIZ)) 1075 __set_bit(i, inuse); 1076 } 1077 1078 i = find_first_zero_bit(inuse, max_netdevices); 1079 free_page((unsigned long) inuse); 1080 } 1081 1082 snprintf(buf, IFNAMSIZ, name, i); 1083 if (!netdev_name_in_use(net, buf)) 1084 return i; 1085 1086 /* It is possible to run out of possible slots 1087 * when the name is long and there isn't enough space left 1088 * for the digits, or if all bits are used. 1089 */ 1090 return -ENFILE; 1091 } 1092 1093 static int dev_alloc_name_ns(struct net *net, 1094 struct net_device *dev, 1095 const char *name) 1096 { 1097 char buf[IFNAMSIZ]; 1098 int ret; 1099 1100 BUG_ON(!net); 1101 ret = __dev_alloc_name(net, name, buf); 1102 if (ret >= 0) 1103 strscpy(dev->name, buf, IFNAMSIZ); 1104 return ret; 1105 } 1106 1107 /** 1108 * dev_alloc_name - allocate a name for a device 1109 * @dev: device 1110 * @name: name format string 1111 * 1112 * Passed a format string - eg "lt%d" it will try and find a suitable 1113 * id. It scans list of devices to build up a free map, then chooses 1114 * the first empty slot. The caller must hold the dev_base or rtnl lock 1115 * while allocating the name and adding the device in order to avoid 1116 * duplicates. 1117 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1118 * Returns the number of the unit assigned or a negative errno code. 1119 */ 1120 1121 int dev_alloc_name(struct net_device *dev, const char *name) 1122 { 1123 return dev_alloc_name_ns(dev_net(dev), dev, name); 1124 } 1125 EXPORT_SYMBOL(dev_alloc_name); 1126 1127 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1128 const char *name) 1129 { 1130 BUG_ON(!net); 1131 1132 if (!dev_valid_name(name)) 1133 return -EINVAL; 1134 1135 if (strchr(name, '%')) 1136 return dev_alloc_name_ns(net, dev, name); 1137 else if (netdev_name_in_use(net, name)) 1138 return -EEXIST; 1139 else if (dev->name != name) 1140 strscpy(dev->name, name, IFNAMSIZ); 1141 1142 return 0; 1143 } 1144 1145 /** 1146 * dev_change_name - change name of a device 1147 * @dev: device 1148 * @newname: name (or format string) must be at least IFNAMSIZ 1149 * 1150 * Change name of a device, can pass format strings "eth%d". 1151 * for wildcarding. 1152 */ 1153 int dev_change_name(struct net_device *dev, const char *newname) 1154 { 1155 unsigned char old_assign_type; 1156 char oldname[IFNAMSIZ]; 1157 int err = 0; 1158 int ret; 1159 struct net *net; 1160 1161 ASSERT_RTNL(); 1162 BUG_ON(!dev_net(dev)); 1163 1164 net = dev_net(dev); 1165 1166 down_write(&devnet_rename_sem); 1167 1168 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1169 up_write(&devnet_rename_sem); 1170 return 0; 1171 } 1172 1173 memcpy(oldname, dev->name, IFNAMSIZ); 1174 1175 err = dev_get_valid_name(net, dev, newname); 1176 if (err < 0) { 1177 up_write(&devnet_rename_sem); 1178 return err; 1179 } 1180 1181 if (oldname[0] && !strchr(oldname, '%')) 1182 netdev_info(dev, "renamed from %s%s\n", oldname, 1183 dev->flags & IFF_UP ? " (while UP)" : ""); 1184 1185 old_assign_type = dev->name_assign_type; 1186 dev->name_assign_type = NET_NAME_RENAMED; 1187 1188 rollback: 1189 ret = device_rename(&dev->dev, dev->name); 1190 if (ret) { 1191 memcpy(dev->name, oldname, IFNAMSIZ); 1192 dev->name_assign_type = old_assign_type; 1193 up_write(&devnet_rename_sem); 1194 return ret; 1195 } 1196 1197 up_write(&devnet_rename_sem); 1198 1199 netdev_adjacent_rename_links(dev, oldname); 1200 1201 write_lock(&dev_base_lock); 1202 netdev_name_node_del(dev->name_node); 1203 write_unlock(&dev_base_lock); 1204 1205 synchronize_rcu(); 1206 1207 write_lock(&dev_base_lock); 1208 netdev_name_node_add(net, dev->name_node); 1209 write_unlock(&dev_base_lock); 1210 1211 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1212 ret = notifier_to_errno(ret); 1213 1214 if (ret) { 1215 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1216 if (err >= 0) { 1217 err = ret; 1218 down_write(&devnet_rename_sem); 1219 memcpy(dev->name, oldname, IFNAMSIZ); 1220 memcpy(oldname, newname, IFNAMSIZ); 1221 dev->name_assign_type = old_assign_type; 1222 old_assign_type = NET_NAME_RENAMED; 1223 goto rollback; 1224 } else { 1225 netdev_err(dev, "name change rollback failed: %d\n", 1226 ret); 1227 } 1228 } 1229 1230 return err; 1231 } 1232 1233 /** 1234 * dev_set_alias - change ifalias of a device 1235 * @dev: device 1236 * @alias: name up to IFALIASZ 1237 * @len: limit of bytes to copy from info 1238 * 1239 * Set ifalias for a device, 1240 */ 1241 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1242 { 1243 struct dev_ifalias *new_alias = NULL; 1244 1245 if (len >= IFALIASZ) 1246 return -EINVAL; 1247 1248 if (len) { 1249 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1250 if (!new_alias) 1251 return -ENOMEM; 1252 1253 memcpy(new_alias->ifalias, alias, len); 1254 new_alias->ifalias[len] = 0; 1255 } 1256 1257 mutex_lock(&ifalias_mutex); 1258 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1259 mutex_is_locked(&ifalias_mutex)); 1260 mutex_unlock(&ifalias_mutex); 1261 1262 if (new_alias) 1263 kfree_rcu(new_alias, rcuhead); 1264 1265 return len; 1266 } 1267 EXPORT_SYMBOL(dev_set_alias); 1268 1269 /** 1270 * dev_get_alias - get ifalias of a device 1271 * @dev: device 1272 * @name: buffer to store name of ifalias 1273 * @len: size of buffer 1274 * 1275 * get ifalias for a device. Caller must make sure dev cannot go 1276 * away, e.g. rcu read lock or own a reference count to device. 1277 */ 1278 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1279 { 1280 const struct dev_ifalias *alias; 1281 int ret = 0; 1282 1283 rcu_read_lock(); 1284 alias = rcu_dereference(dev->ifalias); 1285 if (alias) 1286 ret = snprintf(name, len, "%s", alias->ifalias); 1287 rcu_read_unlock(); 1288 1289 return ret; 1290 } 1291 1292 /** 1293 * netdev_features_change - device changes features 1294 * @dev: device to cause notification 1295 * 1296 * Called to indicate a device has changed features. 1297 */ 1298 void netdev_features_change(struct net_device *dev) 1299 { 1300 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1301 } 1302 EXPORT_SYMBOL(netdev_features_change); 1303 1304 /** 1305 * netdev_state_change - device changes state 1306 * @dev: device to cause notification 1307 * 1308 * Called to indicate a device has changed state. This function calls 1309 * the notifier chains for netdev_chain and sends a NEWLINK message 1310 * to the routing socket. 1311 */ 1312 void netdev_state_change(struct net_device *dev) 1313 { 1314 if (dev->flags & IFF_UP) { 1315 struct netdev_notifier_change_info change_info = { 1316 .info.dev = dev, 1317 }; 1318 1319 call_netdevice_notifiers_info(NETDEV_CHANGE, 1320 &change_info.info); 1321 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1322 } 1323 } 1324 EXPORT_SYMBOL(netdev_state_change); 1325 1326 /** 1327 * __netdev_notify_peers - notify network peers about existence of @dev, 1328 * to be called when rtnl lock is already held. 1329 * @dev: network device 1330 * 1331 * Generate traffic such that interested network peers are aware of 1332 * @dev, such as by generating a gratuitous ARP. This may be used when 1333 * a device wants to inform the rest of the network about some sort of 1334 * reconfiguration such as a failover event or virtual machine 1335 * migration. 1336 */ 1337 void __netdev_notify_peers(struct net_device *dev) 1338 { 1339 ASSERT_RTNL(); 1340 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1341 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1342 } 1343 EXPORT_SYMBOL(__netdev_notify_peers); 1344 1345 /** 1346 * netdev_notify_peers - notify network peers about existence of @dev 1347 * @dev: network device 1348 * 1349 * Generate traffic such that interested network peers are aware of 1350 * @dev, such as by generating a gratuitous ARP. This may be used when 1351 * a device wants to inform the rest of the network about some sort of 1352 * reconfiguration such as a failover event or virtual machine 1353 * migration. 1354 */ 1355 void netdev_notify_peers(struct net_device *dev) 1356 { 1357 rtnl_lock(); 1358 __netdev_notify_peers(dev); 1359 rtnl_unlock(); 1360 } 1361 EXPORT_SYMBOL(netdev_notify_peers); 1362 1363 static int napi_threaded_poll(void *data); 1364 1365 static int napi_kthread_create(struct napi_struct *n) 1366 { 1367 int err = 0; 1368 1369 /* Create and wake up the kthread once to put it in 1370 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1371 * warning and work with loadavg. 1372 */ 1373 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1374 n->dev->name, n->napi_id); 1375 if (IS_ERR(n->thread)) { 1376 err = PTR_ERR(n->thread); 1377 pr_err("kthread_run failed with err %d\n", err); 1378 n->thread = NULL; 1379 } 1380 1381 return err; 1382 } 1383 1384 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1385 { 1386 const struct net_device_ops *ops = dev->netdev_ops; 1387 int ret; 1388 1389 ASSERT_RTNL(); 1390 dev_addr_check(dev); 1391 1392 if (!netif_device_present(dev)) { 1393 /* may be detached because parent is runtime-suspended */ 1394 if (dev->dev.parent) 1395 pm_runtime_resume(dev->dev.parent); 1396 if (!netif_device_present(dev)) 1397 return -ENODEV; 1398 } 1399 1400 /* Block netpoll from trying to do any rx path servicing. 1401 * If we don't do this there is a chance ndo_poll_controller 1402 * or ndo_poll may be running while we open the device 1403 */ 1404 netpoll_poll_disable(dev); 1405 1406 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1407 ret = notifier_to_errno(ret); 1408 if (ret) 1409 return ret; 1410 1411 set_bit(__LINK_STATE_START, &dev->state); 1412 1413 if (ops->ndo_validate_addr) 1414 ret = ops->ndo_validate_addr(dev); 1415 1416 if (!ret && ops->ndo_open) 1417 ret = ops->ndo_open(dev); 1418 1419 netpoll_poll_enable(dev); 1420 1421 if (ret) 1422 clear_bit(__LINK_STATE_START, &dev->state); 1423 else { 1424 dev->flags |= IFF_UP; 1425 dev_set_rx_mode(dev); 1426 dev_activate(dev); 1427 add_device_randomness(dev->dev_addr, dev->addr_len); 1428 } 1429 1430 return ret; 1431 } 1432 1433 /** 1434 * dev_open - prepare an interface for use. 1435 * @dev: device to open 1436 * @extack: netlink extended ack 1437 * 1438 * Takes a device from down to up state. The device's private open 1439 * function is invoked and then the multicast lists are loaded. Finally 1440 * the device is moved into the up state and a %NETDEV_UP message is 1441 * sent to the netdev notifier chain. 1442 * 1443 * Calling this function on an active interface is a nop. On a failure 1444 * a negative errno code is returned. 1445 */ 1446 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1447 { 1448 int ret; 1449 1450 if (dev->flags & IFF_UP) 1451 return 0; 1452 1453 ret = __dev_open(dev, extack); 1454 if (ret < 0) 1455 return ret; 1456 1457 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1458 call_netdevice_notifiers(NETDEV_UP, dev); 1459 1460 return ret; 1461 } 1462 EXPORT_SYMBOL(dev_open); 1463 1464 static void __dev_close_many(struct list_head *head) 1465 { 1466 struct net_device *dev; 1467 1468 ASSERT_RTNL(); 1469 might_sleep(); 1470 1471 list_for_each_entry(dev, head, close_list) { 1472 /* Temporarily disable netpoll until the interface is down */ 1473 netpoll_poll_disable(dev); 1474 1475 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1476 1477 clear_bit(__LINK_STATE_START, &dev->state); 1478 1479 /* Synchronize to scheduled poll. We cannot touch poll list, it 1480 * can be even on different cpu. So just clear netif_running(). 1481 * 1482 * dev->stop() will invoke napi_disable() on all of it's 1483 * napi_struct instances on this device. 1484 */ 1485 smp_mb__after_atomic(); /* Commit netif_running(). */ 1486 } 1487 1488 dev_deactivate_many(head); 1489 1490 list_for_each_entry(dev, head, close_list) { 1491 const struct net_device_ops *ops = dev->netdev_ops; 1492 1493 /* 1494 * Call the device specific close. This cannot fail. 1495 * Only if device is UP 1496 * 1497 * We allow it to be called even after a DETACH hot-plug 1498 * event. 1499 */ 1500 if (ops->ndo_stop) 1501 ops->ndo_stop(dev); 1502 1503 dev->flags &= ~IFF_UP; 1504 netpoll_poll_enable(dev); 1505 } 1506 } 1507 1508 static void __dev_close(struct net_device *dev) 1509 { 1510 LIST_HEAD(single); 1511 1512 list_add(&dev->close_list, &single); 1513 __dev_close_many(&single); 1514 list_del(&single); 1515 } 1516 1517 void dev_close_many(struct list_head *head, bool unlink) 1518 { 1519 struct net_device *dev, *tmp; 1520 1521 /* Remove the devices that don't need to be closed */ 1522 list_for_each_entry_safe(dev, tmp, head, close_list) 1523 if (!(dev->flags & IFF_UP)) 1524 list_del_init(&dev->close_list); 1525 1526 __dev_close_many(head); 1527 1528 list_for_each_entry_safe(dev, tmp, head, close_list) { 1529 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1530 call_netdevice_notifiers(NETDEV_DOWN, dev); 1531 if (unlink) 1532 list_del_init(&dev->close_list); 1533 } 1534 } 1535 EXPORT_SYMBOL(dev_close_many); 1536 1537 /** 1538 * dev_close - shutdown an interface. 1539 * @dev: device to shutdown 1540 * 1541 * This function moves an active device into down state. A 1542 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1543 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1544 * chain. 1545 */ 1546 void dev_close(struct net_device *dev) 1547 { 1548 if (dev->flags & IFF_UP) { 1549 LIST_HEAD(single); 1550 1551 list_add(&dev->close_list, &single); 1552 dev_close_many(&single, true); 1553 list_del(&single); 1554 } 1555 } 1556 EXPORT_SYMBOL(dev_close); 1557 1558 1559 /** 1560 * dev_disable_lro - disable Large Receive Offload on a device 1561 * @dev: device 1562 * 1563 * Disable Large Receive Offload (LRO) on a net device. Must be 1564 * called under RTNL. This is needed if received packets may be 1565 * forwarded to another interface. 1566 */ 1567 void dev_disable_lro(struct net_device *dev) 1568 { 1569 struct net_device *lower_dev; 1570 struct list_head *iter; 1571 1572 dev->wanted_features &= ~NETIF_F_LRO; 1573 netdev_update_features(dev); 1574 1575 if (unlikely(dev->features & NETIF_F_LRO)) 1576 netdev_WARN(dev, "failed to disable LRO!\n"); 1577 1578 netdev_for_each_lower_dev(dev, lower_dev, iter) 1579 dev_disable_lro(lower_dev); 1580 } 1581 EXPORT_SYMBOL(dev_disable_lro); 1582 1583 /** 1584 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1585 * @dev: device 1586 * 1587 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1588 * called under RTNL. This is needed if Generic XDP is installed on 1589 * the device. 1590 */ 1591 static void dev_disable_gro_hw(struct net_device *dev) 1592 { 1593 dev->wanted_features &= ~NETIF_F_GRO_HW; 1594 netdev_update_features(dev); 1595 1596 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1597 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1598 } 1599 1600 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1601 { 1602 #define N(val) \ 1603 case NETDEV_##val: \ 1604 return "NETDEV_" __stringify(val); 1605 switch (cmd) { 1606 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1607 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1608 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1609 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1610 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1611 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1612 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1613 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1614 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1615 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1616 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1617 } 1618 #undef N 1619 return "UNKNOWN_NETDEV_EVENT"; 1620 } 1621 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1622 1623 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1624 struct net_device *dev) 1625 { 1626 struct netdev_notifier_info info = { 1627 .dev = dev, 1628 }; 1629 1630 return nb->notifier_call(nb, val, &info); 1631 } 1632 1633 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1634 struct net_device *dev) 1635 { 1636 int err; 1637 1638 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1639 err = notifier_to_errno(err); 1640 if (err) 1641 return err; 1642 1643 if (!(dev->flags & IFF_UP)) 1644 return 0; 1645 1646 call_netdevice_notifier(nb, NETDEV_UP, dev); 1647 return 0; 1648 } 1649 1650 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1651 struct net_device *dev) 1652 { 1653 if (dev->flags & IFF_UP) { 1654 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1655 dev); 1656 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1657 } 1658 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1659 } 1660 1661 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1662 struct net *net) 1663 { 1664 struct net_device *dev; 1665 int err; 1666 1667 for_each_netdev(net, dev) { 1668 err = call_netdevice_register_notifiers(nb, dev); 1669 if (err) 1670 goto rollback; 1671 } 1672 return 0; 1673 1674 rollback: 1675 for_each_netdev_continue_reverse(net, dev) 1676 call_netdevice_unregister_notifiers(nb, dev); 1677 return err; 1678 } 1679 1680 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1681 struct net *net) 1682 { 1683 struct net_device *dev; 1684 1685 for_each_netdev(net, dev) 1686 call_netdevice_unregister_notifiers(nb, dev); 1687 } 1688 1689 static int dev_boot_phase = 1; 1690 1691 /** 1692 * register_netdevice_notifier - register a network notifier block 1693 * @nb: notifier 1694 * 1695 * Register a notifier to be called when network device events occur. 1696 * The notifier passed is linked into the kernel structures and must 1697 * not be reused until it has been unregistered. A negative errno code 1698 * is returned on a failure. 1699 * 1700 * When registered all registration and up events are replayed 1701 * to the new notifier to allow device to have a race free 1702 * view of the network device list. 1703 */ 1704 1705 int register_netdevice_notifier(struct notifier_block *nb) 1706 { 1707 struct net *net; 1708 int err; 1709 1710 /* Close race with setup_net() and cleanup_net() */ 1711 down_write(&pernet_ops_rwsem); 1712 rtnl_lock(); 1713 err = raw_notifier_chain_register(&netdev_chain, nb); 1714 if (err) 1715 goto unlock; 1716 if (dev_boot_phase) 1717 goto unlock; 1718 for_each_net(net) { 1719 err = call_netdevice_register_net_notifiers(nb, net); 1720 if (err) 1721 goto rollback; 1722 } 1723 1724 unlock: 1725 rtnl_unlock(); 1726 up_write(&pernet_ops_rwsem); 1727 return err; 1728 1729 rollback: 1730 for_each_net_continue_reverse(net) 1731 call_netdevice_unregister_net_notifiers(nb, net); 1732 1733 raw_notifier_chain_unregister(&netdev_chain, nb); 1734 goto unlock; 1735 } 1736 EXPORT_SYMBOL(register_netdevice_notifier); 1737 1738 /** 1739 * unregister_netdevice_notifier - unregister a network notifier block 1740 * @nb: notifier 1741 * 1742 * Unregister a notifier previously registered by 1743 * register_netdevice_notifier(). The notifier is unlinked into the 1744 * kernel structures and may then be reused. A negative errno code 1745 * is returned on a failure. 1746 * 1747 * After unregistering unregister and down device events are synthesized 1748 * for all devices on the device list to the removed notifier to remove 1749 * the need for special case cleanup code. 1750 */ 1751 1752 int unregister_netdevice_notifier(struct notifier_block *nb) 1753 { 1754 struct net *net; 1755 int err; 1756 1757 /* Close race with setup_net() and cleanup_net() */ 1758 down_write(&pernet_ops_rwsem); 1759 rtnl_lock(); 1760 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1761 if (err) 1762 goto unlock; 1763 1764 for_each_net(net) 1765 call_netdevice_unregister_net_notifiers(nb, net); 1766 1767 unlock: 1768 rtnl_unlock(); 1769 up_write(&pernet_ops_rwsem); 1770 return err; 1771 } 1772 EXPORT_SYMBOL(unregister_netdevice_notifier); 1773 1774 static int __register_netdevice_notifier_net(struct net *net, 1775 struct notifier_block *nb, 1776 bool ignore_call_fail) 1777 { 1778 int err; 1779 1780 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1781 if (err) 1782 return err; 1783 if (dev_boot_phase) 1784 return 0; 1785 1786 err = call_netdevice_register_net_notifiers(nb, net); 1787 if (err && !ignore_call_fail) 1788 goto chain_unregister; 1789 1790 return 0; 1791 1792 chain_unregister: 1793 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1794 return err; 1795 } 1796 1797 static int __unregister_netdevice_notifier_net(struct net *net, 1798 struct notifier_block *nb) 1799 { 1800 int err; 1801 1802 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1803 if (err) 1804 return err; 1805 1806 call_netdevice_unregister_net_notifiers(nb, net); 1807 return 0; 1808 } 1809 1810 /** 1811 * register_netdevice_notifier_net - register a per-netns network notifier block 1812 * @net: network namespace 1813 * @nb: notifier 1814 * 1815 * Register a notifier to be called when network device events occur. 1816 * The notifier passed is linked into the kernel structures and must 1817 * not be reused until it has been unregistered. A negative errno code 1818 * is returned on a failure. 1819 * 1820 * When registered all registration and up events are replayed 1821 * to the new notifier to allow device to have a race free 1822 * view of the network device list. 1823 */ 1824 1825 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1826 { 1827 int err; 1828 1829 rtnl_lock(); 1830 err = __register_netdevice_notifier_net(net, nb, false); 1831 rtnl_unlock(); 1832 return err; 1833 } 1834 EXPORT_SYMBOL(register_netdevice_notifier_net); 1835 1836 /** 1837 * unregister_netdevice_notifier_net - unregister a per-netns 1838 * network notifier block 1839 * @net: network namespace 1840 * @nb: notifier 1841 * 1842 * Unregister a notifier previously registered by 1843 * register_netdevice_notifier_net(). The notifier is unlinked from the 1844 * kernel structures and may then be reused. A negative errno code 1845 * is returned on a failure. 1846 * 1847 * After unregistering unregister and down device events are synthesized 1848 * for all devices on the device list to the removed notifier to remove 1849 * the need for special case cleanup code. 1850 */ 1851 1852 int unregister_netdevice_notifier_net(struct net *net, 1853 struct notifier_block *nb) 1854 { 1855 int err; 1856 1857 rtnl_lock(); 1858 err = __unregister_netdevice_notifier_net(net, nb); 1859 rtnl_unlock(); 1860 return err; 1861 } 1862 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1863 1864 static void __move_netdevice_notifier_net(struct net *src_net, 1865 struct net *dst_net, 1866 struct notifier_block *nb) 1867 { 1868 __unregister_netdevice_notifier_net(src_net, nb); 1869 __register_netdevice_notifier_net(dst_net, nb, true); 1870 } 1871 1872 void move_netdevice_notifier_net(struct net *src_net, struct net *dst_net, 1873 struct notifier_block *nb) 1874 { 1875 rtnl_lock(); 1876 __move_netdevice_notifier_net(src_net, dst_net, nb); 1877 rtnl_unlock(); 1878 } 1879 1880 int register_netdevice_notifier_dev_net(struct net_device *dev, 1881 struct notifier_block *nb, 1882 struct netdev_net_notifier *nn) 1883 { 1884 int err; 1885 1886 rtnl_lock(); 1887 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1888 if (!err) { 1889 nn->nb = nb; 1890 list_add(&nn->list, &dev->net_notifier_list); 1891 } 1892 rtnl_unlock(); 1893 return err; 1894 } 1895 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1896 1897 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1898 struct notifier_block *nb, 1899 struct netdev_net_notifier *nn) 1900 { 1901 int err; 1902 1903 rtnl_lock(); 1904 list_del(&nn->list); 1905 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1906 rtnl_unlock(); 1907 return err; 1908 } 1909 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1910 1911 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1912 struct net *net) 1913 { 1914 struct netdev_net_notifier *nn; 1915 1916 list_for_each_entry(nn, &dev->net_notifier_list, list) 1917 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1918 } 1919 1920 /** 1921 * call_netdevice_notifiers_info - call all network notifier blocks 1922 * @val: value passed unmodified to notifier function 1923 * @info: notifier information data 1924 * 1925 * Call all network notifier blocks. Parameters and return value 1926 * are as for raw_notifier_call_chain(). 1927 */ 1928 1929 static int call_netdevice_notifiers_info(unsigned long val, 1930 struct netdev_notifier_info *info) 1931 { 1932 struct net *net = dev_net(info->dev); 1933 int ret; 1934 1935 ASSERT_RTNL(); 1936 1937 /* Run per-netns notifier block chain first, then run the global one. 1938 * Hopefully, one day, the global one is going to be removed after 1939 * all notifier block registrators get converted to be per-netns. 1940 */ 1941 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1942 if (ret & NOTIFY_STOP_MASK) 1943 return ret; 1944 return raw_notifier_call_chain(&netdev_chain, val, info); 1945 } 1946 1947 /** 1948 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1949 * for and rollback on error 1950 * @val_up: value passed unmodified to notifier function 1951 * @val_down: value passed unmodified to the notifier function when 1952 * recovering from an error on @val_up 1953 * @info: notifier information data 1954 * 1955 * Call all per-netns network notifier blocks, but not notifier blocks on 1956 * the global notifier chain. Parameters and return value are as for 1957 * raw_notifier_call_chain_robust(). 1958 */ 1959 1960 static int 1961 call_netdevice_notifiers_info_robust(unsigned long val_up, 1962 unsigned long val_down, 1963 struct netdev_notifier_info *info) 1964 { 1965 struct net *net = dev_net(info->dev); 1966 1967 ASSERT_RTNL(); 1968 1969 return raw_notifier_call_chain_robust(&net->netdev_chain, 1970 val_up, val_down, info); 1971 } 1972 1973 static int call_netdevice_notifiers_extack(unsigned long val, 1974 struct net_device *dev, 1975 struct netlink_ext_ack *extack) 1976 { 1977 struct netdev_notifier_info info = { 1978 .dev = dev, 1979 .extack = extack, 1980 }; 1981 1982 return call_netdevice_notifiers_info(val, &info); 1983 } 1984 1985 /** 1986 * call_netdevice_notifiers - call all network notifier blocks 1987 * @val: value passed unmodified to notifier function 1988 * @dev: net_device pointer passed unmodified to notifier function 1989 * 1990 * Call all network notifier blocks. Parameters and return value 1991 * are as for raw_notifier_call_chain(). 1992 */ 1993 1994 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1995 { 1996 return call_netdevice_notifiers_extack(val, dev, NULL); 1997 } 1998 EXPORT_SYMBOL(call_netdevice_notifiers); 1999 2000 /** 2001 * call_netdevice_notifiers_mtu - call all network notifier blocks 2002 * @val: value passed unmodified to notifier function 2003 * @dev: net_device pointer passed unmodified to notifier function 2004 * @arg: additional u32 argument passed to the notifier function 2005 * 2006 * Call all network notifier blocks. Parameters and return value 2007 * are as for raw_notifier_call_chain(). 2008 */ 2009 static int call_netdevice_notifiers_mtu(unsigned long val, 2010 struct net_device *dev, u32 arg) 2011 { 2012 struct netdev_notifier_info_ext info = { 2013 .info.dev = dev, 2014 .ext.mtu = arg, 2015 }; 2016 2017 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2018 2019 return call_netdevice_notifiers_info(val, &info.info); 2020 } 2021 2022 #ifdef CONFIG_NET_INGRESS 2023 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2024 2025 void net_inc_ingress_queue(void) 2026 { 2027 static_branch_inc(&ingress_needed_key); 2028 } 2029 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2030 2031 void net_dec_ingress_queue(void) 2032 { 2033 static_branch_dec(&ingress_needed_key); 2034 } 2035 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2036 #endif 2037 2038 #ifdef CONFIG_NET_EGRESS 2039 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2040 2041 void net_inc_egress_queue(void) 2042 { 2043 static_branch_inc(&egress_needed_key); 2044 } 2045 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2046 2047 void net_dec_egress_queue(void) 2048 { 2049 static_branch_dec(&egress_needed_key); 2050 } 2051 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2052 #endif 2053 2054 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2055 EXPORT_SYMBOL(netstamp_needed_key); 2056 #ifdef CONFIG_JUMP_LABEL 2057 static atomic_t netstamp_needed_deferred; 2058 static atomic_t netstamp_wanted; 2059 static void netstamp_clear(struct work_struct *work) 2060 { 2061 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2062 int wanted; 2063 2064 wanted = atomic_add_return(deferred, &netstamp_wanted); 2065 if (wanted > 0) 2066 static_branch_enable(&netstamp_needed_key); 2067 else 2068 static_branch_disable(&netstamp_needed_key); 2069 } 2070 static DECLARE_WORK(netstamp_work, netstamp_clear); 2071 #endif 2072 2073 void net_enable_timestamp(void) 2074 { 2075 #ifdef CONFIG_JUMP_LABEL 2076 int wanted = atomic_read(&netstamp_wanted); 2077 2078 while (wanted > 0) { 2079 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2080 return; 2081 } 2082 atomic_inc(&netstamp_needed_deferred); 2083 schedule_work(&netstamp_work); 2084 #else 2085 static_branch_inc(&netstamp_needed_key); 2086 #endif 2087 } 2088 EXPORT_SYMBOL(net_enable_timestamp); 2089 2090 void net_disable_timestamp(void) 2091 { 2092 #ifdef CONFIG_JUMP_LABEL 2093 int wanted = atomic_read(&netstamp_wanted); 2094 2095 while (wanted > 1) { 2096 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2097 return; 2098 } 2099 atomic_dec(&netstamp_needed_deferred); 2100 schedule_work(&netstamp_work); 2101 #else 2102 static_branch_dec(&netstamp_needed_key); 2103 #endif 2104 } 2105 EXPORT_SYMBOL(net_disable_timestamp); 2106 2107 static inline void net_timestamp_set(struct sk_buff *skb) 2108 { 2109 skb->tstamp = 0; 2110 skb->mono_delivery_time = 0; 2111 if (static_branch_unlikely(&netstamp_needed_key)) 2112 skb->tstamp = ktime_get_real(); 2113 } 2114 2115 #define net_timestamp_check(COND, SKB) \ 2116 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2117 if ((COND) && !(SKB)->tstamp) \ 2118 (SKB)->tstamp = ktime_get_real(); \ 2119 } \ 2120 2121 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2122 { 2123 return __is_skb_forwardable(dev, skb, true); 2124 } 2125 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2126 2127 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2128 bool check_mtu) 2129 { 2130 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2131 2132 if (likely(!ret)) { 2133 skb->protocol = eth_type_trans(skb, dev); 2134 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2135 } 2136 2137 return ret; 2138 } 2139 2140 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2141 { 2142 return __dev_forward_skb2(dev, skb, true); 2143 } 2144 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2145 2146 /** 2147 * dev_forward_skb - loopback an skb to another netif 2148 * 2149 * @dev: destination network device 2150 * @skb: buffer to forward 2151 * 2152 * return values: 2153 * NET_RX_SUCCESS (no congestion) 2154 * NET_RX_DROP (packet was dropped, but freed) 2155 * 2156 * dev_forward_skb can be used for injecting an skb from the 2157 * start_xmit function of one device into the receive queue 2158 * of another device. 2159 * 2160 * The receiving device may be in another namespace, so 2161 * we have to clear all information in the skb that could 2162 * impact namespace isolation. 2163 */ 2164 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2165 { 2166 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2167 } 2168 EXPORT_SYMBOL_GPL(dev_forward_skb); 2169 2170 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2171 { 2172 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2173 } 2174 2175 static inline int deliver_skb(struct sk_buff *skb, 2176 struct packet_type *pt_prev, 2177 struct net_device *orig_dev) 2178 { 2179 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2180 return -ENOMEM; 2181 refcount_inc(&skb->users); 2182 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2183 } 2184 2185 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2186 struct packet_type **pt, 2187 struct net_device *orig_dev, 2188 __be16 type, 2189 struct list_head *ptype_list) 2190 { 2191 struct packet_type *ptype, *pt_prev = *pt; 2192 2193 list_for_each_entry_rcu(ptype, ptype_list, list) { 2194 if (ptype->type != type) 2195 continue; 2196 if (pt_prev) 2197 deliver_skb(skb, pt_prev, orig_dev); 2198 pt_prev = ptype; 2199 } 2200 *pt = pt_prev; 2201 } 2202 2203 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2204 { 2205 if (!ptype->af_packet_priv || !skb->sk) 2206 return false; 2207 2208 if (ptype->id_match) 2209 return ptype->id_match(ptype, skb->sk); 2210 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2211 return true; 2212 2213 return false; 2214 } 2215 2216 /** 2217 * dev_nit_active - return true if any network interface taps are in use 2218 * 2219 * @dev: network device to check for the presence of taps 2220 */ 2221 bool dev_nit_active(struct net_device *dev) 2222 { 2223 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2224 } 2225 EXPORT_SYMBOL_GPL(dev_nit_active); 2226 2227 /* 2228 * Support routine. Sends outgoing frames to any network 2229 * taps currently in use. 2230 */ 2231 2232 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2233 { 2234 struct packet_type *ptype; 2235 struct sk_buff *skb2 = NULL; 2236 struct packet_type *pt_prev = NULL; 2237 struct list_head *ptype_list = &ptype_all; 2238 2239 rcu_read_lock(); 2240 again: 2241 list_for_each_entry_rcu(ptype, ptype_list, list) { 2242 if (ptype->ignore_outgoing) 2243 continue; 2244 2245 /* Never send packets back to the socket 2246 * they originated from - MvS (miquels@drinkel.ow.org) 2247 */ 2248 if (skb_loop_sk(ptype, skb)) 2249 continue; 2250 2251 if (pt_prev) { 2252 deliver_skb(skb2, pt_prev, skb->dev); 2253 pt_prev = ptype; 2254 continue; 2255 } 2256 2257 /* need to clone skb, done only once */ 2258 skb2 = skb_clone(skb, GFP_ATOMIC); 2259 if (!skb2) 2260 goto out_unlock; 2261 2262 net_timestamp_set(skb2); 2263 2264 /* skb->nh should be correctly 2265 * set by sender, so that the second statement is 2266 * just protection against buggy protocols. 2267 */ 2268 skb_reset_mac_header(skb2); 2269 2270 if (skb_network_header(skb2) < skb2->data || 2271 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2272 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2273 ntohs(skb2->protocol), 2274 dev->name); 2275 skb_reset_network_header(skb2); 2276 } 2277 2278 skb2->transport_header = skb2->network_header; 2279 skb2->pkt_type = PACKET_OUTGOING; 2280 pt_prev = ptype; 2281 } 2282 2283 if (ptype_list == &ptype_all) { 2284 ptype_list = &dev->ptype_all; 2285 goto again; 2286 } 2287 out_unlock: 2288 if (pt_prev) { 2289 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2290 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2291 else 2292 kfree_skb(skb2); 2293 } 2294 rcu_read_unlock(); 2295 } 2296 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2297 2298 /** 2299 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2300 * @dev: Network device 2301 * @txq: number of queues available 2302 * 2303 * If real_num_tx_queues is changed the tc mappings may no longer be 2304 * valid. To resolve this verify the tc mapping remains valid and if 2305 * not NULL the mapping. With no priorities mapping to this 2306 * offset/count pair it will no longer be used. In the worst case TC0 2307 * is invalid nothing can be done so disable priority mappings. If is 2308 * expected that drivers will fix this mapping if they can before 2309 * calling netif_set_real_num_tx_queues. 2310 */ 2311 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2312 { 2313 int i; 2314 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2315 2316 /* If TC0 is invalidated disable TC mapping */ 2317 if (tc->offset + tc->count > txq) { 2318 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2319 dev->num_tc = 0; 2320 return; 2321 } 2322 2323 /* Invalidated prio to tc mappings set to TC0 */ 2324 for (i = 1; i < TC_BITMASK + 1; i++) { 2325 int q = netdev_get_prio_tc_map(dev, i); 2326 2327 tc = &dev->tc_to_txq[q]; 2328 if (tc->offset + tc->count > txq) { 2329 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", 2330 i, q); 2331 netdev_set_prio_tc_map(dev, i, 0); 2332 } 2333 } 2334 } 2335 2336 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2337 { 2338 if (dev->num_tc) { 2339 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2340 int i; 2341 2342 /* walk through the TCs and see if it falls into any of them */ 2343 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2344 if ((txq - tc->offset) < tc->count) 2345 return i; 2346 } 2347 2348 /* didn't find it, just return -1 to indicate no match */ 2349 return -1; 2350 } 2351 2352 return 0; 2353 } 2354 EXPORT_SYMBOL(netdev_txq_to_tc); 2355 2356 #ifdef CONFIG_XPS 2357 static struct static_key xps_needed __read_mostly; 2358 static struct static_key xps_rxqs_needed __read_mostly; 2359 static DEFINE_MUTEX(xps_map_mutex); 2360 #define xmap_dereference(P) \ 2361 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2362 2363 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2364 struct xps_dev_maps *old_maps, int tci, u16 index) 2365 { 2366 struct xps_map *map = NULL; 2367 int pos; 2368 2369 if (dev_maps) 2370 map = xmap_dereference(dev_maps->attr_map[tci]); 2371 if (!map) 2372 return false; 2373 2374 for (pos = map->len; pos--;) { 2375 if (map->queues[pos] != index) 2376 continue; 2377 2378 if (map->len > 1) { 2379 map->queues[pos] = map->queues[--map->len]; 2380 break; 2381 } 2382 2383 if (old_maps) 2384 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2385 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2386 kfree_rcu(map, rcu); 2387 return false; 2388 } 2389 2390 return true; 2391 } 2392 2393 static bool remove_xps_queue_cpu(struct net_device *dev, 2394 struct xps_dev_maps *dev_maps, 2395 int cpu, u16 offset, u16 count) 2396 { 2397 int num_tc = dev_maps->num_tc; 2398 bool active = false; 2399 int tci; 2400 2401 for (tci = cpu * num_tc; num_tc--; tci++) { 2402 int i, j; 2403 2404 for (i = count, j = offset; i--; j++) { 2405 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2406 break; 2407 } 2408 2409 active |= i < 0; 2410 } 2411 2412 return active; 2413 } 2414 2415 static void reset_xps_maps(struct net_device *dev, 2416 struct xps_dev_maps *dev_maps, 2417 enum xps_map_type type) 2418 { 2419 static_key_slow_dec_cpuslocked(&xps_needed); 2420 if (type == XPS_RXQS) 2421 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2422 2423 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2424 2425 kfree_rcu(dev_maps, rcu); 2426 } 2427 2428 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2429 u16 offset, u16 count) 2430 { 2431 struct xps_dev_maps *dev_maps; 2432 bool active = false; 2433 int i, j; 2434 2435 dev_maps = xmap_dereference(dev->xps_maps[type]); 2436 if (!dev_maps) 2437 return; 2438 2439 for (j = 0; j < dev_maps->nr_ids; j++) 2440 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2441 if (!active) 2442 reset_xps_maps(dev, dev_maps, type); 2443 2444 if (type == XPS_CPUS) { 2445 for (i = offset + (count - 1); count--; i--) 2446 netdev_queue_numa_node_write( 2447 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2448 } 2449 } 2450 2451 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2452 u16 count) 2453 { 2454 if (!static_key_false(&xps_needed)) 2455 return; 2456 2457 cpus_read_lock(); 2458 mutex_lock(&xps_map_mutex); 2459 2460 if (static_key_false(&xps_rxqs_needed)) 2461 clean_xps_maps(dev, XPS_RXQS, offset, count); 2462 2463 clean_xps_maps(dev, XPS_CPUS, offset, count); 2464 2465 mutex_unlock(&xps_map_mutex); 2466 cpus_read_unlock(); 2467 } 2468 2469 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2470 { 2471 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2472 } 2473 2474 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2475 u16 index, bool is_rxqs_map) 2476 { 2477 struct xps_map *new_map; 2478 int alloc_len = XPS_MIN_MAP_ALLOC; 2479 int i, pos; 2480 2481 for (pos = 0; map && pos < map->len; pos++) { 2482 if (map->queues[pos] != index) 2483 continue; 2484 return map; 2485 } 2486 2487 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2488 if (map) { 2489 if (pos < map->alloc_len) 2490 return map; 2491 2492 alloc_len = map->alloc_len * 2; 2493 } 2494 2495 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2496 * map 2497 */ 2498 if (is_rxqs_map) 2499 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2500 else 2501 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2502 cpu_to_node(attr_index)); 2503 if (!new_map) 2504 return NULL; 2505 2506 for (i = 0; i < pos; i++) 2507 new_map->queues[i] = map->queues[i]; 2508 new_map->alloc_len = alloc_len; 2509 new_map->len = pos; 2510 2511 return new_map; 2512 } 2513 2514 /* Copy xps maps at a given index */ 2515 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2516 struct xps_dev_maps *new_dev_maps, int index, 2517 int tc, bool skip_tc) 2518 { 2519 int i, tci = index * dev_maps->num_tc; 2520 struct xps_map *map; 2521 2522 /* copy maps belonging to foreign traffic classes */ 2523 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2524 if (i == tc && skip_tc) 2525 continue; 2526 2527 /* fill in the new device map from the old device map */ 2528 map = xmap_dereference(dev_maps->attr_map[tci]); 2529 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2530 } 2531 } 2532 2533 /* Must be called under cpus_read_lock */ 2534 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2535 u16 index, enum xps_map_type type) 2536 { 2537 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2538 const unsigned long *online_mask = NULL; 2539 bool active = false, copy = false; 2540 int i, j, tci, numa_node_id = -2; 2541 int maps_sz, num_tc = 1, tc = 0; 2542 struct xps_map *map, *new_map; 2543 unsigned int nr_ids; 2544 2545 if (dev->num_tc) { 2546 /* Do not allow XPS on subordinate device directly */ 2547 num_tc = dev->num_tc; 2548 if (num_tc < 0) 2549 return -EINVAL; 2550 2551 /* If queue belongs to subordinate dev use its map */ 2552 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2553 2554 tc = netdev_txq_to_tc(dev, index); 2555 if (tc < 0) 2556 return -EINVAL; 2557 } 2558 2559 mutex_lock(&xps_map_mutex); 2560 2561 dev_maps = xmap_dereference(dev->xps_maps[type]); 2562 if (type == XPS_RXQS) { 2563 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2564 nr_ids = dev->num_rx_queues; 2565 } else { 2566 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2567 if (num_possible_cpus() > 1) 2568 online_mask = cpumask_bits(cpu_online_mask); 2569 nr_ids = nr_cpu_ids; 2570 } 2571 2572 if (maps_sz < L1_CACHE_BYTES) 2573 maps_sz = L1_CACHE_BYTES; 2574 2575 /* The old dev_maps could be larger or smaller than the one we're 2576 * setting up now, as dev->num_tc or nr_ids could have been updated in 2577 * between. We could try to be smart, but let's be safe instead and only 2578 * copy foreign traffic classes if the two map sizes match. 2579 */ 2580 if (dev_maps && 2581 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2582 copy = true; 2583 2584 /* allocate memory for queue storage */ 2585 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2586 j < nr_ids;) { 2587 if (!new_dev_maps) { 2588 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2589 if (!new_dev_maps) { 2590 mutex_unlock(&xps_map_mutex); 2591 return -ENOMEM; 2592 } 2593 2594 new_dev_maps->nr_ids = nr_ids; 2595 new_dev_maps->num_tc = num_tc; 2596 } 2597 2598 tci = j * num_tc + tc; 2599 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2600 2601 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2602 if (!map) 2603 goto error; 2604 2605 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2606 } 2607 2608 if (!new_dev_maps) 2609 goto out_no_new_maps; 2610 2611 if (!dev_maps) { 2612 /* Increment static keys at most once per type */ 2613 static_key_slow_inc_cpuslocked(&xps_needed); 2614 if (type == XPS_RXQS) 2615 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2616 } 2617 2618 for (j = 0; j < nr_ids; j++) { 2619 bool skip_tc = false; 2620 2621 tci = j * num_tc + tc; 2622 if (netif_attr_test_mask(j, mask, nr_ids) && 2623 netif_attr_test_online(j, online_mask, nr_ids)) { 2624 /* add tx-queue to CPU/rx-queue maps */ 2625 int pos = 0; 2626 2627 skip_tc = true; 2628 2629 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2630 while ((pos < map->len) && (map->queues[pos] != index)) 2631 pos++; 2632 2633 if (pos == map->len) 2634 map->queues[map->len++] = index; 2635 #ifdef CONFIG_NUMA 2636 if (type == XPS_CPUS) { 2637 if (numa_node_id == -2) 2638 numa_node_id = cpu_to_node(j); 2639 else if (numa_node_id != cpu_to_node(j)) 2640 numa_node_id = -1; 2641 } 2642 #endif 2643 } 2644 2645 if (copy) 2646 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2647 skip_tc); 2648 } 2649 2650 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2651 2652 /* Cleanup old maps */ 2653 if (!dev_maps) 2654 goto out_no_old_maps; 2655 2656 for (j = 0; j < dev_maps->nr_ids; j++) { 2657 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2658 map = xmap_dereference(dev_maps->attr_map[tci]); 2659 if (!map) 2660 continue; 2661 2662 if (copy) { 2663 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2664 if (map == new_map) 2665 continue; 2666 } 2667 2668 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2669 kfree_rcu(map, rcu); 2670 } 2671 } 2672 2673 old_dev_maps = dev_maps; 2674 2675 out_no_old_maps: 2676 dev_maps = new_dev_maps; 2677 active = true; 2678 2679 out_no_new_maps: 2680 if (type == XPS_CPUS) 2681 /* update Tx queue numa node */ 2682 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2683 (numa_node_id >= 0) ? 2684 numa_node_id : NUMA_NO_NODE); 2685 2686 if (!dev_maps) 2687 goto out_no_maps; 2688 2689 /* removes tx-queue from unused CPUs/rx-queues */ 2690 for (j = 0; j < dev_maps->nr_ids; j++) { 2691 tci = j * dev_maps->num_tc; 2692 2693 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2694 if (i == tc && 2695 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2696 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2697 continue; 2698 2699 active |= remove_xps_queue(dev_maps, 2700 copy ? old_dev_maps : NULL, 2701 tci, index); 2702 } 2703 } 2704 2705 if (old_dev_maps) 2706 kfree_rcu(old_dev_maps, rcu); 2707 2708 /* free map if not active */ 2709 if (!active) 2710 reset_xps_maps(dev, dev_maps, type); 2711 2712 out_no_maps: 2713 mutex_unlock(&xps_map_mutex); 2714 2715 return 0; 2716 error: 2717 /* remove any maps that we added */ 2718 for (j = 0; j < nr_ids; j++) { 2719 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2720 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2721 map = copy ? 2722 xmap_dereference(dev_maps->attr_map[tci]) : 2723 NULL; 2724 if (new_map && new_map != map) 2725 kfree(new_map); 2726 } 2727 } 2728 2729 mutex_unlock(&xps_map_mutex); 2730 2731 kfree(new_dev_maps); 2732 return -ENOMEM; 2733 } 2734 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2735 2736 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2737 u16 index) 2738 { 2739 int ret; 2740 2741 cpus_read_lock(); 2742 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2743 cpus_read_unlock(); 2744 2745 return ret; 2746 } 2747 EXPORT_SYMBOL(netif_set_xps_queue); 2748 2749 #endif 2750 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2751 { 2752 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2753 2754 /* Unbind any subordinate channels */ 2755 while (txq-- != &dev->_tx[0]) { 2756 if (txq->sb_dev) 2757 netdev_unbind_sb_channel(dev, txq->sb_dev); 2758 } 2759 } 2760 2761 void netdev_reset_tc(struct net_device *dev) 2762 { 2763 #ifdef CONFIG_XPS 2764 netif_reset_xps_queues_gt(dev, 0); 2765 #endif 2766 netdev_unbind_all_sb_channels(dev); 2767 2768 /* Reset TC configuration of device */ 2769 dev->num_tc = 0; 2770 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2771 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2772 } 2773 EXPORT_SYMBOL(netdev_reset_tc); 2774 2775 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2776 { 2777 if (tc >= dev->num_tc) 2778 return -EINVAL; 2779 2780 #ifdef CONFIG_XPS 2781 netif_reset_xps_queues(dev, offset, count); 2782 #endif 2783 dev->tc_to_txq[tc].count = count; 2784 dev->tc_to_txq[tc].offset = offset; 2785 return 0; 2786 } 2787 EXPORT_SYMBOL(netdev_set_tc_queue); 2788 2789 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2790 { 2791 if (num_tc > TC_MAX_QUEUE) 2792 return -EINVAL; 2793 2794 #ifdef CONFIG_XPS 2795 netif_reset_xps_queues_gt(dev, 0); 2796 #endif 2797 netdev_unbind_all_sb_channels(dev); 2798 2799 dev->num_tc = num_tc; 2800 return 0; 2801 } 2802 EXPORT_SYMBOL(netdev_set_num_tc); 2803 2804 void netdev_unbind_sb_channel(struct net_device *dev, 2805 struct net_device *sb_dev) 2806 { 2807 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2808 2809 #ifdef CONFIG_XPS 2810 netif_reset_xps_queues_gt(sb_dev, 0); 2811 #endif 2812 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2813 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2814 2815 while (txq-- != &dev->_tx[0]) { 2816 if (txq->sb_dev == sb_dev) 2817 txq->sb_dev = NULL; 2818 } 2819 } 2820 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2821 2822 int netdev_bind_sb_channel_queue(struct net_device *dev, 2823 struct net_device *sb_dev, 2824 u8 tc, u16 count, u16 offset) 2825 { 2826 /* Make certain the sb_dev and dev are already configured */ 2827 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2828 return -EINVAL; 2829 2830 /* We cannot hand out queues we don't have */ 2831 if ((offset + count) > dev->real_num_tx_queues) 2832 return -EINVAL; 2833 2834 /* Record the mapping */ 2835 sb_dev->tc_to_txq[tc].count = count; 2836 sb_dev->tc_to_txq[tc].offset = offset; 2837 2838 /* Provide a way for Tx queue to find the tc_to_txq map or 2839 * XPS map for itself. 2840 */ 2841 while (count--) 2842 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2843 2844 return 0; 2845 } 2846 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2847 2848 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2849 { 2850 /* Do not use a multiqueue device to represent a subordinate channel */ 2851 if (netif_is_multiqueue(dev)) 2852 return -ENODEV; 2853 2854 /* We allow channels 1 - 32767 to be used for subordinate channels. 2855 * Channel 0 is meant to be "native" mode and used only to represent 2856 * the main root device. We allow writing 0 to reset the device back 2857 * to normal mode after being used as a subordinate channel. 2858 */ 2859 if (channel > S16_MAX) 2860 return -EINVAL; 2861 2862 dev->num_tc = -channel; 2863 2864 return 0; 2865 } 2866 EXPORT_SYMBOL(netdev_set_sb_channel); 2867 2868 /* 2869 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2870 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2871 */ 2872 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2873 { 2874 bool disabling; 2875 int rc; 2876 2877 disabling = txq < dev->real_num_tx_queues; 2878 2879 if (txq < 1 || txq > dev->num_tx_queues) 2880 return -EINVAL; 2881 2882 if (dev->reg_state == NETREG_REGISTERED || 2883 dev->reg_state == NETREG_UNREGISTERING) { 2884 ASSERT_RTNL(); 2885 2886 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2887 txq); 2888 if (rc) 2889 return rc; 2890 2891 if (dev->num_tc) 2892 netif_setup_tc(dev, txq); 2893 2894 dev_qdisc_change_real_num_tx(dev, txq); 2895 2896 dev->real_num_tx_queues = txq; 2897 2898 if (disabling) { 2899 synchronize_net(); 2900 qdisc_reset_all_tx_gt(dev, txq); 2901 #ifdef CONFIG_XPS 2902 netif_reset_xps_queues_gt(dev, txq); 2903 #endif 2904 } 2905 } else { 2906 dev->real_num_tx_queues = txq; 2907 } 2908 2909 return 0; 2910 } 2911 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2912 2913 #ifdef CONFIG_SYSFS 2914 /** 2915 * netif_set_real_num_rx_queues - set actual number of RX queues used 2916 * @dev: Network device 2917 * @rxq: Actual number of RX queues 2918 * 2919 * This must be called either with the rtnl_lock held or before 2920 * registration of the net device. Returns 0 on success, or a 2921 * negative error code. If called before registration, it always 2922 * succeeds. 2923 */ 2924 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2925 { 2926 int rc; 2927 2928 if (rxq < 1 || rxq > dev->num_rx_queues) 2929 return -EINVAL; 2930 2931 if (dev->reg_state == NETREG_REGISTERED) { 2932 ASSERT_RTNL(); 2933 2934 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2935 rxq); 2936 if (rc) 2937 return rc; 2938 } 2939 2940 dev->real_num_rx_queues = rxq; 2941 return 0; 2942 } 2943 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2944 #endif 2945 2946 /** 2947 * netif_set_real_num_queues - set actual number of RX and TX queues used 2948 * @dev: Network device 2949 * @txq: Actual number of TX queues 2950 * @rxq: Actual number of RX queues 2951 * 2952 * Set the real number of both TX and RX queues. 2953 * Does nothing if the number of queues is already correct. 2954 */ 2955 int netif_set_real_num_queues(struct net_device *dev, 2956 unsigned int txq, unsigned int rxq) 2957 { 2958 unsigned int old_rxq = dev->real_num_rx_queues; 2959 int err; 2960 2961 if (txq < 1 || txq > dev->num_tx_queues || 2962 rxq < 1 || rxq > dev->num_rx_queues) 2963 return -EINVAL; 2964 2965 /* Start from increases, so the error path only does decreases - 2966 * decreases can't fail. 2967 */ 2968 if (rxq > dev->real_num_rx_queues) { 2969 err = netif_set_real_num_rx_queues(dev, rxq); 2970 if (err) 2971 return err; 2972 } 2973 if (txq > dev->real_num_tx_queues) { 2974 err = netif_set_real_num_tx_queues(dev, txq); 2975 if (err) 2976 goto undo_rx; 2977 } 2978 if (rxq < dev->real_num_rx_queues) 2979 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2980 if (txq < dev->real_num_tx_queues) 2981 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2982 2983 return 0; 2984 undo_rx: 2985 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2986 return err; 2987 } 2988 EXPORT_SYMBOL(netif_set_real_num_queues); 2989 2990 /** 2991 * netif_set_tso_max_size() - set the max size of TSO frames supported 2992 * @dev: netdev to update 2993 * @size: max skb->len of a TSO frame 2994 * 2995 * Set the limit on the size of TSO super-frames the device can handle. 2996 * Unless explicitly set the stack will assume the value of 2997 * %GSO_LEGACY_MAX_SIZE. 2998 */ 2999 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3000 { 3001 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3002 if (size < READ_ONCE(dev->gso_max_size)) 3003 netif_set_gso_max_size(dev, size); 3004 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3005 netif_set_gso_ipv4_max_size(dev, size); 3006 } 3007 EXPORT_SYMBOL(netif_set_tso_max_size); 3008 3009 /** 3010 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3011 * @dev: netdev to update 3012 * @segs: max number of TCP segments 3013 * 3014 * Set the limit on the number of TCP segments the device can generate from 3015 * a single TSO super-frame. 3016 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3017 */ 3018 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3019 { 3020 dev->tso_max_segs = segs; 3021 if (segs < READ_ONCE(dev->gso_max_segs)) 3022 netif_set_gso_max_segs(dev, segs); 3023 } 3024 EXPORT_SYMBOL(netif_set_tso_max_segs); 3025 3026 /** 3027 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3028 * @to: netdev to update 3029 * @from: netdev from which to copy the limits 3030 */ 3031 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3032 { 3033 netif_set_tso_max_size(to, from->tso_max_size); 3034 netif_set_tso_max_segs(to, from->tso_max_segs); 3035 } 3036 EXPORT_SYMBOL(netif_inherit_tso_max); 3037 3038 /** 3039 * netif_get_num_default_rss_queues - default number of RSS queues 3040 * 3041 * Default value is the number of physical cores if there are only 1 or 2, or 3042 * divided by 2 if there are more. 3043 */ 3044 int netif_get_num_default_rss_queues(void) 3045 { 3046 cpumask_var_t cpus; 3047 int cpu, count = 0; 3048 3049 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3050 return 1; 3051 3052 cpumask_copy(cpus, cpu_online_mask); 3053 for_each_cpu(cpu, cpus) { 3054 ++count; 3055 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3056 } 3057 free_cpumask_var(cpus); 3058 3059 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3060 } 3061 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3062 3063 static void __netif_reschedule(struct Qdisc *q) 3064 { 3065 struct softnet_data *sd; 3066 unsigned long flags; 3067 3068 local_irq_save(flags); 3069 sd = this_cpu_ptr(&softnet_data); 3070 q->next_sched = NULL; 3071 *sd->output_queue_tailp = q; 3072 sd->output_queue_tailp = &q->next_sched; 3073 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3074 local_irq_restore(flags); 3075 } 3076 3077 void __netif_schedule(struct Qdisc *q) 3078 { 3079 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3080 __netif_reschedule(q); 3081 } 3082 EXPORT_SYMBOL(__netif_schedule); 3083 3084 struct dev_kfree_skb_cb { 3085 enum skb_free_reason reason; 3086 }; 3087 3088 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3089 { 3090 return (struct dev_kfree_skb_cb *)skb->cb; 3091 } 3092 3093 void netif_schedule_queue(struct netdev_queue *txq) 3094 { 3095 rcu_read_lock(); 3096 if (!netif_xmit_stopped(txq)) { 3097 struct Qdisc *q = rcu_dereference(txq->qdisc); 3098 3099 __netif_schedule(q); 3100 } 3101 rcu_read_unlock(); 3102 } 3103 EXPORT_SYMBOL(netif_schedule_queue); 3104 3105 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3106 { 3107 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3108 struct Qdisc *q; 3109 3110 rcu_read_lock(); 3111 q = rcu_dereference(dev_queue->qdisc); 3112 __netif_schedule(q); 3113 rcu_read_unlock(); 3114 } 3115 } 3116 EXPORT_SYMBOL(netif_tx_wake_queue); 3117 3118 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3119 { 3120 unsigned long flags; 3121 3122 if (unlikely(!skb)) 3123 return; 3124 3125 if (likely(refcount_read(&skb->users) == 1)) { 3126 smp_rmb(); 3127 refcount_set(&skb->users, 0); 3128 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3129 return; 3130 } 3131 get_kfree_skb_cb(skb)->reason = reason; 3132 local_irq_save(flags); 3133 skb->next = __this_cpu_read(softnet_data.completion_queue); 3134 __this_cpu_write(softnet_data.completion_queue, skb); 3135 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3136 local_irq_restore(flags); 3137 } 3138 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3139 3140 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3141 { 3142 if (in_hardirq() || irqs_disabled()) 3143 __dev_kfree_skb_irq(skb, reason); 3144 else 3145 dev_kfree_skb(skb); 3146 } 3147 EXPORT_SYMBOL(__dev_kfree_skb_any); 3148 3149 3150 /** 3151 * netif_device_detach - mark device as removed 3152 * @dev: network device 3153 * 3154 * Mark device as removed from system and therefore no longer available. 3155 */ 3156 void netif_device_detach(struct net_device *dev) 3157 { 3158 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3159 netif_running(dev)) { 3160 netif_tx_stop_all_queues(dev); 3161 } 3162 } 3163 EXPORT_SYMBOL(netif_device_detach); 3164 3165 /** 3166 * netif_device_attach - mark device as attached 3167 * @dev: network device 3168 * 3169 * Mark device as attached from system and restart if needed. 3170 */ 3171 void netif_device_attach(struct net_device *dev) 3172 { 3173 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3174 netif_running(dev)) { 3175 netif_tx_wake_all_queues(dev); 3176 __netdev_watchdog_up(dev); 3177 } 3178 } 3179 EXPORT_SYMBOL(netif_device_attach); 3180 3181 /* 3182 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3183 * to be used as a distribution range. 3184 */ 3185 static u16 skb_tx_hash(const struct net_device *dev, 3186 const struct net_device *sb_dev, 3187 struct sk_buff *skb) 3188 { 3189 u32 hash; 3190 u16 qoffset = 0; 3191 u16 qcount = dev->real_num_tx_queues; 3192 3193 if (dev->num_tc) { 3194 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3195 3196 qoffset = sb_dev->tc_to_txq[tc].offset; 3197 qcount = sb_dev->tc_to_txq[tc].count; 3198 if (unlikely(!qcount)) { 3199 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3200 sb_dev->name, qoffset, tc); 3201 qoffset = 0; 3202 qcount = dev->real_num_tx_queues; 3203 } 3204 } 3205 3206 if (skb_rx_queue_recorded(skb)) { 3207 hash = skb_get_rx_queue(skb); 3208 if (hash >= qoffset) 3209 hash -= qoffset; 3210 while (unlikely(hash >= qcount)) 3211 hash -= qcount; 3212 return hash + qoffset; 3213 } 3214 3215 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3216 } 3217 3218 static void skb_warn_bad_offload(const struct sk_buff *skb) 3219 { 3220 static const netdev_features_t null_features; 3221 struct net_device *dev = skb->dev; 3222 const char *name = ""; 3223 3224 if (!net_ratelimit()) 3225 return; 3226 3227 if (dev) { 3228 if (dev->dev.parent) 3229 name = dev_driver_string(dev->dev.parent); 3230 else 3231 name = netdev_name(dev); 3232 } 3233 skb_dump(KERN_WARNING, skb, false); 3234 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3235 name, dev ? &dev->features : &null_features, 3236 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3237 } 3238 3239 /* 3240 * Invalidate hardware checksum when packet is to be mangled, and 3241 * complete checksum manually on outgoing path. 3242 */ 3243 int skb_checksum_help(struct sk_buff *skb) 3244 { 3245 __wsum csum; 3246 int ret = 0, offset; 3247 3248 if (skb->ip_summed == CHECKSUM_COMPLETE) 3249 goto out_set_summed; 3250 3251 if (unlikely(skb_is_gso(skb))) { 3252 skb_warn_bad_offload(skb); 3253 return -EINVAL; 3254 } 3255 3256 /* Before computing a checksum, we should make sure no frag could 3257 * be modified by an external entity : checksum could be wrong. 3258 */ 3259 if (skb_has_shared_frag(skb)) { 3260 ret = __skb_linearize(skb); 3261 if (ret) 3262 goto out; 3263 } 3264 3265 offset = skb_checksum_start_offset(skb); 3266 ret = -EINVAL; 3267 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3268 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3269 goto out; 3270 } 3271 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3272 3273 offset += skb->csum_offset; 3274 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3275 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3276 goto out; 3277 } 3278 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3279 if (ret) 3280 goto out; 3281 3282 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3283 out_set_summed: 3284 skb->ip_summed = CHECKSUM_NONE; 3285 out: 3286 return ret; 3287 } 3288 EXPORT_SYMBOL(skb_checksum_help); 3289 3290 int skb_crc32c_csum_help(struct sk_buff *skb) 3291 { 3292 __le32 crc32c_csum; 3293 int ret = 0, offset, start; 3294 3295 if (skb->ip_summed != CHECKSUM_PARTIAL) 3296 goto out; 3297 3298 if (unlikely(skb_is_gso(skb))) 3299 goto out; 3300 3301 /* Before computing a checksum, we should make sure no frag could 3302 * be modified by an external entity : checksum could be wrong. 3303 */ 3304 if (unlikely(skb_has_shared_frag(skb))) { 3305 ret = __skb_linearize(skb); 3306 if (ret) 3307 goto out; 3308 } 3309 start = skb_checksum_start_offset(skb); 3310 offset = start + offsetof(struct sctphdr, checksum); 3311 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3312 ret = -EINVAL; 3313 goto out; 3314 } 3315 3316 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3317 if (ret) 3318 goto out; 3319 3320 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3321 skb->len - start, ~(__u32)0, 3322 crc32c_csum_stub)); 3323 *(__le32 *)(skb->data + offset) = crc32c_csum; 3324 skb->ip_summed = CHECKSUM_NONE; 3325 skb->csum_not_inet = 0; 3326 out: 3327 return ret; 3328 } 3329 3330 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3331 { 3332 __be16 type = skb->protocol; 3333 3334 /* Tunnel gso handlers can set protocol to ethernet. */ 3335 if (type == htons(ETH_P_TEB)) { 3336 struct ethhdr *eth; 3337 3338 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3339 return 0; 3340 3341 eth = (struct ethhdr *)skb->data; 3342 type = eth->h_proto; 3343 } 3344 3345 return __vlan_get_protocol(skb, type, depth); 3346 } 3347 3348 /* openvswitch calls this on rx path, so we need a different check. 3349 */ 3350 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3351 { 3352 if (tx_path) 3353 return skb->ip_summed != CHECKSUM_PARTIAL && 3354 skb->ip_summed != CHECKSUM_UNNECESSARY; 3355 3356 return skb->ip_summed == CHECKSUM_NONE; 3357 } 3358 3359 /** 3360 * __skb_gso_segment - Perform segmentation on skb. 3361 * @skb: buffer to segment 3362 * @features: features for the output path (see dev->features) 3363 * @tx_path: whether it is called in TX path 3364 * 3365 * This function segments the given skb and returns a list of segments. 3366 * 3367 * It may return NULL if the skb requires no segmentation. This is 3368 * only possible when GSO is used for verifying header integrity. 3369 * 3370 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3371 */ 3372 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3373 netdev_features_t features, bool tx_path) 3374 { 3375 struct sk_buff *segs; 3376 3377 if (unlikely(skb_needs_check(skb, tx_path))) { 3378 int err; 3379 3380 /* We're going to init ->check field in TCP or UDP header */ 3381 err = skb_cow_head(skb, 0); 3382 if (err < 0) 3383 return ERR_PTR(err); 3384 } 3385 3386 /* Only report GSO partial support if it will enable us to 3387 * support segmentation on this frame without needing additional 3388 * work. 3389 */ 3390 if (features & NETIF_F_GSO_PARTIAL) { 3391 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3392 struct net_device *dev = skb->dev; 3393 3394 partial_features |= dev->features & dev->gso_partial_features; 3395 if (!skb_gso_ok(skb, features | partial_features)) 3396 features &= ~NETIF_F_GSO_PARTIAL; 3397 } 3398 3399 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3400 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3401 3402 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3403 SKB_GSO_CB(skb)->encap_level = 0; 3404 3405 skb_reset_mac_header(skb); 3406 skb_reset_mac_len(skb); 3407 3408 segs = skb_mac_gso_segment(skb, features); 3409 3410 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3411 skb_warn_bad_offload(skb); 3412 3413 return segs; 3414 } 3415 EXPORT_SYMBOL(__skb_gso_segment); 3416 3417 /* Take action when hardware reception checksum errors are detected. */ 3418 #ifdef CONFIG_BUG 3419 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3420 { 3421 netdev_err(dev, "hw csum failure\n"); 3422 skb_dump(KERN_ERR, skb, true); 3423 dump_stack(); 3424 } 3425 3426 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3427 { 3428 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3429 } 3430 EXPORT_SYMBOL(netdev_rx_csum_fault); 3431 #endif 3432 3433 /* XXX: check that highmem exists at all on the given machine. */ 3434 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3435 { 3436 #ifdef CONFIG_HIGHMEM 3437 int i; 3438 3439 if (!(dev->features & NETIF_F_HIGHDMA)) { 3440 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3441 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3442 3443 if (PageHighMem(skb_frag_page(frag))) 3444 return 1; 3445 } 3446 } 3447 #endif 3448 return 0; 3449 } 3450 3451 /* If MPLS offload request, verify we are testing hardware MPLS features 3452 * instead of standard features for the netdev. 3453 */ 3454 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3455 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3456 netdev_features_t features, 3457 __be16 type) 3458 { 3459 if (eth_p_mpls(type)) 3460 features &= skb->dev->mpls_features; 3461 3462 return features; 3463 } 3464 #else 3465 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3466 netdev_features_t features, 3467 __be16 type) 3468 { 3469 return features; 3470 } 3471 #endif 3472 3473 static netdev_features_t harmonize_features(struct sk_buff *skb, 3474 netdev_features_t features) 3475 { 3476 __be16 type; 3477 3478 type = skb_network_protocol(skb, NULL); 3479 features = net_mpls_features(skb, features, type); 3480 3481 if (skb->ip_summed != CHECKSUM_NONE && 3482 !can_checksum_protocol(features, type)) { 3483 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3484 } 3485 if (illegal_highdma(skb->dev, skb)) 3486 features &= ~NETIF_F_SG; 3487 3488 return features; 3489 } 3490 3491 netdev_features_t passthru_features_check(struct sk_buff *skb, 3492 struct net_device *dev, 3493 netdev_features_t features) 3494 { 3495 return features; 3496 } 3497 EXPORT_SYMBOL(passthru_features_check); 3498 3499 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3500 struct net_device *dev, 3501 netdev_features_t features) 3502 { 3503 return vlan_features_check(skb, features); 3504 } 3505 3506 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3507 struct net_device *dev, 3508 netdev_features_t features) 3509 { 3510 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3511 3512 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3513 return features & ~NETIF_F_GSO_MASK; 3514 3515 if (!skb_shinfo(skb)->gso_type) { 3516 skb_warn_bad_offload(skb); 3517 return features & ~NETIF_F_GSO_MASK; 3518 } 3519 3520 /* Support for GSO partial features requires software 3521 * intervention before we can actually process the packets 3522 * so we need to strip support for any partial features now 3523 * and we can pull them back in after we have partially 3524 * segmented the frame. 3525 */ 3526 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3527 features &= ~dev->gso_partial_features; 3528 3529 /* Make sure to clear the IPv4 ID mangling feature if the 3530 * IPv4 header has the potential to be fragmented. 3531 */ 3532 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3533 struct iphdr *iph = skb->encapsulation ? 3534 inner_ip_hdr(skb) : ip_hdr(skb); 3535 3536 if (!(iph->frag_off & htons(IP_DF))) 3537 features &= ~NETIF_F_TSO_MANGLEID; 3538 } 3539 3540 return features; 3541 } 3542 3543 netdev_features_t netif_skb_features(struct sk_buff *skb) 3544 { 3545 struct net_device *dev = skb->dev; 3546 netdev_features_t features = dev->features; 3547 3548 if (skb_is_gso(skb)) 3549 features = gso_features_check(skb, dev, features); 3550 3551 /* If encapsulation offload request, verify we are testing 3552 * hardware encapsulation features instead of standard 3553 * features for the netdev 3554 */ 3555 if (skb->encapsulation) 3556 features &= dev->hw_enc_features; 3557 3558 if (skb_vlan_tagged(skb)) 3559 features = netdev_intersect_features(features, 3560 dev->vlan_features | 3561 NETIF_F_HW_VLAN_CTAG_TX | 3562 NETIF_F_HW_VLAN_STAG_TX); 3563 3564 if (dev->netdev_ops->ndo_features_check) 3565 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3566 features); 3567 else 3568 features &= dflt_features_check(skb, dev, features); 3569 3570 return harmonize_features(skb, features); 3571 } 3572 EXPORT_SYMBOL(netif_skb_features); 3573 3574 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3575 struct netdev_queue *txq, bool more) 3576 { 3577 unsigned int len; 3578 int rc; 3579 3580 if (dev_nit_active(dev)) 3581 dev_queue_xmit_nit(skb, dev); 3582 3583 len = skb->len; 3584 trace_net_dev_start_xmit(skb, dev); 3585 rc = netdev_start_xmit(skb, dev, txq, more); 3586 trace_net_dev_xmit(skb, rc, dev, len); 3587 3588 return rc; 3589 } 3590 3591 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3592 struct netdev_queue *txq, int *ret) 3593 { 3594 struct sk_buff *skb = first; 3595 int rc = NETDEV_TX_OK; 3596 3597 while (skb) { 3598 struct sk_buff *next = skb->next; 3599 3600 skb_mark_not_on_list(skb); 3601 rc = xmit_one(skb, dev, txq, next != NULL); 3602 if (unlikely(!dev_xmit_complete(rc))) { 3603 skb->next = next; 3604 goto out; 3605 } 3606 3607 skb = next; 3608 if (netif_tx_queue_stopped(txq) && skb) { 3609 rc = NETDEV_TX_BUSY; 3610 break; 3611 } 3612 } 3613 3614 out: 3615 *ret = rc; 3616 return skb; 3617 } 3618 3619 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3620 netdev_features_t features) 3621 { 3622 if (skb_vlan_tag_present(skb) && 3623 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3624 skb = __vlan_hwaccel_push_inside(skb); 3625 return skb; 3626 } 3627 3628 int skb_csum_hwoffload_help(struct sk_buff *skb, 3629 const netdev_features_t features) 3630 { 3631 if (unlikely(skb_csum_is_sctp(skb))) 3632 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3633 skb_crc32c_csum_help(skb); 3634 3635 if (features & NETIF_F_HW_CSUM) 3636 return 0; 3637 3638 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3639 switch (skb->csum_offset) { 3640 case offsetof(struct tcphdr, check): 3641 case offsetof(struct udphdr, check): 3642 return 0; 3643 } 3644 } 3645 3646 return skb_checksum_help(skb); 3647 } 3648 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3649 3650 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3651 { 3652 netdev_features_t features; 3653 3654 features = netif_skb_features(skb); 3655 skb = validate_xmit_vlan(skb, features); 3656 if (unlikely(!skb)) 3657 goto out_null; 3658 3659 skb = sk_validate_xmit_skb(skb, dev); 3660 if (unlikely(!skb)) 3661 goto out_null; 3662 3663 if (netif_needs_gso(skb, features)) { 3664 struct sk_buff *segs; 3665 3666 segs = skb_gso_segment(skb, features); 3667 if (IS_ERR(segs)) { 3668 goto out_kfree_skb; 3669 } else if (segs) { 3670 consume_skb(skb); 3671 skb = segs; 3672 } 3673 } else { 3674 if (skb_needs_linearize(skb, features) && 3675 __skb_linearize(skb)) 3676 goto out_kfree_skb; 3677 3678 /* If packet is not checksummed and device does not 3679 * support checksumming for this protocol, complete 3680 * checksumming here. 3681 */ 3682 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3683 if (skb->encapsulation) 3684 skb_set_inner_transport_header(skb, 3685 skb_checksum_start_offset(skb)); 3686 else 3687 skb_set_transport_header(skb, 3688 skb_checksum_start_offset(skb)); 3689 if (skb_csum_hwoffload_help(skb, features)) 3690 goto out_kfree_skb; 3691 } 3692 } 3693 3694 skb = validate_xmit_xfrm(skb, features, again); 3695 3696 return skb; 3697 3698 out_kfree_skb: 3699 kfree_skb(skb); 3700 out_null: 3701 dev_core_stats_tx_dropped_inc(dev); 3702 return NULL; 3703 } 3704 3705 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3706 { 3707 struct sk_buff *next, *head = NULL, *tail; 3708 3709 for (; skb != NULL; skb = next) { 3710 next = skb->next; 3711 skb_mark_not_on_list(skb); 3712 3713 /* in case skb wont be segmented, point to itself */ 3714 skb->prev = skb; 3715 3716 skb = validate_xmit_skb(skb, dev, again); 3717 if (!skb) 3718 continue; 3719 3720 if (!head) 3721 head = skb; 3722 else 3723 tail->next = skb; 3724 /* If skb was segmented, skb->prev points to 3725 * the last segment. If not, it still contains skb. 3726 */ 3727 tail = skb->prev; 3728 } 3729 return head; 3730 } 3731 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3732 3733 static void qdisc_pkt_len_init(struct sk_buff *skb) 3734 { 3735 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3736 3737 qdisc_skb_cb(skb)->pkt_len = skb->len; 3738 3739 /* To get more precise estimation of bytes sent on wire, 3740 * we add to pkt_len the headers size of all segments 3741 */ 3742 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3743 unsigned int hdr_len; 3744 u16 gso_segs = shinfo->gso_segs; 3745 3746 /* mac layer + network layer */ 3747 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3748 3749 /* + transport layer */ 3750 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3751 const struct tcphdr *th; 3752 struct tcphdr _tcphdr; 3753 3754 th = skb_header_pointer(skb, skb_transport_offset(skb), 3755 sizeof(_tcphdr), &_tcphdr); 3756 if (likely(th)) 3757 hdr_len += __tcp_hdrlen(th); 3758 } else { 3759 struct udphdr _udphdr; 3760 3761 if (skb_header_pointer(skb, skb_transport_offset(skb), 3762 sizeof(_udphdr), &_udphdr)) 3763 hdr_len += sizeof(struct udphdr); 3764 } 3765 3766 if (shinfo->gso_type & SKB_GSO_DODGY) 3767 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3768 shinfo->gso_size); 3769 3770 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3771 } 3772 } 3773 3774 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3775 struct sk_buff **to_free, 3776 struct netdev_queue *txq) 3777 { 3778 int rc; 3779 3780 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3781 if (rc == NET_XMIT_SUCCESS) 3782 trace_qdisc_enqueue(q, txq, skb); 3783 return rc; 3784 } 3785 3786 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3787 struct net_device *dev, 3788 struct netdev_queue *txq) 3789 { 3790 spinlock_t *root_lock = qdisc_lock(q); 3791 struct sk_buff *to_free = NULL; 3792 bool contended; 3793 int rc; 3794 3795 qdisc_calculate_pkt_len(skb, q); 3796 3797 if (q->flags & TCQ_F_NOLOCK) { 3798 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3799 qdisc_run_begin(q)) { 3800 /* Retest nolock_qdisc_is_empty() within the protection 3801 * of q->seqlock to protect from racing with requeuing. 3802 */ 3803 if (unlikely(!nolock_qdisc_is_empty(q))) { 3804 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3805 __qdisc_run(q); 3806 qdisc_run_end(q); 3807 3808 goto no_lock_out; 3809 } 3810 3811 qdisc_bstats_cpu_update(q, skb); 3812 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3813 !nolock_qdisc_is_empty(q)) 3814 __qdisc_run(q); 3815 3816 qdisc_run_end(q); 3817 return NET_XMIT_SUCCESS; 3818 } 3819 3820 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3821 qdisc_run(q); 3822 3823 no_lock_out: 3824 if (unlikely(to_free)) 3825 kfree_skb_list_reason(to_free, 3826 SKB_DROP_REASON_QDISC_DROP); 3827 return rc; 3828 } 3829 3830 /* 3831 * Heuristic to force contended enqueues to serialize on a 3832 * separate lock before trying to get qdisc main lock. 3833 * This permits qdisc->running owner to get the lock more 3834 * often and dequeue packets faster. 3835 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3836 * and then other tasks will only enqueue packets. The packets will be 3837 * sent after the qdisc owner is scheduled again. To prevent this 3838 * scenario the task always serialize on the lock. 3839 */ 3840 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3841 if (unlikely(contended)) 3842 spin_lock(&q->busylock); 3843 3844 spin_lock(root_lock); 3845 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3846 __qdisc_drop(skb, &to_free); 3847 rc = NET_XMIT_DROP; 3848 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3849 qdisc_run_begin(q)) { 3850 /* 3851 * This is a work-conserving queue; there are no old skbs 3852 * waiting to be sent out; and the qdisc is not running - 3853 * xmit the skb directly. 3854 */ 3855 3856 qdisc_bstats_update(q, skb); 3857 3858 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3859 if (unlikely(contended)) { 3860 spin_unlock(&q->busylock); 3861 contended = false; 3862 } 3863 __qdisc_run(q); 3864 } 3865 3866 qdisc_run_end(q); 3867 rc = NET_XMIT_SUCCESS; 3868 } else { 3869 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3870 if (qdisc_run_begin(q)) { 3871 if (unlikely(contended)) { 3872 spin_unlock(&q->busylock); 3873 contended = false; 3874 } 3875 __qdisc_run(q); 3876 qdisc_run_end(q); 3877 } 3878 } 3879 spin_unlock(root_lock); 3880 if (unlikely(to_free)) 3881 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3882 if (unlikely(contended)) 3883 spin_unlock(&q->busylock); 3884 return rc; 3885 } 3886 3887 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3888 static void skb_update_prio(struct sk_buff *skb) 3889 { 3890 const struct netprio_map *map; 3891 const struct sock *sk; 3892 unsigned int prioidx; 3893 3894 if (skb->priority) 3895 return; 3896 map = rcu_dereference_bh(skb->dev->priomap); 3897 if (!map) 3898 return; 3899 sk = skb_to_full_sk(skb); 3900 if (!sk) 3901 return; 3902 3903 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3904 3905 if (prioidx < map->priomap_len) 3906 skb->priority = map->priomap[prioidx]; 3907 } 3908 #else 3909 #define skb_update_prio(skb) 3910 #endif 3911 3912 /** 3913 * dev_loopback_xmit - loop back @skb 3914 * @net: network namespace this loopback is happening in 3915 * @sk: sk needed to be a netfilter okfn 3916 * @skb: buffer to transmit 3917 */ 3918 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3919 { 3920 skb_reset_mac_header(skb); 3921 __skb_pull(skb, skb_network_offset(skb)); 3922 skb->pkt_type = PACKET_LOOPBACK; 3923 if (skb->ip_summed == CHECKSUM_NONE) 3924 skb->ip_summed = CHECKSUM_UNNECESSARY; 3925 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3926 skb_dst_force(skb); 3927 netif_rx(skb); 3928 return 0; 3929 } 3930 EXPORT_SYMBOL(dev_loopback_xmit); 3931 3932 #ifdef CONFIG_NET_EGRESS 3933 static struct sk_buff * 3934 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3935 { 3936 #ifdef CONFIG_NET_CLS_ACT 3937 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3938 struct tcf_result cl_res; 3939 3940 if (!miniq) 3941 return skb; 3942 3943 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3944 tc_skb_cb(skb)->mru = 0; 3945 tc_skb_cb(skb)->post_ct = false; 3946 mini_qdisc_bstats_cpu_update(miniq, skb); 3947 3948 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3949 case TC_ACT_OK: 3950 case TC_ACT_RECLASSIFY: 3951 skb->tc_index = TC_H_MIN(cl_res.classid); 3952 break; 3953 case TC_ACT_SHOT: 3954 mini_qdisc_qstats_cpu_drop(miniq); 3955 *ret = NET_XMIT_DROP; 3956 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3957 return NULL; 3958 case TC_ACT_STOLEN: 3959 case TC_ACT_QUEUED: 3960 case TC_ACT_TRAP: 3961 *ret = NET_XMIT_SUCCESS; 3962 consume_skb(skb); 3963 return NULL; 3964 case TC_ACT_REDIRECT: 3965 /* No need to push/pop skb's mac_header here on egress! */ 3966 skb_do_redirect(skb); 3967 *ret = NET_XMIT_SUCCESS; 3968 return NULL; 3969 default: 3970 break; 3971 } 3972 #endif /* CONFIG_NET_CLS_ACT */ 3973 3974 return skb; 3975 } 3976 3977 static struct netdev_queue * 3978 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3979 { 3980 int qm = skb_get_queue_mapping(skb); 3981 3982 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3983 } 3984 3985 static bool netdev_xmit_txqueue_skipped(void) 3986 { 3987 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3988 } 3989 3990 void netdev_xmit_skip_txqueue(bool skip) 3991 { 3992 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3993 } 3994 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3995 #endif /* CONFIG_NET_EGRESS */ 3996 3997 #ifdef CONFIG_XPS 3998 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3999 struct xps_dev_maps *dev_maps, unsigned int tci) 4000 { 4001 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4002 struct xps_map *map; 4003 int queue_index = -1; 4004 4005 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4006 return queue_index; 4007 4008 tci *= dev_maps->num_tc; 4009 tci += tc; 4010 4011 map = rcu_dereference(dev_maps->attr_map[tci]); 4012 if (map) { 4013 if (map->len == 1) 4014 queue_index = map->queues[0]; 4015 else 4016 queue_index = map->queues[reciprocal_scale( 4017 skb_get_hash(skb), map->len)]; 4018 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4019 queue_index = -1; 4020 } 4021 return queue_index; 4022 } 4023 #endif 4024 4025 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4026 struct sk_buff *skb) 4027 { 4028 #ifdef CONFIG_XPS 4029 struct xps_dev_maps *dev_maps; 4030 struct sock *sk = skb->sk; 4031 int queue_index = -1; 4032 4033 if (!static_key_false(&xps_needed)) 4034 return -1; 4035 4036 rcu_read_lock(); 4037 if (!static_key_false(&xps_rxqs_needed)) 4038 goto get_cpus_map; 4039 4040 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4041 if (dev_maps) { 4042 int tci = sk_rx_queue_get(sk); 4043 4044 if (tci >= 0) 4045 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4046 tci); 4047 } 4048 4049 get_cpus_map: 4050 if (queue_index < 0) { 4051 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4052 if (dev_maps) { 4053 unsigned int tci = skb->sender_cpu - 1; 4054 4055 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4056 tci); 4057 } 4058 } 4059 rcu_read_unlock(); 4060 4061 return queue_index; 4062 #else 4063 return -1; 4064 #endif 4065 } 4066 4067 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4068 struct net_device *sb_dev) 4069 { 4070 return 0; 4071 } 4072 EXPORT_SYMBOL(dev_pick_tx_zero); 4073 4074 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4075 struct net_device *sb_dev) 4076 { 4077 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4078 } 4079 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4080 4081 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4082 struct net_device *sb_dev) 4083 { 4084 struct sock *sk = skb->sk; 4085 int queue_index = sk_tx_queue_get(sk); 4086 4087 sb_dev = sb_dev ? : dev; 4088 4089 if (queue_index < 0 || skb->ooo_okay || 4090 queue_index >= dev->real_num_tx_queues) { 4091 int new_index = get_xps_queue(dev, sb_dev, skb); 4092 4093 if (new_index < 0) 4094 new_index = skb_tx_hash(dev, sb_dev, skb); 4095 4096 if (queue_index != new_index && sk && 4097 sk_fullsock(sk) && 4098 rcu_access_pointer(sk->sk_dst_cache)) 4099 sk_tx_queue_set(sk, new_index); 4100 4101 queue_index = new_index; 4102 } 4103 4104 return queue_index; 4105 } 4106 EXPORT_SYMBOL(netdev_pick_tx); 4107 4108 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4109 struct sk_buff *skb, 4110 struct net_device *sb_dev) 4111 { 4112 int queue_index = 0; 4113 4114 #ifdef CONFIG_XPS 4115 u32 sender_cpu = skb->sender_cpu - 1; 4116 4117 if (sender_cpu >= (u32)NR_CPUS) 4118 skb->sender_cpu = raw_smp_processor_id() + 1; 4119 #endif 4120 4121 if (dev->real_num_tx_queues != 1) { 4122 const struct net_device_ops *ops = dev->netdev_ops; 4123 4124 if (ops->ndo_select_queue) 4125 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4126 else 4127 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4128 4129 queue_index = netdev_cap_txqueue(dev, queue_index); 4130 } 4131 4132 skb_set_queue_mapping(skb, queue_index); 4133 return netdev_get_tx_queue(dev, queue_index); 4134 } 4135 4136 /** 4137 * __dev_queue_xmit() - transmit a buffer 4138 * @skb: buffer to transmit 4139 * @sb_dev: suboordinate device used for L2 forwarding offload 4140 * 4141 * Queue a buffer for transmission to a network device. The caller must 4142 * have set the device and priority and built the buffer before calling 4143 * this function. The function can be called from an interrupt. 4144 * 4145 * When calling this method, interrupts MUST be enabled. This is because 4146 * the BH enable code must have IRQs enabled so that it will not deadlock. 4147 * 4148 * Regardless of the return value, the skb is consumed, so it is currently 4149 * difficult to retry a send to this method. (You can bump the ref count 4150 * before sending to hold a reference for retry if you are careful.) 4151 * 4152 * Return: 4153 * * 0 - buffer successfully transmitted 4154 * * positive qdisc return code - NET_XMIT_DROP etc. 4155 * * negative errno - other errors 4156 */ 4157 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4158 { 4159 struct net_device *dev = skb->dev; 4160 struct netdev_queue *txq = NULL; 4161 struct Qdisc *q; 4162 int rc = -ENOMEM; 4163 bool again = false; 4164 4165 skb_reset_mac_header(skb); 4166 skb_assert_len(skb); 4167 4168 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4169 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4170 4171 /* Disable soft irqs for various locks below. Also 4172 * stops preemption for RCU. 4173 */ 4174 rcu_read_lock_bh(); 4175 4176 skb_update_prio(skb); 4177 4178 qdisc_pkt_len_init(skb); 4179 #ifdef CONFIG_NET_CLS_ACT 4180 skb->tc_at_ingress = 0; 4181 #endif 4182 #ifdef CONFIG_NET_EGRESS 4183 if (static_branch_unlikely(&egress_needed_key)) { 4184 if (nf_hook_egress_active()) { 4185 skb = nf_hook_egress(skb, &rc, dev); 4186 if (!skb) 4187 goto out; 4188 } 4189 4190 netdev_xmit_skip_txqueue(false); 4191 4192 nf_skip_egress(skb, true); 4193 skb = sch_handle_egress(skb, &rc, dev); 4194 if (!skb) 4195 goto out; 4196 nf_skip_egress(skb, false); 4197 4198 if (netdev_xmit_txqueue_skipped()) 4199 txq = netdev_tx_queue_mapping(dev, skb); 4200 } 4201 #endif 4202 /* If device/qdisc don't need skb->dst, release it right now while 4203 * its hot in this cpu cache. 4204 */ 4205 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4206 skb_dst_drop(skb); 4207 else 4208 skb_dst_force(skb); 4209 4210 if (!txq) 4211 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4212 4213 q = rcu_dereference_bh(txq->qdisc); 4214 4215 trace_net_dev_queue(skb); 4216 if (q->enqueue) { 4217 rc = __dev_xmit_skb(skb, q, dev, txq); 4218 goto out; 4219 } 4220 4221 /* The device has no queue. Common case for software devices: 4222 * loopback, all the sorts of tunnels... 4223 4224 * Really, it is unlikely that netif_tx_lock protection is necessary 4225 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4226 * counters.) 4227 * However, it is possible, that they rely on protection 4228 * made by us here. 4229 4230 * Check this and shot the lock. It is not prone from deadlocks. 4231 *Either shot noqueue qdisc, it is even simpler 8) 4232 */ 4233 if (dev->flags & IFF_UP) { 4234 int cpu = smp_processor_id(); /* ok because BHs are off */ 4235 4236 /* Other cpus might concurrently change txq->xmit_lock_owner 4237 * to -1 or to their cpu id, but not to our id. 4238 */ 4239 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4240 if (dev_xmit_recursion()) 4241 goto recursion_alert; 4242 4243 skb = validate_xmit_skb(skb, dev, &again); 4244 if (!skb) 4245 goto out; 4246 4247 HARD_TX_LOCK(dev, txq, cpu); 4248 4249 if (!netif_xmit_stopped(txq)) { 4250 dev_xmit_recursion_inc(); 4251 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4252 dev_xmit_recursion_dec(); 4253 if (dev_xmit_complete(rc)) { 4254 HARD_TX_UNLOCK(dev, txq); 4255 goto out; 4256 } 4257 } 4258 HARD_TX_UNLOCK(dev, txq); 4259 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4260 dev->name); 4261 } else { 4262 /* Recursion is detected! It is possible, 4263 * unfortunately 4264 */ 4265 recursion_alert: 4266 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4267 dev->name); 4268 } 4269 } 4270 4271 rc = -ENETDOWN; 4272 rcu_read_unlock_bh(); 4273 4274 dev_core_stats_tx_dropped_inc(dev); 4275 kfree_skb_list(skb); 4276 return rc; 4277 out: 4278 rcu_read_unlock_bh(); 4279 return rc; 4280 } 4281 EXPORT_SYMBOL(__dev_queue_xmit); 4282 4283 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4284 { 4285 struct net_device *dev = skb->dev; 4286 struct sk_buff *orig_skb = skb; 4287 struct netdev_queue *txq; 4288 int ret = NETDEV_TX_BUSY; 4289 bool again = false; 4290 4291 if (unlikely(!netif_running(dev) || 4292 !netif_carrier_ok(dev))) 4293 goto drop; 4294 4295 skb = validate_xmit_skb_list(skb, dev, &again); 4296 if (skb != orig_skb) 4297 goto drop; 4298 4299 skb_set_queue_mapping(skb, queue_id); 4300 txq = skb_get_tx_queue(dev, skb); 4301 4302 local_bh_disable(); 4303 4304 dev_xmit_recursion_inc(); 4305 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4306 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4307 ret = netdev_start_xmit(skb, dev, txq, false); 4308 HARD_TX_UNLOCK(dev, txq); 4309 dev_xmit_recursion_dec(); 4310 4311 local_bh_enable(); 4312 return ret; 4313 drop: 4314 dev_core_stats_tx_dropped_inc(dev); 4315 kfree_skb_list(skb); 4316 return NET_XMIT_DROP; 4317 } 4318 EXPORT_SYMBOL(__dev_direct_xmit); 4319 4320 /************************************************************************* 4321 * Receiver routines 4322 *************************************************************************/ 4323 4324 int netdev_max_backlog __read_mostly = 1000; 4325 EXPORT_SYMBOL(netdev_max_backlog); 4326 4327 int netdev_tstamp_prequeue __read_mostly = 1; 4328 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4329 int netdev_budget __read_mostly = 300; 4330 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4331 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4332 int weight_p __read_mostly = 64; /* old backlog weight */ 4333 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4334 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4335 int dev_rx_weight __read_mostly = 64; 4336 int dev_tx_weight __read_mostly = 64; 4337 4338 /* Called with irq disabled */ 4339 static inline void ____napi_schedule(struct softnet_data *sd, 4340 struct napi_struct *napi) 4341 { 4342 struct task_struct *thread; 4343 4344 lockdep_assert_irqs_disabled(); 4345 4346 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4347 /* Paired with smp_mb__before_atomic() in 4348 * napi_enable()/dev_set_threaded(). 4349 * Use READ_ONCE() to guarantee a complete 4350 * read on napi->thread. Only call 4351 * wake_up_process() when it's not NULL. 4352 */ 4353 thread = READ_ONCE(napi->thread); 4354 if (thread) { 4355 /* Avoid doing set_bit() if the thread is in 4356 * INTERRUPTIBLE state, cause napi_thread_wait() 4357 * makes sure to proceed with napi polling 4358 * if the thread is explicitly woken from here. 4359 */ 4360 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4361 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4362 wake_up_process(thread); 4363 return; 4364 } 4365 } 4366 4367 list_add_tail(&napi->poll_list, &sd->poll_list); 4368 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4369 } 4370 4371 #ifdef CONFIG_RPS 4372 4373 /* One global table that all flow-based protocols share. */ 4374 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4375 EXPORT_SYMBOL(rps_sock_flow_table); 4376 u32 rps_cpu_mask __read_mostly; 4377 EXPORT_SYMBOL(rps_cpu_mask); 4378 4379 struct static_key_false rps_needed __read_mostly; 4380 EXPORT_SYMBOL(rps_needed); 4381 struct static_key_false rfs_needed __read_mostly; 4382 EXPORT_SYMBOL(rfs_needed); 4383 4384 static struct rps_dev_flow * 4385 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4386 struct rps_dev_flow *rflow, u16 next_cpu) 4387 { 4388 if (next_cpu < nr_cpu_ids) { 4389 #ifdef CONFIG_RFS_ACCEL 4390 struct netdev_rx_queue *rxqueue; 4391 struct rps_dev_flow_table *flow_table; 4392 struct rps_dev_flow *old_rflow; 4393 u32 flow_id; 4394 u16 rxq_index; 4395 int rc; 4396 4397 /* Should we steer this flow to a different hardware queue? */ 4398 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4399 !(dev->features & NETIF_F_NTUPLE)) 4400 goto out; 4401 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4402 if (rxq_index == skb_get_rx_queue(skb)) 4403 goto out; 4404 4405 rxqueue = dev->_rx + rxq_index; 4406 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4407 if (!flow_table) 4408 goto out; 4409 flow_id = skb_get_hash(skb) & flow_table->mask; 4410 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4411 rxq_index, flow_id); 4412 if (rc < 0) 4413 goto out; 4414 old_rflow = rflow; 4415 rflow = &flow_table->flows[flow_id]; 4416 rflow->filter = rc; 4417 if (old_rflow->filter == rflow->filter) 4418 old_rflow->filter = RPS_NO_FILTER; 4419 out: 4420 #endif 4421 rflow->last_qtail = 4422 per_cpu(softnet_data, next_cpu).input_queue_head; 4423 } 4424 4425 rflow->cpu = next_cpu; 4426 return rflow; 4427 } 4428 4429 /* 4430 * get_rps_cpu is called from netif_receive_skb and returns the target 4431 * CPU from the RPS map of the receiving queue for a given skb. 4432 * rcu_read_lock must be held on entry. 4433 */ 4434 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4435 struct rps_dev_flow **rflowp) 4436 { 4437 const struct rps_sock_flow_table *sock_flow_table; 4438 struct netdev_rx_queue *rxqueue = dev->_rx; 4439 struct rps_dev_flow_table *flow_table; 4440 struct rps_map *map; 4441 int cpu = -1; 4442 u32 tcpu; 4443 u32 hash; 4444 4445 if (skb_rx_queue_recorded(skb)) { 4446 u16 index = skb_get_rx_queue(skb); 4447 4448 if (unlikely(index >= dev->real_num_rx_queues)) { 4449 WARN_ONCE(dev->real_num_rx_queues > 1, 4450 "%s received packet on queue %u, but number " 4451 "of RX queues is %u\n", 4452 dev->name, index, dev->real_num_rx_queues); 4453 goto done; 4454 } 4455 rxqueue += index; 4456 } 4457 4458 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4459 4460 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4461 map = rcu_dereference(rxqueue->rps_map); 4462 if (!flow_table && !map) 4463 goto done; 4464 4465 skb_reset_network_header(skb); 4466 hash = skb_get_hash(skb); 4467 if (!hash) 4468 goto done; 4469 4470 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4471 if (flow_table && sock_flow_table) { 4472 struct rps_dev_flow *rflow; 4473 u32 next_cpu; 4474 u32 ident; 4475 4476 /* First check into global flow table if there is a match */ 4477 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4478 if ((ident ^ hash) & ~rps_cpu_mask) 4479 goto try_rps; 4480 4481 next_cpu = ident & rps_cpu_mask; 4482 4483 /* OK, now we know there is a match, 4484 * we can look at the local (per receive queue) flow table 4485 */ 4486 rflow = &flow_table->flows[hash & flow_table->mask]; 4487 tcpu = rflow->cpu; 4488 4489 /* 4490 * If the desired CPU (where last recvmsg was done) is 4491 * different from current CPU (one in the rx-queue flow 4492 * table entry), switch if one of the following holds: 4493 * - Current CPU is unset (>= nr_cpu_ids). 4494 * - Current CPU is offline. 4495 * - The current CPU's queue tail has advanced beyond the 4496 * last packet that was enqueued using this table entry. 4497 * This guarantees that all previous packets for the flow 4498 * have been dequeued, thus preserving in order delivery. 4499 */ 4500 if (unlikely(tcpu != next_cpu) && 4501 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4502 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4503 rflow->last_qtail)) >= 0)) { 4504 tcpu = next_cpu; 4505 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4506 } 4507 4508 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4509 *rflowp = rflow; 4510 cpu = tcpu; 4511 goto done; 4512 } 4513 } 4514 4515 try_rps: 4516 4517 if (map) { 4518 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4519 if (cpu_online(tcpu)) { 4520 cpu = tcpu; 4521 goto done; 4522 } 4523 } 4524 4525 done: 4526 return cpu; 4527 } 4528 4529 #ifdef CONFIG_RFS_ACCEL 4530 4531 /** 4532 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4533 * @dev: Device on which the filter was set 4534 * @rxq_index: RX queue index 4535 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4536 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4537 * 4538 * Drivers that implement ndo_rx_flow_steer() should periodically call 4539 * this function for each installed filter and remove the filters for 4540 * which it returns %true. 4541 */ 4542 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4543 u32 flow_id, u16 filter_id) 4544 { 4545 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4546 struct rps_dev_flow_table *flow_table; 4547 struct rps_dev_flow *rflow; 4548 bool expire = true; 4549 unsigned int cpu; 4550 4551 rcu_read_lock(); 4552 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4553 if (flow_table && flow_id <= flow_table->mask) { 4554 rflow = &flow_table->flows[flow_id]; 4555 cpu = READ_ONCE(rflow->cpu); 4556 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4557 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4558 rflow->last_qtail) < 4559 (int)(10 * flow_table->mask))) 4560 expire = false; 4561 } 4562 rcu_read_unlock(); 4563 return expire; 4564 } 4565 EXPORT_SYMBOL(rps_may_expire_flow); 4566 4567 #endif /* CONFIG_RFS_ACCEL */ 4568 4569 /* Called from hardirq (IPI) context */ 4570 static void rps_trigger_softirq(void *data) 4571 { 4572 struct softnet_data *sd = data; 4573 4574 ____napi_schedule(sd, &sd->backlog); 4575 sd->received_rps++; 4576 } 4577 4578 #endif /* CONFIG_RPS */ 4579 4580 /* Called from hardirq (IPI) context */ 4581 static void trigger_rx_softirq(void *data) 4582 { 4583 struct softnet_data *sd = data; 4584 4585 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4586 smp_store_release(&sd->defer_ipi_scheduled, 0); 4587 } 4588 4589 /* 4590 * Check if this softnet_data structure is another cpu one 4591 * If yes, queue it to our IPI list and return 1 4592 * If no, return 0 4593 */ 4594 static int napi_schedule_rps(struct softnet_data *sd) 4595 { 4596 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4597 4598 #ifdef CONFIG_RPS 4599 if (sd != mysd) { 4600 sd->rps_ipi_next = mysd->rps_ipi_list; 4601 mysd->rps_ipi_list = sd; 4602 4603 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4604 return 1; 4605 } 4606 #endif /* CONFIG_RPS */ 4607 __napi_schedule_irqoff(&mysd->backlog); 4608 return 0; 4609 } 4610 4611 #ifdef CONFIG_NET_FLOW_LIMIT 4612 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4613 #endif 4614 4615 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4616 { 4617 #ifdef CONFIG_NET_FLOW_LIMIT 4618 struct sd_flow_limit *fl; 4619 struct softnet_data *sd; 4620 unsigned int old_flow, new_flow; 4621 4622 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4623 return false; 4624 4625 sd = this_cpu_ptr(&softnet_data); 4626 4627 rcu_read_lock(); 4628 fl = rcu_dereference(sd->flow_limit); 4629 if (fl) { 4630 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4631 old_flow = fl->history[fl->history_head]; 4632 fl->history[fl->history_head] = new_flow; 4633 4634 fl->history_head++; 4635 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4636 4637 if (likely(fl->buckets[old_flow])) 4638 fl->buckets[old_flow]--; 4639 4640 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4641 fl->count++; 4642 rcu_read_unlock(); 4643 return true; 4644 } 4645 } 4646 rcu_read_unlock(); 4647 #endif 4648 return false; 4649 } 4650 4651 /* 4652 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4653 * queue (may be a remote CPU queue). 4654 */ 4655 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4656 unsigned int *qtail) 4657 { 4658 enum skb_drop_reason reason; 4659 struct softnet_data *sd; 4660 unsigned long flags; 4661 unsigned int qlen; 4662 4663 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4664 sd = &per_cpu(softnet_data, cpu); 4665 4666 rps_lock_irqsave(sd, &flags); 4667 if (!netif_running(skb->dev)) 4668 goto drop; 4669 qlen = skb_queue_len(&sd->input_pkt_queue); 4670 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4671 if (qlen) { 4672 enqueue: 4673 __skb_queue_tail(&sd->input_pkt_queue, skb); 4674 input_queue_tail_incr_save(sd, qtail); 4675 rps_unlock_irq_restore(sd, &flags); 4676 return NET_RX_SUCCESS; 4677 } 4678 4679 /* Schedule NAPI for backlog device 4680 * We can use non atomic operation since we own the queue lock 4681 */ 4682 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4683 napi_schedule_rps(sd); 4684 goto enqueue; 4685 } 4686 reason = SKB_DROP_REASON_CPU_BACKLOG; 4687 4688 drop: 4689 sd->dropped++; 4690 rps_unlock_irq_restore(sd, &flags); 4691 4692 dev_core_stats_rx_dropped_inc(skb->dev); 4693 kfree_skb_reason(skb, reason); 4694 return NET_RX_DROP; 4695 } 4696 4697 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4698 { 4699 struct net_device *dev = skb->dev; 4700 struct netdev_rx_queue *rxqueue; 4701 4702 rxqueue = dev->_rx; 4703 4704 if (skb_rx_queue_recorded(skb)) { 4705 u16 index = skb_get_rx_queue(skb); 4706 4707 if (unlikely(index >= dev->real_num_rx_queues)) { 4708 WARN_ONCE(dev->real_num_rx_queues > 1, 4709 "%s received packet on queue %u, but number " 4710 "of RX queues is %u\n", 4711 dev->name, index, dev->real_num_rx_queues); 4712 4713 return rxqueue; /* Return first rxqueue */ 4714 } 4715 rxqueue += index; 4716 } 4717 return rxqueue; 4718 } 4719 4720 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4721 struct bpf_prog *xdp_prog) 4722 { 4723 void *orig_data, *orig_data_end, *hard_start; 4724 struct netdev_rx_queue *rxqueue; 4725 bool orig_bcast, orig_host; 4726 u32 mac_len, frame_sz; 4727 __be16 orig_eth_type; 4728 struct ethhdr *eth; 4729 u32 metalen, act; 4730 int off; 4731 4732 /* The XDP program wants to see the packet starting at the MAC 4733 * header. 4734 */ 4735 mac_len = skb->data - skb_mac_header(skb); 4736 hard_start = skb->data - skb_headroom(skb); 4737 4738 /* SKB "head" area always have tailroom for skb_shared_info */ 4739 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4740 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4741 4742 rxqueue = netif_get_rxqueue(skb); 4743 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4744 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4745 skb_headlen(skb) + mac_len, true); 4746 4747 orig_data_end = xdp->data_end; 4748 orig_data = xdp->data; 4749 eth = (struct ethhdr *)xdp->data; 4750 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4751 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4752 orig_eth_type = eth->h_proto; 4753 4754 act = bpf_prog_run_xdp(xdp_prog, xdp); 4755 4756 /* check if bpf_xdp_adjust_head was used */ 4757 off = xdp->data - orig_data; 4758 if (off) { 4759 if (off > 0) 4760 __skb_pull(skb, off); 4761 else if (off < 0) 4762 __skb_push(skb, -off); 4763 4764 skb->mac_header += off; 4765 skb_reset_network_header(skb); 4766 } 4767 4768 /* check if bpf_xdp_adjust_tail was used */ 4769 off = xdp->data_end - orig_data_end; 4770 if (off != 0) { 4771 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4772 skb->len += off; /* positive on grow, negative on shrink */ 4773 } 4774 4775 /* check if XDP changed eth hdr such SKB needs update */ 4776 eth = (struct ethhdr *)xdp->data; 4777 if ((orig_eth_type != eth->h_proto) || 4778 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4779 skb->dev->dev_addr)) || 4780 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4781 __skb_push(skb, ETH_HLEN); 4782 skb->pkt_type = PACKET_HOST; 4783 skb->protocol = eth_type_trans(skb, skb->dev); 4784 } 4785 4786 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4787 * before calling us again on redirect path. We do not call do_redirect 4788 * as we leave that up to the caller. 4789 * 4790 * Caller is responsible for managing lifetime of skb (i.e. calling 4791 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4792 */ 4793 switch (act) { 4794 case XDP_REDIRECT: 4795 case XDP_TX: 4796 __skb_push(skb, mac_len); 4797 break; 4798 case XDP_PASS: 4799 metalen = xdp->data - xdp->data_meta; 4800 if (metalen) 4801 skb_metadata_set(skb, metalen); 4802 break; 4803 } 4804 4805 return act; 4806 } 4807 4808 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4809 struct xdp_buff *xdp, 4810 struct bpf_prog *xdp_prog) 4811 { 4812 u32 act = XDP_DROP; 4813 4814 /* Reinjected packets coming from act_mirred or similar should 4815 * not get XDP generic processing. 4816 */ 4817 if (skb_is_redirected(skb)) 4818 return XDP_PASS; 4819 4820 /* XDP packets must be linear and must have sufficient headroom 4821 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4822 * native XDP provides, thus we need to do it here as well. 4823 */ 4824 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4825 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4826 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4827 int troom = skb->tail + skb->data_len - skb->end; 4828 4829 /* In case we have to go down the path and also linearize, 4830 * then lets do the pskb_expand_head() work just once here. 4831 */ 4832 if (pskb_expand_head(skb, 4833 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4834 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4835 goto do_drop; 4836 if (skb_linearize(skb)) 4837 goto do_drop; 4838 } 4839 4840 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4841 switch (act) { 4842 case XDP_REDIRECT: 4843 case XDP_TX: 4844 case XDP_PASS: 4845 break; 4846 default: 4847 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4848 fallthrough; 4849 case XDP_ABORTED: 4850 trace_xdp_exception(skb->dev, xdp_prog, act); 4851 fallthrough; 4852 case XDP_DROP: 4853 do_drop: 4854 kfree_skb(skb); 4855 break; 4856 } 4857 4858 return act; 4859 } 4860 4861 /* When doing generic XDP we have to bypass the qdisc layer and the 4862 * network taps in order to match in-driver-XDP behavior. This also means 4863 * that XDP packets are able to starve other packets going through a qdisc, 4864 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4865 * queues, so they do not have this starvation issue. 4866 */ 4867 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4868 { 4869 struct net_device *dev = skb->dev; 4870 struct netdev_queue *txq; 4871 bool free_skb = true; 4872 int cpu, rc; 4873 4874 txq = netdev_core_pick_tx(dev, skb, NULL); 4875 cpu = smp_processor_id(); 4876 HARD_TX_LOCK(dev, txq, cpu); 4877 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4878 rc = netdev_start_xmit(skb, dev, txq, 0); 4879 if (dev_xmit_complete(rc)) 4880 free_skb = false; 4881 } 4882 HARD_TX_UNLOCK(dev, txq); 4883 if (free_skb) { 4884 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4885 dev_core_stats_tx_dropped_inc(dev); 4886 kfree_skb(skb); 4887 } 4888 } 4889 4890 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4891 4892 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4893 { 4894 if (xdp_prog) { 4895 struct xdp_buff xdp; 4896 u32 act; 4897 int err; 4898 4899 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4900 if (act != XDP_PASS) { 4901 switch (act) { 4902 case XDP_REDIRECT: 4903 err = xdp_do_generic_redirect(skb->dev, skb, 4904 &xdp, xdp_prog); 4905 if (err) 4906 goto out_redir; 4907 break; 4908 case XDP_TX: 4909 generic_xdp_tx(skb, xdp_prog); 4910 break; 4911 } 4912 return XDP_DROP; 4913 } 4914 } 4915 return XDP_PASS; 4916 out_redir: 4917 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4918 return XDP_DROP; 4919 } 4920 EXPORT_SYMBOL_GPL(do_xdp_generic); 4921 4922 static int netif_rx_internal(struct sk_buff *skb) 4923 { 4924 int ret; 4925 4926 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4927 4928 trace_netif_rx(skb); 4929 4930 #ifdef CONFIG_RPS 4931 if (static_branch_unlikely(&rps_needed)) { 4932 struct rps_dev_flow voidflow, *rflow = &voidflow; 4933 int cpu; 4934 4935 rcu_read_lock(); 4936 4937 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4938 if (cpu < 0) 4939 cpu = smp_processor_id(); 4940 4941 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4942 4943 rcu_read_unlock(); 4944 } else 4945 #endif 4946 { 4947 unsigned int qtail; 4948 4949 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4950 } 4951 return ret; 4952 } 4953 4954 /** 4955 * __netif_rx - Slightly optimized version of netif_rx 4956 * @skb: buffer to post 4957 * 4958 * This behaves as netif_rx except that it does not disable bottom halves. 4959 * As a result this function may only be invoked from the interrupt context 4960 * (either hard or soft interrupt). 4961 */ 4962 int __netif_rx(struct sk_buff *skb) 4963 { 4964 int ret; 4965 4966 lockdep_assert_once(hardirq_count() | softirq_count()); 4967 4968 trace_netif_rx_entry(skb); 4969 ret = netif_rx_internal(skb); 4970 trace_netif_rx_exit(ret); 4971 return ret; 4972 } 4973 EXPORT_SYMBOL(__netif_rx); 4974 4975 /** 4976 * netif_rx - post buffer to the network code 4977 * @skb: buffer to post 4978 * 4979 * This function receives a packet from a device driver and queues it for 4980 * the upper (protocol) levels to process via the backlog NAPI device. It 4981 * always succeeds. The buffer may be dropped during processing for 4982 * congestion control or by the protocol layers. 4983 * The network buffer is passed via the backlog NAPI device. Modern NIC 4984 * driver should use NAPI and GRO. 4985 * This function can used from interrupt and from process context. The 4986 * caller from process context must not disable interrupts before invoking 4987 * this function. 4988 * 4989 * return values: 4990 * NET_RX_SUCCESS (no congestion) 4991 * NET_RX_DROP (packet was dropped) 4992 * 4993 */ 4994 int netif_rx(struct sk_buff *skb) 4995 { 4996 bool need_bh_off = !(hardirq_count() | softirq_count()); 4997 int ret; 4998 4999 if (need_bh_off) 5000 local_bh_disable(); 5001 trace_netif_rx_entry(skb); 5002 ret = netif_rx_internal(skb); 5003 trace_netif_rx_exit(ret); 5004 if (need_bh_off) 5005 local_bh_enable(); 5006 return ret; 5007 } 5008 EXPORT_SYMBOL(netif_rx); 5009 5010 static __latent_entropy void net_tx_action(struct softirq_action *h) 5011 { 5012 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5013 5014 if (sd->completion_queue) { 5015 struct sk_buff *clist; 5016 5017 local_irq_disable(); 5018 clist = sd->completion_queue; 5019 sd->completion_queue = NULL; 5020 local_irq_enable(); 5021 5022 while (clist) { 5023 struct sk_buff *skb = clist; 5024 5025 clist = clist->next; 5026 5027 WARN_ON(refcount_read(&skb->users)); 5028 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5029 trace_consume_skb(skb); 5030 else 5031 trace_kfree_skb(skb, net_tx_action, 5032 SKB_DROP_REASON_NOT_SPECIFIED); 5033 5034 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5035 __kfree_skb(skb); 5036 else 5037 __kfree_skb_defer(skb); 5038 } 5039 } 5040 5041 if (sd->output_queue) { 5042 struct Qdisc *head; 5043 5044 local_irq_disable(); 5045 head = sd->output_queue; 5046 sd->output_queue = NULL; 5047 sd->output_queue_tailp = &sd->output_queue; 5048 local_irq_enable(); 5049 5050 rcu_read_lock(); 5051 5052 while (head) { 5053 struct Qdisc *q = head; 5054 spinlock_t *root_lock = NULL; 5055 5056 head = head->next_sched; 5057 5058 /* We need to make sure head->next_sched is read 5059 * before clearing __QDISC_STATE_SCHED 5060 */ 5061 smp_mb__before_atomic(); 5062 5063 if (!(q->flags & TCQ_F_NOLOCK)) { 5064 root_lock = qdisc_lock(q); 5065 spin_lock(root_lock); 5066 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5067 &q->state))) { 5068 /* There is a synchronize_net() between 5069 * STATE_DEACTIVATED flag being set and 5070 * qdisc_reset()/some_qdisc_is_busy() in 5071 * dev_deactivate(), so we can safely bail out 5072 * early here to avoid data race between 5073 * qdisc_deactivate() and some_qdisc_is_busy() 5074 * for lockless qdisc. 5075 */ 5076 clear_bit(__QDISC_STATE_SCHED, &q->state); 5077 continue; 5078 } 5079 5080 clear_bit(__QDISC_STATE_SCHED, &q->state); 5081 qdisc_run(q); 5082 if (root_lock) 5083 spin_unlock(root_lock); 5084 } 5085 5086 rcu_read_unlock(); 5087 } 5088 5089 xfrm_dev_backlog(sd); 5090 } 5091 5092 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5093 /* This hook is defined here for ATM LANE */ 5094 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5095 unsigned char *addr) __read_mostly; 5096 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5097 #endif 5098 5099 static inline struct sk_buff * 5100 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5101 struct net_device *orig_dev, bool *another) 5102 { 5103 #ifdef CONFIG_NET_CLS_ACT 5104 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5105 struct tcf_result cl_res; 5106 5107 /* If there's at least one ingress present somewhere (so 5108 * we get here via enabled static key), remaining devices 5109 * that are not configured with an ingress qdisc will bail 5110 * out here. 5111 */ 5112 if (!miniq) 5113 return skb; 5114 5115 if (*pt_prev) { 5116 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5117 *pt_prev = NULL; 5118 } 5119 5120 qdisc_skb_cb(skb)->pkt_len = skb->len; 5121 tc_skb_cb(skb)->mru = 0; 5122 tc_skb_cb(skb)->post_ct = false; 5123 skb->tc_at_ingress = 1; 5124 mini_qdisc_bstats_cpu_update(miniq, skb); 5125 5126 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5127 case TC_ACT_OK: 5128 case TC_ACT_RECLASSIFY: 5129 skb->tc_index = TC_H_MIN(cl_res.classid); 5130 break; 5131 case TC_ACT_SHOT: 5132 mini_qdisc_qstats_cpu_drop(miniq); 5133 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5134 *ret = NET_RX_DROP; 5135 return NULL; 5136 case TC_ACT_STOLEN: 5137 case TC_ACT_QUEUED: 5138 case TC_ACT_TRAP: 5139 consume_skb(skb); 5140 *ret = NET_RX_SUCCESS; 5141 return NULL; 5142 case TC_ACT_REDIRECT: 5143 /* skb_mac_header check was done by cls/act_bpf, so 5144 * we can safely push the L2 header back before 5145 * redirecting to another netdev 5146 */ 5147 __skb_push(skb, skb->mac_len); 5148 if (skb_do_redirect(skb) == -EAGAIN) { 5149 __skb_pull(skb, skb->mac_len); 5150 *another = true; 5151 break; 5152 } 5153 *ret = NET_RX_SUCCESS; 5154 return NULL; 5155 case TC_ACT_CONSUMED: 5156 *ret = NET_RX_SUCCESS; 5157 return NULL; 5158 default: 5159 break; 5160 } 5161 #endif /* CONFIG_NET_CLS_ACT */ 5162 return skb; 5163 } 5164 5165 /** 5166 * netdev_is_rx_handler_busy - check if receive handler is registered 5167 * @dev: device to check 5168 * 5169 * Check if a receive handler is already registered for a given device. 5170 * Return true if there one. 5171 * 5172 * The caller must hold the rtnl_mutex. 5173 */ 5174 bool netdev_is_rx_handler_busy(struct net_device *dev) 5175 { 5176 ASSERT_RTNL(); 5177 return dev && rtnl_dereference(dev->rx_handler); 5178 } 5179 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5180 5181 /** 5182 * netdev_rx_handler_register - register receive handler 5183 * @dev: device to register a handler for 5184 * @rx_handler: receive handler to register 5185 * @rx_handler_data: data pointer that is used by rx handler 5186 * 5187 * Register a receive handler for a device. This handler will then be 5188 * called from __netif_receive_skb. A negative errno code is returned 5189 * on a failure. 5190 * 5191 * The caller must hold the rtnl_mutex. 5192 * 5193 * For a general description of rx_handler, see enum rx_handler_result. 5194 */ 5195 int netdev_rx_handler_register(struct net_device *dev, 5196 rx_handler_func_t *rx_handler, 5197 void *rx_handler_data) 5198 { 5199 if (netdev_is_rx_handler_busy(dev)) 5200 return -EBUSY; 5201 5202 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5203 return -EINVAL; 5204 5205 /* Note: rx_handler_data must be set before rx_handler */ 5206 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5207 rcu_assign_pointer(dev->rx_handler, rx_handler); 5208 5209 return 0; 5210 } 5211 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5212 5213 /** 5214 * netdev_rx_handler_unregister - unregister receive handler 5215 * @dev: device to unregister a handler from 5216 * 5217 * Unregister a receive handler from a device. 5218 * 5219 * The caller must hold the rtnl_mutex. 5220 */ 5221 void netdev_rx_handler_unregister(struct net_device *dev) 5222 { 5223 5224 ASSERT_RTNL(); 5225 RCU_INIT_POINTER(dev->rx_handler, NULL); 5226 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5227 * section has a guarantee to see a non NULL rx_handler_data 5228 * as well. 5229 */ 5230 synchronize_net(); 5231 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5232 } 5233 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5234 5235 /* 5236 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5237 * the special handling of PFMEMALLOC skbs. 5238 */ 5239 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5240 { 5241 switch (skb->protocol) { 5242 case htons(ETH_P_ARP): 5243 case htons(ETH_P_IP): 5244 case htons(ETH_P_IPV6): 5245 case htons(ETH_P_8021Q): 5246 case htons(ETH_P_8021AD): 5247 return true; 5248 default: 5249 return false; 5250 } 5251 } 5252 5253 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5254 int *ret, struct net_device *orig_dev) 5255 { 5256 if (nf_hook_ingress_active(skb)) { 5257 int ingress_retval; 5258 5259 if (*pt_prev) { 5260 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5261 *pt_prev = NULL; 5262 } 5263 5264 rcu_read_lock(); 5265 ingress_retval = nf_hook_ingress(skb); 5266 rcu_read_unlock(); 5267 return ingress_retval; 5268 } 5269 return 0; 5270 } 5271 5272 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5273 struct packet_type **ppt_prev) 5274 { 5275 struct packet_type *ptype, *pt_prev; 5276 rx_handler_func_t *rx_handler; 5277 struct sk_buff *skb = *pskb; 5278 struct net_device *orig_dev; 5279 bool deliver_exact = false; 5280 int ret = NET_RX_DROP; 5281 __be16 type; 5282 5283 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5284 5285 trace_netif_receive_skb(skb); 5286 5287 orig_dev = skb->dev; 5288 5289 skb_reset_network_header(skb); 5290 if (!skb_transport_header_was_set(skb)) 5291 skb_reset_transport_header(skb); 5292 skb_reset_mac_len(skb); 5293 5294 pt_prev = NULL; 5295 5296 another_round: 5297 skb->skb_iif = skb->dev->ifindex; 5298 5299 __this_cpu_inc(softnet_data.processed); 5300 5301 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5302 int ret2; 5303 5304 migrate_disable(); 5305 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5306 migrate_enable(); 5307 5308 if (ret2 != XDP_PASS) { 5309 ret = NET_RX_DROP; 5310 goto out; 5311 } 5312 } 5313 5314 if (eth_type_vlan(skb->protocol)) { 5315 skb = skb_vlan_untag(skb); 5316 if (unlikely(!skb)) 5317 goto out; 5318 } 5319 5320 if (skb_skip_tc_classify(skb)) 5321 goto skip_classify; 5322 5323 if (pfmemalloc) 5324 goto skip_taps; 5325 5326 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5327 if (pt_prev) 5328 ret = deliver_skb(skb, pt_prev, orig_dev); 5329 pt_prev = ptype; 5330 } 5331 5332 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5333 if (pt_prev) 5334 ret = deliver_skb(skb, pt_prev, orig_dev); 5335 pt_prev = ptype; 5336 } 5337 5338 skip_taps: 5339 #ifdef CONFIG_NET_INGRESS 5340 if (static_branch_unlikely(&ingress_needed_key)) { 5341 bool another = false; 5342 5343 nf_skip_egress(skb, true); 5344 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5345 &another); 5346 if (another) 5347 goto another_round; 5348 if (!skb) 5349 goto out; 5350 5351 nf_skip_egress(skb, false); 5352 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5353 goto out; 5354 } 5355 #endif 5356 skb_reset_redirect(skb); 5357 skip_classify: 5358 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5359 goto drop; 5360 5361 if (skb_vlan_tag_present(skb)) { 5362 if (pt_prev) { 5363 ret = deliver_skb(skb, pt_prev, orig_dev); 5364 pt_prev = NULL; 5365 } 5366 if (vlan_do_receive(&skb)) 5367 goto another_round; 5368 else if (unlikely(!skb)) 5369 goto out; 5370 } 5371 5372 rx_handler = rcu_dereference(skb->dev->rx_handler); 5373 if (rx_handler) { 5374 if (pt_prev) { 5375 ret = deliver_skb(skb, pt_prev, orig_dev); 5376 pt_prev = NULL; 5377 } 5378 switch (rx_handler(&skb)) { 5379 case RX_HANDLER_CONSUMED: 5380 ret = NET_RX_SUCCESS; 5381 goto out; 5382 case RX_HANDLER_ANOTHER: 5383 goto another_round; 5384 case RX_HANDLER_EXACT: 5385 deliver_exact = true; 5386 break; 5387 case RX_HANDLER_PASS: 5388 break; 5389 default: 5390 BUG(); 5391 } 5392 } 5393 5394 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5395 check_vlan_id: 5396 if (skb_vlan_tag_get_id(skb)) { 5397 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5398 * find vlan device. 5399 */ 5400 skb->pkt_type = PACKET_OTHERHOST; 5401 } else if (eth_type_vlan(skb->protocol)) { 5402 /* Outer header is 802.1P with vlan 0, inner header is 5403 * 802.1Q or 802.1AD and vlan_do_receive() above could 5404 * not find vlan dev for vlan id 0. 5405 */ 5406 __vlan_hwaccel_clear_tag(skb); 5407 skb = skb_vlan_untag(skb); 5408 if (unlikely(!skb)) 5409 goto out; 5410 if (vlan_do_receive(&skb)) 5411 /* After stripping off 802.1P header with vlan 0 5412 * vlan dev is found for inner header. 5413 */ 5414 goto another_round; 5415 else if (unlikely(!skb)) 5416 goto out; 5417 else 5418 /* We have stripped outer 802.1P vlan 0 header. 5419 * But could not find vlan dev. 5420 * check again for vlan id to set OTHERHOST. 5421 */ 5422 goto check_vlan_id; 5423 } 5424 /* Note: we might in the future use prio bits 5425 * and set skb->priority like in vlan_do_receive() 5426 * For the time being, just ignore Priority Code Point 5427 */ 5428 __vlan_hwaccel_clear_tag(skb); 5429 } 5430 5431 type = skb->protocol; 5432 5433 /* deliver only exact match when indicated */ 5434 if (likely(!deliver_exact)) { 5435 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5436 &ptype_base[ntohs(type) & 5437 PTYPE_HASH_MASK]); 5438 } 5439 5440 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5441 &orig_dev->ptype_specific); 5442 5443 if (unlikely(skb->dev != orig_dev)) { 5444 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5445 &skb->dev->ptype_specific); 5446 } 5447 5448 if (pt_prev) { 5449 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5450 goto drop; 5451 *ppt_prev = pt_prev; 5452 } else { 5453 drop: 5454 if (!deliver_exact) 5455 dev_core_stats_rx_dropped_inc(skb->dev); 5456 else 5457 dev_core_stats_rx_nohandler_inc(skb->dev); 5458 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5459 /* Jamal, now you will not able to escape explaining 5460 * me how you were going to use this. :-) 5461 */ 5462 ret = NET_RX_DROP; 5463 } 5464 5465 out: 5466 /* The invariant here is that if *ppt_prev is not NULL 5467 * then skb should also be non-NULL. 5468 * 5469 * Apparently *ppt_prev assignment above holds this invariant due to 5470 * skb dereferencing near it. 5471 */ 5472 *pskb = skb; 5473 return ret; 5474 } 5475 5476 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5477 { 5478 struct net_device *orig_dev = skb->dev; 5479 struct packet_type *pt_prev = NULL; 5480 int ret; 5481 5482 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5483 if (pt_prev) 5484 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5485 skb->dev, pt_prev, orig_dev); 5486 return ret; 5487 } 5488 5489 /** 5490 * netif_receive_skb_core - special purpose version of netif_receive_skb 5491 * @skb: buffer to process 5492 * 5493 * More direct receive version of netif_receive_skb(). It should 5494 * only be used by callers that have a need to skip RPS and Generic XDP. 5495 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5496 * 5497 * This function may only be called from softirq context and interrupts 5498 * should be enabled. 5499 * 5500 * Return values (usually ignored): 5501 * NET_RX_SUCCESS: no congestion 5502 * NET_RX_DROP: packet was dropped 5503 */ 5504 int netif_receive_skb_core(struct sk_buff *skb) 5505 { 5506 int ret; 5507 5508 rcu_read_lock(); 5509 ret = __netif_receive_skb_one_core(skb, false); 5510 rcu_read_unlock(); 5511 5512 return ret; 5513 } 5514 EXPORT_SYMBOL(netif_receive_skb_core); 5515 5516 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5517 struct packet_type *pt_prev, 5518 struct net_device *orig_dev) 5519 { 5520 struct sk_buff *skb, *next; 5521 5522 if (!pt_prev) 5523 return; 5524 if (list_empty(head)) 5525 return; 5526 if (pt_prev->list_func != NULL) 5527 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5528 ip_list_rcv, head, pt_prev, orig_dev); 5529 else 5530 list_for_each_entry_safe(skb, next, head, list) { 5531 skb_list_del_init(skb); 5532 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5533 } 5534 } 5535 5536 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5537 { 5538 /* Fast-path assumptions: 5539 * - There is no RX handler. 5540 * - Only one packet_type matches. 5541 * If either of these fails, we will end up doing some per-packet 5542 * processing in-line, then handling the 'last ptype' for the whole 5543 * sublist. This can't cause out-of-order delivery to any single ptype, 5544 * because the 'last ptype' must be constant across the sublist, and all 5545 * other ptypes are handled per-packet. 5546 */ 5547 /* Current (common) ptype of sublist */ 5548 struct packet_type *pt_curr = NULL; 5549 /* Current (common) orig_dev of sublist */ 5550 struct net_device *od_curr = NULL; 5551 struct list_head sublist; 5552 struct sk_buff *skb, *next; 5553 5554 INIT_LIST_HEAD(&sublist); 5555 list_for_each_entry_safe(skb, next, head, list) { 5556 struct net_device *orig_dev = skb->dev; 5557 struct packet_type *pt_prev = NULL; 5558 5559 skb_list_del_init(skb); 5560 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5561 if (!pt_prev) 5562 continue; 5563 if (pt_curr != pt_prev || od_curr != orig_dev) { 5564 /* dispatch old sublist */ 5565 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5566 /* start new sublist */ 5567 INIT_LIST_HEAD(&sublist); 5568 pt_curr = pt_prev; 5569 od_curr = orig_dev; 5570 } 5571 list_add_tail(&skb->list, &sublist); 5572 } 5573 5574 /* dispatch final sublist */ 5575 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5576 } 5577 5578 static int __netif_receive_skb(struct sk_buff *skb) 5579 { 5580 int ret; 5581 5582 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5583 unsigned int noreclaim_flag; 5584 5585 /* 5586 * PFMEMALLOC skbs are special, they should 5587 * - be delivered to SOCK_MEMALLOC sockets only 5588 * - stay away from userspace 5589 * - have bounded memory usage 5590 * 5591 * Use PF_MEMALLOC as this saves us from propagating the allocation 5592 * context down to all allocation sites. 5593 */ 5594 noreclaim_flag = memalloc_noreclaim_save(); 5595 ret = __netif_receive_skb_one_core(skb, true); 5596 memalloc_noreclaim_restore(noreclaim_flag); 5597 } else 5598 ret = __netif_receive_skb_one_core(skb, false); 5599 5600 return ret; 5601 } 5602 5603 static void __netif_receive_skb_list(struct list_head *head) 5604 { 5605 unsigned long noreclaim_flag = 0; 5606 struct sk_buff *skb, *next; 5607 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5608 5609 list_for_each_entry_safe(skb, next, head, list) { 5610 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5611 struct list_head sublist; 5612 5613 /* Handle the previous sublist */ 5614 list_cut_before(&sublist, head, &skb->list); 5615 if (!list_empty(&sublist)) 5616 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5617 pfmemalloc = !pfmemalloc; 5618 /* See comments in __netif_receive_skb */ 5619 if (pfmemalloc) 5620 noreclaim_flag = memalloc_noreclaim_save(); 5621 else 5622 memalloc_noreclaim_restore(noreclaim_flag); 5623 } 5624 } 5625 /* Handle the remaining sublist */ 5626 if (!list_empty(head)) 5627 __netif_receive_skb_list_core(head, pfmemalloc); 5628 /* Restore pflags */ 5629 if (pfmemalloc) 5630 memalloc_noreclaim_restore(noreclaim_flag); 5631 } 5632 5633 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5634 { 5635 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5636 struct bpf_prog *new = xdp->prog; 5637 int ret = 0; 5638 5639 switch (xdp->command) { 5640 case XDP_SETUP_PROG: 5641 rcu_assign_pointer(dev->xdp_prog, new); 5642 if (old) 5643 bpf_prog_put(old); 5644 5645 if (old && !new) { 5646 static_branch_dec(&generic_xdp_needed_key); 5647 } else if (new && !old) { 5648 static_branch_inc(&generic_xdp_needed_key); 5649 dev_disable_lro(dev); 5650 dev_disable_gro_hw(dev); 5651 } 5652 break; 5653 5654 default: 5655 ret = -EINVAL; 5656 break; 5657 } 5658 5659 return ret; 5660 } 5661 5662 static int netif_receive_skb_internal(struct sk_buff *skb) 5663 { 5664 int ret; 5665 5666 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5667 5668 if (skb_defer_rx_timestamp(skb)) 5669 return NET_RX_SUCCESS; 5670 5671 rcu_read_lock(); 5672 #ifdef CONFIG_RPS 5673 if (static_branch_unlikely(&rps_needed)) { 5674 struct rps_dev_flow voidflow, *rflow = &voidflow; 5675 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5676 5677 if (cpu >= 0) { 5678 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5679 rcu_read_unlock(); 5680 return ret; 5681 } 5682 } 5683 #endif 5684 ret = __netif_receive_skb(skb); 5685 rcu_read_unlock(); 5686 return ret; 5687 } 5688 5689 void netif_receive_skb_list_internal(struct list_head *head) 5690 { 5691 struct sk_buff *skb, *next; 5692 struct list_head sublist; 5693 5694 INIT_LIST_HEAD(&sublist); 5695 list_for_each_entry_safe(skb, next, head, list) { 5696 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5697 skb_list_del_init(skb); 5698 if (!skb_defer_rx_timestamp(skb)) 5699 list_add_tail(&skb->list, &sublist); 5700 } 5701 list_splice_init(&sublist, head); 5702 5703 rcu_read_lock(); 5704 #ifdef CONFIG_RPS 5705 if (static_branch_unlikely(&rps_needed)) { 5706 list_for_each_entry_safe(skb, next, head, list) { 5707 struct rps_dev_flow voidflow, *rflow = &voidflow; 5708 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5709 5710 if (cpu >= 0) { 5711 /* Will be handled, remove from list */ 5712 skb_list_del_init(skb); 5713 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5714 } 5715 } 5716 } 5717 #endif 5718 __netif_receive_skb_list(head); 5719 rcu_read_unlock(); 5720 } 5721 5722 /** 5723 * netif_receive_skb - process receive buffer from network 5724 * @skb: buffer to process 5725 * 5726 * netif_receive_skb() is the main receive data processing function. 5727 * It always succeeds. The buffer may be dropped during processing 5728 * for congestion control or by the protocol layers. 5729 * 5730 * This function may only be called from softirq context and interrupts 5731 * should be enabled. 5732 * 5733 * Return values (usually ignored): 5734 * NET_RX_SUCCESS: no congestion 5735 * NET_RX_DROP: packet was dropped 5736 */ 5737 int netif_receive_skb(struct sk_buff *skb) 5738 { 5739 int ret; 5740 5741 trace_netif_receive_skb_entry(skb); 5742 5743 ret = netif_receive_skb_internal(skb); 5744 trace_netif_receive_skb_exit(ret); 5745 5746 return ret; 5747 } 5748 EXPORT_SYMBOL(netif_receive_skb); 5749 5750 /** 5751 * netif_receive_skb_list - process many receive buffers from network 5752 * @head: list of skbs to process. 5753 * 5754 * Since return value of netif_receive_skb() is normally ignored, and 5755 * wouldn't be meaningful for a list, this function returns void. 5756 * 5757 * This function may only be called from softirq context and interrupts 5758 * should be enabled. 5759 */ 5760 void netif_receive_skb_list(struct list_head *head) 5761 { 5762 struct sk_buff *skb; 5763 5764 if (list_empty(head)) 5765 return; 5766 if (trace_netif_receive_skb_list_entry_enabled()) { 5767 list_for_each_entry(skb, head, list) 5768 trace_netif_receive_skb_list_entry(skb); 5769 } 5770 netif_receive_skb_list_internal(head); 5771 trace_netif_receive_skb_list_exit(0); 5772 } 5773 EXPORT_SYMBOL(netif_receive_skb_list); 5774 5775 static DEFINE_PER_CPU(struct work_struct, flush_works); 5776 5777 /* Network device is going away, flush any packets still pending */ 5778 static void flush_backlog(struct work_struct *work) 5779 { 5780 struct sk_buff *skb, *tmp; 5781 struct softnet_data *sd; 5782 5783 local_bh_disable(); 5784 sd = this_cpu_ptr(&softnet_data); 5785 5786 rps_lock_irq_disable(sd); 5787 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5788 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5789 __skb_unlink(skb, &sd->input_pkt_queue); 5790 dev_kfree_skb_irq(skb); 5791 input_queue_head_incr(sd); 5792 } 5793 } 5794 rps_unlock_irq_enable(sd); 5795 5796 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5797 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5798 __skb_unlink(skb, &sd->process_queue); 5799 kfree_skb(skb); 5800 input_queue_head_incr(sd); 5801 } 5802 } 5803 local_bh_enable(); 5804 } 5805 5806 static bool flush_required(int cpu) 5807 { 5808 #if IS_ENABLED(CONFIG_RPS) 5809 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5810 bool do_flush; 5811 5812 rps_lock_irq_disable(sd); 5813 5814 /* as insertion into process_queue happens with the rps lock held, 5815 * process_queue access may race only with dequeue 5816 */ 5817 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5818 !skb_queue_empty_lockless(&sd->process_queue); 5819 rps_unlock_irq_enable(sd); 5820 5821 return do_flush; 5822 #endif 5823 /* without RPS we can't safely check input_pkt_queue: during a 5824 * concurrent remote skb_queue_splice() we can detect as empty both 5825 * input_pkt_queue and process_queue even if the latter could end-up 5826 * containing a lot of packets. 5827 */ 5828 return true; 5829 } 5830 5831 static void flush_all_backlogs(void) 5832 { 5833 static cpumask_t flush_cpus; 5834 unsigned int cpu; 5835 5836 /* since we are under rtnl lock protection we can use static data 5837 * for the cpumask and avoid allocating on stack the possibly 5838 * large mask 5839 */ 5840 ASSERT_RTNL(); 5841 5842 cpus_read_lock(); 5843 5844 cpumask_clear(&flush_cpus); 5845 for_each_online_cpu(cpu) { 5846 if (flush_required(cpu)) { 5847 queue_work_on(cpu, system_highpri_wq, 5848 per_cpu_ptr(&flush_works, cpu)); 5849 cpumask_set_cpu(cpu, &flush_cpus); 5850 } 5851 } 5852 5853 /* we can have in flight packet[s] on the cpus we are not flushing, 5854 * synchronize_net() in unregister_netdevice_many() will take care of 5855 * them 5856 */ 5857 for_each_cpu(cpu, &flush_cpus) 5858 flush_work(per_cpu_ptr(&flush_works, cpu)); 5859 5860 cpus_read_unlock(); 5861 } 5862 5863 static void net_rps_send_ipi(struct softnet_data *remsd) 5864 { 5865 #ifdef CONFIG_RPS 5866 while (remsd) { 5867 struct softnet_data *next = remsd->rps_ipi_next; 5868 5869 if (cpu_online(remsd->cpu)) 5870 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5871 remsd = next; 5872 } 5873 #endif 5874 } 5875 5876 /* 5877 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5878 * Note: called with local irq disabled, but exits with local irq enabled. 5879 */ 5880 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5881 { 5882 #ifdef CONFIG_RPS 5883 struct softnet_data *remsd = sd->rps_ipi_list; 5884 5885 if (remsd) { 5886 sd->rps_ipi_list = NULL; 5887 5888 local_irq_enable(); 5889 5890 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5891 net_rps_send_ipi(remsd); 5892 } else 5893 #endif 5894 local_irq_enable(); 5895 } 5896 5897 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5898 { 5899 #ifdef CONFIG_RPS 5900 return sd->rps_ipi_list != NULL; 5901 #else 5902 return false; 5903 #endif 5904 } 5905 5906 static int process_backlog(struct napi_struct *napi, int quota) 5907 { 5908 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5909 bool again = true; 5910 int work = 0; 5911 5912 /* Check if we have pending ipi, its better to send them now, 5913 * not waiting net_rx_action() end. 5914 */ 5915 if (sd_has_rps_ipi_waiting(sd)) { 5916 local_irq_disable(); 5917 net_rps_action_and_irq_enable(sd); 5918 } 5919 5920 napi->weight = READ_ONCE(dev_rx_weight); 5921 while (again) { 5922 struct sk_buff *skb; 5923 5924 while ((skb = __skb_dequeue(&sd->process_queue))) { 5925 rcu_read_lock(); 5926 __netif_receive_skb(skb); 5927 rcu_read_unlock(); 5928 input_queue_head_incr(sd); 5929 if (++work >= quota) 5930 return work; 5931 5932 } 5933 5934 rps_lock_irq_disable(sd); 5935 if (skb_queue_empty(&sd->input_pkt_queue)) { 5936 /* 5937 * Inline a custom version of __napi_complete(). 5938 * only current cpu owns and manipulates this napi, 5939 * and NAPI_STATE_SCHED is the only possible flag set 5940 * on backlog. 5941 * We can use a plain write instead of clear_bit(), 5942 * and we dont need an smp_mb() memory barrier. 5943 */ 5944 napi->state = 0; 5945 again = false; 5946 } else { 5947 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5948 &sd->process_queue); 5949 } 5950 rps_unlock_irq_enable(sd); 5951 } 5952 5953 return work; 5954 } 5955 5956 /** 5957 * __napi_schedule - schedule for receive 5958 * @n: entry to schedule 5959 * 5960 * The entry's receive function will be scheduled to run. 5961 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5962 */ 5963 void __napi_schedule(struct napi_struct *n) 5964 { 5965 unsigned long flags; 5966 5967 local_irq_save(flags); 5968 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5969 local_irq_restore(flags); 5970 } 5971 EXPORT_SYMBOL(__napi_schedule); 5972 5973 /** 5974 * napi_schedule_prep - check if napi can be scheduled 5975 * @n: napi context 5976 * 5977 * Test if NAPI routine is already running, and if not mark 5978 * it as running. This is used as a condition variable to 5979 * insure only one NAPI poll instance runs. We also make 5980 * sure there is no pending NAPI disable. 5981 */ 5982 bool napi_schedule_prep(struct napi_struct *n) 5983 { 5984 unsigned long new, val = READ_ONCE(n->state); 5985 5986 do { 5987 if (unlikely(val & NAPIF_STATE_DISABLE)) 5988 return false; 5989 new = val | NAPIF_STATE_SCHED; 5990 5991 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5992 * This was suggested by Alexander Duyck, as compiler 5993 * emits better code than : 5994 * if (val & NAPIF_STATE_SCHED) 5995 * new |= NAPIF_STATE_MISSED; 5996 */ 5997 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5998 NAPIF_STATE_MISSED; 5999 } while (!try_cmpxchg(&n->state, &val, new)); 6000 6001 return !(val & NAPIF_STATE_SCHED); 6002 } 6003 EXPORT_SYMBOL(napi_schedule_prep); 6004 6005 /** 6006 * __napi_schedule_irqoff - schedule for receive 6007 * @n: entry to schedule 6008 * 6009 * Variant of __napi_schedule() assuming hard irqs are masked. 6010 * 6011 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6012 * because the interrupt disabled assumption might not be true 6013 * due to force-threaded interrupts and spinlock substitution. 6014 */ 6015 void __napi_schedule_irqoff(struct napi_struct *n) 6016 { 6017 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6018 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6019 else 6020 __napi_schedule(n); 6021 } 6022 EXPORT_SYMBOL(__napi_schedule_irqoff); 6023 6024 bool napi_complete_done(struct napi_struct *n, int work_done) 6025 { 6026 unsigned long flags, val, new, timeout = 0; 6027 bool ret = true; 6028 6029 /* 6030 * 1) Don't let napi dequeue from the cpu poll list 6031 * just in case its running on a different cpu. 6032 * 2) If we are busy polling, do nothing here, we have 6033 * the guarantee we will be called later. 6034 */ 6035 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6036 NAPIF_STATE_IN_BUSY_POLL))) 6037 return false; 6038 6039 if (work_done) { 6040 if (n->gro_bitmask) 6041 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6042 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6043 } 6044 if (n->defer_hard_irqs_count > 0) { 6045 n->defer_hard_irqs_count--; 6046 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6047 if (timeout) 6048 ret = false; 6049 } 6050 if (n->gro_bitmask) { 6051 /* When the NAPI instance uses a timeout and keeps postponing 6052 * it, we need to bound somehow the time packets are kept in 6053 * the GRO layer 6054 */ 6055 napi_gro_flush(n, !!timeout); 6056 } 6057 6058 gro_normal_list(n); 6059 6060 if (unlikely(!list_empty(&n->poll_list))) { 6061 /* If n->poll_list is not empty, we need to mask irqs */ 6062 local_irq_save(flags); 6063 list_del_init(&n->poll_list); 6064 local_irq_restore(flags); 6065 } 6066 6067 val = READ_ONCE(n->state); 6068 do { 6069 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6070 6071 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6072 NAPIF_STATE_SCHED_THREADED | 6073 NAPIF_STATE_PREFER_BUSY_POLL); 6074 6075 /* If STATE_MISSED was set, leave STATE_SCHED set, 6076 * because we will call napi->poll() one more time. 6077 * This C code was suggested by Alexander Duyck to help gcc. 6078 */ 6079 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6080 NAPIF_STATE_SCHED; 6081 } while (!try_cmpxchg(&n->state, &val, new)); 6082 6083 if (unlikely(val & NAPIF_STATE_MISSED)) { 6084 __napi_schedule(n); 6085 return false; 6086 } 6087 6088 if (timeout) 6089 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6090 HRTIMER_MODE_REL_PINNED); 6091 return ret; 6092 } 6093 EXPORT_SYMBOL(napi_complete_done); 6094 6095 /* must be called under rcu_read_lock(), as we dont take a reference */ 6096 static struct napi_struct *napi_by_id(unsigned int napi_id) 6097 { 6098 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6099 struct napi_struct *napi; 6100 6101 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6102 if (napi->napi_id == napi_id) 6103 return napi; 6104 6105 return NULL; 6106 } 6107 6108 #if defined(CONFIG_NET_RX_BUSY_POLL) 6109 6110 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6111 { 6112 if (!skip_schedule) { 6113 gro_normal_list(napi); 6114 __napi_schedule(napi); 6115 return; 6116 } 6117 6118 if (napi->gro_bitmask) { 6119 /* flush too old packets 6120 * If HZ < 1000, flush all packets. 6121 */ 6122 napi_gro_flush(napi, HZ >= 1000); 6123 } 6124 6125 gro_normal_list(napi); 6126 clear_bit(NAPI_STATE_SCHED, &napi->state); 6127 } 6128 6129 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6130 u16 budget) 6131 { 6132 bool skip_schedule = false; 6133 unsigned long timeout; 6134 int rc; 6135 6136 /* Busy polling means there is a high chance device driver hard irq 6137 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6138 * set in napi_schedule_prep(). 6139 * Since we are about to call napi->poll() once more, we can safely 6140 * clear NAPI_STATE_MISSED. 6141 * 6142 * Note: x86 could use a single "lock and ..." instruction 6143 * to perform these two clear_bit() 6144 */ 6145 clear_bit(NAPI_STATE_MISSED, &napi->state); 6146 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6147 6148 local_bh_disable(); 6149 6150 if (prefer_busy_poll) { 6151 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6152 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6153 if (napi->defer_hard_irqs_count && timeout) { 6154 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6155 skip_schedule = true; 6156 } 6157 } 6158 6159 /* All we really want here is to re-enable device interrupts. 6160 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6161 */ 6162 rc = napi->poll(napi, budget); 6163 /* We can't gro_normal_list() here, because napi->poll() might have 6164 * rearmed the napi (napi_complete_done()) in which case it could 6165 * already be running on another CPU. 6166 */ 6167 trace_napi_poll(napi, rc, budget); 6168 netpoll_poll_unlock(have_poll_lock); 6169 if (rc == budget) 6170 __busy_poll_stop(napi, skip_schedule); 6171 local_bh_enable(); 6172 } 6173 6174 void napi_busy_loop(unsigned int napi_id, 6175 bool (*loop_end)(void *, unsigned long), 6176 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6177 { 6178 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6179 int (*napi_poll)(struct napi_struct *napi, int budget); 6180 void *have_poll_lock = NULL; 6181 struct napi_struct *napi; 6182 6183 restart: 6184 napi_poll = NULL; 6185 6186 rcu_read_lock(); 6187 6188 napi = napi_by_id(napi_id); 6189 if (!napi) 6190 goto out; 6191 6192 preempt_disable(); 6193 for (;;) { 6194 int work = 0; 6195 6196 local_bh_disable(); 6197 if (!napi_poll) { 6198 unsigned long val = READ_ONCE(napi->state); 6199 6200 /* If multiple threads are competing for this napi, 6201 * we avoid dirtying napi->state as much as we can. 6202 */ 6203 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6204 NAPIF_STATE_IN_BUSY_POLL)) { 6205 if (prefer_busy_poll) 6206 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6207 goto count; 6208 } 6209 if (cmpxchg(&napi->state, val, 6210 val | NAPIF_STATE_IN_BUSY_POLL | 6211 NAPIF_STATE_SCHED) != val) { 6212 if (prefer_busy_poll) 6213 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6214 goto count; 6215 } 6216 have_poll_lock = netpoll_poll_lock(napi); 6217 napi_poll = napi->poll; 6218 } 6219 work = napi_poll(napi, budget); 6220 trace_napi_poll(napi, work, budget); 6221 gro_normal_list(napi); 6222 count: 6223 if (work > 0) 6224 __NET_ADD_STATS(dev_net(napi->dev), 6225 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6226 local_bh_enable(); 6227 6228 if (!loop_end || loop_end(loop_end_arg, start_time)) 6229 break; 6230 6231 if (unlikely(need_resched())) { 6232 if (napi_poll) 6233 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6234 preempt_enable(); 6235 rcu_read_unlock(); 6236 cond_resched(); 6237 if (loop_end(loop_end_arg, start_time)) 6238 return; 6239 goto restart; 6240 } 6241 cpu_relax(); 6242 } 6243 if (napi_poll) 6244 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6245 preempt_enable(); 6246 out: 6247 rcu_read_unlock(); 6248 } 6249 EXPORT_SYMBOL(napi_busy_loop); 6250 6251 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6252 6253 static void napi_hash_add(struct napi_struct *napi) 6254 { 6255 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6256 return; 6257 6258 spin_lock(&napi_hash_lock); 6259 6260 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6261 do { 6262 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6263 napi_gen_id = MIN_NAPI_ID; 6264 } while (napi_by_id(napi_gen_id)); 6265 napi->napi_id = napi_gen_id; 6266 6267 hlist_add_head_rcu(&napi->napi_hash_node, 6268 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6269 6270 spin_unlock(&napi_hash_lock); 6271 } 6272 6273 /* Warning : caller is responsible to make sure rcu grace period 6274 * is respected before freeing memory containing @napi 6275 */ 6276 static void napi_hash_del(struct napi_struct *napi) 6277 { 6278 spin_lock(&napi_hash_lock); 6279 6280 hlist_del_init_rcu(&napi->napi_hash_node); 6281 6282 spin_unlock(&napi_hash_lock); 6283 } 6284 6285 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6286 { 6287 struct napi_struct *napi; 6288 6289 napi = container_of(timer, struct napi_struct, timer); 6290 6291 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6292 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6293 */ 6294 if (!napi_disable_pending(napi) && 6295 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6296 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6297 __napi_schedule_irqoff(napi); 6298 } 6299 6300 return HRTIMER_NORESTART; 6301 } 6302 6303 static void init_gro_hash(struct napi_struct *napi) 6304 { 6305 int i; 6306 6307 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6308 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6309 napi->gro_hash[i].count = 0; 6310 } 6311 napi->gro_bitmask = 0; 6312 } 6313 6314 int dev_set_threaded(struct net_device *dev, bool threaded) 6315 { 6316 struct napi_struct *napi; 6317 int err = 0; 6318 6319 if (dev->threaded == threaded) 6320 return 0; 6321 6322 if (threaded) { 6323 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6324 if (!napi->thread) { 6325 err = napi_kthread_create(napi); 6326 if (err) { 6327 threaded = false; 6328 break; 6329 } 6330 } 6331 } 6332 } 6333 6334 dev->threaded = threaded; 6335 6336 /* Make sure kthread is created before THREADED bit 6337 * is set. 6338 */ 6339 smp_mb__before_atomic(); 6340 6341 /* Setting/unsetting threaded mode on a napi might not immediately 6342 * take effect, if the current napi instance is actively being 6343 * polled. In this case, the switch between threaded mode and 6344 * softirq mode will happen in the next round of napi_schedule(). 6345 * This should not cause hiccups/stalls to the live traffic. 6346 */ 6347 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6348 if (threaded) 6349 set_bit(NAPI_STATE_THREADED, &napi->state); 6350 else 6351 clear_bit(NAPI_STATE_THREADED, &napi->state); 6352 } 6353 6354 return err; 6355 } 6356 EXPORT_SYMBOL(dev_set_threaded); 6357 6358 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6359 int (*poll)(struct napi_struct *, int), int weight) 6360 { 6361 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6362 return; 6363 6364 INIT_LIST_HEAD(&napi->poll_list); 6365 INIT_HLIST_NODE(&napi->napi_hash_node); 6366 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6367 napi->timer.function = napi_watchdog; 6368 init_gro_hash(napi); 6369 napi->skb = NULL; 6370 INIT_LIST_HEAD(&napi->rx_list); 6371 napi->rx_count = 0; 6372 napi->poll = poll; 6373 if (weight > NAPI_POLL_WEIGHT) 6374 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6375 weight); 6376 napi->weight = weight; 6377 napi->dev = dev; 6378 #ifdef CONFIG_NETPOLL 6379 napi->poll_owner = -1; 6380 #endif 6381 set_bit(NAPI_STATE_SCHED, &napi->state); 6382 set_bit(NAPI_STATE_NPSVC, &napi->state); 6383 list_add_rcu(&napi->dev_list, &dev->napi_list); 6384 napi_hash_add(napi); 6385 napi_get_frags_check(napi); 6386 /* Create kthread for this napi if dev->threaded is set. 6387 * Clear dev->threaded if kthread creation failed so that 6388 * threaded mode will not be enabled in napi_enable(). 6389 */ 6390 if (dev->threaded && napi_kthread_create(napi)) 6391 dev->threaded = 0; 6392 } 6393 EXPORT_SYMBOL(netif_napi_add_weight); 6394 6395 void napi_disable(struct napi_struct *n) 6396 { 6397 unsigned long val, new; 6398 6399 might_sleep(); 6400 set_bit(NAPI_STATE_DISABLE, &n->state); 6401 6402 val = READ_ONCE(n->state); 6403 do { 6404 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6405 usleep_range(20, 200); 6406 val = READ_ONCE(n->state); 6407 } 6408 6409 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6410 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6411 } while (!try_cmpxchg(&n->state, &val, new)); 6412 6413 hrtimer_cancel(&n->timer); 6414 6415 clear_bit(NAPI_STATE_DISABLE, &n->state); 6416 } 6417 EXPORT_SYMBOL(napi_disable); 6418 6419 /** 6420 * napi_enable - enable NAPI scheduling 6421 * @n: NAPI context 6422 * 6423 * Resume NAPI from being scheduled on this context. 6424 * Must be paired with napi_disable. 6425 */ 6426 void napi_enable(struct napi_struct *n) 6427 { 6428 unsigned long new, val = READ_ONCE(n->state); 6429 6430 do { 6431 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6432 6433 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6434 if (n->dev->threaded && n->thread) 6435 new |= NAPIF_STATE_THREADED; 6436 } while (!try_cmpxchg(&n->state, &val, new)); 6437 } 6438 EXPORT_SYMBOL(napi_enable); 6439 6440 static void flush_gro_hash(struct napi_struct *napi) 6441 { 6442 int i; 6443 6444 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6445 struct sk_buff *skb, *n; 6446 6447 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6448 kfree_skb(skb); 6449 napi->gro_hash[i].count = 0; 6450 } 6451 } 6452 6453 /* Must be called in process context */ 6454 void __netif_napi_del(struct napi_struct *napi) 6455 { 6456 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6457 return; 6458 6459 napi_hash_del(napi); 6460 list_del_rcu(&napi->dev_list); 6461 napi_free_frags(napi); 6462 6463 flush_gro_hash(napi); 6464 napi->gro_bitmask = 0; 6465 6466 if (napi->thread) { 6467 kthread_stop(napi->thread); 6468 napi->thread = NULL; 6469 } 6470 } 6471 EXPORT_SYMBOL(__netif_napi_del); 6472 6473 static int __napi_poll(struct napi_struct *n, bool *repoll) 6474 { 6475 int work, weight; 6476 6477 weight = n->weight; 6478 6479 /* This NAPI_STATE_SCHED test is for avoiding a race 6480 * with netpoll's poll_napi(). Only the entity which 6481 * obtains the lock and sees NAPI_STATE_SCHED set will 6482 * actually make the ->poll() call. Therefore we avoid 6483 * accidentally calling ->poll() when NAPI is not scheduled. 6484 */ 6485 work = 0; 6486 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6487 work = n->poll(n, weight); 6488 trace_napi_poll(n, work, weight); 6489 } 6490 6491 if (unlikely(work > weight)) 6492 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6493 n->poll, work, weight); 6494 6495 if (likely(work < weight)) 6496 return work; 6497 6498 /* Drivers must not modify the NAPI state if they 6499 * consume the entire weight. In such cases this code 6500 * still "owns" the NAPI instance and therefore can 6501 * move the instance around on the list at-will. 6502 */ 6503 if (unlikely(napi_disable_pending(n))) { 6504 napi_complete(n); 6505 return work; 6506 } 6507 6508 /* The NAPI context has more processing work, but busy-polling 6509 * is preferred. Exit early. 6510 */ 6511 if (napi_prefer_busy_poll(n)) { 6512 if (napi_complete_done(n, work)) { 6513 /* If timeout is not set, we need to make sure 6514 * that the NAPI is re-scheduled. 6515 */ 6516 napi_schedule(n); 6517 } 6518 return work; 6519 } 6520 6521 if (n->gro_bitmask) { 6522 /* flush too old packets 6523 * If HZ < 1000, flush all packets. 6524 */ 6525 napi_gro_flush(n, HZ >= 1000); 6526 } 6527 6528 gro_normal_list(n); 6529 6530 /* Some drivers may have called napi_schedule 6531 * prior to exhausting their budget. 6532 */ 6533 if (unlikely(!list_empty(&n->poll_list))) { 6534 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6535 n->dev ? n->dev->name : "backlog"); 6536 return work; 6537 } 6538 6539 *repoll = true; 6540 6541 return work; 6542 } 6543 6544 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6545 { 6546 bool do_repoll = false; 6547 void *have; 6548 int work; 6549 6550 list_del_init(&n->poll_list); 6551 6552 have = netpoll_poll_lock(n); 6553 6554 work = __napi_poll(n, &do_repoll); 6555 6556 if (do_repoll) 6557 list_add_tail(&n->poll_list, repoll); 6558 6559 netpoll_poll_unlock(have); 6560 6561 return work; 6562 } 6563 6564 static int napi_thread_wait(struct napi_struct *napi) 6565 { 6566 bool woken = false; 6567 6568 set_current_state(TASK_INTERRUPTIBLE); 6569 6570 while (!kthread_should_stop()) { 6571 /* Testing SCHED_THREADED bit here to make sure the current 6572 * kthread owns this napi and could poll on this napi. 6573 * Testing SCHED bit is not enough because SCHED bit might be 6574 * set by some other busy poll thread or by napi_disable(). 6575 */ 6576 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6577 WARN_ON(!list_empty(&napi->poll_list)); 6578 __set_current_state(TASK_RUNNING); 6579 return 0; 6580 } 6581 6582 schedule(); 6583 /* woken being true indicates this thread owns this napi. */ 6584 woken = true; 6585 set_current_state(TASK_INTERRUPTIBLE); 6586 } 6587 __set_current_state(TASK_RUNNING); 6588 6589 return -1; 6590 } 6591 6592 static int napi_threaded_poll(void *data) 6593 { 6594 struct napi_struct *napi = data; 6595 void *have; 6596 6597 while (!napi_thread_wait(napi)) { 6598 for (;;) { 6599 bool repoll = false; 6600 6601 local_bh_disable(); 6602 6603 have = netpoll_poll_lock(napi); 6604 __napi_poll(napi, &repoll); 6605 netpoll_poll_unlock(have); 6606 6607 local_bh_enable(); 6608 6609 if (!repoll) 6610 break; 6611 6612 cond_resched(); 6613 } 6614 } 6615 return 0; 6616 } 6617 6618 static void skb_defer_free_flush(struct softnet_data *sd) 6619 { 6620 struct sk_buff *skb, *next; 6621 6622 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6623 if (!READ_ONCE(sd->defer_list)) 6624 return; 6625 6626 spin_lock_irq(&sd->defer_lock); 6627 skb = sd->defer_list; 6628 sd->defer_list = NULL; 6629 sd->defer_count = 0; 6630 spin_unlock_irq(&sd->defer_lock); 6631 6632 while (skb != NULL) { 6633 next = skb->next; 6634 napi_consume_skb(skb, 1); 6635 skb = next; 6636 } 6637 } 6638 6639 static __latent_entropy void net_rx_action(struct softirq_action *h) 6640 { 6641 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6642 unsigned long time_limit = jiffies + 6643 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6644 int budget = READ_ONCE(netdev_budget); 6645 LIST_HEAD(list); 6646 LIST_HEAD(repoll); 6647 6648 local_irq_disable(); 6649 list_splice_init(&sd->poll_list, &list); 6650 local_irq_enable(); 6651 6652 for (;;) { 6653 struct napi_struct *n; 6654 6655 skb_defer_free_flush(sd); 6656 6657 if (list_empty(&list)) { 6658 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6659 goto end; 6660 break; 6661 } 6662 6663 n = list_first_entry(&list, struct napi_struct, poll_list); 6664 budget -= napi_poll(n, &repoll); 6665 6666 /* If softirq window is exhausted then punt. 6667 * Allow this to run for 2 jiffies since which will allow 6668 * an average latency of 1.5/HZ. 6669 */ 6670 if (unlikely(budget <= 0 || 6671 time_after_eq(jiffies, time_limit))) { 6672 sd->time_squeeze++; 6673 break; 6674 } 6675 } 6676 6677 local_irq_disable(); 6678 6679 list_splice_tail_init(&sd->poll_list, &list); 6680 list_splice_tail(&repoll, &list); 6681 list_splice(&list, &sd->poll_list); 6682 if (!list_empty(&sd->poll_list)) 6683 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6684 6685 net_rps_action_and_irq_enable(sd); 6686 end:; 6687 } 6688 6689 struct netdev_adjacent { 6690 struct net_device *dev; 6691 netdevice_tracker dev_tracker; 6692 6693 /* upper master flag, there can only be one master device per list */ 6694 bool master; 6695 6696 /* lookup ignore flag */ 6697 bool ignore; 6698 6699 /* counter for the number of times this device was added to us */ 6700 u16 ref_nr; 6701 6702 /* private field for the users */ 6703 void *private; 6704 6705 struct list_head list; 6706 struct rcu_head rcu; 6707 }; 6708 6709 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6710 struct list_head *adj_list) 6711 { 6712 struct netdev_adjacent *adj; 6713 6714 list_for_each_entry(adj, adj_list, list) { 6715 if (adj->dev == adj_dev) 6716 return adj; 6717 } 6718 return NULL; 6719 } 6720 6721 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6722 struct netdev_nested_priv *priv) 6723 { 6724 struct net_device *dev = (struct net_device *)priv->data; 6725 6726 return upper_dev == dev; 6727 } 6728 6729 /** 6730 * netdev_has_upper_dev - Check if device is linked to an upper device 6731 * @dev: device 6732 * @upper_dev: upper device to check 6733 * 6734 * Find out if a device is linked to specified upper device and return true 6735 * in case it is. Note that this checks only immediate upper device, 6736 * not through a complete stack of devices. The caller must hold the RTNL lock. 6737 */ 6738 bool netdev_has_upper_dev(struct net_device *dev, 6739 struct net_device *upper_dev) 6740 { 6741 struct netdev_nested_priv priv = { 6742 .data = (void *)upper_dev, 6743 }; 6744 6745 ASSERT_RTNL(); 6746 6747 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6748 &priv); 6749 } 6750 EXPORT_SYMBOL(netdev_has_upper_dev); 6751 6752 /** 6753 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6754 * @dev: device 6755 * @upper_dev: upper device to check 6756 * 6757 * Find out if a device is linked to specified upper device and return true 6758 * in case it is. Note that this checks the entire upper device chain. 6759 * The caller must hold rcu lock. 6760 */ 6761 6762 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6763 struct net_device *upper_dev) 6764 { 6765 struct netdev_nested_priv priv = { 6766 .data = (void *)upper_dev, 6767 }; 6768 6769 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6770 &priv); 6771 } 6772 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6773 6774 /** 6775 * netdev_has_any_upper_dev - Check if device is linked to some device 6776 * @dev: device 6777 * 6778 * Find out if a device is linked to an upper device and return true in case 6779 * it is. The caller must hold the RTNL lock. 6780 */ 6781 bool netdev_has_any_upper_dev(struct net_device *dev) 6782 { 6783 ASSERT_RTNL(); 6784 6785 return !list_empty(&dev->adj_list.upper); 6786 } 6787 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6788 6789 /** 6790 * netdev_master_upper_dev_get - Get master upper device 6791 * @dev: device 6792 * 6793 * Find a master upper device and return pointer to it or NULL in case 6794 * it's not there. The caller must hold the RTNL lock. 6795 */ 6796 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6797 { 6798 struct netdev_adjacent *upper; 6799 6800 ASSERT_RTNL(); 6801 6802 if (list_empty(&dev->adj_list.upper)) 6803 return NULL; 6804 6805 upper = list_first_entry(&dev->adj_list.upper, 6806 struct netdev_adjacent, list); 6807 if (likely(upper->master)) 6808 return upper->dev; 6809 return NULL; 6810 } 6811 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6812 6813 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6814 { 6815 struct netdev_adjacent *upper; 6816 6817 ASSERT_RTNL(); 6818 6819 if (list_empty(&dev->adj_list.upper)) 6820 return NULL; 6821 6822 upper = list_first_entry(&dev->adj_list.upper, 6823 struct netdev_adjacent, list); 6824 if (likely(upper->master) && !upper->ignore) 6825 return upper->dev; 6826 return NULL; 6827 } 6828 6829 /** 6830 * netdev_has_any_lower_dev - Check if device is linked to some device 6831 * @dev: device 6832 * 6833 * Find out if a device is linked to a lower device and return true in case 6834 * it is. The caller must hold the RTNL lock. 6835 */ 6836 static bool netdev_has_any_lower_dev(struct net_device *dev) 6837 { 6838 ASSERT_RTNL(); 6839 6840 return !list_empty(&dev->adj_list.lower); 6841 } 6842 6843 void *netdev_adjacent_get_private(struct list_head *adj_list) 6844 { 6845 struct netdev_adjacent *adj; 6846 6847 adj = list_entry(adj_list, struct netdev_adjacent, list); 6848 6849 return adj->private; 6850 } 6851 EXPORT_SYMBOL(netdev_adjacent_get_private); 6852 6853 /** 6854 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6855 * @dev: device 6856 * @iter: list_head ** of the current position 6857 * 6858 * Gets the next device from the dev's upper list, starting from iter 6859 * position. The caller must hold RCU read lock. 6860 */ 6861 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6862 struct list_head **iter) 6863 { 6864 struct netdev_adjacent *upper; 6865 6866 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6867 6868 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6869 6870 if (&upper->list == &dev->adj_list.upper) 6871 return NULL; 6872 6873 *iter = &upper->list; 6874 6875 return upper->dev; 6876 } 6877 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6878 6879 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6880 struct list_head **iter, 6881 bool *ignore) 6882 { 6883 struct netdev_adjacent *upper; 6884 6885 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6886 6887 if (&upper->list == &dev->adj_list.upper) 6888 return NULL; 6889 6890 *iter = &upper->list; 6891 *ignore = upper->ignore; 6892 6893 return upper->dev; 6894 } 6895 6896 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6897 struct list_head **iter) 6898 { 6899 struct netdev_adjacent *upper; 6900 6901 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6902 6903 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6904 6905 if (&upper->list == &dev->adj_list.upper) 6906 return NULL; 6907 6908 *iter = &upper->list; 6909 6910 return upper->dev; 6911 } 6912 6913 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6914 int (*fn)(struct net_device *dev, 6915 struct netdev_nested_priv *priv), 6916 struct netdev_nested_priv *priv) 6917 { 6918 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6919 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6920 int ret, cur = 0; 6921 bool ignore; 6922 6923 now = dev; 6924 iter = &dev->adj_list.upper; 6925 6926 while (1) { 6927 if (now != dev) { 6928 ret = fn(now, priv); 6929 if (ret) 6930 return ret; 6931 } 6932 6933 next = NULL; 6934 while (1) { 6935 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6936 if (!udev) 6937 break; 6938 if (ignore) 6939 continue; 6940 6941 next = udev; 6942 niter = &udev->adj_list.upper; 6943 dev_stack[cur] = now; 6944 iter_stack[cur++] = iter; 6945 break; 6946 } 6947 6948 if (!next) { 6949 if (!cur) 6950 return 0; 6951 next = dev_stack[--cur]; 6952 niter = iter_stack[cur]; 6953 } 6954 6955 now = next; 6956 iter = niter; 6957 } 6958 6959 return 0; 6960 } 6961 6962 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6963 int (*fn)(struct net_device *dev, 6964 struct netdev_nested_priv *priv), 6965 struct netdev_nested_priv *priv) 6966 { 6967 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6968 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6969 int ret, cur = 0; 6970 6971 now = dev; 6972 iter = &dev->adj_list.upper; 6973 6974 while (1) { 6975 if (now != dev) { 6976 ret = fn(now, priv); 6977 if (ret) 6978 return ret; 6979 } 6980 6981 next = NULL; 6982 while (1) { 6983 udev = netdev_next_upper_dev_rcu(now, &iter); 6984 if (!udev) 6985 break; 6986 6987 next = udev; 6988 niter = &udev->adj_list.upper; 6989 dev_stack[cur] = now; 6990 iter_stack[cur++] = iter; 6991 break; 6992 } 6993 6994 if (!next) { 6995 if (!cur) 6996 return 0; 6997 next = dev_stack[--cur]; 6998 niter = iter_stack[cur]; 6999 } 7000 7001 now = next; 7002 iter = niter; 7003 } 7004 7005 return 0; 7006 } 7007 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7008 7009 static bool __netdev_has_upper_dev(struct net_device *dev, 7010 struct net_device *upper_dev) 7011 { 7012 struct netdev_nested_priv priv = { 7013 .flags = 0, 7014 .data = (void *)upper_dev, 7015 }; 7016 7017 ASSERT_RTNL(); 7018 7019 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7020 &priv); 7021 } 7022 7023 /** 7024 * netdev_lower_get_next_private - Get the next ->private from the 7025 * lower neighbour list 7026 * @dev: device 7027 * @iter: list_head ** of the current position 7028 * 7029 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7030 * list, starting from iter position. The caller must hold either hold the 7031 * RTNL lock or its own locking that guarantees that the neighbour lower 7032 * list will remain unchanged. 7033 */ 7034 void *netdev_lower_get_next_private(struct net_device *dev, 7035 struct list_head **iter) 7036 { 7037 struct netdev_adjacent *lower; 7038 7039 lower = list_entry(*iter, struct netdev_adjacent, list); 7040 7041 if (&lower->list == &dev->adj_list.lower) 7042 return NULL; 7043 7044 *iter = lower->list.next; 7045 7046 return lower->private; 7047 } 7048 EXPORT_SYMBOL(netdev_lower_get_next_private); 7049 7050 /** 7051 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7052 * lower neighbour list, RCU 7053 * variant 7054 * @dev: device 7055 * @iter: list_head ** of the current position 7056 * 7057 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7058 * list, starting from iter position. The caller must hold RCU read lock. 7059 */ 7060 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7061 struct list_head **iter) 7062 { 7063 struct netdev_adjacent *lower; 7064 7065 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7066 7067 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7068 7069 if (&lower->list == &dev->adj_list.lower) 7070 return NULL; 7071 7072 *iter = &lower->list; 7073 7074 return lower->private; 7075 } 7076 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7077 7078 /** 7079 * netdev_lower_get_next - Get the next device from the lower neighbour 7080 * list 7081 * @dev: device 7082 * @iter: list_head ** of the current position 7083 * 7084 * Gets the next netdev_adjacent from the dev's lower neighbour 7085 * list, starting from iter position. The caller must hold RTNL lock or 7086 * its own locking that guarantees that the neighbour lower 7087 * list will remain unchanged. 7088 */ 7089 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7090 { 7091 struct netdev_adjacent *lower; 7092 7093 lower = list_entry(*iter, struct netdev_adjacent, list); 7094 7095 if (&lower->list == &dev->adj_list.lower) 7096 return NULL; 7097 7098 *iter = lower->list.next; 7099 7100 return lower->dev; 7101 } 7102 EXPORT_SYMBOL(netdev_lower_get_next); 7103 7104 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7105 struct list_head **iter) 7106 { 7107 struct netdev_adjacent *lower; 7108 7109 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7110 7111 if (&lower->list == &dev->adj_list.lower) 7112 return NULL; 7113 7114 *iter = &lower->list; 7115 7116 return lower->dev; 7117 } 7118 7119 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7120 struct list_head **iter, 7121 bool *ignore) 7122 { 7123 struct netdev_adjacent *lower; 7124 7125 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7126 7127 if (&lower->list == &dev->adj_list.lower) 7128 return NULL; 7129 7130 *iter = &lower->list; 7131 *ignore = lower->ignore; 7132 7133 return lower->dev; 7134 } 7135 7136 int netdev_walk_all_lower_dev(struct net_device *dev, 7137 int (*fn)(struct net_device *dev, 7138 struct netdev_nested_priv *priv), 7139 struct netdev_nested_priv *priv) 7140 { 7141 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7142 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7143 int ret, cur = 0; 7144 7145 now = dev; 7146 iter = &dev->adj_list.lower; 7147 7148 while (1) { 7149 if (now != dev) { 7150 ret = fn(now, priv); 7151 if (ret) 7152 return ret; 7153 } 7154 7155 next = NULL; 7156 while (1) { 7157 ldev = netdev_next_lower_dev(now, &iter); 7158 if (!ldev) 7159 break; 7160 7161 next = ldev; 7162 niter = &ldev->adj_list.lower; 7163 dev_stack[cur] = now; 7164 iter_stack[cur++] = iter; 7165 break; 7166 } 7167 7168 if (!next) { 7169 if (!cur) 7170 return 0; 7171 next = dev_stack[--cur]; 7172 niter = iter_stack[cur]; 7173 } 7174 7175 now = next; 7176 iter = niter; 7177 } 7178 7179 return 0; 7180 } 7181 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7182 7183 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7184 int (*fn)(struct net_device *dev, 7185 struct netdev_nested_priv *priv), 7186 struct netdev_nested_priv *priv) 7187 { 7188 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7189 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7190 int ret, cur = 0; 7191 bool ignore; 7192 7193 now = dev; 7194 iter = &dev->adj_list.lower; 7195 7196 while (1) { 7197 if (now != dev) { 7198 ret = fn(now, priv); 7199 if (ret) 7200 return ret; 7201 } 7202 7203 next = NULL; 7204 while (1) { 7205 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7206 if (!ldev) 7207 break; 7208 if (ignore) 7209 continue; 7210 7211 next = ldev; 7212 niter = &ldev->adj_list.lower; 7213 dev_stack[cur] = now; 7214 iter_stack[cur++] = iter; 7215 break; 7216 } 7217 7218 if (!next) { 7219 if (!cur) 7220 return 0; 7221 next = dev_stack[--cur]; 7222 niter = iter_stack[cur]; 7223 } 7224 7225 now = next; 7226 iter = niter; 7227 } 7228 7229 return 0; 7230 } 7231 7232 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7233 struct list_head **iter) 7234 { 7235 struct netdev_adjacent *lower; 7236 7237 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7238 if (&lower->list == &dev->adj_list.lower) 7239 return NULL; 7240 7241 *iter = &lower->list; 7242 7243 return lower->dev; 7244 } 7245 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7246 7247 static u8 __netdev_upper_depth(struct net_device *dev) 7248 { 7249 struct net_device *udev; 7250 struct list_head *iter; 7251 u8 max_depth = 0; 7252 bool ignore; 7253 7254 for (iter = &dev->adj_list.upper, 7255 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7256 udev; 7257 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7258 if (ignore) 7259 continue; 7260 if (max_depth < udev->upper_level) 7261 max_depth = udev->upper_level; 7262 } 7263 7264 return max_depth; 7265 } 7266 7267 static u8 __netdev_lower_depth(struct net_device *dev) 7268 { 7269 struct net_device *ldev; 7270 struct list_head *iter; 7271 u8 max_depth = 0; 7272 bool ignore; 7273 7274 for (iter = &dev->adj_list.lower, 7275 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7276 ldev; 7277 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7278 if (ignore) 7279 continue; 7280 if (max_depth < ldev->lower_level) 7281 max_depth = ldev->lower_level; 7282 } 7283 7284 return max_depth; 7285 } 7286 7287 static int __netdev_update_upper_level(struct net_device *dev, 7288 struct netdev_nested_priv *__unused) 7289 { 7290 dev->upper_level = __netdev_upper_depth(dev) + 1; 7291 return 0; 7292 } 7293 7294 #ifdef CONFIG_LOCKDEP 7295 static LIST_HEAD(net_unlink_list); 7296 7297 static void net_unlink_todo(struct net_device *dev) 7298 { 7299 if (list_empty(&dev->unlink_list)) 7300 list_add_tail(&dev->unlink_list, &net_unlink_list); 7301 } 7302 #endif 7303 7304 static int __netdev_update_lower_level(struct net_device *dev, 7305 struct netdev_nested_priv *priv) 7306 { 7307 dev->lower_level = __netdev_lower_depth(dev) + 1; 7308 7309 #ifdef CONFIG_LOCKDEP 7310 if (!priv) 7311 return 0; 7312 7313 if (priv->flags & NESTED_SYNC_IMM) 7314 dev->nested_level = dev->lower_level - 1; 7315 if (priv->flags & NESTED_SYNC_TODO) 7316 net_unlink_todo(dev); 7317 #endif 7318 return 0; 7319 } 7320 7321 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7322 int (*fn)(struct net_device *dev, 7323 struct netdev_nested_priv *priv), 7324 struct netdev_nested_priv *priv) 7325 { 7326 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7327 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7328 int ret, cur = 0; 7329 7330 now = dev; 7331 iter = &dev->adj_list.lower; 7332 7333 while (1) { 7334 if (now != dev) { 7335 ret = fn(now, priv); 7336 if (ret) 7337 return ret; 7338 } 7339 7340 next = NULL; 7341 while (1) { 7342 ldev = netdev_next_lower_dev_rcu(now, &iter); 7343 if (!ldev) 7344 break; 7345 7346 next = ldev; 7347 niter = &ldev->adj_list.lower; 7348 dev_stack[cur] = now; 7349 iter_stack[cur++] = iter; 7350 break; 7351 } 7352 7353 if (!next) { 7354 if (!cur) 7355 return 0; 7356 next = dev_stack[--cur]; 7357 niter = iter_stack[cur]; 7358 } 7359 7360 now = next; 7361 iter = niter; 7362 } 7363 7364 return 0; 7365 } 7366 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7367 7368 /** 7369 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7370 * lower neighbour list, RCU 7371 * variant 7372 * @dev: device 7373 * 7374 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7375 * list. The caller must hold RCU read lock. 7376 */ 7377 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7378 { 7379 struct netdev_adjacent *lower; 7380 7381 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7382 struct netdev_adjacent, list); 7383 if (lower) 7384 return lower->private; 7385 return NULL; 7386 } 7387 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7388 7389 /** 7390 * netdev_master_upper_dev_get_rcu - Get master upper device 7391 * @dev: device 7392 * 7393 * Find a master upper device and return pointer to it or NULL in case 7394 * it's not there. The caller must hold the RCU read lock. 7395 */ 7396 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7397 { 7398 struct netdev_adjacent *upper; 7399 7400 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7401 struct netdev_adjacent, list); 7402 if (upper && likely(upper->master)) 7403 return upper->dev; 7404 return NULL; 7405 } 7406 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7407 7408 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7409 struct net_device *adj_dev, 7410 struct list_head *dev_list) 7411 { 7412 char linkname[IFNAMSIZ+7]; 7413 7414 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7415 "upper_%s" : "lower_%s", adj_dev->name); 7416 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7417 linkname); 7418 } 7419 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7420 char *name, 7421 struct list_head *dev_list) 7422 { 7423 char linkname[IFNAMSIZ+7]; 7424 7425 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7426 "upper_%s" : "lower_%s", name); 7427 sysfs_remove_link(&(dev->dev.kobj), linkname); 7428 } 7429 7430 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7431 struct net_device *adj_dev, 7432 struct list_head *dev_list) 7433 { 7434 return (dev_list == &dev->adj_list.upper || 7435 dev_list == &dev->adj_list.lower) && 7436 net_eq(dev_net(dev), dev_net(adj_dev)); 7437 } 7438 7439 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7440 struct net_device *adj_dev, 7441 struct list_head *dev_list, 7442 void *private, bool master) 7443 { 7444 struct netdev_adjacent *adj; 7445 int ret; 7446 7447 adj = __netdev_find_adj(adj_dev, dev_list); 7448 7449 if (adj) { 7450 adj->ref_nr += 1; 7451 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7452 dev->name, adj_dev->name, adj->ref_nr); 7453 7454 return 0; 7455 } 7456 7457 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7458 if (!adj) 7459 return -ENOMEM; 7460 7461 adj->dev = adj_dev; 7462 adj->master = master; 7463 adj->ref_nr = 1; 7464 adj->private = private; 7465 adj->ignore = false; 7466 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7467 7468 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7469 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7470 7471 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7472 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7473 if (ret) 7474 goto free_adj; 7475 } 7476 7477 /* Ensure that master link is always the first item in list. */ 7478 if (master) { 7479 ret = sysfs_create_link(&(dev->dev.kobj), 7480 &(adj_dev->dev.kobj), "master"); 7481 if (ret) 7482 goto remove_symlinks; 7483 7484 list_add_rcu(&adj->list, dev_list); 7485 } else { 7486 list_add_tail_rcu(&adj->list, dev_list); 7487 } 7488 7489 return 0; 7490 7491 remove_symlinks: 7492 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7493 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7494 free_adj: 7495 netdev_put(adj_dev, &adj->dev_tracker); 7496 kfree(adj); 7497 7498 return ret; 7499 } 7500 7501 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7502 struct net_device *adj_dev, 7503 u16 ref_nr, 7504 struct list_head *dev_list) 7505 { 7506 struct netdev_adjacent *adj; 7507 7508 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7509 dev->name, adj_dev->name, ref_nr); 7510 7511 adj = __netdev_find_adj(adj_dev, dev_list); 7512 7513 if (!adj) { 7514 pr_err("Adjacency does not exist for device %s from %s\n", 7515 dev->name, adj_dev->name); 7516 WARN_ON(1); 7517 return; 7518 } 7519 7520 if (adj->ref_nr > ref_nr) { 7521 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7522 dev->name, adj_dev->name, ref_nr, 7523 adj->ref_nr - ref_nr); 7524 adj->ref_nr -= ref_nr; 7525 return; 7526 } 7527 7528 if (adj->master) 7529 sysfs_remove_link(&(dev->dev.kobj), "master"); 7530 7531 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7532 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7533 7534 list_del_rcu(&adj->list); 7535 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7536 adj_dev->name, dev->name, adj_dev->name); 7537 netdev_put(adj_dev, &adj->dev_tracker); 7538 kfree_rcu(adj, rcu); 7539 } 7540 7541 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7542 struct net_device *upper_dev, 7543 struct list_head *up_list, 7544 struct list_head *down_list, 7545 void *private, bool master) 7546 { 7547 int ret; 7548 7549 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7550 private, master); 7551 if (ret) 7552 return ret; 7553 7554 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7555 private, false); 7556 if (ret) { 7557 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7558 return ret; 7559 } 7560 7561 return 0; 7562 } 7563 7564 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7565 struct net_device *upper_dev, 7566 u16 ref_nr, 7567 struct list_head *up_list, 7568 struct list_head *down_list) 7569 { 7570 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7571 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7572 } 7573 7574 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7575 struct net_device *upper_dev, 7576 void *private, bool master) 7577 { 7578 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7579 &dev->adj_list.upper, 7580 &upper_dev->adj_list.lower, 7581 private, master); 7582 } 7583 7584 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7585 struct net_device *upper_dev) 7586 { 7587 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7588 &dev->adj_list.upper, 7589 &upper_dev->adj_list.lower); 7590 } 7591 7592 static int __netdev_upper_dev_link(struct net_device *dev, 7593 struct net_device *upper_dev, bool master, 7594 void *upper_priv, void *upper_info, 7595 struct netdev_nested_priv *priv, 7596 struct netlink_ext_ack *extack) 7597 { 7598 struct netdev_notifier_changeupper_info changeupper_info = { 7599 .info = { 7600 .dev = dev, 7601 .extack = extack, 7602 }, 7603 .upper_dev = upper_dev, 7604 .master = master, 7605 .linking = true, 7606 .upper_info = upper_info, 7607 }; 7608 struct net_device *master_dev; 7609 int ret = 0; 7610 7611 ASSERT_RTNL(); 7612 7613 if (dev == upper_dev) 7614 return -EBUSY; 7615 7616 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7617 if (__netdev_has_upper_dev(upper_dev, dev)) 7618 return -EBUSY; 7619 7620 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7621 return -EMLINK; 7622 7623 if (!master) { 7624 if (__netdev_has_upper_dev(dev, upper_dev)) 7625 return -EEXIST; 7626 } else { 7627 master_dev = __netdev_master_upper_dev_get(dev); 7628 if (master_dev) 7629 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7630 } 7631 7632 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7633 &changeupper_info.info); 7634 ret = notifier_to_errno(ret); 7635 if (ret) 7636 return ret; 7637 7638 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7639 master); 7640 if (ret) 7641 return ret; 7642 7643 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7644 &changeupper_info.info); 7645 ret = notifier_to_errno(ret); 7646 if (ret) 7647 goto rollback; 7648 7649 __netdev_update_upper_level(dev, NULL); 7650 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7651 7652 __netdev_update_lower_level(upper_dev, priv); 7653 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7654 priv); 7655 7656 return 0; 7657 7658 rollback: 7659 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7660 7661 return ret; 7662 } 7663 7664 /** 7665 * netdev_upper_dev_link - Add a link to the upper device 7666 * @dev: device 7667 * @upper_dev: new upper device 7668 * @extack: netlink extended ack 7669 * 7670 * Adds a link to device which is upper to this one. The caller must hold 7671 * the RTNL lock. On a failure a negative errno code is returned. 7672 * On success the reference counts are adjusted and the function 7673 * returns zero. 7674 */ 7675 int netdev_upper_dev_link(struct net_device *dev, 7676 struct net_device *upper_dev, 7677 struct netlink_ext_ack *extack) 7678 { 7679 struct netdev_nested_priv priv = { 7680 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7681 .data = NULL, 7682 }; 7683 7684 return __netdev_upper_dev_link(dev, upper_dev, false, 7685 NULL, NULL, &priv, extack); 7686 } 7687 EXPORT_SYMBOL(netdev_upper_dev_link); 7688 7689 /** 7690 * netdev_master_upper_dev_link - Add a master link to the upper device 7691 * @dev: device 7692 * @upper_dev: new upper device 7693 * @upper_priv: upper device private 7694 * @upper_info: upper info to be passed down via notifier 7695 * @extack: netlink extended ack 7696 * 7697 * Adds a link to device which is upper to this one. In this case, only 7698 * one master upper device can be linked, although other non-master devices 7699 * might be linked as well. The caller must hold the RTNL lock. 7700 * On a failure a negative errno code is returned. On success the reference 7701 * counts are adjusted and the function returns zero. 7702 */ 7703 int netdev_master_upper_dev_link(struct net_device *dev, 7704 struct net_device *upper_dev, 7705 void *upper_priv, void *upper_info, 7706 struct netlink_ext_ack *extack) 7707 { 7708 struct netdev_nested_priv priv = { 7709 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7710 .data = NULL, 7711 }; 7712 7713 return __netdev_upper_dev_link(dev, upper_dev, true, 7714 upper_priv, upper_info, &priv, extack); 7715 } 7716 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7717 7718 static void __netdev_upper_dev_unlink(struct net_device *dev, 7719 struct net_device *upper_dev, 7720 struct netdev_nested_priv *priv) 7721 { 7722 struct netdev_notifier_changeupper_info changeupper_info = { 7723 .info = { 7724 .dev = dev, 7725 }, 7726 .upper_dev = upper_dev, 7727 .linking = false, 7728 }; 7729 7730 ASSERT_RTNL(); 7731 7732 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7733 7734 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7735 &changeupper_info.info); 7736 7737 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7738 7739 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7740 &changeupper_info.info); 7741 7742 __netdev_update_upper_level(dev, NULL); 7743 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7744 7745 __netdev_update_lower_level(upper_dev, priv); 7746 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7747 priv); 7748 } 7749 7750 /** 7751 * netdev_upper_dev_unlink - Removes a link to upper device 7752 * @dev: device 7753 * @upper_dev: new upper device 7754 * 7755 * Removes a link to device which is upper to this one. The caller must hold 7756 * the RTNL lock. 7757 */ 7758 void netdev_upper_dev_unlink(struct net_device *dev, 7759 struct net_device *upper_dev) 7760 { 7761 struct netdev_nested_priv priv = { 7762 .flags = NESTED_SYNC_TODO, 7763 .data = NULL, 7764 }; 7765 7766 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7767 } 7768 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7769 7770 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7771 struct net_device *lower_dev, 7772 bool val) 7773 { 7774 struct netdev_adjacent *adj; 7775 7776 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7777 if (adj) 7778 adj->ignore = val; 7779 7780 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7781 if (adj) 7782 adj->ignore = val; 7783 } 7784 7785 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7786 struct net_device *lower_dev) 7787 { 7788 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7789 } 7790 7791 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7792 struct net_device *lower_dev) 7793 { 7794 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7795 } 7796 7797 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7798 struct net_device *new_dev, 7799 struct net_device *dev, 7800 struct netlink_ext_ack *extack) 7801 { 7802 struct netdev_nested_priv priv = { 7803 .flags = 0, 7804 .data = NULL, 7805 }; 7806 int err; 7807 7808 if (!new_dev) 7809 return 0; 7810 7811 if (old_dev && new_dev != old_dev) 7812 netdev_adjacent_dev_disable(dev, old_dev); 7813 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7814 extack); 7815 if (err) { 7816 if (old_dev && new_dev != old_dev) 7817 netdev_adjacent_dev_enable(dev, old_dev); 7818 return err; 7819 } 7820 7821 return 0; 7822 } 7823 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7824 7825 void netdev_adjacent_change_commit(struct net_device *old_dev, 7826 struct net_device *new_dev, 7827 struct net_device *dev) 7828 { 7829 struct netdev_nested_priv priv = { 7830 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7831 .data = NULL, 7832 }; 7833 7834 if (!new_dev || !old_dev) 7835 return; 7836 7837 if (new_dev == old_dev) 7838 return; 7839 7840 netdev_adjacent_dev_enable(dev, old_dev); 7841 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7842 } 7843 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7844 7845 void netdev_adjacent_change_abort(struct net_device *old_dev, 7846 struct net_device *new_dev, 7847 struct net_device *dev) 7848 { 7849 struct netdev_nested_priv priv = { 7850 .flags = 0, 7851 .data = NULL, 7852 }; 7853 7854 if (!new_dev) 7855 return; 7856 7857 if (old_dev && new_dev != old_dev) 7858 netdev_adjacent_dev_enable(dev, old_dev); 7859 7860 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7861 } 7862 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7863 7864 /** 7865 * netdev_bonding_info_change - Dispatch event about slave change 7866 * @dev: device 7867 * @bonding_info: info to dispatch 7868 * 7869 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7870 * The caller must hold the RTNL lock. 7871 */ 7872 void netdev_bonding_info_change(struct net_device *dev, 7873 struct netdev_bonding_info *bonding_info) 7874 { 7875 struct netdev_notifier_bonding_info info = { 7876 .info.dev = dev, 7877 }; 7878 7879 memcpy(&info.bonding_info, bonding_info, 7880 sizeof(struct netdev_bonding_info)); 7881 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7882 &info.info); 7883 } 7884 EXPORT_SYMBOL(netdev_bonding_info_change); 7885 7886 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7887 struct netlink_ext_ack *extack) 7888 { 7889 struct netdev_notifier_offload_xstats_info info = { 7890 .info.dev = dev, 7891 .info.extack = extack, 7892 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7893 }; 7894 int err; 7895 int rc; 7896 7897 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7898 GFP_KERNEL); 7899 if (!dev->offload_xstats_l3) 7900 return -ENOMEM; 7901 7902 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7903 NETDEV_OFFLOAD_XSTATS_DISABLE, 7904 &info.info); 7905 err = notifier_to_errno(rc); 7906 if (err) 7907 goto free_stats; 7908 7909 return 0; 7910 7911 free_stats: 7912 kfree(dev->offload_xstats_l3); 7913 dev->offload_xstats_l3 = NULL; 7914 return err; 7915 } 7916 7917 int netdev_offload_xstats_enable(struct net_device *dev, 7918 enum netdev_offload_xstats_type type, 7919 struct netlink_ext_ack *extack) 7920 { 7921 ASSERT_RTNL(); 7922 7923 if (netdev_offload_xstats_enabled(dev, type)) 7924 return -EALREADY; 7925 7926 switch (type) { 7927 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7928 return netdev_offload_xstats_enable_l3(dev, extack); 7929 } 7930 7931 WARN_ON(1); 7932 return -EINVAL; 7933 } 7934 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7935 7936 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7937 { 7938 struct netdev_notifier_offload_xstats_info info = { 7939 .info.dev = dev, 7940 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7941 }; 7942 7943 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7944 &info.info); 7945 kfree(dev->offload_xstats_l3); 7946 dev->offload_xstats_l3 = NULL; 7947 } 7948 7949 int netdev_offload_xstats_disable(struct net_device *dev, 7950 enum netdev_offload_xstats_type type) 7951 { 7952 ASSERT_RTNL(); 7953 7954 if (!netdev_offload_xstats_enabled(dev, type)) 7955 return -EALREADY; 7956 7957 switch (type) { 7958 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7959 netdev_offload_xstats_disable_l3(dev); 7960 return 0; 7961 } 7962 7963 WARN_ON(1); 7964 return -EINVAL; 7965 } 7966 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7967 7968 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7969 { 7970 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7971 } 7972 7973 static struct rtnl_hw_stats64 * 7974 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7975 enum netdev_offload_xstats_type type) 7976 { 7977 switch (type) { 7978 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7979 return dev->offload_xstats_l3; 7980 } 7981 7982 WARN_ON(1); 7983 return NULL; 7984 } 7985 7986 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7987 enum netdev_offload_xstats_type type) 7988 { 7989 ASSERT_RTNL(); 7990 7991 return netdev_offload_xstats_get_ptr(dev, type); 7992 } 7993 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7994 7995 struct netdev_notifier_offload_xstats_ru { 7996 bool used; 7997 }; 7998 7999 struct netdev_notifier_offload_xstats_rd { 8000 struct rtnl_hw_stats64 stats; 8001 bool used; 8002 }; 8003 8004 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8005 const struct rtnl_hw_stats64 *src) 8006 { 8007 dest->rx_packets += src->rx_packets; 8008 dest->tx_packets += src->tx_packets; 8009 dest->rx_bytes += src->rx_bytes; 8010 dest->tx_bytes += src->tx_bytes; 8011 dest->rx_errors += src->rx_errors; 8012 dest->tx_errors += src->tx_errors; 8013 dest->rx_dropped += src->rx_dropped; 8014 dest->tx_dropped += src->tx_dropped; 8015 dest->multicast += src->multicast; 8016 } 8017 8018 static int netdev_offload_xstats_get_used(struct net_device *dev, 8019 enum netdev_offload_xstats_type type, 8020 bool *p_used, 8021 struct netlink_ext_ack *extack) 8022 { 8023 struct netdev_notifier_offload_xstats_ru report_used = {}; 8024 struct netdev_notifier_offload_xstats_info info = { 8025 .info.dev = dev, 8026 .info.extack = extack, 8027 .type = type, 8028 .report_used = &report_used, 8029 }; 8030 int rc; 8031 8032 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8033 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8034 &info.info); 8035 *p_used = report_used.used; 8036 return notifier_to_errno(rc); 8037 } 8038 8039 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8040 enum netdev_offload_xstats_type type, 8041 struct rtnl_hw_stats64 *p_stats, 8042 bool *p_used, 8043 struct netlink_ext_ack *extack) 8044 { 8045 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8046 struct netdev_notifier_offload_xstats_info info = { 8047 .info.dev = dev, 8048 .info.extack = extack, 8049 .type = type, 8050 .report_delta = &report_delta, 8051 }; 8052 struct rtnl_hw_stats64 *stats; 8053 int rc; 8054 8055 stats = netdev_offload_xstats_get_ptr(dev, type); 8056 if (WARN_ON(!stats)) 8057 return -EINVAL; 8058 8059 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8060 &info.info); 8061 8062 /* Cache whatever we got, even if there was an error, otherwise the 8063 * successful stats retrievals would get lost. 8064 */ 8065 netdev_hw_stats64_add(stats, &report_delta.stats); 8066 8067 if (p_stats) 8068 *p_stats = *stats; 8069 *p_used = report_delta.used; 8070 8071 return notifier_to_errno(rc); 8072 } 8073 8074 int netdev_offload_xstats_get(struct net_device *dev, 8075 enum netdev_offload_xstats_type type, 8076 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8077 struct netlink_ext_ack *extack) 8078 { 8079 ASSERT_RTNL(); 8080 8081 if (p_stats) 8082 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8083 p_used, extack); 8084 else 8085 return netdev_offload_xstats_get_used(dev, type, p_used, 8086 extack); 8087 } 8088 EXPORT_SYMBOL(netdev_offload_xstats_get); 8089 8090 void 8091 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8092 const struct rtnl_hw_stats64 *stats) 8093 { 8094 report_delta->used = true; 8095 netdev_hw_stats64_add(&report_delta->stats, stats); 8096 } 8097 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8098 8099 void 8100 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8101 { 8102 report_used->used = true; 8103 } 8104 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8105 8106 void netdev_offload_xstats_push_delta(struct net_device *dev, 8107 enum netdev_offload_xstats_type type, 8108 const struct rtnl_hw_stats64 *p_stats) 8109 { 8110 struct rtnl_hw_stats64 *stats; 8111 8112 ASSERT_RTNL(); 8113 8114 stats = netdev_offload_xstats_get_ptr(dev, type); 8115 if (WARN_ON(!stats)) 8116 return; 8117 8118 netdev_hw_stats64_add(stats, p_stats); 8119 } 8120 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8121 8122 /** 8123 * netdev_get_xmit_slave - Get the xmit slave of master device 8124 * @dev: device 8125 * @skb: The packet 8126 * @all_slaves: assume all the slaves are active 8127 * 8128 * The reference counters are not incremented so the caller must be 8129 * careful with locks. The caller must hold RCU lock. 8130 * %NULL is returned if no slave is found. 8131 */ 8132 8133 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8134 struct sk_buff *skb, 8135 bool all_slaves) 8136 { 8137 const struct net_device_ops *ops = dev->netdev_ops; 8138 8139 if (!ops->ndo_get_xmit_slave) 8140 return NULL; 8141 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8142 } 8143 EXPORT_SYMBOL(netdev_get_xmit_slave); 8144 8145 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8146 struct sock *sk) 8147 { 8148 const struct net_device_ops *ops = dev->netdev_ops; 8149 8150 if (!ops->ndo_sk_get_lower_dev) 8151 return NULL; 8152 return ops->ndo_sk_get_lower_dev(dev, sk); 8153 } 8154 8155 /** 8156 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8157 * @dev: device 8158 * @sk: the socket 8159 * 8160 * %NULL is returned if no lower device is found. 8161 */ 8162 8163 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8164 struct sock *sk) 8165 { 8166 struct net_device *lower; 8167 8168 lower = netdev_sk_get_lower_dev(dev, sk); 8169 while (lower) { 8170 dev = lower; 8171 lower = netdev_sk_get_lower_dev(dev, sk); 8172 } 8173 8174 return dev; 8175 } 8176 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8177 8178 static void netdev_adjacent_add_links(struct net_device *dev) 8179 { 8180 struct netdev_adjacent *iter; 8181 8182 struct net *net = dev_net(dev); 8183 8184 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8185 if (!net_eq(net, dev_net(iter->dev))) 8186 continue; 8187 netdev_adjacent_sysfs_add(iter->dev, dev, 8188 &iter->dev->adj_list.lower); 8189 netdev_adjacent_sysfs_add(dev, iter->dev, 8190 &dev->adj_list.upper); 8191 } 8192 8193 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8194 if (!net_eq(net, dev_net(iter->dev))) 8195 continue; 8196 netdev_adjacent_sysfs_add(iter->dev, dev, 8197 &iter->dev->adj_list.upper); 8198 netdev_adjacent_sysfs_add(dev, iter->dev, 8199 &dev->adj_list.lower); 8200 } 8201 } 8202 8203 static void netdev_adjacent_del_links(struct net_device *dev) 8204 { 8205 struct netdev_adjacent *iter; 8206 8207 struct net *net = dev_net(dev); 8208 8209 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8210 if (!net_eq(net, dev_net(iter->dev))) 8211 continue; 8212 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8213 &iter->dev->adj_list.lower); 8214 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8215 &dev->adj_list.upper); 8216 } 8217 8218 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8219 if (!net_eq(net, dev_net(iter->dev))) 8220 continue; 8221 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8222 &iter->dev->adj_list.upper); 8223 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8224 &dev->adj_list.lower); 8225 } 8226 } 8227 8228 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8229 { 8230 struct netdev_adjacent *iter; 8231 8232 struct net *net = dev_net(dev); 8233 8234 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8235 if (!net_eq(net, dev_net(iter->dev))) 8236 continue; 8237 netdev_adjacent_sysfs_del(iter->dev, oldname, 8238 &iter->dev->adj_list.lower); 8239 netdev_adjacent_sysfs_add(iter->dev, dev, 8240 &iter->dev->adj_list.lower); 8241 } 8242 8243 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8244 if (!net_eq(net, dev_net(iter->dev))) 8245 continue; 8246 netdev_adjacent_sysfs_del(iter->dev, oldname, 8247 &iter->dev->adj_list.upper); 8248 netdev_adjacent_sysfs_add(iter->dev, dev, 8249 &iter->dev->adj_list.upper); 8250 } 8251 } 8252 8253 void *netdev_lower_dev_get_private(struct net_device *dev, 8254 struct net_device *lower_dev) 8255 { 8256 struct netdev_adjacent *lower; 8257 8258 if (!lower_dev) 8259 return NULL; 8260 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8261 if (!lower) 8262 return NULL; 8263 8264 return lower->private; 8265 } 8266 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8267 8268 8269 /** 8270 * netdev_lower_state_changed - Dispatch event about lower device state change 8271 * @lower_dev: device 8272 * @lower_state_info: state to dispatch 8273 * 8274 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8275 * The caller must hold the RTNL lock. 8276 */ 8277 void netdev_lower_state_changed(struct net_device *lower_dev, 8278 void *lower_state_info) 8279 { 8280 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8281 .info.dev = lower_dev, 8282 }; 8283 8284 ASSERT_RTNL(); 8285 changelowerstate_info.lower_state_info = lower_state_info; 8286 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8287 &changelowerstate_info.info); 8288 } 8289 EXPORT_SYMBOL(netdev_lower_state_changed); 8290 8291 static void dev_change_rx_flags(struct net_device *dev, int flags) 8292 { 8293 const struct net_device_ops *ops = dev->netdev_ops; 8294 8295 if (ops->ndo_change_rx_flags) 8296 ops->ndo_change_rx_flags(dev, flags); 8297 } 8298 8299 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8300 { 8301 unsigned int old_flags = dev->flags; 8302 kuid_t uid; 8303 kgid_t gid; 8304 8305 ASSERT_RTNL(); 8306 8307 dev->flags |= IFF_PROMISC; 8308 dev->promiscuity += inc; 8309 if (dev->promiscuity == 0) { 8310 /* 8311 * Avoid overflow. 8312 * If inc causes overflow, untouch promisc and return error. 8313 */ 8314 if (inc < 0) 8315 dev->flags &= ~IFF_PROMISC; 8316 else { 8317 dev->promiscuity -= inc; 8318 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8319 return -EOVERFLOW; 8320 } 8321 } 8322 if (dev->flags != old_flags) { 8323 pr_info("device %s %s promiscuous mode\n", 8324 dev->name, 8325 dev->flags & IFF_PROMISC ? "entered" : "left"); 8326 if (audit_enabled) { 8327 current_uid_gid(&uid, &gid); 8328 audit_log(audit_context(), GFP_ATOMIC, 8329 AUDIT_ANOM_PROMISCUOUS, 8330 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8331 dev->name, (dev->flags & IFF_PROMISC), 8332 (old_flags & IFF_PROMISC), 8333 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8334 from_kuid(&init_user_ns, uid), 8335 from_kgid(&init_user_ns, gid), 8336 audit_get_sessionid(current)); 8337 } 8338 8339 dev_change_rx_flags(dev, IFF_PROMISC); 8340 } 8341 if (notify) 8342 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8343 return 0; 8344 } 8345 8346 /** 8347 * dev_set_promiscuity - update promiscuity count on a device 8348 * @dev: device 8349 * @inc: modifier 8350 * 8351 * Add or remove promiscuity from a device. While the count in the device 8352 * remains above zero the interface remains promiscuous. Once it hits zero 8353 * the device reverts back to normal filtering operation. A negative inc 8354 * value is used to drop promiscuity on the device. 8355 * Return 0 if successful or a negative errno code on error. 8356 */ 8357 int dev_set_promiscuity(struct net_device *dev, int inc) 8358 { 8359 unsigned int old_flags = dev->flags; 8360 int err; 8361 8362 err = __dev_set_promiscuity(dev, inc, true); 8363 if (err < 0) 8364 return err; 8365 if (dev->flags != old_flags) 8366 dev_set_rx_mode(dev); 8367 return err; 8368 } 8369 EXPORT_SYMBOL(dev_set_promiscuity); 8370 8371 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8372 { 8373 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8374 8375 ASSERT_RTNL(); 8376 8377 dev->flags |= IFF_ALLMULTI; 8378 dev->allmulti += inc; 8379 if (dev->allmulti == 0) { 8380 /* 8381 * Avoid overflow. 8382 * If inc causes overflow, untouch allmulti and return error. 8383 */ 8384 if (inc < 0) 8385 dev->flags &= ~IFF_ALLMULTI; 8386 else { 8387 dev->allmulti -= inc; 8388 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8389 return -EOVERFLOW; 8390 } 8391 } 8392 if (dev->flags ^ old_flags) { 8393 dev_change_rx_flags(dev, IFF_ALLMULTI); 8394 dev_set_rx_mode(dev); 8395 if (notify) 8396 __dev_notify_flags(dev, old_flags, 8397 dev->gflags ^ old_gflags, 0, NULL); 8398 } 8399 return 0; 8400 } 8401 8402 /** 8403 * dev_set_allmulti - update allmulti count on a device 8404 * @dev: device 8405 * @inc: modifier 8406 * 8407 * Add or remove reception of all multicast frames to a device. While the 8408 * count in the device remains above zero the interface remains listening 8409 * to all interfaces. Once it hits zero the device reverts back to normal 8410 * filtering operation. A negative @inc value is used to drop the counter 8411 * when releasing a resource needing all multicasts. 8412 * Return 0 if successful or a negative errno code on error. 8413 */ 8414 8415 int dev_set_allmulti(struct net_device *dev, int inc) 8416 { 8417 return __dev_set_allmulti(dev, inc, true); 8418 } 8419 EXPORT_SYMBOL(dev_set_allmulti); 8420 8421 /* 8422 * Upload unicast and multicast address lists to device and 8423 * configure RX filtering. When the device doesn't support unicast 8424 * filtering it is put in promiscuous mode while unicast addresses 8425 * are present. 8426 */ 8427 void __dev_set_rx_mode(struct net_device *dev) 8428 { 8429 const struct net_device_ops *ops = dev->netdev_ops; 8430 8431 /* dev_open will call this function so the list will stay sane. */ 8432 if (!(dev->flags&IFF_UP)) 8433 return; 8434 8435 if (!netif_device_present(dev)) 8436 return; 8437 8438 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8439 /* Unicast addresses changes may only happen under the rtnl, 8440 * therefore calling __dev_set_promiscuity here is safe. 8441 */ 8442 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8443 __dev_set_promiscuity(dev, 1, false); 8444 dev->uc_promisc = true; 8445 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8446 __dev_set_promiscuity(dev, -1, false); 8447 dev->uc_promisc = false; 8448 } 8449 } 8450 8451 if (ops->ndo_set_rx_mode) 8452 ops->ndo_set_rx_mode(dev); 8453 } 8454 8455 void dev_set_rx_mode(struct net_device *dev) 8456 { 8457 netif_addr_lock_bh(dev); 8458 __dev_set_rx_mode(dev); 8459 netif_addr_unlock_bh(dev); 8460 } 8461 8462 /** 8463 * dev_get_flags - get flags reported to userspace 8464 * @dev: device 8465 * 8466 * Get the combination of flag bits exported through APIs to userspace. 8467 */ 8468 unsigned int dev_get_flags(const struct net_device *dev) 8469 { 8470 unsigned int flags; 8471 8472 flags = (dev->flags & ~(IFF_PROMISC | 8473 IFF_ALLMULTI | 8474 IFF_RUNNING | 8475 IFF_LOWER_UP | 8476 IFF_DORMANT)) | 8477 (dev->gflags & (IFF_PROMISC | 8478 IFF_ALLMULTI)); 8479 8480 if (netif_running(dev)) { 8481 if (netif_oper_up(dev)) 8482 flags |= IFF_RUNNING; 8483 if (netif_carrier_ok(dev)) 8484 flags |= IFF_LOWER_UP; 8485 if (netif_dormant(dev)) 8486 flags |= IFF_DORMANT; 8487 } 8488 8489 return flags; 8490 } 8491 EXPORT_SYMBOL(dev_get_flags); 8492 8493 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8494 struct netlink_ext_ack *extack) 8495 { 8496 unsigned int old_flags = dev->flags; 8497 int ret; 8498 8499 ASSERT_RTNL(); 8500 8501 /* 8502 * Set the flags on our device. 8503 */ 8504 8505 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8506 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8507 IFF_AUTOMEDIA)) | 8508 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8509 IFF_ALLMULTI)); 8510 8511 /* 8512 * Load in the correct multicast list now the flags have changed. 8513 */ 8514 8515 if ((old_flags ^ flags) & IFF_MULTICAST) 8516 dev_change_rx_flags(dev, IFF_MULTICAST); 8517 8518 dev_set_rx_mode(dev); 8519 8520 /* 8521 * Have we downed the interface. We handle IFF_UP ourselves 8522 * according to user attempts to set it, rather than blindly 8523 * setting it. 8524 */ 8525 8526 ret = 0; 8527 if ((old_flags ^ flags) & IFF_UP) { 8528 if (old_flags & IFF_UP) 8529 __dev_close(dev); 8530 else 8531 ret = __dev_open(dev, extack); 8532 } 8533 8534 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8535 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8536 unsigned int old_flags = dev->flags; 8537 8538 dev->gflags ^= IFF_PROMISC; 8539 8540 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8541 if (dev->flags != old_flags) 8542 dev_set_rx_mode(dev); 8543 } 8544 8545 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8546 * is important. Some (broken) drivers set IFF_PROMISC, when 8547 * IFF_ALLMULTI is requested not asking us and not reporting. 8548 */ 8549 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8550 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8551 8552 dev->gflags ^= IFF_ALLMULTI; 8553 __dev_set_allmulti(dev, inc, false); 8554 } 8555 8556 return ret; 8557 } 8558 8559 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8560 unsigned int gchanges, u32 portid, 8561 const struct nlmsghdr *nlh) 8562 { 8563 unsigned int changes = dev->flags ^ old_flags; 8564 8565 if (gchanges) 8566 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8567 8568 if (changes & IFF_UP) { 8569 if (dev->flags & IFF_UP) 8570 call_netdevice_notifiers(NETDEV_UP, dev); 8571 else 8572 call_netdevice_notifiers(NETDEV_DOWN, dev); 8573 } 8574 8575 if (dev->flags & IFF_UP && 8576 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8577 struct netdev_notifier_change_info change_info = { 8578 .info = { 8579 .dev = dev, 8580 }, 8581 .flags_changed = changes, 8582 }; 8583 8584 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8585 } 8586 } 8587 8588 /** 8589 * dev_change_flags - change device settings 8590 * @dev: device 8591 * @flags: device state flags 8592 * @extack: netlink extended ack 8593 * 8594 * Change settings on device based state flags. The flags are 8595 * in the userspace exported format. 8596 */ 8597 int dev_change_flags(struct net_device *dev, unsigned int flags, 8598 struct netlink_ext_ack *extack) 8599 { 8600 int ret; 8601 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8602 8603 ret = __dev_change_flags(dev, flags, extack); 8604 if (ret < 0) 8605 return ret; 8606 8607 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8608 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8609 return ret; 8610 } 8611 EXPORT_SYMBOL(dev_change_flags); 8612 8613 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8614 { 8615 const struct net_device_ops *ops = dev->netdev_ops; 8616 8617 if (ops->ndo_change_mtu) 8618 return ops->ndo_change_mtu(dev, new_mtu); 8619 8620 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8621 WRITE_ONCE(dev->mtu, new_mtu); 8622 return 0; 8623 } 8624 EXPORT_SYMBOL(__dev_set_mtu); 8625 8626 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8627 struct netlink_ext_ack *extack) 8628 { 8629 /* MTU must be positive, and in range */ 8630 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8631 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8632 return -EINVAL; 8633 } 8634 8635 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8636 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8637 return -EINVAL; 8638 } 8639 return 0; 8640 } 8641 8642 /** 8643 * dev_set_mtu_ext - Change maximum transfer unit 8644 * @dev: device 8645 * @new_mtu: new transfer unit 8646 * @extack: netlink extended ack 8647 * 8648 * Change the maximum transfer size of the network device. 8649 */ 8650 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8651 struct netlink_ext_ack *extack) 8652 { 8653 int err, orig_mtu; 8654 8655 if (new_mtu == dev->mtu) 8656 return 0; 8657 8658 err = dev_validate_mtu(dev, new_mtu, extack); 8659 if (err) 8660 return err; 8661 8662 if (!netif_device_present(dev)) 8663 return -ENODEV; 8664 8665 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8666 err = notifier_to_errno(err); 8667 if (err) 8668 return err; 8669 8670 orig_mtu = dev->mtu; 8671 err = __dev_set_mtu(dev, new_mtu); 8672 8673 if (!err) { 8674 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8675 orig_mtu); 8676 err = notifier_to_errno(err); 8677 if (err) { 8678 /* setting mtu back and notifying everyone again, 8679 * so that they have a chance to revert changes. 8680 */ 8681 __dev_set_mtu(dev, orig_mtu); 8682 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8683 new_mtu); 8684 } 8685 } 8686 return err; 8687 } 8688 8689 int dev_set_mtu(struct net_device *dev, int new_mtu) 8690 { 8691 struct netlink_ext_ack extack; 8692 int err; 8693 8694 memset(&extack, 0, sizeof(extack)); 8695 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8696 if (err && extack._msg) 8697 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8698 return err; 8699 } 8700 EXPORT_SYMBOL(dev_set_mtu); 8701 8702 /** 8703 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8704 * @dev: device 8705 * @new_len: new tx queue length 8706 */ 8707 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8708 { 8709 unsigned int orig_len = dev->tx_queue_len; 8710 int res; 8711 8712 if (new_len != (unsigned int)new_len) 8713 return -ERANGE; 8714 8715 if (new_len != orig_len) { 8716 dev->tx_queue_len = new_len; 8717 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8718 res = notifier_to_errno(res); 8719 if (res) 8720 goto err_rollback; 8721 res = dev_qdisc_change_tx_queue_len(dev); 8722 if (res) 8723 goto err_rollback; 8724 } 8725 8726 return 0; 8727 8728 err_rollback: 8729 netdev_err(dev, "refused to change device tx_queue_len\n"); 8730 dev->tx_queue_len = orig_len; 8731 return res; 8732 } 8733 8734 /** 8735 * dev_set_group - Change group this device belongs to 8736 * @dev: device 8737 * @new_group: group this device should belong to 8738 */ 8739 void dev_set_group(struct net_device *dev, int new_group) 8740 { 8741 dev->group = new_group; 8742 } 8743 8744 /** 8745 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8746 * @dev: device 8747 * @addr: new address 8748 * @extack: netlink extended ack 8749 */ 8750 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8751 struct netlink_ext_ack *extack) 8752 { 8753 struct netdev_notifier_pre_changeaddr_info info = { 8754 .info.dev = dev, 8755 .info.extack = extack, 8756 .dev_addr = addr, 8757 }; 8758 int rc; 8759 8760 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8761 return notifier_to_errno(rc); 8762 } 8763 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8764 8765 /** 8766 * dev_set_mac_address - Change Media Access Control Address 8767 * @dev: device 8768 * @sa: new address 8769 * @extack: netlink extended ack 8770 * 8771 * Change the hardware (MAC) address of the device 8772 */ 8773 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8774 struct netlink_ext_ack *extack) 8775 { 8776 const struct net_device_ops *ops = dev->netdev_ops; 8777 int err; 8778 8779 if (!ops->ndo_set_mac_address) 8780 return -EOPNOTSUPP; 8781 if (sa->sa_family != dev->type) 8782 return -EINVAL; 8783 if (!netif_device_present(dev)) 8784 return -ENODEV; 8785 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8786 if (err) 8787 return err; 8788 err = ops->ndo_set_mac_address(dev, sa); 8789 if (err) 8790 return err; 8791 dev->addr_assign_type = NET_ADDR_SET; 8792 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8793 add_device_randomness(dev->dev_addr, dev->addr_len); 8794 return 0; 8795 } 8796 EXPORT_SYMBOL(dev_set_mac_address); 8797 8798 static DECLARE_RWSEM(dev_addr_sem); 8799 8800 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8801 struct netlink_ext_ack *extack) 8802 { 8803 int ret; 8804 8805 down_write(&dev_addr_sem); 8806 ret = dev_set_mac_address(dev, sa, extack); 8807 up_write(&dev_addr_sem); 8808 return ret; 8809 } 8810 EXPORT_SYMBOL(dev_set_mac_address_user); 8811 8812 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8813 { 8814 size_t size = sizeof(sa->sa_data_min); 8815 struct net_device *dev; 8816 int ret = 0; 8817 8818 down_read(&dev_addr_sem); 8819 rcu_read_lock(); 8820 8821 dev = dev_get_by_name_rcu(net, dev_name); 8822 if (!dev) { 8823 ret = -ENODEV; 8824 goto unlock; 8825 } 8826 if (!dev->addr_len) 8827 memset(sa->sa_data, 0, size); 8828 else 8829 memcpy(sa->sa_data, dev->dev_addr, 8830 min_t(size_t, size, dev->addr_len)); 8831 sa->sa_family = dev->type; 8832 8833 unlock: 8834 rcu_read_unlock(); 8835 up_read(&dev_addr_sem); 8836 return ret; 8837 } 8838 EXPORT_SYMBOL(dev_get_mac_address); 8839 8840 /** 8841 * dev_change_carrier - Change device carrier 8842 * @dev: device 8843 * @new_carrier: new value 8844 * 8845 * Change device carrier 8846 */ 8847 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8848 { 8849 const struct net_device_ops *ops = dev->netdev_ops; 8850 8851 if (!ops->ndo_change_carrier) 8852 return -EOPNOTSUPP; 8853 if (!netif_device_present(dev)) 8854 return -ENODEV; 8855 return ops->ndo_change_carrier(dev, new_carrier); 8856 } 8857 8858 /** 8859 * dev_get_phys_port_id - Get device physical port ID 8860 * @dev: device 8861 * @ppid: port ID 8862 * 8863 * Get device physical port ID 8864 */ 8865 int dev_get_phys_port_id(struct net_device *dev, 8866 struct netdev_phys_item_id *ppid) 8867 { 8868 const struct net_device_ops *ops = dev->netdev_ops; 8869 8870 if (!ops->ndo_get_phys_port_id) 8871 return -EOPNOTSUPP; 8872 return ops->ndo_get_phys_port_id(dev, ppid); 8873 } 8874 8875 /** 8876 * dev_get_phys_port_name - Get device physical port name 8877 * @dev: device 8878 * @name: port name 8879 * @len: limit of bytes to copy to name 8880 * 8881 * Get device physical port name 8882 */ 8883 int dev_get_phys_port_name(struct net_device *dev, 8884 char *name, size_t len) 8885 { 8886 const struct net_device_ops *ops = dev->netdev_ops; 8887 int err; 8888 8889 if (ops->ndo_get_phys_port_name) { 8890 err = ops->ndo_get_phys_port_name(dev, name, len); 8891 if (err != -EOPNOTSUPP) 8892 return err; 8893 } 8894 return devlink_compat_phys_port_name_get(dev, name, len); 8895 } 8896 8897 /** 8898 * dev_get_port_parent_id - Get the device's port parent identifier 8899 * @dev: network device 8900 * @ppid: pointer to a storage for the port's parent identifier 8901 * @recurse: allow/disallow recursion to lower devices 8902 * 8903 * Get the devices's port parent identifier 8904 */ 8905 int dev_get_port_parent_id(struct net_device *dev, 8906 struct netdev_phys_item_id *ppid, 8907 bool recurse) 8908 { 8909 const struct net_device_ops *ops = dev->netdev_ops; 8910 struct netdev_phys_item_id first = { }; 8911 struct net_device *lower_dev; 8912 struct list_head *iter; 8913 int err; 8914 8915 if (ops->ndo_get_port_parent_id) { 8916 err = ops->ndo_get_port_parent_id(dev, ppid); 8917 if (err != -EOPNOTSUPP) 8918 return err; 8919 } 8920 8921 err = devlink_compat_switch_id_get(dev, ppid); 8922 if (!recurse || err != -EOPNOTSUPP) 8923 return err; 8924 8925 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8926 err = dev_get_port_parent_id(lower_dev, ppid, true); 8927 if (err) 8928 break; 8929 if (!first.id_len) 8930 first = *ppid; 8931 else if (memcmp(&first, ppid, sizeof(*ppid))) 8932 return -EOPNOTSUPP; 8933 } 8934 8935 return err; 8936 } 8937 EXPORT_SYMBOL(dev_get_port_parent_id); 8938 8939 /** 8940 * netdev_port_same_parent_id - Indicate if two network devices have 8941 * the same port parent identifier 8942 * @a: first network device 8943 * @b: second network device 8944 */ 8945 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8946 { 8947 struct netdev_phys_item_id a_id = { }; 8948 struct netdev_phys_item_id b_id = { }; 8949 8950 if (dev_get_port_parent_id(a, &a_id, true) || 8951 dev_get_port_parent_id(b, &b_id, true)) 8952 return false; 8953 8954 return netdev_phys_item_id_same(&a_id, &b_id); 8955 } 8956 EXPORT_SYMBOL(netdev_port_same_parent_id); 8957 8958 /** 8959 * dev_change_proto_down - set carrier according to proto_down. 8960 * 8961 * @dev: device 8962 * @proto_down: new value 8963 */ 8964 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8965 { 8966 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8967 return -EOPNOTSUPP; 8968 if (!netif_device_present(dev)) 8969 return -ENODEV; 8970 if (proto_down) 8971 netif_carrier_off(dev); 8972 else 8973 netif_carrier_on(dev); 8974 dev->proto_down = proto_down; 8975 return 0; 8976 } 8977 8978 /** 8979 * dev_change_proto_down_reason - proto down reason 8980 * 8981 * @dev: device 8982 * @mask: proto down mask 8983 * @value: proto down value 8984 */ 8985 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8986 u32 value) 8987 { 8988 int b; 8989 8990 if (!mask) { 8991 dev->proto_down_reason = value; 8992 } else { 8993 for_each_set_bit(b, &mask, 32) { 8994 if (value & (1 << b)) 8995 dev->proto_down_reason |= BIT(b); 8996 else 8997 dev->proto_down_reason &= ~BIT(b); 8998 } 8999 } 9000 } 9001 9002 struct bpf_xdp_link { 9003 struct bpf_link link; 9004 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9005 int flags; 9006 }; 9007 9008 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9009 { 9010 if (flags & XDP_FLAGS_HW_MODE) 9011 return XDP_MODE_HW; 9012 if (flags & XDP_FLAGS_DRV_MODE) 9013 return XDP_MODE_DRV; 9014 if (flags & XDP_FLAGS_SKB_MODE) 9015 return XDP_MODE_SKB; 9016 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9017 } 9018 9019 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9020 { 9021 switch (mode) { 9022 case XDP_MODE_SKB: 9023 return generic_xdp_install; 9024 case XDP_MODE_DRV: 9025 case XDP_MODE_HW: 9026 return dev->netdev_ops->ndo_bpf; 9027 default: 9028 return NULL; 9029 } 9030 } 9031 9032 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9033 enum bpf_xdp_mode mode) 9034 { 9035 return dev->xdp_state[mode].link; 9036 } 9037 9038 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9039 enum bpf_xdp_mode mode) 9040 { 9041 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9042 9043 if (link) 9044 return link->link.prog; 9045 return dev->xdp_state[mode].prog; 9046 } 9047 9048 u8 dev_xdp_prog_count(struct net_device *dev) 9049 { 9050 u8 count = 0; 9051 int i; 9052 9053 for (i = 0; i < __MAX_XDP_MODE; i++) 9054 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9055 count++; 9056 return count; 9057 } 9058 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9059 9060 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9061 { 9062 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9063 9064 return prog ? prog->aux->id : 0; 9065 } 9066 9067 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9068 struct bpf_xdp_link *link) 9069 { 9070 dev->xdp_state[mode].link = link; 9071 dev->xdp_state[mode].prog = NULL; 9072 } 9073 9074 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9075 struct bpf_prog *prog) 9076 { 9077 dev->xdp_state[mode].link = NULL; 9078 dev->xdp_state[mode].prog = prog; 9079 } 9080 9081 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9082 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9083 u32 flags, struct bpf_prog *prog) 9084 { 9085 struct netdev_bpf xdp; 9086 int err; 9087 9088 memset(&xdp, 0, sizeof(xdp)); 9089 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9090 xdp.extack = extack; 9091 xdp.flags = flags; 9092 xdp.prog = prog; 9093 9094 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9095 * "moved" into driver), so they don't increment it on their own, but 9096 * they do decrement refcnt when program is detached or replaced. 9097 * Given net_device also owns link/prog, we need to bump refcnt here 9098 * to prevent drivers from underflowing it. 9099 */ 9100 if (prog) 9101 bpf_prog_inc(prog); 9102 err = bpf_op(dev, &xdp); 9103 if (err) { 9104 if (prog) 9105 bpf_prog_put(prog); 9106 return err; 9107 } 9108 9109 if (mode != XDP_MODE_HW) 9110 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9111 9112 return 0; 9113 } 9114 9115 static void dev_xdp_uninstall(struct net_device *dev) 9116 { 9117 struct bpf_xdp_link *link; 9118 struct bpf_prog *prog; 9119 enum bpf_xdp_mode mode; 9120 bpf_op_t bpf_op; 9121 9122 ASSERT_RTNL(); 9123 9124 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9125 prog = dev_xdp_prog(dev, mode); 9126 if (!prog) 9127 continue; 9128 9129 bpf_op = dev_xdp_bpf_op(dev, mode); 9130 if (!bpf_op) 9131 continue; 9132 9133 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9134 9135 /* auto-detach link from net device */ 9136 link = dev_xdp_link(dev, mode); 9137 if (link) 9138 link->dev = NULL; 9139 else 9140 bpf_prog_put(prog); 9141 9142 dev_xdp_set_link(dev, mode, NULL); 9143 } 9144 } 9145 9146 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9147 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9148 struct bpf_prog *old_prog, u32 flags) 9149 { 9150 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9151 struct bpf_prog *cur_prog; 9152 struct net_device *upper; 9153 struct list_head *iter; 9154 enum bpf_xdp_mode mode; 9155 bpf_op_t bpf_op; 9156 int err; 9157 9158 ASSERT_RTNL(); 9159 9160 /* either link or prog attachment, never both */ 9161 if (link && (new_prog || old_prog)) 9162 return -EINVAL; 9163 /* link supports only XDP mode flags */ 9164 if (link && (flags & ~XDP_FLAGS_MODES)) { 9165 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9166 return -EINVAL; 9167 } 9168 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9169 if (num_modes > 1) { 9170 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9171 return -EINVAL; 9172 } 9173 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9174 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9175 NL_SET_ERR_MSG(extack, 9176 "More than one program loaded, unset mode is ambiguous"); 9177 return -EINVAL; 9178 } 9179 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9180 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9181 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9182 return -EINVAL; 9183 } 9184 9185 mode = dev_xdp_mode(dev, flags); 9186 /* can't replace attached link */ 9187 if (dev_xdp_link(dev, mode)) { 9188 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9189 return -EBUSY; 9190 } 9191 9192 /* don't allow if an upper device already has a program */ 9193 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9194 if (dev_xdp_prog_count(upper) > 0) { 9195 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9196 return -EEXIST; 9197 } 9198 } 9199 9200 cur_prog = dev_xdp_prog(dev, mode); 9201 /* can't replace attached prog with link */ 9202 if (link && cur_prog) { 9203 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9204 return -EBUSY; 9205 } 9206 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9207 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9208 return -EEXIST; 9209 } 9210 9211 /* put effective new program into new_prog */ 9212 if (link) 9213 new_prog = link->link.prog; 9214 9215 if (new_prog) { 9216 bool offload = mode == XDP_MODE_HW; 9217 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9218 ? XDP_MODE_DRV : XDP_MODE_SKB; 9219 9220 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9221 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9222 return -EBUSY; 9223 } 9224 if (!offload && dev_xdp_prog(dev, other_mode)) { 9225 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9226 return -EEXIST; 9227 } 9228 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9229 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9230 return -EINVAL; 9231 } 9232 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9233 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9234 return -EINVAL; 9235 } 9236 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9237 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9238 return -EINVAL; 9239 } 9240 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9241 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9242 return -EINVAL; 9243 } 9244 } 9245 9246 /* don't call drivers if the effective program didn't change */ 9247 if (new_prog != cur_prog) { 9248 bpf_op = dev_xdp_bpf_op(dev, mode); 9249 if (!bpf_op) { 9250 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9251 return -EOPNOTSUPP; 9252 } 9253 9254 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9255 if (err) 9256 return err; 9257 } 9258 9259 if (link) 9260 dev_xdp_set_link(dev, mode, link); 9261 else 9262 dev_xdp_set_prog(dev, mode, new_prog); 9263 if (cur_prog) 9264 bpf_prog_put(cur_prog); 9265 9266 return 0; 9267 } 9268 9269 static int dev_xdp_attach_link(struct net_device *dev, 9270 struct netlink_ext_ack *extack, 9271 struct bpf_xdp_link *link) 9272 { 9273 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9274 } 9275 9276 static int dev_xdp_detach_link(struct net_device *dev, 9277 struct netlink_ext_ack *extack, 9278 struct bpf_xdp_link *link) 9279 { 9280 enum bpf_xdp_mode mode; 9281 bpf_op_t bpf_op; 9282 9283 ASSERT_RTNL(); 9284 9285 mode = dev_xdp_mode(dev, link->flags); 9286 if (dev_xdp_link(dev, mode) != link) 9287 return -EINVAL; 9288 9289 bpf_op = dev_xdp_bpf_op(dev, mode); 9290 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9291 dev_xdp_set_link(dev, mode, NULL); 9292 return 0; 9293 } 9294 9295 static void bpf_xdp_link_release(struct bpf_link *link) 9296 { 9297 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9298 9299 rtnl_lock(); 9300 9301 /* if racing with net_device's tear down, xdp_link->dev might be 9302 * already NULL, in which case link was already auto-detached 9303 */ 9304 if (xdp_link->dev) { 9305 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9306 xdp_link->dev = NULL; 9307 } 9308 9309 rtnl_unlock(); 9310 } 9311 9312 static int bpf_xdp_link_detach(struct bpf_link *link) 9313 { 9314 bpf_xdp_link_release(link); 9315 return 0; 9316 } 9317 9318 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9319 { 9320 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9321 9322 kfree(xdp_link); 9323 } 9324 9325 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9326 struct seq_file *seq) 9327 { 9328 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9329 u32 ifindex = 0; 9330 9331 rtnl_lock(); 9332 if (xdp_link->dev) 9333 ifindex = xdp_link->dev->ifindex; 9334 rtnl_unlock(); 9335 9336 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9337 } 9338 9339 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9340 struct bpf_link_info *info) 9341 { 9342 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9343 u32 ifindex = 0; 9344 9345 rtnl_lock(); 9346 if (xdp_link->dev) 9347 ifindex = xdp_link->dev->ifindex; 9348 rtnl_unlock(); 9349 9350 info->xdp.ifindex = ifindex; 9351 return 0; 9352 } 9353 9354 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9355 struct bpf_prog *old_prog) 9356 { 9357 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9358 enum bpf_xdp_mode mode; 9359 bpf_op_t bpf_op; 9360 int err = 0; 9361 9362 rtnl_lock(); 9363 9364 /* link might have been auto-released already, so fail */ 9365 if (!xdp_link->dev) { 9366 err = -ENOLINK; 9367 goto out_unlock; 9368 } 9369 9370 if (old_prog && link->prog != old_prog) { 9371 err = -EPERM; 9372 goto out_unlock; 9373 } 9374 old_prog = link->prog; 9375 if (old_prog->type != new_prog->type || 9376 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9377 err = -EINVAL; 9378 goto out_unlock; 9379 } 9380 9381 if (old_prog == new_prog) { 9382 /* no-op, don't disturb drivers */ 9383 bpf_prog_put(new_prog); 9384 goto out_unlock; 9385 } 9386 9387 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9388 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9389 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9390 xdp_link->flags, new_prog); 9391 if (err) 9392 goto out_unlock; 9393 9394 old_prog = xchg(&link->prog, new_prog); 9395 bpf_prog_put(old_prog); 9396 9397 out_unlock: 9398 rtnl_unlock(); 9399 return err; 9400 } 9401 9402 static const struct bpf_link_ops bpf_xdp_link_lops = { 9403 .release = bpf_xdp_link_release, 9404 .dealloc = bpf_xdp_link_dealloc, 9405 .detach = bpf_xdp_link_detach, 9406 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9407 .fill_link_info = bpf_xdp_link_fill_link_info, 9408 .update_prog = bpf_xdp_link_update, 9409 }; 9410 9411 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9412 { 9413 struct net *net = current->nsproxy->net_ns; 9414 struct bpf_link_primer link_primer; 9415 struct bpf_xdp_link *link; 9416 struct net_device *dev; 9417 int err, fd; 9418 9419 rtnl_lock(); 9420 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9421 if (!dev) { 9422 rtnl_unlock(); 9423 return -EINVAL; 9424 } 9425 9426 link = kzalloc(sizeof(*link), GFP_USER); 9427 if (!link) { 9428 err = -ENOMEM; 9429 goto unlock; 9430 } 9431 9432 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9433 link->dev = dev; 9434 link->flags = attr->link_create.flags; 9435 9436 err = bpf_link_prime(&link->link, &link_primer); 9437 if (err) { 9438 kfree(link); 9439 goto unlock; 9440 } 9441 9442 err = dev_xdp_attach_link(dev, NULL, link); 9443 rtnl_unlock(); 9444 9445 if (err) { 9446 link->dev = NULL; 9447 bpf_link_cleanup(&link_primer); 9448 goto out_put_dev; 9449 } 9450 9451 fd = bpf_link_settle(&link_primer); 9452 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9453 dev_put(dev); 9454 return fd; 9455 9456 unlock: 9457 rtnl_unlock(); 9458 9459 out_put_dev: 9460 dev_put(dev); 9461 return err; 9462 } 9463 9464 /** 9465 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9466 * @dev: device 9467 * @extack: netlink extended ack 9468 * @fd: new program fd or negative value to clear 9469 * @expected_fd: old program fd that userspace expects to replace or clear 9470 * @flags: xdp-related flags 9471 * 9472 * Set or clear a bpf program for a device 9473 */ 9474 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9475 int fd, int expected_fd, u32 flags) 9476 { 9477 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9478 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9479 int err; 9480 9481 ASSERT_RTNL(); 9482 9483 if (fd >= 0) { 9484 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9485 mode != XDP_MODE_SKB); 9486 if (IS_ERR(new_prog)) 9487 return PTR_ERR(new_prog); 9488 } 9489 9490 if (expected_fd >= 0) { 9491 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9492 mode != XDP_MODE_SKB); 9493 if (IS_ERR(old_prog)) { 9494 err = PTR_ERR(old_prog); 9495 old_prog = NULL; 9496 goto err_out; 9497 } 9498 } 9499 9500 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9501 9502 err_out: 9503 if (err && new_prog) 9504 bpf_prog_put(new_prog); 9505 if (old_prog) 9506 bpf_prog_put(old_prog); 9507 return err; 9508 } 9509 9510 /** 9511 * dev_new_index - allocate an ifindex 9512 * @net: the applicable net namespace 9513 * 9514 * Returns a suitable unique value for a new device interface 9515 * number. The caller must hold the rtnl semaphore or the 9516 * dev_base_lock to be sure it remains unique. 9517 */ 9518 static int dev_new_index(struct net *net) 9519 { 9520 int ifindex = net->ifindex; 9521 9522 for (;;) { 9523 if (++ifindex <= 0) 9524 ifindex = 1; 9525 if (!__dev_get_by_index(net, ifindex)) 9526 return net->ifindex = ifindex; 9527 } 9528 } 9529 9530 /* Delayed registration/unregisteration */ 9531 LIST_HEAD(net_todo_list); 9532 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9533 9534 static void net_set_todo(struct net_device *dev) 9535 { 9536 list_add_tail(&dev->todo_list, &net_todo_list); 9537 atomic_inc(&dev_net(dev)->dev_unreg_count); 9538 } 9539 9540 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9541 struct net_device *upper, netdev_features_t features) 9542 { 9543 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9544 netdev_features_t feature; 9545 int feature_bit; 9546 9547 for_each_netdev_feature(upper_disables, feature_bit) { 9548 feature = __NETIF_F_BIT(feature_bit); 9549 if (!(upper->wanted_features & feature) 9550 && (features & feature)) { 9551 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9552 &feature, upper->name); 9553 features &= ~feature; 9554 } 9555 } 9556 9557 return features; 9558 } 9559 9560 static void netdev_sync_lower_features(struct net_device *upper, 9561 struct net_device *lower, netdev_features_t features) 9562 { 9563 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9564 netdev_features_t feature; 9565 int feature_bit; 9566 9567 for_each_netdev_feature(upper_disables, feature_bit) { 9568 feature = __NETIF_F_BIT(feature_bit); 9569 if (!(features & feature) && (lower->features & feature)) { 9570 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9571 &feature, lower->name); 9572 lower->wanted_features &= ~feature; 9573 __netdev_update_features(lower); 9574 9575 if (unlikely(lower->features & feature)) 9576 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9577 &feature, lower->name); 9578 else 9579 netdev_features_change(lower); 9580 } 9581 } 9582 } 9583 9584 static netdev_features_t netdev_fix_features(struct net_device *dev, 9585 netdev_features_t features) 9586 { 9587 /* Fix illegal checksum combinations */ 9588 if ((features & NETIF_F_HW_CSUM) && 9589 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9590 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9591 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9592 } 9593 9594 /* TSO requires that SG is present as well. */ 9595 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9596 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9597 features &= ~NETIF_F_ALL_TSO; 9598 } 9599 9600 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9601 !(features & NETIF_F_IP_CSUM)) { 9602 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9603 features &= ~NETIF_F_TSO; 9604 features &= ~NETIF_F_TSO_ECN; 9605 } 9606 9607 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9608 !(features & NETIF_F_IPV6_CSUM)) { 9609 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9610 features &= ~NETIF_F_TSO6; 9611 } 9612 9613 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9614 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9615 features &= ~NETIF_F_TSO_MANGLEID; 9616 9617 /* TSO ECN requires that TSO is present as well. */ 9618 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9619 features &= ~NETIF_F_TSO_ECN; 9620 9621 /* Software GSO depends on SG. */ 9622 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9623 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9624 features &= ~NETIF_F_GSO; 9625 } 9626 9627 /* GSO partial features require GSO partial be set */ 9628 if ((features & dev->gso_partial_features) && 9629 !(features & NETIF_F_GSO_PARTIAL)) { 9630 netdev_dbg(dev, 9631 "Dropping partially supported GSO features since no GSO partial.\n"); 9632 features &= ~dev->gso_partial_features; 9633 } 9634 9635 if (!(features & NETIF_F_RXCSUM)) { 9636 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9637 * successfully merged by hardware must also have the 9638 * checksum verified by hardware. If the user does not 9639 * want to enable RXCSUM, logically, we should disable GRO_HW. 9640 */ 9641 if (features & NETIF_F_GRO_HW) { 9642 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9643 features &= ~NETIF_F_GRO_HW; 9644 } 9645 } 9646 9647 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9648 if (features & NETIF_F_RXFCS) { 9649 if (features & NETIF_F_LRO) { 9650 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9651 features &= ~NETIF_F_LRO; 9652 } 9653 9654 if (features & NETIF_F_GRO_HW) { 9655 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9656 features &= ~NETIF_F_GRO_HW; 9657 } 9658 } 9659 9660 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9661 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9662 features &= ~NETIF_F_LRO; 9663 } 9664 9665 if (features & NETIF_F_HW_TLS_TX) { 9666 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9667 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9668 bool hw_csum = features & NETIF_F_HW_CSUM; 9669 9670 if (!ip_csum && !hw_csum) { 9671 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9672 features &= ~NETIF_F_HW_TLS_TX; 9673 } 9674 } 9675 9676 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9677 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9678 features &= ~NETIF_F_HW_TLS_RX; 9679 } 9680 9681 return features; 9682 } 9683 9684 int __netdev_update_features(struct net_device *dev) 9685 { 9686 struct net_device *upper, *lower; 9687 netdev_features_t features; 9688 struct list_head *iter; 9689 int err = -1; 9690 9691 ASSERT_RTNL(); 9692 9693 features = netdev_get_wanted_features(dev); 9694 9695 if (dev->netdev_ops->ndo_fix_features) 9696 features = dev->netdev_ops->ndo_fix_features(dev, features); 9697 9698 /* driver might be less strict about feature dependencies */ 9699 features = netdev_fix_features(dev, features); 9700 9701 /* some features can't be enabled if they're off on an upper device */ 9702 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9703 features = netdev_sync_upper_features(dev, upper, features); 9704 9705 if (dev->features == features) 9706 goto sync_lower; 9707 9708 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9709 &dev->features, &features); 9710 9711 if (dev->netdev_ops->ndo_set_features) 9712 err = dev->netdev_ops->ndo_set_features(dev, features); 9713 else 9714 err = 0; 9715 9716 if (unlikely(err < 0)) { 9717 netdev_err(dev, 9718 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9719 err, &features, &dev->features); 9720 /* return non-0 since some features might have changed and 9721 * it's better to fire a spurious notification than miss it 9722 */ 9723 return -1; 9724 } 9725 9726 sync_lower: 9727 /* some features must be disabled on lower devices when disabled 9728 * on an upper device (think: bonding master or bridge) 9729 */ 9730 netdev_for_each_lower_dev(dev, lower, iter) 9731 netdev_sync_lower_features(dev, lower, features); 9732 9733 if (!err) { 9734 netdev_features_t diff = features ^ dev->features; 9735 9736 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9737 /* udp_tunnel_{get,drop}_rx_info both need 9738 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9739 * device, or they won't do anything. 9740 * Thus we need to update dev->features 9741 * *before* calling udp_tunnel_get_rx_info, 9742 * but *after* calling udp_tunnel_drop_rx_info. 9743 */ 9744 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9745 dev->features = features; 9746 udp_tunnel_get_rx_info(dev); 9747 } else { 9748 udp_tunnel_drop_rx_info(dev); 9749 } 9750 } 9751 9752 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9753 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9754 dev->features = features; 9755 err |= vlan_get_rx_ctag_filter_info(dev); 9756 } else { 9757 vlan_drop_rx_ctag_filter_info(dev); 9758 } 9759 } 9760 9761 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9762 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9763 dev->features = features; 9764 err |= vlan_get_rx_stag_filter_info(dev); 9765 } else { 9766 vlan_drop_rx_stag_filter_info(dev); 9767 } 9768 } 9769 9770 dev->features = features; 9771 } 9772 9773 return err < 0 ? 0 : 1; 9774 } 9775 9776 /** 9777 * netdev_update_features - recalculate device features 9778 * @dev: the device to check 9779 * 9780 * Recalculate dev->features set and send notifications if it 9781 * has changed. Should be called after driver or hardware dependent 9782 * conditions might have changed that influence the features. 9783 */ 9784 void netdev_update_features(struct net_device *dev) 9785 { 9786 if (__netdev_update_features(dev)) 9787 netdev_features_change(dev); 9788 } 9789 EXPORT_SYMBOL(netdev_update_features); 9790 9791 /** 9792 * netdev_change_features - recalculate device features 9793 * @dev: the device to check 9794 * 9795 * Recalculate dev->features set and send notifications even 9796 * if they have not changed. Should be called instead of 9797 * netdev_update_features() if also dev->vlan_features might 9798 * have changed to allow the changes to be propagated to stacked 9799 * VLAN devices. 9800 */ 9801 void netdev_change_features(struct net_device *dev) 9802 { 9803 __netdev_update_features(dev); 9804 netdev_features_change(dev); 9805 } 9806 EXPORT_SYMBOL(netdev_change_features); 9807 9808 /** 9809 * netif_stacked_transfer_operstate - transfer operstate 9810 * @rootdev: the root or lower level device to transfer state from 9811 * @dev: the device to transfer operstate to 9812 * 9813 * Transfer operational state from root to device. This is normally 9814 * called when a stacking relationship exists between the root 9815 * device and the device(a leaf device). 9816 */ 9817 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9818 struct net_device *dev) 9819 { 9820 if (rootdev->operstate == IF_OPER_DORMANT) 9821 netif_dormant_on(dev); 9822 else 9823 netif_dormant_off(dev); 9824 9825 if (rootdev->operstate == IF_OPER_TESTING) 9826 netif_testing_on(dev); 9827 else 9828 netif_testing_off(dev); 9829 9830 if (netif_carrier_ok(rootdev)) 9831 netif_carrier_on(dev); 9832 else 9833 netif_carrier_off(dev); 9834 } 9835 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9836 9837 static int netif_alloc_rx_queues(struct net_device *dev) 9838 { 9839 unsigned int i, count = dev->num_rx_queues; 9840 struct netdev_rx_queue *rx; 9841 size_t sz = count * sizeof(*rx); 9842 int err = 0; 9843 9844 BUG_ON(count < 1); 9845 9846 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9847 if (!rx) 9848 return -ENOMEM; 9849 9850 dev->_rx = rx; 9851 9852 for (i = 0; i < count; i++) { 9853 rx[i].dev = dev; 9854 9855 /* XDP RX-queue setup */ 9856 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9857 if (err < 0) 9858 goto err_rxq_info; 9859 } 9860 return 0; 9861 9862 err_rxq_info: 9863 /* Rollback successful reg's and free other resources */ 9864 while (i--) 9865 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9866 kvfree(dev->_rx); 9867 dev->_rx = NULL; 9868 return err; 9869 } 9870 9871 static void netif_free_rx_queues(struct net_device *dev) 9872 { 9873 unsigned int i, count = dev->num_rx_queues; 9874 9875 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9876 if (!dev->_rx) 9877 return; 9878 9879 for (i = 0; i < count; i++) 9880 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9881 9882 kvfree(dev->_rx); 9883 } 9884 9885 static void netdev_init_one_queue(struct net_device *dev, 9886 struct netdev_queue *queue, void *_unused) 9887 { 9888 /* Initialize queue lock */ 9889 spin_lock_init(&queue->_xmit_lock); 9890 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9891 queue->xmit_lock_owner = -1; 9892 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9893 queue->dev = dev; 9894 #ifdef CONFIG_BQL 9895 dql_init(&queue->dql, HZ); 9896 #endif 9897 } 9898 9899 static void netif_free_tx_queues(struct net_device *dev) 9900 { 9901 kvfree(dev->_tx); 9902 } 9903 9904 static int netif_alloc_netdev_queues(struct net_device *dev) 9905 { 9906 unsigned int count = dev->num_tx_queues; 9907 struct netdev_queue *tx; 9908 size_t sz = count * sizeof(*tx); 9909 9910 if (count < 1 || count > 0xffff) 9911 return -EINVAL; 9912 9913 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9914 if (!tx) 9915 return -ENOMEM; 9916 9917 dev->_tx = tx; 9918 9919 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9920 spin_lock_init(&dev->tx_global_lock); 9921 9922 return 0; 9923 } 9924 9925 void netif_tx_stop_all_queues(struct net_device *dev) 9926 { 9927 unsigned int i; 9928 9929 for (i = 0; i < dev->num_tx_queues; i++) { 9930 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9931 9932 netif_tx_stop_queue(txq); 9933 } 9934 } 9935 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9936 9937 /** 9938 * register_netdevice() - register a network device 9939 * @dev: device to register 9940 * 9941 * Take a prepared network device structure and make it externally accessible. 9942 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9943 * Callers must hold the rtnl lock - you may want register_netdev() 9944 * instead of this. 9945 */ 9946 int register_netdevice(struct net_device *dev) 9947 { 9948 int ret; 9949 struct net *net = dev_net(dev); 9950 9951 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9952 NETDEV_FEATURE_COUNT); 9953 BUG_ON(dev_boot_phase); 9954 ASSERT_RTNL(); 9955 9956 might_sleep(); 9957 9958 /* When net_device's are persistent, this will be fatal. */ 9959 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9960 BUG_ON(!net); 9961 9962 ret = ethtool_check_ops(dev->ethtool_ops); 9963 if (ret) 9964 return ret; 9965 9966 spin_lock_init(&dev->addr_list_lock); 9967 netdev_set_addr_lockdep_class(dev); 9968 9969 ret = dev_get_valid_name(net, dev, dev->name); 9970 if (ret < 0) 9971 goto out; 9972 9973 ret = -ENOMEM; 9974 dev->name_node = netdev_name_node_head_alloc(dev); 9975 if (!dev->name_node) 9976 goto out; 9977 9978 /* Init, if this function is available */ 9979 if (dev->netdev_ops->ndo_init) { 9980 ret = dev->netdev_ops->ndo_init(dev); 9981 if (ret) { 9982 if (ret > 0) 9983 ret = -EIO; 9984 goto err_free_name; 9985 } 9986 } 9987 9988 if (((dev->hw_features | dev->features) & 9989 NETIF_F_HW_VLAN_CTAG_FILTER) && 9990 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9991 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9992 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9993 ret = -EINVAL; 9994 goto err_uninit; 9995 } 9996 9997 ret = -EBUSY; 9998 if (!dev->ifindex) 9999 dev->ifindex = dev_new_index(net); 10000 else if (__dev_get_by_index(net, dev->ifindex)) 10001 goto err_uninit; 10002 10003 /* Transfer changeable features to wanted_features and enable 10004 * software offloads (GSO and GRO). 10005 */ 10006 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10007 dev->features |= NETIF_F_SOFT_FEATURES; 10008 10009 if (dev->udp_tunnel_nic_info) { 10010 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10011 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10012 } 10013 10014 dev->wanted_features = dev->features & dev->hw_features; 10015 10016 if (!(dev->flags & IFF_LOOPBACK)) 10017 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10018 10019 /* If IPv4 TCP segmentation offload is supported we should also 10020 * allow the device to enable segmenting the frame with the option 10021 * of ignoring a static IP ID value. This doesn't enable the 10022 * feature itself but allows the user to enable it later. 10023 */ 10024 if (dev->hw_features & NETIF_F_TSO) 10025 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10026 if (dev->vlan_features & NETIF_F_TSO) 10027 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10028 if (dev->mpls_features & NETIF_F_TSO) 10029 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10030 if (dev->hw_enc_features & NETIF_F_TSO) 10031 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10032 10033 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10034 */ 10035 dev->vlan_features |= NETIF_F_HIGHDMA; 10036 10037 /* Make NETIF_F_SG inheritable to tunnel devices. 10038 */ 10039 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10040 10041 /* Make NETIF_F_SG inheritable to MPLS. 10042 */ 10043 dev->mpls_features |= NETIF_F_SG; 10044 10045 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10046 ret = notifier_to_errno(ret); 10047 if (ret) 10048 goto err_uninit; 10049 10050 ret = netdev_register_kobject(dev); 10051 write_lock(&dev_base_lock); 10052 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10053 write_unlock(&dev_base_lock); 10054 if (ret) 10055 goto err_uninit_notify; 10056 10057 __netdev_update_features(dev); 10058 10059 /* 10060 * Default initial state at registry is that the 10061 * device is present. 10062 */ 10063 10064 set_bit(__LINK_STATE_PRESENT, &dev->state); 10065 10066 linkwatch_init_dev(dev); 10067 10068 dev_init_scheduler(dev); 10069 10070 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10071 list_netdevice(dev); 10072 10073 add_device_randomness(dev->dev_addr, dev->addr_len); 10074 10075 /* If the device has permanent device address, driver should 10076 * set dev_addr and also addr_assign_type should be set to 10077 * NET_ADDR_PERM (default value). 10078 */ 10079 if (dev->addr_assign_type == NET_ADDR_PERM) 10080 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10081 10082 /* Notify protocols, that a new device appeared. */ 10083 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10084 ret = notifier_to_errno(ret); 10085 if (ret) { 10086 /* Expect explicit free_netdev() on failure */ 10087 dev->needs_free_netdev = false; 10088 unregister_netdevice_queue(dev, NULL); 10089 goto out; 10090 } 10091 /* 10092 * Prevent userspace races by waiting until the network 10093 * device is fully setup before sending notifications. 10094 */ 10095 if (!dev->rtnl_link_ops || 10096 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10097 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10098 10099 out: 10100 return ret; 10101 10102 err_uninit_notify: 10103 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10104 err_uninit: 10105 if (dev->netdev_ops->ndo_uninit) 10106 dev->netdev_ops->ndo_uninit(dev); 10107 if (dev->priv_destructor) 10108 dev->priv_destructor(dev); 10109 err_free_name: 10110 netdev_name_node_free(dev->name_node); 10111 goto out; 10112 } 10113 EXPORT_SYMBOL(register_netdevice); 10114 10115 /** 10116 * init_dummy_netdev - init a dummy network device for NAPI 10117 * @dev: device to init 10118 * 10119 * This takes a network device structure and initialize the minimum 10120 * amount of fields so it can be used to schedule NAPI polls without 10121 * registering a full blown interface. This is to be used by drivers 10122 * that need to tie several hardware interfaces to a single NAPI 10123 * poll scheduler due to HW limitations. 10124 */ 10125 int init_dummy_netdev(struct net_device *dev) 10126 { 10127 /* Clear everything. Note we don't initialize spinlocks 10128 * are they aren't supposed to be taken by any of the 10129 * NAPI code and this dummy netdev is supposed to be 10130 * only ever used for NAPI polls 10131 */ 10132 memset(dev, 0, sizeof(struct net_device)); 10133 10134 /* make sure we BUG if trying to hit standard 10135 * register/unregister code path 10136 */ 10137 dev->reg_state = NETREG_DUMMY; 10138 10139 /* NAPI wants this */ 10140 INIT_LIST_HEAD(&dev->napi_list); 10141 10142 /* a dummy interface is started by default */ 10143 set_bit(__LINK_STATE_PRESENT, &dev->state); 10144 set_bit(__LINK_STATE_START, &dev->state); 10145 10146 /* napi_busy_loop stats accounting wants this */ 10147 dev_net_set(dev, &init_net); 10148 10149 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10150 * because users of this 'device' dont need to change 10151 * its refcount. 10152 */ 10153 10154 return 0; 10155 } 10156 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10157 10158 10159 /** 10160 * register_netdev - register a network device 10161 * @dev: device to register 10162 * 10163 * Take a completed network device structure and add it to the kernel 10164 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10165 * chain. 0 is returned on success. A negative errno code is returned 10166 * on a failure to set up the device, or if the name is a duplicate. 10167 * 10168 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10169 * and expands the device name if you passed a format string to 10170 * alloc_netdev. 10171 */ 10172 int register_netdev(struct net_device *dev) 10173 { 10174 int err; 10175 10176 if (rtnl_lock_killable()) 10177 return -EINTR; 10178 err = register_netdevice(dev); 10179 rtnl_unlock(); 10180 return err; 10181 } 10182 EXPORT_SYMBOL(register_netdev); 10183 10184 int netdev_refcnt_read(const struct net_device *dev) 10185 { 10186 #ifdef CONFIG_PCPU_DEV_REFCNT 10187 int i, refcnt = 0; 10188 10189 for_each_possible_cpu(i) 10190 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10191 return refcnt; 10192 #else 10193 return refcount_read(&dev->dev_refcnt); 10194 #endif 10195 } 10196 EXPORT_SYMBOL(netdev_refcnt_read); 10197 10198 int netdev_unregister_timeout_secs __read_mostly = 10; 10199 10200 #define WAIT_REFS_MIN_MSECS 1 10201 #define WAIT_REFS_MAX_MSECS 250 10202 /** 10203 * netdev_wait_allrefs_any - wait until all references are gone. 10204 * @list: list of net_devices to wait on 10205 * 10206 * This is called when unregistering network devices. 10207 * 10208 * Any protocol or device that holds a reference should register 10209 * for netdevice notification, and cleanup and put back the 10210 * reference if they receive an UNREGISTER event. 10211 * We can get stuck here if buggy protocols don't correctly 10212 * call dev_put. 10213 */ 10214 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10215 { 10216 unsigned long rebroadcast_time, warning_time; 10217 struct net_device *dev; 10218 int wait = 0; 10219 10220 rebroadcast_time = warning_time = jiffies; 10221 10222 list_for_each_entry(dev, list, todo_list) 10223 if (netdev_refcnt_read(dev) == 1) 10224 return dev; 10225 10226 while (true) { 10227 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10228 rtnl_lock(); 10229 10230 /* Rebroadcast unregister notification */ 10231 list_for_each_entry(dev, list, todo_list) 10232 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10233 10234 __rtnl_unlock(); 10235 rcu_barrier(); 10236 rtnl_lock(); 10237 10238 list_for_each_entry(dev, list, todo_list) 10239 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10240 &dev->state)) { 10241 /* We must not have linkwatch events 10242 * pending on unregister. If this 10243 * happens, we simply run the queue 10244 * unscheduled, resulting in a noop 10245 * for this device. 10246 */ 10247 linkwatch_run_queue(); 10248 break; 10249 } 10250 10251 __rtnl_unlock(); 10252 10253 rebroadcast_time = jiffies; 10254 } 10255 10256 if (!wait) { 10257 rcu_barrier(); 10258 wait = WAIT_REFS_MIN_MSECS; 10259 } else { 10260 msleep(wait); 10261 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10262 } 10263 10264 list_for_each_entry(dev, list, todo_list) 10265 if (netdev_refcnt_read(dev) == 1) 10266 return dev; 10267 10268 if (time_after(jiffies, warning_time + 10269 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10270 list_for_each_entry(dev, list, todo_list) { 10271 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10272 dev->name, netdev_refcnt_read(dev)); 10273 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10274 } 10275 10276 warning_time = jiffies; 10277 } 10278 } 10279 } 10280 10281 /* The sequence is: 10282 * 10283 * rtnl_lock(); 10284 * ... 10285 * register_netdevice(x1); 10286 * register_netdevice(x2); 10287 * ... 10288 * unregister_netdevice(y1); 10289 * unregister_netdevice(y2); 10290 * ... 10291 * rtnl_unlock(); 10292 * free_netdev(y1); 10293 * free_netdev(y2); 10294 * 10295 * We are invoked by rtnl_unlock(). 10296 * This allows us to deal with problems: 10297 * 1) We can delete sysfs objects which invoke hotplug 10298 * without deadlocking with linkwatch via keventd. 10299 * 2) Since we run with the RTNL semaphore not held, we can sleep 10300 * safely in order to wait for the netdev refcnt to drop to zero. 10301 * 10302 * We must not return until all unregister events added during 10303 * the interval the lock was held have been completed. 10304 */ 10305 void netdev_run_todo(void) 10306 { 10307 struct net_device *dev, *tmp; 10308 struct list_head list; 10309 #ifdef CONFIG_LOCKDEP 10310 struct list_head unlink_list; 10311 10312 list_replace_init(&net_unlink_list, &unlink_list); 10313 10314 while (!list_empty(&unlink_list)) { 10315 struct net_device *dev = list_first_entry(&unlink_list, 10316 struct net_device, 10317 unlink_list); 10318 list_del_init(&dev->unlink_list); 10319 dev->nested_level = dev->lower_level - 1; 10320 } 10321 #endif 10322 10323 /* Snapshot list, allow later requests */ 10324 list_replace_init(&net_todo_list, &list); 10325 10326 __rtnl_unlock(); 10327 10328 /* Wait for rcu callbacks to finish before next phase */ 10329 if (!list_empty(&list)) 10330 rcu_barrier(); 10331 10332 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10333 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10334 netdev_WARN(dev, "run_todo but not unregistering\n"); 10335 list_del(&dev->todo_list); 10336 continue; 10337 } 10338 10339 write_lock(&dev_base_lock); 10340 dev->reg_state = NETREG_UNREGISTERED; 10341 write_unlock(&dev_base_lock); 10342 linkwatch_forget_dev(dev); 10343 } 10344 10345 while (!list_empty(&list)) { 10346 dev = netdev_wait_allrefs_any(&list); 10347 list_del(&dev->todo_list); 10348 10349 /* paranoia */ 10350 BUG_ON(netdev_refcnt_read(dev) != 1); 10351 BUG_ON(!list_empty(&dev->ptype_all)); 10352 BUG_ON(!list_empty(&dev->ptype_specific)); 10353 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10354 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10355 10356 if (dev->priv_destructor) 10357 dev->priv_destructor(dev); 10358 if (dev->needs_free_netdev) 10359 free_netdev(dev); 10360 10361 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10362 wake_up(&netdev_unregistering_wq); 10363 10364 /* Free network device */ 10365 kobject_put(&dev->dev.kobj); 10366 } 10367 } 10368 10369 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10370 * all the same fields in the same order as net_device_stats, with only 10371 * the type differing, but rtnl_link_stats64 may have additional fields 10372 * at the end for newer counters. 10373 */ 10374 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10375 const struct net_device_stats *netdev_stats) 10376 { 10377 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10378 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10379 u64 *dst = (u64 *)stats64; 10380 10381 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10382 for (i = 0; i < n; i++) 10383 dst[i] = atomic_long_read(&src[i]); 10384 /* zero out counters that only exist in rtnl_link_stats64 */ 10385 memset((char *)stats64 + n * sizeof(u64), 0, 10386 sizeof(*stats64) - n * sizeof(u64)); 10387 } 10388 EXPORT_SYMBOL(netdev_stats_to_stats64); 10389 10390 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10391 { 10392 struct net_device_core_stats __percpu *p; 10393 10394 p = alloc_percpu_gfp(struct net_device_core_stats, 10395 GFP_ATOMIC | __GFP_NOWARN); 10396 10397 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10398 free_percpu(p); 10399 10400 /* This READ_ONCE() pairs with the cmpxchg() above */ 10401 return READ_ONCE(dev->core_stats); 10402 } 10403 EXPORT_SYMBOL(netdev_core_stats_alloc); 10404 10405 /** 10406 * dev_get_stats - get network device statistics 10407 * @dev: device to get statistics from 10408 * @storage: place to store stats 10409 * 10410 * Get network statistics from device. Return @storage. 10411 * The device driver may provide its own method by setting 10412 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10413 * otherwise the internal statistics structure is used. 10414 */ 10415 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10416 struct rtnl_link_stats64 *storage) 10417 { 10418 const struct net_device_ops *ops = dev->netdev_ops; 10419 const struct net_device_core_stats __percpu *p; 10420 10421 if (ops->ndo_get_stats64) { 10422 memset(storage, 0, sizeof(*storage)); 10423 ops->ndo_get_stats64(dev, storage); 10424 } else if (ops->ndo_get_stats) { 10425 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10426 } else { 10427 netdev_stats_to_stats64(storage, &dev->stats); 10428 } 10429 10430 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10431 p = READ_ONCE(dev->core_stats); 10432 if (p) { 10433 const struct net_device_core_stats *core_stats; 10434 int i; 10435 10436 for_each_possible_cpu(i) { 10437 core_stats = per_cpu_ptr(p, i); 10438 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10439 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10440 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10441 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10442 } 10443 } 10444 return storage; 10445 } 10446 EXPORT_SYMBOL(dev_get_stats); 10447 10448 /** 10449 * dev_fetch_sw_netstats - get per-cpu network device statistics 10450 * @s: place to store stats 10451 * @netstats: per-cpu network stats to read from 10452 * 10453 * Read per-cpu network statistics and populate the related fields in @s. 10454 */ 10455 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10456 const struct pcpu_sw_netstats __percpu *netstats) 10457 { 10458 int cpu; 10459 10460 for_each_possible_cpu(cpu) { 10461 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10462 const struct pcpu_sw_netstats *stats; 10463 unsigned int start; 10464 10465 stats = per_cpu_ptr(netstats, cpu); 10466 do { 10467 start = u64_stats_fetch_begin(&stats->syncp); 10468 rx_packets = u64_stats_read(&stats->rx_packets); 10469 rx_bytes = u64_stats_read(&stats->rx_bytes); 10470 tx_packets = u64_stats_read(&stats->tx_packets); 10471 tx_bytes = u64_stats_read(&stats->tx_bytes); 10472 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10473 10474 s->rx_packets += rx_packets; 10475 s->rx_bytes += rx_bytes; 10476 s->tx_packets += tx_packets; 10477 s->tx_bytes += tx_bytes; 10478 } 10479 } 10480 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10481 10482 /** 10483 * dev_get_tstats64 - ndo_get_stats64 implementation 10484 * @dev: device to get statistics from 10485 * @s: place to store stats 10486 * 10487 * Populate @s from dev->stats and dev->tstats. Can be used as 10488 * ndo_get_stats64() callback. 10489 */ 10490 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10491 { 10492 netdev_stats_to_stats64(s, &dev->stats); 10493 dev_fetch_sw_netstats(s, dev->tstats); 10494 } 10495 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10496 10497 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10498 { 10499 struct netdev_queue *queue = dev_ingress_queue(dev); 10500 10501 #ifdef CONFIG_NET_CLS_ACT 10502 if (queue) 10503 return queue; 10504 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10505 if (!queue) 10506 return NULL; 10507 netdev_init_one_queue(dev, queue, NULL); 10508 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10509 queue->qdisc_sleeping = &noop_qdisc; 10510 rcu_assign_pointer(dev->ingress_queue, queue); 10511 #endif 10512 return queue; 10513 } 10514 10515 static const struct ethtool_ops default_ethtool_ops; 10516 10517 void netdev_set_default_ethtool_ops(struct net_device *dev, 10518 const struct ethtool_ops *ops) 10519 { 10520 if (dev->ethtool_ops == &default_ethtool_ops) 10521 dev->ethtool_ops = ops; 10522 } 10523 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10524 10525 /** 10526 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10527 * @dev: netdev to enable the IRQ coalescing on 10528 * 10529 * Sets a conservative default for SW IRQ coalescing. Users can use 10530 * sysfs attributes to override the default values. 10531 */ 10532 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10533 { 10534 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10535 10536 dev->gro_flush_timeout = 20000; 10537 dev->napi_defer_hard_irqs = 1; 10538 } 10539 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10540 10541 void netdev_freemem(struct net_device *dev) 10542 { 10543 char *addr = (char *)dev - dev->padded; 10544 10545 kvfree(addr); 10546 } 10547 10548 /** 10549 * alloc_netdev_mqs - allocate network device 10550 * @sizeof_priv: size of private data to allocate space for 10551 * @name: device name format string 10552 * @name_assign_type: origin of device name 10553 * @setup: callback to initialize device 10554 * @txqs: the number of TX subqueues to allocate 10555 * @rxqs: the number of RX subqueues to allocate 10556 * 10557 * Allocates a struct net_device with private data area for driver use 10558 * and performs basic initialization. Also allocates subqueue structs 10559 * for each queue on the device. 10560 */ 10561 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10562 unsigned char name_assign_type, 10563 void (*setup)(struct net_device *), 10564 unsigned int txqs, unsigned int rxqs) 10565 { 10566 struct net_device *dev; 10567 unsigned int alloc_size; 10568 struct net_device *p; 10569 10570 BUG_ON(strlen(name) >= sizeof(dev->name)); 10571 10572 if (txqs < 1) { 10573 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10574 return NULL; 10575 } 10576 10577 if (rxqs < 1) { 10578 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10579 return NULL; 10580 } 10581 10582 alloc_size = sizeof(struct net_device); 10583 if (sizeof_priv) { 10584 /* ensure 32-byte alignment of private area */ 10585 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10586 alloc_size += sizeof_priv; 10587 } 10588 /* ensure 32-byte alignment of whole construct */ 10589 alloc_size += NETDEV_ALIGN - 1; 10590 10591 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10592 if (!p) 10593 return NULL; 10594 10595 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10596 dev->padded = (char *)dev - (char *)p; 10597 10598 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10599 #ifdef CONFIG_PCPU_DEV_REFCNT 10600 dev->pcpu_refcnt = alloc_percpu(int); 10601 if (!dev->pcpu_refcnt) 10602 goto free_dev; 10603 __dev_hold(dev); 10604 #else 10605 refcount_set(&dev->dev_refcnt, 1); 10606 #endif 10607 10608 if (dev_addr_init(dev)) 10609 goto free_pcpu; 10610 10611 dev_mc_init(dev); 10612 dev_uc_init(dev); 10613 10614 dev_net_set(dev, &init_net); 10615 10616 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10617 dev->gso_max_segs = GSO_MAX_SEGS; 10618 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10619 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10620 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10621 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10622 dev->tso_max_segs = TSO_MAX_SEGS; 10623 dev->upper_level = 1; 10624 dev->lower_level = 1; 10625 #ifdef CONFIG_LOCKDEP 10626 dev->nested_level = 0; 10627 INIT_LIST_HEAD(&dev->unlink_list); 10628 #endif 10629 10630 INIT_LIST_HEAD(&dev->napi_list); 10631 INIT_LIST_HEAD(&dev->unreg_list); 10632 INIT_LIST_HEAD(&dev->close_list); 10633 INIT_LIST_HEAD(&dev->link_watch_list); 10634 INIT_LIST_HEAD(&dev->adj_list.upper); 10635 INIT_LIST_HEAD(&dev->adj_list.lower); 10636 INIT_LIST_HEAD(&dev->ptype_all); 10637 INIT_LIST_HEAD(&dev->ptype_specific); 10638 INIT_LIST_HEAD(&dev->net_notifier_list); 10639 #ifdef CONFIG_NET_SCHED 10640 hash_init(dev->qdisc_hash); 10641 #endif 10642 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10643 setup(dev); 10644 10645 if (!dev->tx_queue_len) { 10646 dev->priv_flags |= IFF_NO_QUEUE; 10647 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10648 } 10649 10650 dev->num_tx_queues = txqs; 10651 dev->real_num_tx_queues = txqs; 10652 if (netif_alloc_netdev_queues(dev)) 10653 goto free_all; 10654 10655 dev->num_rx_queues = rxqs; 10656 dev->real_num_rx_queues = rxqs; 10657 if (netif_alloc_rx_queues(dev)) 10658 goto free_all; 10659 10660 strcpy(dev->name, name); 10661 dev->name_assign_type = name_assign_type; 10662 dev->group = INIT_NETDEV_GROUP; 10663 if (!dev->ethtool_ops) 10664 dev->ethtool_ops = &default_ethtool_ops; 10665 10666 nf_hook_netdev_init(dev); 10667 10668 return dev; 10669 10670 free_all: 10671 free_netdev(dev); 10672 return NULL; 10673 10674 free_pcpu: 10675 #ifdef CONFIG_PCPU_DEV_REFCNT 10676 free_percpu(dev->pcpu_refcnt); 10677 free_dev: 10678 #endif 10679 netdev_freemem(dev); 10680 return NULL; 10681 } 10682 EXPORT_SYMBOL(alloc_netdev_mqs); 10683 10684 /** 10685 * free_netdev - free network device 10686 * @dev: device 10687 * 10688 * This function does the last stage of destroying an allocated device 10689 * interface. The reference to the device object is released. If this 10690 * is the last reference then it will be freed.Must be called in process 10691 * context. 10692 */ 10693 void free_netdev(struct net_device *dev) 10694 { 10695 struct napi_struct *p, *n; 10696 10697 might_sleep(); 10698 10699 /* When called immediately after register_netdevice() failed the unwind 10700 * handling may still be dismantling the device. Handle that case by 10701 * deferring the free. 10702 */ 10703 if (dev->reg_state == NETREG_UNREGISTERING) { 10704 ASSERT_RTNL(); 10705 dev->needs_free_netdev = true; 10706 return; 10707 } 10708 10709 netif_free_tx_queues(dev); 10710 netif_free_rx_queues(dev); 10711 10712 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10713 10714 /* Flush device addresses */ 10715 dev_addr_flush(dev); 10716 10717 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10718 netif_napi_del(p); 10719 10720 ref_tracker_dir_exit(&dev->refcnt_tracker); 10721 #ifdef CONFIG_PCPU_DEV_REFCNT 10722 free_percpu(dev->pcpu_refcnt); 10723 dev->pcpu_refcnt = NULL; 10724 #endif 10725 free_percpu(dev->core_stats); 10726 dev->core_stats = NULL; 10727 free_percpu(dev->xdp_bulkq); 10728 dev->xdp_bulkq = NULL; 10729 10730 /* Compatibility with error handling in drivers */ 10731 if (dev->reg_state == NETREG_UNINITIALIZED) { 10732 netdev_freemem(dev); 10733 return; 10734 } 10735 10736 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10737 dev->reg_state = NETREG_RELEASED; 10738 10739 /* will free via device release */ 10740 put_device(&dev->dev); 10741 } 10742 EXPORT_SYMBOL(free_netdev); 10743 10744 /** 10745 * synchronize_net - Synchronize with packet receive processing 10746 * 10747 * Wait for packets currently being received to be done. 10748 * Does not block later packets from starting. 10749 */ 10750 void synchronize_net(void) 10751 { 10752 might_sleep(); 10753 if (rtnl_is_locked()) 10754 synchronize_rcu_expedited(); 10755 else 10756 synchronize_rcu(); 10757 } 10758 EXPORT_SYMBOL(synchronize_net); 10759 10760 /** 10761 * unregister_netdevice_queue - remove device from the kernel 10762 * @dev: device 10763 * @head: list 10764 * 10765 * This function shuts down a device interface and removes it 10766 * from the kernel tables. 10767 * If head not NULL, device is queued to be unregistered later. 10768 * 10769 * Callers must hold the rtnl semaphore. You may want 10770 * unregister_netdev() instead of this. 10771 */ 10772 10773 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10774 { 10775 ASSERT_RTNL(); 10776 10777 if (head) { 10778 list_move_tail(&dev->unreg_list, head); 10779 } else { 10780 LIST_HEAD(single); 10781 10782 list_add(&dev->unreg_list, &single); 10783 unregister_netdevice_many(&single); 10784 } 10785 } 10786 EXPORT_SYMBOL(unregister_netdevice_queue); 10787 10788 void unregister_netdevice_many_notify(struct list_head *head, 10789 u32 portid, const struct nlmsghdr *nlh) 10790 { 10791 struct net_device *dev, *tmp; 10792 LIST_HEAD(close_head); 10793 10794 BUG_ON(dev_boot_phase); 10795 ASSERT_RTNL(); 10796 10797 if (list_empty(head)) 10798 return; 10799 10800 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10801 /* Some devices call without registering 10802 * for initialization unwind. Remove those 10803 * devices and proceed with the remaining. 10804 */ 10805 if (dev->reg_state == NETREG_UNINITIALIZED) { 10806 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10807 dev->name, dev); 10808 10809 WARN_ON(1); 10810 list_del(&dev->unreg_list); 10811 continue; 10812 } 10813 dev->dismantle = true; 10814 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10815 } 10816 10817 /* If device is running, close it first. */ 10818 list_for_each_entry(dev, head, unreg_list) 10819 list_add_tail(&dev->close_list, &close_head); 10820 dev_close_many(&close_head, true); 10821 10822 list_for_each_entry(dev, head, unreg_list) { 10823 /* And unlink it from device chain. */ 10824 write_lock(&dev_base_lock); 10825 unlist_netdevice(dev, false); 10826 dev->reg_state = NETREG_UNREGISTERING; 10827 write_unlock(&dev_base_lock); 10828 } 10829 flush_all_backlogs(); 10830 10831 synchronize_net(); 10832 10833 list_for_each_entry(dev, head, unreg_list) { 10834 struct sk_buff *skb = NULL; 10835 10836 /* Shutdown queueing discipline. */ 10837 dev_shutdown(dev); 10838 10839 dev_xdp_uninstall(dev); 10840 bpf_dev_bound_netdev_unregister(dev); 10841 10842 netdev_offload_xstats_disable_all(dev); 10843 10844 /* Notify protocols, that we are about to destroy 10845 * this device. They should clean all the things. 10846 */ 10847 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10848 10849 if (!dev->rtnl_link_ops || 10850 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10851 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10852 GFP_KERNEL, NULL, 0, 10853 portid, nlmsg_seq(nlh)); 10854 10855 /* 10856 * Flush the unicast and multicast chains 10857 */ 10858 dev_uc_flush(dev); 10859 dev_mc_flush(dev); 10860 10861 netdev_name_node_alt_flush(dev); 10862 netdev_name_node_free(dev->name_node); 10863 10864 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10865 10866 if (dev->netdev_ops->ndo_uninit) 10867 dev->netdev_ops->ndo_uninit(dev); 10868 10869 if (skb) 10870 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10871 10872 /* Notifier chain MUST detach us all upper devices. */ 10873 WARN_ON(netdev_has_any_upper_dev(dev)); 10874 WARN_ON(netdev_has_any_lower_dev(dev)); 10875 10876 /* Remove entries from kobject tree */ 10877 netdev_unregister_kobject(dev); 10878 #ifdef CONFIG_XPS 10879 /* Remove XPS queueing entries */ 10880 netif_reset_xps_queues_gt(dev, 0); 10881 #endif 10882 } 10883 10884 synchronize_net(); 10885 10886 list_for_each_entry(dev, head, unreg_list) { 10887 netdev_put(dev, &dev->dev_registered_tracker); 10888 net_set_todo(dev); 10889 } 10890 10891 list_del(head); 10892 } 10893 10894 /** 10895 * unregister_netdevice_many - unregister many devices 10896 * @head: list of devices 10897 * 10898 * Note: As most callers use a stack allocated list_head, 10899 * we force a list_del() to make sure stack wont be corrupted later. 10900 */ 10901 void unregister_netdevice_many(struct list_head *head) 10902 { 10903 unregister_netdevice_many_notify(head, 0, NULL); 10904 } 10905 EXPORT_SYMBOL(unregister_netdevice_many); 10906 10907 /** 10908 * unregister_netdev - remove device from the kernel 10909 * @dev: device 10910 * 10911 * This function shuts down a device interface and removes it 10912 * from the kernel tables. 10913 * 10914 * This is just a wrapper for unregister_netdevice that takes 10915 * the rtnl semaphore. In general you want to use this and not 10916 * unregister_netdevice. 10917 */ 10918 void unregister_netdev(struct net_device *dev) 10919 { 10920 rtnl_lock(); 10921 unregister_netdevice(dev); 10922 rtnl_unlock(); 10923 } 10924 EXPORT_SYMBOL(unregister_netdev); 10925 10926 /** 10927 * __dev_change_net_namespace - move device to different nethost namespace 10928 * @dev: device 10929 * @net: network namespace 10930 * @pat: If not NULL name pattern to try if the current device name 10931 * is already taken in the destination network namespace. 10932 * @new_ifindex: If not zero, specifies device index in the target 10933 * namespace. 10934 * 10935 * This function shuts down a device interface and moves it 10936 * to a new network namespace. On success 0 is returned, on 10937 * a failure a netagive errno code is returned. 10938 * 10939 * Callers must hold the rtnl semaphore. 10940 */ 10941 10942 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10943 const char *pat, int new_ifindex) 10944 { 10945 struct net *net_old = dev_net(dev); 10946 int err, new_nsid; 10947 10948 ASSERT_RTNL(); 10949 10950 /* Don't allow namespace local devices to be moved. */ 10951 err = -EINVAL; 10952 if (dev->features & NETIF_F_NETNS_LOCAL) 10953 goto out; 10954 10955 /* Ensure the device has been registrered */ 10956 if (dev->reg_state != NETREG_REGISTERED) 10957 goto out; 10958 10959 /* Get out if there is nothing todo */ 10960 err = 0; 10961 if (net_eq(net_old, net)) 10962 goto out; 10963 10964 /* Pick the destination device name, and ensure 10965 * we can use it in the destination network namespace. 10966 */ 10967 err = -EEXIST; 10968 if (netdev_name_in_use(net, dev->name)) { 10969 /* We get here if we can't use the current device name */ 10970 if (!pat) 10971 goto out; 10972 err = dev_get_valid_name(net, dev, pat); 10973 if (err < 0) 10974 goto out; 10975 } 10976 10977 /* Check that new_ifindex isn't used yet. */ 10978 err = -EBUSY; 10979 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10980 goto out; 10981 10982 /* 10983 * And now a mini version of register_netdevice unregister_netdevice. 10984 */ 10985 10986 /* If device is running close it first. */ 10987 dev_close(dev); 10988 10989 /* And unlink it from device chain */ 10990 unlist_netdevice(dev, true); 10991 10992 synchronize_net(); 10993 10994 /* Shutdown queueing discipline. */ 10995 dev_shutdown(dev); 10996 10997 /* Notify protocols, that we are about to destroy 10998 * this device. They should clean all the things. 10999 * 11000 * Note that dev->reg_state stays at NETREG_REGISTERED. 11001 * This is wanted because this way 8021q and macvlan know 11002 * the device is just moving and can keep their slaves up. 11003 */ 11004 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11005 rcu_barrier(); 11006 11007 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11008 /* If there is an ifindex conflict assign a new one */ 11009 if (!new_ifindex) { 11010 if (__dev_get_by_index(net, dev->ifindex)) 11011 new_ifindex = dev_new_index(net); 11012 else 11013 new_ifindex = dev->ifindex; 11014 } 11015 11016 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11017 new_ifindex); 11018 11019 /* 11020 * Flush the unicast and multicast chains 11021 */ 11022 dev_uc_flush(dev); 11023 dev_mc_flush(dev); 11024 11025 /* Send a netdev-removed uevent to the old namespace */ 11026 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11027 netdev_adjacent_del_links(dev); 11028 11029 /* Move per-net netdevice notifiers that are following the netdevice */ 11030 move_netdevice_notifiers_dev_net(dev, net); 11031 11032 /* Actually switch the network namespace */ 11033 dev_net_set(dev, net); 11034 dev->ifindex = new_ifindex; 11035 11036 /* Send a netdev-add uevent to the new namespace */ 11037 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11038 netdev_adjacent_add_links(dev); 11039 11040 /* Fixup kobjects */ 11041 err = device_rename(&dev->dev, dev->name); 11042 WARN_ON(err); 11043 11044 /* Adapt owner in case owning user namespace of target network 11045 * namespace is different from the original one. 11046 */ 11047 err = netdev_change_owner(dev, net_old, net); 11048 WARN_ON(err); 11049 11050 /* Add the device back in the hashes */ 11051 list_netdevice(dev); 11052 11053 /* Notify protocols, that a new device appeared. */ 11054 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11055 11056 /* 11057 * Prevent userspace races by waiting until the network 11058 * device is fully setup before sending notifications. 11059 */ 11060 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11061 11062 synchronize_net(); 11063 err = 0; 11064 out: 11065 return err; 11066 } 11067 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11068 11069 static int dev_cpu_dead(unsigned int oldcpu) 11070 { 11071 struct sk_buff **list_skb; 11072 struct sk_buff *skb; 11073 unsigned int cpu; 11074 struct softnet_data *sd, *oldsd, *remsd = NULL; 11075 11076 local_irq_disable(); 11077 cpu = smp_processor_id(); 11078 sd = &per_cpu(softnet_data, cpu); 11079 oldsd = &per_cpu(softnet_data, oldcpu); 11080 11081 /* Find end of our completion_queue. */ 11082 list_skb = &sd->completion_queue; 11083 while (*list_skb) 11084 list_skb = &(*list_skb)->next; 11085 /* Append completion queue from offline CPU. */ 11086 *list_skb = oldsd->completion_queue; 11087 oldsd->completion_queue = NULL; 11088 11089 /* Append output queue from offline CPU. */ 11090 if (oldsd->output_queue) { 11091 *sd->output_queue_tailp = oldsd->output_queue; 11092 sd->output_queue_tailp = oldsd->output_queue_tailp; 11093 oldsd->output_queue = NULL; 11094 oldsd->output_queue_tailp = &oldsd->output_queue; 11095 } 11096 /* Append NAPI poll list from offline CPU, with one exception : 11097 * process_backlog() must be called by cpu owning percpu backlog. 11098 * We properly handle process_queue & input_pkt_queue later. 11099 */ 11100 while (!list_empty(&oldsd->poll_list)) { 11101 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11102 struct napi_struct, 11103 poll_list); 11104 11105 list_del_init(&napi->poll_list); 11106 if (napi->poll == process_backlog) 11107 napi->state = 0; 11108 else 11109 ____napi_schedule(sd, napi); 11110 } 11111 11112 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11113 local_irq_enable(); 11114 11115 #ifdef CONFIG_RPS 11116 remsd = oldsd->rps_ipi_list; 11117 oldsd->rps_ipi_list = NULL; 11118 #endif 11119 /* send out pending IPI's on offline CPU */ 11120 net_rps_send_ipi(remsd); 11121 11122 /* Process offline CPU's input_pkt_queue */ 11123 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11124 netif_rx(skb); 11125 input_queue_head_incr(oldsd); 11126 } 11127 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11128 netif_rx(skb); 11129 input_queue_head_incr(oldsd); 11130 } 11131 11132 return 0; 11133 } 11134 11135 /** 11136 * netdev_increment_features - increment feature set by one 11137 * @all: current feature set 11138 * @one: new feature set 11139 * @mask: mask feature set 11140 * 11141 * Computes a new feature set after adding a device with feature set 11142 * @one to the master device with current feature set @all. Will not 11143 * enable anything that is off in @mask. Returns the new feature set. 11144 */ 11145 netdev_features_t netdev_increment_features(netdev_features_t all, 11146 netdev_features_t one, netdev_features_t mask) 11147 { 11148 if (mask & NETIF_F_HW_CSUM) 11149 mask |= NETIF_F_CSUM_MASK; 11150 mask |= NETIF_F_VLAN_CHALLENGED; 11151 11152 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11153 all &= one | ~NETIF_F_ALL_FOR_ALL; 11154 11155 /* If one device supports hw checksumming, set for all. */ 11156 if (all & NETIF_F_HW_CSUM) 11157 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11158 11159 return all; 11160 } 11161 EXPORT_SYMBOL(netdev_increment_features); 11162 11163 static struct hlist_head * __net_init netdev_create_hash(void) 11164 { 11165 int i; 11166 struct hlist_head *hash; 11167 11168 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11169 if (hash != NULL) 11170 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11171 INIT_HLIST_HEAD(&hash[i]); 11172 11173 return hash; 11174 } 11175 11176 /* Initialize per network namespace state */ 11177 static int __net_init netdev_init(struct net *net) 11178 { 11179 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11180 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11181 11182 INIT_LIST_HEAD(&net->dev_base_head); 11183 11184 net->dev_name_head = netdev_create_hash(); 11185 if (net->dev_name_head == NULL) 11186 goto err_name; 11187 11188 net->dev_index_head = netdev_create_hash(); 11189 if (net->dev_index_head == NULL) 11190 goto err_idx; 11191 11192 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11193 11194 return 0; 11195 11196 err_idx: 11197 kfree(net->dev_name_head); 11198 err_name: 11199 return -ENOMEM; 11200 } 11201 11202 /** 11203 * netdev_drivername - network driver for the device 11204 * @dev: network device 11205 * 11206 * Determine network driver for device. 11207 */ 11208 const char *netdev_drivername(const struct net_device *dev) 11209 { 11210 const struct device_driver *driver; 11211 const struct device *parent; 11212 const char *empty = ""; 11213 11214 parent = dev->dev.parent; 11215 if (!parent) 11216 return empty; 11217 11218 driver = parent->driver; 11219 if (driver && driver->name) 11220 return driver->name; 11221 return empty; 11222 } 11223 11224 static void __netdev_printk(const char *level, const struct net_device *dev, 11225 struct va_format *vaf) 11226 { 11227 if (dev && dev->dev.parent) { 11228 dev_printk_emit(level[1] - '0', 11229 dev->dev.parent, 11230 "%s %s %s%s: %pV", 11231 dev_driver_string(dev->dev.parent), 11232 dev_name(dev->dev.parent), 11233 netdev_name(dev), netdev_reg_state(dev), 11234 vaf); 11235 } else if (dev) { 11236 printk("%s%s%s: %pV", 11237 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11238 } else { 11239 printk("%s(NULL net_device): %pV", level, vaf); 11240 } 11241 } 11242 11243 void netdev_printk(const char *level, const struct net_device *dev, 11244 const char *format, ...) 11245 { 11246 struct va_format vaf; 11247 va_list args; 11248 11249 va_start(args, format); 11250 11251 vaf.fmt = format; 11252 vaf.va = &args; 11253 11254 __netdev_printk(level, dev, &vaf); 11255 11256 va_end(args); 11257 } 11258 EXPORT_SYMBOL(netdev_printk); 11259 11260 #define define_netdev_printk_level(func, level) \ 11261 void func(const struct net_device *dev, const char *fmt, ...) \ 11262 { \ 11263 struct va_format vaf; \ 11264 va_list args; \ 11265 \ 11266 va_start(args, fmt); \ 11267 \ 11268 vaf.fmt = fmt; \ 11269 vaf.va = &args; \ 11270 \ 11271 __netdev_printk(level, dev, &vaf); \ 11272 \ 11273 va_end(args); \ 11274 } \ 11275 EXPORT_SYMBOL(func); 11276 11277 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11278 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11279 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11280 define_netdev_printk_level(netdev_err, KERN_ERR); 11281 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11282 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11283 define_netdev_printk_level(netdev_info, KERN_INFO); 11284 11285 static void __net_exit netdev_exit(struct net *net) 11286 { 11287 kfree(net->dev_name_head); 11288 kfree(net->dev_index_head); 11289 if (net != &init_net) 11290 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11291 } 11292 11293 static struct pernet_operations __net_initdata netdev_net_ops = { 11294 .init = netdev_init, 11295 .exit = netdev_exit, 11296 }; 11297 11298 static void __net_exit default_device_exit_net(struct net *net) 11299 { 11300 struct net_device *dev, *aux; 11301 /* 11302 * Push all migratable network devices back to the 11303 * initial network namespace 11304 */ 11305 ASSERT_RTNL(); 11306 for_each_netdev_safe(net, dev, aux) { 11307 int err; 11308 char fb_name[IFNAMSIZ]; 11309 11310 /* Ignore unmoveable devices (i.e. loopback) */ 11311 if (dev->features & NETIF_F_NETNS_LOCAL) 11312 continue; 11313 11314 /* Leave virtual devices for the generic cleanup */ 11315 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11316 continue; 11317 11318 /* Push remaining network devices to init_net */ 11319 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11320 if (netdev_name_in_use(&init_net, fb_name)) 11321 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11322 err = dev_change_net_namespace(dev, &init_net, fb_name); 11323 if (err) { 11324 pr_emerg("%s: failed to move %s to init_net: %d\n", 11325 __func__, dev->name, err); 11326 BUG(); 11327 } 11328 } 11329 } 11330 11331 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11332 { 11333 /* At exit all network devices most be removed from a network 11334 * namespace. Do this in the reverse order of registration. 11335 * Do this across as many network namespaces as possible to 11336 * improve batching efficiency. 11337 */ 11338 struct net_device *dev; 11339 struct net *net; 11340 LIST_HEAD(dev_kill_list); 11341 11342 rtnl_lock(); 11343 list_for_each_entry(net, net_list, exit_list) { 11344 default_device_exit_net(net); 11345 cond_resched(); 11346 } 11347 11348 list_for_each_entry(net, net_list, exit_list) { 11349 for_each_netdev_reverse(net, dev) { 11350 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11351 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11352 else 11353 unregister_netdevice_queue(dev, &dev_kill_list); 11354 } 11355 } 11356 unregister_netdevice_many(&dev_kill_list); 11357 rtnl_unlock(); 11358 } 11359 11360 static struct pernet_operations __net_initdata default_device_ops = { 11361 .exit_batch = default_device_exit_batch, 11362 }; 11363 11364 /* 11365 * Initialize the DEV module. At boot time this walks the device list and 11366 * unhooks any devices that fail to initialise (normally hardware not 11367 * present) and leaves us with a valid list of present and active devices. 11368 * 11369 */ 11370 11371 /* 11372 * This is called single threaded during boot, so no need 11373 * to take the rtnl semaphore. 11374 */ 11375 static int __init net_dev_init(void) 11376 { 11377 int i, rc = -ENOMEM; 11378 11379 BUG_ON(!dev_boot_phase); 11380 11381 if (dev_proc_init()) 11382 goto out; 11383 11384 if (netdev_kobject_init()) 11385 goto out; 11386 11387 INIT_LIST_HEAD(&ptype_all); 11388 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11389 INIT_LIST_HEAD(&ptype_base[i]); 11390 11391 if (register_pernet_subsys(&netdev_net_ops)) 11392 goto out; 11393 11394 /* 11395 * Initialise the packet receive queues. 11396 */ 11397 11398 for_each_possible_cpu(i) { 11399 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11400 struct softnet_data *sd = &per_cpu(softnet_data, i); 11401 11402 INIT_WORK(flush, flush_backlog); 11403 11404 skb_queue_head_init(&sd->input_pkt_queue); 11405 skb_queue_head_init(&sd->process_queue); 11406 #ifdef CONFIG_XFRM_OFFLOAD 11407 skb_queue_head_init(&sd->xfrm_backlog); 11408 #endif 11409 INIT_LIST_HEAD(&sd->poll_list); 11410 sd->output_queue_tailp = &sd->output_queue; 11411 #ifdef CONFIG_RPS 11412 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11413 sd->cpu = i; 11414 #endif 11415 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11416 spin_lock_init(&sd->defer_lock); 11417 11418 init_gro_hash(&sd->backlog); 11419 sd->backlog.poll = process_backlog; 11420 sd->backlog.weight = weight_p; 11421 } 11422 11423 dev_boot_phase = 0; 11424 11425 /* The loopback device is special if any other network devices 11426 * is present in a network namespace the loopback device must 11427 * be present. Since we now dynamically allocate and free the 11428 * loopback device ensure this invariant is maintained by 11429 * keeping the loopback device as the first device on the 11430 * list of network devices. Ensuring the loopback devices 11431 * is the first device that appears and the last network device 11432 * that disappears. 11433 */ 11434 if (register_pernet_device(&loopback_net_ops)) 11435 goto out; 11436 11437 if (register_pernet_device(&default_device_ops)) 11438 goto out; 11439 11440 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11441 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11442 11443 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11444 NULL, dev_cpu_dead); 11445 WARN_ON(rc < 0); 11446 rc = 0; 11447 out: 11448 return rc; 11449 } 11450 11451 subsys_initcall(net_dev_init); 11452