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