1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "dev.h" 155 #include "net-sysfs.h" 156 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_info(unsigned long val, 164 struct netdev_notifier_info *info); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 netdev_name_node_del(name_node); 349 kfree(name_node->name); 350 netdev_name_node_free(name_node); 351 } 352 353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 354 { 355 struct netdev_name_node *name_node; 356 struct net *net = dev_net(dev); 357 358 name_node = netdev_name_node_lookup(net, name); 359 if (!name_node) 360 return -ENOENT; 361 /* lookup might have found our primary name or a name belonging 362 * to another device. 363 */ 364 if (name_node == dev->name_node || name_node->dev != dev) 365 return -EINVAL; 366 367 __netdev_name_node_alt_destroy(name_node); 368 369 return 0; 370 } 371 372 static void netdev_name_node_alt_flush(struct net_device *dev) 373 { 374 struct netdev_name_node *name_node, *tmp; 375 376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 377 __netdev_name_node_alt_destroy(name_node); 378 } 379 380 /* Device list insertion */ 381 static void list_netdevice(struct net_device *dev) 382 { 383 struct net *net = dev_net(dev); 384 385 ASSERT_RTNL(); 386 387 write_lock(&dev_base_lock); 388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 389 netdev_name_node_add(net, dev->name_node); 390 hlist_add_head_rcu(&dev->index_hlist, 391 dev_index_hash(net, dev->ifindex)); 392 write_unlock(&dev_base_lock); 393 394 dev_base_seq_inc(net); 395 } 396 397 /* Device list removal 398 * caller must respect a RCU grace period before freeing/reusing dev 399 */ 400 static void unlist_netdevice(struct net_device *dev) 401 { 402 ASSERT_RTNL(); 403 404 /* Unlink dev from the device chain */ 405 write_lock(&dev_base_lock); 406 list_del_rcu(&dev->dev_list); 407 netdev_name_node_del(dev->name_node); 408 hlist_del_rcu(&dev->index_hlist); 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 strlcpy(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 strlcpy(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 /* Some auto-enslaved devices e.g. failover slaves are 1165 * special, as userspace might rename the device after 1166 * the interface had been brought up and running since 1167 * the point kernel initiated auto-enslavement. Allow 1168 * live name change even when these slave devices are 1169 * up and running. 1170 * 1171 * Typically, users of these auto-enslaving devices 1172 * don't actually care about slave name change, as 1173 * they are supposed to operate on master interface 1174 * directly. 1175 */ 1176 if (dev->flags & IFF_UP && 1177 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1178 return -EBUSY; 1179 1180 down_write(&devnet_rename_sem); 1181 1182 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1183 up_write(&devnet_rename_sem); 1184 return 0; 1185 } 1186 1187 memcpy(oldname, dev->name, IFNAMSIZ); 1188 1189 err = dev_get_valid_name(net, dev, newname); 1190 if (err < 0) { 1191 up_write(&devnet_rename_sem); 1192 return err; 1193 } 1194 1195 if (oldname[0] && !strchr(oldname, '%')) 1196 netdev_info(dev, "renamed from %s\n", oldname); 1197 1198 old_assign_type = dev->name_assign_type; 1199 dev->name_assign_type = NET_NAME_RENAMED; 1200 1201 rollback: 1202 ret = device_rename(&dev->dev, dev->name); 1203 if (ret) { 1204 memcpy(dev->name, oldname, IFNAMSIZ); 1205 dev->name_assign_type = old_assign_type; 1206 up_write(&devnet_rename_sem); 1207 return ret; 1208 } 1209 1210 up_write(&devnet_rename_sem); 1211 1212 netdev_adjacent_rename_links(dev, oldname); 1213 1214 write_lock(&dev_base_lock); 1215 netdev_name_node_del(dev->name_node); 1216 write_unlock(&dev_base_lock); 1217 1218 synchronize_rcu(); 1219 1220 write_lock(&dev_base_lock); 1221 netdev_name_node_add(net, dev->name_node); 1222 write_unlock(&dev_base_lock); 1223 1224 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1225 ret = notifier_to_errno(ret); 1226 1227 if (ret) { 1228 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1229 if (err >= 0) { 1230 err = ret; 1231 down_write(&devnet_rename_sem); 1232 memcpy(dev->name, oldname, IFNAMSIZ); 1233 memcpy(oldname, newname, IFNAMSIZ); 1234 dev->name_assign_type = old_assign_type; 1235 old_assign_type = NET_NAME_RENAMED; 1236 goto rollback; 1237 } else { 1238 netdev_err(dev, "name change rollback failed: %d\n", 1239 ret); 1240 } 1241 } 1242 1243 return err; 1244 } 1245 1246 /** 1247 * dev_set_alias - change ifalias of a device 1248 * @dev: device 1249 * @alias: name up to IFALIASZ 1250 * @len: limit of bytes to copy from info 1251 * 1252 * Set ifalias for a device, 1253 */ 1254 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1255 { 1256 struct dev_ifalias *new_alias = NULL; 1257 1258 if (len >= IFALIASZ) 1259 return -EINVAL; 1260 1261 if (len) { 1262 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1263 if (!new_alias) 1264 return -ENOMEM; 1265 1266 memcpy(new_alias->ifalias, alias, len); 1267 new_alias->ifalias[len] = 0; 1268 } 1269 1270 mutex_lock(&ifalias_mutex); 1271 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1272 mutex_is_locked(&ifalias_mutex)); 1273 mutex_unlock(&ifalias_mutex); 1274 1275 if (new_alias) 1276 kfree_rcu(new_alias, rcuhead); 1277 1278 return len; 1279 } 1280 EXPORT_SYMBOL(dev_set_alias); 1281 1282 /** 1283 * dev_get_alias - get ifalias of a device 1284 * @dev: device 1285 * @name: buffer to store name of ifalias 1286 * @len: size of buffer 1287 * 1288 * get ifalias for a device. Caller must make sure dev cannot go 1289 * away, e.g. rcu read lock or own a reference count to device. 1290 */ 1291 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1292 { 1293 const struct dev_ifalias *alias; 1294 int ret = 0; 1295 1296 rcu_read_lock(); 1297 alias = rcu_dereference(dev->ifalias); 1298 if (alias) 1299 ret = snprintf(name, len, "%s", alias->ifalias); 1300 rcu_read_unlock(); 1301 1302 return ret; 1303 } 1304 1305 /** 1306 * netdev_features_change - device changes features 1307 * @dev: device to cause notification 1308 * 1309 * Called to indicate a device has changed features. 1310 */ 1311 void netdev_features_change(struct net_device *dev) 1312 { 1313 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1314 } 1315 EXPORT_SYMBOL(netdev_features_change); 1316 1317 /** 1318 * netdev_state_change - device changes state 1319 * @dev: device to cause notification 1320 * 1321 * Called to indicate a device has changed state. This function calls 1322 * the notifier chains for netdev_chain and sends a NEWLINK message 1323 * to the routing socket. 1324 */ 1325 void netdev_state_change(struct net_device *dev) 1326 { 1327 if (dev->flags & IFF_UP) { 1328 struct netdev_notifier_change_info change_info = { 1329 .info.dev = dev, 1330 }; 1331 1332 call_netdevice_notifiers_info(NETDEV_CHANGE, 1333 &change_info.info); 1334 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1335 } 1336 } 1337 EXPORT_SYMBOL(netdev_state_change); 1338 1339 /** 1340 * __netdev_notify_peers - notify network peers about existence of @dev, 1341 * to be called when rtnl lock is already held. 1342 * @dev: network device 1343 * 1344 * Generate traffic such that interested network peers are aware of 1345 * @dev, such as by generating a gratuitous ARP. This may be used when 1346 * a device wants to inform the rest of the network about some sort of 1347 * reconfiguration such as a failover event or virtual machine 1348 * migration. 1349 */ 1350 void __netdev_notify_peers(struct net_device *dev) 1351 { 1352 ASSERT_RTNL(); 1353 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1354 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1355 } 1356 EXPORT_SYMBOL(__netdev_notify_peers); 1357 1358 /** 1359 * netdev_notify_peers - notify network peers about existence of @dev 1360 * @dev: network device 1361 * 1362 * Generate traffic such that interested network peers are aware of 1363 * @dev, such as by generating a gratuitous ARP. This may be used when 1364 * a device wants to inform the rest of the network about some sort of 1365 * reconfiguration such as a failover event or virtual machine 1366 * migration. 1367 */ 1368 void netdev_notify_peers(struct net_device *dev) 1369 { 1370 rtnl_lock(); 1371 __netdev_notify_peers(dev); 1372 rtnl_unlock(); 1373 } 1374 EXPORT_SYMBOL(netdev_notify_peers); 1375 1376 static int napi_threaded_poll(void *data); 1377 1378 static int napi_kthread_create(struct napi_struct *n) 1379 { 1380 int err = 0; 1381 1382 /* Create and wake up the kthread once to put it in 1383 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1384 * warning and work with loadavg. 1385 */ 1386 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1387 n->dev->name, n->napi_id); 1388 if (IS_ERR(n->thread)) { 1389 err = PTR_ERR(n->thread); 1390 pr_err("kthread_run failed with err %d\n", err); 1391 n->thread = NULL; 1392 } 1393 1394 return err; 1395 } 1396 1397 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1398 { 1399 const struct net_device_ops *ops = dev->netdev_ops; 1400 int ret; 1401 1402 ASSERT_RTNL(); 1403 dev_addr_check(dev); 1404 1405 if (!netif_device_present(dev)) { 1406 /* may be detached because parent is runtime-suspended */ 1407 if (dev->dev.parent) 1408 pm_runtime_resume(dev->dev.parent); 1409 if (!netif_device_present(dev)) 1410 return -ENODEV; 1411 } 1412 1413 /* Block netpoll from trying to do any rx path servicing. 1414 * If we don't do this there is a chance ndo_poll_controller 1415 * or ndo_poll may be running while we open the device 1416 */ 1417 netpoll_poll_disable(dev); 1418 1419 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1420 ret = notifier_to_errno(ret); 1421 if (ret) 1422 return ret; 1423 1424 set_bit(__LINK_STATE_START, &dev->state); 1425 1426 if (ops->ndo_validate_addr) 1427 ret = ops->ndo_validate_addr(dev); 1428 1429 if (!ret && ops->ndo_open) 1430 ret = ops->ndo_open(dev); 1431 1432 netpoll_poll_enable(dev); 1433 1434 if (ret) 1435 clear_bit(__LINK_STATE_START, &dev->state); 1436 else { 1437 dev->flags |= IFF_UP; 1438 dev_set_rx_mode(dev); 1439 dev_activate(dev); 1440 add_device_randomness(dev->dev_addr, dev->addr_len); 1441 } 1442 1443 return ret; 1444 } 1445 1446 /** 1447 * dev_open - prepare an interface for use. 1448 * @dev: device to open 1449 * @extack: netlink extended ack 1450 * 1451 * Takes a device from down to up state. The device's private open 1452 * function is invoked and then the multicast lists are loaded. Finally 1453 * the device is moved into the up state and a %NETDEV_UP message is 1454 * sent to the netdev notifier chain. 1455 * 1456 * Calling this function on an active interface is a nop. On a failure 1457 * a negative errno code is returned. 1458 */ 1459 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1460 { 1461 int ret; 1462 1463 if (dev->flags & IFF_UP) 1464 return 0; 1465 1466 ret = __dev_open(dev, extack); 1467 if (ret < 0) 1468 return ret; 1469 1470 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1471 call_netdevice_notifiers(NETDEV_UP, dev); 1472 1473 return ret; 1474 } 1475 EXPORT_SYMBOL(dev_open); 1476 1477 static void __dev_close_many(struct list_head *head) 1478 { 1479 struct net_device *dev; 1480 1481 ASSERT_RTNL(); 1482 might_sleep(); 1483 1484 list_for_each_entry(dev, head, close_list) { 1485 /* Temporarily disable netpoll until the interface is down */ 1486 netpoll_poll_disable(dev); 1487 1488 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1489 1490 clear_bit(__LINK_STATE_START, &dev->state); 1491 1492 /* Synchronize to scheduled poll. We cannot touch poll list, it 1493 * can be even on different cpu. So just clear netif_running(). 1494 * 1495 * dev->stop() will invoke napi_disable() on all of it's 1496 * napi_struct instances on this device. 1497 */ 1498 smp_mb__after_atomic(); /* Commit netif_running(). */ 1499 } 1500 1501 dev_deactivate_many(head); 1502 1503 list_for_each_entry(dev, head, close_list) { 1504 const struct net_device_ops *ops = dev->netdev_ops; 1505 1506 /* 1507 * Call the device specific close. This cannot fail. 1508 * Only if device is UP 1509 * 1510 * We allow it to be called even after a DETACH hot-plug 1511 * event. 1512 */ 1513 if (ops->ndo_stop) 1514 ops->ndo_stop(dev); 1515 1516 dev->flags &= ~IFF_UP; 1517 netpoll_poll_enable(dev); 1518 } 1519 } 1520 1521 static void __dev_close(struct net_device *dev) 1522 { 1523 LIST_HEAD(single); 1524 1525 list_add(&dev->close_list, &single); 1526 __dev_close_many(&single); 1527 list_del(&single); 1528 } 1529 1530 void dev_close_many(struct list_head *head, bool unlink) 1531 { 1532 struct net_device *dev, *tmp; 1533 1534 /* Remove the devices that don't need to be closed */ 1535 list_for_each_entry_safe(dev, tmp, head, close_list) 1536 if (!(dev->flags & IFF_UP)) 1537 list_del_init(&dev->close_list); 1538 1539 __dev_close_many(head); 1540 1541 list_for_each_entry_safe(dev, tmp, head, close_list) { 1542 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1543 call_netdevice_notifiers(NETDEV_DOWN, dev); 1544 if (unlink) 1545 list_del_init(&dev->close_list); 1546 } 1547 } 1548 EXPORT_SYMBOL(dev_close_many); 1549 1550 /** 1551 * dev_close - shutdown an interface. 1552 * @dev: device to shutdown 1553 * 1554 * This function moves an active device into down state. A 1555 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1556 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1557 * chain. 1558 */ 1559 void dev_close(struct net_device *dev) 1560 { 1561 if (dev->flags & IFF_UP) { 1562 LIST_HEAD(single); 1563 1564 list_add(&dev->close_list, &single); 1565 dev_close_many(&single, true); 1566 list_del(&single); 1567 } 1568 } 1569 EXPORT_SYMBOL(dev_close); 1570 1571 1572 /** 1573 * dev_disable_lro - disable Large Receive Offload on a device 1574 * @dev: device 1575 * 1576 * Disable Large Receive Offload (LRO) on a net device. Must be 1577 * called under RTNL. This is needed if received packets may be 1578 * forwarded to another interface. 1579 */ 1580 void dev_disable_lro(struct net_device *dev) 1581 { 1582 struct net_device *lower_dev; 1583 struct list_head *iter; 1584 1585 dev->wanted_features &= ~NETIF_F_LRO; 1586 netdev_update_features(dev); 1587 1588 if (unlikely(dev->features & NETIF_F_LRO)) 1589 netdev_WARN(dev, "failed to disable LRO!\n"); 1590 1591 netdev_for_each_lower_dev(dev, lower_dev, iter) 1592 dev_disable_lro(lower_dev); 1593 } 1594 EXPORT_SYMBOL(dev_disable_lro); 1595 1596 /** 1597 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1598 * @dev: device 1599 * 1600 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1601 * called under RTNL. This is needed if Generic XDP is installed on 1602 * the device. 1603 */ 1604 static void dev_disable_gro_hw(struct net_device *dev) 1605 { 1606 dev->wanted_features &= ~NETIF_F_GRO_HW; 1607 netdev_update_features(dev); 1608 1609 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1610 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1611 } 1612 1613 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1614 { 1615 #define N(val) \ 1616 case NETDEV_##val: \ 1617 return "NETDEV_" __stringify(val); 1618 switch (cmd) { 1619 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1620 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1621 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1622 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1623 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1624 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1625 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1626 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1627 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1628 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1629 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1630 } 1631 #undef N 1632 return "UNKNOWN_NETDEV_EVENT"; 1633 } 1634 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1635 1636 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1637 struct net_device *dev) 1638 { 1639 struct netdev_notifier_info info = { 1640 .dev = dev, 1641 }; 1642 1643 return nb->notifier_call(nb, val, &info); 1644 } 1645 1646 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1647 struct net_device *dev) 1648 { 1649 int err; 1650 1651 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1652 err = notifier_to_errno(err); 1653 if (err) 1654 return err; 1655 1656 if (!(dev->flags & IFF_UP)) 1657 return 0; 1658 1659 call_netdevice_notifier(nb, NETDEV_UP, dev); 1660 return 0; 1661 } 1662 1663 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1664 struct net_device *dev) 1665 { 1666 if (dev->flags & IFF_UP) { 1667 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1668 dev); 1669 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1670 } 1671 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1672 } 1673 1674 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1675 struct net *net) 1676 { 1677 struct net_device *dev; 1678 int err; 1679 1680 for_each_netdev(net, dev) { 1681 err = call_netdevice_register_notifiers(nb, dev); 1682 if (err) 1683 goto rollback; 1684 } 1685 return 0; 1686 1687 rollback: 1688 for_each_netdev_continue_reverse(net, dev) 1689 call_netdevice_unregister_notifiers(nb, dev); 1690 return err; 1691 } 1692 1693 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1694 struct net *net) 1695 { 1696 struct net_device *dev; 1697 1698 for_each_netdev(net, dev) 1699 call_netdevice_unregister_notifiers(nb, dev); 1700 } 1701 1702 static int dev_boot_phase = 1; 1703 1704 /** 1705 * register_netdevice_notifier - register a network notifier block 1706 * @nb: notifier 1707 * 1708 * Register a notifier to be called when network device events occur. 1709 * The notifier passed is linked into the kernel structures and must 1710 * not be reused until it has been unregistered. A negative errno code 1711 * is returned on a failure. 1712 * 1713 * When registered all registration and up events are replayed 1714 * to the new notifier to allow device to have a race free 1715 * view of the network device list. 1716 */ 1717 1718 int register_netdevice_notifier(struct notifier_block *nb) 1719 { 1720 struct net *net; 1721 int err; 1722 1723 /* Close race with setup_net() and cleanup_net() */ 1724 down_write(&pernet_ops_rwsem); 1725 rtnl_lock(); 1726 err = raw_notifier_chain_register(&netdev_chain, nb); 1727 if (err) 1728 goto unlock; 1729 if (dev_boot_phase) 1730 goto unlock; 1731 for_each_net(net) { 1732 err = call_netdevice_register_net_notifiers(nb, net); 1733 if (err) 1734 goto rollback; 1735 } 1736 1737 unlock: 1738 rtnl_unlock(); 1739 up_write(&pernet_ops_rwsem); 1740 return err; 1741 1742 rollback: 1743 for_each_net_continue_reverse(net) 1744 call_netdevice_unregister_net_notifiers(nb, net); 1745 1746 raw_notifier_chain_unregister(&netdev_chain, nb); 1747 goto unlock; 1748 } 1749 EXPORT_SYMBOL(register_netdevice_notifier); 1750 1751 /** 1752 * unregister_netdevice_notifier - unregister a network notifier block 1753 * @nb: notifier 1754 * 1755 * Unregister a notifier previously registered by 1756 * register_netdevice_notifier(). The notifier is unlinked into the 1757 * kernel structures and may then be reused. A negative errno code 1758 * is returned on a failure. 1759 * 1760 * After unregistering unregister and down device events are synthesized 1761 * for all devices on the device list to the removed notifier to remove 1762 * the need for special case cleanup code. 1763 */ 1764 1765 int unregister_netdevice_notifier(struct notifier_block *nb) 1766 { 1767 struct net *net; 1768 int err; 1769 1770 /* Close race with setup_net() and cleanup_net() */ 1771 down_write(&pernet_ops_rwsem); 1772 rtnl_lock(); 1773 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1774 if (err) 1775 goto unlock; 1776 1777 for_each_net(net) 1778 call_netdevice_unregister_net_notifiers(nb, net); 1779 1780 unlock: 1781 rtnl_unlock(); 1782 up_write(&pernet_ops_rwsem); 1783 return err; 1784 } 1785 EXPORT_SYMBOL(unregister_netdevice_notifier); 1786 1787 static int __register_netdevice_notifier_net(struct net *net, 1788 struct notifier_block *nb, 1789 bool ignore_call_fail) 1790 { 1791 int err; 1792 1793 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1794 if (err) 1795 return err; 1796 if (dev_boot_phase) 1797 return 0; 1798 1799 err = call_netdevice_register_net_notifiers(nb, net); 1800 if (err && !ignore_call_fail) 1801 goto chain_unregister; 1802 1803 return 0; 1804 1805 chain_unregister: 1806 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1807 return err; 1808 } 1809 1810 static int __unregister_netdevice_notifier_net(struct net *net, 1811 struct notifier_block *nb) 1812 { 1813 int err; 1814 1815 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1816 if (err) 1817 return err; 1818 1819 call_netdevice_unregister_net_notifiers(nb, net); 1820 return 0; 1821 } 1822 1823 /** 1824 * register_netdevice_notifier_net - register a per-netns network notifier block 1825 * @net: network namespace 1826 * @nb: notifier 1827 * 1828 * Register a notifier to be called when network device events occur. 1829 * The notifier passed is linked into the kernel structures and must 1830 * not be reused until it has been unregistered. A negative errno code 1831 * is returned on a failure. 1832 * 1833 * When registered all registration and up events are replayed 1834 * to the new notifier to allow device to have a race free 1835 * view of the network device list. 1836 */ 1837 1838 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1839 { 1840 int err; 1841 1842 rtnl_lock(); 1843 err = __register_netdevice_notifier_net(net, nb, false); 1844 rtnl_unlock(); 1845 return err; 1846 } 1847 EXPORT_SYMBOL(register_netdevice_notifier_net); 1848 1849 /** 1850 * unregister_netdevice_notifier_net - unregister a per-netns 1851 * network notifier block 1852 * @net: network namespace 1853 * @nb: notifier 1854 * 1855 * Unregister a notifier previously registered by 1856 * register_netdevice_notifier(). The notifier is unlinked into the 1857 * kernel structures and may then be reused. A negative errno code 1858 * is returned on a failure. 1859 * 1860 * After unregistering unregister and down device events are synthesized 1861 * for all devices on the device list to the removed notifier to remove 1862 * the need for special case cleanup code. 1863 */ 1864 1865 int unregister_netdevice_notifier_net(struct net *net, 1866 struct notifier_block *nb) 1867 { 1868 int err; 1869 1870 rtnl_lock(); 1871 err = __unregister_netdevice_notifier_net(net, nb); 1872 rtnl_unlock(); 1873 return err; 1874 } 1875 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1876 1877 int register_netdevice_notifier_dev_net(struct net_device *dev, 1878 struct notifier_block *nb, 1879 struct netdev_net_notifier *nn) 1880 { 1881 int err; 1882 1883 rtnl_lock(); 1884 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1885 if (!err) { 1886 nn->nb = nb; 1887 list_add(&nn->list, &dev->net_notifier_list); 1888 } 1889 rtnl_unlock(); 1890 return err; 1891 } 1892 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1893 1894 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1895 struct notifier_block *nb, 1896 struct netdev_net_notifier *nn) 1897 { 1898 int err; 1899 1900 rtnl_lock(); 1901 list_del(&nn->list); 1902 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1903 rtnl_unlock(); 1904 return err; 1905 } 1906 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1907 1908 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1909 struct net *net) 1910 { 1911 struct netdev_net_notifier *nn; 1912 1913 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1914 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1915 __register_netdevice_notifier_net(net, nn->nb, true); 1916 } 1917 } 1918 1919 /** 1920 * call_netdevice_notifiers_info - call all network notifier blocks 1921 * @val: value passed unmodified to notifier function 1922 * @info: notifier information data 1923 * 1924 * Call all network notifier blocks. Parameters and return value 1925 * are as for raw_notifier_call_chain(). 1926 */ 1927 1928 static int call_netdevice_notifiers_info(unsigned long val, 1929 struct netdev_notifier_info *info) 1930 { 1931 struct net *net = dev_net(info->dev); 1932 int ret; 1933 1934 ASSERT_RTNL(); 1935 1936 /* Run per-netns notifier block chain first, then run the global one. 1937 * Hopefully, one day, the global one is going to be removed after 1938 * all notifier block registrators get converted to be per-netns. 1939 */ 1940 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1941 if (ret & NOTIFY_STOP_MASK) 1942 return ret; 1943 return raw_notifier_call_chain(&netdev_chain, val, info); 1944 } 1945 1946 /** 1947 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1948 * for and rollback on error 1949 * @val_up: value passed unmodified to notifier function 1950 * @val_down: value passed unmodified to the notifier function when 1951 * recovering from an error on @val_up 1952 * @info: notifier information data 1953 * 1954 * Call all per-netns network notifier blocks, but not notifier blocks on 1955 * the global notifier chain. Parameters and return value are as for 1956 * raw_notifier_call_chain_robust(). 1957 */ 1958 1959 static int 1960 call_netdevice_notifiers_info_robust(unsigned long val_up, 1961 unsigned long val_down, 1962 struct netdev_notifier_info *info) 1963 { 1964 struct net *net = dev_net(info->dev); 1965 1966 ASSERT_RTNL(); 1967 1968 return raw_notifier_call_chain_robust(&net->netdev_chain, 1969 val_up, val_down, info); 1970 } 1971 1972 static int call_netdevice_notifiers_extack(unsigned long val, 1973 struct net_device *dev, 1974 struct netlink_ext_ack *extack) 1975 { 1976 struct netdev_notifier_info info = { 1977 .dev = dev, 1978 .extack = extack, 1979 }; 1980 1981 return call_netdevice_notifiers_info(val, &info); 1982 } 1983 1984 /** 1985 * call_netdevice_notifiers - call all network notifier blocks 1986 * @val: value passed unmodified to notifier function 1987 * @dev: net_device pointer passed unmodified to notifier function 1988 * 1989 * Call all network notifier blocks. Parameters and return value 1990 * are as for raw_notifier_call_chain(). 1991 */ 1992 1993 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1994 { 1995 return call_netdevice_notifiers_extack(val, dev, NULL); 1996 } 1997 EXPORT_SYMBOL(call_netdevice_notifiers); 1998 1999 /** 2000 * call_netdevice_notifiers_mtu - call all network notifier blocks 2001 * @val: value passed unmodified to notifier function 2002 * @dev: net_device pointer passed unmodified to notifier function 2003 * @arg: additional u32 argument passed to the notifier function 2004 * 2005 * Call all network notifier blocks. Parameters and return value 2006 * are as for raw_notifier_call_chain(). 2007 */ 2008 static int call_netdevice_notifiers_mtu(unsigned long val, 2009 struct net_device *dev, u32 arg) 2010 { 2011 struct netdev_notifier_info_ext info = { 2012 .info.dev = dev, 2013 .ext.mtu = arg, 2014 }; 2015 2016 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2017 2018 return call_netdevice_notifiers_info(val, &info.info); 2019 } 2020 2021 #ifdef CONFIG_NET_INGRESS 2022 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2023 2024 void net_inc_ingress_queue(void) 2025 { 2026 static_branch_inc(&ingress_needed_key); 2027 } 2028 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2029 2030 void net_dec_ingress_queue(void) 2031 { 2032 static_branch_dec(&ingress_needed_key); 2033 } 2034 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2035 #endif 2036 2037 #ifdef CONFIG_NET_EGRESS 2038 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2039 2040 void net_inc_egress_queue(void) 2041 { 2042 static_branch_inc(&egress_needed_key); 2043 } 2044 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2045 2046 void net_dec_egress_queue(void) 2047 { 2048 static_branch_dec(&egress_needed_key); 2049 } 2050 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2051 #endif 2052 2053 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2054 EXPORT_SYMBOL(netstamp_needed_key); 2055 #ifdef CONFIG_JUMP_LABEL 2056 static atomic_t netstamp_needed_deferred; 2057 static atomic_t netstamp_wanted; 2058 static void netstamp_clear(struct work_struct *work) 2059 { 2060 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2061 int wanted; 2062 2063 wanted = atomic_add_return(deferred, &netstamp_wanted); 2064 if (wanted > 0) 2065 static_branch_enable(&netstamp_needed_key); 2066 else 2067 static_branch_disable(&netstamp_needed_key); 2068 } 2069 static DECLARE_WORK(netstamp_work, netstamp_clear); 2070 #endif 2071 2072 void net_enable_timestamp(void) 2073 { 2074 #ifdef CONFIG_JUMP_LABEL 2075 int wanted; 2076 2077 while (1) { 2078 wanted = atomic_read(&netstamp_wanted); 2079 if (wanted <= 0) 2080 break; 2081 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2082 return; 2083 } 2084 atomic_inc(&netstamp_needed_deferred); 2085 schedule_work(&netstamp_work); 2086 #else 2087 static_branch_inc(&netstamp_needed_key); 2088 #endif 2089 } 2090 EXPORT_SYMBOL(net_enable_timestamp); 2091 2092 void net_disable_timestamp(void) 2093 { 2094 #ifdef CONFIG_JUMP_LABEL 2095 int wanted; 2096 2097 while (1) { 2098 wanted = atomic_read(&netstamp_wanted); 2099 if (wanted <= 1) 2100 break; 2101 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2102 return; 2103 } 2104 atomic_dec(&netstamp_needed_deferred); 2105 schedule_work(&netstamp_work); 2106 #else 2107 static_branch_dec(&netstamp_needed_key); 2108 #endif 2109 } 2110 EXPORT_SYMBOL(net_disable_timestamp); 2111 2112 static inline void net_timestamp_set(struct sk_buff *skb) 2113 { 2114 skb->tstamp = 0; 2115 skb->mono_delivery_time = 0; 2116 if (static_branch_unlikely(&netstamp_needed_key)) 2117 skb->tstamp = ktime_get_real(); 2118 } 2119 2120 #define net_timestamp_check(COND, SKB) \ 2121 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2122 if ((COND) && !(SKB)->tstamp) \ 2123 (SKB)->tstamp = ktime_get_real(); \ 2124 } \ 2125 2126 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2127 { 2128 return __is_skb_forwardable(dev, skb, true); 2129 } 2130 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2131 2132 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2133 bool check_mtu) 2134 { 2135 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2136 2137 if (likely(!ret)) { 2138 skb->protocol = eth_type_trans(skb, dev); 2139 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2140 } 2141 2142 return ret; 2143 } 2144 2145 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2146 { 2147 return __dev_forward_skb2(dev, skb, true); 2148 } 2149 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2150 2151 /** 2152 * dev_forward_skb - loopback an skb to another netif 2153 * 2154 * @dev: destination network device 2155 * @skb: buffer to forward 2156 * 2157 * return values: 2158 * NET_RX_SUCCESS (no congestion) 2159 * NET_RX_DROP (packet was dropped, but freed) 2160 * 2161 * dev_forward_skb can be used for injecting an skb from the 2162 * start_xmit function of one device into the receive queue 2163 * of another device. 2164 * 2165 * The receiving device may be in another namespace, so 2166 * we have to clear all information in the skb that could 2167 * impact namespace isolation. 2168 */ 2169 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2170 { 2171 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2172 } 2173 EXPORT_SYMBOL_GPL(dev_forward_skb); 2174 2175 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2176 { 2177 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2178 } 2179 2180 static inline int deliver_skb(struct sk_buff *skb, 2181 struct packet_type *pt_prev, 2182 struct net_device *orig_dev) 2183 { 2184 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2185 return -ENOMEM; 2186 refcount_inc(&skb->users); 2187 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2188 } 2189 2190 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2191 struct packet_type **pt, 2192 struct net_device *orig_dev, 2193 __be16 type, 2194 struct list_head *ptype_list) 2195 { 2196 struct packet_type *ptype, *pt_prev = *pt; 2197 2198 list_for_each_entry_rcu(ptype, ptype_list, list) { 2199 if (ptype->type != type) 2200 continue; 2201 if (pt_prev) 2202 deliver_skb(skb, pt_prev, orig_dev); 2203 pt_prev = ptype; 2204 } 2205 *pt = pt_prev; 2206 } 2207 2208 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2209 { 2210 if (!ptype->af_packet_priv || !skb->sk) 2211 return false; 2212 2213 if (ptype->id_match) 2214 return ptype->id_match(ptype, skb->sk); 2215 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2216 return true; 2217 2218 return false; 2219 } 2220 2221 /** 2222 * dev_nit_active - return true if any network interface taps are in use 2223 * 2224 * @dev: network device to check for the presence of taps 2225 */ 2226 bool dev_nit_active(struct net_device *dev) 2227 { 2228 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2229 } 2230 EXPORT_SYMBOL_GPL(dev_nit_active); 2231 2232 /* 2233 * Support routine. Sends outgoing frames to any network 2234 * taps currently in use. 2235 */ 2236 2237 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2238 { 2239 struct packet_type *ptype; 2240 struct sk_buff *skb2 = NULL; 2241 struct packet_type *pt_prev = NULL; 2242 struct list_head *ptype_list = &ptype_all; 2243 2244 rcu_read_lock(); 2245 again: 2246 list_for_each_entry_rcu(ptype, ptype_list, list) { 2247 if (ptype->ignore_outgoing) 2248 continue; 2249 2250 /* Never send packets back to the socket 2251 * they originated from - MvS (miquels@drinkel.ow.org) 2252 */ 2253 if (skb_loop_sk(ptype, skb)) 2254 continue; 2255 2256 if (pt_prev) { 2257 deliver_skb(skb2, pt_prev, skb->dev); 2258 pt_prev = ptype; 2259 continue; 2260 } 2261 2262 /* need to clone skb, done only once */ 2263 skb2 = skb_clone(skb, GFP_ATOMIC); 2264 if (!skb2) 2265 goto out_unlock; 2266 2267 net_timestamp_set(skb2); 2268 2269 /* skb->nh should be correctly 2270 * set by sender, so that the second statement is 2271 * just protection against buggy protocols. 2272 */ 2273 skb_reset_mac_header(skb2); 2274 2275 if (skb_network_header(skb2) < skb2->data || 2276 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2277 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2278 ntohs(skb2->protocol), 2279 dev->name); 2280 skb_reset_network_header(skb2); 2281 } 2282 2283 skb2->transport_header = skb2->network_header; 2284 skb2->pkt_type = PACKET_OUTGOING; 2285 pt_prev = ptype; 2286 } 2287 2288 if (ptype_list == &ptype_all) { 2289 ptype_list = &dev->ptype_all; 2290 goto again; 2291 } 2292 out_unlock: 2293 if (pt_prev) { 2294 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2295 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2296 else 2297 kfree_skb(skb2); 2298 } 2299 rcu_read_unlock(); 2300 } 2301 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2302 2303 /** 2304 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2305 * @dev: Network device 2306 * @txq: number of queues available 2307 * 2308 * If real_num_tx_queues is changed the tc mappings may no longer be 2309 * valid. To resolve this verify the tc mapping remains valid and if 2310 * not NULL the mapping. With no priorities mapping to this 2311 * offset/count pair it will no longer be used. In the worst case TC0 2312 * is invalid nothing can be done so disable priority mappings. If is 2313 * expected that drivers will fix this mapping if they can before 2314 * calling netif_set_real_num_tx_queues. 2315 */ 2316 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2317 { 2318 int i; 2319 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2320 2321 /* If TC0 is invalidated disable TC mapping */ 2322 if (tc->offset + tc->count > txq) { 2323 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2324 dev->num_tc = 0; 2325 return; 2326 } 2327 2328 /* Invalidated prio to tc mappings set to TC0 */ 2329 for (i = 1; i < TC_BITMASK + 1; i++) { 2330 int q = netdev_get_prio_tc_map(dev, i); 2331 2332 tc = &dev->tc_to_txq[q]; 2333 if (tc->offset + tc->count > txq) { 2334 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", 2335 i, q); 2336 netdev_set_prio_tc_map(dev, i, 0); 2337 } 2338 } 2339 } 2340 2341 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2342 { 2343 if (dev->num_tc) { 2344 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2345 int i; 2346 2347 /* walk through the TCs and see if it falls into any of them */ 2348 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2349 if ((txq - tc->offset) < tc->count) 2350 return i; 2351 } 2352 2353 /* didn't find it, just return -1 to indicate no match */ 2354 return -1; 2355 } 2356 2357 return 0; 2358 } 2359 EXPORT_SYMBOL(netdev_txq_to_tc); 2360 2361 #ifdef CONFIG_XPS 2362 static struct static_key xps_needed __read_mostly; 2363 static struct static_key xps_rxqs_needed __read_mostly; 2364 static DEFINE_MUTEX(xps_map_mutex); 2365 #define xmap_dereference(P) \ 2366 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2367 2368 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2369 struct xps_dev_maps *old_maps, int tci, u16 index) 2370 { 2371 struct xps_map *map = NULL; 2372 int pos; 2373 2374 if (dev_maps) 2375 map = xmap_dereference(dev_maps->attr_map[tci]); 2376 if (!map) 2377 return false; 2378 2379 for (pos = map->len; pos--;) { 2380 if (map->queues[pos] != index) 2381 continue; 2382 2383 if (map->len > 1) { 2384 map->queues[pos] = map->queues[--map->len]; 2385 break; 2386 } 2387 2388 if (old_maps) 2389 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2390 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2391 kfree_rcu(map, rcu); 2392 return false; 2393 } 2394 2395 return true; 2396 } 2397 2398 static bool remove_xps_queue_cpu(struct net_device *dev, 2399 struct xps_dev_maps *dev_maps, 2400 int cpu, u16 offset, u16 count) 2401 { 2402 int num_tc = dev_maps->num_tc; 2403 bool active = false; 2404 int tci; 2405 2406 for (tci = cpu * num_tc; num_tc--; tci++) { 2407 int i, j; 2408 2409 for (i = count, j = offset; i--; j++) { 2410 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2411 break; 2412 } 2413 2414 active |= i < 0; 2415 } 2416 2417 return active; 2418 } 2419 2420 static void reset_xps_maps(struct net_device *dev, 2421 struct xps_dev_maps *dev_maps, 2422 enum xps_map_type type) 2423 { 2424 static_key_slow_dec_cpuslocked(&xps_needed); 2425 if (type == XPS_RXQS) 2426 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2427 2428 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2429 2430 kfree_rcu(dev_maps, rcu); 2431 } 2432 2433 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2434 u16 offset, u16 count) 2435 { 2436 struct xps_dev_maps *dev_maps; 2437 bool active = false; 2438 int i, j; 2439 2440 dev_maps = xmap_dereference(dev->xps_maps[type]); 2441 if (!dev_maps) 2442 return; 2443 2444 for (j = 0; j < dev_maps->nr_ids; j++) 2445 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2446 if (!active) 2447 reset_xps_maps(dev, dev_maps, type); 2448 2449 if (type == XPS_CPUS) { 2450 for (i = offset + (count - 1); count--; i--) 2451 netdev_queue_numa_node_write( 2452 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2453 } 2454 } 2455 2456 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2457 u16 count) 2458 { 2459 if (!static_key_false(&xps_needed)) 2460 return; 2461 2462 cpus_read_lock(); 2463 mutex_lock(&xps_map_mutex); 2464 2465 if (static_key_false(&xps_rxqs_needed)) 2466 clean_xps_maps(dev, XPS_RXQS, offset, count); 2467 2468 clean_xps_maps(dev, XPS_CPUS, offset, count); 2469 2470 mutex_unlock(&xps_map_mutex); 2471 cpus_read_unlock(); 2472 } 2473 2474 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2475 { 2476 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2477 } 2478 2479 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2480 u16 index, bool is_rxqs_map) 2481 { 2482 struct xps_map *new_map; 2483 int alloc_len = XPS_MIN_MAP_ALLOC; 2484 int i, pos; 2485 2486 for (pos = 0; map && pos < map->len; pos++) { 2487 if (map->queues[pos] != index) 2488 continue; 2489 return map; 2490 } 2491 2492 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2493 if (map) { 2494 if (pos < map->alloc_len) 2495 return map; 2496 2497 alloc_len = map->alloc_len * 2; 2498 } 2499 2500 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2501 * map 2502 */ 2503 if (is_rxqs_map) 2504 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2505 else 2506 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2507 cpu_to_node(attr_index)); 2508 if (!new_map) 2509 return NULL; 2510 2511 for (i = 0; i < pos; i++) 2512 new_map->queues[i] = map->queues[i]; 2513 new_map->alloc_len = alloc_len; 2514 new_map->len = pos; 2515 2516 return new_map; 2517 } 2518 2519 /* Copy xps maps at a given index */ 2520 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2521 struct xps_dev_maps *new_dev_maps, int index, 2522 int tc, bool skip_tc) 2523 { 2524 int i, tci = index * dev_maps->num_tc; 2525 struct xps_map *map; 2526 2527 /* copy maps belonging to foreign traffic classes */ 2528 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2529 if (i == tc && skip_tc) 2530 continue; 2531 2532 /* fill in the new device map from the old device map */ 2533 map = xmap_dereference(dev_maps->attr_map[tci]); 2534 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2535 } 2536 } 2537 2538 /* Must be called under cpus_read_lock */ 2539 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2540 u16 index, enum xps_map_type type) 2541 { 2542 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2543 const unsigned long *online_mask = NULL; 2544 bool active = false, copy = false; 2545 int i, j, tci, numa_node_id = -2; 2546 int maps_sz, num_tc = 1, tc = 0; 2547 struct xps_map *map, *new_map; 2548 unsigned int nr_ids; 2549 2550 if (dev->num_tc) { 2551 /* Do not allow XPS on subordinate device directly */ 2552 num_tc = dev->num_tc; 2553 if (num_tc < 0) 2554 return -EINVAL; 2555 2556 /* If queue belongs to subordinate dev use its map */ 2557 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2558 2559 tc = netdev_txq_to_tc(dev, index); 2560 if (tc < 0) 2561 return -EINVAL; 2562 } 2563 2564 mutex_lock(&xps_map_mutex); 2565 2566 dev_maps = xmap_dereference(dev->xps_maps[type]); 2567 if (type == XPS_RXQS) { 2568 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2569 nr_ids = dev->num_rx_queues; 2570 } else { 2571 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2572 if (num_possible_cpus() > 1) 2573 online_mask = cpumask_bits(cpu_online_mask); 2574 nr_ids = nr_cpu_ids; 2575 } 2576 2577 if (maps_sz < L1_CACHE_BYTES) 2578 maps_sz = L1_CACHE_BYTES; 2579 2580 /* The old dev_maps could be larger or smaller than the one we're 2581 * setting up now, as dev->num_tc or nr_ids could have been updated in 2582 * between. We could try to be smart, but let's be safe instead and only 2583 * copy foreign traffic classes if the two map sizes match. 2584 */ 2585 if (dev_maps && 2586 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2587 copy = true; 2588 2589 /* allocate memory for queue storage */ 2590 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2591 j < nr_ids;) { 2592 if (!new_dev_maps) { 2593 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2594 if (!new_dev_maps) { 2595 mutex_unlock(&xps_map_mutex); 2596 return -ENOMEM; 2597 } 2598 2599 new_dev_maps->nr_ids = nr_ids; 2600 new_dev_maps->num_tc = num_tc; 2601 } 2602 2603 tci = j * num_tc + tc; 2604 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2605 2606 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2607 if (!map) 2608 goto error; 2609 2610 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2611 } 2612 2613 if (!new_dev_maps) 2614 goto out_no_new_maps; 2615 2616 if (!dev_maps) { 2617 /* Increment static keys at most once per type */ 2618 static_key_slow_inc_cpuslocked(&xps_needed); 2619 if (type == XPS_RXQS) 2620 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2621 } 2622 2623 for (j = 0; j < nr_ids; j++) { 2624 bool skip_tc = false; 2625 2626 tci = j * num_tc + tc; 2627 if (netif_attr_test_mask(j, mask, nr_ids) && 2628 netif_attr_test_online(j, online_mask, nr_ids)) { 2629 /* add tx-queue to CPU/rx-queue maps */ 2630 int pos = 0; 2631 2632 skip_tc = true; 2633 2634 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2635 while ((pos < map->len) && (map->queues[pos] != index)) 2636 pos++; 2637 2638 if (pos == map->len) 2639 map->queues[map->len++] = index; 2640 #ifdef CONFIG_NUMA 2641 if (type == XPS_CPUS) { 2642 if (numa_node_id == -2) 2643 numa_node_id = cpu_to_node(j); 2644 else if (numa_node_id != cpu_to_node(j)) 2645 numa_node_id = -1; 2646 } 2647 #endif 2648 } 2649 2650 if (copy) 2651 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2652 skip_tc); 2653 } 2654 2655 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2656 2657 /* Cleanup old maps */ 2658 if (!dev_maps) 2659 goto out_no_old_maps; 2660 2661 for (j = 0; j < dev_maps->nr_ids; j++) { 2662 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2663 map = xmap_dereference(dev_maps->attr_map[tci]); 2664 if (!map) 2665 continue; 2666 2667 if (copy) { 2668 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2669 if (map == new_map) 2670 continue; 2671 } 2672 2673 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2674 kfree_rcu(map, rcu); 2675 } 2676 } 2677 2678 old_dev_maps = dev_maps; 2679 2680 out_no_old_maps: 2681 dev_maps = new_dev_maps; 2682 active = true; 2683 2684 out_no_new_maps: 2685 if (type == XPS_CPUS) 2686 /* update Tx queue numa node */ 2687 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2688 (numa_node_id >= 0) ? 2689 numa_node_id : NUMA_NO_NODE); 2690 2691 if (!dev_maps) 2692 goto out_no_maps; 2693 2694 /* removes tx-queue from unused CPUs/rx-queues */ 2695 for (j = 0; j < dev_maps->nr_ids; j++) { 2696 tci = j * dev_maps->num_tc; 2697 2698 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2699 if (i == tc && 2700 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2701 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2702 continue; 2703 2704 active |= remove_xps_queue(dev_maps, 2705 copy ? old_dev_maps : NULL, 2706 tci, index); 2707 } 2708 } 2709 2710 if (old_dev_maps) 2711 kfree_rcu(old_dev_maps, rcu); 2712 2713 /* free map if not active */ 2714 if (!active) 2715 reset_xps_maps(dev, dev_maps, type); 2716 2717 out_no_maps: 2718 mutex_unlock(&xps_map_mutex); 2719 2720 return 0; 2721 error: 2722 /* remove any maps that we added */ 2723 for (j = 0; j < nr_ids; j++) { 2724 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2725 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2726 map = copy ? 2727 xmap_dereference(dev_maps->attr_map[tci]) : 2728 NULL; 2729 if (new_map && new_map != map) 2730 kfree(new_map); 2731 } 2732 } 2733 2734 mutex_unlock(&xps_map_mutex); 2735 2736 kfree(new_dev_maps); 2737 return -ENOMEM; 2738 } 2739 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2740 2741 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2742 u16 index) 2743 { 2744 int ret; 2745 2746 cpus_read_lock(); 2747 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2748 cpus_read_unlock(); 2749 2750 return ret; 2751 } 2752 EXPORT_SYMBOL(netif_set_xps_queue); 2753 2754 #endif 2755 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2756 { 2757 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2758 2759 /* Unbind any subordinate channels */ 2760 while (txq-- != &dev->_tx[0]) { 2761 if (txq->sb_dev) 2762 netdev_unbind_sb_channel(dev, txq->sb_dev); 2763 } 2764 } 2765 2766 void netdev_reset_tc(struct net_device *dev) 2767 { 2768 #ifdef CONFIG_XPS 2769 netif_reset_xps_queues_gt(dev, 0); 2770 #endif 2771 netdev_unbind_all_sb_channels(dev); 2772 2773 /* Reset TC configuration of device */ 2774 dev->num_tc = 0; 2775 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2776 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2777 } 2778 EXPORT_SYMBOL(netdev_reset_tc); 2779 2780 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2781 { 2782 if (tc >= dev->num_tc) 2783 return -EINVAL; 2784 2785 #ifdef CONFIG_XPS 2786 netif_reset_xps_queues(dev, offset, count); 2787 #endif 2788 dev->tc_to_txq[tc].count = count; 2789 dev->tc_to_txq[tc].offset = offset; 2790 return 0; 2791 } 2792 EXPORT_SYMBOL(netdev_set_tc_queue); 2793 2794 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2795 { 2796 if (num_tc > TC_MAX_QUEUE) 2797 return -EINVAL; 2798 2799 #ifdef CONFIG_XPS 2800 netif_reset_xps_queues_gt(dev, 0); 2801 #endif 2802 netdev_unbind_all_sb_channels(dev); 2803 2804 dev->num_tc = num_tc; 2805 return 0; 2806 } 2807 EXPORT_SYMBOL(netdev_set_num_tc); 2808 2809 void netdev_unbind_sb_channel(struct net_device *dev, 2810 struct net_device *sb_dev) 2811 { 2812 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2813 2814 #ifdef CONFIG_XPS 2815 netif_reset_xps_queues_gt(sb_dev, 0); 2816 #endif 2817 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2818 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2819 2820 while (txq-- != &dev->_tx[0]) { 2821 if (txq->sb_dev == sb_dev) 2822 txq->sb_dev = NULL; 2823 } 2824 } 2825 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2826 2827 int netdev_bind_sb_channel_queue(struct net_device *dev, 2828 struct net_device *sb_dev, 2829 u8 tc, u16 count, u16 offset) 2830 { 2831 /* Make certain the sb_dev and dev are already configured */ 2832 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2833 return -EINVAL; 2834 2835 /* We cannot hand out queues we don't have */ 2836 if ((offset + count) > dev->real_num_tx_queues) 2837 return -EINVAL; 2838 2839 /* Record the mapping */ 2840 sb_dev->tc_to_txq[tc].count = count; 2841 sb_dev->tc_to_txq[tc].offset = offset; 2842 2843 /* Provide a way for Tx queue to find the tc_to_txq map or 2844 * XPS map for itself. 2845 */ 2846 while (count--) 2847 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2848 2849 return 0; 2850 } 2851 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2852 2853 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2854 { 2855 /* Do not use a multiqueue device to represent a subordinate channel */ 2856 if (netif_is_multiqueue(dev)) 2857 return -ENODEV; 2858 2859 /* We allow channels 1 - 32767 to be used for subordinate channels. 2860 * Channel 0 is meant to be "native" mode and used only to represent 2861 * the main root device. We allow writing 0 to reset the device back 2862 * to normal mode after being used as a subordinate channel. 2863 */ 2864 if (channel > S16_MAX) 2865 return -EINVAL; 2866 2867 dev->num_tc = -channel; 2868 2869 return 0; 2870 } 2871 EXPORT_SYMBOL(netdev_set_sb_channel); 2872 2873 /* 2874 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2875 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2876 */ 2877 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2878 { 2879 bool disabling; 2880 int rc; 2881 2882 disabling = txq < dev->real_num_tx_queues; 2883 2884 if (txq < 1 || txq > dev->num_tx_queues) 2885 return -EINVAL; 2886 2887 if (dev->reg_state == NETREG_REGISTERED || 2888 dev->reg_state == NETREG_UNREGISTERING) { 2889 ASSERT_RTNL(); 2890 2891 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2892 txq); 2893 if (rc) 2894 return rc; 2895 2896 if (dev->num_tc) 2897 netif_setup_tc(dev, txq); 2898 2899 dev_qdisc_change_real_num_tx(dev, txq); 2900 2901 dev->real_num_tx_queues = txq; 2902 2903 if (disabling) { 2904 synchronize_net(); 2905 qdisc_reset_all_tx_gt(dev, txq); 2906 #ifdef CONFIG_XPS 2907 netif_reset_xps_queues_gt(dev, txq); 2908 #endif 2909 } 2910 } else { 2911 dev->real_num_tx_queues = txq; 2912 } 2913 2914 return 0; 2915 } 2916 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2917 2918 #ifdef CONFIG_SYSFS 2919 /** 2920 * netif_set_real_num_rx_queues - set actual number of RX queues used 2921 * @dev: Network device 2922 * @rxq: Actual number of RX queues 2923 * 2924 * This must be called either with the rtnl_lock held or before 2925 * registration of the net device. Returns 0 on success, or a 2926 * negative error code. If called before registration, it always 2927 * succeeds. 2928 */ 2929 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2930 { 2931 int rc; 2932 2933 if (rxq < 1 || rxq > dev->num_rx_queues) 2934 return -EINVAL; 2935 2936 if (dev->reg_state == NETREG_REGISTERED) { 2937 ASSERT_RTNL(); 2938 2939 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2940 rxq); 2941 if (rc) 2942 return rc; 2943 } 2944 2945 dev->real_num_rx_queues = rxq; 2946 return 0; 2947 } 2948 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2949 #endif 2950 2951 /** 2952 * netif_set_real_num_queues - set actual number of RX and TX queues used 2953 * @dev: Network device 2954 * @txq: Actual number of TX queues 2955 * @rxq: Actual number of RX queues 2956 * 2957 * Set the real number of both TX and RX queues. 2958 * Does nothing if the number of queues is already correct. 2959 */ 2960 int netif_set_real_num_queues(struct net_device *dev, 2961 unsigned int txq, unsigned int rxq) 2962 { 2963 unsigned int old_rxq = dev->real_num_rx_queues; 2964 int err; 2965 2966 if (txq < 1 || txq > dev->num_tx_queues || 2967 rxq < 1 || rxq > dev->num_rx_queues) 2968 return -EINVAL; 2969 2970 /* Start from increases, so the error path only does decreases - 2971 * decreases can't fail. 2972 */ 2973 if (rxq > dev->real_num_rx_queues) { 2974 err = netif_set_real_num_rx_queues(dev, rxq); 2975 if (err) 2976 return err; 2977 } 2978 if (txq > dev->real_num_tx_queues) { 2979 err = netif_set_real_num_tx_queues(dev, txq); 2980 if (err) 2981 goto undo_rx; 2982 } 2983 if (rxq < dev->real_num_rx_queues) 2984 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2985 if (txq < dev->real_num_tx_queues) 2986 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2987 2988 return 0; 2989 undo_rx: 2990 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2991 return err; 2992 } 2993 EXPORT_SYMBOL(netif_set_real_num_queues); 2994 2995 /** 2996 * netif_set_tso_max_size() - set the max size of TSO frames supported 2997 * @dev: netdev to update 2998 * @size: max skb->len of a TSO frame 2999 * 3000 * Set the limit on the size of TSO super-frames the device can handle. 3001 * Unless explicitly set the stack will assume the value of 3002 * %GSO_LEGACY_MAX_SIZE. 3003 */ 3004 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3005 { 3006 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3007 if (size < READ_ONCE(dev->gso_max_size)) 3008 netif_set_gso_max_size(dev, size); 3009 } 3010 EXPORT_SYMBOL(netif_set_tso_max_size); 3011 3012 /** 3013 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3014 * @dev: netdev to update 3015 * @segs: max number of TCP segments 3016 * 3017 * Set the limit on the number of TCP segments the device can generate from 3018 * a single TSO super-frame. 3019 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3020 */ 3021 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3022 { 3023 dev->tso_max_segs = segs; 3024 if (segs < READ_ONCE(dev->gso_max_segs)) 3025 netif_set_gso_max_segs(dev, segs); 3026 } 3027 EXPORT_SYMBOL(netif_set_tso_max_segs); 3028 3029 /** 3030 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3031 * @to: netdev to update 3032 * @from: netdev from which to copy the limits 3033 */ 3034 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3035 { 3036 netif_set_tso_max_size(to, from->tso_max_size); 3037 netif_set_tso_max_segs(to, from->tso_max_segs); 3038 } 3039 EXPORT_SYMBOL(netif_inherit_tso_max); 3040 3041 /** 3042 * netif_get_num_default_rss_queues - default number of RSS queues 3043 * 3044 * Default value is the number of physical cores if there are only 1 or 2, or 3045 * divided by 2 if there are more. 3046 */ 3047 int netif_get_num_default_rss_queues(void) 3048 { 3049 cpumask_var_t cpus; 3050 int cpu, count = 0; 3051 3052 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3053 return 1; 3054 3055 cpumask_copy(cpus, cpu_online_mask); 3056 for_each_cpu(cpu, cpus) { 3057 ++count; 3058 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3059 } 3060 free_cpumask_var(cpus); 3061 3062 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3063 } 3064 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3065 3066 static void __netif_reschedule(struct Qdisc *q) 3067 { 3068 struct softnet_data *sd; 3069 unsigned long flags; 3070 3071 local_irq_save(flags); 3072 sd = this_cpu_ptr(&softnet_data); 3073 q->next_sched = NULL; 3074 *sd->output_queue_tailp = q; 3075 sd->output_queue_tailp = &q->next_sched; 3076 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3077 local_irq_restore(flags); 3078 } 3079 3080 void __netif_schedule(struct Qdisc *q) 3081 { 3082 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3083 __netif_reschedule(q); 3084 } 3085 EXPORT_SYMBOL(__netif_schedule); 3086 3087 struct dev_kfree_skb_cb { 3088 enum skb_free_reason reason; 3089 }; 3090 3091 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3092 { 3093 return (struct dev_kfree_skb_cb *)skb->cb; 3094 } 3095 3096 void netif_schedule_queue(struct netdev_queue *txq) 3097 { 3098 rcu_read_lock(); 3099 if (!netif_xmit_stopped(txq)) { 3100 struct Qdisc *q = rcu_dereference(txq->qdisc); 3101 3102 __netif_schedule(q); 3103 } 3104 rcu_read_unlock(); 3105 } 3106 EXPORT_SYMBOL(netif_schedule_queue); 3107 3108 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3109 { 3110 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3111 struct Qdisc *q; 3112 3113 rcu_read_lock(); 3114 q = rcu_dereference(dev_queue->qdisc); 3115 __netif_schedule(q); 3116 rcu_read_unlock(); 3117 } 3118 } 3119 EXPORT_SYMBOL(netif_tx_wake_queue); 3120 3121 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3122 { 3123 unsigned long flags; 3124 3125 if (unlikely(!skb)) 3126 return; 3127 3128 if (likely(refcount_read(&skb->users) == 1)) { 3129 smp_rmb(); 3130 refcount_set(&skb->users, 0); 3131 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3132 return; 3133 } 3134 get_kfree_skb_cb(skb)->reason = reason; 3135 local_irq_save(flags); 3136 skb->next = __this_cpu_read(softnet_data.completion_queue); 3137 __this_cpu_write(softnet_data.completion_queue, skb); 3138 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3139 local_irq_restore(flags); 3140 } 3141 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3142 3143 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3144 { 3145 if (in_hardirq() || irqs_disabled()) 3146 __dev_kfree_skb_irq(skb, reason); 3147 else 3148 dev_kfree_skb(skb); 3149 } 3150 EXPORT_SYMBOL(__dev_kfree_skb_any); 3151 3152 3153 /** 3154 * netif_device_detach - mark device as removed 3155 * @dev: network device 3156 * 3157 * Mark device as removed from system and therefore no longer available. 3158 */ 3159 void netif_device_detach(struct net_device *dev) 3160 { 3161 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3162 netif_running(dev)) { 3163 netif_tx_stop_all_queues(dev); 3164 } 3165 } 3166 EXPORT_SYMBOL(netif_device_detach); 3167 3168 /** 3169 * netif_device_attach - mark device as attached 3170 * @dev: network device 3171 * 3172 * Mark device as attached from system and restart if needed. 3173 */ 3174 void netif_device_attach(struct net_device *dev) 3175 { 3176 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3177 netif_running(dev)) { 3178 netif_tx_wake_all_queues(dev); 3179 __netdev_watchdog_up(dev); 3180 } 3181 } 3182 EXPORT_SYMBOL(netif_device_attach); 3183 3184 /* 3185 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3186 * to be used as a distribution range. 3187 */ 3188 static u16 skb_tx_hash(const struct net_device *dev, 3189 const struct net_device *sb_dev, 3190 struct sk_buff *skb) 3191 { 3192 u32 hash; 3193 u16 qoffset = 0; 3194 u16 qcount = dev->real_num_tx_queues; 3195 3196 if (dev->num_tc) { 3197 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3198 3199 qoffset = sb_dev->tc_to_txq[tc].offset; 3200 qcount = sb_dev->tc_to_txq[tc].count; 3201 if (unlikely(!qcount)) { 3202 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3203 sb_dev->name, qoffset, tc); 3204 qoffset = 0; 3205 qcount = dev->real_num_tx_queues; 3206 } 3207 } 3208 3209 if (skb_rx_queue_recorded(skb)) { 3210 hash = skb_get_rx_queue(skb); 3211 if (hash >= qoffset) 3212 hash -= qoffset; 3213 while (unlikely(hash >= qcount)) 3214 hash -= qcount; 3215 return hash + qoffset; 3216 } 3217 3218 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3219 } 3220 3221 static void skb_warn_bad_offload(const struct sk_buff *skb) 3222 { 3223 static const netdev_features_t null_features; 3224 struct net_device *dev = skb->dev; 3225 const char *name = ""; 3226 3227 if (!net_ratelimit()) 3228 return; 3229 3230 if (dev) { 3231 if (dev->dev.parent) 3232 name = dev_driver_string(dev->dev.parent); 3233 else 3234 name = netdev_name(dev); 3235 } 3236 skb_dump(KERN_WARNING, skb, false); 3237 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3238 name, dev ? &dev->features : &null_features, 3239 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3240 } 3241 3242 /* 3243 * Invalidate hardware checksum when packet is to be mangled, and 3244 * complete checksum manually on outgoing path. 3245 */ 3246 int skb_checksum_help(struct sk_buff *skb) 3247 { 3248 __wsum csum; 3249 int ret = 0, offset; 3250 3251 if (skb->ip_summed == CHECKSUM_COMPLETE) 3252 goto out_set_summed; 3253 3254 if (unlikely(skb_is_gso(skb))) { 3255 skb_warn_bad_offload(skb); 3256 return -EINVAL; 3257 } 3258 3259 /* Before computing a checksum, we should make sure no frag could 3260 * be modified by an external entity : checksum could be wrong. 3261 */ 3262 if (skb_has_shared_frag(skb)) { 3263 ret = __skb_linearize(skb); 3264 if (ret) 3265 goto out; 3266 } 3267 3268 offset = skb_checksum_start_offset(skb); 3269 ret = -EINVAL; 3270 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3271 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3272 goto out; 3273 } 3274 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3275 3276 offset += skb->csum_offset; 3277 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3278 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3279 goto out; 3280 } 3281 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3282 if (ret) 3283 goto out; 3284 3285 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3286 out_set_summed: 3287 skb->ip_summed = CHECKSUM_NONE; 3288 out: 3289 return ret; 3290 } 3291 EXPORT_SYMBOL(skb_checksum_help); 3292 3293 int skb_crc32c_csum_help(struct sk_buff *skb) 3294 { 3295 __le32 crc32c_csum; 3296 int ret = 0, offset, start; 3297 3298 if (skb->ip_summed != CHECKSUM_PARTIAL) 3299 goto out; 3300 3301 if (unlikely(skb_is_gso(skb))) 3302 goto out; 3303 3304 /* Before computing a checksum, we should make sure no frag could 3305 * be modified by an external entity : checksum could be wrong. 3306 */ 3307 if (unlikely(skb_has_shared_frag(skb))) { 3308 ret = __skb_linearize(skb); 3309 if (ret) 3310 goto out; 3311 } 3312 start = skb_checksum_start_offset(skb); 3313 offset = start + offsetof(struct sctphdr, checksum); 3314 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3315 ret = -EINVAL; 3316 goto out; 3317 } 3318 3319 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3320 if (ret) 3321 goto out; 3322 3323 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3324 skb->len - start, ~(__u32)0, 3325 crc32c_csum_stub)); 3326 *(__le32 *)(skb->data + offset) = crc32c_csum; 3327 skb->ip_summed = CHECKSUM_NONE; 3328 skb->csum_not_inet = 0; 3329 out: 3330 return ret; 3331 } 3332 3333 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3334 { 3335 __be16 type = skb->protocol; 3336 3337 /* Tunnel gso handlers can set protocol to ethernet. */ 3338 if (type == htons(ETH_P_TEB)) { 3339 struct ethhdr *eth; 3340 3341 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3342 return 0; 3343 3344 eth = (struct ethhdr *)skb->data; 3345 type = eth->h_proto; 3346 } 3347 3348 return __vlan_get_protocol(skb, type, depth); 3349 } 3350 3351 /* openvswitch calls this on rx path, so we need a different check. 3352 */ 3353 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3354 { 3355 if (tx_path) 3356 return skb->ip_summed != CHECKSUM_PARTIAL && 3357 skb->ip_summed != CHECKSUM_UNNECESSARY; 3358 3359 return skb->ip_summed == CHECKSUM_NONE; 3360 } 3361 3362 /** 3363 * __skb_gso_segment - Perform segmentation on skb. 3364 * @skb: buffer to segment 3365 * @features: features for the output path (see dev->features) 3366 * @tx_path: whether it is called in TX path 3367 * 3368 * This function segments the given skb and returns a list of segments. 3369 * 3370 * It may return NULL if the skb requires no segmentation. This is 3371 * only possible when GSO is used for verifying header integrity. 3372 * 3373 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3374 */ 3375 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3376 netdev_features_t features, bool tx_path) 3377 { 3378 struct sk_buff *segs; 3379 3380 if (unlikely(skb_needs_check(skb, tx_path))) { 3381 int err; 3382 3383 /* We're going to init ->check field in TCP or UDP header */ 3384 err = skb_cow_head(skb, 0); 3385 if (err < 0) 3386 return ERR_PTR(err); 3387 } 3388 3389 /* Only report GSO partial support if it will enable us to 3390 * support segmentation on this frame without needing additional 3391 * work. 3392 */ 3393 if (features & NETIF_F_GSO_PARTIAL) { 3394 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3395 struct net_device *dev = skb->dev; 3396 3397 partial_features |= dev->features & dev->gso_partial_features; 3398 if (!skb_gso_ok(skb, features | partial_features)) 3399 features &= ~NETIF_F_GSO_PARTIAL; 3400 } 3401 3402 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3403 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3404 3405 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3406 SKB_GSO_CB(skb)->encap_level = 0; 3407 3408 skb_reset_mac_header(skb); 3409 skb_reset_mac_len(skb); 3410 3411 segs = skb_mac_gso_segment(skb, features); 3412 3413 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3414 skb_warn_bad_offload(skb); 3415 3416 return segs; 3417 } 3418 EXPORT_SYMBOL(__skb_gso_segment); 3419 3420 /* Take action when hardware reception checksum errors are detected. */ 3421 #ifdef CONFIG_BUG 3422 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3423 { 3424 netdev_err(dev, "hw csum failure\n"); 3425 skb_dump(KERN_ERR, skb, true); 3426 dump_stack(); 3427 } 3428 3429 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3430 { 3431 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3432 } 3433 EXPORT_SYMBOL(netdev_rx_csum_fault); 3434 #endif 3435 3436 /* XXX: check that highmem exists at all on the given machine. */ 3437 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3438 { 3439 #ifdef CONFIG_HIGHMEM 3440 int i; 3441 3442 if (!(dev->features & NETIF_F_HIGHDMA)) { 3443 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3444 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3445 3446 if (PageHighMem(skb_frag_page(frag))) 3447 return 1; 3448 } 3449 } 3450 #endif 3451 return 0; 3452 } 3453 3454 /* If MPLS offload request, verify we are testing hardware MPLS features 3455 * instead of standard features for the netdev. 3456 */ 3457 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3458 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3459 netdev_features_t features, 3460 __be16 type) 3461 { 3462 if (eth_p_mpls(type)) 3463 features &= skb->dev->mpls_features; 3464 3465 return features; 3466 } 3467 #else 3468 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3469 netdev_features_t features, 3470 __be16 type) 3471 { 3472 return features; 3473 } 3474 #endif 3475 3476 static netdev_features_t harmonize_features(struct sk_buff *skb, 3477 netdev_features_t features) 3478 { 3479 __be16 type; 3480 3481 type = skb_network_protocol(skb, NULL); 3482 features = net_mpls_features(skb, features, type); 3483 3484 if (skb->ip_summed != CHECKSUM_NONE && 3485 !can_checksum_protocol(features, type)) { 3486 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3487 } 3488 if (illegal_highdma(skb->dev, skb)) 3489 features &= ~NETIF_F_SG; 3490 3491 return features; 3492 } 3493 3494 netdev_features_t passthru_features_check(struct sk_buff *skb, 3495 struct net_device *dev, 3496 netdev_features_t features) 3497 { 3498 return features; 3499 } 3500 EXPORT_SYMBOL(passthru_features_check); 3501 3502 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3503 struct net_device *dev, 3504 netdev_features_t features) 3505 { 3506 return vlan_features_check(skb, features); 3507 } 3508 3509 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3510 struct net_device *dev, 3511 netdev_features_t features) 3512 { 3513 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3514 3515 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3516 return features & ~NETIF_F_GSO_MASK; 3517 3518 if (!skb_shinfo(skb)->gso_type) { 3519 skb_warn_bad_offload(skb); 3520 return features & ~NETIF_F_GSO_MASK; 3521 } 3522 3523 /* Support for GSO partial features requires software 3524 * intervention before we can actually process the packets 3525 * so we need to strip support for any partial features now 3526 * and we can pull them back in after we have partially 3527 * segmented the frame. 3528 */ 3529 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3530 features &= ~dev->gso_partial_features; 3531 3532 /* Make sure to clear the IPv4 ID mangling feature if the 3533 * IPv4 header has the potential to be fragmented. 3534 */ 3535 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3536 struct iphdr *iph = skb->encapsulation ? 3537 inner_ip_hdr(skb) : ip_hdr(skb); 3538 3539 if (!(iph->frag_off & htons(IP_DF))) 3540 features &= ~NETIF_F_TSO_MANGLEID; 3541 } 3542 3543 return features; 3544 } 3545 3546 netdev_features_t netif_skb_features(struct sk_buff *skb) 3547 { 3548 struct net_device *dev = skb->dev; 3549 netdev_features_t features = dev->features; 3550 3551 if (skb_is_gso(skb)) 3552 features = gso_features_check(skb, dev, features); 3553 3554 /* If encapsulation offload request, verify we are testing 3555 * hardware encapsulation features instead of standard 3556 * features for the netdev 3557 */ 3558 if (skb->encapsulation) 3559 features &= dev->hw_enc_features; 3560 3561 if (skb_vlan_tagged(skb)) 3562 features = netdev_intersect_features(features, 3563 dev->vlan_features | 3564 NETIF_F_HW_VLAN_CTAG_TX | 3565 NETIF_F_HW_VLAN_STAG_TX); 3566 3567 if (dev->netdev_ops->ndo_features_check) 3568 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3569 features); 3570 else 3571 features &= dflt_features_check(skb, dev, features); 3572 3573 return harmonize_features(skb, features); 3574 } 3575 EXPORT_SYMBOL(netif_skb_features); 3576 3577 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3578 struct netdev_queue *txq, bool more) 3579 { 3580 unsigned int len; 3581 int rc; 3582 3583 if (dev_nit_active(dev)) 3584 dev_queue_xmit_nit(skb, dev); 3585 3586 len = skb->len; 3587 trace_net_dev_start_xmit(skb, dev); 3588 rc = netdev_start_xmit(skb, dev, txq, more); 3589 trace_net_dev_xmit(skb, rc, dev, len); 3590 3591 return rc; 3592 } 3593 3594 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3595 struct netdev_queue *txq, int *ret) 3596 { 3597 struct sk_buff *skb = first; 3598 int rc = NETDEV_TX_OK; 3599 3600 while (skb) { 3601 struct sk_buff *next = skb->next; 3602 3603 skb_mark_not_on_list(skb); 3604 rc = xmit_one(skb, dev, txq, next != NULL); 3605 if (unlikely(!dev_xmit_complete(rc))) { 3606 skb->next = next; 3607 goto out; 3608 } 3609 3610 skb = next; 3611 if (netif_tx_queue_stopped(txq) && skb) { 3612 rc = NETDEV_TX_BUSY; 3613 break; 3614 } 3615 } 3616 3617 out: 3618 *ret = rc; 3619 return skb; 3620 } 3621 3622 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3623 netdev_features_t features) 3624 { 3625 if (skb_vlan_tag_present(skb) && 3626 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3627 skb = __vlan_hwaccel_push_inside(skb); 3628 return skb; 3629 } 3630 3631 int skb_csum_hwoffload_help(struct sk_buff *skb, 3632 const netdev_features_t features) 3633 { 3634 if (unlikely(skb_csum_is_sctp(skb))) 3635 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3636 skb_crc32c_csum_help(skb); 3637 3638 if (features & NETIF_F_HW_CSUM) 3639 return 0; 3640 3641 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3642 switch (skb->csum_offset) { 3643 case offsetof(struct tcphdr, check): 3644 case offsetof(struct udphdr, check): 3645 return 0; 3646 } 3647 } 3648 3649 return skb_checksum_help(skb); 3650 } 3651 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3652 3653 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3654 { 3655 netdev_features_t features; 3656 3657 features = netif_skb_features(skb); 3658 skb = validate_xmit_vlan(skb, features); 3659 if (unlikely(!skb)) 3660 goto out_null; 3661 3662 skb = sk_validate_xmit_skb(skb, dev); 3663 if (unlikely(!skb)) 3664 goto out_null; 3665 3666 if (netif_needs_gso(skb, features)) { 3667 struct sk_buff *segs; 3668 3669 segs = skb_gso_segment(skb, features); 3670 if (IS_ERR(segs)) { 3671 goto out_kfree_skb; 3672 } else if (segs) { 3673 consume_skb(skb); 3674 skb = segs; 3675 } 3676 } else { 3677 if (skb_needs_linearize(skb, features) && 3678 __skb_linearize(skb)) 3679 goto out_kfree_skb; 3680 3681 /* If packet is not checksummed and device does not 3682 * support checksumming for this protocol, complete 3683 * checksumming here. 3684 */ 3685 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3686 if (skb->encapsulation) 3687 skb_set_inner_transport_header(skb, 3688 skb_checksum_start_offset(skb)); 3689 else 3690 skb_set_transport_header(skb, 3691 skb_checksum_start_offset(skb)); 3692 if (skb_csum_hwoffload_help(skb, features)) 3693 goto out_kfree_skb; 3694 } 3695 } 3696 3697 skb = validate_xmit_xfrm(skb, features, again); 3698 3699 return skb; 3700 3701 out_kfree_skb: 3702 kfree_skb(skb); 3703 out_null: 3704 dev_core_stats_tx_dropped_inc(dev); 3705 return NULL; 3706 } 3707 3708 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3709 { 3710 struct sk_buff *next, *head = NULL, *tail; 3711 3712 for (; skb != NULL; skb = next) { 3713 next = skb->next; 3714 skb_mark_not_on_list(skb); 3715 3716 /* in case skb wont be segmented, point to itself */ 3717 skb->prev = skb; 3718 3719 skb = validate_xmit_skb(skb, dev, again); 3720 if (!skb) 3721 continue; 3722 3723 if (!head) 3724 head = skb; 3725 else 3726 tail->next = skb; 3727 /* If skb was segmented, skb->prev points to 3728 * the last segment. If not, it still contains skb. 3729 */ 3730 tail = skb->prev; 3731 } 3732 return head; 3733 } 3734 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3735 3736 static void qdisc_pkt_len_init(struct sk_buff *skb) 3737 { 3738 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3739 3740 qdisc_skb_cb(skb)->pkt_len = skb->len; 3741 3742 /* To get more precise estimation of bytes sent on wire, 3743 * we add to pkt_len the headers size of all segments 3744 */ 3745 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3746 unsigned int hdr_len; 3747 u16 gso_segs = shinfo->gso_segs; 3748 3749 /* mac layer + network layer */ 3750 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3751 3752 /* + transport layer */ 3753 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3754 const struct tcphdr *th; 3755 struct tcphdr _tcphdr; 3756 3757 th = skb_header_pointer(skb, skb_transport_offset(skb), 3758 sizeof(_tcphdr), &_tcphdr); 3759 if (likely(th)) 3760 hdr_len += __tcp_hdrlen(th); 3761 } else { 3762 struct udphdr _udphdr; 3763 3764 if (skb_header_pointer(skb, skb_transport_offset(skb), 3765 sizeof(_udphdr), &_udphdr)) 3766 hdr_len += sizeof(struct udphdr); 3767 } 3768 3769 if (shinfo->gso_type & SKB_GSO_DODGY) 3770 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3771 shinfo->gso_size); 3772 3773 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3774 } 3775 } 3776 3777 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3778 struct sk_buff **to_free, 3779 struct netdev_queue *txq) 3780 { 3781 int rc; 3782 3783 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3784 if (rc == NET_XMIT_SUCCESS) 3785 trace_qdisc_enqueue(q, txq, skb); 3786 return rc; 3787 } 3788 3789 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3790 struct net_device *dev, 3791 struct netdev_queue *txq) 3792 { 3793 spinlock_t *root_lock = qdisc_lock(q); 3794 struct sk_buff *to_free = NULL; 3795 bool contended; 3796 int rc; 3797 3798 qdisc_calculate_pkt_len(skb, q); 3799 3800 if (q->flags & TCQ_F_NOLOCK) { 3801 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3802 qdisc_run_begin(q)) { 3803 /* Retest nolock_qdisc_is_empty() within the protection 3804 * of q->seqlock to protect from racing with requeuing. 3805 */ 3806 if (unlikely(!nolock_qdisc_is_empty(q))) { 3807 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3808 __qdisc_run(q); 3809 qdisc_run_end(q); 3810 3811 goto no_lock_out; 3812 } 3813 3814 qdisc_bstats_cpu_update(q, skb); 3815 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3816 !nolock_qdisc_is_empty(q)) 3817 __qdisc_run(q); 3818 3819 qdisc_run_end(q); 3820 return NET_XMIT_SUCCESS; 3821 } 3822 3823 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3824 qdisc_run(q); 3825 3826 no_lock_out: 3827 if (unlikely(to_free)) 3828 kfree_skb_list_reason(to_free, 3829 SKB_DROP_REASON_QDISC_DROP); 3830 return rc; 3831 } 3832 3833 /* 3834 * Heuristic to force contended enqueues to serialize on a 3835 * separate lock before trying to get qdisc main lock. 3836 * This permits qdisc->running owner to get the lock more 3837 * often and dequeue packets faster. 3838 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3839 * and then other tasks will only enqueue packets. The packets will be 3840 * sent after the qdisc owner is scheduled again. To prevent this 3841 * scenario the task always serialize on the lock. 3842 */ 3843 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3844 if (unlikely(contended)) 3845 spin_lock(&q->busylock); 3846 3847 spin_lock(root_lock); 3848 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3849 __qdisc_drop(skb, &to_free); 3850 rc = NET_XMIT_DROP; 3851 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3852 qdisc_run_begin(q)) { 3853 /* 3854 * This is a work-conserving queue; there are no old skbs 3855 * waiting to be sent out; and the qdisc is not running - 3856 * xmit the skb directly. 3857 */ 3858 3859 qdisc_bstats_update(q, skb); 3860 3861 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3862 if (unlikely(contended)) { 3863 spin_unlock(&q->busylock); 3864 contended = false; 3865 } 3866 __qdisc_run(q); 3867 } 3868 3869 qdisc_run_end(q); 3870 rc = NET_XMIT_SUCCESS; 3871 } else { 3872 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3873 if (qdisc_run_begin(q)) { 3874 if (unlikely(contended)) { 3875 spin_unlock(&q->busylock); 3876 contended = false; 3877 } 3878 __qdisc_run(q); 3879 qdisc_run_end(q); 3880 } 3881 } 3882 spin_unlock(root_lock); 3883 if (unlikely(to_free)) 3884 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3885 if (unlikely(contended)) 3886 spin_unlock(&q->busylock); 3887 return rc; 3888 } 3889 3890 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3891 static void skb_update_prio(struct sk_buff *skb) 3892 { 3893 const struct netprio_map *map; 3894 const struct sock *sk; 3895 unsigned int prioidx; 3896 3897 if (skb->priority) 3898 return; 3899 map = rcu_dereference_bh(skb->dev->priomap); 3900 if (!map) 3901 return; 3902 sk = skb_to_full_sk(skb); 3903 if (!sk) 3904 return; 3905 3906 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3907 3908 if (prioidx < map->priomap_len) 3909 skb->priority = map->priomap[prioidx]; 3910 } 3911 #else 3912 #define skb_update_prio(skb) 3913 #endif 3914 3915 /** 3916 * dev_loopback_xmit - loop back @skb 3917 * @net: network namespace this loopback is happening in 3918 * @sk: sk needed to be a netfilter okfn 3919 * @skb: buffer to transmit 3920 */ 3921 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3922 { 3923 skb_reset_mac_header(skb); 3924 __skb_pull(skb, skb_network_offset(skb)); 3925 skb->pkt_type = PACKET_LOOPBACK; 3926 if (skb->ip_summed == CHECKSUM_NONE) 3927 skb->ip_summed = CHECKSUM_UNNECESSARY; 3928 WARN_ON(!skb_dst(skb)); 3929 skb_dst_force(skb); 3930 netif_rx(skb); 3931 return 0; 3932 } 3933 EXPORT_SYMBOL(dev_loopback_xmit); 3934 3935 #ifdef CONFIG_NET_EGRESS 3936 static struct sk_buff * 3937 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3938 { 3939 #ifdef CONFIG_NET_CLS_ACT 3940 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3941 struct tcf_result cl_res; 3942 3943 if (!miniq) 3944 return skb; 3945 3946 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3947 tc_skb_cb(skb)->mru = 0; 3948 tc_skb_cb(skb)->post_ct = false; 3949 mini_qdisc_bstats_cpu_update(miniq, skb); 3950 3951 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3952 case TC_ACT_OK: 3953 case TC_ACT_RECLASSIFY: 3954 skb->tc_index = TC_H_MIN(cl_res.classid); 3955 break; 3956 case TC_ACT_SHOT: 3957 mini_qdisc_qstats_cpu_drop(miniq); 3958 *ret = NET_XMIT_DROP; 3959 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3960 return NULL; 3961 case TC_ACT_STOLEN: 3962 case TC_ACT_QUEUED: 3963 case TC_ACT_TRAP: 3964 *ret = NET_XMIT_SUCCESS; 3965 consume_skb(skb); 3966 return NULL; 3967 case TC_ACT_REDIRECT: 3968 /* No need to push/pop skb's mac_header here on egress! */ 3969 skb_do_redirect(skb); 3970 *ret = NET_XMIT_SUCCESS; 3971 return NULL; 3972 default: 3973 break; 3974 } 3975 #endif /* CONFIG_NET_CLS_ACT */ 3976 3977 return skb; 3978 } 3979 3980 static struct netdev_queue * 3981 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3982 { 3983 int qm = skb_get_queue_mapping(skb); 3984 3985 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3986 } 3987 3988 static bool netdev_xmit_txqueue_skipped(void) 3989 { 3990 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3991 } 3992 3993 void netdev_xmit_skip_txqueue(bool skip) 3994 { 3995 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3996 } 3997 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3998 #endif /* CONFIG_NET_EGRESS */ 3999 4000 #ifdef CONFIG_XPS 4001 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4002 struct xps_dev_maps *dev_maps, unsigned int tci) 4003 { 4004 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4005 struct xps_map *map; 4006 int queue_index = -1; 4007 4008 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4009 return queue_index; 4010 4011 tci *= dev_maps->num_tc; 4012 tci += tc; 4013 4014 map = rcu_dereference(dev_maps->attr_map[tci]); 4015 if (map) { 4016 if (map->len == 1) 4017 queue_index = map->queues[0]; 4018 else 4019 queue_index = map->queues[reciprocal_scale( 4020 skb_get_hash(skb), map->len)]; 4021 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4022 queue_index = -1; 4023 } 4024 return queue_index; 4025 } 4026 #endif 4027 4028 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4029 struct sk_buff *skb) 4030 { 4031 #ifdef CONFIG_XPS 4032 struct xps_dev_maps *dev_maps; 4033 struct sock *sk = skb->sk; 4034 int queue_index = -1; 4035 4036 if (!static_key_false(&xps_needed)) 4037 return -1; 4038 4039 rcu_read_lock(); 4040 if (!static_key_false(&xps_rxqs_needed)) 4041 goto get_cpus_map; 4042 4043 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4044 if (dev_maps) { 4045 int tci = sk_rx_queue_get(sk); 4046 4047 if (tci >= 0) 4048 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4049 tci); 4050 } 4051 4052 get_cpus_map: 4053 if (queue_index < 0) { 4054 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4055 if (dev_maps) { 4056 unsigned int tci = skb->sender_cpu - 1; 4057 4058 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4059 tci); 4060 } 4061 } 4062 rcu_read_unlock(); 4063 4064 return queue_index; 4065 #else 4066 return -1; 4067 #endif 4068 } 4069 4070 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4071 struct net_device *sb_dev) 4072 { 4073 return 0; 4074 } 4075 EXPORT_SYMBOL(dev_pick_tx_zero); 4076 4077 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4078 struct net_device *sb_dev) 4079 { 4080 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4081 } 4082 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4083 4084 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4085 struct net_device *sb_dev) 4086 { 4087 struct sock *sk = skb->sk; 4088 int queue_index = sk_tx_queue_get(sk); 4089 4090 sb_dev = sb_dev ? : dev; 4091 4092 if (queue_index < 0 || skb->ooo_okay || 4093 queue_index >= dev->real_num_tx_queues) { 4094 int new_index = get_xps_queue(dev, sb_dev, skb); 4095 4096 if (new_index < 0) 4097 new_index = skb_tx_hash(dev, sb_dev, skb); 4098 4099 if (queue_index != new_index && sk && 4100 sk_fullsock(sk) && 4101 rcu_access_pointer(sk->sk_dst_cache)) 4102 sk_tx_queue_set(sk, new_index); 4103 4104 queue_index = new_index; 4105 } 4106 4107 return queue_index; 4108 } 4109 EXPORT_SYMBOL(netdev_pick_tx); 4110 4111 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4112 struct sk_buff *skb, 4113 struct net_device *sb_dev) 4114 { 4115 int queue_index = 0; 4116 4117 #ifdef CONFIG_XPS 4118 u32 sender_cpu = skb->sender_cpu - 1; 4119 4120 if (sender_cpu >= (u32)NR_CPUS) 4121 skb->sender_cpu = raw_smp_processor_id() + 1; 4122 #endif 4123 4124 if (dev->real_num_tx_queues != 1) { 4125 const struct net_device_ops *ops = dev->netdev_ops; 4126 4127 if (ops->ndo_select_queue) 4128 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4129 else 4130 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4131 4132 queue_index = netdev_cap_txqueue(dev, queue_index); 4133 } 4134 4135 skb_set_queue_mapping(skb, queue_index); 4136 return netdev_get_tx_queue(dev, queue_index); 4137 } 4138 4139 /** 4140 * __dev_queue_xmit() - transmit a buffer 4141 * @skb: buffer to transmit 4142 * @sb_dev: suboordinate device used for L2 forwarding offload 4143 * 4144 * Queue a buffer for transmission to a network device. The caller must 4145 * have set the device and priority and built the buffer before calling 4146 * this function. The function can be called from an interrupt. 4147 * 4148 * When calling this method, interrupts MUST be enabled. This is because 4149 * the BH enable code must have IRQs enabled so that it will not deadlock. 4150 * 4151 * Regardless of the return value, the skb is consumed, so it is currently 4152 * difficult to retry a send to this method. (You can bump the ref count 4153 * before sending to hold a reference for retry if you are careful.) 4154 * 4155 * Return: 4156 * * 0 - buffer successfully transmitted 4157 * * positive qdisc return code - NET_XMIT_DROP etc. 4158 * * negative errno - other errors 4159 */ 4160 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4161 { 4162 struct net_device *dev = skb->dev; 4163 struct netdev_queue *txq = NULL; 4164 struct Qdisc *q; 4165 int rc = -ENOMEM; 4166 bool again = false; 4167 4168 skb_reset_mac_header(skb); 4169 4170 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4171 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4172 4173 /* Disable soft irqs for various locks below. Also 4174 * stops preemption for RCU. 4175 */ 4176 rcu_read_lock_bh(); 4177 4178 skb_update_prio(skb); 4179 4180 qdisc_pkt_len_init(skb); 4181 #ifdef CONFIG_NET_CLS_ACT 4182 skb->tc_at_ingress = 0; 4183 #endif 4184 #ifdef CONFIG_NET_EGRESS 4185 if (static_branch_unlikely(&egress_needed_key)) { 4186 if (nf_hook_egress_active()) { 4187 skb = nf_hook_egress(skb, &rc, dev); 4188 if (!skb) 4189 goto out; 4190 } 4191 4192 netdev_xmit_skip_txqueue(false); 4193 4194 nf_skip_egress(skb, true); 4195 skb = sch_handle_egress(skb, &rc, dev); 4196 if (!skb) 4197 goto out; 4198 nf_skip_egress(skb, false); 4199 4200 if (netdev_xmit_txqueue_skipped()) 4201 txq = netdev_tx_queue_mapping(dev, skb); 4202 } 4203 #endif 4204 /* If device/qdisc don't need skb->dst, release it right now while 4205 * its hot in this cpu cache. 4206 */ 4207 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4208 skb_dst_drop(skb); 4209 else 4210 skb_dst_force(skb); 4211 4212 if (!txq) 4213 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4214 4215 q = rcu_dereference_bh(txq->qdisc); 4216 4217 trace_net_dev_queue(skb); 4218 if (q->enqueue) { 4219 rc = __dev_xmit_skb(skb, q, dev, txq); 4220 goto out; 4221 } 4222 4223 /* The device has no queue. Common case for software devices: 4224 * loopback, all the sorts of tunnels... 4225 4226 * Really, it is unlikely that netif_tx_lock protection is necessary 4227 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4228 * counters.) 4229 * However, it is possible, that they rely on protection 4230 * made by us here. 4231 4232 * Check this and shot the lock. It is not prone from deadlocks. 4233 *Either shot noqueue qdisc, it is even simpler 8) 4234 */ 4235 if (dev->flags & IFF_UP) { 4236 int cpu = smp_processor_id(); /* ok because BHs are off */ 4237 4238 /* Other cpus might concurrently change txq->xmit_lock_owner 4239 * to -1 or to their cpu id, but not to our id. 4240 */ 4241 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4242 if (dev_xmit_recursion()) 4243 goto recursion_alert; 4244 4245 skb = validate_xmit_skb(skb, dev, &again); 4246 if (!skb) 4247 goto out; 4248 4249 HARD_TX_LOCK(dev, txq, cpu); 4250 4251 if (!netif_xmit_stopped(txq)) { 4252 dev_xmit_recursion_inc(); 4253 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4254 dev_xmit_recursion_dec(); 4255 if (dev_xmit_complete(rc)) { 4256 HARD_TX_UNLOCK(dev, txq); 4257 goto out; 4258 } 4259 } 4260 HARD_TX_UNLOCK(dev, txq); 4261 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4262 dev->name); 4263 } else { 4264 /* Recursion is detected! It is possible, 4265 * unfortunately 4266 */ 4267 recursion_alert: 4268 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4269 dev->name); 4270 } 4271 } 4272 4273 rc = -ENETDOWN; 4274 rcu_read_unlock_bh(); 4275 4276 dev_core_stats_tx_dropped_inc(dev); 4277 kfree_skb_list(skb); 4278 return rc; 4279 out: 4280 rcu_read_unlock_bh(); 4281 return rc; 4282 } 4283 EXPORT_SYMBOL(__dev_queue_xmit); 4284 4285 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4286 { 4287 struct net_device *dev = skb->dev; 4288 struct sk_buff *orig_skb = skb; 4289 struct netdev_queue *txq; 4290 int ret = NETDEV_TX_BUSY; 4291 bool again = false; 4292 4293 if (unlikely(!netif_running(dev) || 4294 !netif_carrier_ok(dev))) 4295 goto drop; 4296 4297 skb = validate_xmit_skb_list(skb, dev, &again); 4298 if (skb != orig_skb) 4299 goto drop; 4300 4301 skb_set_queue_mapping(skb, queue_id); 4302 txq = skb_get_tx_queue(dev, skb); 4303 4304 local_bh_disable(); 4305 4306 dev_xmit_recursion_inc(); 4307 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4308 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4309 ret = netdev_start_xmit(skb, dev, txq, false); 4310 HARD_TX_UNLOCK(dev, txq); 4311 dev_xmit_recursion_dec(); 4312 4313 local_bh_enable(); 4314 return ret; 4315 drop: 4316 dev_core_stats_tx_dropped_inc(dev); 4317 kfree_skb_list(skb); 4318 return NET_XMIT_DROP; 4319 } 4320 EXPORT_SYMBOL(__dev_direct_xmit); 4321 4322 /************************************************************************* 4323 * Receiver routines 4324 *************************************************************************/ 4325 4326 int netdev_max_backlog __read_mostly = 1000; 4327 EXPORT_SYMBOL(netdev_max_backlog); 4328 4329 int netdev_tstamp_prequeue __read_mostly = 1; 4330 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4331 int netdev_budget __read_mostly = 300; 4332 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4333 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4334 int weight_p __read_mostly = 64; /* old backlog weight */ 4335 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4336 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4337 int dev_rx_weight __read_mostly = 64; 4338 int dev_tx_weight __read_mostly = 64; 4339 4340 /* Called with irq disabled */ 4341 static inline void ____napi_schedule(struct softnet_data *sd, 4342 struct napi_struct *napi) 4343 { 4344 struct task_struct *thread; 4345 4346 lockdep_assert_irqs_disabled(); 4347 4348 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4349 /* Paired with smp_mb__before_atomic() in 4350 * napi_enable()/dev_set_threaded(). 4351 * Use READ_ONCE() to guarantee a complete 4352 * read on napi->thread. Only call 4353 * wake_up_process() when it's not NULL. 4354 */ 4355 thread = READ_ONCE(napi->thread); 4356 if (thread) { 4357 /* Avoid doing set_bit() if the thread is in 4358 * INTERRUPTIBLE state, cause napi_thread_wait() 4359 * makes sure to proceed with napi polling 4360 * if the thread is explicitly woken from here. 4361 */ 4362 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4363 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4364 wake_up_process(thread); 4365 return; 4366 } 4367 } 4368 4369 list_add_tail(&napi->poll_list, &sd->poll_list); 4370 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4371 } 4372 4373 #ifdef CONFIG_RPS 4374 4375 /* One global table that all flow-based protocols share. */ 4376 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4377 EXPORT_SYMBOL(rps_sock_flow_table); 4378 u32 rps_cpu_mask __read_mostly; 4379 EXPORT_SYMBOL(rps_cpu_mask); 4380 4381 struct static_key_false rps_needed __read_mostly; 4382 EXPORT_SYMBOL(rps_needed); 4383 struct static_key_false rfs_needed __read_mostly; 4384 EXPORT_SYMBOL(rfs_needed); 4385 4386 static struct rps_dev_flow * 4387 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4388 struct rps_dev_flow *rflow, u16 next_cpu) 4389 { 4390 if (next_cpu < nr_cpu_ids) { 4391 #ifdef CONFIG_RFS_ACCEL 4392 struct netdev_rx_queue *rxqueue; 4393 struct rps_dev_flow_table *flow_table; 4394 struct rps_dev_flow *old_rflow; 4395 u32 flow_id; 4396 u16 rxq_index; 4397 int rc; 4398 4399 /* Should we steer this flow to a different hardware queue? */ 4400 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4401 !(dev->features & NETIF_F_NTUPLE)) 4402 goto out; 4403 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4404 if (rxq_index == skb_get_rx_queue(skb)) 4405 goto out; 4406 4407 rxqueue = dev->_rx + rxq_index; 4408 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4409 if (!flow_table) 4410 goto out; 4411 flow_id = skb_get_hash(skb) & flow_table->mask; 4412 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4413 rxq_index, flow_id); 4414 if (rc < 0) 4415 goto out; 4416 old_rflow = rflow; 4417 rflow = &flow_table->flows[flow_id]; 4418 rflow->filter = rc; 4419 if (old_rflow->filter == rflow->filter) 4420 old_rflow->filter = RPS_NO_FILTER; 4421 out: 4422 #endif 4423 rflow->last_qtail = 4424 per_cpu(softnet_data, next_cpu).input_queue_head; 4425 } 4426 4427 rflow->cpu = next_cpu; 4428 return rflow; 4429 } 4430 4431 /* 4432 * get_rps_cpu is called from netif_receive_skb and returns the target 4433 * CPU from the RPS map of the receiving queue for a given skb. 4434 * rcu_read_lock must be held on entry. 4435 */ 4436 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4437 struct rps_dev_flow **rflowp) 4438 { 4439 const struct rps_sock_flow_table *sock_flow_table; 4440 struct netdev_rx_queue *rxqueue = dev->_rx; 4441 struct rps_dev_flow_table *flow_table; 4442 struct rps_map *map; 4443 int cpu = -1; 4444 u32 tcpu; 4445 u32 hash; 4446 4447 if (skb_rx_queue_recorded(skb)) { 4448 u16 index = skb_get_rx_queue(skb); 4449 4450 if (unlikely(index >= dev->real_num_rx_queues)) { 4451 WARN_ONCE(dev->real_num_rx_queues > 1, 4452 "%s received packet on queue %u, but number " 4453 "of RX queues is %u\n", 4454 dev->name, index, dev->real_num_rx_queues); 4455 goto done; 4456 } 4457 rxqueue += index; 4458 } 4459 4460 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4461 4462 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4463 map = rcu_dereference(rxqueue->rps_map); 4464 if (!flow_table && !map) 4465 goto done; 4466 4467 skb_reset_network_header(skb); 4468 hash = skb_get_hash(skb); 4469 if (!hash) 4470 goto done; 4471 4472 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4473 if (flow_table && sock_flow_table) { 4474 struct rps_dev_flow *rflow; 4475 u32 next_cpu; 4476 u32 ident; 4477 4478 /* First check into global flow table if there is a match */ 4479 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4480 if ((ident ^ hash) & ~rps_cpu_mask) 4481 goto try_rps; 4482 4483 next_cpu = ident & rps_cpu_mask; 4484 4485 /* OK, now we know there is a match, 4486 * we can look at the local (per receive queue) flow table 4487 */ 4488 rflow = &flow_table->flows[hash & flow_table->mask]; 4489 tcpu = rflow->cpu; 4490 4491 /* 4492 * If the desired CPU (where last recvmsg was done) is 4493 * different from current CPU (one in the rx-queue flow 4494 * table entry), switch if one of the following holds: 4495 * - Current CPU is unset (>= nr_cpu_ids). 4496 * - Current CPU is offline. 4497 * - The current CPU's queue tail has advanced beyond the 4498 * last packet that was enqueued using this table entry. 4499 * This guarantees that all previous packets for the flow 4500 * have been dequeued, thus preserving in order delivery. 4501 */ 4502 if (unlikely(tcpu != next_cpu) && 4503 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4504 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4505 rflow->last_qtail)) >= 0)) { 4506 tcpu = next_cpu; 4507 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4508 } 4509 4510 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4511 *rflowp = rflow; 4512 cpu = tcpu; 4513 goto done; 4514 } 4515 } 4516 4517 try_rps: 4518 4519 if (map) { 4520 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4521 if (cpu_online(tcpu)) { 4522 cpu = tcpu; 4523 goto done; 4524 } 4525 } 4526 4527 done: 4528 return cpu; 4529 } 4530 4531 #ifdef CONFIG_RFS_ACCEL 4532 4533 /** 4534 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4535 * @dev: Device on which the filter was set 4536 * @rxq_index: RX queue index 4537 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4538 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4539 * 4540 * Drivers that implement ndo_rx_flow_steer() should periodically call 4541 * this function for each installed filter and remove the filters for 4542 * which it returns %true. 4543 */ 4544 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4545 u32 flow_id, u16 filter_id) 4546 { 4547 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4548 struct rps_dev_flow_table *flow_table; 4549 struct rps_dev_flow *rflow; 4550 bool expire = true; 4551 unsigned int cpu; 4552 4553 rcu_read_lock(); 4554 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4555 if (flow_table && flow_id <= flow_table->mask) { 4556 rflow = &flow_table->flows[flow_id]; 4557 cpu = READ_ONCE(rflow->cpu); 4558 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4559 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4560 rflow->last_qtail) < 4561 (int)(10 * flow_table->mask))) 4562 expire = false; 4563 } 4564 rcu_read_unlock(); 4565 return expire; 4566 } 4567 EXPORT_SYMBOL(rps_may_expire_flow); 4568 4569 #endif /* CONFIG_RFS_ACCEL */ 4570 4571 /* Called from hardirq (IPI) context */ 4572 static void rps_trigger_softirq(void *data) 4573 { 4574 struct softnet_data *sd = data; 4575 4576 ____napi_schedule(sd, &sd->backlog); 4577 sd->received_rps++; 4578 } 4579 4580 #endif /* CONFIG_RPS */ 4581 4582 /* Called from hardirq (IPI) context */ 4583 static void trigger_rx_softirq(void *data) 4584 { 4585 struct softnet_data *sd = data; 4586 4587 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4588 smp_store_release(&sd->defer_ipi_scheduled, 0); 4589 } 4590 4591 /* 4592 * Check if this softnet_data structure is another cpu one 4593 * If yes, queue it to our IPI list and return 1 4594 * If no, return 0 4595 */ 4596 static int napi_schedule_rps(struct softnet_data *sd) 4597 { 4598 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4599 4600 #ifdef CONFIG_RPS 4601 if (sd != mysd) { 4602 sd->rps_ipi_next = mysd->rps_ipi_list; 4603 mysd->rps_ipi_list = sd; 4604 4605 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4606 return 1; 4607 } 4608 #endif /* CONFIG_RPS */ 4609 __napi_schedule_irqoff(&mysd->backlog); 4610 return 0; 4611 } 4612 4613 #ifdef CONFIG_NET_FLOW_LIMIT 4614 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4615 #endif 4616 4617 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4618 { 4619 #ifdef CONFIG_NET_FLOW_LIMIT 4620 struct sd_flow_limit *fl; 4621 struct softnet_data *sd; 4622 unsigned int old_flow, new_flow; 4623 4624 if (qlen < (netdev_max_backlog >> 1)) 4625 return false; 4626 4627 sd = this_cpu_ptr(&softnet_data); 4628 4629 rcu_read_lock(); 4630 fl = rcu_dereference(sd->flow_limit); 4631 if (fl) { 4632 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4633 old_flow = fl->history[fl->history_head]; 4634 fl->history[fl->history_head] = new_flow; 4635 4636 fl->history_head++; 4637 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4638 4639 if (likely(fl->buckets[old_flow])) 4640 fl->buckets[old_flow]--; 4641 4642 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4643 fl->count++; 4644 rcu_read_unlock(); 4645 return true; 4646 } 4647 } 4648 rcu_read_unlock(); 4649 #endif 4650 return false; 4651 } 4652 4653 /* 4654 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4655 * queue (may be a remote CPU queue). 4656 */ 4657 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4658 unsigned int *qtail) 4659 { 4660 enum skb_drop_reason reason; 4661 struct softnet_data *sd; 4662 unsigned long flags; 4663 unsigned int qlen; 4664 4665 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4666 sd = &per_cpu(softnet_data, cpu); 4667 4668 rps_lock_irqsave(sd, &flags); 4669 if (!netif_running(skb->dev)) 4670 goto drop; 4671 qlen = skb_queue_len(&sd->input_pkt_queue); 4672 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4673 if (qlen) { 4674 enqueue: 4675 __skb_queue_tail(&sd->input_pkt_queue, skb); 4676 input_queue_tail_incr_save(sd, qtail); 4677 rps_unlock_irq_restore(sd, &flags); 4678 return NET_RX_SUCCESS; 4679 } 4680 4681 /* Schedule NAPI for backlog device 4682 * We can use non atomic operation since we own the queue lock 4683 */ 4684 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4685 napi_schedule_rps(sd); 4686 goto enqueue; 4687 } 4688 reason = SKB_DROP_REASON_CPU_BACKLOG; 4689 4690 drop: 4691 sd->dropped++; 4692 rps_unlock_irq_restore(sd, &flags); 4693 4694 dev_core_stats_rx_dropped_inc(skb->dev); 4695 kfree_skb_reason(skb, reason); 4696 return NET_RX_DROP; 4697 } 4698 4699 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4700 { 4701 struct net_device *dev = skb->dev; 4702 struct netdev_rx_queue *rxqueue; 4703 4704 rxqueue = dev->_rx; 4705 4706 if (skb_rx_queue_recorded(skb)) { 4707 u16 index = skb_get_rx_queue(skb); 4708 4709 if (unlikely(index >= dev->real_num_rx_queues)) { 4710 WARN_ONCE(dev->real_num_rx_queues > 1, 4711 "%s received packet on queue %u, but number " 4712 "of RX queues is %u\n", 4713 dev->name, index, dev->real_num_rx_queues); 4714 4715 return rxqueue; /* Return first rxqueue */ 4716 } 4717 rxqueue += index; 4718 } 4719 return rxqueue; 4720 } 4721 4722 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4723 struct bpf_prog *xdp_prog) 4724 { 4725 void *orig_data, *orig_data_end, *hard_start; 4726 struct netdev_rx_queue *rxqueue; 4727 bool orig_bcast, orig_host; 4728 u32 mac_len, frame_sz; 4729 __be16 orig_eth_type; 4730 struct ethhdr *eth; 4731 u32 metalen, act; 4732 int off; 4733 4734 /* The XDP program wants to see the packet starting at the MAC 4735 * header. 4736 */ 4737 mac_len = skb->data - skb_mac_header(skb); 4738 hard_start = skb->data - skb_headroom(skb); 4739 4740 /* SKB "head" area always have tailroom for skb_shared_info */ 4741 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4742 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4743 4744 rxqueue = netif_get_rxqueue(skb); 4745 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4746 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4747 skb_headlen(skb) + mac_len, true); 4748 4749 orig_data_end = xdp->data_end; 4750 orig_data = xdp->data; 4751 eth = (struct ethhdr *)xdp->data; 4752 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4753 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4754 orig_eth_type = eth->h_proto; 4755 4756 act = bpf_prog_run_xdp(xdp_prog, xdp); 4757 4758 /* check if bpf_xdp_adjust_head was used */ 4759 off = xdp->data - orig_data; 4760 if (off) { 4761 if (off > 0) 4762 __skb_pull(skb, off); 4763 else if (off < 0) 4764 __skb_push(skb, -off); 4765 4766 skb->mac_header += off; 4767 skb_reset_network_header(skb); 4768 } 4769 4770 /* check if bpf_xdp_adjust_tail was used */ 4771 off = xdp->data_end - orig_data_end; 4772 if (off != 0) { 4773 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4774 skb->len += off; /* positive on grow, negative on shrink */ 4775 } 4776 4777 /* check if XDP changed eth hdr such SKB needs update */ 4778 eth = (struct ethhdr *)xdp->data; 4779 if ((orig_eth_type != eth->h_proto) || 4780 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4781 skb->dev->dev_addr)) || 4782 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4783 __skb_push(skb, ETH_HLEN); 4784 skb->pkt_type = PACKET_HOST; 4785 skb->protocol = eth_type_trans(skb, skb->dev); 4786 } 4787 4788 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4789 * before calling us again on redirect path. We do not call do_redirect 4790 * as we leave that up to the caller. 4791 * 4792 * Caller is responsible for managing lifetime of skb (i.e. calling 4793 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4794 */ 4795 switch (act) { 4796 case XDP_REDIRECT: 4797 case XDP_TX: 4798 __skb_push(skb, mac_len); 4799 break; 4800 case XDP_PASS: 4801 metalen = xdp->data - xdp->data_meta; 4802 if (metalen) 4803 skb_metadata_set(skb, metalen); 4804 break; 4805 } 4806 4807 return act; 4808 } 4809 4810 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4811 struct xdp_buff *xdp, 4812 struct bpf_prog *xdp_prog) 4813 { 4814 u32 act = XDP_DROP; 4815 4816 /* Reinjected packets coming from act_mirred or similar should 4817 * not get XDP generic processing. 4818 */ 4819 if (skb_is_redirected(skb)) 4820 return XDP_PASS; 4821 4822 /* XDP packets must be linear and must have sufficient headroom 4823 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4824 * native XDP provides, thus we need to do it here as well. 4825 */ 4826 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4827 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4828 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4829 int troom = skb->tail + skb->data_len - skb->end; 4830 4831 /* In case we have to go down the path and also linearize, 4832 * then lets do the pskb_expand_head() work just once here. 4833 */ 4834 if (pskb_expand_head(skb, 4835 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4836 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4837 goto do_drop; 4838 if (skb_linearize(skb)) 4839 goto do_drop; 4840 } 4841 4842 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4843 switch (act) { 4844 case XDP_REDIRECT: 4845 case XDP_TX: 4846 case XDP_PASS: 4847 break; 4848 default: 4849 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4850 fallthrough; 4851 case XDP_ABORTED: 4852 trace_xdp_exception(skb->dev, xdp_prog, act); 4853 fallthrough; 4854 case XDP_DROP: 4855 do_drop: 4856 kfree_skb(skb); 4857 break; 4858 } 4859 4860 return act; 4861 } 4862 4863 /* When doing generic XDP we have to bypass the qdisc layer and the 4864 * network taps in order to match in-driver-XDP behavior. 4865 */ 4866 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4867 { 4868 struct net_device *dev = skb->dev; 4869 struct netdev_queue *txq; 4870 bool free_skb = true; 4871 int cpu, rc; 4872 4873 txq = netdev_core_pick_tx(dev, skb, NULL); 4874 cpu = smp_processor_id(); 4875 HARD_TX_LOCK(dev, txq, cpu); 4876 if (!netif_xmit_stopped(txq)) { 4877 rc = netdev_start_xmit(skb, dev, txq, 0); 4878 if (dev_xmit_complete(rc)) 4879 free_skb = false; 4880 } 4881 HARD_TX_UNLOCK(dev, txq); 4882 if (free_skb) { 4883 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4884 kfree_skb(skb); 4885 } 4886 } 4887 4888 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4889 4890 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4891 { 4892 if (xdp_prog) { 4893 struct xdp_buff xdp; 4894 u32 act; 4895 int err; 4896 4897 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4898 if (act != XDP_PASS) { 4899 switch (act) { 4900 case XDP_REDIRECT: 4901 err = xdp_do_generic_redirect(skb->dev, skb, 4902 &xdp, xdp_prog); 4903 if (err) 4904 goto out_redir; 4905 break; 4906 case XDP_TX: 4907 generic_xdp_tx(skb, xdp_prog); 4908 break; 4909 } 4910 return XDP_DROP; 4911 } 4912 } 4913 return XDP_PASS; 4914 out_redir: 4915 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4916 return XDP_DROP; 4917 } 4918 EXPORT_SYMBOL_GPL(do_xdp_generic); 4919 4920 static int netif_rx_internal(struct sk_buff *skb) 4921 { 4922 int ret; 4923 4924 net_timestamp_check(netdev_tstamp_prequeue, skb); 4925 4926 trace_netif_rx(skb); 4927 4928 #ifdef CONFIG_RPS 4929 if (static_branch_unlikely(&rps_needed)) { 4930 struct rps_dev_flow voidflow, *rflow = &voidflow; 4931 int cpu; 4932 4933 rcu_read_lock(); 4934 4935 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4936 if (cpu < 0) 4937 cpu = smp_processor_id(); 4938 4939 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4940 4941 rcu_read_unlock(); 4942 } else 4943 #endif 4944 { 4945 unsigned int qtail; 4946 4947 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4948 } 4949 return ret; 4950 } 4951 4952 /** 4953 * __netif_rx - Slightly optimized version of netif_rx 4954 * @skb: buffer to post 4955 * 4956 * This behaves as netif_rx except that it does not disable bottom halves. 4957 * As a result this function may only be invoked from the interrupt context 4958 * (either hard or soft interrupt). 4959 */ 4960 int __netif_rx(struct sk_buff *skb) 4961 { 4962 int ret; 4963 4964 lockdep_assert_once(hardirq_count() | softirq_count()); 4965 4966 trace_netif_rx_entry(skb); 4967 ret = netif_rx_internal(skb); 4968 trace_netif_rx_exit(ret); 4969 return ret; 4970 } 4971 EXPORT_SYMBOL(__netif_rx); 4972 4973 /** 4974 * netif_rx - post buffer to the network code 4975 * @skb: buffer to post 4976 * 4977 * This function receives a packet from a device driver and queues it for 4978 * the upper (protocol) levels to process via the backlog NAPI device. It 4979 * always succeeds. The buffer may be dropped during processing for 4980 * congestion control or by the protocol layers. 4981 * The network buffer is passed via the backlog NAPI device. Modern NIC 4982 * driver should use NAPI and GRO. 4983 * This function can used from interrupt and from process context. The 4984 * caller from process context must not disable interrupts before invoking 4985 * this function. 4986 * 4987 * return values: 4988 * NET_RX_SUCCESS (no congestion) 4989 * NET_RX_DROP (packet was dropped) 4990 * 4991 */ 4992 int netif_rx(struct sk_buff *skb) 4993 { 4994 bool need_bh_off = !(hardirq_count() | softirq_count()); 4995 int ret; 4996 4997 if (need_bh_off) 4998 local_bh_disable(); 4999 trace_netif_rx_entry(skb); 5000 ret = netif_rx_internal(skb); 5001 trace_netif_rx_exit(ret); 5002 if (need_bh_off) 5003 local_bh_enable(); 5004 return ret; 5005 } 5006 EXPORT_SYMBOL(netif_rx); 5007 5008 static __latent_entropy void net_tx_action(struct softirq_action *h) 5009 { 5010 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5011 5012 if (sd->completion_queue) { 5013 struct sk_buff *clist; 5014 5015 local_irq_disable(); 5016 clist = sd->completion_queue; 5017 sd->completion_queue = NULL; 5018 local_irq_enable(); 5019 5020 while (clist) { 5021 struct sk_buff *skb = clist; 5022 5023 clist = clist->next; 5024 5025 WARN_ON(refcount_read(&skb->users)); 5026 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5027 trace_consume_skb(skb); 5028 else 5029 trace_kfree_skb(skb, net_tx_action, 5030 SKB_DROP_REASON_NOT_SPECIFIED); 5031 5032 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5033 __kfree_skb(skb); 5034 else 5035 __kfree_skb_defer(skb); 5036 } 5037 } 5038 5039 if (sd->output_queue) { 5040 struct Qdisc *head; 5041 5042 local_irq_disable(); 5043 head = sd->output_queue; 5044 sd->output_queue = NULL; 5045 sd->output_queue_tailp = &sd->output_queue; 5046 local_irq_enable(); 5047 5048 rcu_read_lock(); 5049 5050 while (head) { 5051 struct Qdisc *q = head; 5052 spinlock_t *root_lock = NULL; 5053 5054 head = head->next_sched; 5055 5056 /* We need to make sure head->next_sched is read 5057 * before clearing __QDISC_STATE_SCHED 5058 */ 5059 smp_mb__before_atomic(); 5060 5061 if (!(q->flags & TCQ_F_NOLOCK)) { 5062 root_lock = qdisc_lock(q); 5063 spin_lock(root_lock); 5064 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5065 &q->state))) { 5066 /* There is a synchronize_net() between 5067 * STATE_DEACTIVATED flag being set and 5068 * qdisc_reset()/some_qdisc_is_busy() in 5069 * dev_deactivate(), so we can safely bail out 5070 * early here to avoid data race between 5071 * qdisc_deactivate() and some_qdisc_is_busy() 5072 * for lockless qdisc. 5073 */ 5074 clear_bit(__QDISC_STATE_SCHED, &q->state); 5075 continue; 5076 } 5077 5078 clear_bit(__QDISC_STATE_SCHED, &q->state); 5079 qdisc_run(q); 5080 if (root_lock) 5081 spin_unlock(root_lock); 5082 } 5083 5084 rcu_read_unlock(); 5085 } 5086 5087 xfrm_dev_backlog(sd); 5088 } 5089 5090 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5091 /* This hook is defined here for ATM LANE */ 5092 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5093 unsigned char *addr) __read_mostly; 5094 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5095 #endif 5096 5097 static inline struct sk_buff * 5098 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5099 struct net_device *orig_dev, bool *another) 5100 { 5101 #ifdef CONFIG_NET_CLS_ACT 5102 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5103 struct tcf_result cl_res; 5104 5105 /* If there's at least one ingress present somewhere (so 5106 * we get here via enabled static key), remaining devices 5107 * that are not configured with an ingress qdisc will bail 5108 * out here. 5109 */ 5110 if (!miniq) 5111 return skb; 5112 5113 if (*pt_prev) { 5114 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5115 *pt_prev = NULL; 5116 } 5117 5118 qdisc_skb_cb(skb)->pkt_len = skb->len; 5119 tc_skb_cb(skb)->mru = 0; 5120 tc_skb_cb(skb)->post_ct = false; 5121 skb->tc_at_ingress = 1; 5122 mini_qdisc_bstats_cpu_update(miniq, skb); 5123 5124 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5125 case TC_ACT_OK: 5126 case TC_ACT_RECLASSIFY: 5127 skb->tc_index = TC_H_MIN(cl_res.classid); 5128 break; 5129 case TC_ACT_SHOT: 5130 mini_qdisc_qstats_cpu_drop(miniq); 5131 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5132 return NULL; 5133 case TC_ACT_STOLEN: 5134 case TC_ACT_QUEUED: 5135 case TC_ACT_TRAP: 5136 consume_skb(skb); 5137 return NULL; 5138 case TC_ACT_REDIRECT: 5139 /* skb_mac_header check was done by cls/act_bpf, so 5140 * we can safely push the L2 header back before 5141 * redirecting to another netdev 5142 */ 5143 __skb_push(skb, skb->mac_len); 5144 if (skb_do_redirect(skb) == -EAGAIN) { 5145 __skb_pull(skb, skb->mac_len); 5146 *another = true; 5147 break; 5148 } 5149 return NULL; 5150 case TC_ACT_CONSUMED: 5151 return NULL; 5152 default: 5153 break; 5154 } 5155 #endif /* CONFIG_NET_CLS_ACT */ 5156 return skb; 5157 } 5158 5159 /** 5160 * netdev_is_rx_handler_busy - check if receive handler is registered 5161 * @dev: device to check 5162 * 5163 * Check if a receive handler is already registered for a given device. 5164 * Return true if there one. 5165 * 5166 * The caller must hold the rtnl_mutex. 5167 */ 5168 bool netdev_is_rx_handler_busy(struct net_device *dev) 5169 { 5170 ASSERT_RTNL(); 5171 return dev && rtnl_dereference(dev->rx_handler); 5172 } 5173 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5174 5175 /** 5176 * netdev_rx_handler_register - register receive handler 5177 * @dev: device to register a handler for 5178 * @rx_handler: receive handler to register 5179 * @rx_handler_data: data pointer that is used by rx handler 5180 * 5181 * Register a receive handler for a device. This handler will then be 5182 * called from __netif_receive_skb. A negative errno code is returned 5183 * on a failure. 5184 * 5185 * The caller must hold the rtnl_mutex. 5186 * 5187 * For a general description of rx_handler, see enum rx_handler_result. 5188 */ 5189 int netdev_rx_handler_register(struct net_device *dev, 5190 rx_handler_func_t *rx_handler, 5191 void *rx_handler_data) 5192 { 5193 if (netdev_is_rx_handler_busy(dev)) 5194 return -EBUSY; 5195 5196 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5197 return -EINVAL; 5198 5199 /* Note: rx_handler_data must be set before rx_handler */ 5200 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5201 rcu_assign_pointer(dev->rx_handler, rx_handler); 5202 5203 return 0; 5204 } 5205 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5206 5207 /** 5208 * netdev_rx_handler_unregister - unregister receive handler 5209 * @dev: device to unregister a handler from 5210 * 5211 * Unregister a receive handler from a device. 5212 * 5213 * The caller must hold the rtnl_mutex. 5214 */ 5215 void netdev_rx_handler_unregister(struct net_device *dev) 5216 { 5217 5218 ASSERT_RTNL(); 5219 RCU_INIT_POINTER(dev->rx_handler, NULL); 5220 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5221 * section has a guarantee to see a non NULL rx_handler_data 5222 * as well. 5223 */ 5224 synchronize_net(); 5225 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5226 } 5227 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5228 5229 /* 5230 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5231 * the special handling of PFMEMALLOC skbs. 5232 */ 5233 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5234 { 5235 switch (skb->protocol) { 5236 case htons(ETH_P_ARP): 5237 case htons(ETH_P_IP): 5238 case htons(ETH_P_IPV6): 5239 case htons(ETH_P_8021Q): 5240 case htons(ETH_P_8021AD): 5241 return true; 5242 default: 5243 return false; 5244 } 5245 } 5246 5247 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5248 int *ret, struct net_device *orig_dev) 5249 { 5250 if (nf_hook_ingress_active(skb)) { 5251 int ingress_retval; 5252 5253 if (*pt_prev) { 5254 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5255 *pt_prev = NULL; 5256 } 5257 5258 rcu_read_lock(); 5259 ingress_retval = nf_hook_ingress(skb); 5260 rcu_read_unlock(); 5261 return ingress_retval; 5262 } 5263 return 0; 5264 } 5265 5266 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5267 struct packet_type **ppt_prev) 5268 { 5269 struct packet_type *ptype, *pt_prev; 5270 rx_handler_func_t *rx_handler; 5271 struct sk_buff *skb = *pskb; 5272 struct net_device *orig_dev; 5273 bool deliver_exact = false; 5274 int ret = NET_RX_DROP; 5275 __be16 type; 5276 5277 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5278 5279 trace_netif_receive_skb(skb); 5280 5281 orig_dev = skb->dev; 5282 5283 skb_reset_network_header(skb); 5284 if (!skb_transport_header_was_set(skb)) 5285 skb_reset_transport_header(skb); 5286 skb_reset_mac_len(skb); 5287 5288 pt_prev = NULL; 5289 5290 another_round: 5291 skb->skb_iif = skb->dev->ifindex; 5292 5293 __this_cpu_inc(softnet_data.processed); 5294 5295 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5296 int ret2; 5297 5298 migrate_disable(); 5299 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5300 migrate_enable(); 5301 5302 if (ret2 != XDP_PASS) { 5303 ret = NET_RX_DROP; 5304 goto out; 5305 } 5306 } 5307 5308 if (eth_type_vlan(skb->protocol)) { 5309 skb = skb_vlan_untag(skb); 5310 if (unlikely(!skb)) 5311 goto out; 5312 } 5313 5314 if (skb_skip_tc_classify(skb)) 5315 goto skip_classify; 5316 5317 if (pfmemalloc) 5318 goto skip_taps; 5319 5320 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5321 if (pt_prev) 5322 ret = deliver_skb(skb, pt_prev, orig_dev); 5323 pt_prev = ptype; 5324 } 5325 5326 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5327 if (pt_prev) 5328 ret = deliver_skb(skb, pt_prev, orig_dev); 5329 pt_prev = ptype; 5330 } 5331 5332 skip_taps: 5333 #ifdef CONFIG_NET_INGRESS 5334 if (static_branch_unlikely(&ingress_needed_key)) { 5335 bool another = false; 5336 5337 nf_skip_egress(skb, true); 5338 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5339 &another); 5340 if (another) 5341 goto another_round; 5342 if (!skb) 5343 goto out; 5344 5345 nf_skip_egress(skb, false); 5346 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5347 goto out; 5348 } 5349 #endif 5350 skb_reset_redirect(skb); 5351 skip_classify: 5352 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5353 goto drop; 5354 5355 if (skb_vlan_tag_present(skb)) { 5356 if (pt_prev) { 5357 ret = deliver_skb(skb, pt_prev, orig_dev); 5358 pt_prev = NULL; 5359 } 5360 if (vlan_do_receive(&skb)) 5361 goto another_round; 5362 else if (unlikely(!skb)) 5363 goto out; 5364 } 5365 5366 rx_handler = rcu_dereference(skb->dev->rx_handler); 5367 if (rx_handler) { 5368 if (pt_prev) { 5369 ret = deliver_skb(skb, pt_prev, orig_dev); 5370 pt_prev = NULL; 5371 } 5372 switch (rx_handler(&skb)) { 5373 case RX_HANDLER_CONSUMED: 5374 ret = NET_RX_SUCCESS; 5375 goto out; 5376 case RX_HANDLER_ANOTHER: 5377 goto another_round; 5378 case RX_HANDLER_EXACT: 5379 deliver_exact = true; 5380 break; 5381 case RX_HANDLER_PASS: 5382 break; 5383 default: 5384 BUG(); 5385 } 5386 } 5387 5388 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5389 check_vlan_id: 5390 if (skb_vlan_tag_get_id(skb)) { 5391 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5392 * find vlan device. 5393 */ 5394 skb->pkt_type = PACKET_OTHERHOST; 5395 } else if (eth_type_vlan(skb->protocol)) { 5396 /* Outer header is 802.1P with vlan 0, inner header is 5397 * 802.1Q or 802.1AD and vlan_do_receive() above could 5398 * not find vlan dev for vlan id 0. 5399 */ 5400 __vlan_hwaccel_clear_tag(skb); 5401 skb = skb_vlan_untag(skb); 5402 if (unlikely(!skb)) 5403 goto out; 5404 if (vlan_do_receive(&skb)) 5405 /* After stripping off 802.1P header with vlan 0 5406 * vlan dev is found for inner header. 5407 */ 5408 goto another_round; 5409 else if (unlikely(!skb)) 5410 goto out; 5411 else 5412 /* We have stripped outer 802.1P vlan 0 header. 5413 * But could not find vlan dev. 5414 * check again for vlan id to set OTHERHOST. 5415 */ 5416 goto check_vlan_id; 5417 } 5418 /* Note: we might in the future use prio bits 5419 * and set skb->priority like in vlan_do_receive() 5420 * For the time being, just ignore Priority Code Point 5421 */ 5422 __vlan_hwaccel_clear_tag(skb); 5423 } 5424 5425 type = skb->protocol; 5426 5427 /* deliver only exact match when indicated */ 5428 if (likely(!deliver_exact)) { 5429 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5430 &ptype_base[ntohs(type) & 5431 PTYPE_HASH_MASK]); 5432 } 5433 5434 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5435 &orig_dev->ptype_specific); 5436 5437 if (unlikely(skb->dev != orig_dev)) { 5438 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5439 &skb->dev->ptype_specific); 5440 } 5441 5442 if (pt_prev) { 5443 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5444 goto drop; 5445 *ppt_prev = pt_prev; 5446 } else { 5447 drop: 5448 if (!deliver_exact) 5449 dev_core_stats_rx_dropped_inc(skb->dev); 5450 else 5451 dev_core_stats_rx_nohandler_inc(skb->dev); 5452 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5453 /* Jamal, now you will not able to escape explaining 5454 * me how you were going to use this. :-) 5455 */ 5456 ret = NET_RX_DROP; 5457 } 5458 5459 out: 5460 /* The invariant here is that if *ppt_prev is not NULL 5461 * then skb should also be non-NULL. 5462 * 5463 * Apparently *ppt_prev assignment above holds this invariant due to 5464 * skb dereferencing near it. 5465 */ 5466 *pskb = skb; 5467 return ret; 5468 } 5469 5470 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5471 { 5472 struct net_device *orig_dev = skb->dev; 5473 struct packet_type *pt_prev = NULL; 5474 int ret; 5475 5476 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5477 if (pt_prev) 5478 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5479 skb->dev, pt_prev, orig_dev); 5480 return ret; 5481 } 5482 5483 /** 5484 * netif_receive_skb_core - special purpose version of netif_receive_skb 5485 * @skb: buffer to process 5486 * 5487 * More direct receive version of netif_receive_skb(). It should 5488 * only be used by callers that have a need to skip RPS and Generic XDP. 5489 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5490 * 5491 * This function may only be called from softirq context and interrupts 5492 * should be enabled. 5493 * 5494 * Return values (usually ignored): 5495 * NET_RX_SUCCESS: no congestion 5496 * NET_RX_DROP: packet was dropped 5497 */ 5498 int netif_receive_skb_core(struct sk_buff *skb) 5499 { 5500 int ret; 5501 5502 rcu_read_lock(); 5503 ret = __netif_receive_skb_one_core(skb, false); 5504 rcu_read_unlock(); 5505 5506 return ret; 5507 } 5508 EXPORT_SYMBOL(netif_receive_skb_core); 5509 5510 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5511 struct packet_type *pt_prev, 5512 struct net_device *orig_dev) 5513 { 5514 struct sk_buff *skb, *next; 5515 5516 if (!pt_prev) 5517 return; 5518 if (list_empty(head)) 5519 return; 5520 if (pt_prev->list_func != NULL) 5521 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5522 ip_list_rcv, head, pt_prev, orig_dev); 5523 else 5524 list_for_each_entry_safe(skb, next, head, list) { 5525 skb_list_del_init(skb); 5526 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5527 } 5528 } 5529 5530 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5531 { 5532 /* Fast-path assumptions: 5533 * - There is no RX handler. 5534 * - Only one packet_type matches. 5535 * If either of these fails, we will end up doing some per-packet 5536 * processing in-line, then handling the 'last ptype' for the whole 5537 * sublist. This can't cause out-of-order delivery to any single ptype, 5538 * because the 'last ptype' must be constant across the sublist, and all 5539 * other ptypes are handled per-packet. 5540 */ 5541 /* Current (common) ptype of sublist */ 5542 struct packet_type *pt_curr = NULL; 5543 /* Current (common) orig_dev of sublist */ 5544 struct net_device *od_curr = NULL; 5545 struct list_head sublist; 5546 struct sk_buff *skb, *next; 5547 5548 INIT_LIST_HEAD(&sublist); 5549 list_for_each_entry_safe(skb, next, head, list) { 5550 struct net_device *orig_dev = skb->dev; 5551 struct packet_type *pt_prev = NULL; 5552 5553 skb_list_del_init(skb); 5554 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5555 if (!pt_prev) 5556 continue; 5557 if (pt_curr != pt_prev || od_curr != orig_dev) { 5558 /* dispatch old sublist */ 5559 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5560 /* start new sublist */ 5561 INIT_LIST_HEAD(&sublist); 5562 pt_curr = pt_prev; 5563 od_curr = orig_dev; 5564 } 5565 list_add_tail(&skb->list, &sublist); 5566 } 5567 5568 /* dispatch final sublist */ 5569 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5570 } 5571 5572 static int __netif_receive_skb(struct sk_buff *skb) 5573 { 5574 int ret; 5575 5576 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5577 unsigned int noreclaim_flag; 5578 5579 /* 5580 * PFMEMALLOC skbs are special, they should 5581 * - be delivered to SOCK_MEMALLOC sockets only 5582 * - stay away from userspace 5583 * - have bounded memory usage 5584 * 5585 * Use PF_MEMALLOC as this saves us from propagating the allocation 5586 * context down to all allocation sites. 5587 */ 5588 noreclaim_flag = memalloc_noreclaim_save(); 5589 ret = __netif_receive_skb_one_core(skb, true); 5590 memalloc_noreclaim_restore(noreclaim_flag); 5591 } else 5592 ret = __netif_receive_skb_one_core(skb, false); 5593 5594 return ret; 5595 } 5596 5597 static void __netif_receive_skb_list(struct list_head *head) 5598 { 5599 unsigned long noreclaim_flag = 0; 5600 struct sk_buff *skb, *next; 5601 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5602 5603 list_for_each_entry_safe(skb, next, head, list) { 5604 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5605 struct list_head sublist; 5606 5607 /* Handle the previous sublist */ 5608 list_cut_before(&sublist, head, &skb->list); 5609 if (!list_empty(&sublist)) 5610 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5611 pfmemalloc = !pfmemalloc; 5612 /* See comments in __netif_receive_skb */ 5613 if (pfmemalloc) 5614 noreclaim_flag = memalloc_noreclaim_save(); 5615 else 5616 memalloc_noreclaim_restore(noreclaim_flag); 5617 } 5618 } 5619 /* Handle the remaining sublist */ 5620 if (!list_empty(head)) 5621 __netif_receive_skb_list_core(head, pfmemalloc); 5622 /* Restore pflags */ 5623 if (pfmemalloc) 5624 memalloc_noreclaim_restore(noreclaim_flag); 5625 } 5626 5627 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5628 { 5629 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5630 struct bpf_prog *new = xdp->prog; 5631 int ret = 0; 5632 5633 switch (xdp->command) { 5634 case XDP_SETUP_PROG: 5635 rcu_assign_pointer(dev->xdp_prog, new); 5636 if (old) 5637 bpf_prog_put(old); 5638 5639 if (old && !new) { 5640 static_branch_dec(&generic_xdp_needed_key); 5641 } else if (new && !old) { 5642 static_branch_inc(&generic_xdp_needed_key); 5643 dev_disable_lro(dev); 5644 dev_disable_gro_hw(dev); 5645 } 5646 break; 5647 5648 default: 5649 ret = -EINVAL; 5650 break; 5651 } 5652 5653 return ret; 5654 } 5655 5656 static int netif_receive_skb_internal(struct sk_buff *skb) 5657 { 5658 int ret; 5659 5660 net_timestamp_check(netdev_tstamp_prequeue, skb); 5661 5662 if (skb_defer_rx_timestamp(skb)) 5663 return NET_RX_SUCCESS; 5664 5665 rcu_read_lock(); 5666 #ifdef CONFIG_RPS 5667 if (static_branch_unlikely(&rps_needed)) { 5668 struct rps_dev_flow voidflow, *rflow = &voidflow; 5669 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5670 5671 if (cpu >= 0) { 5672 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5673 rcu_read_unlock(); 5674 return ret; 5675 } 5676 } 5677 #endif 5678 ret = __netif_receive_skb(skb); 5679 rcu_read_unlock(); 5680 return ret; 5681 } 5682 5683 void netif_receive_skb_list_internal(struct list_head *head) 5684 { 5685 struct sk_buff *skb, *next; 5686 struct list_head sublist; 5687 5688 INIT_LIST_HEAD(&sublist); 5689 list_for_each_entry_safe(skb, next, head, list) { 5690 net_timestamp_check(netdev_tstamp_prequeue, skb); 5691 skb_list_del_init(skb); 5692 if (!skb_defer_rx_timestamp(skb)) 5693 list_add_tail(&skb->list, &sublist); 5694 } 5695 list_splice_init(&sublist, head); 5696 5697 rcu_read_lock(); 5698 #ifdef CONFIG_RPS 5699 if (static_branch_unlikely(&rps_needed)) { 5700 list_for_each_entry_safe(skb, next, head, list) { 5701 struct rps_dev_flow voidflow, *rflow = &voidflow; 5702 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5703 5704 if (cpu >= 0) { 5705 /* Will be handled, remove from list */ 5706 skb_list_del_init(skb); 5707 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5708 } 5709 } 5710 } 5711 #endif 5712 __netif_receive_skb_list(head); 5713 rcu_read_unlock(); 5714 } 5715 5716 /** 5717 * netif_receive_skb - process receive buffer from network 5718 * @skb: buffer to process 5719 * 5720 * netif_receive_skb() is the main receive data processing function. 5721 * It always succeeds. The buffer may be dropped during processing 5722 * for congestion control or by the protocol layers. 5723 * 5724 * This function may only be called from softirq context and interrupts 5725 * should be enabled. 5726 * 5727 * Return values (usually ignored): 5728 * NET_RX_SUCCESS: no congestion 5729 * NET_RX_DROP: packet was dropped 5730 */ 5731 int netif_receive_skb(struct sk_buff *skb) 5732 { 5733 int ret; 5734 5735 trace_netif_receive_skb_entry(skb); 5736 5737 ret = netif_receive_skb_internal(skb); 5738 trace_netif_receive_skb_exit(ret); 5739 5740 return ret; 5741 } 5742 EXPORT_SYMBOL(netif_receive_skb); 5743 5744 /** 5745 * netif_receive_skb_list - process many receive buffers from network 5746 * @head: list of skbs to process. 5747 * 5748 * Since return value of netif_receive_skb() is normally ignored, and 5749 * wouldn't be meaningful for a list, this function returns void. 5750 * 5751 * This function may only be called from softirq context and interrupts 5752 * should be enabled. 5753 */ 5754 void netif_receive_skb_list(struct list_head *head) 5755 { 5756 struct sk_buff *skb; 5757 5758 if (list_empty(head)) 5759 return; 5760 if (trace_netif_receive_skb_list_entry_enabled()) { 5761 list_for_each_entry(skb, head, list) 5762 trace_netif_receive_skb_list_entry(skb); 5763 } 5764 netif_receive_skb_list_internal(head); 5765 trace_netif_receive_skb_list_exit(0); 5766 } 5767 EXPORT_SYMBOL(netif_receive_skb_list); 5768 5769 static DEFINE_PER_CPU(struct work_struct, flush_works); 5770 5771 /* Network device is going away, flush any packets still pending */ 5772 static void flush_backlog(struct work_struct *work) 5773 { 5774 struct sk_buff *skb, *tmp; 5775 struct softnet_data *sd; 5776 5777 local_bh_disable(); 5778 sd = this_cpu_ptr(&softnet_data); 5779 5780 rps_lock_irq_disable(sd); 5781 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5782 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5783 __skb_unlink(skb, &sd->input_pkt_queue); 5784 dev_kfree_skb_irq(skb); 5785 input_queue_head_incr(sd); 5786 } 5787 } 5788 rps_unlock_irq_enable(sd); 5789 5790 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5791 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5792 __skb_unlink(skb, &sd->process_queue); 5793 kfree_skb(skb); 5794 input_queue_head_incr(sd); 5795 } 5796 } 5797 local_bh_enable(); 5798 } 5799 5800 static bool flush_required(int cpu) 5801 { 5802 #if IS_ENABLED(CONFIG_RPS) 5803 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5804 bool do_flush; 5805 5806 rps_lock_irq_disable(sd); 5807 5808 /* as insertion into process_queue happens with the rps lock held, 5809 * process_queue access may race only with dequeue 5810 */ 5811 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5812 !skb_queue_empty_lockless(&sd->process_queue); 5813 rps_unlock_irq_enable(sd); 5814 5815 return do_flush; 5816 #endif 5817 /* without RPS we can't safely check input_pkt_queue: during a 5818 * concurrent remote skb_queue_splice() we can detect as empty both 5819 * input_pkt_queue and process_queue even if the latter could end-up 5820 * containing a lot of packets. 5821 */ 5822 return true; 5823 } 5824 5825 static void flush_all_backlogs(void) 5826 { 5827 static cpumask_t flush_cpus; 5828 unsigned int cpu; 5829 5830 /* since we are under rtnl lock protection we can use static data 5831 * for the cpumask and avoid allocating on stack the possibly 5832 * large mask 5833 */ 5834 ASSERT_RTNL(); 5835 5836 cpus_read_lock(); 5837 5838 cpumask_clear(&flush_cpus); 5839 for_each_online_cpu(cpu) { 5840 if (flush_required(cpu)) { 5841 queue_work_on(cpu, system_highpri_wq, 5842 per_cpu_ptr(&flush_works, cpu)); 5843 cpumask_set_cpu(cpu, &flush_cpus); 5844 } 5845 } 5846 5847 /* we can have in flight packet[s] on the cpus we are not flushing, 5848 * synchronize_net() in unregister_netdevice_many() will take care of 5849 * them 5850 */ 5851 for_each_cpu(cpu, &flush_cpus) 5852 flush_work(per_cpu_ptr(&flush_works, cpu)); 5853 5854 cpus_read_unlock(); 5855 } 5856 5857 static void net_rps_send_ipi(struct softnet_data *remsd) 5858 { 5859 #ifdef CONFIG_RPS 5860 while (remsd) { 5861 struct softnet_data *next = remsd->rps_ipi_next; 5862 5863 if (cpu_online(remsd->cpu)) 5864 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5865 remsd = next; 5866 } 5867 #endif 5868 } 5869 5870 /* 5871 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5872 * Note: called with local irq disabled, but exits with local irq enabled. 5873 */ 5874 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5875 { 5876 #ifdef CONFIG_RPS 5877 struct softnet_data *remsd = sd->rps_ipi_list; 5878 5879 if (remsd) { 5880 sd->rps_ipi_list = NULL; 5881 5882 local_irq_enable(); 5883 5884 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5885 net_rps_send_ipi(remsd); 5886 } else 5887 #endif 5888 local_irq_enable(); 5889 } 5890 5891 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5892 { 5893 #ifdef CONFIG_RPS 5894 return sd->rps_ipi_list != NULL; 5895 #else 5896 return false; 5897 #endif 5898 } 5899 5900 static int process_backlog(struct napi_struct *napi, int quota) 5901 { 5902 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5903 bool again = true; 5904 int work = 0; 5905 5906 /* Check if we have pending ipi, its better to send them now, 5907 * not waiting net_rx_action() end. 5908 */ 5909 if (sd_has_rps_ipi_waiting(sd)) { 5910 local_irq_disable(); 5911 net_rps_action_and_irq_enable(sd); 5912 } 5913 5914 napi->weight = dev_rx_weight; 5915 while (again) { 5916 struct sk_buff *skb; 5917 5918 while ((skb = __skb_dequeue(&sd->process_queue))) { 5919 rcu_read_lock(); 5920 __netif_receive_skb(skb); 5921 rcu_read_unlock(); 5922 input_queue_head_incr(sd); 5923 if (++work >= quota) 5924 return work; 5925 5926 } 5927 5928 rps_lock_irq_disable(sd); 5929 if (skb_queue_empty(&sd->input_pkt_queue)) { 5930 /* 5931 * Inline a custom version of __napi_complete(). 5932 * only current cpu owns and manipulates this napi, 5933 * and NAPI_STATE_SCHED is the only possible flag set 5934 * on backlog. 5935 * We can use a plain write instead of clear_bit(), 5936 * and we dont need an smp_mb() memory barrier. 5937 */ 5938 napi->state = 0; 5939 again = false; 5940 } else { 5941 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5942 &sd->process_queue); 5943 } 5944 rps_unlock_irq_enable(sd); 5945 } 5946 5947 return work; 5948 } 5949 5950 /** 5951 * __napi_schedule - schedule for receive 5952 * @n: entry to schedule 5953 * 5954 * The entry's receive function will be scheduled to run. 5955 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5956 */ 5957 void __napi_schedule(struct napi_struct *n) 5958 { 5959 unsigned long flags; 5960 5961 local_irq_save(flags); 5962 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5963 local_irq_restore(flags); 5964 } 5965 EXPORT_SYMBOL(__napi_schedule); 5966 5967 /** 5968 * napi_schedule_prep - check if napi can be scheduled 5969 * @n: napi context 5970 * 5971 * Test if NAPI routine is already running, and if not mark 5972 * it as running. This is used as a condition variable to 5973 * insure only one NAPI poll instance runs. We also make 5974 * sure there is no pending NAPI disable. 5975 */ 5976 bool napi_schedule_prep(struct napi_struct *n) 5977 { 5978 unsigned long val, new; 5979 5980 do { 5981 val = READ_ONCE(n->state); 5982 if (unlikely(val & NAPIF_STATE_DISABLE)) 5983 return false; 5984 new = val | NAPIF_STATE_SCHED; 5985 5986 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5987 * This was suggested by Alexander Duyck, as compiler 5988 * emits better code than : 5989 * if (val & NAPIF_STATE_SCHED) 5990 * new |= NAPIF_STATE_MISSED; 5991 */ 5992 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5993 NAPIF_STATE_MISSED; 5994 } while (cmpxchg(&n->state, val, new) != val); 5995 5996 return !(val & NAPIF_STATE_SCHED); 5997 } 5998 EXPORT_SYMBOL(napi_schedule_prep); 5999 6000 /** 6001 * __napi_schedule_irqoff - schedule for receive 6002 * @n: entry to schedule 6003 * 6004 * Variant of __napi_schedule() assuming hard irqs are masked. 6005 * 6006 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6007 * because the interrupt disabled assumption might not be true 6008 * due to force-threaded interrupts and spinlock substitution. 6009 */ 6010 void __napi_schedule_irqoff(struct napi_struct *n) 6011 { 6012 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6013 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6014 else 6015 __napi_schedule(n); 6016 } 6017 EXPORT_SYMBOL(__napi_schedule_irqoff); 6018 6019 bool napi_complete_done(struct napi_struct *n, int work_done) 6020 { 6021 unsigned long flags, val, new, timeout = 0; 6022 bool ret = true; 6023 6024 /* 6025 * 1) Don't let napi dequeue from the cpu poll list 6026 * just in case its running on a different cpu. 6027 * 2) If we are busy polling, do nothing here, we have 6028 * the guarantee we will be called later. 6029 */ 6030 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6031 NAPIF_STATE_IN_BUSY_POLL))) 6032 return false; 6033 6034 if (work_done) { 6035 if (n->gro_bitmask) 6036 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6037 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6038 } 6039 if (n->defer_hard_irqs_count > 0) { 6040 n->defer_hard_irqs_count--; 6041 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6042 if (timeout) 6043 ret = false; 6044 } 6045 if (n->gro_bitmask) { 6046 /* When the NAPI instance uses a timeout and keeps postponing 6047 * it, we need to bound somehow the time packets are kept in 6048 * the GRO layer 6049 */ 6050 napi_gro_flush(n, !!timeout); 6051 } 6052 6053 gro_normal_list(n); 6054 6055 if (unlikely(!list_empty(&n->poll_list))) { 6056 /* If n->poll_list is not empty, we need to mask irqs */ 6057 local_irq_save(flags); 6058 list_del_init(&n->poll_list); 6059 local_irq_restore(flags); 6060 } 6061 6062 do { 6063 val = READ_ONCE(n->state); 6064 6065 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6066 6067 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6068 NAPIF_STATE_SCHED_THREADED | 6069 NAPIF_STATE_PREFER_BUSY_POLL); 6070 6071 /* If STATE_MISSED was set, leave STATE_SCHED set, 6072 * because we will call napi->poll() one more time. 6073 * This C code was suggested by Alexander Duyck to help gcc. 6074 */ 6075 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6076 NAPIF_STATE_SCHED; 6077 } while (cmpxchg(&n->state, val, new) != val); 6078 6079 if (unlikely(val & NAPIF_STATE_MISSED)) { 6080 __napi_schedule(n); 6081 return false; 6082 } 6083 6084 if (timeout) 6085 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6086 HRTIMER_MODE_REL_PINNED); 6087 return ret; 6088 } 6089 EXPORT_SYMBOL(napi_complete_done); 6090 6091 /* must be called under rcu_read_lock(), as we dont take a reference */ 6092 static struct napi_struct *napi_by_id(unsigned int napi_id) 6093 { 6094 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6095 struct napi_struct *napi; 6096 6097 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6098 if (napi->napi_id == napi_id) 6099 return napi; 6100 6101 return NULL; 6102 } 6103 6104 #if defined(CONFIG_NET_RX_BUSY_POLL) 6105 6106 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6107 { 6108 if (!skip_schedule) { 6109 gro_normal_list(napi); 6110 __napi_schedule(napi); 6111 return; 6112 } 6113 6114 if (napi->gro_bitmask) { 6115 /* flush too old packets 6116 * If HZ < 1000, flush all packets. 6117 */ 6118 napi_gro_flush(napi, HZ >= 1000); 6119 } 6120 6121 gro_normal_list(napi); 6122 clear_bit(NAPI_STATE_SCHED, &napi->state); 6123 } 6124 6125 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6126 u16 budget) 6127 { 6128 bool skip_schedule = false; 6129 unsigned long timeout; 6130 int rc; 6131 6132 /* Busy polling means there is a high chance device driver hard irq 6133 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6134 * set in napi_schedule_prep(). 6135 * Since we are about to call napi->poll() once more, we can safely 6136 * clear NAPI_STATE_MISSED. 6137 * 6138 * Note: x86 could use a single "lock and ..." instruction 6139 * to perform these two clear_bit() 6140 */ 6141 clear_bit(NAPI_STATE_MISSED, &napi->state); 6142 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6143 6144 local_bh_disable(); 6145 6146 if (prefer_busy_poll) { 6147 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6148 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6149 if (napi->defer_hard_irqs_count && timeout) { 6150 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6151 skip_schedule = true; 6152 } 6153 } 6154 6155 /* All we really want here is to re-enable device interrupts. 6156 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6157 */ 6158 rc = napi->poll(napi, budget); 6159 /* We can't gro_normal_list() here, because napi->poll() might have 6160 * rearmed the napi (napi_complete_done()) in which case it could 6161 * already be running on another CPU. 6162 */ 6163 trace_napi_poll(napi, rc, budget); 6164 netpoll_poll_unlock(have_poll_lock); 6165 if (rc == budget) 6166 __busy_poll_stop(napi, skip_schedule); 6167 local_bh_enable(); 6168 } 6169 6170 void napi_busy_loop(unsigned int napi_id, 6171 bool (*loop_end)(void *, unsigned long), 6172 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6173 { 6174 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6175 int (*napi_poll)(struct napi_struct *napi, int budget); 6176 void *have_poll_lock = NULL; 6177 struct napi_struct *napi; 6178 6179 restart: 6180 napi_poll = NULL; 6181 6182 rcu_read_lock(); 6183 6184 napi = napi_by_id(napi_id); 6185 if (!napi) 6186 goto out; 6187 6188 preempt_disable(); 6189 for (;;) { 6190 int work = 0; 6191 6192 local_bh_disable(); 6193 if (!napi_poll) { 6194 unsigned long val = READ_ONCE(napi->state); 6195 6196 /* If multiple threads are competing for this napi, 6197 * we avoid dirtying napi->state as much as we can. 6198 */ 6199 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6200 NAPIF_STATE_IN_BUSY_POLL)) { 6201 if (prefer_busy_poll) 6202 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6203 goto count; 6204 } 6205 if (cmpxchg(&napi->state, val, 6206 val | NAPIF_STATE_IN_BUSY_POLL | 6207 NAPIF_STATE_SCHED) != val) { 6208 if (prefer_busy_poll) 6209 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6210 goto count; 6211 } 6212 have_poll_lock = netpoll_poll_lock(napi); 6213 napi_poll = napi->poll; 6214 } 6215 work = napi_poll(napi, budget); 6216 trace_napi_poll(napi, work, budget); 6217 gro_normal_list(napi); 6218 count: 6219 if (work > 0) 6220 __NET_ADD_STATS(dev_net(napi->dev), 6221 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6222 local_bh_enable(); 6223 6224 if (!loop_end || loop_end(loop_end_arg, start_time)) 6225 break; 6226 6227 if (unlikely(need_resched())) { 6228 if (napi_poll) 6229 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6230 preempt_enable(); 6231 rcu_read_unlock(); 6232 cond_resched(); 6233 if (loop_end(loop_end_arg, start_time)) 6234 return; 6235 goto restart; 6236 } 6237 cpu_relax(); 6238 } 6239 if (napi_poll) 6240 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6241 preempt_enable(); 6242 out: 6243 rcu_read_unlock(); 6244 } 6245 EXPORT_SYMBOL(napi_busy_loop); 6246 6247 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6248 6249 static void napi_hash_add(struct napi_struct *napi) 6250 { 6251 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6252 return; 6253 6254 spin_lock(&napi_hash_lock); 6255 6256 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6257 do { 6258 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6259 napi_gen_id = MIN_NAPI_ID; 6260 } while (napi_by_id(napi_gen_id)); 6261 napi->napi_id = napi_gen_id; 6262 6263 hlist_add_head_rcu(&napi->napi_hash_node, 6264 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6265 6266 spin_unlock(&napi_hash_lock); 6267 } 6268 6269 /* Warning : caller is responsible to make sure rcu grace period 6270 * is respected before freeing memory containing @napi 6271 */ 6272 static void napi_hash_del(struct napi_struct *napi) 6273 { 6274 spin_lock(&napi_hash_lock); 6275 6276 hlist_del_init_rcu(&napi->napi_hash_node); 6277 6278 spin_unlock(&napi_hash_lock); 6279 } 6280 6281 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6282 { 6283 struct napi_struct *napi; 6284 6285 napi = container_of(timer, struct napi_struct, timer); 6286 6287 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6288 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6289 */ 6290 if (!napi_disable_pending(napi) && 6291 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6292 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6293 __napi_schedule_irqoff(napi); 6294 } 6295 6296 return HRTIMER_NORESTART; 6297 } 6298 6299 static void init_gro_hash(struct napi_struct *napi) 6300 { 6301 int i; 6302 6303 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6304 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6305 napi->gro_hash[i].count = 0; 6306 } 6307 napi->gro_bitmask = 0; 6308 } 6309 6310 int dev_set_threaded(struct net_device *dev, bool threaded) 6311 { 6312 struct napi_struct *napi; 6313 int err = 0; 6314 6315 if (dev->threaded == threaded) 6316 return 0; 6317 6318 if (threaded) { 6319 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6320 if (!napi->thread) { 6321 err = napi_kthread_create(napi); 6322 if (err) { 6323 threaded = false; 6324 break; 6325 } 6326 } 6327 } 6328 } 6329 6330 dev->threaded = threaded; 6331 6332 /* Make sure kthread is created before THREADED bit 6333 * is set. 6334 */ 6335 smp_mb__before_atomic(); 6336 6337 /* Setting/unsetting threaded mode on a napi might not immediately 6338 * take effect, if the current napi instance is actively being 6339 * polled. In this case, the switch between threaded mode and 6340 * softirq mode will happen in the next round of napi_schedule(). 6341 * This should not cause hiccups/stalls to the live traffic. 6342 */ 6343 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6344 if (threaded) 6345 set_bit(NAPI_STATE_THREADED, &napi->state); 6346 else 6347 clear_bit(NAPI_STATE_THREADED, &napi->state); 6348 } 6349 6350 return err; 6351 } 6352 EXPORT_SYMBOL(dev_set_threaded); 6353 6354 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6355 int (*poll)(struct napi_struct *, int), int weight) 6356 { 6357 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6358 return; 6359 6360 INIT_LIST_HEAD(&napi->poll_list); 6361 INIT_HLIST_NODE(&napi->napi_hash_node); 6362 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6363 napi->timer.function = napi_watchdog; 6364 init_gro_hash(napi); 6365 napi->skb = NULL; 6366 INIT_LIST_HEAD(&napi->rx_list); 6367 napi->rx_count = 0; 6368 napi->poll = poll; 6369 if (weight > NAPI_POLL_WEIGHT) 6370 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6371 weight); 6372 napi->weight = weight; 6373 napi->dev = dev; 6374 #ifdef CONFIG_NETPOLL 6375 napi->poll_owner = -1; 6376 #endif 6377 set_bit(NAPI_STATE_SCHED, &napi->state); 6378 set_bit(NAPI_STATE_NPSVC, &napi->state); 6379 list_add_rcu(&napi->dev_list, &dev->napi_list); 6380 napi_hash_add(napi); 6381 /* Create kthread for this napi if dev->threaded is set. 6382 * Clear dev->threaded if kthread creation failed so that 6383 * threaded mode will not be enabled in napi_enable(). 6384 */ 6385 if (dev->threaded && napi_kthread_create(napi)) 6386 dev->threaded = 0; 6387 } 6388 EXPORT_SYMBOL(netif_napi_add_weight); 6389 6390 void napi_disable(struct napi_struct *n) 6391 { 6392 unsigned long val, new; 6393 6394 might_sleep(); 6395 set_bit(NAPI_STATE_DISABLE, &n->state); 6396 6397 for ( ; ; ) { 6398 val = READ_ONCE(n->state); 6399 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6400 usleep_range(20, 200); 6401 continue; 6402 } 6403 6404 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6405 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6406 6407 if (cmpxchg(&n->state, val, new) == val) 6408 break; 6409 } 6410 6411 hrtimer_cancel(&n->timer); 6412 6413 clear_bit(NAPI_STATE_DISABLE, &n->state); 6414 } 6415 EXPORT_SYMBOL(napi_disable); 6416 6417 /** 6418 * napi_enable - enable NAPI scheduling 6419 * @n: NAPI context 6420 * 6421 * Resume NAPI from being scheduled on this context. 6422 * Must be paired with napi_disable. 6423 */ 6424 void napi_enable(struct napi_struct *n) 6425 { 6426 unsigned long val, new; 6427 6428 do { 6429 val = READ_ONCE(n->state); 6430 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6431 6432 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6433 if (n->dev->threaded && n->thread) 6434 new |= NAPIF_STATE_THREADED; 6435 } while (cmpxchg(&n->state, val, new) != val); 6436 } 6437 EXPORT_SYMBOL(napi_enable); 6438 6439 static void flush_gro_hash(struct napi_struct *napi) 6440 { 6441 int i; 6442 6443 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6444 struct sk_buff *skb, *n; 6445 6446 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6447 kfree_skb(skb); 6448 napi->gro_hash[i].count = 0; 6449 } 6450 } 6451 6452 /* Must be called in process context */ 6453 void __netif_napi_del(struct napi_struct *napi) 6454 { 6455 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6456 return; 6457 6458 napi_hash_del(napi); 6459 list_del_rcu(&napi->dev_list); 6460 napi_free_frags(napi); 6461 6462 flush_gro_hash(napi); 6463 napi->gro_bitmask = 0; 6464 6465 if (napi->thread) { 6466 kthread_stop(napi->thread); 6467 napi->thread = NULL; 6468 } 6469 } 6470 EXPORT_SYMBOL(__netif_napi_del); 6471 6472 static int __napi_poll(struct napi_struct *n, bool *repoll) 6473 { 6474 int work, weight; 6475 6476 weight = n->weight; 6477 6478 /* This NAPI_STATE_SCHED test is for avoiding a race 6479 * with netpoll's poll_napi(). Only the entity which 6480 * obtains the lock and sees NAPI_STATE_SCHED set will 6481 * actually make the ->poll() call. Therefore we avoid 6482 * accidentally calling ->poll() when NAPI is not scheduled. 6483 */ 6484 work = 0; 6485 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6486 work = n->poll(n, weight); 6487 trace_napi_poll(n, work, weight); 6488 } 6489 6490 if (unlikely(work > weight)) 6491 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6492 n->poll, work, weight); 6493 6494 if (likely(work < weight)) 6495 return work; 6496 6497 /* Drivers must not modify the NAPI state if they 6498 * consume the entire weight. In such cases this code 6499 * still "owns" the NAPI instance and therefore can 6500 * move the instance around on the list at-will. 6501 */ 6502 if (unlikely(napi_disable_pending(n))) { 6503 napi_complete(n); 6504 return work; 6505 } 6506 6507 /* The NAPI context has more processing work, but busy-polling 6508 * is preferred. Exit early. 6509 */ 6510 if (napi_prefer_busy_poll(n)) { 6511 if (napi_complete_done(n, work)) { 6512 /* If timeout is not set, we need to make sure 6513 * that the NAPI is re-scheduled. 6514 */ 6515 napi_schedule(n); 6516 } 6517 return work; 6518 } 6519 6520 if (n->gro_bitmask) { 6521 /* flush too old packets 6522 * If HZ < 1000, flush all packets. 6523 */ 6524 napi_gro_flush(n, HZ >= 1000); 6525 } 6526 6527 gro_normal_list(n); 6528 6529 /* Some drivers may have called napi_schedule 6530 * prior to exhausting their budget. 6531 */ 6532 if (unlikely(!list_empty(&n->poll_list))) { 6533 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6534 n->dev ? n->dev->name : "backlog"); 6535 return work; 6536 } 6537 6538 *repoll = true; 6539 6540 return work; 6541 } 6542 6543 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6544 { 6545 bool do_repoll = false; 6546 void *have; 6547 int work; 6548 6549 list_del_init(&n->poll_list); 6550 6551 have = netpoll_poll_lock(n); 6552 6553 work = __napi_poll(n, &do_repoll); 6554 6555 if (do_repoll) 6556 list_add_tail(&n->poll_list, repoll); 6557 6558 netpoll_poll_unlock(have); 6559 6560 return work; 6561 } 6562 6563 static int napi_thread_wait(struct napi_struct *napi) 6564 { 6565 bool woken = false; 6566 6567 set_current_state(TASK_INTERRUPTIBLE); 6568 6569 while (!kthread_should_stop()) { 6570 /* Testing SCHED_THREADED bit here to make sure the current 6571 * kthread owns this napi and could poll on this napi. 6572 * Testing SCHED bit is not enough because SCHED bit might be 6573 * set by some other busy poll thread or by napi_disable(). 6574 */ 6575 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6576 WARN_ON(!list_empty(&napi->poll_list)); 6577 __set_current_state(TASK_RUNNING); 6578 return 0; 6579 } 6580 6581 schedule(); 6582 /* woken being true indicates this thread owns this napi. */ 6583 woken = true; 6584 set_current_state(TASK_INTERRUPTIBLE); 6585 } 6586 __set_current_state(TASK_RUNNING); 6587 6588 return -1; 6589 } 6590 6591 static int napi_threaded_poll(void *data) 6592 { 6593 struct napi_struct *napi = data; 6594 void *have; 6595 6596 while (!napi_thread_wait(napi)) { 6597 for (;;) { 6598 bool repoll = false; 6599 6600 local_bh_disable(); 6601 6602 have = netpoll_poll_lock(napi); 6603 __napi_poll(napi, &repoll); 6604 netpoll_poll_unlock(have); 6605 6606 local_bh_enable(); 6607 6608 if (!repoll) 6609 break; 6610 6611 cond_resched(); 6612 } 6613 } 6614 return 0; 6615 } 6616 6617 static void skb_defer_free_flush(struct softnet_data *sd) 6618 { 6619 struct sk_buff *skb, *next; 6620 unsigned long flags; 6621 6622 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6623 if (!READ_ONCE(sd->defer_list)) 6624 return; 6625 6626 spin_lock_irqsave(&sd->defer_lock, flags); 6627 skb = sd->defer_list; 6628 sd->defer_list = NULL; 6629 sd->defer_count = 0; 6630 spin_unlock_irqrestore(&sd->defer_lock, flags); 6631 6632 while (skb != NULL) { 6633 next = skb->next; 6634 napi_consume_skb(skb, 1); 6635 skb = next; 6636 } 6637 } 6638 6639 static __latent_entropy void net_rx_action(struct softirq_action *h) 6640 { 6641 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6642 unsigned long time_limit = jiffies + 6643 usecs_to_jiffies(netdev_budget_usecs); 6644 int budget = netdev_budget; 6645 LIST_HEAD(list); 6646 LIST_HEAD(repoll); 6647 6648 local_irq_disable(); 6649 list_splice_init(&sd->poll_list, &list); 6650 local_irq_enable(); 6651 6652 for (;;) { 6653 struct napi_struct *n; 6654 6655 skb_defer_free_flush(sd); 6656 6657 if (list_empty(&list)) { 6658 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6659 goto end; 6660 break; 6661 } 6662 6663 n = list_first_entry(&list, struct napi_struct, poll_list); 6664 budget -= napi_poll(n, &repoll); 6665 6666 /* If softirq window is exhausted then punt. 6667 * Allow this to run for 2 jiffies since which will allow 6668 * an average latency of 1.5/HZ. 6669 */ 6670 if (unlikely(budget <= 0 || 6671 time_after_eq(jiffies, time_limit))) { 6672 sd->time_squeeze++; 6673 break; 6674 } 6675 } 6676 6677 local_irq_disable(); 6678 6679 list_splice_tail_init(&sd->poll_list, &list); 6680 list_splice_tail(&repoll, &list); 6681 list_splice(&list, &sd->poll_list); 6682 if (!list_empty(&sd->poll_list)) 6683 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6684 6685 net_rps_action_and_irq_enable(sd); 6686 end:; 6687 } 6688 6689 struct netdev_adjacent { 6690 struct net_device *dev; 6691 netdevice_tracker dev_tracker; 6692 6693 /* upper master flag, there can only be one master device per list */ 6694 bool master; 6695 6696 /* lookup ignore flag */ 6697 bool ignore; 6698 6699 /* counter for the number of times this device was added to us */ 6700 u16 ref_nr; 6701 6702 /* private field for the users */ 6703 void *private; 6704 6705 struct list_head list; 6706 struct rcu_head rcu; 6707 }; 6708 6709 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6710 struct list_head *adj_list) 6711 { 6712 struct netdev_adjacent *adj; 6713 6714 list_for_each_entry(adj, adj_list, list) { 6715 if (adj->dev == adj_dev) 6716 return adj; 6717 } 6718 return NULL; 6719 } 6720 6721 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6722 struct netdev_nested_priv *priv) 6723 { 6724 struct net_device *dev = (struct net_device *)priv->data; 6725 6726 return upper_dev == dev; 6727 } 6728 6729 /** 6730 * netdev_has_upper_dev - Check if device is linked to an upper device 6731 * @dev: device 6732 * @upper_dev: upper device to check 6733 * 6734 * Find out if a device is linked to specified upper device and return true 6735 * in case it is. Note that this checks only immediate upper device, 6736 * not through a complete stack of devices. The caller must hold the RTNL lock. 6737 */ 6738 bool netdev_has_upper_dev(struct net_device *dev, 6739 struct net_device *upper_dev) 6740 { 6741 struct netdev_nested_priv priv = { 6742 .data = (void *)upper_dev, 6743 }; 6744 6745 ASSERT_RTNL(); 6746 6747 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6748 &priv); 6749 } 6750 EXPORT_SYMBOL(netdev_has_upper_dev); 6751 6752 /** 6753 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6754 * @dev: device 6755 * @upper_dev: upper device to check 6756 * 6757 * Find out if a device is linked to specified upper device and return true 6758 * in case it is. Note that this checks the entire upper device chain. 6759 * The caller must hold rcu lock. 6760 */ 6761 6762 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6763 struct net_device *upper_dev) 6764 { 6765 struct netdev_nested_priv priv = { 6766 .data = (void *)upper_dev, 6767 }; 6768 6769 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6770 &priv); 6771 } 6772 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6773 6774 /** 6775 * netdev_has_any_upper_dev - Check if device is linked to some device 6776 * @dev: device 6777 * 6778 * Find out if a device is linked to an upper device and return true in case 6779 * it is. The caller must hold the RTNL lock. 6780 */ 6781 bool netdev_has_any_upper_dev(struct net_device *dev) 6782 { 6783 ASSERT_RTNL(); 6784 6785 return !list_empty(&dev->adj_list.upper); 6786 } 6787 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6788 6789 /** 6790 * netdev_master_upper_dev_get - Get master upper device 6791 * @dev: device 6792 * 6793 * Find a master upper device and return pointer to it or NULL in case 6794 * it's not there. The caller must hold the RTNL lock. 6795 */ 6796 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6797 { 6798 struct netdev_adjacent *upper; 6799 6800 ASSERT_RTNL(); 6801 6802 if (list_empty(&dev->adj_list.upper)) 6803 return NULL; 6804 6805 upper = list_first_entry(&dev->adj_list.upper, 6806 struct netdev_adjacent, list); 6807 if (likely(upper->master)) 6808 return upper->dev; 6809 return NULL; 6810 } 6811 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6812 6813 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6814 { 6815 struct netdev_adjacent *upper; 6816 6817 ASSERT_RTNL(); 6818 6819 if (list_empty(&dev->adj_list.upper)) 6820 return NULL; 6821 6822 upper = list_first_entry(&dev->adj_list.upper, 6823 struct netdev_adjacent, list); 6824 if (likely(upper->master) && !upper->ignore) 6825 return upper->dev; 6826 return NULL; 6827 } 6828 6829 /** 6830 * netdev_has_any_lower_dev - Check if device is linked to some device 6831 * @dev: device 6832 * 6833 * Find out if a device is linked to a lower device and return true in case 6834 * it is. The caller must hold the RTNL lock. 6835 */ 6836 static bool netdev_has_any_lower_dev(struct net_device *dev) 6837 { 6838 ASSERT_RTNL(); 6839 6840 return !list_empty(&dev->adj_list.lower); 6841 } 6842 6843 void *netdev_adjacent_get_private(struct list_head *adj_list) 6844 { 6845 struct netdev_adjacent *adj; 6846 6847 adj = list_entry(adj_list, struct netdev_adjacent, list); 6848 6849 return adj->private; 6850 } 6851 EXPORT_SYMBOL(netdev_adjacent_get_private); 6852 6853 /** 6854 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6855 * @dev: device 6856 * @iter: list_head ** of the current position 6857 * 6858 * Gets the next device from the dev's upper list, starting from iter 6859 * position. The caller must hold RCU read lock. 6860 */ 6861 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6862 struct list_head **iter) 6863 { 6864 struct netdev_adjacent *upper; 6865 6866 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6867 6868 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6869 6870 if (&upper->list == &dev->adj_list.upper) 6871 return NULL; 6872 6873 *iter = &upper->list; 6874 6875 return upper->dev; 6876 } 6877 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6878 6879 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6880 struct list_head **iter, 6881 bool *ignore) 6882 { 6883 struct netdev_adjacent *upper; 6884 6885 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6886 6887 if (&upper->list == &dev->adj_list.upper) 6888 return NULL; 6889 6890 *iter = &upper->list; 6891 *ignore = upper->ignore; 6892 6893 return upper->dev; 6894 } 6895 6896 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6897 struct list_head **iter) 6898 { 6899 struct netdev_adjacent *upper; 6900 6901 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6902 6903 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6904 6905 if (&upper->list == &dev->adj_list.upper) 6906 return NULL; 6907 6908 *iter = &upper->list; 6909 6910 return upper->dev; 6911 } 6912 6913 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6914 int (*fn)(struct net_device *dev, 6915 struct netdev_nested_priv *priv), 6916 struct netdev_nested_priv *priv) 6917 { 6918 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6919 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6920 int ret, cur = 0; 6921 bool ignore; 6922 6923 now = dev; 6924 iter = &dev->adj_list.upper; 6925 6926 while (1) { 6927 if (now != dev) { 6928 ret = fn(now, priv); 6929 if (ret) 6930 return ret; 6931 } 6932 6933 next = NULL; 6934 while (1) { 6935 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6936 if (!udev) 6937 break; 6938 if (ignore) 6939 continue; 6940 6941 next = udev; 6942 niter = &udev->adj_list.upper; 6943 dev_stack[cur] = now; 6944 iter_stack[cur++] = iter; 6945 break; 6946 } 6947 6948 if (!next) { 6949 if (!cur) 6950 return 0; 6951 next = dev_stack[--cur]; 6952 niter = iter_stack[cur]; 6953 } 6954 6955 now = next; 6956 iter = niter; 6957 } 6958 6959 return 0; 6960 } 6961 6962 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6963 int (*fn)(struct net_device *dev, 6964 struct netdev_nested_priv *priv), 6965 struct netdev_nested_priv *priv) 6966 { 6967 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6968 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6969 int ret, cur = 0; 6970 6971 now = dev; 6972 iter = &dev->adj_list.upper; 6973 6974 while (1) { 6975 if (now != dev) { 6976 ret = fn(now, priv); 6977 if (ret) 6978 return ret; 6979 } 6980 6981 next = NULL; 6982 while (1) { 6983 udev = netdev_next_upper_dev_rcu(now, &iter); 6984 if (!udev) 6985 break; 6986 6987 next = udev; 6988 niter = &udev->adj_list.upper; 6989 dev_stack[cur] = now; 6990 iter_stack[cur++] = iter; 6991 break; 6992 } 6993 6994 if (!next) { 6995 if (!cur) 6996 return 0; 6997 next = dev_stack[--cur]; 6998 niter = iter_stack[cur]; 6999 } 7000 7001 now = next; 7002 iter = niter; 7003 } 7004 7005 return 0; 7006 } 7007 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7008 7009 static bool __netdev_has_upper_dev(struct net_device *dev, 7010 struct net_device *upper_dev) 7011 { 7012 struct netdev_nested_priv priv = { 7013 .flags = 0, 7014 .data = (void *)upper_dev, 7015 }; 7016 7017 ASSERT_RTNL(); 7018 7019 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7020 &priv); 7021 } 7022 7023 /** 7024 * netdev_lower_get_next_private - Get the next ->private from the 7025 * lower neighbour list 7026 * @dev: device 7027 * @iter: list_head ** of the current position 7028 * 7029 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7030 * list, starting from iter position. The caller must hold either hold the 7031 * RTNL lock or its own locking that guarantees that the neighbour lower 7032 * list will remain unchanged. 7033 */ 7034 void *netdev_lower_get_next_private(struct net_device *dev, 7035 struct list_head **iter) 7036 { 7037 struct netdev_adjacent *lower; 7038 7039 lower = list_entry(*iter, struct netdev_adjacent, list); 7040 7041 if (&lower->list == &dev->adj_list.lower) 7042 return NULL; 7043 7044 *iter = lower->list.next; 7045 7046 return lower->private; 7047 } 7048 EXPORT_SYMBOL(netdev_lower_get_next_private); 7049 7050 /** 7051 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7052 * lower neighbour list, RCU 7053 * variant 7054 * @dev: device 7055 * @iter: list_head ** of the current position 7056 * 7057 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7058 * list, starting from iter position. The caller must hold RCU read lock. 7059 */ 7060 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7061 struct list_head **iter) 7062 { 7063 struct netdev_adjacent *lower; 7064 7065 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7066 7067 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7068 7069 if (&lower->list == &dev->adj_list.lower) 7070 return NULL; 7071 7072 *iter = &lower->list; 7073 7074 return lower->private; 7075 } 7076 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7077 7078 /** 7079 * netdev_lower_get_next - Get the next device from the lower neighbour 7080 * list 7081 * @dev: device 7082 * @iter: list_head ** of the current position 7083 * 7084 * Gets the next netdev_adjacent from the dev's lower neighbour 7085 * list, starting from iter position. The caller must hold RTNL lock or 7086 * its own locking that guarantees that the neighbour lower 7087 * list will remain unchanged. 7088 */ 7089 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7090 { 7091 struct netdev_adjacent *lower; 7092 7093 lower = list_entry(*iter, struct netdev_adjacent, list); 7094 7095 if (&lower->list == &dev->adj_list.lower) 7096 return NULL; 7097 7098 *iter = lower->list.next; 7099 7100 return lower->dev; 7101 } 7102 EXPORT_SYMBOL(netdev_lower_get_next); 7103 7104 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7105 struct list_head **iter) 7106 { 7107 struct netdev_adjacent *lower; 7108 7109 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7110 7111 if (&lower->list == &dev->adj_list.lower) 7112 return NULL; 7113 7114 *iter = &lower->list; 7115 7116 return lower->dev; 7117 } 7118 7119 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7120 struct list_head **iter, 7121 bool *ignore) 7122 { 7123 struct netdev_adjacent *lower; 7124 7125 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7126 7127 if (&lower->list == &dev->adj_list.lower) 7128 return NULL; 7129 7130 *iter = &lower->list; 7131 *ignore = lower->ignore; 7132 7133 return lower->dev; 7134 } 7135 7136 int netdev_walk_all_lower_dev(struct net_device *dev, 7137 int (*fn)(struct net_device *dev, 7138 struct netdev_nested_priv *priv), 7139 struct netdev_nested_priv *priv) 7140 { 7141 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7142 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7143 int ret, cur = 0; 7144 7145 now = dev; 7146 iter = &dev->adj_list.lower; 7147 7148 while (1) { 7149 if (now != dev) { 7150 ret = fn(now, priv); 7151 if (ret) 7152 return ret; 7153 } 7154 7155 next = NULL; 7156 while (1) { 7157 ldev = netdev_next_lower_dev(now, &iter); 7158 if (!ldev) 7159 break; 7160 7161 next = ldev; 7162 niter = &ldev->adj_list.lower; 7163 dev_stack[cur] = now; 7164 iter_stack[cur++] = iter; 7165 break; 7166 } 7167 7168 if (!next) { 7169 if (!cur) 7170 return 0; 7171 next = dev_stack[--cur]; 7172 niter = iter_stack[cur]; 7173 } 7174 7175 now = next; 7176 iter = niter; 7177 } 7178 7179 return 0; 7180 } 7181 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7182 7183 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7184 int (*fn)(struct net_device *dev, 7185 struct netdev_nested_priv *priv), 7186 struct netdev_nested_priv *priv) 7187 { 7188 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7189 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7190 int ret, cur = 0; 7191 bool ignore; 7192 7193 now = dev; 7194 iter = &dev->adj_list.lower; 7195 7196 while (1) { 7197 if (now != dev) { 7198 ret = fn(now, priv); 7199 if (ret) 7200 return ret; 7201 } 7202 7203 next = NULL; 7204 while (1) { 7205 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7206 if (!ldev) 7207 break; 7208 if (ignore) 7209 continue; 7210 7211 next = ldev; 7212 niter = &ldev->adj_list.lower; 7213 dev_stack[cur] = now; 7214 iter_stack[cur++] = iter; 7215 break; 7216 } 7217 7218 if (!next) { 7219 if (!cur) 7220 return 0; 7221 next = dev_stack[--cur]; 7222 niter = iter_stack[cur]; 7223 } 7224 7225 now = next; 7226 iter = niter; 7227 } 7228 7229 return 0; 7230 } 7231 7232 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7233 struct list_head **iter) 7234 { 7235 struct netdev_adjacent *lower; 7236 7237 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7238 if (&lower->list == &dev->adj_list.lower) 7239 return NULL; 7240 7241 *iter = &lower->list; 7242 7243 return lower->dev; 7244 } 7245 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7246 7247 static u8 __netdev_upper_depth(struct net_device *dev) 7248 { 7249 struct net_device *udev; 7250 struct list_head *iter; 7251 u8 max_depth = 0; 7252 bool ignore; 7253 7254 for (iter = &dev->adj_list.upper, 7255 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7256 udev; 7257 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7258 if (ignore) 7259 continue; 7260 if (max_depth < udev->upper_level) 7261 max_depth = udev->upper_level; 7262 } 7263 7264 return max_depth; 7265 } 7266 7267 static u8 __netdev_lower_depth(struct net_device *dev) 7268 { 7269 struct net_device *ldev; 7270 struct list_head *iter; 7271 u8 max_depth = 0; 7272 bool ignore; 7273 7274 for (iter = &dev->adj_list.lower, 7275 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7276 ldev; 7277 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7278 if (ignore) 7279 continue; 7280 if (max_depth < ldev->lower_level) 7281 max_depth = ldev->lower_level; 7282 } 7283 7284 return max_depth; 7285 } 7286 7287 static int __netdev_update_upper_level(struct net_device *dev, 7288 struct netdev_nested_priv *__unused) 7289 { 7290 dev->upper_level = __netdev_upper_depth(dev) + 1; 7291 return 0; 7292 } 7293 7294 #ifdef CONFIG_LOCKDEP 7295 static LIST_HEAD(net_unlink_list); 7296 7297 static void net_unlink_todo(struct net_device *dev) 7298 { 7299 if (list_empty(&dev->unlink_list)) 7300 list_add_tail(&dev->unlink_list, &net_unlink_list); 7301 } 7302 #endif 7303 7304 static int __netdev_update_lower_level(struct net_device *dev, 7305 struct netdev_nested_priv *priv) 7306 { 7307 dev->lower_level = __netdev_lower_depth(dev) + 1; 7308 7309 #ifdef CONFIG_LOCKDEP 7310 if (!priv) 7311 return 0; 7312 7313 if (priv->flags & NESTED_SYNC_IMM) 7314 dev->nested_level = dev->lower_level - 1; 7315 if (priv->flags & NESTED_SYNC_TODO) 7316 net_unlink_todo(dev); 7317 #endif 7318 return 0; 7319 } 7320 7321 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7322 int (*fn)(struct net_device *dev, 7323 struct netdev_nested_priv *priv), 7324 struct netdev_nested_priv *priv) 7325 { 7326 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7327 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7328 int ret, cur = 0; 7329 7330 now = dev; 7331 iter = &dev->adj_list.lower; 7332 7333 while (1) { 7334 if (now != dev) { 7335 ret = fn(now, priv); 7336 if (ret) 7337 return ret; 7338 } 7339 7340 next = NULL; 7341 while (1) { 7342 ldev = netdev_next_lower_dev_rcu(now, &iter); 7343 if (!ldev) 7344 break; 7345 7346 next = ldev; 7347 niter = &ldev->adj_list.lower; 7348 dev_stack[cur] = now; 7349 iter_stack[cur++] = iter; 7350 break; 7351 } 7352 7353 if (!next) { 7354 if (!cur) 7355 return 0; 7356 next = dev_stack[--cur]; 7357 niter = iter_stack[cur]; 7358 } 7359 7360 now = next; 7361 iter = niter; 7362 } 7363 7364 return 0; 7365 } 7366 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7367 7368 /** 7369 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7370 * lower neighbour list, RCU 7371 * variant 7372 * @dev: device 7373 * 7374 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7375 * list. The caller must hold RCU read lock. 7376 */ 7377 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7378 { 7379 struct netdev_adjacent *lower; 7380 7381 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7382 struct netdev_adjacent, list); 7383 if (lower) 7384 return lower->private; 7385 return NULL; 7386 } 7387 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7388 7389 /** 7390 * netdev_master_upper_dev_get_rcu - Get master upper device 7391 * @dev: device 7392 * 7393 * Find a master upper device and return pointer to it or NULL in case 7394 * it's not there. The caller must hold the RCU read lock. 7395 */ 7396 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7397 { 7398 struct netdev_adjacent *upper; 7399 7400 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7401 struct netdev_adjacent, list); 7402 if (upper && likely(upper->master)) 7403 return upper->dev; 7404 return NULL; 7405 } 7406 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7407 7408 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7409 struct net_device *adj_dev, 7410 struct list_head *dev_list) 7411 { 7412 char linkname[IFNAMSIZ+7]; 7413 7414 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7415 "upper_%s" : "lower_%s", adj_dev->name); 7416 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7417 linkname); 7418 } 7419 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7420 char *name, 7421 struct list_head *dev_list) 7422 { 7423 char linkname[IFNAMSIZ+7]; 7424 7425 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7426 "upper_%s" : "lower_%s", name); 7427 sysfs_remove_link(&(dev->dev.kobj), linkname); 7428 } 7429 7430 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7431 struct net_device *adj_dev, 7432 struct list_head *dev_list) 7433 { 7434 return (dev_list == &dev->adj_list.upper || 7435 dev_list == &dev->adj_list.lower) && 7436 net_eq(dev_net(dev), dev_net(adj_dev)); 7437 } 7438 7439 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7440 struct net_device *adj_dev, 7441 struct list_head *dev_list, 7442 void *private, bool master) 7443 { 7444 struct netdev_adjacent *adj; 7445 int ret; 7446 7447 adj = __netdev_find_adj(adj_dev, dev_list); 7448 7449 if (adj) { 7450 adj->ref_nr += 1; 7451 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7452 dev->name, adj_dev->name, adj->ref_nr); 7453 7454 return 0; 7455 } 7456 7457 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7458 if (!adj) 7459 return -ENOMEM; 7460 7461 adj->dev = adj_dev; 7462 adj->master = master; 7463 adj->ref_nr = 1; 7464 adj->private = private; 7465 adj->ignore = false; 7466 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7467 7468 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7469 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7470 7471 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7472 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7473 if (ret) 7474 goto free_adj; 7475 } 7476 7477 /* Ensure that master link is always the first item in list. */ 7478 if (master) { 7479 ret = sysfs_create_link(&(dev->dev.kobj), 7480 &(adj_dev->dev.kobj), "master"); 7481 if (ret) 7482 goto remove_symlinks; 7483 7484 list_add_rcu(&adj->list, dev_list); 7485 } else { 7486 list_add_tail_rcu(&adj->list, dev_list); 7487 } 7488 7489 return 0; 7490 7491 remove_symlinks: 7492 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7493 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7494 free_adj: 7495 dev_put_track(adj_dev, &adj->dev_tracker); 7496 kfree(adj); 7497 7498 return ret; 7499 } 7500 7501 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7502 struct net_device *adj_dev, 7503 u16 ref_nr, 7504 struct list_head *dev_list) 7505 { 7506 struct netdev_adjacent *adj; 7507 7508 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7509 dev->name, adj_dev->name, ref_nr); 7510 7511 adj = __netdev_find_adj(adj_dev, dev_list); 7512 7513 if (!adj) { 7514 pr_err("Adjacency does not exist for device %s from %s\n", 7515 dev->name, adj_dev->name); 7516 WARN_ON(1); 7517 return; 7518 } 7519 7520 if (adj->ref_nr > ref_nr) { 7521 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7522 dev->name, adj_dev->name, ref_nr, 7523 adj->ref_nr - ref_nr); 7524 adj->ref_nr -= ref_nr; 7525 return; 7526 } 7527 7528 if (adj->master) 7529 sysfs_remove_link(&(dev->dev.kobj), "master"); 7530 7531 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7532 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7533 7534 list_del_rcu(&adj->list); 7535 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7536 adj_dev->name, dev->name, adj_dev->name); 7537 dev_put_track(adj_dev, &adj->dev_tracker); 7538 kfree_rcu(adj, rcu); 7539 } 7540 7541 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7542 struct net_device *upper_dev, 7543 struct list_head *up_list, 7544 struct list_head *down_list, 7545 void *private, bool master) 7546 { 7547 int ret; 7548 7549 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7550 private, master); 7551 if (ret) 7552 return ret; 7553 7554 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7555 private, false); 7556 if (ret) { 7557 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7558 return ret; 7559 } 7560 7561 return 0; 7562 } 7563 7564 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7565 struct net_device *upper_dev, 7566 u16 ref_nr, 7567 struct list_head *up_list, 7568 struct list_head *down_list) 7569 { 7570 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7571 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7572 } 7573 7574 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7575 struct net_device *upper_dev, 7576 void *private, bool master) 7577 { 7578 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7579 &dev->adj_list.upper, 7580 &upper_dev->adj_list.lower, 7581 private, master); 7582 } 7583 7584 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7585 struct net_device *upper_dev) 7586 { 7587 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7588 &dev->adj_list.upper, 7589 &upper_dev->adj_list.lower); 7590 } 7591 7592 static int __netdev_upper_dev_link(struct net_device *dev, 7593 struct net_device *upper_dev, bool master, 7594 void *upper_priv, void *upper_info, 7595 struct netdev_nested_priv *priv, 7596 struct netlink_ext_ack *extack) 7597 { 7598 struct netdev_notifier_changeupper_info changeupper_info = { 7599 .info = { 7600 .dev = dev, 7601 .extack = extack, 7602 }, 7603 .upper_dev = upper_dev, 7604 .master = master, 7605 .linking = true, 7606 .upper_info = upper_info, 7607 }; 7608 struct net_device *master_dev; 7609 int ret = 0; 7610 7611 ASSERT_RTNL(); 7612 7613 if (dev == upper_dev) 7614 return -EBUSY; 7615 7616 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7617 if (__netdev_has_upper_dev(upper_dev, dev)) 7618 return -EBUSY; 7619 7620 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7621 return -EMLINK; 7622 7623 if (!master) { 7624 if (__netdev_has_upper_dev(dev, upper_dev)) 7625 return -EEXIST; 7626 } else { 7627 master_dev = __netdev_master_upper_dev_get(dev); 7628 if (master_dev) 7629 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7630 } 7631 7632 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7633 &changeupper_info.info); 7634 ret = notifier_to_errno(ret); 7635 if (ret) 7636 return ret; 7637 7638 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7639 master); 7640 if (ret) 7641 return ret; 7642 7643 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7644 &changeupper_info.info); 7645 ret = notifier_to_errno(ret); 7646 if (ret) 7647 goto rollback; 7648 7649 __netdev_update_upper_level(dev, NULL); 7650 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7651 7652 __netdev_update_lower_level(upper_dev, priv); 7653 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7654 priv); 7655 7656 return 0; 7657 7658 rollback: 7659 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7660 7661 return ret; 7662 } 7663 7664 /** 7665 * netdev_upper_dev_link - Add a link to the upper device 7666 * @dev: device 7667 * @upper_dev: new upper device 7668 * @extack: netlink extended ack 7669 * 7670 * Adds a link to device which is upper to this one. The caller must hold 7671 * the RTNL lock. On a failure a negative errno code is returned. 7672 * On success the reference counts are adjusted and the function 7673 * returns zero. 7674 */ 7675 int netdev_upper_dev_link(struct net_device *dev, 7676 struct net_device *upper_dev, 7677 struct netlink_ext_ack *extack) 7678 { 7679 struct netdev_nested_priv priv = { 7680 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7681 .data = NULL, 7682 }; 7683 7684 return __netdev_upper_dev_link(dev, upper_dev, false, 7685 NULL, NULL, &priv, extack); 7686 } 7687 EXPORT_SYMBOL(netdev_upper_dev_link); 7688 7689 /** 7690 * netdev_master_upper_dev_link - Add a master link to the upper device 7691 * @dev: device 7692 * @upper_dev: new upper device 7693 * @upper_priv: upper device private 7694 * @upper_info: upper info to be passed down via notifier 7695 * @extack: netlink extended ack 7696 * 7697 * Adds a link to device which is upper to this one. In this case, only 7698 * one master upper device can be linked, although other non-master devices 7699 * might be linked as well. The caller must hold the RTNL lock. 7700 * On a failure a negative errno code is returned. On success the reference 7701 * counts are adjusted and the function returns zero. 7702 */ 7703 int netdev_master_upper_dev_link(struct net_device *dev, 7704 struct net_device *upper_dev, 7705 void *upper_priv, void *upper_info, 7706 struct netlink_ext_ack *extack) 7707 { 7708 struct netdev_nested_priv priv = { 7709 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7710 .data = NULL, 7711 }; 7712 7713 return __netdev_upper_dev_link(dev, upper_dev, true, 7714 upper_priv, upper_info, &priv, extack); 7715 } 7716 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7717 7718 static void __netdev_upper_dev_unlink(struct net_device *dev, 7719 struct net_device *upper_dev, 7720 struct netdev_nested_priv *priv) 7721 { 7722 struct netdev_notifier_changeupper_info changeupper_info = { 7723 .info = { 7724 .dev = dev, 7725 }, 7726 .upper_dev = upper_dev, 7727 .linking = false, 7728 }; 7729 7730 ASSERT_RTNL(); 7731 7732 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7733 7734 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7735 &changeupper_info.info); 7736 7737 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7738 7739 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7740 &changeupper_info.info); 7741 7742 __netdev_update_upper_level(dev, NULL); 7743 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7744 7745 __netdev_update_lower_level(upper_dev, priv); 7746 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7747 priv); 7748 } 7749 7750 /** 7751 * netdev_upper_dev_unlink - Removes a link to upper device 7752 * @dev: device 7753 * @upper_dev: new upper device 7754 * 7755 * Removes a link to device which is upper to this one. The caller must hold 7756 * the RTNL lock. 7757 */ 7758 void netdev_upper_dev_unlink(struct net_device *dev, 7759 struct net_device *upper_dev) 7760 { 7761 struct netdev_nested_priv priv = { 7762 .flags = NESTED_SYNC_TODO, 7763 .data = NULL, 7764 }; 7765 7766 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7767 } 7768 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7769 7770 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7771 struct net_device *lower_dev, 7772 bool val) 7773 { 7774 struct netdev_adjacent *adj; 7775 7776 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7777 if (adj) 7778 adj->ignore = val; 7779 7780 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7781 if (adj) 7782 adj->ignore = val; 7783 } 7784 7785 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7786 struct net_device *lower_dev) 7787 { 7788 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7789 } 7790 7791 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7792 struct net_device *lower_dev) 7793 { 7794 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7795 } 7796 7797 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7798 struct net_device *new_dev, 7799 struct net_device *dev, 7800 struct netlink_ext_ack *extack) 7801 { 7802 struct netdev_nested_priv priv = { 7803 .flags = 0, 7804 .data = NULL, 7805 }; 7806 int err; 7807 7808 if (!new_dev) 7809 return 0; 7810 7811 if (old_dev && new_dev != old_dev) 7812 netdev_adjacent_dev_disable(dev, old_dev); 7813 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7814 extack); 7815 if (err) { 7816 if (old_dev && new_dev != old_dev) 7817 netdev_adjacent_dev_enable(dev, old_dev); 7818 return err; 7819 } 7820 7821 return 0; 7822 } 7823 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7824 7825 void netdev_adjacent_change_commit(struct net_device *old_dev, 7826 struct net_device *new_dev, 7827 struct net_device *dev) 7828 { 7829 struct netdev_nested_priv priv = { 7830 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7831 .data = NULL, 7832 }; 7833 7834 if (!new_dev || !old_dev) 7835 return; 7836 7837 if (new_dev == old_dev) 7838 return; 7839 7840 netdev_adjacent_dev_enable(dev, old_dev); 7841 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7842 } 7843 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7844 7845 void netdev_adjacent_change_abort(struct net_device *old_dev, 7846 struct net_device *new_dev, 7847 struct net_device *dev) 7848 { 7849 struct netdev_nested_priv priv = { 7850 .flags = 0, 7851 .data = NULL, 7852 }; 7853 7854 if (!new_dev) 7855 return; 7856 7857 if (old_dev && new_dev != old_dev) 7858 netdev_adjacent_dev_enable(dev, old_dev); 7859 7860 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7861 } 7862 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7863 7864 /** 7865 * netdev_bonding_info_change - Dispatch event about slave change 7866 * @dev: device 7867 * @bonding_info: info to dispatch 7868 * 7869 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7870 * The caller must hold the RTNL lock. 7871 */ 7872 void netdev_bonding_info_change(struct net_device *dev, 7873 struct netdev_bonding_info *bonding_info) 7874 { 7875 struct netdev_notifier_bonding_info info = { 7876 .info.dev = dev, 7877 }; 7878 7879 memcpy(&info.bonding_info, bonding_info, 7880 sizeof(struct netdev_bonding_info)); 7881 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7882 &info.info); 7883 } 7884 EXPORT_SYMBOL(netdev_bonding_info_change); 7885 7886 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7887 struct netlink_ext_ack *extack) 7888 { 7889 struct netdev_notifier_offload_xstats_info info = { 7890 .info.dev = dev, 7891 .info.extack = extack, 7892 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7893 }; 7894 int err; 7895 int rc; 7896 7897 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7898 GFP_KERNEL); 7899 if (!dev->offload_xstats_l3) 7900 return -ENOMEM; 7901 7902 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7903 NETDEV_OFFLOAD_XSTATS_DISABLE, 7904 &info.info); 7905 err = notifier_to_errno(rc); 7906 if (err) 7907 goto free_stats; 7908 7909 return 0; 7910 7911 free_stats: 7912 kfree(dev->offload_xstats_l3); 7913 dev->offload_xstats_l3 = NULL; 7914 return err; 7915 } 7916 7917 int netdev_offload_xstats_enable(struct net_device *dev, 7918 enum netdev_offload_xstats_type type, 7919 struct netlink_ext_ack *extack) 7920 { 7921 ASSERT_RTNL(); 7922 7923 if (netdev_offload_xstats_enabled(dev, type)) 7924 return -EALREADY; 7925 7926 switch (type) { 7927 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7928 return netdev_offload_xstats_enable_l3(dev, extack); 7929 } 7930 7931 WARN_ON(1); 7932 return -EINVAL; 7933 } 7934 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7935 7936 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7937 { 7938 struct netdev_notifier_offload_xstats_info info = { 7939 .info.dev = dev, 7940 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7941 }; 7942 7943 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7944 &info.info); 7945 kfree(dev->offload_xstats_l3); 7946 dev->offload_xstats_l3 = NULL; 7947 } 7948 7949 int netdev_offload_xstats_disable(struct net_device *dev, 7950 enum netdev_offload_xstats_type type) 7951 { 7952 ASSERT_RTNL(); 7953 7954 if (!netdev_offload_xstats_enabled(dev, type)) 7955 return -EALREADY; 7956 7957 switch (type) { 7958 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7959 netdev_offload_xstats_disable_l3(dev); 7960 return 0; 7961 } 7962 7963 WARN_ON(1); 7964 return -EINVAL; 7965 } 7966 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7967 7968 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7969 { 7970 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7971 } 7972 7973 static struct rtnl_hw_stats64 * 7974 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7975 enum netdev_offload_xstats_type type) 7976 { 7977 switch (type) { 7978 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7979 return dev->offload_xstats_l3; 7980 } 7981 7982 WARN_ON(1); 7983 return NULL; 7984 } 7985 7986 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7987 enum netdev_offload_xstats_type type) 7988 { 7989 ASSERT_RTNL(); 7990 7991 return netdev_offload_xstats_get_ptr(dev, type); 7992 } 7993 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7994 7995 struct netdev_notifier_offload_xstats_ru { 7996 bool used; 7997 }; 7998 7999 struct netdev_notifier_offload_xstats_rd { 8000 struct rtnl_hw_stats64 stats; 8001 bool used; 8002 }; 8003 8004 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8005 const struct rtnl_hw_stats64 *src) 8006 { 8007 dest->rx_packets += src->rx_packets; 8008 dest->tx_packets += src->tx_packets; 8009 dest->rx_bytes += src->rx_bytes; 8010 dest->tx_bytes += src->tx_bytes; 8011 dest->rx_errors += src->rx_errors; 8012 dest->tx_errors += src->tx_errors; 8013 dest->rx_dropped += src->rx_dropped; 8014 dest->tx_dropped += src->tx_dropped; 8015 dest->multicast += src->multicast; 8016 } 8017 8018 static int netdev_offload_xstats_get_used(struct net_device *dev, 8019 enum netdev_offload_xstats_type type, 8020 bool *p_used, 8021 struct netlink_ext_ack *extack) 8022 { 8023 struct netdev_notifier_offload_xstats_ru report_used = {}; 8024 struct netdev_notifier_offload_xstats_info info = { 8025 .info.dev = dev, 8026 .info.extack = extack, 8027 .type = type, 8028 .report_used = &report_used, 8029 }; 8030 int rc; 8031 8032 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8033 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8034 &info.info); 8035 *p_used = report_used.used; 8036 return notifier_to_errno(rc); 8037 } 8038 8039 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8040 enum netdev_offload_xstats_type type, 8041 struct rtnl_hw_stats64 *p_stats, 8042 bool *p_used, 8043 struct netlink_ext_ack *extack) 8044 { 8045 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8046 struct netdev_notifier_offload_xstats_info info = { 8047 .info.dev = dev, 8048 .info.extack = extack, 8049 .type = type, 8050 .report_delta = &report_delta, 8051 }; 8052 struct rtnl_hw_stats64 *stats; 8053 int rc; 8054 8055 stats = netdev_offload_xstats_get_ptr(dev, type); 8056 if (WARN_ON(!stats)) 8057 return -EINVAL; 8058 8059 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8060 &info.info); 8061 8062 /* Cache whatever we got, even if there was an error, otherwise the 8063 * successful stats retrievals would get lost. 8064 */ 8065 netdev_hw_stats64_add(stats, &report_delta.stats); 8066 8067 if (p_stats) 8068 *p_stats = *stats; 8069 *p_used = report_delta.used; 8070 8071 return notifier_to_errno(rc); 8072 } 8073 8074 int netdev_offload_xstats_get(struct net_device *dev, 8075 enum netdev_offload_xstats_type type, 8076 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8077 struct netlink_ext_ack *extack) 8078 { 8079 ASSERT_RTNL(); 8080 8081 if (p_stats) 8082 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8083 p_used, extack); 8084 else 8085 return netdev_offload_xstats_get_used(dev, type, p_used, 8086 extack); 8087 } 8088 EXPORT_SYMBOL(netdev_offload_xstats_get); 8089 8090 void 8091 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8092 const struct rtnl_hw_stats64 *stats) 8093 { 8094 report_delta->used = true; 8095 netdev_hw_stats64_add(&report_delta->stats, stats); 8096 } 8097 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8098 8099 void 8100 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8101 { 8102 report_used->used = true; 8103 } 8104 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8105 8106 void netdev_offload_xstats_push_delta(struct net_device *dev, 8107 enum netdev_offload_xstats_type type, 8108 const struct rtnl_hw_stats64 *p_stats) 8109 { 8110 struct rtnl_hw_stats64 *stats; 8111 8112 ASSERT_RTNL(); 8113 8114 stats = netdev_offload_xstats_get_ptr(dev, type); 8115 if (WARN_ON(!stats)) 8116 return; 8117 8118 netdev_hw_stats64_add(stats, p_stats); 8119 } 8120 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8121 8122 /** 8123 * netdev_get_xmit_slave - Get the xmit slave of master device 8124 * @dev: device 8125 * @skb: The packet 8126 * @all_slaves: assume all the slaves are active 8127 * 8128 * The reference counters are not incremented so the caller must be 8129 * careful with locks. The caller must hold RCU lock. 8130 * %NULL is returned if no slave is found. 8131 */ 8132 8133 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8134 struct sk_buff *skb, 8135 bool all_slaves) 8136 { 8137 const struct net_device_ops *ops = dev->netdev_ops; 8138 8139 if (!ops->ndo_get_xmit_slave) 8140 return NULL; 8141 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8142 } 8143 EXPORT_SYMBOL(netdev_get_xmit_slave); 8144 8145 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8146 struct sock *sk) 8147 { 8148 const struct net_device_ops *ops = dev->netdev_ops; 8149 8150 if (!ops->ndo_sk_get_lower_dev) 8151 return NULL; 8152 return ops->ndo_sk_get_lower_dev(dev, sk); 8153 } 8154 8155 /** 8156 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8157 * @dev: device 8158 * @sk: the socket 8159 * 8160 * %NULL is returned if no lower device is found. 8161 */ 8162 8163 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8164 struct sock *sk) 8165 { 8166 struct net_device *lower; 8167 8168 lower = netdev_sk_get_lower_dev(dev, sk); 8169 while (lower) { 8170 dev = lower; 8171 lower = netdev_sk_get_lower_dev(dev, sk); 8172 } 8173 8174 return dev; 8175 } 8176 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8177 8178 static void netdev_adjacent_add_links(struct net_device *dev) 8179 { 8180 struct netdev_adjacent *iter; 8181 8182 struct net *net = dev_net(dev); 8183 8184 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8185 if (!net_eq(net, dev_net(iter->dev))) 8186 continue; 8187 netdev_adjacent_sysfs_add(iter->dev, dev, 8188 &iter->dev->adj_list.lower); 8189 netdev_adjacent_sysfs_add(dev, iter->dev, 8190 &dev->adj_list.upper); 8191 } 8192 8193 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8194 if (!net_eq(net, dev_net(iter->dev))) 8195 continue; 8196 netdev_adjacent_sysfs_add(iter->dev, dev, 8197 &iter->dev->adj_list.upper); 8198 netdev_adjacent_sysfs_add(dev, iter->dev, 8199 &dev->adj_list.lower); 8200 } 8201 } 8202 8203 static void netdev_adjacent_del_links(struct net_device *dev) 8204 { 8205 struct netdev_adjacent *iter; 8206 8207 struct net *net = dev_net(dev); 8208 8209 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8210 if (!net_eq(net, dev_net(iter->dev))) 8211 continue; 8212 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8213 &iter->dev->adj_list.lower); 8214 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8215 &dev->adj_list.upper); 8216 } 8217 8218 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8219 if (!net_eq(net, dev_net(iter->dev))) 8220 continue; 8221 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8222 &iter->dev->adj_list.upper); 8223 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8224 &dev->adj_list.lower); 8225 } 8226 } 8227 8228 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8229 { 8230 struct netdev_adjacent *iter; 8231 8232 struct net *net = dev_net(dev); 8233 8234 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8235 if (!net_eq(net, dev_net(iter->dev))) 8236 continue; 8237 netdev_adjacent_sysfs_del(iter->dev, oldname, 8238 &iter->dev->adj_list.lower); 8239 netdev_adjacent_sysfs_add(iter->dev, dev, 8240 &iter->dev->adj_list.lower); 8241 } 8242 8243 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8244 if (!net_eq(net, dev_net(iter->dev))) 8245 continue; 8246 netdev_adjacent_sysfs_del(iter->dev, oldname, 8247 &iter->dev->adj_list.upper); 8248 netdev_adjacent_sysfs_add(iter->dev, dev, 8249 &iter->dev->adj_list.upper); 8250 } 8251 } 8252 8253 void *netdev_lower_dev_get_private(struct net_device *dev, 8254 struct net_device *lower_dev) 8255 { 8256 struct netdev_adjacent *lower; 8257 8258 if (!lower_dev) 8259 return NULL; 8260 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8261 if (!lower) 8262 return NULL; 8263 8264 return lower->private; 8265 } 8266 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8267 8268 8269 /** 8270 * netdev_lower_state_changed - Dispatch event about lower device state change 8271 * @lower_dev: device 8272 * @lower_state_info: state to dispatch 8273 * 8274 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8275 * The caller must hold the RTNL lock. 8276 */ 8277 void netdev_lower_state_changed(struct net_device *lower_dev, 8278 void *lower_state_info) 8279 { 8280 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8281 .info.dev = lower_dev, 8282 }; 8283 8284 ASSERT_RTNL(); 8285 changelowerstate_info.lower_state_info = lower_state_info; 8286 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8287 &changelowerstate_info.info); 8288 } 8289 EXPORT_SYMBOL(netdev_lower_state_changed); 8290 8291 static void dev_change_rx_flags(struct net_device *dev, int flags) 8292 { 8293 const struct net_device_ops *ops = dev->netdev_ops; 8294 8295 if (ops->ndo_change_rx_flags) 8296 ops->ndo_change_rx_flags(dev, flags); 8297 } 8298 8299 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8300 { 8301 unsigned int old_flags = dev->flags; 8302 kuid_t uid; 8303 kgid_t gid; 8304 8305 ASSERT_RTNL(); 8306 8307 dev->flags |= IFF_PROMISC; 8308 dev->promiscuity += inc; 8309 if (dev->promiscuity == 0) { 8310 /* 8311 * Avoid overflow. 8312 * If inc causes overflow, untouch promisc and return error. 8313 */ 8314 if (inc < 0) 8315 dev->flags &= ~IFF_PROMISC; 8316 else { 8317 dev->promiscuity -= inc; 8318 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8319 return -EOVERFLOW; 8320 } 8321 } 8322 if (dev->flags != old_flags) { 8323 pr_info("device %s %s promiscuous mode\n", 8324 dev->name, 8325 dev->flags & IFF_PROMISC ? "entered" : "left"); 8326 if (audit_enabled) { 8327 current_uid_gid(&uid, &gid); 8328 audit_log(audit_context(), GFP_ATOMIC, 8329 AUDIT_ANOM_PROMISCUOUS, 8330 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8331 dev->name, (dev->flags & IFF_PROMISC), 8332 (old_flags & IFF_PROMISC), 8333 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8334 from_kuid(&init_user_ns, uid), 8335 from_kgid(&init_user_ns, gid), 8336 audit_get_sessionid(current)); 8337 } 8338 8339 dev_change_rx_flags(dev, IFF_PROMISC); 8340 } 8341 if (notify) 8342 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8343 return 0; 8344 } 8345 8346 /** 8347 * dev_set_promiscuity - update promiscuity count on a device 8348 * @dev: device 8349 * @inc: modifier 8350 * 8351 * Add or remove promiscuity from a device. While the count in the device 8352 * remains above zero the interface remains promiscuous. Once it hits zero 8353 * the device reverts back to normal filtering operation. A negative inc 8354 * value is used to drop promiscuity on the device. 8355 * Return 0 if successful or a negative errno code on error. 8356 */ 8357 int dev_set_promiscuity(struct net_device *dev, int inc) 8358 { 8359 unsigned int old_flags = dev->flags; 8360 int err; 8361 8362 err = __dev_set_promiscuity(dev, inc, true); 8363 if (err < 0) 8364 return err; 8365 if (dev->flags != old_flags) 8366 dev_set_rx_mode(dev); 8367 return err; 8368 } 8369 EXPORT_SYMBOL(dev_set_promiscuity); 8370 8371 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8372 { 8373 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8374 8375 ASSERT_RTNL(); 8376 8377 dev->flags |= IFF_ALLMULTI; 8378 dev->allmulti += inc; 8379 if (dev->allmulti == 0) { 8380 /* 8381 * Avoid overflow. 8382 * If inc causes overflow, untouch allmulti and return error. 8383 */ 8384 if (inc < 0) 8385 dev->flags &= ~IFF_ALLMULTI; 8386 else { 8387 dev->allmulti -= inc; 8388 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8389 return -EOVERFLOW; 8390 } 8391 } 8392 if (dev->flags ^ old_flags) { 8393 dev_change_rx_flags(dev, IFF_ALLMULTI); 8394 dev_set_rx_mode(dev); 8395 if (notify) 8396 __dev_notify_flags(dev, old_flags, 8397 dev->gflags ^ old_gflags); 8398 } 8399 return 0; 8400 } 8401 8402 /** 8403 * dev_set_allmulti - update allmulti count on a device 8404 * @dev: device 8405 * @inc: modifier 8406 * 8407 * Add or remove reception of all multicast frames to a device. While the 8408 * count in the device remains above zero the interface remains listening 8409 * to all interfaces. Once it hits zero the device reverts back to normal 8410 * filtering operation. A negative @inc value is used to drop the counter 8411 * when releasing a resource needing all multicasts. 8412 * Return 0 if successful or a negative errno code on error. 8413 */ 8414 8415 int dev_set_allmulti(struct net_device *dev, int inc) 8416 { 8417 return __dev_set_allmulti(dev, inc, true); 8418 } 8419 EXPORT_SYMBOL(dev_set_allmulti); 8420 8421 /* 8422 * Upload unicast and multicast address lists to device and 8423 * configure RX filtering. When the device doesn't support unicast 8424 * filtering it is put in promiscuous mode while unicast addresses 8425 * are present. 8426 */ 8427 void __dev_set_rx_mode(struct net_device *dev) 8428 { 8429 const struct net_device_ops *ops = dev->netdev_ops; 8430 8431 /* dev_open will call this function so the list will stay sane. */ 8432 if (!(dev->flags&IFF_UP)) 8433 return; 8434 8435 if (!netif_device_present(dev)) 8436 return; 8437 8438 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8439 /* Unicast addresses changes may only happen under the rtnl, 8440 * therefore calling __dev_set_promiscuity here is safe. 8441 */ 8442 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8443 __dev_set_promiscuity(dev, 1, false); 8444 dev->uc_promisc = true; 8445 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8446 __dev_set_promiscuity(dev, -1, false); 8447 dev->uc_promisc = false; 8448 } 8449 } 8450 8451 if (ops->ndo_set_rx_mode) 8452 ops->ndo_set_rx_mode(dev); 8453 } 8454 8455 void dev_set_rx_mode(struct net_device *dev) 8456 { 8457 netif_addr_lock_bh(dev); 8458 __dev_set_rx_mode(dev); 8459 netif_addr_unlock_bh(dev); 8460 } 8461 8462 /** 8463 * dev_get_flags - get flags reported to userspace 8464 * @dev: device 8465 * 8466 * Get the combination of flag bits exported through APIs to userspace. 8467 */ 8468 unsigned int dev_get_flags(const struct net_device *dev) 8469 { 8470 unsigned int flags; 8471 8472 flags = (dev->flags & ~(IFF_PROMISC | 8473 IFF_ALLMULTI | 8474 IFF_RUNNING | 8475 IFF_LOWER_UP | 8476 IFF_DORMANT)) | 8477 (dev->gflags & (IFF_PROMISC | 8478 IFF_ALLMULTI)); 8479 8480 if (netif_running(dev)) { 8481 if (netif_oper_up(dev)) 8482 flags |= IFF_RUNNING; 8483 if (netif_carrier_ok(dev)) 8484 flags |= IFF_LOWER_UP; 8485 if (netif_dormant(dev)) 8486 flags |= IFF_DORMANT; 8487 } 8488 8489 return flags; 8490 } 8491 EXPORT_SYMBOL(dev_get_flags); 8492 8493 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8494 struct netlink_ext_ack *extack) 8495 { 8496 unsigned int old_flags = dev->flags; 8497 int ret; 8498 8499 ASSERT_RTNL(); 8500 8501 /* 8502 * Set the flags on our device. 8503 */ 8504 8505 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8506 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8507 IFF_AUTOMEDIA)) | 8508 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8509 IFF_ALLMULTI)); 8510 8511 /* 8512 * Load in the correct multicast list now the flags have changed. 8513 */ 8514 8515 if ((old_flags ^ flags) & IFF_MULTICAST) 8516 dev_change_rx_flags(dev, IFF_MULTICAST); 8517 8518 dev_set_rx_mode(dev); 8519 8520 /* 8521 * Have we downed the interface. We handle IFF_UP ourselves 8522 * according to user attempts to set it, rather than blindly 8523 * setting it. 8524 */ 8525 8526 ret = 0; 8527 if ((old_flags ^ flags) & IFF_UP) { 8528 if (old_flags & IFF_UP) 8529 __dev_close(dev); 8530 else 8531 ret = __dev_open(dev, extack); 8532 } 8533 8534 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8535 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8536 unsigned int old_flags = dev->flags; 8537 8538 dev->gflags ^= IFF_PROMISC; 8539 8540 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8541 if (dev->flags != old_flags) 8542 dev_set_rx_mode(dev); 8543 } 8544 8545 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8546 * is important. Some (broken) drivers set IFF_PROMISC, when 8547 * IFF_ALLMULTI is requested not asking us and not reporting. 8548 */ 8549 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8550 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8551 8552 dev->gflags ^= IFF_ALLMULTI; 8553 __dev_set_allmulti(dev, inc, false); 8554 } 8555 8556 return ret; 8557 } 8558 8559 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8560 unsigned int gchanges) 8561 { 8562 unsigned int changes = dev->flags ^ old_flags; 8563 8564 if (gchanges) 8565 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8566 8567 if (changes & IFF_UP) { 8568 if (dev->flags & IFF_UP) 8569 call_netdevice_notifiers(NETDEV_UP, dev); 8570 else 8571 call_netdevice_notifiers(NETDEV_DOWN, dev); 8572 } 8573 8574 if (dev->flags & IFF_UP && 8575 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8576 struct netdev_notifier_change_info change_info = { 8577 .info = { 8578 .dev = dev, 8579 }, 8580 .flags_changed = changes, 8581 }; 8582 8583 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8584 } 8585 } 8586 8587 /** 8588 * dev_change_flags - change device settings 8589 * @dev: device 8590 * @flags: device state flags 8591 * @extack: netlink extended ack 8592 * 8593 * Change settings on device based state flags. The flags are 8594 * in the userspace exported format. 8595 */ 8596 int dev_change_flags(struct net_device *dev, unsigned int flags, 8597 struct netlink_ext_ack *extack) 8598 { 8599 int ret; 8600 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8601 8602 ret = __dev_change_flags(dev, flags, extack); 8603 if (ret < 0) 8604 return ret; 8605 8606 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8607 __dev_notify_flags(dev, old_flags, changes); 8608 return ret; 8609 } 8610 EXPORT_SYMBOL(dev_change_flags); 8611 8612 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8613 { 8614 const struct net_device_ops *ops = dev->netdev_ops; 8615 8616 if (ops->ndo_change_mtu) 8617 return ops->ndo_change_mtu(dev, new_mtu); 8618 8619 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8620 WRITE_ONCE(dev->mtu, new_mtu); 8621 return 0; 8622 } 8623 EXPORT_SYMBOL(__dev_set_mtu); 8624 8625 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8626 struct netlink_ext_ack *extack) 8627 { 8628 /* MTU must be positive, and in range */ 8629 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8630 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8631 return -EINVAL; 8632 } 8633 8634 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8635 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8636 return -EINVAL; 8637 } 8638 return 0; 8639 } 8640 8641 /** 8642 * dev_set_mtu_ext - Change maximum transfer unit 8643 * @dev: device 8644 * @new_mtu: new transfer unit 8645 * @extack: netlink extended ack 8646 * 8647 * Change the maximum transfer size of the network device. 8648 */ 8649 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8650 struct netlink_ext_ack *extack) 8651 { 8652 int err, orig_mtu; 8653 8654 if (new_mtu == dev->mtu) 8655 return 0; 8656 8657 err = dev_validate_mtu(dev, new_mtu, extack); 8658 if (err) 8659 return err; 8660 8661 if (!netif_device_present(dev)) 8662 return -ENODEV; 8663 8664 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8665 err = notifier_to_errno(err); 8666 if (err) 8667 return err; 8668 8669 orig_mtu = dev->mtu; 8670 err = __dev_set_mtu(dev, new_mtu); 8671 8672 if (!err) { 8673 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8674 orig_mtu); 8675 err = notifier_to_errno(err); 8676 if (err) { 8677 /* setting mtu back and notifying everyone again, 8678 * so that they have a chance to revert changes. 8679 */ 8680 __dev_set_mtu(dev, orig_mtu); 8681 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8682 new_mtu); 8683 } 8684 } 8685 return err; 8686 } 8687 8688 int dev_set_mtu(struct net_device *dev, int new_mtu) 8689 { 8690 struct netlink_ext_ack extack; 8691 int err; 8692 8693 memset(&extack, 0, sizeof(extack)); 8694 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8695 if (err && extack._msg) 8696 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8697 return err; 8698 } 8699 EXPORT_SYMBOL(dev_set_mtu); 8700 8701 /** 8702 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8703 * @dev: device 8704 * @new_len: new tx queue length 8705 */ 8706 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8707 { 8708 unsigned int orig_len = dev->tx_queue_len; 8709 int res; 8710 8711 if (new_len != (unsigned int)new_len) 8712 return -ERANGE; 8713 8714 if (new_len != orig_len) { 8715 dev->tx_queue_len = new_len; 8716 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8717 res = notifier_to_errno(res); 8718 if (res) 8719 goto err_rollback; 8720 res = dev_qdisc_change_tx_queue_len(dev); 8721 if (res) 8722 goto err_rollback; 8723 } 8724 8725 return 0; 8726 8727 err_rollback: 8728 netdev_err(dev, "refused to change device tx_queue_len\n"); 8729 dev->tx_queue_len = orig_len; 8730 return res; 8731 } 8732 8733 /** 8734 * dev_set_group - Change group this device belongs to 8735 * @dev: device 8736 * @new_group: group this device should belong to 8737 */ 8738 void dev_set_group(struct net_device *dev, int new_group) 8739 { 8740 dev->group = new_group; 8741 } 8742 8743 /** 8744 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8745 * @dev: device 8746 * @addr: new address 8747 * @extack: netlink extended ack 8748 */ 8749 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8750 struct netlink_ext_ack *extack) 8751 { 8752 struct netdev_notifier_pre_changeaddr_info info = { 8753 .info.dev = dev, 8754 .info.extack = extack, 8755 .dev_addr = addr, 8756 }; 8757 int rc; 8758 8759 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8760 return notifier_to_errno(rc); 8761 } 8762 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8763 8764 /** 8765 * dev_set_mac_address - Change Media Access Control Address 8766 * @dev: device 8767 * @sa: new address 8768 * @extack: netlink extended ack 8769 * 8770 * Change the hardware (MAC) address of the device 8771 */ 8772 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8773 struct netlink_ext_ack *extack) 8774 { 8775 const struct net_device_ops *ops = dev->netdev_ops; 8776 int err; 8777 8778 if (!ops->ndo_set_mac_address) 8779 return -EOPNOTSUPP; 8780 if (sa->sa_family != dev->type) 8781 return -EINVAL; 8782 if (!netif_device_present(dev)) 8783 return -ENODEV; 8784 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8785 if (err) 8786 return err; 8787 err = ops->ndo_set_mac_address(dev, sa); 8788 if (err) 8789 return err; 8790 dev->addr_assign_type = NET_ADDR_SET; 8791 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8792 add_device_randomness(dev->dev_addr, dev->addr_len); 8793 return 0; 8794 } 8795 EXPORT_SYMBOL(dev_set_mac_address); 8796 8797 static DECLARE_RWSEM(dev_addr_sem); 8798 8799 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8800 struct netlink_ext_ack *extack) 8801 { 8802 int ret; 8803 8804 down_write(&dev_addr_sem); 8805 ret = dev_set_mac_address(dev, sa, extack); 8806 up_write(&dev_addr_sem); 8807 return ret; 8808 } 8809 EXPORT_SYMBOL(dev_set_mac_address_user); 8810 8811 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8812 { 8813 size_t size = sizeof(sa->sa_data); 8814 struct net_device *dev; 8815 int ret = 0; 8816 8817 down_read(&dev_addr_sem); 8818 rcu_read_lock(); 8819 8820 dev = dev_get_by_name_rcu(net, dev_name); 8821 if (!dev) { 8822 ret = -ENODEV; 8823 goto unlock; 8824 } 8825 if (!dev->addr_len) 8826 memset(sa->sa_data, 0, size); 8827 else 8828 memcpy(sa->sa_data, dev->dev_addr, 8829 min_t(size_t, size, dev->addr_len)); 8830 sa->sa_family = dev->type; 8831 8832 unlock: 8833 rcu_read_unlock(); 8834 up_read(&dev_addr_sem); 8835 return ret; 8836 } 8837 EXPORT_SYMBOL(dev_get_mac_address); 8838 8839 /** 8840 * dev_change_carrier - Change device carrier 8841 * @dev: device 8842 * @new_carrier: new value 8843 * 8844 * Change device carrier 8845 */ 8846 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8847 { 8848 const struct net_device_ops *ops = dev->netdev_ops; 8849 8850 if (!ops->ndo_change_carrier) 8851 return -EOPNOTSUPP; 8852 if (!netif_device_present(dev)) 8853 return -ENODEV; 8854 return ops->ndo_change_carrier(dev, new_carrier); 8855 } 8856 8857 /** 8858 * dev_get_phys_port_id - Get device physical port ID 8859 * @dev: device 8860 * @ppid: port ID 8861 * 8862 * Get device physical port ID 8863 */ 8864 int dev_get_phys_port_id(struct net_device *dev, 8865 struct netdev_phys_item_id *ppid) 8866 { 8867 const struct net_device_ops *ops = dev->netdev_ops; 8868 8869 if (!ops->ndo_get_phys_port_id) 8870 return -EOPNOTSUPP; 8871 return ops->ndo_get_phys_port_id(dev, ppid); 8872 } 8873 8874 /** 8875 * dev_get_phys_port_name - Get device physical port name 8876 * @dev: device 8877 * @name: port name 8878 * @len: limit of bytes to copy to name 8879 * 8880 * Get device physical port name 8881 */ 8882 int dev_get_phys_port_name(struct net_device *dev, 8883 char *name, size_t len) 8884 { 8885 const struct net_device_ops *ops = dev->netdev_ops; 8886 int err; 8887 8888 if (ops->ndo_get_phys_port_name) { 8889 err = ops->ndo_get_phys_port_name(dev, name, len); 8890 if (err != -EOPNOTSUPP) 8891 return err; 8892 } 8893 return devlink_compat_phys_port_name_get(dev, name, len); 8894 } 8895 8896 /** 8897 * dev_get_port_parent_id - Get the device's port parent identifier 8898 * @dev: network device 8899 * @ppid: pointer to a storage for the port's parent identifier 8900 * @recurse: allow/disallow recursion to lower devices 8901 * 8902 * Get the devices's port parent identifier 8903 */ 8904 int dev_get_port_parent_id(struct net_device *dev, 8905 struct netdev_phys_item_id *ppid, 8906 bool recurse) 8907 { 8908 const struct net_device_ops *ops = dev->netdev_ops; 8909 struct netdev_phys_item_id first = { }; 8910 struct net_device *lower_dev; 8911 struct list_head *iter; 8912 int err; 8913 8914 if (ops->ndo_get_port_parent_id) { 8915 err = ops->ndo_get_port_parent_id(dev, ppid); 8916 if (err != -EOPNOTSUPP) 8917 return err; 8918 } 8919 8920 err = devlink_compat_switch_id_get(dev, ppid); 8921 if (!recurse || err != -EOPNOTSUPP) 8922 return err; 8923 8924 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8925 err = dev_get_port_parent_id(lower_dev, ppid, true); 8926 if (err) 8927 break; 8928 if (!first.id_len) 8929 first = *ppid; 8930 else if (memcmp(&first, ppid, sizeof(*ppid))) 8931 return -EOPNOTSUPP; 8932 } 8933 8934 return err; 8935 } 8936 EXPORT_SYMBOL(dev_get_port_parent_id); 8937 8938 /** 8939 * netdev_port_same_parent_id - Indicate if two network devices have 8940 * the same port parent identifier 8941 * @a: first network device 8942 * @b: second network device 8943 */ 8944 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8945 { 8946 struct netdev_phys_item_id a_id = { }; 8947 struct netdev_phys_item_id b_id = { }; 8948 8949 if (dev_get_port_parent_id(a, &a_id, true) || 8950 dev_get_port_parent_id(b, &b_id, true)) 8951 return false; 8952 8953 return netdev_phys_item_id_same(&a_id, &b_id); 8954 } 8955 EXPORT_SYMBOL(netdev_port_same_parent_id); 8956 8957 /** 8958 * dev_change_proto_down - set carrier according to proto_down. 8959 * 8960 * @dev: device 8961 * @proto_down: new value 8962 */ 8963 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8964 { 8965 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8966 return -EOPNOTSUPP; 8967 if (!netif_device_present(dev)) 8968 return -ENODEV; 8969 if (proto_down) 8970 netif_carrier_off(dev); 8971 else 8972 netif_carrier_on(dev); 8973 dev->proto_down = proto_down; 8974 return 0; 8975 } 8976 8977 /** 8978 * dev_change_proto_down_reason - proto down reason 8979 * 8980 * @dev: device 8981 * @mask: proto down mask 8982 * @value: proto down value 8983 */ 8984 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8985 u32 value) 8986 { 8987 int b; 8988 8989 if (!mask) { 8990 dev->proto_down_reason = value; 8991 } else { 8992 for_each_set_bit(b, &mask, 32) { 8993 if (value & (1 << b)) 8994 dev->proto_down_reason |= BIT(b); 8995 else 8996 dev->proto_down_reason &= ~BIT(b); 8997 } 8998 } 8999 } 9000 9001 struct bpf_xdp_link { 9002 struct bpf_link link; 9003 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9004 int flags; 9005 }; 9006 9007 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9008 { 9009 if (flags & XDP_FLAGS_HW_MODE) 9010 return XDP_MODE_HW; 9011 if (flags & XDP_FLAGS_DRV_MODE) 9012 return XDP_MODE_DRV; 9013 if (flags & XDP_FLAGS_SKB_MODE) 9014 return XDP_MODE_SKB; 9015 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9016 } 9017 9018 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9019 { 9020 switch (mode) { 9021 case XDP_MODE_SKB: 9022 return generic_xdp_install; 9023 case XDP_MODE_DRV: 9024 case XDP_MODE_HW: 9025 return dev->netdev_ops->ndo_bpf; 9026 default: 9027 return NULL; 9028 } 9029 } 9030 9031 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9032 enum bpf_xdp_mode mode) 9033 { 9034 return dev->xdp_state[mode].link; 9035 } 9036 9037 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9038 enum bpf_xdp_mode mode) 9039 { 9040 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9041 9042 if (link) 9043 return link->link.prog; 9044 return dev->xdp_state[mode].prog; 9045 } 9046 9047 u8 dev_xdp_prog_count(struct net_device *dev) 9048 { 9049 u8 count = 0; 9050 int i; 9051 9052 for (i = 0; i < __MAX_XDP_MODE; i++) 9053 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9054 count++; 9055 return count; 9056 } 9057 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9058 9059 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9060 { 9061 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9062 9063 return prog ? prog->aux->id : 0; 9064 } 9065 9066 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9067 struct bpf_xdp_link *link) 9068 { 9069 dev->xdp_state[mode].link = link; 9070 dev->xdp_state[mode].prog = NULL; 9071 } 9072 9073 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9074 struct bpf_prog *prog) 9075 { 9076 dev->xdp_state[mode].link = NULL; 9077 dev->xdp_state[mode].prog = prog; 9078 } 9079 9080 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9081 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9082 u32 flags, struct bpf_prog *prog) 9083 { 9084 struct netdev_bpf xdp; 9085 int err; 9086 9087 memset(&xdp, 0, sizeof(xdp)); 9088 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9089 xdp.extack = extack; 9090 xdp.flags = flags; 9091 xdp.prog = prog; 9092 9093 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9094 * "moved" into driver), so they don't increment it on their own, but 9095 * they do decrement refcnt when program is detached or replaced. 9096 * Given net_device also owns link/prog, we need to bump refcnt here 9097 * to prevent drivers from underflowing it. 9098 */ 9099 if (prog) 9100 bpf_prog_inc(prog); 9101 err = bpf_op(dev, &xdp); 9102 if (err) { 9103 if (prog) 9104 bpf_prog_put(prog); 9105 return err; 9106 } 9107 9108 if (mode != XDP_MODE_HW) 9109 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9110 9111 return 0; 9112 } 9113 9114 static void dev_xdp_uninstall(struct net_device *dev) 9115 { 9116 struct bpf_xdp_link *link; 9117 struct bpf_prog *prog; 9118 enum bpf_xdp_mode mode; 9119 bpf_op_t bpf_op; 9120 9121 ASSERT_RTNL(); 9122 9123 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9124 prog = dev_xdp_prog(dev, mode); 9125 if (!prog) 9126 continue; 9127 9128 bpf_op = dev_xdp_bpf_op(dev, mode); 9129 if (!bpf_op) 9130 continue; 9131 9132 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9133 9134 /* auto-detach link from net device */ 9135 link = dev_xdp_link(dev, mode); 9136 if (link) 9137 link->dev = NULL; 9138 else 9139 bpf_prog_put(prog); 9140 9141 dev_xdp_set_link(dev, mode, NULL); 9142 } 9143 } 9144 9145 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9146 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9147 struct bpf_prog *old_prog, u32 flags) 9148 { 9149 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9150 struct bpf_prog *cur_prog; 9151 struct net_device *upper; 9152 struct list_head *iter; 9153 enum bpf_xdp_mode mode; 9154 bpf_op_t bpf_op; 9155 int err; 9156 9157 ASSERT_RTNL(); 9158 9159 /* either link or prog attachment, never both */ 9160 if (link && (new_prog || old_prog)) 9161 return -EINVAL; 9162 /* link supports only XDP mode flags */ 9163 if (link && (flags & ~XDP_FLAGS_MODES)) { 9164 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9165 return -EINVAL; 9166 } 9167 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9168 if (num_modes > 1) { 9169 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9170 return -EINVAL; 9171 } 9172 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9173 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9174 NL_SET_ERR_MSG(extack, 9175 "More than one program loaded, unset mode is ambiguous"); 9176 return -EINVAL; 9177 } 9178 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9179 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9180 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9181 return -EINVAL; 9182 } 9183 9184 mode = dev_xdp_mode(dev, flags); 9185 /* can't replace attached link */ 9186 if (dev_xdp_link(dev, mode)) { 9187 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9188 return -EBUSY; 9189 } 9190 9191 /* don't allow if an upper device already has a program */ 9192 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9193 if (dev_xdp_prog_count(upper) > 0) { 9194 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9195 return -EEXIST; 9196 } 9197 } 9198 9199 cur_prog = dev_xdp_prog(dev, mode); 9200 /* can't replace attached prog with link */ 9201 if (link && cur_prog) { 9202 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9203 return -EBUSY; 9204 } 9205 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9206 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9207 return -EEXIST; 9208 } 9209 9210 /* put effective new program into new_prog */ 9211 if (link) 9212 new_prog = link->link.prog; 9213 9214 if (new_prog) { 9215 bool offload = mode == XDP_MODE_HW; 9216 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9217 ? XDP_MODE_DRV : XDP_MODE_SKB; 9218 9219 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9220 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9221 return -EBUSY; 9222 } 9223 if (!offload && dev_xdp_prog(dev, other_mode)) { 9224 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9225 return -EEXIST; 9226 } 9227 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9228 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9229 return -EINVAL; 9230 } 9231 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9232 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9233 return -EINVAL; 9234 } 9235 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9236 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9237 return -EINVAL; 9238 } 9239 } 9240 9241 /* don't call drivers if the effective program didn't change */ 9242 if (new_prog != cur_prog) { 9243 bpf_op = dev_xdp_bpf_op(dev, mode); 9244 if (!bpf_op) { 9245 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9246 return -EOPNOTSUPP; 9247 } 9248 9249 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9250 if (err) 9251 return err; 9252 } 9253 9254 if (link) 9255 dev_xdp_set_link(dev, mode, link); 9256 else 9257 dev_xdp_set_prog(dev, mode, new_prog); 9258 if (cur_prog) 9259 bpf_prog_put(cur_prog); 9260 9261 return 0; 9262 } 9263 9264 static int dev_xdp_attach_link(struct net_device *dev, 9265 struct netlink_ext_ack *extack, 9266 struct bpf_xdp_link *link) 9267 { 9268 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9269 } 9270 9271 static int dev_xdp_detach_link(struct net_device *dev, 9272 struct netlink_ext_ack *extack, 9273 struct bpf_xdp_link *link) 9274 { 9275 enum bpf_xdp_mode mode; 9276 bpf_op_t bpf_op; 9277 9278 ASSERT_RTNL(); 9279 9280 mode = dev_xdp_mode(dev, link->flags); 9281 if (dev_xdp_link(dev, mode) != link) 9282 return -EINVAL; 9283 9284 bpf_op = dev_xdp_bpf_op(dev, mode); 9285 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9286 dev_xdp_set_link(dev, mode, NULL); 9287 return 0; 9288 } 9289 9290 static void bpf_xdp_link_release(struct bpf_link *link) 9291 { 9292 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9293 9294 rtnl_lock(); 9295 9296 /* if racing with net_device's tear down, xdp_link->dev might be 9297 * already NULL, in which case link was already auto-detached 9298 */ 9299 if (xdp_link->dev) { 9300 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9301 xdp_link->dev = NULL; 9302 } 9303 9304 rtnl_unlock(); 9305 } 9306 9307 static int bpf_xdp_link_detach(struct bpf_link *link) 9308 { 9309 bpf_xdp_link_release(link); 9310 return 0; 9311 } 9312 9313 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9314 { 9315 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9316 9317 kfree(xdp_link); 9318 } 9319 9320 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9321 struct seq_file *seq) 9322 { 9323 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9324 u32 ifindex = 0; 9325 9326 rtnl_lock(); 9327 if (xdp_link->dev) 9328 ifindex = xdp_link->dev->ifindex; 9329 rtnl_unlock(); 9330 9331 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9332 } 9333 9334 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9335 struct bpf_link_info *info) 9336 { 9337 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9338 u32 ifindex = 0; 9339 9340 rtnl_lock(); 9341 if (xdp_link->dev) 9342 ifindex = xdp_link->dev->ifindex; 9343 rtnl_unlock(); 9344 9345 info->xdp.ifindex = ifindex; 9346 return 0; 9347 } 9348 9349 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9350 struct bpf_prog *old_prog) 9351 { 9352 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9353 enum bpf_xdp_mode mode; 9354 bpf_op_t bpf_op; 9355 int err = 0; 9356 9357 rtnl_lock(); 9358 9359 /* link might have been auto-released already, so fail */ 9360 if (!xdp_link->dev) { 9361 err = -ENOLINK; 9362 goto out_unlock; 9363 } 9364 9365 if (old_prog && link->prog != old_prog) { 9366 err = -EPERM; 9367 goto out_unlock; 9368 } 9369 old_prog = link->prog; 9370 if (old_prog->type != new_prog->type || 9371 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9372 err = -EINVAL; 9373 goto out_unlock; 9374 } 9375 9376 if (old_prog == new_prog) { 9377 /* no-op, don't disturb drivers */ 9378 bpf_prog_put(new_prog); 9379 goto out_unlock; 9380 } 9381 9382 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9383 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9384 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9385 xdp_link->flags, new_prog); 9386 if (err) 9387 goto out_unlock; 9388 9389 old_prog = xchg(&link->prog, new_prog); 9390 bpf_prog_put(old_prog); 9391 9392 out_unlock: 9393 rtnl_unlock(); 9394 return err; 9395 } 9396 9397 static const struct bpf_link_ops bpf_xdp_link_lops = { 9398 .release = bpf_xdp_link_release, 9399 .dealloc = bpf_xdp_link_dealloc, 9400 .detach = bpf_xdp_link_detach, 9401 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9402 .fill_link_info = bpf_xdp_link_fill_link_info, 9403 .update_prog = bpf_xdp_link_update, 9404 }; 9405 9406 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9407 { 9408 struct net *net = current->nsproxy->net_ns; 9409 struct bpf_link_primer link_primer; 9410 struct bpf_xdp_link *link; 9411 struct net_device *dev; 9412 int err, fd; 9413 9414 rtnl_lock(); 9415 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9416 if (!dev) { 9417 rtnl_unlock(); 9418 return -EINVAL; 9419 } 9420 9421 link = kzalloc(sizeof(*link), GFP_USER); 9422 if (!link) { 9423 err = -ENOMEM; 9424 goto unlock; 9425 } 9426 9427 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9428 link->dev = dev; 9429 link->flags = attr->link_create.flags; 9430 9431 err = bpf_link_prime(&link->link, &link_primer); 9432 if (err) { 9433 kfree(link); 9434 goto unlock; 9435 } 9436 9437 err = dev_xdp_attach_link(dev, NULL, link); 9438 rtnl_unlock(); 9439 9440 if (err) { 9441 link->dev = NULL; 9442 bpf_link_cleanup(&link_primer); 9443 goto out_put_dev; 9444 } 9445 9446 fd = bpf_link_settle(&link_primer); 9447 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9448 dev_put(dev); 9449 return fd; 9450 9451 unlock: 9452 rtnl_unlock(); 9453 9454 out_put_dev: 9455 dev_put(dev); 9456 return err; 9457 } 9458 9459 /** 9460 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9461 * @dev: device 9462 * @extack: netlink extended ack 9463 * @fd: new program fd or negative value to clear 9464 * @expected_fd: old program fd that userspace expects to replace or clear 9465 * @flags: xdp-related flags 9466 * 9467 * Set or clear a bpf program for a device 9468 */ 9469 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9470 int fd, int expected_fd, u32 flags) 9471 { 9472 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9473 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9474 int err; 9475 9476 ASSERT_RTNL(); 9477 9478 if (fd >= 0) { 9479 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9480 mode != XDP_MODE_SKB); 9481 if (IS_ERR(new_prog)) 9482 return PTR_ERR(new_prog); 9483 } 9484 9485 if (expected_fd >= 0) { 9486 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9487 mode != XDP_MODE_SKB); 9488 if (IS_ERR(old_prog)) { 9489 err = PTR_ERR(old_prog); 9490 old_prog = NULL; 9491 goto err_out; 9492 } 9493 } 9494 9495 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9496 9497 err_out: 9498 if (err && new_prog) 9499 bpf_prog_put(new_prog); 9500 if (old_prog) 9501 bpf_prog_put(old_prog); 9502 return err; 9503 } 9504 9505 /** 9506 * dev_new_index - allocate an ifindex 9507 * @net: the applicable net namespace 9508 * 9509 * Returns a suitable unique value for a new device interface 9510 * number. The caller must hold the rtnl semaphore or the 9511 * dev_base_lock to be sure it remains unique. 9512 */ 9513 static int dev_new_index(struct net *net) 9514 { 9515 int ifindex = net->ifindex; 9516 9517 for (;;) { 9518 if (++ifindex <= 0) 9519 ifindex = 1; 9520 if (!__dev_get_by_index(net, ifindex)) 9521 return net->ifindex = ifindex; 9522 } 9523 } 9524 9525 /* Delayed registration/unregisteration */ 9526 LIST_HEAD(net_todo_list); 9527 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9528 9529 static void net_set_todo(struct net_device *dev) 9530 { 9531 list_add_tail(&dev->todo_list, &net_todo_list); 9532 atomic_inc(&dev_net(dev)->dev_unreg_count); 9533 } 9534 9535 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9536 struct net_device *upper, netdev_features_t features) 9537 { 9538 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9539 netdev_features_t feature; 9540 int feature_bit; 9541 9542 for_each_netdev_feature(upper_disables, feature_bit) { 9543 feature = __NETIF_F_BIT(feature_bit); 9544 if (!(upper->wanted_features & feature) 9545 && (features & feature)) { 9546 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9547 &feature, upper->name); 9548 features &= ~feature; 9549 } 9550 } 9551 9552 return features; 9553 } 9554 9555 static void netdev_sync_lower_features(struct net_device *upper, 9556 struct net_device *lower, netdev_features_t features) 9557 { 9558 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9559 netdev_features_t feature; 9560 int feature_bit; 9561 9562 for_each_netdev_feature(upper_disables, feature_bit) { 9563 feature = __NETIF_F_BIT(feature_bit); 9564 if (!(features & feature) && (lower->features & feature)) { 9565 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9566 &feature, lower->name); 9567 lower->wanted_features &= ~feature; 9568 __netdev_update_features(lower); 9569 9570 if (unlikely(lower->features & feature)) 9571 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9572 &feature, lower->name); 9573 else 9574 netdev_features_change(lower); 9575 } 9576 } 9577 } 9578 9579 static netdev_features_t netdev_fix_features(struct net_device *dev, 9580 netdev_features_t features) 9581 { 9582 /* Fix illegal checksum combinations */ 9583 if ((features & NETIF_F_HW_CSUM) && 9584 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9585 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9586 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9587 } 9588 9589 /* TSO requires that SG is present as well. */ 9590 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9591 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9592 features &= ~NETIF_F_ALL_TSO; 9593 } 9594 9595 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9596 !(features & NETIF_F_IP_CSUM)) { 9597 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9598 features &= ~NETIF_F_TSO; 9599 features &= ~NETIF_F_TSO_ECN; 9600 } 9601 9602 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9603 !(features & NETIF_F_IPV6_CSUM)) { 9604 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9605 features &= ~NETIF_F_TSO6; 9606 } 9607 9608 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9609 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9610 features &= ~NETIF_F_TSO_MANGLEID; 9611 9612 /* TSO ECN requires that TSO is present as well. */ 9613 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9614 features &= ~NETIF_F_TSO_ECN; 9615 9616 /* Software GSO depends on SG. */ 9617 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9618 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9619 features &= ~NETIF_F_GSO; 9620 } 9621 9622 /* GSO partial features require GSO partial be set */ 9623 if ((features & dev->gso_partial_features) && 9624 !(features & NETIF_F_GSO_PARTIAL)) { 9625 netdev_dbg(dev, 9626 "Dropping partially supported GSO features since no GSO partial.\n"); 9627 features &= ~dev->gso_partial_features; 9628 } 9629 9630 if (!(features & NETIF_F_RXCSUM)) { 9631 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9632 * successfully merged by hardware must also have the 9633 * checksum verified by hardware. If the user does not 9634 * want to enable RXCSUM, logically, we should disable GRO_HW. 9635 */ 9636 if (features & NETIF_F_GRO_HW) { 9637 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9638 features &= ~NETIF_F_GRO_HW; 9639 } 9640 } 9641 9642 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9643 if (features & NETIF_F_RXFCS) { 9644 if (features & NETIF_F_LRO) { 9645 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9646 features &= ~NETIF_F_LRO; 9647 } 9648 9649 if (features & NETIF_F_GRO_HW) { 9650 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9651 features &= ~NETIF_F_GRO_HW; 9652 } 9653 } 9654 9655 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9656 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9657 features &= ~NETIF_F_LRO; 9658 } 9659 9660 if (features & NETIF_F_HW_TLS_TX) { 9661 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9662 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9663 bool hw_csum = features & NETIF_F_HW_CSUM; 9664 9665 if (!ip_csum && !hw_csum) { 9666 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9667 features &= ~NETIF_F_HW_TLS_TX; 9668 } 9669 } 9670 9671 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9672 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9673 features &= ~NETIF_F_HW_TLS_RX; 9674 } 9675 9676 return features; 9677 } 9678 9679 int __netdev_update_features(struct net_device *dev) 9680 { 9681 struct net_device *upper, *lower; 9682 netdev_features_t features; 9683 struct list_head *iter; 9684 int err = -1; 9685 9686 ASSERT_RTNL(); 9687 9688 features = netdev_get_wanted_features(dev); 9689 9690 if (dev->netdev_ops->ndo_fix_features) 9691 features = dev->netdev_ops->ndo_fix_features(dev, features); 9692 9693 /* driver might be less strict about feature dependencies */ 9694 features = netdev_fix_features(dev, features); 9695 9696 /* some features can't be enabled if they're off on an upper device */ 9697 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9698 features = netdev_sync_upper_features(dev, upper, features); 9699 9700 if (dev->features == features) 9701 goto sync_lower; 9702 9703 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9704 &dev->features, &features); 9705 9706 if (dev->netdev_ops->ndo_set_features) 9707 err = dev->netdev_ops->ndo_set_features(dev, features); 9708 else 9709 err = 0; 9710 9711 if (unlikely(err < 0)) { 9712 netdev_err(dev, 9713 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9714 err, &features, &dev->features); 9715 /* return non-0 since some features might have changed and 9716 * it's better to fire a spurious notification than miss it 9717 */ 9718 return -1; 9719 } 9720 9721 sync_lower: 9722 /* some features must be disabled on lower devices when disabled 9723 * on an upper device (think: bonding master or bridge) 9724 */ 9725 netdev_for_each_lower_dev(dev, lower, iter) 9726 netdev_sync_lower_features(dev, lower, features); 9727 9728 if (!err) { 9729 netdev_features_t diff = features ^ dev->features; 9730 9731 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9732 /* udp_tunnel_{get,drop}_rx_info both need 9733 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9734 * device, or they won't do anything. 9735 * Thus we need to update dev->features 9736 * *before* calling udp_tunnel_get_rx_info, 9737 * but *after* calling udp_tunnel_drop_rx_info. 9738 */ 9739 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9740 dev->features = features; 9741 udp_tunnel_get_rx_info(dev); 9742 } else { 9743 udp_tunnel_drop_rx_info(dev); 9744 } 9745 } 9746 9747 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9748 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9749 dev->features = features; 9750 err |= vlan_get_rx_ctag_filter_info(dev); 9751 } else { 9752 vlan_drop_rx_ctag_filter_info(dev); 9753 } 9754 } 9755 9756 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9757 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9758 dev->features = features; 9759 err |= vlan_get_rx_stag_filter_info(dev); 9760 } else { 9761 vlan_drop_rx_stag_filter_info(dev); 9762 } 9763 } 9764 9765 dev->features = features; 9766 } 9767 9768 return err < 0 ? 0 : 1; 9769 } 9770 9771 /** 9772 * netdev_update_features - recalculate device features 9773 * @dev: the device to check 9774 * 9775 * Recalculate dev->features set and send notifications if it 9776 * has changed. Should be called after driver or hardware dependent 9777 * conditions might have changed that influence the features. 9778 */ 9779 void netdev_update_features(struct net_device *dev) 9780 { 9781 if (__netdev_update_features(dev)) 9782 netdev_features_change(dev); 9783 } 9784 EXPORT_SYMBOL(netdev_update_features); 9785 9786 /** 9787 * netdev_change_features - recalculate device features 9788 * @dev: the device to check 9789 * 9790 * Recalculate dev->features set and send notifications even 9791 * if they have not changed. Should be called instead of 9792 * netdev_update_features() if also dev->vlan_features might 9793 * have changed to allow the changes to be propagated to stacked 9794 * VLAN devices. 9795 */ 9796 void netdev_change_features(struct net_device *dev) 9797 { 9798 __netdev_update_features(dev); 9799 netdev_features_change(dev); 9800 } 9801 EXPORT_SYMBOL(netdev_change_features); 9802 9803 /** 9804 * netif_stacked_transfer_operstate - transfer operstate 9805 * @rootdev: the root or lower level device to transfer state from 9806 * @dev: the device to transfer operstate to 9807 * 9808 * Transfer operational state from root to device. This is normally 9809 * called when a stacking relationship exists between the root 9810 * device and the device(a leaf device). 9811 */ 9812 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9813 struct net_device *dev) 9814 { 9815 if (rootdev->operstate == IF_OPER_DORMANT) 9816 netif_dormant_on(dev); 9817 else 9818 netif_dormant_off(dev); 9819 9820 if (rootdev->operstate == IF_OPER_TESTING) 9821 netif_testing_on(dev); 9822 else 9823 netif_testing_off(dev); 9824 9825 if (netif_carrier_ok(rootdev)) 9826 netif_carrier_on(dev); 9827 else 9828 netif_carrier_off(dev); 9829 } 9830 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9831 9832 static int netif_alloc_rx_queues(struct net_device *dev) 9833 { 9834 unsigned int i, count = dev->num_rx_queues; 9835 struct netdev_rx_queue *rx; 9836 size_t sz = count * sizeof(*rx); 9837 int err = 0; 9838 9839 BUG_ON(count < 1); 9840 9841 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9842 if (!rx) 9843 return -ENOMEM; 9844 9845 dev->_rx = rx; 9846 9847 for (i = 0; i < count; i++) { 9848 rx[i].dev = dev; 9849 9850 /* XDP RX-queue setup */ 9851 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9852 if (err < 0) 9853 goto err_rxq_info; 9854 } 9855 return 0; 9856 9857 err_rxq_info: 9858 /* Rollback successful reg's and free other resources */ 9859 while (i--) 9860 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9861 kvfree(dev->_rx); 9862 dev->_rx = NULL; 9863 return err; 9864 } 9865 9866 static void netif_free_rx_queues(struct net_device *dev) 9867 { 9868 unsigned int i, count = dev->num_rx_queues; 9869 9870 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9871 if (!dev->_rx) 9872 return; 9873 9874 for (i = 0; i < count; i++) 9875 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9876 9877 kvfree(dev->_rx); 9878 } 9879 9880 static void netdev_init_one_queue(struct net_device *dev, 9881 struct netdev_queue *queue, void *_unused) 9882 { 9883 /* Initialize queue lock */ 9884 spin_lock_init(&queue->_xmit_lock); 9885 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9886 queue->xmit_lock_owner = -1; 9887 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9888 queue->dev = dev; 9889 #ifdef CONFIG_BQL 9890 dql_init(&queue->dql, HZ); 9891 #endif 9892 } 9893 9894 static void netif_free_tx_queues(struct net_device *dev) 9895 { 9896 kvfree(dev->_tx); 9897 } 9898 9899 static int netif_alloc_netdev_queues(struct net_device *dev) 9900 { 9901 unsigned int count = dev->num_tx_queues; 9902 struct netdev_queue *tx; 9903 size_t sz = count * sizeof(*tx); 9904 9905 if (count < 1 || count > 0xffff) 9906 return -EINVAL; 9907 9908 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9909 if (!tx) 9910 return -ENOMEM; 9911 9912 dev->_tx = tx; 9913 9914 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9915 spin_lock_init(&dev->tx_global_lock); 9916 9917 return 0; 9918 } 9919 9920 void netif_tx_stop_all_queues(struct net_device *dev) 9921 { 9922 unsigned int i; 9923 9924 for (i = 0; i < dev->num_tx_queues; i++) { 9925 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9926 9927 netif_tx_stop_queue(txq); 9928 } 9929 } 9930 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9931 9932 /** 9933 * register_netdevice() - register a network device 9934 * @dev: device to register 9935 * 9936 * Take a prepared network device structure and make it externally accessible. 9937 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9938 * Callers must hold the rtnl lock - you may want register_netdev() 9939 * instead of this. 9940 */ 9941 int register_netdevice(struct net_device *dev) 9942 { 9943 int ret; 9944 struct net *net = dev_net(dev); 9945 9946 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9947 NETDEV_FEATURE_COUNT); 9948 BUG_ON(dev_boot_phase); 9949 ASSERT_RTNL(); 9950 9951 might_sleep(); 9952 9953 /* When net_device's are persistent, this will be fatal. */ 9954 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9955 BUG_ON(!net); 9956 9957 ret = ethtool_check_ops(dev->ethtool_ops); 9958 if (ret) 9959 return ret; 9960 9961 spin_lock_init(&dev->addr_list_lock); 9962 netdev_set_addr_lockdep_class(dev); 9963 9964 ret = dev_get_valid_name(net, dev, dev->name); 9965 if (ret < 0) 9966 goto out; 9967 9968 ret = -ENOMEM; 9969 dev->name_node = netdev_name_node_head_alloc(dev); 9970 if (!dev->name_node) 9971 goto out; 9972 9973 /* Init, if this function is available */ 9974 if (dev->netdev_ops->ndo_init) { 9975 ret = dev->netdev_ops->ndo_init(dev); 9976 if (ret) { 9977 if (ret > 0) 9978 ret = -EIO; 9979 goto err_free_name; 9980 } 9981 } 9982 9983 if (((dev->hw_features | dev->features) & 9984 NETIF_F_HW_VLAN_CTAG_FILTER) && 9985 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9986 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9987 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9988 ret = -EINVAL; 9989 goto err_uninit; 9990 } 9991 9992 ret = -EBUSY; 9993 if (!dev->ifindex) 9994 dev->ifindex = dev_new_index(net); 9995 else if (__dev_get_by_index(net, dev->ifindex)) 9996 goto err_uninit; 9997 9998 /* Transfer changeable features to wanted_features and enable 9999 * software offloads (GSO and GRO). 10000 */ 10001 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10002 dev->features |= NETIF_F_SOFT_FEATURES; 10003 10004 if (dev->udp_tunnel_nic_info) { 10005 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10006 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10007 } 10008 10009 dev->wanted_features = dev->features & dev->hw_features; 10010 10011 if (!(dev->flags & IFF_LOOPBACK)) 10012 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10013 10014 /* If IPv4 TCP segmentation offload is supported we should also 10015 * allow the device to enable segmenting the frame with the option 10016 * of ignoring a static IP ID value. This doesn't enable the 10017 * feature itself but allows the user to enable it later. 10018 */ 10019 if (dev->hw_features & NETIF_F_TSO) 10020 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10021 if (dev->vlan_features & NETIF_F_TSO) 10022 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10023 if (dev->mpls_features & NETIF_F_TSO) 10024 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10025 if (dev->hw_enc_features & NETIF_F_TSO) 10026 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10027 10028 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10029 */ 10030 dev->vlan_features |= NETIF_F_HIGHDMA; 10031 10032 /* Make NETIF_F_SG inheritable to tunnel devices. 10033 */ 10034 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10035 10036 /* Make NETIF_F_SG inheritable to MPLS. 10037 */ 10038 dev->mpls_features |= NETIF_F_SG; 10039 10040 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10041 ret = notifier_to_errno(ret); 10042 if (ret) 10043 goto err_uninit; 10044 10045 ret = netdev_register_kobject(dev); 10046 if (ret) { 10047 dev->reg_state = NETREG_UNREGISTERED; 10048 goto err_uninit; 10049 } 10050 dev->reg_state = NETREG_REGISTERED; 10051 10052 __netdev_update_features(dev); 10053 10054 /* 10055 * Default initial state at registry is that the 10056 * device is present. 10057 */ 10058 10059 set_bit(__LINK_STATE_PRESENT, &dev->state); 10060 10061 linkwatch_init_dev(dev); 10062 10063 dev_init_scheduler(dev); 10064 10065 dev_hold_track(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10066 list_netdevice(dev); 10067 10068 add_device_randomness(dev->dev_addr, dev->addr_len); 10069 10070 /* If the device has permanent device address, driver should 10071 * set dev_addr and also addr_assign_type should be set to 10072 * NET_ADDR_PERM (default value). 10073 */ 10074 if (dev->addr_assign_type == NET_ADDR_PERM) 10075 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10076 10077 /* Notify protocols, that a new device appeared. */ 10078 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10079 ret = notifier_to_errno(ret); 10080 if (ret) { 10081 /* Expect explicit free_netdev() on failure */ 10082 dev->needs_free_netdev = false; 10083 unregister_netdevice_queue(dev, NULL); 10084 goto out; 10085 } 10086 /* 10087 * Prevent userspace races by waiting until the network 10088 * device is fully setup before sending notifications. 10089 */ 10090 if (!dev->rtnl_link_ops || 10091 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10092 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10093 10094 out: 10095 return ret; 10096 10097 err_uninit: 10098 if (dev->netdev_ops->ndo_uninit) 10099 dev->netdev_ops->ndo_uninit(dev); 10100 if (dev->priv_destructor) 10101 dev->priv_destructor(dev); 10102 err_free_name: 10103 netdev_name_node_free(dev->name_node); 10104 goto out; 10105 } 10106 EXPORT_SYMBOL(register_netdevice); 10107 10108 /** 10109 * init_dummy_netdev - init a dummy network device for NAPI 10110 * @dev: device to init 10111 * 10112 * This takes a network device structure and initialize the minimum 10113 * amount of fields so it can be used to schedule NAPI polls without 10114 * registering a full blown interface. This is to be used by drivers 10115 * that need to tie several hardware interfaces to a single NAPI 10116 * poll scheduler due to HW limitations. 10117 */ 10118 int init_dummy_netdev(struct net_device *dev) 10119 { 10120 /* Clear everything. Note we don't initialize spinlocks 10121 * are they aren't supposed to be taken by any of the 10122 * NAPI code and this dummy netdev is supposed to be 10123 * only ever used for NAPI polls 10124 */ 10125 memset(dev, 0, sizeof(struct net_device)); 10126 10127 /* make sure we BUG if trying to hit standard 10128 * register/unregister code path 10129 */ 10130 dev->reg_state = NETREG_DUMMY; 10131 10132 /* NAPI wants this */ 10133 INIT_LIST_HEAD(&dev->napi_list); 10134 10135 /* a dummy interface is started by default */ 10136 set_bit(__LINK_STATE_PRESENT, &dev->state); 10137 set_bit(__LINK_STATE_START, &dev->state); 10138 10139 /* napi_busy_loop stats accounting wants this */ 10140 dev_net_set(dev, &init_net); 10141 10142 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10143 * because users of this 'device' dont need to change 10144 * its refcount. 10145 */ 10146 10147 return 0; 10148 } 10149 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10150 10151 10152 /** 10153 * register_netdev - register a network device 10154 * @dev: device to register 10155 * 10156 * Take a completed network device structure and add it to the kernel 10157 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10158 * chain. 0 is returned on success. A negative errno code is returned 10159 * on a failure to set up the device, or if the name is a duplicate. 10160 * 10161 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10162 * and expands the device name if you passed a format string to 10163 * alloc_netdev. 10164 */ 10165 int register_netdev(struct net_device *dev) 10166 { 10167 int err; 10168 10169 if (rtnl_lock_killable()) 10170 return -EINTR; 10171 err = register_netdevice(dev); 10172 rtnl_unlock(); 10173 return err; 10174 } 10175 EXPORT_SYMBOL(register_netdev); 10176 10177 int netdev_refcnt_read(const struct net_device *dev) 10178 { 10179 #ifdef CONFIG_PCPU_DEV_REFCNT 10180 int i, refcnt = 0; 10181 10182 for_each_possible_cpu(i) 10183 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10184 return refcnt; 10185 #else 10186 return refcount_read(&dev->dev_refcnt); 10187 #endif 10188 } 10189 EXPORT_SYMBOL(netdev_refcnt_read); 10190 10191 int netdev_unregister_timeout_secs __read_mostly = 10; 10192 10193 #define WAIT_REFS_MIN_MSECS 1 10194 #define WAIT_REFS_MAX_MSECS 250 10195 /** 10196 * netdev_wait_allrefs_any - wait until all references are gone. 10197 * @list: list of net_devices to wait on 10198 * 10199 * This is called when unregistering network devices. 10200 * 10201 * Any protocol or device that holds a reference should register 10202 * for netdevice notification, and cleanup and put back the 10203 * reference if they receive an UNREGISTER event. 10204 * We can get stuck here if buggy protocols don't correctly 10205 * call dev_put. 10206 */ 10207 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10208 { 10209 unsigned long rebroadcast_time, warning_time; 10210 struct net_device *dev; 10211 int wait = 0; 10212 10213 rebroadcast_time = warning_time = jiffies; 10214 10215 list_for_each_entry(dev, list, todo_list) 10216 if (netdev_refcnt_read(dev) == 1) 10217 return dev; 10218 10219 while (true) { 10220 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10221 rtnl_lock(); 10222 10223 /* Rebroadcast unregister notification */ 10224 list_for_each_entry(dev, list, todo_list) 10225 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10226 10227 __rtnl_unlock(); 10228 rcu_barrier(); 10229 rtnl_lock(); 10230 10231 list_for_each_entry(dev, list, todo_list) 10232 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10233 &dev->state)) { 10234 /* We must not have linkwatch events 10235 * pending on unregister. If this 10236 * happens, we simply run the queue 10237 * unscheduled, resulting in a noop 10238 * for this device. 10239 */ 10240 linkwatch_run_queue(); 10241 break; 10242 } 10243 10244 __rtnl_unlock(); 10245 10246 rebroadcast_time = jiffies; 10247 } 10248 10249 if (!wait) { 10250 rcu_barrier(); 10251 wait = WAIT_REFS_MIN_MSECS; 10252 } else { 10253 msleep(wait); 10254 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10255 } 10256 10257 list_for_each_entry(dev, list, todo_list) 10258 if (netdev_refcnt_read(dev) == 1) 10259 return dev; 10260 10261 if (time_after(jiffies, warning_time + 10262 netdev_unregister_timeout_secs * HZ)) { 10263 list_for_each_entry(dev, list, todo_list) { 10264 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10265 dev->name, netdev_refcnt_read(dev)); 10266 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10267 } 10268 10269 warning_time = jiffies; 10270 } 10271 } 10272 } 10273 10274 /* The sequence is: 10275 * 10276 * rtnl_lock(); 10277 * ... 10278 * register_netdevice(x1); 10279 * register_netdevice(x2); 10280 * ... 10281 * unregister_netdevice(y1); 10282 * unregister_netdevice(y2); 10283 * ... 10284 * rtnl_unlock(); 10285 * free_netdev(y1); 10286 * free_netdev(y2); 10287 * 10288 * We are invoked by rtnl_unlock(). 10289 * This allows us to deal with problems: 10290 * 1) We can delete sysfs objects which invoke hotplug 10291 * without deadlocking with linkwatch via keventd. 10292 * 2) Since we run with the RTNL semaphore not held, we can sleep 10293 * safely in order to wait for the netdev refcnt to drop to zero. 10294 * 10295 * We must not return until all unregister events added during 10296 * the interval the lock was held have been completed. 10297 */ 10298 void netdev_run_todo(void) 10299 { 10300 struct net_device *dev, *tmp; 10301 struct list_head list; 10302 #ifdef CONFIG_LOCKDEP 10303 struct list_head unlink_list; 10304 10305 list_replace_init(&net_unlink_list, &unlink_list); 10306 10307 while (!list_empty(&unlink_list)) { 10308 struct net_device *dev = list_first_entry(&unlink_list, 10309 struct net_device, 10310 unlink_list); 10311 list_del_init(&dev->unlink_list); 10312 dev->nested_level = dev->lower_level - 1; 10313 } 10314 #endif 10315 10316 /* Snapshot list, allow later requests */ 10317 list_replace_init(&net_todo_list, &list); 10318 10319 __rtnl_unlock(); 10320 10321 /* Wait for rcu callbacks to finish before next phase */ 10322 if (!list_empty(&list)) 10323 rcu_barrier(); 10324 10325 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10326 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10327 netdev_WARN(dev, "run_todo but not unregistering\n"); 10328 list_del(&dev->todo_list); 10329 continue; 10330 } 10331 10332 dev->reg_state = NETREG_UNREGISTERED; 10333 linkwatch_forget_dev(dev); 10334 } 10335 10336 while (!list_empty(&list)) { 10337 dev = netdev_wait_allrefs_any(&list); 10338 list_del(&dev->todo_list); 10339 10340 /* paranoia */ 10341 BUG_ON(netdev_refcnt_read(dev) != 1); 10342 BUG_ON(!list_empty(&dev->ptype_all)); 10343 BUG_ON(!list_empty(&dev->ptype_specific)); 10344 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10345 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10346 #if IS_ENABLED(CONFIG_DECNET) 10347 WARN_ON(dev->dn_ptr); 10348 #endif 10349 if (dev->priv_destructor) 10350 dev->priv_destructor(dev); 10351 if (dev->needs_free_netdev) 10352 free_netdev(dev); 10353 10354 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10355 wake_up(&netdev_unregistering_wq); 10356 10357 /* Free network device */ 10358 kobject_put(&dev->dev.kobj); 10359 } 10360 } 10361 10362 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10363 * all the same fields in the same order as net_device_stats, with only 10364 * the type differing, but rtnl_link_stats64 may have additional fields 10365 * at the end for newer counters. 10366 */ 10367 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10368 const struct net_device_stats *netdev_stats) 10369 { 10370 #if BITS_PER_LONG == 64 10371 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10372 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10373 /* zero out counters that only exist in rtnl_link_stats64 */ 10374 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10375 sizeof(*stats64) - sizeof(*netdev_stats)); 10376 #else 10377 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10378 const unsigned long *src = (const unsigned long *)netdev_stats; 10379 u64 *dst = (u64 *)stats64; 10380 10381 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10382 for (i = 0; i < n; i++) 10383 dst[i] = src[i]; 10384 /* zero out counters that only exist in rtnl_link_stats64 */ 10385 memset((char *)stats64 + n * sizeof(u64), 0, 10386 sizeof(*stats64) - n * sizeof(u64)); 10387 #endif 10388 } 10389 EXPORT_SYMBOL(netdev_stats_to_stats64); 10390 10391 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10392 { 10393 struct net_device_core_stats __percpu *p; 10394 10395 p = alloc_percpu_gfp(struct net_device_core_stats, 10396 GFP_ATOMIC | __GFP_NOWARN); 10397 10398 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10399 free_percpu(p); 10400 10401 /* This READ_ONCE() pairs with the cmpxchg() above */ 10402 return READ_ONCE(dev->core_stats); 10403 } 10404 EXPORT_SYMBOL(netdev_core_stats_alloc); 10405 10406 /** 10407 * dev_get_stats - get network device statistics 10408 * @dev: device to get statistics from 10409 * @storage: place to store stats 10410 * 10411 * Get network statistics from device. Return @storage. 10412 * The device driver may provide its own method by setting 10413 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10414 * otherwise the internal statistics structure is used. 10415 */ 10416 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10417 struct rtnl_link_stats64 *storage) 10418 { 10419 const struct net_device_ops *ops = dev->netdev_ops; 10420 const struct net_device_core_stats __percpu *p; 10421 10422 if (ops->ndo_get_stats64) { 10423 memset(storage, 0, sizeof(*storage)); 10424 ops->ndo_get_stats64(dev, storage); 10425 } else if (ops->ndo_get_stats) { 10426 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10427 } else { 10428 netdev_stats_to_stats64(storage, &dev->stats); 10429 } 10430 10431 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10432 p = READ_ONCE(dev->core_stats); 10433 if (p) { 10434 const struct net_device_core_stats *core_stats; 10435 int i; 10436 10437 for_each_possible_cpu(i) { 10438 core_stats = per_cpu_ptr(p, i); 10439 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10440 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10441 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10442 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10443 } 10444 } 10445 return storage; 10446 } 10447 EXPORT_SYMBOL(dev_get_stats); 10448 10449 /** 10450 * dev_fetch_sw_netstats - get per-cpu network device statistics 10451 * @s: place to store stats 10452 * @netstats: per-cpu network stats to read from 10453 * 10454 * Read per-cpu network statistics and populate the related fields in @s. 10455 */ 10456 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10457 const struct pcpu_sw_netstats __percpu *netstats) 10458 { 10459 int cpu; 10460 10461 for_each_possible_cpu(cpu) { 10462 const struct pcpu_sw_netstats *stats; 10463 struct pcpu_sw_netstats tmp; 10464 unsigned int start; 10465 10466 stats = per_cpu_ptr(netstats, cpu); 10467 do { 10468 start = u64_stats_fetch_begin_irq(&stats->syncp); 10469 tmp.rx_packets = stats->rx_packets; 10470 tmp.rx_bytes = stats->rx_bytes; 10471 tmp.tx_packets = stats->tx_packets; 10472 tmp.tx_bytes = stats->tx_bytes; 10473 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10474 10475 s->rx_packets += tmp.rx_packets; 10476 s->rx_bytes += tmp.rx_bytes; 10477 s->tx_packets += tmp.tx_packets; 10478 s->tx_bytes += tmp.tx_bytes; 10479 } 10480 } 10481 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10482 10483 /** 10484 * dev_get_tstats64 - ndo_get_stats64 implementation 10485 * @dev: device to get statistics from 10486 * @s: place to store stats 10487 * 10488 * Populate @s from dev->stats and dev->tstats. Can be used as 10489 * ndo_get_stats64() callback. 10490 */ 10491 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10492 { 10493 netdev_stats_to_stats64(s, &dev->stats); 10494 dev_fetch_sw_netstats(s, dev->tstats); 10495 } 10496 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10497 10498 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10499 { 10500 struct netdev_queue *queue = dev_ingress_queue(dev); 10501 10502 #ifdef CONFIG_NET_CLS_ACT 10503 if (queue) 10504 return queue; 10505 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10506 if (!queue) 10507 return NULL; 10508 netdev_init_one_queue(dev, queue, NULL); 10509 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10510 queue->qdisc_sleeping = &noop_qdisc; 10511 rcu_assign_pointer(dev->ingress_queue, queue); 10512 #endif 10513 return queue; 10514 } 10515 10516 static const struct ethtool_ops default_ethtool_ops; 10517 10518 void netdev_set_default_ethtool_ops(struct net_device *dev, 10519 const struct ethtool_ops *ops) 10520 { 10521 if (dev->ethtool_ops == &default_ethtool_ops) 10522 dev->ethtool_ops = ops; 10523 } 10524 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10525 10526 void netdev_freemem(struct net_device *dev) 10527 { 10528 char *addr = (char *)dev - dev->padded; 10529 10530 kvfree(addr); 10531 } 10532 10533 /** 10534 * alloc_netdev_mqs - allocate network device 10535 * @sizeof_priv: size of private data to allocate space for 10536 * @name: device name format string 10537 * @name_assign_type: origin of device name 10538 * @setup: callback to initialize device 10539 * @txqs: the number of TX subqueues to allocate 10540 * @rxqs: the number of RX subqueues to allocate 10541 * 10542 * Allocates a struct net_device with private data area for driver use 10543 * and performs basic initialization. Also allocates subqueue structs 10544 * for each queue on the device. 10545 */ 10546 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10547 unsigned char name_assign_type, 10548 void (*setup)(struct net_device *), 10549 unsigned int txqs, unsigned int rxqs) 10550 { 10551 struct net_device *dev; 10552 unsigned int alloc_size; 10553 struct net_device *p; 10554 10555 BUG_ON(strlen(name) >= sizeof(dev->name)); 10556 10557 if (txqs < 1) { 10558 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10559 return NULL; 10560 } 10561 10562 if (rxqs < 1) { 10563 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10564 return NULL; 10565 } 10566 10567 alloc_size = sizeof(struct net_device); 10568 if (sizeof_priv) { 10569 /* ensure 32-byte alignment of private area */ 10570 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10571 alloc_size += sizeof_priv; 10572 } 10573 /* ensure 32-byte alignment of whole construct */ 10574 alloc_size += NETDEV_ALIGN - 1; 10575 10576 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10577 if (!p) 10578 return NULL; 10579 10580 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10581 dev->padded = (char *)dev - (char *)p; 10582 10583 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10584 #ifdef CONFIG_PCPU_DEV_REFCNT 10585 dev->pcpu_refcnt = alloc_percpu(int); 10586 if (!dev->pcpu_refcnt) 10587 goto free_dev; 10588 __dev_hold(dev); 10589 #else 10590 refcount_set(&dev->dev_refcnt, 1); 10591 #endif 10592 10593 if (dev_addr_init(dev)) 10594 goto free_pcpu; 10595 10596 dev_mc_init(dev); 10597 dev_uc_init(dev); 10598 10599 dev_net_set(dev, &init_net); 10600 10601 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10602 dev->gso_max_segs = GSO_MAX_SEGS; 10603 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10604 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10605 dev->tso_max_segs = TSO_MAX_SEGS; 10606 dev->upper_level = 1; 10607 dev->lower_level = 1; 10608 #ifdef CONFIG_LOCKDEP 10609 dev->nested_level = 0; 10610 INIT_LIST_HEAD(&dev->unlink_list); 10611 #endif 10612 10613 INIT_LIST_HEAD(&dev->napi_list); 10614 INIT_LIST_HEAD(&dev->unreg_list); 10615 INIT_LIST_HEAD(&dev->close_list); 10616 INIT_LIST_HEAD(&dev->link_watch_list); 10617 INIT_LIST_HEAD(&dev->adj_list.upper); 10618 INIT_LIST_HEAD(&dev->adj_list.lower); 10619 INIT_LIST_HEAD(&dev->ptype_all); 10620 INIT_LIST_HEAD(&dev->ptype_specific); 10621 INIT_LIST_HEAD(&dev->net_notifier_list); 10622 #ifdef CONFIG_NET_SCHED 10623 hash_init(dev->qdisc_hash); 10624 #endif 10625 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10626 setup(dev); 10627 10628 if (!dev->tx_queue_len) { 10629 dev->priv_flags |= IFF_NO_QUEUE; 10630 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10631 } 10632 10633 dev->num_tx_queues = txqs; 10634 dev->real_num_tx_queues = txqs; 10635 if (netif_alloc_netdev_queues(dev)) 10636 goto free_all; 10637 10638 dev->num_rx_queues = rxqs; 10639 dev->real_num_rx_queues = rxqs; 10640 if (netif_alloc_rx_queues(dev)) 10641 goto free_all; 10642 10643 strcpy(dev->name, name); 10644 dev->name_assign_type = name_assign_type; 10645 dev->group = INIT_NETDEV_GROUP; 10646 if (!dev->ethtool_ops) 10647 dev->ethtool_ops = &default_ethtool_ops; 10648 10649 nf_hook_netdev_init(dev); 10650 10651 return dev; 10652 10653 free_all: 10654 free_netdev(dev); 10655 return NULL; 10656 10657 free_pcpu: 10658 #ifdef CONFIG_PCPU_DEV_REFCNT 10659 free_percpu(dev->pcpu_refcnt); 10660 free_dev: 10661 #endif 10662 netdev_freemem(dev); 10663 return NULL; 10664 } 10665 EXPORT_SYMBOL(alloc_netdev_mqs); 10666 10667 /** 10668 * free_netdev - free network device 10669 * @dev: device 10670 * 10671 * This function does the last stage of destroying an allocated device 10672 * interface. The reference to the device object is released. If this 10673 * is the last reference then it will be freed.Must be called in process 10674 * context. 10675 */ 10676 void free_netdev(struct net_device *dev) 10677 { 10678 struct napi_struct *p, *n; 10679 10680 might_sleep(); 10681 10682 /* When called immediately after register_netdevice() failed the unwind 10683 * handling may still be dismantling the device. Handle that case by 10684 * deferring the free. 10685 */ 10686 if (dev->reg_state == NETREG_UNREGISTERING) { 10687 ASSERT_RTNL(); 10688 dev->needs_free_netdev = true; 10689 return; 10690 } 10691 10692 netif_free_tx_queues(dev); 10693 netif_free_rx_queues(dev); 10694 10695 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10696 10697 /* Flush device addresses */ 10698 dev_addr_flush(dev); 10699 10700 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10701 netif_napi_del(p); 10702 10703 ref_tracker_dir_exit(&dev->refcnt_tracker); 10704 #ifdef CONFIG_PCPU_DEV_REFCNT 10705 free_percpu(dev->pcpu_refcnt); 10706 dev->pcpu_refcnt = NULL; 10707 #endif 10708 free_percpu(dev->core_stats); 10709 dev->core_stats = NULL; 10710 free_percpu(dev->xdp_bulkq); 10711 dev->xdp_bulkq = NULL; 10712 10713 /* Compatibility with error handling in drivers */ 10714 if (dev->reg_state == NETREG_UNINITIALIZED) { 10715 netdev_freemem(dev); 10716 return; 10717 } 10718 10719 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10720 dev->reg_state = NETREG_RELEASED; 10721 10722 /* will free via device release */ 10723 put_device(&dev->dev); 10724 } 10725 EXPORT_SYMBOL(free_netdev); 10726 10727 /** 10728 * synchronize_net - Synchronize with packet receive processing 10729 * 10730 * Wait for packets currently being received to be done. 10731 * Does not block later packets from starting. 10732 */ 10733 void synchronize_net(void) 10734 { 10735 might_sleep(); 10736 if (rtnl_is_locked()) 10737 synchronize_rcu_expedited(); 10738 else 10739 synchronize_rcu(); 10740 } 10741 EXPORT_SYMBOL(synchronize_net); 10742 10743 /** 10744 * unregister_netdevice_queue - remove device from the kernel 10745 * @dev: device 10746 * @head: list 10747 * 10748 * This function shuts down a device interface and removes it 10749 * from the kernel tables. 10750 * If head not NULL, device is queued to be unregistered later. 10751 * 10752 * Callers must hold the rtnl semaphore. You may want 10753 * unregister_netdev() instead of this. 10754 */ 10755 10756 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10757 { 10758 ASSERT_RTNL(); 10759 10760 if (head) { 10761 list_move_tail(&dev->unreg_list, head); 10762 } else { 10763 LIST_HEAD(single); 10764 10765 list_add(&dev->unreg_list, &single); 10766 unregister_netdevice_many(&single); 10767 } 10768 } 10769 EXPORT_SYMBOL(unregister_netdevice_queue); 10770 10771 /** 10772 * unregister_netdevice_many - unregister many devices 10773 * @head: list of devices 10774 * 10775 * Note: As most callers use a stack allocated list_head, 10776 * we force a list_del() to make sure stack wont be corrupted later. 10777 */ 10778 void unregister_netdevice_many(struct list_head *head) 10779 { 10780 struct net_device *dev, *tmp; 10781 LIST_HEAD(close_head); 10782 10783 BUG_ON(dev_boot_phase); 10784 ASSERT_RTNL(); 10785 10786 if (list_empty(head)) 10787 return; 10788 10789 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10790 /* Some devices call without registering 10791 * for initialization unwind. Remove those 10792 * devices and proceed with the remaining. 10793 */ 10794 if (dev->reg_state == NETREG_UNINITIALIZED) { 10795 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10796 dev->name, dev); 10797 10798 WARN_ON(1); 10799 list_del(&dev->unreg_list); 10800 continue; 10801 } 10802 dev->dismantle = true; 10803 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10804 } 10805 10806 /* If device is running, close it first. */ 10807 list_for_each_entry(dev, head, unreg_list) 10808 list_add_tail(&dev->close_list, &close_head); 10809 dev_close_many(&close_head, true); 10810 10811 list_for_each_entry(dev, head, unreg_list) { 10812 /* And unlink it from device chain. */ 10813 unlist_netdevice(dev); 10814 10815 dev->reg_state = NETREG_UNREGISTERING; 10816 } 10817 flush_all_backlogs(); 10818 10819 synchronize_net(); 10820 10821 list_for_each_entry(dev, head, unreg_list) { 10822 struct sk_buff *skb = NULL; 10823 10824 /* Shutdown queueing discipline. */ 10825 dev_shutdown(dev); 10826 10827 dev_xdp_uninstall(dev); 10828 10829 netdev_offload_xstats_disable_all(dev); 10830 10831 /* Notify protocols, that we are about to destroy 10832 * this device. They should clean all the things. 10833 */ 10834 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10835 10836 if (!dev->rtnl_link_ops || 10837 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10838 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10839 GFP_KERNEL, NULL, 0); 10840 10841 /* 10842 * Flush the unicast and multicast chains 10843 */ 10844 dev_uc_flush(dev); 10845 dev_mc_flush(dev); 10846 10847 netdev_name_node_alt_flush(dev); 10848 netdev_name_node_free(dev->name_node); 10849 10850 if (dev->netdev_ops->ndo_uninit) 10851 dev->netdev_ops->ndo_uninit(dev); 10852 10853 if (skb) 10854 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10855 10856 /* Notifier chain MUST detach us all upper devices. */ 10857 WARN_ON(netdev_has_any_upper_dev(dev)); 10858 WARN_ON(netdev_has_any_lower_dev(dev)); 10859 10860 /* Remove entries from kobject tree */ 10861 netdev_unregister_kobject(dev); 10862 #ifdef CONFIG_XPS 10863 /* Remove XPS queueing entries */ 10864 netif_reset_xps_queues_gt(dev, 0); 10865 #endif 10866 } 10867 10868 synchronize_net(); 10869 10870 list_for_each_entry(dev, head, unreg_list) { 10871 dev_put_track(dev, &dev->dev_registered_tracker); 10872 net_set_todo(dev); 10873 } 10874 10875 list_del(head); 10876 } 10877 EXPORT_SYMBOL(unregister_netdevice_many); 10878 10879 /** 10880 * unregister_netdev - remove device from the kernel 10881 * @dev: device 10882 * 10883 * This function shuts down a device interface and removes it 10884 * from the kernel tables. 10885 * 10886 * This is just a wrapper for unregister_netdevice that takes 10887 * the rtnl semaphore. In general you want to use this and not 10888 * unregister_netdevice. 10889 */ 10890 void unregister_netdev(struct net_device *dev) 10891 { 10892 rtnl_lock(); 10893 unregister_netdevice(dev); 10894 rtnl_unlock(); 10895 } 10896 EXPORT_SYMBOL(unregister_netdev); 10897 10898 /** 10899 * __dev_change_net_namespace - move device to different nethost namespace 10900 * @dev: device 10901 * @net: network namespace 10902 * @pat: If not NULL name pattern to try if the current device name 10903 * is already taken in the destination network namespace. 10904 * @new_ifindex: If not zero, specifies device index in the target 10905 * namespace. 10906 * 10907 * This function shuts down a device interface and moves it 10908 * to a new network namespace. On success 0 is returned, on 10909 * a failure a netagive errno code is returned. 10910 * 10911 * Callers must hold the rtnl semaphore. 10912 */ 10913 10914 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10915 const char *pat, int new_ifindex) 10916 { 10917 struct net *net_old = dev_net(dev); 10918 int err, new_nsid; 10919 10920 ASSERT_RTNL(); 10921 10922 /* Don't allow namespace local devices to be moved. */ 10923 err = -EINVAL; 10924 if (dev->features & NETIF_F_NETNS_LOCAL) 10925 goto out; 10926 10927 /* Ensure the device has been registrered */ 10928 if (dev->reg_state != NETREG_REGISTERED) 10929 goto out; 10930 10931 /* Get out if there is nothing todo */ 10932 err = 0; 10933 if (net_eq(net_old, net)) 10934 goto out; 10935 10936 /* Pick the destination device name, and ensure 10937 * we can use it in the destination network namespace. 10938 */ 10939 err = -EEXIST; 10940 if (netdev_name_in_use(net, dev->name)) { 10941 /* We get here if we can't use the current device name */ 10942 if (!pat) 10943 goto out; 10944 err = dev_get_valid_name(net, dev, pat); 10945 if (err < 0) 10946 goto out; 10947 } 10948 10949 /* Check that new_ifindex isn't used yet. */ 10950 err = -EBUSY; 10951 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10952 goto out; 10953 10954 /* 10955 * And now a mini version of register_netdevice unregister_netdevice. 10956 */ 10957 10958 /* If device is running close it first. */ 10959 dev_close(dev); 10960 10961 /* And unlink it from device chain */ 10962 unlist_netdevice(dev); 10963 10964 synchronize_net(); 10965 10966 /* Shutdown queueing discipline. */ 10967 dev_shutdown(dev); 10968 10969 /* Notify protocols, that we are about to destroy 10970 * this device. They should clean all the things. 10971 * 10972 * Note that dev->reg_state stays at NETREG_REGISTERED. 10973 * This is wanted because this way 8021q and macvlan know 10974 * the device is just moving and can keep their slaves up. 10975 */ 10976 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10977 rcu_barrier(); 10978 10979 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10980 /* If there is an ifindex conflict assign a new one */ 10981 if (!new_ifindex) { 10982 if (__dev_get_by_index(net, dev->ifindex)) 10983 new_ifindex = dev_new_index(net); 10984 else 10985 new_ifindex = dev->ifindex; 10986 } 10987 10988 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10989 new_ifindex); 10990 10991 /* 10992 * Flush the unicast and multicast chains 10993 */ 10994 dev_uc_flush(dev); 10995 dev_mc_flush(dev); 10996 10997 /* Send a netdev-removed uevent to the old namespace */ 10998 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10999 netdev_adjacent_del_links(dev); 11000 11001 /* Move per-net netdevice notifiers that are following the netdevice */ 11002 move_netdevice_notifiers_dev_net(dev, net); 11003 11004 /* Actually switch the network namespace */ 11005 dev_net_set(dev, net); 11006 dev->ifindex = new_ifindex; 11007 11008 /* Send a netdev-add uevent to the new namespace */ 11009 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11010 netdev_adjacent_add_links(dev); 11011 11012 /* Fixup kobjects */ 11013 err = device_rename(&dev->dev, dev->name); 11014 WARN_ON(err); 11015 11016 /* Adapt owner in case owning user namespace of target network 11017 * namespace is different from the original one. 11018 */ 11019 err = netdev_change_owner(dev, net_old, net); 11020 WARN_ON(err); 11021 11022 /* Add the device back in the hashes */ 11023 list_netdevice(dev); 11024 11025 /* Notify protocols, that a new device appeared. */ 11026 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11027 11028 /* 11029 * Prevent userspace races by waiting until the network 11030 * device is fully setup before sending notifications. 11031 */ 11032 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11033 11034 synchronize_net(); 11035 err = 0; 11036 out: 11037 return err; 11038 } 11039 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11040 11041 static int dev_cpu_dead(unsigned int oldcpu) 11042 { 11043 struct sk_buff **list_skb; 11044 struct sk_buff *skb; 11045 unsigned int cpu; 11046 struct softnet_data *sd, *oldsd, *remsd = NULL; 11047 11048 local_irq_disable(); 11049 cpu = smp_processor_id(); 11050 sd = &per_cpu(softnet_data, cpu); 11051 oldsd = &per_cpu(softnet_data, oldcpu); 11052 11053 /* Find end of our completion_queue. */ 11054 list_skb = &sd->completion_queue; 11055 while (*list_skb) 11056 list_skb = &(*list_skb)->next; 11057 /* Append completion queue from offline CPU. */ 11058 *list_skb = oldsd->completion_queue; 11059 oldsd->completion_queue = NULL; 11060 11061 /* Append output queue from offline CPU. */ 11062 if (oldsd->output_queue) { 11063 *sd->output_queue_tailp = oldsd->output_queue; 11064 sd->output_queue_tailp = oldsd->output_queue_tailp; 11065 oldsd->output_queue = NULL; 11066 oldsd->output_queue_tailp = &oldsd->output_queue; 11067 } 11068 /* Append NAPI poll list from offline CPU, with one exception : 11069 * process_backlog() must be called by cpu owning percpu backlog. 11070 * We properly handle process_queue & input_pkt_queue later. 11071 */ 11072 while (!list_empty(&oldsd->poll_list)) { 11073 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11074 struct napi_struct, 11075 poll_list); 11076 11077 list_del_init(&napi->poll_list); 11078 if (napi->poll == process_backlog) 11079 napi->state = 0; 11080 else 11081 ____napi_schedule(sd, napi); 11082 } 11083 11084 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11085 local_irq_enable(); 11086 11087 #ifdef CONFIG_RPS 11088 remsd = oldsd->rps_ipi_list; 11089 oldsd->rps_ipi_list = NULL; 11090 #endif 11091 /* send out pending IPI's on offline CPU */ 11092 net_rps_send_ipi(remsd); 11093 11094 /* Process offline CPU's input_pkt_queue */ 11095 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11096 netif_rx(skb); 11097 input_queue_head_incr(oldsd); 11098 } 11099 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11100 netif_rx(skb); 11101 input_queue_head_incr(oldsd); 11102 } 11103 11104 return 0; 11105 } 11106 11107 /** 11108 * netdev_increment_features - increment feature set by one 11109 * @all: current feature set 11110 * @one: new feature set 11111 * @mask: mask feature set 11112 * 11113 * Computes a new feature set after adding a device with feature set 11114 * @one to the master device with current feature set @all. Will not 11115 * enable anything that is off in @mask. Returns the new feature set. 11116 */ 11117 netdev_features_t netdev_increment_features(netdev_features_t all, 11118 netdev_features_t one, netdev_features_t mask) 11119 { 11120 if (mask & NETIF_F_HW_CSUM) 11121 mask |= NETIF_F_CSUM_MASK; 11122 mask |= NETIF_F_VLAN_CHALLENGED; 11123 11124 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11125 all &= one | ~NETIF_F_ALL_FOR_ALL; 11126 11127 /* If one device supports hw checksumming, set for all. */ 11128 if (all & NETIF_F_HW_CSUM) 11129 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11130 11131 return all; 11132 } 11133 EXPORT_SYMBOL(netdev_increment_features); 11134 11135 static struct hlist_head * __net_init netdev_create_hash(void) 11136 { 11137 int i; 11138 struct hlist_head *hash; 11139 11140 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11141 if (hash != NULL) 11142 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11143 INIT_HLIST_HEAD(&hash[i]); 11144 11145 return hash; 11146 } 11147 11148 /* Initialize per network namespace state */ 11149 static int __net_init netdev_init(struct net *net) 11150 { 11151 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11152 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11153 11154 INIT_LIST_HEAD(&net->dev_base_head); 11155 11156 net->dev_name_head = netdev_create_hash(); 11157 if (net->dev_name_head == NULL) 11158 goto err_name; 11159 11160 net->dev_index_head = netdev_create_hash(); 11161 if (net->dev_index_head == NULL) 11162 goto err_idx; 11163 11164 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11165 11166 return 0; 11167 11168 err_idx: 11169 kfree(net->dev_name_head); 11170 err_name: 11171 return -ENOMEM; 11172 } 11173 11174 /** 11175 * netdev_drivername - network driver for the device 11176 * @dev: network device 11177 * 11178 * Determine network driver for device. 11179 */ 11180 const char *netdev_drivername(const struct net_device *dev) 11181 { 11182 const struct device_driver *driver; 11183 const struct device *parent; 11184 const char *empty = ""; 11185 11186 parent = dev->dev.parent; 11187 if (!parent) 11188 return empty; 11189 11190 driver = parent->driver; 11191 if (driver && driver->name) 11192 return driver->name; 11193 return empty; 11194 } 11195 11196 static void __netdev_printk(const char *level, const struct net_device *dev, 11197 struct va_format *vaf) 11198 { 11199 if (dev && dev->dev.parent) { 11200 dev_printk_emit(level[1] - '0', 11201 dev->dev.parent, 11202 "%s %s %s%s: %pV", 11203 dev_driver_string(dev->dev.parent), 11204 dev_name(dev->dev.parent), 11205 netdev_name(dev), netdev_reg_state(dev), 11206 vaf); 11207 } else if (dev) { 11208 printk("%s%s%s: %pV", 11209 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11210 } else { 11211 printk("%s(NULL net_device): %pV", level, vaf); 11212 } 11213 } 11214 11215 void netdev_printk(const char *level, const struct net_device *dev, 11216 const char *format, ...) 11217 { 11218 struct va_format vaf; 11219 va_list args; 11220 11221 va_start(args, format); 11222 11223 vaf.fmt = format; 11224 vaf.va = &args; 11225 11226 __netdev_printk(level, dev, &vaf); 11227 11228 va_end(args); 11229 } 11230 EXPORT_SYMBOL(netdev_printk); 11231 11232 #define define_netdev_printk_level(func, level) \ 11233 void func(const struct net_device *dev, const char *fmt, ...) \ 11234 { \ 11235 struct va_format vaf; \ 11236 va_list args; \ 11237 \ 11238 va_start(args, fmt); \ 11239 \ 11240 vaf.fmt = fmt; \ 11241 vaf.va = &args; \ 11242 \ 11243 __netdev_printk(level, dev, &vaf); \ 11244 \ 11245 va_end(args); \ 11246 } \ 11247 EXPORT_SYMBOL(func); 11248 11249 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11250 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11251 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11252 define_netdev_printk_level(netdev_err, KERN_ERR); 11253 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11254 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11255 define_netdev_printk_level(netdev_info, KERN_INFO); 11256 11257 static void __net_exit netdev_exit(struct net *net) 11258 { 11259 kfree(net->dev_name_head); 11260 kfree(net->dev_index_head); 11261 if (net != &init_net) 11262 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11263 } 11264 11265 static struct pernet_operations __net_initdata netdev_net_ops = { 11266 .init = netdev_init, 11267 .exit = netdev_exit, 11268 }; 11269 11270 static void __net_exit default_device_exit_net(struct net *net) 11271 { 11272 struct net_device *dev, *aux; 11273 /* 11274 * Push all migratable network devices back to the 11275 * initial network namespace 11276 */ 11277 ASSERT_RTNL(); 11278 for_each_netdev_safe(net, dev, aux) { 11279 int err; 11280 char fb_name[IFNAMSIZ]; 11281 11282 /* Ignore unmoveable devices (i.e. loopback) */ 11283 if (dev->features & NETIF_F_NETNS_LOCAL) 11284 continue; 11285 11286 /* Leave virtual devices for the generic cleanup */ 11287 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11288 continue; 11289 11290 /* Push remaining network devices to init_net */ 11291 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11292 if (netdev_name_in_use(&init_net, fb_name)) 11293 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11294 err = dev_change_net_namespace(dev, &init_net, fb_name); 11295 if (err) { 11296 pr_emerg("%s: failed to move %s to init_net: %d\n", 11297 __func__, dev->name, err); 11298 BUG(); 11299 } 11300 } 11301 } 11302 11303 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11304 { 11305 /* At exit all network devices most be removed from a network 11306 * namespace. Do this in the reverse order of registration. 11307 * Do this across as many network namespaces as possible to 11308 * improve batching efficiency. 11309 */ 11310 struct net_device *dev; 11311 struct net *net; 11312 LIST_HEAD(dev_kill_list); 11313 11314 rtnl_lock(); 11315 list_for_each_entry(net, net_list, exit_list) { 11316 default_device_exit_net(net); 11317 cond_resched(); 11318 } 11319 11320 list_for_each_entry(net, net_list, exit_list) { 11321 for_each_netdev_reverse(net, dev) { 11322 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11323 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11324 else 11325 unregister_netdevice_queue(dev, &dev_kill_list); 11326 } 11327 } 11328 unregister_netdevice_many(&dev_kill_list); 11329 rtnl_unlock(); 11330 } 11331 11332 static struct pernet_operations __net_initdata default_device_ops = { 11333 .exit_batch = default_device_exit_batch, 11334 }; 11335 11336 /* 11337 * Initialize the DEV module. At boot time this walks the device list and 11338 * unhooks any devices that fail to initialise (normally hardware not 11339 * present) and leaves us with a valid list of present and active devices. 11340 * 11341 */ 11342 11343 /* 11344 * This is called single threaded during boot, so no need 11345 * to take the rtnl semaphore. 11346 */ 11347 static int __init net_dev_init(void) 11348 { 11349 int i, rc = -ENOMEM; 11350 11351 BUG_ON(!dev_boot_phase); 11352 11353 if (dev_proc_init()) 11354 goto out; 11355 11356 if (netdev_kobject_init()) 11357 goto out; 11358 11359 INIT_LIST_HEAD(&ptype_all); 11360 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11361 INIT_LIST_HEAD(&ptype_base[i]); 11362 11363 if (register_pernet_subsys(&netdev_net_ops)) 11364 goto out; 11365 11366 /* 11367 * Initialise the packet receive queues. 11368 */ 11369 11370 for_each_possible_cpu(i) { 11371 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11372 struct softnet_data *sd = &per_cpu(softnet_data, i); 11373 11374 INIT_WORK(flush, flush_backlog); 11375 11376 skb_queue_head_init(&sd->input_pkt_queue); 11377 skb_queue_head_init(&sd->process_queue); 11378 #ifdef CONFIG_XFRM_OFFLOAD 11379 skb_queue_head_init(&sd->xfrm_backlog); 11380 #endif 11381 INIT_LIST_HEAD(&sd->poll_list); 11382 sd->output_queue_tailp = &sd->output_queue; 11383 #ifdef CONFIG_RPS 11384 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11385 sd->cpu = i; 11386 #endif 11387 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11388 spin_lock_init(&sd->defer_lock); 11389 11390 init_gro_hash(&sd->backlog); 11391 sd->backlog.poll = process_backlog; 11392 sd->backlog.weight = weight_p; 11393 } 11394 11395 dev_boot_phase = 0; 11396 11397 /* The loopback device is special if any other network devices 11398 * is present in a network namespace the loopback device must 11399 * be present. Since we now dynamically allocate and free the 11400 * loopback device ensure this invariant is maintained by 11401 * keeping the loopback device as the first device on the 11402 * list of network devices. Ensuring the loopback devices 11403 * is the first device that appears and the last network device 11404 * that disappears. 11405 */ 11406 if (register_pernet_device(&loopback_net_ops)) 11407 goto out; 11408 11409 if (register_pernet_device(&default_device_ops)) 11410 goto out; 11411 11412 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11413 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11414 11415 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11416 NULL, dev_cpu_dead); 11417 WARN_ON(rc < 0); 11418 rc = 0; 11419 out: 11420 return rc; 11421 } 11422 11423 subsys_initcall(net_dev_init); 11424