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 .daddr = daddr, 686 }; 687 struct net_device_path *path; 688 int ret = 0; 689 690 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_inherit_tso_max() - copy all TSO limits from a lower device to an upper 2997 * @to: netdev to update 2998 * @from: netdev from which to copy the limits 2999 */ 3000 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3001 { 3002 netif_set_gso_max_size(to, from->gso_max_size); 3003 netif_set_gso_max_segs(to, from->gso_max_segs); 3004 } 3005 EXPORT_SYMBOL(netif_inherit_tso_max); 3006 3007 /** 3008 * netif_get_num_default_rss_queues - default number of RSS queues 3009 * 3010 * Default value is the number of physical cores if there are only 1 or 2, or 3011 * divided by 2 if there are more. 3012 */ 3013 int netif_get_num_default_rss_queues(void) 3014 { 3015 cpumask_var_t cpus; 3016 int cpu, count = 0; 3017 3018 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3019 return 1; 3020 3021 cpumask_copy(cpus, cpu_online_mask); 3022 for_each_cpu(cpu, cpus) { 3023 ++count; 3024 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3025 } 3026 free_cpumask_var(cpus); 3027 3028 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3029 } 3030 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3031 3032 static void __netif_reschedule(struct Qdisc *q) 3033 { 3034 struct softnet_data *sd; 3035 unsigned long flags; 3036 3037 local_irq_save(flags); 3038 sd = this_cpu_ptr(&softnet_data); 3039 q->next_sched = NULL; 3040 *sd->output_queue_tailp = q; 3041 sd->output_queue_tailp = &q->next_sched; 3042 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3043 local_irq_restore(flags); 3044 } 3045 3046 void __netif_schedule(struct Qdisc *q) 3047 { 3048 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3049 __netif_reschedule(q); 3050 } 3051 EXPORT_SYMBOL(__netif_schedule); 3052 3053 struct dev_kfree_skb_cb { 3054 enum skb_free_reason reason; 3055 }; 3056 3057 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3058 { 3059 return (struct dev_kfree_skb_cb *)skb->cb; 3060 } 3061 3062 void netif_schedule_queue(struct netdev_queue *txq) 3063 { 3064 rcu_read_lock(); 3065 if (!netif_xmit_stopped(txq)) { 3066 struct Qdisc *q = rcu_dereference(txq->qdisc); 3067 3068 __netif_schedule(q); 3069 } 3070 rcu_read_unlock(); 3071 } 3072 EXPORT_SYMBOL(netif_schedule_queue); 3073 3074 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3075 { 3076 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3077 struct Qdisc *q; 3078 3079 rcu_read_lock(); 3080 q = rcu_dereference(dev_queue->qdisc); 3081 __netif_schedule(q); 3082 rcu_read_unlock(); 3083 } 3084 } 3085 EXPORT_SYMBOL(netif_tx_wake_queue); 3086 3087 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3088 { 3089 unsigned long flags; 3090 3091 if (unlikely(!skb)) 3092 return; 3093 3094 if (likely(refcount_read(&skb->users) == 1)) { 3095 smp_rmb(); 3096 refcount_set(&skb->users, 0); 3097 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3098 return; 3099 } 3100 get_kfree_skb_cb(skb)->reason = reason; 3101 local_irq_save(flags); 3102 skb->next = __this_cpu_read(softnet_data.completion_queue); 3103 __this_cpu_write(softnet_data.completion_queue, skb); 3104 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3105 local_irq_restore(flags); 3106 } 3107 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3108 3109 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3110 { 3111 if (in_hardirq() || irqs_disabled()) 3112 __dev_kfree_skb_irq(skb, reason); 3113 else 3114 dev_kfree_skb(skb); 3115 } 3116 EXPORT_SYMBOL(__dev_kfree_skb_any); 3117 3118 3119 /** 3120 * netif_device_detach - mark device as removed 3121 * @dev: network device 3122 * 3123 * Mark device as removed from system and therefore no longer available. 3124 */ 3125 void netif_device_detach(struct net_device *dev) 3126 { 3127 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3128 netif_running(dev)) { 3129 netif_tx_stop_all_queues(dev); 3130 } 3131 } 3132 EXPORT_SYMBOL(netif_device_detach); 3133 3134 /** 3135 * netif_device_attach - mark device as attached 3136 * @dev: network device 3137 * 3138 * Mark device as attached from system and restart if needed. 3139 */ 3140 void netif_device_attach(struct net_device *dev) 3141 { 3142 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3143 netif_running(dev)) { 3144 netif_tx_wake_all_queues(dev); 3145 __netdev_watchdog_up(dev); 3146 } 3147 } 3148 EXPORT_SYMBOL(netif_device_attach); 3149 3150 /* 3151 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3152 * to be used as a distribution range. 3153 */ 3154 static u16 skb_tx_hash(const struct net_device *dev, 3155 const struct net_device *sb_dev, 3156 struct sk_buff *skb) 3157 { 3158 u32 hash; 3159 u16 qoffset = 0; 3160 u16 qcount = dev->real_num_tx_queues; 3161 3162 if (dev->num_tc) { 3163 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3164 3165 qoffset = sb_dev->tc_to_txq[tc].offset; 3166 qcount = sb_dev->tc_to_txq[tc].count; 3167 if (unlikely(!qcount)) { 3168 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3169 sb_dev->name, qoffset, tc); 3170 qoffset = 0; 3171 qcount = dev->real_num_tx_queues; 3172 } 3173 } 3174 3175 if (skb_rx_queue_recorded(skb)) { 3176 hash = skb_get_rx_queue(skb); 3177 if (hash >= qoffset) 3178 hash -= qoffset; 3179 while (unlikely(hash >= qcount)) 3180 hash -= qcount; 3181 return hash + qoffset; 3182 } 3183 3184 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3185 } 3186 3187 static void skb_warn_bad_offload(const struct sk_buff *skb) 3188 { 3189 static const netdev_features_t null_features; 3190 struct net_device *dev = skb->dev; 3191 const char *name = ""; 3192 3193 if (!net_ratelimit()) 3194 return; 3195 3196 if (dev) { 3197 if (dev->dev.parent) 3198 name = dev_driver_string(dev->dev.parent); 3199 else 3200 name = netdev_name(dev); 3201 } 3202 skb_dump(KERN_WARNING, skb, false); 3203 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3204 name, dev ? &dev->features : &null_features, 3205 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3206 } 3207 3208 /* 3209 * Invalidate hardware checksum when packet is to be mangled, and 3210 * complete checksum manually on outgoing path. 3211 */ 3212 int skb_checksum_help(struct sk_buff *skb) 3213 { 3214 __wsum csum; 3215 int ret = 0, offset; 3216 3217 if (skb->ip_summed == CHECKSUM_COMPLETE) 3218 goto out_set_summed; 3219 3220 if (unlikely(skb_is_gso(skb))) { 3221 skb_warn_bad_offload(skb); 3222 return -EINVAL; 3223 } 3224 3225 /* Before computing a checksum, we should make sure no frag could 3226 * be modified by an external entity : checksum could be wrong. 3227 */ 3228 if (skb_has_shared_frag(skb)) { 3229 ret = __skb_linearize(skb); 3230 if (ret) 3231 goto out; 3232 } 3233 3234 offset = skb_checksum_start_offset(skb); 3235 BUG_ON(offset >= skb_headlen(skb)); 3236 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3237 3238 offset += skb->csum_offset; 3239 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3240 3241 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3242 if (ret) 3243 goto out; 3244 3245 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3246 out_set_summed: 3247 skb->ip_summed = CHECKSUM_NONE; 3248 out: 3249 return ret; 3250 } 3251 EXPORT_SYMBOL(skb_checksum_help); 3252 3253 int skb_crc32c_csum_help(struct sk_buff *skb) 3254 { 3255 __le32 crc32c_csum; 3256 int ret = 0, offset, start; 3257 3258 if (skb->ip_summed != CHECKSUM_PARTIAL) 3259 goto out; 3260 3261 if (unlikely(skb_is_gso(skb))) 3262 goto out; 3263 3264 /* Before computing a checksum, we should make sure no frag could 3265 * be modified by an external entity : checksum could be wrong. 3266 */ 3267 if (unlikely(skb_has_shared_frag(skb))) { 3268 ret = __skb_linearize(skb); 3269 if (ret) 3270 goto out; 3271 } 3272 start = skb_checksum_start_offset(skb); 3273 offset = start + offsetof(struct sctphdr, checksum); 3274 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3275 ret = -EINVAL; 3276 goto out; 3277 } 3278 3279 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3280 if (ret) 3281 goto out; 3282 3283 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3284 skb->len - start, ~(__u32)0, 3285 crc32c_csum_stub)); 3286 *(__le32 *)(skb->data + offset) = crc32c_csum; 3287 skb->ip_summed = CHECKSUM_NONE; 3288 skb->csum_not_inet = 0; 3289 out: 3290 return ret; 3291 } 3292 3293 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3294 { 3295 __be16 type = skb->protocol; 3296 3297 /* Tunnel gso handlers can set protocol to ethernet. */ 3298 if (type == htons(ETH_P_TEB)) { 3299 struct ethhdr *eth; 3300 3301 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3302 return 0; 3303 3304 eth = (struct ethhdr *)skb->data; 3305 type = eth->h_proto; 3306 } 3307 3308 return __vlan_get_protocol(skb, type, depth); 3309 } 3310 3311 /* openvswitch calls this on rx path, so we need a different check. 3312 */ 3313 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3314 { 3315 if (tx_path) 3316 return skb->ip_summed != CHECKSUM_PARTIAL && 3317 skb->ip_summed != CHECKSUM_UNNECESSARY; 3318 3319 return skb->ip_summed == CHECKSUM_NONE; 3320 } 3321 3322 /** 3323 * __skb_gso_segment - Perform segmentation on skb. 3324 * @skb: buffer to segment 3325 * @features: features for the output path (see dev->features) 3326 * @tx_path: whether it is called in TX path 3327 * 3328 * This function segments the given skb and returns a list of segments. 3329 * 3330 * It may return NULL if the skb requires no segmentation. This is 3331 * only possible when GSO is used for verifying header integrity. 3332 * 3333 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3334 */ 3335 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3336 netdev_features_t features, bool tx_path) 3337 { 3338 struct sk_buff *segs; 3339 3340 if (unlikely(skb_needs_check(skb, tx_path))) { 3341 int err; 3342 3343 /* We're going to init ->check field in TCP or UDP header */ 3344 err = skb_cow_head(skb, 0); 3345 if (err < 0) 3346 return ERR_PTR(err); 3347 } 3348 3349 /* Only report GSO partial support if it will enable us to 3350 * support segmentation on this frame without needing additional 3351 * work. 3352 */ 3353 if (features & NETIF_F_GSO_PARTIAL) { 3354 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3355 struct net_device *dev = skb->dev; 3356 3357 partial_features |= dev->features & dev->gso_partial_features; 3358 if (!skb_gso_ok(skb, features | partial_features)) 3359 features &= ~NETIF_F_GSO_PARTIAL; 3360 } 3361 3362 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3363 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3364 3365 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3366 SKB_GSO_CB(skb)->encap_level = 0; 3367 3368 skb_reset_mac_header(skb); 3369 skb_reset_mac_len(skb); 3370 3371 segs = skb_mac_gso_segment(skb, features); 3372 3373 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3374 skb_warn_bad_offload(skb); 3375 3376 return segs; 3377 } 3378 EXPORT_SYMBOL(__skb_gso_segment); 3379 3380 /* Take action when hardware reception checksum errors are detected. */ 3381 #ifdef CONFIG_BUG 3382 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3383 { 3384 netdev_err(dev, "hw csum failure\n"); 3385 skb_dump(KERN_ERR, skb, true); 3386 dump_stack(); 3387 } 3388 3389 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3390 { 3391 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3392 } 3393 EXPORT_SYMBOL(netdev_rx_csum_fault); 3394 #endif 3395 3396 /* XXX: check that highmem exists at all on the given machine. */ 3397 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3398 { 3399 #ifdef CONFIG_HIGHMEM 3400 int i; 3401 3402 if (!(dev->features & NETIF_F_HIGHDMA)) { 3403 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3404 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3405 3406 if (PageHighMem(skb_frag_page(frag))) 3407 return 1; 3408 } 3409 } 3410 #endif 3411 return 0; 3412 } 3413 3414 /* If MPLS offload request, verify we are testing hardware MPLS features 3415 * instead of standard features for the netdev. 3416 */ 3417 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3418 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3419 netdev_features_t features, 3420 __be16 type) 3421 { 3422 if (eth_p_mpls(type)) 3423 features &= skb->dev->mpls_features; 3424 3425 return features; 3426 } 3427 #else 3428 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3429 netdev_features_t features, 3430 __be16 type) 3431 { 3432 return features; 3433 } 3434 #endif 3435 3436 static netdev_features_t harmonize_features(struct sk_buff *skb, 3437 netdev_features_t features) 3438 { 3439 __be16 type; 3440 3441 type = skb_network_protocol(skb, NULL); 3442 features = net_mpls_features(skb, features, type); 3443 3444 if (skb->ip_summed != CHECKSUM_NONE && 3445 !can_checksum_protocol(features, type)) { 3446 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3447 } 3448 if (illegal_highdma(skb->dev, skb)) 3449 features &= ~NETIF_F_SG; 3450 3451 return features; 3452 } 3453 3454 netdev_features_t passthru_features_check(struct sk_buff *skb, 3455 struct net_device *dev, 3456 netdev_features_t features) 3457 { 3458 return features; 3459 } 3460 EXPORT_SYMBOL(passthru_features_check); 3461 3462 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3463 struct net_device *dev, 3464 netdev_features_t features) 3465 { 3466 return vlan_features_check(skb, features); 3467 } 3468 3469 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3470 struct net_device *dev, 3471 netdev_features_t features) 3472 { 3473 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3474 3475 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3476 return features & ~NETIF_F_GSO_MASK; 3477 3478 if (!skb_shinfo(skb)->gso_type) { 3479 skb_warn_bad_offload(skb); 3480 return features & ~NETIF_F_GSO_MASK; 3481 } 3482 3483 /* Support for GSO partial features requires software 3484 * intervention before we can actually process the packets 3485 * so we need to strip support for any partial features now 3486 * and we can pull them back in after we have partially 3487 * segmented the frame. 3488 */ 3489 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3490 features &= ~dev->gso_partial_features; 3491 3492 /* Make sure to clear the IPv4 ID mangling feature if the 3493 * IPv4 header has the potential to be fragmented. 3494 */ 3495 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3496 struct iphdr *iph = skb->encapsulation ? 3497 inner_ip_hdr(skb) : ip_hdr(skb); 3498 3499 if (!(iph->frag_off & htons(IP_DF))) 3500 features &= ~NETIF_F_TSO_MANGLEID; 3501 } 3502 3503 return features; 3504 } 3505 3506 netdev_features_t netif_skb_features(struct sk_buff *skb) 3507 { 3508 struct net_device *dev = skb->dev; 3509 netdev_features_t features = dev->features; 3510 3511 if (skb_is_gso(skb)) 3512 features = gso_features_check(skb, dev, features); 3513 3514 /* If encapsulation offload request, verify we are testing 3515 * hardware encapsulation features instead of standard 3516 * features for the netdev 3517 */ 3518 if (skb->encapsulation) 3519 features &= dev->hw_enc_features; 3520 3521 if (skb_vlan_tagged(skb)) 3522 features = netdev_intersect_features(features, 3523 dev->vlan_features | 3524 NETIF_F_HW_VLAN_CTAG_TX | 3525 NETIF_F_HW_VLAN_STAG_TX); 3526 3527 if (dev->netdev_ops->ndo_features_check) 3528 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3529 features); 3530 else 3531 features &= dflt_features_check(skb, dev, features); 3532 3533 return harmonize_features(skb, features); 3534 } 3535 EXPORT_SYMBOL(netif_skb_features); 3536 3537 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3538 struct netdev_queue *txq, bool more) 3539 { 3540 unsigned int len; 3541 int rc; 3542 3543 if (dev_nit_active(dev)) 3544 dev_queue_xmit_nit(skb, dev); 3545 3546 len = skb->len; 3547 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3548 trace_net_dev_start_xmit(skb, dev); 3549 rc = netdev_start_xmit(skb, dev, txq, more); 3550 trace_net_dev_xmit(skb, rc, dev, len); 3551 3552 return rc; 3553 } 3554 3555 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3556 struct netdev_queue *txq, int *ret) 3557 { 3558 struct sk_buff *skb = first; 3559 int rc = NETDEV_TX_OK; 3560 3561 while (skb) { 3562 struct sk_buff *next = skb->next; 3563 3564 skb_mark_not_on_list(skb); 3565 rc = xmit_one(skb, dev, txq, next != NULL); 3566 if (unlikely(!dev_xmit_complete(rc))) { 3567 skb->next = next; 3568 goto out; 3569 } 3570 3571 skb = next; 3572 if (netif_tx_queue_stopped(txq) && skb) { 3573 rc = NETDEV_TX_BUSY; 3574 break; 3575 } 3576 } 3577 3578 out: 3579 *ret = rc; 3580 return skb; 3581 } 3582 3583 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3584 netdev_features_t features) 3585 { 3586 if (skb_vlan_tag_present(skb) && 3587 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3588 skb = __vlan_hwaccel_push_inside(skb); 3589 return skb; 3590 } 3591 3592 int skb_csum_hwoffload_help(struct sk_buff *skb, 3593 const netdev_features_t features) 3594 { 3595 if (unlikely(skb_csum_is_sctp(skb))) 3596 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3597 skb_crc32c_csum_help(skb); 3598 3599 if (features & NETIF_F_HW_CSUM) 3600 return 0; 3601 3602 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3603 switch (skb->csum_offset) { 3604 case offsetof(struct tcphdr, check): 3605 case offsetof(struct udphdr, check): 3606 return 0; 3607 } 3608 } 3609 3610 return skb_checksum_help(skb); 3611 } 3612 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3613 3614 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3615 { 3616 netdev_features_t features; 3617 3618 features = netif_skb_features(skb); 3619 skb = validate_xmit_vlan(skb, features); 3620 if (unlikely(!skb)) 3621 goto out_null; 3622 3623 skb = sk_validate_xmit_skb(skb, dev); 3624 if (unlikely(!skb)) 3625 goto out_null; 3626 3627 if (netif_needs_gso(skb, features)) { 3628 struct sk_buff *segs; 3629 3630 segs = skb_gso_segment(skb, features); 3631 if (IS_ERR(segs)) { 3632 goto out_kfree_skb; 3633 } else if (segs) { 3634 consume_skb(skb); 3635 skb = segs; 3636 } 3637 } else { 3638 if (skb_needs_linearize(skb, features) && 3639 __skb_linearize(skb)) 3640 goto out_kfree_skb; 3641 3642 /* If packet is not checksummed and device does not 3643 * support checksumming for this protocol, complete 3644 * checksumming here. 3645 */ 3646 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3647 if (skb->encapsulation) 3648 skb_set_inner_transport_header(skb, 3649 skb_checksum_start_offset(skb)); 3650 else 3651 skb_set_transport_header(skb, 3652 skb_checksum_start_offset(skb)); 3653 if (skb_csum_hwoffload_help(skb, features)) 3654 goto out_kfree_skb; 3655 } 3656 } 3657 3658 skb = validate_xmit_xfrm(skb, features, again); 3659 3660 return skb; 3661 3662 out_kfree_skb: 3663 kfree_skb(skb); 3664 out_null: 3665 dev_core_stats_tx_dropped_inc(dev); 3666 return NULL; 3667 } 3668 3669 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3670 { 3671 struct sk_buff *next, *head = NULL, *tail; 3672 3673 for (; skb != NULL; skb = next) { 3674 next = skb->next; 3675 skb_mark_not_on_list(skb); 3676 3677 /* in case skb wont be segmented, point to itself */ 3678 skb->prev = skb; 3679 3680 skb = validate_xmit_skb(skb, dev, again); 3681 if (!skb) 3682 continue; 3683 3684 if (!head) 3685 head = skb; 3686 else 3687 tail->next = skb; 3688 /* If skb was segmented, skb->prev points to 3689 * the last segment. If not, it still contains skb. 3690 */ 3691 tail = skb->prev; 3692 } 3693 return head; 3694 } 3695 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3696 3697 static void qdisc_pkt_len_init(struct sk_buff *skb) 3698 { 3699 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3700 3701 qdisc_skb_cb(skb)->pkt_len = skb->len; 3702 3703 /* To get more precise estimation of bytes sent on wire, 3704 * we add to pkt_len the headers size of all segments 3705 */ 3706 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3707 unsigned int hdr_len; 3708 u16 gso_segs = shinfo->gso_segs; 3709 3710 /* mac layer + network layer */ 3711 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3712 3713 /* + transport layer */ 3714 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3715 const struct tcphdr *th; 3716 struct tcphdr _tcphdr; 3717 3718 th = skb_header_pointer(skb, skb_transport_offset(skb), 3719 sizeof(_tcphdr), &_tcphdr); 3720 if (likely(th)) 3721 hdr_len += __tcp_hdrlen(th); 3722 } else { 3723 struct udphdr _udphdr; 3724 3725 if (skb_header_pointer(skb, skb_transport_offset(skb), 3726 sizeof(_udphdr), &_udphdr)) 3727 hdr_len += sizeof(struct udphdr); 3728 } 3729 3730 if (shinfo->gso_type & SKB_GSO_DODGY) 3731 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3732 shinfo->gso_size); 3733 3734 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3735 } 3736 } 3737 3738 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3739 struct sk_buff **to_free, 3740 struct netdev_queue *txq) 3741 { 3742 int rc; 3743 3744 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3745 if (rc == NET_XMIT_SUCCESS) 3746 trace_qdisc_enqueue(q, txq, skb); 3747 return rc; 3748 } 3749 3750 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3751 struct net_device *dev, 3752 struct netdev_queue *txq) 3753 { 3754 spinlock_t *root_lock = qdisc_lock(q); 3755 struct sk_buff *to_free = NULL; 3756 bool contended; 3757 int rc; 3758 3759 qdisc_calculate_pkt_len(skb, q); 3760 3761 if (q->flags & TCQ_F_NOLOCK) { 3762 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3763 qdisc_run_begin(q)) { 3764 /* Retest nolock_qdisc_is_empty() within the protection 3765 * of q->seqlock to protect from racing with requeuing. 3766 */ 3767 if (unlikely(!nolock_qdisc_is_empty(q))) { 3768 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3769 __qdisc_run(q); 3770 qdisc_run_end(q); 3771 3772 goto no_lock_out; 3773 } 3774 3775 qdisc_bstats_cpu_update(q, skb); 3776 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3777 !nolock_qdisc_is_empty(q)) 3778 __qdisc_run(q); 3779 3780 qdisc_run_end(q); 3781 return NET_XMIT_SUCCESS; 3782 } 3783 3784 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3785 qdisc_run(q); 3786 3787 no_lock_out: 3788 if (unlikely(to_free)) 3789 kfree_skb_list_reason(to_free, 3790 SKB_DROP_REASON_QDISC_DROP); 3791 return rc; 3792 } 3793 3794 /* 3795 * Heuristic to force contended enqueues to serialize on a 3796 * separate lock before trying to get qdisc main lock. 3797 * This permits qdisc->running owner to get the lock more 3798 * often and dequeue packets faster. 3799 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3800 * and then other tasks will only enqueue packets. The packets will be 3801 * sent after the qdisc owner is scheduled again. To prevent this 3802 * scenario the task always serialize on the lock. 3803 */ 3804 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3805 if (unlikely(contended)) 3806 spin_lock(&q->busylock); 3807 3808 spin_lock(root_lock); 3809 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3810 __qdisc_drop(skb, &to_free); 3811 rc = NET_XMIT_DROP; 3812 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3813 qdisc_run_begin(q)) { 3814 /* 3815 * This is a work-conserving queue; there are no old skbs 3816 * waiting to be sent out; and the qdisc is not running - 3817 * xmit the skb directly. 3818 */ 3819 3820 qdisc_bstats_update(q, skb); 3821 3822 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3823 if (unlikely(contended)) { 3824 spin_unlock(&q->busylock); 3825 contended = false; 3826 } 3827 __qdisc_run(q); 3828 } 3829 3830 qdisc_run_end(q); 3831 rc = NET_XMIT_SUCCESS; 3832 } else { 3833 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3834 if (qdisc_run_begin(q)) { 3835 if (unlikely(contended)) { 3836 spin_unlock(&q->busylock); 3837 contended = false; 3838 } 3839 __qdisc_run(q); 3840 qdisc_run_end(q); 3841 } 3842 } 3843 spin_unlock(root_lock); 3844 if (unlikely(to_free)) 3845 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3846 if (unlikely(contended)) 3847 spin_unlock(&q->busylock); 3848 return rc; 3849 } 3850 3851 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3852 static void skb_update_prio(struct sk_buff *skb) 3853 { 3854 const struct netprio_map *map; 3855 const struct sock *sk; 3856 unsigned int prioidx; 3857 3858 if (skb->priority) 3859 return; 3860 map = rcu_dereference_bh(skb->dev->priomap); 3861 if (!map) 3862 return; 3863 sk = skb_to_full_sk(skb); 3864 if (!sk) 3865 return; 3866 3867 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3868 3869 if (prioidx < map->priomap_len) 3870 skb->priority = map->priomap[prioidx]; 3871 } 3872 #else 3873 #define skb_update_prio(skb) 3874 #endif 3875 3876 /** 3877 * dev_loopback_xmit - loop back @skb 3878 * @net: network namespace this loopback is happening in 3879 * @sk: sk needed to be a netfilter okfn 3880 * @skb: buffer to transmit 3881 */ 3882 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3883 { 3884 skb_reset_mac_header(skb); 3885 __skb_pull(skb, skb_network_offset(skb)); 3886 skb->pkt_type = PACKET_LOOPBACK; 3887 if (skb->ip_summed == CHECKSUM_NONE) 3888 skb->ip_summed = CHECKSUM_UNNECESSARY; 3889 WARN_ON(!skb_dst(skb)); 3890 skb_dst_force(skb); 3891 netif_rx(skb); 3892 return 0; 3893 } 3894 EXPORT_SYMBOL(dev_loopback_xmit); 3895 3896 #ifdef CONFIG_NET_EGRESS 3897 static struct sk_buff * 3898 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3899 { 3900 #ifdef CONFIG_NET_CLS_ACT 3901 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3902 struct tcf_result cl_res; 3903 3904 if (!miniq) 3905 return skb; 3906 3907 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3908 tc_skb_cb(skb)->mru = 0; 3909 tc_skb_cb(skb)->post_ct = false; 3910 mini_qdisc_bstats_cpu_update(miniq, skb); 3911 3912 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3913 case TC_ACT_OK: 3914 case TC_ACT_RECLASSIFY: 3915 skb->tc_index = TC_H_MIN(cl_res.classid); 3916 break; 3917 case TC_ACT_SHOT: 3918 mini_qdisc_qstats_cpu_drop(miniq); 3919 *ret = NET_XMIT_DROP; 3920 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3921 return NULL; 3922 case TC_ACT_STOLEN: 3923 case TC_ACT_QUEUED: 3924 case TC_ACT_TRAP: 3925 *ret = NET_XMIT_SUCCESS; 3926 consume_skb(skb); 3927 return NULL; 3928 case TC_ACT_REDIRECT: 3929 /* No need to push/pop skb's mac_header here on egress! */ 3930 skb_do_redirect(skb); 3931 *ret = NET_XMIT_SUCCESS; 3932 return NULL; 3933 default: 3934 break; 3935 } 3936 #endif /* CONFIG_NET_CLS_ACT */ 3937 3938 return skb; 3939 } 3940 3941 static struct netdev_queue * 3942 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3943 { 3944 int qm = skb_get_queue_mapping(skb); 3945 3946 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3947 } 3948 3949 static bool netdev_xmit_txqueue_skipped(void) 3950 { 3951 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3952 } 3953 3954 void netdev_xmit_skip_txqueue(bool skip) 3955 { 3956 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3957 } 3958 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3959 #endif /* CONFIG_NET_EGRESS */ 3960 3961 #ifdef CONFIG_XPS 3962 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3963 struct xps_dev_maps *dev_maps, unsigned int tci) 3964 { 3965 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3966 struct xps_map *map; 3967 int queue_index = -1; 3968 3969 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3970 return queue_index; 3971 3972 tci *= dev_maps->num_tc; 3973 tci += tc; 3974 3975 map = rcu_dereference(dev_maps->attr_map[tci]); 3976 if (map) { 3977 if (map->len == 1) 3978 queue_index = map->queues[0]; 3979 else 3980 queue_index = map->queues[reciprocal_scale( 3981 skb_get_hash(skb), map->len)]; 3982 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3983 queue_index = -1; 3984 } 3985 return queue_index; 3986 } 3987 #endif 3988 3989 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3990 struct sk_buff *skb) 3991 { 3992 #ifdef CONFIG_XPS 3993 struct xps_dev_maps *dev_maps; 3994 struct sock *sk = skb->sk; 3995 int queue_index = -1; 3996 3997 if (!static_key_false(&xps_needed)) 3998 return -1; 3999 4000 rcu_read_lock(); 4001 if (!static_key_false(&xps_rxqs_needed)) 4002 goto get_cpus_map; 4003 4004 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4005 if (dev_maps) { 4006 int tci = sk_rx_queue_get(sk); 4007 4008 if (tci >= 0) 4009 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4010 tci); 4011 } 4012 4013 get_cpus_map: 4014 if (queue_index < 0) { 4015 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4016 if (dev_maps) { 4017 unsigned int tci = skb->sender_cpu - 1; 4018 4019 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4020 tci); 4021 } 4022 } 4023 rcu_read_unlock(); 4024 4025 return queue_index; 4026 #else 4027 return -1; 4028 #endif 4029 } 4030 4031 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4032 struct net_device *sb_dev) 4033 { 4034 return 0; 4035 } 4036 EXPORT_SYMBOL(dev_pick_tx_zero); 4037 4038 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4039 struct net_device *sb_dev) 4040 { 4041 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4042 } 4043 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4044 4045 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4046 struct net_device *sb_dev) 4047 { 4048 struct sock *sk = skb->sk; 4049 int queue_index = sk_tx_queue_get(sk); 4050 4051 sb_dev = sb_dev ? : dev; 4052 4053 if (queue_index < 0 || skb->ooo_okay || 4054 queue_index >= dev->real_num_tx_queues) { 4055 int new_index = get_xps_queue(dev, sb_dev, skb); 4056 4057 if (new_index < 0) 4058 new_index = skb_tx_hash(dev, sb_dev, skb); 4059 4060 if (queue_index != new_index && sk && 4061 sk_fullsock(sk) && 4062 rcu_access_pointer(sk->sk_dst_cache)) 4063 sk_tx_queue_set(sk, new_index); 4064 4065 queue_index = new_index; 4066 } 4067 4068 return queue_index; 4069 } 4070 EXPORT_SYMBOL(netdev_pick_tx); 4071 4072 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4073 struct sk_buff *skb, 4074 struct net_device *sb_dev) 4075 { 4076 int queue_index = 0; 4077 4078 #ifdef CONFIG_XPS 4079 u32 sender_cpu = skb->sender_cpu - 1; 4080 4081 if (sender_cpu >= (u32)NR_CPUS) 4082 skb->sender_cpu = raw_smp_processor_id() + 1; 4083 #endif 4084 4085 if (dev->real_num_tx_queues != 1) { 4086 const struct net_device_ops *ops = dev->netdev_ops; 4087 4088 if (ops->ndo_select_queue) 4089 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4090 else 4091 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4092 4093 queue_index = netdev_cap_txqueue(dev, queue_index); 4094 } 4095 4096 skb_set_queue_mapping(skb, queue_index); 4097 return netdev_get_tx_queue(dev, queue_index); 4098 } 4099 4100 /** 4101 * __dev_queue_xmit - transmit a buffer 4102 * @skb: buffer to transmit 4103 * @sb_dev: suboordinate device used for L2 forwarding offload 4104 * 4105 * Queue a buffer for transmission to a network device. The caller must 4106 * have set the device and priority and built the buffer before calling 4107 * this function. The function can be called from an interrupt. 4108 * 4109 * A negative errno code is returned on a failure. A success does not 4110 * guarantee the frame will be transmitted as it may be dropped due 4111 * to congestion or traffic shaping. 4112 * 4113 * ----------------------------------------------------------------------------------- 4114 * I notice this method can also return errors from the queue disciplines, 4115 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4116 * be positive. 4117 * 4118 * Regardless of the return value, the skb is consumed, so it is currently 4119 * difficult to retry a send to this method. (You can bump the ref count 4120 * before sending to hold a reference for retry if you are careful.) 4121 * 4122 * When calling this method, interrupts MUST be enabled. This is because 4123 * the BH enable code must have IRQs enabled so that it will not deadlock. 4124 * --BLG 4125 */ 4126 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4127 { 4128 struct net_device *dev = skb->dev; 4129 struct netdev_queue *txq = NULL; 4130 struct Qdisc *q; 4131 int rc = -ENOMEM; 4132 bool again = false; 4133 4134 skb_reset_mac_header(skb); 4135 4136 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4137 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4138 4139 /* Disable soft irqs for various locks below. Also 4140 * stops preemption for RCU. 4141 */ 4142 rcu_read_lock_bh(); 4143 4144 skb_update_prio(skb); 4145 4146 qdisc_pkt_len_init(skb); 4147 #ifdef CONFIG_NET_CLS_ACT 4148 skb->tc_at_ingress = 0; 4149 #endif 4150 #ifdef CONFIG_NET_EGRESS 4151 if (static_branch_unlikely(&egress_needed_key)) { 4152 if (nf_hook_egress_active()) { 4153 skb = nf_hook_egress(skb, &rc, dev); 4154 if (!skb) 4155 goto out; 4156 } 4157 4158 netdev_xmit_skip_txqueue(false); 4159 4160 nf_skip_egress(skb, true); 4161 skb = sch_handle_egress(skb, &rc, dev); 4162 if (!skb) 4163 goto out; 4164 nf_skip_egress(skb, false); 4165 4166 if (netdev_xmit_txqueue_skipped()) 4167 txq = netdev_tx_queue_mapping(dev, skb); 4168 } 4169 #endif 4170 /* If device/qdisc don't need skb->dst, release it right now while 4171 * its hot in this cpu cache. 4172 */ 4173 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4174 skb_dst_drop(skb); 4175 else 4176 skb_dst_force(skb); 4177 4178 if (!txq) 4179 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4180 4181 q = rcu_dereference_bh(txq->qdisc); 4182 4183 trace_net_dev_queue(skb); 4184 if (q->enqueue) { 4185 rc = __dev_xmit_skb(skb, q, dev, txq); 4186 goto out; 4187 } 4188 4189 /* The device has no queue. Common case for software devices: 4190 * loopback, all the sorts of tunnels... 4191 4192 * Really, it is unlikely that netif_tx_lock protection is necessary 4193 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4194 * counters.) 4195 * However, it is possible, that they rely on protection 4196 * made by us here. 4197 4198 * Check this and shot the lock. It is not prone from deadlocks. 4199 *Either shot noqueue qdisc, it is even simpler 8) 4200 */ 4201 if (dev->flags & IFF_UP) { 4202 int cpu = smp_processor_id(); /* ok because BHs are off */ 4203 4204 /* Other cpus might concurrently change txq->xmit_lock_owner 4205 * to -1 or to their cpu id, but not to our id. 4206 */ 4207 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4208 if (dev_xmit_recursion()) 4209 goto recursion_alert; 4210 4211 skb = validate_xmit_skb(skb, dev, &again); 4212 if (!skb) 4213 goto out; 4214 4215 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4216 HARD_TX_LOCK(dev, txq, cpu); 4217 4218 if (!netif_xmit_stopped(txq)) { 4219 dev_xmit_recursion_inc(); 4220 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4221 dev_xmit_recursion_dec(); 4222 if (dev_xmit_complete(rc)) { 4223 HARD_TX_UNLOCK(dev, txq); 4224 goto out; 4225 } 4226 } 4227 HARD_TX_UNLOCK(dev, txq); 4228 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4229 dev->name); 4230 } else { 4231 /* Recursion is detected! It is possible, 4232 * unfortunately 4233 */ 4234 recursion_alert: 4235 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4236 dev->name); 4237 } 4238 } 4239 4240 rc = -ENETDOWN; 4241 rcu_read_unlock_bh(); 4242 4243 dev_core_stats_tx_dropped_inc(dev); 4244 kfree_skb_list(skb); 4245 return rc; 4246 out: 4247 rcu_read_unlock_bh(); 4248 return rc; 4249 } 4250 EXPORT_SYMBOL(__dev_queue_xmit); 4251 4252 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4253 { 4254 struct net_device *dev = skb->dev; 4255 struct sk_buff *orig_skb = skb; 4256 struct netdev_queue *txq; 4257 int ret = NETDEV_TX_BUSY; 4258 bool again = false; 4259 4260 if (unlikely(!netif_running(dev) || 4261 !netif_carrier_ok(dev))) 4262 goto drop; 4263 4264 skb = validate_xmit_skb_list(skb, dev, &again); 4265 if (skb != orig_skb) 4266 goto drop; 4267 4268 skb_set_queue_mapping(skb, queue_id); 4269 txq = skb_get_tx_queue(dev, skb); 4270 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4271 4272 local_bh_disable(); 4273 4274 dev_xmit_recursion_inc(); 4275 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4276 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4277 ret = netdev_start_xmit(skb, dev, txq, false); 4278 HARD_TX_UNLOCK(dev, txq); 4279 dev_xmit_recursion_dec(); 4280 4281 local_bh_enable(); 4282 return ret; 4283 drop: 4284 dev_core_stats_tx_dropped_inc(dev); 4285 kfree_skb_list(skb); 4286 return NET_XMIT_DROP; 4287 } 4288 EXPORT_SYMBOL(__dev_direct_xmit); 4289 4290 /************************************************************************* 4291 * Receiver routines 4292 *************************************************************************/ 4293 4294 int netdev_max_backlog __read_mostly = 1000; 4295 EXPORT_SYMBOL(netdev_max_backlog); 4296 4297 int netdev_tstamp_prequeue __read_mostly = 1; 4298 int netdev_budget __read_mostly = 300; 4299 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4300 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4301 int weight_p __read_mostly = 64; /* old backlog weight */ 4302 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4303 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4304 int dev_rx_weight __read_mostly = 64; 4305 int dev_tx_weight __read_mostly = 64; 4306 4307 /* Called with irq disabled */ 4308 static inline void ____napi_schedule(struct softnet_data *sd, 4309 struct napi_struct *napi) 4310 { 4311 struct task_struct *thread; 4312 4313 lockdep_assert_irqs_disabled(); 4314 4315 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4316 /* Paired with smp_mb__before_atomic() in 4317 * napi_enable()/dev_set_threaded(). 4318 * Use READ_ONCE() to guarantee a complete 4319 * read on napi->thread. Only call 4320 * wake_up_process() when it's not NULL. 4321 */ 4322 thread = READ_ONCE(napi->thread); 4323 if (thread) { 4324 /* Avoid doing set_bit() if the thread is in 4325 * INTERRUPTIBLE state, cause napi_thread_wait() 4326 * makes sure to proceed with napi polling 4327 * if the thread is explicitly woken from here. 4328 */ 4329 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4330 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4331 wake_up_process(thread); 4332 return; 4333 } 4334 } 4335 4336 list_add_tail(&napi->poll_list, &sd->poll_list); 4337 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4338 } 4339 4340 #ifdef CONFIG_RPS 4341 4342 /* One global table that all flow-based protocols share. */ 4343 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4344 EXPORT_SYMBOL(rps_sock_flow_table); 4345 u32 rps_cpu_mask __read_mostly; 4346 EXPORT_SYMBOL(rps_cpu_mask); 4347 4348 struct static_key_false rps_needed __read_mostly; 4349 EXPORT_SYMBOL(rps_needed); 4350 struct static_key_false rfs_needed __read_mostly; 4351 EXPORT_SYMBOL(rfs_needed); 4352 4353 static struct rps_dev_flow * 4354 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4355 struct rps_dev_flow *rflow, u16 next_cpu) 4356 { 4357 if (next_cpu < nr_cpu_ids) { 4358 #ifdef CONFIG_RFS_ACCEL 4359 struct netdev_rx_queue *rxqueue; 4360 struct rps_dev_flow_table *flow_table; 4361 struct rps_dev_flow *old_rflow; 4362 u32 flow_id; 4363 u16 rxq_index; 4364 int rc; 4365 4366 /* Should we steer this flow to a different hardware queue? */ 4367 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4368 !(dev->features & NETIF_F_NTUPLE)) 4369 goto out; 4370 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4371 if (rxq_index == skb_get_rx_queue(skb)) 4372 goto out; 4373 4374 rxqueue = dev->_rx + rxq_index; 4375 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4376 if (!flow_table) 4377 goto out; 4378 flow_id = skb_get_hash(skb) & flow_table->mask; 4379 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4380 rxq_index, flow_id); 4381 if (rc < 0) 4382 goto out; 4383 old_rflow = rflow; 4384 rflow = &flow_table->flows[flow_id]; 4385 rflow->filter = rc; 4386 if (old_rflow->filter == rflow->filter) 4387 old_rflow->filter = RPS_NO_FILTER; 4388 out: 4389 #endif 4390 rflow->last_qtail = 4391 per_cpu(softnet_data, next_cpu).input_queue_head; 4392 } 4393 4394 rflow->cpu = next_cpu; 4395 return rflow; 4396 } 4397 4398 /* 4399 * get_rps_cpu is called from netif_receive_skb and returns the target 4400 * CPU from the RPS map of the receiving queue for a given skb. 4401 * rcu_read_lock must be held on entry. 4402 */ 4403 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4404 struct rps_dev_flow **rflowp) 4405 { 4406 const struct rps_sock_flow_table *sock_flow_table; 4407 struct netdev_rx_queue *rxqueue = dev->_rx; 4408 struct rps_dev_flow_table *flow_table; 4409 struct rps_map *map; 4410 int cpu = -1; 4411 u32 tcpu; 4412 u32 hash; 4413 4414 if (skb_rx_queue_recorded(skb)) { 4415 u16 index = skb_get_rx_queue(skb); 4416 4417 if (unlikely(index >= dev->real_num_rx_queues)) { 4418 WARN_ONCE(dev->real_num_rx_queues > 1, 4419 "%s received packet on queue %u, but number " 4420 "of RX queues is %u\n", 4421 dev->name, index, dev->real_num_rx_queues); 4422 goto done; 4423 } 4424 rxqueue += index; 4425 } 4426 4427 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4428 4429 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4430 map = rcu_dereference(rxqueue->rps_map); 4431 if (!flow_table && !map) 4432 goto done; 4433 4434 skb_reset_network_header(skb); 4435 hash = skb_get_hash(skb); 4436 if (!hash) 4437 goto done; 4438 4439 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4440 if (flow_table && sock_flow_table) { 4441 struct rps_dev_flow *rflow; 4442 u32 next_cpu; 4443 u32 ident; 4444 4445 /* First check into global flow table if there is a match */ 4446 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4447 if ((ident ^ hash) & ~rps_cpu_mask) 4448 goto try_rps; 4449 4450 next_cpu = ident & rps_cpu_mask; 4451 4452 /* OK, now we know there is a match, 4453 * we can look at the local (per receive queue) flow table 4454 */ 4455 rflow = &flow_table->flows[hash & flow_table->mask]; 4456 tcpu = rflow->cpu; 4457 4458 /* 4459 * If the desired CPU (where last recvmsg was done) is 4460 * different from current CPU (one in the rx-queue flow 4461 * table entry), switch if one of the following holds: 4462 * - Current CPU is unset (>= nr_cpu_ids). 4463 * - Current CPU is offline. 4464 * - The current CPU's queue tail has advanced beyond the 4465 * last packet that was enqueued using this table entry. 4466 * This guarantees that all previous packets for the flow 4467 * have been dequeued, thus preserving in order delivery. 4468 */ 4469 if (unlikely(tcpu != next_cpu) && 4470 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4471 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4472 rflow->last_qtail)) >= 0)) { 4473 tcpu = next_cpu; 4474 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4475 } 4476 4477 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4478 *rflowp = rflow; 4479 cpu = tcpu; 4480 goto done; 4481 } 4482 } 4483 4484 try_rps: 4485 4486 if (map) { 4487 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4488 if (cpu_online(tcpu)) { 4489 cpu = tcpu; 4490 goto done; 4491 } 4492 } 4493 4494 done: 4495 return cpu; 4496 } 4497 4498 #ifdef CONFIG_RFS_ACCEL 4499 4500 /** 4501 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4502 * @dev: Device on which the filter was set 4503 * @rxq_index: RX queue index 4504 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4505 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4506 * 4507 * Drivers that implement ndo_rx_flow_steer() should periodically call 4508 * this function for each installed filter and remove the filters for 4509 * which it returns %true. 4510 */ 4511 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4512 u32 flow_id, u16 filter_id) 4513 { 4514 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4515 struct rps_dev_flow_table *flow_table; 4516 struct rps_dev_flow *rflow; 4517 bool expire = true; 4518 unsigned int cpu; 4519 4520 rcu_read_lock(); 4521 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4522 if (flow_table && flow_id <= flow_table->mask) { 4523 rflow = &flow_table->flows[flow_id]; 4524 cpu = READ_ONCE(rflow->cpu); 4525 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4526 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4527 rflow->last_qtail) < 4528 (int)(10 * flow_table->mask))) 4529 expire = false; 4530 } 4531 rcu_read_unlock(); 4532 return expire; 4533 } 4534 EXPORT_SYMBOL(rps_may_expire_flow); 4535 4536 #endif /* CONFIG_RFS_ACCEL */ 4537 4538 /* Called from hardirq (IPI) context */ 4539 static void rps_trigger_softirq(void *data) 4540 { 4541 struct softnet_data *sd = data; 4542 4543 ____napi_schedule(sd, &sd->backlog); 4544 sd->received_rps++; 4545 } 4546 4547 #endif /* CONFIG_RPS */ 4548 4549 /* Called from hardirq (IPI) context */ 4550 static void trigger_rx_softirq(void *data __always_unused) 4551 { 4552 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4553 } 4554 4555 /* 4556 * Check if this softnet_data structure is another cpu one 4557 * If yes, queue it to our IPI list and return 1 4558 * If no, return 0 4559 */ 4560 static int napi_schedule_rps(struct softnet_data *sd) 4561 { 4562 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4563 4564 #ifdef CONFIG_RPS 4565 if (sd != mysd) { 4566 sd->rps_ipi_next = mysd->rps_ipi_list; 4567 mysd->rps_ipi_list = sd; 4568 4569 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4570 return 1; 4571 } 4572 #endif /* CONFIG_RPS */ 4573 __napi_schedule_irqoff(&mysd->backlog); 4574 return 0; 4575 } 4576 4577 #ifdef CONFIG_NET_FLOW_LIMIT 4578 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4579 #endif 4580 4581 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4582 { 4583 #ifdef CONFIG_NET_FLOW_LIMIT 4584 struct sd_flow_limit *fl; 4585 struct softnet_data *sd; 4586 unsigned int old_flow, new_flow; 4587 4588 if (qlen < (netdev_max_backlog >> 1)) 4589 return false; 4590 4591 sd = this_cpu_ptr(&softnet_data); 4592 4593 rcu_read_lock(); 4594 fl = rcu_dereference(sd->flow_limit); 4595 if (fl) { 4596 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4597 old_flow = fl->history[fl->history_head]; 4598 fl->history[fl->history_head] = new_flow; 4599 4600 fl->history_head++; 4601 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4602 4603 if (likely(fl->buckets[old_flow])) 4604 fl->buckets[old_flow]--; 4605 4606 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4607 fl->count++; 4608 rcu_read_unlock(); 4609 return true; 4610 } 4611 } 4612 rcu_read_unlock(); 4613 #endif 4614 return false; 4615 } 4616 4617 /* 4618 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4619 * queue (may be a remote CPU queue). 4620 */ 4621 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4622 unsigned int *qtail) 4623 { 4624 enum skb_drop_reason reason; 4625 struct softnet_data *sd; 4626 unsigned long flags; 4627 unsigned int qlen; 4628 4629 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4630 sd = &per_cpu(softnet_data, cpu); 4631 4632 rps_lock_irqsave(sd, &flags); 4633 if (!netif_running(skb->dev)) 4634 goto drop; 4635 qlen = skb_queue_len(&sd->input_pkt_queue); 4636 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4637 if (qlen) { 4638 enqueue: 4639 __skb_queue_tail(&sd->input_pkt_queue, skb); 4640 input_queue_tail_incr_save(sd, qtail); 4641 rps_unlock_irq_restore(sd, &flags); 4642 return NET_RX_SUCCESS; 4643 } 4644 4645 /* Schedule NAPI for backlog device 4646 * We can use non atomic operation since we own the queue lock 4647 */ 4648 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4649 napi_schedule_rps(sd); 4650 goto enqueue; 4651 } 4652 reason = SKB_DROP_REASON_CPU_BACKLOG; 4653 4654 drop: 4655 sd->dropped++; 4656 rps_unlock_irq_restore(sd, &flags); 4657 4658 dev_core_stats_rx_dropped_inc(skb->dev); 4659 kfree_skb_reason(skb, reason); 4660 return NET_RX_DROP; 4661 } 4662 4663 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4664 { 4665 struct net_device *dev = skb->dev; 4666 struct netdev_rx_queue *rxqueue; 4667 4668 rxqueue = dev->_rx; 4669 4670 if (skb_rx_queue_recorded(skb)) { 4671 u16 index = skb_get_rx_queue(skb); 4672 4673 if (unlikely(index >= dev->real_num_rx_queues)) { 4674 WARN_ONCE(dev->real_num_rx_queues > 1, 4675 "%s received packet on queue %u, but number " 4676 "of RX queues is %u\n", 4677 dev->name, index, dev->real_num_rx_queues); 4678 4679 return rxqueue; /* Return first rxqueue */ 4680 } 4681 rxqueue += index; 4682 } 4683 return rxqueue; 4684 } 4685 4686 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4687 struct bpf_prog *xdp_prog) 4688 { 4689 void *orig_data, *orig_data_end, *hard_start; 4690 struct netdev_rx_queue *rxqueue; 4691 bool orig_bcast, orig_host; 4692 u32 mac_len, frame_sz; 4693 __be16 orig_eth_type; 4694 struct ethhdr *eth; 4695 u32 metalen, act; 4696 int off; 4697 4698 /* The XDP program wants to see the packet starting at the MAC 4699 * header. 4700 */ 4701 mac_len = skb->data - skb_mac_header(skb); 4702 hard_start = skb->data - skb_headroom(skb); 4703 4704 /* SKB "head" area always have tailroom for skb_shared_info */ 4705 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4706 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4707 4708 rxqueue = netif_get_rxqueue(skb); 4709 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4710 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4711 skb_headlen(skb) + mac_len, true); 4712 4713 orig_data_end = xdp->data_end; 4714 orig_data = xdp->data; 4715 eth = (struct ethhdr *)xdp->data; 4716 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4717 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4718 orig_eth_type = eth->h_proto; 4719 4720 act = bpf_prog_run_xdp(xdp_prog, xdp); 4721 4722 /* check if bpf_xdp_adjust_head was used */ 4723 off = xdp->data - orig_data; 4724 if (off) { 4725 if (off > 0) 4726 __skb_pull(skb, off); 4727 else if (off < 0) 4728 __skb_push(skb, -off); 4729 4730 skb->mac_header += off; 4731 skb_reset_network_header(skb); 4732 } 4733 4734 /* check if bpf_xdp_adjust_tail was used */ 4735 off = xdp->data_end - orig_data_end; 4736 if (off != 0) { 4737 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4738 skb->len += off; /* positive on grow, negative on shrink */ 4739 } 4740 4741 /* check if XDP changed eth hdr such SKB needs update */ 4742 eth = (struct ethhdr *)xdp->data; 4743 if ((orig_eth_type != eth->h_proto) || 4744 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4745 skb->dev->dev_addr)) || 4746 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4747 __skb_push(skb, ETH_HLEN); 4748 skb->pkt_type = PACKET_HOST; 4749 skb->protocol = eth_type_trans(skb, skb->dev); 4750 } 4751 4752 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4753 * before calling us again on redirect path. We do not call do_redirect 4754 * as we leave that up to the caller. 4755 * 4756 * Caller is responsible for managing lifetime of skb (i.e. calling 4757 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4758 */ 4759 switch (act) { 4760 case XDP_REDIRECT: 4761 case XDP_TX: 4762 __skb_push(skb, mac_len); 4763 break; 4764 case XDP_PASS: 4765 metalen = xdp->data - xdp->data_meta; 4766 if (metalen) 4767 skb_metadata_set(skb, metalen); 4768 break; 4769 } 4770 4771 return act; 4772 } 4773 4774 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4775 struct xdp_buff *xdp, 4776 struct bpf_prog *xdp_prog) 4777 { 4778 u32 act = XDP_DROP; 4779 4780 /* Reinjected packets coming from act_mirred or similar should 4781 * not get XDP generic processing. 4782 */ 4783 if (skb_is_redirected(skb)) 4784 return XDP_PASS; 4785 4786 /* XDP packets must be linear and must have sufficient headroom 4787 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4788 * native XDP provides, thus we need to do it here as well. 4789 */ 4790 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4791 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4792 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4793 int troom = skb->tail + skb->data_len - skb->end; 4794 4795 /* In case we have to go down the path and also linearize, 4796 * then lets do the pskb_expand_head() work just once here. 4797 */ 4798 if (pskb_expand_head(skb, 4799 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4800 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4801 goto do_drop; 4802 if (skb_linearize(skb)) 4803 goto do_drop; 4804 } 4805 4806 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4807 switch (act) { 4808 case XDP_REDIRECT: 4809 case XDP_TX: 4810 case XDP_PASS: 4811 break; 4812 default: 4813 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4814 fallthrough; 4815 case XDP_ABORTED: 4816 trace_xdp_exception(skb->dev, xdp_prog, act); 4817 fallthrough; 4818 case XDP_DROP: 4819 do_drop: 4820 kfree_skb(skb); 4821 break; 4822 } 4823 4824 return act; 4825 } 4826 4827 /* When doing generic XDP we have to bypass the qdisc layer and the 4828 * network taps in order to match in-driver-XDP behavior. 4829 */ 4830 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4831 { 4832 struct net_device *dev = skb->dev; 4833 struct netdev_queue *txq; 4834 bool free_skb = true; 4835 int cpu, rc; 4836 4837 txq = netdev_core_pick_tx(dev, skb, NULL); 4838 cpu = smp_processor_id(); 4839 HARD_TX_LOCK(dev, txq, cpu); 4840 if (!netif_xmit_stopped(txq)) { 4841 rc = netdev_start_xmit(skb, dev, txq, 0); 4842 if (dev_xmit_complete(rc)) 4843 free_skb = false; 4844 } 4845 HARD_TX_UNLOCK(dev, txq); 4846 if (free_skb) { 4847 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4848 kfree_skb(skb); 4849 } 4850 } 4851 4852 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4853 4854 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4855 { 4856 if (xdp_prog) { 4857 struct xdp_buff xdp; 4858 u32 act; 4859 int err; 4860 4861 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4862 if (act != XDP_PASS) { 4863 switch (act) { 4864 case XDP_REDIRECT: 4865 err = xdp_do_generic_redirect(skb->dev, skb, 4866 &xdp, xdp_prog); 4867 if (err) 4868 goto out_redir; 4869 break; 4870 case XDP_TX: 4871 generic_xdp_tx(skb, xdp_prog); 4872 break; 4873 } 4874 return XDP_DROP; 4875 } 4876 } 4877 return XDP_PASS; 4878 out_redir: 4879 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4880 return XDP_DROP; 4881 } 4882 EXPORT_SYMBOL_GPL(do_xdp_generic); 4883 4884 static int netif_rx_internal(struct sk_buff *skb) 4885 { 4886 int ret; 4887 4888 net_timestamp_check(netdev_tstamp_prequeue, skb); 4889 4890 trace_netif_rx(skb); 4891 4892 #ifdef CONFIG_RPS 4893 if (static_branch_unlikely(&rps_needed)) { 4894 struct rps_dev_flow voidflow, *rflow = &voidflow; 4895 int cpu; 4896 4897 rcu_read_lock(); 4898 4899 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4900 if (cpu < 0) 4901 cpu = smp_processor_id(); 4902 4903 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4904 4905 rcu_read_unlock(); 4906 } else 4907 #endif 4908 { 4909 unsigned int qtail; 4910 4911 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4912 } 4913 return ret; 4914 } 4915 4916 /** 4917 * __netif_rx - Slightly optimized version of netif_rx 4918 * @skb: buffer to post 4919 * 4920 * This behaves as netif_rx except that it does not disable bottom halves. 4921 * As a result this function may only be invoked from the interrupt context 4922 * (either hard or soft interrupt). 4923 */ 4924 int __netif_rx(struct sk_buff *skb) 4925 { 4926 int ret; 4927 4928 lockdep_assert_once(hardirq_count() | softirq_count()); 4929 4930 trace_netif_rx_entry(skb); 4931 ret = netif_rx_internal(skb); 4932 trace_netif_rx_exit(ret); 4933 return ret; 4934 } 4935 EXPORT_SYMBOL(__netif_rx); 4936 4937 /** 4938 * netif_rx - post buffer to the network code 4939 * @skb: buffer to post 4940 * 4941 * This function receives a packet from a device driver and queues it for 4942 * the upper (protocol) levels to process via the backlog NAPI device. It 4943 * always succeeds. The buffer may be dropped during processing for 4944 * congestion control or by the protocol layers. 4945 * The network buffer is passed via the backlog NAPI device. Modern NIC 4946 * driver should use NAPI and GRO. 4947 * This function can used from interrupt and from process context. The 4948 * caller from process context must not disable interrupts before invoking 4949 * this function. 4950 * 4951 * return values: 4952 * NET_RX_SUCCESS (no congestion) 4953 * NET_RX_DROP (packet was dropped) 4954 * 4955 */ 4956 int netif_rx(struct sk_buff *skb) 4957 { 4958 bool need_bh_off = !(hardirq_count() | softirq_count()); 4959 int ret; 4960 4961 if (need_bh_off) 4962 local_bh_disable(); 4963 trace_netif_rx_entry(skb); 4964 ret = netif_rx_internal(skb); 4965 trace_netif_rx_exit(ret); 4966 if (need_bh_off) 4967 local_bh_enable(); 4968 return ret; 4969 } 4970 EXPORT_SYMBOL(netif_rx); 4971 4972 static __latent_entropy void net_tx_action(struct softirq_action *h) 4973 { 4974 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4975 4976 if (sd->completion_queue) { 4977 struct sk_buff *clist; 4978 4979 local_irq_disable(); 4980 clist = sd->completion_queue; 4981 sd->completion_queue = NULL; 4982 local_irq_enable(); 4983 4984 while (clist) { 4985 struct sk_buff *skb = clist; 4986 4987 clist = clist->next; 4988 4989 WARN_ON(refcount_read(&skb->users)); 4990 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4991 trace_consume_skb(skb); 4992 else 4993 trace_kfree_skb(skb, net_tx_action, 4994 SKB_DROP_REASON_NOT_SPECIFIED); 4995 4996 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4997 __kfree_skb(skb); 4998 else 4999 __kfree_skb_defer(skb); 5000 } 5001 } 5002 5003 if (sd->output_queue) { 5004 struct Qdisc *head; 5005 5006 local_irq_disable(); 5007 head = sd->output_queue; 5008 sd->output_queue = NULL; 5009 sd->output_queue_tailp = &sd->output_queue; 5010 local_irq_enable(); 5011 5012 rcu_read_lock(); 5013 5014 while (head) { 5015 struct Qdisc *q = head; 5016 spinlock_t *root_lock = NULL; 5017 5018 head = head->next_sched; 5019 5020 /* We need to make sure head->next_sched is read 5021 * before clearing __QDISC_STATE_SCHED 5022 */ 5023 smp_mb__before_atomic(); 5024 5025 if (!(q->flags & TCQ_F_NOLOCK)) { 5026 root_lock = qdisc_lock(q); 5027 spin_lock(root_lock); 5028 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5029 &q->state))) { 5030 /* There is a synchronize_net() between 5031 * STATE_DEACTIVATED flag being set and 5032 * qdisc_reset()/some_qdisc_is_busy() in 5033 * dev_deactivate(), so we can safely bail out 5034 * early here to avoid data race between 5035 * qdisc_deactivate() and some_qdisc_is_busy() 5036 * for lockless qdisc. 5037 */ 5038 clear_bit(__QDISC_STATE_SCHED, &q->state); 5039 continue; 5040 } 5041 5042 clear_bit(__QDISC_STATE_SCHED, &q->state); 5043 qdisc_run(q); 5044 if (root_lock) 5045 spin_unlock(root_lock); 5046 } 5047 5048 rcu_read_unlock(); 5049 } 5050 5051 xfrm_dev_backlog(sd); 5052 } 5053 5054 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5055 /* This hook is defined here for ATM LANE */ 5056 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5057 unsigned char *addr) __read_mostly; 5058 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5059 #endif 5060 5061 static inline struct sk_buff * 5062 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5063 struct net_device *orig_dev, bool *another) 5064 { 5065 #ifdef CONFIG_NET_CLS_ACT 5066 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5067 struct tcf_result cl_res; 5068 5069 /* If there's at least one ingress present somewhere (so 5070 * we get here via enabled static key), remaining devices 5071 * that are not configured with an ingress qdisc will bail 5072 * out here. 5073 */ 5074 if (!miniq) 5075 return skb; 5076 5077 if (*pt_prev) { 5078 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5079 *pt_prev = NULL; 5080 } 5081 5082 qdisc_skb_cb(skb)->pkt_len = skb->len; 5083 tc_skb_cb(skb)->mru = 0; 5084 tc_skb_cb(skb)->post_ct = false; 5085 skb->tc_at_ingress = 1; 5086 mini_qdisc_bstats_cpu_update(miniq, skb); 5087 5088 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5089 case TC_ACT_OK: 5090 case TC_ACT_RECLASSIFY: 5091 skb->tc_index = TC_H_MIN(cl_res.classid); 5092 break; 5093 case TC_ACT_SHOT: 5094 mini_qdisc_qstats_cpu_drop(miniq); 5095 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5096 return NULL; 5097 case TC_ACT_STOLEN: 5098 case TC_ACT_QUEUED: 5099 case TC_ACT_TRAP: 5100 consume_skb(skb); 5101 return NULL; 5102 case TC_ACT_REDIRECT: 5103 /* skb_mac_header check was done by cls/act_bpf, so 5104 * we can safely push the L2 header back before 5105 * redirecting to another netdev 5106 */ 5107 __skb_push(skb, skb->mac_len); 5108 if (skb_do_redirect(skb) == -EAGAIN) { 5109 __skb_pull(skb, skb->mac_len); 5110 *another = true; 5111 break; 5112 } 5113 return NULL; 5114 case TC_ACT_CONSUMED: 5115 return NULL; 5116 default: 5117 break; 5118 } 5119 #endif /* CONFIG_NET_CLS_ACT */ 5120 return skb; 5121 } 5122 5123 /** 5124 * netdev_is_rx_handler_busy - check if receive handler is registered 5125 * @dev: device to check 5126 * 5127 * Check if a receive handler is already registered for a given device. 5128 * Return true if there one. 5129 * 5130 * The caller must hold the rtnl_mutex. 5131 */ 5132 bool netdev_is_rx_handler_busy(struct net_device *dev) 5133 { 5134 ASSERT_RTNL(); 5135 return dev && rtnl_dereference(dev->rx_handler); 5136 } 5137 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5138 5139 /** 5140 * netdev_rx_handler_register - register receive handler 5141 * @dev: device to register a handler for 5142 * @rx_handler: receive handler to register 5143 * @rx_handler_data: data pointer that is used by rx handler 5144 * 5145 * Register a receive handler for a device. This handler will then be 5146 * called from __netif_receive_skb. A negative errno code is returned 5147 * on a failure. 5148 * 5149 * The caller must hold the rtnl_mutex. 5150 * 5151 * For a general description of rx_handler, see enum rx_handler_result. 5152 */ 5153 int netdev_rx_handler_register(struct net_device *dev, 5154 rx_handler_func_t *rx_handler, 5155 void *rx_handler_data) 5156 { 5157 if (netdev_is_rx_handler_busy(dev)) 5158 return -EBUSY; 5159 5160 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5161 return -EINVAL; 5162 5163 /* Note: rx_handler_data must be set before rx_handler */ 5164 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5165 rcu_assign_pointer(dev->rx_handler, rx_handler); 5166 5167 return 0; 5168 } 5169 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5170 5171 /** 5172 * netdev_rx_handler_unregister - unregister receive handler 5173 * @dev: device to unregister a handler from 5174 * 5175 * Unregister a receive handler from a device. 5176 * 5177 * The caller must hold the rtnl_mutex. 5178 */ 5179 void netdev_rx_handler_unregister(struct net_device *dev) 5180 { 5181 5182 ASSERT_RTNL(); 5183 RCU_INIT_POINTER(dev->rx_handler, NULL); 5184 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5185 * section has a guarantee to see a non NULL rx_handler_data 5186 * as well. 5187 */ 5188 synchronize_net(); 5189 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5190 } 5191 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5192 5193 /* 5194 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5195 * the special handling of PFMEMALLOC skbs. 5196 */ 5197 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5198 { 5199 switch (skb->protocol) { 5200 case htons(ETH_P_ARP): 5201 case htons(ETH_P_IP): 5202 case htons(ETH_P_IPV6): 5203 case htons(ETH_P_8021Q): 5204 case htons(ETH_P_8021AD): 5205 return true; 5206 default: 5207 return false; 5208 } 5209 } 5210 5211 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5212 int *ret, struct net_device *orig_dev) 5213 { 5214 if (nf_hook_ingress_active(skb)) { 5215 int ingress_retval; 5216 5217 if (*pt_prev) { 5218 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5219 *pt_prev = NULL; 5220 } 5221 5222 rcu_read_lock(); 5223 ingress_retval = nf_hook_ingress(skb); 5224 rcu_read_unlock(); 5225 return ingress_retval; 5226 } 5227 return 0; 5228 } 5229 5230 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5231 struct packet_type **ppt_prev) 5232 { 5233 struct packet_type *ptype, *pt_prev; 5234 rx_handler_func_t *rx_handler; 5235 struct sk_buff *skb = *pskb; 5236 struct net_device *orig_dev; 5237 bool deliver_exact = false; 5238 int ret = NET_RX_DROP; 5239 __be16 type; 5240 5241 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5242 5243 trace_netif_receive_skb(skb); 5244 5245 orig_dev = skb->dev; 5246 5247 skb_reset_network_header(skb); 5248 if (!skb_transport_header_was_set(skb)) 5249 skb_reset_transport_header(skb); 5250 skb_reset_mac_len(skb); 5251 5252 pt_prev = NULL; 5253 5254 another_round: 5255 skb->skb_iif = skb->dev->ifindex; 5256 5257 __this_cpu_inc(softnet_data.processed); 5258 5259 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5260 int ret2; 5261 5262 migrate_disable(); 5263 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5264 migrate_enable(); 5265 5266 if (ret2 != XDP_PASS) { 5267 ret = NET_RX_DROP; 5268 goto out; 5269 } 5270 } 5271 5272 if (eth_type_vlan(skb->protocol)) { 5273 skb = skb_vlan_untag(skb); 5274 if (unlikely(!skb)) 5275 goto out; 5276 } 5277 5278 if (skb_skip_tc_classify(skb)) 5279 goto skip_classify; 5280 5281 if (pfmemalloc) 5282 goto skip_taps; 5283 5284 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5285 if (pt_prev) 5286 ret = deliver_skb(skb, pt_prev, orig_dev); 5287 pt_prev = ptype; 5288 } 5289 5290 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5291 if (pt_prev) 5292 ret = deliver_skb(skb, pt_prev, orig_dev); 5293 pt_prev = ptype; 5294 } 5295 5296 skip_taps: 5297 #ifdef CONFIG_NET_INGRESS 5298 if (static_branch_unlikely(&ingress_needed_key)) { 5299 bool another = false; 5300 5301 nf_skip_egress(skb, true); 5302 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5303 &another); 5304 if (another) 5305 goto another_round; 5306 if (!skb) 5307 goto out; 5308 5309 nf_skip_egress(skb, false); 5310 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5311 goto out; 5312 } 5313 #endif 5314 skb_reset_redirect(skb); 5315 skip_classify: 5316 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5317 goto drop; 5318 5319 if (skb_vlan_tag_present(skb)) { 5320 if (pt_prev) { 5321 ret = deliver_skb(skb, pt_prev, orig_dev); 5322 pt_prev = NULL; 5323 } 5324 if (vlan_do_receive(&skb)) 5325 goto another_round; 5326 else if (unlikely(!skb)) 5327 goto out; 5328 } 5329 5330 rx_handler = rcu_dereference(skb->dev->rx_handler); 5331 if (rx_handler) { 5332 if (pt_prev) { 5333 ret = deliver_skb(skb, pt_prev, orig_dev); 5334 pt_prev = NULL; 5335 } 5336 switch (rx_handler(&skb)) { 5337 case RX_HANDLER_CONSUMED: 5338 ret = NET_RX_SUCCESS; 5339 goto out; 5340 case RX_HANDLER_ANOTHER: 5341 goto another_round; 5342 case RX_HANDLER_EXACT: 5343 deliver_exact = true; 5344 break; 5345 case RX_HANDLER_PASS: 5346 break; 5347 default: 5348 BUG(); 5349 } 5350 } 5351 5352 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5353 check_vlan_id: 5354 if (skb_vlan_tag_get_id(skb)) { 5355 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5356 * find vlan device. 5357 */ 5358 skb->pkt_type = PACKET_OTHERHOST; 5359 } else if (eth_type_vlan(skb->protocol)) { 5360 /* Outer header is 802.1P with vlan 0, inner header is 5361 * 802.1Q or 802.1AD and vlan_do_receive() above could 5362 * not find vlan dev for vlan id 0. 5363 */ 5364 __vlan_hwaccel_clear_tag(skb); 5365 skb = skb_vlan_untag(skb); 5366 if (unlikely(!skb)) 5367 goto out; 5368 if (vlan_do_receive(&skb)) 5369 /* After stripping off 802.1P header with vlan 0 5370 * vlan dev is found for inner header. 5371 */ 5372 goto another_round; 5373 else if (unlikely(!skb)) 5374 goto out; 5375 else 5376 /* We have stripped outer 802.1P vlan 0 header. 5377 * But could not find vlan dev. 5378 * check again for vlan id to set OTHERHOST. 5379 */ 5380 goto check_vlan_id; 5381 } 5382 /* Note: we might in the future use prio bits 5383 * and set skb->priority like in vlan_do_receive() 5384 * For the time being, just ignore Priority Code Point 5385 */ 5386 __vlan_hwaccel_clear_tag(skb); 5387 } 5388 5389 type = skb->protocol; 5390 5391 /* deliver only exact match when indicated */ 5392 if (likely(!deliver_exact)) { 5393 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5394 &ptype_base[ntohs(type) & 5395 PTYPE_HASH_MASK]); 5396 } 5397 5398 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5399 &orig_dev->ptype_specific); 5400 5401 if (unlikely(skb->dev != orig_dev)) { 5402 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5403 &skb->dev->ptype_specific); 5404 } 5405 5406 if (pt_prev) { 5407 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5408 goto drop; 5409 *ppt_prev = pt_prev; 5410 } else { 5411 drop: 5412 if (!deliver_exact) 5413 dev_core_stats_rx_dropped_inc(skb->dev); 5414 else 5415 dev_core_stats_rx_nohandler_inc(skb->dev); 5416 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5417 /* Jamal, now you will not able to escape explaining 5418 * me how you were going to use this. :-) 5419 */ 5420 ret = NET_RX_DROP; 5421 } 5422 5423 out: 5424 /* The invariant here is that if *ppt_prev is not NULL 5425 * then skb should also be non-NULL. 5426 * 5427 * Apparently *ppt_prev assignment above holds this invariant due to 5428 * skb dereferencing near it. 5429 */ 5430 *pskb = skb; 5431 return ret; 5432 } 5433 5434 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5435 { 5436 struct net_device *orig_dev = skb->dev; 5437 struct packet_type *pt_prev = NULL; 5438 int ret; 5439 5440 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5441 if (pt_prev) 5442 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5443 skb->dev, pt_prev, orig_dev); 5444 return ret; 5445 } 5446 5447 /** 5448 * netif_receive_skb_core - special purpose version of netif_receive_skb 5449 * @skb: buffer to process 5450 * 5451 * More direct receive version of netif_receive_skb(). It should 5452 * only be used by callers that have a need to skip RPS and Generic XDP. 5453 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5454 * 5455 * This function may only be called from softirq context and interrupts 5456 * should be enabled. 5457 * 5458 * Return values (usually ignored): 5459 * NET_RX_SUCCESS: no congestion 5460 * NET_RX_DROP: packet was dropped 5461 */ 5462 int netif_receive_skb_core(struct sk_buff *skb) 5463 { 5464 int ret; 5465 5466 rcu_read_lock(); 5467 ret = __netif_receive_skb_one_core(skb, false); 5468 rcu_read_unlock(); 5469 5470 return ret; 5471 } 5472 EXPORT_SYMBOL(netif_receive_skb_core); 5473 5474 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5475 struct packet_type *pt_prev, 5476 struct net_device *orig_dev) 5477 { 5478 struct sk_buff *skb, *next; 5479 5480 if (!pt_prev) 5481 return; 5482 if (list_empty(head)) 5483 return; 5484 if (pt_prev->list_func != NULL) 5485 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5486 ip_list_rcv, head, pt_prev, orig_dev); 5487 else 5488 list_for_each_entry_safe(skb, next, head, list) { 5489 skb_list_del_init(skb); 5490 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5491 } 5492 } 5493 5494 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5495 { 5496 /* Fast-path assumptions: 5497 * - There is no RX handler. 5498 * - Only one packet_type matches. 5499 * If either of these fails, we will end up doing some per-packet 5500 * processing in-line, then handling the 'last ptype' for the whole 5501 * sublist. This can't cause out-of-order delivery to any single ptype, 5502 * because the 'last ptype' must be constant across the sublist, and all 5503 * other ptypes are handled per-packet. 5504 */ 5505 /* Current (common) ptype of sublist */ 5506 struct packet_type *pt_curr = NULL; 5507 /* Current (common) orig_dev of sublist */ 5508 struct net_device *od_curr = NULL; 5509 struct list_head sublist; 5510 struct sk_buff *skb, *next; 5511 5512 INIT_LIST_HEAD(&sublist); 5513 list_for_each_entry_safe(skb, next, head, list) { 5514 struct net_device *orig_dev = skb->dev; 5515 struct packet_type *pt_prev = NULL; 5516 5517 skb_list_del_init(skb); 5518 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5519 if (!pt_prev) 5520 continue; 5521 if (pt_curr != pt_prev || od_curr != orig_dev) { 5522 /* dispatch old sublist */ 5523 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5524 /* start new sublist */ 5525 INIT_LIST_HEAD(&sublist); 5526 pt_curr = pt_prev; 5527 od_curr = orig_dev; 5528 } 5529 list_add_tail(&skb->list, &sublist); 5530 } 5531 5532 /* dispatch final sublist */ 5533 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5534 } 5535 5536 static int __netif_receive_skb(struct sk_buff *skb) 5537 { 5538 int ret; 5539 5540 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5541 unsigned int noreclaim_flag; 5542 5543 /* 5544 * PFMEMALLOC skbs are special, they should 5545 * - be delivered to SOCK_MEMALLOC sockets only 5546 * - stay away from userspace 5547 * - have bounded memory usage 5548 * 5549 * Use PF_MEMALLOC as this saves us from propagating the allocation 5550 * context down to all allocation sites. 5551 */ 5552 noreclaim_flag = memalloc_noreclaim_save(); 5553 ret = __netif_receive_skb_one_core(skb, true); 5554 memalloc_noreclaim_restore(noreclaim_flag); 5555 } else 5556 ret = __netif_receive_skb_one_core(skb, false); 5557 5558 return ret; 5559 } 5560 5561 static void __netif_receive_skb_list(struct list_head *head) 5562 { 5563 unsigned long noreclaim_flag = 0; 5564 struct sk_buff *skb, *next; 5565 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5566 5567 list_for_each_entry_safe(skb, next, head, list) { 5568 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5569 struct list_head sublist; 5570 5571 /* Handle the previous sublist */ 5572 list_cut_before(&sublist, head, &skb->list); 5573 if (!list_empty(&sublist)) 5574 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5575 pfmemalloc = !pfmemalloc; 5576 /* See comments in __netif_receive_skb */ 5577 if (pfmemalloc) 5578 noreclaim_flag = memalloc_noreclaim_save(); 5579 else 5580 memalloc_noreclaim_restore(noreclaim_flag); 5581 } 5582 } 5583 /* Handle the remaining sublist */ 5584 if (!list_empty(head)) 5585 __netif_receive_skb_list_core(head, pfmemalloc); 5586 /* Restore pflags */ 5587 if (pfmemalloc) 5588 memalloc_noreclaim_restore(noreclaim_flag); 5589 } 5590 5591 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5592 { 5593 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5594 struct bpf_prog *new = xdp->prog; 5595 int ret = 0; 5596 5597 switch (xdp->command) { 5598 case XDP_SETUP_PROG: 5599 rcu_assign_pointer(dev->xdp_prog, new); 5600 if (old) 5601 bpf_prog_put(old); 5602 5603 if (old && !new) { 5604 static_branch_dec(&generic_xdp_needed_key); 5605 } else if (new && !old) { 5606 static_branch_inc(&generic_xdp_needed_key); 5607 dev_disable_lro(dev); 5608 dev_disable_gro_hw(dev); 5609 } 5610 break; 5611 5612 default: 5613 ret = -EINVAL; 5614 break; 5615 } 5616 5617 return ret; 5618 } 5619 5620 static int netif_receive_skb_internal(struct sk_buff *skb) 5621 { 5622 int ret; 5623 5624 net_timestamp_check(netdev_tstamp_prequeue, skb); 5625 5626 if (skb_defer_rx_timestamp(skb)) 5627 return NET_RX_SUCCESS; 5628 5629 rcu_read_lock(); 5630 #ifdef CONFIG_RPS 5631 if (static_branch_unlikely(&rps_needed)) { 5632 struct rps_dev_flow voidflow, *rflow = &voidflow; 5633 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5634 5635 if (cpu >= 0) { 5636 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5637 rcu_read_unlock(); 5638 return ret; 5639 } 5640 } 5641 #endif 5642 ret = __netif_receive_skb(skb); 5643 rcu_read_unlock(); 5644 return ret; 5645 } 5646 5647 void netif_receive_skb_list_internal(struct list_head *head) 5648 { 5649 struct sk_buff *skb, *next; 5650 struct list_head sublist; 5651 5652 INIT_LIST_HEAD(&sublist); 5653 list_for_each_entry_safe(skb, next, head, list) { 5654 net_timestamp_check(netdev_tstamp_prequeue, skb); 5655 skb_list_del_init(skb); 5656 if (!skb_defer_rx_timestamp(skb)) 5657 list_add_tail(&skb->list, &sublist); 5658 } 5659 list_splice_init(&sublist, head); 5660 5661 rcu_read_lock(); 5662 #ifdef CONFIG_RPS 5663 if (static_branch_unlikely(&rps_needed)) { 5664 list_for_each_entry_safe(skb, next, head, list) { 5665 struct rps_dev_flow voidflow, *rflow = &voidflow; 5666 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5667 5668 if (cpu >= 0) { 5669 /* Will be handled, remove from list */ 5670 skb_list_del_init(skb); 5671 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5672 } 5673 } 5674 } 5675 #endif 5676 __netif_receive_skb_list(head); 5677 rcu_read_unlock(); 5678 } 5679 5680 /** 5681 * netif_receive_skb - process receive buffer from network 5682 * @skb: buffer to process 5683 * 5684 * netif_receive_skb() is the main receive data processing function. 5685 * It always succeeds. The buffer may be dropped during processing 5686 * for congestion control or by the protocol layers. 5687 * 5688 * This function may only be called from softirq context and interrupts 5689 * should be enabled. 5690 * 5691 * Return values (usually ignored): 5692 * NET_RX_SUCCESS: no congestion 5693 * NET_RX_DROP: packet was dropped 5694 */ 5695 int netif_receive_skb(struct sk_buff *skb) 5696 { 5697 int ret; 5698 5699 trace_netif_receive_skb_entry(skb); 5700 5701 ret = netif_receive_skb_internal(skb); 5702 trace_netif_receive_skb_exit(ret); 5703 5704 return ret; 5705 } 5706 EXPORT_SYMBOL(netif_receive_skb); 5707 5708 /** 5709 * netif_receive_skb_list - process many receive buffers from network 5710 * @head: list of skbs to process. 5711 * 5712 * Since return value of netif_receive_skb() is normally ignored, and 5713 * wouldn't be meaningful for a list, this function returns void. 5714 * 5715 * This function may only be called from softirq context and interrupts 5716 * should be enabled. 5717 */ 5718 void netif_receive_skb_list(struct list_head *head) 5719 { 5720 struct sk_buff *skb; 5721 5722 if (list_empty(head)) 5723 return; 5724 if (trace_netif_receive_skb_list_entry_enabled()) { 5725 list_for_each_entry(skb, head, list) 5726 trace_netif_receive_skb_list_entry(skb); 5727 } 5728 netif_receive_skb_list_internal(head); 5729 trace_netif_receive_skb_list_exit(0); 5730 } 5731 EXPORT_SYMBOL(netif_receive_skb_list); 5732 5733 static DEFINE_PER_CPU(struct work_struct, flush_works); 5734 5735 /* Network device is going away, flush any packets still pending */ 5736 static void flush_backlog(struct work_struct *work) 5737 { 5738 struct sk_buff *skb, *tmp; 5739 struct softnet_data *sd; 5740 5741 local_bh_disable(); 5742 sd = this_cpu_ptr(&softnet_data); 5743 5744 rps_lock_irq_disable(sd); 5745 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5746 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5747 __skb_unlink(skb, &sd->input_pkt_queue); 5748 dev_kfree_skb_irq(skb); 5749 input_queue_head_incr(sd); 5750 } 5751 } 5752 rps_unlock_irq_enable(sd); 5753 5754 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5755 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5756 __skb_unlink(skb, &sd->process_queue); 5757 kfree_skb(skb); 5758 input_queue_head_incr(sd); 5759 } 5760 } 5761 local_bh_enable(); 5762 } 5763 5764 static bool flush_required(int cpu) 5765 { 5766 #if IS_ENABLED(CONFIG_RPS) 5767 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5768 bool do_flush; 5769 5770 rps_lock_irq_disable(sd); 5771 5772 /* as insertion into process_queue happens with the rps lock held, 5773 * process_queue access may race only with dequeue 5774 */ 5775 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5776 !skb_queue_empty_lockless(&sd->process_queue); 5777 rps_unlock_irq_enable(sd); 5778 5779 return do_flush; 5780 #endif 5781 /* without RPS we can't safely check input_pkt_queue: during a 5782 * concurrent remote skb_queue_splice() we can detect as empty both 5783 * input_pkt_queue and process_queue even if the latter could end-up 5784 * containing a lot of packets. 5785 */ 5786 return true; 5787 } 5788 5789 static void flush_all_backlogs(void) 5790 { 5791 static cpumask_t flush_cpus; 5792 unsigned int cpu; 5793 5794 /* since we are under rtnl lock protection we can use static data 5795 * for the cpumask and avoid allocating on stack the possibly 5796 * large mask 5797 */ 5798 ASSERT_RTNL(); 5799 5800 cpus_read_lock(); 5801 5802 cpumask_clear(&flush_cpus); 5803 for_each_online_cpu(cpu) { 5804 if (flush_required(cpu)) { 5805 queue_work_on(cpu, system_highpri_wq, 5806 per_cpu_ptr(&flush_works, cpu)); 5807 cpumask_set_cpu(cpu, &flush_cpus); 5808 } 5809 } 5810 5811 /* we can have in flight packet[s] on the cpus we are not flushing, 5812 * synchronize_net() in unregister_netdevice_many() will take care of 5813 * them 5814 */ 5815 for_each_cpu(cpu, &flush_cpus) 5816 flush_work(per_cpu_ptr(&flush_works, cpu)); 5817 5818 cpus_read_unlock(); 5819 } 5820 5821 static void net_rps_send_ipi(struct softnet_data *remsd) 5822 { 5823 #ifdef CONFIG_RPS 5824 while (remsd) { 5825 struct softnet_data *next = remsd->rps_ipi_next; 5826 5827 if (cpu_online(remsd->cpu)) 5828 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5829 remsd = next; 5830 } 5831 #endif 5832 } 5833 5834 /* 5835 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5836 * Note: called with local irq disabled, but exits with local irq enabled. 5837 */ 5838 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5839 { 5840 #ifdef CONFIG_RPS 5841 struct softnet_data *remsd = sd->rps_ipi_list; 5842 5843 if (remsd) { 5844 sd->rps_ipi_list = NULL; 5845 5846 local_irq_enable(); 5847 5848 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5849 net_rps_send_ipi(remsd); 5850 } else 5851 #endif 5852 local_irq_enable(); 5853 } 5854 5855 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5856 { 5857 #ifdef CONFIG_RPS 5858 return sd->rps_ipi_list != NULL; 5859 #else 5860 return false; 5861 #endif 5862 } 5863 5864 static int process_backlog(struct napi_struct *napi, int quota) 5865 { 5866 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5867 bool again = true; 5868 int work = 0; 5869 5870 /* Check if we have pending ipi, its better to send them now, 5871 * not waiting net_rx_action() end. 5872 */ 5873 if (sd_has_rps_ipi_waiting(sd)) { 5874 local_irq_disable(); 5875 net_rps_action_and_irq_enable(sd); 5876 } 5877 5878 napi->weight = dev_rx_weight; 5879 while (again) { 5880 struct sk_buff *skb; 5881 5882 while ((skb = __skb_dequeue(&sd->process_queue))) { 5883 rcu_read_lock(); 5884 __netif_receive_skb(skb); 5885 rcu_read_unlock(); 5886 input_queue_head_incr(sd); 5887 if (++work >= quota) 5888 return work; 5889 5890 } 5891 5892 rps_lock_irq_disable(sd); 5893 if (skb_queue_empty(&sd->input_pkt_queue)) { 5894 /* 5895 * Inline a custom version of __napi_complete(). 5896 * only current cpu owns and manipulates this napi, 5897 * and NAPI_STATE_SCHED is the only possible flag set 5898 * on backlog. 5899 * We can use a plain write instead of clear_bit(), 5900 * and we dont need an smp_mb() memory barrier. 5901 */ 5902 napi->state = 0; 5903 again = false; 5904 } else { 5905 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5906 &sd->process_queue); 5907 } 5908 rps_unlock_irq_enable(sd); 5909 } 5910 5911 return work; 5912 } 5913 5914 /** 5915 * __napi_schedule - schedule for receive 5916 * @n: entry to schedule 5917 * 5918 * The entry's receive function will be scheduled to run. 5919 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5920 */ 5921 void __napi_schedule(struct napi_struct *n) 5922 { 5923 unsigned long flags; 5924 5925 local_irq_save(flags); 5926 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5927 local_irq_restore(flags); 5928 } 5929 EXPORT_SYMBOL(__napi_schedule); 5930 5931 /** 5932 * napi_schedule_prep - check if napi can be scheduled 5933 * @n: napi context 5934 * 5935 * Test if NAPI routine is already running, and if not mark 5936 * it as running. This is used as a condition variable to 5937 * insure only one NAPI poll instance runs. We also make 5938 * sure there is no pending NAPI disable. 5939 */ 5940 bool napi_schedule_prep(struct napi_struct *n) 5941 { 5942 unsigned long val, new; 5943 5944 do { 5945 val = READ_ONCE(n->state); 5946 if (unlikely(val & NAPIF_STATE_DISABLE)) 5947 return false; 5948 new = val | NAPIF_STATE_SCHED; 5949 5950 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5951 * This was suggested by Alexander Duyck, as compiler 5952 * emits better code than : 5953 * if (val & NAPIF_STATE_SCHED) 5954 * new |= NAPIF_STATE_MISSED; 5955 */ 5956 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5957 NAPIF_STATE_MISSED; 5958 } while (cmpxchg(&n->state, val, new) != val); 5959 5960 return !(val & NAPIF_STATE_SCHED); 5961 } 5962 EXPORT_SYMBOL(napi_schedule_prep); 5963 5964 /** 5965 * __napi_schedule_irqoff - schedule for receive 5966 * @n: entry to schedule 5967 * 5968 * Variant of __napi_schedule() assuming hard irqs are masked. 5969 * 5970 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5971 * because the interrupt disabled assumption might not be true 5972 * due to force-threaded interrupts and spinlock substitution. 5973 */ 5974 void __napi_schedule_irqoff(struct napi_struct *n) 5975 { 5976 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5977 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5978 else 5979 __napi_schedule(n); 5980 } 5981 EXPORT_SYMBOL(__napi_schedule_irqoff); 5982 5983 bool napi_complete_done(struct napi_struct *n, int work_done) 5984 { 5985 unsigned long flags, val, new, timeout = 0; 5986 bool ret = true; 5987 5988 /* 5989 * 1) Don't let napi dequeue from the cpu poll list 5990 * just in case its running on a different cpu. 5991 * 2) If we are busy polling, do nothing here, we have 5992 * the guarantee we will be called later. 5993 */ 5994 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5995 NAPIF_STATE_IN_BUSY_POLL))) 5996 return false; 5997 5998 if (work_done) { 5999 if (n->gro_bitmask) 6000 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6001 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6002 } 6003 if (n->defer_hard_irqs_count > 0) { 6004 n->defer_hard_irqs_count--; 6005 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6006 if (timeout) 6007 ret = false; 6008 } 6009 if (n->gro_bitmask) { 6010 /* When the NAPI instance uses a timeout and keeps postponing 6011 * it, we need to bound somehow the time packets are kept in 6012 * the GRO layer 6013 */ 6014 napi_gro_flush(n, !!timeout); 6015 } 6016 6017 gro_normal_list(n); 6018 6019 if (unlikely(!list_empty(&n->poll_list))) { 6020 /* If n->poll_list is not empty, we need to mask irqs */ 6021 local_irq_save(flags); 6022 list_del_init(&n->poll_list); 6023 local_irq_restore(flags); 6024 } 6025 6026 do { 6027 val = READ_ONCE(n->state); 6028 6029 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6030 6031 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6032 NAPIF_STATE_SCHED_THREADED | 6033 NAPIF_STATE_PREFER_BUSY_POLL); 6034 6035 /* If STATE_MISSED was set, leave STATE_SCHED set, 6036 * because we will call napi->poll() one more time. 6037 * This C code was suggested by Alexander Duyck to help gcc. 6038 */ 6039 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6040 NAPIF_STATE_SCHED; 6041 } while (cmpxchg(&n->state, val, new) != val); 6042 6043 if (unlikely(val & NAPIF_STATE_MISSED)) { 6044 __napi_schedule(n); 6045 return false; 6046 } 6047 6048 if (timeout) 6049 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6050 HRTIMER_MODE_REL_PINNED); 6051 return ret; 6052 } 6053 EXPORT_SYMBOL(napi_complete_done); 6054 6055 /* must be called under rcu_read_lock(), as we dont take a reference */ 6056 static struct napi_struct *napi_by_id(unsigned int napi_id) 6057 { 6058 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6059 struct napi_struct *napi; 6060 6061 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6062 if (napi->napi_id == napi_id) 6063 return napi; 6064 6065 return NULL; 6066 } 6067 6068 #if defined(CONFIG_NET_RX_BUSY_POLL) 6069 6070 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6071 { 6072 if (!skip_schedule) { 6073 gro_normal_list(napi); 6074 __napi_schedule(napi); 6075 return; 6076 } 6077 6078 if (napi->gro_bitmask) { 6079 /* flush too old packets 6080 * If HZ < 1000, flush all packets. 6081 */ 6082 napi_gro_flush(napi, HZ >= 1000); 6083 } 6084 6085 gro_normal_list(napi); 6086 clear_bit(NAPI_STATE_SCHED, &napi->state); 6087 } 6088 6089 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6090 u16 budget) 6091 { 6092 bool skip_schedule = false; 6093 unsigned long timeout; 6094 int rc; 6095 6096 /* Busy polling means there is a high chance device driver hard irq 6097 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6098 * set in napi_schedule_prep(). 6099 * Since we are about to call napi->poll() once more, we can safely 6100 * clear NAPI_STATE_MISSED. 6101 * 6102 * Note: x86 could use a single "lock and ..." instruction 6103 * to perform these two clear_bit() 6104 */ 6105 clear_bit(NAPI_STATE_MISSED, &napi->state); 6106 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6107 6108 local_bh_disable(); 6109 6110 if (prefer_busy_poll) { 6111 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6112 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6113 if (napi->defer_hard_irqs_count && timeout) { 6114 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6115 skip_schedule = true; 6116 } 6117 } 6118 6119 /* All we really want here is to re-enable device interrupts. 6120 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6121 */ 6122 rc = napi->poll(napi, budget); 6123 /* We can't gro_normal_list() here, because napi->poll() might have 6124 * rearmed the napi (napi_complete_done()) in which case it could 6125 * already be running on another CPU. 6126 */ 6127 trace_napi_poll(napi, rc, budget); 6128 netpoll_poll_unlock(have_poll_lock); 6129 if (rc == budget) 6130 __busy_poll_stop(napi, skip_schedule); 6131 local_bh_enable(); 6132 } 6133 6134 void napi_busy_loop(unsigned int napi_id, 6135 bool (*loop_end)(void *, unsigned long), 6136 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6137 { 6138 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6139 int (*napi_poll)(struct napi_struct *napi, int budget); 6140 void *have_poll_lock = NULL; 6141 struct napi_struct *napi; 6142 6143 restart: 6144 napi_poll = NULL; 6145 6146 rcu_read_lock(); 6147 6148 napi = napi_by_id(napi_id); 6149 if (!napi) 6150 goto out; 6151 6152 preempt_disable(); 6153 for (;;) { 6154 int work = 0; 6155 6156 local_bh_disable(); 6157 if (!napi_poll) { 6158 unsigned long val = READ_ONCE(napi->state); 6159 6160 /* If multiple threads are competing for this napi, 6161 * we avoid dirtying napi->state as much as we can. 6162 */ 6163 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6164 NAPIF_STATE_IN_BUSY_POLL)) { 6165 if (prefer_busy_poll) 6166 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6167 goto count; 6168 } 6169 if (cmpxchg(&napi->state, val, 6170 val | NAPIF_STATE_IN_BUSY_POLL | 6171 NAPIF_STATE_SCHED) != val) { 6172 if (prefer_busy_poll) 6173 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6174 goto count; 6175 } 6176 have_poll_lock = netpoll_poll_lock(napi); 6177 napi_poll = napi->poll; 6178 } 6179 work = napi_poll(napi, budget); 6180 trace_napi_poll(napi, work, budget); 6181 gro_normal_list(napi); 6182 count: 6183 if (work > 0) 6184 __NET_ADD_STATS(dev_net(napi->dev), 6185 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6186 local_bh_enable(); 6187 6188 if (!loop_end || loop_end(loop_end_arg, start_time)) 6189 break; 6190 6191 if (unlikely(need_resched())) { 6192 if (napi_poll) 6193 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6194 preempt_enable(); 6195 rcu_read_unlock(); 6196 cond_resched(); 6197 if (loop_end(loop_end_arg, start_time)) 6198 return; 6199 goto restart; 6200 } 6201 cpu_relax(); 6202 } 6203 if (napi_poll) 6204 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6205 preempt_enable(); 6206 out: 6207 rcu_read_unlock(); 6208 } 6209 EXPORT_SYMBOL(napi_busy_loop); 6210 6211 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6212 6213 static void napi_hash_add(struct napi_struct *napi) 6214 { 6215 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6216 return; 6217 6218 spin_lock(&napi_hash_lock); 6219 6220 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6221 do { 6222 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6223 napi_gen_id = MIN_NAPI_ID; 6224 } while (napi_by_id(napi_gen_id)); 6225 napi->napi_id = napi_gen_id; 6226 6227 hlist_add_head_rcu(&napi->napi_hash_node, 6228 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6229 6230 spin_unlock(&napi_hash_lock); 6231 } 6232 6233 /* Warning : caller is responsible to make sure rcu grace period 6234 * is respected before freeing memory containing @napi 6235 */ 6236 static void napi_hash_del(struct napi_struct *napi) 6237 { 6238 spin_lock(&napi_hash_lock); 6239 6240 hlist_del_init_rcu(&napi->napi_hash_node); 6241 6242 spin_unlock(&napi_hash_lock); 6243 } 6244 6245 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6246 { 6247 struct napi_struct *napi; 6248 6249 napi = container_of(timer, struct napi_struct, timer); 6250 6251 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6252 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6253 */ 6254 if (!napi_disable_pending(napi) && 6255 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6256 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6257 __napi_schedule_irqoff(napi); 6258 } 6259 6260 return HRTIMER_NORESTART; 6261 } 6262 6263 static void init_gro_hash(struct napi_struct *napi) 6264 { 6265 int i; 6266 6267 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6268 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6269 napi->gro_hash[i].count = 0; 6270 } 6271 napi->gro_bitmask = 0; 6272 } 6273 6274 int dev_set_threaded(struct net_device *dev, bool threaded) 6275 { 6276 struct napi_struct *napi; 6277 int err = 0; 6278 6279 if (dev->threaded == threaded) 6280 return 0; 6281 6282 if (threaded) { 6283 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6284 if (!napi->thread) { 6285 err = napi_kthread_create(napi); 6286 if (err) { 6287 threaded = false; 6288 break; 6289 } 6290 } 6291 } 6292 } 6293 6294 dev->threaded = threaded; 6295 6296 /* Make sure kthread is created before THREADED bit 6297 * is set. 6298 */ 6299 smp_mb__before_atomic(); 6300 6301 /* Setting/unsetting threaded mode on a napi might not immediately 6302 * take effect, if the current napi instance is actively being 6303 * polled. In this case, the switch between threaded mode and 6304 * softirq mode will happen in the next round of napi_schedule(). 6305 * This should not cause hiccups/stalls to the live traffic. 6306 */ 6307 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6308 if (threaded) 6309 set_bit(NAPI_STATE_THREADED, &napi->state); 6310 else 6311 clear_bit(NAPI_STATE_THREADED, &napi->state); 6312 } 6313 6314 return err; 6315 } 6316 EXPORT_SYMBOL(dev_set_threaded); 6317 6318 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6319 int (*poll)(struct napi_struct *, int), int weight) 6320 { 6321 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6322 return; 6323 6324 INIT_LIST_HEAD(&napi->poll_list); 6325 INIT_HLIST_NODE(&napi->napi_hash_node); 6326 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6327 napi->timer.function = napi_watchdog; 6328 init_gro_hash(napi); 6329 napi->skb = NULL; 6330 INIT_LIST_HEAD(&napi->rx_list); 6331 napi->rx_count = 0; 6332 napi->poll = poll; 6333 if (weight > NAPI_POLL_WEIGHT) 6334 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6335 weight); 6336 napi->weight = weight; 6337 napi->dev = dev; 6338 #ifdef CONFIG_NETPOLL 6339 napi->poll_owner = -1; 6340 #endif 6341 set_bit(NAPI_STATE_SCHED, &napi->state); 6342 set_bit(NAPI_STATE_NPSVC, &napi->state); 6343 list_add_rcu(&napi->dev_list, &dev->napi_list); 6344 napi_hash_add(napi); 6345 /* Create kthread for this napi if dev->threaded is set. 6346 * Clear dev->threaded if kthread creation failed so that 6347 * threaded mode will not be enabled in napi_enable(). 6348 */ 6349 if (dev->threaded && napi_kthread_create(napi)) 6350 dev->threaded = 0; 6351 } 6352 EXPORT_SYMBOL(netif_napi_add_weight); 6353 6354 void napi_disable(struct napi_struct *n) 6355 { 6356 unsigned long val, new; 6357 6358 might_sleep(); 6359 set_bit(NAPI_STATE_DISABLE, &n->state); 6360 6361 for ( ; ; ) { 6362 val = READ_ONCE(n->state); 6363 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6364 usleep_range(20, 200); 6365 continue; 6366 } 6367 6368 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6369 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6370 6371 if (cmpxchg(&n->state, val, new) == val) 6372 break; 6373 } 6374 6375 hrtimer_cancel(&n->timer); 6376 6377 clear_bit(NAPI_STATE_DISABLE, &n->state); 6378 } 6379 EXPORT_SYMBOL(napi_disable); 6380 6381 /** 6382 * napi_enable - enable NAPI scheduling 6383 * @n: NAPI context 6384 * 6385 * Resume NAPI from being scheduled on this context. 6386 * Must be paired with napi_disable. 6387 */ 6388 void napi_enable(struct napi_struct *n) 6389 { 6390 unsigned long val, new; 6391 6392 do { 6393 val = READ_ONCE(n->state); 6394 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6395 6396 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6397 if (n->dev->threaded && n->thread) 6398 new |= NAPIF_STATE_THREADED; 6399 } while (cmpxchg(&n->state, val, new) != val); 6400 } 6401 EXPORT_SYMBOL(napi_enable); 6402 6403 static void flush_gro_hash(struct napi_struct *napi) 6404 { 6405 int i; 6406 6407 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6408 struct sk_buff *skb, *n; 6409 6410 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6411 kfree_skb(skb); 6412 napi->gro_hash[i].count = 0; 6413 } 6414 } 6415 6416 /* Must be called in process context */ 6417 void __netif_napi_del(struct napi_struct *napi) 6418 { 6419 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6420 return; 6421 6422 napi_hash_del(napi); 6423 list_del_rcu(&napi->dev_list); 6424 napi_free_frags(napi); 6425 6426 flush_gro_hash(napi); 6427 napi->gro_bitmask = 0; 6428 6429 if (napi->thread) { 6430 kthread_stop(napi->thread); 6431 napi->thread = NULL; 6432 } 6433 } 6434 EXPORT_SYMBOL(__netif_napi_del); 6435 6436 static int __napi_poll(struct napi_struct *n, bool *repoll) 6437 { 6438 int work, weight; 6439 6440 weight = n->weight; 6441 6442 /* This NAPI_STATE_SCHED test is for avoiding a race 6443 * with netpoll's poll_napi(). Only the entity which 6444 * obtains the lock and sees NAPI_STATE_SCHED set will 6445 * actually make the ->poll() call. Therefore we avoid 6446 * accidentally calling ->poll() when NAPI is not scheduled. 6447 */ 6448 work = 0; 6449 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6450 work = n->poll(n, weight); 6451 trace_napi_poll(n, work, weight); 6452 } 6453 6454 if (unlikely(work > weight)) 6455 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6456 n->poll, work, weight); 6457 6458 if (likely(work < weight)) 6459 return work; 6460 6461 /* Drivers must not modify the NAPI state if they 6462 * consume the entire weight. In such cases this code 6463 * still "owns" the NAPI instance and therefore can 6464 * move the instance around on the list at-will. 6465 */ 6466 if (unlikely(napi_disable_pending(n))) { 6467 napi_complete(n); 6468 return work; 6469 } 6470 6471 /* The NAPI context has more processing work, but busy-polling 6472 * is preferred. Exit early. 6473 */ 6474 if (napi_prefer_busy_poll(n)) { 6475 if (napi_complete_done(n, work)) { 6476 /* If timeout is not set, we need to make sure 6477 * that the NAPI is re-scheduled. 6478 */ 6479 napi_schedule(n); 6480 } 6481 return work; 6482 } 6483 6484 if (n->gro_bitmask) { 6485 /* flush too old packets 6486 * If HZ < 1000, flush all packets. 6487 */ 6488 napi_gro_flush(n, HZ >= 1000); 6489 } 6490 6491 gro_normal_list(n); 6492 6493 /* Some drivers may have called napi_schedule 6494 * prior to exhausting their budget. 6495 */ 6496 if (unlikely(!list_empty(&n->poll_list))) { 6497 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6498 n->dev ? n->dev->name : "backlog"); 6499 return work; 6500 } 6501 6502 *repoll = true; 6503 6504 return work; 6505 } 6506 6507 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6508 { 6509 bool do_repoll = false; 6510 void *have; 6511 int work; 6512 6513 list_del_init(&n->poll_list); 6514 6515 have = netpoll_poll_lock(n); 6516 6517 work = __napi_poll(n, &do_repoll); 6518 6519 if (do_repoll) 6520 list_add_tail(&n->poll_list, repoll); 6521 6522 netpoll_poll_unlock(have); 6523 6524 return work; 6525 } 6526 6527 static int napi_thread_wait(struct napi_struct *napi) 6528 { 6529 bool woken = false; 6530 6531 set_current_state(TASK_INTERRUPTIBLE); 6532 6533 while (!kthread_should_stop()) { 6534 /* Testing SCHED_THREADED bit here to make sure the current 6535 * kthread owns this napi and could poll on this napi. 6536 * Testing SCHED bit is not enough because SCHED bit might be 6537 * set by some other busy poll thread or by napi_disable(). 6538 */ 6539 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6540 WARN_ON(!list_empty(&napi->poll_list)); 6541 __set_current_state(TASK_RUNNING); 6542 return 0; 6543 } 6544 6545 schedule(); 6546 /* woken being true indicates this thread owns this napi. */ 6547 woken = true; 6548 set_current_state(TASK_INTERRUPTIBLE); 6549 } 6550 __set_current_state(TASK_RUNNING); 6551 6552 return -1; 6553 } 6554 6555 static int napi_threaded_poll(void *data) 6556 { 6557 struct napi_struct *napi = data; 6558 void *have; 6559 6560 while (!napi_thread_wait(napi)) { 6561 for (;;) { 6562 bool repoll = false; 6563 6564 local_bh_disable(); 6565 6566 have = netpoll_poll_lock(napi); 6567 __napi_poll(napi, &repoll); 6568 netpoll_poll_unlock(have); 6569 6570 local_bh_enable(); 6571 6572 if (!repoll) 6573 break; 6574 6575 cond_resched(); 6576 } 6577 } 6578 return 0; 6579 } 6580 6581 static void skb_defer_free_flush(struct softnet_data *sd) 6582 { 6583 struct sk_buff *skb, *next; 6584 unsigned long flags; 6585 6586 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6587 if (!READ_ONCE(sd->defer_list)) 6588 return; 6589 6590 spin_lock_irqsave(&sd->defer_lock, flags); 6591 skb = sd->defer_list; 6592 sd->defer_list = NULL; 6593 sd->defer_count = 0; 6594 spin_unlock_irqrestore(&sd->defer_lock, flags); 6595 6596 while (skb != NULL) { 6597 next = skb->next; 6598 __kfree_skb(skb); 6599 skb = next; 6600 } 6601 } 6602 6603 static __latent_entropy void net_rx_action(struct softirq_action *h) 6604 { 6605 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6606 unsigned long time_limit = jiffies + 6607 usecs_to_jiffies(netdev_budget_usecs); 6608 int budget = netdev_budget; 6609 LIST_HEAD(list); 6610 LIST_HEAD(repoll); 6611 6612 local_irq_disable(); 6613 list_splice_init(&sd->poll_list, &list); 6614 local_irq_enable(); 6615 6616 for (;;) { 6617 struct napi_struct *n; 6618 6619 if (list_empty(&list)) { 6620 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6621 goto end; 6622 break; 6623 } 6624 6625 n = list_first_entry(&list, struct napi_struct, poll_list); 6626 budget -= napi_poll(n, &repoll); 6627 6628 /* If softirq window is exhausted then punt. 6629 * Allow this to run for 2 jiffies since which will allow 6630 * an average latency of 1.5/HZ. 6631 */ 6632 if (unlikely(budget <= 0 || 6633 time_after_eq(jiffies, time_limit))) { 6634 sd->time_squeeze++; 6635 break; 6636 } 6637 } 6638 6639 local_irq_disable(); 6640 6641 list_splice_tail_init(&sd->poll_list, &list); 6642 list_splice_tail(&repoll, &list); 6643 list_splice(&list, &sd->poll_list); 6644 if (!list_empty(&sd->poll_list)) 6645 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6646 6647 net_rps_action_and_irq_enable(sd); 6648 end: 6649 skb_defer_free_flush(sd); 6650 } 6651 6652 struct netdev_adjacent { 6653 struct net_device *dev; 6654 netdevice_tracker dev_tracker; 6655 6656 /* upper master flag, there can only be one master device per list */ 6657 bool master; 6658 6659 /* lookup ignore flag */ 6660 bool ignore; 6661 6662 /* counter for the number of times this device was added to us */ 6663 u16 ref_nr; 6664 6665 /* private field for the users */ 6666 void *private; 6667 6668 struct list_head list; 6669 struct rcu_head rcu; 6670 }; 6671 6672 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6673 struct list_head *adj_list) 6674 { 6675 struct netdev_adjacent *adj; 6676 6677 list_for_each_entry(adj, adj_list, list) { 6678 if (adj->dev == adj_dev) 6679 return adj; 6680 } 6681 return NULL; 6682 } 6683 6684 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6685 struct netdev_nested_priv *priv) 6686 { 6687 struct net_device *dev = (struct net_device *)priv->data; 6688 6689 return upper_dev == dev; 6690 } 6691 6692 /** 6693 * netdev_has_upper_dev - Check if device is linked to an upper device 6694 * @dev: device 6695 * @upper_dev: upper device to check 6696 * 6697 * Find out if a device is linked to specified upper device and return true 6698 * in case it is. Note that this checks only immediate upper device, 6699 * not through a complete stack of devices. The caller must hold the RTNL lock. 6700 */ 6701 bool netdev_has_upper_dev(struct net_device *dev, 6702 struct net_device *upper_dev) 6703 { 6704 struct netdev_nested_priv priv = { 6705 .data = (void *)upper_dev, 6706 }; 6707 6708 ASSERT_RTNL(); 6709 6710 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6711 &priv); 6712 } 6713 EXPORT_SYMBOL(netdev_has_upper_dev); 6714 6715 /** 6716 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6717 * @dev: device 6718 * @upper_dev: upper device to check 6719 * 6720 * Find out if a device is linked to specified upper device and return true 6721 * in case it is. Note that this checks the entire upper device chain. 6722 * The caller must hold rcu lock. 6723 */ 6724 6725 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6726 struct net_device *upper_dev) 6727 { 6728 struct netdev_nested_priv priv = { 6729 .data = (void *)upper_dev, 6730 }; 6731 6732 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6733 &priv); 6734 } 6735 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6736 6737 /** 6738 * netdev_has_any_upper_dev - Check if device is linked to some device 6739 * @dev: device 6740 * 6741 * Find out if a device is linked to an upper device and return true in case 6742 * it is. The caller must hold the RTNL lock. 6743 */ 6744 bool netdev_has_any_upper_dev(struct net_device *dev) 6745 { 6746 ASSERT_RTNL(); 6747 6748 return !list_empty(&dev->adj_list.upper); 6749 } 6750 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6751 6752 /** 6753 * netdev_master_upper_dev_get - Get master upper device 6754 * @dev: device 6755 * 6756 * Find a master upper device and return pointer to it or NULL in case 6757 * it's not there. The caller must hold the RTNL lock. 6758 */ 6759 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6760 { 6761 struct netdev_adjacent *upper; 6762 6763 ASSERT_RTNL(); 6764 6765 if (list_empty(&dev->adj_list.upper)) 6766 return NULL; 6767 6768 upper = list_first_entry(&dev->adj_list.upper, 6769 struct netdev_adjacent, list); 6770 if (likely(upper->master)) 6771 return upper->dev; 6772 return NULL; 6773 } 6774 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6775 6776 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6777 { 6778 struct netdev_adjacent *upper; 6779 6780 ASSERT_RTNL(); 6781 6782 if (list_empty(&dev->adj_list.upper)) 6783 return NULL; 6784 6785 upper = list_first_entry(&dev->adj_list.upper, 6786 struct netdev_adjacent, list); 6787 if (likely(upper->master) && !upper->ignore) 6788 return upper->dev; 6789 return NULL; 6790 } 6791 6792 /** 6793 * netdev_has_any_lower_dev - Check if device is linked to some device 6794 * @dev: device 6795 * 6796 * Find out if a device is linked to a lower device and return true in case 6797 * it is. The caller must hold the RTNL lock. 6798 */ 6799 static bool netdev_has_any_lower_dev(struct net_device *dev) 6800 { 6801 ASSERT_RTNL(); 6802 6803 return !list_empty(&dev->adj_list.lower); 6804 } 6805 6806 void *netdev_adjacent_get_private(struct list_head *adj_list) 6807 { 6808 struct netdev_adjacent *adj; 6809 6810 adj = list_entry(adj_list, struct netdev_adjacent, list); 6811 6812 return adj->private; 6813 } 6814 EXPORT_SYMBOL(netdev_adjacent_get_private); 6815 6816 /** 6817 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6818 * @dev: device 6819 * @iter: list_head ** of the current position 6820 * 6821 * Gets the next device from the dev's upper list, starting from iter 6822 * position. The caller must hold RCU read lock. 6823 */ 6824 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6825 struct list_head **iter) 6826 { 6827 struct netdev_adjacent *upper; 6828 6829 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6830 6831 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6832 6833 if (&upper->list == &dev->adj_list.upper) 6834 return NULL; 6835 6836 *iter = &upper->list; 6837 6838 return upper->dev; 6839 } 6840 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6841 6842 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6843 struct list_head **iter, 6844 bool *ignore) 6845 { 6846 struct netdev_adjacent *upper; 6847 6848 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6849 6850 if (&upper->list == &dev->adj_list.upper) 6851 return NULL; 6852 6853 *iter = &upper->list; 6854 *ignore = upper->ignore; 6855 6856 return upper->dev; 6857 } 6858 6859 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6860 struct list_head **iter) 6861 { 6862 struct netdev_adjacent *upper; 6863 6864 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6865 6866 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6867 6868 if (&upper->list == &dev->adj_list.upper) 6869 return NULL; 6870 6871 *iter = &upper->list; 6872 6873 return upper->dev; 6874 } 6875 6876 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6877 int (*fn)(struct net_device *dev, 6878 struct netdev_nested_priv *priv), 6879 struct netdev_nested_priv *priv) 6880 { 6881 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6882 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6883 int ret, cur = 0; 6884 bool ignore; 6885 6886 now = dev; 6887 iter = &dev->adj_list.upper; 6888 6889 while (1) { 6890 if (now != dev) { 6891 ret = fn(now, priv); 6892 if (ret) 6893 return ret; 6894 } 6895 6896 next = NULL; 6897 while (1) { 6898 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6899 if (!udev) 6900 break; 6901 if (ignore) 6902 continue; 6903 6904 next = udev; 6905 niter = &udev->adj_list.upper; 6906 dev_stack[cur] = now; 6907 iter_stack[cur++] = iter; 6908 break; 6909 } 6910 6911 if (!next) { 6912 if (!cur) 6913 return 0; 6914 next = dev_stack[--cur]; 6915 niter = iter_stack[cur]; 6916 } 6917 6918 now = next; 6919 iter = niter; 6920 } 6921 6922 return 0; 6923 } 6924 6925 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6926 int (*fn)(struct net_device *dev, 6927 struct netdev_nested_priv *priv), 6928 struct netdev_nested_priv *priv) 6929 { 6930 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6931 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6932 int ret, cur = 0; 6933 6934 now = dev; 6935 iter = &dev->adj_list.upper; 6936 6937 while (1) { 6938 if (now != dev) { 6939 ret = fn(now, priv); 6940 if (ret) 6941 return ret; 6942 } 6943 6944 next = NULL; 6945 while (1) { 6946 udev = netdev_next_upper_dev_rcu(now, &iter); 6947 if (!udev) 6948 break; 6949 6950 next = udev; 6951 niter = &udev->adj_list.upper; 6952 dev_stack[cur] = now; 6953 iter_stack[cur++] = iter; 6954 break; 6955 } 6956 6957 if (!next) { 6958 if (!cur) 6959 return 0; 6960 next = dev_stack[--cur]; 6961 niter = iter_stack[cur]; 6962 } 6963 6964 now = next; 6965 iter = niter; 6966 } 6967 6968 return 0; 6969 } 6970 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6971 6972 static bool __netdev_has_upper_dev(struct net_device *dev, 6973 struct net_device *upper_dev) 6974 { 6975 struct netdev_nested_priv priv = { 6976 .flags = 0, 6977 .data = (void *)upper_dev, 6978 }; 6979 6980 ASSERT_RTNL(); 6981 6982 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 6983 &priv); 6984 } 6985 6986 /** 6987 * netdev_lower_get_next_private - Get the next ->private from the 6988 * lower neighbour list 6989 * @dev: device 6990 * @iter: list_head ** of the current position 6991 * 6992 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6993 * list, starting from iter position. The caller must hold either hold the 6994 * RTNL lock or its own locking that guarantees that the neighbour lower 6995 * list will remain unchanged. 6996 */ 6997 void *netdev_lower_get_next_private(struct net_device *dev, 6998 struct list_head **iter) 6999 { 7000 struct netdev_adjacent *lower; 7001 7002 lower = list_entry(*iter, struct netdev_adjacent, list); 7003 7004 if (&lower->list == &dev->adj_list.lower) 7005 return NULL; 7006 7007 *iter = lower->list.next; 7008 7009 return lower->private; 7010 } 7011 EXPORT_SYMBOL(netdev_lower_get_next_private); 7012 7013 /** 7014 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7015 * lower neighbour list, RCU 7016 * variant 7017 * @dev: device 7018 * @iter: list_head ** of the current position 7019 * 7020 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7021 * list, starting from iter position. The caller must hold RCU read lock. 7022 */ 7023 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7024 struct list_head **iter) 7025 { 7026 struct netdev_adjacent *lower; 7027 7028 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7029 7030 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7031 7032 if (&lower->list == &dev->adj_list.lower) 7033 return NULL; 7034 7035 *iter = &lower->list; 7036 7037 return lower->private; 7038 } 7039 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7040 7041 /** 7042 * netdev_lower_get_next - Get the next device from the lower neighbour 7043 * list 7044 * @dev: device 7045 * @iter: list_head ** of the current position 7046 * 7047 * Gets the next netdev_adjacent from the dev's lower neighbour 7048 * list, starting from iter position. The caller must hold RTNL lock or 7049 * its own locking that guarantees that the neighbour lower 7050 * list will remain unchanged. 7051 */ 7052 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7053 { 7054 struct netdev_adjacent *lower; 7055 7056 lower = list_entry(*iter, struct netdev_adjacent, list); 7057 7058 if (&lower->list == &dev->adj_list.lower) 7059 return NULL; 7060 7061 *iter = lower->list.next; 7062 7063 return lower->dev; 7064 } 7065 EXPORT_SYMBOL(netdev_lower_get_next); 7066 7067 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7068 struct list_head **iter) 7069 { 7070 struct netdev_adjacent *lower; 7071 7072 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7073 7074 if (&lower->list == &dev->adj_list.lower) 7075 return NULL; 7076 7077 *iter = &lower->list; 7078 7079 return lower->dev; 7080 } 7081 7082 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7083 struct list_head **iter, 7084 bool *ignore) 7085 { 7086 struct netdev_adjacent *lower; 7087 7088 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7089 7090 if (&lower->list == &dev->adj_list.lower) 7091 return NULL; 7092 7093 *iter = &lower->list; 7094 *ignore = lower->ignore; 7095 7096 return lower->dev; 7097 } 7098 7099 int netdev_walk_all_lower_dev(struct net_device *dev, 7100 int (*fn)(struct net_device *dev, 7101 struct netdev_nested_priv *priv), 7102 struct netdev_nested_priv *priv) 7103 { 7104 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7105 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7106 int ret, cur = 0; 7107 7108 now = dev; 7109 iter = &dev->adj_list.lower; 7110 7111 while (1) { 7112 if (now != dev) { 7113 ret = fn(now, priv); 7114 if (ret) 7115 return ret; 7116 } 7117 7118 next = NULL; 7119 while (1) { 7120 ldev = netdev_next_lower_dev(now, &iter); 7121 if (!ldev) 7122 break; 7123 7124 next = ldev; 7125 niter = &ldev->adj_list.lower; 7126 dev_stack[cur] = now; 7127 iter_stack[cur++] = iter; 7128 break; 7129 } 7130 7131 if (!next) { 7132 if (!cur) 7133 return 0; 7134 next = dev_stack[--cur]; 7135 niter = iter_stack[cur]; 7136 } 7137 7138 now = next; 7139 iter = niter; 7140 } 7141 7142 return 0; 7143 } 7144 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7145 7146 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7147 int (*fn)(struct net_device *dev, 7148 struct netdev_nested_priv *priv), 7149 struct netdev_nested_priv *priv) 7150 { 7151 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7152 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7153 int ret, cur = 0; 7154 bool ignore; 7155 7156 now = dev; 7157 iter = &dev->adj_list.lower; 7158 7159 while (1) { 7160 if (now != dev) { 7161 ret = fn(now, priv); 7162 if (ret) 7163 return ret; 7164 } 7165 7166 next = NULL; 7167 while (1) { 7168 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7169 if (!ldev) 7170 break; 7171 if (ignore) 7172 continue; 7173 7174 next = ldev; 7175 niter = &ldev->adj_list.lower; 7176 dev_stack[cur] = now; 7177 iter_stack[cur++] = iter; 7178 break; 7179 } 7180 7181 if (!next) { 7182 if (!cur) 7183 return 0; 7184 next = dev_stack[--cur]; 7185 niter = iter_stack[cur]; 7186 } 7187 7188 now = next; 7189 iter = niter; 7190 } 7191 7192 return 0; 7193 } 7194 7195 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7196 struct list_head **iter) 7197 { 7198 struct netdev_adjacent *lower; 7199 7200 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7201 if (&lower->list == &dev->adj_list.lower) 7202 return NULL; 7203 7204 *iter = &lower->list; 7205 7206 return lower->dev; 7207 } 7208 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7209 7210 static u8 __netdev_upper_depth(struct net_device *dev) 7211 { 7212 struct net_device *udev; 7213 struct list_head *iter; 7214 u8 max_depth = 0; 7215 bool ignore; 7216 7217 for (iter = &dev->adj_list.upper, 7218 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7219 udev; 7220 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7221 if (ignore) 7222 continue; 7223 if (max_depth < udev->upper_level) 7224 max_depth = udev->upper_level; 7225 } 7226 7227 return max_depth; 7228 } 7229 7230 static u8 __netdev_lower_depth(struct net_device *dev) 7231 { 7232 struct net_device *ldev; 7233 struct list_head *iter; 7234 u8 max_depth = 0; 7235 bool ignore; 7236 7237 for (iter = &dev->adj_list.lower, 7238 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7239 ldev; 7240 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7241 if (ignore) 7242 continue; 7243 if (max_depth < ldev->lower_level) 7244 max_depth = ldev->lower_level; 7245 } 7246 7247 return max_depth; 7248 } 7249 7250 static int __netdev_update_upper_level(struct net_device *dev, 7251 struct netdev_nested_priv *__unused) 7252 { 7253 dev->upper_level = __netdev_upper_depth(dev) + 1; 7254 return 0; 7255 } 7256 7257 #ifdef CONFIG_LOCKDEP 7258 static LIST_HEAD(net_unlink_list); 7259 7260 static void net_unlink_todo(struct net_device *dev) 7261 { 7262 if (list_empty(&dev->unlink_list)) 7263 list_add_tail(&dev->unlink_list, &net_unlink_list); 7264 } 7265 #endif 7266 7267 static int __netdev_update_lower_level(struct net_device *dev, 7268 struct netdev_nested_priv *priv) 7269 { 7270 dev->lower_level = __netdev_lower_depth(dev) + 1; 7271 7272 #ifdef CONFIG_LOCKDEP 7273 if (!priv) 7274 return 0; 7275 7276 if (priv->flags & NESTED_SYNC_IMM) 7277 dev->nested_level = dev->lower_level - 1; 7278 if (priv->flags & NESTED_SYNC_TODO) 7279 net_unlink_todo(dev); 7280 #endif 7281 return 0; 7282 } 7283 7284 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7285 int (*fn)(struct net_device *dev, 7286 struct netdev_nested_priv *priv), 7287 struct netdev_nested_priv *priv) 7288 { 7289 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7290 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7291 int ret, cur = 0; 7292 7293 now = dev; 7294 iter = &dev->adj_list.lower; 7295 7296 while (1) { 7297 if (now != dev) { 7298 ret = fn(now, priv); 7299 if (ret) 7300 return ret; 7301 } 7302 7303 next = NULL; 7304 while (1) { 7305 ldev = netdev_next_lower_dev_rcu(now, &iter); 7306 if (!ldev) 7307 break; 7308 7309 next = ldev; 7310 niter = &ldev->adj_list.lower; 7311 dev_stack[cur] = now; 7312 iter_stack[cur++] = iter; 7313 break; 7314 } 7315 7316 if (!next) { 7317 if (!cur) 7318 return 0; 7319 next = dev_stack[--cur]; 7320 niter = iter_stack[cur]; 7321 } 7322 7323 now = next; 7324 iter = niter; 7325 } 7326 7327 return 0; 7328 } 7329 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7330 7331 /** 7332 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7333 * lower neighbour list, RCU 7334 * variant 7335 * @dev: device 7336 * 7337 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7338 * list. The caller must hold RCU read lock. 7339 */ 7340 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7341 { 7342 struct netdev_adjacent *lower; 7343 7344 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7345 struct netdev_adjacent, list); 7346 if (lower) 7347 return lower->private; 7348 return NULL; 7349 } 7350 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7351 7352 /** 7353 * netdev_master_upper_dev_get_rcu - Get master upper device 7354 * @dev: device 7355 * 7356 * Find a master upper device and return pointer to it or NULL in case 7357 * it's not there. The caller must hold the RCU read lock. 7358 */ 7359 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7360 { 7361 struct netdev_adjacent *upper; 7362 7363 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7364 struct netdev_adjacent, list); 7365 if (upper && likely(upper->master)) 7366 return upper->dev; 7367 return NULL; 7368 } 7369 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7370 7371 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7372 struct net_device *adj_dev, 7373 struct list_head *dev_list) 7374 { 7375 char linkname[IFNAMSIZ+7]; 7376 7377 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7378 "upper_%s" : "lower_%s", adj_dev->name); 7379 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7380 linkname); 7381 } 7382 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7383 char *name, 7384 struct list_head *dev_list) 7385 { 7386 char linkname[IFNAMSIZ+7]; 7387 7388 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7389 "upper_%s" : "lower_%s", name); 7390 sysfs_remove_link(&(dev->dev.kobj), linkname); 7391 } 7392 7393 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7394 struct net_device *adj_dev, 7395 struct list_head *dev_list) 7396 { 7397 return (dev_list == &dev->adj_list.upper || 7398 dev_list == &dev->adj_list.lower) && 7399 net_eq(dev_net(dev), dev_net(adj_dev)); 7400 } 7401 7402 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7403 struct net_device *adj_dev, 7404 struct list_head *dev_list, 7405 void *private, bool master) 7406 { 7407 struct netdev_adjacent *adj; 7408 int ret; 7409 7410 adj = __netdev_find_adj(adj_dev, dev_list); 7411 7412 if (adj) { 7413 adj->ref_nr += 1; 7414 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7415 dev->name, adj_dev->name, adj->ref_nr); 7416 7417 return 0; 7418 } 7419 7420 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7421 if (!adj) 7422 return -ENOMEM; 7423 7424 adj->dev = adj_dev; 7425 adj->master = master; 7426 adj->ref_nr = 1; 7427 adj->private = private; 7428 adj->ignore = false; 7429 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7430 7431 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7432 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7433 7434 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7435 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7436 if (ret) 7437 goto free_adj; 7438 } 7439 7440 /* Ensure that master link is always the first item in list. */ 7441 if (master) { 7442 ret = sysfs_create_link(&(dev->dev.kobj), 7443 &(adj_dev->dev.kobj), "master"); 7444 if (ret) 7445 goto remove_symlinks; 7446 7447 list_add_rcu(&adj->list, dev_list); 7448 } else { 7449 list_add_tail_rcu(&adj->list, dev_list); 7450 } 7451 7452 return 0; 7453 7454 remove_symlinks: 7455 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7456 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7457 free_adj: 7458 dev_put_track(adj_dev, &adj->dev_tracker); 7459 kfree(adj); 7460 7461 return ret; 7462 } 7463 7464 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7465 struct net_device *adj_dev, 7466 u16 ref_nr, 7467 struct list_head *dev_list) 7468 { 7469 struct netdev_adjacent *adj; 7470 7471 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7472 dev->name, adj_dev->name, ref_nr); 7473 7474 adj = __netdev_find_adj(adj_dev, dev_list); 7475 7476 if (!adj) { 7477 pr_err("Adjacency does not exist for device %s from %s\n", 7478 dev->name, adj_dev->name); 7479 WARN_ON(1); 7480 return; 7481 } 7482 7483 if (adj->ref_nr > ref_nr) { 7484 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7485 dev->name, adj_dev->name, ref_nr, 7486 adj->ref_nr - ref_nr); 7487 adj->ref_nr -= ref_nr; 7488 return; 7489 } 7490 7491 if (adj->master) 7492 sysfs_remove_link(&(dev->dev.kobj), "master"); 7493 7494 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7495 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7496 7497 list_del_rcu(&adj->list); 7498 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7499 adj_dev->name, dev->name, adj_dev->name); 7500 dev_put_track(adj_dev, &adj->dev_tracker); 7501 kfree_rcu(adj, rcu); 7502 } 7503 7504 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7505 struct net_device *upper_dev, 7506 struct list_head *up_list, 7507 struct list_head *down_list, 7508 void *private, bool master) 7509 { 7510 int ret; 7511 7512 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7513 private, master); 7514 if (ret) 7515 return ret; 7516 7517 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7518 private, false); 7519 if (ret) { 7520 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7521 return ret; 7522 } 7523 7524 return 0; 7525 } 7526 7527 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7528 struct net_device *upper_dev, 7529 u16 ref_nr, 7530 struct list_head *up_list, 7531 struct list_head *down_list) 7532 { 7533 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7534 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7535 } 7536 7537 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7538 struct net_device *upper_dev, 7539 void *private, bool master) 7540 { 7541 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7542 &dev->adj_list.upper, 7543 &upper_dev->adj_list.lower, 7544 private, master); 7545 } 7546 7547 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7548 struct net_device *upper_dev) 7549 { 7550 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7551 &dev->adj_list.upper, 7552 &upper_dev->adj_list.lower); 7553 } 7554 7555 static int __netdev_upper_dev_link(struct net_device *dev, 7556 struct net_device *upper_dev, bool master, 7557 void *upper_priv, void *upper_info, 7558 struct netdev_nested_priv *priv, 7559 struct netlink_ext_ack *extack) 7560 { 7561 struct netdev_notifier_changeupper_info changeupper_info = { 7562 .info = { 7563 .dev = dev, 7564 .extack = extack, 7565 }, 7566 .upper_dev = upper_dev, 7567 .master = master, 7568 .linking = true, 7569 .upper_info = upper_info, 7570 }; 7571 struct net_device *master_dev; 7572 int ret = 0; 7573 7574 ASSERT_RTNL(); 7575 7576 if (dev == upper_dev) 7577 return -EBUSY; 7578 7579 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7580 if (__netdev_has_upper_dev(upper_dev, dev)) 7581 return -EBUSY; 7582 7583 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7584 return -EMLINK; 7585 7586 if (!master) { 7587 if (__netdev_has_upper_dev(dev, upper_dev)) 7588 return -EEXIST; 7589 } else { 7590 master_dev = __netdev_master_upper_dev_get(dev); 7591 if (master_dev) 7592 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7593 } 7594 7595 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7596 &changeupper_info.info); 7597 ret = notifier_to_errno(ret); 7598 if (ret) 7599 return ret; 7600 7601 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7602 master); 7603 if (ret) 7604 return ret; 7605 7606 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7607 &changeupper_info.info); 7608 ret = notifier_to_errno(ret); 7609 if (ret) 7610 goto rollback; 7611 7612 __netdev_update_upper_level(dev, NULL); 7613 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7614 7615 __netdev_update_lower_level(upper_dev, priv); 7616 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7617 priv); 7618 7619 return 0; 7620 7621 rollback: 7622 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7623 7624 return ret; 7625 } 7626 7627 /** 7628 * netdev_upper_dev_link - Add a link to the upper device 7629 * @dev: device 7630 * @upper_dev: new upper device 7631 * @extack: netlink extended ack 7632 * 7633 * Adds a link to device which is upper to this one. The caller must hold 7634 * the RTNL lock. On a failure a negative errno code is returned. 7635 * On success the reference counts are adjusted and the function 7636 * returns zero. 7637 */ 7638 int netdev_upper_dev_link(struct net_device *dev, 7639 struct net_device *upper_dev, 7640 struct netlink_ext_ack *extack) 7641 { 7642 struct netdev_nested_priv priv = { 7643 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7644 .data = NULL, 7645 }; 7646 7647 return __netdev_upper_dev_link(dev, upper_dev, false, 7648 NULL, NULL, &priv, extack); 7649 } 7650 EXPORT_SYMBOL(netdev_upper_dev_link); 7651 7652 /** 7653 * netdev_master_upper_dev_link - Add a master link to the upper device 7654 * @dev: device 7655 * @upper_dev: new upper device 7656 * @upper_priv: upper device private 7657 * @upper_info: upper info to be passed down via notifier 7658 * @extack: netlink extended ack 7659 * 7660 * Adds a link to device which is upper to this one. In this case, only 7661 * one master upper device can be linked, although other non-master devices 7662 * might be linked as well. The caller must hold the RTNL lock. 7663 * On a failure a negative errno code is returned. On success the reference 7664 * counts are adjusted and the function returns zero. 7665 */ 7666 int netdev_master_upper_dev_link(struct net_device *dev, 7667 struct net_device *upper_dev, 7668 void *upper_priv, void *upper_info, 7669 struct netlink_ext_ack *extack) 7670 { 7671 struct netdev_nested_priv priv = { 7672 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7673 .data = NULL, 7674 }; 7675 7676 return __netdev_upper_dev_link(dev, upper_dev, true, 7677 upper_priv, upper_info, &priv, extack); 7678 } 7679 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7680 7681 static void __netdev_upper_dev_unlink(struct net_device *dev, 7682 struct net_device *upper_dev, 7683 struct netdev_nested_priv *priv) 7684 { 7685 struct netdev_notifier_changeupper_info changeupper_info = { 7686 .info = { 7687 .dev = dev, 7688 }, 7689 .upper_dev = upper_dev, 7690 .linking = false, 7691 }; 7692 7693 ASSERT_RTNL(); 7694 7695 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7696 7697 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7698 &changeupper_info.info); 7699 7700 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7701 7702 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7703 &changeupper_info.info); 7704 7705 __netdev_update_upper_level(dev, NULL); 7706 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7707 7708 __netdev_update_lower_level(upper_dev, priv); 7709 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7710 priv); 7711 } 7712 7713 /** 7714 * netdev_upper_dev_unlink - Removes a link to upper device 7715 * @dev: device 7716 * @upper_dev: new upper device 7717 * 7718 * Removes a link to device which is upper to this one. The caller must hold 7719 * the RTNL lock. 7720 */ 7721 void netdev_upper_dev_unlink(struct net_device *dev, 7722 struct net_device *upper_dev) 7723 { 7724 struct netdev_nested_priv priv = { 7725 .flags = NESTED_SYNC_TODO, 7726 .data = NULL, 7727 }; 7728 7729 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7730 } 7731 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7732 7733 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7734 struct net_device *lower_dev, 7735 bool val) 7736 { 7737 struct netdev_adjacent *adj; 7738 7739 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7740 if (adj) 7741 adj->ignore = val; 7742 7743 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7744 if (adj) 7745 adj->ignore = val; 7746 } 7747 7748 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7749 struct net_device *lower_dev) 7750 { 7751 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7752 } 7753 7754 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7755 struct net_device *lower_dev) 7756 { 7757 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7758 } 7759 7760 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7761 struct net_device *new_dev, 7762 struct net_device *dev, 7763 struct netlink_ext_ack *extack) 7764 { 7765 struct netdev_nested_priv priv = { 7766 .flags = 0, 7767 .data = NULL, 7768 }; 7769 int err; 7770 7771 if (!new_dev) 7772 return 0; 7773 7774 if (old_dev && new_dev != old_dev) 7775 netdev_adjacent_dev_disable(dev, old_dev); 7776 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7777 extack); 7778 if (err) { 7779 if (old_dev && new_dev != old_dev) 7780 netdev_adjacent_dev_enable(dev, old_dev); 7781 return err; 7782 } 7783 7784 return 0; 7785 } 7786 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7787 7788 void netdev_adjacent_change_commit(struct net_device *old_dev, 7789 struct net_device *new_dev, 7790 struct net_device *dev) 7791 { 7792 struct netdev_nested_priv priv = { 7793 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7794 .data = NULL, 7795 }; 7796 7797 if (!new_dev || !old_dev) 7798 return; 7799 7800 if (new_dev == old_dev) 7801 return; 7802 7803 netdev_adjacent_dev_enable(dev, old_dev); 7804 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7805 } 7806 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7807 7808 void netdev_adjacent_change_abort(struct net_device *old_dev, 7809 struct net_device *new_dev, 7810 struct net_device *dev) 7811 { 7812 struct netdev_nested_priv priv = { 7813 .flags = 0, 7814 .data = NULL, 7815 }; 7816 7817 if (!new_dev) 7818 return; 7819 7820 if (old_dev && new_dev != old_dev) 7821 netdev_adjacent_dev_enable(dev, old_dev); 7822 7823 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7824 } 7825 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7826 7827 /** 7828 * netdev_bonding_info_change - Dispatch event about slave change 7829 * @dev: device 7830 * @bonding_info: info to dispatch 7831 * 7832 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7833 * The caller must hold the RTNL lock. 7834 */ 7835 void netdev_bonding_info_change(struct net_device *dev, 7836 struct netdev_bonding_info *bonding_info) 7837 { 7838 struct netdev_notifier_bonding_info info = { 7839 .info.dev = dev, 7840 }; 7841 7842 memcpy(&info.bonding_info, bonding_info, 7843 sizeof(struct netdev_bonding_info)); 7844 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7845 &info.info); 7846 } 7847 EXPORT_SYMBOL(netdev_bonding_info_change); 7848 7849 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7850 struct netlink_ext_ack *extack) 7851 { 7852 struct netdev_notifier_offload_xstats_info info = { 7853 .info.dev = dev, 7854 .info.extack = extack, 7855 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7856 }; 7857 int err; 7858 int rc; 7859 7860 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7861 GFP_KERNEL); 7862 if (!dev->offload_xstats_l3) 7863 return -ENOMEM; 7864 7865 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7866 NETDEV_OFFLOAD_XSTATS_DISABLE, 7867 &info.info); 7868 err = notifier_to_errno(rc); 7869 if (err) 7870 goto free_stats; 7871 7872 return 0; 7873 7874 free_stats: 7875 kfree(dev->offload_xstats_l3); 7876 dev->offload_xstats_l3 = NULL; 7877 return err; 7878 } 7879 7880 int netdev_offload_xstats_enable(struct net_device *dev, 7881 enum netdev_offload_xstats_type type, 7882 struct netlink_ext_ack *extack) 7883 { 7884 ASSERT_RTNL(); 7885 7886 if (netdev_offload_xstats_enabled(dev, type)) 7887 return -EALREADY; 7888 7889 switch (type) { 7890 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7891 return netdev_offload_xstats_enable_l3(dev, extack); 7892 } 7893 7894 WARN_ON(1); 7895 return -EINVAL; 7896 } 7897 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7898 7899 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7900 { 7901 struct netdev_notifier_offload_xstats_info info = { 7902 .info.dev = dev, 7903 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7904 }; 7905 7906 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7907 &info.info); 7908 kfree(dev->offload_xstats_l3); 7909 dev->offload_xstats_l3 = NULL; 7910 } 7911 7912 int netdev_offload_xstats_disable(struct net_device *dev, 7913 enum netdev_offload_xstats_type type) 7914 { 7915 ASSERT_RTNL(); 7916 7917 if (!netdev_offload_xstats_enabled(dev, type)) 7918 return -EALREADY; 7919 7920 switch (type) { 7921 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7922 netdev_offload_xstats_disable_l3(dev); 7923 return 0; 7924 } 7925 7926 WARN_ON(1); 7927 return -EINVAL; 7928 } 7929 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7930 7931 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7932 { 7933 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7934 } 7935 7936 static struct rtnl_hw_stats64 * 7937 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7938 enum netdev_offload_xstats_type type) 7939 { 7940 switch (type) { 7941 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7942 return dev->offload_xstats_l3; 7943 } 7944 7945 WARN_ON(1); 7946 return NULL; 7947 } 7948 7949 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7950 enum netdev_offload_xstats_type type) 7951 { 7952 ASSERT_RTNL(); 7953 7954 return netdev_offload_xstats_get_ptr(dev, type); 7955 } 7956 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7957 7958 struct netdev_notifier_offload_xstats_ru { 7959 bool used; 7960 }; 7961 7962 struct netdev_notifier_offload_xstats_rd { 7963 struct rtnl_hw_stats64 stats; 7964 bool used; 7965 }; 7966 7967 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 7968 const struct rtnl_hw_stats64 *src) 7969 { 7970 dest->rx_packets += src->rx_packets; 7971 dest->tx_packets += src->tx_packets; 7972 dest->rx_bytes += src->rx_bytes; 7973 dest->tx_bytes += src->tx_bytes; 7974 dest->rx_errors += src->rx_errors; 7975 dest->tx_errors += src->tx_errors; 7976 dest->rx_dropped += src->rx_dropped; 7977 dest->tx_dropped += src->tx_dropped; 7978 dest->multicast += src->multicast; 7979 } 7980 7981 static int netdev_offload_xstats_get_used(struct net_device *dev, 7982 enum netdev_offload_xstats_type type, 7983 bool *p_used, 7984 struct netlink_ext_ack *extack) 7985 { 7986 struct netdev_notifier_offload_xstats_ru report_used = {}; 7987 struct netdev_notifier_offload_xstats_info info = { 7988 .info.dev = dev, 7989 .info.extack = extack, 7990 .type = type, 7991 .report_used = &report_used, 7992 }; 7993 int rc; 7994 7995 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 7996 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 7997 &info.info); 7998 *p_used = report_used.used; 7999 return notifier_to_errno(rc); 8000 } 8001 8002 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8003 enum netdev_offload_xstats_type type, 8004 struct rtnl_hw_stats64 *p_stats, 8005 bool *p_used, 8006 struct netlink_ext_ack *extack) 8007 { 8008 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8009 struct netdev_notifier_offload_xstats_info info = { 8010 .info.dev = dev, 8011 .info.extack = extack, 8012 .type = type, 8013 .report_delta = &report_delta, 8014 }; 8015 struct rtnl_hw_stats64 *stats; 8016 int rc; 8017 8018 stats = netdev_offload_xstats_get_ptr(dev, type); 8019 if (WARN_ON(!stats)) 8020 return -EINVAL; 8021 8022 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8023 &info.info); 8024 8025 /* Cache whatever we got, even if there was an error, otherwise the 8026 * successful stats retrievals would get lost. 8027 */ 8028 netdev_hw_stats64_add(stats, &report_delta.stats); 8029 8030 if (p_stats) 8031 *p_stats = *stats; 8032 *p_used = report_delta.used; 8033 8034 return notifier_to_errno(rc); 8035 } 8036 8037 int netdev_offload_xstats_get(struct net_device *dev, 8038 enum netdev_offload_xstats_type type, 8039 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8040 struct netlink_ext_ack *extack) 8041 { 8042 ASSERT_RTNL(); 8043 8044 if (p_stats) 8045 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8046 p_used, extack); 8047 else 8048 return netdev_offload_xstats_get_used(dev, type, p_used, 8049 extack); 8050 } 8051 EXPORT_SYMBOL(netdev_offload_xstats_get); 8052 8053 void 8054 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8055 const struct rtnl_hw_stats64 *stats) 8056 { 8057 report_delta->used = true; 8058 netdev_hw_stats64_add(&report_delta->stats, stats); 8059 } 8060 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8061 8062 void 8063 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8064 { 8065 report_used->used = true; 8066 } 8067 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8068 8069 void netdev_offload_xstats_push_delta(struct net_device *dev, 8070 enum netdev_offload_xstats_type type, 8071 const struct rtnl_hw_stats64 *p_stats) 8072 { 8073 struct rtnl_hw_stats64 *stats; 8074 8075 ASSERT_RTNL(); 8076 8077 stats = netdev_offload_xstats_get_ptr(dev, type); 8078 if (WARN_ON(!stats)) 8079 return; 8080 8081 netdev_hw_stats64_add(stats, p_stats); 8082 } 8083 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8084 8085 /** 8086 * netdev_get_xmit_slave - Get the xmit slave of master device 8087 * @dev: device 8088 * @skb: The packet 8089 * @all_slaves: assume all the slaves are active 8090 * 8091 * The reference counters are not incremented so the caller must be 8092 * careful with locks. The caller must hold RCU lock. 8093 * %NULL is returned if no slave is found. 8094 */ 8095 8096 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8097 struct sk_buff *skb, 8098 bool all_slaves) 8099 { 8100 const struct net_device_ops *ops = dev->netdev_ops; 8101 8102 if (!ops->ndo_get_xmit_slave) 8103 return NULL; 8104 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8105 } 8106 EXPORT_SYMBOL(netdev_get_xmit_slave); 8107 8108 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8109 struct sock *sk) 8110 { 8111 const struct net_device_ops *ops = dev->netdev_ops; 8112 8113 if (!ops->ndo_sk_get_lower_dev) 8114 return NULL; 8115 return ops->ndo_sk_get_lower_dev(dev, sk); 8116 } 8117 8118 /** 8119 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8120 * @dev: device 8121 * @sk: the socket 8122 * 8123 * %NULL is returned if no lower device is found. 8124 */ 8125 8126 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8127 struct sock *sk) 8128 { 8129 struct net_device *lower; 8130 8131 lower = netdev_sk_get_lower_dev(dev, sk); 8132 while (lower) { 8133 dev = lower; 8134 lower = netdev_sk_get_lower_dev(dev, sk); 8135 } 8136 8137 return dev; 8138 } 8139 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8140 8141 static void netdev_adjacent_add_links(struct net_device *dev) 8142 { 8143 struct netdev_adjacent *iter; 8144 8145 struct net *net = dev_net(dev); 8146 8147 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8148 if (!net_eq(net, dev_net(iter->dev))) 8149 continue; 8150 netdev_adjacent_sysfs_add(iter->dev, dev, 8151 &iter->dev->adj_list.lower); 8152 netdev_adjacent_sysfs_add(dev, iter->dev, 8153 &dev->adj_list.upper); 8154 } 8155 8156 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8157 if (!net_eq(net, dev_net(iter->dev))) 8158 continue; 8159 netdev_adjacent_sysfs_add(iter->dev, dev, 8160 &iter->dev->adj_list.upper); 8161 netdev_adjacent_sysfs_add(dev, iter->dev, 8162 &dev->adj_list.lower); 8163 } 8164 } 8165 8166 static void netdev_adjacent_del_links(struct net_device *dev) 8167 { 8168 struct netdev_adjacent *iter; 8169 8170 struct net *net = dev_net(dev); 8171 8172 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8173 if (!net_eq(net, dev_net(iter->dev))) 8174 continue; 8175 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8176 &iter->dev->adj_list.lower); 8177 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8178 &dev->adj_list.upper); 8179 } 8180 8181 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8182 if (!net_eq(net, dev_net(iter->dev))) 8183 continue; 8184 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8185 &iter->dev->adj_list.upper); 8186 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8187 &dev->adj_list.lower); 8188 } 8189 } 8190 8191 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8192 { 8193 struct netdev_adjacent *iter; 8194 8195 struct net *net = dev_net(dev); 8196 8197 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8198 if (!net_eq(net, dev_net(iter->dev))) 8199 continue; 8200 netdev_adjacent_sysfs_del(iter->dev, oldname, 8201 &iter->dev->adj_list.lower); 8202 netdev_adjacent_sysfs_add(iter->dev, dev, 8203 &iter->dev->adj_list.lower); 8204 } 8205 8206 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8207 if (!net_eq(net, dev_net(iter->dev))) 8208 continue; 8209 netdev_adjacent_sysfs_del(iter->dev, oldname, 8210 &iter->dev->adj_list.upper); 8211 netdev_adjacent_sysfs_add(iter->dev, dev, 8212 &iter->dev->adj_list.upper); 8213 } 8214 } 8215 8216 void *netdev_lower_dev_get_private(struct net_device *dev, 8217 struct net_device *lower_dev) 8218 { 8219 struct netdev_adjacent *lower; 8220 8221 if (!lower_dev) 8222 return NULL; 8223 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8224 if (!lower) 8225 return NULL; 8226 8227 return lower->private; 8228 } 8229 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8230 8231 8232 /** 8233 * netdev_lower_state_changed - Dispatch event about lower device state change 8234 * @lower_dev: device 8235 * @lower_state_info: state to dispatch 8236 * 8237 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8238 * The caller must hold the RTNL lock. 8239 */ 8240 void netdev_lower_state_changed(struct net_device *lower_dev, 8241 void *lower_state_info) 8242 { 8243 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8244 .info.dev = lower_dev, 8245 }; 8246 8247 ASSERT_RTNL(); 8248 changelowerstate_info.lower_state_info = lower_state_info; 8249 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8250 &changelowerstate_info.info); 8251 } 8252 EXPORT_SYMBOL(netdev_lower_state_changed); 8253 8254 static void dev_change_rx_flags(struct net_device *dev, int flags) 8255 { 8256 const struct net_device_ops *ops = dev->netdev_ops; 8257 8258 if (ops->ndo_change_rx_flags) 8259 ops->ndo_change_rx_flags(dev, flags); 8260 } 8261 8262 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8263 { 8264 unsigned int old_flags = dev->flags; 8265 kuid_t uid; 8266 kgid_t gid; 8267 8268 ASSERT_RTNL(); 8269 8270 dev->flags |= IFF_PROMISC; 8271 dev->promiscuity += inc; 8272 if (dev->promiscuity == 0) { 8273 /* 8274 * Avoid overflow. 8275 * If inc causes overflow, untouch promisc and return error. 8276 */ 8277 if (inc < 0) 8278 dev->flags &= ~IFF_PROMISC; 8279 else { 8280 dev->promiscuity -= inc; 8281 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8282 return -EOVERFLOW; 8283 } 8284 } 8285 if (dev->flags != old_flags) { 8286 pr_info("device %s %s promiscuous mode\n", 8287 dev->name, 8288 dev->flags & IFF_PROMISC ? "entered" : "left"); 8289 if (audit_enabled) { 8290 current_uid_gid(&uid, &gid); 8291 audit_log(audit_context(), GFP_ATOMIC, 8292 AUDIT_ANOM_PROMISCUOUS, 8293 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8294 dev->name, (dev->flags & IFF_PROMISC), 8295 (old_flags & IFF_PROMISC), 8296 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8297 from_kuid(&init_user_ns, uid), 8298 from_kgid(&init_user_ns, gid), 8299 audit_get_sessionid(current)); 8300 } 8301 8302 dev_change_rx_flags(dev, IFF_PROMISC); 8303 } 8304 if (notify) 8305 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8306 return 0; 8307 } 8308 8309 /** 8310 * dev_set_promiscuity - update promiscuity count on a device 8311 * @dev: device 8312 * @inc: modifier 8313 * 8314 * Add or remove promiscuity from a device. While the count in the device 8315 * remains above zero the interface remains promiscuous. Once it hits zero 8316 * the device reverts back to normal filtering operation. A negative inc 8317 * value is used to drop promiscuity on the device. 8318 * Return 0 if successful or a negative errno code on error. 8319 */ 8320 int dev_set_promiscuity(struct net_device *dev, int inc) 8321 { 8322 unsigned int old_flags = dev->flags; 8323 int err; 8324 8325 err = __dev_set_promiscuity(dev, inc, true); 8326 if (err < 0) 8327 return err; 8328 if (dev->flags != old_flags) 8329 dev_set_rx_mode(dev); 8330 return err; 8331 } 8332 EXPORT_SYMBOL(dev_set_promiscuity); 8333 8334 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8335 { 8336 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8337 8338 ASSERT_RTNL(); 8339 8340 dev->flags |= IFF_ALLMULTI; 8341 dev->allmulti += inc; 8342 if (dev->allmulti == 0) { 8343 /* 8344 * Avoid overflow. 8345 * If inc causes overflow, untouch allmulti and return error. 8346 */ 8347 if (inc < 0) 8348 dev->flags &= ~IFF_ALLMULTI; 8349 else { 8350 dev->allmulti -= inc; 8351 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8352 return -EOVERFLOW; 8353 } 8354 } 8355 if (dev->flags ^ old_flags) { 8356 dev_change_rx_flags(dev, IFF_ALLMULTI); 8357 dev_set_rx_mode(dev); 8358 if (notify) 8359 __dev_notify_flags(dev, old_flags, 8360 dev->gflags ^ old_gflags); 8361 } 8362 return 0; 8363 } 8364 8365 /** 8366 * dev_set_allmulti - update allmulti count on a device 8367 * @dev: device 8368 * @inc: modifier 8369 * 8370 * Add or remove reception of all multicast frames to a device. While the 8371 * count in the device remains above zero the interface remains listening 8372 * to all interfaces. Once it hits zero the device reverts back to normal 8373 * filtering operation. A negative @inc value is used to drop the counter 8374 * when releasing a resource needing all multicasts. 8375 * Return 0 if successful or a negative errno code on error. 8376 */ 8377 8378 int dev_set_allmulti(struct net_device *dev, int inc) 8379 { 8380 return __dev_set_allmulti(dev, inc, true); 8381 } 8382 EXPORT_SYMBOL(dev_set_allmulti); 8383 8384 /* 8385 * Upload unicast and multicast address lists to device and 8386 * configure RX filtering. When the device doesn't support unicast 8387 * filtering it is put in promiscuous mode while unicast addresses 8388 * are present. 8389 */ 8390 void __dev_set_rx_mode(struct net_device *dev) 8391 { 8392 const struct net_device_ops *ops = dev->netdev_ops; 8393 8394 /* dev_open will call this function so the list will stay sane. */ 8395 if (!(dev->flags&IFF_UP)) 8396 return; 8397 8398 if (!netif_device_present(dev)) 8399 return; 8400 8401 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8402 /* Unicast addresses changes may only happen under the rtnl, 8403 * therefore calling __dev_set_promiscuity here is safe. 8404 */ 8405 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8406 __dev_set_promiscuity(dev, 1, false); 8407 dev->uc_promisc = true; 8408 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8409 __dev_set_promiscuity(dev, -1, false); 8410 dev->uc_promisc = false; 8411 } 8412 } 8413 8414 if (ops->ndo_set_rx_mode) 8415 ops->ndo_set_rx_mode(dev); 8416 } 8417 8418 void dev_set_rx_mode(struct net_device *dev) 8419 { 8420 netif_addr_lock_bh(dev); 8421 __dev_set_rx_mode(dev); 8422 netif_addr_unlock_bh(dev); 8423 } 8424 8425 /** 8426 * dev_get_flags - get flags reported to userspace 8427 * @dev: device 8428 * 8429 * Get the combination of flag bits exported through APIs to userspace. 8430 */ 8431 unsigned int dev_get_flags(const struct net_device *dev) 8432 { 8433 unsigned int flags; 8434 8435 flags = (dev->flags & ~(IFF_PROMISC | 8436 IFF_ALLMULTI | 8437 IFF_RUNNING | 8438 IFF_LOWER_UP | 8439 IFF_DORMANT)) | 8440 (dev->gflags & (IFF_PROMISC | 8441 IFF_ALLMULTI)); 8442 8443 if (netif_running(dev)) { 8444 if (netif_oper_up(dev)) 8445 flags |= IFF_RUNNING; 8446 if (netif_carrier_ok(dev)) 8447 flags |= IFF_LOWER_UP; 8448 if (netif_dormant(dev)) 8449 flags |= IFF_DORMANT; 8450 } 8451 8452 return flags; 8453 } 8454 EXPORT_SYMBOL(dev_get_flags); 8455 8456 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8457 struct netlink_ext_ack *extack) 8458 { 8459 unsigned int old_flags = dev->flags; 8460 int ret; 8461 8462 ASSERT_RTNL(); 8463 8464 /* 8465 * Set the flags on our device. 8466 */ 8467 8468 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8469 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8470 IFF_AUTOMEDIA)) | 8471 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8472 IFF_ALLMULTI)); 8473 8474 /* 8475 * Load in the correct multicast list now the flags have changed. 8476 */ 8477 8478 if ((old_flags ^ flags) & IFF_MULTICAST) 8479 dev_change_rx_flags(dev, IFF_MULTICAST); 8480 8481 dev_set_rx_mode(dev); 8482 8483 /* 8484 * Have we downed the interface. We handle IFF_UP ourselves 8485 * according to user attempts to set it, rather than blindly 8486 * setting it. 8487 */ 8488 8489 ret = 0; 8490 if ((old_flags ^ flags) & IFF_UP) { 8491 if (old_flags & IFF_UP) 8492 __dev_close(dev); 8493 else 8494 ret = __dev_open(dev, extack); 8495 } 8496 8497 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8498 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8499 unsigned int old_flags = dev->flags; 8500 8501 dev->gflags ^= IFF_PROMISC; 8502 8503 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8504 if (dev->flags != old_flags) 8505 dev_set_rx_mode(dev); 8506 } 8507 8508 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8509 * is important. Some (broken) drivers set IFF_PROMISC, when 8510 * IFF_ALLMULTI is requested not asking us and not reporting. 8511 */ 8512 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8513 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8514 8515 dev->gflags ^= IFF_ALLMULTI; 8516 __dev_set_allmulti(dev, inc, false); 8517 } 8518 8519 return ret; 8520 } 8521 8522 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8523 unsigned int gchanges) 8524 { 8525 unsigned int changes = dev->flags ^ old_flags; 8526 8527 if (gchanges) 8528 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8529 8530 if (changes & IFF_UP) { 8531 if (dev->flags & IFF_UP) 8532 call_netdevice_notifiers(NETDEV_UP, dev); 8533 else 8534 call_netdevice_notifiers(NETDEV_DOWN, dev); 8535 } 8536 8537 if (dev->flags & IFF_UP && 8538 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8539 struct netdev_notifier_change_info change_info = { 8540 .info = { 8541 .dev = dev, 8542 }, 8543 .flags_changed = changes, 8544 }; 8545 8546 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8547 } 8548 } 8549 8550 /** 8551 * dev_change_flags - change device settings 8552 * @dev: device 8553 * @flags: device state flags 8554 * @extack: netlink extended ack 8555 * 8556 * Change settings on device based state flags. The flags are 8557 * in the userspace exported format. 8558 */ 8559 int dev_change_flags(struct net_device *dev, unsigned int flags, 8560 struct netlink_ext_ack *extack) 8561 { 8562 int ret; 8563 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8564 8565 ret = __dev_change_flags(dev, flags, extack); 8566 if (ret < 0) 8567 return ret; 8568 8569 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8570 __dev_notify_flags(dev, old_flags, changes); 8571 return ret; 8572 } 8573 EXPORT_SYMBOL(dev_change_flags); 8574 8575 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8576 { 8577 const struct net_device_ops *ops = dev->netdev_ops; 8578 8579 if (ops->ndo_change_mtu) 8580 return ops->ndo_change_mtu(dev, new_mtu); 8581 8582 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8583 WRITE_ONCE(dev->mtu, new_mtu); 8584 return 0; 8585 } 8586 EXPORT_SYMBOL(__dev_set_mtu); 8587 8588 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8589 struct netlink_ext_ack *extack) 8590 { 8591 /* MTU must be positive, and in range */ 8592 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8593 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8594 return -EINVAL; 8595 } 8596 8597 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8598 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8599 return -EINVAL; 8600 } 8601 return 0; 8602 } 8603 8604 /** 8605 * dev_set_mtu_ext - Change maximum transfer unit 8606 * @dev: device 8607 * @new_mtu: new transfer unit 8608 * @extack: netlink extended ack 8609 * 8610 * Change the maximum transfer size of the network device. 8611 */ 8612 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8613 struct netlink_ext_ack *extack) 8614 { 8615 int err, orig_mtu; 8616 8617 if (new_mtu == dev->mtu) 8618 return 0; 8619 8620 err = dev_validate_mtu(dev, new_mtu, extack); 8621 if (err) 8622 return err; 8623 8624 if (!netif_device_present(dev)) 8625 return -ENODEV; 8626 8627 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8628 err = notifier_to_errno(err); 8629 if (err) 8630 return err; 8631 8632 orig_mtu = dev->mtu; 8633 err = __dev_set_mtu(dev, new_mtu); 8634 8635 if (!err) { 8636 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8637 orig_mtu); 8638 err = notifier_to_errno(err); 8639 if (err) { 8640 /* setting mtu back and notifying everyone again, 8641 * so that they have a chance to revert changes. 8642 */ 8643 __dev_set_mtu(dev, orig_mtu); 8644 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8645 new_mtu); 8646 } 8647 } 8648 return err; 8649 } 8650 8651 int dev_set_mtu(struct net_device *dev, int new_mtu) 8652 { 8653 struct netlink_ext_ack extack; 8654 int err; 8655 8656 memset(&extack, 0, sizeof(extack)); 8657 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8658 if (err && extack._msg) 8659 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8660 return err; 8661 } 8662 EXPORT_SYMBOL(dev_set_mtu); 8663 8664 /** 8665 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8666 * @dev: device 8667 * @new_len: new tx queue length 8668 */ 8669 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8670 { 8671 unsigned int orig_len = dev->tx_queue_len; 8672 int res; 8673 8674 if (new_len != (unsigned int)new_len) 8675 return -ERANGE; 8676 8677 if (new_len != orig_len) { 8678 dev->tx_queue_len = new_len; 8679 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8680 res = notifier_to_errno(res); 8681 if (res) 8682 goto err_rollback; 8683 res = dev_qdisc_change_tx_queue_len(dev); 8684 if (res) 8685 goto err_rollback; 8686 } 8687 8688 return 0; 8689 8690 err_rollback: 8691 netdev_err(dev, "refused to change device tx_queue_len\n"); 8692 dev->tx_queue_len = orig_len; 8693 return res; 8694 } 8695 8696 /** 8697 * dev_set_group - Change group this device belongs to 8698 * @dev: device 8699 * @new_group: group this device should belong to 8700 */ 8701 void dev_set_group(struct net_device *dev, int new_group) 8702 { 8703 dev->group = new_group; 8704 } 8705 8706 /** 8707 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8708 * @dev: device 8709 * @addr: new address 8710 * @extack: netlink extended ack 8711 */ 8712 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8713 struct netlink_ext_ack *extack) 8714 { 8715 struct netdev_notifier_pre_changeaddr_info info = { 8716 .info.dev = dev, 8717 .info.extack = extack, 8718 .dev_addr = addr, 8719 }; 8720 int rc; 8721 8722 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8723 return notifier_to_errno(rc); 8724 } 8725 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8726 8727 /** 8728 * dev_set_mac_address - Change Media Access Control Address 8729 * @dev: device 8730 * @sa: new address 8731 * @extack: netlink extended ack 8732 * 8733 * Change the hardware (MAC) address of the device 8734 */ 8735 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8736 struct netlink_ext_ack *extack) 8737 { 8738 const struct net_device_ops *ops = dev->netdev_ops; 8739 int err; 8740 8741 if (!ops->ndo_set_mac_address) 8742 return -EOPNOTSUPP; 8743 if (sa->sa_family != dev->type) 8744 return -EINVAL; 8745 if (!netif_device_present(dev)) 8746 return -ENODEV; 8747 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8748 if (err) 8749 return err; 8750 err = ops->ndo_set_mac_address(dev, sa); 8751 if (err) 8752 return err; 8753 dev->addr_assign_type = NET_ADDR_SET; 8754 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8755 add_device_randomness(dev->dev_addr, dev->addr_len); 8756 return 0; 8757 } 8758 EXPORT_SYMBOL(dev_set_mac_address); 8759 8760 static DECLARE_RWSEM(dev_addr_sem); 8761 8762 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8763 struct netlink_ext_ack *extack) 8764 { 8765 int ret; 8766 8767 down_write(&dev_addr_sem); 8768 ret = dev_set_mac_address(dev, sa, extack); 8769 up_write(&dev_addr_sem); 8770 return ret; 8771 } 8772 EXPORT_SYMBOL(dev_set_mac_address_user); 8773 8774 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8775 { 8776 size_t size = sizeof(sa->sa_data); 8777 struct net_device *dev; 8778 int ret = 0; 8779 8780 down_read(&dev_addr_sem); 8781 rcu_read_lock(); 8782 8783 dev = dev_get_by_name_rcu(net, dev_name); 8784 if (!dev) { 8785 ret = -ENODEV; 8786 goto unlock; 8787 } 8788 if (!dev->addr_len) 8789 memset(sa->sa_data, 0, size); 8790 else 8791 memcpy(sa->sa_data, dev->dev_addr, 8792 min_t(size_t, size, dev->addr_len)); 8793 sa->sa_family = dev->type; 8794 8795 unlock: 8796 rcu_read_unlock(); 8797 up_read(&dev_addr_sem); 8798 return ret; 8799 } 8800 EXPORT_SYMBOL(dev_get_mac_address); 8801 8802 /** 8803 * dev_change_carrier - Change device carrier 8804 * @dev: device 8805 * @new_carrier: new value 8806 * 8807 * Change device carrier 8808 */ 8809 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8810 { 8811 const struct net_device_ops *ops = dev->netdev_ops; 8812 8813 if (!ops->ndo_change_carrier) 8814 return -EOPNOTSUPP; 8815 if (!netif_device_present(dev)) 8816 return -ENODEV; 8817 return ops->ndo_change_carrier(dev, new_carrier); 8818 } 8819 8820 /** 8821 * dev_get_phys_port_id - Get device physical port ID 8822 * @dev: device 8823 * @ppid: port ID 8824 * 8825 * Get device physical port ID 8826 */ 8827 int dev_get_phys_port_id(struct net_device *dev, 8828 struct netdev_phys_item_id *ppid) 8829 { 8830 const struct net_device_ops *ops = dev->netdev_ops; 8831 8832 if (!ops->ndo_get_phys_port_id) 8833 return -EOPNOTSUPP; 8834 return ops->ndo_get_phys_port_id(dev, ppid); 8835 } 8836 8837 /** 8838 * dev_get_phys_port_name - Get device physical port name 8839 * @dev: device 8840 * @name: port name 8841 * @len: limit of bytes to copy to name 8842 * 8843 * Get device physical port name 8844 */ 8845 int dev_get_phys_port_name(struct net_device *dev, 8846 char *name, size_t len) 8847 { 8848 const struct net_device_ops *ops = dev->netdev_ops; 8849 int err; 8850 8851 if (ops->ndo_get_phys_port_name) { 8852 err = ops->ndo_get_phys_port_name(dev, name, len); 8853 if (err != -EOPNOTSUPP) 8854 return err; 8855 } 8856 return devlink_compat_phys_port_name_get(dev, name, len); 8857 } 8858 8859 /** 8860 * dev_get_port_parent_id - Get the device's port parent identifier 8861 * @dev: network device 8862 * @ppid: pointer to a storage for the port's parent identifier 8863 * @recurse: allow/disallow recursion to lower devices 8864 * 8865 * Get the devices's port parent identifier 8866 */ 8867 int dev_get_port_parent_id(struct net_device *dev, 8868 struct netdev_phys_item_id *ppid, 8869 bool recurse) 8870 { 8871 const struct net_device_ops *ops = dev->netdev_ops; 8872 struct netdev_phys_item_id first = { }; 8873 struct net_device *lower_dev; 8874 struct list_head *iter; 8875 int err; 8876 8877 if (ops->ndo_get_port_parent_id) { 8878 err = ops->ndo_get_port_parent_id(dev, ppid); 8879 if (err != -EOPNOTSUPP) 8880 return err; 8881 } 8882 8883 err = devlink_compat_switch_id_get(dev, ppid); 8884 if (!recurse || err != -EOPNOTSUPP) 8885 return err; 8886 8887 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8888 err = dev_get_port_parent_id(lower_dev, ppid, true); 8889 if (err) 8890 break; 8891 if (!first.id_len) 8892 first = *ppid; 8893 else if (memcmp(&first, ppid, sizeof(*ppid))) 8894 return -EOPNOTSUPP; 8895 } 8896 8897 return err; 8898 } 8899 EXPORT_SYMBOL(dev_get_port_parent_id); 8900 8901 /** 8902 * netdev_port_same_parent_id - Indicate if two network devices have 8903 * the same port parent identifier 8904 * @a: first network device 8905 * @b: second network device 8906 */ 8907 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8908 { 8909 struct netdev_phys_item_id a_id = { }; 8910 struct netdev_phys_item_id b_id = { }; 8911 8912 if (dev_get_port_parent_id(a, &a_id, true) || 8913 dev_get_port_parent_id(b, &b_id, true)) 8914 return false; 8915 8916 return netdev_phys_item_id_same(&a_id, &b_id); 8917 } 8918 EXPORT_SYMBOL(netdev_port_same_parent_id); 8919 8920 /** 8921 * dev_change_proto_down - set carrier according to proto_down. 8922 * 8923 * @dev: device 8924 * @proto_down: new value 8925 */ 8926 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8927 { 8928 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8929 return -EOPNOTSUPP; 8930 if (!netif_device_present(dev)) 8931 return -ENODEV; 8932 if (proto_down) 8933 netif_carrier_off(dev); 8934 else 8935 netif_carrier_on(dev); 8936 dev->proto_down = proto_down; 8937 return 0; 8938 } 8939 8940 /** 8941 * dev_change_proto_down_reason - proto down reason 8942 * 8943 * @dev: device 8944 * @mask: proto down mask 8945 * @value: proto down value 8946 */ 8947 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8948 u32 value) 8949 { 8950 int b; 8951 8952 if (!mask) { 8953 dev->proto_down_reason = value; 8954 } else { 8955 for_each_set_bit(b, &mask, 32) { 8956 if (value & (1 << b)) 8957 dev->proto_down_reason |= BIT(b); 8958 else 8959 dev->proto_down_reason &= ~BIT(b); 8960 } 8961 } 8962 } 8963 8964 struct bpf_xdp_link { 8965 struct bpf_link link; 8966 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8967 int flags; 8968 }; 8969 8970 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8971 { 8972 if (flags & XDP_FLAGS_HW_MODE) 8973 return XDP_MODE_HW; 8974 if (flags & XDP_FLAGS_DRV_MODE) 8975 return XDP_MODE_DRV; 8976 if (flags & XDP_FLAGS_SKB_MODE) 8977 return XDP_MODE_SKB; 8978 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8979 } 8980 8981 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8982 { 8983 switch (mode) { 8984 case XDP_MODE_SKB: 8985 return generic_xdp_install; 8986 case XDP_MODE_DRV: 8987 case XDP_MODE_HW: 8988 return dev->netdev_ops->ndo_bpf; 8989 default: 8990 return NULL; 8991 } 8992 } 8993 8994 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8995 enum bpf_xdp_mode mode) 8996 { 8997 return dev->xdp_state[mode].link; 8998 } 8999 9000 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9001 enum bpf_xdp_mode mode) 9002 { 9003 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9004 9005 if (link) 9006 return link->link.prog; 9007 return dev->xdp_state[mode].prog; 9008 } 9009 9010 u8 dev_xdp_prog_count(struct net_device *dev) 9011 { 9012 u8 count = 0; 9013 int i; 9014 9015 for (i = 0; i < __MAX_XDP_MODE; i++) 9016 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9017 count++; 9018 return count; 9019 } 9020 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9021 9022 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9023 { 9024 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9025 9026 return prog ? prog->aux->id : 0; 9027 } 9028 9029 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9030 struct bpf_xdp_link *link) 9031 { 9032 dev->xdp_state[mode].link = link; 9033 dev->xdp_state[mode].prog = NULL; 9034 } 9035 9036 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9037 struct bpf_prog *prog) 9038 { 9039 dev->xdp_state[mode].link = NULL; 9040 dev->xdp_state[mode].prog = prog; 9041 } 9042 9043 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9044 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9045 u32 flags, struct bpf_prog *prog) 9046 { 9047 struct netdev_bpf xdp; 9048 int err; 9049 9050 memset(&xdp, 0, sizeof(xdp)); 9051 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9052 xdp.extack = extack; 9053 xdp.flags = flags; 9054 xdp.prog = prog; 9055 9056 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9057 * "moved" into driver), so they don't increment it on their own, but 9058 * they do decrement refcnt when program is detached or replaced. 9059 * Given net_device also owns link/prog, we need to bump refcnt here 9060 * to prevent drivers from underflowing it. 9061 */ 9062 if (prog) 9063 bpf_prog_inc(prog); 9064 err = bpf_op(dev, &xdp); 9065 if (err) { 9066 if (prog) 9067 bpf_prog_put(prog); 9068 return err; 9069 } 9070 9071 if (mode != XDP_MODE_HW) 9072 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9073 9074 return 0; 9075 } 9076 9077 static void dev_xdp_uninstall(struct net_device *dev) 9078 { 9079 struct bpf_xdp_link *link; 9080 struct bpf_prog *prog; 9081 enum bpf_xdp_mode mode; 9082 bpf_op_t bpf_op; 9083 9084 ASSERT_RTNL(); 9085 9086 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9087 prog = dev_xdp_prog(dev, mode); 9088 if (!prog) 9089 continue; 9090 9091 bpf_op = dev_xdp_bpf_op(dev, mode); 9092 if (!bpf_op) 9093 continue; 9094 9095 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9096 9097 /* auto-detach link from net device */ 9098 link = dev_xdp_link(dev, mode); 9099 if (link) 9100 link->dev = NULL; 9101 else 9102 bpf_prog_put(prog); 9103 9104 dev_xdp_set_link(dev, mode, NULL); 9105 } 9106 } 9107 9108 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9109 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9110 struct bpf_prog *old_prog, u32 flags) 9111 { 9112 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9113 struct bpf_prog *cur_prog; 9114 struct net_device *upper; 9115 struct list_head *iter; 9116 enum bpf_xdp_mode mode; 9117 bpf_op_t bpf_op; 9118 int err; 9119 9120 ASSERT_RTNL(); 9121 9122 /* either link or prog attachment, never both */ 9123 if (link && (new_prog || old_prog)) 9124 return -EINVAL; 9125 /* link supports only XDP mode flags */ 9126 if (link && (flags & ~XDP_FLAGS_MODES)) { 9127 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9128 return -EINVAL; 9129 } 9130 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9131 if (num_modes > 1) { 9132 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9133 return -EINVAL; 9134 } 9135 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9136 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9137 NL_SET_ERR_MSG(extack, 9138 "More than one program loaded, unset mode is ambiguous"); 9139 return -EINVAL; 9140 } 9141 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9142 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9143 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9144 return -EINVAL; 9145 } 9146 9147 mode = dev_xdp_mode(dev, flags); 9148 /* can't replace attached link */ 9149 if (dev_xdp_link(dev, mode)) { 9150 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9151 return -EBUSY; 9152 } 9153 9154 /* don't allow if an upper device already has a program */ 9155 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9156 if (dev_xdp_prog_count(upper) > 0) { 9157 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9158 return -EEXIST; 9159 } 9160 } 9161 9162 cur_prog = dev_xdp_prog(dev, mode); 9163 /* can't replace attached prog with link */ 9164 if (link && cur_prog) { 9165 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9166 return -EBUSY; 9167 } 9168 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9169 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9170 return -EEXIST; 9171 } 9172 9173 /* put effective new program into new_prog */ 9174 if (link) 9175 new_prog = link->link.prog; 9176 9177 if (new_prog) { 9178 bool offload = mode == XDP_MODE_HW; 9179 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9180 ? XDP_MODE_DRV : XDP_MODE_SKB; 9181 9182 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9183 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9184 return -EBUSY; 9185 } 9186 if (!offload && dev_xdp_prog(dev, other_mode)) { 9187 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9188 return -EEXIST; 9189 } 9190 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9191 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9192 return -EINVAL; 9193 } 9194 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9195 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9196 return -EINVAL; 9197 } 9198 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9199 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9200 return -EINVAL; 9201 } 9202 } 9203 9204 /* don't call drivers if the effective program didn't change */ 9205 if (new_prog != cur_prog) { 9206 bpf_op = dev_xdp_bpf_op(dev, mode); 9207 if (!bpf_op) { 9208 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9209 return -EOPNOTSUPP; 9210 } 9211 9212 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9213 if (err) 9214 return err; 9215 } 9216 9217 if (link) 9218 dev_xdp_set_link(dev, mode, link); 9219 else 9220 dev_xdp_set_prog(dev, mode, new_prog); 9221 if (cur_prog) 9222 bpf_prog_put(cur_prog); 9223 9224 return 0; 9225 } 9226 9227 static int dev_xdp_attach_link(struct net_device *dev, 9228 struct netlink_ext_ack *extack, 9229 struct bpf_xdp_link *link) 9230 { 9231 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9232 } 9233 9234 static int dev_xdp_detach_link(struct net_device *dev, 9235 struct netlink_ext_ack *extack, 9236 struct bpf_xdp_link *link) 9237 { 9238 enum bpf_xdp_mode mode; 9239 bpf_op_t bpf_op; 9240 9241 ASSERT_RTNL(); 9242 9243 mode = dev_xdp_mode(dev, link->flags); 9244 if (dev_xdp_link(dev, mode) != link) 9245 return -EINVAL; 9246 9247 bpf_op = dev_xdp_bpf_op(dev, mode); 9248 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9249 dev_xdp_set_link(dev, mode, NULL); 9250 return 0; 9251 } 9252 9253 static void bpf_xdp_link_release(struct bpf_link *link) 9254 { 9255 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9256 9257 rtnl_lock(); 9258 9259 /* if racing with net_device's tear down, xdp_link->dev might be 9260 * already NULL, in which case link was already auto-detached 9261 */ 9262 if (xdp_link->dev) { 9263 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9264 xdp_link->dev = NULL; 9265 } 9266 9267 rtnl_unlock(); 9268 } 9269 9270 static int bpf_xdp_link_detach(struct bpf_link *link) 9271 { 9272 bpf_xdp_link_release(link); 9273 return 0; 9274 } 9275 9276 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9277 { 9278 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9279 9280 kfree(xdp_link); 9281 } 9282 9283 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9284 struct seq_file *seq) 9285 { 9286 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9287 u32 ifindex = 0; 9288 9289 rtnl_lock(); 9290 if (xdp_link->dev) 9291 ifindex = xdp_link->dev->ifindex; 9292 rtnl_unlock(); 9293 9294 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9295 } 9296 9297 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9298 struct bpf_link_info *info) 9299 { 9300 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9301 u32 ifindex = 0; 9302 9303 rtnl_lock(); 9304 if (xdp_link->dev) 9305 ifindex = xdp_link->dev->ifindex; 9306 rtnl_unlock(); 9307 9308 info->xdp.ifindex = ifindex; 9309 return 0; 9310 } 9311 9312 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9313 struct bpf_prog *old_prog) 9314 { 9315 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9316 enum bpf_xdp_mode mode; 9317 bpf_op_t bpf_op; 9318 int err = 0; 9319 9320 rtnl_lock(); 9321 9322 /* link might have been auto-released already, so fail */ 9323 if (!xdp_link->dev) { 9324 err = -ENOLINK; 9325 goto out_unlock; 9326 } 9327 9328 if (old_prog && link->prog != old_prog) { 9329 err = -EPERM; 9330 goto out_unlock; 9331 } 9332 old_prog = link->prog; 9333 if (old_prog->type != new_prog->type || 9334 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9335 err = -EINVAL; 9336 goto out_unlock; 9337 } 9338 9339 if (old_prog == new_prog) { 9340 /* no-op, don't disturb drivers */ 9341 bpf_prog_put(new_prog); 9342 goto out_unlock; 9343 } 9344 9345 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9346 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9347 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9348 xdp_link->flags, new_prog); 9349 if (err) 9350 goto out_unlock; 9351 9352 old_prog = xchg(&link->prog, new_prog); 9353 bpf_prog_put(old_prog); 9354 9355 out_unlock: 9356 rtnl_unlock(); 9357 return err; 9358 } 9359 9360 static const struct bpf_link_ops bpf_xdp_link_lops = { 9361 .release = bpf_xdp_link_release, 9362 .dealloc = bpf_xdp_link_dealloc, 9363 .detach = bpf_xdp_link_detach, 9364 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9365 .fill_link_info = bpf_xdp_link_fill_link_info, 9366 .update_prog = bpf_xdp_link_update, 9367 }; 9368 9369 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9370 { 9371 struct net *net = current->nsproxy->net_ns; 9372 struct bpf_link_primer link_primer; 9373 struct bpf_xdp_link *link; 9374 struct net_device *dev; 9375 int err, fd; 9376 9377 rtnl_lock(); 9378 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9379 if (!dev) { 9380 rtnl_unlock(); 9381 return -EINVAL; 9382 } 9383 9384 link = kzalloc(sizeof(*link), GFP_USER); 9385 if (!link) { 9386 err = -ENOMEM; 9387 goto unlock; 9388 } 9389 9390 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9391 link->dev = dev; 9392 link->flags = attr->link_create.flags; 9393 9394 err = bpf_link_prime(&link->link, &link_primer); 9395 if (err) { 9396 kfree(link); 9397 goto unlock; 9398 } 9399 9400 err = dev_xdp_attach_link(dev, NULL, link); 9401 rtnl_unlock(); 9402 9403 if (err) { 9404 link->dev = NULL; 9405 bpf_link_cleanup(&link_primer); 9406 goto out_put_dev; 9407 } 9408 9409 fd = bpf_link_settle(&link_primer); 9410 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9411 dev_put(dev); 9412 return fd; 9413 9414 unlock: 9415 rtnl_unlock(); 9416 9417 out_put_dev: 9418 dev_put(dev); 9419 return err; 9420 } 9421 9422 /** 9423 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9424 * @dev: device 9425 * @extack: netlink extended ack 9426 * @fd: new program fd or negative value to clear 9427 * @expected_fd: old program fd that userspace expects to replace or clear 9428 * @flags: xdp-related flags 9429 * 9430 * Set or clear a bpf program for a device 9431 */ 9432 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9433 int fd, int expected_fd, u32 flags) 9434 { 9435 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9436 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9437 int err; 9438 9439 ASSERT_RTNL(); 9440 9441 if (fd >= 0) { 9442 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9443 mode != XDP_MODE_SKB); 9444 if (IS_ERR(new_prog)) 9445 return PTR_ERR(new_prog); 9446 } 9447 9448 if (expected_fd >= 0) { 9449 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9450 mode != XDP_MODE_SKB); 9451 if (IS_ERR(old_prog)) { 9452 err = PTR_ERR(old_prog); 9453 old_prog = NULL; 9454 goto err_out; 9455 } 9456 } 9457 9458 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9459 9460 err_out: 9461 if (err && new_prog) 9462 bpf_prog_put(new_prog); 9463 if (old_prog) 9464 bpf_prog_put(old_prog); 9465 return err; 9466 } 9467 9468 /** 9469 * dev_new_index - allocate an ifindex 9470 * @net: the applicable net namespace 9471 * 9472 * Returns a suitable unique value for a new device interface 9473 * number. The caller must hold the rtnl semaphore or the 9474 * dev_base_lock to be sure it remains unique. 9475 */ 9476 static int dev_new_index(struct net *net) 9477 { 9478 int ifindex = net->ifindex; 9479 9480 for (;;) { 9481 if (++ifindex <= 0) 9482 ifindex = 1; 9483 if (!__dev_get_by_index(net, ifindex)) 9484 return net->ifindex = ifindex; 9485 } 9486 } 9487 9488 /* Delayed registration/unregisteration */ 9489 LIST_HEAD(net_todo_list); 9490 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9491 9492 static void net_set_todo(struct net_device *dev) 9493 { 9494 list_add_tail(&dev->todo_list, &net_todo_list); 9495 atomic_inc(&dev_net(dev)->dev_unreg_count); 9496 } 9497 9498 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9499 struct net_device *upper, netdev_features_t features) 9500 { 9501 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9502 netdev_features_t feature; 9503 int feature_bit; 9504 9505 for_each_netdev_feature(upper_disables, feature_bit) { 9506 feature = __NETIF_F_BIT(feature_bit); 9507 if (!(upper->wanted_features & feature) 9508 && (features & feature)) { 9509 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9510 &feature, upper->name); 9511 features &= ~feature; 9512 } 9513 } 9514 9515 return features; 9516 } 9517 9518 static void netdev_sync_lower_features(struct net_device *upper, 9519 struct net_device *lower, netdev_features_t features) 9520 { 9521 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9522 netdev_features_t feature; 9523 int feature_bit; 9524 9525 for_each_netdev_feature(upper_disables, feature_bit) { 9526 feature = __NETIF_F_BIT(feature_bit); 9527 if (!(features & feature) && (lower->features & feature)) { 9528 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9529 &feature, lower->name); 9530 lower->wanted_features &= ~feature; 9531 __netdev_update_features(lower); 9532 9533 if (unlikely(lower->features & feature)) 9534 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9535 &feature, lower->name); 9536 else 9537 netdev_features_change(lower); 9538 } 9539 } 9540 } 9541 9542 static netdev_features_t netdev_fix_features(struct net_device *dev, 9543 netdev_features_t features) 9544 { 9545 /* Fix illegal checksum combinations */ 9546 if ((features & NETIF_F_HW_CSUM) && 9547 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9548 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9549 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9550 } 9551 9552 /* TSO requires that SG is present as well. */ 9553 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9554 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9555 features &= ~NETIF_F_ALL_TSO; 9556 } 9557 9558 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9559 !(features & NETIF_F_IP_CSUM)) { 9560 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9561 features &= ~NETIF_F_TSO; 9562 features &= ~NETIF_F_TSO_ECN; 9563 } 9564 9565 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9566 !(features & NETIF_F_IPV6_CSUM)) { 9567 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9568 features &= ~NETIF_F_TSO6; 9569 } 9570 9571 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9572 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9573 features &= ~NETIF_F_TSO_MANGLEID; 9574 9575 /* TSO ECN requires that TSO is present as well. */ 9576 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9577 features &= ~NETIF_F_TSO_ECN; 9578 9579 /* Software GSO depends on SG. */ 9580 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9581 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9582 features &= ~NETIF_F_GSO; 9583 } 9584 9585 /* GSO partial features require GSO partial be set */ 9586 if ((features & dev->gso_partial_features) && 9587 !(features & NETIF_F_GSO_PARTIAL)) { 9588 netdev_dbg(dev, 9589 "Dropping partially supported GSO features since no GSO partial.\n"); 9590 features &= ~dev->gso_partial_features; 9591 } 9592 9593 if (!(features & NETIF_F_RXCSUM)) { 9594 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9595 * successfully merged by hardware must also have the 9596 * checksum verified by hardware. If the user does not 9597 * want to enable RXCSUM, logically, we should disable GRO_HW. 9598 */ 9599 if (features & NETIF_F_GRO_HW) { 9600 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9601 features &= ~NETIF_F_GRO_HW; 9602 } 9603 } 9604 9605 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9606 if (features & NETIF_F_RXFCS) { 9607 if (features & NETIF_F_LRO) { 9608 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9609 features &= ~NETIF_F_LRO; 9610 } 9611 9612 if (features & NETIF_F_GRO_HW) { 9613 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9614 features &= ~NETIF_F_GRO_HW; 9615 } 9616 } 9617 9618 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9619 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9620 features &= ~NETIF_F_LRO; 9621 } 9622 9623 if (features & NETIF_F_HW_TLS_TX) { 9624 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9625 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9626 bool hw_csum = features & NETIF_F_HW_CSUM; 9627 9628 if (!ip_csum && !hw_csum) { 9629 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9630 features &= ~NETIF_F_HW_TLS_TX; 9631 } 9632 } 9633 9634 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9635 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9636 features &= ~NETIF_F_HW_TLS_RX; 9637 } 9638 9639 return features; 9640 } 9641 9642 int __netdev_update_features(struct net_device *dev) 9643 { 9644 struct net_device *upper, *lower; 9645 netdev_features_t features; 9646 struct list_head *iter; 9647 int err = -1; 9648 9649 ASSERT_RTNL(); 9650 9651 features = netdev_get_wanted_features(dev); 9652 9653 if (dev->netdev_ops->ndo_fix_features) 9654 features = dev->netdev_ops->ndo_fix_features(dev, features); 9655 9656 /* driver might be less strict about feature dependencies */ 9657 features = netdev_fix_features(dev, features); 9658 9659 /* some features can't be enabled if they're off on an upper device */ 9660 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9661 features = netdev_sync_upper_features(dev, upper, features); 9662 9663 if (dev->features == features) 9664 goto sync_lower; 9665 9666 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9667 &dev->features, &features); 9668 9669 if (dev->netdev_ops->ndo_set_features) 9670 err = dev->netdev_ops->ndo_set_features(dev, features); 9671 else 9672 err = 0; 9673 9674 if (unlikely(err < 0)) { 9675 netdev_err(dev, 9676 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9677 err, &features, &dev->features); 9678 /* return non-0 since some features might have changed and 9679 * it's better to fire a spurious notification than miss it 9680 */ 9681 return -1; 9682 } 9683 9684 sync_lower: 9685 /* some features must be disabled on lower devices when disabled 9686 * on an upper device (think: bonding master or bridge) 9687 */ 9688 netdev_for_each_lower_dev(dev, lower, iter) 9689 netdev_sync_lower_features(dev, lower, features); 9690 9691 if (!err) { 9692 netdev_features_t diff = features ^ dev->features; 9693 9694 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9695 /* udp_tunnel_{get,drop}_rx_info both need 9696 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9697 * device, or they won't do anything. 9698 * Thus we need to update dev->features 9699 * *before* calling udp_tunnel_get_rx_info, 9700 * but *after* calling udp_tunnel_drop_rx_info. 9701 */ 9702 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9703 dev->features = features; 9704 udp_tunnel_get_rx_info(dev); 9705 } else { 9706 udp_tunnel_drop_rx_info(dev); 9707 } 9708 } 9709 9710 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9711 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9712 dev->features = features; 9713 err |= vlan_get_rx_ctag_filter_info(dev); 9714 } else { 9715 vlan_drop_rx_ctag_filter_info(dev); 9716 } 9717 } 9718 9719 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9720 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9721 dev->features = features; 9722 err |= vlan_get_rx_stag_filter_info(dev); 9723 } else { 9724 vlan_drop_rx_stag_filter_info(dev); 9725 } 9726 } 9727 9728 dev->features = features; 9729 } 9730 9731 return err < 0 ? 0 : 1; 9732 } 9733 9734 /** 9735 * netdev_update_features - recalculate device features 9736 * @dev: the device to check 9737 * 9738 * Recalculate dev->features set and send notifications if it 9739 * has changed. Should be called after driver or hardware dependent 9740 * conditions might have changed that influence the features. 9741 */ 9742 void netdev_update_features(struct net_device *dev) 9743 { 9744 if (__netdev_update_features(dev)) 9745 netdev_features_change(dev); 9746 } 9747 EXPORT_SYMBOL(netdev_update_features); 9748 9749 /** 9750 * netdev_change_features - recalculate device features 9751 * @dev: the device to check 9752 * 9753 * Recalculate dev->features set and send notifications even 9754 * if they have not changed. Should be called instead of 9755 * netdev_update_features() if also dev->vlan_features might 9756 * have changed to allow the changes to be propagated to stacked 9757 * VLAN devices. 9758 */ 9759 void netdev_change_features(struct net_device *dev) 9760 { 9761 __netdev_update_features(dev); 9762 netdev_features_change(dev); 9763 } 9764 EXPORT_SYMBOL(netdev_change_features); 9765 9766 /** 9767 * netif_stacked_transfer_operstate - transfer operstate 9768 * @rootdev: the root or lower level device to transfer state from 9769 * @dev: the device to transfer operstate to 9770 * 9771 * Transfer operational state from root to device. This is normally 9772 * called when a stacking relationship exists between the root 9773 * device and the device(a leaf device). 9774 */ 9775 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9776 struct net_device *dev) 9777 { 9778 if (rootdev->operstate == IF_OPER_DORMANT) 9779 netif_dormant_on(dev); 9780 else 9781 netif_dormant_off(dev); 9782 9783 if (rootdev->operstate == IF_OPER_TESTING) 9784 netif_testing_on(dev); 9785 else 9786 netif_testing_off(dev); 9787 9788 if (netif_carrier_ok(rootdev)) 9789 netif_carrier_on(dev); 9790 else 9791 netif_carrier_off(dev); 9792 } 9793 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9794 9795 static int netif_alloc_rx_queues(struct net_device *dev) 9796 { 9797 unsigned int i, count = dev->num_rx_queues; 9798 struct netdev_rx_queue *rx; 9799 size_t sz = count * sizeof(*rx); 9800 int err = 0; 9801 9802 BUG_ON(count < 1); 9803 9804 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9805 if (!rx) 9806 return -ENOMEM; 9807 9808 dev->_rx = rx; 9809 9810 for (i = 0; i < count; i++) { 9811 rx[i].dev = dev; 9812 9813 /* XDP RX-queue setup */ 9814 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9815 if (err < 0) 9816 goto err_rxq_info; 9817 } 9818 return 0; 9819 9820 err_rxq_info: 9821 /* Rollback successful reg's and free other resources */ 9822 while (i--) 9823 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9824 kvfree(dev->_rx); 9825 dev->_rx = NULL; 9826 return err; 9827 } 9828 9829 static void netif_free_rx_queues(struct net_device *dev) 9830 { 9831 unsigned int i, count = dev->num_rx_queues; 9832 9833 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9834 if (!dev->_rx) 9835 return; 9836 9837 for (i = 0; i < count; i++) 9838 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9839 9840 kvfree(dev->_rx); 9841 } 9842 9843 static void netdev_init_one_queue(struct net_device *dev, 9844 struct netdev_queue *queue, void *_unused) 9845 { 9846 /* Initialize queue lock */ 9847 spin_lock_init(&queue->_xmit_lock); 9848 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9849 queue->xmit_lock_owner = -1; 9850 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9851 queue->dev = dev; 9852 #ifdef CONFIG_BQL 9853 dql_init(&queue->dql, HZ); 9854 #endif 9855 } 9856 9857 static void netif_free_tx_queues(struct net_device *dev) 9858 { 9859 kvfree(dev->_tx); 9860 } 9861 9862 static int netif_alloc_netdev_queues(struct net_device *dev) 9863 { 9864 unsigned int count = dev->num_tx_queues; 9865 struct netdev_queue *tx; 9866 size_t sz = count * sizeof(*tx); 9867 9868 if (count < 1 || count > 0xffff) 9869 return -EINVAL; 9870 9871 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9872 if (!tx) 9873 return -ENOMEM; 9874 9875 dev->_tx = tx; 9876 9877 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9878 spin_lock_init(&dev->tx_global_lock); 9879 9880 return 0; 9881 } 9882 9883 void netif_tx_stop_all_queues(struct net_device *dev) 9884 { 9885 unsigned int i; 9886 9887 for (i = 0; i < dev->num_tx_queues; i++) { 9888 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9889 9890 netif_tx_stop_queue(txq); 9891 } 9892 } 9893 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9894 9895 /** 9896 * register_netdevice - register a network device 9897 * @dev: device to register 9898 * 9899 * Take a completed network device structure and add it to the kernel 9900 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9901 * chain. 0 is returned on success. A negative errno code is returned 9902 * on a failure to set up the device, or if the name is a duplicate. 9903 * 9904 * Callers must hold the rtnl semaphore. You may want 9905 * register_netdev() instead of this. 9906 * 9907 * BUGS: 9908 * The locking appears insufficient to guarantee two parallel registers 9909 * will not get the same name. 9910 */ 9911 9912 int register_netdevice(struct net_device *dev) 9913 { 9914 int ret; 9915 struct net *net = dev_net(dev); 9916 9917 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9918 NETDEV_FEATURE_COUNT); 9919 BUG_ON(dev_boot_phase); 9920 ASSERT_RTNL(); 9921 9922 might_sleep(); 9923 9924 /* When net_device's are persistent, this will be fatal. */ 9925 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9926 BUG_ON(!net); 9927 9928 ret = ethtool_check_ops(dev->ethtool_ops); 9929 if (ret) 9930 return ret; 9931 9932 spin_lock_init(&dev->addr_list_lock); 9933 netdev_set_addr_lockdep_class(dev); 9934 9935 ret = dev_get_valid_name(net, dev, dev->name); 9936 if (ret < 0) 9937 goto out; 9938 9939 ret = -ENOMEM; 9940 dev->name_node = netdev_name_node_head_alloc(dev); 9941 if (!dev->name_node) 9942 goto out; 9943 9944 /* Init, if this function is available */ 9945 if (dev->netdev_ops->ndo_init) { 9946 ret = dev->netdev_ops->ndo_init(dev); 9947 if (ret) { 9948 if (ret > 0) 9949 ret = -EIO; 9950 goto err_free_name; 9951 } 9952 } 9953 9954 if (((dev->hw_features | dev->features) & 9955 NETIF_F_HW_VLAN_CTAG_FILTER) && 9956 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9957 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9958 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9959 ret = -EINVAL; 9960 goto err_uninit; 9961 } 9962 9963 ret = -EBUSY; 9964 if (!dev->ifindex) 9965 dev->ifindex = dev_new_index(net); 9966 else if (__dev_get_by_index(net, dev->ifindex)) 9967 goto err_uninit; 9968 9969 /* Transfer changeable features to wanted_features and enable 9970 * software offloads (GSO and GRO). 9971 */ 9972 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 9973 dev->features |= NETIF_F_SOFT_FEATURES; 9974 9975 if (dev->udp_tunnel_nic_info) { 9976 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9977 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9978 } 9979 9980 dev->wanted_features = dev->features & dev->hw_features; 9981 9982 if (!(dev->flags & IFF_LOOPBACK)) 9983 dev->hw_features |= NETIF_F_NOCACHE_COPY; 9984 9985 /* If IPv4 TCP segmentation offload is supported we should also 9986 * allow the device to enable segmenting the frame with the option 9987 * of ignoring a static IP ID value. This doesn't enable the 9988 * feature itself but allows the user to enable it later. 9989 */ 9990 if (dev->hw_features & NETIF_F_TSO) 9991 dev->hw_features |= NETIF_F_TSO_MANGLEID; 9992 if (dev->vlan_features & NETIF_F_TSO) 9993 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 9994 if (dev->mpls_features & NETIF_F_TSO) 9995 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 9996 if (dev->hw_enc_features & NETIF_F_TSO) 9997 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 9998 9999 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10000 */ 10001 dev->vlan_features |= NETIF_F_HIGHDMA; 10002 10003 /* Make NETIF_F_SG inheritable to tunnel devices. 10004 */ 10005 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10006 10007 /* Make NETIF_F_SG inheritable to MPLS. 10008 */ 10009 dev->mpls_features |= NETIF_F_SG; 10010 10011 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10012 ret = notifier_to_errno(ret); 10013 if (ret) 10014 goto err_uninit; 10015 10016 ret = netdev_register_kobject(dev); 10017 if (ret) { 10018 dev->reg_state = NETREG_UNREGISTERED; 10019 goto err_uninit; 10020 } 10021 dev->reg_state = NETREG_REGISTERED; 10022 10023 __netdev_update_features(dev); 10024 10025 /* 10026 * Default initial state at registry is that the 10027 * device is present. 10028 */ 10029 10030 set_bit(__LINK_STATE_PRESENT, &dev->state); 10031 10032 linkwatch_init_dev(dev); 10033 10034 dev_init_scheduler(dev); 10035 10036 dev_hold_track(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10037 list_netdevice(dev); 10038 10039 add_device_randomness(dev->dev_addr, dev->addr_len); 10040 10041 /* If the device has permanent device address, driver should 10042 * set dev_addr and also addr_assign_type should be set to 10043 * NET_ADDR_PERM (default value). 10044 */ 10045 if (dev->addr_assign_type == NET_ADDR_PERM) 10046 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10047 10048 /* Notify protocols, that a new device appeared. */ 10049 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10050 ret = notifier_to_errno(ret); 10051 if (ret) { 10052 /* Expect explicit free_netdev() on failure */ 10053 dev->needs_free_netdev = false; 10054 unregister_netdevice_queue(dev, NULL); 10055 goto out; 10056 } 10057 /* 10058 * Prevent userspace races by waiting until the network 10059 * device is fully setup before sending notifications. 10060 */ 10061 if (!dev->rtnl_link_ops || 10062 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10063 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10064 10065 out: 10066 return ret; 10067 10068 err_uninit: 10069 if (dev->netdev_ops->ndo_uninit) 10070 dev->netdev_ops->ndo_uninit(dev); 10071 if (dev->priv_destructor) 10072 dev->priv_destructor(dev); 10073 err_free_name: 10074 netdev_name_node_free(dev->name_node); 10075 goto out; 10076 } 10077 EXPORT_SYMBOL(register_netdevice); 10078 10079 /** 10080 * init_dummy_netdev - init a dummy network device for NAPI 10081 * @dev: device to init 10082 * 10083 * This takes a network device structure and initialize the minimum 10084 * amount of fields so it can be used to schedule NAPI polls without 10085 * registering a full blown interface. This is to be used by drivers 10086 * that need to tie several hardware interfaces to a single NAPI 10087 * poll scheduler due to HW limitations. 10088 */ 10089 int init_dummy_netdev(struct net_device *dev) 10090 { 10091 /* Clear everything. Note we don't initialize spinlocks 10092 * are they aren't supposed to be taken by any of the 10093 * NAPI code and this dummy netdev is supposed to be 10094 * only ever used for NAPI polls 10095 */ 10096 memset(dev, 0, sizeof(struct net_device)); 10097 10098 /* make sure we BUG if trying to hit standard 10099 * register/unregister code path 10100 */ 10101 dev->reg_state = NETREG_DUMMY; 10102 10103 /* NAPI wants this */ 10104 INIT_LIST_HEAD(&dev->napi_list); 10105 10106 /* a dummy interface is started by default */ 10107 set_bit(__LINK_STATE_PRESENT, &dev->state); 10108 set_bit(__LINK_STATE_START, &dev->state); 10109 10110 /* napi_busy_loop stats accounting wants this */ 10111 dev_net_set(dev, &init_net); 10112 10113 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10114 * because users of this 'device' dont need to change 10115 * its refcount. 10116 */ 10117 10118 return 0; 10119 } 10120 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10121 10122 10123 /** 10124 * register_netdev - register a network device 10125 * @dev: device to register 10126 * 10127 * Take a completed network device structure and add it to the kernel 10128 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10129 * chain. 0 is returned on success. A negative errno code is returned 10130 * on a failure to set up the device, or if the name is a duplicate. 10131 * 10132 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10133 * and expands the device name if you passed a format string to 10134 * alloc_netdev. 10135 */ 10136 int register_netdev(struct net_device *dev) 10137 { 10138 int err; 10139 10140 if (rtnl_lock_killable()) 10141 return -EINTR; 10142 err = register_netdevice(dev); 10143 rtnl_unlock(); 10144 return err; 10145 } 10146 EXPORT_SYMBOL(register_netdev); 10147 10148 int netdev_refcnt_read(const struct net_device *dev) 10149 { 10150 #ifdef CONFIG_PCPU_DEV_REFCNT 10151 int i, refcnt = 0; 10152 10153 for_each_possible_cpu(i) 10154 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10155 return refcnt; 10156 #else 10157 return refcount_read(&dev->dev_refcnt); 10158 #endif 10159 } 10160 EXPORT_SYMBOL(netdev_refcnt_read); 10161 10162 int netdev_unregister_timeout_secs __read_mostly = 10; 10163 10164 #define WAIT_REFS_MIN_MSECS 1 10165 #define WAIT_REFS_MAX_MSECS 250 10166 /** 10167 * netdev_wait_allrefs_any - wait until all references are gone. 10168 * @list: list of net_devices to wait on 10169 * 10170 * This is called when unregistering network devices. 10171 * 10172 * Any protocol or device that holds a reference should register 10173 * for netdevice notification, and cleanup and put back the 10174 * reference if they receive an UNREGISTER event. 10175 * We can get stuck here if buggy protocols don't correctly 10176 * call dev_put. 10177 */ 10178 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10179 { 10180 unsigned long rebroadcast_time, warning_time; 10181 struct net_device *dev; 10182 int wait = 0; 10183 10184 rebroadcast_time = warning_time = jiffies; 10185 10186 list_for_each_entry(dev, list, todo_list) 10187 if (netdev_refcnt_read(dev) == 1) 10188 return dev; 10189 10190 while (true) { 10191 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10192 rtnl_lock(); 10193 10194 /* Rebroadcast unregister notification */ 10195 list_for_each_entry(dev, list, todo_list) 10196 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10197 10198 __rtnl_unlock(); 10199 rcu_barrier(); 10200 rtnl_lock(); 10201 10202 list_for_each_entry(dev, list, todo_list) 10203 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10204 &dev->state)) { 10205 /* We must not have linkwatch events 10206 * pending on unregister. If this 10207 * happens, we simply run the queue 10208 * unscheduled, resulting in a noop 10209 * for this device. 10210 */ 10211 linkwatch_run_queue(); 10212 break; 10213 } 10214 10215 __rtnl_unlock(); 10216 10217 rebroadcast_time = jiffies; 10218 } 10219 10220 if (!wait) { 10221 rcu_barrier(); 10222 wait = WAIT_REFS_MIN_MSECS; 10223 } else { 10224 msleep(wait); 10225 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10226 } 10227 10228 list_for_each_entry(dev, list, todo_list) 10229 if (netdev_refcnt_read(dev) == 1) 10230 return dev; 10231 10232 if (time_after(jiffies, warning_time + 10233 netdev_unregister_timeout_secs * HZ)) { 10234 list_for_each_entry(dev, list, todo_list) { 10235 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10236 dev->name, netdev_refcnt_read(dev)); 10237 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10238 } 10239 10240 warning_time = jiffies; 10241 } 10242 } 10243 } 10244 10245 /* The sequence is: 10246 * 10247 * rtnl_lock(); 10248 * ... 10249 * register_netdevice(x1); 10250 * register_netdevice(x2); 10251 * ... 10252 * unregister_netdevice(y1); 10253 * unregister_netdevice(y2); 10254 * ... 10255 * rtnl_unlock(); 10256 * free_netdev(y1); 10257 * free_netdev(y2); 10258 * 10259 * We are invoked by rtnl_unlock(). 10260 * This allows us to deal with problems: 10261 * 1) We can delete sysfs objects which invoke hotplug 10262 * without deadlocking with linkwatch via keventd. 10263 * 2) Since we run with the RTNL semaphore not held, we can sleep 10264 * safely in order to wait for the netdev refcnt to drop to zero. 10265 * 10266 * We must not return until all unregister events added during 10267 * the interval the lock was held have been completed. 10268 */ 10269 void netdev_run_todo(void) 10270 { 10271 struct net_device *dev, *tmp; 10272 struct list_head list; 10273 #ifdef CONFIG_LOCKDEP 10274 struct list_head unlink_list; 10275 10276 list_replace_init(&net_unlink_list, &unlink_list); 10277 10278 while (!list_empty(&unlink_list)) { 10279 struct net_device *dev = list_first_entry(&unlink_list, 10280 struct net_device, 10281 unlink_list); 10282 list_del_init(&dev->unlink_list); 10283 dev->nested_level = dev->lower_level - 1; 10284 } 10285 #endif 10286 10287 /* Snapshot list, allow later requests */ 10288 list_replace_init(&net_todo_list, &list); 10289 10290 __rtnl_unlock(); 10291 10292 /* Wait for rcu callbacks to finish before next phase */ 10293 if (!list_empty(&list)) 10294 rcu_barrier(); 10295 10296 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10297 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10298 netdev_WARN(dev, "run_todo but not unregistering\n"); 10299 list_del(&dev->todo_list); 10300 continue; 10301 } 10302 10303 dev->reg_state = NETREG_UNREGISTERED; 10304 linkwatch_forget_dev(dev); 10305 } 10306 10307 while (!list_empty(&list)) { 10308 dev = netdev_wait_allrefs_any(&list); 10309 list_del(&dev->todo_list); 10310 10311 /* paranoia */ 10312 BUG_ON(netdev_refcnt_read(dev) != 1); 10313 BUG_ON(!list_empty(&dev->ptype_all)); 10314 BUG_ON(!list_empty(&dev->ptype_specific)); 10315 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10316 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10317 #if IS_ENABLED(CONFIG_DECNET) 10318 WARN_ON(dev->dn_ptr); 10319 #endif 10320 if (dev->priv_destructor) 10321 dev->priv_destructor(dev); 10322 if (dev->needs_free_netdev) 10323 free_netdev(dev); 10324 10325 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10326 wake_up(&netdev_unregistering_wq); 10327 10328 /* Free network device */ 10329 kobject_put(&dev->dev.kobj); 10330 } 10331 } 10332 10333 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10334 * all the same fields in the same order as net_device_stats, with only 10335 * the type differing, but rtnl_link_stats64 may have additional fields 10336 * at the end for newer counters. 10337 */ 10338 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10339 const struct net_device_stats *netdev_stats) 10340 { 10341 #if BITS_PER_LONG == 64 10342 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10343 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10344 /* zero out counters that only exist in rtnl_link_stats64 */ 10345 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10346 sizeof(*stats64) - sizeof(*netdev_stats)); 10347 #else 10348 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10349 const unsigned long *src = (const unsigned long *)netdev_stats; 10350 u64 *dst = (u64 *)stats64; 10351 10352 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10353 for (i = 0; i < n; i++) 10354 dst[i] = src[i]; 10355 /* zero out counters that only exist in rtnl_link_stats64 */ 10356 memset((char *)stats64 + n * sizeof(u64), 0, 10357 sizeof(*stats64) - n * sizeof(u64)); 10358 #endif 10359 } 10360 EXPORT_SYMBOL(netdev_stats_to_stats64); 10361 10362 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10363 { 10364 struct net_device_core_stats __percpu *p; 10365 10366 p = alloc_percpu_gfp(struct net_device_core_stats, 10367 GFP_ATOMIC | __GFP_NOWARN); 10368 10369 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10370 free_percpu(p); 10371 10372 /* This READ_ONCE() pairs with the cmpxchg() above */ 10373 return READ_ONCE(dev->core_stats); 10374 } 10375 EXPORT_SYMBOL(netdev_core_stats_alloc); 10376 10377 /** 10378 * dev_get_stats - get network device statistics 10379 * @dev: device to get statistics from 10380 * @storage: place to store stats 10381 * 10382 * Get network statistics from device. Return @storage. 10383 * The device driver may provide its own method by setting 10384 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10385 * otherwise the internal statistics structure is used. 10386 */ 10387 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10388 struct rtnl_link_stats64 *storage) 10389 { 10390 const struct net_device_ops *ops = dev->netdev_ops; 10391 const struct net_device_core_stats __percpu *p; 10392 10393 if (ops->ndo_get_stats64) { 10394 memset(storage, 0, sizeof(*storage)); 10395 ops->ndo_get_stats64(dev, storage); 10396 } else if (ops->ndo_get_stats) { 10397 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10398 } else { 10399 netdev_stats_to_stats64(storage, &dev->stats); 10400 } 10401 10402 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10403 p = READ_ONCE(dev->core_stats); 10404 if (p) { 10405 const struct net_device_core_stats *core_stats; 10406 int i; 10407 10408 for_each_possible_cpu(i) { 10409 core_stats = per_cpu_ptr(p, i); 10410 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10411 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10412 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10413 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10414 } 10415 } 10416 return storage; 10417 } 10418 EXPORT_SYMBOL(dev_get_stats); 10419 10420 /** 10421 * dev_fetch_sw_netstats - get per-cpu network device statistics 10422 * @s: place to store stats 10423 * @netstats: per-cpu network stats to read from 10424 * 10425 * Read per-cpu network statistics and populate the related fields in @s. 10426 */ 10427 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10428 const struct pcpu_sw_netstats __percpu *netstats) 10429 { 10430 int cpu; 10431 10432 for_each_possible_cpu(cpu) { 10433 const struct pcpu_sw_netstats *stats; 10434 struct pcpu_sw_netstats tmp; 10435 unsigned int start; 10436 10437 stats = per_cpu_ptr(netstats, cpu); 10438 do { 10439 start = u64_stats_fetch_begin_irq(&stats->syncp); 10440 tmp.rx_packets = stats->rx_packets; 10441 tmp.rx_bytes = stats->rx_bytes; 10442 tmp.tx_packets = stats->tx_packets; 10443 tmp.tx_bytes = stats->tx_bytes; 10444 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10445 10446 s->rx_packets += tmp.rx_packets; 10447 s->rx_bytes += tmp.rx_bytes; 10448 s->tx_packets += tmp.tx_packets; 10449 s->tx_bytes += tmp.tx_bytes; 10450 } 10451 } 10452 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10453 10454 /** 10455 * dev_get_tstats64 - ndo_get_stats64 implementation 10456 * @dev: device to get statistics from 10457 * @s: place to store stats 10458 * 10459 * Populate @s from dev->stats and dev->tstats. Can be used as 10460 * ndo_get_stats64() callback. 10461 */ 10462 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10463 { 10464 netdev_stats_to_stats64(s, &dev->stats); 10465 dev_fetch_sw_netstats(s, dev->tstats); 10466 } 10467 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10468 10469 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10470 { 10471 struct netdev_queue *queue = dev_ingress_queue(dev); 10472 10473 #ifdef CONFIG_NET_CLS_ACT 10474 if (queue) 10475 return queue; 10476 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10477 if (!queue) 10478 return NULL; 10479 netdev_init_one_queue(dev, queue, NULL); 10480 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10481 queue->qdisc_sleeping = &noop_qdisc; 10482 rcu_assign_pointer(dev->ingress_queue, queue); 10483 #endif 10484 return queue; 10485 } 10486 10487 static const struct ethtool_ops default_ethtool_ops; 10488 10489 void netdev_set_default_ethtool_ops(struct net_device *dev, 10490 const struct ethtool_ops *ops) 10491 { 10492 if (dev->ethtool_ops == &default_ethtool_ops) 10493 dev->ethtool_ops = ops; 10494 } 10495 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10496 10497 void netdev_freemem(struct net_device *dev) 10498 { 10499 char *addr = (char *)dev - dev->padded; 10500 10501 kvfree(addr); 10502 } 10503 10504 /** 10505 * alloc_netdev_mqs - allocate network device 10506 * @sizeof_priv: size of private data to allocate space for 10507 * @name: device name format string 10508 * @name_assign_type: origin of device name 10509 * @setup: callback to initialize device 10510 * @txqs: the number of TX subqueues to allocate 10511 * @rxqs: the number of RX subqueues to allocate 10512 * 10513 * Allocates a struct net_device with private data area for driver use 10514 * and performs basic initialization. Also allocates subqueue structs 10515 * for each queue on the device. 10516 */ 10517 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10518 unsigned char name_assign_type, 10519 void (*setup)(struct net_device *), 10520 unsigned int txqs, unsigned int rxqs) 10521 { 10522 struct net_device *dev; 10523 unsigned int alloc_size; 10524 struct net_device *p; 10525 10526 BUG_ON(strlen(name) >= sizeof(dev->name)); 10527 10528 if (txqs < 1) { 10529 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10530 return NULL; 10531 } 10532 10533 if (rxqs < 1) { 10534 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10535 return NULL; 10536 } 10537 10538 alloc_size = sizeof(struct net_device); 10539 if (sizeof_priv) { 10540 /* ensure 32-byte alignment of private area */ 10541 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10542 alloc_size += sizeof_priv; 10543 } 10544 /* ensure 32-byte alignment of whole construct */ 10545 alloc_size += NETDEV_ALIGN - 1; 10546 10547 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10548 if (!p) 10549 return NULL; 10550 10551 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10552 dev->padded = (char *)dev - (char *)p; 10553 10554 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10555 #ifdef CONFIG_PCPU_DEV_REFCNT 10556 dev->pcpu_refcnt = alloc_percpu(int); 10557 if (!dev->pcpu_refcnt) 10558 goto free_dev; 10559 __dev_hold(dev); 10560 #else 10561 refcount_set(&dev->dev_refcnt, 1); 10562 #endif 10563 10564 if (dev_addr_init(dev)) 10565 goto free_pcpu; 10566 10567 dev_mc_init(dev); 10568 dev_uc_init(dev); 10569 10570 dev_net_set(dev, &init_net); 10571 10572 dev->gso_max_size = GSO_MAX_SIZE; 10573 dev->gso_max_segs = GSO_MAX_SEGS; 10574 dev->gro_max_size = GRO_MAX_SIZE; 10575 dev->upper_level = 1; 10576 dev->lower_level = 1; 10577 #ifdef CONFIG_LOCKDEP 10578 dev->nested_level = 0; 10579 INIT_LIST_HEAD(&dev->unlink_list); 10580 #endif 10581 10582 INIT_LIST_HEAD(&dev->napi_list); 10583 INIT_LIST_HEAD(&dev->unreg_list); 10584 INIT_LIST_HEAD(&dev->close_list); 10585 INIT_LIST_HEAD(&dev->link_watch_list); 10586 INIT_LIST_HEAD(&dev->adj_list.upper); 10587 INIT_LIST_HEAD(&dev->adj_list.lower); 10588 INIT_LIST_HEAD(&dev->ptype_all); 10589 INIT_LIST_HEAD(&dev->ptype_specific); 10590 INIT_LIST_HEAD(&dev->net_notifier_list); 10591 #ifdef CONFIG_NET_SCHED 10592 hash_init(dev->qdisc_hash); 10593 #endif 10594 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10595 setup(dev); 10596 10597 if (!dev->tx_queue_len) { 10598 dev->priv_flags |= IFF_NO_QUEUE; 10599 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10600 } 10601 10602 dev->num_tx_queues = txqs; 10603 dev->real_num_tx_queues = txqs; 10604 if (netif_alloc_netdev_queues(dev)) 10605 goto free_all; 10606 10607 dev->num_rx_queues = rxqs; 10608 dev->real_num_rx_queues = rxqs; 10609 if (netif_alloc_rx_queues(dev)) 10610 goto free_all; 10611 10612 strcpy(dev->name, name); 10613 dev->name_assign_type = name_assign_type; 10614 dev->group = INIT_NETDEV_GROUP; 10615 if (!dev->ethtool_ops) 10616 dev->ethtool_ops = &default_ethtool_ops; 10617 10618 nf_hook_netdev_init(dev); 10619 10620 return dev; 10621 10622 free_all: 10623 free_netdev(dev); 10624 return NULL; 10625 10626 free_pcpu: 10627 #ifdef CONFIG_PCPU_DEV_REFCNT 10628 free_percpu(dev->pcpu_refcnt); 10629 free_dev: 10630 #endif 10631 netdev_freemem(dev); 10632 return NULL; 10633 } 10634 EXPORT_SYMBOL(alloc_netdev_mqs); 10635 10636 /** 10637 * free_netdev - free network device 10638 * @dev: device 10639 * 10640 * This function does the last stage of destroying an allocated device 10641 * interface. The reference to the device object is released. If this 10642 * is the last reference then it will be freed.Must be called in process 10643 * context. 10644 */ 10645 void free_netdev(struct net_device *dev) 10646 { 10647 struct napi_struct *p, *n; 10648 10649 might_sleep(); 10650 10651 /* When called immediately after register_netdevice() failed the unwind 10652 * handling may still be dismantling the device. Handle that case by 10653 * deferring the free. 10654 */ 10655 if (dev->reg_state == NETREG_UNREGISTERING) { 10656 ASSERT_RTNL(); 10657 dev->needs_free_netdev = true; 10658 return; 10659 } 10660 10661 netif_free_tx_queues(dev); 10662 netif_free_rx_queues(dev); 10663 10664 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10665 10666 /* Flush device addresses */ 10667 dev_addr_flush(dev); 10668 10669 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10670 netif_napi_del(p); 10671 10672 ref_tracker_dir_exit(&dev->refcnt_tracker); 10673 #ifdef CONFIG_PCPU_DEV_REFCNT 10674 free_percpu(dev->pcpu_refcnt); 10675 dev->pcpu_refcnt = NULL; 10676 #endif 10677 free_percpu(dev->core_stats); 10678 dev->core_stats = NULL; 10679 free_percpu(dev->xdp_bulkq); 10680 dev->xdp_bulkq = NULL; 10681 10682 /* Compatibility with error handling in drivers */ 10683 if (dev->reg_state == NETREG_UNINITIALIZED) { 10684 netdev_freemem(dev); 10685 return; 10686 } 10687 10688 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10689 dev->reg_state = NETREG_RELEASED; 10690 10691 /* will free via device release */ 10692 put_device(&dev->dev); 10693 } 10694 EXPORT_SYMBOL(free_netdev); 10695 10696 /** 10697 * synchronize_net - Synchronize with packet receive processing 10698 * 10699 * Wait for packets currently being received to be done. 10700 * Does not block later packets from starting. 10701 */ 10702 void synchronize_net(void) 10703 { 10704 might_sleep(); 10705 if (rtnl_is_locked()) 10706 synchronize_rcu_expedited(); 10707 else 10708 synchronize_rcu(); 10709 } 10710 EXPORT_SYMBOL(synchronize_net); 10711 10712 /** 10713 * unregister_netdevice_queue - remove device from the kernel 10714 * @dev: device 10715 * @head: list 10716 * 10717 * This function shuts down a device interface and removes it 10718 * from the kernel tables. 10719 * If head not NULL, device is queued to be unregistered later. 10720 * 10721 * Callers must hold the rtnl semaphore. You may want 10722 * unregister_netdev() instead of this. 10723 */ 10724 10725 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10726 { 10727 ASSERT_RTNL(); 10728 10729 if (head) { 10730 list_move_tail(&dev->unreg_list, head); 10731 } else { 10732 LIST_HEAD(single); 10733 10734 list_add(&dev->unreg_list, &single); 10735 unregister_netdevice_many(&single); 10736 } 10737 } 10738 EXPORT_SYMBOL(unregister_netdevice_queue); 10739 10740 /** 10741 * unregister_netdevice_many - unregister many devices 10742 * @head: list of devices 10743 * 10744 * Note: As most callers use a stack allocated list_head, 10745 * we force a list_del() to make sure stack wont be corrupted later. 10746 */ 10747 void unregister_netdevice_many(struct list_head *head) 10748 { 10749 struct net_device *dev, *tmp; 10750 LIST_HEAD(close_head); 10751 10752 BUG_ON(dev_boot_phase); 10753 ASSERT_RTNL(); 10754 10755 if (list_empty(head)) 10756 return; 10757 10758 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10759 /* Some devices call without registering 10760 * for initialization unwind. Remove those 10761 * devices and proceed with the remaining. 10762 */ 10763 if (dev->reg_state == NETREG_UNINITIALIZED) { 10764 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10765 dev->name, dev); 10766 10767 WARN_ON(1); 10768 list_del(&dev->unreg_list); 10769 continue; 10770 } 10771 dev->dismantle = true; 10772 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10773 } 10774 10775 /* If device is running, close it first. */ 10776 list_for_each_entry(dev, head, unreg_list) 10777 list_add_tail(&dev->close_list, &close_head); 10778 dev_close_many(&close_head, true); 10779 10780 list_for_each_entry(dev, head, unreg_list) { 10781 /* And unlink it from device chain. */ 10782 unlist_netdevice(dev); 10783 10784 dev->reg_state = NETREG_UNREGISTERING; 10785 } 10786 flush_all_backlogs(); 10787 10788 synchronize_net(); 10789 10790 list_for_each_entry(dev, head, unreg_list) { 10791 struct sk_buff *skb = NULL; 10792 10793 /* Shutdown queueing discipline. */ 10794 dev_shutdown(dev); 10795 10796 dev_xdp_uninstall(dev); 10797 10798 netdev_offload_xstats_disable_all(dev); 10799 10800 /* Notify protocols, that we are about to destroy 10801 * this device. They should clean all the things. 10802 */ 10803 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10804 10805 if (!dev->rtnl_link_ops || 10806 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10807 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10808 GFP_KERNEL, NULL, 0); 10809 10810 /* 10811 * Flush the unicast and multicast chains 10812 */ 10813 dev_uc_flush(dev); 10814 dev_mc_flush(dev); 10815 10816 netdev_name_node_alt_flush(dev); 10817 netdev_name_node_free(dev->name_node); 10818 10819 if (dev->netdev_ops->ndo_uninit) 10820 dev->netdev_ops->ndo_uninit(dev); 10821 10822 if (skb) 10823 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10824 10825 /* Notifier chain MUST detach us all upper devices. */ 10826 WARN_ON(netdev_has_any_upper_dev(dev)); 10827 WARN_ON(netdev_has_any_lower_dev(dev)); 10828 10829 /* Remove entries from kobject tree */ 10830 netdev_unregister_kobject(dev); 10831 #ifdef CONFIG_XPS 10832 /* Remove XPS queueing entries */ 10833 netif_reset_xps_queues_gt(dev, 0); 10834 #endif 10835 } 10836 10837 synchronize_net(); 10838 10839 list_for_each_entry(dev, head, unreg_list) { 10840 dev_put_track(dev, &dev->dev_registered_tracker); 10841 net_set_todo(dev); 10842 } 10843 10844 list_del(head); 10845 } 10846 EXPORT_SYMBOL(unregister_netdevice_many); 10847 10848 /** 10849 * unregister_netdev - remove device from the kernel 10850 * @dev: device 10851 * 10852 * This function shuts down a device interface and removes it 10853 * from the kernel tables. 10854 * 10855 * This is just a wrapper for unregister_netdevice that takes 10856 * the rtnl semaphore. In general you want to use this and not 10857 * unregister_netdevice. 10858 */ 10859 void unregister_netdev(struct net_device *dev) 10860 { 10861 rtnl_lock(); 10862 unregister_netdevice(dev); 10863 rtnl_unlock(); 10864 } 10865 EXPORT_SYMBOL(unregister_netdev); 10866 10867 /** 10868 * __dev_change_net_namespace - move device to different nethost namespace 10869 * @dev: device 10870 * @net: network namespace 10871 * @pat: If not NULL name pattern to try if the current device name 10872 * is already taken in the destination network namespace. 10873 * @new_ifindex: If not zero, specifies device index in the target 10874 * namespace. 10875 * 10876 * This function shuts down a device interface and moves it 10877 * to a new network namespace. On success 0 is returned, on 10878 * a failure a netagive errno code is returned. 10879 * 10880 * Callers must hold the rtnl semaphore. 10881 */ 10882 10883 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10884 const char *pat, int new_ifindex) 10885 { 10886 struct net *net_old = dev_net(dev); 10887 int err, new_nsid; 10888 10889 ASSERT_RTNL(); 10890 10891 /* Don't allow namespace local devices to be moved. */ 10892 err = -EINVAL; 10893 if (dev->features & NETIF_F_NETNS_LOCAL) 10894 goto out; 10895 10896 /* Ensure the device has been registrered */ 10897 if (dev->reg_state != NETREG_REGISTERED) 10898 goto out; 10899 10900 /* Get out if there is nothing todo */ 10901 err = 0; 10902 if (net_eq(net_old, net)) 10903 goto out; 10904 10905 /* Pick the destination device name, and ensure 10906 * we can use it in the destination network namespace. 10907 */ 10908 err = -EEXIST; 10909 if (netdev_name_in_use(net, dev->name)) { 10910 /* We get here if we can't use the current device name */ 10911 if (!pat) 10912 goto out; 10913 err = dev_get_valid_name(net, dev, pat); 10914 if (err < 0) 10915 goto out; 10916 } 10917 10918 /* Check that new_ifindex isn't used yet. */ 10919 err = -EBUSY; 10920 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10921 goto out; 10922 10923 /* 10924 * And now a mini version of register_netdevice unregister_netdevice. 10925 */ 10926 10927 /* If device is running close it first. */ 10928 dev_close(dev); 10929 10930 /* And unlink it from device chain */ 10931 unlist_netdevice(dev); 10932 10933 synchronize_net(); 10934 10935 /* Shutdown queueing discipline. */ 10936 dev_shutdown(dev); 10937 10938 /* Notify protocols, that we are about to destroy 10939 * this device. They should clean all the things. 10940 * 10941 * Note that dev->reg_state stays at NETREG_REGISTERED. 10942 * This is wanted because this way 8021q and macvlan know 10943 * the device is just moving and can keep their slaves up. 10944 */ 10945 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10946 rcu_barrier(); 10947 10948 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10949 /* If there is an ifindex conflict assign a new one */ 10950 if (!new_ifindex) { 10951 if (__dev_get_by_index(net, dev->ifindex)) 10952 new_ifindex = dev_new_index(net); 10953 else 10954 new_ifindex = dev->ifindex; 10955 } 10956 10957 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10958 new_ifindex); 10959 10960 /* 10961 * Flush the unicast and multicast chains 10962 */ 10963 dev_uc_flush(dev); 10964 dev_mc_flush(dev); 10965 10966 /* Send a netdev-removed uevent to the old namespace */ 10967 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10968 netdev_adjacent_del_links(dev); 10969 10970 /* Move per-net netdevice notifiers that are following the netdevice */ 10971 move_netdevice_notifiers_dev_net(dev, net); 10972 10973 /* Actually switch the network namespace */ 10974 dev_net_set(dev, net); 10975 dev->ifindex = new_ifindex; 10976 10977 /* Send a netdev-add uevent to the new namespace */ 10978 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10979 netdev_adjacent_add_links(dev); 10980 10981 /* Fixup kobjects */ 10982 err = device_rename(&dev->dev, dev->name); 10983 WARN_ON(err); 10984 10985 /* Adapt owner in case owning user namespace of target network 10986 * namespace is different from the original one. 10987 */ 10988 err = netdev_change_owner(dev, net_old, net); 10989 WARN_ON(err); 10990 10991 /* Add the device back in the hashes */ 10992 list_netdevice(dev); 10993 10994 /* Notify protocols, that a new device appeared. */ 10995 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10996 10997 /* 10998 * Prevent userspace races by waiting until the network 10999 * device is fully setup before sending notifications. 11000 */ 11001 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11002 11003 synchronize_net(); 11004 err = 0; 11005 out: 11006 return err; 11007 } 11008 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11009 11010 static int dev_cpu_dead(unsigned int oldcpu) 11011 { 11012 struct sk_buff **list_skb; 11013 struct sk_buff *skb; 11014 unsigned int cpu; 11015 struct softnet_data *sd, *oldsd, *remsd = NULL; 11016 11017 local_irq_disable(); 11018 cpu = smp_processor_id(); 11019 sd = &per_cpu(softnet_data, cpu); 11020 oldsd = &per_cpu(softnet_data, oldcpu); 11021 11022 /* Find end of our completion_queue. */ 11023 list_skb = &sd->completion_queue; 11024 while (*list_skb) 11025 list_skb = &(*list_skb)->next; 11026 /* Append completion queue from offline CPU. */ 11027 *list_skb = oldsd->completion_queue; 11028 oldsd->completion_queue = NULL; 11029 11030 /* Append output queue from offline CPU. */ 11031 if (oldsd->output_queue) { 11032 *sd->output_queue_tailp = oldsd->output_queue; 11033 sd->output_queue_tailp = oldsd->output_queue_tailp; 11034 oldsd->output_queue = NULL; 11035 oldsd->output_queue_tailp = &oldsd->output_queue; 11036 } 11037 /* Append NAPI poll list from offline CPU, with one exception : 11038 * process_backlog() must be called by cpu owning percpu backlog. 11039 * We properly handle process_queue & input_pkt_queue later. 11040 */ 11041 while (!list_empty(&oldsd->poll_list)) { 11042 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11043 struct napi_struct, 11044 poll_list); 11045 11046 list_del_init(&napi->poll_list); 11047 if (napi->poll == process_backlog) 11048 napi->state = 0; 11049 else 11050 ____napi_schedule(sd, napi); 11051 } 11052 11053 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11054 local_irq_enable(); 11055 11056 #ifdef CONFIG_RPS 11057 remsd = oldsd->rps_ipi_list; 11058 oldsd->rps_ipi_list = NULL; 11059 #endif 11060 /* send out pending IPI's on offline CPU */ 11061 net_rps_send_ipi(remsd); 11062 11063 /* Process offline CPU's input_pkt_queue */ 11064 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11065 netif_rx(skb); 11066 input_queue_head_incr(oldsd); 11067 } 11068 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11069 netif_rx(skb); 11070 input_queue_head_incr(oldsd); 11071 } 11072 11073 return 0; 11074 } 11075 11076 /** 11077 * netdev_increment_features - increment feature set by one 11078 * @all: current feature set 11079 * @one: new feature set 11080 * @mask: mask feature set 11081 * 11082 * Computes a new feature set after adding a device with feature set 11083 * @one to the master device with current feature set @all. Will not 11084 * enable anything that is off in @mask. Returns the new feature set. 11085 */ 11086 netdev_features_t netdev_increment_features(netdev_features_t all, 11087 netdev_features_t one, netdev_features_t mask) 11088 { 11089 if (mask & NETIF_F_HW_CSUM) 11090 mask |= NETIF_F_CSUM_MASK; 11091 mask |= NETIF_F_VLAN_CHALLENGED; 11092 11093 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11094 all &= one | ~NETIF_F_ALL_FOR_ALL; 11095 11096 /* If one device supports hw checksumming, set for all. */ 11097 if (all & NETIF_F_HW_CSUM) 11098 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11099 11100 return all; 11101 } 11102 EXPORT_SYMBOL(netdev_increment_features); 11103 11104 static struct hlist_head * __net_init netdev_create_hash(void) 11105 { 11106 int i; 11107 struct hlist_head *hash; 11108 11109 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11110 if (hash != NULL) 11111 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11112 INIT_HLIST_HEAD(&hash[i]); 11113 11114 return hash; 11115 } 11116 11117 /* Initialize per network namespace state */ 11118 static int __net_init netdev_init(struct net *net) 11119 { 11120 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11121 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11122 11123 INIT_LIST_HEAD(&net->dev_base_head); 11124 11125 net->dev_name_head = netdev_create_hash(); 11126 if (net->dev_name_head == NULL) 11127 goto err_name; 11128 11129 net->dev_index_head = netdev_create_hash(); 11130 if (net->dev_index_head == NULL) 11131 goto err_idx; 11132 11133 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11134 11135 return 0; 11136 11137 err_idx: 11138 kfree(net->dev_name_head); 11139 err_name: 11140 return -ENOMEM; 11141 } 11142 11143 /** 11144 * netdev_drivername - network driver for the device 11145 * @dev: network device 11146 * 11147 * Determine network driver for device. 11148 */ 11149 const char *netdev_drivername(const struct net_device *dev) 11150 { 11151 const struct device_driver *driver; 11152 const struct device *parent; 11153 const char *empty = ""; 11154 11155 parent = dev->dev.parent; 11156 if (!parent) 11157 return empty; 11158 11159 driver = parent->driver; 11160 if (driver && driver->name) 11161 return driver->name; 11162 return empty; 11163 } 11164 11165 static void __netdev_printk(const char *level, const struct net_device *dev, 11166 struct va_format *vaf) 11167 { 11168 if (dev && dev->dev.parent) { 11169 dev_printk_emit(level[1] - '0', 11170 dev->dev.parent, 11171 "%s %s %s%s: %pV", 11172 dev_driver_string(dev->dev.parent), 11173 dev_name(dev->dev.parent), 11174 netdev_name(dev), netdev_reg_state(dev), 11175 vaf); 11176 } else if (dev) { 11177 printk("%s%s%s: %pV", 11178 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11179 } else { 11180 printk("%s(NULL net_device): %pV", level, vaf); 11181 } 11182 } 11183 11184 void netdev_printk(const char *level, const struct net_device *dev, 11185 const char *format, ...) 11186 { 11187 struct va_format vaf; 11188 va_list args; 11189 11190 va_start(args, format); 11191 11192 vaf.fmt = format; 11193 vaf.va = &args; 11194 11195 __netdev_printk(level, dev, &vaf); 11196 11197 va_end(args); 11198 } 11199 EXPORT_SYMBOL(netdev_printk); 11200 11201 #define define_netdev_printk_level(func, level) \ 11202 void func(const struct net_device *dev, const char *fmt, ...) \ 11203 { \ 11204 struct va_format vaf; \ 11205 va_list args; \ 11206 \ 11207 va_start(args, fmt); \ 11208 \ 11209 vaf.fmt = fmt; \ 11210 vaf.va = &args; \ 11211 \ 11212 __netdev_printk(level, dev, &vaf); \ 11213 \ 11214 va_end(args); \ 11215 } \ 11216 EXPORT_SYMBOL(func); 11217 11218 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11219 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11220 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11221 define_netdev_printk_level(netdev_err, KERN_ERR); 11222 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11223 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11224 define_netdev_printk_level(netdev_info, KERN_INFO); 11225 11226 static void __net_exit netdev_exit(struct net *net) 11227 { 11228 kfree(net->dev_name_head); 11229 kfree(net->dev_index_head); 11230 if (net != &init_net) 11231 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11232 } 11233 11234 static struct pernet_operations __net_initdata netdev_net_ops = { 11235 .init = netdev_init, 11236 .exit = netdev_exit, 11237 }; 11238 11239 static void __net_exit default_device_exit_net(struct net *net) 11240 { 11241 struct net_device *dev, *aux; 11242 /* 11243 * Push all migratable network devices back to the 11244 * initial network namespace 11245 */ 11246 ASSERT_RTNL(); 11247 for_each_netdev_safe(net, dev, aux) { 11248 int err; 11249 char fb_name[IFNAMSIZ]; 11250 11251 /* Ignore unmoveable devices (i.e. loopback) */ 11252 if (dev->features & NETIF_F_NETNS_LOCAL) 11253 continue; 11254 11255 /* Leave virtual devices for the generic cleanup */ 11256 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11257 continue; 11258 11259 /* Push remaining network devices to init_net */ 11260 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11261 if (netdev_name_in_use(&init_net, fb_name)) 11262 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11263 err = dev_change_net_namespace(dev, &init_net, fb_name); 11264 if (err) { 11265 pr_emerg("%s: failed to move %s to init_net: %d\n", 11266 __func__, dev->name, err); 11267 BUG(); 11268 } 11269 } 11270 } 11271 11272 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11273 { 11274 /* At exit all network devices most be removed from a network 11275 * namespace. Do this in the reverse order of registration. 11276 * Do this across as many network namespaces as possible to 11277 * improve batching efficiency. 11278 */ 11279 struct net_device *dev; 11280 struct net *net; 11281 LIST_HEAD(dev_kill_list); 11282 11283 rtnl_lock(); 11284 list_for_each_entry(net, net_list, exit_list) { 11285 default_device_exit_net(net); 11286 cond_resched(); 11287 } 11288 11289 list_for_each_entry(net, net_list, exit_list) { 11290 for_each_netdev_reverse(net, dev) { 11291 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11292 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11293 else 11294 unregister_netdevice_queue(dev, &dev_kill_list); 11295 } 11296 } 11297 unregister_netdevice_many(&dev_kill_list); 11298 rtnl_unlock(); 11299 } 11300 11301 static struct pernet_operations __net_initdata default_device_ops = { 11302 .exit_batch = default_device_exit_batch, 11303 }; 11304 11305 /* 11306 * Initialize the DEV module. At boot time this walks the device list and 11307 * unhooks any devices that fail to initialise (normally hardware not 11308 * present) and leaves us with a valid list of present and active devices. 11309 * 11310 */ 11311 11312 /* 11313 * This is called single threaded during boot, so no need 11314 * to take the rtnl semaphore. 11315 */ 11316 static int __init net_dev_init(void) 11317 { 11318 int i, rc = -ENOMEM; 11319 11320 BUG_ON(!dev_boot_phase); 11321 11322 if (dev_proc_init()) 11323 goto out; 11324 11325 if (netdev_kobject_init()) 11326 goto out; 11327 11328 INIT_LIST_HEAD(&ptype_all); 11329 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11330 INIT_LIST_HEAD(&ptype_base[i]); 11331 11332 if (register_pernet_subsys(&netdev_net_ops)) 11333 goto out; 11334 11335 /* 11336 * Initialise the packet receive queues. 11337 */ 11338 11339 for_each_possible_cpu(i) { 11340 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11341 struct softnet_data *sd = &per_cpu(softnet_data, i); 11342 11343 INIT_WORK(flush, flush_backlog); 11344 11345 skb_queue_head_init(&sd->input_pkt_queue); 11346 skb_queue_head_init(&sd->process_queue); 11347 #ifdef CONFIG_XFRM_OFFLOAD 11348 skb_queue_head_init(&sd->xfrm_backlog); 11349 #endif 11350 INIT_LIST_HEAD(&sd->poll_list); 11351 sd->output_queue_tailp = &sd->output_queue; 11352 #ifdef CONFIG_RPS 11353 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11354 sd->cpu = i; 11355 #endif 11356 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, NULL); 11357 spin_lock_init(&sd->defer_lock); 11358 11359 init_gro_hash(&sd->backlog); 11360 sd->backlog.poll = process_backlog; 11361 sd->backlog.weight = weight_p; 11362 } 11363 11364 dev_boot_phase = 0; 11365 11366 /* The loopback device is special if any other network devices 11367 * is present in a network namespace the loopback device must 11368 * be present. Since we now dynamically allocate and free the 11369 * loopback device ensure this invariant is maintained by 11370 * keeping the loopback device as the first device on the 11371 * list of network devices. Ensuring the loopback devices 11372 * is the first device that appears and the last network device 11373 * that disappears. 11374 */ 11375 if (register_pernet_device(&loopback_net_ops)) 11376 goto out; 11377 11378 if (register_pernet_device(&default_device_ops)) 11379 goto out; 11380 11381 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11382 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11383 11384 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11385 NULL, dev_cpu_dead); 11386 WARN_ON(rc < 0); 11387 rc = 0; 11388 out: 11389 return rc; 11390 } 11391 11392 subsys_initcall(net_dev_init); 11393