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_set_tso_max_size() - set the max size of TSO frames supported 2997 * @dev: netdev to update 2998 * @size: max skb->len of a TSO frame 2999 * 3000 * Set the limit on the size of TSO super-frames the device can handle. 3001 * Unless explicitly set the stack will assume the value of %GSO_MAX_SIZE. 3002 */ 3003 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3004 { 3005 dev->tso_max_size = size; 3006 if (size < READ_ONCE(dev->gso_max_size)) 3007 netif_set_gso_max_size(dev, size); 3008 } 3009 EXPORT_SYMBOL(netif_set_tso_max_size); 3010 3011 /** 3012 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3013 * @dev: netdev to update 3014 * @segs: max number of TCP segments 3015 * 3016 * Set the limit on the number of TCP segments the device can generate from 3017 * a single TSO super-frame. 3018 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3019 */ 3020 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3021 { 3022 dev->tso_max_segs = segs; 3023 if (segs < READ_ONCE(dev->gso_max_segs)) 3024 netif_set_gso_max_segs(dev, segs); 3025 } 3026 EXPORT_SYMBOL(netif_set_tso_max_segs); 3027 3028 /** 3029 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3030 * @to: netdev to update 3031 * @from: netdev from which to copy the limits 3032 */ 3033 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3034 { 3035 netif_set_gso_max_size(to, from->gso_max_size); 3036 netif_set_gso_max_segs(to, from->gso_max_segs); 3037 } 3038 EXPORT_SYMBOL(netif_inherit_tso_max); 3039 3040 /** 3041 * netif_get_num_default_rss_queues - default number of RSS queues 3042 * 3043 * Default value is the number of physical cores if there are only 1 or 2, or 3044 * divided by 2 if there are more. 3045 */ 3046 int netif_get_num_default_rss_queues(void) 3047 { 3048 cpumask_var_t cpus; 3049 int cpu, count = 0; 3050 3051 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3052 return 1; 3053 3054 cpumask_copy(cpus, cpu_online_mask); 3055 for_each_cpu(cpu, cpus) { 3056 ++count; 3057 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3058 } 3059 free_cpumask_var(cpus); 3060 3061 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3062 } 3063 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3064 3065 static void __netif_reschedule(struct Qdisc *q) 3066 { 3067 struct softnet_data *sd; 3068 unsigned long flags; 3069 3070 local_irq_save(flags); 3071 sd = this_cpu_ptr(&softnet_data); 3072 q->next_sched = NULL; 3073 *sd->output_queue_tailp = q; 3074 sd->output_queue_tailp = &q->next_sched; 3075 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3076 local_irq_restore(flags); 3077 } 3078 3079 void __netif_schedule(struct Qdisc *q) 3080 { 3081 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3082 __netif_reschedule(q); 3083 } 3084 EXPORT_SYMBOL(__netif_schedule); 3085 3086 struct dev_kfree_skb_cb { 3087 enum skb_free_reason reason; 3088 }; 3089 3090 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3091 { 3092 return (struct dev_kfree_skb_cb *)skb->cb; 3093 } 3094 3095 void netif_schedule_queue(struct netdev_queue *txq) 3096 { 3097 rcu_read_lock(); 3098 if (!netif_xmit_stopped(txq)) { 3099 struct Qdisc *q = rcu_dereference(txq->qdisc); 3100 3101 __netif_schedule(q); 3102 } 3103 rcu_read_unlock(); 3104 } 3105 EXPORT_SYMBOL(netif_schedule_queue); 3106 3107 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3108 { 3109 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3110 struct Qdisc *q; 3111 3112 rcu_read_lock(); 3113 q = rcu_dereference(dev_queue->qdisc); 3114 __netif_schedule(q); 3115 rcu_read_unlock(); 3116 } 3117 } 3118 EXPORT_SYMBOL(netif_tx_wake_queue); 3119 3120 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3121 { 3122 unsigned long flags; 3123 3124 if (unlikely(!skb)) 3125 return; 3126 3127 if (likely(refcount_read(&skb->users) == 1)) { 3128 smp_rmb(); 3129 refcount_set(&skb->users, 0); 3130 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3131 return; 3132 } 3133 get_kfree_skb_cb(skb)->reason = reason; 3134 local_irq_save(flags); 3135 skb->next = __this_cpu_read(softnet_data.completion_queue); 3136 __this_cpu_write(softnet_data.completion_queue, skb); 3137 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3138 local_irq_restore(flags); 3139 } 3140 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3141 3142 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3143 { 3144 if (in_hardirq() || irqs_disabled()) 3145 __dev_kfree_skb_irq(skb, reason); 3146 else 3147 dev_kfree_skb(skb); 3148 } 3149 EXPORT_SYMBOL(__dev_kfree_skb_any); 3150 3151 3152 /** 3153 * netif_device_detach - mark device as removed 3154 * @dev: network device 3155 * 3156 * Mark device as removed from system and therefore no longer available. 3157 */ 3158 void netif_device_detach(struct net_device *dev) 3159 { 3160 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3161 netif_running(dev)) { 3162 netif_tx_stop_all_queues(dev); 3163 } 3164 } 3165 EXPORT_SYMBOL(netif_device_detach); 3166 3167 /** 3168 * netif_device_attach - mark device as attached 3169 * @dev: network device 3170 * 3171 * Mark device as attached from system and restart if needed. 3172 */ 3173 void netif_device_attach(struct net_device *dev) 3174 { 3175 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3176 netif_running(dev)) { 3177 netif_tx_wake_all_queues(dev); 3178 __netdev_watchdog_up(dev); 3179 } 3180 } 3181 EXPORT_SYMBOL(netif_device_attach); 3182 3183 /* 3184 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3185 * to be used as a distribution range. 3186 */ 3187 static u16 skb_tx_hash(const struct net_device *dev, 3188 const struct net_device *sb_dev, 3189 struct sk_buff *skb) 3190 { 3191 u32 hash; 3192 u16 qoffset = 0; 3193 u16 qcount = dev->real_num_tx_queues; 3194 3195 if (dev->num_tc) { 3196 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3197 3198 qoffset = sb_dev->tc_to_txq[tc].offset; 3199 qcount = sb_dev->tc_to_txq[tc].count; 3200 if (unlikely(!qcount)) { 3201 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3202 sb_dev->name, qoffset, tc); 3203 qoffset = 0; 3204 qcount = dev->real_num_tx_queues; 3205 } 3206 } 3207 3208 if (skb_rx_queue_recorded(skb)) { 3209 hash = skb_get_rx_queue(skb); 3210 if (hash >= qoffset) 3211 hash -= qoffset; 3212 while (unlikely(hash >= qcount)) 3213 hash -= qcount; 3214 return hash + qoffset; 3215 } 3216 3217 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3218 } 3219 3220 static void skb_warn_bad_offload(const struct sk_buff *skb) 3221 { 3222 static const netdev_features_t null_features; 3223 struct net_device *dev = skb->dev; 3224 const char *name = ""; 3225 3226 if (!net_ratelimit()) 3227 return; 3228 3229 if (dev) { 3230 if (dev->dev.parent) 3231 name = dev_driver_string(dev->dev.parent); 3232 else 3233 name = netdev_name(dev); 3234 } 3235 skb_dump(KERN_WARNING, skb, false); 3236 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3237 name, dev ? &dev->features : &null_features, 3238 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3239 } 3240 3241 /* 3242 * Invalidate hardware checksum when packet is to be mangled, and 3243 * complete checksum manually on outgoing path. 3244 */ 3245 int skb_checksum_help(struct sk_buff *skb) 3246 { 3247 __wsum csum; 3248 int ret = 0, offset; 3249 3250 if (skb->ip_summed == CHECKSUM_COMPLETE) 3251 goto out_set_summed; 3252 3253 if (unlikely(skb_is_gso(skb))) { 3254 skb_warn_bad_offload(skb); 3255 return -EINVAL; 3256 } 3257 3258 /* Before computing a checksum, we should make sure no frag could 3259 * be modified by an external entity : checksum could be wrong. 3260 */ 3261 if (skb_has_shared_frag(skb)) { 3262 ret = __skb_linearize(skb); 3263 if (ret) 3264 goto out; 3265 } 3266 3267 offset = skb_checksum_start_offset(skb); 3268 BUG_ON(offset >= skb_headlen(skb)); 3269 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3270 3271 offset += skb->csum_offset; 3272 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3273 3274 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3275 if (ret) 3276 goto out; 3277 3278 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3279 out_set_summed: 3280 skb->ip_summed = CHECKSUM_NONE; 3281 out: 3282 return ret; 3283 } 3284 EXPORT_SYMBOL(skb_checksum_help); 3285 3286 int skb_crc32c_csum_help(struct sk_buff *skb) 3287 { 3288 __le32 crc32c_csum; 3289 int ret = 0, offset, start; 3290 3291 if (skb->ip_summed != CHECKSUM_PARTIAL) 3292 goto out; 3293 3294 if (unlikely(skb_is_gso(skb))) 3295 goto out; 3296 3297 /* Before computing a checksum, we should make sure no frag could 3298 * be modified by an external entity : checksum could be wrong. 3299 */ 3300 if (unlikely(skb_has_shared_frag(skb))) { 3301 ret = __skb_linearize(skb); 3302 if (ret) 3303 goto out; 3304 } 3305 start = skb_checksum_start_offset(skb); 3306 offset = start + offsetof(struct sctphdr, checksum); 3307 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3308 ret = -EINVAL; 3309 goto out; 3310 } 3311 3312 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3313 if (ret) 3314 goto out; 3315 3316 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3317 skb->len - start, ~(__u32)0, 3318 crc32c_csum_stub)); 3319 *(__le32 *)(skb->data + offset) = crc32c_csum; 3320 skb->ip_summed = CHECKSUM_NONE; 3321 skb->csum_not_inet = 0; 3322 out: 3323 return ret; 3324 } 3325 3326 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3327 { 3328 __be16 type = skb->protocol; 3329 3330 /* Tunnel gso handlers can set protocol to ethernet. */ 3331 if (type == htons(ETH_P_TEB)) { 3332 struct ethhdr *eth; 3333 3334 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3335 return 0; 3336 3337 eth = (struct ethhdr *)skb->data; 3338 type = eth->h_proto; 3339 } 3340 3341 return __vlan_get_protocol(skb, type, depth); 3342 } 3343 3344 /* openvswitch calls this on rx path, so we need a different check. 3345 */ 3346 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3347 { 3348 if (tx_path) 3349 return skb->ip_summed != CHECKSUM_PARTIAL && 3350 skb->ip_summed != CHECKSUM_UNNECESSARY; 3351 3352 return skb->ip_summed == CHECKSUM_NONE; 3353 } 3354 3355 /** 3356 * __skb_gso_segment - Perform segmentation on skb. 3357 * @skb: buffer to segment 3358 * @features: features for the output path (see dev->features) 3359 * @tx_path: whether it is called in TX path 3360 * 3361 * This function segments the given skb and returns a list of segments. 3362 * 3363 * It may return NULL if the skb requires no segmentation. This is 3364 * only possible when GSO is used for verifying header integrity. 3365 * 3366 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3367 */ 3368 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3369 netdev_features_t features, bool tx_path) 3370 { 3371 struct sk_buff *segs; 3372 3373 if (unlikely(skb_needs_check(skb, tx_path))) { 3374 int err; 3375 3376 /* We're going to init ->check field in TCP or UDP header */ 3377 err = skb_cow_head(skb, 0); 3378 if (err < 0) 3379 return ERR_PTR(err); 3380 } 3381 3382 /* Only report GSO partial support if it will enable us to 3383 * support segmentation on this frame without needing additional 3384 * work. 3385 */ 3386 if (features & NETIF_F_GSO_PARTIAL) { 3387 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3388 struct net_device *dev = skb->dev; 3389 3390 partial_features |= dev->features & dev->gso_partial_features; 3391 if (!skb_gso_ok(skb, features | partial_features)) 3392 features &= ~NETIF_F_GSO_PARTIAL; 3393 } 3394 3395 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3396 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3397 3398 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3399 SKB_GSO_CB(skb)->encap_level = 0; 3400 3401 skb_reset_mac_header(skb); 3402 skb_reset_mac_len(skb); 3403 3404 segs = skb_mac_gso_segment(skb, features); 3405 3406 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3407 skb_warn_bad_offload(skb); 3408 3409 return segs; 3410 } 3411 EXPORT_SYMBOL(__skb_gso_segment); 3412 3413 /* Take action when hardware reception checksum errors are detected. */ 3414 #ifdef CONFIG_BUG 3415 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3416 { 3417 netdev_err(dev, "hw csum failure\n"); 3418 skb_dump(KERN_ERR, skb, true); 3419 dump_stack(); 3420 } 3421 3422 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3423 { 3424 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3425 } 3426 EXPORT_SYMBOL(netdev_rx_csum_fault); 3427 #endif 3428 3429 /* XXX: check that highmem exists at all on the given machine. */ 3430 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3431 { 3432 #ifdef CONFIG_HIGHMEM 3433 int i; 3434 3435 if (!(dev->features & NETIF_F_HIGHDMA)) { 3436 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3437 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3438 3439 if (PageHighMem(skb_frag_page(frag))) 3440 return 1; 3441 } 3442 } 3443 #endif 3444 return 0; 3445 } 3446 3447 /* If MPLS offload request, verify we are testing hardware MPLS features 3448 * instead of standard features for the netdev. 3449 */ 3450 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3451 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3452 netdev_features_t features, 3453 __be16 type) 3454 { 3455 if (eth_p_mpls(type)) 3456 features &= skb->dev->mpls_features; 3457 3458 return features; 3459 } 3460 #else 3461 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3462 netdev_features_t features, 3463 __be16 type) 3464 { 3465 return features; 3466 } 3467 #endif 3468 3469 static netdev_features_t harmonize_features(struct sk_buff *skb, 3470 netdev_features_t features) 3471 { 3472 __be16 type; 3473 3474 type = skb_network_protocol(skb, NULL); 3475 features = net_mpls_features(skb, features, type); 3476 3477 if (skb->ip_summed != CHECKSUM_NONE && 3478 !can_checksum_protocol(features, type)) { 3479 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3480 } 3481 if (illegal_highdma(skb->dev, skb)) 3482 features &= ~NETIF_F_SG; 3483 3484 return features; 3485 } 3486 3487 netdev_features_t passthru_features_check(struct sk_buff *skb, 3488 struct net_device *dev, 3489 netdev_features_t features) 3490 { 3491 return features; 3492 } 3493 EXPORT_SYMBOL(passthru_features_check); 3494 3495 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3496 struct net_device *dev, 3497 netdev_features_t features) 3498 { 3499 return vlan_features_check(skb, features); 3500 } 3501 3502 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3503 struct net_device *dev, 3504 netdev_features_t features) 3505 { 3506 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3507 3508 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3509 return features & ~NETIF_F_GSO_MASK; 3510 3511 if (!skb_shinfo(skb)->gso_type) { 3512 skb_warn_bad_offload(skb); 3513 return features & ~NETIF_F_GSO_MASK; 3514 } 3515 3516 /* Support for GSO partial features requires software 3517 * intervention before we can actually process the packets 3518 * so we need to strip support for any partial features now 3519 * and we can pull them back in after we have partially 3520 * segmented the frame. 3521 */ 3522 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3523 features &= ~dev->gso_partial_features; 3524 3525 /* Make sure to clear the IPv4 ID mangling feature if the 3526 * IPv4 header has the potential to be fragmented. 3527 */ 3528 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3529 struct iphdr *iph = skb->encapsulation ? 3530 inner_ip_hdr(skb) : ip_hdr(skb); 3531 3532 if (!(iph->frag_off & htons(IP_DF))) 3533 features &= ~NETIF_F_TSO_MANGLEID; 3534 } 3535 3536 return features; 3537 } 3538 3539 netdev_features_t netif_skb_features(struct sk_buff *skb) 3540 { 3541 struct net_device *dev = skb->dev; 3542 netdev_features_t features = dev->features; 3543 3544 if (skb_is_gso(skb)) 3545 features = gso_features_check(skb, dev, features); 3546 3547 /* If encapsulation offload request, verify we are testing 3548 * hardware encapsulation features instead of standard 3549 * features for the netdev 3550 */ 3551 if (skb->encapsulation) 3552 features &= dev->hw_enc_features; 3553 3554 if (skb_vlan_tagged(skb)) 3555 features = netdev_intersect_features(features, 3556 dev->vlan_features | 3557 NETIF_F_HW_VLAN_CTAG_TX | 3558 NETIF_F_HW_VLAN_STAG_TX); 3559 3560 if (dev->netdev_ops->ndo_features_check) 3561 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3562 features); 3563 else 3564 features &= dflt_features_check(skb, dev, features); 3565 3566 return harmonize_features(skb, features); 3567 } 3568 EXPORT_SYMBOL(netif_skb_features); 3569 3570 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3571 struct netdev_queue *txq, bool more) 3572 { 3573 unsigned int len; 3574 int rc; 3575 3576 if (dev_nit_active(dev)) 3577 dev_queue_xmit_nit(skb, dev); 3578 3579 len = skb->len; 3580 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3581 trace_net_dev_start_xmit(skb, dev); 3582 rc = netdev_start_xmit(skb, dev, txq, more); 3583 trace_net_dev_xmit(skb, rc, dev, len); 3584 3585 return rc; 3586 } 3587 3588 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3589 struct netdev_queue *txq, int *ret) 3590 { 3591 struct sk_buff *skb = first; 3592 int rc = NETDEV_TX_OK; 3593 3594 while (skb) { 3595 struct sk_buff *next = skb->next; 3596 3597 skb_mark_not_on_list(skb); 3598 rc = xmit_one(skb, dev, txq, next != NULL); 3599 if (unlikely(!dev_xmit_complete(rc))) { 3600 skb->next = next; 3601 goto out; 3602 } 3603 3604 skb = next; 3605 if (netif_tx_queue_stopped(txq) && skb) { 3606 rc = NETDEV_TX_BUSY; 3607 break; 3608 } 3609 } 3610 3611 out: 3612 *ret = rc; 3613 return skb; 3614 } 3615 3616 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3617 netdev_features_t features) 3618 { 3619 if (skb_vlan_tag_present(skb) && 3620 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3621 skb = __vlan_hwaccel_push_inside(skb); 3622 return skb; 3623 } 3624 3625 int skb_csum_hwoffload_help(struct sk_buff *skb, 3626 const netdev_features_t features) 3627 { 3628 if (unlikely(skb_csum_is_sctp(skb))) 3629 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3630 skb_crc32c_csum_help(skb); 3631 3632 if (features & NETIF_F_HW_CSUM) 3633 return 0; 3634 3635 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3636 switch (skb->csum_offset) { 3637 case offsetof(struct tcphdr, check): 3638 case offsetof(struct udphdr, check): 3639 return 0; 3640 } 3641 } 3642 3643 return skb_checksum_help(skb); 3644 } 3645 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3646 3647 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3648 { 3649 netdev_features_t features; 3650 3651 features = netif_skb_features(skb); 3652 skb = validate_xmit_vlan(skb, features); 3653 if (unlikely(!skb)) 3654 goto out_null; 3655 3656 skb = sk_validate_xmit_skb(skb, dev); 3657 if (unlikely(!skb)) 3658 goto out_null; 3659 3660 if (netif_needs_gso(skb, features)) { 3661 struct sk_buff *segs; 3662 3663 segs = skb_gso_segment(skb, features); 3664 if (IS_ERR(segs)) { 3665 goto out_kfree_skb; 3666 } else if (segs) { 3667 consume_skb(skb); 3668 skb = segs; 3669 } 3670 } else { 3671 if (skb_needs_linearize(skb, features) && 3672 __skb_linearize(skb)) 3673 goto out_kfree_skb; 3674 3675 /* If packet is not checksummed and device does not 3676 * support checksumming for this protocol, complete 3677 * checksumming here. 3678 */ 3679 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3680 if (skb->encapsulation) 3681 skb_set_inner_transport_header(skb, 3682 skb_checksum_start_offset(skb)); 3683 else 3684 skb_set_transport_header(skb, 3685 skb_checksum_start_offset(skb)); 3686 if (skb_csum_hwoffload_help(skb, features)) 3687 goto out_kfree_skb; 3688 } 3689 } 3690 3691 skb = validate_xmit_xfrm(skb, features, again); 3692 3693 return skb; 3694 3695 out_kfree_skb: 3696 kfree_skb(skb); 3697 out_null: 3698 dev_core_stats_tx_dropped_inc(dev); 3699 return NULL; 3700 } 3701 3702 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3703 { 3704 struct sk_buff *next, *head = NULL, *tail; 3705 3706 for (; skb != NULL; skb = next) { 3707 next = skb->next; 3708 skb_mark_not_on_list(skb); 3709 3710 /* in case skb wont be segmented, point to itself */ 3711 skb->prev = skb; 3712 3713 skb = validate_xmit_skb(skb, dev, again); 3714 if (!skb) 3715 continue; 3716 3717 if (!head) 3718 head = skb; 3719 else 3720 tail->next = skb; 3721 /* If skb was segmented, skb->prev points to 3722 * the last segment. If not, it still contains skb. 3723 */ 3724 tail = skb->prev; 3725 } 3726 return head; 3727 } 3728 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3729 3730 static void qdisc_pkt_len_init(struct sk_buff *skb) 3731 { 3732 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3733 3734 qdisc_skb_cb(skb)->pkt_len = skb->len; 3735 3736 /* To get more precise estimation of bytes sent on wire, 3737 * we add to pkt_len the headers size of all segments 3738 */ 3739 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3740 unsigned int hdr_len; 3741 u16 gso_segs = shinfo->gso_segs; 3742 3743 /* mac layer + network layer */ 3744 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3745 3746 /* + transport layer */ 3747 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3748 const struct tcphdr *th; 3749 struct tcphdr _tcphdr; 3750 3751 th = skb_header_pointer(skb, skb_transport_offset(skb), 3752 sizeof(_tcphdr), &_tcphdr); 3753 if (likely(th)) 3754 hdr_len += __tcp_hdrlen(th); 3755 } else { 3756 struct udphdr _udphdr; 3757 3758 if (skb_header_pointer(skb, skb_transport_offset(skb), 3759 sizeof(_udphdr), &_udphdr)) 3760 hdr_len += sizeof(struct udphdr); 3761 } 3762 3763 if (shinfo->gso_type & SKB_GSO_DODGY) 3764 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3765 shinfo->gso_size); 3766 3767 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3768 } 3769 } 3770 3771 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3772 struct sk_buff **to_free, 3773 struct netdev_queue *txq) 3774 { 3775 int rc; 3776 3777 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3778 if (rc == NET_XMIT_SUCCESS) 3779 trace_qdisc_enqueue(q, txq, skb); 3780 return rc; 3781 } 3782 3783 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3784 struct net_device *dev, 3785 struct netdev_queue *txq) 3786 { 3787 spinlock_t *root_lock = qdisc_lock(q); 3788 struct sk_buff *to_free = NULL; 3789 bool contended; 3790 int rc; 3791 3792 qdisc_calculate_pkt_len(skb, q); 3793 3794 if (q->flags & TCQ_F_NOLOCK) { 3795 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3796 qdisc_run_begin(q)) { 3797 /* Retest nolock_qdisc_is_empty() within the protection 3798 * of q->seqlock to protect from racing with requeuing. 3799 */ 3800 if (unlikely(!nolock_qdisc_is_empty(q))) { 3801 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3802 __qdisc_run(q); 3803 qdisc_run_end(q); 3804 3805 goto no_lock_out; 3806 } 3807 3808 qdisc_bstats_cpu_update(q, skb); 3809 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3810 !nolock_qdisc_is_empty(q)) 3811 __qdisc_run(q); 3812 3813 qdisc_run_end(q); 3814 return NET_XMIT_SUCCESS; 3815 } 3816 3817 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3818 qdisc_run(q); 3819 3820 no_lock_out: 3821 if (unlikely(to_free)) 3822 kfree_skb_list_reason(to_free, 3823 SKB_DROP_REASON_QDISC_DROP); 3824 return rc; 3825 } 3826 3827 /* 3828 * Heuristic to force contended enqueues to serialize on a 3829 * separate lock before trying to get qdisc main lock. 3830 * This permits qdisc->running owner to get the lock more 3831 * often and dequeue packets faster. 3832 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3833 * and then other tasks will only enqueue packets. The packets will be 3834 * sent after the qdisc owner is scheduled again. To prevent this 3835 * scenario the task always serialize on the lock. 3836 */ 3837 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3838 if (unlikely(contended)) 3839 spin_lock(&q->busylock); 3840 3841 spin_lock(root_lock); 3842 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3843 __qdisc_drop(skb, &to_free); 3844 rc = NET_XMIT_DROP; 3845 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3846 qdisc_run_begin(q)) { 3847 /* 3848 * This is a work-conserving queue; there are no old skbs 3849 * waiting to be sent out; and the qdisc is not running - 3850 * xmit the skb directly. 3851 */ 3852 3853 qdisc_bstats_update(q, skb); 3854 3855 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3856 if (unlikely(contended)) { 3857 spin_unlock(&q->busylock); 3858 contended = false; 3859 } 3860 __qdisc_run(q); 3861 } 3862 3863 qdisc_run_end(q); 3864 rc = NET_XMIT_SUCCESS; 3865 } else { 3866 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3867 if (qdisc_run_begin(q)) { 3868 if (unlikely(contended)) { 3869 spin_unlock(&q->busylock); 3870 contended = false; 3871 } 3872 __qdisc_run(q); 3873 qdisc_run_end(q); 3874 } 3875 } 3876 spin_unlock(root_lock); 3877 if (unlikely(to_free)) 3878 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3879 if (unlikely(contended)) 3880 spin_unlock(&q->busylock); 3881 return rc; 3882 } 3883 3884 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3885 static void skb_update_prio(struct sk_buff *skb) 3886 { 3887 const struct netprio_map *map; 3888 const struct sock *sk; 3889 unsigned int prioidx; 3890 3891 if (skb->priority) 3892 return; 3893 map = rcu_dereference_bh(skb->dev->priomap); 3894 if (!map) 3895 return; 3896 sk = skb_to_full_sk(skb); 3897 if (!sk) 3898 return; 3899 3900 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3901 3902 if (prioidx < map->priomap_len) 3903 skb->priority = map->priomap[prioidx]; 3904 } 3905 #else 3906 #define skb_update_prio(skb) 3907 #endif 3908 3909 /** 3910 * dev_loopback_xmit - loop back @skb 3911 * @net: network namespace this loopback is happening in 3912 * @sk: sk needed to be a netfilter okfn 3913 * @skb: buffer to transmit 3914 */ 3915 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3916 { 3917 skb_reset_mac_header(skb); 3918 __skb_pull(skb, skb_network_offset(skb)); 3919 skb->pkt_type = PACKET_LOOPBACK; 3920 if (skb->ip_summed == CHECKSUM_NONE) 3921 skb->ip_summed = CHECKSUM_UNNECESSARY; 3922 WARN_ON(!skb_dst(skb)); 3923 skb_dst_force(skb); 3924 netif_rx(skb); 3925 return 0; 3926 } 3927 EXPORT_SYMBOL(dev_loopback_xmit); 3928 3929 #ifdef CONFIG_NET_EGRESS 3930 static struct sk_buff * 3931 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3932 { 3933 #ifdef CONFIG_NET_CLS_ACT 3934 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3935 struct tcf_result cl_res; 3936 3937 if (!miniq) 3938 return skb; 3939 3940 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3941 tc_skb_cb(skb)->mru = 0; 3942 tc_skb_cb(skb)->post_ct = false; 3943 mini_qdisc_bstats_cpu_update(miniq, skb); 3944 3945 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3946 case TC_ACT_OK: 3947 case TC_ACT_RECLASSIFY: 3948 skb->tc_index = TC_H_MIN(cl_res.classid); 3949 break; 3950 case TC_ACT_SHOT: 3951 mini_qdisc_qstats_cpu_drop(miniq); 3952 *ret = NET_XMIT_DROP; 3953 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3954 return NULL; 3955 case TC_ACT_STOLEN: 3956 case TC_ACT_QUEUED: 3957 case TC_ACT_TRAP: 3958 *ret = NET_XMIT_SUCCESS; 3959 consume_skb(skb); 3960 return NULL; 3961 case TC_ACT_REDIRECT: 3962 /* No need to push/pop skb's mac_header here on egress! */ 3963 skb_do_redirect(skb); 3964 *ret = NET_XMIT_SUCCESS; 3965 return NULL; 3966 default: 3967 break; 3968 } 3969 #endif /* CONFIG_NET_CLS_ACT */ 3970 3971 return skb; 3972 } 3973 3974 static struct netdev_queue * 3975 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3976 { 3977 int qm = skb_get_queue_mapping(skb); 3978 3979 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3980 } 3981 3982 static bool netdev_xmit_txqueue_skipped(void) 3983 { 3984 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3985 } 3986 3987 void netdev_xmit_skip_txqueue(bool skip) 3988 { 3989 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3990 } 3991 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3992 #endif /* CONFIG_NET_EGRESS */ 3993 3994 #ifdef CONFIG_XPS 3995 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3996 struct xps_dev_maps *dev_maps, unsigned int tci) 3997 { 3998 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3999 struct xps_map *map; 4000 int queue_index = -1; 4001 4002 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4003 return queue_index; 4004 4005 tci *= dev_maps->num_tc; 4006 tci += tc; 4007 4008 map = rcu_dereference(dev_maps->attr_map[tci]); 4009 if (map) { 4010 if (map->len == 1) 4011 queue_index = map->queues[0]; 4012 else 4013 queue_index = map->queues[reciprocal_scale( 4014 skb_get_hash(skb), map->len)]; 4015 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4016 queue_index = -1; 4017 } 4018 return queue_index; 4019 } 4020 #endif 4021 4022 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4023 struct sk_buff *skb) 4024 { 4025 #ifdef CONFIG_XPS 4026 struct xps_dev_maps *dev_maps; 4027 struct sock *sk = skb->sk; 4028 int queue_index = -1; 4029 4030 if (!static_key_false(&xps_needed)) 4031 return -1; 4032 4033 rcu_read_lock(); 4034 if (!static_key_false(&xps_rxqs_needed)) 4035 goto get_cpus_map; 4036 4037 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4038 if (dev_maps) { 4039 int tci = sk_rx_queue_get(sk); 4040 4041 if (tci >= 0) 4042 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4043 tci); 4044 } 4045 4046 get_cpus_map: 4047 if (queue_index < 0) { 4048 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4049 if (dev_maps) { 4050 unsigned int tci = skb->sender_cpu - 1; 4051 4052 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4053 tci); 4054 } 4055 } 4056 rcu_read_unlock(); 4057 4058 return queue_index; 4059 #else 4060 return -1; 4061 #endif 4062 } 4063 4064 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4065 struct net_device *sb_dev) 4066 { 4067 return 0; 4068 } 4069 EXPORT_SYMBOL(dev_pick_tx_zero); 4070 4071 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4072 struct net_device *sb_dev) 4073 { 4074 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4075 } 4076 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4077 4078 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4079 struct net_device *sb_dev) 4080 { 4081 struct sock *sk = skb->sk; 4082 int queue_index = sk_tx_queue_get(sk); 4083 4084 sb_dev = sb_dev ? : dev; 4085 4086 if (queue_index < 0 || skb->ooo_okay || 4087 queue_index >= dev->real_num_tx_queues) { 4088 int new_index = get_xps_queue(dev, sb_dev, skb); 4089 4090 if (new_index < 0) 4091 new_index = skb_tx_hash(dev, sb_dev, skb); 4092 4093 if (queue_index != new_index && sk && 4094 sk_fullsock(sk) && 4095 rcu_access_pointer(sk->sk_dst_cache)) 4096 sk_tx_queue_set(sk, new_index); 4097 4098 queue_index = new_index; 4099 } 4100 4101 return queue_index; 4102 } 4103 EXPORT_SYMBOL(netdev_pick_tx); 4104 4105 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4106 struct sk_buff *skb, 4107 struct net_device *sb_dev) 4108 { 4109 int queue_index = 0; 4110 4111 #ifdef CONFIG_XPS 4112 u32 sender_cpu = skb->sender_cpu - 1; 4113 4114 if (sender_cpu >= (u32)NR_CPUS) 4115 skb->sender_cpu = raw_smp_processor_id() + 1; 4116 #endif 4117 4118 if (dev->real_num_tx_queues != 1) { 4119 const struct net_device_ops *ops = dev->netdev_ops; 4120 4121 if (ops->ndo_select_queue) 4122 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4123 else 4124 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4125 4126 queue_index = netdev_cap_txqueue(dev, queue_index); 4127 } 4128 4129 skb_set_queue_mapping(skb, queue_index); 4130 return netdev_get_tx_queue(dev, queue_index); 4131 } 4132 4133 /** 4134 * __dev_queue_xmit - transmit a buffer 4135 * @skb: buffer to transmit 4136 * @sb_dev: suboordinate device used for L2 forwarding offload 4137 * 4138 * Queue a buffer for transmission to a network device. The caller must 4139 * have set the device and priority and built the buffer before calling 4140 * this function. The function can be called from an interrupt. 4141 * 4142 * A negative errno code is returned on a failure. A success does not 4143 * guarantee the frame will be transmitted as it may be dropped due 4144 * to congestion or traffic shaping. 4145 * 4146 * ----------------------------------------------------------------------------------- 4147 * I notice this method can also return errors from the queue disciplines, 4148 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4149 * be positive. 4150 * 4151 * Regardless of the return value, the skb is consumed, so it is currently 4152 * difficult to retry a send to this method. (You can bump the ref count 4153 * before sending to hold a reference for retry if you are careful.) 4154 * 4155 * When calling this method, interrupts MUST be enabled. This is because 4156 * the BH enable code must have IRQs enabled so that it will not deadlock. 4157 * --BLG 4158 */ 4159 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4160 { 4161 struct net_device *dev = skb->dev; 4162 struct netdev_queue *txq = NULL; 4163 struct Qdisc *q; 4164 int rc = -ENOMEM; 4165 bool again = false; 4166 4167 skb_reset_mac_header(skb); 4168 4169 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4170 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4171 4172 /* Disable soft irqs for various locks below. Also 4173 * stops preemption for RCU. 4174 */ 4175 rcu_read_lock_bh(); 4176 4177 skb_update_prio(skb); 4178 4179 qdisc_pkt_len_init(skb); 4180 #ifdef CONFIG_NET_CLS_ACT 4181 skb->tc_at_ingress = 0; 4182 #endif 4183 #ifdef CONFIG_NET_EGRESS 4184 if (static_branch_unlikely(&egress_needed_key)) { 4185 if (nf_hook_egress_active()) { 4186 skb = nf_hook_egress(skb, &rc, dev); 4187 if (!skb) 4188 goto out; 4189 } 4190 4191 netdev_xmit_skip_txqueue(false); 4192 4193 nf_skip_egress(skb, true); 4194 skb = sch_handle_egress(skb, &rc, dev); 4195 if (!skb) 4196 goto out; 4197 nf_skip_egress(skb, false); 4198 4199 if (netdev_xmit_txqueue_skipped()) 4200 txq = netdev_tx_queue_mapping(dev, skb); 4201 } 4202 #endif 4203 /* If device/qdisc don't need skb->dst, release it right now while 4204 * its hot in this cpu cache. 4205 */ 4206 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4207 skb_dst_drop(skb); 4208 else 4209 skb_dst_force(skb); 4210 4211 if (!txq) 4212 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4213 4214 q = rcu_dereference_bh(txq->qdisc); 4215 4216 trace_net_dev_queue(skb); 4217 if (q->enqueue) { 4218 rc = __dev_xmit_skb(skb, q, dev, txq); 4219 goto out; 4220 } 4221 4222 /* The device has no queue. Common case for software devices: 4223 * loopback, all the sorts of tunnels... 4224 4225 * Really, it is unlikely that netif_tx_lock protection is necessary 4226 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4227 * counters.) 4228 * However, it is possible, that they rely on protection 4229 * made by us here. 4230 4231 * Check this and shot the lock. It is not prone from deadlocks. 4232 *Either shot noqueue qdisc, it is even simpler 8) 4233 */ 4234 if (dev->flags & IFF_UP) { 4235 int cpu = smp_processor_id(); /* ok because BHs are off */ 4236 4237 /* Other cpus might concurrently change txq->xmit_lock_owner 4238 * to -1 or to their cpu id, but not to our id. 4239 */ 4240 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4241 if (dev_xmit_recursion()) 4242 goto recursion_alert; 4243 4244 skb = validate_xmit_skb(skb, dev, &again); 4245 if (!skb) 4246 goto out; 4247 4248 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4249 HARD_TX_LOCK(dev, txq, cpu); 4250 4251 if (!netif_xmit_stopped(txq)) { 4252 dev_xmit_recursion_inc(); 4253 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4254 dev_xmit_recursion_dec(); 4255 if (dev_xmit_complete(rc)) { 4256 HARD_TX_UNLOCK(dev, txq); 4257 goto out; 4258 } 4259 } 4260 HARD_TX_UNLOCK(dev, txq); 4261 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4262 dev->name); 4263 } else { 4264 /* Recursion is detected! It is possible, 4265 * unfortunately 4266 */ 4267 recursion_alert: 4268 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4269 dev->name); 4270 } 4271 } 4272 4273 rc = -ENETDOWN; 4274 rcu_read_unlock_bh(); 4275 4276 dev_core_stats_tx_dropped_inc(dev); 4277 kfree_skb_list(skb); 4278 return rc; 4279 out: 4280 rcu_read_unlock_bh(); 4281 return rc; 4282 } 4283 EXPORT_SYMBOL(__dev_queue_xmit); 4284 4285 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4286 { 4287 struct net_device *dev = skb->dev; 4288 struct sk_buff *orig_skb = skb; 4289 struct netdev_queue *txq; 4290 int ret = NETDEV_TX_BUSY; 4291 bool again = false; 4292 4293 if (unlikely(!netif_running(dev) || 4294 !netif_carrier_ok(dev))) 4295 goto drop; 4296 4297 skb = validate_xmit_skb_list(skb, dev, &again); 4298 if (skb != orig_skb) 4299 goto drop; 4300 4301 skb_set_queue_mapping(skb, queue_id); 4302 txq = skb_get_tx_queue(dev, skb); 4303 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4304 4305 local_bh_disable(); 4306 4307 dev_xmit_recursion_inc(); 4308 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4309 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4310 ret = netdev_start_xmit(skb, dev, txq, false); 4311 HARD_TX_UNLOCK(dev, txq); 4312 dev_xmit_recursion_dec(); 4313 4314 local_bh_enable(); 4315 return ret; 4316 drop: 4317 dev_core_stats_tx_dropped_inc(dev); 4318 kfree_skb_list(skb); 4319 return NET_XMIT_DROP; 4320 } 4321 EXPORT_SYMBOL(__dev_direct_xmit); 4322 4323 /************************************************************************* 4324 * Receiver routines 4325 *************************************************************************/ 4326 4327 int netdev_max_backlog __read_mostly = 1000; 4328 EXPORT_SYMBOL(netdev_max_backlog); 4329 4330 int netdev_tstamp_prequeue __read_mostly = 1; 4331 int netdev_budget __read_mostly = 300; 4332 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4333 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4334 int weight_p __read_mostly = 64; /* old backlog weight */ 4335 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4336 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4337 int dev_rx_weight __read_mostly = 64; 4338 int dev_tx_weight __read_mostly = 64; 4339 4340 /* Called with irq disabled */ 4341 static inline void ____napi_schedule(struct softnet_data *sd, 4342 struct napi_struct *napi) 4343 { 4344 struct task_struct *thread; 4345 4346 lockdep_assert_irqs_disabled(); 4347 4348 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4349 /* Paired with smp_mb__before_atomic() in 4350 * napi_enable()/dev_set_threaded(). 4351 * Use READ_ONCE() to guarantee a complete 4352 * read on napi->thread. Only call 4353 * wake_up_process() when it's not NULL. 4354 */ 4355 thread = READ_ONCE(napi->thread); 4356 if (thread) { 4357 /* Avoid doing set_bit() if the thread is in 4358 * INTERRUPTIBLE state, cause napi_thread_wait() 4359 * makes sure to proceed with napi polling 4360 * if the thread is explicitly woken from here. 4361 */ 4362 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4363 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4364 wake_up_process(thread); 4365 return; 4366 } 4367 } 4368 4369 list_add_tail(&napi->poll_list, &sd->poll_list); 4370 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4371 } 4372 4373 #ifdef CONFIG_RPS 4374 4375 /* One global table that all flow-based protocols share. */ 4376 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4377 EXPORT_SYMBOL(rps_sock_flow_table); 4378 u32 rps_cpu_mask __read_mostly; 4379 EXPORT_SYMBOL(rps_cpu_mask); 4380 4381 struct static_key_false rps_needed __read_mostly; 4382 EXPORT_SYMBOL(rps_needed); 4383 struct static_key_false rfs_needed __read_mostly; 4384 EXPORT_SYMBOL(rfs_needed); 4385 4386 static struct rps_dev_flow * 4387 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4388 struct rps_dev_flow *rflow, u16 next_cpu) 4389 { 4390 if (next_cpu < nr_cpu_ids) { 4391 #ifdef CONFIG_RFS_ACCEL 4392 struct netdev_rx_queue *rxqueue; 4393 struct rps_dev_flow_table *flow_table; 4394 struct rps_dev_flow *old_rflow; 4395 u32 flow_id; 4396 u16 rxq_index; 4397 int rc; 4398 4399 /* Should we steer this flow to a different hardware queue? */ 4400 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4401 !(dev->features & NETIF_F_NTUPLE)) 4402 goto out; 4403 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4404 if (rxq_index == skb_get_rx_queue(skb)) 4405 goto out; 4406 4407 rxqueue = dev->_rx + rxq_index; 4408 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4409 if (!flow_table) 4410 goto out; 4411 flow_id = skb_get_hash(skb) & flow_table->mask; 4412 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4413 rxq_index, flow_id); 4414 if (rc < 0) 4415 goto out; 4416 old_rflow = rflow; 4417 rflow = &flow_table->flows[flow_id]; 4418 rflow->filter = rc; 4419 if (old_rflow->filter == rflow->filter) 4420 old_rflow->filter = RPS_NO_FILTER; 4421 out: 4422 #endif 4423 rflow->last_qtail = 4424 per_cpu(softnet_data, next_cpu).input_queue_head; 4425 } 4426 4427 rflow->cpu = next_cpu; 4428 return rflow; 4429 } 4430 4431 /* 4432 * get_rps_cpu is called from netif_receive_skb and returns the target 4433 * CPU from the RPS map of the receiving queue for a given skb. 4434 * rcu_read_lock must be held on entry. 4435 */ 4436 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4437 struct rps_dev_flow **rflowp) 4438 { 4439 const struct rps_sock_flow_table *sock_flow_table; 4440 struct netdev_rx_queue *rxqueue = dev->_rx; 4441 struct rps_dev_flow_table *flow_table; 4442 struct rps_map *map; 4443 int cpu = -1; 4444 u32 tcpu; 4445 u32 hash; 4446 4447 if (skb_rx_queue_recorded(skb)) { 4448 u16 index = skb_get_rx_queue(skb); 4449 4450 if (unlikely(index >= dev->real_num_rx_queues)) { 4451 WARN_ONCE(dev->real_num_rx_queues > 1, 4452 "%s received packet on queue %u, but number " 4453 "of RX queues is %u\n", 4454 dev->name, index, dev->real_num_rx_queues); 4455 goto done; 4456 } 4457 rxqueue += index; 4458 } 4459 4460 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4461 4462 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4463 map = rcu_dereference(rxqueue->rps_map); 4464 if (!flow_table && !map) 4465 goto done; 4466 4467 skb_reset_network_header(skb); 4468 hash = skb_get_hash(skb); 4469 if (!hash) 4470 goto done; 4471 4472 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4473 if (flow_table && sock_flow_table) { 4474 struct rps_dev_flow *rflow; 4475 u32 next_cpu; 4476 u32 ident; 4477 4478 /* First check into global flow table if there is a match */ 4479 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4480 if ((ident ^ hash) & ~rps_cpu_mask) 4481 goto try_rps; 4482 4483 next_cpu = ident & rps_cpu_mask; 4484 4485 /* OK, now we know there is a match, 4486 * we can look at the local (per receive queue) flow table 4487 */ 4488 rflow = &flow_table->flows[hash & flow_table->mask]; 4489 tcpu = rflow->cpu; 4490 4491 /* 4492 * If the desired CPU (where last recvmsg was done) is 4493 * different from current CPU (one in the rx-queue flow 4494 * table entry), switch if one of the following holds: 4495 * - Current CPU is unset (>= nr_cpu_ids). 4496 * - Current CPU is offline. 4497 * - The current CPU's queue tail has advanced beyond the 4498 * last packet that was enqueued using this table entry. 4499 * This guarantees that all previous packets for the flow 4500 * have been dequeued, thus preserving in order delivery. 4501 */ 4502 if (unlikely(tcpu != next_cpu) && 4503 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4504 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4505 rflow->last_qtail)) >= 0)) { 4506 tcpu = next_cpu; 4507 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4508 } 4509 4510 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4511 *rflowp = rflow; 4512 cpu = tcpu; 4513 goto done; 4514 } 4515 } 4516 4517 try_rps: 4518 4519 if (map) { 4520 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4521 if (cpu_online(tcpu)) { 4522 cpu = tcpu; 4523 goto done; 4524 } 4525 } 4526 4527 done: 4528 return cpu; 4529 } 4530 4531 #ifdef CONFIG_RFS_ACCEL 4532 4533 /** 4534 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4535 * @dev: Device on which the filter was set 4536 * @rxq_index: RX queue index 4537 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4538 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4539 * 4540 * Drivers that implement ndo_rx_flow_steer() should periodically call 4541 * this function for each installed filter and remove the filters for 4542 * which it returns %true. 4543 */ 4544 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4545 u32 flow_id, u16 filter_id) 4546 { 4547 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4548 struct rps_dev_flow_table *flow_table; 4549 struct rps_dev_flow *rflow; 4550 bool expire = true; 4551 unsigned int cpu; 4552 4553 rcu_read_lock(); 4554 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4555 if (flow_table && flow_id <= flow_table->mask) { 4556 rflow = &flow_table->flows[flow_id]; 4557 cpu = READ_ONCE(rflow->cpu); 4558 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4559 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4560 rflow->last_qtail) < 4561 (int)(10 * flow_table->mask))) 4562 expire = false; 4563 } 4564 rcu_read_unlock(); 4565 return expire; 4566 } 4567 EXPORT_SYMBOL(rps_may_expire_flow); 4568 4569 #endif /* CONFIG_RFS_ACCEL */ 4570 4571 /* Called from hardirq (IPI) context */ 4572 static void rps_trigger_softirq(void *data) 4573 { 4574 struct softnet_data *sd = data; 4575 4576 ____napi_schedule(sd, &sd->backlog); 4577 sd->received_rps++; 4578 } 4579 4580 #endif /* CONFIG_RPS */ 4581 4582 /* Called from hardirq (IPI) context */ 4583 static void trigger_rx_softirq(void *data __always_unused) 4584 { 4585 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4586 } 4587 4588 /* 4589 * Check if this softnet_data structure is another cpu one 4590 * If yes, queue it to our IPI list and return 1 4591 * If no, return 0 4592 */ 4593 static int napi_schedule_rps(struct softnet_data *sd) 4594 { 4595 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4596 4597 #ifdef CONFIG_RPS 4598 if (sd != mysd) { 4599 sd->rps_ipi_next = mysd->rps_ipi_list; 4600 mysd->rps_ipi_list = sd; 4601 4602 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4603 return 1; 4604 } 4605 #endif /* CONFIG_RPS */ 4606 __napi_schedule_irqoff(&mysd->backlog); 4607 return 0; 4608 } 4609 4610 #ifdef CONFIG_NET_FLOW_LIMIT 4611 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4612 #endif 4613 4614 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4615 { 4616 #ifdef CONFIG_NET_FLOW_LIMIT 4617 struct sd_flow_limit *fl; 4618 struct softnet_data *sd; 4619 unsigned int old_flow, new_flow; 4620 4621 if (qlen < (netdev_max_backlog >> 1)) 4622 return false; 4623 4624 sd = this_cpu_ptr(&softnet_data); 4625 4626 rcu_read_lock(); 4627 fl = rcu_dereference(sd->flow_limit); 4628 if (fl) { 4629 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4630 old_flow = fl->history[fl->history_head]; 4631 fl->history[fl->history_head] = new_flow; 4632 4633 fl->history_head++; 4634 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4635 4636 if (likely(fl->buckets[old_flow])) 4637 fl->buckets[old_flow]--; 4638 4639 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4640 fl->count++; 4641 rcu_read_unlock(); 4642 return true; 4643 } 4644 } 4645 rcu_read_unlock(); 4646 #endif 4647 return false; 4648 } 4649 4650 /* 4651 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4652 * queue (may be a remote CPU queue). 4653 */ 4654 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4655 unsigned int *qtail) 4656 { 4657 enum skb_drop_reason reason; 4658 struct softnet_data *sd; 4659 unsigned long flags; 4660 unsigned int qlen; 4661 4662 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4663 sd = &per_cpu(softnet_data, cpu); 4664 4665 rps_lock_irqsave(sd, &flags); 4666 if (!netif_running(skb->dev)) 4667 goto drop; 4668 qlen = skb_queue_len(&sd->input_pkt_queue); 4669 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4670 if (qlen) { 4671 enqueue: 4672 __skb_queue_tail(&sd->input_pkt_queue, skb); 4673 input_queue_tail_incr_save(sd, qtail); 4674 rps_unlock_irq_restore(sd, &flags); 4675 return NET_RX_SUCCESS; 4676 } 4677 4678 /* Schedule NAPI for backlog device 4679 * We can use non atomic operation since we own the queue lock 4680 */ 4681 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4682 napi_schedule_rps(sd); 4683 goto enqueue; 4684 } 4685 reason = SKB_DROP_REASON_CPU_BACKLOG; 4686 4687 drop: 4688 sd->dropped++; 4689 rps_unlock_irq_restore(sd, &flags); 4690 4691 dev_core_stats_rx_dropped_inc(skb->dev); 4692 kfree_skb_reason(skb, reason); 4693 return NET_RX_DROP; 4694 } 4695 4696 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4697 { 4698 struct net_device *dev = skb->dev; 4699 struct netdev_rx_queue *rxqueue; 4700 4701 rxqueue = dev->_rx; 4702 4703 if (skb_rx_queue_recorded(skb)) { 4704 u16 index = skb_get_rx_queue(skb); 4705 4706 if (unlikely(index >= dev->real_num_rx_queues)) { 4707 WARN_ONCE(dev->real_num_rx_queues > 1, 4708 "%s received packet on queue %u, but number " 4709 "of RX queues is %u\n", 4710 dev->name, index, dev->real_num_rx_queues); 4711 4712 return rxqueue; /* Return first rxqueue */ 4713 } 4714 rxqueue += index; 4715 } 4716 return rxqueue; 4717 } 4718 4719 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4720 struct bpf_prog *xdp_prog) 4721 { 4722 void *orig_data, *orig_data_end, *hard_start; 4723 struct netdev_rx_queue *rxqueue; 4724 bool orig_bcast, orig_host; 4725 u32 mac_len, frame_sz; 4726 __be16 orig_eth_type; 4727 struct ethhdr *eth; 4728 u32 metalen, act; 4729 int off; 4730 4731 /* The XDP program wants to see the packet starting at the MAC 4732 * header. 4733 */ 4734 mac_len = skb->data - skb_mac_header(skb); 4735 hard_start = skb->data - skb_headroom(skb); 4736 4737 /* SKB "head" area always have tailroom for skb_shared_info */ 4738 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4739 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4740 4741 rxqueue = netif_get_rxqueue(skb); 4742 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4743 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4744 skb_headlen(skb) + mac_len, true); 4745 4746 orig_data_end = xdp->data_end; 4747 orig_data = xdp->data; 4748 eth = (struct ethhdr *)xdp->data; 4749 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4750 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4751 orig_eth_type = eth->h_proto; 4752 4753 act = bpf_prog_run_xdp(xdp_prog, xdp); 4754 4755 /* check if bpf_xdp_adjust_head was used */ 4756 off = xdp->data - orig_data; 4757 if (off) { 4758 if (off > 0) 4759 __skb_pull(skb, off); 4760 else if (off < 0) 4761 __skb_push(skb, -off); 4762 4763 skb->mac_header += off; 4764 skb_reset_network_header(skb); 4765 } 4766 4767 /* check if bpf_xdp_adjust_tail was used */ 4768 off = xdp->data_end - orig_data_end; 4769 if (off != 0) { 4770 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4771 skb->len += off; /* positive on grow, negative on shrink */ 4772 } 4773 4774 /* check if XDP changed eth hdr such SKB needs update */ 4775 eth = (struct ethhdr *)xdp->data; 4776 if ((orig_eth_type != eth->h_proto) || 4777 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4778 skb->dev->dev_addr)) || 4779 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4780 __skb_push(skb, ETH_HLEN); 4781 skb->pkt_type = PACKET_HOST; 4782 skb->protocol = eth_type_trans(skb, skb->dev); 4783 } 4784 4785 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4786 * before calling us again on redirect path. We do not call do_redirect 4787 * as we leave that up to the caller. 4788 * 4789 * Caller is responsible for managing lifetime of skb (i.e. calling 4790 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4791 */ 4792 switch (act) { 4793 case XDP_REDIRECT: 4794 case XDP_TX: 4795 __skb_push(skb, mac_len); 4796 break; 4797 case XDP_PASS: 4798 metalen = xdp->data - xdp->data_meta; 4799 if (metalen) 4800 skb_metadata_set(skb, metalen); 4801 break; 4802 } 4803 4804 return act; 4805 } 4806 4807 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4808 struct xdp_buff *xdp, 4809 struct bpf_prog *xdp_prog) 4810 { 4811 u32 act = XDP_DROP; 4812 4813 /* Reinjected packets coming from act_mirred or similar should 4814 * not get XDP generic processing. 4815 */ 4816 if (skb_is_redirected(skb)) 4817 return XDP_PASS; 4818 4819 /* XDP packets must be linear and must have sufficient headroom 4820 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4821 * native XDP provides, thus we need to do it here as well. 4822 */ 4823 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4824 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4825 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4826 int troom = skb->tail + skb->data_len - skb->end; 4827 4828 /* In case we have to go down the path and also linearize, 4829 * then lets do the pskb_expand_head() work just once here. 4830 */ 4831 if (pskb_expand_head(skb, 4832 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4833 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4834 goto do_drop; 4835 if (skb_linearize(skb)) 4836 goto do_drop; 4837 } 4838 4839 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4840 switch (act) { 4841 case XDP_REDIRECT: 4842 case XDP_TX: 4843 case XDP_PASS: 4844 break; 4845 default: 4846 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4847 fallthrough; 4848 case XDP_ABORTED: 4849 trace_xdp_exception(skb->dev, xdp_prog, act); 4850 fallthrough; 4851 case XDP_DROP: 4852 do_drop: 4853 kfree_skb(skb); 4854 break; 4855 } 4856 4857 return act; 4858 } 4859 4860 /* When doing generic XDP we have to bypass the qdisc layer and the 4861 * network taps in order to match in-driver-XDP behavior. 4862 */ 4863 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4864 { 4865 struct net_device *dev = skb->dev; 4866 struct netdev_queue *txq; 4867 bool free_skb = true; 4868 int cpu, rc; 4869 4870 txq = netdev_core_pick_tx(dev, skb, NULL); 4871 cpu = smp_processor_id(); 4872 HARD_TX_LOCK(dev, txq, cpu); 4873 if (!netif_xmit_stopped(txq)) { 4874 rc = netdev_start_xmit(skb, dev, txq, 0); 4875 if (dev_xmit_complete(rc)) 4876 free_skb = false; 4877 } 4878 HARD_TX_UNLOCK(dev, txq); 4879 if (free_skb) { 4880 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4881 kfree_skb(skb); 4882 } 4883 } 4884 4885 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4886 4887 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4888 { 4889 if (xdp_prog) { 4890 struct xdp_buff xdp; 4891 u32 act; 4892 int err; 4893 4894 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4895 if (act != XDP_PASS) { 4896 switch (act) { 4897 case XDP_REDIRECT: 4898 err = xdp_do_generic_redirect(skb->dev, skb, 4899 &xdp, xdp_prog); 4900 if (err) 4901 goto out_redir; 4902 break; 4903 case XDP_TX: 4904 generic_xdp_tx(skb, xdp_prog); 4905 break; 4906 } 4907 return XDP_DROP; 4908 } 4909 } 4910 return XDP_PASS; 4911 out_redir: 4912 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4913 return XDP_DROP; 4914 } 4915 EXPORT_SYMBOL_GPL(do_xdp_generic); 4916 4917 static int netif_rx_internal(struct sk_buff *skb) 4918 { 4919 int ret; 4920 4921 net_timestamp_check(netdev_tstamp_prequeue, skb); 4922 4923 trace_netif_rx(skb); 4924 4925 #ifdef CONFIG_RPS 4926 if (static_branch_unlikely(&rps_needed)) { 4927 struct rps_dev_flow voidflow, *rflow = &voidflow; 4928 int cpu; 4929 4930 rcu_read_lock(); 4931 4932 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4933 if (cpu < 0) 4934 cpu = smp_processor_id(); 4935 4936 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4937 4938 rcu_read_unlock(); 4939 } else 4940 #endif 4941 { 4942 unsigned int qtail; 4943 4944 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4945 } 4946 return ret; 4947 } 4948 4949 /** 4950 * __netif_rx - Slightly optimized version of netif_rx 4951 * @skb: buffer to post 4952 * 4953 * This behaves as netif_rx except that it does not disable bottom halves. 4954 * As a result this function may only be invoked from the interrupt context 4955 * (either hard or soft interrupt). 4956 */ 4957 int __netif_rx(struct sk_buff *skb) 4958 { 4959 int ret; 4960 4961 lockdep_assert_once(hardirq_count() | softirq_count()); 4962 4963 trace_netif_rx_entry(skb); 4964 ret = netif_rx_internal(skb); 4965 trace_netif_rx_exit(ret); 4966 return ret; 4967 } 4968 EXPORT_SYMBOL(__netif_rx); 4969 4970 /** 4971 * netif_rx - post buffer to the network code 4972 * @skb: buffer to post 4973 * 4974 * This function receives a packet from a device driver and queues it for 4975 * the upper (protocol) levels to process via the backlog NAPI device. It 4976 * always succeeds. The buffer may be dropped during processing for 4977 * congestion control or by the protocol layers. 4978 * The network buffer is passed via the backlog NAPI device. Modern NIC 4979 * driver should use NAPI and GRO. 4980 * This function can used from interrupt and from process context. The 4981 * caller from process context must not disable interrupts before invoking 4982 * this function. 4983 * 4984 * return values: 4985 * NET_RX_SUCCESS (no congestion) 4986 * NET_RX_DROP (packet was dropped) 4987 * 4988 */ 4989 int netif_rx(struct sk_buff *skb) 4990 { 4991 bool need_bh_off = !(hardirq_count() | softirq_count()); 4992 int ret; 4993 4994 if (need_bh_off) 4995 local_bh_disable(); 4996 trace_netif_rx_entry(skb); 4997 ret = netif_rx_internal(skb); 4998 trace_netif_rx_exit(ret); 4999 if (need_bh_off) 5000 local_bh_enable(); 5001 return ret; 5002 } 5003 EXPORT_SYMBOL(netif_rx); 5004 5005 static __latent_entropy void net_tx_action(struct softirq_action *h) 5006 { 5007 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5008 5009 if (sd->completion_queue) { 5010 struct sk_buff *clist; 5011 5012 local_irq_disable(); 5013 clist = sd->completion_queue; 5014 sd->completion_queue = NULL; 5015 local_irq_enable(); 5016 5017 while (clist) { 5018 struct sk_buff *skb = clist; 5019 5020 clist = clist->next; 5021 5022 WARN_ON(refcount_read(&skb->users)); 5023 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5024 trace_consume_skb(skb); 5025 else 5026 trace_kfree_skb(skb, net_tx_action, 5027 SKB_DROP_REASON_NOT_SPECIFIED); 5028 5029 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5030 __kfree_skb(skb); 5031 else 5032 __kfree_skb_defer(skb); 5033 } 5034 } 5035 5036 if (sd->output_queue) { 5037 struct Qdisc *head; 5038 5039 local_irq_disable(); 5040 head = sd->output_queue; 5041 sd->output_queue = NULL; 5042 sd->output_queue_tailp = &sd->output_queue; 5043 local_irq_enable(); 5044 5045 rcu_read_lock(); 5046 5047 while (head) { 5048 struct Qdisc *q = head; 5049 spinlock_t *root_lock = NULL; 5050 5051 head = head->next_sched; 5052 5053 /* We need to make sure head->next_sched is read 5054 * before clearing __QDISC_STATE_SCHED 5055 */ 5056 smp_mb__before_atomic(); 5057 5058 if (!(q->flags & TCQ_F_NOLOCK)) { 5059 root_lock = qdisc_lock(q); 5060 spin_lock(root_lock); 5061 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5062 &q->state))) { 5063 /* There is a synchronize_net() between 5064 * STATE_DEACTIVATED flag being set and 5065 * qdisc_reset()/some_qdisc_is_busy() in 5066 * dev_deactivate(), so we can safely bail out 5067 * early here to avoid data race between 5068 * qdisc_deactivate() and some_qdisc_is_busy() 5069 * for lockless qdisc. 5070 */ 5071 clear_bit(__QDISC_STATE_SCHED, &q->state); 5072 continue; 5073 } 5074 5075 clear_bit(__QDISC_STATE_SCHED, &q->state); 5076 qdisc_run(q); 5077 if (root_lock) 5078 spin_unlock(root_lock); 5079 } 5080 5081 rcu_read_unlock(); 5082 } 5083 5084 xfrm_dev_backlog(sd); 5085 } 5086 5087 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5088 /* This hook is defined here for ATM LANE */ 5089 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5090 unsigned char *addr) __read_mostly; 5091 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5092 #endif 5093 5094 static inline struct sk_buff * 5095 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5096 struct net_device *orig_dev, bool *another) 5097 { 5098 #ifdef CONFIG_NET_CLS_ACT 5099 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5100 struct tcf_result cl_res; 5101 5102 /* If there's at least one ingress present somewhere (so 5103 * we get here via enabled static key), remaining devices 5104 * that are not configured with an ingress qdisc will bail 5105 * out here. 5106 */ 5107 if (!miniq) 5108 return skb; 5109 5110 if (*pt_prev) { 5111 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5112 *pt_prev = NULL; 5113 } 5114 5115 qdisc_skb_cb(skb)->pkt_len = skb->len; 5116 tc_skb_cb(skb)->mru = 0; 5117 tc_skb_cb(skb)->post_ct = false; 5118 skb->tc_at_ingress = 1; 5119 mini_qdisc_bstats_cpu_update(miniq, skb); 5120 5121 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5122 case TC_ACT_OK: 5123 case TC_ACT_RECLASSIFY: 5124 skb->tc_index = TC_H_MIN(cl_res.classid); 5125 break; 5126 case TC_ACT_SHOT: 5127 mini_qdisc_qstats_cpu_drop(miniq); 5128 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5129 return NULL; 5130 case TC_ACT_STOLEN: 5131 case TC_ACT_QUEUED: 5132 case TC_ACT_TRAP: 5133 consume_skb(skb); 5134 return NULL; 5135 case TC_ACT_REDIRECT: 5136 /* skb_mac_header check was done by cls/act_bpf, so 5137 * we can safely push the L2 header back before 5138 * redirecting to another netdev 5139 */ 5140 __skb_push(skb, skb->mac_len); 5141 if (skb_do_redirect(skb) == -EAGAIN) { 5142 __skb_pull(skb, skb->mac_len); 5143 *another = true; 5144 break; 5145 } 5146 return NULL; 5147 case TC_ACT_CONSUMED: 5148 return NULL; 5149 default: 5150 break; 5151 } 5152 #endif /* CONFIG_NET_CLS_ACT */ 5153 return skb; 5154 } 5155 5156 /** 5157 * netdev_is_rx_handler_busy - check if receive handler is registered 5158 * @dev: device to check 5159 * 5160 * Check if a receive handler is already registered for a given device. 5161 * Return true if there one. 5162 * 5163 * The caller must hold the rtnl_mutex. 5164 */ 5165 bool netdev_is_rx_handler_busy(struct net_device *dev) 5166 { 5167 ASSERT_RTNL(); 5168 return dev && rtnl_dereference(dev->rx_handler); 5169 } 5170 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5171 5172 /** 5173 * netdev_rx_handler_register - register receive handler 5174 * @dev: device to register a handler for 5175 * @rx_handler: receive handler to register 5176 * @rx_handler_data: data pointer that is used by rx handler 5177 * 5178 * Register a receive handler for a device. This handler will then be 5179 * called from __netif_receive_skb. A negative errno code is returned 5180 * on a failure. 5181 * 5182 * The caller must hold the rtnl_mutex. 5183 * 5184 * For a general description of rx_handler, see enum rx_handler_result. 5185 */ 5186 int netdev_rx_handler_register(struct net_device *dev, 5187 rx_handler_func_t *rx_handler, 5188 void *rx_handler_data) 5189 { 5190 if (netdev_is_rx_handler_busy(dev)) 5191 return -EBUSY; 5192 5193 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5194 return -EINVAL; 5195 5196 /* Note: rx_handler_data must be set before rx_handler */ 5197 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5198 rcu_assign_pointer(dev->rx_handler, rx_handler); 5199 5200 return 0; 5201 } 5202 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5203 5204 /** 5205 * netdev_rx_handler_unregister - unregister receive handler 5206 * @dev: device to unregister a handler from 5207 * 5208 * Unregister a receive handler from a device. 5209 * 5210 * The caller must hold the rtnl_mutex. 5211 */ 5212 void netdev_rx_handler_unregister(struct net_device *dev) 5213 { 5214 5215 ASSERT_RTNL(); 5216 RCU_INIT_POINTER(dev->rx_handler, NULL); 5217 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5218 * section has a guarantee to see a non NULL rx_handler_data 5219 * as well. 5220 */ 5221 synchronize_net(); 5222 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5223 } 5224 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5225 5226 /* 5227 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5228 * the special handling of PFMEMALLOC skbs. 5229 */ 5230 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5231 { 5232 switch (skb->protocol) { 5233 case htons(ETH_P_ARP): 5234 case htons(ETH_P_IP): 5235 case htons(ETH_P_IPV6): 5236 case htons(ETH_P_8021Q): 5237 case htons(ETH_P_8021AD): 5238 return true; 5239 default: 5240 return false; 5241 } 5242 } 5243 5244 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5245 int *ret, struct net_device *orig_dev) 5246 { 5247 if (nf_hook_ingress_active(skb)) { 5248 int ingress_retval; 5249 5250 if (*pt_prev) { 5251 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5252 *pt_prev = NULL; 5253 } 5254 5255 rcu_read_lock(); 5256 ingress_retval = nf_hook_ingress(skb); 5257 rcu_read_unlock(); 5258 return ingress_retval; 5259 } 5260 return 0; 5261 } 5262 5263 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5264 struct packet_type **ppt_prev) 5265 { 5266 struct packet_type *ptype, *pt_prev; 5267 rx_handler_func_t *rx_handler; 5268 struct sk_buff *skb = *pskb; 5269 struct net_device *orig_dev; 5270 bool deliver_exact = false; 5271 int ret = NET_RX_DROP; 5272 __be16 type; 5273 5274 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5275 5276 trace_netif_receive_skb(skb); 5277 5278 orig_dev = skb->dev; 5279 5280 skb_reset_network_header(skb); 5281 if (!skb_transport_header_was_set(skb)) 5282 skb_reset_transport_header(skb); 5283 skb_reset_mac_len(skb); 5284 5285 pt_prev = NULL; 5286 5287 another_round: 5288 skb->skb_iif = skb->dev->ifindex; 5289 5290 __this_cpu_inc(softnet_data.processed); 5291 5292 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5293 int ret2; 5294 5295 migrate_disable(); 5296 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5297 migrate_enable(); 5298 5299 if (ret2 != XDP_PASS) { 5300 ret = NET_RX_DROP; 5301 goto out; 5302 } 5303 } 5304 5305 if (eth_type_vlan(skb->protocol)) { 5306 skb = skb_vlan_untag(skb); 5307 if (unlikely(!skb)) 5308 goto out; 5309 } 5310 5311 if (skb_skip_tc_classify(skb)) 5312 goto skip_classify; 5313 5314 if (pfmemalloc) 5315 goto skip_taps; 5316 5317 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5318 if (pt_prev) 5319 ret = deliver_skb(skb, pt_prev, orig_dev); 5320 pt_prev = ptype; 5321 } 5322 5323 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5324 if (pt_prev) 5325 ret = deliver_skb(skb, pt_prev, orig_dev); 5326 pt_prev = ptype; 5327 } 5328 5329 skip_taps: 5330 #ifdef CONFIG_NET_INGRESS 5331 if (static_branch_unlikely(&ingress_needed_key)) { 5332 bool another = false; 5333 5334 nf_skip_egress(skb, true); 5335 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5336 &another); 5337 if (another) 5338 goto another_round; 5339 if (!skb) 5340 goto out; 5341 5342 nf_skip_egress(skb, false); 5343 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5344 goto out; 5345 } 5346 #endif 5347 skb_reset_redirect(skb); 5348 skip_classify: 5349 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5350 goto drop; 5351 5352 if (skb_vlan_tag_present(skb)) { 5353 if (pt_prev) { 5354 ret = deliver_skb(skb, pt_prev, orig_dev); 5355 pt_prev = NULL; 5356 } 5357 if (vlan_do_receive(&skb)) 5358 goto another_round; 5359 else if (unlikely(!skb)) 5360 goto out; 5361 } 5362 5363 rx_handler = rcu_dereference(skb->dev->rx_handler); 5364 if (rx_handler) { 5365 if (pt_prev) { 5366 ret = deliver_skb(skb, pt_prev, orig_dev); 5367 pt_prev = NULL; 5368 } 5369 switch (rx_handler(&skb)) { 5370 case RX_HANDLER_CONSUMED: 5371 ret = NET_RX_SUCCESS; 5372 goto out; 5373 case RX_HANDLER_ANOTHER: 5374 goto another_round; 5375 case RX_HANDLER_EXACT: 5376 deliver_exact = true; 5377 break; 5378 case RX_HANDLER_PASS: 5379 break; 5380 default: 5381 BUG(); 5382 } 5383 } 5384 5385 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5386 check_vlan_id: 5387 if (skb_vlan_tag_get_id(skb)) { 5388 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5389 * find vlan device. 5390 */ 5391 skb->pkt_type = PACKET_OTHERHOST; 5392 } else if (eth_type_vlan(skb->protocol)) { 5393 /* Outer header is 802.1P with vlan 0, inner header is 5394 * 802.1Q or 802.1AD and vlan_do_receive() above could 5395 * not find vlan dev for vlan id 0. 5396 */ 5397 __vlan_hwaccel_clear_tag(skb); 5398 skb = skb_vlan_untag(skb); 5399 if (unlikely(!skb)) 5400 goto out; 5401 if (vlan_do_receive(&skb)) 5402 /* After stripping off 802.1P header with vlan 0 5403 * vlan dev is found for inner header. 5404 */ 5405 goto another_round; 5406 else if (unlikely(!skb)) 5407 goto out; 5408 else 5409 /* We have stripped outer 802.1P vlan 0 header. 5410 * But could not find vlan dev. 5411 * check again for vlan id to set OTHERHOST. 5412 */ 5413 goto check_vlan_id; 5414 } 5415 /* Note: we might in the future use prio bits 5416 * and set skb->priority like in vlan_do_receive() 5417 * For the time being, just ignore Priority Code Point 5418 */ 5419 __vlan_hwaccel_clear_tag(skb); 5420 } 5421 5422 type = skb->protocol; 5423 5424 /* deliver only exact match when indicated */ 5425 if (likely(!deliver_exact)) { 5426 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5427 &ptype_base[ntohs(type) & 5428 PTYPE_HASH_MASK]); 5429 } 5430 5431 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5432 &orig_dev->ptype_specific); 5433 5434 if (unlikely(skb->dev != orig_dev)) { 5435 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5436 &skb->dev->ptype_specific); 5437 } 5438 5439 if (pt_prev) { 5440 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5441 goto drop; 5442 *ppt_prev = pt_prev; 5443 } else { 5444 drop: 5445 if (!deliver_exact) 5446 dev_core_stats_rx_dropped_inc(skb->dev); 5447 else 5448 dev_core_stats_rx_nohandler_inc(skb->dev); 5449 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5450 /* Jamal, now you will not able to escape explaining 5451 * me how you were going to use this. :-) 5452 */ 5453 ret = NET_RX_DROP; 5454 } 5455 5456 out: 5457 /* The invariant here is that if *ppt_prev is not NULL 5458 * then skb should also be non-NULL. 5459 * 5460 * Apparently *ppt_prev assignment above holds this invariant due to 5461 * skb dereferencing near it. 5462 */ 5463 *pskb = skb; 5464 return ret; 5465 } 5466 5467 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5468 { 5469 struct net_device *orig_dev = skb->dev; 5470 struct packet_type *pt_prev = NULL; 5471 int ret; 5472 5473 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5474 if (pt_prev) 5475 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5476 skb->dev, pt_prev, orig_dev); 5477 return ret; 5478 } 5479 5480 /** 5481 * netif_receive_skb_core - special purpose version of netif_receive_skb 5482 * @skb: buffer to process 5483 * 5484 * More direct receive version of netif_receive_skb(). It should 5485 * only be used by callers that have a need to skip RPS and Generic XDP. 5486 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5487 * 5488 * This function may only be called from softirq context and interrupts 5489 * should be enabled. 5490 * 5491 * Return values (usually ignored): 5492 * NET_RX_SUCCESS: no congestion 5493 * NET_RX_DROP: packet was dropped 5494 */ 5495 int netif_receive_skb_core(struct sk_buff *skb) 5496 { 5497 int ret; 5498 5499 rcu_read_lock(); 5500 ret = __netif_receive_skb_one_core(skb, false); 5501 rcu_read_unlock(); 5502 5503 return ret; 5504 } 5505 EXPORT_SYMBOL(netif_receive_skb_core); 5506 5507 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5508 struct packet_type *pt_prev, 5509 struct net_device *orig_dev) 5510 { 5511 struct sk_buff *skb, *next; 5512 5513 if (!pt_prev) 5514 return; 5515 if (list_empty(head)) 5516 return; 5517 if (pt_prev->list_func != NULL) 5518 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5519 ip_list_rcv, head, pt_prev, orig_dev); 5520 else 5521 list_for_each_entry_safe(skb, next, head, list) { 5522 skb_list_del_init(skb); 5523 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5524 } 5525 } 5526 5527 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5528 { 5529 /* Fast-path assumptions: 5530 * - There is no RX handler. 5531 * - Only one packet_type matches. 5532 * If either of these fails, we will end up doing some per-packet 5533 * processing in-line, then handling the 'last ptype' for the whole 5534 * sublist. This can't cause out-of-order delivery to any single ptype, 5535 * because the 'last ptype' must be constant across the sublist, and all 5536 * other ptypes are handled per-packet. 5537 */ 5538 /* Current (common) ptype of sublist */ 5539 struct packet_type *pt_curr = NULL; 5540 /* Current (common) orig_dev of sublist */ 5541 struct net_device *od_curr = NULL; 5542 struct list_head sublist; 5543 struct sk_buff *skb, *next; 5544 5545 INIT_LIST_HEAD(&sublist); 5546 list_for_each_entry_safe(skb, next, head, list) { 5547 struct net_device *orig_dev = skb->dev; 5548 struct packet_type *pt_prev = NULL; 5549 5550 skb_list_del_init(skb); 5551 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5552 if (!pt_prev) 5553 continue; 5554 if (pt_curr != pt_prev || od_curr != orig_dev) { 5555 /* dispatch old sublist */ 5556 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5557 /* start new sublist */ 5558 INIT_LIST_HEAD(&sublist); 5559 pt_curr = pt_prev; 5560 od_curr = orig_dev; 5561 } 5562 list_add_tail(&skb->list, &sublist); 5563 } 5564 5565 /* dispatch final sublist */ 5566 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5567 } 5568 5569 static int __netif_receive_skb(struct sk_buff *skb) 5570 { 5571 int ret; 5572 5573 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5574 unsigned int noreclaim_flag; 5575 5576 /* 5577 * PFMEMALLOC skbs are special, they should 5578 * - be delivered to SOCK_MEMALLOC sockets only 5579 * - stay away from userspace 5580 * - have bounded memory usage 5581 * 5582 * Use PF_MEMALLOC as this saves us from propagating the allocation 5583 * context down to all allocation sites. 5584 */ 5585 noreclaim_flag = memalloc_noreclaim_save(); 5586 ret = __netif_receive_skb_one_core(skb, true); 5587 memalloc_noreclaim_restore(noreclaim_flag); 5588 } else 5589 ret = __netif_receive_skb_one_core(skb, false); 5590 5591 return ret; 5592 } 5593 5594 static void __netif_receive_skb_list(struct list_head *head) 5595 { 5596 unsigned long noreclaim_flag = 0; 5597 struct sk_buff *skb, *next; 5598 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5599 5600 list_for_each_entry_safe(skb, next, head, list) { 5601 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5602 struct list_head sublist; 5603 5604 /* Handle the previous sublist */ 5605 list_cut_before(&sublist, head, &skb->list); 5606 if (!list_empty(&sublist)) 5607 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5608 pfmemalloc = !pfmemalloc; 5609 /* See comments in __netif_receive_skb */ 5610 if (pfmemalloc) 5611 noreclaim_flag = memalloc_noreclaim_save(); 5612 else 5613 memalloc_noreclaim_restore(noreclaim_flag); 5614 } 5615 } 5616 /* Handle the remaining sublist */ 5617 if (!list_empty(head)) 5618 __netif_receive_skb_list_core(head, pfmemalloc); 5619 /* Restore pflags */ 5620 if (pfmemalloc) 5621 memalloc_noreclaim_restore(noreclaim_flag); 5622 } 5623 5624 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5625 { 5626 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5627 struct bpf_prog *new = xdp->prog; 5628 int ret = 0; 5629 5630 switch (xdp->command) { 5631 case XDP_SETUP_PROG: 5632 rcu_assign_pointer(dev->xdp_prog, new); 5633 if (old) 5634 bpf_prog_put(old); 5635 5636 if (old && !new) { 5637 static_branch_dec(&generic_xdp_needed_key); 5638 } else if (new && !old) { 5639 static_branch_inc(&generic_xdp_needed_key); 5640 dev_disable_lro(dev); 5641 dev_disable_gro_hw(dev); 5642 } 5643 break; 5644 5645 default: 5646 ret = -EINVAL; 5647 break; 5648 } 5649 5650 return ret; 5651 } 5652 5653 static int netif_receive_skb_internal(struct sk_buff *skb) 5654 { 5655 int ret; 5656 5657 net_timestamp_check(netdev_tstamp_prequeue, skb); 5658 5659 if (skb_defer_rx_timestamp(skb)) 5660 return NET_RX_SUCCESS; 5661 5662 rcu_read_lock(); 5663 #ifdef CONFIG_RPS 5664 if (static_branch_unlikely(&rps_needed)) { 5665 struct rps_dev_flow voidflow, *rflow = &voidflow; 5666 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5667 5668 if (cpu >= 0) { 5669 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5670 rcu_read_unlock(); 5671 return ret; 5672 } 5673 } 5674 #endif 5675 ret = __netif_receive_skb(skb); 5676 rcu_read_unlock(); 5677 return ret; 5678 } 5679 5680 void netif_receive_skb_list_internal(struct list_head *head) 5681 { 5682 struct sk_buff *skb, *next; 5683 struct list_head sublist; 5684 5685 INIT_LIST_HEAD(&sublist); 5686 list_for_each_entry_safe(skb, next, head, list) { 5687 net_timestamp_check(netdev_tstamp_prequeue, skb); 5688 skb_list_del_init(skb); 5689 if (!skb_defer_rx_timestamp(skb)) 5690 list_add_tail(&skb->list, &sublist); 5691 } 5692 list_splice_init(&sublist, head); 5693 5694 rcu_read_lock(); 5695 #ifdef CONFIG_RPS 5696 if (static_branch_unlikely(&rps_needed)) { 5697 list_for_each_entry_safe(skb, next, head, list) { 5698 struct rps_dev_flow voidflow, *rflow = &voidflow; 5699 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5700 5701 if (cpu >= 0) { 5702 /* Will be handled, remove from list */ 5703 skb_list_del_init(skb); 5704 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5705 } 5706 } 5707 } 5708 #endif 5709 __netif_receive_skb_list(head); 5710 rcu_read_unlock(); 5711 } 5712 5713 /** 5714 * netif_receive_skb - process receive buffer from network 5715 * @skb: buffer to process 5716 * 5717 * netif_receive_skb() is the main receive data processing function. 5718 * It always succeeds. The buffer may be dropped during processing 5719 * for congestion control or by the protocol layers. 5720 * 5721 * This function may only be called from softirq context and interrupts 5722 * should be enabled. 5723 * 5724 * Return values (usually ignored): 5725 * NET_RX_SUCCESS: no congestion 5726 * NET_RX_DROP: packet was dropped 5727 */ 5728 int netif_receive_skb(struct sk_buff *skb) 5729 { 5730 int ret; 5731 5732 trace_netif_receive_skb_entry(skb); 5733 5734 ret = netif_receive_skb_internal(skb); 5735 trace_netif_receive_skb_exit(ret); 5736 5737 return ret; 5738 } 5739 EXPORT_SYMBOL(netif_receive_skb); 5740 5741 /** 5742 * netif_receive_skb_list - process many receive buffers from network 5743 * @head: list of skbs to process. 5744 * 5745 * Since return value of netif_receive_skb() is normally ignored, and 5746 * wouldn't be meaningful for a list, this function returns void. 5747 * 5748 * This function may only be called from softirq context and interrupts 5749 * should be enabled. 5750 */ 5751 void netif_receive_skb_list(struct list_head *head) 5752 { 5753 struct sk_buff *skb; 5754 5755 if (list_empty(head)) 5756 return; 5757 if (trace_netif_receive_skb_list_entry_enabled()) { 5758 list_for_each_entry(skb, head, list) 5759 trace_netif_receive_skb_list_entry(skb); 5760 } 5761 netif_receive_skb_list_internal(head); 5762 trace_netif_receive_skb_list_exit(0); 5763 } 5764 EXPORT_SYMBOL(netif_receive_skb_list); 5765 5766 static DEFINE_PER_CPU(struct work_struct, flush_works); 5767 5768 /* Network device is going away, flush any packets still pending */ 5769 static void flush_backlog(struct work_struct *work) 5770 { 5771 struct sk_buff *skb, *tmp; 5772 struct softnet_data *sd; 5773 5774 local_bh_disable(); 5775 sd = this_cpu_ptr(&softnet_data); 5776 5777 rps_lock_irq_disable(sd); 5778 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5779 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5780 __skb_unlink(skb, &sd->input_pkt_queue); 5781 dev_kfree_skb_irq(skb); 5782 input_queue_head_incr(sd); 5783 } 5784 } 5785 rps_unlock_irq_enable(sd); 5786 5787 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5788 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5789 __skb_unlink(skb, &sd->process_queue); 5790 kfree_skb(skb); 5791 input_queue_head_incr(sd); 5792 } 5793 } 5794 local_bh_enable(); 5795 } 5796 5797 static bool flush_required(int cpu) 5798 { 5799 #if IS_ENABLED(CONFIG_RPS) 5800 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5801 bool do_flush; 5802 5803 rps_lock_irq_disable(sd); 5804 5805 /* as insertion into process_queue happens with the rps lock held, 5806 * process_queue access may race only with dequeue 5807 */ 5808 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5809 !skb_queue_empty_lockless(&sd->process_queue); 5810 rps_unlock_irq_enable(sd); 5811 5812 return do_flush; 5813 #endif 5814 /* without RPS we can't safely check input_pkt_queue: during a 5815 * concurrent remote skb_queue_splice() we can detect as empty both 5816 * input_pkt_queue and process_queue even if the latter could end-up 5817 * containing a lot of packets. 5818 */ 5819 return true; 5820 } 5821 5822 static void flush_all_backlogs(void) 5823 { 5824 static cpumask_t flush_cpus; 5825 unsigned int cpu; 5826 5827 /* since we are under rtnl lock protection we can use static data 5828 * for the cpumask and avoid allocating on stack the possibly 5829 * large mask 5830 */ 5831 ASSERT_RTNL(); 5832 5833 cpus_read_lock(); 5834 5835 cpumask_clear(&flush_cpus); 5836 for_each_online_cpu(cpu) { 5837 if (flush_required(cpu)) { 5838 queue_work_on(cpu, system_highpri_wq, 5839 per_cpu_ptr(&flush_works, cpu)); 5840 cpumask_set_cpu(cpu, &flush_cpus); 5841 } 5842 } 5843 5844 /* we can have in flight packet[s] on the cpus we are not flushing, 5845 * synchronize_net() in unregister_netdevice_many() will take care of 5846 * them 5847 */ 5848 for_each_cpu(cpu, &flush_cpus) 5849 flush_work(per_cpu_ptr(&flush_works, cpu)); 5850 5851 cpus_read_unlock(); 5852 } 5853 5854 static void net_rps_send_ipi(struct softnet_data *remsd) 5855 { 5856 #ifdef CONFIG_RPS 5857 while (remsd) { 5858 struct softnet_data *next = remsd->rps_ipi_next; 5859 5860 if (cpu_online(remsd->cpu)) 5861 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5862 remsd = next; 5863 } 5864 #endif 5865 } 5866 5867 /* 5868 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5869 * Note: called with local irq disabled, but exits with local irq enabled. 5870 */ 5871 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5872 { 5873 #ifdef CONFIG_RPS 5874 struct softnet_data *remsd = sd->rps_ipi_list; 5875 5876 if (remsd) { 5877 sd->rps_ipi_list = NULL; 5878 5879 local_irq_enable(); 5880 5881 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5882 net_rps_send_ipi(remsd); 5883 } else 5884 #endif 5885 local_irq_enable(); 5886 } 5887 5888 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5889 { 5890 #ifdef CONFIG_RPS 5891 return sd->rps_ipi_list != NULL; 5892 #else 5893 return false; 5894 #endif 5895 } 5896 5897 static int process_backlog(struct napi_struct *napi, int quota) 5898 { 5899 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5900 bool again = true; 5901 int work = 0; 5902 5903 /* Check if we have pending ipi, its better to send them now, 5904 * not waiting net_rx_action() end. 5905 */ 5906 if (sd_has_rps_ipi_waiting(sd)) { 5907 local_irq_disable(); 5908 net_rps_action_and_irq_enable(sd); 5909 } 5910 5911 napi->weight = dev_rx_weight; 5912 while (again) { 5913 struct sk_buff *skb; 5914 5915 while ((skb = __skb_dequeue(&sd->process_queue))) { 5916 rcu_read_lock(); 5917 __netif_receive_skb(skb); 5918 rcu_read_unlock(); 5919 input_queue_head_incr(sd); 5920 if (++work >= quota) 5921 return work; 5922 5923 } 5924 5925 rps_lock_irq_disable(sd); 5926 if (skb_queue_empty(&sd->input_pkt_queue)) { 5927 /* 5928 * Inline a custom version of __napi_complete(). 5929 * only current cpu owns and manipulates this napi, 5930 * and NAPI_STATE_SCHED is the only possible flag set 5931 * on backlog. 5932 * We can use a plain write instead of clear_bit(), 5933 * and we dont need an smp_mb() memory barrier. 5934 */ 5935 napi->state = 0; 5936 again = false; 5937 } else { 5938 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5939 &sd->process_queue); 5940 } 5941 rps_unlock_irq_enable(sd); 5942 } 5943 5944 return work; 5945 } 5946 5947 /** 5948 * __napi_schedule - schedule for receive 5949 * @n: entry to schedule 5950 * 5951 * The entry's receive function will be scheduled to run. 5952 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5953 */ 5954 void __napi_schedule(struct napi_struct *n) 5955 { 5956 unsigned long flags; 5957 5958 local_irq_save(flags); 5959 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5960 local_irq_restore(flags); 5961 } 5962 EXPORT_SYMBOL(__napi_schedule); 5963 5964 /** 5965 * napi_schedule_prep - check if napi can be scheduled 5966 * @n: napi context 5967 * 5968 * Test if NAPI routine is already running, and if not mark 5969 * it as running. This is used as a condition variable to 5970 * insure only one NAPI poll instance runs. We also make 5971 * sure there is no pending NAPI disable. 5972 */ 5973 bool napi_schedule_prep(struct napi_struct *n) 5974 { 5975 unsigned long val, new; 5976 5977 do { 5978 val = READ_ONCE(n->state); 5979 if (unlikely(val & NAPIF_STATE_DISABLE)) 5980 return false; 5981 new = val | NAPIF_STATE_SCHED; 5982 5983 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5984 * This was suggested by Alexander Duyck, as compiler 5985 * emits better code than : 5986 * if (val & NAPIF_STATE_SCHED) 5987 * new |= NAPIF_STATE_MISSED; 5988 */ 5989 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5990 NAPIF_STATE_MISSED; 5991 } while (cmpxchg(&n->state, val, new) != val); 5992 5993 return !(val & NAPIF_STATE_SCHED); 5994 } 5995 EXPORT_SYMBOL(napi_schedule_prep); 5996 5997 /** 5998 * __napi_schedule_irqoff - schedule for receive 5999 * @n: entry to schedule 6000 * 6001 * Variant of __napi_schedule() assuming hard irqs are masked. 6002 * 6003 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6004 * because the interrupt disabled assumption might not be true 6005 * due to force-threaded interrupts and spinlock substitution. 6006 */ 6007 void __napi_schedule_irqoff(struct napi_struct *n) 6008 { 6009 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6010 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6011 else 6012 __napi_schedule(n); 6013 } 6014 EXPORT_SYMBOL(__napi_schedule_irqoff); 6015 6016 bool napi_complete_done(struct napi_struct *n, int work_done) 6017 { 6018 unsigned long flags, val, new, timeout = 0; 6019 bool ret = true; 6020 6021 /* 6022 * 1) Don't let napi dequeue from the cpu poll list 6023 * just in case its running on a different cpu. 6024 * 2) If we are busy polling, do nothing here, we have 6025 * the guarantee we will be called later. 6026 */ 6027 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6028 NAPIF_STATE_IN_BUSY_POLL))) 6029 return false; 6030 6031 if (work_done) { 6032 if (n->gro_bitmask) 6033 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6034 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6035 } 6036 if (n->defer_hard_irqs_count > 0) { 6037 n->defer_hard_irqs_count--; 6038 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6039 if (timeout) 6040 ret = false; 6041 } 6042 if (n->gro_bitmask) { 6043 /* When the NAPI instance uses a timeout and keeps postponing 6044 * it, we need to bound somehow the time packets are kept in 6045 * the GRO layer 6046 */ 6047 napi_gro_flush(n, !!timeout); 6048 } 6049 6050 gro_normal_list(n); 6051 6052 if (unlikely(!list_empty(&n->poll_list))) { 6053 /* If n->poll_list is not empty, we need to mask irqs */ 6054 local_irq_save(flags); 6055 list_del_init(&n->poll_list); 6056 local_irq_restore(flags); 6057 } 6058 6059 do { 6060 val = READ_ONCE(n->state); 6061 6062 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6063 6064 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6065 NAPIF_STATE_SCHED_THREADED | 6066 NAPIF_STATE_PREFER_BUSY_POLL); 6067 6068 /* If STATE_MISSED was set, leave STATE_SCHED set, 6069 * because we will call napi->poll() one more time. 6070 * This C code was suggested by Alexander Duyck to help gcc. 6071 */ 6072 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6073 NAPIF_STATE_SCHED; 6074 } while (cmpxchg(&n->state, val, new) != val); 6075 6076 if (unlikely(val & NAPIF_STATE_MISSED)) { 6077 __napi_schedule(n); 6078 return false; 6079 } 6080 6081 if (timeout) 6082 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6083 HRTIMER_MODE_REL_PINNED); 6084 return ret; 6085 } 6086 EXPORT_SYMBOL(napi_complete_done); 6087 6088 /* must be called under rcu_read_lock(), as we dont take a reference */ 6089 static struct napi_struct *napi_by_id(unsigned int napi_id) 6090 { 6091 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6092 struct napi_struct *napi; 6093 6094 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6095 if (napi->napi_id == napi_id) 6096 return napi; 6097 6098 return NULL; 6099 } 6100 6101 #if defined(CONFIG_NET_RX_BUSY_POLL) 6102 6103 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6104 { 6105 if (!skip_schedule) { 6106 gro_normal_list(napi); 6107 __napi_schedule(napi); 6108 return; 6109 } 6110 6111 if (napi->gro_bitmask) { 6112 /* flush too old packets 6113 * If HZ < 1000, flush all packets. 6114 */ 6115 napi_gro_flush(napi, HZ >= 1000); 6116 } 6117 6118 gro_normal_list(napi); 6119 clear_bit(NAPI_STATE_SCHED, &napi->state); 6120 } 6121 6122 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6123 u16 budget) 6124 { 6125 bool skip_schedule = false; 6126 unsigned long timeout; 6127 int rc; 6128 6129 /* Busy polling means there is a high chance device driver hard irq 6130 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6131 * set in napi_schedule_prep(). 6132 * Since we are about to call napi->poll() once more, we can safely 6133 * clear NAPI_STATE_MISSED. 6134 * 6135 * Note: x86 could use a single "lock and ..." instruction 6136 * to perform these two clear_bit() 6137 */ 6138 clear_bit(NAPI_STATE_MISSED, &napi->state); 6139 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6140 6141 local_bh_disable(); 6142 6143 if (prefer_busy_poll) { 6144 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6145 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6146 if (napi->defer_hard_irqs_count && timeout) { 6147 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6148 skip_schedule = true; 6149 } 6150 } 6151 6152 /* All we really want here is to re-enable device interrupts. 6153 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6154 */ 6155 rc = napi->poll(napi, budget); 6156 /* We can't gro_normal_list() here, because napi->poll() might have 6157 * rearmed the napi (napi_complete_done()) in which case it could 6158 * already be running on another CPU. 6159 */ 6160 trace_napi_poll(napi, rc, budget); 6161 netpoll_poll_unlock(have_poll_lock); 6162 if (rc == budget) 6163 __busy_poll_stop(napi, skip_schedule); 6164 local_bh_enable(); 6165 } 6166 6167 void napi_busy_loop(unsigned int napi_id, 6168 bool (*loop_end)(void *, unsigned long), 6169 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6170 { 6171 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6172 int (*napi_poll)(struct napi_struct *napi, int budget); 6173 void *have_poll_lock = NULL; 6174 struct napi_struct *napi; 6175 6176 restart: 6177 napi_poll = NULL; 6178 6179 rcu_read_lock(); 6180 6181 napi = napi_by_id(napi_id); 6182 if (!napi) 6183 goto out; 6184 6185 preempt_disable(); 6186 for (;;) { 6187 int work = 0; 6188 6189 local_bh_disable(); 6190 if (!napi_poll) { 6191 unsigned long val = READ_ONCE(napi->state); 6192 6193 /* If multiple threads are competing for this napi, 6194 * we avoid dirtying napi->state as much as we can. 6195 */ 6196 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6197 NAPIF_STATE_IN_BUSY_POLL)) { 6198 if (prefer_busy_poll) 6199 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6200 goto count; 6201 } 6202 if (cmpxchg(&napi->state, val, 6203 val | NAPIF_STATE_IN_BUSY_POLL | 6204 NAPIF_STATE_SCHED) != val) { 6205 if (prefer_busy_poll) 6206 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6207 goto count; 6208 } 6209 have_poll_lock = netpoll_poll_lock(napi); 6210 napi_poll = napi->poll; 6211 } 6212 work = napi_poll(napi, budget); 6213 trace_napi_poll(napi, work, budget); 6214 gro_normal_list(napi); 6215 count: 6216 if (work > 0) 6217 __NET_ADD_STATS(dev_net(napi->dev), 6218 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6219 local_bh_enable(); 6220 6221 if (!loop_end || loop_end(loop_end_arg, start_time)) 6222 break; 6223 6224 if (unlikely(need_resched())) { 6225 if (napi_poll) 6226 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6227 preempt_enable(); 6228 rcu_read_unlock(); 6229 cond_resched(); 6230 if (loop_end(loop_end_arg, start_time)) 6231 return; 6232 goto restart; 6233 } 6234 cpu_relax(); 6235 } 6236 if (napi_poll) 6237 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6238 preempt_enable(); 6239 out: 6240 rcu_read_unlock(); 6241 } 6242 EXPORT_SYMBOL(napi_busy_loop); 6243 6244 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6245 6246 static void napi_hash_add(struct napi_struct *napi) 6247 { 6248 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6249 return; 6250 6251 spin_lock(&napi_hash_lock); 6252 6253 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6254 do { 6255 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6256 napi_gen_id = MIN_NAPI_ID; 6257 } while (napi_by_id(napi_gen_id)); 6258 napi->napi_id = napi_gen_id; 6259 6260 hlist_add_head_rcu(&napi->napi_hash_node, 6261 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6262 6263 spin_unlock(&napi_hash_lock); 6264 } 6265 6266 /* Warning : caller is responsible to make sure rcu grace period 6267 * is respected before freeing memory containing @napi 6268 */ 6269 static void napi_hash_del(struct napi_struct *napi) 6270 { 6271 spin_lock(&napi_hash_lock); 6272 6273 hlist_del_init_rcu(&napi->napi_hash_node); 6274 6275 spin_unlock(&napi_hash_lock); 6276 } 6277 6278 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6279 { 6280 struct napi_struct *napi; 6281 6282 napi = container_of(timer, struct napi_struct, timer); 6283 6284 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6285 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6286 */ 6287 if (!napi_disable_pending(napi) && 6288 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6289 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6290 __napi_schedule_irqoff(napi); 6291 } 6292 6293 return HRTIMER_NORESTART; 6294 } 6295 6296 static void init_gro_hash(struct napi_struct *napi) 6297 { 6298 int i; 6299 6300 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6301 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6302 napi->gro_hash[i].count = 0; 6303 } 6304 napi->gro_bitmask = 0; 6305 } 6306 6307 int dev_set_threaded(struct net_device *dev, bool threaded) 6308 { 6309 struct napi_struct *napi; 6310 int err = 0; 6311 6312 if (dev->threaded == threaded) 6313 return 0; 6314 6315 if (threaded) { 6316 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6317 if (!napi->thread) { 6318 err = napi_kthread_create(napi); 6319 if (err) { 6320 threaded = false; 6321 break; 6322 } 6323 } 6324 } 6325 } 6326 6327 dev->threaded = threaded; 6328 6329 /* Make sure kthread is created before THREADED bit 6330 * is set. 6331 */ 6332 smp_mb__before_atomic(); 6333 6334 /* Setting/unsetting threaded mode on a napi might not immediately 6335 * take effect, if the current napi instance is actively being 6336 * polled. In this case, the switch between threaded mode and 6337 * softirq mode will happen in the next round of napi_schedule(). 6338 * This should not cause hiccups/stalls to the live traffic. 6339 */ 6340 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6341 if (threaded) 6342 set_bit(NAPI_STATE_THREADED, &napi->state); 6343 else 6344 clear_bit(NAPI_STATE_THREADED, &napi->state); 6345 } 6346 6347 return err; 6348 } 6349 EXPORT_SYMBOL(dev_set_threaded); 6350 6351 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6352 int (*poll)(struct napi_struct *, int), int weight) 6353 { 6354 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6355 return; 6356 6357 INIT_LIST_HEAD(&napi->poll_list); 6358 INIT_HLIST_NODE(&napi->napi_hash_node); 6359 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6360 napi->timer.function = napi_watchdog; 6361 init_gro_hash(napi); 6362 napi->skb = NULL; 6363 INIT_LIST_HEAD(&napi->rx_list); 6364 napi->rx_count = 0; 6365 napi->poll = poll; 6366 if (weight > NAPI_POLL_WEIGHT) 6367 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6368 weight); 6369 napi->weight = weight; 6370 napi->dev = dev; 6371 #ifdef CONFIG_NETPOLL 6372 napi->poll_owner = -1; 6373 #endif 6374 set_bit(NAPI_STATE_SCHED, &napi->state); 6375 set_bit(NAPI_STATE_NPSVC, &napi->state); 6376 list_add_rcu(&napi->dev_list, &dev->napi_list); 6377 napi_hash_add(napi); 6378 /* Create kthread for this napi if dev->threaded is set. 6379 * Clear dev->threaded if kthread creation failed so that 6380 * threaded mode will not be enabled in napi_enable(). 6381 */ 6382 if (dev->threaded && napi_kthread_create(napi)) 6383 dev->threaded = 0; 6384 } 6385 EXPORT_SYMBOL(netif_napi_add_weight); 6386 6387 void napi_disable(struct napi_struct *n) 6388 { 6389 unsigned long val, new; 6390 6391 might_sleep(); 6392 set_bit(NAPI_STATE_DISABLE, &n->state); 6393 6394 for ( ; ; ) { 6395 val = READ_ONCE(n->state); 6396 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6397 usleep_range(20, 200); 6398 continue; 6399 } 6400 6401 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6402 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6403 6404 if (cmpxchg(&n->state, val, new) == val) 6405 break; 6406 } 6407 6408 hrtimer_cancel(&n->timer); 6409 6410 clear_bit(NAPI_STATE_DISABLE, &n->state); 6411 } 6412 EXPORT_SYMBOL(napi_disable); 6413 6414 /** 6415 * napi_enable - enable NAPI scheduling 6416 * @n: NAPI context 6417 * 6418 * Resume NAPI from being scheduled on this context. 6419 * Must be paired with napi_disable. 6420 */ 6421 void napi_enable(struct napi_struct *n) 6422 { 6423 unsigned long val, new; 6424 6425 do { 6426 val = READ_ONCE(n->state); 6427 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6428 6429 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6430 if (n->dev->threaded && n->thread) 6431 new |= NAPIF_STATE_THREADED; 6432 } while (cmpxchg(&n->state, val, new) != val); 6433 } 6434 EXPORT_SYMBOL(napi_enable); 6435 6436 static void flush_gro_hash(struct napi_struct *napi) 6437 { 6438 int i; 6439 6440 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6441 struct sk_buff *skb, *n; 6442 6443 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6444 kfree_skb(skb); 6445 napi->gro_hash[i].count = 0; 6446 } 6447 } 6448 6449 /* Must be called in process context */ 6450 void __netif_napi_del(struct napi_struct *napi) 6451 { 6452 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6453 return; 6454 6455 napi_hash_del(napi); 6456 list_del_rcu(&napi->dev_list); 6457 napi_free_frags(napi); 6458 6459 flush_gro_hash(napi); 6460 napi->gro_bitmask = 0; 6461 6462 if (napi->thread) { 6463 kthread_stop(napi->thread); 6464 napi->thread = NULL; 6465 } 6466 } 6467 EXPORT_SYMBOL(__netif_napi_del); 6468 6469 static int __napi_poll(struct napi_struct *n, bool *repoll) 6470 { 6471 int work, weight; 6472 6473 weight = n->weight; 6474 6475 /* This NAPI_STATE_SCHED test is for avoiding a race 6476 * with netpoll's poll_napi(). Only the entity which 6477 * obtains the lock and sees NAPI_STATE_SCHED set will 6478 * actually make the ->poll() call. Therefore we avoid 6479 * accidentally calling ->poll() when NAPI is not scheduled. 6480 */ 6481 work = 0; 6482 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6483 work = n->poll(n, weight); 6484 trace_napi_poll(n, work, weight); 6485 } 6486 6487 if (unlikely(work > weight)) 6488 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6489 n->poll, work, weight); 6490 6491 if (likely(work < weight)) 6492 return work; 6493 6494 /* Drivers must not modify the NAPI state if they 6495 * consume the entire weight. In such cases this code 6496 * still "owns" the NAPI instance and therefore can 6497 * move the instance around on the list at-will. 6498 */ 6499 if (unlikely(napi_disable_pending(n))) { 6500 napi_complete(n); 6501 return work; 6502 } 6503 6504 /* The NAPI context has more processing work, but busy-polling 6505 * is preferred. Exit early. 6506 */ 6507 if (napi_prefer_busy_poll(n)) { 6508 if (napi_complete_done(n, work)) { 6509 /* If timeout is not set, we need to make sure 6510 * that the NAPI is re-scheduled. 6511 */ 6512 napi_schedule(n); 6513 } 6514 return work; 6515 } 6516 6517 if (n->gro_bitmask) { 6518 /* flush too old packets 6519 * If HZ < 1000, flush all packets. 6520 */ 6521 napi_gro_flush(n, HZ >= 1000); 6522 } 6523 6524 gro_normal_list(n); 6525 6526 /* Some drivers may have called napi_schedule 6527 * prior to exhausting their budget. 6528 */ 6529 if (unlikely(!list_empty(&n->poll_list))) { 6530 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6531 n->dev ? n->dev->name : "backlog"); 6532 return work; 6533 } 6534 6535 *repoll = true; 6536 6537 return work; 6538 } 6539 6540 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6541 { 6542 bool do_repoll = false; 6543 void *have; 6544 int work; 6545 6546 list_del_init(&n->poll_list); 6547 6548 have = netpoll_poll_lock(n); 6549 6550 work = __napi_poll(n, &do_repoll); 6551 6552 if (do_repoll) 6553 list_add_tail(&n->poll_list, repoll); 6554 6555 netpoll_poll_unlock(have); 6556 6557 return work; 6558 } 6559 6560 static int napi_thread_wait(struct napi_struct *napi) 6561 { 6562 bool woken = false; 6563 6564 set_current_state(TASK_INTERRUPTIBLE); 6565 6566 while (!kthread_should_stop()) { 6567 /* Testing SCHED_THREADED bit here to make sure the current 6568 * kthread owns this napi and could poll on this napi. 6569 * Testing SCHED bit is not enough because SCHED bit might be 6570 * set by some other busy poll thread or by napi_disable(). 6571 */ 6572 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6573 WARN_ON(!list_empty(&napi->poll_list)); 6574 __set_current_state(TASK_RUNNING); 6575 return 0; 6576 } 6577 6578 schedule(); 6579 /* woken being true indicates this thread owns this napi. */ 6580 woken = true; 6581 set_current_state(TASK_INTERRUPTIBLE); 6582 } 6583 __set_current_state(TASK_RUNNING); 6584 6585 return -1; 6586 } 6587 6588 static int napi_threaded_poll(void *data) 6589 { 6590 struct napi_struct *napi = data; 6591 void *have; 6592 6593 while (!napi_thread_wait(napi)) { 6594 for (;;) { 6595 bool repoll = false; 6596 6597 local_bh_disable(); 6598 6599 have = netpoll_poll_lock(napi); 6600 __napi_poll(napi, &repoll); 6601 netpoll_poll_unlock(have); 6602 6603 local_bh_enable(); 6604 6605 if (!repoll) 6606 break; 6607 6608 cond_resched(); 6609 } 6610 } 6611 return 0; 6612 } 6613 6614 static void skb_defer_free_flush(struct softnet_data *sd) 6615 { 6616 struct sk_buff *skb, *next; 6617 unsigned long flags; 6618 6619 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6620 if (!READ_ONCE(sd->defer_list)) 6621 return; 6622 6623 spin_lock_irqsave(&sd->defer_lock, flags); 6624 skb = sd->defer_list; 6625 sd->defer_list = NULL; 6626 sd->defer_count = 0; 6627 spin_unlock_irqrestore(&sd->defer_lock, flags); 6628 6629 while (skb != NULL) { 6630 next = skb->next; 6631 __kfree_skb(skb); 6632 skb = next; 6633 } 6634 } 6635 6636 static __latent_entropy void net_rx_action(struct softirq_action *h) 6637 { 6638 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6639 unsigned long time_limit = jiffies + 6640 usecs_to_jiffies(netdev_budget_usecs); 6641 int budget = netdev_budget; 6642 LIST_HEAD(list); 6643 LIST_HEAD(repoll); 6644 6645 local_irq_disable(); 6646 list_splice_init(&sd->poll_list, &list); 6647 local_irq_enable(); 6648 6649 for (;;) { 6650 struct napi_struct *n; 6651 6652 if (list_empty(&list)) { 6653 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6654 goto end; 6655 break; 6656 } 6657 6658 n = list_first_entry(&list, struct napi_struct, poll_list); 6659 budget -= napi_poll(n, &repoll); 6660 6661 /* If softirq window is exhausted then punt. 6662 * Allow this to run for 2 jiffies since which will allow 6663 * an average latency of 1.5/HZ. 6664 */ 6665 if (unlikely(budget <= 0 || 6666 time_after_eq(jiffies, time_limit))) { 6667 sd->time_squeeze++; 6668 break; 6669 } 6670 } 6671 6672 local_irq_disable(); 6673 6674 list_splice_tail_init(&sd->poll_list, &list); 6675 list_splice_tail(&repoll, &list); 6676 list_splice(&list, &sd->poll_list); 6677 if (!list_empty(&sd->poll_list)) 6678 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6679 6680 net_rps_action_and_irq_enable(sd); 6681 end: 6682 skb_defer_free_flush(sd); 6683 } 6684 6685 struct netdev_adjacent { 6686 struct net_device *dev; 6687 netdevice_tracker dev_tracker; 6688 6689 /* upper master flag, there can only be one master device per list */ 6690 bool master; 6691 6692 /* lookup ignore flag */ 6693 bool ignore; 6694 6695 /* counter for the number of times this device was added to us */ 6696 u16 ref_nr; 6697 6698 /* private field for the users */ 6699 void *private; 6700 6701 struct list_head list; 6702 struct rcu_head rcu; 6703 }; 6704 6705 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6706 struct list_head *adj_list) 6707 { 6708 struct netdev_adjacent *adj; 6709 6710 list_for_each_entry(adj, adj_list, list) { 6711 if (adj->dev == adj_dev) 6712 return adj; 6713 } 6714 return NULL; 6715 } 6716 6717 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6718 struct netdev_nested_priv *priv) 6719 { 6720 struct net_device *dev = (struct net_device *)priv->data; 6721 6722 return upper_dev == dev; 6723 } 6724 6725 /** 6726 * netdev_has_upper_dev - Check if device is linked to an upper device 6727 * @dev: device 6728 * @upper_dev: upper device to check 6729 * 6730 * Find out if a device is linked to specified upper device and return true 6731 * in case it is. Note that this checks only immediate upper device, 6732 * not through a complete stack of devices. The caller must hold the RTNL lock. 6733 */ 6734 bool netdev_has_upper_dev(struct net_device *dev, 6735 struct net_device *upper_dev) 6736 { 6737 struct netdev_nested_priv priv = { 6738 .data = (void *)upper_dev, 6739 }; 6740 6741 ASSERT_RTNL(); 6742 6743 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6744 &priv); 6745 } 6746 EXPORT_SYMBOL(netdev_has_upper_dev); 6747 6748 /** 6749 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6750 * @dev: device 6751 * @upper_dev: upper device to check 6752 * 6753 * Find out if a device is linked to specified upper device and return true 6754 * in case it is. Note that this checks the entire upper device chain. 6755 * The caller must hold rcu lock. 6756 */ 6757 6758 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6759 struct net_device *upper_dev) 6760 { 6761 struct netdev_nested_priv priv = { 6762 .data = (void *)upper_dev, 6763 }; 6764 6765 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6766 &priv); 6767 } 6768 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6769 6770 /** 6771 * netdev_has_any_upper_dev - Check if device is linked to some device 6772 * @dev: device 6773 * 6774 * Find out if a device is linked to an upper device and return true in case 6775 * it is. The caller must hold the RTNL lock. 6776 */ 6777 bool netdev_has_any_upper_dev(struct net_device *dev) 6778 { 6779 ASSERT_RTNL(); 6780 6781 return !list_empty(&dev->adj_list.upper); 6782 } 6783 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6784 6785 /** 6786 * netdev_master_upper_dev_get - Get master upper device 6787 * @dev: device 6788 * 6789 * Find a master upper device and return pointer to it or NULL in case 6790 * it's not there. The caller must hold the RTNL lock. 6791 */ 6792 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6793 { 6794 struct netdev_adjacent *upper; 6795 6796 ASSERT_RTNL(); 6797 6798 if (list_empty(&dev->adj_list.upper)) 6799 return NULL; 6800 6801 upper = list_first_entry(&dev->adj_list.upper, 6802 struct netdev_adjacent, list); 6803 if (likely(upper->master)) 6804 return upper->dev; 6805 return NULL; 6806 } 6807 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6808 6809 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6810 { 6811 struct netdev_adjacent *upper; 6812 6813 ASSERT_RTNL(); 6814 6815 if (list_empty(&dev->adj_list.upper)) 6816 return NULL; 6817 6818 upper = list_first_entry(&dev->adj_list.upper, 6819 struct netdev_adjacent, list); 6820 if (likely(upper->master) && !upper->ignore) 6821 return upper->dev; 6822 return NULL; 6823 } 6824 6825 /** 6826 * netdev_has_any_lower_dev - Check if device is linked to some device 6827 * @dev: device 6828 * 6829 * Find out if a device is linked to a lower device and return true in case 6830 * it is. The caller must hold the RTNL lock. 6831 */ 6832 static bool netdev_has_any_lower_dev(struct net_device *dev) 6833 { 6834 ASSERT_RTNL(); 6835 6836 return !list_empty(&dev->adj_list.lower); 6837 } 6838 6839 void *netdev_adjacent_get_private(struct list_head *adj_list) 6840 { 6841 struct netdev_adjacent *adj; 6842 6843 adj = list_entry(adj_list, struct netdev_adjacent, list); 6844 6845 return adj->private; 6846 } 6847 EXPORT_SYMBOL(netdev_adjacent_get_private); 6848 6849 /** 6850 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6851 * @dev: device 6852 * @iter: list_head ** of the current position 6853 * 6854 * Gets the next device from the dev's upper list, starting from iter 6855 * position. The caller must hold RCU read lock. 6856 */ 6857 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6858 struct list_head **iter) 6859 { 6860 struct netdev_adjacent *upper; 6861 6862 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6863 6864 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6865 6866 if (&upper->list == &dev->adj_list.upper) 6867 return NULL; 6868 6869 *iter = &upper->list; 6870 6871 return upper->dev; 6872 } 6873 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6874 6875 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6876 struct list_head **iter, 6877 bool *ignore) 6878 { 6879 struct netdev_adjacent *upper; 6880 6881 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6882 6883 if (&upper->list == &dev->adj_list.upper) 6884 return NULL; 6885 6886 *iter = &upper->list; 6887 *ignore = upper->ignore; 6888 6889 return upper->dev; 6890 } 6891 6892 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6893 struct list_head **iter) 6894 { 6895 struct netdev_adjacent *upper; 6896 6897 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6898 6899 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6900 6901 if (&upper->list == &dev->adj_list.upper) 6902 return NULL; 6903 6904 *iter = &upper->list; 6905 6906 return upper->dev; 6907 } 6908 6909 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6910 int (*fn)(struct net_device *dev, 6911 struct netdev_nested_priv *priv), 6912 struct netdev_nested_priv *priv) 6913 { 6914 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6915 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6916 int ret, cur = 0; 6917 bool ignore; 6918 6919 now = dev; 6920 iter = &dev->adj_list.upper; 6921 6922 while (1) { 6923 if (now != dev) { 6924 ret = fn(now, priv); 6925 if (ret) 6926 return ret; 6927 } 6928 6929 next = NULL; 6930 while (1) { 6931 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6932 if (!udev) 6933 break; 6934 if (ignore) 6935 continue; 6936 6937 next = udev; 6938 niter = &udev->adj_list.upper; 6939 dev_stack[cur] = now; 6940 iter_stack[cur++] = iter; 6941 break; 6942 } 6943 6944 if (!next) { 6945 if (!cur) 6946 return 0; 6947 next = dev_stack[--cur]; 6948 niter = iter_stack[cur]; 6949 } 6950 6951 now = next; 6952 iter = niter; 6953 } 6954 6955 return 0; 6956 } 6957 6958 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6959 int (*fn)(struct net_device *dev, 6960 struct netdev_nested_priv *priv), 6961 struct netdev_nested_priv *priv) 6962 { 6963 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6964 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6965 int ret, cur = 0; 6966 6967 now = dev; 6968 iter = &dev->adj_list.upper; 6969 6970 while (1) { 6971 if (now != dev) { 6972 ret = fn(now, priv); 6973 if (ret) 6974 return ret; 6975 } 6976 6977 next = NULL; 6978 while (1) { 6979 udev = netdev_next_upper_dev_rcu(now, &iter); 6980 if (!udev) 6981 break; 6982 6983 next = udev; 6984 niter = &udev->adj_list.upper; 6985 dev_stack[cur] = now; 6986 iter_stack[cur++] = iter; 6987 break; 6988 } 6989 6990 if (!next) { 6991 if (!cur) 6992 return 0; 6993 next = dev_stack[--cur]; 6994 niter = iter_stack[cur]; 6995 } 6996 6997 now = next; 6998 iter = niter; 6999 } 7000 7001 return 0; 7002 } 7003 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7004 7005 static bool __netdev_has_upper_dev(struct net_device *dev, 7006 struct net_device *upper_dev) 7007 { 7008 struct netdev_nested_priv priv = { 7009 .flags = 0, 7010 .data = (void *)upper_dev, 7011 }; 7012 7013 ASSERT_RTNL(); 7014 7015 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7016 &priv); 7017 } 7018 7019 /** 7020 * netdev_lower_get_next_private - Get the next ->private from the 7021 * lower neighbour list 7022 * @dev: device 7023 * @iter: list_head ** of the current position 7024 * 7025 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7026 * list, starting from iter position. The caller must hold either hold the 7027 * RTNL lock or its own locking that guarantees that the neighbour lower 7028 * list will remain unchanged. 7029 */ 7030 void *netdev_lower_get_next_private(struct net_device *dev, 7031 struct list_head **iter) 7032 { 7033 struct netdev_adjacent *lower; 7034 7035 lower = list_entry(*iter, struct netdev_adjacent, list); 7036 7037 if (&lower->list == &dev->adj_list.lower) 7038 return NULL; 7039 7040 *iter = lower->list.next; 7041 7042 return lower->private; 7043 } 7044 EXPORT_SYMBOL(netdev_lower_get_next_private); 7045 7046 /** 7047 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7048 * lower neighbour list, RCU 7049 * variant 7050 * @dev: device 7051 * @iter: list_head ** of the current position 7052 * 7053 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7054 * list, starting from iter position. The caller must hold RCU read lock. 7055 */ 7056 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7057 struct list_head **iter) 7058 { 7059 struct netdev_adjacent *lower; 7060 7061 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7062 7063 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7064 7065 if (&lower->list == &dev->adj_list.lower) 7066 return NULL; 7067 7068 *iter = &lower->list; 7069 7070 return lower->private; 7071 } 7072 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7073 7074 /** 7075 * netdev_lower_get_next - Get the next device from the lower neighbour 7076 * list 7077 * @dev: device 7078 * @iter: list_head ** of the current position 7079 * 7080 * Gets the next netdev_adjacent from the dev's lower neighbour 7081 * list, starting from iter position. The caller must hold RTNL lock or 7082 * its own locking that guarantees that the neighbour lower 7083 * list will remain unchanged. 7084 */ 7085 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7086 { 7087 struct netdev_adjacent *lower; 7088 7089 lower = list_entry(*iter, struct netdev_adjacent, list); 7090 7091 if (&lower->list == &dev->adj_list.lower) 7092 return NULL; 7093 7094 *iter = lower->list.next; 7095 7096 return lower->dev; 7097 } 7098 EXPORT_SYMBOL(netdev_lower_get_next); 7099 7100 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7101 struct list_head **iter) 7102 { 7103 struct netdev_adjacent *lower; 7104 7105 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7106 7107 if (&lower->list == &dev->adj_list.lower) 7108 return NULL; 7109 7110 *iter = &lower->list; 7111 7112 return lower->dev; 7113 } 7114 7115 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7116 struct list_head **iter, 7117 bool *ignore) 7118 { 7119 struct netdev_adjacent *lower; 7120 7121 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7122 7123 if (&lower->list == &dev->adj_list.lower) 7124 return NULL; 7125 7126 *iter = &lower->list; 7127 *ignore = lower->ignore; 7128 7129 return lower->dev; 7130 } 7131 7132 int netdev_walk_all_lower_dev(struct net_device *dev, 7133 int (*fn)(struct net_device *dev, 7134 struct netdev_nested_priv *priv), 7135 struct netdev_nested_priv *priv) 7136 { 7137 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7138 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7139 int ret, cur = 0; 7140 7141 now = dev; 7142 iter = &dev->adj_list.lower; 7143 7144 while (1) { 7145 if (now != dev) { 7146 ret = fn(now, priv); 7147 if (ret) 7148 return ret; 7149 } 7150 7151 next = NULL; 7152 while (1) { 7153 ldev = netdev_next_lower_dev(now, &iter); 7154 if (!ldev) 7155 break; 7156 7157 next = ldev; 7158 niter = &ldev->adj_list.lower; 7159 dev_stack[cur] = now; 7160 iter_stack[cur++] = iter; 7161 break; 7162 } 7163 7164 if (!next) { 7165 if (!cur) 7166 return 0; 7167 next = dev_stack[--cur]; 7168 niter = iter_stack[cur]; 7169 } 7170 7171 now = next; 7172 iter = niter; 7173 } 7174 7175 return 0; 7176 } 7177 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7178 7179 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7180 int (*fn)(struct net_device *dev, 7181 struct netdev_nested_priv *priv), 7182 struct netdev_nested_priv *priv) 7183 { 7184 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7185 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7186 int ret, cur = 0; 7187 bool ignore; 7188 7189 now = dev; 7190 iter = &dev->adj_list.lower; 7191 7192 while (1) { 7193 if (now != dev) { 7194 ret = fn(now, priv); 7195 if (ret) 7196 return ret; 7197 } 7198 7199 next = NULL; 7200 while (1) { 7201 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7202 if (!ldev) 7203 break; 7204 if (ignore) 7205 continue; 7206 7207 next = ldev; 7208 niter = &ldev->adj_list.lower; 7209 dev_stack[cur] = now; 7210 iter_stack[cur++] = iter; 7211 break; 7212 } 7213 7214 if (!next) { 7215 if (!cur) 7216 return 0; 7217 next = dev_stack[--cur]; 7218 niter = iter_stack[cur]; 7219 } 7220 7221 now = next; 7222 iter = niter; 7223 } 7224 7225 return 0; 7226 } 7227 7228 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7229 struct list_head **iter) 7230 { 7231 struct netdev_adjacent *lower; 7232 7233 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7234 if (&lower->list == &dev->adj_list.lower) 7235 return NULL; 7236 7237 *iter = &lower->list; 7238 7239 return lower->dev; 7240 } 7241 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7242 7243 static u8 __netdev_upper_depth(struct net_device *dev) 7244 { 7245 struct net_device *udev; 7246 struct list_head *iter; 7247 u8 max_depth = 0; 7248 bool ignore; 7249 7250 for (iter = &dev->adj_list.upper, 7251 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7252 udev; 7253 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7254 if (ignore) 7255 continue; 7256 if (max_depth < udev->upper_level) 7257 max_depth = udev->upper_level; 7258 } 7259 7260 return max_depth; 7261 } 7262 7263 static u8 __netdev_lower_depth(struct net_device *dev) 7264 { 7265 struct net_device *ldev; 7266 struct list_head *iter; 7267 u8 max_depth = 0; 7268 bool ignore; 7269 7270 for (iter = &dev->adj_list.lower, 7271 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7272 ldev; 7273 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7274 if (ignore) 7275 continue; 7276 if (max_depth < ldev->lower_level) 7277 max_depth = ldev->lower_level; 7278 } 7279 7280 return max_depth; 7281 } 7282 7283 static int __netdev_update_upper_level(struct net_device *dev, 7284 struct netdev_nested_priv *__unused) 7285 { 7286 dev->upper_level = __netdev_upper_depth(dev) + 1; 7287 return 0; 7288 } 7289 7290 #ifdef CONFIG_LOCKDEP 7291 static LIST_HEAD(net_unlink_list); 7292 7293 static void net_unlink_todo(struct net_device *dev) 7294 { 7295 if (list_empty(&dev->unlink_list)) 7296 list_add_tail(&dev->unlink_list, &net_unlink_list); 7297 } 7298 #endif 7299 7300 static int __netdev_update_lower_level(struct net_device *dev, 7301 struct netdev_nested_priv *priv) 7302 { 7303 dev->lower_level = __netdev_lower_depth(dev) + 1; 7304 7305 #ifdef CONFIG_LOCKDEP 7306 if (!priv) 7307 return 0; 7308 7309 if (priv->flags & NESTED_SYNC_IMM) 7310 dev->nested_level = dev->lower_level - 1; 7311 if (priv->flags & NESTED_SYNC_TODO) 7312 net_unlink_todo(dev); 7313 #endif 7314 return 0; 7315 } 7316 7317 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7318 int (*fn)(struct net_device *dev, 7319 struct netdev_nested_priv *priv), 7320 struct netdev_nested_priv *priv) 7321 { 7322 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7323 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7324 int ret, cur = 0; 7325 7326 now = dev; 7327 iter = &dev->adj_list.lower; 7328 7329 while (1) { 7330 if (now != dev) { 7331 ret = fn(now, priv); 7332 if (ret) 7333 return ret; 7334 } 7335 7336 next = NULL; 7337 while (1) { 7338 ldev = netdev_next_lower_dev_rcu(now, &iter); 7339 if (!ldev) 7340 break; 7341 7342 next = ldev; 7343 niter = &ldev->adj_list.lower; 7344 dev_stack[cur] = now; 7345 iter_stack[cur++] = iter; 7346 break; 7347 } 7348 7349 if (!next) { 7350 if (!cur) 7351 return 0; 7352 next = dev_stack[--cur]; 7353 niter = iter_stack[cur]; 7354 } 7355 7356 now = next; 7357 iter = niter; 7358 } 7359 7360 return 0; 7361 } 7362 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7363 7364 /** 7365 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7366 * lower neighbour list, RCU 7367 * variant 7368 * @dev: device 7369 * 7370 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7371 * list. The caller must hold RCU read lock. 7372 */ 7373 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7374 { 7375 struct netdev_adjacent *lower; 7376 7377 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7378 struct netdev_adjacent, list); 7379 if (lower) 7380 return lower->private; 7381 return NULL; 7382 } 7383 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7384 7385 /** 7386 * netdev_master_upper_dev_get_rcu - Get master upper device 7387 * @dev: device 7388 * 7389 * Find a master upper device and return pointer to it or NULL in case 7390 * it's not there. The caller must hold the RCU read lock. 7391 */ 7392 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7393 { 7394 struct netdev_adjacent *upper; 7395 7396 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7397 struct netdev_adjacent, list); 7398 if (upper && likely(upper->master)) 7399 return upper->dev; 7400 return NULL; 7401 } 7402 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7403 7404 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7405 struct net_device *adj_dev, 7406 struct list_head *dev_list) 7407 { 7408 char linkname[IFNAMSIZ+7]; 7409 7410 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7411 "upper_%s" : "lower_%s", adj_dev->name); 7412 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7413 linkname); 7414 } 7415 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7416 char *name, 7417 struct list_head *dev_list) 7418 { 7419 char linkname[IFNAMSIZ+7]; 7420 7421 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7422 "upper_%s" : "lower_%s", name); 7423 sysfs_remove_link(&(dev->dev.kobj), linkname); 7424 } 7425 7426 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7427 struct net_device *adj_dev, 7428 struct list_head *dev_list) 7429 { 7430 return (dev_list == &dev->adj_list.upper || 7431 dev_list == &dev->adj_list.lower) && 7432 net_eq(dev_net(dev), dev_net(adj_dev)); 7433 } 7434 7435 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7436 struct net_device *adj_dev, 7437 struct list_head *dev_list, 7438 void *private, bool master) 7439 { 7440 struct netdev_adjacent *adj; 7441 int ret; 7442 7443 adj = __netdev_find_adj(adj_dev, dev_list); 7444 7445 if (adj) { 7446 adj->ref_nr += 1; 7447 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7448 dev->name, adj_dev->name, adj->ref_nr); 7449 7450 return 0; 7451 } 7452 7453 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7454 if (!adj) 7455 return -ENOMEM; 7456 7457 adj->dev = adj_dev; 7458 adj->master = master; 7459 adj->ref_nr = 1; 7460 adj->private = private; 7461 adj->ignore = false; 7462 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7463 7464 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7465 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7466 7467 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7468 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7469 if (ret) 7470 goto free_adj; 7471 } 7472 7473 /* Ensure that master link is always the first item in list. */ 7474 if (master) { 7475 ret = sysfs_create_link(&(dev->dev.kobj), 7476 &(adj_dev->dev.kobj), "master"); 7477 if (ret) 7478 goto remove_symlinks; 7479 7480 list_add_rcu(&adj->list, dev_list); 7481 } else { 7482 list_add_tail_rcu(&adj->list, dev_list); 7483 } 7484 7485 return 0; 7486 7487 remove_symlinks: 7488 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7489 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7490 free_adj: 7491 dev_put_track(adj_dev, &adj->dev_tracker); 7492 kfree(adj); 7493 7494 return ret; 7495 } 7496 7497 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7498 struct net_device *adj_dev, 7499 u16 ref_nr, 7500 struct list_head *dev_list) 7501 { 7502 struct netdev_adjacent *adj; 7503 7504 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7505 dev->name, adj_dev->name, ref_nr); 7506 7507 adj = __netdev_find_adj(adj_dev, dev_list); 7508 7509 if (!adj) { 7510 pr_err("Adjacency does not exist for device %s from %s\n", 7511 dev->name, adj_dev->name); 7512 WARN_ON(1); 7513 return; 7514 } 7515 7516 if (adj->ref_nr > ref_nr) { 7517 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7518 dev->name, adj_dev->name, ref_nr, 7519 adj->ref_nr - ref_nr); 7520 adj->ref_nr -= ref_nr; 7521 return; 7522 } 7523 7524 if (adj->master) 7525 sysfs_remove_link(&(dev->dev.kobj), "master"); 7526 7527 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7528 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7529 7530 list_del_rcu(&adj->list); 7531 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7532 adj_dev->name, dev->name, adj_dev->name); 7533 dev_put_track(adj_dev, &adj->dev_tracker); 7534 kfree_rcu(adj, rcu); 7535 } 7536 7537 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7538 struct net_device *upper_dev, 7539 struct list_head *up_list, 7540 struct list_head *down_list, 7541 void *private, bool master) 7542 { 7543 int ret; 7544 7545 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7546 private, master); 7547 if (ret) 7548 return ret; 7549 7550 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7551 private, false); 7552 if (ret) { 7553 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7554 return ret; 7555 } 7556 7557 return 0; 7558 } 7559 7560 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7561 struct net_device *upper_dev, 7562 u16 ref_nr, 7563 struct list_head *up_list, 7564 struct list_head *down_list) 7565 { 7566 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7567 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7568 } 7569 7570 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7571 struct net_device *upper_dev, 7572 void *private, bool master) 7573 { 7574 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7575 &dev->adj_list.upper, 7576 &upper_dev->adj_list.lower, 7577 private, master); 7578 } 7579 7580 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7581 struct net_device *upper_dev) 7582 { 7583 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7584 &dev->adj_list.upper, 7585 &upper_dev->adj_list.lower); 7586 } 7587 7588 static int __netdev_upper_dev_link(struct net_device *dev, 7589 struct net_device *upper_dev, bool master, 7590 void *upper_priv, void *upper_info, 7591 struct netdev_nested_priv *priv, 7592 struct netlink_ext_ack *extack) 7593 { 7594 struct netdev_notifier_changeupper_info changeupper_info = { 7595 .info = { 7596 .dev = dev, 7597 .extack = extack, 7598 }, 7599 .upper_dev = upper_dev, 7600 .master = master, 7601 .linking = true, 7602 .upper_info = upper_info, 7603 }; 7604 struct net_device *master_dev; 7605 int ret = 0; 7606 7607 ASSERT_RTNL(); 7608 7609 if (dev == upper_dev) 7610 return -EBUSY; 7611 7612 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7613 if (__netdev_has_upper_dev(upper_dev, dev)) 7614 return -EBUSY; 7615 7616 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7617 return -EMLINK; 7618 7619 if (!master) { 7620 if (__netdev_has_upper_dev(dev, upper_dev)) 7621 return -EEXIST; 7622 } else { 7623 master_dev = __netdev_master_upper_dev_get(dev); 7624 if (master_dev) 7625 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7626 } 7627 7628 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7629 &changeupper_info.info); 7630 ret = notifier_to_errno(ret); 7631 if (ret) 7632 return ret; 7633 7634 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7635 master); 7636 if (ret) 7637 return ret; 7638 7639 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7640 &changeupper_info.info); 7641 ret = notifier_to_errno(ret); 7642 if (ret) 7643 goto rollback; 7644 7645 __netdev_update_upper_level(dev, NULL); 7646 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7647 7648 __netdev_update_lower_level(upper_dev, priv); 7649 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7650 priv); 7651 7652 return 0; 7653 7654 rollback: 7655 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7656 7657 return ret; 7658 } 7659 7660 /** 7661 * netdev_upper_dev_link - Add a link to the upper device 7662 * @dev: device 7663 * @upper_dev: new upper device 7664 * @extack: netlink extended ack 7665 * 7666 * Adds a link to device which is upper to this one. The caller must hold 7667 * the RTNL lock. On a failure a negative errno code is returned. 7668 * On success the reference counts are adjusted and the function 7669 * returns zero. 7670 */ 7671 int netdev_upper_dev_link(struct net_device *dev, 7672 struct net_device *upper_dev, 7673 struct netlink_ext_ack *extack) 7674 { 7675 struct netdev_nested_priv priv = { 7676 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7677 .data = NULL, 7678 }; 7679 7680 return __netdev_upper_dev_link(dev, upper_dev, false, 7681 NULL, NULL, &priv, extack); 7682 } 7683 EXPORT_SYMBOL(netdev_upper_dev_link); 7684 7685 /** 7686 * netdev_master_upper_dev_link - Add a master link to the upper device 7687 * @dev: device 7688 * @upper_dev: new upper device 7689 * @upper_priv: upper device private 7690 * @upper_info: upper info to be passed down via notifier 7691 * @extack: netlink extended ack 7692 * 7693 * Adds a link to device which is upper to this one. In this case, only 7694 * one master upper device can be linked, although other non-master devices 7695 * might be linked as well. The caller must hold the RTNL lock. 7696 * On a failure a negative errno code is returned. On success the reference 7697 * counts are adjusted and the function returns zero. 7698 */ 7699 int netdev_master_upper_dev_link(struct net_device *dev, 7700 struct net_device *upper_dev, 7701 void *upper_priv, void *upper_info, 7702 struct netlink_ext_ack *extack) 7703 { 7704 struct netdev_nested_priv priv = { 7705 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7706 .data = NULL, 7707 }; 7708 7709 return __netdev_upper_dev_link(dev, upper_dev, true, 7710 upper_priv, upper_info, &priv, extack); 7711 } 7712 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7713 7714 static void __netdev_upper_dev_unlink(struct net_device *dev, 7715 struct net_device *upper_dev, 7716 struct netdev_nested_priv *priv) 7717 { 7718 struct netdev_notifier_changeupper_info changeupper_info = { 7719 .info = { 7720 .dev = dev, 7721 }, 7722 .upper_dev = upper_dev, 7723 .linking = false, 7724 }; 7725 7726 ASSERT_RTNL(); 7727 7728 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7729 7730 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7731 &changeupper_info.info); 7732 7733 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7734 7735 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7736 &changeupper_info.info); 7737 7738 __netdev_update_upper_level(dev, NULL); 7739 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7740 7741 __netdev_update_lower_level(upper_dev, priv); 7742 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7743 priv); 7744 } 7745 7746 /** 7747 * netdev_upper_dev_unlink - Removes a link to upper device 7748 * @dev: device 7749 * @upper_dev: new upper device 7750 * 7751 * Removes a link to device which is upper to this one. The caller must hold 7752 * the RTNL lock. 7753 */ 7754 void netdev_upper_dev_unlink(struct net_device *dev, 7755 struct net_device *upper_dev) 7756 { 7757 struct netdev_nested_priv priv = { 7758 .flags = NESTED_SYNC_TODO, 7759 .data = NULL, 7760 }; 7761 7762 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7763 } 7764 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7765 7766 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7767 struct net_device *lower_dev, 7768 bool val) 7769 { 7770 struct netdev_adjacent *adj; 7771 7772 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7773 if (adj) 7774 adj->ignore = val; 7775 7776 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7777 if (adj) 7778 adj->ignore = val; 7779 } 7780 7781 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7782 struct net_device *lower_dev) 7783 { 7784 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7785 } 7786 7787 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7788 struct net_device *lower_dev) 7789 { 7790 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7791 } 7792 7793 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7794 struct net_device *new_dev, 7795 struct net_device *dev, 7796 struct netlink_ext_ack *extack) 7797 { 7798 struct netdev_nested_priv priv = { 7799 .flags = 0, 7800 .data = NULL, 7801 }; 7802 int err; 7803 7804 if (!new_dev) 7805 return 0; 7806 7807 if (old_dev && new_dev != old_dev) 7808 netdev_adjacent_dev_disable(dev, old_dev); 7809 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7810 extack); 7811 if (err) { 7812 if (old_dev && new_dev != old_dev) 7813 netdev_adjacent_dev_enable(dev, old_dev); 7814 return err; 7815 } 7816 7817 return 0; 7818 } 7819 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7820 7821 void netdev_adjacent_change_commit(struct net_device *old_dev, 7822 struct net_device *new_dev, 7823 struct net_device *dev) 7824 { 7825 struct netdev_nested_priv priv = { 7826 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7827 .data = NULL, 7828 }; 7829 7830 if (!new_dev || !old_dev) 7831 return; 7832 7833 if (new_dev == old_dev) 7834 return; 7835 7836 netdev_adjacent_dev_enable(dev, old_dev); 7837 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7838 } 7839 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7840 7841 void netdev_adjacent_change_abort(struct net_device *old_dev, 7842 struct net_device *new_dev, 7843 struct net_device *dev) 7844 { 7845 struct netdev_nested_priv priv = { 7846 .flags = 0, 7847 .data = NULL, 7848 }; 7849 7850 if (!new_dev) 7851 return; 7852 7853 if (old_dev && new_dev != old_dev) 7854 netdev_adjacent_dev_enable(dev, old_dev); 7855 7856 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7857 } 7858 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7859 7860 /** 7861 * netdev_bonding_info_change - Dispatch event about slave change 7862 * @dev: device 7863 * @bonding_info: info to dispatch 7864 * 7865 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7866 * The caller must hold the RTNL lock. 7867 */ 7868 void netdev_bonding_info_change(struct net_device *dev, 7869 struct netdev_bonding_info *bonding_info) 7870 { 7871 struct netdev_notifier_bonding_info info = { 7872 .info.dev = dev, 7873 }; 7874 7875 memcpy(&info.bonding_info, bonding_info, 7876 sizeof(struct netdev_bonding_info)); 7877 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7878 &info.info); 7879 } 7880 EXPORT_SYMBOL(netdev_bonding_info_change); 7881 7882 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7883 struct netlink_ext_ack *extack) 7884 { 7885 struct netdev_notifier_offload_xstats_info info = { 7886 .info.dev = dev, 7887 .info.extack = extack, 7888 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7889 }; 7890 int err; 7891 int rc; 7892 7893 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7894 GFP_KERNEL); 7895 if (!dev->offload_xstats_l3) 7896 return -ENOMEM; 7897 7898 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7899 NETDEV_OFFLOAD_XSTATS_DISABLE, 7900 &info.info); 7901 err = notifier_to_errno(rc); 7902 if (err) 7903 goto free_stats; 7904 7905 return 0; 7906 7907 free_stats: 7908 kfree(dev->offload_xstats_l3); 7909 dev->offload_xstats_l3 = NULL; 7910 return err; 7911 } 7912 7913 int netdev_offload_xstats_enable(struct net_device *dev, 7914 enum netdev_offload_xstats_type type, 7915 struct netlink_ext_ack *extack) 7916 { 7917 ASSERT_RTNL(); 7918 7919 if (netdev_offload_xstats_enabled(dev, type)) 7920 return -EALREADY; 7921 7922 switch (type) { 7923 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7924 return netdev_offload_xstats_enable_l3(dev, extack); 7925 } 7926 7927 WARN_ON(1); 7928 return -EINVAL; 7929 } 7930 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7931 7932 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7933 { 7934 struct netdev_notifier_offload_xstats_info info = { 7935 .info.dev = dev, 7936 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7937 }; 7938 7939 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7940 &info.info); 7941 kfree(dev->offload_xstats_l3); 7942 dev->offload_xstats_l3 = NULL; 7943 } 7944 7945 int netdev_offload_xstats_disable(struct net_device *dev, 7946 enum netdev_offload_xstats_type type) 7947 { 7948 ASSERT_RTNL(); 7949 7950 if (!netdev_offload_xstats_enabled(dev, type)) 7951 return -EALREADY; 7952 7953 switch (type) { 7954 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7955 netdev_offload_xstats_disable_l3(dev); 7956 return 0; 7957 } 7958 7959 WARN_ON(1); 7960 return -EINVAL; 7961 } 7962 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7963 7964 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7965 { 7966 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7967 } 7968 7969 static struct rtnl_hw_stats64 * 7970 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7971 enum netdev_offload_xstats_type type) 7972 { 7973 switch (type) { 7974 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7975 return dev->offload_xstats_l3; 7976 } 7977 7978 WARN_ON(1); 7979 return NULL; 7980 } 7981 7982 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7983 enum netdev_offload_xstats_type type) 7984 { 7985 ASSERT_RTNL(); 7986 7987 return netdev_offload_xstats_get_ptr(dev, type); 7988 } 7989 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7990 7991 struct netdev_notifier_offload_xstats_ru { 7992 bool used; 7993 }; 7994 7995 struct netdev_notifier_offload_xstats_rd { 7996 struct rtnl_hw_stats64 stats; 7997 bool used; 7998 }; 7999 8000 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8001 const struct rtnl_hw_stats64 *src) 8002 { 8003 dest->rx_packets += src->rx_packets; 8004 dest->tx_packets += src->tx_packets; 8005 dest->rx_bytes += src->rx_bytes; 8006 dest->tx_bytes += src->tx_bytes; 8007 dest->rx_errors += src->rx_errors; 8008 dest->tx_errors += src->tx_errors; 8009 dest->rx_dropped += src->rx_dropped; 8010 dest->tx_dropped += src->tx_dropped; 8011 dest->multicast += src->multicast; 8012 } 8013 8014 static int netdev_offload_xstats_get_used(struct net_device *dev, 8015 enum netdev_offload_xstats_type type, 8016 bool *p_used, 8017 struct netlink_ext_ack *extack) 8018 { 8019 struct netdev_notifier_offload_xstats_ru report_used = {}; 8020 struct netdev_notifier_offload_xstats_info info = { 8021 .info.dev = dev, 8022 .info.extack = extack, 8023 .type = type, 8024 .report_used = &report_used, 8025 }; 8026 int rc; 8027 8028 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8029 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8030 &info.info); 8031 *p_used = report_used.used; 8032 return notifier_to_errno(rc); 8033 } 8034 8035 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8036 enum netdev_offload_xstats_type type, 8037 struct rtnl_hw_stats64 *p_stats, 8038 bool *p_used, 8039 struct netlink_ext_ack *extack) 8040 { 8041 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8042 struct netdev_notifier_offload_xstats_info info = { 8043 .info.dev = dev, 8044 .info.extack = extack, 8045 .type = type, 8046 .report_delta = &report_delta, 8047 }; 8048 struct rtnl_hw_stats64 *stats; 8049 int rc; 8050 8051 stats = netdev_offload_xstats_get_ptr(dev, type); 8052 if (WARN_ON(!stats)) 8053 return -EINVAL; 8054 8055 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8056 &info.info); 8057 8058 /* Cache whatever we got, even if there was an error, otherwise the 8059 * successful stats retrievals would get lost. 8060 */ 8061 netdev_hw_stats64_add(stats, &report_delta.stats); 8062 8063 if (p_stats) 8064 *p_stats = *stats; 8065 *p_used = report_delta.used; 8066 8067 return notifier_to_errno(rc); 8068 } 8069 8070 int netdev_offload_xstats_get(struct net_device *dev, 8071 enum netdev_offload_xstats_type type, 8072 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8073 struct netlink_ext_ack *extack) 8074 { 8075 ASSERT_RTNL(); 8076 8077 if (p_stats) 8078 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8079 p_used, extack); 8080 else 8081 return netdev_offload_xstats_get_used(dev, type, p_used, 8082 extack); 8083 } 8084 EXPORT_SYMBOL(netdev_offload_xstats_get); 8085 8086 void 8087 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8088 const struct rtnl_hw_stats64 *stats) 8089 { 8090 report_delta->used = true; 8091 netdev_hw_stats64_add(&report_delta->stats, stats); 8092 } 8093 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8094 8095 void 8096 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8097 { 8098 report_used->used = true; 8099 } 8100 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8101 8102 void netdev_offload_xstats_push_delta(struct net_device *dev, 8103 enum netdev_offload_xstats_type type, 8104 const struct rtnl_hw_stats64 *p_stats) 8105 { 8106 struct rtnl_hw_stats64 *stats; 8107 8108 ASSERT_RTNL(); 8109 8110 stats = netdev_offload_xstats_get_ptr(dev, type); 8111 if (WARN_ON(!stats)) 8112 return; 8113 8114 netdev_hw_stats64_add(stats, p_stats); 8115 } 8116 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8117 8118 /** 8119 * netdev_get_xmit_slave - Get the xmit slave of master device 8120 * @dev: device 8121 * @skb: The packet 8122 * @all_slaves: assume all the slaves are active 8123 * 8124 * The reference counters are not incremented so the caller must be 8125 * careful with locks. The caller must hold RCU lock. 8126 * %NULL is returned if no slave is found. 8127 */ 8128 8129 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8130 struct sk_buff *skb, 8131 bool all_slaves) 8132 { 8133 const struct net_device_ops *ops = dev->netdev_ops; 8134 8135 if (!ops->ndo_get_xmit_slave) 8136 return NULL; 8137 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8138 } 8139 EXPORT_SYMBOL(netdev_get_xmit_slave); 8140 8141 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8142 struct sock *sk) 8143 { 8144 const struct net_device_ops *ops = dev->netdev_ops; 8145 8146 if (!ops->ndo_sk_get_lower_dev) 8147 return NULL; 8148 return ops->ndo_sk_get_lower_dev(dev, sk); 8149 } 8150 8151 /** 8152 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8153 * @dev: device 8154 * @sk: the socket 8155 * 8156 * %NULL is returned if no lower device is found. 8157 */ 8158 8159 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8160 struct sock *sk) 8161 { 8162 struct net_device *lower; 8163 8164 lower = netdev_sk_get_lower_dev(dev, sk); 8165 while (lower) { 8166 dev = lower; 8167 lower = netdev_sk_get_lower_dev(dev, sk); 8168 } 8169 8170 return dev; 8171 } 8172 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8173 8174 static void netdev_adjacent_add_links(struct net_device *dev) 8175 { 8176 struct netdev_adjacent *iter; 8177 8178 struct net *net = dev_net(dev); 8179 8180 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8181 if (!net_eq(net, dev_net(iter->dev))) 8182 continue; 8183 netdev_adjacent_sysfs_add(iter->dev, dev, 8184 &iter->dev->adj_list.lower); 8185 netdev_adjacent_sysfs_add(dev, iter->dev, 8186 &dev->adj_list.upper); 8187 } 8188 8189 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8190 if (!net_eq(net, dev_net(iter->dev))) 8191 continue; 8192 netdev_adjacent_sysfs_add(iter->dev, dev, 8193 &iter->dev->adj_list.upper); 8194 netdev_adjacent_sysfs_add(dev, iter->dev, 8195 &dev->adj_list.lower); 8196 } 8197 } 8198 8199 static void netdev_adjacent_del_links(struct net_device *dev) 8200 { 8201 struct netdev_adjacent *iter; 8202 8203 struct net *net = dev_net(dev); 8204 8205 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8206 if (!net_eq(net, dev_net(iter->dev))) 8207 continue; 8208 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8209 &iter->dev->adj_list.lower); 8210 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8211 &dev->adj_list.upper); 8212 } 8213 8214 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8215 if (!net_eq(net, dev_net(iter->dev))) 8216 continue; 8217 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8218 &iter->dev->adj_list.upper); 8219 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8220 &dev->adj_list.lower); 8221 } 8222 } 8223 8224 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8225 { 8226 struct netdev_adjacent *iter; 8227 8228 struct net *net = dev_net(dev); 8229 8230 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8231 if (!net_eq(net, dev_net(iter->dev))) 8232 continue; 8233 netdev_adjacent_sysfs_del(iter->dev, oldname, 8234 &iter->dev->adj_list.lower); 8235 netdev_adjacent_sysfs_add(iter->dev, dev, 8236 &iter->dev->adj_list.lower); 8237 } 8238 8239 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8240 if (!net_eq(net, dev_net(iter->dev))) 8241 continue; 8242 netdev_adjacent_sysfs_del(iter->dev, oldname, 8243 &iter->dev->adj_list.upper); 8244 netdev_adjacent_sysfs_add(iter->dev, dev, 8245 &iter->dev->adj_list.upper); 8246 } 8247 } 8248 8249 void *netdev_lower_dev_get_private(struct net_device *dev, 8250 struct net_device *lower_dev) 8251 { 8252 struct netdev_adjacent *lower; 8253 8254 if (!lower_dev) 8255 return NULL; 8256 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8257 if (!lower) 8258 return NULL; 8259 8260 return lower->private; 8261 } 8262 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8263 8264 8265 /** 8266 * netdev_lower_state_changed - Dispatch event about lower device state change 8267 * @lower_dev: device 8268 * @lower_state_info: state to dispatch 8269 * 8270 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8271 * The caller must hold the RTNL lock. 8272 */ 8273 void netdev_lower_state_changed(struct net_device *lower_dev, 8274 void *lower_state_info) 8275 { 8276 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8277 .info.dev = lower_dev, 8278 }; 8279 8280 ASSERT_RTNL(); 8281 changelowerstate_info.lower_state_info = lower_state_info; 8282 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8283 &changelowerstate_info.info); 8284 } 8285 EXPORT_SYMBOL(netdev_lower_state_changed); 8286 8287 static void dev_change_rx_flags(struct net_device *dev, int flags) 8288 { 8289 const struct net_device_ops *ops = dev->netdev_ops; 8290 8291 if (ops->ndo_change_rx_flags) 8292 ops->ndo_change_rx_flags(dev, flags); 8293 } 8294 8295 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8296 { 8297 unsigned int old_flags = dev->flags; 8298 kuid_t uid; 8299 kgid_t gid; 8300 8301 ASSERT_RTNL(); 8302 8303 dev->flags |= IFF_PROMISC; 8304 dev->promiscuity += inc; 8305 if (dev->promiscuity == 0) { 8306 /* 8307 * Avoid overflow. 8308 * If inc causes overflow, untouch promisc and return error. 8309 */ 8310 if (inc < 0) 8311 dev->flags &= ~IFF_PROMISC; 8312 else { 8313 dev->promiscuity -= inc; 8314 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8315 return -EOVERFLOW; 8316 } 8317 } 8318 if (dev->flags != old_flags) { 8319 pr_info("device %s %s promiscuous mode\n", 8320 dev->name, 8321 dev->flags & IFF_PROMISC ? "entered" : "left"); 8322 if (audit_enabled) { 8323 current_uid_gid(&uid, &gid); 8324 audit_log(audit_context(), GFP_ATOMIC, 8325 AUDIT_ANOM_PROMISCUOUS, 8326 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8327 dev->name, (dev->flags & IFF_PROMISC), 8328 (old_flags & IFF_PROMISC), 8329 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8330 from_kuid(&init_user_ns, uid), 8331 from_kgid(&init_user_ns, gid), 8332 audit_get_sessionid(current)); 8333 } 8334 8335 dev_change_rx_flags(dev, IFF_PROMISC); 8336 } 8337 if (notify) 8338 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8339 return 0; 8340 } 8341 8342 /** 8343 * dev_set_promiscuity - update promiscuity count on a device 8344 * @dev: device 8345 * @inc: modifier 8346 * 8347 * Add or remove promiscuity from a device. While the count in the device 8348 * remains above zero the interface remains promiscuous. Once it hits zero 8349 * the device reverts back to normal filtering operation. A negative inc 8350 * value is used to drop promiscuity on the device. 8351 * Return 0 if successful or a negative errno code on error. 8352 */ 8353 int dev_set_promiscuity(struct net_device *dev, int inc) 8354 { 8355 unsigned int old_flags = dev->flags; 8356 int err; 8357 8358 err = __dev_set_promiscuity(dev, inc, true); 8359 if (err < 0) 8360 return err; 8361 if (dev->flags != old_flags) 8362 dev_set_rx_mode(dev); 8363 return err; 8364 } 8365 EXPORT_SYMBOL(dev_set_promiscuity); 8366 8367 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8368 { 8369 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8370 8371 ASSERT_RTNL(); 8372 8373 dev->flags |= IFF_ALLMULTI; 8374 dev->allmulti += inc; 8375 if (dev->allmulti == 0) { 8376 /* 8377 * Avoid overflow. 8378 * If inc causes overflow, untouch allmulti and return error. 8379 */ 8380 if (inc < 0) 8381 dev->flags &= ~IFF_ALLMULTI; 8382 else { 8383 dev->allmulti -= inc; 8384 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8385 return -EOVERFLOW; 8386 } 8387 } 8388 if (dev->flags ^ old_flags) { 8389 dev_change_rx_flags(dev, IFF_ALLMULTI); 8390 dev_set_rx_mode(dev); 8391 if (notify) 8392 __dev_notify_flags(dev, old_flags, 8393 dev->gflags ^ old_gflags); 8394 } 8395 return 0; 8396 } 8397 8398 /** 8399 * dev_set_allmulti - update allmulti count on a device 8400 * @dev: device 8401 * @inc: modifier 8402 * 8403 * Add or remove reception of all multicast frames to a device. While the 8404 * count in the device remains above zero the interface remains listening 8405 * to all interfaces. Once it hits zero the device reverts back to normal 8406 * filtering operation. A negative @inc value is used to drop the counter 8407 * when releasing a resource needing all multicasts. 8408 * Return 0 if successful or a negative errno code on error. 8409 */ 8410 8411 int dev_set_allmulti(struct net_device *dev, int inc) 8412 { 8413 return __dev_set_allmulti(dev, inc, true); 8414 } 8415 EXPORT_SYMBOL(dev_set_allmulti); 8416 8417 /* 8418 * Upload unicast and multicast address lists to device and 8419 * configure RX filtering. When the device doesn't support unicast 8420 * filtering it is put in promiscuous mode while unicast addresses 8421 * are present. 8422 */ 8423 void __dev_set_rx_mode(struct net_device *dev) 8424 { 8425 const struct net_device_ops *ops = dev->netdev_ops; 8426 8427 /* dev_open will call this function so the list will stay sane. */ 8428 if (!(dev->flags&IFF_UP)) 8429 return; 8430 8431 if (!netif_device_present(dev)) 8432 return; 8433 8434 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8435 /* Unicast addresses changes may only happen under the rtnl, 8436 * therefore calling __dev_set_promiscuity here is safe. 8437 */ 8438 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8439 __dev_set_promiscuity(dev, 1, false); 8440 dev->uc_promisc = true; 8441 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8442 __dev_set_promiscuity(dev, -1, false); 8443 dev->uc_promisc = false; 8444 } 8445 } 8446 8447 if (ops->ndo_set_rx_mode) 8448 ops->ndo_set_rx_mode(dev); 8449 } 8450 8451 void dev_set_rx_mode(struct net_device *dev) 8452 { 8453 netif_addr_lock_bh(dev); 8454 __dev_set_rx_mode(dev); 8455 netif_addr_unlock_bh(dev); 8456 } 8457 8458 /** 8459 * dev_get_flags - get flags reported to userspace 8460 * @dev: device 8461 * 8462 * Get the combination of flag bits exported through APIs to userspace. 8463 */ 8464 unsigned int dev_get_flags(const struct net_device *dev) 8465 { 8466 unsigned int flags; 8467 8468 flags = (dev->flags & ~(IFF_PROMISC | 8469 IFF_ALLMULTI | 8470 IFF_RUNNING | 8471 IFF_LOWER_UP | 8472 IFF_DORMANT)) | 8473 (dev->gflags & (IFF_PROMISC | 8474 IFF_ALLMULTI)); 8475 8476 if (netif_running(dev)) { 8477 if (netif_oper_up(dev)) 8478 flags |= IFF_RUNNING; 8479 if (netif_carrier_ok(dev)) 8480 flags |= IFF_LOWER_UP; 8481 if (netif_dormant(dev)) 8482 flags |= IFF_DORMANT; 8483 } 8484 8485 return flags; 8486 } 8487 EXPORT_SYMBOL(dev_get_flags); 8488 8489 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8490 struct netlink_ext_ack *extack) 8491 { 8492 unsigned int old_flags = dev->flags; 8493 int ret; 8494 8495 ASSERT_RTNL(); 8496 8497 /* 8498 * Set the flags on our device. 8499 */ 8500 8501 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8502 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8503 IFF_AUTOMEDIA)) | 8504 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8505 IFF_ALLMULTI)); 8506 8507 /* 8508 * Load in the correct multicast list now the flags have changed. 8509 */ 8510 8511 if ((old_flags ^ flags) & IFF_MULTICAST) 8512 dev_change_rx_flags(dev, IFF_MULTICAST); 8513 8514 dev_set_rx_mode(dev); 8515 8516 /* 8517 * Have we downed the interface. We handle IFF_UP ourselves 8518 * according to user attempts to set it, rather than blindly 8519 * setting it. 8520 */ 8521 8522 ret = 0; 8523 if ((old_flags ^ flags) & IFF_UP) { 8524 if (old_flags & IFF_UP) 8525 __dev_close(dev); 8526 else 8527 ret = __dev_open(dev, extack); 8528 } 8529 8530 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8531 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8532 unsigned int old_flags = dev->flags; 8533 8534 dev->gflags ^= IFF_PROMISC; 8535 8536 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8537 if (dev->flags != old_flags) 8538 dev_set_rx_mode(dev); 8539 } 8540 8541 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8542 * is important. Some (broken) drivers set IFF_PROMISC, when 8543 * IFF_ALLMULTI is requested not asking us and not reporting. 8544 */ 8545 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8546 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8547 8548 dev->gflags ^= IFF_ALLMULTI; 8549 __dev_set_allmulti(dev, inc, false); 8550 } 8551 8552 return ret; 8553 } 8554 8555 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8556 unsigned int gchanges) 8557 { 8558 unsigned int changes = dev->flags ^ old_flags; 8559 8560 if (gchanges) 8561 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8562 8563 if (changes & IFF_UP) { 8564 if (dev->flags & IFF_UP) 8565 call_netdevice_notifiers(NETDEV_UP, dev); 8566 else 8567 call_netdevice_notifiers(NETDEV_DOWN, dev); 8568 } 8569 8570 if (dev->flags & IFF_UP && 8571 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8572 struct netdev_notifier_change_info change_info = { 8573 .info = { 8574 .dev = dev, 8575 }, 8576 .flags_changed = changes, 8577 }; 8578 8579 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8580 } 8581 } 8582 8583 /** 8584 * dev_change_flags - change device settings 8585 * @dev: device 8586 * @flags: device state flags 8587 * @extack: netlink extended ack 8588 * 8589 * Change settings on device based state flags. The flags are 8590 * in the userspace exported format. 8591 */ 8592 int dev_change_flags(struct net_device *dev, unsigned int flags, 8593 struct netlink_ext_ack *extack) 8594 { 8595 int ret; 8596 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8597 8598 ret = __dev_change_flags(dev, flags, extack); 8599 if (ret < 0) 8600 return ret; 8601 8602 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8603 __dev_notify_flags(dev, old_flags, changes); 8604 return ret; 8605 } 8606 EXPORT_SYMBOL(dev_change_flags); 8607 8608 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8609 { 8610 const struct net_device_ops *ops = dev->netdev_ops; 8611 8612 if (ops->ndo_change_mtu) 8613 return ops->ndo_change_mtu(dev, new_mtu); 8614 8615 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8616 WRITE_ONCE(dev->mtu, new_mtu); 8617 return 0; 8618 } 8619 EXPORT_SYMBOL(__dev_set_mtu); 8620 8621 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8622 struct netlink_ext_ack *extack) 8623 { 8624 /* MTU must be positive, and in range */ 8625 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8626 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8627 return -EINVAL; 8628 } 8629 8630 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8631 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8632 return -EINVAL; 8633 } 8634 return 0; 8635 } 8636 8637 /** 8638 * dev_set_mtu_ext - Change maximum transfer unit 8639 * @dev: device 8640 * @new_mtu: new transfer unit 8641 * @extack: netlink extended ack 8642 * 8643 * Change the maximum transfer size of the network device. 8644 */ 8645 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8646 struct netlink_ext_ack *extack) 8647 { 8648 int err, orig_mtu; 8649 8650 if (new_mtu == dev->mtu) 8651 return 0; 8652 8653 err = dev_validate_mtu(dev, new_mtu, extack); 8654 if (err) 8655 return err; 8656 8657 if (!netif_device_present(dev)) 8658 return -ENODEV; 8659 8660 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8661 err = notifier_to_errno(err); 8662 if (err) 8663 return err; 8664 8665 orig_mtu = dev->mtu; 8666 err = __dev_set_mtu(dev, new_mtu); 8667 8668 if (!err) { 8669 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8670 orig_mtu); 8671 err = notifier_to_errno(err); 8672 if (err) { 8673 /* setting mtu back and notifying everyone again, 8674 * so that they have a chance to revert changes. 8675 */ 8676 __dev_set_mtu(dev, orig_mtu); 8677 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8678 new_mtu); 8679 } 8680 } 8681 return err; 8682 } 8683 8684 int dev_set_mtu(struct net_device *dev, int new_mtu) 8685 { 8686 struct netlink_ext_ack extack; 8687 int err; 8688 8689 memset(&extack, 0, sizeof(extack)); 8690 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8691 if (err && extack._msg) 8692 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8693 return err; 8694 } 8695 EXPORT_SYMBOL(dev_set_mtu); 8696 8697 /** 8698 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8699 * @dev: device 8700 * @new_len: new tx queue length 8701 */ 8702 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8703 { 8704 unsigned int orig_len = dev->tx_queue_len; 8705 int res; 8706 8707 if (new_len != (unsigned int)new_len) 8708 return -ERANGE; 8709 8710 if (new_len != orig_len) { 8711 dev->tx_queue_len = new_len; 8712 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8713 res = notifier_to_errno(res); 8714 if (res) 8715 goto err_rollback; 8716 res = dev_qdisc_change_tx_queue_len(dev); 8717 if (res) 8718 goto err_rollback; 8719 } 8720 8721 return 0; 8722 8723 err_rollback: 8724 netdev_err(dev, "refused to change device tx_queue_len\n"); 8725 dev->tx_queue_len = orig_len; 8726 return res; 8727 } 8728 8729 /** 8730 * dev_set_group - Change group this device belongs to 8731 * @dev: device 8732 * @new_group: group this device should belong to 8733 */ 8734 void dev_set_group(struct net_device *dev, int new_group) 8735 { 8736 dev->group = new_group; 8737 } 8738 8739 /** 8740 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8741 * @dev: device 8742 * @addr: new address 8743 * @extack: netlink extended ack 8744 */ 8745 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8746 struct netlink_ext_ack *extack) 8747 { 8748 struct netdev_notifier_pre_changeaddr_info info = { 8749 .info.dev = dev, 8750 .info.extack = extack, 8751 .dev_addr = addr, 8752 }; 8753 int rc; 8754 8755 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8756 return notifier_to_errno(rc); 8757 } 8758 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8759 8760 /** 8761 * dev_set_mac_address - Change Media Access Control Address 8762 * @dev: device 8763 * @sa: new address 8764 * @extack: netlink extended ack 8765 * 8766 * Change the hardware (MAC) address of the device 8767 */ 8768 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8769 struct netlink_ext_ack *extack) 8770 { 8771 const struct net_device_ops *ops = dev->netdev_ops; 8772 int err; 8773 8774 if (!ops->ndo_set_mac_address) 8775 return -EOPNOTSUPP; 8776 if (sa->sa_family != dev->type) 8777 return -EINVAL; 8778 if (!netif_device_present(dev)) 8779 return -ENODEV; 8780 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8781 if (err) 8782 return err; 8783 err = ops->ndo_set_mac_address(dev, sa); 8784 if (err) 8785 return err; 8786 dev->addr_assign_type = NET_ADDR_SET; 8787 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8788 add_device_randomness(dev->dev_addr, dev->addr_len); 8789 return 0; 8790 } 8791 EXPORT_SYMBOL(dev_set_mac_address); 8792 8793 static DECLARE_RWSEM(dev_addr_sem); 8794 8795 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8796 struct netlink_ext_ack *extack) 8797 { 8798 int ret; 8799 8800 down_write(&dev_addr_sem); 8801 ret = dev_set_mac_address(dev, sa, extack); 8802 up_write(&dev_addr_sem); 8803 return ret; 8804 } 8805 EXPORT_SYMBOL(dev_set_mac_address_user); 8806 8807 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8808 { 8809 size_t size = sizeof(sa->sa_data); 8810 struct net_device *dev; 8811 int ret = 0; 8812 8813 down_read(&dev_addr_sem); 8814 rcu_read_lock(); 8815 8816 dev = dev_get_by_name_rcu(net, dev_name); 8817 if (!dev) { 8818 ret = -ENODEV; 8819 goto unlock; 8820 } 8821 if (!dev->addr_len) 8822 memset(sa->sa_data, 0, size); 8823 else 8824 memcpy(sa->sa_data, dev->dev_addr, 8825 min_t(size_t, size, dev->addr_len)); 8826 sa->sa_family = dev->type; 8827 8828 unlock: 8829 rcu_read_unlock(); 8830 up_read(&dev_addr_sem); 8831 return ret; 8832 } 8833 EXPORT_SYMBOL(dev_get_mac_address); 8834 8835 /** 8836 * dev_change_carrier - Change device carrier 8837 * @dev: device 8838 * @new_carrier: new value 8839 * 8840 * Change device carrier 8841 */ 8842 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8843 { 8844 const struct net_device_ops *ops = dev->netdev_ops; 8845 8846 if (!ops->ndo_change_carrier) 8847 return -EOPNOTSUPP; 8848 if (!netif_device_present(dev)) 8849 return -ENODEV; 8850 return ops->ndo_change_carrier(dev, new_carrier); 8851 } 8852 8853 /** 8854 * dev_get_phys_port_id - Get device physical port ID 8855 * @dev: device 8856 * @ppid: port ID 8857 * 8858 * Get device physical port ID 8859 */ 8860 int dev_get_phys_port_id(struct net_device *dev, 8861 struct netdev_phys_item_id *ppid) 8862 { 8863 const struct net_device_ops *ops = dev->netdev_ops; 8864 8865 if (!ops->ndo_get_phys_port_id) 8866 return -EOPNOTSUPP; 8867 return ops->ndo_get_phys_port_id(dev, ppid); 8868 } 8869 8870 /** 8871 * dev_get_phys_port_name - Get device physical port name 8872 * @dev: device 8873 * @name: port name 8874 * @len: limit of bytes to copy to name 8875 * 8876 * Get device physical port name 8877 */ 8878 int dev_get_phys_port_name(struct net_device *dev, 8879 char *name, size_t len) 8880 { 8881 const struct net_device_ops *ops = dev->netdev_ops; 8882 int err; 8883 8884 if (ops->ndo_get_phys_port_name) { 8885 err = ops->ndo_get_phys_port_name(dev, name, len); 8886 if (err != -EOPNOTSUPP) 8887 return err; 8888 } 8889 return devlink_compat_phys_port_name_get(dev, name, len); 8890 } 8891 8892 /** 8893 * dev_get_port_parent_id - Get the device's port parent identifier 8894 * @dev: network device 8895 * @ppid: pointer to a storage for the port's parent identifier 8896 * @recurse: allow/disallow recursion to lower devices 8897 * 8898 * Get the devices's port parent identifier 8899 */ 8900 int dev_get_port_parent_id(struct net_device *dev, 8901 struct netdev_phys_item_id *ppid, 8902 bool recurse) 8903 { 8904 const struct net_device_ops *ops = dev->netdev_ops; 8905 struct netdev_phys_item_id first = { }; 8906 struct net_device *lower_dev; 8907 struct list_head *iter; 8908 int err; 8909 8910 if (ops->ndo_get_port_parent_id) { 8911 err = ops->ndo_get_port_parent_id(dev, ppid); 8912 if (err != -EOPNOTSUPP) 8913 return err; 8914 } 8915 8916 err = devlink_compat_switch_id_get(dev, ppid); 8917 if (!recurse || err != -EOPNOTSUPP) 8918 return err; 8919 8920 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8921 err = dev_get_port_parent_id(lower_dev, ppid, true); 8922 if (err) 8923 break; 8924 if (!first.id_len) 8925 first = *ppid; 8926 else if (memcmp(&first, ppid, sizeof(*ppid))) 8927 return -EOPNOTSUPP; 8928 } 8929 8930 return err; 8931 } 8932 EXPORT_SYMBOL(dev_get_port_parent_id); 8933 8934 /** 8935 * netdev_port_same_parent_id - Indicate if two network devices have 8936 * the same port parent identifier 8937 * @a: first network device 8938 * @b: second network device 8939 */ 8940 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8941 { 8942 struct netdev_phys_item_id a_id = { }; 8943 struct netdev_phys_item_id b_id = { }; 8944 8945 if (dev_get_port_parent_id(a, &a_id, true) || 8946 dev_get_port_parent_id(b, &b_id, true)) 8947 return false; 8948 8949 return netdev_phys_item_id_same(&a_id, &b_id); 8950 } 8951 EXPORT_SYMBOL(netdev_port_same_parent_id); 8952 8953 /** 8954 * dev_change_proto_down - set carrier according to proto_down. 8955 * 8956 * @dev: device 8957 * @proto_down: new value 8958 */ 8959 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8960 { 8961 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8962 return -EOPNOTSUPP; 8963 if (!netif_device_present(dev)) 8964 return -ENODEV; 8965 if (proto_down) 8966 netif_carrier_off(dev); 8967 else 8968 netif_carrier_on(dev); 8969 dev->proto_down = proto_down; 8970 return 0; 8971 } 8972 8973 /** 8974 * dev_change_proto_down_reason - proto down reason 8975 * 8976 * @dev: device 8977 * @mask: proto down mask 8978 * @value: proto down value 8979 */ 8980 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8981 u32 value) 8982 { 8983 int b; 8984 8985 if (!mask) { 8986 dev->proto_down_reason = value; 8987 } else { 8988 for_each_set_bit(b, &mask, 32) { 8989 if (value & (1 << b)) 8990 dev->proto_down_reason |= BIT(b); 8991 else 8992 dev->proto_down_reason &= ~BIT(b); 8993 } 8994 } 8995 } 8996 8997 struct bpf_xdp_link { 8998 struct bpf_link link; 8999 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9000 int flags; 9001 }; 9002 9003 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9004 { 9005 if (flags & XDP_FLAGS_HW_MODE) 9006 return XDP_MODE_HW; 9007 if (flags & XDP_FLAGS_DRV_MODE) 9008 return XDP_MODE_DRV; 9009 if (flags & XDP_FLAGS_SKB_MODE) 9010 return XDP_MODE_SKB; 9011 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9012 } 9013 9014 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9015 { 9016 switch (mode) { 9017 case XDP_MODE_SKB: 9018 return generic_xdp_install; 9019 case XDP_MODE_DRV: 9020 case XDP_MODE_HW: 9021 return dev->netdev_ops->ndo_bpf; 9022 default: 9023 return NULL; 9024 } 9025 } 9026 9027 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9028 enum bpf_xdp_mode mode) 9029 { 9030 return dev->xdp_state[mode].link; 9031 } 9032 9033 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9034 enum bpf_xdp_mode mode) 9035 { 9036 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9037 9038 if (link) 9039 return link->link.prog; 9040 return dev->xdp_state[mode].prog; 9041 } 9042 9043 u8 dev_xdp_prog_count(struct net_device *dev) 9044 { 9045 u8 count = 0; 9046 int i; 9047 9048 for (i = 0; i < __MAX_XDP_MODE; i++) 9049 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9050 count++; 9051 return count; 9052 } 9053 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9054 9055 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9056 { 9057 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9058 9059 return prog ? prog->aux->id : 0; 9060 } 9061 9062 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9063 struct bpf_xdp_link *link) 9064 { 9065 dev->xdp_state[mode].link = link; 9066 dev->xdp_state[mode].prog = NULL; 9067 } 9068 9069 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9070 struct bpf_prog *prog) 9071 { 9072 dev->xdp_state[mode].link = NULL; 9073 dev->xdp_state[mode].prog = prog; 9074 } 9075 9076 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9077 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9078 u32 flags, struct bpf_prog *prog) 9079 { 9080 struct netdev_bpf xdp; 9081 int err; 9082 9083 memset(&xdp, 0, sizeof(xdp)); 9084 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9085 xdp.extack = extack; 9086 xdp.flags = flags; 9087 xdp.prog = prog; 9088 9089 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9090 * "moved" into driver), so they don't increment it on their own, but 9091 * they do decrement refcnt when program is detached or replaced. 9092 * Given net_device also owns link/prog, we need to bump refcnt here 9093 * to prevent drivers from underflowing it. 9094 */ 9095 if (prog) 9096 bpf_prog_inc(prog); 9097 err = bpf_op(dev, &xdp); 9098 if (err) { 9099 if (prog) 9100 bpf_prog_put(prog); 9101 return err; 9102 } 9103 9104 if (mode != XDP_MODE_HW) 9105 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9106 9107 return 0; 9108 } 9109 9110 static void dev_xdp_uninstall(struct net_device *dev) 9111 { 9112 struct bpf_xdp_link *link; 9113 struct bpf_prog *prog; 9114 enum bpf_xdp_mode mode; 9115 bpf_op_t bpf_op; 9116 9117 ASSERT_RTNL(); 9118 9119 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9120 prog = dev_xdp_prog(dev, mode); 9121 if (!prog) 9122 continue; 9123 9124 bpf_op = dev_xdp_bpf_op(dev, mode); 9125 if (!bpf_op) 9126 continue; 9127 9128 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9129 9130 /* auto-detach link from net device */ 9131 link = dev_xdp_link(dev, mode); 9132 if (link) 9133 link->dev = NULL; 9134 else 9135 bpf_prog_put(prog); 9136 9137 dev_xdp_set_link(dev, mode, NULL); 9138 } 9139 } 9140 9141 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9142 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9143 struct bpf_prog *old_prog, u32 flags) 9144 { 9145 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9146 struct bpf_prog *cur_prog; 9147 struct net_device *upper; 9148 struct list_head *iter; 9149 enum bpf_xdp_mode mode; 9150 bpf_op_t bpf_op; 9151 int err; 9152 9153 ASSERT_RTNL(); 9154 9155 /* either link or prog attachment, never both */ 9156 if (link && (new_prog || old_prog)) 9157 return -EINVAL; 9158 /* link supports only XDP mode flags */ 9159 if (link && (flags & ~XDP_FLAGS_MODES)) { 9160 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9161 return -EINVAL; 9162 } 9163 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9164 if (num_modes > 1) { 9165 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9166 return -EINVAL; 9167 } 9168 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9169 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9170 NL_SET_ERR_MSG(extack, 9171 "More than one program loaded, unset mode is ambiguous"); 9172 return -EINVAL; 9173 } 9174 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9175 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9176 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9177 return -EINVAL; 9178 } 9179 9180 mode = dev_xdp_mode(dev, flags); 9181 /* can't replace attached link */ 9182 if (dev_xdp_link(dev, mode)) { 9183 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9184 return -EBUSY; 9185 } 9186 9187 /* don't allow if an upper device already has a program */ 9188 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9189 if (dev_xdp_prog_count(upper) > 0) { 9190 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9191 return -EEXIST; 9192 } 9193 } 9194 9195 cur_prog = dev_xdp_prog(dev, mode); 9196 /* can't replace attached prog with link */ 9197 if (link && cur_prog) { 9198 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9199 return -EBUSY; 9200 } 9201 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9202 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9203 return -EEXIST; 9204 } 9205 9206 /* put effective new program into new_prog */ 9207 if (link) 9208 new_prog = link->link.prog; 9209 9210 if (new_prog) { 9211 bool offload = mode == XDP_MODE_HW; 9212 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9213 ? XDP_MODE_DRV : XDP_MODE_SKB; 9214 9215 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9216 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9217 return -EBUSY; 9218 } 9219 if (!offload && dev_xdp_prog(dev, other_mode)) { 9220 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9221 return -EEXIST; 9222 } 9223 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9224 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9225 return -EINVAL; 9226 } 9227 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9228 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9229 return -EINVAL; 9230 } 9231 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9232 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9233 return -EINVAL; 9234 } 9235 } 9236 9237 /* don't call drivers if the effective program didn't change */ 9238 if (new_prog != cur_prog) { 9239 bpf_op = dev_xdp_bpf_op(dev, mode); 9240 if (!bpf_op) { 9241 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9242 return -EOPNOTSUPP; 9243 } 9244 9245 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9246 if (err) 9247 return err; 9248 } 9249 9250 if (link) 9251 dev_xdp_set_link(dev, mode, link); 9252 else 9253 dev_xdp_set_prog(dev, mode, new_prog); 9254 if (cur_prog) 9255 bpf_prog_put(cur_prog); 9256 9257 return 0; 9258 } 9259 9260 static int dev_xdp_attach_link(struct net_device *dev, 9261 struct netlink_ext_ack *extack, 9262 struct bpf_xdp_link *link) 9263 { 9264 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9265 } 9266 9267 static int dev_xdp_detach_link(struct net_device *dev, 9268 struct netlink_ext_ack *extack, 9269 struct bpf_xdp_link *link) 9270 { 9271 enum bpf_xdp_mode mode; 9272 bpf_op_t bpf_op; 9273 9274 ASSERT_RTNL(); 9275 9276 mode = dev_xdp_mode(dev, link->flags); 9277 if (dev_xdp_link(dev, mode) != link) 9278 return -EINVAL; 9279 9280 bpf_op = dev_xdp_bpf_op(dev, mode); 9281 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9282 dev_xdp_set_link(dev, mode, NULL); 9283 return 0; 9284 } 9285 9286 static void bpf_xdp_link_release(struct bpf_link *link) 9287 { 9288 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9289 9290 rtnl_lock(); 9291 9292 /* if racing with net_device's tear down, xdp_link->dev might be 9293 * already NULL, in which case link was already auto-detached 9294 */ 9295 if (xdp_link->dev) { 9296 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9297 xdp_link->dev = NULL; 9298 } 9299 9300 rtnl_unlock(); 9301 } 9302 9303 static int bpf_xdp_link_detach(struct bpf_link *link) 9304 { 9305 bpf_xdp_link_release(link); 9306 return 0; 9307 } 9308 9309 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9310 { 9311 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9312 9313 kfree(xdp_link); 9314 } 9315 9316 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9317 struct seq_file *seq) 9318 { 9319 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9320 u32 ifindex = 0; 9321 9322 rtnl_lock(); 9323 if (xdp_link->dev) 9324 ifindex = xdp_link->dev->ifindex; 9325 rtnl_unlock(); 9326 9327 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9328 } 9329 9330 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9331 struct bpf_link_info *info) 9332 { 9333 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9334 u32 ifindex = 0; 9335 9336 rtnl_lock(); 9337 if (xdp_link->dev) 9338 ifindex = xdp_link->dev->ifindex; 9339 rtnl_unlock(); 9340 9341 info->xdp.ifindex = ifindex; 9342 return 0; 9343 } 9344 9345 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9346 struct bpf_prog *old_prog) 9347 { 9348 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9349 enum bpf_xdp_mode mode; 9350 bpf_op_t bpf_op; 9351 int err = 0; 9352 9353 rtnl_lock(); 9354 9355 /* link might have been auto-released already, so fail */ 9356 if (!xdp_link->dev) { 9357 err = -ENOLINK; 9358 goto out_unlock; 9359 } 9360 9361 if (old_prog && link->prog != old_prog) { 9362 err = -EPERM; 9363 goto out_unlock; 9364 } 9365 old_prog = link->prog; 9366 if (old_prog->type != new_prog->type || 9367 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9368 err = -EINVAL; 9369 goto out_unlock; 9370 } 9371 9372 if (old_prog == new_prog) { 9373 /* no-op, don't disturb drivers */ 9374 bpf_prog_put(new_prog); 9375 goto out_unlock; 9376 } 9377 9378 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9379 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9380 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9381 xdp_link->flags, new_prog); 9382 if (err) 9383 goto out_unlock; 9384 9385 old_prog = xchg(&link->prog, new_prog); 9386 bpf_prog_put(old_prog); 9387 9388 out_unlock: 9389 rtnl_unlock(); 9390 return err; 9391 } 9392 9393 static const struct bpf_link_ops bpf_xdp_link_lops = { 9394 .release = bpf_xdp_link_release, 9395 .dealloc = bpf_xdp_link_dealloc, 9396 .detach = bpf_xdp_link_detach, 9397 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9398 .fill_link_info = bpf_xdp_link_fill_link_info, 9399 .update_prog = bpf_xdp_link_update, 9400 }; 9401 9402 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9403 { 9404 struct net *net = current->nsproxy->net_ns; 9405 struct bpf_link_primer link_primer; 9406 struct bpf_xdp_link *link; 9407 struct net_device *dev; 9408 int err, fd; 9409 9410 rtnl_lock(); 9411 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9412 if (!dev) { 9413 rtnl_unlock(); 9414 return -EINVAL; 9415 } 9416 9417 link = kzalloc(sizeof(*link), GFP_USER); 9418 if (!link) { 9419 err = -ENOMEM; 9420 goto unlock; 9421 } 9422 9423 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9424 link->dev = dev; 9425 link->flags = attr->link_create.flags; 9426 9427 err = bpf_link_prime(&link->link, &link_primer); 9428 if (err) { 9429 kfree(link); 9430 goto unlock; 9431 } 9432 9433 err = dev_xdp_attach_link(dev, NULL, link); 9434 rtnl_unlock(); 9435 9436 if (err) { 9437 link->dev = NULL; 9438 bpf_link_cleanup(&link_primer); 9439 goto out_put_dev; 9440 } 9441 9442 fd = bpf_link_settle(&link_primer); 9443 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9444 dev_put(dev); 9445 return fd; 9446 9447 unlock: 9448 rtnl_unlock(); 9449 9450 out_put_dev: 9451 dev_put(dev); 9452 return err; 9453 } 9454 9455 /** 9456 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9457 * @dev: device 9458 * @extack: netlink extended ack 9459 * @fd: new program fd or negative value to clear 9460 * @expected_fd: old program fd that userspace expects to replace or clear 9461 * @flags: xdp-related flags 9462 * 9463 * Set or clear a bpf program for a device 9464 */ 9465 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9466 int fd, int expected_fd, u32 flags) 9467 { 9468 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9469 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9470 int err; 9471 9472 ASSERT_RTNL(); 9473 9474 if (fd >= 0) { 9475 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9476 mode != XDP_MODE_SKB); 9477 if (IS_ERR(new_prog)) 9478 return PTR_ERR(new_prog); 9479 } 9480 9481 if (expected_fd >= 0) { 9482 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9483 mode != XDP_MODE_SKB); 9484 if (IS_ERR(old_prog)) { 9485 err = PTR_ERR(old_prog); 9486 old_prog = NULL; 9487 goto err_out; 9488 } 9489 } 9490 9491 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9492 9493 err_out: 9494 if (err && new_prog) 9495 bpf_prog_put(new_prog); 9496 if (old_prog) 9497 bpf_prog_put(old_prog); 9498 return err; 9499 } 9500 9501 /** 9502 * dev_new_index - allocate an ifindex 9503 * @net: the applicable net namespace 9504 * 9505 * Returns a suitable unique value for a new device interface 9506 * number. The caller must hold the rtnl semaphore or the 9507 * dev_base_lock to be sure it remains unique. 9508 */ 9509 static int dev_new_index(struct net *net) 9510 { 9511 int ifindex = net->ifindex; 9512 9513 for (;;) { 9514 if (++ifindex <= 0) 9515 ifindex = 1; 9516 if (!__dev_get_by_index(net, ifindex)) 9517 return net->ifindex = ifindex; 9518 } 9519 } 9520 9521 /* Delayed registration/unregisteration */ 9522 LIST_HEAD(net_todo_list); 9523 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9524 9525 static void net_set_todo(struct net_device *dev) 9526 { 9527 list_add_tail(&dev->todo_list, &net_todo_list); 9528 atomic_inc(&dev_net(dev)->dev_unreg_count); 9529 } 9530 9531 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9532 struct net_device *upper, netdev_features_t features) 9533 { 9534 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9535 netdev_features_t feature; 9536 int feature_bit; 9537 9538 for_each_netdev_feature(upper_disables, feature_bit) { 9539 feature = __NETIF_F_BIT(feature_bit); 9540 if (!(upper->wanted_features & feature) 9541 && (features & feature)) { 9542 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9543 &feature, upper->name); 9544 features &= ~feature; 9545 } 9546 } 9547 9548 return features; 9549 } 9550 9551 static void netdev_sync_lower_features(struct net_device *upper, 9552 struct net_device *lower, netdev_features_t features) 9553 { 9554 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9555 netdev_features_t feature; 9556 int feature_bit; 9557 9558 for_each_netdev_feature(upper_disables, feature_bit) { 9559 feature = __NETIF_F_BIT(feature_bit); 9560 if (!(features & feature) && (lower->features & feature)) { 9561 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9562 &feature, lower->name); 9563 lower->wanted_features &= ~feature; 9564 __netdev_update_features(lower); 9565 9566 if (unlikely(lower->features & feature)) 9567 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9568 &feature, lower->name); 9569 else 9570 netdev_features_change(lower); 9571 } 9572 } 9573 } 9574 9575 static netdev_features_t netdev_fix_features(struct net_device *dev, 9576 netdev_features_t features) 9577 { 9578 /* Fix illegal checksum combinations */ 9579 if ((features & NETIF_F_HW_CSUM) && 9580 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9581 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9582 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9583 } 9584 9585 /* TSO requires that SG is present as well. */ 9586 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9587 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9588 features &= ~NETIF_F_ALL_TSO; 9589 } 9590 9591 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9592 !(features & NETIF_F_IP_CSUM)) { 9593 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9594 features &= ~NETIF_F_TSO; 9595 features &= ~NETIF_F_TSO_ECN; 9596 } 9597 9598 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9599 !(features & NETIF_F_IPV6_CSUM)) { 9600 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9601 features &= ~NETIF_F_TSO6; 9602 } 9603 9604 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9605 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9606 features &= ~NETIF_F_TSO_MANGLEID; 9607 9608 /* TSO ECN requires that TSO is present as well. */ 9609 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9610 features &= ~NETIF_F_TSO_ECN; 9611 9612 /* Software GSO depends on SG. */ 9613 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9614 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9615 features &= ~NETIF_F_GSO; 9616 } 9617 9618 /* GSO partial features require GSO partial be set */ 9619 if ((features & dev->gso_partial_features) && 9620 !(features & NETIF_F_GSO_PARTIAL)) { 9621 netdev_dbg(dev, 9622 "Dropping partially supported GSO features since no GSO partial.\n"); 9623 features &= ~dev->gso_partial_features; 9624 } 9625 9626 if (!(features & NETIF_F_RXCSUM)) { 9627 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9628 * successfully merged by hardware must also have the 9629 * checksum verified by hardware. If the user does not 9630 * want to enable RXCSUM, logically, we should disable GRO_HW. 9631 */ 9632 if (features & NETIF_F_GRO_HW) { 9633 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9634 features &= ~NETIF_F_GRO_HW; 9635 } 9636 } 9637 9638 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9639 if (features & NETIF_F_RXFCS) { 9640 if (features & NETIF_F_LRO) { 9641 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9642 features &= ~NETIF_F_LRO; 9643 } 9644 9645 if (features & NETIF_F_GRO_HW) { 9646 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9647 features &= ~NETIF_F_GRO_HW; 9648 } 9649 } 9650 9651 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9652 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9653 features &= ~NETIF_F_LRO; 9654 } 9655 9656 if (features & NETIF_F_HW_TLS_TX) { 9657 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9658 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9659 bool hw_csum = features & NETIF_F_HW_CSUM; 9660 9661 if (!ip_csum && !hw_csum) { 9662 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9663 features &= ~NETIF_F_HW_TLS_TX; 9664 } 9665 } 9666 9667 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9668 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9669 features &= ~NETIF_F_HW_TLS_RX; 9670 } 9671 9672 return features; 9673 } 9674 9675 int __netdev_update_features(struct net_device *dev) 9676 { 9677 struct net_device *upper, *lower; 9678 netdev_features_t features; 9679 struct list_head *iter; 9680 int err = -1; 9681 9682 ASSERT_RTNL(); 9683 9684 features = netdev_get_wanted_features(dev); 9685 9686 if (dev->netdev_ops->ndo_fix_features) 9687 features = dev->netdev_ops->ndo_fix_features(dev, features); 9688 9689 /* driver might be less strict about feature dependencies */ 9690 features = netdev_fix_features(dev, features); 9691 9692 /* some features can't be enabled if they're off on an upper device */ 9693 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9694 features = netdev_sync_upper_features(dev, upper, features); 9695 9696 if (dev->features == features) 9697 goto sync_lower; 9698 9699 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9700 &dev->features, &features); 9701 9702 if (dev->netdev_ops->ndo_set_features) 9703 err = dev->netdev_ops->ndo_set_features(dev, features); 9704 else 9705 err = 0; 9706 9707 if (unlikely(err < 0)) { 9708 netdev_err(dev, 9709 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9710 err, &features, &dev->features); 9711 /* return non-0 since some features might have changed and 9712 * it's better to fire a spurious notification than miss it 9713 */ 9714 return -1; 9715 } 9716 9717 sync_lower: 9718 /* some features must be disabled on lower devices when disabled 9719 * on an upper device (think: bonding master or bridge) 9720 */ 9721 netdev_for_each_lower_dev(dev, lower, iter) 9722 netdev_sync_lower_features(dev, lower, features); 9723 9724 if (!err) { 9725 netdev_features_t diff = features ^ dev->features; 9726 9727 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9728 /* udp_tunnel_{get,drop}_rx_info both need 9729 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9730 * device, or they won't do anything. 9731 * Thus we need to update dev->features 9732 * *before* calling udp_tunnel_get_rx_info, 9733 * but *after* calling udp_tunnel_drop_rx_info. 9734 */ 9735 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9736 dev->features = features; 9737 udp_tunnel_get_rx_info(dev); 9738 } else { 9739 udp_tunnel_drop_rx_info(dev); 9740 } 9741 } 9742 9743 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9744 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9745 dev->features = features; 9746 err |= vlan_get_rx_ctag_filter_info(dev); 9747 } else { 9748 vlan_drop_rx_ctag_filter_info(dev); 9749 } 9750 } 9751 9752 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9753 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9754 dev->features = features; 9755 err |= vlan_get_rx_stag_filter_info(dev); 9756 } else { 9757 vlan_drop_rx_stag_filter_info(dev); 9758 } 9759 } 9760 9761 dev->features = features; 9762 } 9763 9764 return err < 0 ? 0 : 1; 9765 } 9766 9767 /** 9768 * netdev_update_features - recalculate device features 9769 * @dev: the device to check 9770 * 9771 * Recalculate dev->features set and send notifications if it 9772 * has changed. Should be called after driver or hardware dependent 9773 * conditions might have changed that influence the features. 9774 */ 9775 void netdev_update_features(struct net_device *dev) 9776 { 9777 if (__netdev_update_features(dev)) 9778 netdev_features_change(dev); 9779 } 9780 EXPORT_SYMBOL(netdev_update_features); 9781 9782 /** 9783 * netdev_change_features - recalculate device features 9784 * @dev: the device to check 9785 * 9786 * Recalculate dev->features set and send notifications even 9787 * if they have not changed. Should be called instead of 9788 * netdev_update_features() if also dev->vlan_features might 9789 * have changed to allow the changes to be propagated to stacked 9790 * VLAN devices. 9791 */ 9792 void netdev_change_features(struct net_device *dev) 9793 { 9794 __netdev_update_features(dev); 9795 netdev_features_change(dev); 9796 } 9797 EXPORT_SYMBOL(netdev_change_features); 9798 9799 /** 9800 * netif_stacked_transfer_operstate - transfer operstate 9801 * @rootdev: the root or lower level device to transfer state from 9802 * @dev: the device to transfer operstate to 9803 * 9804 * Transfer operational state from root to device. This is normally 9805 * called when a stacking relationship exists between the root 9806 * device and the device(a leaf device). 9807 */ 9808 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9809 struct net_device *dev) 9810 { 9811 if (rootdev->operstate == IF_OPER_DORMANT) 9812 netif_dormant_on(dev); 9813 else 9814 netif_dormant_off(dev); 9815 9816 if (rootdev->operstate == IF_OPER_TESTING) 9817 netif_testing_on(dev); 9818 else 9819 netif_testing_off(dev); 9820 9821 if (netif_carrier_ok(rootdev)) 9822 netif_carrier_on(dev); 9823 else 9824 netif_carrier_off(dev); 9825 } 9826 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9827 9828 static int netif_alloc_rx_queues(struct net_device *dev) 9829 { 9830 unsigned int i, count = dev->num_rx_queues; 9831 struct netdev_rx_queue *rx; 9832 size_t sz = count * sizeof(*rx); 9833 int err = 0; 9834 9835 BUG_ON(count < 1); 9836 9837 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9838 if (!rx) 9839 return -ENOMEM; 9840 9841 dev->_rx = rx; 9842 9843 for (i = 0; i < count; i++) { 9844 rx[i].dev = dev; 9845 9846 /* XDP RX-queue setup */ 9847 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9848 if (err < 0) 9849 goto err_rxq_info; 9850 } 9851 return 0; 9852 9853 err_rxq_info: 9854 /* Rollback successful reg's and free other resources */ 9855 while (i--) 9856 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9857 kvfree(dev->_rx); 9858 dev->_rx = NULL; 9859 return err; 9860 } 9861 9862 static void netif_free_rx_queues(struct net_device *dev) 9863 { 9864 unsigned int i, count = dev->num_rx_queues; 9865 9866 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9867 if (!dev->_rx) 9868 return; 9869 9870 for (i = 0; i < count; i++) 9871 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9872 9873 kvfree(dev->_rx); 9874 } 9875 9876 static void netdev_init_one_queue(struct net_device *dev, 9877 struct netdev_queue *queue, void *_unused) 9878 { 9879 /* Initialize queue lock */ 9880 spin_lock_init(&queue->_xmit_lock); 9881 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9882 queue->xmit_lock_owner = -1; 9883 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9884 queue->dev = dev; 9885 #ifdef CONFIG_BQL 9886 dql_init(&queue->dql, HZ); 9887 #endif 9888 } 9889 9890 static void netif_free_tx_queues(struct net_device *dev) 9891 { 9892 kvfree(dev->_tx); 9893 } 9894 9895 static int netif_alloc_netdev_queues(struct net_device *dev) 9896 { 9897 unsigned int count = dev->num_tx_queues; 9898 struct netdev_queue *tx; 9899 size_t sz = count * sizeof(*tx); 9900 9901 if (count < 1 || count > 0xffff) 9902 return -EINVAL; 9903 9904 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9905 if (!tx) 9906 return -ENOMEM; 9907 9908 dev->_tx = tx; 9909 9910 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9911 spin_lock_init(&dev->tx_global_lock); 9912 9913 return 0; 9914 } 9915 9916 void netif_tx_stop_all_queues(struct net_device *dev) 9917 { 9918 unsigned int i; 9919 9920 for (i = 0; i < dev->num_tx_queues; i++) { 9921 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9922 9923 netif_tx_stop_queue(txq); 9924 } 9925 } 9926 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9927 9928 /** 9929 * register_netdevice - register a network device 9930 * @dev: device to register 9931 * 9932 * Take a completed network device structure and add it to the kernel 9933 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9934 * chain. 0 is returned on success. A negative errno code is returned 9935 * on a failure to set up the device, or if the name is a duplicate. 9936 * 9937 * Callers must hold the rtnl semaphore. You may want 9938 * register_netdev() instead of this. 9939 * 9940 * BUGS: 9941 * The locking appears insufficient to guarantee two parallel registers 9942 * will not get the same name. 9943 */ 9944 9945 int register_netdevice(struct net_device *dev) 9946 { 9947 int ret; 9948 struct net *net = dev_net(dev); 9949 9950 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9951 NETDEV_FEATURE_COUNT); 9952 BUG_ON(dev_boot_phase); 9953 ASSERT_RTNL(); 9954 9955 might_sleep(); 9956 9957 /* When net_device's are persistent, this will be fatal. */ 9958 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9959 BUG_ON(!net); 9960 9961 ret = ethtool_check_ops(dev->ethtool_ops); 9962 if (ret) 9963 return ret; 9964 9965 spin_lock_init(&dev->addr_list_lock); 9966 netdev_set_addr_lockdep_class(dev); 9967 9968 ret = dev_get_valid_name(net, dev, dev->name); 9969 if (ret < 0) 9970 goto out; 9971 9972 ret = -ENOMEM; 9973 dev->name_node = netdev_name_node_head_alloc(dev); 9974 if (!dev->name_node) 9975 goto out; 9976 9977 /* Init, if this function is available */ 9978 if (dev->netdev_ops->ndo_init) { 9979 ret = dev->netdev_ops->ndo_init(dev); 9980 if (ret) { 9981 if (ret > 0) 9982 ret = -EIO; 9983 goto err_free_name; 9984 } 9985 } 9986 9987 if (((dev->hw_features | dev->features) & 9988 NETIF_F_HW_VLAN_CTAG_FILTER) && 9989 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9990 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9991 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9992 ret = -EINVAL; 9993 goto err_uninit; 9994 } 9995 9996 ret = -EBUSY; 9997 if (!dev->ifindex) 9998 dev->ifindex = dev_new_index(net); 9999 else if (__dev_get_by_index(net, dev->ifindex)) 10000 goto err_uninit; 10001 10002 /* Transfer changeable features to wanted_features and enable 10003 * software offloads (GSO and GRO). 10004 */ 10005 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10006 dev->features |= NETIF_F_SOFT_FEATURES; 10007 10008 if (dev->udp_tunnel_nic_info) { 10009 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10010 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10011 } 10012 10013 dev->wanted_features = dev->features & dev->hw_features; 10014 10015 if (!(dev->flags & IFF_LOOPBACK)) 10016 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10017 10018 /* If IPv4 TCP segmentation offload is supported we should also 10019 * allow the device to enable segmenting the frame with the option 10020 * of ignoring a static IP ID value. This doesn't enable the 10021 * feature itself but allows the user to enable it later. 10022 */ 10023 if (dev->hw_features & NETIF_F_TSO) 10024 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10025 if (dev->vlan_features & NETIF_F_TSO) 10026 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10027 if (dev->mpls_features & NETIF_F_TSO) 10028 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10029 if (dev->hw_enc_features & NETIF_F_TSO) 10030 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10031 10032 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10033 */ 10034 dev->vlan_features |= NETIF_F_HIGHDMA; 10035 10036 /* Make NETIF_F_SG inheritable to tunnel devices. 10037 */ 10038 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10039 10040 /* Make NETIF_F_SG inheritable to MPLS. 10041 */ 10042 dev->mpls_features |= NETIF_F_SG; 10043 10044 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10045 ret = notifier_to_errno(ret); 10046 if (ret) 10047 goto err_uninit; 10048 10049 ret = netdev_register_kobject(dev); 10050 if (ret) { 10051 dev->reg_state = NETREG_UNREGISTERED; 10052 goto err_uninit; 10053 } 10054 dev->reg_state = NETREG_REGISTERED; 10055 10056 __netdev_update_features(dev); 10057 10058 /* 10059 * Default initial state at registry is that the 10060 * device is present. 10061 */ 10062 10063 set_bit(__LINK_STATE_PRESENT, &dev->state); 10064 10065 linkwatch_init_dev(dev); 10066 10067 dev_init_scheduler(dev); 10068 10069 dev_hold_track(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10070 list_netdevice(dev); 10071 10072 add_device_randomness(dev->dev_addr, dev->addr_len); 10073 10074 /* If the device has permanent device address, driver should 10075 * set dev_addr and also addr_assign_type should be set to 10076 * NET_ADDR_PERM (default value). 10077 */ 10078 if (dev->addr_assign_type == NET_ADDR_PERM) 10079 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10080 10081 /* Notify protocols, that a new device appeared. */ 10082 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10083 ret = notifier_to_errno(ret); 10084 if (ret) { 10085 /* Expect explicit free_netdev() on failure */ 10086 dev->needs_free_netdev = false; 10087 unregister_netdevice_queue(dev, NULL); 10088 goto out; 10089 } 10090 /* 10091 * Prevent userspace races by waiting until the network 10092 * device is fully setup before sending notifications. 10093 */ 10094 if (!dev->rtnl_link_ops || 10095 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10096 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10097 10098 out: 10099 return ret; 10100 10101 err_uninit: 10102 if (dev->netdev_ops->ndo_uninit) 10103 dev->netdev_ops->ndo_uninit(dev); 10104 if (dev->priv_destructor) 10105 dev->priv_destructor(dev); 10106 err_free_name: 10107 netdev_name_node_free(dev->name_node); 10108 goto out; 10109 } 10110 EXPORT_SYMBOL(register_netdevice); 10111 10112 /** 10113 * init_dummy_netdev - init a dummy network device for NAPI 10114 * @dev: device to init 10115 * 10116 * This takes a network device structure and initialize the minimum 10117 * amount of fields so it can be used to schedule NAPI polls without 10118 * registering a full blown interface. This is to be used by drivers 10119 * that need to tie several hardware interfaces to a single NAPI 10120 * poll scheduler due to HW limitations. 10121 */ 10122 int init_dummy_netdev(struct net_device *dev) 10123 { 10124 /* Clear everything. Note we don't initialize spinlocks 10125 * are they aren't supposed to be taken by any of the 10126 * NAPI code and this dummy netdev is supposed to be 10127 * only ever used for NAPI polls 10128 */ 10129 memset(dev, 0, sizeof(struct net_device)); 10130 10131 /* make sure we BUG if trying to hit standard 10132 * register/unregister code path 10133 */ 10134 dev->reg_state = NETREG_DUMMY; 10135 10136 /* NAPI wants this */ 10137 INIT_LIST_HEAD(&dev->napi_list); 10138 10139 /* a dummy interface is started by default */ 10140 set_bit(__LINK_STATE_PRESENT, &dev->state); 10141 set_bit(__LINK_STATE_START, &dev->state); 10142 10143 /* napi_busy_loop stats accounting wants this */ 10144 dev_net_set(dev, &init_net); 10145 10146 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10147 * because users of this 'device' dont need to change 10148 * its refcount. 10149 */ 10150 10151 return 0; 10152 } 10153 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10154 10155 10156 /** 10157 * register_netdev - register a network device 10158 * @dev: device to register 10159 * 10160 * Take a completed network device structure and add it to the kernel 10161 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10162 * chain. 0 is returned on success. A negative errno code is returned 10163 * on a failure to set up the device, or if the name is a duplicate. 10164 * 10165 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10166 * and expands the device name if you passed a format string to 10167 * alloc_netdev. 10168 */ 10169 int register_netdev(struct net_device *dev) 10170 { 10171 int err; 10172 10173 if (rtnl_lock_killable()) 10174 return -EINTR; 10175 err = register_netdevice(dev); 10176 rtnl_unlock(); 10177 return err; 10178 } 10179 EXPORT_SYMBOL(register_netdev); 10180 10181 int netdev_refcnt_read(const struct net_device *dev) 10182 { 10183 #ifdef CONFIG_PCPU_DEV_REFCNT 10184 int i, refcnt = 0; 10185 10186 for_each_possible_cpu(i) 10187 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10188 return refcnt; 10189 #else 10190 return refcount_read(&dev->dev_refcnt); 10191 #endif 10192 } 10193 EXPORT_SYMBOL(netdev_refcnt_read); 10194 10195 int netdev_unregister_timeout_secs __read_mostly = 10; 10196 10197 #define WAIT_REFS_MIN_MSECS 1 10198 #define WAIT_REFS_MAX_MSECS 250 10199 /** 10200 * netdev_wait_allrefs_any - wait until all references are gone. 10201 * @list: list of net_devices to wait on 10202 * 10203 * This is called when unregistering network devices. 10204 * 10205 * Any protocol or device that holds a reference should register 10206 * for netdevice notification, and cleanup and put back the 10207 * reference if they receive an UNREGISTER event. 10208 * We can get stuck here if buggy protocols don't correctly 10209 * call dev_put. 10210 */ 10211 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10212 { 10213 unsigned long rebroadcast_time, warning_time; 10214 struct net_device *dev; 10215 int wait = 0; 10216 10217 rebroadcast_time = warning_time = jiffies; 10218 10219 list_for_each_entry(dev, list, todo_list) 10220 if (netdev_refcnt_read(dev) == 1) 10221 return dev; 10222 10223 while (true) { 10224 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10225 rtnl_lock(); 10226 10227 /* Rebroadcast unregister notification */ 10228 list_for_each_entry(dev, list, todo_list) 10229 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10230 10231 __rtnl_unlock(); 10232 rcu_barrier(); 10233 rtnl_lock(); 10234 10235 list_for_each_entry(dev, list, todo_list) 10236 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10237 &dev->state)) { 10238 /* We must not have linkwatch events 10239 * pending on unregister. If this 10240 * happens, we simply run the queue 10241 * unscheduled, resulting in a noop 10242 * for this device. 10243 */ 10244 linkwatch_run_queue(); 10245 break; 10246 } 10247 10248 __rtnl_unlock(); 10249 10250 rebroadcast_time = jiffies; 10251 } 10252 10253 if (!wait) { 10254 rcu_barrier(); 10255 wait = WAIT_REFS_MIN_MSECS; 10256 } else { 10257 msleep(wait); 10258 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10259 } 10260 10261 list_for_each_entry(dev, list, todo_list) 10262 if (netdev_refcnt_read(dev) == 1) 10263 return dev; 10264 10265 if (time_after(jiffies, warning_time + 10266 netdev_unregister_timeout_secs * HZ)) { 10267 list_for_each_entry(dev, list, todo_list) { 10268 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10269 dev->name, netdev_refcnt_read(dev)); 10270 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10271 } 10272 10273 warning_time = jiffies; 10274 } 10275 } 10276 } 10277 10278 /* The sequence is: 10279 * 10280 * rtnl_lock(); 10281 * ... 10282 * register_netdevice(x1); 10283 * register_netdevice(x2); 10284 * ... 10285 * unregister_netdevice(y1); 10286 * unregister_netdevice(y2); 10287 * ... 10288 * rtnl_unlock(); 10289 * free_netdev(y1); 10290 * free_netdev(y2); 10291 * 10292 * We are invoked by rtnl_unlock(). 10293 * This allows us to deal with problems: 10294 * 1) We can delete sysfs objects which invoke hotplug 10295 * without deadlocking with linkwatch via keventd. 10296 * 2) Since we run with the RTNL semaphore not held, we can sleep 10297 * safely in order to wait for the netdev refcnt to drop to zero. 10298 * 10299 * We must not return until all unregister events added during 10300 * the interval the lock was held have been completed. 10301 */ 10302 void netdev_run_todo(void) 10303 { 10304 struct net_device *dev, *tmp; 10305 struct list_head list; 10306 #ifdef CONFIG_LOCKDEP 10307 struct list_head unlink_list; 10308 10309 list_replace_init(&net_unlink_list, &unlink_list); 10310 10311 while (!list_empty(&unlink_list)) { 10312 struct net_device *dev = list_first_entry(&unlink_list, 10313 struct net_device, 10314 unlink_list); 10315 list_del_init(&dev->unlink_list); 10316 dev->nested_level = dev->lower_level - 1; 10317 } 10318 #endif 10319 10320 /* Snapshot list, allow later requests */ 10321 list_replace_init(&net_todo_list, &list); 10322 10323 __rtnl_unlock(); 10324 10325 /* Wait for rcu callbacks to finish before next phase */ 10326 if (!list_empty(&list)) 10327 rcu_barrier(); 10328 10329 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10330 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10331 netdev_WARN(dev, "run_todo but not unregistering\n"); 10332 list_del(&dev->todo_list); 10333 continue; 10334 } 10335 10336 dev->reg_state = NETREG_UNREGISTERED; 10337 linkwatch_forget_dev(dev); 10338 } 10339 10340 while (!list_empty(&list)) { 10341 dev = netdev_wait_allrefs_any(&list); 10342 list_del(&dev->todo_list); 10343 10344 /* paranoia */ 10345 BUG_ON(netdev_refcnt_read(dev) != 1); 10346 BUG_ON(!list_empty(&dev->ptype_all)); 10347 BUG_ON(!list_empty(&dev->ptype_specific)); 10348 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10349 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10350 #if IS_ENABLED(CONFIG_DECNET) 10351 WARN_ON(dev->dn_ptr); 10352 #endif 10353 if (dev->priv_destructor) 10354 dev->priv_destructor(dev); 10355 if (dev->needs_free_netdev) 10356 free_netdev(dev); 10357 10358 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10359 wake_up(&netdev_unregistering_wq); 10360 10361 /* Free network device */ 10362 kobject_put(&dev->dev.kobj); 10363 } 10364 } 10365 10366 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10367 * all the same fields in the same order as net_device_stats, with only 10368 * the type differing, but rtnl_link_stats64 may have additional fields 10369 * at the end for newer counters. 10370 */ 10371 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10372 const struct net_device_stats *netdev_stats) 10373 { 10374 #if BITS_PER_LONG == 64 10375 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10376 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10377 /* zero out counters that only exist in rtnl_link_stats64 */ 10378 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10379 sizeof(*stats64) - sizeof(*netdev_stats)); 10380 #else 10381 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10382 const unsigned long *src = (const unsigned long *)netdev_stats; 10383 u64 *dst = (u64 *)stats64; 10384 10385 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10386 for (i = 0; i < n; i++) 10387 dst[i] = src[i]; 10388 /* zero out counters that only exist in rtnl_link_stats64 */ 10389 memset((char *)stats64 + n * sizeof(u64), 0, 10390 sizeof(*stats64) - n * sizeof(u64)); 10391 #endif 10392 } 10393 EXPORT_SYMBOL(netdev_stats_to_stats64); 10394 10395 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10396 { 10397 struct net_device_core_stats __percpu *p; 10398 10399 p = alloc_percpu_gfp(struct net_device_core_stats, 10400 GFP_ATOMIC | __GFP_NOWARN); 10401 10402 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10403 free_percpu(p); 10404 10405 /* This READ_ONCE() pairs with the cmpxchg() above */ 10406 return READ_ONCE(dev->core_stats); 10407 } 10408 EXPORT_SYMBOL(netdev_core_stats_alloc); 10409 10410 /** 10411 * dev_get_stats - get network device statistics 10412 * @dev: device to get statistics from 10413 * @storage: place to store stats 10414 * 10415 * Get network statistics from device. Return @storage. 10416 * The device driver may provide its own method by setting 10417 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10418 * otherwise the internal statistics structure is used. 10419 */ 10420 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10421 struct rtnl_link_stats64 *storage) 10422 { 10423 const struct net_device_ops *ops = dev->netdev_ops; 10424 const struct net_device_core_stats __percpu *p; 10425 10426 if (ops->ndo_get_stats64) { 10427 memset(storage, 0, sizeof(*storage)); 10428 ops->ndo_get_stats64(dev, storage); 10429 } else if (ops->ndo_get_stats) { 10430 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10431 } else { 10432 netdev_stats_to_stats64(storage, &dev->stats); 10433 } 10434 10435 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10436 p = READ_ONCE(dev->core_stats); 10437 if (p) { 10438 const struct net_device_core_stats *core_stats; 10439 int i; 10440 10441 for_each_possible_cpu(i) { 10442 core_stats = per_cpu_ptr(p, i); 10443 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10444 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10445 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10446 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10447 } 10448 } 10449 return storage; 10450 } 10451 EXPORT_SYMBOL(dev_get_stats); 10452 10453 /** 10454 * dev_fetch_sw_netstats - get per-cpu network device statistics 10455 * @s: place to store stats 10456 * @netstats: per-cpu network stats to read from 10457 * 10458 * Read per-cpu network statistics and populate the related fields in @s. 10459 */ 10460 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10461 const struct pcpu_sw_netstats __percpu *netstats) 10462 { 10463 int cpu; 10464 10465 for_each_possible_cpu(cpu) { 10466 const struct pcpu_sw_netstats *stats; 10467 struct pcpu_sw_netstats tmp; 10468 unsigned int start; 10469 10470 stats = per_cpu_ptr(netstats, cpu); 10471 do { 10472 start = u64_stats_fetch_begin_irq(&stats->syncp); 10473 tmp.rx_packets = stats->rx_packets; 10474 tmp.rx_bytes = stats->rx_bytes; 10475 tmp.tx_packets = stats->tx_packets; 10476 tmp.tx_bytes = stats->tx_bytes; 10477 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10478 10479 s->rx_packets += tmp.rx_packets; 10480 s->rx_bytes += tmp.rx_bytes; 10481 s->tx_packets += tmp.tx_packets; 10482 s->tx_bytes += tmp.tx_bytes; 10483 } 10484 } 10485 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10486 10487 /** 10488 * dev_get_tstats64 - ndo_get_stats64 implementation 10489 * @dev: device to get statistics from 10490 * @s: place to store stats 10491 * 10492 * Populate @s from dev->stats and dev->tstats. Can be used as 10493 * ndo_get_stats64() callback. 10494 */ 10495 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10496 { 10497 netdev_stats_to_stats64(s, &dev->stats); 10498 dev_fetch_sw_netstats(s, dev->tstats); 10499 } 10500 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10501 10502 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10503 { 10504 struct netdev_queue *queue = dev_ingress_queue(dev); 10505 10506 #ifdef CONFIG_NET_CLS_ACT 10507 if (queue) 10508 return queue; 10509 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10510 if (!queue) 10511 return NULL; 10512 netdev_init_one_queue(dev, queue, NULL); 10513 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10514 queue->qdisc_sleeping = &noop_qdisc; 10515 rcu_assign_pointer(dev->ingress_queue, queue); 10516 #endif 10517 return queue; 10518 } 10519 10520 static const struct ethtool_ops default_ethtool_ops; 10521 10522 void netdev_set_default_ethtool_ops(struct net_device *dev, 10523 const struct ethtool_ops *ops) 10524 { 10525 if (dev->ethtool_ops == &default_ethtool_ops) 10526 dev->ethtool_ops = ops; 10527 } 10528 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10529 10530 void netdev_freemem(struct net_device *dev) 10531 { 10532 char *addr = (char *)dev - dev->padded; 10533 10534 kvfree(addr); 10535 } 10536 10537 /** 10538 * alloc_netdev_mqs - allocate network device 10539 * @sizeof_priv: size of private data to allocate space for 10540 * @name: device name format string 10541 * @name_assign_type: origin of device name 10542 * @setup: callback to initialize device 10543 * @txqs: the number of TX subqueues to allocate 10544 * @rxqs: the number of RX subqueues to allocate 10545 * 10546 * Allocates a struct net_device with private data area for driver use 10547 * and performs basic initialization. Also allocates subqueue structs 10548 * for each queue on the device. 10549 */ 10550 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10551 unsigned char name_assign_type, 10552 void (*setup)(struct net_device *), 10553 unsigned int txqs, unsigned int rxqs) 10554 { 10555 struct net_device *dev; 10556 unsigned int alloc_size; 10557 struct net_device *p; 10558 10559 BUG_ON(strlen(name) >= sizeof(dev->name)); 10560 10561 if (txqs < 1) { 10562 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10563 return NULL; 10564 } 10565 10566 if (rxqs < 1) { 10567 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10568 return NULL; 10569 } 10570 10571 alloc_size = sizeof(struct net_device); 10572 if (sizeof_priv) { 10573 /* ensure 32-byte alignment of private area */ 10574 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10575 alloc_size += sizeof_priv; 10576 } 10577 /* ensure 32-byte alignment of whole construct */ 10578 alloc_size += NETDEV_ALIGN - 1; 10579 10580 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10581 if (!p) 10582 return NULL; 10583 10584 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10585 dev->padded = (char *)dev - (char *)p; 10586 10587 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10588 #ifdef CONFIG_PCPU_DEV_REFCNT 10589 dev->pcpu_refcnt = alloc_percpu(int); 10590 if (!dev->pcpu_refcnt) 10591 goto free_dev; 10592 __dev_hold(dev); 10593 #else 10594 refcount_set(&dev->dev_refcnt, 1); 10595 #endif 10596 10597 if (dev_addr_init(dev)) 10598 goto free_pcpu; 10599 10600 dev_mc_init(dev); 10601 dev_uc_init(dev); 10602 10603 dev_net_set(dev, &init_net); 10604 10605 dev->gso_max_size = GSO_MAX_SIZE; 10606 dev->gso_max_segs = GSO_MAX_SEGS; 10607 dev->gro_max_size = GRO_MAX_SIZE; 10608 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10609 dev->tso_max_segs = TSO_MAX_SEGS; 10610 dev->upper_level = 1; 10611 dev->lower_level = 1; 10612 #ifdef CONFIG_LOCKDEP 10613 dev->nested_level = 0; 10614 INIT_LIST_HEAD(&dev->unlink_list); 10615 #endif 10616 10617 INIT_LIST_HEAD(&dev->napi_list); 10618 INIT_LIST_HEAD(&dev->unreg_list); 10619 INIT_LIST_HEAD(&dev->close_list); 10620 INIT_LIST_HEAD(&dev->link_watch_list); 10621 INIT_LIST_HEAD(&dev->adj_list.upper); 10622 INIT_LIST_HEAD(&dev->adj_list.lower); 10623 INIT_LIST_HEAD(&dev->ptype_all); 10624 INIT_LIST_HEAD(&dev->ptype_specific); 10625 INIT_LIST_HEAD(&dev->net_notifier_list); 10626 #ifdef CONFIG_NET_SCHED 10627 hash_init(dev->qdisc_hash); 10628 #endif 10629 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10630 setup(dev); 10631 10632 if (!dev->tx_queue_len) { 10633 dev->priv_flags |= IFF_NO_QUEUE; 10634 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10635 } 10636 10637 dev->num_tx_queues = txqs; 10638 dev->real_num_tx_queues = txqs; 10639 if (netif_alloc_netdev_queues(dev)) 10640 goto free_all; 10641 10642 dev->num_rx_queues = rxqs; 10643 dev->real_num_rx_queues = rxqs; 10644 if (netif_alloc_rx_queues(dev)) 10645 goto free_all; 10646 10647 strcpy(dev->name, name); 10648 dev->name_assign_type = name_assign_type; 10649 dev->group = INIT_NETDEV_GROUP; 10650 if (!dev->ethtool_ops) 10651 dev->ethtool_ops = &default_ethtool_ops; 10652 10653 nf_hook_netdev_init(dev); 10654 10655 return dev; 10656 10657 free_all: 10658 free_netdev(dev); 10659 return NULL; 10660 10661 free_pcpu: 10662 #ifdef CONFIG_PCPU_DEV_REFCNT 10663 free_percpu(dev->pcpu_refcnt); 10664 free_dev: 10665 #endif 10666 netdev_freemem(dev); 10667 return NULL; 10668 } 10669 EXPORT_SYMBOL(alloc_netdev_mqs); 10670 10671 /** 10672 * free_netdev - free network device 10673 * @dev: device 10674 * 10675 * This function does the last stage of destroying an allocated device 10676 * interface. The reference to the device object is released. If this 10677 * is the last reference then it will be freed.Must be called in process 10678 * context. 10679 */ 10680 void free_netdev(struct net_device *dev) 10681 { 10682 struct napi_struct *p, *n; 10683 10684 might_sleep(); 10685 10686 /* When called immediately after register_netdevice() failed the unwind 10687 * handling may still be dismantling the device. Handle that case by 10688 * deferring the free. 10689 */ 10690 if (dev->reg_state == NETREG_UNREGISTERING) { 10691 ASSERT_RTNL(); 10692 dev->needs_free_netdev = true; 10693 return; 10694 } 10695 10696 netif_free_tx_queues(dev); 10697 netif_free_rx_queues(dev); 10698 10699 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10700 10701 /* Flush device addresses */ 10702 dev_addr_flush(dev); 10703 10704 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10705 netif_napi_del(p); 10706 10707 ref_tracker_dir_exit(&dev->refcnt_tracker); 10708 #ifdef CONFIG_PCPU_DEV_REFCNT 10709 free_percpu(dev->pcpu_refcnt); 10710 dev->pcpu_refcnt = NULL; 10711 #endif 10712 free_percpu(dev->core_stats); 10713 dev->core_stats = NULL; 10714 free_percpu(dev->xdp_bulkq); 10715 dev->xdp_bulkq = NULL; 10716 10717 /* Compatibility with error handling in drivers */ 10718 if (dev->reg_state == NETREG_UNINITIALIZED) { 10719 netdev_freemem(dev); 10720 return; 10721 } 10722 10723 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10724 dev->reg_state = NETREG_RELEASED; 10725 10726 /* will free via device release */ 10727 put_device(&dev->dev); 10728 } 10729 EXPORT_SYMBOL(free_netdev); 10730 10731 /** 10732 * synchronize_net - Synchronize with packet receive processing 10733 * 10734 * Wait for packets currently being received to be done. 10735 * Does not block later packets from starting. 10736 */ 10737 void synchronize_net(void) 10738 { 10739 might_sleep(); 10740 if (rtnl_is_locked()) 10741 synchronize_rcu_expedited(); 10742 else 10743 synchronize_rcu(); 10744 } 10745 EXPORT_SYMBOL(synchronize_net); 10746 10747 /** 10748 * unregister_netdevice_queue - remove device from the kernel 10749 * @dev: device 10750 * @head: list 10751 * 10752 * This function shuts down a device interface and removes it 10753 * from the kernel tables. 10754 * If head not NULL, device is queued to be unregistered later. 10755 * 10756 * Callers must hold the rtnl semaphore. You may want 10757 * unregister_netdev() instead of this. 10758 */ 10759 10760 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10761 { 10762 ASSERT_RTNL(); 10763 10764 if (head) { 10765 list_move_tail(&dev->unreg_list, head); 10766 } else { 10767 LIST_HEAD(single); 10768 10769 list_add(&dev->unreg_list, &single); 10770 unregister_netdevice_many(&single); 10771 } 10772 } 10773 EXPORT_SYMBOL(unregister_netdevice_queue); 10774 10775 /** 10776 * unregister_netdevice_many - unregister many devices 10777 * @head: list of devices 10778 * 10779 * Note: As most callers use a stack allocated list_head, 10780 * we force a list_del() to make sure stack wont be corrupted later. 10781 */ 10782 void unregister_netdevice_many(struct list_head *head) 10783 { 10784 struct net_device *dev, *tmp; 10785 LIST_HEAD(close_head); 10786 10787 BUG_ON(dev_boot_phase); 10788 ASSERT_RTNL(); 10789 10790 if (list_empty(head)) 10791 return; 10792 10793 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10794 /* Some devices call without registering 10795 * for initialization unwind. Remove those 10796 * devices and proceed with the remaining. 10797 */ 10798 if (dev->reg_state == NETREG_UNINITIALIZED) { 10799 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10800 dev->name, dev); 10801 10802 WARN_ON(1); 10803 list_del(&dev->unreg_list); 10804 continue; 10805 } 10806 dev->dismantle = true; 10807 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10808 } 10809 10810 /* If device is running, close it first. */ 10811 list_for_each_entry(dev, head, unreg_list) 10812 list_add_tail(&dev->close_list, &close_head); 10813 dev_close_many(&close_head, true); 10814 10815 list_for_each_entry(dev, head, unreg_list) { 10816 /* And unlink it from device chain. */ 10817 unlist_netdevice(dev); 10818 10819 dev->reg_state = NETREG_UNREGISTERING; 10820 } 10821 flush_all_backlogs(); 10822 10823 synchronize_net(); 10824 10825 list_for_each_entry(dev, head, unreg_list) { 10826 struct sk_buff *skb = NULL; 10827 10828 /* Shutdown queueing discipline. */ 10829 dev_shutdown(dev); 10830 10831 dev_xdp_uninstall(dev); 10832 10833 netdev_offload_xstats_disable_all(dev); 10834 10835 /* Notify protocols, that we are about to destroy 10836 * this device. They should clean all the things. 10837 */ 10838 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10839 10840 if (!dev->rtnl_link_ops || 10841 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10842 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10843 GFP_KERNEL, NULL, 0); 10844 10845 /* 10846 * Flush the unicast and multicast chains 10847 */ 10848 dev_uc_flush(dev); 10849 dev_mc_flush(dev); 10850 10851 netdev_name_node_alt_flush(dev); 10852 netdev_name_node_free(dev->name_node); 10853 10854 if (dev->netdev_ops->ndo_uninit) 10855 dev->netdev_ops->ndo_uninit(dev); 10856 10857 if (skb) 10858 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10859 10860 /* Notifier chain MUST detach us all upper devices. */ 10861 WARN_ON(netdev_has_any_upper_dev(dev)); 10862 WARN_ON(netdev_has_any_lower_dev(dev)); 10863 10864 /* Remove entries from kobject tree */ 10865 netdev_unregister_kobject(dev); 10866 #ifdef CONFIG_XPS 10867 /* Remove XPS queueing entries */ 10868 netif_reset_xps_queues_gt(dev, 0); 10869 #endif 10870 } 10871 10872 synchronize_net(); 10873 10874 list_for_each_entry(dev, head, unreg_list) { 10875 dev_put_track(dev, &dev->dev_registered_tracker); 10876 net_set_todo(dev); 10877 } 10878 10879 list_del(head); 10880 } 10881 EXPORT_SYMBOL(unregister_netdevice_many); 10882 10883 /** 10884 * unregister_netdev - remove device from the kernel 10885 * @dev: device 10886 * 10887 * This function shuts down a device interface and removes it 10888 * from the kernel tables. 10889 * 10890 * This is just a wrapper for unregister_netdevice that takes 10891 * the rtnl semaphore. In general you want to use this and not 10892 * unregister_netdevice. 10893 */ 10894 void unregister_netdev(struct net_device *dev) 10895 { 10896 rtnl_lock(); 10897 unregister_netdevice(dev); 10898 rtnl_unlock(); 10899 } 10900 EXPORT_SYMBOL(unregister_netdev); 10901 10902 /** 10903 * __dev_change_net_namespace - move device to different nethost namespace 10904 * @dev: device 10905 * @net: network namespace 10906 * @pat: If not NULL name pattern to try if the current device name 10907 * is already taken in the destination network namespace. 10908 * @new_ifindex: If not zero, specifies device index in the target 10909 * namespace. 10910 * 10911 * This function shuts down a device interface and moves it 10912 * to a new network namespace. On success 0 is returned, on 10913 * a failure a netagive errno code is returned. 10914 * 10915 * Callers must hold the rtnl semaphore. 10916 */ 10917 10918 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10919 const char *pat, int new_ifindex) 10920 { 10921 struct net *net_old = dev_net(dev); 10922 int err, new_nsid; 10923 10924 ASSERT_RTNL(); 10925 10926 /* Don't allow namespace local devices to be moved. */ 10927 err = -EINVAL; 10928 if (dev->features & NETIF_F_NETNS_LOCAL) 10929 goto out; 10930 10931 /* Ensure the device has been registrered */ 10932 if (dev->reg_state != NETREG_REGISTERED) 10933 goto out; 10934 10935 /* Get out if there is nothing todo */ 10936 err = 0; 10937 if (net_eq(net_old, net)) 10938 goto out; 10939 10940 /* Pick the destination device name, and ensure 10941 * we can use it in the destination network namespace. 10942 */ 10943 err = -EEXIST; 10944 if (netdev_name_in_use(net, dev->name)) { 10945 /* We get here if we can't use the current device name */ 10946 if (!pat) 10947 goto out; 10948 err = dev_get_valid_name(net, dev, pat); 10949 if (err < 0) 10950 goto out; 10951 } 10952 10953 /* Check that new_ifindex isn't used yet. */ 10954 err = -EBUSY; 10955 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10956 goto out; 10957 10958 /* 10959 * And now a mini version of register_netdevice unregister_netdevice. 10960 */ 10961 10962 /* If device is running close it first. */ 10963 dev_close(dev); 10964 10965 /* And unlink it from device chain */ 10966 unlist_netdevice(dev); 10967 10968 synchronize_net(); 10969 10970 /* Shutdown queueing discipline. */ 10971 dev_shutdown(dev); 10972 10973 /* Notify protocols, that we are about to destroy 10974 * this device. They should clean all the things. 10975 * 10976 * Note that dev->reg_state stays at NETREG_REGISTERED. 10977 * This is wanted because this way 8021q and macvlan know 10978 * the device is just moving and can keep their slaves up. 10979 */ 10980 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10981 rcu_barrier(); 10982 10983 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10984 /* If there is an ifindex conflict assign a new one */ 10985 if (!new_ifindex) { 10986 if (__dev_get_by_index(net, dev->ifindex)) 10987 new_ifindex = dev_new_index(net); 10988 else 10989 new_ifindex = dev->ifindex; 10990 } 10991 10992 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10993 new_ifindex); 10994 10995 /* 10996 * Flush the unicast and multicast chains 10997 */ 10998 dev_uc_flush(dev); 10999 dev_mc_flush(dev); 11000 11001 /* Send a netdev-removed uevent to the old namespace */ 11002 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11003 netdev_adjacent_del_links(dev); 11004 11005 /* Move per-net netdevice notifiers that are following the netdevice */ 11006 move_netdevice_notifiers_dev_net(dev, net); 11007 11008 /* Actually switch the network namespace */ 11009 dev_net_set(dev, net); 11010 dev->ifindex = new_ifindex; 11011 11012 /* Send a netdev-add uevent to the new namespace */ 11013 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11014 netdev_adjacent_add_links(dev); 11015 11016 /* Fixup kobjects */ 11017 err = device_rename(&dev->dev, dev->name); 11018 WARN_ON(err); 11019 11020 /* Adapt owner in case owning user namespace of target network 11021 * namespace is different from the original one. 11022 */ 11023 err = netdev_change_owner(dev, net_old, net); 11024 WARN_ON(err); 11025 11026 /* Add the device back in the hashes */ 11027 list_netdevice(dev); 11028 11029 /* Notify protocols, that a new device appeared. */ 11030 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11031 11032 /* 11033 * Prevent userspace races by waiting until the network 11034 * device is fully setup before sending notifications. 11035 */ 11036 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11037 11038 synchronize_net(); 11039 err = 0; 11040 out: 11041 return err; 11042 } 11043 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11044 11045 static int dev_cpu_dead(unsigned int oldcpu) 11046 { 11047 struct sk_buff **list_skb; 11048 struct sk_buff *skb; 11049 unsigned int cpu; 11050 struct softnet_data *sd, *oldsd, *remsd = NULL; 11051 11052 local_irq_disable(); 11053 cpu = smp_processor_id(); 11054 sd = &per_cpu(softnet_data, cpu); 11055 oldsd = &per_cpu(softnet_data, oldcpu); 11056 11057 /* Find end of our completion_queue. */ 11058 list_skb = &sd->completion_queue; 11059 while (*list_skb) 11060 list_skb = &(*list_skb)->next; 11061 /* Append completion queue from offline CPU. */ 11062 *list_skb = oldsd->completion_queue; 11063 oldsd->completion_queue = NULL; 11064 11065 /* Append output queue from offline CPU. */ 11066 if (oldsd->output_queue) { 11067 *sd->output_queue_tailp = oldsd->output_queue; 11068 sd->output_queue_tailp = oldsd->output_queue_tailp; 11069 oldsd->output_queue = NULL; 11070 oldsd->output_queue_tailp = &oldsd->output_queue; 11071 } 11072 /* Append NAPI poll list from offline CPU, with one exception : 11073 * process_backlog() must be called by cpu owning percpu backlog. 11074 * We properly handle process_queue & input_pkt_queue later. 11075 */ 11076 while (!list_empty(&oldsd->poll_list)) { 11077 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11078 struct napi_struct, 11079 poll_list); 11080 11081 list_del_init(&napi->poll_list); 11082 if (napi->poll == process_backlog) 11083 napi->state = 0; 11084 else 11085 ____napi_schedule(sd, napi); 11086 } 11087 11088 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11089 local_irq_enable(); 11090 11091 #ifdef CONFIG_RPS 11092 remsd = oldsd->rps_ipi_list; 11093 oldsd->rps_ipi_list = NULL; 11094 #endif 11095 /* send out pending IPI's on offline CPU */ 11096 net_rps_send_ipi(remsd); 11097 11098 /* Process offline CPU's input_pkt_queue */ 11099 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11100 netif_rx(skb); 11101 input_queue_head_incr(oldsd); 11102 } 11103 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11104 netif_rx(skb); 11105 input_queue_head_incr(oldsd); 11106 } 11107 11108 return 0; 11109 } 11110 11111 /** 11112 * netdev_increment_features - increment feature set by one 11113 * @all: current feature set 11114 * @one: new feature set 11115 * @mask: mask feature set 11116 * 11117 * Computes a new feature set after adding a device with feature set 11118 * @one to the master device with current feature set @all. Will not 11119 * enable anything that is off in @mask. Returns the new feature set. 11120 */ 11121 netdev_features_t netdev_increment_features(netdev_features_t all, 11122 netdev_features_t one, netdev_features_t mask) 11123 { 11124 if (mask & NETIF_F_HW_CSUM) 11125 mask |= NETIF_F_CSUM_MASK; 11126 mask |= NETIF_F_VLAN_CHALLENGED; 11127 11128 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11129 all &= one | ~NETIF_F_ALL_FOR_ALL; 11130 11131 /* If one device supports hw checksumming, set for all. */ 11132 if (all & NETIF_F_HW_CSUM) 11133 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11134 11135 return all; 11136 } 11137 EXPORT_SYMBOL(netdev_increment_features); 11138 11139 static struct hlist_head * __net_init netdev_create_hash(void) 11140 { 11141 int i; 11142 struct hlist_head *hash; 11143 11144 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11145 if (hash != NULL) 11146 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11147 INIT_HLIST_HEAD(&hash[i]); 11148 11149 return hash; 11150 } 11151 11152 /* Initialize per network namespace state */ 11153 static int __net_init netdev_init(struct net *net) 11154 { 11155 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11156 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11157 11158 INIT_LIST_HEAD(&net->dev_base_head); 11159 11160 net->dev_name_head = netdev_create_hash(); 11161 if (net->dev_name_head == NULL) 11162 goto err_name; 11163 11164 net->dev_index_head = netdev_create_hash(); 11165 if (net->dev_index_head == NULL) 11166 goto err_idx; 11167 11168 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11169 11170 return 0; 11171 11172 err_idx: 11173 kfree(net->dev_name_head); 11174 err_name: 11175 return -ENOMEM; 11176 } 11177 11178 /** 11179 * netdev_drivername - network driver for the device 11180 * @dev: network device 11181 * 11182 * Determine network driver for device. 11183 */ 11184 const char *netdev_drivername(const struct net_device *dev) 11185 { 11186 const struct device_driver *driver; 11187 const struct device *parent; 11188 const char *empty = ""; 11189 11190 parent = dev->dev.parent; 11191 if (!parent) 11192 return empty; 11193 11194 driver = parent->driver; 11195 if (driver && driver->name) 11196 return driver->name; 11197 return empty; 11198 } 11199 11200 static void __netdev_printk(const char *level, const struct net_device *dev, 11201 struct va_format *vaf) 11202 { 11203 if (dev && dev->dev.parent) { 11204 dev_printk_emit(level[1] - '0', 11205 dev->dev.parent, 11206 "%s %s %s%s: %pV", 11207 dev_driver_string(dev->dev.parent), 11208 dev_name(dev->dev.parent), 11209 netdev_name(dev), netdev_reg_state(dev), 11210 vaf); 11211 } else if (dev) { 11212 printk("%s%s%s: %pV", 11213 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11214 } else { 11215 printk("%s(NULL net_device): %pV", level, vaf); 11216 } 11217 } 11218 11219 void netdev_printk(const char *level, const struct net_device *dev, 11220 const char *format, ...) 11221 { 11222 struct va_format vaf; 11223 va_list args; 11224 11225 va_start(args, format); 11226 11227 vaf.fmt = format; 11228 vaf.va = &args; 11229 11230 __netdev_printk(level, dev, &vaf); 11231 11232 va_end(args); 11233 } 11234 EXPORT_SYMBOL(netdev_printk); 11235 11236 #define define_netdev_printk_level(func, level) \ 11237 void func(const struct net_device *dev, const char *fmt, ...) \ 11238 { \ 11239 struct va_format vaf; \ 11240 va_list args; \ 11241 \ 11242 va_start(args, fmt); \ 11243 \ 11244 vaf.fmt = fmt; \ 11245 vaf.va = &args; \ 11246 \ 11247 __netdev_printk(level, dev, &vaf); \ 11248 \ 11249 va_end(args); \ 11250 } \ 11251 EXPORT_SYMBOL(func); 11252 11253 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11254 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11255 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11256 define_netdev_printk_level(netdev_err, KERN_ERR); 11257 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11258 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11259 define_netdev_printk_level(netdev_info, KERN_INFO); 11260 11261 static void __net_exit netdev_exit(struct net *net) 11262 { 11263 kfree(net->dev_name_head); 11264 kfree(net->dev_index_head); 11265 if (net != &init_net) 11266 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11267 } 11268 11269 static struct pernet_operations __net_initdata netdev_net_ops = { 11270 .init = netdev_init, 11271 .exit = netdev_exit, 11272 }; 11273 11274 static void __net_exit default_device_exit_net(struct net *net) 11275 { 11276 struct net_device *dev, *aux; 11277 /* 11278 * Push all migratable network devices back to the 11279 * initial network namespace 11280 */ 11281 ASSERT_RTNL(); 11282 for_each_netdev_safe(net, dev, aux) { 11283 int err; 11284 char fb_name[IFNAMSIZ]; 11285 11286 /* Ignore unmoveable devices (i.e. loopback) */ 11287 if (dev->features & NETIF_F_NETNS_LOCAL) 11288 continue; 11289 11290 /* Leave virtual devices for the generic cleanup */ 11291 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11292 continue; 11293 11294 /* Push remaining network devices to init_net */ 11295 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11296 if (netdev_name_in_use(&init_net, fb_name)) 11297 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11298 err = dev_change_net_namespace(dev, &init_net, fb_name); 11299 if (err) { 11300 pr_emerg("%s: failed to move %s to init_net: %d\n", 11301 __func__, dev->name, err); 11302 BUG(); 11303 } 11304 } 11305 } 11306 11307 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11308 { 11309 /* At exit all network devices most be removed from a network 11310 * namespace. Do this in the reverse order of registration. 11311 * Do this across as many network namespaces as possible to 11312 * improve batching efficiency. 11313 */ 11314 struct net_device *dev; 11315 struct net *net; 11316 LIST_HEAD(dev_kill_list); 11317 11318 rtnl_lock(); 11319 list_for_each_entry(net, net_list, exit_list) { 11320 default_device_exit_net(net); 11321 cond_resched(); 11322 } 11323 11324 list_for_each_entry(net, net_list, exit_list) { 11325 for_each_netdev_reverse(net, dev) { 11326 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11327 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11328 else 11329 unregister_netdevice_queue(dev, &dev_kill_list); 11330 } 11331 } 11332 unregister_netdevice_many(&dev_kill_list); 11333 rtnl_unlock(); 11334 } 11335 11336 static struct pernet_operations __net_initdata default_device_ops = { 11337 .exit_batch = default_device_exit_batch, 11338 }; 11339 11340 /* 11341 * Initialize the DEV module. At boot time this walks the device list and 11342 * unhooks any devices that fail to initialise (normally hardware not 11343 * present) and leaves us with a valid list of present and active devices. 11344 * 11345 */ 11346 11347 /* 11348 * This is called single threaded during boot, so no need 11349 * to take the rtnl semaphore. 11350 */ 11351 static int __init net_dev_init(void) 11352 { 11353 int i, rc = -ENOMEM; 11354 11355 BUG_ON(!dev_boot_phase); 11356 11357 if (dev_proc_init()) 11358 goto out; 11359 11360 if (netdev_kobject_init()) 11361 goto out; 11362 11363 INIT_LIST_HEAD(&ptype_all); 11364 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11365 INIT_LIST_HEAD(&ptype_base[i]); 11366 11367 if (register_pernet_subsys(&netdev_net_ops)) 11368 goto out; 11369 11370 /* 11371 * Initialise the packet receive queues. 11372 */ 11373 11374 for_each_possible_cpu(i) { 11375 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11376 struct softnet_data *sd = &per_cpu(softnet_data, i); 11377 11378 INIT_WORK(flush, flush_backlog); 11379 11380 skb_queue_head_init(&sd->input_pkt_queue); 11381 skb_queue_head_init(&sd->process_queue); 11382 #ifdef CONFIG_XFRM_OFFLOAD 11383 skb_queue_head_init(&sd->xfrm_backlog); 11384 #endif 11385 INIT_LIST_HEAD(&sd->poll_list); 11386 sd->output_queue_tailp = &sd->output_queue; 11387 #ifdef CONFIG_RPS 11388 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11389 sd->cpu = i; 11390 #endif 11391 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, NULL); 11392 spin_lock_init(&sd->defer_lock); 11393 11394 init_gro_hash(&sd->backlog); 11395 sd->backlog.poll = process_backlog; 11396 sd->backlog.weight = weight_p; 11397 } 11398 11399 dev_boot_phase = 0; 11400 11401 /* The loopback device is special if any other network devices 11402 * is present in a network namespace the loopback device must 11403 * be present. Since we now dynamically allocate and free the 11404 * loopback device ensure this invariant is maintained by 11405 * keeping the loopback device as the first device on the 11406 * list of network devices. Ensuring the loopback devices 11407 * is the first device that appears and the last network device 11408 * that disappears. 11409 */ 11410 if (register_pernet_device(&loopback_net_ops)) 11411 goto out; 11412 11413 if (register_pernet_device(&default_device_ops)) 11414 goto out; 11415 11416 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11417 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11418 11419 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11420 NULL, dev_cpu_dead); 11421 WARN_ON(rc < 0); 11422 rc = 0; 11423 out: 11424 return rc; 11425 } 11426 11427 subsys_initcall(net_dev_init); 11428