1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "dev.h" 155 #include "net-sysfs.h" 156 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_extack(unsigned long val, 164 struct net_device *dev, 165 struct netlink_ext_ack *extack); 166 static struct napi_struct *napi_by_id(unsigned int napi_id); 167 168 /* 169 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 170 * semaphore. 171 * 172 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 173 * 174 * Writers must hold the rtnl semaphore while they loop through the 175 * dev_base_head list, and hold dev_base_lock for writing when they do the 176 * actual updates. This allows pure readers to access the list even 177 * while a writer is preparing to update it. 178 * 179 * To put it another way, dev_base_lock is held for writing only to 180 * protect against pure readers; the rtnl semaphore provides the 181 * protection against other writers. 182 * 183 * See, for example usages, register_netdevice() and 184 * unregister_netdevice(), which must be called with the rtnl 185 * semaphore held. 186 */ 187 DEFINE_RWLOCK(dev_base_lock); 188 EXPORT_SYMBOL(dev_base_lock); 189 190 static DEFINE_MUTEX(ifalias_mutex); 191 192 /* protects napi_hash addition/deletion and napi_gen_id */ 193 static DEFINE_SPINLOCK(napi_hash_lock); 194 195 static unsigned int napi_gen_id = NR_CPUS; 196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 197 198 static DECLARE_RWSEM(devnet_rename_sem); 199 200 static inline void dev_base_seq_inc(struct net *net) 201 { 202 while (++net->dev_base_seq == 0) 203 ; 204 } 205 206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 207 { 208 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 209 210 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 211 } 212 213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 214 { 215 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 216 } 217 218 static inline void rps_lock_irqsave(struct softnet_data *sd, 219 unsigned long *flags) 220 { 221 if (IS_ENABLED(CONFIG_RPS)) 222 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 223 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 224 local_irq_save(*flags); 225 } 226 227 static inline void rps_lock_irq_disable(struct softnet_data *sd) 228 { 229 if (IS_ENABLED(CONFIG_RPS)) 230 spin_lock_irq(&sd->input_pkt_queue.lock); 231 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 232 local_irq_disable(); 233 } 234 235 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 236 unsigned long *flags) 237 { 238 if (IS_ENABLED(CONFIG_RPS)) 239 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 240 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 241 local_irq_restore(*flags); 242 } 243 244 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 245 { 246 if (IS_ENABLED(CONFIG_RPS)) 247 spin_unlock_irq(&sd->input_pkt_queue.lock); 248 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 249 local_irq_enable(); 250 } 251 252 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 253 const char *name) 254 { 255 struct netdev_name_node *name_node; 256 257 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 258 if (!name_node) 259 return NULL; 260 INIT_HLIST_NODE(&name_node->hlist); 261 name_node->dev = dev; 262 name_node->name = name; 263 return name_node; 264 } 265 266 static struct netdev_name_node * 267 netdev_name_node_head_alloc(struct net_device *dev) 268 { 269 struct netdev_name_node *name_node; 270 271 name_node = netdev_name_node_alloc(dev, dev->name); 272 if (!name_node) 273 return NULL; 274 INIT_LIST_HEAD(&name_node->list); 275 return name_node; 276 } 277 278 static void netdev_name_node_free(struct netdev_name_node *name_node) 279 { 280 kfree(name_node); 281 } 282 283 static void netdev_name_node_add(struct net *net, 284 struct netdev_name_node *name_node) 285 { 286 hlist_add_head_rcu(&name_node->hlist, 287 dev_name_hash(net, name_node->name)); 288 } 289 290 static void netdev_name_node_del(struct netdev_name_node *name_node) 291 { 292 hlist_del_rcu(&name_node->hlist); 293 } 294 295 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 296 const char *name) 297 { 298 struct hlist_head *head = dev_name_hash(net, name); 299 struct netdev_name_node *name_node; 300 301 hlist_for_each_entry(name_node, head, hlist) 302 if (!strcmp(name_node->name, name)) 303 return name_node; 304 return NULL; 305 } 306 307 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 308 const char *name) 309 { 310 struct hlist_head *head = dev_name_hash(net, name); 311 struct netdev_name_node *name_node; 312 313 hlist_for_each_entry_rcu(name_node, head, hlist) 314 if (!strcmp(name_node->name, name)) 315 return name_node; 316 return NULL; 317 } 318 319 bool netdev_name_in_use(struct net *net, const char *name) 320 { 321 return netdev_name_node_lookup(net, name); 322 } 323 EXPORT_SYMBOL(netdev_name_in_use); 324 325 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 326 { 327 struct netdev_name_node *name_node; 328 struct net *net = dev_net(dev); 329 330 name_node = netdev_name_node_lookup(net, name); 331 if (name_node) 332 return -EEXIST; 333 name_node = netdev_name_node_alloc(dev, name); 334 if (!name_node) 335 return -ENOMEM; 336 netdev_name_node_add(net, name_node); 337 /* The node that holds dev->name acts as a head of per-device list. */ 338 list_add_tail(&name_node->list, &dev->name_node->list); 339 340 return 0; 341 } 342 343 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 344 { 345 list_del(&name_node->list); 346 netdev_name_node_del(name_node); 347 kfree(name_node->name); 348 netdev_name_node_free(name_node); 349 } 350 351 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 352 { 353 struct netdev_name_node *name_node; 354 struct net *net = dev_net(dev); 355 356 name_node = netdev_name_node_lookup(net, name); 357 if (!name_node) 358 return -ENOENT; 359 /* lookup might have found our primary name or a name belonging 360 * to another device. 361 */ 362 if (name_node == dev->name_node || name_node->dev != dev) 363 return -EINVAL; 364 365 __netdev_name_node_alt_destroy(name_node); 366 367 return 0; 368 } 369 370 static void netdev_name_node_alt_flush(struct net_device *dev) 371 { 372 struct netdev_name_node *name_node, *tmp; 373 374 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 375 __netdev_name_node_alt_destroy(name_node); 376 } 377 378 /* Device list insertion */ 379 static void list_netdevice(struct net_device *dev) 380 { 381 struct net *net = dev_net(dev); 382 383 ASSERT_RTNL(); 384 385 write_lock(&dev_base_lock); 386 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 387 netdev_name_node_add(net, dev->name_node); 388 hlist_add_head_rcu(&dev->index_hlist, 389 dev_index_hash(net, dev->ifindex)); 390 write_unlock(&dev_base_lock); 391 392 dev_base_seq_inc(net); 393 } 394 395 /* Device list removal 396 * caller must respect a RCU grace period before freeing/reusing dev 397 */ 398 static void unlist_netdevice(struct net_device *dev, bool lock) 399 { 400 ASSERT_RTNL(); 401 402 /* Unlink dev from the device chain */ 403 if (lock) 404 write_lock(&dev_base_lock); 405 list_del_rcu(&dev->dev_list); 406 netdev_name_node_del(dev->name_node); 407 hlist_del_rcu(&dev->index_hlist); 408 if (lock) 409 write_unlock(&dev_base_lock); 410 411 dev_base_seq_inc(dev_net(dev)); 412 } 413 414 /* 415 * Our notifier list 416 */ 417 418 static RAW_NOTIFIER_HEAD(netdev_chain); 419 420 /* 421 * Device drivers call our routines to queue packets here. We empty the 422 * queue in the local softnet handler. 423 */ 424 425 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 426 EXPORT_PER_CPU_SYMBOL(softnet_data); 427 428 #ifdef CONFIG_LOCKDEP 429 /* 430 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 431 * according to dev->type 432 */ 433 static const unsigned short netdev_lock_type[] = { 434 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 435 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 436 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 437 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 438 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 439 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 440 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 441 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 442 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 443 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 444 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 445 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 446 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 447 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 448 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 449 450 static const char *const netdev_lock_name[] = { 451 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 452 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 453 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 454 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 455 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 456 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 457 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 458 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 459 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 460 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 461 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 462 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 463 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 464 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 465 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 466 467 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 468 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 469 470 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 471 { 472 int i; 473 474 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 475 if (netdev_lock_type[i] == dev_type) 476 return i; 477 /* the last key is used by default */ 478 return ARRAY_SIZE(netdev_lock_type) - 1; 479 } 480 481 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 482 unsigned short dev_type) 483 { 484 int i; 485 486 i = netdev_lock_pos(dev_type); 487 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 488 netdev_lock_name[i]); 489 } 490 491 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 492 { 493 int i; 494 495 i = netdev_lock_pos(dev->type); 496 lockdep_set_class_and_name(&dev->addr_list_lock, 497 &netdev_addr_lock_key[i], 498 netdev_lock_name[i]); 499 } 500 #else 501 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 502 unsigned short dev_type) 503 { 504 } 505 506 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 507 { 508 } 509 #endif 510 511 /******************************************************************************* 512 * 513 * Protocol management and registration routines 514 * 515 *******************************************************************************/ 516 517 518 /* 519 * Add a protocol ID to the list. Now that the input handler is 520 * smarter we can dispense with all the messy stuff that used to be 521 * here. 522 * 523 * BEWARE!!! Protocol handlers, mangling input packets, 524 * MUST BE last in hash buckets and checking protocol handlers 525 * MUST start from promiscuous ptype_all chain in net_bh. 526 * It is true now, do not change it. 527 * Explanation follows: if protocol handler, mangling packet, will 528 * be the first on list, it is not able to sense, that packet 529 * is cloned and should be copied-on-write, so that it will 530 * change it and subsequent readers will get broken packet. 531 * --ANK (980803) 532 */ 533 534 static inline struct list_head *ptype_head(const struct packet_type *pt) 535 { 536 if (pt->type == htons(ETH_P_ALL)) 537 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 538 else 539 return pt->dev ? &pt->dev->ptype_specific : 540 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 541 } 542 543 /** 544 * dev_add_pack - add packet handler 545 * @pt: packet type declaration 546 * 547 * Add a protocol handler to the networking stack. The passed &packet_type 548 * is linked into kernel lists and may not be freed until it has been 549 * removed from the kernel lists. 550 * 551 * This call does not sleep therefore it can not 552 * guarantee all CPU's that are in middle of receiving packets 553 * will see the new packet type (until the next received packet). 554 */ 555 556 void dev_add_pack(struct packet_type *pt) 557 { 558 struct list_head *head = ptype_head(pt); 559 560 spin_lock(&ptype_lock); 561 list_add_rcu(&pt->list, head); 562 spin_unlock(&ptype_lock); 563 } 564 EXPORT_SYMBOL(dev_add_pack); 565 566 /** 567 * __dev_remove_pack - remove packet handler 568 * @pt: packet type declaration 569 * 570 * Remove a protocol handler that was previously added to the kernel 571 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 572 * from the kernel lists and can be freed or reused once this function 573 * returns. 574 * 575 * The packet type might still be in use by receivers 576 * and must not be freed until after all the CPU's have gone 577 * through a quiescent state. 578 */ 579 void __dev_remove_pack(struct packet_type *pt) 580 { 581 struct list_head *head = ptype_head(pt); 582 struct packet_type *pt1; 583 584 spin_lock(&ptype_lock); 585 586 list_for_each_entry(pt1, head, list) { 587 if (pt == pt1) { 588 list_del_rcu(&pt->list); 589 goto out; 590 } 591 } 592 593 pr_warn("dev_remove_pack: %p not found\n", pt); 594 out: 595 spin_unlock(&ptype_lock); 596 } 597 EXPORT_SYMBOL(__dev_remove_pack); 598 599 /** 600 * dev_remove_pack - remove packet handler 601 * @pt: packet type declaration 602 * 603 * Remove a protocol handler that was previously added to the kernel 604 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 605 * from the kernel lists and can be freed or reused once this function 606 * returns. 607 * 608 * This call sleeps to guarantee that no CPU is looking at the packet 609 * type after return. 610 */ 611 void dev_remove_pack(struct packet_type *pt) 612 { 613 __dev_remove_pack(pt); 614 615 synchronize_net(); 616 } 617 EXPORT_SYMBOL(dev_remove_pack); 618 619 620 /******************************************************************************* 621 * 622 * Device Interface Subroutines 623 * 624 *******************************************************************************/ 625 626 /** 627 * dev_get_iflink - get 'iflink' value of a interface 628 * @dev: targeted interface 629 * 630 * Indicates the ifindex the interface is linked to. 631 * Physical interfaces have the same 'ifindex' and 'iflink' values. 632 */ 633 634 int dev_get_iflink(const struct net_device *dev) 635 { 636 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 637 return dev->netdev_ops->ndo_get_iflink(dev); 638 639 return dev->ifindex; 640 } 641 EXPORT_SYMBOL(dev_get_iflink); 642 643 /** 644 * dev_fill_metadata_dst - Retrieve tunnel egress information. 645 * @dev: targeted interface 646 * @skb: The packet. 647 * 648 * For better visibility of tunnel traffic OVS needs to retrieve 649 * egress tunnel information for a packet. Following API allows 650 * user to get this info. 651 */ 652 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 653 { 654 struct ip_tunnel_info *info; 655 656 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 657 return -EINVAL; 658 659 info = skb_tunnel_info_unclone(skb); 660 if (!info) 661 return -ENOMEM; 662 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 663 return -EINVAL; 664 665 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 666 } 667 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 668 669 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 670 { 671 int k = stack->num_paths++; 672 673 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 674 return NULL; 675 676 return &stack->path[k]; 677 } 678 679 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 680 struct net_device_path_stack *stack) 681 { 682 const struct net_device *last_dev; 683 struct net_device_path_ctx ctx = { 684 .dev = dev, 685 }; 686 struct net_device_path *path; 687 int ret = 0; 688 689 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 690 stack->num_paths = 0; 691 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 692 last_dev = ctx.dev; 693 path = dev_fwd_path(stack); 694 if (!path) 695 return -1; 696 697 memset(path, 0, sizeof(struct net_device_path)); 698 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 699 if (ret < 0) 700 return -1; 701 702 if (WARN_ON_ONCE(last_dev == ctx.dev)) 703 return -1; 704 } 705 706 if (!ctx.dev) 707 return ret; 708 709 path = dev_fwd_path(stack); 710 if (!path) 711 return -1; 712 path->type = DEV_PATH_ETHERNET; 713 path->dev = ctx.dev; 714 715 return ret; 716 } 717 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 718 719 /** 720 * __dev_get_by_name - find a device by its name 721 * @net: the applicable net namespace 722 * @name: name to find 723 * 724 * Find an interface by name. Must be called under RTNL semaphore 725 * or @dev_base_lock. If the name is found a pointer to the device 726 * is returned. If the name is not found then %NULL is returned. The 727 * reference counters are not incremented so the caller must be 728 * careful with locks. 729 */ 730 731 struct net_device *__dev_get_by_name(struct net *net, const char *name) 732 { 733 struct netdev_name_node *node_name; 734 735 node_name = netdev_name_node_lookup(net, name); 736 return node_name ? node_name->dev : NULL; 737 } 738 EXPORT_SYMBOL(__dev_get_by_name); 739 740 /** 741 * dev_get_by_name_rcu - find a device by its name 742 * @net: the applicable net namespace 743 * @name: name to find 744 * 745 * Find an interface by name. 746 * If the name is found a pointer to the device is returned. 747 * If the name is not found then %NULL is returned. 748 * The reference counters are not incremented so the caller must be 749 * careful with locks. The caller must hold RCU lock. 750 */ 751 752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 753 { 754 struct netdev_name_node *node_name; 755 756 node_name = netdev_name_node_lookup_rcu(net, name); 757 return node_name ? node_name->dev : NULL; 758 } 759 EXPORT_SYMBOL(dev_get_by_name_rcu); 760 761 /** 762 * dev_get_by_name - find a device by its name 763 * @net: the applicable net namespace 764 * @name: name to find 765 * 766 * Find an interface by name. This can be called from any 767 * context and does its own locking. The returned handle has 768 * the usage count incremented and the caller must use dev_put() to 769 * release it when it is no longer needed. %NULL is returned if no 770 * matching device is found. 771 */ 772 773 struct net_device *dev_get_by_name(struct net *net, const char *name) 774 { 775 struct net_device *dev; 776 777 rcu_read_lock(); 778 dev = dev_get_by_name_rcu(net, name); 779 dev_hold(dev); 780 rcu_read_unlock(); 781 return dev; 782 } 783 EXPORT_SYMBOL(dev_get_by_name); 784 785 /** 786 * __dev_get_by_index - find a device by its ifindex 787 * @net: the applicable net namespace 788 * @ifindex: index of device 789 * 790 * Search for an interface by index. Returns %NULL if the device 791 * is not found or a pointer to the device. The device has not 792 * had its reference counter increased so the caller must be careful 793 * about locking. The caller must hold either the RTNL semaphore 794 * or @dev_base_lock. 795 */ 796 797 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 798 { 799 struct net_device *dev; 800 struct hlist_head *head = dev_index_hash(net, ifindex); 801 802 hlist_for_each_entry(dev, head, index_hlist) 803 if (dev->ifindex == ifindex) 804 return dev; 805 806 return NULL; 807 } 808 EXPORT_SYMBOL(__dev_get_by_index); 809 810 /** 811 * dev_get_by_index_rcu - find a device by its ifindex 812 * @net: the applicable net namespace 813 * @ifindex: index of device 814 * 815 * Search for an interface by index. Returns %NULL if the device 816 * is not found or a pointer to the device. The device has not 817 * had its reference counter increased so the caller must be careful 818 * about locking. The caller must hold RCU lock. 819 */ 820 821 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 822 { 823 struct net_device *dev; 824 struct hlist_head *head = dev_index_hash(net, ifindex); 825 826 hlist_for_each_entry_rcu(dev, head, index_hlist) 827 if (dev->ifindex == ifindex) 828 return dev; 829 830 return NULL; 831 } 832 EXPORT_SYMBOL(dev_get_by_index_rcu); 833 834 835 /** 836 * dev_get_by_index - find a device by its ifindex 837 * @net: the applicable net namespace 838 * @ifindex: index of device 839 * 840 * Search for an interface by index. Returns NULL if the device 841 * is not found or a pointer to the device. The device returned has 842 * had a reference added and the pointer is safe until the user calls 843 * dev_put to indicate they have finished with it. 844 */ 845 846 struct net_device *dev_get_by_index(struct net *net, int ifindex) 847 { 848 struct net_device *dev; 849 850 rcu_read_lock(); 851 dev = dev_get_by_index_rcu(net, ifindex); 852 dev_hold(dev); 853 rcu_read_unlock(); 854 return dev; 855 } 856 EXPORT_SYMBOL(dev_get_by_index); 857 858 /** 859 * dev_get_by_napi_id - find a device by napi_id 860 * @napi_id: ID of the NAPI struct 861 * 862 * Search for an interface by NAPI ID. Returns %NULL if the device 863 * is not found or a pointer to the device. The device has not had 864 * its reference counter increased so the caller must be careful 865 * about locking. The caller must hold RCU lock. 866 */ 867 868 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 869 { 870 struct napi_struct *napi; 871 872 WARN_ON_ONCE(!rcu_read_lock_held()); 873 874 if (napi_id < MIN_NAPI_ID) 875 return NULL; 876 877 napi = napi_by_id(napi_id); 878 879 return napi ? napi->dev : NULL; 880 } 881 EXPORT_SYMBOL(dev_get_by_napi_id); 882 883 /** 884 * netdev_get_name - get a netdevice name, knowing its ifindex. 885 * @net: network namespace 886 * @name: a pointer to the buffer where the name will be stored. 887 * @ifindex: the ifindex of the interface to get the name from. 888 */ 889 int netdev_get_name(struct net *net, char *name, int ifindex) 890 { 891 struct net_device *dev; 892 int ret; 893 894 down_read(&devnet_rename_sem); 895 rcu_read_lock(); 896 897 dev = dev_get_by_index_rcu(net, ifindex); 898 if (!dev) { 899 ret = -ENODEV; 900 goto out; 901 } 902 903 strcpy(name, dev->name); 904 905 ret = 0; 906 out: 907 rcu_read_unlock(); 908 up_read(&devnet_rename_sem); 909 return ret; 910 } 911 912 /** 913 * dev_getbyhwaddr_rcu - find a device by its hardware address 914 * @net: the applicable net namespace 915 * @type: media type of device 916 * @ha: hardware address 917 * 918 * Search for an interface by MAC address. Returns NULL if the device 919 * is not found or a pointer to the device. 920 * The caller must hold RCU or RTNL. 921 * The returned device has not had its ref count increased 922 * and the caller must therefore be careful about locking 923 * 924 */ 925 926 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 927 const char *ha) 928 { 929 struct net_device *dev; 930 931 for_each_netdev_rcu(net, dev) 932 if (dev->type == type && 933 !memcmp(dev->dev_addr, ha, dev->addr_len)) 934 return dev; 935 936 return NULL; 937 } 938 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 939 940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 941 { 942 struct net_device *dev, *ret = NULL; 943 944 rcu_read_lock(); 945 for_each_netdev_rcu(net, dev) 946 if (dev->type == type) { 947 dev_hold(dev); 948 ret = dev; 949 break; 950 } 951 rcu_read_unlock(); 952 return ret; 953 } 954 EXPORT_SYMBOL(dev_getfirstbyhwtype); 955 956 /** 957 * __dev_get_by_flags - find any device with given flags 958 * @net: the applicable net namespace 959 * @if_flags: IFF_* values 960 * @mask: bitmask of bits in if_flags to check 961 * 962 * Search for any interface with the given flags. Returns NULL if a device 963 * is not found or a pointer to the device. Must be called inside 964 * rtnl_lock(), and result refcount is unchanged. 965 */ 966 967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 968 unsigned short mask) 969 { 970 struct net_device *dev, *ret; 971 972 ASSERT_RTNL(); 973 974 ret = NULL; 975 for_each_netdev(net, dev) { 976 if (((dev->flags ^ if_flags) & mask) == 0) { 977 ret = dev; 978 break; 979 } 980 } 981 return ret; 982 } 983 EXPORT_SYMBOL(__dev_get_by_flags); 984 985 /** 986 * dev_valid_name - check if name is okay for network device 987 * @name: name string 988 * 989 * Network device names need to be valid file names to 990 * allow sysfs to work. We also disallow any kind of 991 * whitespace. 992 */ 993 bool dev_valid_name(const char *name) 994 { 995 if (*name == '\0') 996 return false; 997 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 998 return false; 999 if (!strcmp(name, ".") || !strcmp(name, "..")) 1000 return false; 1001 1002 while (*name) { 1003 if (*name == '/' || *name == ':' || isspace(*name)) 1004 return false; 1005 name++; 1006 } 1007 return true; 1008 } 1009 EXPORT_SYMBOL(dev_valid_name); 1010 1011 /** 1012 * __dev_alloc_name - allocate a name for a device 1013 * @net: network namespace to allocate the device name in 1014 * @name: name format string 1015 * @buf: scratch buffer and result name string 1016 * 1017 * Passed a format string - eg "lt%d" it will try and find a suitable 1018 * id. It scans list of devices to build up a free map, then chooses 1019 * the first empty slot. The caller must hold the dev_base or rtnl lock 1020 * while allocating the name and adding the device in order to avoid 1021 * duplicates. 1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1023 * Returns the number of the unit assigned or a negative errno code. 1024 */ 1025 1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1027 { 1028 int i = 0; 1029 const char *p; 1030 const int max_netdevices = 8*PAGE_SIZE; 1031 unsigned long *inuse; 1032 struct net_device *d; 1033 1034 if (!dev_valid_name(name)) 1035 return -EINVAL; 1036 1037 p = strchr(name, '%'); 1038 if (p) { 1039 /* 1040 * Verify the string as this thing may have come from 1041 * the user. There must be either one "%d" and no other "%" 1042 * characters. 1043 */ 1044 if (p[1] != 'd' || strchr(p + 2, '%')) 1045 return -EINVAL; 1046 1047 /* Use one page as a bit array of possible slots */ 1048 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1049 if (!inuse) 1050 return -ENOMEM; 1051 1052 for_each_netdev(net, d) { 1053 struct netdev_name_node *name_node; 1054 list_for_each_entry(name_node, &d->name_node->list, list) { 1055 if (!sscanf(name_node->name, name, &i)) 1056 continue; 1057 if (i < 0 || i >= max_netdevices) 1058 continue; 1059 1060 /* avoid cases where sscanf is not exact inverse of printf */ 1061 snprintf(buf, IFNAMSIZ, name, i); 1062 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1063 __set_bit(i, inuse); 1064 } 1065 if (!sscanf(d->name, name, &i)) 1066 continue; 1067 if (i < 0 || i >= max_netdevices) 1068 continue; 1069 1070 /* avoid cases where sscanf is not exact inverse of printf */ 1071 snprintf(buf, IFNAMSIZ, name, i); 1072 if (!strncmp(buf, d->name, IFNAMSIZ)) 1073 __set_bit(i, inuse); 1074 } 1075 1076 i = find_first_zero_bit(inuse, max_netdevices); 1077 free_page((unsigned long) inuse); 1078 } 1079 1080 snprintf(buf, IFNAMSIZ, name, i); 1081 if (!netdev_name_in_use(net, buf)) 1082 return i; 1083 1084 /* It is possible to run out of possible slots 1085 * when the name is long and there isn't enough space left 1086 * for the digits, or if all bits are used. 1087 */ 1088 return -ENFILE; 1089 } 1090 1091 static int dev_alloc_name_ns(struct net *net, 1092 struct net_device *dev, 1093 const char *name) 1094 { 1095 char buf[IFNAMSIZ]; 1096 int ret; 1097 1098 BUG_ON(!net); 1099 ret = __dev_alloc_name(net, name, buf); 1100 if (ret >= 0) 1101 strscpy(dev->name, buf, IFNAMSIZ); 1102 return ret; 1103 } 1104 1105 /** 1106 * dev_alloc_name - allocate a name for a device 1107 * @dev: device 1108 * @name: name format string 1109 * 1110 * Passed a format string - eg "lt%d" it will try and find a suitable 1111 * id. It scans list of devices to build up a free map, then chooses 1112 * the first empty slot. The caller must hold the dev_base or rtnl lock 1113 * while allocating the name and adding the device in order to avoid 1114 * duplicates. 1115 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1116 * Returns the number of the unit assigned or a negative errno code. 1117 */ 1118 1119 int dev_alloc_name(struct net_device *dev, const char *name) 1120 { 1121 return dev_alloc_name_ns(dev_net(dev), dev, name); 1122 } 1123 EXPORT_SYMBOL(dev_alloc_name); 1124 1125 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1126 const char *name) 1127 { 1128 BUG_ON(!net); 1129 1130 if (!dev_valid_name(name)) 1131 return -EINVAL; 1132 1133 if (strchr(name, '%')) 1134 return dev_alloc_name_ns(net, dev, name); 1135 else if (netdev_name_in_use(net, name)) 1136 return -EEXIST; 1137 else if (dev->name != name) 1138 strscpy(dev->name, name, IFNAMSIZ); 1139 1140 return 0; 1141 } 1142 1143 /** 1144 * dev_change_name - change name of a device 1145 * @dev: device 1146 * @newname: name (or format string) must be at least IFNAMSIZ 1147 * 1148 * Change name of a device, can pass format strings "eth%d". 1149 * for wildcarding. 1150 */ 1151 int dev_change_name(struct net_device *dev, const char *newname) 1152 { 1153 unsigned char old_assign_type; 1154 char oldname[IFNAMSIZ]; 1155 int err = 0; 1156 int ret; 1157 struct net *net; 1158 1159 ASSERT_RTNL(); 1160 BUG_ON(!dev_net(dev)); 1161 1162 net = dev_net(dev); 1163 1164 down_write(&devnet_rename_sem); 1165 1166 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1167 up_write(&devnet_rename_sem); 1168 return 0; 1169 } 1170 1171 memcpy(oldname, dev->name, IFNAMSIZ); 1172 1173 err = dev_get_valid_name(net, dev, newname); 1174 if (err < 0) { 1175 up_write(&devnet_rename_sem); 1176 return err; 1177 } 1178 1179 if (oldname[0] && !strchr(oldname, '%')) 1180 netdev_info(dev, "renamed from %s%s\n", oldname, 1181 dev->flags & IFF_UP ? " (while UP)" : ""); 1182 1183 old_assign_type = dev->name_assign_type; 1184 dev->name_assign_type = NET_NAME_RENAMED; 1185 1186 rollback: 1187 ret = device_rename(&dev->dev, dev->name); 1188 if (ret) { 1189 memcpy(dev->name, oldname, IFNAMSIZ); 1190 dev->name_assign_type = old_assign_type; 1191 up_write(&devnet_rename_sem); 1192 return ret; 1193 } 1194 1195 up_write(&devnet_rename_sem); 1196 1197 netdev_adjacent_rename_links(dev, oldname); 1198 1199 write_lock(&dev_base_lock); 1200 netdev_name_node_del(dev->name_node); 1201 write_unlock(&dev_base_lock); 1202 1203 synchronize_rcu(); 1204 1205 write_lock(&dev_base_lock); 1206 netdev_name_node_add(net, dev->name_node); 1207 write_unlock(&dev_base_lock); 1208 1209 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1210 ret = notifier_to_errno(ret); 1211 1212 if (ret) { 1213 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1214 if (err >= 0) { 1215 err = ret; 1216 down_write(&devnet_rename_sem); 1217 memcpy(dev->name, oldname, IFNAMSIZ); 1218 memcpy(oldname, newname, IFNAMSIZ); 1219 dev->name_assign_type = old_assign_type; 1220 old_assign_type = NET_NAME_RENAMED; 1221 goto rollback; 1222 } else { 1223 netdev_err(dev, "name change rollback failed: %d\n", 1224 ret); 1225 } 1226 } 1227 1228 return err; 1229 } 1230 1231 /** 1232 * dev_set_alias - change ifalias of a device 1233 * @dev: device 1234 * @alias: name up to IFALIASZ 1235 * @len: limit of bytes to copy from info 1236 * 1237 * Set ifalias for a device, 1238 */ 1239 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1240 { 1241 struct dev_ifalias *new_alias = NULL; 1242 1243 if (len >= IFALIASZ) 1244 return -EINVAL; 1245 1246 if (len) { 1247 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1248 if (!new_alias) 1249 return -ENOMEM; 1250 1251 memcpy(new_alias->ifalias, alias, len); 1252 new_alias->ifalias[len] = 0; 1253 } 1254 1255 mutex_lock(&ifalias_mutex); 1256 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1257 mutex_is_locked(&ifalias_mutex)); 1258 mutex_unlock(&ifalias_mutex); 1259 1260 if (new_alias) 1261 kfree_rcu(new_alias, rcuhead); 1262 1263 return len; 1264 } 1265 EXPORT_SYMBOL(dev_set_alias); 1266 1267 /** 1268 * dev_get_alias - get ifalias of a device 1269 * @dev: device 1270 * @name: buffer to store name of ifalias 1271 * @len: size of buffer 1272 * 1273 * get ifalias for a device. Caller must make sure dev cannot go 1274 * away, e.g. rcu read lock or own a reference count to device. 1275 */ 1276 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1277 { 1278 const struct dev_ifalias *alias; 1279 int ret = 0; 1280 1281 rcu_read_lock(); 1282 alias = rcu_dereference(dev->ifalias); 1283 if (alias) 1284 ret = snprintf(name, len, "%s", alias->ifalias); 1285 rcu_read_unlock(); 1286 1287 return ret; 1288 } 1289 1290 /** 1291 * netdev_features_change - device changes features 1292 * @dev: device to cause notification 1293 * 1294 * Called to indicate a device has changed features. 1295 */ 1296 void netdev_features_change(struct net_device *dev) 1297 { 1298 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1299 } 1300 EXPORT_SYMBOL(netdev_features_change); 1301 1302 /** 1303 * netdev_state_change - device changes state 1304 * @dev: device to cause notification 1305 * 1306 * Called to indicate a device has changed state. This function calls 1307 * the notifier chains for netdev_chain and sends a NEWLINK message 1308 * to the routing socket. 1309 */ 1310 void netdev_state_change(struct net_device *dev) 1311 { 1312 if (dev->flags & IFF_UP) { 1313 struct netdev_notifier_change_info change_info = { 1314 .info.dev = dev, 1315 }; 1316 1317 call_netdevice_notifiers_info(NETDEV_CHANGE, 1318 &change_info.info); 1319 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1320 } 1321 } 1322 EXPORT_SYMBOL(netdev_state_change); 1323 1324 /** 1325 * __netdev_notify_peers - notify network peers about existence of @dev, 1326 * to be called when rtnl lock is already held. 1327 * @dev: network device 1328 * 1329 * Generate traffic such that interested network peers are aware of 1330 * @dev, such as by generating a gratuitous ARP. This may be used when 1331 * a device wants to inform the rest of the network about some sort of 1332 * reconfiguration such as a failover event or virtual machine 1333 * migration. 1334 */ 1335 void __netdev_notify_peers(struct net_device *dev) 1336 { 1337 ASSERT_RTNL(); 1338 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1339 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1340 } 1341 EXPORT_SYMBOL(__netdev_notify_peers); 1342 1343 /** 1344 * netdev_notify_peers - notify network peers about existence of @dev 1345 * @dev: network device 1346 * 1347 * Generate traffic such that interested network peers are aware of 1348 * @dev, such as by generating a gratuitous ARP. This may be used when 1349 * a device wants to inform the rest of the network about some sort of 1350 * reconfiguration such as a failover event or virtual machine 1351 * migration. 1352 */ 1353 void netdev_notify_peers(struct net_device *dev) 1354 { 1355 rtnl_lock(); 1356 __netdev_notify_peers(dev); 1357 rtnl_unlock(); 1358 } 1359 EXPORT_SYMBOL(netdev_notify_peers); 1360 1361 static int napi_threaded_poll(void *data); 1362 1363 static int napi_kthread_create(struct napi_struct *n) 1364 { 1365 int err = 0; 1366 1367 /* Create and wake up the kthread once to put it in 1368 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1369 * warning and work with loadavg. 1370 */ 1371 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1372 n->dev->name, n->napi_id); 1373 if (IS_ERR(n->thread)) { 1374 err = PTR_ERR(n->thread); 1375 pr_err("kthread_run failed with err %d\n", err); 1376 n->thread = NULL; 1377 } 1378 1379 return err; 1380 } 1381 1382 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1383 { 1384 const struct net_device_ops *ops = dev->netdev_ops; 1385 int ret; 1386 1387 ASSERT_RTNL(); 1388 dev_addr_check(dev); 1389 1390 if (!netif_device_present(dev)) { 1391 /* may be detached because parent is runtime-suspended */ 1392 if (dev->dev.parent) 1393 pm_runtime_resume(dev->dev.parent); 1394 if (!netif_device_present(dev)) 1395 return -ENODEV; 1396 } 1397 1398 /* Block netpoll from trying to do any rx path servicing. 1399 * If we don't do this there is a chance ndo_poll_controller 1400 * or ndo_poll may be running while we open the device 1401 */ 1402 netpoll_poll_disable(dev); 1403 1404 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1405 ret = notifier_to_errno(ret); 1406 if (ret) 1407 return ret; 1408 1409 set_bit(__LINK_STATE_START, &dev->state); 1410 1411 if (ops->ndo_validate_addr) 1412 ret = ops->ndo_validate_addr(dev); 1413 1414 if (!ret && ops->ndo_open) 1415 ret = ops->ndo_open(dev); 1416 1417 netpoll_poll_enable(dev); 1418 1419 if (ret) 1420 clear_bit(__LINK_STATE_START, &dev->state); 1421 else { 1422 dev->flags |= IFF_UP; 1423 dev_set_rx_mode(dev); 1424 dev_activate(dev); 1425 add_device_randomness(dev->dev_addr, dev->addr_len); 1426 } 1427 1428 return ret; 1429 } 1430 1431 /** 1432 * dev_open - prepare an interface for use. 1433 * @dev: device to open 1434 * @extack: netlink extended ack 1435 * 1436 * Takes a device from down to up state. The device's private open 1437 * function is invoked and then the multicast lists are loaded. Finally 1438 * the device is moved into the up state and a %NETDEV_UP message is 1439 * sent to the netdev notifier chain. 1440 * 1441 * Calling this function on an active interface is a nop. On a failure 1442 * a negative errno code is returned. 1443 */ 1444 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1445 { 1446 int ret; 1447 1448 if (dev->flags & IFF_UP) 1449 return 0; 1450 1451 ret = __dev_open(dev, extack); 1452 if (ret < 0) 1453 return ret; 1454 1455 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1456 call_netdevice_notifiers(NETDEV_UP, dev); 1457 1458 return ret; 1459 } 1460 EXPORT_SYMBOL(dev_open); 1461 1462 static void __dev_close_many(struct list_head *head) 1463 { 1464 struct net_device *dev; 1465 1466 ASSERT_RTNL(); 1467 might_sleep(); 1468 1469 list_for_each_entry(dev, head, close_list) { 1470 /* Temporarily disable netpoll until the interface is down */ 1471 netpoll_poll_disable(dev); 1472 1473 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1474 1475 clear_bit(__LINK_STATE_START, &dev->state); 1476 1477 /* Synchronize to scheduled poll. We cannot touch poll list, it 1478 * can be even on different cpu. So just clear netif_running(). 1479 * 1480 * dev->stop() will invoke napi_disable() on all of it's 1481 * napi_struct instances on this device. 1482 */ 1483 smp_mb__after_atomic(); /* Commit netif_running(). */ 1484 } 1485 1486 dev_deactivate_many(head); 1487 1488 list_for_each_entry(dev, head, close_list) { 1489 const struct net_device_ops *ops = dev->netdev_ops; 1490 1491 /* 1492 * Call the device specific close. This cannot fail. 1493 * Only if device is UP 1494 * 1495 * We allow it to be called even after a DETACH hot-plug 1496 * event. 1497 */ 1498 if (ops->ndo_stop) 1499 ops->ndo_stop(dev); 1500 1501 dev->flags &= ~IFF_UP; 1502 netpoll_poll_enable(dev); 1503 } 1504 } 1505 1506 static void __dev_close(struct net_device *dev) 1507 { 1508 LIST_HEAD(single); 1509 1510 list_add(&dev->close_list, &single); 1511 __dev_close_many(&single); 1512 list_del(&single); 1513 } 1514 1515 void dev_close_many(struct list_head *head, bool unlink) 1516 { 1517 struct net_device *dev, *tmp; 1518 1519 /* Remove the devices that don't need to be closed */ 1520 list_for_each_entry_safe(dev, tmp, head, close_list) 1521 if (!(dev->flags & IFF_UP)) 1522 list_del_init(&dev->close_list); 1523 1524 __dev_close_many(head); 1525 1526 list_for_each_entry_safe(dev, tmp, head, close_list) { 1527 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1528 call_netdevice_notifiers(NETDEV_DOWN, dev); 1529 if (unlink) 1530 list_del_init(&dev->close_list); 1531 } 1532 } 1533 EXPORT_SYMBOL(dev_close_many); 1534 1535 /** 1536 * dev_close - shutdown an interface. 1537 * @dev: device to shutdown 1538 * 1539 * This function moves an active device into down state. A 1540 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1541 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1542 * chain. 1543 */ 1544 void dev_close(struct net_device *dev) 1545 { 1546 if (dev->flags & IFF_UP) { 1547 LIST_HEAD(single); 1548 1549 list_add(&dev->close_list, &single); 1550 dev_close_many(&single, true); 1551 list_del(&single); 1552 } 1553 } 1554 EXPORT_SYMBOL(dev_close); 1555 1556 1557 /** 1558 * dev_disable_lro - disable Large Receive Offload on a device 1559 * @dev: device 1560 * 1561 * Disable Large Receive Offload (LRO) on a net device. Must be 1562 * called under RTNL. This is needed if received packets may be 1563 * forwarded to another interface. 1564 */ 1565 void dev_disable_lro(struct net_device *dev) 1566 { 1567 struct net_device *lower_dev; 1568 struct list_head *iter; 1569 1570 dev->wanted_features &= ~NETIF_F_LRO; 1571 netdev_update_features(dev); 1572 1573 if (unlikely(dev->features & NETIF_F_LRO)) 1574 netdev_WARN(dev, "failed to disable LRO!\n"); 1575 1576 netdev_for_each_lower_dev(dev, lower_dev, iter) 1577 dev_disable_lro(lower_dev); 1578 } 1579 EXPORT_SYMBOL(dev_disable_lro); 1580 1581 /** 1582 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1583 * @dev: device 1584 * 1585 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1586 * called under RTNL. This is needed if Generic XDP is installed on 1587 * the device. 1588 */ 1589 static void dev_disable_gro_hw(struct net_device *dev) 1590 { 1591 dev->wanted_features &= ~NETIF_F_GRO_HW; 1592 netdev_update_features(dev); 1593 1594 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1595 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1596 } 1597 1598 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1599 { 1600 #define N(val) \ 1601 case NETDEV_##val: \ 1602 return "NETDEV_" __stringify(val); 1603 switch (cmd) { 1604 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1605 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1606 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1607 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1608 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1609 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1610 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1611 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1612 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1613 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1614 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1615 N(XDP_FEAT_CHANGE) 1616 } 1617 #undef N 1618 return "UNKNOWN_NETDEV_EVENT"; 1619 } 1620 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1621 1622 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1623 struct net_device *dev) 1624 { 1625 struct netdev_notifier_info info = { 1626 .dev = dev, 1627 }; 1628 1629 return nb->notifier_call(nb, val, &info); 1630 } 1631 1632 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1633 struct net_device *dev) 1634 { 1635 int err; 1636 1637 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1638 err = notifier_to_errno(err); 1639 if (err) 1640 return err; 1641 1642 if (!(dev->flags & IFF_UP)) 1643 return 0; 1644 1645 call_netdevice_notifier(nb, NETDEV_UP, dev); 1646 return 0; 1647 } 1648 1649 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1650 struct net_device *dev) 1651 { 1652 if (dev->flags & IFF_UP) { 1653 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1654 dev); 1655 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1656 } 1657 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1658 } 1659 1660 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1661 struct net *net) 1662 { 1663 struct net_device *dev; 1664 int err; 1665 1666 for_each_netdev(net, dev) { 1667 err = call_netdevice_register_notifiers(nb, dev); 1668 if (err) 1669 goto rollback; 1670 } 1671 return 0; 1672 1673 rollback: 1674 for_each_netdev_continue_reverse(net, dev) 1675 call_netdevice_unregister_notifiers(nb, dev); 1676 return err; 1677 } 1678 1679 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1680 struct net *net) 1681 { 1682 struct net_device *dev; 1683 1684 for_each_netdev(net, dev) 1685 call_netdevice_unregister_notifiers(nb, dev); 1686 } 1687 1688 static int dev_boot_phase = 1; 1689 1690 /** 1691 * register_netdevice_notifier - register a network notifier block 1692 * @nb: notifier 1693 * 1694 * Register a notifier to be called when network device events occur. 1695 * The notifier passed is linked into the kernel structures and must 1696 * not be reused until it has been unregistered. A negative errno code 1697 * is returned on a failure. 1698 * 1699 * When registered all registration and up events are replayed 1700 * to the new notifier to allow device to have a race free 1701 * view of the network device list. 1702 */ 1703 1704 int register_netdevice_notifier(struct notifier_block *nb) 1705 { 1706 struct net *net; 1707 int err; 1708 1709 /* Close race with setup_net() and cleanup_net() */ 1710 down_write(&pernet_ops_rwsem); 1711 rtnl_lock(); 1712 err = raw_notifier_chain_register(&netdev_chain, nb); 1713 if (err) 1714 goto unlock; 1715 if (dev_boot_phase) 1716 goto unlock; 1717 for_each_net(net) { 1718 err = call_netdevice_register_net_notifiers(nb, net); 1719 if (err) 1720 goto rollback; 1721 } 1722 1723 unlock: 1724 rtnl_unlock(); 1725 up_write(&pernet_ops_rwsem); 1726 return err; 1727 1728 rollback: 1729 for_each_net_continue_reverse(net) 1730 call_netdevice_unregister_net_notifiers(nb, net); 1731 1732 raw_notifier_chain_unregister(&netdev_chain, nb); 1733 goto unlock; 1734 } 1735 EXPORT_SYMBOL(register_netdevice_notifier); 1736 1737 /** 1738 * unregister_netdevice_notifier - unregister a network notifier block 1739 * @nb: notifier 1740 * 1741 * Unregister a notifier previously registered by 1742 * register_netdevice_notifier(). The notifier is unlinked into the 1743 * kernel structures and may then be reused. A negative errno code 1744 * is returned on a failure. 1745 * 1746 * After unregistering unregister and down device events are synthesized 1747 * for all devices on the device list to the removed notifier to remove 1748 * the need for special case cleanup code. 1749 */ 1750 1751 int unregister_netdevice_notifier(struct notifier_block *nb) 1752 { 1753 struct net *net; 1754 int err; 1755 1756 /* Close race with setup_net() and cleanup_net() */ 1757 down_write(&pernet_ops_rwsem); 1758 rtnl_lock(); 1759 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1760 if (err) 1761 goto unlock; 1762 1763 for_each_net(net) 1764 call_netdevice_unregister_net_notifiers(nb, net); 1765 1766 unlock: 1767 rtnl_unlock(); 1768 up_write(&pernet_ops_rwsem); 1769 return err; 1770 } 1771 EXPORT_SYMBOL(unregister_netdevice_notifier); 1772 1773 static int __register_netdevice_notifier_net(struct net *net, 1774 struct notifier_block *nb, 1775 bool ignore_call_fail) 1776 { 1777 int err; 1778 1779 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1780 if (err) 1781 return err; 1782 if (dev_boot_phase) 1783 return 0; 1784 1785 err = call_netdevice_register_net_notifiers(nb, net); 1786 if (err && !ignore_call_fail) 1787 goto chain_unregister; 1788 1789 return 0; 1790 1791 chain_unregister: 1792 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1793 return err; 1794 } 1795 1796 static int __unregister_netdevice_notifier_net(struct net *net, 1797 struct notifier_block *nb) 1798 { 1799 int err; 1800 1801 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1802 if (err) 1803 return err; 1804 1805 call_netdevice_unregister_net_notifiers(nb, net); 1806 return 0; 1807 } 1808 1809 /** 1810 * register_netdevice_notifier_net - register a per-netns network notifier block 1811 * @net: network namespace 1812 * @nb: notifier 1813 * 1814 * Register a notifier to be called when network device events occur. 1815 * The notifier passed is linked into the kernel structures and must 1816 * not be reused until it has been unregistered. A negative errno code 1817 * is returned on a failure. 1818 * 1819 * When registered all registration and up events are replayed 1820 * to the new notifier to allow device to have a race free 1821 * view of the network device list. 1822 */ 1823 1824 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1825 { 1826 int err; 1827 1828 rtnl_lock(); 1829 err = __register_netdevice_notifier_net(net, nb, false); 1830 rtnl_unlock(); 1831 return err; 1832 } 1833 EXPORT_SYMBOL(register_netdevice_notifier_net); 1834 1835 /** 1836 * unregister_netdevice_notifier_net - unregister a per-netns 1837 * network notifier block 1838 * @net: network namespace 1839 * @nb: notifier 1840 * 1841 * Unregister a notifier previously registered by 1842 * register_netdevice_notifier_net(). The notifier is unlinked from the 1843 * kernel structures and may then be reused. A negative errno code 1844 * is returned on a failure. 1845 * 1846 * After unregistering unregister and down device events are synthesized 1847 * for all devices on the device list to the removed notifier to remove 1848 * the need for special case cleanup code. 1849 */ 1850 1851 int unregister_netdevice_notifier_net(struct net *net, 1852 struct notifier_block *nb) 1853 { 1854 int err; 1855 1856 rtnl_lock(); 1857 err = __unregister_netdevice_notifier_net(net, nb); 1858 rtnl_unlock(); 1859 return err; 1860 } 1861 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1862 1863 static void __move_netdevice_notifier_net(struct net *src_net, 1864 struct net *dst_net, 1865 struct notifier_block *nb) 1866 { 1867 __unregister_netdevice_notifier_net(src_net, nb); 1868 __register_netdevice_notifier_net(dst_net, nb, true); 1869 } 1870 1871 int register_netdevice_notifier_dev_net(struct net_device *dev, 1872 struct notifier_block *nb, 1873 struct netdev_net_notifier *nn) 1874 { 1875 int err; 1876 1877 rtnl_lock(); 1878 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1879 if (!err) { 1880 nn->nb = nb; 1881 list_add(&nn->list, &dev->net_notifier_list); 1882 } 1883 rtnl_unlock(); 1884 return err; 1885 } 1886 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1887 1888 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1889 struct notifier_block *nb, 1890 struct netdev_net_notifier *nn) 1891 { 1892 int err; 1893 1894 rtnl_lock(); 1895 list_del(&nn->list); 1896 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1897 rtnl_unlock(); 1898 return err; 1899 } 1900 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1901 1902 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1903 struct net *net) 1904 { 1905 struct netdev_net_notifier *nn; 1906 1907 list_for_each_entry(nn, &dev->net_notifier_list, list) 1908 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1909 } 1910 1911 /** 1912 * call_netdevice_notifiers_info - call all network notifier blocks 1913 * @val: value passed unmodified to notifier function 1914 * @info: notifier information data 1915 * 1916 * Call all network notifier blocks. Parameters and return value 1917 * are as for raw_notifier_call_chain(). 1918 */ 1919 1920 int call_netdevice_notifiers_info(unsigned long val, 1921 struct netdev_notifier_info *info) 1922 { 1923 struct net *net = dev_net(info->dev); 1924 int ret; 1925 1926 ASSERT_RTNL(); 1927 1928 /* Run per-netns notifier block chain first, then run the global one. 1929 * Hopefully, one day, the global one is going to be removed after 1930 * all notifier block registrators get converted to be per-netns. 1931 */ 1932 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1933 if (ret & NOTIFY_STOP_MASK) 1934 return ret; 1935 return raw_notifier_call_chain(&netdev_chain, val, info); 1936 } 1937 1938 /** 1939 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1940 * for and rollback on error 1941 * @val_up: value passed unmodified to notifier function 1942 * @val_down: value passed unmodified to the notifier function when 1943 * recovering from an error on @val_up 1944 * @info: notifier information data 1945 * 1946 * Call all per-netns network notifier blocks, but not notifier blocks on 1947 * the global notifier chain. Parameters and return value are as for 1948 * raw_notifier_call_chain_robust(). 1949 */ 1950 1951 static int 1952 call_netdevice_notifiers_info_robust(unsigned long val_up, 1953 unsigned long val_down, 1954 struct netdev_notifier_info *info) 1955 { 1956 struct net *net = dev_net(info->dev); 1957 1958 ASSERT_RTNL(); 1959 1960 return raw_notifier_call_chain_robust(&net->netdev_chain, 1961 val_up, val_down, info); 1962 } 1963 1964 static int call_netdevice_notifiers_extack(unsigned long val, 1965 struct net_device *dev, 1966 struct netlink_ext_ack *extack) 1967 { 1968 struct netdev_notifier_info info = { 1969 .dev = dev, 1970 .extack = extack, 1971 }; 1972 1973 return call_netdevice_notifiers_info(val, &info); 1974 } 1975 1976 /** 1977 * call_netdevice_notifiers - call all network notifier blocks 1978 * @val: value passed unmodified to notifier function 1979 * @dev: net_device pointer passed unmodified to notifier function 1980 * 1981 * Call all network notifier blocks. Parameters and return value 1982 * are as for raw_notifier_call_chain(). 1983 */ 1984 1985 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1986 { 1987 return call_netdevice_notifiers_extack(val, dev, NULL); 1988 } 1989 EXPORT_SYMBOL(call_netdevice_notifiers); 1990 1991 /** 1992 * call_netdevice_notifiers_mtu - call all network notifier blocks 1993 * @val: value passed unmodified to notifier function 1994 * @dev: net_device pointer passed unmodified to notifier function 1995 * @arg: additional u32 argument passed to the notifier function 1996 * 1997 * Call all network notifier blocks. Parameters and return value 1998 * are as for raw_notifier_call_chain(). 1999 */ 2000 static int call_netdevice_notifiers_mtu(unsigned long val, 2001 struct net_device *dev, u32 arg) 2002 { 2003 struct netdev_notifier_info_ext info = { 2004 .info.dev = dev, 2005 .ext.mtu = arg, 2006 }; 2007 2008 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2009 2010 return call_netdevice_notifiers_info(val, &info.info); 2011 } 2012 2013 #ifdef CONFIG_NET_INGRESS 2014 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2015 2016 void net_inc_ingress_queue(void) 2017 { 2018 static_branch_inc(&ingress_needed_key); 2019 } 2020 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2021 2022 void net_dec_ingress_queue(void) 2023 { 2024 static_branch_dec(&ingress_needed_key); 2025 } 2026 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2027 #endif 2028 2029 #ifdef CONFIG_NET_EGRESS 2030 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2031 2032 void net_inc_egress_queue(void) 2033 { 2034 static_branch_inc(&egress_needed_key); 2035 } 2036 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2037 2038 void net_dec_egress_queue(void) 2039 { 2040 static_branch_dec(&egress_needed_key); 2041 } 2042 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2043 #endif 2044 2045 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2046 EXPORT_SYMBOL(netstamp_needed_key); 2047 #ifdef CONFIG_JUMP_LABEL 2048 static atomic_t netstamp_needed_deferred; 2049 static atomic_t netstamp_wanted; 2050 static void netstamp_clear(struct work_struct *work) 2051 { 2052 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2053 int wanted; 2054 2055 wanted = atomic_add_return(deferred, &netstamp_wanted); 2056 if (wanted > 0) 2057 static_branch_enable(&netstamp_needed_key); 2058 else 2059 static_branch_disable(&netstamp_needed_key); 2060 } 2061 static DECLARE_WORK(netstamp_work, netstamp_clear); 2062 #endif 2063 2064 void net_enable_timestamp(void) 2065 { 2066 #ifdef CONFIG_JUMP_LABEL 2067 int wanted = atomic_read(&netstamp_wanted); 2068 2069 while (wanted > 0) { 2070 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2071 return; 2072 } 2073 atomic_inc(&netstamp_needed_deferred); 2074 schedule_work(&netstamp_work); 2075 #else 2076 static_branch_inc(&netstamp_needed_key); 2077 #endif 2078 } 2079 EXPORT_SYMBOL(net_enable_timestamp); 2080 2081 void net_disable_timestamp(void) 2082 { 2083 #ifdef CONFIG_JUMP_LABEL 2084 int wanted = atomic_read(&netstamp_wanted); 2085 2086 while (wanted > 1) { 2087 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2088 return; 2089 } 2090 atomic_dec(&netstamp_needed_deferred); 2091 schedule_work(&netstamp_work); 2092 #else 2093 static_branch_dec(&netstamp_needed_key); 2094 #endif 2095 } 2096 EXPORT_SYMBOL(net_disable_timestamp); 2097 2098 static inline void net_timestamp_set(struct sk_buff *skb) 2099 { 2100 skb->tstamp = 0; 2101 skb->mono_delivery_time = 0; 2102 if (static_branch_unlikely(&netstamp_needed_key)) 2103 skb->tstamp = ktime_get_real(); 2104 } 2105 2106 #define net_timestamp_check(COND, SKB) \ 2107 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2108 if ((COND) && !(SKB)->tstamp) \ 2109 (SKB)->tstamp = ktime_get_real(); \ 2110 } \ 2111 2112 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2113 { 2114 return __is_skb_forwardable(dev, skb, true); 2115 } 2116 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2117 2118 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2119 bool check_mtu) 2120 { 2121 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2122 2123 if (likely(!ret)) { 2124 skb->protocol = eth_type_trans(skb, dev); 2125 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2126 } 2127 2128 return ret; 2129 } 2130 2131 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2132 { 2133 return __dev_forward_skb2(dev, skb, true); 2134 } 2135 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2136 2137 /** 2138 * dev_forward_skb - loopback an skb to another netif 2139 * 2140 * @dev: destination network device 2141 * @skb: buffer to forward 2142 * 2143 * return values: 2144 * NET_RX_SUCCESS (no congestion) 2145 * NET_RX_DROP (packet was dropped, but freed) 2146 * 2147 * dev_forward_skb can be used for injecting an skb from the 2148 * start_xmit function of one device into the receive queue 2149 * of another device. 2150 * 2151 * The receiving device may be in another namespace, so 2152 * we have to clear all information in the skb that could 2153 * impact namespace isolation. 2154 */ 2155 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2156 { 2157 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2158 } 2159 EXPORT_SYMBOL_GPL(dev_forward_skb); 2160 2161 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2162 { 2163 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2164 } 2165 2166 static inline int deliver_skb(struct sk_buff *skb, 2167 struct packet_type *pt_prev, 2168 struct net_device *orig_dev) 2169 { 2170 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2171 return -ENOMEM; 2172 refcount_inc(&skb->users); 2173 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2174 } 2175 2176 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2177 struct packet_type **pt, 2178 struct net_device *orig_dev, 2179 __be16 type, 2180 struct list_head *ptype_list) 2181 { 2182 struct packet_type *ptype, *pt_prev = *pt; 2183 2184 list_for_each_entry_rcu(ptype, ptype_list, list) { 2185 if (ptype->type != type) 2186 continue; 2187 if (pt_prev) 2188 deliver_skb(skb, pt_prev, orig_dev); 2189 pt_prev = ptype; 2190 } 2191 *pt = pt_prev; 2192 } 2193 2194 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2195 { 2196 if (!ptype->af_packet_priv || !skb->sk) 2197 return false; 2198 2199 if (ptype->id_match) 2200 return ptype->id_match(ptype, skb->sk); 2201 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2202 return true; 2203 2204 return false; 2205 } 2206 2207 /** 2208 * dev_nit_active - return true if any network interface taps are in use 2209 * 2210 * @dev: network device to check for the presence of taps 2211 */ 2212 bool dev_nit_active(struct net_device *dev) 2213 { 2214 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2215 } 2216 EXPORT_SYMBOL_GPL(dev_nit_active); 2217 2218 /* 2219 * Support routine. Sends outgoing frames to any network 2220 * taps currently in use. 2221 */ 2222 2223 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2224 { 2225 struct packet_type *ptype; 2226 struct sk_buff *skb2 = NULL; 2227 struct packet_type *pt_prev = NULL; 2228 struct list_head *ptype_list = &ptype_all; 2229 2230 rcu_read_lock(); 2231 again: 2232 list_for_each_entry_rcu(ptype, ptype_list, list) { 2233 if (ptype->ignore_outgoing) 2234 continue; 2235 2236 /* Never send packets back to the socket 2237 * they originated from - MvS (miquels@drinkel.ow.org) 2238 */ 2239 if (skb_loop_sk(ptype, skb)) 2240 continue; 2241 2242 if (pt_prev) { 2243 deliver_skb(skb2, pt_prev, skb->dev); 2244 pt_prev = ptype; 2245 continue; 2246 } 2247 2248 /* need to clone skb, done only once */ 2249 skb2 = skb_clone(skb, GFP_ATOMIC); 2250 if (!skb2) 2251 goto out_unlock; 2252 2253 net_timestamp_set(skb2); 2254 2255 /* skb->nh should be correctly 2256 * set by sender, so that the second statement is 2257 * just protection against buggy protocols. 2258 */ 2259 skb_reset_mac_header(skb2); 2260 2261 if (skb_network_header(skb2) < skb2->data || 2262 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2263 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2264 ntohs(skb2->protocol), 2265 dev->name); 2266 skb_reset_network_header(skb2); 2267 } 2268 2269 skb2->transport_header = skb2->network_header; 2270 skb2->pkt_type = PACKET_OUTGOING; 2271 pt_prev = ptype; 2272 } 2273 2274 if (ptype_list == &ptype_all) { 2275 ptype_list = &dev->ptype_all; 2276 goto again; 2277 } 2278 out_unlock: 2279 if (pt_prev) { 2280 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2281 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2282 else 2283 kfree_skb(skb2); 2284 } 2285 rcu_read_unlock(); 2286 } 2287 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2288 2289 /** 2290 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2291 * @dev: Network device 2292 * @txq: number of queues available 2293 * 2294 * If real_num_tx_queues is changed the tc mappings may no longer be 2295 * valid. To resolve this verify the tc mapping remains valid and if 2296 * not NULL the mapping. With no priorities mapping to this 2297 * offset/count pair it will no longer be used. In the worst case TC0 2298 * is invalid nothing can be done so disable priority mappings. If is 2299 * expected that drivers will fix this mapping if they can before 2300 * calling netif_set_real_num_tx_queues. 2301 */ 2302 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2303 { 2304 int i; 2305 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2306 2307 /* If TC0 is invalidated disable TC mapping */ 2308 if (tc->offset + tc->count > txq) { 2309 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2310 dev->num_tc = 0; 2311 return; 2312 } 2313 2314 /* Invalidated prio to tc mappings set to TC0 */ 2315 for (i = 1; i < TC_BITMASK + 1; i++) { 2316 int q = netdev_get_prio_tc_map(dev, i); 2317 2318 tc = &dev->tc_to_txq[q]; 2319 if (tc->offset + tc->count > txq) { 2320 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2321 i, q); 2322 netdev_set_prio_tc_map(dev, i, 0); 2323 } 2324 } 2325 } 2326 2327 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2328 { 2329 if (dev->num_tc) { 2330 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2331 int i; 2332 2333 /* walk through the TCs and see if it falls into any of them */ 2334 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2335 if ((txq - tc->offset) < tc->count) 2336 return i; 2337 } 2338 2339 /* didn't find it, just return -1 to indicate no match */ 2340 return -1; 2341 } 2342 2343 return 0; 2344 } 2345 EXPORT_SYMBOL(netdev_txq_to_tc); 2346 2347 #ifdef CONFIG_XPS 2348 static struct static_key xps_needed __read_mostly; 2349 static struct static_key xps_rxqs_needed __read_mostly; 2350 static DEFINE_MUTEX(xps_map_mutex); 2351 #define xmap_dereference(P) \ 2352 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2353 2354 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2355 struct xps_dev_maps *old_maps, int tci, u16 index) 2356 { 2357 struct xps_map *map = NULL; 2358 int pos; 2359 2360 if (dev_maps) 2361 map = xmap_dereference(dev_maps->attr_map[tci]); 2362 if (!map) 2363 return false; 2364 2365 for (pos = map->len; pos--;) { 2366 if (map->queues[pos] != index) 2367 continue; 2368 2369 if (map->len > 1) { 2370 map->queues[pos] = map->queues[--map->len]; 2371 break; 2372 } 2373 2374 if (old_maps) 2375 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2376 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2377 kfree_rcu(map, rcu); 2378 return false; 2379 } 2380 2381 return true; 2382 } 2383 2384 static bool remove_xps_queue_cpu(struct net_device *dev, 2385 struct xps_dev_maps *dev_maps, 2386 int cpu, u16 offset, u16 count) 2387 { 2388 int num_tc = dev_maps->num_tc; 2389 bool active = false; 2390 int tci; 2391 2392 for (tci = cpu * num_tc; num_tc--; tci++) { 2393 int i, j; 2394 2395 for (i = count, j = offset; i--; j++) { 2396 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2397 break; 2398 } 2399 2400 active |= i < 0; 2401 } 2402 2403 return active; 2404 } 2405 2406 static void reset_xps_maps(struct net_device *dev, 2407 struct xps_dev_maps *dev_maps, 2408 enum xps_map_type type) 2409 { 2410 static_key_slow_dec_cpuslocked(&xps_needed); 2411 if (type == XPS_RXQS) 2412 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2413 2414 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2415 2416 kfree_rcu(dev_maps, rcu); 2417 } 2418 2419 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2420 u16 offset, u16 count) 2421 { 2422 struct xps_dev_maps *dev_maps; 2423 bool active = false; 2424 int i, j; 2425 2426 dev_maps = xmap_dereference(dev->xps_maps[type]); 2427 if (!dev_maps) 2428 return; 2429 2430 for (j = 0; j < dev_maps->nr_ids; j++) 2431 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2432 if (!active) 2433 reset_xps_maps(dev, dev_maps, type); 2434 2435 if (type == XPS_CPUS) { 2436 for (i = offset + (count - 1); count--; i--) 2437 netdev_queue_numa_node_write( 2438 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2439 } 2440 } 2441 2442 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2443 u16 count) 2444 { 2445 if (!static_key_false(&xps_needed)) 2446 return; 2447 2448 cpus_read_lock(); 2449 mutex_lock(&xps_map_mutex); 2450 2451 if (static_key_false(&xps_rxqs_needed)) 2452 clean_xps_maps(dev, XPS_RXQS, offset, count); 2453 2454 clean_xps_maps(dev, XPS_CPUS, offset, count); 2455 2456 mutex_unlock(&xps_map_mutex); 2457 cpus_read_unlock(); 2458 } 2459 2460 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2461 { 2462 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2463 } 2464 2465 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2466 u16 index, bool is_rxqs_map) 2467 { 2468 struct xps_map *new_map; 2469 int alloc_len = XPS_MIN_MAP_ALLOC; 2470 int i, pos; 2471 2472 for (pos = 0; map && pos < map->len; pos++) { 2473 if (map->queues[pos] != index) 2474 continue; 2475 return map; 2476 } 2477 2478 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2479 if (map) { 2480 if (pos < map->alloc_len) 2481 return map; 2482 2483 alloc_len = map->alloc_len * 2; 2484 } 2485 2486 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2487 * map 2488 */ 2489 if (is_rxqs_map) 2490 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2491 else 2492 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2493 cpu_to_node(attr_index)); 2494 if (!new_map) 2495 return NULL; 2496 2497 for (i = 0; i < pos; i++) 2498 new_map->queues[i] = map->queues[i]; 2499 new_map->alloc_len = alloc_len; 2500 new_map->len = pos; 2501 2502 return new_map; 2503 } 2504 2505 /* Copy xps maps at a given index */ 2506 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2507 struct xps_dev_maps *new_dev_maps, int index, 2508 int tc, bool skip_tc) 2509 { 2510 int i, tci = index * dev_maps->num_tc; 2511 struct xps_map *map; 2512 2513 /* copy maps belonging to foreign traffic classes */ 2514 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2515 if (i == tc && skip_tc) 2516 continue; 2517 2518 /* fill in the new device map from the old device map */ 2519 map = xmap_dereference(dev_maps->attr_map[tci]); 2520 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2521 } 2522 } 2523 2524 /* Must be called under cpus_read_lock */ 2525 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2526 u16 index, enum xps_map_type type) 2527 { 2528 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2529 const unsigned long *online_mask = NULL; 2530 bool active = false, copy = false; 2531 int i, j, tci, numa_node_id = -2; 2532 int maps_sz, num_tc = 1, tc = 0; 2533 struct xps_map *map, *new_map; 2534 unsigned int nr_ids; 2535 2536 WARN_ON_ONCE(index >= dev->num_tx_queues); 2537 2538 if (dev->num_tc) { 2539 /* Do not allow XPS on subordinate device directly */ 2540 num_tc = dev->num_tc; 2541 if (num_tc < 0) 2542 return -EINVAL; 2543 2544 /* If queue belongs to subordinate dev use its map */ 2545 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2546 2547 tc = netdev_txq_to_tc(dev, index); 2548 if (tc < 0) 2549 return -EINVAL; 2550 } 2551 2552 mutex_lock(&xps_map_mutex); 2553 2554 dev_maps = xmap_dereference(dev->xps_maps[type]); 2555 if (type == XPS_RXQS) { 2556 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2557 nr_ids = dev->num_rx_queues; 2558 } else { 2559 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2560 if (num_possible_cpus() > 1) 2561 online_mask = cpumask_bits(cpu_online_mask); 2562 nr_ids = nr_cpu_ids; 2563 } 2564 2565 if (maps_sz < L1_CACHE_BYTES) 2566 maps_sz = L1_CACHE_BYTES; 2567 2568 /* The old dev_maps could be larger or smaller than the one we're 2569 * setting up now, as dev->num_tc or nr_ids could have been updated in 2570 * between. We could try to be smart, but let's be safe instead and only 2571 * copy foreign traffic classes if the two map sizes match. 2572 */ 2573 if (dev_maps && 2574 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2575 copy = true; 2576 2577 /* allocate memory for queue storage */ 2578 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2579 j < nr_ids;) { 2580 if (!new_dev_maps) { 2581 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2582 if (!new_dev_maps) { 2583 mutex_unlock(&xps_map_mutex); 2584 return -ENOMEM; 2585 } 2586 2587 new_dev_maps->nr_ids = nr_ids; 2588 new_dev_maps->num_tc = num_tc; 2589 } 2590 2591 tci = j * num_tc + tc; 2592 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2593 2594 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2595 if (!map) 2596 goto error; 2597 2598 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2599 } 2600 2601 if (!new_dev_maps) 2602 goto out_no_new_maps; 2603 2604 if (!dev_maps) { 2605 /* Increment static keys at most once per type */ 2606 static_key_slow_inc_cpuslocked(&xps_needed); 2607 if (type == XPS_RXQS) 2608 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2609 } 2610 2611 for (j = 0; j < nr_ids; j++) { 2612 bool skip_tc = false; 2613 2614 tci = j * num_tc + tc; 2615 if (netif_attr_test_mask(j, mask, nr_ids) && 2616 netif_attr_test_online(j, online_mask, nr_ids)) { 2617 /* add tx-queue to CPU/rx-queue maps */ 2618 int pos = 0; 2619 2620 skip_tc = true; 2621 2622 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2623 while ((pos < map->len) && (map->queues[pos] != index)) 2624 pos++; 2625 2626 if (pos == map->len) 2627 map->queues[map->len++] = index; 2628 #ifdef CONFIG_NUMA 2629 if (type == XPS_CPUS) { 2630 if (numa_node_id == -2) 2631 numa_node_id = cpu_to_node(j); 2632 else if (numa_node_id != cpu_to_node(j)) 2633 numa_node_id = -1; 2634 } 2635 #endif 2636 } 2637 2638 if (copy) 2639 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2640 skip_tc); 2641 } 2642 2643 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2644 2645 /* Cleanup old maps */ 2646 if (!dev_maps) 2647 goto out_no_old_maps; 2648 2649 for (j = 0; j < dev_maps->nr_ids; j++) { 2650 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2651 map = xmap_dereference(dev_maps->attr_map[tci]); 2652 if (!map) 2653 continue; 2654 2655 if (copy) { 2656 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2657 if (map == new_map) 2658 continue; 2659 } 2660 2661 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2662 kfree_rcu(map, rcu); 2663 } 2664 } 2665 2666 old_dev_maps = dev_maps; 2667 2668 out_no_old_maps: 2669 dev_maps = new_dev_maps; 2670 active = true; 2671 2672 out_no_new_maps: 2673 if (type == XPS_CPUS) 2674 /* update Tx queue numa node */ 2675 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2676 (numa_node_id >= 0) ? 2677 numa_node_id : NUMA_NO_NODE); 2678 2679 if (!dev_maps) 2680 goto out_no_maps; 2681 2682 /* removes tx-queue from unused CPUs/rx-queues */ 2683 for (j = 0; j < dev_maps->nr_ids; j++) { 2684 tci = j * dev_maps->num_tc; 2685 2686 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2687 if (i == tc && 2688 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2689 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2690 continue; 2691 2692 active |= remove_xps_queue(dev_maps, 2693 copy ? old_dev_maps : NULL, 2694 tci, index); 2695 } 2696 } 2697 2698 if (old_dev_maps) 2699 kfree_rcu(old_dev_maps, rcu); 2700 2701 /* free map if not active */ 2702 if (!active) 2703 reset_xps_maps(dev, dev_maps, type); 2704 2705 out_no_maps: 2706 mutex_unlock(&xps_map_mutex); 2707 2708 return 0; 2709 error: 2710 /* remove any maps that we added */ 2711 for (j = 0; j < nr_ids; j++) { 2712 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2713 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2714 map = copy ? 2715 xmap_dereference(dev_maps->attr_map[tci]) : 2716 NULL; 2717 if (new_map && new_map != map) 2718 kfree(new_map); 2719 } 2720 } 2721 2722 mutex_unlock(&xps_map_mutex); 2723 2724 kfree(new_dev_maps); 2725 return -ENOMEM; 2726 } 2727 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2728 2729 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2730 u16 index) 2731 { 2732 int ret; 2733 2734 cpus_read_lock(); 2735 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2736 cpus_read_unlock(); 2737 2738 return ret; 2739 } 2740 EXPORT_SYMBOL(netif_set_xps_queue); 2741 2742 #endif 2743 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2744 { 2745 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2746 2747 /* Unbind any subordinate channels */ 2748 while (txq-- != &dev->_tx[0]) { 2749 if (txq->sb_dev) 2750 netdev_unbind_sb_channel(dev, txq->sb_dev); 2751 } 2752 } 2753 2754 void netdev_reset_tc(struct net_device *dev) 2755 { 2756 #ifdef CONFIG_XPS 2757 netif_reset_xps_queues_gt(dev, 0); 2758 #endif 2759 netdev_unbind_all_sb_channels(dev); 2760 2761 /* Reset TC configuration of device */ 2762 dev->num_tc = 0; 2763 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2764 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2765 } 2766 EXPORT_SYMBOL(netdev_reset_tc); 2767 2768 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2769 { 2770 if (tc >= dev->num_tc) 2771 return -EINVAL; 2772 2773 #ifdef CONFIG_XPS 2774 netif_reset_xps_queues(dev, offset, count); 2775 #endif 2776 dev->tc_to_txq[tc].count = count; 2777 dev->tc_to_txq[tc].offset = offset; 2778 return 0; 2779 } 2780 EXPORT_SYMBOL(netdev_set_tc_queue); 2781 2782 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2783 { 2784 if (num_tc > TC_MAX_QUEUE) 2785 return -EINVAL; 2786 2787 #ifdef CONFIG_XPS 2788 netif_reset_xps_queues_gt(dev, 0); 2789 #endif 2790 netdev_unbind_all_sb_channels(dev); 2791 2792 dev->num_tc = num_tc; 2793 return 0; 2794 } 2795 EXPORT_SYMBOL(netdev_set_num_tc); 2796 2797 void netdev_unbind_sb_channel(struct net_device *dev, 2798 struct net_device *sb_dev) 2799 { 2800 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2801 2802 #ifdef CONFIG_XPS 2803 netif_reset_xps_queues_gt(sb_dev, 0); 2804 #endif 2805 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2806 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2807 2808 while (txq-- != &dev->_tx[0]) { 2809 if (txq->sb_dev == sb_dev) 2810 txq->sb_dev = NULL; 2811 } 2812 } 2813 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2814 2815 int netdev_bind_sb_channel_queue(struct net_device *dev, 2816 struct net_device *sb_dev, 2817 u8 tc, u16 count, u16 offset) 2818 { 2819 /* Make certain the sb_dev and dev are already configured */ 2820 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2821 return -EINVAL; 2822 2823 /* We cannot hand out queues we don't have */ 2824 if ((offset + count) > dev->real_num_tx_queues) 2825 return -EINVAL; 2826 2827 /* Record the mapping */ 2828 sb_dev->tc_to_txq[tc].count = count; 2829 sb_dev->tc_to_txq[tc].offset = offset; 2830 2831 /* Provide a way for Tx queue to find the tc_to_txq map or 2832 * XPS map for itself. 2833 */ 2834 while (count--) 2835 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2836 2837 return 0; 2838 } 2839 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2840 2841 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2842 { 2843 /* Do not use a multiqueue device to represent a subordinate channel */ 2844 if (netif_is_multiqueue(dev)) 2845 return -ENODEV; 2846 2847 /* We allow channels 1 - 32767 to be used for subordinate channels. 2848 * Channel 0 is meant to be "native" mode and used only to represent 2849 * the main root device. We allow writing 0 to reset the device back 2850 * to normal mode after being used as a subordinate channel. 2851 */ 2852 if (channel > S16_MAX) 2853 return -EINVAL; 2854 2855 dev->num_tc = -channel; 2856 2857 return 0; 2858 } 2859 EXPORT_SYMBOL(netdev_set_sb_channel); 2860 2861 /* 2862 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2863 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2864 */ 2865 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2866 { 2867 bool disabling; 2868 int rc; 2869 2870 disabling = txq < dev->real_num_tx_queues; 2871 2872 if (txq < 1 || txq > dev->num_tx_queues) 2873 return -EINVAL; 2874 2875 if (dev->reg_state == NETREG_REGISTERED || 2876 dev->reg_state == NETREG_UNREGISTERING) { 2877 ASSERT_RTNL(); 2878 2879 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2880 txq); 2881 if (rc) 2882 return rc; 2883 2884 if (dev->num_tc) 2885 netif_setup_tc(dev, txq); 2886 2887 dev_qdisc_change_real_num_tx(dev, txq); 2888 2889 dev->real_num_tx_queues = txq; 2890 2891 if (disabling) { 2892 synchronize_net(); 2893 qdisc_reset_all_tx_gt(dev, txq); 2894 #ifdef CONFIG_XPS 2895 netif_reset_xps_queues_gt(dev, txq); 2896 #endif 2897 } 2898 } else { 2899 dev->real_num_tx_queues = txq; 2900 } 2901 2902 return 0; 2903 } 2904 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2905 2906 #ifdef CONFIG_SYSFS 2907 /** 2908 * netif_set_real_num_rx_queues - set actual number of RX queues used 2909 * @dev: Network device 2910 * @rxq: Actual number of RX queues 2911 * 2912 * This must be called either with the rtnl_lock held or before 2913 * registration of the net device. Returns 0 on success, or a 2914 * negative error code. If called before registration, it always 2915 * succeeds. 2916 */ 2917 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2918 { 2919 int rc; 2920 2921 if (rxq < 1 || rxq > dev->num_rx_queues) 2922 return -EINVAL; 2923 2924 if (dev->reg_state == NETREG_REGISTERED) { 2925 ASSERT_RTNL(); 2926 2927 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2928 rxq); 2929 if (rc) 2930 return rc; 2931 } 2932 2933 dev->real_num_rx_queues = rxq; 2934 return 0; 2935 } 2936 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2937 #endif 2938 2939 /** 2940 * netif_set_real_num_queues - set actual number of RX and TX queues used 2941 * @dev: Network device 2942 * @txq: Actual number of TX queues 2943 * @rxq: Actual number of RX queues 2944 * 2945 * Set the real number of both TX and RX queues. 2946 * Does nothing if the number of queues is already correct. 2947 */ 2948 int netif_set_real_num_queues(struct net_device *dev, 2949 unsigned int txq, unsigned int rxq) 2950 { 2951 unsigned int old_rxq = dev->real_num_rx_queues; 2952 int err; 2953 2954 if (txq < 1 || txq > dev->num_tx_queues || 2955 rxq < 1 || rxq > dev->num_rx_queues) 2956 return -EINVAL; 2957 2958 /* Start from increases, so the error path only does decreases - 2959 * decreases can't fail. 2960 */ 2961 if (rxq > dev->real_num_rx_queues) { 2962 err = netif_set_real_num_rx_queues(dev, rxq); 2963 if (err) 2964 return err; 2965 } 2966 if (txq > dev->real_num_tx_queues) { 2967 err = netif_set_real_num_tx_queues(dev, txq); 2968 if (err) 2969 goto undo_rx; 2970 } 2971 if (rxq < dev->real_num_rx_queues) 2972 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2973 if (txq < dev->real_num_tx_queues) 2974 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2975 2976 return 0; 2977 undo_rx: 2978 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2979 return err; 2980 } 2981 EXPORT_SYMBOL(netif_set_real_num_queues); 2982 2983 /** 2984 * netif_set_tso_max_size() - set the max size of TSO frames supported 2985 * @dev: netdev to update 2986 * @size: max skb->len of a TSO frame 2987 * 2988 * Set the limit on the size of TSO super-frames the device can handle. 2989 * Unless explicitly set the stack will assume the value of 2990 * %GSO_LEGACY_MAX_SIZE. 2991 */ 2992 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 2993 { 2994 dev->tso_max_size = min(GSO_MAX_SIZE, size); 2995 if (size < READ_ONCE(dev->gso_max_size)) 2996 netif_set_gso_max_size(dev, size); 2997 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 2998 netif_set_gso_ipv4_max_size(dev, size); 2999 } 3000 EXPORT_SYMBOL(netif_set_tso_max_size); 3001 3002 /** 3003 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3004 * @dev: netdev to update 3005 * @segs: max number of TCP segments 3006 * 3007 * Set the limit on the number of TCP segments the device can generate from 3008 * a single TSO super-frame. 3009 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3010 */ 3011 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3012 { 3013 dev->tso_max_segs = segs; 3014 if (segs < READ_ONCE(dev->gso_max_segs)) 3015 netif_set_gso_max_segs(dev, segs); 3016 } 3017 EXPORT_SYMBOL(netif_set_tso_max_segs); 3018 3019 /** 3020 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3021 * @to: netdev to update 3022 * @from: netdev from which to copy the limits 3023 */ 3024 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3025 { 3026 netif_set_tso_max_size(to, from->tso_max_size); 3027 netif_set_tso_max_segs(to, from->tso_max_segs); 3028 } 3029 EXPORT_SYMBOL(netif_inherit_tso_max); 3030 3031 /** 3032 * netif_get_num_default_rss_queues - default number of RSS queues 3033 * 3034 * Default value is the number of physical cores if there are only 1 or 2, or 3035 * divided by 2 if there are more. 3036 */ 3037 int netif_get_num_default_rss_queues(void) 3038 { 3039 cpumask_var_t cpus; 3040 int cpu, count = 0; 3041 3042 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3043 return 1; 3044 3045 cpumask_copy(cpus, cpu_online_mask); 3046 for_each_cpu(cpu, cpus) { 3047 ++count; 3048 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3049 } 3050 free_cpumask_var(cpus); 3051 3052 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3053 } 3054 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3055 3056 static void __netif_reschedule(struct Qdisc *q) 3057 { 3058 struct softnet_data *sd; 3059 unsigned long flags; 3060 3061 local_irq_save(flags); 3062 sd = this_cpu_ptr(&softnet_data); 3063 q->next_sched = NULL; 3064 *sd->output_queue_tailp = q; 3065 sd->output_queue_tailp = &q->next_sched; 3066 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3067 local_irq_restore(flags); 3068 } 3069 3070 void __netif_schedule(struct Qdisc *q) 3071 { 3072 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3073 __netif_reschedule(q); 3074 } 3075 EXPORT_SYMBOL(__netif_schedule); 3076 3077 struct dev_kfree_skb_cb { 3078 enum skb_drop_reason reason; 3079 }; 3080 3081 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3082 { 3083 return (struct dev_kfree_skb_cb *)skb->cb; 3084 } 3085 3086 void netif_schedule_queue(struct netdev_queue *txq) 3087 { 3088 rcu_read_lock(); 3089 if (!netif_xmit_stopped(txq)) { 3090 struct Qdisc *q = rcu_dereference(txq->qdisc); 3091 3092 __netif_schedule(q); 3093 } 3094 rcu_read_unlock(); 3095 } 3096 EXPORT_SYMBOL(netif_schedule_queue); 3097 3098 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3099 { 3100 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3101 struct Qdisc *q; 3102 3103 rcu_read_lock(); 3104 q = rcu_dereference(dev_queue->qdisc); 3105 __netif_schedule(q); 3106 rcu_read_unlock(); 3107 } 3108 } 3109 EXPORT_SYMBOL(netif_tx_wake_queue); 3110 3111 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3112 { 3113 unsigned long flags; 3114 3115 if (unlikely(!skb)) 3116 return; 3117 3118 if (likely(refcount_read(&skb->users) == 1)) { 3119 smp_rmb(); 3120 refcount_set(&skb->users, 0); 3121 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3122 return; 3123 } 3124 get_kfree_skb_cb(skb)->reason = reason; 3125 local_irq_save(flags); 3126 skb->next = __this_cpu_read(softnet_data.completion_queue); 3127 __this_cpu_write(softnet_data.completion_queue, skb); 3128 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3129 local_irq_restore(flags); 3130 } 3131 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3132 3133 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3134 { 3135 if (in_hardirq() || irqs_disabled()) 3136 dev_kfree_skb_irq_reason(skb, reason); 3137 else 3138 kfree_skb_reason(skb, reason); 3139 } 3140 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3141 3142 3143 /** 3144 * netif_device_detach - mark device as removed 3145 * @dev: network device 3146 * 3147 * Mark device as removed from system and therefore no longer available. 3148 */ 3149 void netif_device_detach(struct net_device *dev) 3150 { 3151 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3152 netif_running(dev)) { 3153 netif_tx_stop_all_queues(dev); 3154 } 3155 } 3156 EXPORT_SYMBOL(netif_device_detach); 3157 3158 /** 3159 * netif_device_attach - mark device as attached 3160 * @dev: network device 3161 * 3162 * Mark device as attached from system and restart if needed. 3163 */ 3164 void netif_device_attach(struct net_device *dev) 3165 { 3166 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3167 netif_running(dev)) { 3168 netif_tx_wake_all_queues(dev); 3169 __netdev_watchdog_up(dev); 3170 } 3171 } 3172 EXPORT_SYMBOL(netif_device_attach); 3173 3174 /* 3175 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3176 * to be used as a distribution range. 3177 */ 3178 static u16 skb_tx_hash(const struct net_device *dev, 3179 const struct net_device *sb_dev, 3180 struct sk_buff *skb) 3181 { 3182 u32 hash; 3183 u16 qoffset = 0; 3184 u16 qcount = dev->real_num_tx_queues; 3185 3186 if (dev->num_tc) { 3187 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3188 3189 qoffset = sb_dev->tc_to_txq[tc].offset; 3190 qcount = sb_dev->tc_to_txq[tc].count; 3191 if (unlikely(!qcount)) { 3192 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3193 sb_dev->name, qoffset, tc); 3194 qoffset = 0; 3195 qcount = dev->real_num_tx_queues; 3196 } 3197 } 3198 3199 if (skb_rx_queue_recorded(skb)) { 3200 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3201 hash = skb_get_rx_queue(skb); 3202 if (hash >= qoffset) 3203 hash -= qoffset; 3204 while (unlikely(hash >= qcount)) 3205 hash -= qcount; 3206 return hash + qoffset; 3207 } 3208 3209 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3210 } 3211 3212 static void skb_warn_bad_offload(const struct sk_buff *skb) 3213 { 3214 static const netdev_features_t null_features; 3215 struct net_device *dev = skb->dev; 3216 const char *name = ""; 3217 3218 if (!net_ratelimit()) 3219 return; 3220 3221 if (dev) { 3222 if (dev->dev.parent) 3223 name = dev_driver_string(dev->dev.parent); 3224 else 3225 name = netdev_name(dev); 3226 } 3227 skb_dump(KERN_WARNING, skb, false); 3228 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3229 name, dev ? &dev->features : &null_features, 3230 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3231 } 3232 3233 /* 3234 * Invalidate hardware checksum when packet is to be mangled, and 3235 * complete checksum manually on outgoing path. 3236 */ 3237 int skb_checksum_help(struct sk_buff *skb) 3238 { 3239 __wsum csum; 3240 int ret = 0, offset; 3241 3242 if (skb->ip_summed == CHECKSUM_COMPLETE) 3243 goto out_set_summed; 3244 3245 if (unlikely(skb_is_gso(skb))) { 3246 skb_warn_bad_offload(skb); 3247 return -EINVAL; 3248 } 3249 3250 /* Before computing a checksum, we should make sure no frag could 3251 * be modified by an external entity : checksum could be wrong. 3252 */ 3253 if (skb_has_shared_frag(skb)) { 3254 ret = __skb_linearize(skb); 3255 if (ret) 3256 goto out; 3257 } 3258 3259 offset = skb_checksum_start_offset(skb); 3260 ret = -EINVAL; 3261 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3262 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3263 goto out; 3264 } 3265 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3266 3267 offset += skb->csum_offset; 3268 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3269 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3270 goto out; 3271 } 3272 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3273 if (ret) 3274 goto out; 3275 3276 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3277 out_set_summed: 3278 skb->ip_summed = CHECKSUM_NONE; 3279 out: 3280 return ret; 3281 } 3282 EXPORT_SYMBOL(skb_checksum_help); 3283 3284 int skb_crc32c_csum_help(struct sk_buff *skb) 3285 { 3286 __le32 crc32c_csum; 3287 int ret = 0, offset, start; 3288 3289 if (skb->ip_summed != CHECKSUM_PARTIAL) 3290 goto out; 3291 3292 if (unlikely(skb_is_gso(skb))) 3293 goto out; 3294 3295 /* Before computing a checksum, we should make sure no frag could 3296 * be modified by an external entity : checksum could be wrong. 3297 */ 3298 if (unlikely(skb_has_shared_frag(skb))) { 3299 ret = __skb_linearize(skb); 3300 if (ret) 3301 goto out; 3302 } 3303 start = skb_checksum_start_offset(skb); 3304 offset = start + offsetof(struct sctphdr, checksum); 3305 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3306 ret = -EINVAL; 3307 goto out; 3308 } 3309 3310 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3311 if (ret) 3312 goto out; 3313 3314 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3315 skb->len - start, ~(__u32)0, 3316 crc32c_csum_stub)); 3317 *(__le32 *)(skb->data + offset) = crc32c_csum; 3318 skb_reset_csum_not_inet(skb); 3319 out: 3320 return ret; 3321 } 3322 3323 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3324 { 3325 __be16 type = skb->protocol; 3326 3327 /* Tunnel gso handlers can set protocol to ethernet. */ 3328 if (type == htons(ETH_P_TEB)) { 3329 struct ethhdr *eth; 3330 3331 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3332 return 0; 3333 3334 eth = (struct ethhdr *)skb->data; 3335 type = eth->h_proto; 3336 } 3337 3338 return vlan_get_protocol_and_depth(skb, type, depth); 3339 } 3340 3341 /* openvswitch calls this on rx path, so we need a different check. 3342 */ 3343 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3344 { 3345 if (tx_path) 3346 return skb->ip_summed != CHECKSUM_PARTIAL && 3347 skb->ip_summed != CHECKSUM_UNNECESSARY; 3348 3349 return skb->ip_summed == CHECKSUM_NONE; 3350 } 3351 3352 /** 3353 * __skb_gso_segment - Perform segmentation on skb. 3354 * @skb: buffer to segment 3355 * @features: features for the output path (see dev->features) 3356 * @tx_path: whether it is called in TX path 3357 * 3358 * This function segments the given skb and returns a list of segments. 3359 * 3360 * It may return NULL if the skb requires no segmentation. This is 3361 * only possible when GSO is used for verifying header integrity. 3362 * 3363 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3364 */ 3365 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3366 netdev_features_t features, bool tx_path) 3367 { 3368 struct sk_buff *segs; 3369 3370 if (unlikely(skb_needs_check(skb, tx_path))) { 3371 int err; 3372 3373 /* We're going to init ->check field in TCP or UDP header */ 3374 err = skb_cow_head(skb, 0); 3375 if (err < 0) 3376 return ERR_PTR(err); 3377 } 3378 3379 /* Only report GSO partial support if it will enable us to 3380 * support segmentation on this frame without needing additional 3381 * work. 3382 */ 3383 if (features & NETIF_F_GSO_PARTIAL) { 3384 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3385 struct net_device *dev = skb->dev; 3386 3387 partial_features |= dev->features & dev->gso_partial_features; 3388 if (!skb_gso_ok(skb, features | partial_features)) 3389 features &= ~NETIF_F_GSO_PARTIAL; 3390 } 3391 3392 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3393 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3394 3395 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3396 SKB_GSO_CB(skb)->encap_level = 0; 3397 3398 skb_reset_mac_header(skb); 3399 skb_reset_mac_len(skb); 3400 3401 segs = skb_mac_gso_segment(skb, features); 3402 3403 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3404 skb_warn_bad_offload(skb); 3405 3406 return segs; 3407 } 3408 EXPORT_SYMBOL(__skb_gso_segment); 3409 3410 /* Take action when hardware reception checksum errors are detected. */ 3411 #ifdef CONFIG_BUG 3412 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3413 { 3414 netdev_err(dev, "hw csum failure\n"); 3415 skb_dump(KERN_ERR, skb, true); 3416 dump_stack(); 3417 } 3418 3419 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3420 { 3421 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3422 } 3423 EXPORT_SYMBOL(netdev_rx_csum_fault); 3424 #endif 3425 3426 /* XXX: check that highmem exists at all on the given machine. */ 3427 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3428 { 3429 #ifdef CONFIG_HIGHMEM 3430 int i; 3431 3432 if (!(dev->features & NETIF_F_HIGHDMA)) { 3433 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3434 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3435 3436 if (PageHighMem(skb_frag_page(frag))) 3437 return 1; 3438 } 3439 } 3440 #endif 3441 return 0; 3442 } 3443 3444 /* If MPLS offload request, verify we are testing hardware MPLS features 3445 * instead of standard features for the netdev. 3446 */ 3447 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3448 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3449 netdev_features_t features, 3450 __be16 type) 3451 { 3452 if (eth_p_mpls(type)) 3453 features &= skb->dev->mpls_features; 3454 3455 return features; 3456 } 3457 #else 3458 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3459 netdev_features_t features, 3460 __be16 type) 3461 { 3462 return features; 3463 } 3464 #endif 3465 3466 static netdev_features_t harmonize_features(struct sk_buff *skb, 3467 netdev_features_t features) 3468 { 3469 __be16 type; 3470 3471 type = skb_network_protocol(skb, NULL); 3472 features = net_mpls_features(skb, features, type); 3473 3474 if (skb->ip_summed != CHECKSUM_NONE && 3475 !can_checksum_protocol(features, type)) { 3476 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3477 } 3478 if (illegal_highdma(skb->dev, skb)) 3479 features &= ~NETIF_F_SG; 3480 3481 return features; 3482 } 3483 3484 netdev_features_t passthru_features_check(struct sk_buff *skb, 3485 struct net_device *dev, 3486 netdev_features_t features) 3487 { 3488 return features; 3489 } 3490 EXPORT_SYMBOL(passthru_features_check); 3491 3492 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3493 struct net_device *dev, 3494 netdev_features_t features) 3495 { 3496 return vlan_features_check(skb, features); 3497 } 3498 3499 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3500 struct net_device *dev, 3501 netdev_features_t features) 3502 { 3503 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3504 3505 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3506 return features & ~NETIF_F_GSO_MASK; 3507 3508 if (!skb_shinfo(skb)->gso_type) { 3509 skb_warn_bad_offload(skb); 3510 return features & ~NETIF_F_GSO_MASK; 3511 } 3512 3513 /* Support for GSO partial features requires software 3514 * intervention before we can actually process the packets 3515 * so we need to strip support for any partial features now 3516 * and we can pull them back in after we have partially 3517 * segmented the frame. 3518 */ 3519 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3520 features &= ~dev->gso_partial_features; 3521 3522 /* Make sure to clear the IPv4 ID mangling feature if the 3523 * IPv4 header has the potential to be fragmented. 3524 */ 3525 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3526 struct iphdr *iph = skb->encapsulation ? 3527 inner_ip_hdr(skb) : ip_hdr(skb); 3528 3529 if (!(iph->frag_off & htons(IP_DF))) 3530 features &= ~NETIF_F_TSO_MANGLEID; 3531 } 3532 3533 return features; 3534 } 3535 3536 netdev_features_t netif_skb_features(struct sk_buff *skb) 3537 { 3538 struct net_device *dev = skb->dev; 3539 netdev_features_t features = dev->features; 3540 3541 if (skb_is_gso(skb)) 3542 features = gso_features_check(skb, dev, features); 3543 3544 /* If encapsulation offload request, verify we are testing 3545 * hardware encapsulation features instead of standard 3546 * features for the netdev 3547 */ 3548 if (skb->encapsulation) 3549 features &= dev->hw_enc_features; 3550 3551 if (skb_vlan_tagged(skb)) 3552 features = netdev_intersect_features(features, 3553 dev->vlan_features | 3554 NETIF_F_HW_VLAN_CTAG_TX | 3555 NETIF_F_HW_VLAN_STAG_TX); 3556 3557 if (dev->netdev_ops->ndo_features_check) 3558 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3559 features); 3560 else 3561 features &= dflt_features_check(skb, dev, features); 3562 3563 return harmonize_features(skb, features); 3564 } 3565 EXPORT_SYMBOL(netif_skb_features); 3566 3567 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3568 struct netdev_queue *txq, bool more) 3569 { 3570 unsigned int len; 3571 int rc; 3572 3573 if (dev_nit_active(dev)) 3574 dev_queue_xmit_nit(skb, dev); 3575 3576 len = skb->len; 3577 trace_net_dev_start_xmit(skb, dev); 3578 rc = netdev_start_xmit(skb, dev, txq, more); 3579 trace_net_dev_xmit(skb, rc, dev, len); 3580 3581 return rc; 3582 } 3583 3584 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3585 struct netdev_queue *txq, int *ret) 3586 { 3587 struct sk_buff *skb = first; 3588 int rc = NETDEV_TX_OK; 3589 3590 while (skb) { 3591 struct sk_buff *next = skb->next; 3592 3593 skb_mark_not_on_list(skb); 3594 rc = xmit_one(skb, dev, txq, next != NULL); 3595 if (unlikely(!dev_xmit_complete(rc))) { 3596 skb->next = next; 3597 goto out; 3598 } 3599 3600 skb = next; 3601 if (netif_tx_queue_stopped(txq) && skb) { 3602 rc = NETDEV_TX_BUSY; 3603 break; 3604 } 3605 } 3606 3607 out: 3608 *ret = rc; 3609 return skb; 3610 } 3611 3612 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3613 netdev_features_t features) 3614 { 3615 if (skb_vlan_tag_present(skb) && 3616 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3617 skb = __vlan_hwaccel_push_inside(skb); 3618 return skb; 3619 } 3620 3621 int skb_csum_hwoffload_help(struct sk_buff *skb, 3622 const netdev_features_t features) 3623 { 3624 if (unlikely(skb_csum_is_sctp(skb))) 3625 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3626 skb_crc32c_csum_help(skb); 3627 3628 if (features & NETIF_F_HW_CSUM) 3629 return 0; 3630 3631 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3632 switch (skb->csum_offset) { 3633 case offsetof(struct tcphdr, check): 3634 case offsetof(struct udphdr, check): 3635 return 0; 3636 } 3637 } 3638 3639 return skb_checksum_help(skb); 3640 } 3641 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3642 3643 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3644 { 3645 netdev_features_t features; 3646 3647 features = netif_skb_features(skb); 3648 skb = validate_xmit_vlan(skb, features); 3649 if (unlikely(!skb)) 3650 goto out_null; 3651 3652 skb = sk_validate_xmit_skb(skb, dev); 3653 if (unlikely(!skb)) 3654 goto out_null; 3655 3656 if (netif_needs_gso(skb, features)) { 3657 struct sk_buff *segs; 3658 3659 segs = skb_gso_segment(skb, features); 3660 if (IS_ERR(segs)) { 3661 goto out_kfree_skb; 3662 } else if (segs) { 3663 consume_skb(skb); 3664 skb = segs; 3665 } 3666 } else { 3667 if (skb_needs_linearize(skb, features) && 3668 __skb_linearize(skb)) 3669 goto out_kfree_skb; 3670 3671 /* If packet is not checksummed and device does not 3672 * support checksumming for this protocol, complete 3673 * checksumming here. 3674 */ 3675 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3676 if (skb->encapsulation) 3677 skb_set_inner_transport_header(skb, 3678 skb_checksum_start_offset(skb)); 3679 else 3680 skb_set_transport_header(skb, 3681 skb_checksum_start_offset(skb)); 3682 if (skb_csum_hwoffload_help(skb, features)) 3683 goto out_kfree_skb; 3684 } 3685 } 3686 3687 skb = validate_xmit_xfrm(skb, features, again); 3688 3689 return skb; 3690 3691 out_kfree_skb: 3692 kfree_skb(skb); 3693 out_null: 3694 dev_core_stats_tx_dropped_inc(dev); 3695 return NULL; 3696 } 3697 3698 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3699 { 3700 struct sk_buff *next, *head = NULL, *tail; 3701 3702 for (; skb != NULL; skb = next) { 3703 next = skb->next; 3704 skb_mark_not_on_list(skb); 3705 3706 /* in case skb wont be segmented, point to itself */ 3707 skb->prev = skb; 3708 3709 skb = validate_xmit_skb(skb, dev, again); 3710 if (!skb) 3711 continue; 3712 3713 if (!head) 3714 head = skb; 3715 else 3716 tail->next = skb; 3717 /* If skb was segmented, skb->prev points to 3718 * the last segment. If not, it still contains skb. 3719 */ 3720 tail = skb->prev; 3721 } 3722 return head; 3723 } 3724 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3725 3726 static void qdisc_pkt_len_init(struct sk_buff *skb) 3727 { 3728 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3729 3730 qdisc_skb_cb(skb)->pkt_len = skb->len; 3731 3732 /* To get more precise estimation of bytes sent on wire, 3733 * we add to pkt_len the headers size of all segments 3734 */ 3735 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3736 u16 gso_segs = shinfo->gso_segs; 3737 unsigned int hdr_len; 3738 3739 /* mac layer + network layer */ 3740 hdr_len = skb_transport_offset(skb); 3741 3742 /* + transport layer */ 3743 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3744 const struct tcphdr *th; 3745 struct tcphdr _tcphdr; 3746 3747 th = skb_header_pointer(skb, hdr_len, 3748 sizeof(_tcphdr), &_tcphdr); 3749 if (likely(th)) 3750 hdr_len += __tcp_hdrlen(th); 3751 } else { 3752 struct udphdr _udphdr; 3753 3754 if (skb_header_pointer(skb, hdr_len, 3755 sizeof(_udphdr), &_udphdr)) 3756 hdr_len += sizeof(struct udphdr); 3757 } 3758 3759 if (shinfo->gso_type & SKB_GSO_DODGY) 3760 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3761 shinfo->gso_size); 3762 3763 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3764 } 3765 } 3766 3767 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3768 struct sk_buff **to_free, 3769 struct netdev_queue *txq) 3770 { 3771 int rc; 3772 3773 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3774 if (rc == NET_XMIT_SUCCESS) 3775 trace_qdisc_enqueue(q, txq, skb); 3776 return rc; 3777 } 3778 3779 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3780 struct net_device *dev, 3781 struct netdev_queue *txq) 3782 { 3783 spinlock_t *root_lock = qdisc_lock(q); 3784 struct sk_buff *to_free = NULL; 3785 bool contended; 3786 int rc; 3787 3788 qdisc_calculate_pkt_len(skb, q); 3789 3790 if (q->flags & TCQ_F_NOLOCK) { 3791 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3792 qdisc_run_begin(q)) { 3793 /* Retest nolock_qdisc_is_empty() within the protection 3794 * of q->seqlock to protect from racing with requeuing. 3795 */ 3796 if (unlikely(!nolock_qdisc_is_empty(q))) { 3797 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3798 __qdisc_run(q); 3799 qdisc_run_end(q); 3800 3801 goto no_lock_out; 3802 } 3803 3804 qdisc_bstats_cpu_update(q, skb); 3805 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3806 !nolock_qdisc_is_empty(q)) 3807 __qdisc_run(q); 3808 3809 qdisc_run_end(q); 3810 return NET_XMIT_SUCCESS; 3811 } 3812 3813 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3814 qdisc_run(q); 3815 3816 no_lock_out: 3817 if (unlikely(to_free)) 3818 kfree_skb_list_reason(to_free, 3819 SKB_DROP_REASON_QDISC_DROP); 3820 return rc; 3821 } 3822 3823 /* 3824 * Heuristic to force contended enqueues to serialize on a 3825 * separate lock before trying to get qdisc main lock. 3826 * This permits qdisc->running owner to get the lock more 3827 * often and dequeue packets faster. 3828 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3829 * and then other tasks will only enqueue packets. The packets will be 3830 * sent after the qdisc owner is scheduled again. To prevent this 3831 * scenario the task always serialize on the lock. 3832 */ 3833 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3834 if (unlikely(contended)) 3835 spin_lock(&q->busylock); 3836 3837 spin_lock(root_lock); 3838 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3839 __qdisc_drop(skb, &to_free); 3840 rc = NET_XMIT_DROP; 3841 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3842 qdisc_run_begin(q)) { 3843 /* 3844 * This is a work-conserving queue; there are no old skbs 3845 * waiting to be sent out; and the qdisc is not running - 3846 * xmit the skb directly. 3847 */ 3848 3849 qdisc_bstats_update(q, skb); 3850 3851 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3852 if (unlikely(contended)) { 3853 spin_unlock(&q->busylock); 3854 contended = false; 3855 } 3856 __qdisc_run(q); 3857 } 3858 3859 qdisc_run_end(q); 3860 rc = NET_XMIT_SUCCESS; 3861 } else { 3862 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3863 if (qdisc_run_begin(q)) { 3864 if (unlikely(contended)) { 3865 spin_unlock(&q->busylock); 3866 contended = false; 3867 } 3868 __qdisc_run(q); 3869 qdisc_run_end(q); 3870 } 3871 } 3872 spin_unlock(root_lock); 3873 if (unlikely(to_free)) 3874 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3875 if (unlikely(contended)) 3876 spin_unlock(&q->busylock); 3877 return rc; 3878 } 3879 3880 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3881 static void skb_update_prio(struct sk_buff *skb) 3882 { 3883 const struct netprio_map *map; 3884 const struct sock *sk; 3885 unsigned int prioidx; 3886 3887 if (skb->priority) 3888 return; 3889 map = rcu_dereference_bh(skb->dev->priomap); 3890 if (!map) 3891 return; 3892 sk = skb_to_full_sk(skb); 3893 if (!sk) 3894 return; 3895 3896 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3897 3898 if (prioidx < map->priomap_len) 3899 skb->priority = map->priomap[prioidx]; 3900 } 3901 #else 3902 #define skb_update_prio(skb) 3903 #endif 3904 3905 /** 3906 * dev_loopback_xmit - loop back @skb 3907 * @net: network namespace this loopback is happening in 3908 * @sk: sk needed to be a netfilter okfn 3909 * @skb: buffer to transmit 3910 */ 3911 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3912 { 3913 skb_reset_mac_header(skb); 3914 __skb_pull(skb, skb_network_offset(skb)); 3915 skb->pkt_type = PACKET_LOOPBACK; 3916 if (skb->ip_summed == CHECKSUM_NONE) 3917 skb->ip_summed = CHECKSUM_UNNECESSARY; 3918 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3919 skb_dst_force(skb); 3920 netif_rx(skb); 3921 return 0; 3922 } 3923 EXPORT_SYMBOL(dev_loopback_xmit); 3924 3925 #ifdef CONFIG_NET_EGRESS 3926 static struct sk_buff * 3927 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3928 { 3929 #ifdef CONFIG_NET_CLS_ACT 3930 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3931 struct tcf_result cl_res; 3932 3933 if (!miniq) 3934 return skb; 3935 3936 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3937 tc_skb_cb(skb)->mru = 0; 3938 tc_skb_cb(skb)->post_ct = false; 3939 mini_qdisc_bstats_cpu_update(miniq, skb); 3940 3941 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3942 case TC_ACT_OK: 3943 case TC_ACT_RECLASSIFY: 3944 skb->tc_index = TC_H_MIN(cl_res.classid); 3945 break; 3946 case TC_ACT_SHOT: 3947 mini_qdisc_qstats_cpu_drop(miniq); 3948 *ret = NET_XMIT_DROP; 3949 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3950 return NULL; 3951 case TC_ACT_STOLEN: 3952 case TC_ACT_QUEUED: 3953 case TC_ACT_TRAP: 3954 *ret = NET_XMIT_SUCCESS; 3955 consume_skb(skb); 3956 return NULL; 3957 case TC_ACT_REDIRECT: 3958 /* No need to push/pop skb's mac_header here on egress! */ 3959 skb_do_redirect(skb); 3960 *ret = NET_XMIT_SUCCESS; 3961 return NULL; 3962 default: 3963 break; 3964 } 3965 #endif /* CONFIG_NET_CLS_ACT */ 3966 3967 return skb; 3968 } 3969 3970 static struct netdev_queue * 3971 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3972 { 3973 int qm = skb_get_queue_mapping(skb); 3974 3975 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3976 } 3977 3978 static bool netdev_xmit_txqueue_skipped(void) 3979 { 3980 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3981 } 3982 3983 void netdev_xmit_skip_txqueue(bool skip) 3984 { 3985 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3986 } 3987 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3988 #endif /* CONFIG_NET_EGRESS */ 3989 3990 #ifdef CONFIG_XPS 3991 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3992 struct xps_dev_maps *dev_maps, unsigned int tci) 3993 { 3994 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3995 struct xps_map *map; 3996 int queue_index = -1; 3997 3998 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3999 return queue_index; 4000 4001 tci *= dev_maps->num_tc; 4002 tci += tc; 4003 4004 map = rcu_dereference(dev_maps->attr_map[tci]); 4005 if (map) { 4006 if (map->len == 1) 4007 queue_index = map->queues[0]; 4008 else 4009 queue_index = map->queues[reciprocal_scale( 4010 skb_get_hash(skb), map->len)]; 4011 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4012 queue_index = -1; 4013 } 4014 return queue_index; 4015 } 4016 #endif 4017 4018 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4019 struct sk_buff *skb) 4020 { 4021 #ifdef CONFIG_XPS 4022 struct xps_dev_maps *dev_maps; 4023 struct sock *sk = skb->sk; 4024 int queue_index = -1; 4025 4026 if (!static_key_false(&xps_needed)) 4027 return -1; 4028 4029 rcu_read_lock(); 4030 if (!static_key_false(&xps_rxqs_needed)) 4031 goto get_cpus_map; 4032 4033 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4034 if (dev_maps) { 4035 int tci = sk_rx_queue_get(sk); 4036 4037 if (tci >= 0) 4038 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4039 tci); 4040 } 4041 4042 get_cpus_map: 4043 if (queue_index < 0) { 4044 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4045 if (dev_maps) { 4046 unsigned int tci = skb->sender_cpu - 1; 4047 4048 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4049 tci); 4050 } 4051 } 4052 rcu_read_unlock(); 4053 4054 return queue_index; 4055 #else 4056 return -1; 4057 #endif 4058 } 4059 4060 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4061 struct net_device *sb_dev) 4062 { 4063 return 0; 4064 } 4065 EXPORT_SYMBOL(dev_pick_tx_zero); 4066 4067 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4068 struct net_device *sb_dev) 4069 { 4070 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4071 } 4072 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4073 4074 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4075 struct net_device *sb_dev) 4076 { 4077 struct sock *sk = skb->sk; 4078 int queue_index = sk_tx_queue_get(sk); 4079 4080 sb_dev = sb_dev ? : dev; 4081 4082 if (queue_index < 0 || skb->ooo_okay || 4083 queue_index >= dev->real_num_tx_queues) { 4084 int new_index = get_xps_queue(dev, sb_dev, skb); 4085 4086 if (new_index < 0) 4087 new_index = skb_tx_hash(dev, sb_dev, skb); 4088 4089 if (queue_index != new_index && sk && 4090 sk_fullsock(sk) && 4091 rcu_access_pointer(sk->sk_dst_cache)) 4092 sk_tx_queue_set(sk, new_index); 4093 4094 queue_index = new_index; 4095 } 4096 4097 return queue_index; 4098 } 4099 EXPORT_SYMBOL(netdev_pick_tx); 4100 4101 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4102 struct sk_buff *skb, 4103 struct net_device *sb_dev) 4104 { 4105 int queue_index = 0; 4106 4107 #ifdef CONFIG_XPS 4108 u32 sender_cpu = skb->sender_cpu - 1; 4109 4110 if (sender_cpu >= (u32)NR_CPUS) 4111 skb->sender_cpu = raw_smp_processor_id() + 1; 4112 #endif 4113 4114 if (dev->real_num_tx_queues != 1) { 4115 const struct net_device_ops *ops = dev->netdev_ops; 4116 4117 if (ops->ndo_select_queue) 4118 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4119 else 4120 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4121 4122 queue_index = netdev_cap_txqueue(dev, queue_index); 4123 } 4124 4125 skb_set_queue_mapping(skb, queue_index); 4126 return netdev_get_tx_queue(dev, queue_index); 4127 } 4128 4129 /** 4130 * __dev_queue_xmit() - transmit a buffer 4131 * @skb: buffer to transmit 4132 * @sb_dev: suboordinate device used for L2 forwarding offload 4133 * 4134 * Queue a buffer for transmission to a network device. The caller must 4135 * have set the device and priority and built the buffer before calling 4136 * this function. The function can be called from an interrupt. 4137 * 4138 * When calling this method, interrupts MUST be enabled. This is because 4139 * the BH enable code must have IRQs enabled so that it will not deadlock. 4140 * 4141 * Regardless of the return value, the skb is consumed, so it is currently 4142 * difficult to retry a send to this method. (You can bump the ref count 4143 * before sending to hold a reference for retry if you are careful.) 4144 * 4145 * Return: 4146 * * 0 - buffer successfully transmitted 4147 * * positive qdisc return code - NET_XMIT_DROP etc. 4148 * * negative errno - other errors 4149 */ 4150 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4151 { 4152 struct net_device *dev = skb->dev; 4153 struct netdev_queue *txq = NULL; 4154 struct Qdisc *q; 4155 int rc = -ENOMEM; 4156 bool again = false; 4157 4158 skb_reset_mac_header(skb); 4159 skb_assert_len(skb); 4160 4161 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4162 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4163 4164 /* Disable soft irqs for various locks below. Also 4165 * stops preemption for RCU. 4166 */ 4167 rcu_read_lock_bh(); 4168 4169 skb_update_prio(skb); 4170 4171 qdisc_pkt_len_init(skb); 4172 #ifdef CONFIG_NET_CLS_ACT 4173 skb->tc_at_ingress = 0; 4174 #endif 4175 #ifdef CONFIG_NET_EGRESS 4176 if (static_branch_unlikely(&egress_needed_key)) { 4177 if (nf_hook_egress_active()) { 4178 skb = nf_hook_egress(skb, &rc, dev); 4179 if (!skb) 4180 goto out; 4181 } 4182 4183 netdev_xmit_skip_txqueue(false); 4184 4185 nf_skip_egress(skb, true); 4186 skb = sch_handle_egress(skb, &rc, dev); 4187 if (!skb) 4188 goto out; 4189 nf_skip_egress(skb, false); 4190 4191 if (netdev_xmit_txqueue_skipped()) 4192 txq = netdev_tx_queue_mapping(dev, skb); 4193 } 4194 #endif 4195 /* If device/qdisc don't need skb->dst, release it right now while 4196 * its hot in this cpu cache. 4197 */ 4198 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4199 skb_dst_drop(skb); 4200 else 4201 skb_dst_force(skb); 4202 4203 if (!txq) 4204 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4205 4206 q = rcu_dereference_bh(txq->qdisc); 4207 4208 trace_net_dev_queue(skb); 4209 if (q->enqueue) { 4210 rc = __dev_xmit_skb(skb, q, dev, txq); 4211 goto out; 4212 } 4213 4214 /* The device has no queue. Common case for software devices: 4215 * loopback, all the sorts of tunnels... 4216 4217 * Really, it is unlikely that netif_tx_lock protection is necessary 4218 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4219 * counters.) 4220 * However, it is possible, that they rely on protection 4221 * made by us here. 4222 4223 * Check this and shot the lock. It is not prone from deadlocks. 4224 *Either shot noqueue qdisc, it is even simpler 8) 4225 */ 4226 if (dev->flags & IFF_UP) { 4227 int cpu = smp_processor_id(); /* ok because BHs are off */ 4228 4229 /* Other cpus might concurrently change txq->xmit_lock_owner 4230 * to -1 or to their cpu id, but not to our id. 4231 */ 4232 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4233 if (dev_xmit_recursion()) 4234 goto recursion_alert; 4235 4236 skb = validate_xmit_skb(skb, dev, &again); 4237 if (!skb) 4238 goto out; 4239 4240 HARD_TX_LOCK(dev, txq, cpu); 4241 4242 if (!netif_xmit_stopped(txq)) { 4243 dev_xmit_recursion_inc(); 4244 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4245 dev_xmit_recursion_dec(); 4246 if (dev_xmit_complete(rc)) { 4247 HARD_TX_UNLOCK(dev, txq); 4248 goto out; 4249 } 4250 } 4251 HARD_TX_UNLOCK(dev, txq); 4252 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4253 dev->name); 4254 } else { 4255 /* Recursion is detected! It is possible, 4256 * unfortunately 4257 */ 4258 recursion_alert: 4259 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4260 dev->name); 4261 } 4262 } 4263 4264 rc = -ENETDOWN; 4265 rcu_read_unlock_bh(); 4266 4267 dev_core_stats_tx_dropped_inc(dev); 4268 kfree_skb_list(skb); 4269 return rc; 4270 out: 4271 rcu_read_unlock_bh(); 4272 return rc; 4273 } 4274 EXPORT_SYMBOL(__dev_queue_xmit); 4275 4276 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4277 { 4278 struct net_device *dev = skb->dev; 4279 struct sk_buff *orig_skb = skb; 4280 struct netdev_queue *txq; 4281 int ret = NETDEV_TX_BUSY; 4282 bool again = false; 4283 4284 if (unlikely(!netif_running(dev) || 4285 !netif_carrier_ok(dev))) 4286 goto drop; 4287 4288 skb = validate_xmit_skb_list(skb, dev, &again); 4289 if (skb != orig_skb) 4290 goto drop; 4291 4292 skb_set_queue_mapping(skb, queue_id); 4293 txq = skb_get_tx_queue(dev, skb); 4294 4295 local_bh_disable(); 4296 4297 dev_xmit_recursion_inc(); 4298 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4299 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4300 ret = netdev_start_xmit(skb, dev, txq, false); 4301 HARD_TX_UNLOCK(dev, txq); 4302 dev_xmit_recursion_dec(); 4303 4304 local_bh_enable(); 4305 return ret; 4306 drop: 4307 dev_core_stats_tx_dropped_inc(dev); 4308 kfree_skb_list(skb); 4309 return NET_XMIT_DROP; 4310 } 4311 EXPORT_SYMBOL(__dev_direct_xmit); 4312 4313 /************************************************************************* 4314 * Receiver routines 4315 *************************************************************************/ 4316 4317 int netdev_max_backlog __read_mostly = 1000; 4318 EXPORT_SYMBOL(netdev_max_backlog); 4319 4320 int netdev_tstamp_prequeue __read_mostly = 1; 4321 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4322 int netdev_budget __read_mostly = 300; 4323 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4324 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4325 int weight_p __read_mostly = 64; /* old backlog weight */ 4326 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4327 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4328 int dev_rx_weight __read_mostly = 64; 4329 int dev_tx_weight __read_mostly = 64; 4330 4331 /* Called with irq disabled */ 4332 static inline void ____napi_schedule(struct softnet_data *sd, 4333 struct napi_struct *napi) 4334 { 4335 struct task_struct *thread; 4336 4337 lockdep_assert_irqs_disabled(); 4338 4339 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4340 /* Paired with smp_mb__before_atomic() in 4341 * napi_enable()/dev_set_threaded(). 4342 * Use READ_ONCE() to guarantee a complete 4343 * read on napi->thread. Only call 4344 * wake_up_process() when it's not NULL. 4345 */ 4346 thread = READ_ONCE(napi->thread); 4347 if (thread) { 4348 /* Avoid doing set_bit() if the thread is in 4349 * INTERRUPTIBLE state, cause napi_thread_wait() 4350 * makes sure to proceed with napi polling 4351 * if the thread is explicitly woken from here. 4352 */ 4353 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4354 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4355 wake_up_process(thread); 4356 return; 4357 } 4358 } 4359 4360 list_add_tail(&napi->poll_list, &sd->poll_list); 4361 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4362 /* If not called from net_rx_action() 4363 * we have to raise NET_RX_SOFTIRQ. 4364 */ 4365 if (!sd->in_net_rx_action) 4366 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4367 } 4368 4369 #ifdef CONFIG_RPS 4370 4371 /* One global table that all flow-based protocols share. */ 4372 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4373 EXPORT_SYMBOL(rps_sock_flow_table); 4374 u32 rps_cpu_mask __read_mostly; 4375 EXPORT_SYMBOL(rps_cpu_mask); 4376 4377 struct static_key_false rps_needed __read_mostly; 4378 EXPORT_SYMBOL(rps_needed); 4379 struct static_key_false rfs_needed __read_mostly; 4380 EXPORT_SYMBOL(rfs_needed); 4381 4382 static struct rps_dev_flow * 4383 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4384 struct rps_dev_flow *rflow, u16 next_cpu) 4385 { 4386 if (next_cpu < nr_cpu_ids) { 4387 #ifdef CONFIG_RFS_ACCEL 4388 struct netdev_rx_queue *rxqueue; 4389 struct rps_dev_flow_table *flow_table; 4390 struct rps_dev_flow *old_rflow; 4391 u32 flow_id; 4392 u16 rxq_index; 4393 int rc; 4394 4395 /* Should we steer this flow to a different hardware queue? */ 4396 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4397 !(dev->features & NETIF_F_NTUPLE)) 4398 goto out; 4399 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4400 if (rxq_index == skb_get_rx_queue(skb)) 4401 goto out; 4402 4403 rxqueue = dev->_rx + rxq_index; 4404 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4405 if (!flow_table) 4406 goto out; 4407 flow_id = skb_get_hash(skb) & flow_table->mask; 4408 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4409 rxq_index, flow_id); 4410 if (rc < 0) 4411 goto out; 4412 old_rflow = rflow; 4413 rflow = &flow_table->flows[flow_id]; 4414 rflow->filter = rc; 4415 if (old_rflow->filter == rflow->filter) 4416 old_rflow->filter = RPS_NO_FILTER; 4417 out: 4418 #endif 4419 rflow->last_qtail = 4420 per_cpu(softnet_data, next_cpu).input_queue_head; 4421 } 4422 4423 rflow->cpu = next_cpu; 4424 return rflow; 4425 } 4426 4427 /* 4428 * get_rps_cpu is called from netif_receive_skb and returns the target 4429 * CPU from the RPS map of the receiving queue for a given skb. 4430 * rcu_read_lock must be held on entry. 4431 */ 4432 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4433 struct rps_dev_flow **rflowp) 4434 { 4435 const struct rps_sock_flow_table *sock_flow_table; 4436 struct netdev_rx_queue *rxqueue = dev->_rx; 4437 struct rps_dev_flow_table *flow_table; 4438 struct rps_map *map; 4439 int cpu = -1; 4440 u32 tcpu; 4441 u32 hash; 4442 4443 if (skb_rx_queue_recorded(skb)) { 4444 u16 index = skb_get_rx_queue(skb); 4445 4446 if (unlikely(index >= dev->real_num_rx_queues)) { 4447 WARN_ONCE(dev->real_num_rx_queues > 1, 4448 "%s received packet on queue %u, but number " 4449 "of RX queues is %u\n", 4450 dev->name, index, dev->real_num_rx_queues); 4451 goto done; 4452 } 4453 rxqueue += index; 4454 } 4455 4456 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4457 4458 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4459 map = rcu_dereference(rxqueue->rps_map); 4460 if (!flow_table && !map) 4461 goto done; 4462 4463 skb_reset_network_header(skb); 4464 hash = skb_get_hash(skb); 4465 if (!hash) 4466 goto done; 4467 4468 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4469 if (flow_table && sock_flow_table) { 4470 struct rps_dev_flow *rflow; 4471 u32 next_cpu; 4472 u32 ident; 4473 4474 /* First check into global flow table if there is a match. 4475 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4476 */ 4477 ident = READ_ONCE(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 * After we queued a packet into sd->input_pkt_queue, 4591 * we need to make sure this queue is serviced soon. 4592 * 4593 * - If this is another cpu queue, link it to our rps_ipi_list, 4594 * and make sure we will process rps_ipi_list from net_rx_action(). 4595 * 4596 * - If this is our own queue, NAPI schedule our backlog. 4597 * Note that this also raises NET_RX_SOFTIRQ. 4598 */ 4599 static void napi_schedule_rps(struct softnet_data *sd) 4600 { 4601 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4602 4603 #ifdef CONFIG_RPS 4604 if (sd != mysd) { 4605 sd->rps_ipi_next = mysd->rps_ipi_list; 4606 mysd->rps_ipi_list = sd; 4607 4608 /* If not called from net_rx_action() or napi_threaded_poll() 4609 * we have to raise NET_RX_SOFTIRQ. 4610 */ 4611 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4612 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4613 return; 4614 } 4615 #endif /* CONFIG_RPS */ 4616 __napi_schedule_irqoff(&mysd->backlog); 4617 } 4618 4619 #ifdef CONFIG_NET_FLOW_LIMIT 4620 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4621 #endif 4622 4623 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4624 { 4625 #ifdef CONFIG_NET_FLOW_LIMIT 4626 struct sd_flow_limit *fl; 4627 struct softnet_data *sd; 4628 unsigned int old_flow, new_flow; 4629 4630 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4631 return false; 4632 4633 sd = this_cpu_ptr(&softnet_data); 4634 4635 rcu_read_lock(); 4636 fl = rcu_dereference(sd->flow_limit); 4637 if (fl) { 4638 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4639 old_flow = fl->history[fl->history_head]; 4640 fl->history[fl->history_head] = new_flow; 4641 4642 fl->history_head++; 4643 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4644 4645 if (likely(fl->buckets[old_flow])) 4646 fl->buckets[old_flow]--; 4647 4648 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4649 fl->count++; 4650 rcu_read_unlock(); 4651 return true; 4652 } 4653 } 4654 rcu_read_unlock(); 4655 #endif 4656 return false; 4657 } 4658 4659 /* 4660 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4661 * queue (may be a remote CPU queue). 4662 */ 4663 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4664 unsigned int *qtail) 4665 { 4666 enum skb_drop_reason reason; 4667 struct softnet_data *sd; 4668 unsigned long flags; 4669 unsigned int qlen; 4670 4671 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4672 sd = &per_cpu(softnet_data, cpu); 4673 4674 rps_lock_irqsave(sd, &flags); 4675 if (!netif_running(skb->dev)) 4676 goto drop; 4677 qlen = skb_queue_len(&sd->input_pkt_queue); 4678 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4679 if (qlen) { 4680 enqueue: 4681 __skb_queue_tail(&sd->input_pkt_queue, skb); 4682 input_queue_tail_incr_save(sd, qtail); 4683 rps_unlock_irq_restore(sd, &flags); 4684 return NET_RX_SUCCESS; 4685 } 4686 4687 /* Schedule NAPI for backlog device 4688 * We can use non atomic operation since we own the queue lock 4689 */ 4690 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4691 napi_schedule_rps(sd); 4692 goto enqueue; 4693 } 4694 reason = SKB_DROP_REASON_CPU_BACKLOG; 4695 4696 drop: 4697 sd->dropped++; 4698 rps_unlock_irq_restore(sd, &flags); 4699 4700 dev_core_stats_rx_dropped_inc(skb->dev); 4701 kfree_skb_reason(skb, reason); 4702 return NET_RX_DROP; 4703 } 4704 4705 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4706 { 4707 struct net_device *dev = skb->dev; 4708 struct netdev_rx_queue *rxqueue; 4709 4710 rxqueue = dev->_rx; 4711 4712 if (skb_rx_queue_recorded(skb)) { 4713 u16 index = skb_get_rx_queue(skb); 4714 4715 if (unlikely(index >= dev->real_num_rx_queues)) { 4716 WARN_ONCE(dev->real_num_rx_queues > 1, 4717 "%s received packet on queue %u, but number " 4718 "of RX queues is %u\n", 4719 dev->name, index, dev->real_num_rx_queues); 4720 4721 return rxqueue; /* Return first rxqueue */ 4722 } 4723 rxqueue += index; 4724 } 4725 return rxqueue; 4726 } 4727 4728 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4729 struct bpf_prog *xdp_prog) 4730 { 4731 void *orig_data, *orig_data_end, *hard_start; 4732 struct netdev_rx_queue *rxqueue; 4733 bool orig_bcast, orig_host; 4734 u32 mac_len, frame_sz; 4735 __be16 orig_eth_type; 4736 struct ethhdr *eth; 4737 u32 metalen, act; 4738 int off; 4739 4740 /* The XDP program wants to see the packet starting at the MAC 4741 * header. 4742 */ 4743 mac_len = skb->data - skb_mac_header(skb); 4744 hard_start = skb->data - skb_headroom(skb); 4745 4746 /* SKB "head" area always have tailroom for skb_shared_info */ 4747 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4748 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4749 4750 rxqueue = netif_get_rxqueue(skb); 4751 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4752 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4753 skb_headlen(skb) + mac_len, true); 4754 4755 orig_data_end = xdp->data_end; 4756 orig_data = xdp->data; 4757 eth = (struct ethhdr *)xdp->data; 4758 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4759 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4760 orig_eth_type = eth->h_proto; 4761 4762 act = bpf_prog_run_xdp(xdp_prog, xdp); 4763 4764 /* check if bpf_xdp_adjust_head was used */ 4765 off = xdp->data - orig_data; 4766 if (off) { 4767 if (off > 0) 4768 __skb_pull(skb, off); 4769 else if (off < 0) 4770 __skb_push(skb, -off); 4771 4772 skb->mac_header += off; 4773 skb_reset_network_header(skb); 4774 } 4775 4776 /* check if bpf_xdp_adjust_tail was used */ 4777 off = xdp->data_end - orig_data_end; 4778 if (off != 0) { 4779 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4780 skb->len += off; /* positive on grow, negative on shrink */ 4781 } 4782 4783 /* check if XDP changed eth hdr such SKB needs update */ 4784 eth = (struct ethhdr *)xdp->data; 4785 if ((orig_eth_type != eth->h_proto) || 4786 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4787 skb->dev->dev_addr)) || 4788 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4789 __skb_push(skb, ETH_HLEN); 4790 skb->pkt_type = PACKET_HOST; 4791 skb->protocol = eth_type_trans(skb, skb->dev); 4792 } 4793 4794 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4795 * before calling us again on redirect path. We do not call do_redirect 4796 * as we leave that up to the caller. 4797 * 4798 * Caller is responsible for managing lifetime of skb (i.e. calling 4799 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4800 */ 4801 switch (act) { 4802 case XDP_REDIRECT: 4803 case XDP_TX: 4804 __skb_push(skb, mac_len); 4805 break; 4806 case XDP_PASS: 4807 metalen = xdp->data - xdp->data_meta; 4808 if (metalen) 4809 skb_metadata_set(skb, metalen); 4810 break; 4811 } 4812 4813 return act; 4814 } 4815 4816 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4817 struct xdp_buff *xdp, 4818 struct bpf_prog *xdp_prog) 4819 { 4820 u32 act = XDP_DROP; 4821 4822 /* Reinjected packets coming from act_mirred or similar should 4823 * not get XDP generic processing. 4824 */ 4825 if (skb_is_redirected(skb)) 4826 return XDP_PASS; 4827 4828 /* XDP packets must be linear and must have sufficient headroom 4829 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4830 * native XDP provides, thus we need to do it here as well. 4831 */ 4832 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4833 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4834 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4835 int troom = skb->tail + skb->data_len - skb->end; 4836 4837 /* In case we have to go down the path and also linearize, 4838 * then lets do the pskb_expand_head() work just once here. 4839 */ 4840 if (pskb_expand_head(skb, 4841 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4842 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4843 goto do_drop; 4844 if (skb_linearize(skb)) 4845 goto do_drop; 4846 } 4847 4848 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4849 switch (act) { 4850 case XDP_REDIRECT: 4851 case XDP_TX: 4852 case XDP_PASS: 4853 break; 4854 default: 4855 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4856 fallthrough; 4857 case XDP_ABORTED: 4858 trace_xdp_exception(skb->dev, xdp_prog, act); 4859 fallthrough; 4860 case XDP_DROP: 4861 do_drop: 4862 kfree_skb(skb); 4863 break; 4864 } 4865 4866 return act; 4867 } 4868 4869 /* When doing generic XDP we have to bypass the qdisc layer and the 4870 * network taps in order to match in-driver-XDP behavior. This also means 4871 * that XDP packets are able to starve other packets going through a qdisc, 4872 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4873 * queues, so they do not have this starvation issue. 4874 */ 4875 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4876 { 4877 struct net_device *dev = skb->dev; 4878 struct netdev_queue *txq; 4879 bool free_skb = true; 4880 int cpu, rc; 4881 4882 txq = netdev_core_pick_tx(dev, skb, NULL); 4883 cpu = smp_processor_id(); 4884 HARD_TX_LOCK(dev, txq, cpu); 4885 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4886 rc = netdev_start_xmit(skb, dev, txq, 0); 4887 if (dev_xmit_complete(rc)) 4888 free_skb = false; 4889 } 4890 HARD_TX_UNLOCK(dev, txq); 4891 if (free_skb) { 4892 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4893 dev_core_stats_tx_dropped_inc(dev); 4894 kfree_skb(skb); 4895 } 4896 } 4897 4898 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4899 4900 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4901 { 4902 if (xdp_prog) { 4903 struct xdp_buff xdp; 4904 u32 act; 4905 int err; 4906 4907 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4908 if (act != XDP_PASS) { 4909 switch (act) { 4910 case XDP_REDIRECT: 4911 err = xdp_do_generic_redirect(skb->dev, skb, 4912 &xdp, xdp_prog); 4913 if (err) 4914 goto out_redir; 4915 break; 4916 case XDP_TX: 4917 generic_xdp_tx(skb, xdp_prog); 4918 break; 4919 } 4920 return XDP_DROP; 4921 } 4922 } 4923 return XDP_PASS; 4924 out_redir: 4925 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4926 return XDP_DROP; 4927 } 4928 EXPORT_SYMBOL_GPL(do_xdp_generic); 4929 4930 static int netif_rx_internal(struct sk_buff *skb) 4931 { 4932 int ret; 4933 4934 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4935 4936 trace_netif_rx(skb); 4937 4938 #ifdef CONFIG_RPS 4939 if (static_branch_unlikely(&rps_needed)) { 4940 struct rps_dev_flow voidflow, *rflow = &voidflow; 4941 int cpu; 4942 4943 rcu_read_lock(); 4944 4945 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4946 if (cpu < 0) 4947 cpu = smp_processor_id(); 4948 4949 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4950 4951 rcu_read_unlock(); 4952 } else 4953 #endif 4954 { 4955 unsigned int qtail; 4956 4957 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4958 } 4959 return ret; 4960 } 4961 4962 /** 4963 * __netif_rx - Slightly optimized version of netif_rx 4964 * @skb: buffer to post 4965 * 4966 * This behaves as netif_rx except that it does not disable bottom halves. 4967 * As a result this function may only be invoked from the interrupt context 4968 * (either hard or soft interrupt). 4969 */ 4970 int __netif_rx(struct sk_buff *skb) 4971 { 4972 int ret; 4973 4974 lockdep_assert_once(hardirq_count() | softirq_count()); 4975 4976 trace_netif_rx_entry(skb); 4977 ret = netif_rx_internal(skb); 4978 trace_netif_rx_exit(ret); 4979 return ret; 4980 } 4981 EXPORT_SYMBOL(__netif_rx); 4982 4983 /** 4984 * netif_rx - post buffer to the network code 4985 * @skb: buffer to post 4986 * 4987 * This function receives a packet from a device driver and queues it for 4988 * the upper (protocol) levels to process via the backlog NAPI device. It 4989 * always succeeds. The buffer may be dropped during processing for 4990 * congestion control or by the protocol layers. 4991 * The network buffer is passed via the backlog NAPI device. Modern NIC 4992 * driver should use NAPI and GRO. 4993 * This function can used from interrupt and from process context. The 4994 * caller from process context must not disable interrupts before invoking 4995 * this function. 4996 * 4997 * return values: 4998 * NET_RX_SUCCESS (no congestion) 4999 * NET_RX_DROP (packet was dropped) 5000 * 5001 */ 5002 int netif_rx(struct sk_buff *skb) 5003 { 5004 bool need_bh_off = !(hardirq_count() | softirq_count()); 5005 int ret; 5006 5007 if (need_bh_off) 5008 local_bh_disable(); 5009 trace_netif_rx_entry(skb); 5010 ret = netif_rx_internal(skb); 5011 trace_netif_rx_exit(ret); 5012 if (need_bh_off) 5013 local_bh_enable(); 5014 return ret; 5015 } 5016 EXPORT_SYMBOL(netif_rx); 5017 5018 static __latent_entropy void net_tx_action(struct softirq_action *h) 5019 { 5020 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5021 5022 if (sd->completion_queue) { 5023 struct sk_buff *clist; 5024 5025 local_irq_disable(); 5026 clist = sd->completion_queue; 5027 sd->completion_queue = NULL; 5028 local_irq_enable(); 5029 5030 while (clist) { 5031 struct sk_buff *skb = clist; 5032 5033 clist = clist->next; 5034 5035 WARN_ON(refcount_read(&skb->users)); 5036 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5037 trace_consume_skb(skb, net_tx_action); 5038 else 5039 trace_kfree_skb(skb, net_tx_action, 5040 get_kfree_skb_cb(skb)->reason); 5041 5042 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5043 __kfree_skb(skb); 5044 else 5045 __napi_kfree_skb(skb, 5046 get_kfree_skb_cb(skb)->reason); 5047 } 5048 } 5049 5050 if (sd->output_queue) { 5051 struct Qdisc *head; 5052 5053 local_irq_disable(); 5054 head = sd->output_queue; 5055 sd->output_queue = NULL; 5056 sd->output_queue_tailp = &sd->output_queue; 5057 local_irq_enable(); 5058 5059 rcu_read_lock(); 5060 5061 while (head) { 5062 struct Qdisc *q = head; 5063 spinlock_t *root_lock = NULL; 5064 5065 head = head->next_sched; 5066 5067 /* We need to make sure head->next_sched is read 5068 * before clearing __QDISC_STATE_SCHED 5069 */ 5070 smp_mb__before_atomic(); 5071 5072 if (!(q->flags & TCQ_F_NOLOCK)) { 5073 root_lock = qdisc_lock(q); 5074 spin_lock(root_lock); 5075 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5076 &q->state))) { 5077 /* There is a synchronize_net() between 5078 * STATE_DEACTIVATED flag being set and 5079 * qdisc_reset()/some_qdisc_is_busy() in 5080 * dev_deactivate(), so we can safely bail out 5081 * early here to avoid data race between 5082 * qdisc_deactivate() and some_qdisc_is_busy() 5083 * for lockless qdisc. 5084 */ 5085 clear_bit(__QDISC_STATE_SCHED, &q->state); 5086 continue; 5087 } 5088 5089 clear_bit(__QDISC_STATE_SCHED, &q->state); 5090 qdisc_run(q); 5091 if (root_lock) 5092 spin_unlock(root_lock); 5093 } 5094 5095 rcu_read_unlock(); 5096 } 5097 5098 xfrm_dev_backlog(sd); 5099 } 5100 5101 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5102 /* This hook is defined here for ATM LANE */ 5103 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5104 unsigned char *addr) __read_mostly; 5105 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5106 #endif 5107 5108 static inline struct sk_buff * 5109 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5110 struct net_device *orig_dev, bool *another) 5111 { 5112 #ifdef CONFIG_NET_CLS_ACT 5113 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5114 struct tcf_result cl_res; 5115 5116 /* If there's at least one ingress present somewhere (so 5117 * we get here via enabled static key), remaining devices 5118 * that are not configured with an ingress qdisc will bail 5119 * out here. 5120 */ 5121 if (!miniq) 5122 return skb; 5123 5124 if (*pt_prev) { 5125 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5126 *pt_prev = NULL; 5127 } 5128 5129 qdisc_skb_cb(skb)->pkt_len = skb->len; 5130 tc_skb_cb(skb)->mru = 0; 5131 tc_skb_cb(skb)->post_ct = false; 5132 skb->tc_at_ingress = 1; 5133 mini_qdisc_bstats_cpu_update(miniq, skb); 5134 5135 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5136 case TC_ACT_OK: 5137 case TC_ACT_RECLASSIFY: 5138 skb->tc_index = TC_H_MIN(cl_res.classid); 5139 break; 5140 case TC_ACT_SHOT: 5141 mini_qdisc_qstats_cpu_drop(miniq); 5142 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5143 *ret = NET_RX_DROP; 5144 return NULL; 5145 case TC_ACT_STOLEN: 5146 case TC_ACT_QUEUED: 5147 case TC_ACT_TRAP: 5148 consume_skb(skb); 5149 *ret = NET_RX_SUCCESS; 5150 return NULL; 5151 case TC_ACT_REDIRECT: 5152 /* skb_mac_header check was done by cls/act_bpf, so 5153 * we can safely push the L2 header back before 5154 * redirecting to another netdev 5155 */ 5156 __skb_push(skb, skb->mac_len); 5157 if (skb_do_redirect(skb) == -EAGAIN) { 5158 __skb_pull(skb, skb->mac_len); 5159 *another = true; 5160 break; 5161 } 5162 *ret = NET_RX_SUCCESS; 5163 return NULL; 5164 case TC_ACT_CONSUMED: 5165 *ret = NET_RX_SUCCESS; 5166 return NULL; 5167 default: 5168 break; 5169 } 5170 #endif /* CONFIG_NET_CLS_ACT */ 5171 return skb; 5172 } 5173 5174 /** 5175 * netdev_is_rx_handler_busy - check if receive handler is registered 5176 * @dev: device to check 5177 * 5178 * Check if a receive handler is already registered for a given device. 5179 * Return true if there one. 5180 * 5181 * The caller must hold the rtnl_mutex. 5182 */ 5183 bool netdev_is_rx_handler_busy(struct net_device *dev) 5184 { 5185 ASSERT_RTNL(); 5186 return dev && rtnl_dereference(dev->rx_handler); 5187 } 5188 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5189 5190 /** 5191 * netdev_rx_handler_register - register receive handler 5192 * @dev: device to register a handler for 5193 * @rx_handler: receive handler to register 5194 * @rx_handler_data: data pointer that is used by rx handler 5195 * 5196 * Register a receive handler for a device. This handler will then be 5197 * called from __netif_receive_skb. A negative errno code is returned 5198 * on a failure. 5199 * 5200 * The caller must hold the rtnl_mutex. 5201 * 5202 * For a general description of rx_handler, see enum rx_handler_result. 5203 */ 5204 int netdev_rx_handler_register(struct net_device *dev, 5205 rx_handler_func_t *rx_handler, 5206 void *rx_handler_data) 5207 { 5208 if (netdev_is_rx_handler_busy(dev)) 5209 return -EBUSY; 5210 5211 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5212 return -EINVAL; 5213 5214 /* Note: rx_handler_data must be set before rx_handler */ 5215 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5216 rcu_assign_pointer(dev->rx_handler, rx_handler); 5217 5218 return 0; 5219 } 5220 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5221 5222 /** 5223 * netdev_rx_handler_unregister - unregister receive handler 5224 * @dev: device to unregister a handler from 5225 * 5226 * Unregister a receive handler from a device. 5227 * 5228 * The caller must hold the rtnl_mutex. 5229 */ 5230 void netdev_rx_handler_unregister(struct net_device *dev) 5231 { 5232 5233 ASSERT_RTNL(); 5234 RCU_INIT_POINTER(dev->rx_handler, NULL); 5235 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5236 * section has a guarantee to see a non NULL rx_handler_data 5237 * as well. 5238 */ 5239 synchronize_net(); 5240 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5241 } 5242 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5243 5244 /* 5245 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5246 * the special handling of PFMEMALLOC skbs. 5247 */ 5248 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5249 { 5250 switch (skb->protocol) { 5251 case htons(ETH_P_ARP): 5252 case htons(ETH_P_IP): 5253 case htons(ETH_P_IPV6): 5254 case htons(ETH_P_8021Q): 5255 case htons(ETH_P_8021AD): 5256 return true; 5257 default: 5258 return false; 5259 } 5260 } 5261 5262 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5263 int *ret, struct net_device *orig_dev) 5264 { 5265 if (nf_hook_ingress_active(skb)) { 5266 int ingress_retval; 5267 5268 if (*pt_prev) { 5269 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5270 *pt_prev = NULL; 5271 } 5272 5273 rcu_read_lock(); 5274 ingress_retval = nf_hook_ingress(skb); 5275 rcu_read_unlock(); 5276 return ingress_retval; 5277 } 5278 return 0; 5279 } 5280 5281 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5282 struct packet_type **ppt_prev) 5283 { 5284 struct packet_type *ptype, *pt_prev; 5285 rx_handler_func_t *rx_handler; 5286 struct sk_buff *skb = *pskb; 5287 struct net_device *orig_dev; 5288 bool deliver_exact = false; 5289 int ret = NET_RX_DROP; 5290 __be16 type; 5291 5292 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5293 5294 trace_netif_receive_skb(skb); 5295 5296 orig_dev = skb->dev; 5297 5298 skb_reset_network_header(skb); 5299 if (!skb_transport_header_was_set(skb)) 5300 skb_reset_transport_header(skb); 5301 skb_reset_mac_len(skb); 5302 5303 pt_prev = NULL; 5304 5305 another_round: 5306 skb->skb_iif = skb->dev->ifindex; 5307 5308 __this_cpu_inc(softnet_data.processed); 5309 5310 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5311 int ret2; 5312 5313 migrate_disable(); 5314 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5315 migrate_enable(); 5316 5317 if (ret2 != XDP_PASS) { 5318 ret = NET_RX_DROP; 5319 goto out; 5320 } 5321 } 5322 5323 if (eth_type_vlan(skb->protocol)) { 5324 skb = skb_vlan_untag(skb); 5325 if (unlikely(!skb)) 5326 goto out; 5327 } 5328 5329 if (skb_skip_tc_classify(skb)) 5330 goto skip_classify; 5331 5332 if (pfmemalloc) 5333 goto skip_taps; 5334 5335 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5336 if (pt_prev) 5337 ret = deliver_skb(skb, pt_prev, orig_dev); 5338 pt_prev = ptype; 5339 } 5340 5341 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5342 if (pt_prev) 5343 ret = deliver_skb(skb, pt_prev, orig_dev); 5344 pt_prev = ptype; 5345 } 5346 5347 skip_taps: 5348 #ifdef CONFIG_NET_INGRESS 5349 if (static_branch_unlikely(&ingress_needed_key)) { 5350 bool another = false; 5351 5352 nf_skip_egress(skb, true); 5353 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5354 &another); 5355 if (another) 5356 goto another_round; 5357 if (!skb) 5358 goto out; 5359 5360 nf_skip_egress(skb, false); 5361 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5362 goto out; 5363 } 5364 #endif 5365 skb_reset_redirect(skb); 5366 skip_classify: 5367 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5368 goto drop; 5369 5370 if (skb_vlan_tag_present(skb)) { 5371 if (pt_prev) { 5372 ret = deliver_skb(skb, pt_prev, orig_dev); 5373 pt_prev = NULL; 5374 } 5375 if (vlan_do_receive(&skb)) 5376 goto another_round; 5377 else if (unlikely(!skb)) 5378 goto out; 5379 } 5380 5381 rx_handler = rcu_dereference(skb->dev->rx_handler); 5382 if (rx_handler) { 5383 if (pt_prev) { 5384 ret = deliver_skb(skb, pt_prev, orig_dev); 5385 pt_prev = NULL; 5386 } 5387 switch (rx_handler(&skb)) { 5388 case RX_HANDLER_CONSUMED: 5389 ret = NET_RX_SUCCESS; 5390 goto out; 5391 case RX_HANDLER_ANOTHER: 5392 goto another_round; 5393 case RX_HANDLER_EXACT: 5394 deliver_exact = true; 5395 break; 5396 case RX_HANDLER_PASS: 5397 break; 5398 default: 5399 BUG(); 5400 } 5401 } 5402 5403 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5404 check_vlan_id: 5405 if (skb_vlan_tag_get_id(skb)) { 5406 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5407 * find vlan device. 5408 */ 5409 skb->pkt_type = PACKET_OTHERHOST; 5410 } else if (eth_type_vlan(skb->protocol)) { 5411 /* Outer header is 802.1P with vlan 0, inner header is 5412 * 802.1Q or 802.1AD and vlan_do_receive() above could 5413 * not find vlan dev for vlan id 0. 5414 */ 5415 __vlan_hwaccel_clear_tag(skb); 5416 skb = skb_vlan_untag(skb); 5417 if (unlikely(!skb)) 5418 goto out; 5419 if (vlan_do_receive(&skb)) 5420 /* After stripping off 802.1P header with vlan 0 5421 * vlan dev is found for inner header. 5422 */ 5423 goto another_round; 5424 else if (unlikely(!skb)) 5425 goto out; 5426 else 5427 /* We have stripped outer 802.1P vlan 0 header. 5428 * But could not find vlan dev. 5429 * check again for vlan id to set OTHERHOST. 5430 */ 5431 goto check_vlan_id; 5432 } 5433 /* Note: we might in the future use prio bits 5434 * and set skb->priority like in vlan_do_receive() 5435 * For the time being, just ignore Priority Code Point 5436 */ 5437 __vlan_hwaccel_clear_tag(skb); 5438 } 5439 5440 type = skb->protocol; 5441 5442 /* deliver only exact match when indicated */ 5443 if (likely(!deliver_exact)) { 5444 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5445 &ptype_base[ntohs(type) & 5446 PTYPE_HASH_MASK]); 5447 } 5448 5449 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5450 &orig_dev->ptype_specific); 5451 5452 if (unlikely(skb->dev != orig_dev)) { 5453 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5454 &skb->dev->ptype_specific); 5455 } 5456 5457 if (pt_prev) { 5458 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5459 goto drop; 5460 *ppt_prev = pt_prev; 5461 } else { 5462 drop: 5463 if (!deliver_exact) 5464 dev_core_stats_rx_dropped_inc(skb->dev); 5465 else 5466 dev_core_stats_rx_nohandler_inc(skb->dev); 5467 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5468 /* Jamal, now you will not able to escape explaining 5469 * me how you were going to use this. :-) 5470 */ 5471 ret = NET_RX_DROP; 5472 } 5473 5474 out: 5475 /* The invariant here is that if *ppt_prev is not NULL 5476 * then skb should also be non-NULL. 5477 * 5478 * Apparently *ppt_prev assignment above holds this invariant due to 5479 * skb dereferencing near it. 5480 */ 5481 *pskb = skb; 5482 return ret; 5483 } 5484 5485 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5486 { 5487 struct net_device *orig_dev = skb->dev; 5488 struct packet_type *pt_prev = NULL; 5489 int ret; 5490 5491 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5492 if (pt_prev) 5493 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5494 skb->dev, pt_prev, orig_dev); 5495 return ret; 5496 } 5497 5498 /** 5499 * netif_receive_skb_core - special purpose version of netif_receive_skb 5500 * @skb: buffer to process 5501 * 5502 * More direct receive version of netif_receive_skb(). It should 5503 * only be used by callers that have a need to skip RPS and Generic XDP. 5504 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5505 * 5506 * This function may only be called from softirq context and interrupts 5507 * should be enabled. 5508 * 5509 * Return values (usually ignored): 5510 * NET_RX_SUCCESS: no congestion 5511 * NET_RX_DROP: packet was dropped 5512 */ 5513 int netif_receive_skb_core(struct sk_buff *skb) 5514 { 5515 int ret; 5516 5517 rcu_read_lock(); 5518 ret = __netif_receive_skb_one_core(skb, false); 5519 rcu_read_unlock(); 5520 5521 return ret; 5522 } 5523 EXPORT_SYMBOL(netif_receive_skb_core); 5524 5525 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5526 struct packet_type *pt_prev, 5527 struct net_device *orig_dev) 5528 { 5529 struct sk_buff *skb, *next; 5530 5531 if (!pt_prev) 5532 return; 5533 if (list_empty(head)) 5534 return; 5535 if (pt_prev->list_func != NULL) 5536 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5537 ip_list_rcv, head, pt_prev, orig_dev); 5538 else 5539 list_for_each_entry_safe(skb, next, head, list) { 5540 skb_list_del_init(skb); 5541 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5542 } 5543 } 5544 5545 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5546 { 5547 /* Fast-path assumptions: 5548 * - There is no RX handler. 5549 * - Only one packet_type matches. 5550 * If either of these fails, we will end up doing some per-packet 5551 * processing in-line, then handling the 'last ptype' for the whole 5552 * sublist. This can't cause out-of-order delivery to any single ptype, 5553 * because the 'last ptype' must be constant across the sublist, and all 5554 * other ptypes are handled per-packet. 5555 */ 5556 /* Current (common) ptype of sublist */ 5557 struct packet_type *pt_curr = NULL; 5558 /* Current (common) orig_dev of sublist */ 5559 struct net_device *od_curr = NULL; 5560 struct list_head sublist; 5561 struct sk_buff *skb, *next; 5562 5563 INIT_LIST_HEAD(&sublist); 5564 list_for_each_entry_safe(skb, next, head, list) { 5565 struct net_device *orig_dev = skb->dev; 5566 struct packet_type *pt_prev = NULL; 5567 5568 skb_list_del_init(skb); 5569 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5570 if (!pt_prev) 5571 continue; 5572 if (pt_curr != pt_prev || od_curr != orig_dev) { 5573 /* dispatch old sublist */ 5574 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5575 /* start new sublist */ 5576 INIT_LIST_HEAD(&sublist); 5577 pt_curr = pt_prev; 5578 od_curr = orig_dev; 5579 } 5580 list_add_tail(&skb->list, &sublist); 5581 } 5582 5583 /* dispatch final sublist */ 5584 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5585 } 5586 5587 static int __netif_receive_skb(struct sk_buff *skb) 5588 { 5589 int ret; 5590 5591 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5592 unsigned int noreclaim_flag; 5593 5594 /* 5595 * PFMEMALLOC skbs are special, they should 5596 * - be delivered to SOCK_MEMALLOC sockets only 5597 * - stay away from userspace 5598 * - have bounded memory usage 5599 * 5600 * Use PF_MEMALLOC as this saves us from propagating the allocation 5601 * context down to all allocation sites. 5602 */ 5603 noreclaim_flag = memalloc_noreclaim_save(); 5604 ret = __netif_receive_skb_one_core(skb, true); 5605 memalloc_noreclaim_restore(noreclaim_flag); 5606 } else 5607 ret = __netif_receive_skb_one_core(skb, false); 5608 5609 return ret; 5610 } 5611 5612 static void __netif_receive_skb_list(struct list_head *head) 5613 { 5614 unsigned long noreclaim_flag = 0; 5615 struct sk_buff *skb, *next; 5616 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5617 5618 list_for_each_entry_safe(skb, next, head, list) { 5619 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5620 struct list_head sublist; 5621 5622 /* Handle the previous sublist */ 5623 list_cut_before(&sublist, head, &skb->list); 5624 if (!list_empty(&sublist)) 5625 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5626 pfmemalloc = !pfmemalloc; 5627 /* See comments in __netif_receive_skb */ 5628 if (pfmemalloc) 5629 noreclaim_flag = memalloc_noreclaim_save(); 5630 else 5631 memalloc_noreclaim_restore(noreclaim_flag); 5632 } 5633 } 5634 /* Handle the remaining sublist */ 5635 if (!list_empty(head)) 5636 __netif_receive_skb_list_core(head, pfmemalloc); 5637 /* Restore pflags */ 5638 if (pfmemalloc) 5639 memalloc_noreclaim_restore(noreclaim_flag); 5640 } 5641 5642 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5643 { 5644 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5645 struct bpf_prog *new = xdp->prog; 5646 int ret = 0; 5647 5648 switch (xdp->command) { 5649 case XDP_SETUP_PROG: 5650 rcu_assign_pointer(dev->xdp_prog, new); 5651 if (old) 5652 bpf_prog_put(old); 5653 5654 if (old && !new) { 5655 static_branch_dec(&generic_xdp_needed_key); 5656 } else if (new && !old) { 5657 static_branch_inc(&generic_xdp_needed_key); 5658 dev_disable_lro(dev); 5659 dev_disable_gro_hw(dev); 5660 } 5661 break; 5662 5663 default: 5664 ret = -EINVAL; 5665 break; 5666 } 5667 5668 return ret; 5669 } 5670 5671 static int netif_receive_skb_internal(struct sk_buff *skb) 5672 { 5673 int ret; 5674 5675 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5676 5677 if (skb_defer_rx_timestamp(skb)) 5678 return NET_RX_SUCCESS; 5679 5680 rcu_read_lock(); 5681 #ifdef CONFIG_RPS 5682 if (static_branch_unlikely(&rps_needed)) { 5683 struct rps_dev_flow voidflow, *rflow = &voidflow; 5684 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5685 5686 if (cpu >= 0) { 5687 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5688 rcu_read_unlock(); 5689 return ret; 5690 } 5691 } 5692 #endif 5693 ret = __netif_receive_skb(skb); 5694 rcu_read_unlock(); 5695 return ret; 5696 } 5697 5698 void netif_receive_skb_list_internal(struct list_head *head) 5699 { 5700 struct sk_buff *skb, *next; 5701 struct list_head sublist; 5702 5703 INIT_LIST_HEAD(&sublist); 5704 list_for_each_entry_safe(skb, next, head, list) { 5705 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5706 skb_list_del_init(skb); 5707 if (!skb_defer_rx_timestamp(skb)) 5708 list_add_tail(&skb->list, &sublist); 5709 } 5710 list_splice_init(&sublist, head); 5711 5712 rcu_read_lock(); 5713 #ifdef CONFIG_RPS 5714 if (static_branch_unlikely(&rps_needed)) { 5715 list_for_each_entry_safe(skb, next, head, list) { 5716 struct rps_dev_flow voidflow, *rflow = &voidflow; 5717 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5718 5719 if (cpu >= 0) { 5720 /* Will be handled, remove from list */ 5721 skb_list_del_init(skb); 5722 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5723 } 5724 } 5725 } 5726 #endif 5727 __netif_receive_skb_list(head); 5728 rcu_read_unlock(); 5729 } 5730 5731 /** 5732 * netif_receive_skb - process receive buffer from network 5733 * @skb: buffer to process 5734 * 5735 * netif_receive_skb() is the main receive data processing function. 5736 * It always succeeds. The buffer may be dropped during processing 5737 * for congestion control or by the protocol layers. 5738 * 5739 * This function may only be called from softirq context and interrupts 5740 * should be enabled. 5741 * 5742 * Return values (usually ignored): 5743 * NET_RX_SUCCESS: no congestion 5744 * NET_RX_DROP: packet was dropped 5745 */ 5746 int netif_receive_skb(struct sk_buff *skb) 5747 { 5748 int ret; 5749 5750 trace_netif_receive_skb_entry(skb); 5751 5752 ret = netif_receive_skb_internal(skb); 5753 trace_netif_receive_skb_exit(ret); 5754 5755 return ret; 5756 } 5757 EXPORT_SYMBOL(netif_receive_skb); 5758 5759 /** 5760 * netif_receive_skb_list - process many receive buffers from network 5761 * @head: list of skbs to process. 5762 * 5763 * Since return value of netif_receive_skb() is normally ignored, and 5764 * wouldn't be meaningful for a list, this function returns void. 5765 * 5766 * This function may only be called from softirq context and interrupts 5767 * should be enabled. 5768 */ 5769 void netif_receive_skb_list(struct list_head *head) 5770 { 5771 struct sk_buff *skb; 5772 5773 if (list_empty(head)) 5774 return; 5775 if (trace_netif_receive_skb_list_entry_enabled()) { 5776 list_for_each_entry(skb, head, list) 5777 trace_netif_receive_skb_list_entry(skb); 5778 } 5779 netif_receive_skb_list_internal(head); 5780 trace_netif_receive_skb_list_exit(0); 5781 } 5782 EXPORT_SYMBOL(netif_receive_skb_list); 5783 5784 static DEFINE_PER_CPU(struct work_struct, flush_works); 5785 5786 /* Network device is going away, flush any packets still pending */ 5787 static void flush_backlog(struct work_struct *work) 5788 { 5789 struct sk_buff *skb, *tmp; 5790 struct softnet_data *sd; 5791 5792 local_bh_disable(); 5793 sd = this_cpu_ptr(&softnet_data); 5794 5795 rps_lock_irq_disable(sd); 5796 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5797 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5798 __skb_unlink(skb, &sd->input_pkt_queue); 5799 dev_kfree_skb_irq(skb); 5800 input_queue_head_incr(sd); 5801 } 5802 } 5803 rps_unlock_irq_enable(sd); 5804 5805 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5806 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5807 __skb_unlink(skb, &sd->process_queue); 5808 kfree_skb(skb); 5809 input_queue_head_incr(sd); 5810 } 5811 } 5812 local_bh_enable(); 5813 } 5814 5815 static bool flush_required(int cpu) 5816 { 5817 #if IS_ENABLED(CONFIG_RPS) 5818 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5819 bool do_flush; 5820 5821 rps_lock_irq_disable(sd); 5822 5823 /* as insertion into process_queue happens with the rps lock held, 5824 * process_queue access may race only with dequeue 5825 */ 5826 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5827 !skb_queue_empty_lockless(&sd->process_queue); 5828 rps_unlock_irq_enable(sd); 5829 5830 return do_flush; 5831 #endif 5832 /* without RPS we can't safely check input_pkt_queue: during a 5833 * concurrent remote skb_queue_splice() we can detect as empty both 5834 * input_pkt_queue and process_queue even if the latter could end-up 5835 * containing a lot of packets. 5836 */ 5837 return true; 5838 } 5839 5840 static void flush_all_backlogs(void) 5841 { 5842 static cpumask_t flush_cpus; 5843 unsigned int cpu; 5844 5845 /* since we are under rtnl lock protection we can use static data 5846 * for the cpumask and avoid allocating on stack the possibly 5847 * large mask 5848 */ 5849 ASSERT_RTNL(); 5850 5851 cpus_read_lock(); 5852 5853 cpumask_clear(&flush_cpus); 5854 for_each_online_cpu(cpu) { 5855 if (flush_required(cpu)) { 5856 queue_work_on(cpu, system_highpri_wq, 5857 per_cpu_ptr(&flush_works, cpu)); 5858 cpumask_set_cpu(cpu, &flush_cpus); 5859 } 5860 } 5861 5862 /* we can have in flight packet[s] on the cpus we are not flushing, 5863 * synchronize_net() in unregister_netdevice_many() will take care of 5864 * them 5865 */ 5866 for_each_cpu(cpu, &flush_cpus) 5867 flush_work(per_cpu_ptr(&flush_works, cpu)); 5868 5869 cpus_read_unlock(); 5870 } 5871 5872 static void net_rps_send_ipi(struct softnet_data *remsd) 5873 { 5874 #ifdef CONFIG_RPS 5875 while (remsd) { 5876 struct softnet_data *next = remsd->rps_ipi_next; 5877 5878 if (cpu_online(remsd->cpu)) 5879 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5880 remsd = next; 5881 } 5882 #endif 5883 } 5884 5885 /* 5886 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5887 * Note: called with local irq disabled, but exits with local irq enabled. 5888 */ 5889 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5890 { 5891 #ifdef CONFIG_RPS 5892 struct softnet_data *remsd = sd->rps_ipi_list; 5893 5894 if (remsd) { 5895 sd->rps_ipi_list = NULL; 5896 5897 local_irq_enable(); 5898 5899 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5900 net_rps_send_ipi(remsd); 5901 } else 5902 #endif 5903 local_irq_enable(); 5904 } 5905 5906 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5907 { 5908 #ifdef CONFIG_RPS 5909 return sd->rps_ipi_list != NULL; 5910 #else 5911 return false; 5912 #endif 5913 } 5914 5915 static int process_backlog(struct napi_struct *napi, int quota) 5916 { 5917 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5918 bool again = true; 5919 int work = 0; 5920 5921 /* Check if we have pending ipi, its better to send them now, 5922 * not waiting net_rx_action() end. 5923 */ 5924 if (sd_has_rps_ipi_waiting(sd)) { 5925 local_irq_disable(); 5926 net_rps_action_and_irq_enable(sd); 5927 } 5928 5929 napi->weight = READ_ONCE(dev_rx_weight); 5930 while (again) { 5931 struct sk_buff *skb; 5932 5933 while ((skb = __skb_dequeue(&sd->process_queue))) { 5934 rcu_read_lock(); 5935 __netif_receive_skb(skb); 5936 rcu_read_unlock(); 5937 input_queue_head_incr(sd); 5938 if (++work >= quota) 5939 return work; 5940 5941 } 5942 5943 rps_lock_irq_disable(sd); 5944 if (skb_queue_empty(&sd->input_pkt_queue)) { 5945 /* 5946 * Inline a custom version of __napi_complete(). 5947 * only current cpu owns and manipulates this napi, 5948 * and NAPI_STATE_SCHED is the only possible flag set 5949 * on backlog. 5950 * We can use a plain write instead of clear_bit(), 5951 * and we dont need an smp_mb() memory barrier. 5952 */ 5953 napi->state = 0; 5954 again = false; 5955 } else { 5956 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5957 &sd->process_queue); 5958 } 5959 rps_unlock_irq_enable(sd); 5960 } 5961 5962 return work; 5963 } 5964 5965 /** 5966 * __napi_schedule - schedule for receive 5967 * @n: entry to schedule 5968 * 5969 * The entry's receive function will be scheduled to run. 5970 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5971 */ 5972 void __napi_schedule(struct napi_struct *n) 5973 { 5974 unsigned long flags; 5975 5976 local_irq_save(flags); 5977 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5978 local_irq_restore(flags); 5979 } 5980 EXPORT_SYMBOL(__napi_schedule); 5981 5982 /** 5983 * napi_schedule_prep - check if napi can be scheduled 5984 * @n: napi context 5985 * 5986 * Test if NAPI routine is already running, and if not mark 5987 * it as running. This is used as a condition variable to 5988 * insure only one NAPI poll instance runs. We also make 5989 * sure there is no pending NAPI disable. 5990 */ 5991 bool napi_schedule_prep(struct napi_struct *n) 5992 { 5993 unsigned long new, val = READ_ONCE(n->state); 5994 5995 do { 5996 if (unlikely(val & NAPIF_STATE_DISABLE)) 5997 return false; 5998 new = val | NAPIF_STATE_SCHED; 5999 6000 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6001 * This was suggested by Alexander Duyck, as compiler 6002 * emits better code than : 6003 * if (val & NAPIF_STATE_SCHED) 6004 * new |= NAPIF_STATE_MISSED; 6005 */ 6006 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6007 NAPIF_STATE_MISSED; 6008 } while (!try_cmpxchg(&n->state, &val, new)); 6009 6010 return !(val & NAPIF_STATE_SCHED); 6011 } 6012 EXPORT_SYMBOL(napi_schedule_prep); 6013 6014 /** 6015 * __napi_schedule_irqoff - schedule for receive 6016 * @n: entry to schedule 6017 * 6018 * Variant of __napi_schedule() assuming hard irqs are masked. 6019 * 6020 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6021 * because the interrupt disabled assumption might not be true 6022 * due to force-threaded interrupts and spinlock substitution. 6023 */ 6024 void __napi_schedule_irqoff(struct napi_struct *n) 6025 { 6026 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6027 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6028 else 6029 __napi_schedule(n); 6030 } 6031 EXPORT_SYMBOL(__napi_schedule_irqoff); 6032 6033 bool napi_complete_done(struct napi_struct *n, int work_done) 6034 { 6035 unsigned long flags, val, new, timeout = 0; 6036 bool ret = true; 6037 6038 /* 6039 * 1) Don't let napi dequeue from the cpu poll list 6040 * just in case its running on a different cpu. 6041 * 2) If we are busy polling, do nothing here, we have 6042 * the guarantee we will be called later. 6043 */ 6044 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6045 NAPIF_STATE_IN_BUSY_POLL))) 6046 return false; 6047 6048 if (work_done) { 6049 if (n->gro_bitmask) 6050 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6051 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6052 } 6053 if (n->defer_hard_irqs_count > 0) { 6054 n->defer_hard_irqs_count--; 6055 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6056 if (timeout) 6057 ret = false; 6058 } 6059 if (n->gro_bitmask) { 6060 /* When the NAPI instance uses a timeout and keeps postponing 6061 * it, we need to bound somehow the time packets are kept in 6062 * the GRO layer 6063 */ 6064 napi_gro_flush(n, !!timeout); 6065 } 6066 6067 gro_normal_list(n); 6068 6069 if (unlikely(!list_empty(&n->poll_list))) { 6070 /* If n->poll_list is not empty, we need to mask irqs */ 6071 local_irq_save(flags); 6072 list_del_init(&n->poll_list); 6073 local_irq_restore(flags); 6074 } 6075 WRITE_ONCE(n->list_owner, -1); 6076 6077 val = READ_ONCE(n->state); 6078 do { 6079 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6080 6081 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6082 NAPIF_STATE_SCHED_THREADED | 6083 NAPIF_STATE_PREFER_BUSY_POLL); 6084 6085 /* If STATE_MISSED was set, leave STATE_SCHED set, 6086 * because we will call napi->poll() one more time. 6087 * This C code was suggested by Alexander Duyck to help gcc. 6088 */ 6089 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6090 NAPIF_STATE_SCHED; 6091 } while (!try_cmpxchg(&n->state, &val, new)); 6092 6093 if (unlikely(val & NAPIF_STATE_MISSED)) { 6094 __napi_schedule(n); 6095 return false; 6096 } 6097 6098 if (timeout) 6099 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6100 HRTIMER_MODE_REL_PINNED); 6101 return ret; 6102 } 6103 EXPORT_SYMBOL(napi_complete_done); 6104 6105 /* must be called under rcu_read_lock(), as we dont take a reference */ 6106 static struct napi_struct *napi_by_id(unsigned int napi_id) 6107 { 6108 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6109 struct napi_struct *napi; 6110 6111 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6112 if (napi->napi_id == napi_id) 6113 return napi; 6114 6115 return NULL; 6116 } 6117 6118 #if defined(CONFIG_NET_RX_BUSY_POLL) 6119 6120 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6121 { 6122 if (!skip_schedule) { 6123 gro_normal_list(napi); 6124 __napi_schedule(napi); 6125 return; 6126 } 6127 6128 if (napi->gro_bitmask) { 6129 /* flush too old packets 6130 * If HZ < 1000, flush all packets. 6131 */ 6132 napi_gro_flush(napi, HZ >= 1000); 6133 } 6134 6135 gro_normal_list(napi); 6136 clear_bit(NAPI_STATE_SCHED, &napi->state); 6137 } 6138 6139 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6140 u16 budget) 6141 { 6142 bool skip_schedule = false; 6143 unsigned long timeout; 6144 int rc; 6145 6146 /* Busy polling means there is a high chance device driver hard irq 6147 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6148 * set in napi_schedule_prep(). 6149 * Since we are about to call napi->poll() once more, we can safely 6150 * clear NAPI_STATE_MISSED. 6151 * 6152 * Note: x86 could use a single "lock and ..." instruction 6153 * to perform these two clear_bit() 6154 */ 6155 clear_bit(NAPI_STATE_MISSED, &napi->state); 6156 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6157 6158 local_bh_disable(); 6159 6160 if (prefer_busy_poll) { 6161 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6162 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6163 if (napi->defer_hard_irqs_count && timeout) { 6164 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6165 skip_schedule = true; 6166 } 6167 } 6168 6169 /* All we really want here is to re-enable device interrupts. 6170 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6171 */ 6172 rc = napi->poll(napi, budget); 6173 /* We can't gro_normal_list() here, because napi->poll() might have 6174 * rearmed the napi (napi_complete_done()) in which case it could 6175 * already be running on another CPU. 6176 */ 6177 trace_napi_poll(napi, rc, budget); 6178 netpoll_poll_unlock(have_poll_lock); 6179 if (rc == budget) 6180 __busy_poll_stop(napi, skip_schedule); 6181 local_bh_enable(); 6182 } 6183 6184 void napi_busy_loop(unsigned int napi_id, 6185 bool (*loop_end)(void *, unsigned long), 6186 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6187 { 6188 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6189 int (*napi_poll)(struct napi_struct *napi, int budget); 6190 void *have_poll_lock = NULL; 6191 struct napi_struct *napi; 6192 6193 restart: 6194 napi_poll = NULL; 6195 6196 rcu_read_lock(); 6197 6198 napi = napi_by_id(napi_id); 6199 if (!napi) 6200 goto out; 6201 6202 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6203 preempt_disable(); 6204 for (;;) { 6205 int work = 0; 6206 6207 local_bh_disable(); 6208 if (!napi_poll) { 6209 unsigned long val = READ_ONCE(napi->state); 6210 6211 /* If multiple threads are competing for this napi, 6212 * we avoid dirtying napi->state as much as we can. 6213 */ 6214 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6215 NAPIF_STATE_IN_BUSY_POLL)) { 6216 if (prefer_busy_poll) 6217 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6218 goto count; 6219 } 6220 if (cmpxchg(&napi->state, val, 6221 val | NAPIF_STATE_IN_BUSY_POLL | 6222 NAPIF_STATE_SCHED) != val) { 6223 if (prefer_busy_poll) 6224 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6225 goto count; 6226 } 6227 have_poll_lock = netpoll_poll_lock(napi); 6228 napi_poll = napi->poll; 6229 } 6230 work = napi_poll(napi, budget); 6231 trace_napi_poll(napi, work, budget); 6232 gro_normal_list(napi); 6233 count: 6234 if (work > 0) 6235 __NET_ADD_STATS(dev_net(napi->dev), 6236 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6237 local_bh_enable(); 6238 6239 if (!loop_end || loop_end(loop_end_arg, start_time)) 6240 break; 6241 6242 if (unlikely(need_resched())) { 6243 if (napi_poll) 6244 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6245 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6246 preempt_enable(); 6247 rcu_read_unlock(); 6248 cond_resched(); 6249 if (loop_end(loop_end_arg, start_time)) 6250 return; 6251 goto restart; 6252 } 6253 cpu_relax(); 6254 } 6255 if (napi_poll) 6256 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6257 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6258 preempt_enable(); 6259 out: 6260 rcu_read_unlock(); 6261 } 6262 EXPORT_SYMBOL(napi_busy_loop); 6263 6264 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6265 6266 static void napi_hash_add(struct napi_struct *napi) 6267 { 6268 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6269 return; 6270 6271 spin_lock(&napi_hash_lock); 6272 6273 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6274 do { 6275 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6276 napi_gen_id = MIN_NAPI_ID; 6277 } while (napi_by_id(napi_gen_id)); 6278 napi->napi_id = napi_gen_id; 6279 6280 hlist_add_head_rcu(&napi->napi_hash_node, 6281 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6282 6283 spin_unlock(&napi_hash_lock); 6284 } 6285 6286 /* Warning : caller is responsible to make sure rcu grace period 6287 * is respected before freeing memory containing @napi 6288 */ 6289 static void napi_hash_del(struct napi_struct *napi) 6290 { 6291 spin_lock(&napi_hash_lock); 6292 6293 hlist_del_init_rcu(&napi->napi_hash_node); 6294 6295 spin_unlock(&napi_hash_lock); 6296 } 6297 6298 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6299 { 6300 struct napi_struct *napi; 6301 6302 napi = container_of(timer, struct napi_struct, timer); 6303 6304 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6305 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6306 */ 6307 if (!napi_disable_pending(napi) && 6308 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6309 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6310 __napi_schedule_irqoff(napi); 6311 } 6312 6313 return HRTIMER_NORESTART; 6314 } 6315 6316 static void init_gro_hash(struct napi_struct *napi) 6317 { 6318 int i; 6319 6320 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6321 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6322 napi->gro_hash[i].count = 0; 6323 } 6324 napi->gro_bitmask = 0; 6325 } 6326 6327 int dev_set_threaded(struct net_device *dev, bool threaded) 6328 { 6329 struct napi_struct *napi; 6330 int err = 0; 6331 6332 if (dev->threaded == threaded) 6333 return 0; 6334 6335 if (threaded) { 6336 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6337 if (!napi->thread) { 6338 err = napi_kthread_create(napi); 6339 if (err) { 6340 threaded = false; 6341 break; 6342 } 6343 } 6344 } 6345 } 6346 6347 dev->threaded = threaded; 6348 6349 /* Make sure kthread is created before THREADED bit 6350 * is set. 6351 */ 6352 smp_mb__before_atomic(); 6353 6354 /* Setting/unsetting threaded mode on a napi might not immediately 6355 * take effect, if the current napi instance is actively being 6356 * polled. In this case, the switch between threaded mode and 6357 * softirq mode will happen in the next round of napi_schedule(). 6358 * This should not cause hiccups/stalls to the live traffic. 6359 */ 6360 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6361 if (threaded) 6362 set_bit(NAPI_STATE_THREADED, &napi->state); 6363 else 6364 clear_bit(NAPI_STATE_THREADED, &napi->state); 6365 } 6366 6367 return err; 6368 } 6369 EXPORT_SYMBOL(dev_set_threaded); 6370 6371 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6372 int (*poll)(struct napi_struct *, int), int weight) 6373 { 6374 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6375 return; 6376 6377 INIT_LIST_HEAD(&napi->poll_list); 6378 INIT_HLIST_NODE(&napi->napi_hash_node); 6379 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6380 napi->timer.function = napi_watchdog; 6381 init_gro_hash(napi); 6382 napi->skb = NULL; 6383 INIT_LIST_HEAD(&napi->rx_list); 6384 napi->rx_count = 0; 6385 napi->poll = poll; 6386 if (weight > NAPI_POLL_WEIGHT) 6387 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6388 weight); 6389 napi->weight = weight; 6390 napi->dev = dev; 6391 #ifdef CONFIG_NETPOLL 6392 napi->poll_owner = -1; 6393 #endif 6394 napi->list_owner = -1; 6395 set_bit(NAPI_STATE_SCHED, &napi->state); 6396 set_bit(NAPI_STATE_NPSVC, &napi->state); 6397 list_add_rcu(&napi->dev_list, &dev->napi_list); 6398 napi_hash_add(napi); 6399 napi_get_frags_check(napi); 6400 /* Create kthread for this napi if dev->threaded is set. 6401 * Clear dev->threaded if kthread creation failed so that 6402 * threaded mode will not be enabled in napi_enable(). 6403 */ 6404 if (dev->threaded && napi_kthread_create(napi)) 6405 dev->threaded = 0; 6406 } 6407 EXPORT_SYMBOL(netif_napi_add_weight); 6408 6409 void napi_disable(struct napi_struct *n) 6410 { 6411 unsigned long val, new; 6412 6413 might_sleep(); 6414 set_bit(NAPI_STATE_DISABLE, &n->state); 6415 6416 val = READ_ONCE(n->state); 6417 do { 6418 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6419 usleep_range(20, 200); 6420 val = READ_ONCE(n->state); 6421 } 6422 6423 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6424 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6425 } while (!try_cmpxchg(&n->state, &val, new)); 6426 6427 hrtimer_cancel(&n->timer); 6428 6429 clear_bit(NAPI_STATE_DISABLE, &n->state); 6430 } 6431 EXPORT_SYMBOL(napi_disable); 6432 6433 /** 6434 * napi_enable - enable NAPI scheduling 6435 * @n: NAPI context 6436 * 6437 * Resume NAPI from being scheduled on this context. 6438 * Must be paired with napi_disable. 6439 */ 6440 void napi_enable(struct napi_struct *n) 6441 { 6442 unsigned long new, val = READ_ONCE(n->state); 6443 6444 do { 6445 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6446 6447 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6448 if (n->dev->threaded && n->thread) 6449 new |= NAPIF_STATE_THREADED; 6450 } while (!try_cmpxchg(&n->state, &val, new)); 6451 } 6452 EXPORT_SYMBOL(napi_enable); 6453 6454 static void flush_gro_hash(struct napi_struct *napi) 6455 { 6456 int i; 6457 6458 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6459 struct sk_buff *skb, *n; 6460 6461 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6462 kfree_skb(skb); 6463 napi->gro_hash[i].count = 0; 6464 } 6465 } 6466 6467 /* Must be called in process context */ 6468 void __netif_napi_del(struct napi_struct *napi) 6469 { 6470 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6471 return; 6472 6473 napi_hash_del(napi); 6474 list_del_rcu(&napi->dev_list); 6475 napi_free_frags(napi); 6476 6477 flush_gro_hash(napi); 6478 napi->gro_bitmask = 0; 6479 6480 if (napi->thread) { 6481 kthread_stop(napi->thread); 6482 napi->thread = NULL; 6483 } 6484 } 6485 EXPORT_SYMBOL(__netif_napi_del); 6486 6487 static int __napi_poll(struct napi_struct *n, bool *repoll) 6488 { 6489 int work, weight; 6490 6491 weight = n->weight; 6492 6493 /* This NAPI_STATE_SCHED test is for avoiding a race 6494 * with netpoll's poll_napi(). Only the entity which 6495 * obtains the lock and sees NAPI_STATE_SCHED set will 6496 * actually make the ->poll() call. Therefore we avoid 6497 * accidentally calling ->poll() when NAPI is not scheduled. 6498 */ 6499 work = 0; 6500 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6501 work = n->poll(n, weight); 6502 trace_napi_poll(n, work, weight); 6503 } 6504 6505 if (unlikely(work > weight)) 6506 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6507 n->poll, work, weight); 6508 6509 if (likely(work < weight)) 6510 return work; 6511 6512 /* Drivers must not modify the NAPI state if they 6513 * consume the entire weight. In such cases this code 6514 * still "owns" the NAPI instance and therefore can 6515 * move the instance around on the list at-will. 6516 */ 6517 if (unlikely(napi_disable_pending(n))) { 6518 napi_complete(n); 6519 return work; 6520 } 6521 6522 /* The NAPI context has more processing work, but busy-polling 6523 * is preferred. Exit early. 6524 */ 6525 if (napi_prefer_busy_poll(n)) { 6526 if (napi_complete_done(n, work)) { 6527 /* If timeout is not set, we need to make sure 6528 * that the NAPI is re-scheduled. 6529 */ 6530 napi_schedule(n); 6531 } 6532 return work; 6533 } 6534 6535 if (n->gro_bitmask) { 6536 /* flush too old packets 6537 * If HZ < 1000, flush all packets. 6538 */ 6539 napi_gro_flush(n, HZ >= 1000); 6540 } 6541 6542 gro_normal_list(n); 6543 6544 /* Some drivers may have called napi_schedule 6545 * prior to exhausting their budget. 6546 */ 6547 if (unlikely(!list_empty(&n->poll_list))) { 6548 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6549 n->dev ? n->dev->name : "backlog"); 6550 return work; 6551 } 6552 6553 *repoll = true; 6554 6555 return work; 6556 } 6557 6558 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6559 { 6560 bool do_repoll = false; 6561 void *have; 6562 int work; 6563 6564 list_del_init(&n->poll_list); 6565 6566 have = netpoll_poll_lock(n); 6567 6568 work = __napi_poll(n, &do_repoll); 6569 6570 if (do_repoll) 6571 list_add_tail(&n->poll_list, repoll); 6572 6573 netpoll_poll_unlock(have); 6574 6575 return work; 6576 } 6577 6578 static int napi_thread_wait(struct napi_struct *napi) 6579 { 6580 bool woken = false; 6581 6582 set_current_state(TASK_INTERRUPTIBLE); 6583 6584 while (!kthread_should_stop()) { 6585 /* Testing SCHED_THREADED bit here to make sure the current 6586 * kthread owns this napi and could poll on this napi. 6587 * Testing SCHED bit is not enough because SCHED bit might be 6588 * set by some other busy poll thread or by napi_disable(). 6589 */ 6590 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6591 WARN_ON(!list_empty(&napi->poll_list)); 6592 __set_current_state(TASK_RUNNING); 6593 return 0; 6594 } 6595 6596 schedule(); 6597 /* woken being true indicates this thread owns this napi. */ 6598 woken = true; 6599 set_current_state(TASK_INTERRUPTIBLE); 6600 } 6601 __set_current_state(TASK_RUNNING); 6602 6603 return -1; 6604 } 6605 6606 static void skb_defer_free_flush(struct softnet_data *sd) 6607 { 6608 struct sk_buff *skb, *next; 6609 6610 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6611 if (!READ_ONCE(sd->defer_list)) 6612 return; 6613 6614 spin_lock(&sd->defer_lock); 6615 skb = sd->defer_list; 6616 sd->defer_list = NULL; 6617 sd->defer_count = 0; 6618 spin_unlock(&sd->defer_lock); 6619 6620 while (skb != NULL) { 6621 next = skb->next; 6622 napi_consume_skb(skb, 1); 6623 skb = next; 6624 } 6625 } 6626 6627 static int napi_threaded_poll(void *data) 6628 { 6629 struct napi_struct *napi = data; 6630 struct softnet_data *sd; 6631 void *have; 6632 6633 while (!napi_thread_wait(napi)) { 6634 for (;;) { 6635 bool repoll = false; 6636 6637 local_bh_disable(); 6638 sd = this_cpu_ptr(&softnet_data); 6639 sd->in_napi_threaded_poll = true; 6640 6641 have = netpoll_poll_lock(napi); 6642 __napi_poll(napi, &repoll); 6643 netpoll_poll_unlock(have); 6644 6645 sd->in_napi_threaded_poll = false; 6646 barrier(); 6647 6648 if (sd_has_rps_ipi_waiting(sd)) { 6649 local_irq_disable(); 6650 net_rps_action_and_irq_enable(sd); 6651 } 6652 skb_defer_free_flush(sd); 6653 local_bh_enable(); 6654 6655 if (!repoll) 6656 break; 6657 6658 cond_resched(); 6659 } 6660 } 6661 return 0; 6662 } 6663 6664 static __latent_entropy void net_rx_action(struct softirq_action *h) 6665 { 6666 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6667 unsigned long time_limit = jiffies + 6668 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6669 int budget = READ_ONCE(netdev_budget); 6670 LIST_HEAD(list); 6671 LIST_HEAD(repoll); 6672 6673 start: 6674 sd->in_net_rx_action = true; 6675 local_irq_disable(); 6676 list_splice_init(&sd->poll_list, &list); 6677 local_irq_enable(); 6678 6679 for (;;) { 6680 struct napi_struct *n; 6681 6682 skb_defer_free_flush(sd); 6683 6684 if (list_empty(&list)) { 6685 if (list_empty(&repoll)) { 6686 sd->in_net_rx_action = false; 6687 barrier(); 6688 /* We need to check if ____napi_schedule() 6689 * had refilled poll_list while 6690 * sd->in_net_rx_action was true. 6691 */ 6692 if (!list_empty(&sd->poll_list)) 6693 goto start; 6694 if (!sd_has_rps_ipi_waiting(sd)) 6695 goto end; 6696 } 6697 break; 6698 } 6699 6700 n = list_first_entry(&list, struct napi_struct, poll_list); 6701 budget -= napi_poll(n, &repoll); 6702 6703 /* If softirq window is exhausted then punt. 6704 * Allow this to run for 2 jiffies since which will allow 6705 * an average latency of 1.5/HZ. 6706 */ 6707 if (unlikely(budget <= 0 || 6708 time_after_eq(jiffies, time_limit))) { 6709 sd->time_squeeze++; 6710 break; 6711 } 6712 } 6713 6714 local_irq_disable(); 6715 6716 list_splice_tail_init(&sd->poll_list, &list); 6717 list_splice_tail(&repoll, &list); 6718 list_splice(&list, &sd->poll_list); 6719 if (!list_empty(&sd->poll_list)) 6720 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6721 else 6722 sd->in_net_rx_action = false; 6723 6724 net_rps_action_and_irq_enable(sd); 6725 end:; 6726 } 6727 6728 struct netdev_adjacent { 6729 struct net_device *dev; 6730 netdevice_tracker dev_tracker; 6731 6732 /* upper master flag, there can only be one master device per list */ 6733 bool master; 6734 6735 /* lookup ignore flag */ 6736 bool ignore; 6737 6738 /* counter for the number of times this device was added to us */ 6739 u16 ref_nr; 6740 6741 /* private field for the users */ 6742 void *private; 6743 6744 struct list_head list; 6745 struct rcu_head rcu; 6746 }; 6747 6748 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6749 struct list_head *adj_list) 6750 { 6751 struct netdev_adjacent *adj; 6752 6753 list_for_each_entry(adj, adj_list, list) { 6754 if (adj->dev == adj_dev) 6755 return adj; 6756 } 6757 return NULL; 6758 } 6759 6760 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6761 struct netdev_nested_priv *priv) 6762 { 6763 struct net_device *dev = (struct net_device *)priv->data; 6764 6765 return upper_dev == dev; 6766 } 6767 6768 /** 6769 * netdev_has_upper_dev - Check if device is linked to an upper device 6770 * @dev: device 6771 * @upper_dev: upper device to check 6772 * 6773 * Find out if a device is linked to specified upper device and return true 6774 * in case it is. Note that this checks only immediate upper device, 6775 * not through a complete stack of devices. The caller must hold the RTNL lock. 6776 */ 6777 bool netdev_has_upper_dev(struct net_device *dev, 6778 struct net_device *upper_dev) 6779 { 6780 struct netdev_nested_priv priv = { 6781 .data = (void *)upper_dev, 6782 }; 6783 6784 ASSERT_RTNL(); 6785 6786 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6787 &priv); 6788 } 6789 EXPORT_SYMBOL(netdev_has_upper_dev); 6790 6791 /** 6792 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6793 * @dev: device 6794 * @upper_dev: upper device to check 6795 * 6796 * Find out if a device is linked to specified upper device and return true 6797 * in case it is. Note that this checks the entire upper device chain. 6798 * The caller must hold rcu lock. 6799 */ 6800 6801 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6802 struct net_device *upper_dev) 6803 { 6804 struct netdev_nested_priv priv = { 6805 .data = (void *)upper_dev, 6806 }; 6807 6808 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6809 &priv); 6810 } 6811 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6812 6813 /** 6814 * netdev_has_any_upper_dev - Check if device is linked to some device 6815 * @dev: device 6816 * 6817 * Find out if a device is linked to an upper device and return true in case 6818 * it is. The caller must hold the RTNL lock. 6819 */ 6820 bool netdev_has_any_upper_dev(struct net_device *dev) 6821 { 6822 ASSERT_RTNL(); 6823 6824 return !list_empty(&dev->adj_list.upper); 6825 } 6826 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6827 6828 /** 6829 * netdev_master_upper_dev_get - Get master upper device 6830 * @dev: device 6831 * 6832 * Find a master upper device and return pointer to it or NULL in case 6833 * it's not there. The caller must hold the RTNL lock. 6834 */ 6835 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6836 { 6837 struct netdev_adjacent *upper; 6838 6839 ASSERT_RTNL(); 6840 6841 if (list_empty(&dev->adj_list.upper)) 6842 return NULL; 6843 6844 upper = list_first_entry(&dev->adj_list.upper, 6845 struct netdev_adjacent, list); 6846 if (likely(upper->master)) 6847 return upper->dev; 6848 return NULL; 6849 } 6850 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6851 6852 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6853 { 6854 struct netdev_adjacent *upper; 6855 6856 ASSERT_RTNL(); 6857 6858 if (list_empty(&dev->adj_list.upper)) 6859 return NULL; 6860 6861 upper = list_first_entry(&dev->adj_list.upper, 6862 struct netdev_adjacent, list); 6863 if (likely(upper->master) && !upper->ignore) 6864 return upper->dev; 6865 return NULL; 6866 } 6867 6868 /** 6869 * netdev_has_any_lower_dev - Check if device is linked to some device 6870 * @dev: device 6871 * 6872 * Find out if a device is linked to a lower device and return true in case 6873 * it is. The caller must hold the RTNL lock. 6874 */ 6875 static bool netdev_has_any_lower_dev(struct net_device *dev) 6876 { 6877 ASSERT_RTNL(); 6878 6879 return !list_empty(&dev->adj_list.lower); 6880 } 6881 6882 void *netdev_adjacent_get_private(struct list_head *adj_list) 6883 { 6884 struct netdev_adjacent *adj; 6885 6886 adj = list_entry(adj_list, struct netdev_adjacent, list); 6887 6888 return adj->private; 6889 } 6890 EXPORT_SYMBOL(netdev_adjacent_get_private); 6891 6892 /** 6893 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6894 * @dev: device 6895 * @iter: list_head ** of the current position 6896 * 6897 * Gets the next device from the dev's upper list, starting from iter 6898 * position. The caller must hold RCU read lock. 6899 */ 6900 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6901 struct list_head **iter) 6902 { 6903 struct netdev_adjacent *upper; 6904 6905 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6906 6907 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6908 6909 if (&upper->list == &dev->adj_list.upper) 6910 return NULL; 6911 6912 *iter = &upper->list; 6913 6914 return upper->dev; 6915 } 6916 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6917 6918 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6919 struct list_head **iter, 6920 bool *ignore) 6921 { 6922 struct netdev_adjacent *upper; 6923 6924 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6925 6926 if (&upper->list == &dev->adj_list.upper) 6927 return NULL; 6928 6929 *iter = &upper->list; 6930 *ignore = upper->ignore; 6931 6932 return upper->dev; 6933 } 6934 6935 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6936 struct list_head **iter) 6937 { 6938 struct netdev_adjacent *upper; 6939 6940 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6941 6942 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6943 6944 if (&upper->list == &dev->adj_list.upper) 6945 return NULL; 6946 6947 *iter = &upper->list; 6948 6949 return upper->dev; 6950 } 6951 6952 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6953 int (*fn)(struct net_device *dev, 6954 struct netdev_nested_priv *priv), 6955 struct netdev_nested_priv *priv) 6956 { 6957 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6958 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6959 int ret, cur = 0; 6960 bool ignore; 6961 6962 now = dev; 6963 iter = &dev->adj_list.upper; 6964 6965 while (1) { 6966 if (now != dev) { 6967 ret = fn(now, priv); 6968 if (ret) 6969 return ret; 6970 } 6971 6972 next = NULL; 6973 while (1) { 6974 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6975 if (!udev) 6976 break; 6977 if (ignore) 6978 continue; 6979 6980 next = udev; 6981 niter = &udev->adj_list.upper; 6982 dev_stack[cur] = now; 6983 iter_stack[cur++] = iter; 6984 break; 6985 } 6986 6987 if (!next) { 6988 if (!cur) 6989 return 0; 6990 next = dev_stack[--cur]; 6991 niter = iter_stack[cur]; 6992 } 6993 6994 now = next; 6995 iter = niter; 6996 } 6997 6998 return 0; 6999 } 7000 7001 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7002 int (*fn)(struct net_device *dev, 7003 struct netdev_nested_priv *priv), 7004 struct netdev_nested_priv *priv) 7005 { 7006 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7007 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7008 int ret, cur = 0; 7009 7010 now = dev; 7011 iter = &dev->adj_list.upper; 7012 7013 while (1) { 7014 if (now != dev) { 7015 ret = fn(now, priv); 7016 if (ret) 7017 return ret; 7018 } 7019 7020 next = NULL; 7021 while (1) { 7022 udev = netdev_next_upper_dev_rcu(now, &iter); 7023 if (!udev) 7024 break; 7025 7026 next = udev; 7027 niter = &udev->adj_list.upper; 7028 dev_stack[cur] = now; 7029 iter_stack[cur++] = iter; 7030 break; 7031 } 7032 7033 if (!next) { 7034 if (!cur) 7035 return 0; 7036 next = dev_stack[--cur]; 7037 niter = iter_stack[cur]; 7038 } 7039 7040 now = next; 7041 iter = niter; 7042 } 7043 7044 return 0; 7045 } 7046 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7047 7048 static bool __netdev_has_upper_dev(struct net_device *dev, 7049 struct net_device *upper_dev) 7050 { 7051 struct netdev_nested_priv priv = { 7052 .flags = 0, 7053 .data = (void *)upper_dev, 7054 }; 7055 7056 ASSERT_RTNL(); 7057 7058 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7059 &priv); 7060 } 7061 7062 /** 7063 * netdev_lower_get_next_private - Get the next ->private from the 7064 * lower neighbour list 7065 * @dev: device 7066 * @iter: list_head ** of the current position 7067 * 7068 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7069 * list, starting from iter position. The caller must hold either hold the 7070 * RTNL lock or its own locking that guarantees that the neighbour lower 7071 * list will remain unchanged. 7072 */ 7073 void *netdev_lower_get_next_private(struct net_device *dev, 7074 struct list_head **iter) 7075 { 7076 struct netdev_adjacent *lower; 7077 7078 lower = list_entry(*iter, struct netdev_adjacent, list); 7079 7080 if (&lower->list == &dev->adj_list.lower) 7081 return NULL; 7082 7083 *iter = lower->list.next; 7084 7085 return lower->private; 7086 } 7087 EXPORT_SYMBOL(netdev_lower_get_next_private); 7088 7089 /** 7090 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7091 * lower neighbour list, RCU 7092 * variant 7093 * @dev: device 7094 * @iter: list_head ** of the current position 7095 * 7096 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7097 * list, starting from iter position. The caller must hold RCU read lock. 7098 */ 7099 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7100 struct list_head **iter) 7101 { 7102 struct netdev_adjacent *lower; 7103 7104 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7105 7106 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7107 7108 if (&lower->list == &dev->adj_list.lower) 7109 return NULL; 7110 7111 *iter = &lower->list; 7112 7113 return lower->private; 7114 } 7115 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7116 7117 /** 7118 * netdev_lower_get_next - Get the next device from the lower neighbour 7119 * list 7120 * @dev: device 7121 * @iter: list_head ** of the current position 7122 * 7123 * Gets the next netdev_adjacent from the dev's lower neighbour 7124 * list, starting from iter position. The caller must hold RTNL lock or 7125 * its own locking that guarantees that the neighbour lower 7126 * list will remain unchanged. 7127 */ 7128 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7129 { 7130 struct netdev_adjacent *lower; 7131 7132 lower = list_entry(*iter, struct netdev_adjacent, list); 7133 7134 if (&lower->list == &dev->adj_list.lower) 7135 return NULL; 7136 7137 *iter = lower->list.next; 7138 7139 return lower->dev; 7140 } 7141 EXPORT_SYMBOL(netdev_lower_get_next); 7142 7143 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7144 struct list_head **iter) 7145 { 7146 struct netdev_adjacent *lower; 7147 7148 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7149 7150 if (&lower->list == &dev->adj_list.lower) 7151 return NULL; 7152 7153 *iter = &lower->list; 7154 7155 return lower->dev; 7156 } 7157 7158 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7159 struct list_head **iter, 7160 bool *ignore) 7161 { 7162 struct netdev_adjacent *lower; 7163 7164 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7165 7166 if (&lower->list == &dev->adj_list.lower) 7167 return NULL; 7168 7169 *iter = &lower->list; 7170 *ignore = lower->ignore; 7171 7172 return lower->dev; 7173 } 7174 7175 int netdev_walk_all_lower_dev(struct net_device *dev, 7176 int (*fn)(struct net_device *dev, 7177 struct netdev_nested_priv *priv), 7178 struct netdev_nested_priv *priv) 7179 { 7180 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7181 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7182 int ret, cur = 0; 7183 7184 now = dev; 7185 iter = &dev->adj_list.lower; 7186 7187 while (1) { 7188 if (now != dev) { 7189 ret = fn(now, priv); 7190 if (ret) 7191 return ret; 7192 } 7193 7194 next = NULL; 7195 while (1) { 7196 ldev = netdev_next_lower_dev(now, &iter); 7197 if (!ldev) 7198 break; 7199 7200 next = ldev; 7201 niter = &ldev->adj_list.lower; 7202 dev_stack[cur] = now; 7203 iter_stack[cur++] = iter; 7204 break; 7205 } 7206 7207 if (!next) { 7208 if (!cur) 7209 return 0; 7210 next = dev_stack[--cur]; 7211 niter = iter_stack[cur]; 7212 } 7213 7214 now = next; 7215 iter = niter; 7216 } 7217 7218 return 0; 7219 } 7220 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7221 7222 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7223 int (*fn)(struct net_device *dev, 7224 struct netdev_nested_priv *priv), 7225 struct netdev_nested_priv *priv) 7226 { 7227 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7228 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7229 int ret, cur = 0; 7230 bool ignore; 7231 7232 now = dev; 7233 iter = &dev->adj_list.lower; 7234 7235 while (1) { 7236 if (now != dev) { 7237 ret = fn(now, priv); 7238 if (ret) 7239 return ret; 7240 } 7241 7242 next = NULL; 7243 while (1) { 7244 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7245 if (!ldev) 7246 break; 7247 if (ignore) 7248 continue; 7249 7250 next = ldev; 7251 niter = &ldev->adj_list.lower; 7252 dev_stack[cur] = now; 7253 iter_stack[cur++] = iter; 7254 break; 7255 } 7256 7257 if (!next) { 7258 if (!cur) 7259 return 0; 7260 next = dev_stack[--cur]; 7261 niter = iter_stack[cur]; 7262 } 7263 7264 now = next; 7265 iter = niter; 7266 } 7267 7268 return 0; 7269 } 7270 7271 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7272 struct list_head **iter) 7273 { 7274 struct netdev_adjacent *lower; 7275 7276 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7277 if (&lower->list == &dev->adj_list.lower) 7278 return NULL; 7279 7280 *iter = &lower->list; 7281 7282 return lower->dev; 7283 } 7284 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7285 7286 static u8 __netdev_upper_depth(struct net_device *dev) 7287 { 7288 struct net_device *udev; 7289 struct list_head *iter; 7290 u8 max_depth = 0; 7291 bool ignore; 7292 7293 for (iter = &dev->adj_list.upper, 7294 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7295 udev; 7296 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7297 if (ignore) 7298 continue; 7299 if (max_depth < udev->upper_level) 7300 max_depth = udev->upper_level; 7301 } 7302 7303 return max_depth; 7304 } 7305 7306 static u8 __netdev_lower_depth(struct net_device *dev) 7307 { 7308 struct net_device *ldev; 7309 struct list_head *iter; 7310 u8 max_depth = 0; 7311 bool ignore; 7312 7313 for (iter = &dev->adj_list.lower, 7314 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7315 ldev; 7316 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7317 if (ignore) 7318 continue; 7319 if (max_depth < ldev->lower_level) 7320 max_depth = ldev->lower_level; 7321 } 7322 7323 return max_depth; 7324 } 7325 7326 static int __netdev_update_upper_level(struct net_device *dev, 7327 struct netdev_nested_priv *__unused) 7328 { 7329 dev->upper_level = __netdev_upper_depth(dev) + 1; 7330 return 0; 7331 } 7332 7333 #ifdef CONFIG_LOCKDEP 7334 static LIST_HEAD(net_unlink_list); 7335 7336 static void net_unlink_todo(struct net_device *dev) 7337 { 7338 if (list_empty(&dev->unlink_list)) 7339 list_add_tail(&dev->unlink_list, &net_unlink_list); 7340 } 7341 #endif 7342 7343 static int __netdev_update_lower_level(struct net_device *dev, 7344 struct netdev_nested_priv *priv) 7345 { 7346 dev->lower_level = __netdev_lower_depth(dev) + 1; 7347 7348 #ifdef CONFIG_LOCKDEP 7349 if (!priv) 7350 return 0; 7351 7352 if (priv->flags & NESTED_SYNC_IMM) 7353 dev->nested_level = dev->lower_level - 1; 7354 if (priv->flags & NESTED_SYNC_TODO) 7355 net_unlink_todo(dev); 7356 #endif 7357 return 0; 7358 } 7359 7360 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7361 int (*fn)(struct net_device *dev, 7362 struct netdev_nested_priv *priv), 7363 struct netdev_nested_priv *priv) 7364 { 7365 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7366 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7367 int ret, cur = 0; 7368 7369 now = dev; 7370 iter = &dev->adj_list.lower; 7371 7372 while (1) { 7373 if (now != dev) { 7374 ret = fn(now, priv); 7375 if (ret) 7376 return ret; 7377 } 7378 7379 next = NULL; 7380 while (1) { 7381 ldev = netdev_next_lower_dev_rcu(now, &iter); 7382 if (!ldev) 7383 break; 7384 7385 next = ldev; 7386 niter = &ldev->adj_list.lower; 7387 dev_stack[cur] = now; 7388 iter_stack[cur++] = iter; 7389 break; 7390 } 7391 7392 if (!next) { 7393 if (!cur) 7394 return 0; 7395 next = dev_stack[--cur]; 7396 niter = iter_stack[cur]; 7397 } 7398 7399 now = next; 7400 iter = niter; 7401 } 7402 7403 return 0; 7404 } 7405 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7406 7407 /** 7408 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7409 * lower neighbour list, RCU 7410 * variant 7411 * @dev: device 7412 * 7413 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7414 * list. The caller must hold RCU read lock. 7415 */ 7416 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7417 { 7418 struct netdev_adjacent *lower; 7419 7420 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7421 struct netdev_adjacent, list); 7422 if (lower) 7423 return lower->private; 7424 return NULL; 7425 } 7426 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7427 7428 /** 7429 * netdev_master_upper_dev_get_rcu - Get master upper device 7430 * @dev: device 7431 * 7432 * Find a master upper device and return pointer to it or NULL in case 7433 * it's not there. The caller must hold the RCU read lock. 7434 */ 7435 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7436 { 7437 struct netdev_adjacent *upper; 7438 7439 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7440 struct netdev_adjacent, list); 7441 if (upper && likely(upper->master)) 7442 return upper->dev; 7443 return NULL; 7444 } 7445 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7446 7447 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7448 struct net_device *adj_dev, 7449 struct list_head *dev_list) 7450 { 7451 char linkname[IFNAMSIZ+7]; 7452 7453 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7454 "upper_%s" : "lower_%s", adj_dev->name); 7455 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7456 linkname); 7457 } 7458 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7459 char *name, 7460 struct list_head *dev_list) 7461 { 7462 char linkname[IFNAMSIZ+7]; 7463 7464 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7465 "upper_%s" : "lower_%s", name); 7466 sysfs_remove_link(&(dev->dev.kobj), linkname); 7467 } 7468 7469 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7470 struct net_device *adj_dev, 7471 struct list_head *dev_list) 7472 { 7473 return (dev_list == &dev->adj_list.upper || 7474 dev_list == &dev->adj_list.lower) && 7475 net_eq(dev_net(dev), dev_net(adj_dev)); 7476 } 7477 7478 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7479 struct net_device *adj_dev, 7480 struct list_head *dev_list, 7481 void *private, bool master) 7482 { 7483 struct netdev_adjacent *adj; 7484 int ret; 7485 7486 adj = __netdev_find_adj(adj_dev, dev_list); 7487 7488 if (adj) { 7489 adj->ref_nr += 1; 7490 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7491 dev->name, adj_dev->name, adj->ref_nr); 7492 7493 return 0; 7494 } 7495 7496 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7497 if (!adj) 7498 return -ENOMEM; 7499 7500 adj->dev = adj_dev; 7501 adj->master = master; 7502 adj->ref_nr = 1; 7503 adj->private = private; 7504 adj->ignore = false; 7505 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7506 7507 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7508 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7509 7510 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7511 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7512 if (ret) 7513 goto free_adj; 7514 } 7515 7516 /* Ensure that master link is always the first item in list. */ 7517 if (master) { 7518 ret = sysfs_create_link(&(dev->dev.kobj), 7519 &(adj_dev->dev.kobj), "master"); 7520 if (ret) 7521 goto remove_symlinks; 7522 7523 list_add_rcu(&adj->list, dev_list); 7524 } else { 7525 list_add_tail_rcu(&adj->list, dev_list); 7526 } 7527 7528 return 0; 7529 7530 remove_symlinks: 7531 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7532 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7533 free_adj: 7534 netdev_put(adj_dev, &adj->dev_tracker); 7535 kfree(adj); 7536 7537 return ret; 7538 } 7539 7540 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7541 struct net_device *adj_dev, 7542 u16 ref_nr, 7543 struct list_head *dev_list) 7544 { 7545 struct netdev_adjacent *adj; 7546 7547 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7548 dev->name, adj_dev->name, ref_nr); 7549 7550 adj = __netdev_find_adj(adj_dev, dev_list); 7551 7552 if (!adj) { 7553 pr_err("Adjacency does not exist for device %s from %s\n", 7554 dev->name, adj_dev->name); 7555 WARN_ON(1); 7556 return; 7557 } 7558 7559 if (adj->ref_nr > ref_nr) { 7560 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7561 dev->name, adj_dev->name, ref_nr, 7562 adj->ref_nr - ref_nr); 7563 adj->ref_nr -= ref_nr; 7564 return; 7565 } 7566 7567 if (adj->master) 7568 sysfs_remove_link(&(dev->dev.kobj), "master"); 7569 7570 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7571 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7572 7573 list_del_rcu(&adj->list); 7574 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7575 adj_dev->name, dev->name, adj_dev->name); 7576 netdev_put(adj_dev, &adj->dev_tracker); 7577 kfree_rcu(adj, rcu); 7578 } 7579 7580 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7581 struct net_device *upper_dev, 7582 struct list_head *up_list, 7583 struct list_head *down_list, 7584 void *private, bool master) 7585 { 7586 int ret; 7587 7588 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7589 private, master); 7590 if (ret) 7591 return ret; 7592 7593 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7594 private, false); 7595 if (ret) { 7596 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7597 return ret; 7598 } 7599 7600 return 0; 7601 } 7602 7603 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7604 struct net_device *upper_dev, 7605 u16 ref_nr, 7606 struct list_head *up_list, 7607 struct list_head *down_list) 7608 { 7609 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7610 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7611 } 7612 7613 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7614 struct net_device *upper_dev, 7615 void *private, bool master) 7616 { 7617 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7618 &dev->adj_list.upper, 7619 &upper_dev->adj_list.lower, 7620 private, master); 7621 } 7622 7623 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7624 struct net_device *upper_dev) 7625 { 7626 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7627 &dev->adj_list.upper, 7628 &upper_dev->adj_list.lower); 7629 } 7630 7631 static int __netdev_upper_dev_link(struct net_device *dev, 7632 struct net_device *upper_dev, bool master, 7633 void *upper_priv, void *upper_info, 7634 struct netdev_nested_priv *priv, 7635 struct netlink_ext_ack *extack) 7636 { 7637 struct netdev_notifier_changeupper_info changeupper_info = { 7638 .info = { 7639 .dev = dev, 7640 .extack = extack, 7641 }, 7642 .upper_dev = upper_dev, 7643 .master = master, 7644 .linking = true, 7645 .upper_info = upper_info, 7646 }; 7647 struct net_device *master_dev; 7648 int ret = 0; 7649 7650 ASSERT_RTNL(); 7651 7652 if (dev == upper_dev) 7653 return -EBUSY; 7654 7655 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7656 if (__netdev_has_upper_dev(upper_dev, dev)) 7657 return -EBUSY; 7658 7659 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7660 return -EMLINK; 7661 7662 if (!master) { 7663 if (__netdev_has_upper_dev(dev, upper_dev)) 7664 return -EEXIST; 7665 } else { 7666 master_dev = __netdev_master_upper_dev_get(dev); 7667 if (master_dev) 7668 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7669 } 7670 7671 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7672 &changeupper_info.info); 7673 ret = notifier_to_errno(ret); 7674 if (ret) 7675 return ret; 7676 7677 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7678 master); 7679 if (ret) 7680 return ret; 7681 7682 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7683 &changeupper_info.info); 7684 ret = notifier_to_errno(ret); 7685 if (ret) 7686 goto rollback; 7687 7688 __netdev_update_upper_level(dev, NULL); 7689 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7690 7691 __netdev_update_lower_level(upper_dev, priv); 7692 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7693 priv); 7694 7695 return 0; 7696 7697 rollback: 7698 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7699 7700 return ret; 7701 } 7702 7703 /** 7704 * netdev_upper_dev_link - Add a link to the upper device 7705 * @dev: device 7706 * @upper_dev: new upper device 7707 * @extack: netlink extended ack 7708 * 7709 * Adds a link to device which is upper to this one. The caller must hold 7710 * the RTNL lock. On a failure a negative errno code is returned. 7711 * On success the reference counts are adjusted and the function 7712 * returns zero. 7713 */ 7714 int netdev_upper_dev_link(struct net_device *dev, 7715 struct net_device *upper_dev, 7716 struct netlink_ext_ack *extack) 7717 { 7718 struct netdev_nested_priv priv = { 7719 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7720 .data = NULL, 7721 }; 7722 7723 return __netdev_upper_dev_link(dev, upper_dev, false, 7724 NULL, NULL, &priv, extack); 7725 } 7726 EXPORT_SYMBOL(netdev_upper_dev_link); 7727 7728 /** 7729 * netdev_master_upper_dev_link - Add a master link to the upper device 7730 * @dev: device 7731 * @upper_dev: new upper device 7732 * @upper_priv: upper device private 7733 * @upper_info: upper info to be passed down via notifier 7734 * @extack: netlink extended ack 7735 * 7736 * Adds a link to device which is upper to this one. In this case, only 7737 * one master upper device can be linked, although other non-master devices 7738 * might be linked as well. The caller must hold the RTNL lock. 7739 * On a failure a negative errno code is returned. On success the reference 7740 * counts are adjusted and the function returns zero. 7741 */ 7742 int netdev_master_upper_dev_link(struct net_device *dev, 7743 struct net_device *upper_dev, 7744 void *upper_priv, void *upper_info, 7745 struct netlink_ext_ack *extack) 7746 { 7747 struct netdev_nested_priv priv = { 7748 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7749 .data = NULL, 7750 }; 7751 7752 return __netdev_upper_dev_link(dev, upper_dev, true, 7753 upper_priv, upper_info, &priv, extack); 7754 } 7755 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7756 7757 static void __netdev_upper_dev_unlink(struct net_device *dev, 7758 struct net_device *upper_dev, 7759 struct netdev_nested_priv *priv) 7760 { 7761 struct netdev_notifier_changeupper_info changeupper_info = { 7762 .info = { 7763 .dev = dev, 7764 }, 7765 .upper_dev = upper_dev, 7766 .linking = false, 7767 }; 7768 7769 ASSERT_RTNL(); 7770 7771 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7772 7773 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7774 &changeupper_info.info); 7775 7776 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7777 7778 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7779 &changeupper_info.info); 7780 7781 __netdev_update_upper_level(dev, NULL); 7782 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7783 7784 __netdev_update_lower_level(upper_dev, priv); 7785 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7786 priv); 7787 } 7788 7789 /** 7790 * netdev_upper_dev_unlink - Removes a link to upper device 7791 * @dev: device 7792 * @upper_dev: new upper device 7793 * 7794 * Removes a link to device which is upper to this one. The caller must hold 7795 * the RTNL lock. 7796 */ 7797 void netdev_upper_dev_unlink(struct net_device *dev, 7798 struct net_device *upper_dev) 7799 { 7800 struct netdev_nested_priv priv = { 7801 .flags = NESTED_SYNC_TODO, 7802 .data = NULL, 7803 }; 7804 7805 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7806 } 7807 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7808 7809 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7810 struct net_device *lower_dev, 7811 bool val) 7812 { 7813 struct netdev_adjacent *adj; 7814 7815 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7816 if (adj) 7817 adj->ignore = val; 7818 7819 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7820 if (adj) 7821 adj->ignore = val; 7822 } 7823 7824 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7825 struct net_device *lower_dev) 7826 { 7827 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7828 } 7829 7830 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7831 struct net_device *lower_dev) 7832 { 7833 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7834 } 7835 7836 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7837 struct net_device *new_dev, 7838 struct net_device *dev, 7839 struct netlink_ext_ack *extack) 7840 { 7841 struct netdev_nested_priv priv = { 7842 .flags = 0, 7843 .data = NULL, 7844 }; 7845 int err; 7846 7847 if (!new_dev) 7848 return 0; 7849 7850 if (old_dev && new_dev != old_dev) 7851 netdev_adjacent_dev_disable(dev, old_dev); 7852 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7853 extack); 7854 if (err) { 7855 if (old_dev && new_dev != old_dev) 7856 netdev_adjacent_dev_enable(dev, old_dev); 7857 return err; 7858 } 7859 7860 return 0; 7861 } 7862 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7863 7864 void netdev_adjacent_change_commit(struct net_device *old_dev, 7865 struct net_device *new_dev, 7866 struct net_device *dev) 7867 { 7868 struct netdev_nested_priv priv = { 7869 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7870 .data = NULL, 7871 }; 7872 7873 if (!new_dev || !old_dev) 7874 return; 7875 7876 if (new_dev == old_dev) 7877 return; 7878 7879 netdev_adjacent_dev_enable(dev, old_dev); 7880 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7881 } 7882 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7883 7884 void netdev_adjacent_change_abort(struct net_device *old_dev, 7885 struct net_device *new_dev, 7886 struct net_device *dev) 7887 { 7888 struct netdev_nested_priv priv = { 7889 .flags = 0, 7890 .data = NULL, 7891 }; 7892 7893 if (!new_dev) 7894 return; 7895 7896 if (old_dev && new_dev != old_dev) 7897 netdev_adjacent_dev_enable(dev, old_dev); 7898 7899 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7900 } 7901 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7902 7903 /** 7904 * netdev_bonding_info_change - Dispatch event about slave change 7905 * @dev: device 7906 * @bonding_info: info to dispatch 7907 * 7908 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7909 * The caller must hold the RTNL lock. 7910 */ 7911 void netdev_bonding_info_change(struct net_device *dev, 7912 struct netdev_bonding_info *bonding_info) 7913 { 7914 struct netdev_notifier_bonding_info info = { 7915 .info.dev = dev, 7916 }; 7917 7918 memcpy(&info.bonding_info, bonding_info, 7919 sizeof(struct netdev_bonding_info)); 7920 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7921 &info.info); 7922 } 7923 EXPORT_SYMBOL(netdev_bonding_info_change); 7924 7925 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7926 struct netlink_ext_ack *extack) 7927 { 7928 struct netdev_notifier_offload_xstats_info info = { 7929 .info.dev = dev, 7930 .info.extack = extack, 7931 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7932 }; 7933 int err; 7934 int rc; 7935 7936 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7937 GFP_KERNEL); 7938 if (!dev->offload_xstats_l3) 7939 return -ENOMEM; 7940 7941 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7942 NETDEV_OFFLOAD_XSTATS_DISABLE, 7943 &info.info); 7944 err = notifier_to_errno(rc); 7945 if (err) 7946 goto free_stats; 7947 7948 return 0; 7949 7950 free_stats: 7951 kfree(dev->offload_xstats_l3); 7952 dev->offload_xstats_l3 = NULL; 7953 return err; 7954 } 7955 7956 int netdev_offload_xstats_enable(struct net_device *dev, 7957 enum netdev_offload_xstats_type type, 7958 struct netlink_ext_ack *extack) 7959 { 7960 ASSERT_RTNL(); 7961 7962 if (netdev_offload_xstats_enabled(dev, type)) 7963 return -EALREADY; 7964 7965 switch (type) { 7966 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7967 return netdev_offload_xstats_enable_l3(dev, extack); 7968 } 7969 7970 WARN_ON(1); 7971 return -EINVAL; 7972 } 7973 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7974 7975 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7976 { 7977 struct netdev_notifier_offload_xstats_info info = { 7978 .info.dev = dev, 7979 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7980 }; 7981 7982 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7983 &info.info); 7984 kfree(dev->offload_xstats_l3); 7985 dev->offload_xstats_l3 = NULL; 7986 } 7987 7988 int netdev_offload_xstats_disable(struct net_device *dev, 7989 enum netdev_offload_xstats_type type) 7990 { 7991 ASSERT_RTNL(); 7992 7993 if (!netdev_offload_xstats_enabled(dev, type)) 7994 return -EALREADY; 7995 7996 switch (type) { 7997 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7998 netdev_offload_xstats_disable_l3(dev); 7999 return 0; 8000 } 8001 8002 WARN_ON(1); 8003 return -EINVAL; 8004 } 8005 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8006 8007 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8008 { 8009 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8010 } 8011 8012 static struct rtnl_hw_stats64 * 8013 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8014 enum netdev_offload_xstats_type type) 8015 { 8016 switch (type) { 8017 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8018 return dev->offload_xstats_l3; 8019 } 8020 8021 WARN_ON(1); 8022 return NULL; 8023 } 8024 8025 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8026 enum netdev_offload_xstats_type type) 8027 { 8028 ASSERT_RTNL(); 8029 8030 return netdev_offload_xstats_get_ptr(dev, type); 8031 } 8032 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8033 8034 struct netdev_notifier_offload_xstats_ru { 8035 bool used; 8036 }; 8037 8038 struct netdev_notifier_offload_xstats_rd { 8039 struct rtnl_hw_stats64 stats; 8040 bool used; 8041 }; 8042 8043 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8044 const struct rtnl_hw_stats64 *src) 8045 { 8046 dest->rx_packets += src->rx_packets; 8047 dest->tx_packets += src->tx_packets; 8048 dest->rx_bytes += src->rx_bytes; 8049 dest->tx_bytes += src->tx_bytes; 8050 dest->rx_errors += src->rx_errors; 8051 dest->tx_errors += src->tx_errors; 8052 dest->rx_dropped += src->rx_dropped; 8053 dest->tx_dropped += src->tx_dropped; 8054 dest->multicast += src->multicast; 8055 } 8056 8057 static int netdev_offload_xstats_get_used(struct net_device *dev, 8058 enum netdev_offload_xstats_type type, 8059 bool *p_used, 8060 struct netlink_ext_ack *extack) 8061 { 8062 struct netdev_notifier_offload_xstats_ru report_used = {}; 8063 struct netdev_notifier_offload_xstats_info info = { 8064 .info.dev = dev, 8065 .info.extack = extack, 8066 .type = type, 8067 .report_used = &report_used, 8068 }; 8069 int rc; 8070 8071 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8072 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8073 &info.info); 8074 *p_used = report_used.used; 8075 return notifier_to_errno(rc); 8076 } 8077 8078 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8079 enum netdev_offload_xstats_type type, 8080 struct rtnl_hw_stats64 *p_stats, 8081 bool *p_used, 8082 struct netlink_ext_ack *extack) 8083 { 8084 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8085 struct netdev_notifier_offload_xstats_info info = { 8086 .info.dev = dev, 8087 .info.extack = extack, 8088 .type = type, 8089 .report_delta = &report_delta, 8090 }; 8091 struct rtnl_hw_stats64 *stats; 8092 int rc; 8093 8094 stats = netdev_offload_xstats_get_ptr(dev, type); 8095 if (WARN_ON(!stats)) 8096 return -EINVAL; 8097 8098 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8099 &info.info); 8100 8101 /* Cache whatever we got, even if there was an error, otherwise the 8102 * successful stats retrievals would get lost. 8103 */ 8104 netdev_hw_stats64_add(stats, &report_delta.stats); 8105 8106 if (p_stats) 8107 *p_stats = *stats; 8108 *p_used = report_delta.used; 8109 8110 return notifier_to_errno(rc); 8111 } 8112 8113 int netdev_offload_xstats_get(struct net_device *dev, 8114 enum netdev_offload_xstats_type type, 8115 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8116 struct netlink_ext_ack *extack) 8117 { 8118 ASSERT_RTNL(); 8119 8120 if (p_stats) 8121 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8122 p_used, extack); 8123 else 8124 return netdev_offload_xstats_get_used(dev, type, p_used, 8125 extack); 8126 } 8127 EXPORT_SYMBOL(netdev_offload_xstats_get); 8128 8129 void 8130 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8131 const struct rtnl_hw_stats64 *stats) 8132 { 8133 report_delta->used = true; 8134 netdev_hw_stats64_add(&report_delta->stats, stats); 8135 } 8136 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8137 8138 void 8139 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8140 { 8141 report_used->used = true; 8142 } 8143 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8144 8145 void netdev_offload_xstats_push_delta(struct net_device *dev, 8146 enum netdev_offload_xstats_type type, 8147 const struct rtnl_hw_stats64 *p_stats) 8148 { 8149 struct rtnl_hw_stats64 *stats; 8150 8151 ASSERT_RTNL(); 8152 8153 stats = netdev_offload_xstats_get_ptr(dev, type); 8154 if (WARN_ON(!stats)) 8155 return; 8156 8157 netdev_hw_stats64_add(stats, p_stats); 8158 } 8159 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8160 8161 /** 8162 * netdev_get_xmit_slave - Get the xmit slave of master device 8163 * @dev: device 8164 * @skb: The packet 8165 * @all_slaves: assume all the slaves are active 8166 * 8167 * The reference counters are not incremented so the caller must be 8168 * careful with locks. The caller must hold RCU lock. 8169 * %NULL is returned if no slave is found. 8170 */ 8171 8172 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8173 struct sk_buff *skb, 8174 bool all_slaves) 8175 { 8176 const struct net_device_ops *ops = dev->netdev_ops; 8177 8178 if (!ops->ndo_get_xmit_slave) 8179 return NULL; 8180 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8181 } 8182 EXPORT_SYMBOL(netdev_get_xmit_slave); 8183 8184 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8185 struct sock *sk) 8186 { 8187 const struct net_device_ops *ops = dev->netdev_ops; 8188 8189 if (!ops->ndo_sk_get_lower_dev) 8190 return NULL; 8191 return ops->ndo_sk_get_lower_dev(dev, sk); 8192 } 8193 8194 /** 8195 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8196 * @dev: device 8197 * @sk: the socket 8198 * 8199 * %NULL is returned if no lower device is found. 8200 */ 8201 8202 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8203 struct sock *sk) 8204 { 8205 struct net_device *lower; 8206 8207 lower = netdev_sk_get_lower_dev(dev, sk); 8208 while (lower) { 8209 dev = lower; 8210 lower = netdev_sk_get_lower_dev(dev, sk); 8211 } 8212 8213 return dev; 8214 } 8215 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8216 8217 static void netdev_adjacent_add_links(struct net_device *dev) 8218 { 8219 struct netdev_adjacent *iter; 8220 8221 struct net *net = dev_net(dev); 8222 8223 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8224 if (!net_eq(net, dev_net(iter->dev))) 8225 continue; 8226 netdev_adjacent_sysfs_add(iter->dev, dev, 8227 &iter->dev->adj_list.lower); 8228 netdev_adjacent_sysfs_add(dev, iter->dev, 8229 &dev->adj_list.upper); 8230 } 8231 8232 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8233 if (!net_eq(net, dev_net(iter->dev))) 8234 continue; 8235 netdev_adjacent_sysfs_add(iter->dev, dev, 8236 &iter->dev->adj_list.upper); 8237 netdev_adjacent_sysfs_add(dev, iter->dev, 8238 &dev->adj_list.lower); 8239 } 8240 } 8241 8242 static void netdev_adjacent_del_links(struct net_device *dev) 8243 { 8244 struct netdev_adjacent *iter; 8245 8246 struct net *net = dev_net(dev); 8247 8248 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8249 if (!net_eq(net, dev_net(iter->dev))) 8250 continue; 8251 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8252 &iter->dev->adj_list.lower); 8253 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8254 &dev->adj_list.upper); 8255 } 8256 8257 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8258 if (!net_eq(net, dev_net(iter->dev))) 8259 continue; 8260 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8261 &iter->dev->adj_list.upper); 8262 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8263 &dev->adj_list.lower); 8264 } 8265 } 8266 8267 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8268 { 8269 struct netdev_adjacent *iter; 8270 8271 struct net *net = dev_net(dev); 8272 8273 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8274 if (!net_eq(net, dev_net(iter->dev))) 8275 continue; 8276 netdev_adjacent_sysfs_del(iter->dev, oldname, 8277 &iter->dev->adj_list.lower); 8278 netdev_adjacent_sysfs_add(iter->dev, dev, 8279 &iter->dev->adj_list.lower); 8280 } 8281 8282 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8283 if (!net_eq(net, dev_net(iter->dev))) 8284 continue; 8285 netdev_adjacent_sysfs_del(iter->dev, oldname, 8286 &iter->dev->adj_list.upper); 8287 netdev_adjacent_sysfs_add(iter->dev, dev, 8288 &iter->dev->adj_list.upper); 8289 } 8290 } 8291 8292 void *netdev_lower_dev_get_private(struct net_device *dev, 8293 struct net_device *lower_dev) 8294 { 8295 struct netdev_adjacent *lower; 8296 8297 if (!lower_dev) 8298 return NULL; 8299 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8300 if (!lower) 8301 return NULL; 8302 8303 return lower->private; 8304 } 8305 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8306 8307 8308 /** 8309 * netdev_lower_state_changed - Dispatch event about lower device state change 8310 * @lower_dev: device 8311 * @lower_state_info: state to dispatch 8312 * 8313 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8314 * The caller must hold the RTNL lock. 8315 */ 8316 void netdev_lower_state_changed(struct net_device *lower_dev, 8317 void *lower_state_info) 8318 { 8319 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8320 .info.dev = lower_dev, 8321 }; 8322 8323 ASSERT_RTNL(); 8324 changelowerstate_info.lower_state_info = lower_state_info; 8325 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8326 &changelowerstate_info.info); 8327 } 8328 EXPORT_SYMBOL(netdev_lower_state_changed); 8329 8330 static void dev_change_rx_flags(struct net_device *dev, int flags) 8331 { 8332 const struct net_device_ops *ops = dev->netdev_ops; 8333 8334 if (ops->ndo_change_rx_flags) 8335 ops->ndo_change_rx_flags(dev, flags); 8336 } 8337 8338 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8339 { 8340 unsigned int old_flags = dev->flags; 8341 kuid_t uid; 8342 kgid_t gid; 8343 8344 ASSERT_RTNL(); 8345 8346 dev->flags |= IFF_PROMISC; 8347 dev->promiscuity += inc; 8348 if (dev->promiscuity == 0) { 8349 /* 8350 * Avoid overflow. 8351 * If inc causes overflow, untouch promisc and return error. 8352 */ 8353 if (inc < 0) 8354 dev->flags &= ~IFF_PROMISC; 8355 else { 8356 dev->promiscuity -= inc; 8357 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8358 return -EOVERFLOW; 8359 } 8360 } 8361 if (dev->flags != old_flags) { 8362 netdev_info(dev, "%s promiscuous mode\n", 8363 dev->flags & IFF_PROMISC ? "entered" : "left"); 8364 if (audit_enabled) { 8365 current_uid_gid(&uid, &gid); 8366 audit_log(audit_context(), GFP_ATOMIC, 8367 AUDIT_ANOM_PROMISCUOUS, 8368 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8369 dev->name, (dev->flags & IFF_PROMISC), 8370 (old_flags & IFF_PROMISC), 8371 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8372 from_kuid(&init_user_ns, uid), 8373 from_kgid(&init_user_ns, gid), 8374 audit_get_sessionid(current)); 8375 } 8376 8377 dev_change_rx_flags(dev, IFF_PROMISC); 8378 } 8379 if (notify) 8380 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8381 return 0; 8382 } 8383 8384 /** 8385 * dev_set_promiscuity - update promiscuity count on a device 8386 * @dev: device 8387 * @inc: modifier 8388 * 8389 * Add or remove promiscuity from a device. While the count in the device 8390 * remains above zero the interface remains promiscuous. Once it hits zero 8391 * the device reverts back to normal filtering operation. A negative inc 8392 * value is used to drop promiscuity on the device. 8393 * Return 0 if successful or a negative errno code on error. 8394 */ 8395 int dev_set_promiscuity(struct net_device *dev, int inc) 8396 { 8397 unsigned int old_flags = dev->flags; 8398 int err; 8399 8400 err = __dev_set_promiscuity(dev, inc, true); 8401 if (err < 0) 8402 return err; 8403 if (dev->flags != old_flags) 8404 dev_set_rx_mode(dev); 8405 return err; 8406 } 8407 EXPORT_SYMBOL(dev_set_promiscuity); 8408 8409 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8410 { 8411 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8412 8413 ASSERT_RTNL(); 8414 8415 dev->flags |= IFF_ALLMULTI; 8416 dev->allmulti += inc; 8417 if (dev->allmulti == 0) { 8418 /* 8419 * Avoid overflow. 8420 * If inc causes overflow, untouch allmulti and return error. 8421 */ 8422 if (inc < 0) 8423 dev->flags &= ~IFF_ALLMULTI; 8424 else { 8425 dev->allmulti -= inc; 8426 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8427 return -EOVERFLOW; 8428 } 8429 } 8430 if (dev->flags ^ old_flags) { 8431 netdev_info(dev, "%s allmulticast mode\n", 8432 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8433 dev_change_rx_flags(dev, IFF_ALLMULTI); 8434 dev_set_rx_mode(dev); 8435 if (notify) 8436 __dev_notify_flags(dev, old_flags, 8437 dev->gflags ^ old_gflags, 0, NULL); 8438 } 8439 return 0; 8440 } 8441 8442 /** 8443 * dev_set_allmulti - update allmulti count on a device 8444 * @dev: device 8445 * @inc: modifier 8446 * 8447 * Add or remove reception of all multicast frames to a device. While the 8448 * count in the device remains above zero the interface remains listening 8449 * to all interfaces. Once it hits zero the device reverts back to normal 8450 * filtering operation. A negative @inc value is used to drop the counter 8451 * when releasing a resource needing all multicasts. 8452 * Return 0 if successful or a negative errno code on error. 8453 */ 8454 8455 int dev_set_allmulti(struct net_device *dev, int inc) 8456 { 8457 return __dev_set_allmulti(dev, inc, true); 8458 } 8459 EXPORT_SYMBOL(dev_set_allmulti); 8460 8461 /* 8462 * Upload unicast and multicast address lists to device and 8463 * configure RX filtering. When the device doesn't support unicast 8464 * filtering it is put in promiscuous mode while unicast addresses 8465 * are present. 8466 */ 8467 void __dev_set_rx_mode(struct net_device *dev) 8468 { 8469 const struct net_device_ops *ops = dev->netdev_ops; 8470 8471 /* dev_open will call this function so the list will stay sane. */ 8472 if (!(dev->flags&IFF_UP)) 8473 return; 8474 8475 if (!netif_device_present(dev)) 8476 return; 8477 8478 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8479 /* Unicast addresses changes may only happen under the rtnl, 8480 * therefore calling __dev_set_promiscuity here is safe. 8481 */ 8482 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8483 __dev_set_promiscuity(dev, 1, false); 8484 dev->uc_promisc = true; 8485 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8486 __dev_set_promiscuity(dev, -1, false); 8487 dev->uc_promisc = false; 8488 } 8489 } 8490 8491 if (ops->ndo_set_rx_mode) 8492 ops->ndo_set_rx_mode(dev); 8493 } 8494 8495 void dev_set_rx_mode(struct net_device *dev) 8496 { 8497 netif_addr_lock_bh(dev); 8498 __dev_set_rx_mode(dev); 8499 netif_addr_unlock_bh(dev); 8500 } 8501 8502 /** 8503 * dev_get_flags - get flags reported to userspace 8504 * @dev: device 8505 * 8506 * Get the combination of flag bits exported through APIs to userspace. 8507 */ 8508 unsigned int dev_get_flags(const struct net_device *dev) 8509 { 8510 unsigned int flags; 8511 8512 flags = (dev->flags & ~(IFF_PROMISC | 8513 IFF_ALLMULTI | 8514 IFF_RUNNING | 8515 IFF_LOWER_UP | 8516 IFF_DORMANT)) | 8517 (dev->gflags & (IFF_PROMISC | 8518 IFF_ALLMULTI)); 8519 8520 if (netif_running(dev)) { 8521 if (netif_oper_up(dev)) 8522 flags |= IFF_RUNNING; 8523 if (netif_carrier_ok(dev)) 8524 flags |= IFF_LOWER_UP; 8525 if (netif_dormant(dev)) 8526 flags |= IFF_DORMANT; 8527 } 8528 8529 return flags; 8530 } 8531 EXPORT_SYMBOL(dev_get_flags); 8532 8533 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8534 struct netlink_ext_ack *extack) 8535 { 8536 unsigned int old_flags = dev->flags; 8537 int ret; 8538 8539 ASSERT_RTNL(); 8540 8541 /* 8542 * Set the flags on our device. 8543 */ 8544 8545 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8546 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8547 IFF_AUTOMEDIA)) | 8548 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8549 IFF_ALLMULTI)); 8550 8551 /* 8552 * Load in the correct multicast list now the flags have changed. 8553 */ 8554 8555 if ((old_flags ^ flags) & IFF_MULTICAST) 8556 dev_change_rx_flags(dev, IFF_MULTICAST); 8557 8558 dev_set_rx_mode(dev); 8559 8560 /* 8561 * Have we downed the interface. We handle IFF_UP ourselves 8562 * according to user attempts to set it, rather than blindly 8563 * setting it. 8564 */ 8565 8566 ret = 0; 8567 if ((old_flags ^ flags) & IFF_UP) { 8568 if (old_flags & IFF_UP) 8569 __dev_close(dev); 8570 else 8571 ret = __dev_open(dev, extack); 8572 } 8573 8574 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8575 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8576 unsigned int old_flags = dev->flags; 8577 8578 dev->gflags ^= IFF_PROMISC; 8579 8580 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8581 if (dev->flags != old_flags) 8582 dev_set_rx_mode(dev); 8583 } 8584 8585 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8586 * is important. Some (broken) drivers set IFF_PROMISC, when 8587 * IFF_ALLMULTI is requested not asking us and not reporting. 8588 */ 8589 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8590 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8591 8592 dev->gflags ^= IFF_ALLMULTI; 8593 __dev_set_allmulti(dev, inc, false); 8594 } 8595 8596 return ret; 8597 } 8598 8599 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8600 unsigned int gchanges, u32 portid, 8601 const struct nlmsghdr *nlh) 8602 { 8603 unsigned int changes = dev->flags ^ old_flags; 8604 8605 if (gchanges) 8606 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8607 8608 if (changes & IFF_UP) { 8609 if (dev->flags & IFF_UP) 8610 call_netdevice_notifiers(NETDEV_UP, dev); 8611 else 8612 call_netdevice_notifiers(NETDEV_DOWN, dev); 8613 } 8614 8615 if (dev->flags & IFF_UP && 8616 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8617 struct netdev_notifier_change_info change_info = { 8618 .info = { 8619 .dev = dev, 8620 }, 8621 .flags_changed = changes, 8622 }; 8623 8624 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8625 } 8626 } 8627 8628 /** 8629 * dev_change_flags - change device settings 8630 * @dev: device 8631 * @flags: device state flags 8632 * @extack: netlink extended ack 8633 * 8634 * Change settings on device based state flags. The flags are 8635 * in the userspace exported format. 8636 */ 8637 int dev_change_flags(struct net_device *dev, unsigned int flags, 8638 struct netlink_ext_ack *extack) 8639 { 8640 int ret; 8641 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8642 8643 ret = __dev_change_flags(dev, flags, extack); 8644 if (ret < 0) 8645 return ret; 8646 8647 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8648 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8649 return ret; 8650 } 8651 EXPORT_SYMBOL(dev_change_flags); 8652 8653 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8654 { 8655 const struct net_device_ops *ops = dev->netdev_ops; 8656 8657 if (ops->ndo_change_mtu) 8658 return ops->ndo_change_mtu(dev, new_mtu); 8659 8660 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8661 WRITE_ONCE(dev->mtu, new_mtu); 8662 return 0; 8663 } 8664 EXPORT_SYMBOL(__dev_set_mtu); 8665 8666 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8667 struct netlink_ext_ack *extack) 8668 { 8669 /* MTU must be positive, and in range */ 8670 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8671 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8672 return -EINVAL; 8673 } 8674 8675 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8676 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8677 return -EINVAL; 8678 } 8679 return 0; 8680 } 8681 8682 /** 8683 * dev_set_mtu_ext - Change maximum transfer unit 8684 * @dev: device 8685 * @new_mtu: new transfer unit 8686 * @extack: netlink extended ack 8687 * 8688 * Change the maximum transfer size of the network device. 8689 */ 8690 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8691 struct netlink_ext_ack *extack) 8692 { 8693 int err, orig_mtu; 8694 8695 if (new_mtu == dev->mtu) 8696 return 0; 8697 8698 err = dev_validate_mtu(dev, new_mtu, extack); 8699 if (err) 8700 return err; 8701 8702 if (!netif_device_present(dev)) 8703 return -ENODEV; 8704 8705 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8706 err = notifier_to_errno(err); 8707 if (err) 8708 return err; 8709 8710 orig_mtu = dev->mtu; 8711 err = __dev_set_mtu(dev, new_mtu); 8712 8713 if (!err) { 8714 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8715 orig_mtu); 8716 err = notifier_to_errno(err); 8717 if (err) { 8718 /* setting mtu back and notifying everyone again, 8719 * so that they have a chance to revert changes. 8720 */ 8721 __dev_set_mtu(dev, orig_mtu); 8722 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8723 new_mtu); 8724 } 8725 } 8726 return err; 8727 } 8728 8729 int dev_set_mtu(struct net_device *dev, int new_mtu) 8730 { 8731 struct netlink_ext_ack extack; 8732 int err; 8733 8734 memset(&extack, 0, sizeof(extack)); 8735 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8736 if (err && extack._msg) 8737 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8738 return err; 8739 } 8740 EXPORT_SYMBOL(dev_set_mtu); 8741 8742 /** 8743 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8744 * @dev: device 8745 * @new_len: new tx queue length 8746 */ 8747 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8748 { 8749 unsigned int orig_len = dev->tx_queue_len; 8750 int res; 8751 8752 if (new_len != (unsigned int)new_len) 8753 return -ERANGE; 8754 8755 if (new_len != orig_len) { 8756 dev->tx_queue_len = new_len; 8757 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8758 res = notifier_to_errno(res); 8759 if (res) 8760 goto err_rollback; 8761 res = dev_qdisc_change_tx_queue_len(dev); 8762 if (res) 8763 goto err_rollback; 8764 } 8765 8766 return 0; 8767 8768 err_rollback: 8769 netdev_err(dev, "refused to change device tx_queue_len\n"); 8770 dev->tx_queue_len = orig_len; 8771 return res; 8772 } 8773 8774 /** 8775 * dev_set_group - Change group this device belongs to 8776 * @dev: device 8777 * @new_group: group this device should belong to 8778 */ 8779 void dev_set_group(struct net_device *dev, int new_group) 8780 { 8781 dev->group = new_group; 8782 } 8783 8784 /** 8785 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8786 * @dev: device 8787 * @addr: new address 8788 * @extack: netlink extended ack 8789 */ 8790 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8791 struct netlink_ext_ack *extack) 8792 { 8793 struct netdev_notifier_pre_changeaddr_info info = { 8794 .info.dev = dev, 8795 .info.extack = extack, 8796 .dev_addr = addr, 8797 }; 8798 int rc; 8799 8800 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8801 return notifier_to_errno(rc); 8802 } 8803 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8804 8805 /** 8806 * dev_set_mac_address - Change Media Access Control Address 8807 * @dev: device 8808 * @sa: new address 8809 * @extack: netlink extended ack 8810 * 8811 * Change the hardware (MAC) address of the device 8812 */ 8813 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8814 struct netlink_ext_ack *extack) 8815 { 8816 const struct net_device_ops *ops = dev->netdev_ops; 8817 int err; 8818 8819 if (!ops->ndo_set_mac_address) 8820 return -EOPNOTSUPP; 8821 if (sa->sa_family != dev->type) 8822 return -EINVAL; 8823 if (!netif_device_present(dev)) 8824 return -ENODEV; 8825 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8826 if (err) 8827 return err; 8828 err = ops->ndo_set_mac_address(dev, sa); 8829 if (err) 8830 return err; 8831 dev->addr_assign_type = NET_ADDR_SET; 8832 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8833 add_device_randomness(dev->dev_addr, dev->addr_len); 8834 return 0; 8835 } 8836 EXPORT_SYMBOL(dev_set_mac_address); 8837 8838 static DECLARE_RWSEM(dev_addr_sem); 8839 8840 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8841 struct netlink_ext_ack *extack) 8842 { 8843 int ret; 8844 8845 down_write(&dev_addr_sem); 8846 ret = dev_set_mac_address(dev, sa, extack); 8847 up_write(&dev_addr_sem); 8848 return ret; 8849 } 8850 EXPORT_SYMBOL(dev_set_mac_address_user); 8851 8852 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8853 { 8854 size_t size = sizeof(sa->sa_data_min); 8855 struct net_device *dev; 8856 int ret = 0; 8857 8858 down_read(&dev_addr_sem); 8859 rcu_read_lock(); 8860 8861 dev = dev_get_by_name_rcu(net, dev_name); 8862 if (!dev) { 8863 ret = -ENODEV; 8864 goto unlock; 8865 } 8866 if (!dev->addr_len) 8867 memset(sa->sa_data, 0, size); 8868 else 8869 memcpy(sa->sa_data, dev->dev_addr, 8870 min_t(size_t, size, dev->addr_len)); 8871 sa->sa_family = dev->type; 8872 8873 unlock: 8874 rcu_read_unlock(); 8875 up_read(&dev_addr_sem); 8876 return ret; 8877 } 8878 EXPORT_SYMBOL(dev_get_mac_address); 8879 8880 /** 8881 * dev_change_carrier - Change device carrier 8882 * @dev: device 8883 * @new_carrier: new value 8884 * 8885 * Change device carrier 8886 */ 8887 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8888 { 8889 const struct net_device_ops *ops = dev->netdev_ops; 8890 8891 if (!ops->ndo_change_carrier) 8892 return -EOPNOTSUPP; 8893 if (!netif_device_present(dev)) 8894 return -ENODEV; 8895 return ops->ndo_change_carrier(dev, new_carrier); 8896 } 8897 8898 /** 8899 * dev_get_phys_port_id - Get device physical port ID 8900 * @dev: device 8901 * @ppid: port ID 8902 * 8903 * Get device physical port ID 8904 */ 8905 int dev_get_phys_port_id(struct net_device *dev, 8906 struct netdev_phys_item_id *ppid) 8907 { 8908 const struct net_device_ops *ops = dev->netdev_ops; 8909 8910 if (!ops->ndo_get_phys_port_id) 8911 return -EOPNOTSUPP; 8912 return ops->ndo_get_phys_port_id(dev, ppid); 8913 } 8914 8915 /** 8916 * dev_get_phys_port_name - Get device physical port name 8917 * @dev: device 8918 * @name: port name 8919 * @len: limit of bytes to copy to name 8920 * 8921 * Get device physical port name 8922 */ 8923 int dev_get_phys_port_name(struct net_device *dev, 8924 char *name, size_t len) 8925 { 8926 const struct net_device_ops *ops = dev->netdev_ops; 8927 int err; 8928 8929 if (ops->ndo_get_phys_port_name) { 8930 err = ops->ndo_get_phys_port_name(dev, name, len); 8931 if (err != -EOPNOTSUPP) 8932 return err; 8933 } 8934 return devlink_compat_phys_port_name_get(dev, name, len); 8935 } 8936 8937 /** 8938 * dev_get_port_parent_id - Get the device's port parent identifier 8939 * @dev: network device 8940 * @ppid: pointer to a storage for the port's parent identifier 8941 * @recurse: allow/disallow recursion to lower devices 8942 * 8943 * Get the devices's port parent identifier 8944 */ 8945 int dev_get_port_parent_id(struct net_device *dev, 8946 struct netdev_phys_item_id *ppid, 8947 bool recurse) 8948 { 8949 const struct net_device_ops *ops = dev->netdev_ops; 8950 struct netdev_phys_item_id first = { }; 8951 struct net_device *lower_dev; 8952 struct list_head *iter; 8953 int err; 8954 8955 if (ops->ndo_get_port_parent_id) { 8956 err = ops->ndo_get_port_parent_id(dev, ppid); 8957 if (err != -EOPNOTSUPP) 8958 return err; 8959 } 8960 8961 err = devlink_compat_switch_id_get(dev, ppid); 8962 if (!recurse || err != -EOPNOTSUPP) 8963 return err; 8964 8965 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8966 err = dev_get_port_parent_id(lower_dev, ppid, true); 8967 if (err) 8968 break; 8969 if (!first.id_len) 8970 first = *ppid; 8971 else if (memcmp(&first, ppid, sizeof(*ppid))) 8972 return -EOPNOTSUPP; 8973 } 8974 8975 return err; 8976 } 8977 EXPORT_SYMBOL(dev_get_port_parent_id); 8978 8979 /** 8980 * netdev_port_same_parent_id - Indicate if two network devices have 8981 * the same port parent identifier 8982 * @a: first network device 8983 * @b: second network device 8984 */ 8985 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8986 { 8987 struct netdev_phys_item_id a_id = { }; 8988 struct netdev_phys_item_id b_id = { }; 8989 8990 if (dev_get_port_parent_id(a, &a_id, true) || 8991 dev_get_port_parent_id(b, &b_id, true)) 8992 return false; 8993 8994 return netdev_phys_item_id_same(&a_id, &b_id); 8995 } 8996 EXPORT_SYMBOL(netdev_port_same_parent_id); 8997 8998 /** 8999 * dev_change_proto_down - set carrier according to proto_down. 9000 * 9001 * @dev: device 9002 * @proto_down: new value 9003 */ 9004 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9005 { 9006 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9007 return -EOPNOTSUPP; 9008 if (!netif_device_present(dev)) 9009 return -ENODEV; 9010 if (proto_down) 9011 netif_carrier_off(dev); 9012 else 9013 netif_carrier_on(dev); 9014 dev->proto_down = proto_down; 9015 return 0; 9016 } 9017 9018 /** 9019 * dev_change_proto_down_reason - proto down reason 9020 * 9021 * @dev: device 9022 * @mask: proto down mask 9023 * @value: proto down value 9024 */ 9025 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9026 u32 value) 9027 { 9028 int b; 9029 9030 if (!mask) { 9031 dev->proto_down_reason = value; 9032 } else { 9033 for_each_set_bit(b, &mask, 32) { 9034 if (value & (1 << b)) 9035 dev->proto_down_reason |= BIT(b); 9036 else 9037 dev->proto_down_reason &= ~BIT(b); 9038 } 9039 } 9040 } 9041 9042 struct bpf_xdp_link { 9043 struct bpf_link link; 9044 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9045 int flags; 9046 }; 9047 9048 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9049 { 9050 if (flags & XDP_FLAGS_HW_MODE) 9051 return XDP_MODE_HW; 9052 if (flags & XDP_FLAGS_DRV_MODE) 9053 return XDP_MODE_DRV; 9054 if (flags & XDP_FLAGS_SKB_MODE) 9055 return XDP_MODE_SKB; 9056 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9057 } 9058 9059 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9060 { 9061 switch (mode) { 9062 case XDP_MODE_SKB: 9063 return generic_xdp_install; 9064 case XDP_MODE_DRV: 9065 case XDP_MODE_HW: 9066 return dev->netdev_ops->ndo_bpf; 9067 default: 9068 return NULL; 9069 } 9070 } 9071 9072 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9073 enum bpf_xdp_mode mode) 9074 { 9075 return dev->xdp_state[mode].link; 9076 } 9077 9078 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9079 enum bpf_xdp_mode mode) 9080 { 9081 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9082 9083 if (link) 9084 return link->link.prog; 9085 return dev->xdp_state[mode].prog; 9086 } 9087 9088 u8 dev_xdp_prog_count(struct net_device *dev) 9089 { 9090 u8 count = 0; 9091 int i; 9092 9093 for (i = 0; i < __MAX_XDP_MODE; i++) 9094 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9095 count++; 9096 return count; 9097 } 9098 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9099 9100 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9101 { 9102 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9103 9104 return prog ? prog->aux->id : 0; 9105 } 9106 9107 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9108 struct bpf_xdp_link *link) 9109 { 9110 dev->xdp_state[mode].link = link; 9111 dev->xdp_state[mode].prog = NULL; 9112 } 9113 9114 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9115 struct bpf_prog *prog) 9116 { 9117 dev->xdp_state[mode].link = NULL; 9118 dev->xdp_state[mode].prog = prog; 9119 } 9120 9121 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9122 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9123 u32 flags, struct bpf_prog *prog) 9124 { 9125 struct netdev_bpf xdp; 9126 int err; 9127 9128 memset(&xdp, 0, sizeof(xdp)); 9129 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9130 xdp.extack = extack; 9131 xdp.flags = flags; 9132 xdp.prog = prog; 9133 9134 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9135 * "moved" into driver), so they don't increment it on their own, but 9136 * they do decrement refcnt when program is detached or replaced. 9137 * Given net_device also owns link/prog, we need to bump refcnt here 9138 * to prevent drivers from underflowing it. 9139 */ 9140 if (prog) 9141 bpf_prog_inc(prog); 9142 err = bpf_op(dev, &xdp); 9143 if (err) { 9144 if (prog) 9145 bpf_prog_put(prog); 9146 return err; 9147 } 9148 9149 if (mode != XDP_MODE_HW) 9150 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9151 9152 return 0; 9153 } 9154 9155 static void dev_xdp_uninstall(struct net_device *dev) 9156 { 9157 struct bpf_xdp_link *link; 9158 struct bpf_prog *prog; 9159 enum bpf_xdp_mode mode; 9160 bpf_op_t bpf_op; 9161 9162 ASSERT_RTNL(); 9163 9164 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9165 prog = dev_xdp_prog(dev, mode); 9166 if (!prog) 9167 continue; 9168 9169 bpf_op = dev_xdp_bpf_op(dev, mode); 9170 if (!bpf_op) 9171 continue; 9172 9173 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9174 9175 /* auto-detach link from net device */ 9176 link = dev_xdp_link(dev, mode); 9177 if (link) 9178 link->dev = NULL; 9179 else 9180 bpf_prog_put(prog); 9181 9182 dev_xdp_set_link(dev, mode, NULL); 9183 } 9184 } 9185 9186 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9187 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9188 struct bpf_prog *old_prog, u32 flags) 9189 { 9190 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9191 struct bpf_prog *cur_prog; 9192 struct net_device *upper; 9193 struct list_head *iter; 9194 enum bpf_xdp_mode mode; 9195 bpf_op_t bpf_op; 9196 int err; 9197 9198 ASSERT_RTNL(); 9199 9200 /* either link or prog attachment, never both */ 9201 if (link && (new_prog || old_prog)) 9202 return -EINVAL; 9203 /* link supports only XDP mode flags */ 9204 if (link && (flags & ~XDP_FLAGS_MODES)) { 9205 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9206 return -EINVAL; 9207 } 9208 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9209 if (num_modes > 1) { 9210 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9211 return -EINVAL; 9212 } 9213 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9214 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9215 NL_SET_ERR_MSG(extack, 9216 "More than one program loaded, unset mode is ambiguous"); 9217 return -EINVAL; 9218 } 9219 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9220 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9221 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9222 return -EINVAL; 9223 } 9224 9225 mode = dev_xdp_mode(dev, flags); 9226 /* can't replace attached link */ 9227 if (dev_xdp_link(dev, mode)) { 9228 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9229 return -EBUSY; 9230 } 9231 9232 /* don't allow if an upper device already has a program */ 9233 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9234 if (dev_xdp_prog_count(upper) > 0) { 9235 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9236 return -EEXIST; 9237 } 9238 } 9239 9240 cur_prog = dev_xdp_prog(dev, mode); 9241 /* can't replace attached prog with link */ 9242 if (link && cur_prog) { 9243 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9244 return -EBUSY; 9245 } 9246 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9247 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9248 return -EEXIST; 9249 } 9250 9251 /* put effective new program into new_prog */ 9252 if (link) 9253 new_prog = link->link.prog; 9254 9255 if (new_prog) { 9256 bool offload = mode == XDP_MODE_HW; 9257 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9258 ? XDP_MODE_DRV : XDP_MODE_SKB; 9259 9260 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9261 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9262 return -EBUSY; 9263 } 9264 if (!offload && dev_xdp_prog(dev, other_mode)) { 9265 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9266 return -EEXIST; 9267 } 9268 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9269 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9270 return -EINVAL; 9271 } 9272 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9273 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9274 return -EINVAL; 9275 } 9276 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9277 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9278 return -EINVAL; 9279 } 9280 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9281 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9282 return -EINVAL; 9283 } 9284 } 9285 9286 /* don't call drivers if the effective program didn't change */ 9287 if (new_prog != cur_prog) { 9288 bpf_op = dev_xdp_bpf_op(dev, mode); 9289 if (!bpf_op) { 9290 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9291 return -EOPNOTSUPP; 9292 } 9293 9294 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9295 if (err) 9296 return err; 9297 } 9298 9299 if (link) 9300 dev_xdp_set_link(dev, mode, link); 9301 else 9302 dev_xdp_set_prog(dev, mode, new_prog); 9303 if (cur_prog) 9304 bpf_prog_put(cur_prog); 9305 9306 return 0; 9307 } 9308 9309 static int dev_xdp_attach_link(struct net_device *dev, 9310 struct netlink_ext_ack *extack, 9311 struct bpf_xdp_link *link) 9312 { 9313 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9314 } 9315 9316 static int dev_xdp_detach_link(struct net_device *dev, 9317 struct netlink_ext_ack *extack, 9318 struct bpf_xdp_link *link) 9319 { 9320 enum bpf_xdp_mode mode; 9321 bpf_op_t bpf_op; 9322 9323 ASSERT_RTNL(); 9324 9325 mode = dev_xdp_mode(dev, link->flags); 9326 if (dev_xdp_link(dev, mode) != link) 9327 return -EINVAL; 9328 9329 bpf_op = dev_xdp_bpf_op(dev, mode); 9330 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9331 dev_xdp_set_link(dev, mode, NULL); 9332 return 0; 9333 } 9334 9335 static void bpf_xdp_link_release(struct bpf_link *link) 9336 { 9337 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9338 9339 rtnl_lock(); 9340 9341 /* if racing with net_device's tear down, xdp_link->dev might be 9342 * already NULL, in which case link was already auto-detached 9343 */ 9344 if (xdp_link->dev) { 9345 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9346 xdp_link->dev = NULL; 9347 } 9348 9349 rtnl_unlock(); 9350 } 9351 9352 static int bpf_xdp_link_detach(struct bpf_link *link) 9353 { 9354 bpf_xdp_link_release(link); 9355 return 0; 9356 } 9357 9358 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9359 { 9360 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9361 9362 kfree(xdp_link); 9363 } 9364 9365 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9366 struct seq_file *seq) 9367 { 9368 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9369 u32 ifindex = 0; 9370 9371 rtnl_lock(); 9372 if (xdp_link->dev) 9373 ifindex = xdp_link->dev->ifindex; 9374 rtnl_unlock(); 9375 9376 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9377 } 9378 9379 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9380 struct bpf_link_info *info) 9381 { 9382 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9383 u32 ifindex = 0; 9384 9385 rtnl_lock(); 9386 if (xdp_link->dev) 9387 ifindex = xdp_link->dev->ifindex; 9388 rtnl_unlock(); 9389 9390 info->xdp.ifindex = ifindex; 9391 return 0; 9392 } 9393 9394 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9395 struct bpf_prog *old_prog) 9396 { 9397 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9398 enum bpf_xdp_mode mode; 9399 bpf_op_t bpf_op; 9400 int err = 0; 9401 9402 rtnl_lock(); 9403 9404 /* link might have been auto-released already, so fail */ 9405 if (!xdp_link->dev) { 9406 err = -ENOLINK; 9407 goto out_unlock; 9408 } 9409 9410 if (old_prog && link->prog != old_prog) { 9411 err = -EPERM; 9412 goto out_unlock; 9413 } 9414 old_prog = link->prog; 9415 if (old_prog->type != new_prog->type || 9416 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9417 err = -EINVAL; 9418 goto out_unlock; 9419 } 9420 9421 if (old_prog == new_prog) { 9422 /* no-op, don't disturb drivers */ 9423 bpf_prog_put(new_prog); 9424 goto out_unlock; 9425 } 9426 9427 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9428 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9429 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9430 xdp_link->flags, new_prog); 9431 if (err) 9432 goto out_unlock; 9433 9434 old_prog = xchg(&link->prog, new_prog); 9435 bpf_prog_put(old_prog); 9436 9437 out_unlock: 9438 rtnl_unlock(); 9439 return err; 9440 } 9441 9442 static const struct bpf_link_ops bpf_xdp_link_lops = { 9443 .release = bpf_xdp_link_release, 9444 .dealloc = bpf_xdp_link_dealloc, 9445 .detach = bpf_xdp_link_detach, 9446 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9447 .fill_link_info = bpf_xdp_link_fill_link_info, 9448 .update_prog = bpf_xdp_link_update, 9449 }; 9450 9451 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9452 { 9453 struct net *net = current->nsproxy->net_ns; 9454 struct bpf_link_primer link_primer; 9455 struct bpf_xdp_link *link; 9456 struct net_device *dev; 9457 int err, fd; 9458 9459 rtnl_lock(); 9460 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9461 if (!dev) { 9462 rtnl_unlock(); 9463 return -EINVAL; 9464 } 9465 9466 link = kzalloc(sizeof(*link), GFP_USER); 9467 if (!link) { 9468 err = -ENOMEM; 9469 goto unlock; 9470 } 9471 9472 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9473 link->dev = dev; 9474 link->flags = attr->link_create.flags; 9475 9476 err = bpf_link_prime(&link->link, &link_primer); 9477 if (err) { 9478 kfree(link); 9479 goto unlock; 9480 } 9481 9482 err = dev_xdp_attach_link(dev, NULL, link); 9483 rtnl_unlock(); 9484 9485 if (err) { 9486 link->dev = NULL; 9487 bpf_link_cleanup(&link_primer); 9488 goto out_put_dev; 9489 } 9490 9491 fd = bpf_link_settle(&link_primer); 9492 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9493 dev_put(dev); 9494 return fd; 9495 9496 unlock: 9497 rtnl_unlock(); 9498 9499 out_put_dev: 9500 dev_put(dev); 9501 return err; 9502 } 9503 9504 /** 9505 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9506 * @dev: device 9507 * @extack: netlink extended ack 9508 * @fd: new program fd or negative value to clear 9509 * @expected_fd: old program fd that userspace expects to replace or clear 9510 * @flags: xdp-related flags 9511 * 9512 * Set or clear a bpf program for a device 9513 */ 9514 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9515 int fd, int expected_fd, u32 flags) 9516 { 9517 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9518 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9519 int err; 9520 9521 ASSERT_RTNL(); 9522 9523 if (fd >= 0) { 9524 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9525 mode != XDP_MODE_SKB); 9526 if (IS_ERR(new_prog)) 9527 return PTR_ERR(new_prog); 9528 } 9529 9530 if (expected_fd >= 0) { 9531 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9532 mode != XDP_MODE_SKB); 9533 if (IS_ERR(old_prog)) { 9534 err = PTR_ERR(old_prog); 9535 old_prog = NULL; 9536 goto err_out; 9537 } 9538 } 9539 9540 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9541 9542 err_out: 9543 if (err && new_prog) 9544 bpf_prog_put(new_prog); 9545 if (old_prog) 9546 bpf_prog_put(old_prog); 9547 return err; 9548 } 9549 9550 /** 9551 * dev_new_index - allocate an ifindex 9552 * @net: the applicable net namespace 9553 * 9554 * Returns a suitable unique value for a new device interface 9555 * number. The caller must hold the rtnl semaphore or the 9556 * dev_base_lock to be sure it remains unique. 9557 */ 9558 static int dev_new_index(struct net *net) 9559 { 9560 int ifindex = net->ifindex; 9561 9562 for (;;) { 9563 if (++ifindex <= 0) 9564 ifindex = 1; 9565 if (!__dev_get_by_index(net, ifindex)) 9566 return net->ifindex = ifindex; 9567 } 9568 } 9569 9570 /* Delayed registration/unregisteration */ 9571 LIST_HEAD(net_todo_list); 9572 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9573 9574 static void net_set_todo(struct net_device *dev) 9575 { 9576 list_add_tail(&dev->todo_list, &net_todo_list); 9577 atomic_inc(&dev_net(dev)->dev_unreg_count); 9578 } 9579 9580 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9581 struct net_device *upper, netdev_features_t features) 9582 { 9583 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9584 netdev_features_t feature; 9585 int feature_bit; 9586 9587 for_each_netdev_feature(upper_disables, feature_bit) { 9588 feature = __NETIF_F_BIT(feature_bit); 9589 if (!(upper->wanted_features & feature) 9590 && (features & feature)) { 9591 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9592 &feature, upper->name); 9593 features &= ~feature; 9594 } 9595 } 9596 9597 return features; 9598 } 9599 9600 static void netdev_sync_lower_features(struct net_device *upper, 9601 struct net_device *lower, netdev_features_t features) 9602 { 9603 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9604 netdev_features_t feature; 9605 int feature_bit; 9606 9607 for_each_netdev_feature(upper_disables, feature_bit) { 9608 feature = __NETIF_F_BIT(feature_bit); 9609 if (!(features & feature) && (lower->features & feature)) { 9610 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9611 &feature, lower->name); 9612 lower->wanted_features &= ~feature; 9613 __netdev_update_features(lower); 9614 9615 if (unlikely(lower->features & feature)) 9616 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9617 &feature, lower->name); 9618 else 9619 netdev_features_change(lower); 9620 } 9621 } 9622 } 9623 9624 static netdev_features_t netdev_fix_features(struct net_device *dev, 9625 netdev_features_t features) 9626 { 9627 /* Fix illegal checksum combinations */ 9628 if ((features & NETIF_F_HW_CSUM) && 9629 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9630 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9631 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9632 } 9633 9634 /* TSO requires that SG is present as well. */ 9635 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9636 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9637 features &= ~NETIF_F_ALL_TSO; 9638 } 9639 9640 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9641 !(features & NETIF_F_IP_CSUM)) { 9642 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9643 features &= ~NETIF_F_TSO; 9644 features &= ~NETIF_F_TSO_ECN; 9645 } 9646 9647 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9648 !(features & NETIF_F_IPV6_CSUM)) { 9649 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9650 features &= ~NETIF_F_TSO6; 9651 } 9652 9653 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9654 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9655 features &= ~NETIF_F_TSO_MANGLEID; 9656 9657 /* TSO ECN requires that TSO is present as well. */ 9658 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9659 features &= ~NETIF_F_TSO_ECN; 9660 9661 /* Software GSO depends on SG. */ 9662 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9663 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9664 features &= ~NETIF_F_GSO; 9665 } 9666 9667 /* GSO partial features require GSO partial be set */ 9668 if ((features & dev->gso_partial_features) && 9669 !(features & NETIF_F_GSO_PARTIAL)) { 9670 netdev_dbg(dev, 9671 "Dropping partially supported GSO features since no GSO partial.\n"); 9672 features &= ~dev->gso_partial_features; 9673 } 9674 9675 if (!(features & NETIF_F_RXCSUM)) { 9676 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9677 * successfully merged by hardware must also have the 9678 * checksum verified by hardware. If the user does not 9679 * want to enable RXCSUM, logically, we should disable GRO_HW. 9680 */ 9681 if (features & NETIF_F_GRO_HW) { 9682 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9683 features &= ~NETIF_F_GRO_HW; 9684 } 9685 } 9686 9687 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9688 if (features & NETIF_F_RXFCS) { 9689 if (features & NETIF_F_LRO) { 9690 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9691 features &= ~NETIF_F_LRO; 9692 } 9693 9694 if (features & NETIF_F_GRO_HW) { 9695 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9696 features &= ~NETIF_F_GRO_HW; 9697 } 9698 } 9699 9700 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9701 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9702 features &= ~NETIF_F_LRO; 9703 } 9704 9705 if (features & NETIF_F_HW_TLS_TX) { 9706 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9707 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9708 bool hw_csum = features & NETIF_F_HW_CSUM; 9709 9710 if (!ip_csum && !hw_csum) { 9711 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9712 features &= ~NETIF_F_HW_TLS_TX; 9713 } 9714 } 9715 9716 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9717 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9718 features &= ~NETIF_F_HW_TLS_RX; 9719 } 9720 9721 return features; 9722 } 9723 9724 int __netdev_update_features(struct net_device *dev) 9725 { 9726 struct net_device *upper, *lower; 9727 netdev_features_t features; 9728 struct list_head *iter; 9729 int err = -1; 9730 9731 ASSERT_RTNL(); 9732 9733 features = netdev_get_wanted_features(dev); 9734 9735 if (dev->netdev_ops->ndo_fix_features) 9736 features = dev->netdev_ops->ndo_fix_features(dev, features); 9737 9738 /* driver might be less strict about feature dependencies */ 9739 features = netdev_fix_features(dev, features); 9740 9741 /* some features can't be enabled if they're off on an upper device */ 9742 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9743 features = netdev_sync_upper_features(dev, upper, features); 9744 9745 if (dev->features == features) 9746 goto sync_lower; 9747 9748 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9749 &dev->features, &features); 9750 9751 if (dev->netdev_ops->ndo_set_features) 9752 err = dev->netdev_ops->ndo_set_features(dev, features); 9753 else 9754 err = 0; 9755 9756 if (unlikely(err < 0)) { 9757 netdev_err(dev, 9758 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9759 err, &features, &dev->features); 9760 /* return non-0 since some features might have changed and 9761 * it's better to fire a spurious notification than miss it 9762 */ 9763 return -1; 9764 } 9765 9766 sync_lower: 9767 /* some features must be disabled on lower devices when disabled 9768 * on an upper device (think: bonding master or bridge) 9769 */ 9770 netdev_for_each_lower_dev(dev, lower, iter) 9771 netdev_sync_lower_features(dev, lower, features); 9772 9773 if (!err) { 9774 netdev_features_t diff = features ^ dev->features; 9775 9776 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9777 /* udp_tunnel_{get,drop}_rx_info both need 9778 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9779 * device, or they won't do anything. 9780 * Thus we need to update dev->features 9781 * *before* calling udp_tunnel_get_rx_info, 9782 * but *after* calling udp_tunnel_drop_rx_info. 9783 */ 9784 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9785 dev->features = features; 9786 udp_tunnel_get_rx_info(dev); 9787 } else { 9788 udp_tunnel_drop_rx_info(dev); 9789 } 9790 } 9791 9792 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9793 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9794 dev->features = features; 9795 err |= vlan_get_rx_ctag_filter_info(dev); 9796 } else { 9797 vlan_drop_rx_ctag_filter_info(dev); 9798 } 9799 } 9800 9801 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9802 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9803 dev->features = features; 9804 err |= vlan_get_rx_stag_filter_info(dev); 9805 } else { 9806 vlan_drop_rx_stag_filter_info(dev); 9807 } 9808 } 9809 9810 dev->features = features; 9811 } 9812 9813 return err < 0 ? 0 : 1; 9814 } 9815 9816 /** 9817 * netdev_update_features - recalculate device features 9818 * @dev: the device to check 9819 * 9820 * Recalculate dev->features set and send notifications if it 9821 * has changed. Should be called after driver or hardware dependent 9822 * conditions might have changed that influence the features. 9823 */ 9824 void netdev_update_features(struct net_device *dev) 9825 { 9826 if (__netdev_update_features(dev)) 9827 netdev_features_change(dev); 9828 } 9829 EXPORT_SYMBOL(netdev_update_features); 9830 9831 /** 9832 * netdev_change_features - recalculate device features 9833 * @dev: the device to check 9834 * 9835 * Recalculate dev->features set and send notifications even 9836 * if they have not changed. Should be called instead of 9837 * netdev_update_features() if also dev->vlan_features might 9838 * have changed to allow the changes to be propagated to stacked 9839 * VLAN devices. 9840 */ 9841 void netdev_change_features(struct net_device *dev) 9842 { 9843 __netdev_update_features(dev); 9844 netdev_features_change(dev); 9845 } 9846 EXPORT_SYMBOL(netdev_change_features); 9847 9848 /** 9849 * netif_stacked_transfer_operstate - transfer operstate 9850 * @rootdev: the root or lower level device to transfer state from 9851 * @dev: the device to transfer operstate to 9852 * 9853 * Transfer operational state from root to device. This is normally 9854 * called when a stacking relationship exists between the root 9855 * device and the device(a leaf device). 9856 */ 9857 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9858 struct net_device *dev) 9859 { 9860 if (rootdev->operstate == IF_OPER_DORMANT) 9861 netif_dormant_on(dev); 9862 else 9863 netif_dormant_off(dev); 9864 9865 if (rootdev->operstate == IF_OPER_TESTING) 9866 netif_testing_on(dev); 9867 else 9868 netif_testing_off(dev); 9869 9870 if (netif_carrier_ok(rootdev)) 9871 netif_carrier_on(dev); 9872 else 9873 netif_carrier_off(dev); 9874 } 9875 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9876 9877 static int netif_alloc_rx_queues(struct net_device *dev) 9878 { 9879 unsigned int i, count = dev->num_rx_queues; 9880 struct netdev_rx_queue *rx; 9881 size_t sz = count * sizeof(*rx); 9882 int err = 0; 9883 9884 BUG_ON(count < 1); 9885 9886 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9887 if (!rx) 9888 return -ENOMEM; 9889 9890 dev->_rx = rx; 9891 9892 for (i = 0; i < count; i++) { 9893 rx[i].dev = dev; 9894 9895 /* XDP RX-queue setup */ 9896 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9897 if (err < 0) 9898 goto err_rxq_info; 9899 } 9900 return 0; 9901 9902 err_rxq_info: 9903 /* Rollback successful reg's and free other resources */ 9904 while (i--) 9905 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9906 kvfree(dev->_rx); 9907 dev->_rx = NULL; 9908 return err; 9909 } 9910 9911 static void netif_free_rx_queues(struct net_device *dev) 9912 { 9913 unsigned int i, count = dev->num_rx_queues; 9914 9915 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9916 if (!dev->_rx) 9917 return; 9918 9919 for (i = 0; i < count; i++) 9920 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9921 9922 kvfree(dev->_rx); 9923 } 9924 9925 static void netdev_init_one_queue(struct net_device *dev, 9926 struct netdev_queue *queue, void *_unused) 9927 { 9928 /* Initialize queue lock */ 9929 spin_lock_init(&queue->_xmit_lock); 9930 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9931 queue->xmit_lock_owner = -1; 9932 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9933 queue->dev = dev; 9934 #ifdef CONFIG_BQL 9935 dql_init(&queue->dql, HZ); 9936 #endif 9937 } 9938 9939 static void netif_free_tx_queues(struct net_device *dev) 9940 { 9941 kvfree(dev->_tx); 9942 } 9943 9944 static int netif_alloc_netdev_queues(struct net_device *dev) 9945 { 9946 unsigned int count = dev->num_tx_queues; 9947 struct netdev_queue *tx; 9948 size_t sz = count * sizeof(*tx); 9949 9950 if (count < 1 || count > 0xffff) 9951 return -EINVAL; 9952 9953 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9954 if (!tx) 9955 return -ENOMEM; 9956 9957 dev->_tx = tx; 9958 9959 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9960 spin_lock_init(&dev->tx_global_lock); 9961 9962 return 0; 9963 } 9964 9965 void netif_tx_stop_all_queues(struct net_device *dev) 9966 { 9967 unsigned int i; 9968 9969 for (i = 0; i < dev->num_tx_queues; i++) { 9970 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9971 9972 netif_tx_stop_queue(txq); 9973 } 9974 } 9975 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9976 9977 /** 9978 * register_netdevice() - register a network device 9979 * @dev: device to register 9980 * 9981 * Take a prepared network device structure and make it externally accessible. 9982 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9983 * Callers must hold the rtnl lock - you may want register_netdev() 9984 * instead of this. 9985 */ 9986 int register_netdevice(struct net_device *dev) 9987 { 9988 int ret; 9989 struct net *net = dev_net(dev); 9990 9991 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9992 NETDEV_FEATURE_COUNT); 9993 BUG_ON(dev_boot_phase); 9994 ASSERT_RTNL(); 9995 9996 might_sleep(); 9997 9998 /* When net_device's are persistent, this will be fatal. */ 9999 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10000 BUG_ON(!net); 10001 10002 ret = ethtool_check_ops(dev->ethtool_ops); 10003 if (ret) 10004 return ret; 10005 10006 spin_lock_init(&dev->addr_list_lock); 10007 netdev_set_addr_lockdep_class(dev); 10008 10009 ret = dev_get_valid_name(net, dev, dev->name); 10010 if (ret < 0) 10011 goto out; 10012 10013 ret = -ENOMEM; 10014 dev->name_node = netdev_name_node_head_alloc(dev); 10015 if (!dev->name_node) 10016 goto out; 10017 10018 /* Init, if this function is available */ 10019 if (dev->netdev_ops->ndo_init) { 10020 ret = dev->netdev_ops->ndo_init(dev); 10021 if (ret) { 10022 if (ret > 0) 10023 ret = -EIO; 10024 goto err_free_name; 10025 } 10026 } 10027 10028 if (((dev->hw_features | dev->features) & 10029 NETIF_F_HW_VLAN_CTAG_FILTER) && 10030 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10031 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10032 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10033 ret = -EINVAL; 10034 goto err_uninit; 10035 } 10036 10037 ret = -EBUSY; 10038 if (!dev->ifindex) 10039 dev->ifindex = dev_new_index(net); 10040 else if (__dev_get_by_index(net, dev->ifindex)) 10041 goto err_uninit; 10042 10043 /* Transfer changeable features to wanted_features and enable 10044 * software offloads (GSO and GRO). 10045 */ 10046 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10047 dev->features |= NETIF_F_SOFT_FEATURES; 10048 10049 if (dev->udp_tunnel_nic_info) { 10050 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10051 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10052 } 10053 10054 dev->wanted_features = dev->features & dev->hw_features; 10055 10056 if (!(dev->flags & IFF_LOOPBACK)) 10057 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10058 10059 /* If IPv4 TCP segmentation offload is supported we should also 10060 * allow the device to enable segmenting the frame with the option 10061 * of ignoring a static IP ID value. This doesn't enable the 10062 * feature itself but allows the user to enable it later. 10063 */ 10064 if (dev->hw_features & NETIF_F_TSO) 10065 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10066 if (dev->vlan_features & NETIF_F_TSO) 10067 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10068 if (dev->mpls_features & NETIF_F_TSO) 10069 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10070 if (dev->hw_enc_features & NETIF_F_TSO) 10071 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10072 10073 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10074 */ 10075 dev->vlan_features |= NETIF_F_HIGHDMA; 10076 10077 /* Make NETIF_F_SG inheritable to tunnel devices. 10078 */ 10079 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10080 10081 /* Make NETIF_F_SG inheritable to MPLS. 10082 */ 10083 dev->mpls_features |= NETIF_F_SG; 10084 10085 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10086 ret = notifier_to_errno(ret); 10087 if (ret) 10088 goto err_uninit; 10089 10090 ret = netdev_register_kobject(dev); 10091 write_lock(&dev_base_lock); 10092 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10093 write_unlock(&dev_base_lock); 10094 if (ret) 10095 goto err_uninit_notify; 10096 10097 __netdev_update_features(dev); 10098 10099 /* 10100 * Default initial state at registry is that the 10101 * device is present. 10102 */ 10103 10104 set_bit(__LINK_STATE_PRESENT, &dev->state); 10105 10106 linkwatch_init_dev(dev); 10107 10108 dev_init_scheduler(dev); 10109 10110 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10111 list_netdevice(dev); 10112 10113 add_device_randomness(dev->dev_addr, dev->addr_len); 10114 10115 /* If the device has permanent device address, driver should 10116 * set dev_addr and also addr_assign_type should be set to 10117 * NET_ADDR_PERM (default value). 10118 */ 10119 if (dev->addr_assign_type == NET_ADDR_PERM) 10120 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10121 10122 /* Notify protocols, that a new device appeared. */ 10123 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10124 ret = notifier_to_errno(ret); 10125 if (ret) { 10126 /* Expect explicit free_netdev() on failure */ 10127 dev->needs_free_netdev = false; 10128 unregister_netdevice_queue(dev, NULL); 10129 goto out; 10130 } 10131 /* 10132 * Prevent userspace races by waiting until the network 10133 * device is fully setup before sending notifications. 10134 */ 10135 if (!dev->rtnl_link_ops || 10136 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10137 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10138 10139 out: 10140 return ret; 10141 10142 err_uninit_notify: 10143 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10144 err_uninit: 10145 if (dev->netdev_ops->ndo_uninit) 10146 dev->netdev_ops->ndo_uninit(dev); 10147 if (dev->priv_destructor) 10148 dev->priv_destructor(dev); 10149 err_free_name: 10150 netdev_name_node_free(dev->name_node); 10151 goto out; 10152 } 10153 EXPORT_SYMBOL(register_netdevice); 10154 10155 /** 10156 * init_dummy_netdev - init a dummy network device for NAPI 10157 * @dev: device to init 10158 * 10159 * This takes a network device structure and initialize the minimum 10160 * amount of fields so it can be used to schedule NAPI polls without 10161 * registering a full blown interface. This is to be used by drivers 10162 * that need to tie several hardware interfaces to a single NAPI 10163 * poll scheduler due to HW limitations. 10164 */ 10165 int init_dummy_netdev(struct net_device *dev) 10166 { 10167 /* Clear everything. Note we don't initialize spinlocks 10168 * are they aren't supposed to be taken by any of the 10169 * NAPI code and this dummy netdev is supposed to be 10170 * only ever used for NAPI polls 10171 */ 10172 memset(dev, 0, sizeof(struct net_device)); 10173 10174 /* make sure we BUG if trying to hit standard 10175 * register/unregister code path 10176 */ 10177 dev->reg_state = NETREG_DUMMY; 10178 10179 /* NAPI wants this */ 10180 INIT_LIST_HEAD(&dev->napi_list); 10181 10182 /* a dummy interface is started by default */ 10183 set_bit(__LINK_STATE_PRESENT, &dev->state); 10184 set_bit(__LINK_STATE_START, &dev->state); 10185 10186 /* napi_busy_loop stats accounting wants this */ 10187 dev_net_set(dev, &init_net); 10188 10189 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10190 * because users of this 'device' dont need to change 10191 * its refcount. 10192 */ 10193 10194 return 0; 10195 } 10196 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10197 10198 10199 /** 10200 * register_netdev - register a network device 10201 * @dev: device to register 10202 * 10203 * Take a completed network device structure and add it to the kernel 10204 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10205 * chain. 0 is returned on success. A negative errno code is returned 10206 * on a failure to set up the device, or if the name is a duplicate. 10207 * 10208 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10209 * and expands the device name if you passed a format string to 10210 * alloc_netdev. 10211 */ 10212 int register_netdev(struct net_device *dev) 10213 { 10214 int err; 10215 10216 if (rtnl_lock_killable()) 10217 return -EINTR; 10218 err = register_netdevice(dev); 10219 rtnl_unlock(); 10220 return err; 10221 } 10222 EXPORT_SYMBOL(register_netdev); 10223 10224 int netdev_refcnt_read(const struct net_device *dev) 10225 { 10226 #ifdef CONFIG_PCPU_DEV_REFCNT 10227 int i, refcnt = 0; 10228 10229 for_each_possible_cpu(i) 10230 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10231 return refcnt; 10232 #else 10233 return refcount_read(&dev->dev_refcnt); 10234 #endif 10235 } 10236 EXPORT_SYMBOL(netdev_refcnt_read); 10237 10238 int netdev_unregister_timeout_secs __read_mostly = 10; 10239 10240 #define WAIT_REFS_MIN_MSECS 1 10241 #define WAIT_REFS_MAX_MSECS 250 10242 /** 10243 * netdev_wait_allrefs_any - wait until all references are gone. 10244 * @list: list of net_devices to wait on 10245 * 10246 * This is called when unregistering network devices. 10247 * 10248 * Any protocol or device that holds a reference should register 10249 * for netdevice notification, and cleanup and put back the 10250 * reference if they receive an UNREGISTER event. 10251 * We can get stuck here if buggy protocols don't correctly 10252 * call dev_put. 10253 */ 10254 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10255 { 10256 unsigned long rebroadcast_time, warning_time; 10257 struct net_device *dev; 10258 int wait = 0; 10259 10260 rebroadcast_time = warning_time = jiffies; 10261 10262 list_for_each_entry(dev, list, todo_list) 10263 if (netdev_refcnt_read(dev) == 1) 10264 return dev; 10265 10266 while (true) { 10267 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10268 rtnl_lock(); 10269 10270 /* Rebroadcast unregister notification */ 10271 list_for_each_entry(dev, list, todo_list) 10272 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10273 10274 __rtnl_unlock(); 10275 rcu_barrier(); 10276 rtnl_lock(); 10277 10278 list_for_each_entry(dev, list, todo_list) 10279 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10280 &dev->state)) { 10281 /* We must not have linkwatch events 10282 * pending on unregister. If this 10283 * happens, we simply run the queue 10284 * unscheduled, resulting in a noop 10285 * for this device. 10286 */ 10287 linkwatch_run_queue(); 10288 break; 10289 } 10290 10291 __rtnl_unlock(); 10292 10293 rebroadcast_time = jiffies; 10294 } 10295 10296 if (!wait) { 10297 rcu_barrier(); 10298 wait = WAIT_REFS_MIN_MSECS; 10299 } else { 10300 msleep(wait); 10301 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10302 } 10303 10304 list_for_each_entry(dev, list, todo_list) 10305 if (netdev_refcnt_read(dev) == 1) 10306 return dev; 10307 10308 if (time_after(jiffies, warning_time + 10309 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10310 list_for_each_entry(dev, list, todo_list) { 10311 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10312 dev->name, netdev_refcnt_read(dev)); 10313 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10314 } 10315 10316 warning_time = jiffies; 10317 } 10318 } 10319 } 10320 10321 /* The sequence is: 10322 * 10323 * rtnl_lock(); 10324 * ... 10325 * register_netdevice(x1); 10326 * register_netdevice(x2); 10327 * ... 10328 * unregister_netdevice(y1); 10329 * unregister_netdevice(y2); 10330 * ... 10331 * rtnl_unlock(); 10332 * free_netdev(y1); 10333 * free_netdev(y2); 10334 * 10335 * We are invoked by rtnl_unlock(). 10336 * This allows us to deal with problems: 10337 * 1) We can delete sysfs objects which invoke hotplug 10338 * without deadlocking with linkwatch via keventd. 10339 * 2) Since we run with the RTNL semaphore not held, we can sleep 10340 * safely in order to wait for the netdev refcnt to drop to zero. 10341 * 10342 * We must not return until all unregister events added during 10343 * the interval the lock was held have been completed. 10344 */ 10345 void netdev_run_todo(void) 10346 { 10347 struct net_device *dev, *tmp; 10348 struct list_head list; 10349 #ifdef CONFIG_LOCKDEP 10350 struct list_head unlink_list; 10351 10352 list_replace_init(&net_unlink_list, &unlink_list); 10353 10354 while (!list_empty(&unlink_list)) { 10355 struct net_device *dev = list_first_entry(&unlink_list, 10356 struct net_device, 10357 unlink_list); 10358 list_del_init(&dev->unlink_list); 10359 dev->nested_level = dev->lower_level - 1; 10360 } 10361 #endif 10362 10363 /* Snapshot list, allow later requests */ 10364 list_replace_init(&net_todo_list, &list); 10365 10366 __rtnl_unlock(); 10367 10368 /* Wait for rcu callbacks to finish before next phase */ 10369 if (!list_empty(&list)) 10370 rcu_barrier(); 10371 10372 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10373 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10374 netdev_WARN(dev, "run_todo but not unregistering\n"); 10375 list_del(&dev->todo_list); 10376 continue; 10377 } 10378 10379 write_lock(&dev_base_lock); 10380 dev->reg_state = NETREG_UNREGISTERED; 10381 write_unlock(&dev_base_lock); 10382 linkwatch_forget_dev(dev); 10383 } 10384 10385 while (!list_empty(&list)) { 10386 dev = netdev_wait_allrefs_any(&list); 10387 list_del(&dev->todo_list); 10388 10389 /* paranoia */ 10390 BUG_ON(netdev_refcnt_read(dev) != 1); 10391 BUG_ON(!list_empty(&dev->ptype_all)); 10392 BUG_ON(!list_empty(&dev->ptype_specific)); 10393 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10394 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10395 10396 if (dev->priv_destructor) 10397 dev->priv_destructor(dev); 10398 if (dev->needs_free_netdev) 10399 free_netdev(dev); 10400 10401 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10402 wake_up(&netdev_unregistering_wq); 10403 10404 /* Free network device */ 10405 kobject_put(&dev->dev.kobj); 10406 } 10407 } 10408 10409 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10410 * all the same fields in the same order as net_device_stats, with only 10411 * the type differing, but rtnl_link_stats64 may have additional fields 10412 * at the end for newer counters. 10413 */ 10414 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10415 const struct net_device_stats *netdev_stats) 10416 { 10417 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10418 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10419 u64 *dst = (u64 *)stats64; 10420 10421 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10422 for (i = 0; i < n; i++) 10423 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10424 /* zero out counters that only exist in rtnl_link_stats64 */ 10425 memset((char *)stats64 + n * sizeof(u64), 0, 10426 sizeof(*stats64) - n * sizeof(u64)); 10427 } 10428 EXPORT_SYMBOL(netdev_stats_to_stats64); 10429 10430 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10431 { 10432 struct net_device_core_stats __percpu *p; 10433 10434 p = alloc_percpu_gfp(struct net_device_core_stats, 10435 GFP_ATOMIC | __GFP_NOWARN); 10436 10437 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10438 free_percpu(p); 10439 10440 /* This READ_ONCE() pairs with the cmpxchg() above */ 10441 return READ_ONCE(dev->core_stats); 10442 } 10443 EXPORT_SYMBOL(netdev_core_stats_alloc); 10444 10445 /** 10446 * dev_get_stats - get network device statistics 10447 * @dev: device to get statistics from 10448 * @storage: place to store stats 10449 * 10450 * Get network statistics from device. Return @storage. 10451 * The device driver may provide its own method by setting 10452 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10453 * otherwise the internal statistics structure is used. 10454 */ 10455 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10456 struct rtnl_link_stats64 *storage) 10457 { 10458 const struct net_device_ops *ops = dev->netdev_ops; 10459 const struct net_device_core_stats __percpu *p; 10460 10461 if (ops->ndo_get_stats64) { 10462 memset(storage, 0, sizeof(*storage)); 10463 ops->ndo_get_stats64(dev, storage); 10464 } else if (ops->ndo_get_stats) { 10465 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10466 } else { 10467 netdev_stats_to_stats64(storage, &dev->stats); 10468 } 10469 10470 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10471 p = READ_ONCE(dev->core_stats); 10472 if (p) { 10473 const struct net_device_core_stats *core_stats; 10474 int i; 10475 10476 for_each_possible_cpu(i) { 10477 core_stats = per_cpu_ptr(p, i); 10478 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10479 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10480 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10481 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10482 } 10483 } 10484 return storage; 10485 } 10486 EXPORT_SYMBOL(dev_get_stats); 10487 10488 /** 10489 * dev_fetch_sw_netstats - get per-cpu network device statistics 10490 * @s: place to store stats 10491 * @netstats: per-cpu network stats to read from 10492 * 10493 * Read per-cpu network statistics and populate the related fields in @s. 10494 */ 10495 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10496 const struct pcpu_sw_netstats __percpu *netstats) 10497 { 10498 int cpu; 10499 10500 for_each_possible_cpu(cpu) { 10501 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10502 const struct pcpu_sw_netstats *stats; 10503 unsigned int start; 10504 10505 stats = per_cpu_ptr(netstats, cpu); 10506 do { 10507 start = u64_stats_fetch_begin(&stats->syncp); 10508 rx_packets = u64_stats_read(&stats->rx_packets); 10509 rx_bytes = u64_stats_read(&stats->rx_bytes); 10510 tx_packets = u64_stats_read(&stats->tx_packets); 10511 tx_bytes = u64_stats_read(&stats->tx_bytes); 10512 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10513 10514 s->rx_packets += rx_packets; 10515 s->rx_bytes += rx_bytes; 10516 s->tx_packets += tx_packets; 10517 s->tx_bytes += tx_bytes; 10518 } 10519 } 10520 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10521 10522 /** 10523 * dev_get_tstats64 - ndo_get_stats64 implementation 10524 * @dev: device to get statistics from 10525 * @s: place to store stats 10526 * 10527 * Populate @s from dev->stats and dev->tstats. Can be used as 10528 * ndo_get_stats64() callback. 10529 */ 10530 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10531 { 10532 netdev_stats_to_stats64(s, &dev->stats); 10533 dev_fetch_sw_netstats(s, dev->tstats); 10534 } 10535 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10536 10537 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10538 { 10539 struct netdev_queue *queue = dev_ingress_queue(dev); 10540 10541 #ifdef CONFIG_NET_CLS_ACT 10542 if (queue) 10543 return queue; 10544 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10545 if (!queue) 10546 return NULL; 10547 netdev_init_one_queue(dev, queue, NULL); 10548 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10549 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10550 rcu_assign_pointer(dev->ingress_queue, queue); 10551 #endif 10552 return queue; 10553 } 10554 10555 static const struct ethtool_ops default_ethtool_ops; 10556 10557 void netdev_set_default_ethtool_ops(struct net_device *dev, 10558 const struct ethtool_ops *ops) 10559 { 10560 if (dev->ethtool_ops == &default_ethtool_ops) 10561 dev->ethtool_ops = ops; 10562 } 10563 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10564 10565 /** 10566 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10567 * @dev: netdev to enable the IRQ coalescing on 10568 * 10569 * Sets a conservative default for SW IRQ coalescing. Users can use 10570 * sysfs attributes to override the default values. 10571 */ 10572 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10573 { 10574 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10575 10576 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10577 dev->gro_flush_timeout = 20000; 10578 dev->napi_defer_hard_irqs = 1; 10579 } 10580 } 10581 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10582 10583 void netdev_freemem(struct net_device *dev) 10584 { 10585 char *addr = (char *)dev - dev->padded; 10586 10587 kvfree(addr); 10588 } 10589 10590 /** 10591 * alloc_netdev_mqs - allocate network device 10592 * @sizeof_priv: size of private data to allocate space for 10593 * @name: device name format string 10594 * @name_assign_type: origin of device name 10595 * @setup: callback to initialize device 10596 * @txqs: the number of TX subqueues to allocate 10597 * @rxqs: the number of RX subqueues to allocate 10598 * 10599 * Allocates a struct net_device with private data area for driver use 10600 * and performs basic initialization. Also allocates subqueue structs 10601 * for each queue on the device. 10602 */ 10603 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10604 unsigned char name_assign_type, 10605 void (*setup)(struct net_device *), 10606 unsigned int txqs, unsigned int rxqs) 10607 { 10608 struct net_device *dev; 10609 unsigned int alloc_size; 10610 struct net_device *p; 10611 10612 BUG_ON(strlen(name) >= sizeof(dev->name)); 10613 10614 if (txqs < 1) { 10615 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10616 return NULL; 10617 } 10618 10619 if (rxqs < 1) { 10620 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10621 return NULL; 10622 } 10623 10624 alloc_size = sizeof(struct net_device); 10625 if (sizeof_priv) { 10626 /* ensure 32-byte alignment of private area */ 10627 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10628 alloc_size += sizeof_priv; 10629 } 10630 /* ensure 32-byte alignment of whole construct */ 10631 alloc_size += NETDEV_ALIGN - 1; 10632 10633 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10634 if (!p) 10635 return NULL; 10636 10637 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10638 dev->padded = (char *)dev - (char *)p; 10639 10640 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10641 #ifdef CONFIG_PCPU_DEV_REFCNT 10642 dev->pcpu_refcnt = alloc_percpu(int); 10643 if (!dev->pcpu_refcnt) 10644 goto free_dev; 10645 __dev_hold(dev); 10646 #else 10647 refcount_set(&dev->dev_refcnt, 1); 10648 #endif 10649 10650 if (dev_addr_init(dev)) 10651 goto free_pcpu; 10652 10653 dev_mc_init(dev); 10654 dev_uc_init(dev); 10655 10656 dev_net_set(dev, &init_net); 10657 10658 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10659 dev->gso_max_segs = GSO_MAX_SEGS; 10660 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10661 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10662 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10663 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10664 dev->tso_max_segs = TSO_MAX_SEGS; 10665 dev->upper_level = 1; 10666 dev->lower_level = 1; 10667 #ifdef CONFIG_LOCKDEP 10668 dev->nested_level = 0; 10669 INIT_LIST_HEAD(&dev->unlink_list); 10670 #endif 10671 10672 INIT_LIST_HEAD(&dev->napi_list); 10673 INIT_LIST_HEAD(&dev->unreg_list); 10674 INIT_LIST_HEAD(&dev->close_list); 10675 INIT_LIST_HEAD(&dev->link_watch_list); 10676 INIT_LIST_HEAD(&dev->adj_list.upper); 10677 INIT_LIST_HEAD(&dev->adj_list.lower); 10678 INIT_LIST_HEAD(&dev->ptype_all); 10679 INIT_LIST_HEAD(&dev->ptype_specific); 10680 INIT_LIST_HEAD(&dev->net_notifier_list); 10681 #ifdef CONFIG_NET_SCHED 10682 hash_init(dev->qdisc_hash); 10683 #endif 10684 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10685 setup(dev); 10686 10687 if (!dev->tx_queue_len) { 10688 dev->priv_flags |= IFF_NO_QUEUE; 10689 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10690 } 10691 10692 dev->num_tx_queues = txqs; 10693 dev->real_num_tx_queues = txqs; 10694 if (netif_alloc_netdev_queues(dev)) 10695 goto free_all; 10696 10697 dev->num_rx_queues = rxqs; 10698 dev->real_num_rx_queues = rxqs; 10699 if (netif_alloc_rx_queues(dev)) 10700 goto free_all; 10701 10702 strcpy(dev->name, name); 10703 dev->name_assign_type = name_assign_type; 10704 dev->group = INIT_NETDEV_GROUP; 10705 if (!dev->ethtool_ops) 10706 dev->ethtool_ops = &default_ethtool_ops; 10707 10708 nf_hook_netdev_init(dev); 10709 10710 return dev; 10711 10712 free_all: 10713 free_netdev(dev); 10714 return NULL; 10715 10716 free_pcpu: 10717 #ifdef CONFIG_PCPU_DEV_REFCNT 10718 free_percpu(dev->pcpu_refcnt); 10719 free_dev: 10720 #endif 10721 netdev_freemem(dev); 10722 return NULL; 10723 } 10724 EXPORT_SYMBOL(alloc_netdev_mqs); 10725 10726 /** 10727 * free_netdev - free network device 10728 * @dev: device 10729 * 10730 * This function does the last stage of destroying an allocated device 10731 * interface. The reference to the device object is released. If this 10732 * is the last reference then it will be freed.Must be called in process 10733 * context. 10734 */ 10735 void free_netdev(struct net_device *dev) 10736 { 10737 struct napi_struct *p, *n; 10738 10739 might_sleep(); 10740 10741 /* When called immediately after register_netdevice() failed the unwind 10742 * handling may still be dismantling the device. Handle that case by 10743 * deferring the free. 10744 */ 10745 if (dev->reg_state == NETREG_UNREGISTERING) { 10746 ASSERT_RTNL(); 10747 dev->needs_free_netdev = true; 10748 return; 10749 } 10750 10751 netif_free_tx_queues(dev); 10752 netif_free_rx_queues(dev); 10753 10754 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10755 10756 /* Flush device addresses */ 10757 dev_addr_flush(dev); 10758 10759 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10760 netif_napi_del(p); 10761 10762 ref_tracker_dir_exit(&dev->refcnt_tracker); 10763 #ifdef CONFIG_PCPU_DEV_REFCNT 10764 free_percpu(dev->pcpu_refcnt); 10765 dev->pcpu_refcnt = NULL; 10766 #endif 10767 free_percpu(dev->core_stats); 10768 dev->core_stats = NULL; 10769 free_percpu(dev->xdp_bulkq); 10770 dev->xdp_bulkq = NULL; 10771 10772 /* Compatibility with error handling in drivers */ 10773 if (dev->reg_state == NETREG_UNINITIALIZED) { 10774 netdev_freemem(dev); 10775 return; 10776 } 10777 10778 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10779 dev->reg_state = NETREG_RELEASED; 10780 10781 /* will free via device release */ 10782 put_device(&dev->dev); 10783 } 10784 EXPORT_SYMBOL(free_netdev); 10785 10786 /** 10787 * synchronize_net - Synchronize with packet receive processing 10788 * 10789 * Wait for packets currently being received to be done. 10790 * Does not block later packets from starting. 10791 */ 10792 void synchronize_net(void) 10793 { 10794 might_sleep(); 10795 if (rtnl_is_locked()) 10796 synchronize_rcu_expedited(); 10797 else 10798 synchronize_rcu(); 10799 } 10800 EXPORT_SYMBOL(synchronize_net); 10801 10802 /** 10803 * unregister_netdevice_queue - remove device from the kernel 10804 * @dev: device 10805 * @head: list 10806 * 10807 * This function shuts down a device interface and removes it 10808 * from the kernel tables. 10809 * If head not NULL, device is queued to be unregistered later. 10810 * 10811 * Callers must hold the rtnl semaphore. You may want 10812 * unregister_netdev() instead of this. 10813 */ 10814 10815 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10816 { 10817 ASSERT_RTNL(); 10818 10819 if (head) { 10820 list_move_tail(&dev->unreg_list, head); 10821 } else { 10822 LIST_HEAD(single); 10823 10824 list_add(&dev->unreg_list, &single); 10825 unregister_netdevice_many(&single); 10826 } 10827 } 10828 EXPORT_SYMBOL(unregister_netdevice_queue); 10829 10830 void unregister_netdevice_many_notify(struct list_head *head, 10831 u32 portid, const struct nlmsghdr *nlh) 10832 { 10833 struct net_device *dev, *tmp; 10834 LIST_HEAD(close_head); 10835 10836 BUG_ON(dev_boot_phase); 10837 ASSERT_RTNL(); 10838 10839 if (list_empty(head)) 10840 return; 10841 10842 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10843 /* Some devices call without registering 10844 * for initialization unwind. Remove those 10845 * devices and proceed with the remaining. 10846 */ 10847 if (dev->reg_state == NETREG_UNINITIALIZED) { 10848 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10849 dev->name, dev); 10850 10851 WARN_ON(1); 10852 list_del(&dev->unreg_list); 10853 continue; 10854 } 10855 dev->dismantle = true; 10856 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10857 } 10858 10859 /* If device is running, close it first. */ 10860 list_for_each_entry(dev, head, unreg_list) 10861 list_add_tail(&dev->close_list, &close_head); 10862 dev_close_many(&close_head, true); 10863 10864 list_for_each_entry(dev, head, unreg_list) { 10865 /* And unlink it from device chain. */ 10866 write_lock(&dev_base_lock); 10867 unlist_netdevice(dev, false); 10868 dev->reg_state = NETREG_UNREGISTERING; 10869 write_unlock(&dev_base_lock); 10870 } 10871 flush_all_backlogs(); 10872 10873 synchronize_net(); 10874 10875 list_for_each_entry(dev, head, unreg_list) { 10876 struct sk_buff *skb = NULL; 10877 10878 /* Shutdown queueing discipline. */ 10879 dev_shutdown(dev); 10880 10881 dev_xdp_uninstall(dev); 10882 bpf_dev_bound_netdev_unregister(dev); 10883 10884 netdev_offload_xstats_disable_all(dev); 10885 10886 /* Notify protocols, that we are about to destroy 10887 * this device. They should clean all the things. 10888 */ 10889 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10890 10891 if (!dev->rtnl_link_ops || 10892 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10893 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10894 GFP_KERNEL, NULL, 0, 10895 portid, nlh); 10896 10897 /* 10898 * Flush the unicast and multicast chains 10899 */ 10900 dev_uc_flush(dev); 10901 dev_mc_flush(dev); 10902 10903 netdev_name_node_alt_flush(dev); 10904 netdev_name_node_free(dev->name_node); 10905 10906 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10907 10908 if (dev->netdev_ops->ndo_uninit) 10909 dev->netdev_ops->ndo_uninit(dev); 10910 10911 if (skb) 10912 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10913 10914 /* Notifier chain MUST detach us all upper devices. */ 10915 WARN_ON(netdev_has_any_upper_dev(dev)); 10916 WARN_ON(netdev_has_any_lower_dev(dev)); 10917 10918 /* Remove entries from kobject tree */ 10919 netdev_unregister_kobject(dev); 10920 #ifdef CONFIG_XPS 10921 /* Remove XPS queueing entries */ 10922 netif_reset_xps_queues_gt(dev, 0); 10923 #endif 10924 } 10925 10926 synchronize_net(); 10927 10928 list_for_each_entry(dev, head, unreg_list) { 10929 netdev_put(dev, &dev->dev_registered_tracker); 10930 net_set_todo(dev); 10931 } 10932 10933 list_del(head); 10934 } 10935 10936 /** 10937 * unregister_netdevice_many - unregister many devices 10938 * @head: list of devices 10939 * 10940 * Note: As most callers use a stack allocated list_head, 10941 * we force a list_del() to make sure stack wont be corrupted later. 10942 */ 10943 void unregister_netdevice_many(struct list_head *head) 10944 { 10945 unregister_netdevice_many_notify(head, 0, NULL); 10946 } 10947 EXPORT_SYMBOL(unregister_netdevice_many); 10948 10949 /** 10950 * unregister_netdev - remove device from the kernel 10951 * @dev: device 10952 * 10953 * This function shuts down a device interface and removes it 10954 * from the kernel tables. 10955 * 10956 * This is just a wrapper for unregister_netdevice that takes 10957 * the rtnl semaphore. In general you want to use this and not 10958 * unregister_netdevice. 10959 */ 10960 void unregister_netdev(struct net_device *dev) 10961 { 10962 rtnl_lock(); 10963 unregister_netdevice(dev); 10964 rtnl_unlock(); 10965 } 10966 EXPORT_SYMBOL(unregister_netdev); 10967 10968 /** 10969 * __dev_change_net_namespace - move device to different nethost namespace 10970 * @dev: device 10971 * @net: network namespace 10972 * @pat: If not NULL name pattern to try if the current device name 10973 * is already taken in the destination network namespace. 10974 * @new_ifindex: If not zero, specifies device index in the target 10975 * namespace. 10976 * 10977 * This function shuts down a device interface and moves it 10978 * to a new network namespace. On success 0 is returned, on 10979 * a failure a netagive errno code is returned. 10980 * 10981 * Callers must hold the rtnl semaphore. 10982 */ 10983 10984 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10985 const char *pat, int new_ifindex) 10986 { 10987 struct net *net_old = dev_net(dev); 10988 int err, new_nsid; 10989 10990 ASSERT_RTNL(); 10991 10992 /* Don't allow namespace local devices to be moved. */ 10993 err = -EINVAL; 10994 if (dev->features & NETIF_F_NETNS_LOCAL) 10995 goto out; 10996 10997 /* Ensure the device has been registrered */ 10998 if (dev->reg_state != NETREG_REGISTERED) 10999 goto out; 11000 11001 /* Get out if there is nothing todo */ 11002 err = 0; 11003 if (net_eq(net_old, net)) 11004 goto out; 11005 11006 /* Pick the destination device name, and ensure 11007 * we can use it in the destination network namespace. 11008 */ 11009 err = -EEXIST; 11010 if (netdev_name_in_use(net, dev->name)) { 11011 /* We get here if we can't use the current device name */ 11012 if (!pat) 11013 goto out; 11014 err = dev_get_valid_name(net, dev, pat); 11015 if (err < 0) 11016 goto out; 11017 } 11018 11019 /* Check that new_ifindex isn't used yet. */ 11020 err = -EBUSY; 11021 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11022 goto out; 11023 11024 /* 11025 * And now a mini version of register_netdevice unregister_netdevice. 11026 */ 11027 11028 /* If device is running close it first. */ 11029 dev_close(dev); 11030 11031 /* And unlink it from device chain */ 11032 unlist_netdevice(dev, true); 11033 11034 synchronize_net(); 11035 11036 /* Shutdown queueing discipline. */ 11037 dev_shutdown(dev); 11038 11039 /* Notify protocols, that we are about to destroy 11040 * this device. They should clean all the things. 11041 * 11042 * Note that dev->reg_state stays at NETREG_REGISTERED. 11043 * This is wanted because this way 8021q and macvlan know 11044 * the device is just moving and can keep their slaves up. 11045 */ 11046 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11047 rcu_barrier(); 11048 11049 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11050 /* If there is an ifindex conflict assign a new one */ 11051 if (!new_ifindex) { 11052 if (__dev_get_by_index(net, dev->ifindex)) 11053 new_ifindex = dev_new_index(net); 11054 else 11055 new_ifindex = dev->ifindex; 11056 } 11057 11058 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11059 new_ifindex); 11060 11061 /* 11062 * Flush the unicast and multicast chains 11063 */ 11064 dev_uc_flush(dev); 11065 dev_mc_flush(dev); 11066 11067 /* Send a netdev-removed uevent to the old namespace */ 11068 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11069 netdev_adjacent_del_links(dev); 11070 11071 /* Move per-net netdevice notifiers that are following the netdevice */ 11072 move_netdevice_notifiers_dev_net(dev, net); 11073 11074 /* Actually switch the network namespace */ 11075 dev_net_set(dev, net); 11076 dev->ifindex = new_ifindex; 11077 11078 /* Send a netdev-add uevent to the new namespace */ 11079 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11080 netdev_adjacent_add_links(dev); 11081 11082 /* Fixup kobjects */ 11083 err = device_rename(&dev->dev, dev->name); 11084 WARN_ON(err); 11085 11086 /* Adapt owner in case owning user namespace of target network 11087 * namespace is different from the original one. 11088 */ 11089 err = netdev_change_owner(dev, net_old, net); 11090 WARN_ON(err); 11091 11092 /* Add the device back in the hashes */ 11093 list_netdevice(dev); 11094 11095 /* Notify protocols, that a new device appeared. */ 11096 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11097 11098 /* 11099 * Prevent userspace races by waiting until the network 11100 * device is fully setup before sending notifications. 11101 */ 11102 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11103 11104 synchronize_net(); 11105 err = 0; 11106 out: 11107 return err; 11108 } 11109 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11110 11111 static int dev_cpu_dead(unsigned int oldcpu) 11112 { 11113 struct sk_buff **list_skb; 11114 struct sk_buff *skb; 11115 unsigned int cpu; 11116 struct softnet_data *sd, *oldsd, *remsd = NULL; 11117 11118 local_irq_disable(); 11119 cpu = smp_processor_id(); 11120 sd = &per_cpu(softnet_data, cpu); 11121 oldsd = &per_cpu(softnet_data, oldcpu); 11122 11123 /* Find end of our completion_queue. */ 11124 list_skb = &sd->completion_queue; 11125 while (*list_skb) 11126 list_skb = &(*list_skb)->next; 11127 /* Append completion queue from offline CPU. */ 11128 *list_skb = oldsd->completion_queue; 11129 oldsd->completion_queue = NULL; 11130 11131 /* Append output queue from offline CPU. */ 11132 if (oldsd->output_queue) { 11133 *sd->output_queue_tailp = oldsd->output_queue; 11134 sd->output_queue_tailp = oldsd->output_queue_tailp; 11135 oldsd->output_queue = NULL; 11136 oldsd->output_queue_tailp = &oldsd->output_queue; 11137 } 11138 /* Append NAPI poll list from offline CPU, with one exception : 11139 * process_backlog() must be called by cpu owning percpu backlog. 11140 * We properly handle process_queue & input_pkt_queue later. 11141 */ 11142 while (!list_empty(&oldsd->poll_list)) { 11143 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11144 struct napi_struct, 11145 poll_list); 11146 11147 list_del_init(&napi->poll_list); 11148 if (napi->poll == process_backlog) 11149 napi->state = 0; 11150 else 11151 ____napi_schedule(sd, napi); 11152 } 11153 11154 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11155 local_irq_enable(); 11156 11157 #ifdef CONFIG_RPS 11158 remsd = oldsd->rps_ipi_list; 11159 oldsd->rps_ipi_list = NULL; 11160 #endif 11161 /* send out pending IPI's on offline CPU */ 11162 net_rps_send_ipi(remsd); 11163 11164 /* Process offline CPU's input_pkt_queue */ 11165 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11166 netif_rx(skb); 11167 input_queue_head_incr(oldsd); 11168 } 11169 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11170 netif_rx(skb); 11171 input_queue_head_incr(oldsd); 11172 } 11173 11174 return 0; 11175 } 11176 11177 /** 11178 * netdev_increment_features - increment feature set by one 11179 * @all: current feature set 11180 * @one: new feature set 11181 * @mask: mask feature set 11182 * 11183 * Computes a new feature set after adding a device with feature set 11184 * @one to the master device with current feature set @all. Will not 11185 * enable anything that is off in @mask. Returns the new feature set. 11186 */ 11187 netdev_features_t netdev_increment_features(netdev_features_t all, 11188 netdev_features_t one, netdev_features_t mask) 11189 { 11190 if (mask & NETIF_F_HW_CSUM) 11191 mask |= NETIF_F_CSUM_MASK; 11192 mask |= NETIF_F_VLAN_CHALLENGED; 11193 11194 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11195 all &= one | ~NETIF_F_ALL_FOR_ALL; 11196 11197 /* If one device supports hw checksumming, set for all. */ 11198 if (all & NETIF_F_HW_CSUM) 11199 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11200 11201 return all; 11202 } 11203 EXPORT_SYMBOL(netdev_increment_features); 11204 11205 static struct hlist_head * __net_init netdev_create_hash(void) 11206 { 11207 int i; 11208 struct hlist_head *hash; 11209 11210 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11211 if (hash != NULL) 11212 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11213 INIT_HLIST_HEAD(&hash[i]); 11214 11215 return hash; 11216 } 11217 11218 /* Initialize per network namespace state */ 11219 static int __net_init netdev_init(struct net *net) 11220 { 11221 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11222 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11223 11224 INIT_LIST_HEAD(&net->dev_base_head); 11225 11226 net->dev_name_head = netdev_create_hash(); 11227 if (net->dev_name_head == NULL) 11228 goto err_name; 11229 11230 net->dev_index_head = netdev_create_hash(); 11231 if (net->dev_index_head == NULL) 11232 goto err_idx; 11233 11234 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11235 11236 return 0; 11237 11238 err_idx: 11239 kfree(net->dev_name_head); 11240 err_name: 11241 return -ENOMEM; 11242 } 11243 11244 /** 11245 * netdev_drivername - network driver for the device 11246 * @dev: network device 11247 * 11248 * Determine network driver for device. 11249 */ 11250 const char *netdev_drivername(const struct net_device *dev) 11251 { 11252 const struct device_driver *driver; 11253 const struct device *parent; 11254 const char *empty = ""; 11255 11256 parent = dev->dev.parent; 11257 if (!parent) 11258 return empty; 11259 11260 driver = parent->driver; 11261 if (driver && driver->name) 11262 return driver->name; 11263 return empty; 11264 } 11265 11266 static void __netdev_printk(const char *level, const struct net_device *dev, 11267 struct va_format *vaf) 11268 { 11269 if (dev && dev->dev.parent) { 11270 dev_printk_emit(level[1] - '0', 11271 dev->dev.parent, 11272 "%s %s %s%s: %pV", 11273 dev_driver_string(dev->dev.parent), 11274 dev_name(dev->dev.parent), 11275 netdev_name(dev), netdev_reg_state(dev), 11276 vaf); 11277 } else if (dev) { 11278 printk("%s%s%s: %pV", 11279 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11280 } else { 11281 printk("%s(NULL net_device): %pV", level, vaf); 11282 } 11283 } 11284 11285 void netdev_printk(const char *level, const struct net_device *dev, 11286 const char *format, ...) 11287 { 11288 struct va_format vaf; 11289 va_list args; 11290 11291 va_start(args, format); 11292 11293 vaf.fmt = format; 11294 vaf.va = &args; 11295 11296 __netdev_printk(level, dev, &vaf); 11297 11298 va_end(args); 11299 } 11300 EXPORT_SYMBOL(netdev_printk); 11301 11302 #define define_netdev_printk_level(func, level) \ 11303 void func(const struct net_device *dev, const char *fmt, ...) \ 11304 { \ 11305 struct va_format vaf; \ 11306 va_list args; \ 11307 \ 11308 va_start(args, fmt); \ 11309 \ 11310 vaf.fmt = fmt; \ 11311 vaf.va = &args; \ 11312 \ 11313 __netdev_printk(level, dev, &vaf); \ 11314 \ 11315 va_end(args); \ 11316 } \ 11317 EXPORT_SYMBOL(func); 11318 11319 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11320 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11321 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11322 define_netdev_printk_level(netdev_err, KERN_ERR); 11323 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11324 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11325 define_netdev_printk_level(netdev_info, KERN_INFO); 11326 11327 static void __net_exit netdev_exit(struct net *net) 11328 { 11329 kfree(net->dev_name_head); 11330 kfree(net->dev_index_head); 11331 if (net != &init_net) 11332 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11333 } 11334 11335 static struct pernet_operations __net_initdata netdev_net_ops = { 11336 .init = netdev_init, 11337 .exit = netdev_exit, 11338 }; 11339 11340 static void __net_exit default_device_exit_net(struct net *net) 11341 { 11342 struct net_device *dev, *aux; 11343 /* 11344 * Push all migratable network devices back to the 11345 * initial network namespace 11346 */ 11347 ASSERT_RTNL(); 11348 for_each_netdev_safe(net, dev, aux) { 11349 int err; 11350 char fb_name[IFNAMSIZ]; 11351 11352 /* Ignore unmoveable devices (i.e. loopback) */ 11353 if (dev->features & NETIF_F_NETNS_LOCAL) 11354 continue; 11355 11356 /* Leave virtual devices for the generic cleanup */ 11357 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11358 continue; 11359 11360 /* Push remaining network devices to init_net */ 11361 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11362 if (netdev_name_in_use(&init_net, fb_name)) 11363 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11364 err = dev_change_net_namespace(dev, &init_net, fb_name); 11365 if (err) { 11366 pr_emerg("%s: failed to move %s to init_net: %d\n", 11367 __func__, dev->name, err); 11368 BUG(); 11369 } 11370 } 11371 } 11372 11373 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11374 { 11375 /* At exit all network devices most be removed from a network 11376 * namespace. Do this in the reverse order of registration. 11377 * Do this across as many network namespaces as possible to 11378 * improve batching efficiency. 11379 */ 11380 struct net_device *dev; 11381 struct net *net; 11382 LIST_HEAD(dev_kill_list); 11383 11384 rtnl_lock(); 11385 list_for_each_entry(net, net_list, exit_list) { 11386 default_device_exit_net(net); 11387 cond_resched(); 11388 } 11389 11390 list_for_each_entry(net, net_list, exit_list) { 11391 for_each_netdev_reverse(net, dev) { 11392 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11393 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11394 else 11395 unregister_netdevice_queue(dev, &dev_kill_list); 11396 } 11397 } 11398 unregister_netdevice_many(&dev_kill_list); 11399 rtnl_unlock(); 11400 } 11401 11402 static struct pernet_operations __net_initdata default_device_ops = { 11403 .exit_batch = default_device_exit_batch, 11404 }; 11405 11406 /* 11407 * Initialize the DEV module. At boot time this walks the device list and 11408 * unhooks any devices that fail to initialise (normally hardware not 11409 * present) and leaves us with a valid list of present and active devices. 11410 * 11411 */ 11412 11413 /* 11414 * This is called single threaded during boot, so no need 11415 * to take the rtnl semaphore. 11416 */ 11417 static int __init net_dev_init(void) 11418 { 11419 int i, rc = -ENOMEM; 11420 11421 BUG_ON(!dev_boot_phase); 11422 11423 if (dev_proc_init()) 11424 goto out; 11425 11426 if (netdev_kobject_init()) 11427 goto out; 11428 11429 INIT_LIST_HEAD(&ptype_all); 11430 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11431 INIT_LIST_HEAD(&ptype_base[i]); 11432 11433 if (register_pernet_subsys(&netdev_net_ops)) 11434 goto out; 11435 11436 /* 11437 * Initialise the packet receive queues. 11438 */ 11439 11440 for_each_possible_cpu(i) { 11441 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11442 struct softnet_data *sd = &per_cpu(softnet_data, i); 11443 11444 INIT_WORK(flush, flush_backlog); 11445 11446 skb_queue_head_init(&sd->input_pkt_queue); 11447 skb_queue_head_init(&sd->process_queue); 11448 #ifdef CONFIG_XFRM_OFFLOAD 11449 skb_queue_head_init(&sd->xfrm_backlog); 11450 #endif 11451 INIT_LIST_HEAD(&sd->poll_list); 11452 sd->output_queue_tailp = &sd->output_queue; 11453 #ifdef CONFIG_RPS 11454 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11455 sd->cpu = i; 11456 #endif 11457 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11458 spin_lock_init(&sd->defer_lock); 11459 11460 init_gro_hash(&sd->backlog); 11461 sd->backlog.poll = process_backlog; 11462 sd->backlog.weight = weight_p; 11463 } 11464 11465 dev_boot_phase = 0; 11466 11467 /* The loopback device is special if any other network devices 11468 * is present in a network namespace the loopback device must 11469 * be present. Since we now dynamically allocate and free the 11470 * loopback device ensure this invariant is maintained by 11471 * keeping the loopback device as the first device on the 11472 * list of network devices. Ensuring the loopback devices 11473 * is the first device that appears and the last network device 11474 * that disappears. 11475 */ 11476 if (register_pernet_device(&loopback_net_ops)) 11477 goto out; 11478 11479 if (register_pernet_device(&default_device_ops)) 11480 goto out; 11481 11482 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11483 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11484 11485 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11486 NULL, dev_cpu_dead); 11487 WARN_ON(rc < 0); 11488 rc = 0; 11489 out: 11490 return rc; 11491 } 11492 11493 subsys_initcall(net_dev_init); 11494