1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "dev.h" 155 #include "net-sysfs.h" 156 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_info(unsigned long val, 164 struct netdev_notifier_info *info); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 netdev_name_node_del(name_node); 349 kfree(name_node->name); 350 netdev_name_node_free(name_node); 351 } 352 353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 354 { 355 struct netdev_name_node *name_node; 356 struct net *net = dev_net(dev); 357 358 name_node = netdev_name_node_lookup(net, name); 359 if (!name_node) 360 return -ENOENT; 361 /* lookup might have found our primary name or a name belonging 362 * to another device. 363 */ 364 if (name_node == dev->name_node || name_node->dev != dev) 365 return -EINVAL; 366 367 __netdev_name_node_alt_destroy(name_node); 368 369 return 0; 370 } 371 372 static void netdev_name_node_alt_flush(struct net_device *dev) 373 { 374 struct netdev_name_node *name_node, *tmp; 375 376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 377 __netdev_name_node_alt_destroy(name_node); 378 } 379 380 /* Device list insertion */ 381 static void list_netdevice(struct net_device *dev) 382 { 383 struct net *net = dev_net(dev); 384 385 ASSERT_RTNL(); 386 387 write_lock(&dev_base_lock); 388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 389 netdev_name_node_add(net, dev->name_node); 390 hlist_add_head_rcu(&dev->index_hlist, 391 dev_index_hash(net, dev->ifindex)); 392 write_unlock(&dev_base_lock); 393 394 dev_base_seq_inc(net); 395 } 396 397 /* Device list removal 398 * caller must respect a RCU grace period before freeing/reusing dev 399 */ 400 static void unlist_netdevice(struct net_device *dev) 401 { 402 ASSERT_RTNL(); 403 404 /* Unlink dev from the device chain */ 405 write_lock(&dev_base_lock); 406 list_del_rcu(&dev->dev_list); 407 netdev_name_node_del(dev->name_node); 408 hlist_del_rcu(&dev->index_hlist); 409 write_unlock(&dev_base_lock); 410 411 dev_base_seq_inc(dev_net(dev)); 412 } 413 414 /* 415 * Our notifier list 416 */ 417 418 static RAW_NOTIFIER_HEAD(netdev_chain); 419 420 /* 421 * Device drivers call our routines to queue packets here. We empty the 422 * queue in the local softnet handler. 423 */ 424 425 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 426 EXPORT_PER_CPU_SYMBOL(softnet_data); 427 428 #ifdef CONFIG_LOCKDEP 429 /* 430 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 431 * according to dev->type 432 */ 433 static const unsigned short netdev_lock_type[] = { 434 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 435 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 436 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 437 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 438 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 439 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 440 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 441 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 442 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 443 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 444 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 445 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 446 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 447 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 448 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 449 450 static const char *const netdev_lock_name[] = { 451 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 452 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 453 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 454 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 455 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 456 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 457 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 458 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 459 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 460 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 461 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 462 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 463 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 464 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 465 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 466 467 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 468 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 469 470 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 471 { 472 int i; 473 474 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 475 if (netdev_lock_type[i] == dev_type) 476 return i; 477 /* the last key is used by default */ 478 return ARRAY_SIZE(netdev_lock_type) - 1; 479 } 480 481 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 482 unsigned short dev_type) 483 { 484 int i; 485 486 i = netdev_lock_pos(dev_type); 487 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 488 netdev_lock_name[i]); 489 } 490 491 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 492 { 493 int i; 494 495 i = netdev_lock_pos(dev->type); 496 lockdep_set_class_and_name(&dev->addr_list_lock, 497 &netdev_addr_lock_key[i], 498 netdev_lock_name[i]); 499 } 500 #else 501 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 502 unsigned short dev_type) 503 { 504 } 505 506 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 507 { 508 } 509 #endif 510 511 /******************************************************************************* 512 * 513 * Protocol management and registration routines 514 * 515 *******************************************************************************/ 516 517 518 /* 519 * Add a protocol ID to the list. Now that the input handler is 520 * smarter we can dispense with all the messy stuff that used to be 521 * here. 522 * 523 * BEWARE!!! Protocol handlers, mangling input packets, 524 * MUST BE last in hash buckets and checking protocol handlers 525 * MUST start from promiscuous ptype_all chain in net_bh. 526 * It is true now, do not change it. 527 * Explanation follows: if protocol handler, mangling packet, will 528 * be the first on list, it is not able to sense, that packet 529 * is cloned and should be copied-on-write, so that it will 530 * change it and subsequent readers will get broken packet. 531 * --ANK (980803) 532 */ 533 534 static inline struct list_head *ptype_head(const struct packet_type *pt) 535 { 536 if (pt->type == htons(ETH_P_ALL)) 537 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 538 else 539 return pt->dev ? &pt->dev->ptype_specific : 540 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 541 } 542 543 /** 544 * dev_add_pack - add packet handler 545 * @pt: packet type declaration 546 * 547 * Add a protocol handler to the networking stack. The passed &packet_type 548 * is linked into kernel lists and may not be freed until it has been 549 * removed from the kernel lists. 550 * 551 * This call does not sleep therefore it can not 552 * guarantee all CPU's that are in middle of receiving packets 553 * will see the new packet type (until the next received packet). 554 */ 555 556 void dev_add_pack(struct packet_type *pt) 557 { 558 struct list_head *head = ptype_head(pt); 559 560 spin_lock(&ptype_lock); 561 list_add_rcu(&pt->list, head); 562 spin_unlock(&ptype_lock); 563 } 564 EXPORT_SYMBOL(dev_add_pack); 565 566 /** 567 * __dev_remove_pack - remove packet handler 568 * @pt: packet type declaration 569 * 570 * Remove a protocol handler that was previously added to the kernel 571 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 572 * from the kernel lists and can be freed or reused once this function 573 * returns. 574 * 575 * The packet type might still be in use by receivers 576 * and must not be freed until after all the CPU's have gone 577 * through a quiescent state. 578 */ 579 void __dev_remove_pack(struct packet_type *pt) 580 { 581 struct list_head *head = ptype_head(pt); 582 struct packet_type *pt1; 583 584 spin_lock(&ptype_lock); 585 586 list_for_each_entry(pt1, head, list) { 587 if (pt == pt1) { 588 list_del_rcu(&pt->list); 589 goto out; 590 } 591 } 592 593 pr_warn("dev_remove_pack: %p not found\n", pt); 594 out: 595 spin_unlock(&ptype_lock); 596 } 597 EXPORT_SYMBOL(__dev_remove_pack); 598 599 /** 600 * dev_remove_pack - remove packet handler 601 * @pt: packet type declaration 602 * 603 * Remove a protocol handler that was previously added to the kernel 604 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 605 * from the kernel lists and can be freed or reused once this function 606 * returns. 607 * 608 * This call sleeps to guarantee that no CPU is looking at the packet 609 * type after return. 610 */ 611 void dev_remove_pack(struct packet_type *pt) 612 { 613 __dev_remove_pack(pt); 614 615 synchronize_net(); 616 } 617 EXPORT_SYMBOL(dev_remove_pack); 618 619 620 /******************************************************************************* 621 * 622 * Device Interface Subroutines 623 * 624 *******************************************************************************/ 625 626 /** 627 * dev_get_iflink - get 'iflink' value of a interface 628 * @dev: targeted interface 629 * 630 * Indicates the ifindex the interface is linked to. 631 * Physical interfaces have the same 'ifindex' and 'iflink' values. 632 */ 633 634 int dev_get_iflink(const struct net_device *dev) 635 { 636 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 637 return dev->netdev_ops->ndo_get_iflink(dev); 638 639 return dev->ifindex; 640 } 641 EXPORT_SYMBOL(dev_get_iflink); 642 643 /** 644 * dev_fill_metadata_dst - Retrieve tunnel egress information. 645 * @dev: targeted interface 646 * @skb: The packet. 647 * 648 * For better visibility of tunnel traffic OVS needs to retrieve 649 * egress tunnel information for a packet. Following API allows 650 * user to get this info. 651 */ 652 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 653 { 654 struct ip_tunnel_info *info; 655 656 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 657 return -EINVAL; 658 659 info = skb_tunnel_info_unclone(skb); 660 if (!info) 661 return -ENOMEM; 662 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 663 return -EINVAL; 664 665 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 666 } 667 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 668 669 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 670 { 671 int k = stack->num_paths++; 672 673 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 674 return NULL; 675 676 return &stack->path[k]; 677 } 678 679 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 680 struct net_device_path_stack *stack) 681 { 682 const struct net_device *last_dev; 683 struct net_device_path_ctx ctx = { 684 .dev = dev, 685 .daddr = daddr, 686 }; 687 struct net_device_path *path; 688 int ret = 0; 689 690 stack->num_paths = 0; 691 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 692 last_dev = ctx.dev; 693 path = dev_fwd_path(stack); 694 if (!path) 695 return -1; 696 697 memset(path, 0, sizeof(struct net_device_path)); 698 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 699 if (ret < 0) 700 return -1; 701 702 if (WARN_ON_ONCE(last_dev == ctx.dev)) 703 return -1; 704 } 705 706 if (!ctx.dev) 707 return ret; 708 709 path = dev_fwd_path(stack); 710 if (!path) 711 return -1; 712 path->type = DEV_PATH_ETHERNET; 713 path->dev = ctx.dev; 714 715 return ret; 716 } 717 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 718 719 /** 720 * __dev_get_by_name - find a device by its name 721 * @net: the applicable net namespace 722 * @name: name to find 723 * 724 * Find an interface by name. Must be called under RTNL semaphore 725 * or @dev_base_lock. If the name is found a pointer to the device 726 * is returned. If the name is not found then %NULL is returned. The 727 * reference counters are not incremented so the caller must be 728 * careful with locks. 729 */ 730 731 struct net_device *__dev_get_by_name(struct net *net, const char *name) 732 { 733 struct netdev_name_node *node_name; 734 735 node_name = netdev_name_node_lookup(net, name); 736 return node_name ? node_name->dev : NULL; 737 } 738 EXPORT_SYMBOL(__dev_get_by_name); 739 740 /** 741 * dev_get_by_name_rcu - find a device by its name 742 * @net: the applicable net namespace 743 * @name: name to find 744 * 745 * Find an interface by name. 746 * If the name is found a pointer to the device is returned. 747 * If the name is not found then %NULL is returned. 748 * The reference counters are not incremented so the caller must be 749 * careful with locks. The caller must hold RCU lock. 750 */ 751 752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 753 { 754 struct netdev_name_node *node_name; 755 756 node_name = netdev_name_node_lookup_rcu(net, name); 757 return node_name ? node_name->dev : NULL; 758 } 759 EXPORT_SYMBOL(dev_get_by_name_rcu); 760 761 /** 762 * dev_get_by_name - find a device by its name 763 * @net: the applicable net namespace 764 * @name: name to find 765 * 766 * Find an interface by name. This can be called from any 767 * context and does its own locking. The returned handle has 768 * the usage count incremented and the caller must use dev_put() to 769 * release it when it is no longer needed. %NULL is returned if no 770 * matching device is found. 771 */ 772 773 struct net_device *dev_get_by_name(struct net *net, const char *name) 774 { 775 struct net_device *dev; 776 777 rcu_read_lock(); 778 dev = dev_get_by_name_rcu(net, name); 779 dev_hold(dev); 780 rcu_read_unlock(); 781 return dev; 782 } 783 EXPORT_SYMBOL(dev_get_by_name); 784 785 /** 786 * __dev_get_by_index - find a device by its ifindex 787 * @net: the applicable net namespace 788 * @ifindex: index of device 789 * 790 * Search for an interface by index. Returns %NULL if the device 791 * is not found or a pointer to the device. The device has not 792 * had its reference counter increased so the caller must be careful 793 * about locking. The caller must hold either the RTNL semaphore 794 * or @dev_base_lock. 795 */ 796 797 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 798 { 799 struct net_device *dev; 800 struct hlist_head *head = dev_index_hash(net, ifindex); 801 802 hlist_for_each_entry(dev, head, index_hlist) 803 if (dev->ifindex == ifindex) 804 return dev; 805 806 return NULL; 807 } 808 EXPORT_SYMBOL(__dev_get_by_index); 809 810 /** 811 * dev_get_by_index_rcu - find a device by its ifindex 812 * @net: the applicable net namespace 813 * @ifindex: index of device 814 * 815 * Search for an interface by index. Returns %NULL if the device 816 * is not found or a pointer to the device. The device has not 817 * had its reference counter increased so the caller must be careful 818 * about locking. The caller must hold RCU lock. 819 */ 820 821 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 822 { 823 struct net_device *dev; 824 struct hlist_head *head = dev_index_hash(net, ifindex); 825 826 hlist_for_each_entry_rcu(dev, head, index_hlist) 827 if (dev->ifindex == ifindex) 828 return dev; 829 830 return NULL; 831 } 832 EXPORT_SYMBOL(dev_get_by_index_rcu); 833 834 835 /** 836 * dev_get_by_index - find a device by its ifindex 837 * @net: the applicable net namespace 838 * @ifindex: index of device 839 * 840 * Search for an interface by index. Returns NULL if the device 841 * is not found or a pointer to the device. The device returned has 842 * had a reference added and the pointer is safe until the user calls 843 * dev_put to indicate they have finished with it. 844 */ 845 846 struct net_device *dev_get_by_index(struct net *net, int ifindex) 847 { 848 struct net_device *dev; 849 850 rcu_read_lock(); 851 dev = dev_get_by_index_rcu(net, ifindex); 852 dev_hold(dev); 853 rcu_read_unlock(); 854 return dev; 855 } 856 EXPORT_SYMBOL(dev_get_by_index); 857 858 /** 859 * dev_get_by_napi_id - find a device by napi_id 860 * @napi_id: ID of the NAPI struct 861 * 862 * Search for an interface by NAPI ID. Returns %NULL if the device 863 * is not found or a pointer to the device. The device has not had 864 * its reference counter increased so the caller must be careful 865 * about locking. The caller must hold RCU lock. 866 */ 867 868 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 869 { 870 struct napi_struct *napi; 871 872 WARN_ON_ONCE(!rcu_read_lock_held()); 873 874 if (napi_id < MIN_NAPI_ID) 875 return NULL; 876 877 napi = napi_by_id(napi_id); 878 879 return napi ? napi->dev : NULL; 880 } 881 EXPORT_SYMBOL(dev_get_by_napi_id); 882 883 /** 884 * netdev_get_name - get a netdevice name, knowing its ifindex. 885 * @net: network namespace 886 * @name: a pointer to the buffer where the name will be stored. 887 * @ifindex: the ifindex of the interface to get the name from. 888 */ 889 int netdev_get_name(struct net *net, char *name, int ifindex) 890 { 891 struct net_device *dev; 892 int ret; 893 894 down_read(&devnet_rename_sem); 895 rcu_read_lock(); 896 897 dev = dev_get_by_index_rcu(net, ifindex); 898 if (!dev) { 899 ret = -ENODEV; 900 goto out; 901 } 902 903 strcpy(name, dev->name); 904 905 ret = 0; 906 out: 907 rcu_read_unlock(); 908 up_read(&devnet_rename_sem); 909 return ret; 910 } 911 912 /** 913 * dev_getbyhwaddr_rcu - find a device by its hardware address 914 * @net: the applicable net namespace 915 * @type: media type of device 916 * @ha: hardware address 917 * 918 * Search for an interface by MAC address. Returns NULL if the device 919 * is not found or a pointer to the device. 920 * The caller must hold RCU or RTNL. 921 * The returned device has not had its ref count increased 922 * and the caller must therefore be careful about locking 923 * 924 */ 925 926 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 927 const char *ha) 928 { 929 struct net_device *dev; 930 931 for_each_netdev_rcu(net, dev) 932 if (dev->type == type && 933 !memcmp(dev->dev_addr, ha, dev->addr_len)) 934 return dev; 935 936 return NULL; 937 } 938 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 939 940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 941 { 942 struct net_device *dev, *ret = NULL; 943 944 rcu_read_lock(); 945 for_each_netdev_rcu(net, dev) 946 if (dev->type == type) { 947 dev_hold(dev); 948 ret = dev; 949 break; 950 } 951 rcu_read_unlock(); 952 return ret; 953 } 954 EXPORT_SYMBOL(dev_getfirstbyhwtype); 955 956 /** 957 * __dev_get_by_flags - find any device with given flags 958 * @net: the applicable net namespace 959 * @if_flags: IFF_* values 960 * @mask: bitmask of bits in if_flags to check 961 * 962 * Search for any interface with the given flags. Returns NULL if a device 963 * is not found or a pointer to the device. Must be called inside 964 * rtnl_lock(), and result refcount is unchanged. 965 */ 966 967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 968 unsigned short mask) 969 { 970 struct net_device *dev, *ret; 971 972 ASSERT_RTNL(); 973 974 ret = NULL; 975 for_each_netdev(net, dev) { 976 if (((dev->flags ^ if_flags) & mask) == 0) { 977 ret = dev; 978 break; 979 } 980 } 981 return ret; 982 } 983 EXPORT_SYMBOL(__dev_get_by_flags); 984 985 /** 986 * dev_valid_name - check if name is okay for network device 987 * @name: name string 988 * 989 * Network device names need to be valid file names to 990 * allow sysfs to work. We also disallow any kind of 991 * whitespace. 992 */ 993 bool dev_valid_name(const char *name) 994 { 995 if (*name == '\0') 996 return false; 997 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 998 return false; 999 if (!strcmp(name, ".") || !strcmp(name, "..")) 1000 return false; 1001 1002 while (*name) { 1003 if (*name == '/' || *name == ':' || isspace(*name)) 1004 return false; 1005 name++; 1006 } 1007 return true; 1008 } 1009 EXPORT_SYMBOL(dev_valid_name); 1010 1011 /** 1012 * __dev_alloc_name - allocate a name for a device 1013 * @net: network namespace to allocate the device name in 1014 * @name: name format string 1015 * @buf: scratch buffer and result name string 1016 * 1017 * Passed a format string - eg "lt%d" it will try and find a suitable 1018 * id. It scans list of devices to build up a free map, then chooses 1019 * the first empty slot. The caller must hold the dev_base or rtnl lock 1020 * while allocating the name and adding the device in order to avoid 1021 * duplicates. 1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1023 * Returns the number of the unit assigned or a negative errno code. 1024 */ 1025 1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1027 { 1028 int i = 0; 1029 const char *p; 1030 const int max_netdevices = 8*PAGE_SIZE; 1031 unsigned long *inuse; 1032 struct net_device *d; 1033 1034 if (!dev_valid_name(name)) 1035 return -EINVAL; 1036 1037 p = strchr(name, '%'); 1038 if (p) { 1039 /* 1040 * Verify the string as this thing may have come from 1041 * the user. There must be either one "%d" and no other "%" 1042 * characters. 1043 */ 1044 if (p[1] != 'd' || strchr(p + 2, '%')) 1045 return -EINVAL; 1046 1047 /* Use one page as a bit array of possible slots */ 1048 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1049 if (!inuse) 1050 return -ENOMEM; 1051 1052 for_each_netdev(net, d) { 1053 struct netdev_name_node *name_node; 1054 list_for_each_entry(name_node, &d->name_node->list, list) { 1055 if (!sscanf(name_node->name, name, &i)) 1056 continue; 1057 if (i < 0 || i >= max_netdevices) 1058 continue; 1059 1060 /* avoid cases where sscanf is not exact inverse of printf */ 1061 snprintf(buf, IFNAMSIZ, name, i); 1062 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1063 __set_bit(i, inuse); 1064 } 1065 if (!sscanf(d->name, name, &i)) 1066 continue; 1067 if (i < 0 || i >= max_netdevices) 1068 continue; 1069 1070 /* avoid cases where sscanf is not exact inverse of printf */ 1071 snprintf(buf, IFNAMSIZ, name, i); 1072 if (!strncmp(buf, d->name, IFNAMSIZ)) 1073 __set_bit(i, inuse); 1074 } 1075 1076 i = find_first_zero_bit(inuse, max_netdevices); 1077 free_page((unsigned long) inuse); 1078 } 1079 1080 snprintf(buf, IFNAMSIZ, name, i); 1081 if (!netdev_name_in_use(net, buf)) 1082 return i; 1083 1084 /* It is possible to run out of possible slots 1085 * when the name is long and there isn't enough space left 1086 * for the digits, or if all bits are used. 1087 */ 1088 return -ENFILE; 1089 } 1090 1091 static int dev_alloc_name_ns(struct net *net, 1092 struct net_device *dev, 1093 const char *name) 1094 { 1095 char buf[IFNAMSIZ]; 1096 int ret; 1097 1098 BUG_ON(!net); 1099 ret = __dev_alloc_name(net, name, buf); 1100 if (ret >= 0) 1101 strlcpy(dev->name, buf, IFNAMSIZ); 1102 return ret; 1103 } 1104 1105 /** 1106 * dev_alloc_name - allocate a name for a device 1107 * @dev: device 1108 * @name: name format string 1109 * 1110 * Passed a format string - eg "lt%d" it will try and find a suitable 1111 * id. It scans list of devices to build up a free map, then chooses 1112 * the first empty slot. The caller must hold the dev_base or rtnl lock 1113 * while allocating the name and adding the device in order to avoid 1114 * duplicates. 1115 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1116 * Returns the number of the unit assigned or a negative errno code. 1117 */ 1118 1119 int dev_alloc_name(struct net_device *dev, const char *name) 1120 { 1121 return dev_alloc_name_ns(dev_net(dev), dev, name); 1122 } 1123 EXPORT_SYMBOL(dev_alloc_name); 1124 1125 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1126 const char *name) 1127 { 1128 BUG_ON(!net); 1129 1130 if (!dev_valid_name(name)) 1131 return -EINVAL; 1132 1133 if (strchr(name, '%')) 1134 return dev_alloc_name_ns(net, dev, name); 1135 else if (netdev_name_in_use(net, name)) 1136 return -EEXIST; 1137 else if (dev->name != name) 1138 strlcpy(dev->name, name, IFNAMSIZ); 1139 1140 return 0; 1141 } 1142 1143 /** 1144 * dev_change_name - change name of a device 1145 * @dev: device 1146 * @newname: name (or format string) must be at least IFNAMSIZ 1147 * 1148 * Change name of a device, can pass format strings "eth%d". 1149 * for wildcarding. 1150 */ 1151 int dev_change_name(struct net_device *dev, const char *newname) 1152 { 1153 unsigned char old_assign_type; 1154 char oldname[IFNAMSIZ]; 1155 int err = 0; 1156 int ret; 1157 struct net *net; 1158 1159 ASSERT_RTNL(); 1160 BUG_ON(!dev_net(dev)); 1161 1162 net = dev_net(dev); 1163 1164 /* Some auto-enslaved devices e.g. failover slaves are 1165 * special, as userspace might rename the device after 1166 * the interface had been brought up and running since 1167 * the point kernel initiated auto-enslavement. Allow 1168 * live name change even when these slave devices are 1169 * up and running. 1170 * 1171 * Typically, users of these auto-enslaving devices 1172 * don't actually care about slave name change, as 1173 * they are supposed to operate on master interface 1174 * directly. 1175 */ 1176 if (dev->flags & IFF_UP && 1177 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1178 return -EBUSY; 1179 1180 down_write(&devnet_rename_sem); 1181 1182 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1183 up_write(&devnet_rename_sem); 1184 return 0; 1185 } 1186 1187 memcpy(oldname, dev->name, IFNAMSIZ); 1188 1189 err = dev_get_valid_name(net, dev, newname); 1190 if (err < 0) { 1191 up_write(&devnet_rename_sem); 1192 return err; 1193 } 1194 1195 if (oldname[0] && !strchr(oldname, '%')) 1196 netdev_info(dev, "renamed from %s\n", oldname); 1197 1198 old_assign_type = dev->name_assign_type; 1199 dev->name_assign_type = NET_NAME_RENAMED; 1200 1201 rollback: 1202 ret = device_rename(&dev->dev, dev->name); 1203 if (ret) { 1204 memcpy(dev->name, oldname, IFNAMSIZ); 1205 dev->name_assign_type = old_assign_type; 1206 up_write(&devnet_rename_sem); 1207 return ret; 1208 } 1209 1210 up_write(&devnet_rename_sem); 1211 1212 netdev_adjacent_rename_links(dev, oldname); 1213 1214 write_lock(&dev_base_lock); 1215 netdev_name_node_del(dev->name_node); 1216 write_unlock(&dev_base_lock); 1217 1218 synchronize_rcu(); 1219 1220 write_lock(&dev_base_lock); 1221 netdev_name_node_add(net, dev->name_node); 1222 write_unlock(&dev_base_lock); 1223 1224 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1225 ret = notifier_to_errno(ret); 1226 1227 if (ret) { 1228 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1229 if (err >= 0) { 1230 err = ret; 1231 down_write(&devnet_rename_sem); 1232 memcpy(dev->name, oldname, IFNAMSIZ); 1233 memcpy(oldname, newname, IFNAMSIZ); 1234 dev->name_assign_type = old_assign_type; 1235 old_assign_type = NET_NAME_RENAMED; 1236 goto rollback; 1237 } else { 1238 netdev_err(dev, "name change rollback failed: %d\n", 1239 ret); 1240 } 1241 } 1242 1243 return err; 1244 } 1245 1246 /** 1247 * dev_set_alias - change ifalias of a device 1248 * @dev: device 1249 * @alias: name up to IFALIASZ 1250 * @len: limit of bytes to copy from info 1251 * 1252 * Set ifalias for a device, 1253 */ 1254 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1255 { 1256 struct dev_ifalias *new_alias = NULL; 1257 1258 if (len >= IFALIASZ) 1259 return -EINVAL; 1260 1261 if (len) { 1262 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1263 if (!new_alias) 1264 return -ENOMEM; 1265 1266 memcpy(new_alias->ifalias, alias, len); 1267 new_alias->ifalias[len] = 0; 1268 } 1269 1270 mutex_lock(&ifalias_mutex); 1271 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1272 mutex_is_locked(&ifalias_mutex)); 1273 mutex_unlock(&ifalias_mutex); 1274 1275 if (new_alias) 1276 kfree_rcu(new_alias, rcuhead); 1277 1278 return len; 1279 } 1280 EXPORT_SYMBOL(dev_set_alias); 1281 1282 /** 1283 * dev_get_alias - get ifalias of a device 1284 * @dev: device 1285 * @name: buffer to store name of ifalias 1286 * @len: size of buffer 1287 * 1288 * get ifalias for a device. Caller must make sure dev cannot go 1289 * away, e.g. rcu read lock or own a reference count to device. 1290 */ 1291 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1292 { 1293 const struct dev_ifalias *alias; 1294 int ret = 0; 1295 1296 rcu_read_lock(); 1297 alias = rcu_dereference(dev->ifalias); 1298 if (alias) 1299 ret = snprintf(name, len, "%s", alias->ifalias); 1300 rcu_read_unlock(); 1301 1302 return ret; 1303 } 1304 1305 /** 1306 * netdev_features_change - device changes features 1307 * @dev: device to cause notification 1308 * 1309 * Called to indicate a device has changed features. 1310 */ 1311 void netdev_features_change(struct net_device *dev) 1312 { 1313 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1314 } 1315 EXPORT_SYMBOL(netdev_features_change); 1316 1317 /** 1318 * netdev_state_change - device changes state 1319 * @dev: device to cause notification 1320 * 1321 * Called to indicate a device has changed state. This function calls 1322 * the notifier chains for netdev_chain and sends a NEWLINK message 1323 * to the routing socket. 1324 */ 1325 void netdev_state_change(struct net_device *dev) 1326 { 1327 if (dev->flags & IFF_UP) { 1328 struct netdev_notifier_change_info change_info = { 1329 .info.dev = dev, 1330 }; 1331 1332 call_netdevice_notifiers_info(NETDEV_CHANGE, 1333 &change_info.info); 1334 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1335 } 1336 } 1337 EXPORT_SYMBOL(netdev_state_change); 1338 1339 /** 1340 * __netdev_notify_peers - notify network peers about existence of @dev, 1341 * to be called when rtnl lock is already held. 1342 * @dev: network device 1343 * 1344 * Generate traffic such that interested network peers are aware of 1345 * @dev, such as by generating a gratuitous ARP. This may be used when 1346 * a device wants to inform the rest of the network about some sort of 1347 * reconfiguration such as a failover event or virtual machine 1348 * migration. 1349 */ 1350 void __netdev_notify_peers(struct net_device *dev) 1351 { 1352 ASSERT_RTNL(); 1353 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1354 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1355 } 1356 EXPORT_SYMBOL(__netdev_notify_peers); 1357 1358 /** 1359 * netdev_notify_peers - notify network peers about existence of @dev 1360 * @dev: network device 1361 * 1362 * Generate traffic such that interested network peers are aware of 1363 * @dev, such as by generating a gratuitous ARP. This may be used when 1364 * a device wants to inform the rest of the network about some sort of 1365 * reconfiguration such as a failover event or virtual machine 1366 * migration. 1367 */ 1368 void netdev_notify_peers(struct net_device *dev) 1369 { 1370 rtnl_lock(); 1371 __netdev_notify_peers(dev); 1372 rtnl_unlock(); 1373 } 1374 EXPORT_SYMBOL(netdev_notify_peers); 1375 1376 static int napi_threaded_poll(void *data); 1377 1378 static int napi_kthread_create(struct napi_struct *n) 1379 { 1380 int err = 0; 1381 1382 /* Create and wake up the kthread once to put it in 1383 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1384 * warning and work with loadavg. 1385 */ 1386 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1387 n->dev->name, n->napi_id); 1388 if (IS_ERR(n->thread)) { 1389 err = PTR_ERR(n->thread); 1390 pr_err("kthread_run failed with err %d\n", err); 1391 n->thread = NULL; 1392 } 1393 1394 return err; 1395 } 1396 1397 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1398 { 1399 const struct net_device_ops *ops = dev->netdev_ops; 1400 int ret; 1401 1402 ASSERT_RTNL(); 1403 dev_addr_check(dev); 1404 1405 if (!netif_device_present(dev)) { 1406 /* may be detached because parent is runtime-suspended */ 1407 if (dev->dev.parent) 1408 pm_runtime_resume(dev->dev.parent); 1409 if (!netif_device_present(dev)) 1410 return -ENODEV; 1411 } 1412 1413 /* Block netpoll from trying to do any rx path servicing. 1414 * If we don't do this there is a chance ndo_poll_controller 1415 * or ndo_poll may be running while we open the device 1416 */ 1417 netpoll_poll_disable(dev); 1418 1419 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1420 ret = notifier_to_errno(ret); 1421 if (ret) 1422 return ret; 1423 1424 set_bit(__LINK_STATE_START, &dev->state); 1425 1426 if (ops->ndo_validate_addr) 1427 ret = ops->ndo_validate_addr(dev); 1428 1429 if (!ret && ops->ndo_open) 1430 ret = ops->ndo_open(dev); 1431 1432 netpoll_poll_enable(dev); 1433 1434 if (ret) 1435 clear_bit(__LINK_STATE_START, &dev->state); 1436 else { 1437 dev->flags |= IFF_UP; 1438 dev_set_rx_mode(dev); 1439 dev_activate(dev); 1440 add_device_randomness(dev->dev_addr, dev->addr_len); 1441 } 1442 1443 return ret; 1444 } 1445 1446 /** 1447 * dev_open - prepare an interface for use. 1448 * @dev: device to open 1449 * @extack: netlink extended ack 1450 * 1451 * Takes a device from down to up state. The device's private open 1452 * function is invoked and then the multicast lists are loaded. Finally 1453 * the device is moved into the up state and a %NETDEV_UP message is 1454 * sent to the netdev notifier chain. 1455 * 1456 * Calling this function on an active interface is a nop. On a failure 1457 * a negative errno code is returned. 1458 */ 1459 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1460 { 1461 int ret; 1462 1463 if (dev->flags & IFF_UP) 1464 return 0; 1465 1466 ret = __dev_open(dev, extack); 1467 if (ret < 0) 1468 return ret; 1469 1470 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1471 call_netdevice_notifiers(NETDEV_UP, dev); 1472 1473 return ret; 1474 } 1475 EXPORT_SYMBOL(dev_open); 1476 1477 static void __dev_close_many(struct list_head *head) 1478 { 1479 struct net_device *dev; 1480 1481 ASSERT_RTNL(); 1482 might_sleep(); 1483 1484 list_for_each_entry(dev, head, close_list) { 1485 /* Temporarily disable netpoll until the interface is down */ 1486 netpoll_poll_disable(dev); 1487 1488 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1489 1490 clear_bit(__LINK_STATE_START, &dev->state); 1491 1492 /* Synchronize to scheduled poll. We cannot touch poll list, it 1493 * can be even on different cpu. So just clear netif_running(). 1494 * 1495 * dev->stop() will invoke napi_disable() on all of it's 1496 * napi_struct instances on this device. 1497 */ 1498 smp_mb__after_atomic(); /* Commit netif_running(). */ 1499 } 1500 1501 dev_deactivate_many(head); 1502 1503 list_for_each_entry(dev, head, close_list) { 1504 const struct net_device_ops *ops = dev->netdev_ops; 1505 1506 /* 1507 * Call the device specific close. This cannot fail. 1508 * Only if device is UP 1509 * 1510 * We allow it to be called even after a DETACH hot-plug 1511 * event. 1512 */ 1513 if (ops->ndo_stop) 1514 ops->ndo_stop(dev); 1515 1516 dev->flags &= ~IFF_UP; 1517 netpoll_poll_enable(dev); 1518 } 1519 } 1520 1521 static void __dev_close(struct net_device *dev) 1522 { 1523 LIST_HEAD(single); 1524 1525 list_add(&dev->close_list, &single); 1526 __dev_close_many(&single); 1527 list_del(&single); 1528 } 1529 1530 void dev_close_many(struct list_head *head, bool unlink) 1531 { 1532 struct net_device *dev, *tmp; 1533 1534 /* Remove the devices that don't need to be closed */ 1535 list_for_each_entry_safe(dev, tmp, head, close_list) 1536 if (!(dev->flags & IFF_UP)) 1537 list_del_init(&dev->close_list); 1538 1539 __dev_close_many(head); 1540 1541 list_for_each_entry_safe(dev, tmp, head, close_list) { 1542 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1543 call_netdevice_notifiers(NETDEV_DOWN, dev); 1544 if (unlink) 1545 list_del_init(&dev->close_list); 1546 } 1547 } 1548 EXPORT_SYMBOL(dev_close_many); 1549 1550 /** 1551 * dev_close - shutdown an interface. 1552 * @dev: device to shutdown 1553 * 1554 * This function moves an active device into down state. A 1555 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1556 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1557 * chain. 1558 */ 1559 void dev_close(struct net_device *dev) 1560 { 1561 if (dev->flags & IFF_UP) { 1562 LIST_HEAD(single); 1563 1564 list_add(&dev->close_list, &single); 1565 dev_close_many(&single, true); 1566 list_del(&single); 1567 } 1568 } 1569 EXPORT_SYMBOL(dev_close); 1570 1571 1572 /** 1573 * dev_disable_lro - disable Large Receive Offload on a device 1574 * @dev: device 1575 * 1576 * Disable Large Receive Offload (LRO) on a net device. Must be 1577 * called under RTNL. This is needed if received packets may be 1578 * forwarded to another interface. 1579 */ 1580 void dev_disable_lro(struct net_device *dev) 1581 { 1582 struct net_device *lower_dev; 1583 struct list_head *iter; 1584 1585 dev->wanted_features &= ~NETIF_F_LRO; 1586 netdev_update_features(dev); 1587 1588 if (unlikely(dev->features & NETIF_F_LRO)) 1589 netdev_WARN(dev, "failed to disable LRO!\n"); 1590 1591 netdev_for_each_lower_dev(dev, lower_dev, iter) 1592 dev_disable_lro(lower_dev); 1593 } 1594 EXPORT_SYMBOL(dev_disable_lro); 1595 1596 /** 1597 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1598 * @dev: device 1599 * 1600 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1601 * called under RTNL. This is needed if Generic XDP is installed on 1602 * the device. 1603 */ 1604 static void dev_disable_gro_hw(struct net_device *dev) 1605 { 1606 dev->wanted_features &= ~NETIF_F_GRO_HW; 1607 netdev_update_features(dev); 1608 1609 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1610 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1611 } 1612 1613 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1614 { 1615 #define N(val) \ 1616 case NETDEV_##val: \ 1617 return "NETDEV_" __stringify(val); 1618 switch (cmd) { 1619 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1620 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1621 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1622 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1623 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1624 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1625 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1626 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1627 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1628 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1629 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1630 } 1631 #undef N 1632 return "UNKNOWN_NETDEV_EVENT"; 1633 } 1634 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1635 1636 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1637 struct net_device *dev) 1638 { 1639 struct netdev_notifier_info info = { 1640 .dev = dev, 1641 }; 1642 1643 return nb->notifier_call(nb, val, &info); 1644 } 1645 1646 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1647 struct net_device *dev) 1648 { 1649 int err; 1650 1651 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1652 err = notifier_to_errno(err); 1653 if (err) 1654 return err; 1655 1656 if (!(dev->flags & IFF_UP)) 1657 return 0; 1658 1659 call_netdevice_notifier(nb, NETDEV_UP, dev); 1660 return 0; 1661 } 1662 1663 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1664 struct net_device *dev) 1665 { 1666 if (dev->flags & IFF_UP) { 1667 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1668 dev); 1669 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1670 } 1671 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1672 } 1673 1674 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1675 struct net *net) 1676 { 1677 struct net_device *dev; 1678 int err; 1679 1680 for_each_netdev(net, dev) { 1681 err = call_netdevice_register_notifiers(nb, dev); 1682 if (err) 1683 goto rollback; 1684 } 1685 return 0; 1686 1687 rollback: 1688 for_each_netdev_continue_reverse(net, dev) 1689 call_netdevice_unregister_notifiers(nb, dev); 1690 return err; 1691 } 1692 1693 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1694 struct net *net) 1695 { 1696 struct net_device *dev; 1697 1698 for_each_netdev(net, dev) 1699 call_netdevice_unregister_notifiers(nb, dev); 1700 } 1701 1702 static int dev_boot_phase = 1; 1703 1704 /** 1705 * register_netdevice_notifier - register a network notifier block 1706 * @nb: notifier 1707 * 1708 * Register a notifier to be called when network device events occur. 1709 * The notifier passed is linked into the kernel structures and must 1710 * not be reused until it has been unregistered. A negative errno code 1711 * is returned on a failure. 1712 * 1713 * When registered all registration and up events are replayed 1714 * to the new notifier to allow device to have a race free 1715 * view of the network device list. 1716 */ 1717 1718 int register_netdevice_notifier(struct notifier_block *nb) 1719 { 1720 struct net *net; 1721 int err; 1722 1723 /* Close race with setup_net() and cleanup_net() */ 1724 down_write(&pernet_ops_rwsem); 1725 rtnl_lock(); 1726 err = raw_notifier_chain_register(&netdev_chain, nb); 1727 if (err) 1728 goto unlock; 1729 if (dev_boot_phase) 1730 goto unlock; 1731 for_each_net(net) { 1732 err = call_netdevice_register_net_notifiers(nb, net); 1733 if (err) 1734 goto rollback; 1735 } 1736 1737 unlock: 1738 rtnl_unlock(); 1739 up_write(&pernet_ops_rwsem); 1740 return err; 1741 1742 rollback: 1743 for_each_net_continue_reverse(net) 1744 call_netdevice_unregister_net_notifiers(nb, net); 1745 1746 raw_notifier_chain_unregister(&netdev_chain, nb); 1747 goto unlock; 1748 } 1749 EXPORT_SYMBOL(register_netdevice_notifier); 1750 1751 /** 1752 * unregister_netdevice_notifier - unregister a network notifier block 1753 * @nb: notifier 1754 * 1755 * Unregister a notifier previously registered by 1756 * register_netdevice_notifier(). The notifier is unlinked into the 1757 * kernel structures and may then be reused. A negative errno code 1758 * is returned on a failure. 1759 * 1760 * After unregistering unregister and down device events are synthesized 1761 * for all devices on the device list to the removed notifier to remove 1762 * the need for special case cleanup code. 1763 */ 1764 1765 int unregister_netdevice_notifier(struct notifier_block *nb) 1766 { 1767 struct net *net; 1768 int err; 1769 1770 /* Close race with setup_net() and cleanup_net() */ 1771 down_write(&pernet_ops_rwsem); 1772 rtnl_lock(); 1773 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1774 if (err) 1775 goto unlock; 1776 1777 for_each_net(net) 1778 call_netdevice_unregister_net_notifiers(nb, net); 1779 1780 unlock: 1781 rtnl_unlock(); 1782 up_write(&pernet_ops_rwsem); 1783 return err; 1784 } 1785 EXPORT_SYMBOL(unregister_netdevice_notifier); 1786 1787 static int __register_netdevice_notifier_net(struct net *net, 1788 struct notifier_block *nb, 1789 bool ignore_call_fail) 1790 { 1791 int err; 1792 1793 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1794 if (err) 1795 return err; 1796 if (dev_boot_phase) 1797 return 0; 1798 1799 err = call_netdevice_register_net_notifiers(nb, net); 1800 if (err && !ignore_call_fail) 1801 goto chain_unregister; 1802 1803 return 0; 1804 1805 chain_unregister: 1806 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1807 return err; 1808 } 1809 1810 static int __unregister_netdevice_notifier_net(struct net *net, 1811 struct notifier_block *nb) 1812 { 1813 int err; 1814 1815 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1816 if (err) 1817 return err; 1818 1819 call_netdevice_unregister_net_notifiers(nb, net); 1820 return 0; 1821 } 1822 1823 /** 1824 * register_netdevice_notifier_net - register a per-netns network notifier block 1825 * @net: network namespace 1826 * @nb: notifier 1827 * 1828 * Register a notifier to be called when network device events occur. 1829 * The notifier passed is linked into the kernel structures and must 1830 * not be reused until it has been unregistered. A negative errno code 1831 * is returned on a failure. 1832 * 1833 * When registered all registration and up events are replayed 1834 * to the new notifier to allow device to have a race free 1835 * view of the network device list. 1836 */ 1837 1838 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1839 { 1840 int err; 1841 1842 rtnl_lock(); 1843 err = __register_netdevice_notifier_net(net, nb, false); 1844 rtnl_unlock(); 1845 return err; 1846 } 1847 EXPORT_SYMBOL(register_netdevice_notifier_net); 1848 1849 /** 1850 * unregister_netdevice_notifier_net - unregister a per-netns 1851 * network notifier block 1852 * @net: network namespace 1853 * @nb: notifier 1854 * 1855 * Unregister a notifier previously registered by 1856 * register_netdevice_notifier(). The notifier is unlinked into the 1857 * kernel structures and may then be reused. A negative errno code 1858 * is returned on a failure. 1859 * 1860 * After unregistering unregister and down device events are synthesized 1861 * for all devices on the device list to the removed notifier to remove 1862 * the need for special case cleanup code. 1863 */ 1864 1865 int unregister_netdevice_notifier_net(struct net *net, 1866 struct notifier_block *nb) 1867 { 1868 int err; 1869 1870 rtnl_lock(); 1871 err = __unregister_netdevice_notifier_net(net, nb); 1872 rtnl_unlock(); 1873 return err; 1874 } 1875 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1876 1877 int register_netdevice_notifier_dev_net(struct net_device *dev, 1878 struct notifier_block *nb, 1879 struct netdev_net_notifier *nn) 1880 { 1881 int err; 1882 1883 rtnl_lock(); 1884 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1885 if (!err) { 1886 nn->nb = nb; 1887 list_add(&nn->list, &dev->net_notifier_list); 1888 } 1889 rtnl_unlock(); 1890 return err; 1891 } 1892 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1893 1894 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1895 struct notifier_block *nb, 1896 struct netdev_net_notifier *nn) 1897 { 1898 int err; 1899 1900 rtnl_lock(); 1901 list_del(&nn->list); 1902 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1903 rtnl_unlock(); 1904 return err; 1905 } 1906 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1907 1908 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1909 struct net *net) 1910 { 1911 struct netdev_net_notifier *nn; 1912 1913 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1914 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1915 __register_netdevice_notifier_net(net, nn->nb, true); 1916 } 1917 } 1918 1919 /** 1920 * call_netdevice_notifiers_info - call all network notifier blocks 1921 * @val: value passed unmodified to notifier function 1922 * @info: notifier information data 1923 * 1924 * Call all network notifier blocks. Parameters and return value 1925 * are as for raw_notifier_call_chain(). 1926 */ 1927 1928 static int call_netdevice_notifiers_info(unsigned long val, 1929 struct netdev_notifier_info *info) 1930 { 1931 struct net *net = dev_net(info->dev); 1932 int ret; 1933 1934 ASSERT_RTNL(); 1935 1936 /* Run per-netns notifier block chain first, then run the global one. 1937 * Hopefully, one day, the global one is going to be removed after 1938 * all notifier block registrators get converted to be per-netns. 1939 */ 1940 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1941 if (ret & NOTIFY_STOP_MASK) 1942 return ret; 1943 return raw_notifier_call_chain(&netdev_chain, val, info); 1944 } 1945 1946 /** 1947 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1948 * for and rollback on error 1949 * @val_up: value passed unmodified to notifier function 1950 * @val_down: value passed unmodified to the notifier function when 1951 * recovering from an error on @val_up 1952 * @info: notifier information data 1953 * 1954 * Call all per-netns network notifier blocks, but not notifier blocks on 1955 * the global notifier chain. Parameters and return value are as for 1956 * raw_notifier_call_chain_robust(). 1957 */ 1958 1959 static int 1960 call_netdevice_notifiers_info_robust(unsigned long val_up, 1961 unsigned long val_down, 1962 struct netdev_notifier_info *info) 1963 { 1964 struct net *net = dev_net(info->dev); 1965 1966 ASSERT_RTNL(); 1967 1968 return raw_notifier_call_chain_robust(&net->netdev_chain, 1969 val_up, val_down, info); 1970 } 1971 1972 static int call_netdevice_notifiers_extack(unsigned long val, 1973 struct net_device *dev, 1974 struct netlink_ext_ack *extack) 1975 { 1976 struct netdev_notifier_info info = { 1977 .dev = dev, 1978 .extack = extack, 1979 }; 1980 1981 return call_netdevice_notifiers_info(val, &info); 1982 } 1983 1984 /** 1985 * call_netdevice_notifiers - call all network notifier blocks 1986 * @val: value passed unmodified to notifier function 1987 * @dev: net_device pointer passed unmodified to notifier function 1988 * 1989 * Call all network notifier blocks. Parameters and return value 1990 * are as for raw_notifier_call_chain(). 1991 */ 1992 1993 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1994 { 1995 return call_netdevice_notifiers_extack(val, dev, NULL); 1996 } 1997 EXPORT_SYMBOL(call_netdevice_notifiers); 1998 1999 /** 2000 * call_netdevice_notifiers_mtu - call all network notifier blocks 2001 * @val: value passed unmodified to notifier function 2002 * @dev: net_device pointer passed unmodified to notifier function 2003 * @arg: additional u32 argument passed to the notifier function 2004 * 2005 * Call all network notifier blocks. Parameters and return value 2006 * are as for raw_notifier_call_chain(). 2007 */ 2008 static int call_netdevice_notifiers_mtu(unsigned long val, 2009 struct net_device *dev, u32 arg) 2010 { 2011 struct netdev_notifier_info_ext info = { 2012 .info.dev = dev, 2013 .ext.mtu = arg, 2014 }; 2015 2016 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2017 2018 return call_netdevice_notifiers_info(val, &info.info); 2019 } 2020 2021 #ifdef CONFIG_NET_INGRESS 2022 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2023 2024 void net_inc_ingress_queue(void) 2025 { 2026 static_branch_inc(&ingress_needed_key); 2027 } 2028 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2029 2030 void net_dec_ingress_queue(void) 2031 { 2032 static_branch_dec(&ingress_needed_key); 2033 } 2034 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2035 #endif 2036 2037 #ifdef CONFIG_NET_EGRESS 2038 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2039 2040 void net_inc_egress_queue(void) 2041 { 2042 static_branch_inc(&egress_needed_key); 2043 } 2044 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2045 2046 void net_dec_egress_queue(void) 2047 { 2048 static_branch_dec(&egress_needed_key); 2049 } 2050 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2051 #endif 2052 2053 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2054 EXPORT_SYMBOL(netstamp_needed_key); 2055 #ifdef CONFIG_JUMP_LABEL 2056 static atomic_t netstamp_needed_deferred; 2057 static atomic_t netstamp_wanted; 2058 static void netstamp_clear(struct work_struct *work) 2059 { 2060 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2061 int wanted; 2062 2063 wanted = atomic_add_return(deferred, &netstamp_wanted); 2064 if (wanted > 0) 2065 static_branch_enable(&netstamp_needed_key); 2066 else 2067 static_branch_disable(&netstamp_needed_key); 2068 } 2069 static DECLARE_WORK(netstamp_work, netstamp_clear); 2070 #endif 2071 2072 void net_enable_timestamp(void) 2073 { 2074 #ifdef CONFIG_JUMP_LABEL 2075 int wanted; 2076 2077 while (1) { 2078 wanted = atomic_read(&netstamp_wanted); 2079 if (wanted <= 0) 2080 break; 2081 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2082 return; 2083 } 2084 atomic_inc(&netstamp_needed_deferred); 2085 schedule_work(&netstamp_work); 2086 #else 2087 static_branch_inc(&netstamp_needed_key); 2088 #endif 2089 } 2090 EXPORT_SYMBOL(net_enable_timestamp); 2091 2092 void net_disable_timestamp(void) 2093 { 2094 #ifdef CONFIG_JUMP_LABEL 2095 int wanted; 2096 2097 while (1) { 2098 wanted = atomic_read(&netstamp_wanted); 2099 if (wanted <= 1) 2100 break; 2101 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2102 return; 2103 } 2104 atomic_dec(&netstamp_needed_deferred); 2105 schedule_work(&netstamp_work); 2106 #else 2107 static_branch_dec(&netstamp_needed_key); 2108 #endif 2109 } 2110 EXPORT_SYMBOL(net_disable_timestamp); 2111 2112 static inline void net_timestamp_set(struct sk_buff *skb) 2113 { 2114 skb->tstamp = 0; 2115 skb->mono_delivery_time = 0; 2116 if (static_branch_unlikely(&netstamp_needed_key)) 2117 skb->tstamp = ktime_get_real(); 2118 } 2119 2120 #define net_timestamp_check(COND, SKB) \ 2121 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2122 if ((COND) && !(SKB)->tstamp) \ 2123 (SKB)->tstamp = ktime_get_real(); \ 2124 } \ 2125 2126 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2127 { 2128 return __is_skb_forwardable(dev, skb, true); 2129 } 2130 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2131 2132 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2133 bool check_mtu) 2134 { 2135 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2136 2137 if (likely(!ret)) { 2138 skb->protocol = eth_type_trans(skb, dev); 2139 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2140 } 2141 2142 return ret; 2143 } 2144 2145 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2146 { 2147 return __dev_forward_skb2(dev, skb, true); 2148 } 2149 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2150 2151 /** 2152 * dev_forward_skb - loopback an skb to another netif 2153 * 2154 * @dev: destination network device 2155 * @skb: buffer to forward 2156 * 2157 * return values: 2158 * NET_RX_SUCCESS (no congestion) 2159 * NET_RX_DROP (packet was dropped, but freed) 2160 * 2161 * dev_forward_skb can be used for injecting an skb from the 2162 * start_xmit function of one device into the receive queue 2163 * of another device. 2164 * 2165 * The receiving device may be in another namespace, so 2166 * we have to clear all information in the skb that could 2167 * impact namespace isolation. 2168 */ 2169 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2170 { 2171 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2172 } 2173 EXPORT_SYMBOL_GPL(dev_forward_skb); 2174 2175 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2176 { 2177 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2178 } 2179 2180 static inline int deliver_skb(struct sk_buff *skb, 2181 struct packet_type *pt_prev, 2182 struct net_device *orig_dev) 2183 { 2184 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2185 return -ENOMEM; 2186 refcount_inc(&skb->users); 2187 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2188 } 2189 2190 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2191 struct packet_type **pt, 2192 struct net_device *orig_dev, 2193 __be16 type, 2194 struct list_head *ptype_list) 2195 { 2196 struct packet_type *ptype, *pt_prev = *pt; 2197 2198 list_for_each_entry_rcu(ptype, ptype_list, list) { 2199 if (ptype->type != type) 2200 continue; 2201 if (pt_prev) 2202 deliver_skb(skb, pt_prev, orig_dev); 2203 pt_prev = ptype; 2204 } 2205 *pt = pt_prev; 2206 } 2207 2208 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2209 { 2210 if (!ptype->af_packet_priv || !skb->sk) 2211 return false; 2212 2213 if (ptype->id_match) 2214 return ptype->id_match(ptype, skb->sk); 2215 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2216 return true; 2217 2218 return false; 2219 } 2220 2221 /** 2222 * dev_nit_active - return true if any network interface taps are in use 2223 * 2224 * @dev: network device to check for the presence of taps 2225 */ 2226 bool dev_nit_active(struct net_device *dev) 2227 { 2228 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2229 } 2230 EXPORT_SYMBOL_GPL(dev_nit_active); 2231 2232 /* 2233 * Support routine. Sends outgoing frames to any network 2234 * taps currently in use. 2235 */ 2236 2237 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2238 { 2239 struct packet_type *ptype; 2240 struct sk_buff *skb2 = NULL; 2241 struct packet_type *pt_prev = NULL; 2242 struct list_head *ptype_list = &ptype_all; 2243 2244 rcu_read_lock(); 2245 again: 2246 list_for_each_entry_rcu(ptype, ptype_list, list) { 2247 if (ptype->ignore_outgoing) 2248 continue; 2249 2250 /* Never send packets back to the socket 2251 * they originated from - MvS (miquels@drinkel.ow.org) 2252 */ 2253 if (skb_loop_sk(ptype, skb)) 2254 continue; 2255 2256 if (pt_prev) { 2257 deliver_skb(skb2, pt_prev, skb->dev); 2258 pt_prev = ptype; 2259 continue; 2260 } 2261 2262 /* need to clone skb, done only once */ 2263 skb2 = skb_clone(skb, GFP_ATOMIC); 2264 if (!skb2) 2265 goto out_unlock; 2266 2267 net_timestamp_set(skb2); 2268 2269 /* skb->nh should be correctly 2270 * set by sender, so that the second statement is 2271 * just protection against buggy protocols. 2272 */ 2273 skb_reset_mac_header(skb2); 2274 2275 if (skb_network_header(skb2) < skb2->data || 2276 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2277 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2278 ntohs(skb2->protocol), 2279 dev->name); 2280 skb_reset_network_header(skb2); 2281 } 2282 2283 skb2->transport_header = skb2->network_header; 2284 skb2->pkt_type = PACKET_OUTGOING; 2285 pt_prev = ptype; 2286 } 2287 2288 if (ptype_list == &ptype_all) { 2289 ptype_list = &dev->ptype_all; 2290 goto again; 2291 } 2292 out_unlock: 2293 if (pt_prev) { 2294 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2295 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2296 else 2297 kfree_skb(skb2); 2298 } 2299 rcu_read_unlock(); 2300 } 2301 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2302 2303 /** 2304 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2305 * @dev: Network device 2306 * @txq: number of queues available 2307 * 2308 * If real_num_tx_queues is changed the tc mappings may no longer be 2309 * valid. To resolve this verify the tc mapping remains valid and if 2310 * not NULL the mapping. With no priorities mapping to this 2311 * offset/count pair it will no longer be used. In the worst case TC0 2312 * is invalid nothing can be done so disable priority mappings. If is 2313 * expected that drivers will fix this mapping if they can before 2314 * calling netif_set_real_num_tx_queues. 2315 */ 2316 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2317 { 2318 int i; 2319 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2320 2321 /* If TC0 is invalidated disable TC mapping */ 2322 if (tc->offset + tc->count > txq) { 2323 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2324 dev->num_tc = 0; 2325 return; 2326 } 2327 2328 /* Invalidated prio to tc mappings set to TC0 */ 2329 for (i = 1; i < TC_BITMASK + 1; i++) { 2330 int q = netdev_get_prio_tc_map(dev, i); 2331 2332 tc = &dev->tc_to_txq[q]; 2333 if (tc->offset + tc->count > txq) { 2334 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2335 i, q); 2336 netdev_set_prio_tc_map(dev, i, 0); 2337 } 2338 } 2339 } 2340 2341 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2342 { 2343 if (dev->num_tc) { 2344 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2345 int i; 2346 2347 /* walk through the TCs and see if it falls into any of them */ 2348 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2349 if ((txq - tc->offset) < tc->count) 2350 return i; 2351 } 2352 2353 /* didn't find it, just return -1 to indicate no match */ 2354 return -1; 2355 } 2356 2357 return 0; 2358 } 2359 EXPORT_SYMBOL(netdev_txq_to_tc); 2360 2361 #ifdef CONFIG_XPS 2362 static struct static_key xps_needed __read_mostly; 2363 static struct static_key xps_rxqs_needed __read_mostly; 2364 static DEFINE_MUTEX(xps_map_mutex); 2365 #define xmap_dereference(P) \ 2366 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2367 2368 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2369 struct xps_dev_maps *old_maps, int tci, u16 index) 2370 { 2371 struct xps_map *map = NULL; 2372 int pos; 2373 2374 if (dev_maps) 2375 map = xmap_dereference(dev_maps->attr_map[tci]); 2376 if (!map) 2377 return false; 2378 2379 for (pos = map->len; pos--;) { 2380 if (map->queues[pos] != index) 2381 continue; 2382 2383 if (map->len > 1) { 2384 map->queues[pos] = map->queues[--map->len]; 2385 break; 2386 } 2387 2388 if (old_maps) 2389 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2390 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2391 kfree_rcu(map, rcu); 2392 return false; 2393 } 2394 2395 return true; 2396 } 2397 2398 static bool remove_xps_queue_cpu(struct net_device *dev, 2399 struct xps_dev_maps *dev_maps, 2400 int cpu, u16 offset, u16 count) 2401 { 2402 int num_tc = dev_maps->num_tc; 2403 bool active = false; 2404 int tci; 2405 2406 for (tci = cpu * num_tc; num_tc--; tci++) { 2407 int i, j; 2408 2409 for (i = count, j = offset; i--; j++) { 2410 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2411 break; 2412 } 2413 2414 active |= i < 0; 2415 } 2416 2417 return active; 2418 } 2419 2420 static void reset_xps_maps(struct net_device *dev, 2421 struct xps_dev_maps *dev_maps, 2422 enum xps_map_type type) 2423 { 2424 static_key_slow_dec_cpuslocked(&xps_needed); 2425 if (type == XPS_RXQS) 2426 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2427 2428 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2429 2430 kfree_rcu(dev_maps, rcu); 2431 } 2432 2433 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2434 u16 offset, u16 count) 2435 { 2436 struct xps_dev_maps *dev_maps; 2437 bool active = false; 2438 int i, j; 2439 2440 dev_maps = xmap_dereference(dev->xps_maps[type]); 2441 if (!dev_maps) 2442 return; 2443 2444 for (j = 0; j < dev_maps->nr_ids; j++) 2445 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2446 if (!active) 2447 reset_xps_maps(dev, dev_maps, type); 2448 2449 if (type == XPS_CPUS) { 2450 for (i = offset + (count - 1); count--; i--) 2451 netdev_queue_numa_node_write( 2452 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2453 } 2454 } 2455 2456 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2457 u16 count) 2458 { 2459 if (!static_key_false(&xps_needed)) 2460 return; 2461 2462 cpus_read_lock(); 2463 mutex_lock(&xps_map_mutex); 2464 2465 if (static_key_false(&xps_rxqs_needed)) 2466 clean_xps_maps(dev, XPS_RXQS, offset, count); 2467 2468 clean_xps_maps(dev, XPS_CPUS, offset, count); 2469 2470 mutex_unlock(&xps_map_mutex); 2471 cpus_read_unlock(); 2472 } 2473 2474 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2475 { 2476 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2477 } 2478 2479 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2480 u16 index, bool is_rxqs_map) 2481 { 2482 struct xps_map *new_map; 2483 int alloc_len = XPS_MIN_MAP_ALLOC; 2484 int i, pos; 2485 2486 for (pos = 0; map && pos < map->len; pos++) { 2487 if (map->queues[pos] != index) 2488 continue; 2489 return map; 2490 } 2491 2492 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2493 if (map) { 2494 if (pos < map->alloc_len) 2495 return map; 2496 2497 alloc_len = map->alloc_len * 2; 2498 } 2499 2500 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2501 * map 2502 */ 2503 if (is_rxqs_map) 2504 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2505 else 2506 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2507 cpu_to_node(attr_index)); 2508 if (!new_map) 2509 return NULL; 2510 2511 for (i = 0; i < pos; i++) 2512 new_map->queues[i] = map->queues[i]; 2513 new_map->alloc_len = alloc_len; 2514 new_map->len = pos; 2515 2516 return new_map; 2517 } 2518 2519 /* Copy xps maps at a given index */ 2520 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2521 struct xps_dev_maps *new_dev_maps, int index, 2522 int tc, bool skip_tc) 2523 { 2524 int i, tci = index * dev_maps->num_tc; 2525 struct xps_map *map; 2526 2527 /* copy maps belonging to foreign traffic classes */ 2528 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2529 if (i == tc && skip_tc) 2530 continue; 2531 2532 /* fill in the new device map from the old device map */ 2533 map = xmap_dereference(dev_maps->attr_map[tci]); 2534 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2535 } 2536 } 2537 2538 /* Must be called under cpus_read_lock */ 2539 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2540 u16 index, enum xps_map_type type) 2541 { 2542 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2543 const unsigned long *online_mask = NULL; 2544 bool active = false, copy = false; 2545 int i, j, tci, numa_node_id = -2; 2546 int maps_sz, num_tc = 1, tc = 0; 2547 struct xps_map *map, *new_map; 2548 unsigned int nr_ids; 2549 2550 if (dev->num_tc) { 2551 /* Do not allow XPS on subordinate device directly */ 2552 num_tc = dev->num_tc; 2553 if (num_tc < 0) 2554 return -EINVAL; 2555 2556 /* If queue belongs to subordinate dev use its map */ 2557 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2558 2559 tc = netdev_txq_to_tc(dev, index); 2560 if (tc < 0) 2561 return -EINVAL; 2562 } 2563 2564 mutex_lock(&xps_map_mutex); 2565 2566 dev_maps = xmap_dereference(dev->xps_maps[type]); 2567 if (type == XPS_RXQS) { 2568 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2569 nr_ids = dev->num_rx_queues; 2570 } else { 2571 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2572 if (num_possible_cpus() > 1) 2573 online_mask = cpumask_bits(cpu_online_mask); 2574 nr_ids = nr_cpu_ids; 2575 } 2576 2577 if (maps_sz < L1_CACHE_BYTES) 2578 maps_sz = L1_CACHE_BYTES; 2579 2580 /* The old dev_maps could be larger or smaller than the one we're 2581 * setting up now, as dev->num_tc or nr_ids could have been updated in 2582 * between. We could try to be smart, but let's be safe instead and only 2583 * copy foreign traffic classes if the two map sizes match. 2584 */ 2585 if (dev_maps && 2586 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2587 copy = true; 2588 2589 /* allocate memory for queue storage */ 2590 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2591 j < nr_ids;) { 2592 if (!new_dev_maps) { 2593 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2594 if (!new_dev_maps) { 2595 mutex_unlock(&xps_map_mutex); 2596 return -ENOMEM; 2597 } 2598 2599 new_dev_maps->nr_ids = nr_ids; 2600 new_dev_maps->num_tc = num_tc; 2601 } 2602 2603 tci = j * num_tc + tc; 2604 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2605 2606 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2607 if (!map) 2608 goto error; 2609 2610 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2611 } 2612 2613 if (!new_dev_maps) 2614 goto out_no_new_maps; 2615 2616 if (!dev_maps) { 2617 /* Increment static keys at most once per type */ 2618 static_key_slow_inc_cpuslocked(&xps_needed); 2619 if (type == XPS_RXQS) 2620 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2621 } 2622 2623 for (j = 0; j < nr_ids; j++) { 2624 bool skip_tc = false; 2625 2626 tci = j * num_tc + tc; 2627 if (netif_attr_test_mask(j, mask, nr_ids) && 2628 netif_attr_test_online(j, online_mask, nr_ids)) { 2629 /* add tx-queue to CPU/rx-queue maps */ 2630 int pos = 0; 2631 2632 skip_tc = true; 2633 2634 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2635 while ((pos < map->len) && (map->queues[pos] != index)) 2636 pos++; 2637 2638 if (pos == map->len) 2639 map->queues[map->len++] = index; 2640 #ifdef CONFIG_NUMA 2641 if (type == XPS_CPUS) { 2642 if (numa_node_id == -2) 2643 numa_node_id = cpu_to_node(j); 2644 else if (numa_node_id != cpu_to_node(j)) 2645 numa_node_id = -1; 2646 } 2647 #endif 2648 } 2649 2650 if (copy) 2651 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2652 skip_tc); 2653 } 2654 2655 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2656 2657 /* Cleanup old maps */ 2658 if (!dev_maps) 2659 goto out_no_old_maps; 2660 2661 for (j = 0; j < dev_maps->nr_ids; j++) { 2662 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2663 map = xmap_dereference(dev_maps->attr_map[tci]); 2664 if (!map) 2665 continue; 2666 2667 if (copy) { 2668 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2669 if (map == new_map) 2670 continue; 2671 } 2672 2673 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2674 kfree_rcu(map, rcu); 2675 } 2676 } 2677 2678 old_dev_maps = dev_maps; 2679 2680 out_no_old_maps: 2681 dev_maps = new_dev_maps; 2682 active = true; 2683 2684 out_no_new_maps: 2685 if (type == XPS_CPUS) 2686 /* update Tx queue numa node */ 2687 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2688 (numa_node_id >= 0) ? 2689 numa_node_id : NUMA_NO_NODE); 2690 2691 if (!dev_maps) 2692 goto out_no_maps; 2693 2694 /* removes tx-queue from unused CPUs/rx-queues */ 2695 for (j = 0; j < dev_maps->nr_ids; j++) { 2696 tci = j * dev_maps->num_tc; 2697 2698 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2699 if (i == tc && 2700 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2701 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2702 continue; 2703 2704 active |= remove_xps_queue(dev_maps, 2705 copy ? old_dev_maps : NULL, 2706 tci, index); 2707 } 2708 } 2709 2710 if (old_dev_maps) 2711 kfree_rcu(old_dev_maps, rcu); 2712 2713 /* free map if not active */ 2714 if (!active) 2715 reset_xps_maps(dev, dev_maps, type); 2716 2717 out_no_maps: 2718 mutex_unlock(&xps_map_mutex); 2719 2720 return 0; 2721 error: 2722 /* remove any maps that we added */ 2723 for (j = 0; j < nr_ids; j++) { 2724 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2725 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2726 map = copy ? 2727 xmap_dereference(dev_maps->attr_map[tci]) : 2728 NULL; 2729 if (new_map && new_map != map) 2730 kfree(new_map); 2731 } 2732 } 2733 2734 mutex_unlock(&xps_map_mutex); 2735 2736 kfree(new_dev_maps); 2737 return -ENOMEM; 2738 } 2739 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2740 2741 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2742 u16 index) 2743 { 2744 int ret; 2745 2746 cpus_read_lock(); 2747 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2748 cpus_read_unlock(); 2749 2750 return ret; 2751 } 2752 EXPORT_SYMBOL(netif_set_xps_queue); 2753 2754 #endif 2755 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2756 { 2757 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2758 2759 /* Unbind any subordinate channels */ 2760 while (txq-- != &dev->_tx[0]) { 2761 if (txq->sb_dev) 2762 netdev_unbind_sb_channel(dev, txq->sb_dev); 2763 } 2764 } 2765 2766 void netdev_reset_tc(struct net_device *dev) 2767 { 2768 #ifdef CONFIG_XPS 2769 netif_reset_xps_queues_gt(dev, 0); 2770 #endif 2771 netdev_unbind_all_sb_channels(dev); 2772 2773 /* Reset TC configuration of device */ 2774 dev->num_tc = 0; 2775 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2776 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2777 } 2778 EXPORT_SYMBOL(netdev_reset_tc); 2779 2780 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2781 { 2782 if (tc >= dev->num_tc) 2783 return -EINVAL; 2784 2785 #ifdef CONFIG_XPS 2786 netif_reset_xps_queues(dev, offset, count); 2787 #endif 2788 dev->tc_to_txq[tc].count = count; 2789 dev->tc_to_txq[tc].offset = offset; 2790 return 0; 2791 } 2792 EXPORT_SYMBOL(netdev_set_tc_queue); 2793 2794 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2795 { 2796 if (num_tc > TC_MAX_QUEUE) 2797 return -EINVAL; 2798 2799 #ifdef CONFIG_XPS 2800 netif_reset_xps_queues_gt(dev, 0); 2801 #endif 2802 netdev_unbind_all_sb_channels(dev); 2803 2804 dev->num_tc = num_tc; 2805 return 0; 2806 } 2807 EXPORT_SYMBOL(netdev_set_num_tc); 2808 2809 void netdev_unbind_sb_channel(struct net_device *dev, 2810 struct net_device *sb_dev) 2811 { 2812 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2813 2814 #ifdef CONFIG_XPS 2815 netif_reset_xps_queues_gt(sb_dev, 0); 2816 #endif 2817 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2818 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2819 2820 while (txq-- != &dev->_tx[0]) { 2821 if (txq->sb_dev == sb_dev) 2822 txq->sb_dev = NULL; 2823 } 2824 } 2825 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2826 2827 int netdev_bind_sb_channel_queue(struct net_device *dev, 2828 struct net_device *sb_dev, 2829 u8 tc, u16 count, u16 offset) 2830 { 2831 /* Make certain the sb_dev and dev are already configured */ 2832 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2833 return -EINVAL; 2834 2835 /* We cannot hand out queues we don't have */ 2836 if ((offset + count) > dev->real_num_tx_queues) 2837 return -EINVAL; 2838 2839 /* Record the mapping */ 2840 sb_dev->tc_to_txq[tc].count = count; 2841 sb_dev->tc_to_txq[tc].offset = offset; 2842 2843 /* Provide a way for Tx queue to find the tc_to_txq map or 2844 * XPS map for itself. 2845 */ 2846 while (count--) 2847 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2848 2849 return 0; 2850 } 2851 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2852 2853 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2854 { 2855 /* Do not use a multiqueue device to represent a subordinate channel */ 2856 if (netif_is_multiqueue(dev)) 2857 return -ENODEV; 2858 2859 /* We allow channels 1 - 32767 to be used for subordinate channels. 2860 * Channel 0 is meant to be "native" mode and used only to represent 2861 * the main root device. We allow writing 0 to reset the device back 2862 * to normal mode after being used as a subordinate channel. 2863 */ 2864 if (channel > S16_MAX) 2865 return -EINVAL; 2866 2867 dev->num_tc = -channel; 2868 2869 return 0; 2870 } 2871 EXPORT_SYMBOL(netdev_set_sb_channel); 2872 2873 /* 2874 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2875 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2876 */ 2877 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2878 { 2879 bool disabling; 2880 int rc; 2881 2882 disabling = txq < dev->real_num_tx_queues; 2883 2884 if (txq < 1 || txq > dev->num_tx_queues) 2885 return -EINVAL; 2886 2887 if (dev->reg_state == NETREG_REGISTERED || 2888 dev->reg_state == NETREG_UNREGISTERING) { 2889 ASSERT_RTNL(); 2890 2891 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2892 txq); 2893 if (rc) 2894 return rc; 2895 2896 if (dev->num_tc) 2897 netif_setup_tc(dev, txq); 2898 2899 dev_qdisc_change_real_num_tx(dev, txq); 2900 2901 dev->real_num_tx_queues = txq; 2902 2903 if (disabling) { 2904 synchronize_net(); 2905 qdisc_reset_all_tx_gt(dev, txq); 2906 #ifdef CONFIG_XPS 2907 netif_reset_xps_queues_gt(dev, txq); 2908 #endif 2909 } 2910 } else { 2911 dev->real_num_tx_queues = txq; 2912 } 2913 2914 return 0; 2915 } 2916 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2917 2918 #ifdef CONFIG_SYSFS 2919 /** 2920 * netif_set_real_num_rx_queues - set actual number of RX queues used 2921 * @dev: Network device 2922 * @rxq: Actual number of RX queues 2923 * 2924 * This must be called either with the rtnl_lock held or before 2925 * registration of the net device. Returns 0 on success, or a 2926 * negative error code. If called before registration, it always 2927 * succeeds. 2928 */ 2929 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2930 { 2931 int rc; 2932 2933 if (rxq < 1 || rxq > dev->num_rx_queues) 2934 return -EINVAL; 2935 2936 if (dev->reg_state == NETREG_REGISTERED) { 2937 ASSERT_RTNL(); 2938 2939 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2940 rxq); 2941 if (rc) 2942 return rc; 2943 } 2944 2945 dev->real_num_rx_queues = rxq; 2946 return 0; 2947 } 2948 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2949 #endif 2950 2951 /** 2952 * netif_set_real_num_queues - set actual number of RX and TX queues used 2953 * @dev: Network device 2954 * @txq: Actual number of TX queues 2955 * @rxq: Actual number of RX queues 2956 * 2957 * Set the real number of both TX and RX queues. 2958 * Does nothing if the number of queues is already correct. 2959 */ 2960 int netif_set_real_num_queues(struct net_device *dev, 2961 unsigned int txq, unsigned int rxq) 2962 { 2963 unsigned int old_rxq = dev->real_num_rx_queues; 2964 int err; 2965 2966 if (txq < 1 || txq > dev->num_tx_queues || 2967 rxq < 1 || rxq > dev->num_rx_queues) 2968 return -EINVAL; 2969 2970 /* Start from increases, so the error path only does decreases - 2971 * decreases can't fail. 2972 */ 2973 if (rxq > dev->real_num_rx_queues) { 2974 err = netif_set_real_num_rx_queues(dev, rxq); 2975 if (err) 2976 return err; 2977 } 2978 if (txq > dev->real_num_tx_queues) { 2979 err = netif_set_real_num_tx_queues(dev, txq); 2980 if (err) 2981 goto undo_rx; 2982 } 2983 if (rxq < dev->real_num_rx_queues) 2984 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2985 if (txq < dev->real_num_tx_queues) 2986 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2987 2988 return 0; 2989 undo_rx: 2990 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2991 return err; 2992 } 2993 EXPORT_SYMBOL(netif_set_real_num_queues); 2994 2995 /** 2996 * netif_get_num_default_rss_queues - default number of RSS queues 2997 * 2998 * Default value is the number of physical cores if there are only 1 or 2, or 2999 * divided by 2 if there are more. 3000 */ 3001 int netif_get_num_default_rss_queues(void) 3002 { 3003 cpumask_var_t cpus; 3004 int cpu, count = 0; 3005 3006 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3007 return 1; 3008 3009 cpumask_copy(cpus, cpu_online_mask); 3010 for_each_cpu(cpu, cpus) { 3011 ++count; 3012 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3013 } 3014 free_cpumask_var(cpus); 3015 3016 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3017 } 3018 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3019 3020 static void __netif_reschedule(struct Qdisc *q) 3021 { 3022 struct softnet_data *sd; 3023 unsigned long flags; 3024 3025 local_irq_save(flags); 3026 sd = this_cpu_ptr(&softnet_data); 3027 q->next_sched = NULL; 3028 *sd->output_queue_tailp = q; 3029 sd->output_queue_tailp = &q->next_sched; 3030 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3031 local_irq_restore(flags); 3032 } 3033 3034 void __netif_schedule(struct Qdisc *q) 3035 { 3036 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3037 __netif_reschedule(q); 3038 } 3039 EXPORT_SYMBOL(__netif_schedule); 3040 3041 struct dev_kfree_skb_cb { 3042 enum skb_free_reason reason; 3043 }; 3044 3045 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3046 { 3047 return (struct dev_kfree_skb_cb *)skb->cb; 3048 } 3049 3050 void netif_schedule_queue(struct netdev_queue *txq) 3051 { 3052 rcu_read_lock(); 3053 if (!netif_xmit_stopped(txq)) { 3054 struct Qdisc *q = rcu_dereference(txq->qdisc); 3055 3056 __netif_schedule(q); 3057 } 3058 rcu_read_unlock(); 3059 } 3060 EXPORT_SYMBOL(netif_schedule_queue); 3061 3062 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3063 { 3064 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3065 struct Qdisc *q; 3066 3067 rcu_read_lock(); 3068 q = rcu_dereference(dev_queue->qdisc); 3069 __netif_schedule(q); 3070 rcu_read_unlock(); 3071 } 3072 } 3073 EXPORT_SYMBOL(netif_tx_wake_queue); 3074 3075 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3076 { 3077 unsigned long flags; 3078 3079 if (unlikely(!skb)) 3080 return; 3081 3082 if (likely(refcount_read(&skb->users) == 1)) { 3083 smp_rmb(); 3084 refcount_set(&skb->users, 0); 3085 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3086 return; 3087 } 3088 get_kfree_skb_cb(skb)->reason = reason; 3089 local_irq_save(flags); 3090 skb->next = __this_cpu_read(softnet_data.completion_queue); 3091 __this_cpu_write(softnet_data.completion_queue, skb); 3092 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3093 local_irq_restore(flags); 3094 } 3095 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3096 3097 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3098 { 3099 if (in_hardirq() || irqs_disabled()) 3100 __dev_kfree_skb_irq(skb, reason); 3101 else 3102 dev_kfree_skb(skb); 3103 } 3104 EXPORT_SYMBOL(__dev_kfree_skb_any); 3105 3106 3107 /** 3108 * netif_device_detach - mark device as removed 3109 * @dev: network device 3110 * 3111 * Mark device as removed from system and therefore no longer available. 3112 */ 3113 void netif_device_detach(struct net_device *dev) 3114 { 3115 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3116 netif_running(dev)) { 3117 netif_tx_stop_all_queues(dev); 3118 } 3119 } 3120 EXPORT_SYMBOL(netif_device_detach); 3121 3122 /** 3123 * netif_device_attach - mark device as attached 3124 * @dev: network device 3125 * 3126 * Mark device as attached from system and restart if needed. 3127 */ 3128 void netif_device_attach(struct net_device *dev) 3129 { 3130 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3131 netif_running(dev)) { 3132 netif_tx_wake_all_queues(dev); 3133 __netdev_watchdog_up(dev); 3134 } 3135 } 3136 EXPORT_SYMBOL(netif_device_attach); 3137 3138 /* 3139 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3140 * to be used as a distribution range. 3141 */ 3142 static u16 skb_tx_hash(const struct net_device *dev, 3143 const struct net_device *sb_dev, 3144 struct sk_buff *skb) 3145 { 3146 u32 hash; 3147 u16 qoffset = 0; 3148 u16 qcount = dev->real_num_tx_queues; 3149 3150 if (dev->num_tc) { 3151 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3152 3153 qoffset = sb_dev->tc_to_txq[tc].offset; 3154 qcount = sb_dev->tc_to_txq[tc].count; 3155 if (unlikely(!qcount)) { 3156 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3157 sb_dev->name, qoffset, tc); 3158 qoffset = 0; 3159 qcount = dev->real_num_tx_queues; 3160 } 3161 } 3162 3163 if (skb_rx_queue_recorded(skb)) { 3164 hash = skb_get_rx_queue(skb); 3165 if (hash >= qoffset) 3166 hash -= qoffset; 3167 while (unlikely(hash >= qcount)) 3168 hash -= qcount; 3169 return hash + qoffset; 3170 } 3171 3172 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3173 } 3174 3175 static void skb_warn_bad_offload(const struct sk_buff *skb) 3176 { 3177 static const netdev_features_t null_features; 3178 struct net_device *dev = skb->dev; 3179 const char *name = ""; 3180 3181 if (!net_ratelimit()) 3182 return; 3183 3184 if (dev) { 3185 if (dev->dev.parent) 3186 name = dev_driver_string(dev->dev.parent); 3187 else 3188 name = netdev_name(dev); 3189 } 3190 skb_dump(KERN_WARNING, skb, false); 3191 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3192 name, dev ? &dev->features : &null_features, 3193 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3194 } 3195 3196 /* 3197 * Invalidate hardware checksum when packet is to be mangled, and 3198 * complete checksum manually on outgoing path. 3199 */ 3200 int skb_checksum_help(struct sk_buff *skb) 3201 { 3202 __wsum csum; 3203 int ret = 0, offset; 3204 3205 if (skb->ip_summed == CHECKSUM_COMPLETE) 3206 goto out_set_summed; 3207 3208 if (unlikely(skb_is_gso(skb))) { 3209 skb_warn_bad_offload(skb); 3210 return -EINVAL; 3211 } 3212 3213 /* Before computing a checksum, we should make sure no frag could 3214 * be modified by an external entity : checksum could be wrong. 3215 */ 3216 if (skb_has_shared_frag(skb)) { 3217 ret = __skb_linearize(skb); 3218 if (ret) 3219 goto out; 3220 } 3221 3222 offset = skb_checksum_start_offset(skb); 3223 BUG_ON(offset >= skb_headlen(skb)); 3224 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3225 3226 offset += skb->csum_offset; 3227 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3228 3229 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3230 if (ret) 3231 goto out; 3232 3233 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3234 out_set_summed: 3235 skb->ip_summed = CHECKSUM_NONE; 3236 out: 3237 return ret; 3238 } 3239 EXPORT_SYMBOL(skb_checksum_help); 3240 3241 int skb_crc32c_csum_help(struct sk_buff *skb) 3242 { 3243 __le32 crc32c_csum; 3244 int ret = 0, offset, start; 3245 3246 if (skb->ip_summed != CHECKSUM_PARTIAL) 3247 goto out; 3248 3249 if (unlikely(skb_is_gso(skb))) 3250 goto out; 3251 3252 /* Before computing a checksum, we should make sure no frag could 3253 * be modified by an external entity : checksum could be wrong. 3254 */ 3255 if (unlikely(skb_has_shared_frag(skb))) { 3256 ret = __skb_linearize(skb); 3257 if (ret) 3258 goto out; 3259 } 3260 start = skb_checksum_start_offset(skb); 3261 offset = start + offsetof(struct sctphdr, checksum); 3262 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3263 ret = -EINVAL; 3264 goto out; 3265 } 3266 3267 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3268 if (ret) 3269 goto out; 3270 3271 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3272 skb->len - start, ~(__u32)0, 3273 crc32c_csum_stub)); 3274 *(__le32 *)(skb->data + offset) = crc32c_csum; 3275 skb->ip_summed = CHECKSUM_NONE; 3276 skb->csum_not_inet = 0; 3277 out: 3278 return ret; 3279 } 3280 3281 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3282 { 3283 __be16 type = skb->protocol; 3284 3285 /* Tunnel gso handlers can set protocol to ethernet. */ 3286 if (type == htons(ETH_P_TEB)) { 3287 struct ethhdr *eth; 3288 3289 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3290 return 0; 3291 3292 eth = (struct ethhdr *)skb->data; 3293 type = eth->h_proto; 3294 } 3295 3296 return __vlan_get_protocol(skb, type, depth); 3297 } 3298 3299 /* openvswitch calls this on rx path, so we need a different check. 3300 */ 3301 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3302 { 3303 if (tx_path) 3304 return skb->ip_summed != CHECKSUM_PARTIAL && 3305 skb->ip_summed != CHECKSUM_UNNECESSARY; 3306 3307 return skb->ip_summed == CHECKSUM_NONE; 3308 } 3309 3310 /** 3311 * __skb_gso_segment - Perform segmentation on skb. 3312 * @skb: buffer to segment 3313 * @features: features for the output path (see dev->features) 3314 * @tx_path: whether it is called in TX path 3315 * 3316 * This function segments the given skb and returns a list of segments. 3317 * 3318 * It may return NULL if the skb requires no segmentation. This is 3319 * only possible when GSO is used for verifying header integrity. 3320 * 3321 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3322 */ 3323 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3324 netdev_features_t features, bool tx_path) 3325 { 3326 struct sk_buff *segs; 3327 3328 if (unlikely(skb_needs_check(skb, tx_path))) { 3329 int err; 3330 3331 /* We're going to init ->check field in TCP or UDP header */ 3332 err = skb_cow_head(skb, 0); 3333 if (err < 0) 3334 return ERR_PTR(err); 3335 } 3336 3337 /* Only report GSO partial support if it will enable us to 3338 * support segmentation on this frame without needing additional 3339 * work. 3340 */ 3341 if (features & NETIF_F_GSO_PARTIAL) { 3342 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3343 struct net_device *dev = skb->dev; 3344 3345 partial_features |= dev->features & dev->gso_partial_features; 3346 if (!skb_gso_ok(skb, features | partial_features)) 3347 features &= ~NETIF_F_GSO_PARTIAL; 3348 } 3349 3350 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3351 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3352 3353 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3354 SKB_GSO_CB(skb)->encap_level = 0; 3355 3356 skb_reset_mac_header(skb); 3357 skb_reset_mac_len(skb); 3358 3359 segs = skb_mac_gso_segment(skb, features); 3360 3361 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3362 skb_warn_bad_offload(skb); 3363 3364 return segs; 3365 } 3366 EXPORT_SYMBOL(__skb_gso_segment); 3367 3368 /* Take action when hardware reception checksum errors are detected. */ 3369 #ifdef CONFIG_BUG 3370 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3371 { 3372 netdev_err(dev, "hw csum failure\n"); 3373 skb_dump(KERN_ERR, skb, true); 3374 dump_stack(); 3375 } 3376 3377 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3378 { 3379 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3380 } 3381 EXPORT_SYMBOL(netdev_rx_csum_fault); 3382 #endif 3383 3384 /* XXX: check that highmem exists at all on the given machine. */ 3385 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3386 { 3387 #ifdef CONFIG_HIGHMEM 3388 int i; 3389 3390 if (!(dev->features & NETIF_F_HIGHDMA)) { 3391 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3392 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3393 3394 if (PageHighMem(skb_frag_page(frag))) 3395 return 1; 3396 } 3397 } 3398 #endif 3399 return 0; 3400 } 3401 3402 /* If MPLS offload request, verify we are testing hardware MPLS features 3403 * instead of standard features for the netdev. 3404 */ 3405 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3406 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3407 netdev_features_t features, 3408 __be16 type) 3409 { 3410 if (eth_p_mpls(type)) 3411 features &= skb->dev->mpls_features; 3412 3413 return features; 3414 } 3415 #else 3416 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3417 netdev_features_t features, 3418 __be16 type) 3419 { 3420 return features; 3421 } 3422 #endif 3423 3424 static netdev_features_t harmonize_features(struct sk_buff *skb, 3425 netdev_features_t features) 3426 { 3427 __be16 type; 3428 3429 type = skb_network_protocol(skb, NULL); 3430 features = net_mpls_features(skb, features, type); 3431 3432 if (skb->ip_summed != CHECKSUM_NONE && 3433 !can_checksum_protocol(features, type)) { 3434 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3435 } 3436 if (illegal_highdma(skb->dev, skb)) 3437 features &= ~NETIF_F_SG; 3438 3439 return features; 3440 } 3441 3442 netdev_features_t passthru_features_check(struct sk_buff *skb, 3443 struct net_device *dev, 3444 netdev_features_t features) 3445 { 3446 return features; 3447 } 3448 EXPORT_SYMBOL(passthru_features_check); 3449 3450 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3451 struct net_device *dev, 3452 netdev_features_t features) 3453 { 3454 return vlan_features_check(skb, features); 3455 } 3456 3457 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3458 struct net_device *dev, 3459 netdev_features_t features) 3460 { 3461 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3462 3463 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3464 return features & ~NETIF_F_GSO_MASK; 3465 3466 if (!skb_shinfo(skb)->gso_type) { 3467 skb_warn_bad_offload(skb); 3468 return features & ~NETIF_F_GSO_MASK; 3469 } 3470 3471 /* Support for GSO partial features requires software 3472 * intervention before we can actually process the packets 3473 * so we need to strip support for any partial features now 3474 * and we can pull them back in after we have partially 3475 * segmented the frame. 3476 */ 3477 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3478 features &= ~dev->gso_partial_features; 3479 3480 /* Make sure to clear the IPv4 ID mangling feature if the 3481 * IPv4 header has the potential to be fragmented. 3482 */ 3483 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3484 struct iphdr *iph = skb->encapsulation ? 3485 inner_ip_hdr(skb) : ip_hdr(skb); 3486 3487 if (!(iph->frag_off & htons(IP_DF))) 3488 features &= ~NETIF_F_TSO_MANGLEID; 3489 } 3490 3491 return features; 3492 } 3493 3494 netdev_features_t netif_skb_features(struct sk_buff *skb) 3495 { 3496 struct net_device *dev = skb->dev; 3497 netdev_features_t features = dev->features; 3498 3499 if (skb_is_gso(skb)) 3500 features = gso_features_check(skb, dev, features); 3501 3502 /* If encapsulation offload request, verify we are testing 3503 * hardware encapsulation features instead of standard 3504 * features for the netdev 3505 */ 3506 if (skb->encapsulation) 3507 features &= dev->hw_enc_features; 3508 3509 if (skb_vlan_tagged(skb)) 3510 features = netdev_intersect_features(features, 3511 dev->vlan_features | 3512 NETIF_F_HW_VLAN_CTAG_TX | 3513 NETIF_F_HW_VLAN_STAG_TX); 3514 3515 if (dev->netdev_ops->ndo_features_check) 3516 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3517 features); 3518 else 3519 features &= dflt_features_check(skb, dev, features); 3520 3521 return harmonize_features(skb, features); 3522 } 3523 EXPORT_SYMBOL(netif_skb_features); 3524 3525 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3526 struct netdev_queue *txq, bool more) 3527 { 3528 unsigned int len; 3529 int rc; 3530 3531 if (dev_nit_active(dev)) 3532 dev_queue_xmit_nit(skb, dev); 3533 3534 len = skb->len; 3535 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3536 trace_net_dev_start_xmit(skb, dev); 3537 rc = netdev_start_xmit(skb, dev, txq, more); 3538 trace_net_dev_xmit(skb, rc, dev, len); 3539 3540 return rc; 3541 } 3542 3543 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3544 struct netdev_queue *txq, int *ret) 3545 { 3546 struct sk_buff *skb = first; 3547 int rc = NETDEV_TX_OK; 3548 3549 while (skb) { 3550 struct sk_buff *next = skb->next; 3551 3552 skb_mark_not_on_list(skb); 3553 rc = xmit_one(skb, dev, txq, next != NULL); 3554 if (unlikely(!dev_xmit_complete(rc))) { 3555 skb->next = next; 3556 goto out; 3557 } 3558 3559 skb = next; 3560 if (netif_tx_queue_stopped(txq) && skb) { 3561 rc = NETDEV_TX_BUSY; 3562 break; 3563 } 3564 } 3565 3566 out: 3567 *ret = rc; 3568 return skb; 3569 } 3570 3571 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3572 netdev_features_t features) 3573 { 3574 if (skb_vlan_tag_present(skb) && 3575 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3576 skb = __vlan_hwaccel_push_inside(skb); 3577 return skb; 3578 } 3579 3580 int skb_csum_hwoffload_help(struct sk_buff *skb, 3581 const netdev_features_t features) 3582 { 3583 if (unlikely(skb_csum_is_sctp(skb))) 3584 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3585 skb_crc32c_csum_help(skb); 3586 3587 if (features & NETIF_F_HW_CSUM) 3588 return 0; 3589 3590 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3591 switch (skb->csum_offset) { 3592 case offsetof(struct tcphdr, check): 3593 case offsetof(struct udphdr, check): 3594 return 0; 3595 } 3596 } 3597 3598 return skb_checksum_help(skb); 3599 } 3600 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3601 3602 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3603 { 3604 netdev_features_t features; 3605 3606 features = netif_skb_features(skb); 3607 skb = validate_xmit_vlan(skb, features); 3608 if (unlikely(!skb)) 3609 goto out_null; 3610 3611 skb = sk_validate_xmit_skb(skb, dev); 3612 if (unlikely(!skb)) 3613 goto out_null; 3614 3615 if (netif_needs_gso(skb, features)) { 3616 struct sk_buff *segs; 3617 3618 segs = skb_gso_segment(skb, features); 3619 if (IS_ERR(segs)) { 3620 goto out_kfree_skb; 3621 } else if (segs) { 3622 consume_skb(skb); 3623 skb = segs; 3624 } 3625 } else { 3626 if (skb_needs_linearize(skb, features) && 3627 __skb_linearize(skb)) 3628 goto out_kfree_skb; 3629 3630 /* If packet is not checksummed and device does not 3631 * support checksumming for this protocol, complete 3632 * checksumming here. 3633 */ 3634 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3635 if (skb->encapsulation) 3636 skb_set_inner_transport_header(skb, 3637 skb_checksum_start_offset(skb)); 3638 else 3639 skb_set_transport_header(skb, 3640 skb_checksum_start_offset(skb)); 3641 if (skb_csum_hwoffload_help(skb, features)) 3642 goto out_kfree_skb; 3643 } 3644 } 3645 3646 skb = validate_xmit_xfrm(skb, features, again); 3647 3648 return skb; 3649 3650 out_kfree_skb: 3651 kfree_skb(skb); 3652 out_null: 3653 dev_core_stats_tx_dropped_inc(dev); 3654 return NULL; 3655 } 3656 3657 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3658 { 3659 struct sk_buff *next, *head = NULL, *tail; 3660 3661 for (; skb != NULL; skb = next) { 3662 next = skb->next; 3663 skb_mark_not_on_list(skb); 3664 3665 /* in case skb wont be segmented, point to itself */ 3666 skb->prev = skb; 3667 3668 skb = validate_xmit_skb(skb, dev, again); 3669 if (!skb) 3670 continue; 3671 3672 if (!head) 3673 head = skb; 3674 else 3675 tail->next = skb; 3676 /* If skb was segmented, skb->prev points to 3677 * the last segment. If not, it still contains skb. 3678 */ 3679 tail = skb->prev; 3680 } 3681 return head; 3682 } 3683 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3684 3685 static void qdisc_pkt_len_init(struct sk_buff *skb) 3686 { 3687 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3688 3689 qdisc_skb_cb(skb)->pkt_len = skb->len; 3690 3691 /* To get more precise estimation of bytes sent on wire, 3692 * we add to pkt_len the headers size of all segments 3693 */ 3694 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3695 unsigned int hdr_len; 3696 u16 gso_segs = shinfo->gso_segs; 3697 3698 /* mac layer + network layer */ 3699 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3700 3701 /* + transport layer */ 3702 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3703 const struct tcphdr *th; 3704 struct tcphdr _tcphdr; 3705 3706 th = skb_header_pointer(skb, skb_transport_offset(skb), 3707 sizeof(_tcphdr), &_tcphdr); 3708 if (likely(th)) 3709 hdr_len += __tcp_hdrlen(th); 3710 } else { 3711 struct udphdr _udphdr; 3712 3713 if (skb_header_pointer(skb, skb_transport_offset(skb), 3714 sizeof(_udphdr), &_udphdr)) 3715 hdr_len += sizeof(struct udphdr); 3716 } 3717 3718 if (shinfo->gso_type & SKB_GSO_DODGY) 3719 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3720 shinfo->gso_size); 3721 3722 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3723 } 3724 } 3725 3726 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3727 struct sk_buff **to_free, 3728 struct netdev_queue *txq) 3729 { 3730 int rc; 3731 3732 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3733 if (rc == NET_XMIT_SUCCESS) 3734 trace_qdisc_enqueue(q, txq, skb); 3735 return rc; 3736 } 3737 3738 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3739 struct net_device *dev, 3740 struct netdev_queue *txq) 3741 { 3742 spinlock_t *root_lock = qdisc_lock(q); 3743 struct sk_buff *to_free = NULL; 3744 bool contended; 3745 int rc; 3746 3747 qdisc_calculate_pkt_len(skb, q); 3748 3749 if (q->flags & TCQ_F_NOLOCK) { 3750 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3751 qdisc_run_begin(q)) { 3752 /* Retest nolock_qdisc_is_empty() within the protection 3753 * of q->seqlock to protect from racing with requeuing. 3754 */ 3755 if (unlikely(!nolock_qdisc_is_empty(q))) { 3756 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3757 __qdisc_run(q); 3758 qdisc_run_end(q); 3759 3760 goto no_lock_out; 3761 } 3762 3763 qdisc_bstats_cpu_update(q, skb); 3764 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3765 !nolock_qdisc_is_empty(q)) 3766 __qdisc_run(q); 3767 3768 qdisc_run_end(q); 3769 return NET_XMIT_SUCCESS; 3770 } 3771 3772 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3773 qdisc_run(q); 3774 3775 no_lock_out: 3776 if (unlikely(to_free)) 3777 kfree_skb_list_reason(to_free, 3778 SKB_DROP_REASON_QDISC_DROP); 3779 return rc; 3780 } 3781 3782 /* 3783 * Heuristic to force contended enqueues to serialize on a 3784 * separate lock before trying to get qdisc main lock. 3785 * This permits qdisc->running owner to get the lock more 3786 * often and dequeue packets faster. 3787 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3788 * and then other tasks will only enqueue packets. The packets will be 3789 * sent after the qdisc owner is scheduled again. To prevent this 3790 * scenario the task always serialize on the lock. 3791 */ 3792 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3793 if (unlikely(contended)) 3794 spin_lock(&q->busylock); 3795 3796 spin_lock(root_lock); 3797 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3798 __qdisc_drop(skb, &to_free); 3799 rc = NET_XMIT_DROP; 3800 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3801 qdisc_run_begin(q)) { 3802 /* 3803 * This is a work-conserving queue; there are no old skbs 3804 * waiting to be sent out; and the qdisc is not running - 3805 * xmit the skb directly. 3806 */ 3807 3808 qdisc_bstats_update(q, skb); 3809 3810 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3811 if (unlikely(contended)) { 3812 spin_unlock(&q->busylock); 3813 contended = false; 3814 } 3815 __qdisc_run(q); 3816 } 3817 3818 qdisc_run_end(q); 3819 rc = NET_XMIT_SUCCESS; 3820 } else { 3821 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3822 if (qdisc_run_begin(q)) { 3823 if (unlikely(contended)) { 3824 spin_unlock(&q->busylock); 3825 contended = false; 3826 } 3827 __qdisc_run(q); 3828 qdisc_run_end(q); 3829 } 3830 } 3831 spin_unlock(root_lock); 3832 if (unlikely(to_free)) 3833 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3834 if (unlikely(contended)) 3835 spin_unlock(&q->busylock); 3836 return rc; 3837 } 3838 3839 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3840 static void skb_update_prio(struct sk_buff *skb) 3841 { 3842 const struct netprio_map *map; 3843 const struct sock *sk; 3844 unsigned int prioidx; 3845 3846 if (skb->priority) 3847 return; 3848 map = rcu_dereference_bh(skb->dev->priomap); 3849 if (!map) 3850 return; 3851 sk = skb_to_full_sk(skb); 3852 if (!sk) 3853 return; 3854 3855 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3856 3857 if (prioidx < map->priomap_len) 3858 skb->priority = map->priomap[prioidx]; 3859 } 3860 #else 3861 #define skb_update_prio(skb) 3862 #endif 3863 3864 /** 3865 * dev_loopback_xmit - loop back @skb 3866 * @net: network namespace this loopback is happening in 3867 * @sk: sk needed to be a netfilter okfn 3868 * @skb: buffer to transmit 3869 */ 3870 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3871 { 3872 skb_reset_mac_header(skb); 3873 __skb_pull(skb, skb_network_offset(skb)); 3874 skb->pkt_type = PACKET_LOOPBACK; 3875 if (skb->ip_summed == CHECKSUM_NONE) 3876 skb->ip_summed = CHECKSUM_UNNECESSARY; 3877 WARN_ON(!skb_dst(skb)); 3878 skb_dst_force(skb); 3879 netif_rx(skb); 3880 return 0; 3881 } 3882 EXPORT_SYMBOL(dev_loopback_xmit); 3883 3884 #ifdef CONFIG_NET_EGRESS 3885 static struct sk_buff * 3886 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3887 { 3888 #ifdef CONFIG_NET_CLS_ACT 3889 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3890 struct tcf_result cl_res; 3891 3892 if (!miniq) 3893 return skb; 3894 3895 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3896 tc_skb_cb(skb)->mru = 0; 3897 tc_skb_cb(skb)->post_ct = false; 3898 mini_qdisc_bstats_cpu_update(miniq, skb); 3899 3900 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3901 case TC_ACT_OK: 3902 case TC_ACT_RECLASSIFY: 3903 skb->tc_index = TC_H_MIN(cl_res.classid); 3904 break; 3905 case TC_ACT_SHOT: 3906 mini_qdisc_qstats_cpu_drop(miniq); 3907 *ret = NET_XMIT_DROP; 3908 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3909 return NULL; 3910 case TC_ACT_STOLEN: 3911 case TC_ACT_QUEUED: 3912 case TC_ACT_TRAP: 3913 *ret = NET_XMIT_SUCCESS; 3914 consume_skb(skb); 3915 return NULL; 3916 case TC_ACT_REDIRECT: 3917 /* No need to push/pop skb's mac_header here on egress! */ 3918 skb_do_redirect(skb); 3919 *ret = NET_XMIT_SUCCESS; 3920 return NULL; 3921 default: 3922 break; 3923 } 3924 #endif /* CONFIG_NET_CLS_ACT */ 3925 3926 return skb; 3927 } 3928 3929 static struct netdev_queue * 3930 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3931 { 3932 int qm = skb_get_queue_mapping(skb); 3933 3934 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3935 } 3936 3937 static bool netdev_xmit_txqueue_skipped(void) 3938 { 3939 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3940 } 3941 3942 void netdev_xmit_skip_txqueue(bool skip) 3943 { 3944 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3945 } 3946 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3947 #endif /* CONFIG_NET_EGRESS */ 3948 3949 #ifdef CONFIG_XPS 3950 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3951 struct xps_dev_maps *dev_maps, unsigned int tci) 3952 { 3953 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3954 struct xps_map *map; 3955 int queue_index = -1; 3956 3957 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3958 return queue_index; 3959 3960 tci *= dev_maps->num_tc; 3961 tci += tc; 3962 3963 map = rcu_dereference(dev_maps->attr_map[tci]); 3964 if (map) { 3965 if (map->len == 1) 3966 queue_index = map->queues[0]; 3967 else 3968 queue_index = map->queues[reciprocal_scale( 3969 skb_get_hash(skb), map->len)]; 3970 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3971 queue_index = -1; 3972 } 3973 return queue_index; 3974 } 3975 #endif 3976 3977 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3978 struct sk_buff *skb) 3979 { 3980 #ifdef CONFIG_XPS 3981 struct xps_dev_maps *dev_maps; 3982 struct sock *sk = skb->sk; 3983 int queue_index = -1; 3984 3985 if (!static_key_false(&xps_needed)) 3986 return -1; 3987 3988 rcu_read_lock(); 3989 if (!static_key_false(&xps_rxqs_needed)) 3990 goto get_cpus_map; 3991 3992 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3993 if (dev_maps) { 3994 int tci = sk_rx_queue_get(sk); 3995 3996 if (tci >= 0) 3997 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3998 tci); 3999 } 4000 4001 get_cpus_map: 4002 if (queue_index < 0) { 4003 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4004 if (dev_maps) { 4005 unsigned int tci = skb->sender_cpu - 1; 4006 4007 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4008 tci); 4009 } 4010 } 4011 rcu_read_unlock(); 4012 4013 return queue_index; 4014 #else 4015 return -1; 4016 #endif 4017 } 4018 4019 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4020 struct net_device *sb_dev) 4021 { 4022 return 0; 4023 } 4024 EXPORT_SYMBOL(dev_pick_tx_zero); 4025 4026 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4027 struct net_device *sb_dev) 4028 { 4029 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4030 } 4031 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4032 4033 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4034 struct net_device *sb_dev) 4035 { 4036 struct sock *sk = skb->sk; 4037 int queue_index = sk_tx_queue_get(sk); 4038 4039 sb_dev = sb_dev ? : dev; 4040 4041 if (queue_index < 0 || skb->ooo_okay || 4042 queue_index >= dev->real_num_tx_queues) { 4043 int new_index = get_xps_queue(dev, sb_dev, skb); 4044 4045 if (new_index < 0) 4046 new_index = skb_tx_hash(dev, sb_dev, skb); 4047 4048 if (queue_index != new_index && sk && 4049 sk_fullsock(sk) && 4050 rcu_access_pointer(sk->sk_dst_cache)) 4051 sk_tx_queue_set(sk, new_index); 4052 4053 queue_index = new_index; 4054 } 4055 4056 return queue_index; 4057 } 4058 EXPORT_SYMBOL(netdev_pick_tx); 4059 4060 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4061 struct sk_buff *skb, 4062 struct net_device *sb_dev) 4063 { 4064 int queue_index = 0; 4065 4066 #ifdef CONFIG_XPS 4067 u32 sender_cpu = skb->sender_cpu - 1; 4068 4069 if (sender_cpu >= (u32)NR_CPUS) 4070 skb->sender_cpu = raw_smp_processor_id() + 1; 4071 #endif 4072 4073 if (dev->real_num_tx_queues != 1) { 4074 const struct net_device_ops *ops = dev->netdev_ops; 4075 4076 if (ops->ndo_select_queue) 4077 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4078 else 4079 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4080 4081 queue_index = netdev_cap_txqueue(dev, queue_index); 4082 } 4083 4084 skb_set_queue_mapping(skb, queue_index); 4085 return netdev_get_tx_queue(dev, queue_index); 4086 } 4087 4088 /** 4089 * __dev_queue_xmit - transmit a buffer 4090 * @skb: buffer to transmit 4091 * @sb_dev: suboordinate device used for L2 forwarding offload 4092 * 4093 * Queue a buffer for transmission to a network device. The caller must 4094 * have set the device and priority and built the buffer before calling 4095 * this function. The function can be called from an interrupt. 4096 * 4097 * A negative errno code is returned on a failure. A success does not 4098 * guarantee the frame will be transmitted as it may be dropped due 4099 * to congestion or traffic shaping. 4100 * 4101 * ----------------------------------------------------------------------------------- 4102 * I notice this method can also return errors from the queue disciplines, 4103 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4104 * be positive. 4105 * 4106 * Regardless of the return value, the skb is consumed, so it is currently 4107 * difficult to retry a send to this method. (You can bump the ref count 4108 * before sending to hold a reference for retry if you are careful.) 4109 * 4110 * When calling this method, interrupts MUST be enabled. This is because 4111 * the BH enable code must have IRQs enabled so that it will not deadlock. 4112 * --BLG 4113 */ 4114 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4115 { 4116 struct net_device *dev = skb->dev; 4117 struct netdev_queue *txq = NULL; 4118 struct Qdisc *q; 4119 int rc = -ENOMEM; 4120 bool again = false; 4121 4122 skb_reset_mac_header(skb); 4123 4124 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4125 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4126 4127 /* Disable soft irqs for various locks below. Also 4128 * stops preemption for RCU. 4129 */ 4130 rcu_read_lock_bh(); 4131 4132 skb_update_prio(skb); 4133 4134 qdisc_pkt_len_init(skb); 4135 #ifdef CONFIG_NET_CLS_ACT 4136 skb->tc_at_ingress = 0; 4137 #endif 4138 #ifdef CONFIG_NET_EGRESS 4139 if (static_branch_unlikely(&egress_needed_key)) { 4140 if (nf_hook_egress_active()) { 4141 skb = nf_hook_egress(skb, &rc, dev); 4142 if (!skb) 4143 goto out; 4144 } 4145 4146 netdev_xmit_skip_txqueue(false); 4147 4148 nf_skip_egress(skb, true); 4149 skb = sch_handle_egress(skb, &rc, dev); 4150 if (!skb) 4151 goto out; 4152 nf_skip_egress(skb, false); 4153 4154 if (netdev_xmit_txqueue_skipped()) 4155 txq = netdev_tx_queue_mapping(dev, skb); 4156 } 4157 #endif 4158 /* If device/qdisc don't need skb->dst, release it right now while 4159 * its hot in this cpu cache. 4160 */ 4161 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4162 skb_dst_drop(skb); 4163 else 4164 skb_dst_force(skb); 4165 4166 if (!txq) 4167 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4168 4169 q = rcu_dereference_bh(txq->qdisc); 4170 4171 trace_net_dev_queue(skb); 4172 if (q->enqueue) { 4173 rc = __dev_xmit_skb(skb, q, dev, txq); 4174 goto out; 4175 } 4176 4177 /* The device has no queue. Common case for software devices: 4178 * loopback, all the sorts of tunnels... 4179 4180 * Really, it is unlikely that netif_tx_lock protection is necessary 4181 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4182 * counters.) 4183 * However, it is possible, that they rely on protection 4184 * made by us here. 4185 4186 * Check this and shot the lock. It is not prone from deadlocks. 4187 *Either shot noqueue qdisc, it is even simpler 8) 4188 */ 4189 if (dev->flags & IFF_UP) { 4190 int cpu = smp_processor_id(); /* ok because BHs are off */ 4191 4192 /* Other cpus might concurrently change txq->xmit_lock_owner 4193 * to -1 or to their cpu id, but not to our id. 4194 */ 4195 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4196 if (dev_xmit_recursion()) 4197 goto recursion_alert; 4198 4199 skb = validate_xmit_skb(skb, dev, &again); 4200 if (!skb) 4201 goto out; 4202 4203 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4204 HARD_TX_LOCK(dev, txq, cpu); 4205 4206 if (!netif_xmit_stopped(txq)) { 4207 dev_xmit_recursion_inc(); 4208 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4209 dev_xmit_recursion_dec(); 4210 if (dev_xmit_complete(rc)) { 4211 HARD_TX_UNLOCK(dev, txq); 4212 goto out; 4213 } 4214 } 4215 HARD_TX_UNLOCK(dev, txq); 4216 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4217 dev->name); 4218 } else { 4219 /* Recursion is detected! It is possible, 4220 * unfortunately 4221 */ 4222 recursion_alert: 4223 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4224 dev->name); 4225 } 4226 } 4227 4228 rc = -ENETDOWN; 4229 rcu_read_unlock_bh(); 4230 4231 dev_core_stats_tx_dropped_inc(dev); 4232 kfree_skb_list(skb); 4233 return rc; 4234 out: 4235 rcu_read_unlock_bh(); 4236 return rc; 4237 } 4238 4239 int dev_queue_xmit(struct sk_buff *skb) 4240 { 4241 return __dev_queue_xmit(skb, NULL); 4242 } 4243 EXPORT_SYMBOL(dev_queue_xmit); 4244 4245 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4246 { 4247 return __dev_queue_xmit(skb, sb_dev); 4248 } 4249 EXPORT_SYMBOL(dev_queue_xmit_accel); 4250 4251 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4252 { 4253 struct net_device *dev = skb->dev; 4254 struct sk_buff *orig_skb = skb; 4255 struct netdev_queue *txq; 4256 int ret = NETDEV_TX_BUSY; 4257 bool again = false; 4258 4259 if (unlikely(!netif_running(dev) || 4260 !netif_carrier_ok(dev))) 4261 goto drop; 4262 4263 skb = validate_xmit_skb_list(skb, dev, &again); 4264 if (skb != orig_skb) 4265 goto drop; 4266 4267 skb_set_queue_mapping(skb, queue_id); 4268 txq = skb_get_tx_queue(dev, skb); 4269 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4270 4271 local_bh_disable(); 4272 4273 dev_xmit_recursion_inc(); 4274 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4275 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4276 ret = netdev_start_xmit(skb, dev, txq, false); 4277 HARD_TX_UNLOCK(dev, txq); 4278 dev_xmit_recursion_dec(); 4279 4280 local_bh_enable(); 4281 return ret; 4282 drop: 4283 dev_core_stats_tx_dropped_inc(dev); 4284 kfree_skb_list(skb); 4285 return NET_XMIT_DROP; 4286 } 4287 EXPORT_SYMBOL(__dev_direct_xmit); 4288 4289 /************************************************************************* 4290 * Receiver routines 4291 *************************************************************************/ 4292 4293 int netdev_max_backlog __read_mostly = 1000; 4294 EXPORT_SYMBOL(netdev_max_backlog); 4295 4296 int netdev_tstamp_prequeue __read_mostly = 1; 4297 int netdev_budget __read_mostly = 300; 4298 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4299 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4300 int weight_p __read_mostly = 64; /* old backlog weight */ 4301 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4302 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4303 int dev_rx_weight __read_mostly = 64; 4304 int dev_tx_weight __read_mostly = 64; 4305 4306 /* Called with irq disabled */ 4307 static inline void ____napi_schedule(struct softnet_data *sd, 4308 struct napi_struct *napi) 4309 { 4310 struct task_struct *thread; 4311 4312 lockdep_assert_irqs_disabled(); 4313 4314 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4315 /* Paired with smp_mb__before_atomic() in 4316 * napi_enable()/dev_set_threaded(). 4317 * Use READ_ONCE() to guarantee a complete 4318 * read on napi->thread. Only call 4319 * wake_up_process() when it's not NULL. 4320 */ 4321 thread = READ_ONCE(napi->thread); 4322 if (thread) { 4323 /* Avoid doing set_bit() if the thread is in 4324 * INTERRUPTIBLE state, cause napi_thread_wait() 4325 * makes sure to proceed with napi polling 4326 * if the thread is explicitly woken from here. 4327 */ 4328 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4329 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4330 wake_up_process(thread); 4331 return; 4332 } 4333 } 4334 4335 list_add_tail(&napi->poll_list, &sd->poll_list); 4336 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4337 } 4338 4339 #ifdef CONFIG_RPS 4340 4341 /* One global table that all flow-based protocols share. */ 4342 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4343 EXPORT_SYMBOL(rps_sock_flow_table); 4344 u32 rps_cpu_mask __read_mostly; 4345 EXPORT_SYMBOL(rps_cpu_mask); 4346 4347 struct static_key_false rps_needed __read_mostly; 4348 EXPORT_SYMBOL(rps_needed); 4349 struct static_key_false rfs_needed __read_mostly; 4350 EXPORT_SYMBOL(rfs_needed); 4351 4352 static struct rps_dev_flow * 4353 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4354 struct rps_dev_flow *rflow, u16 next_cpu) 4355 { 4356 if (next_cpu < nr_cpu_ids) { 4357 #ifdef CONFIG_RFS_ACCEL 4358 struct netdev_rx_queue *rxqueue; 4359 struct rps_dev_flow_table *flow_table; 4360 struct rps_dev_flow *old_rflow; 4361 u32 flow_id; 4362 u16 rxq_index; 4363 int rc; 4364 4365 /* Should we steer this flow to a different hardware queue? */ 4366 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4367 !(dev->features & NETIF_F_NTUPLE)) 4368 goto out; 4369 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4370 if (rxq_index == skb_get_rx_queue(skb)) 4371 goto out; 4372 4373 rxqueue = dev->_rx + rxq_index; 4374 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4375 if (!flow_table) 4376 goto out; 4377 flow_id = skb_get_hash(skb) & flow_table->mask; 4378 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4379 rxq_index, flow_id); 4380 if (rc < 0) 4381 goto out; 4382 old_rflow = rflow; 4383 rflow = &flow_table->flows[flow_id]; 4384 rflow->filter = rc; 4385 if (old_rflow->filter == rflow->filter) 4386 old_rflow->filter = RPS_NO_FILTER; 4387 out: 4388 #endif 4389 rflow->last_qtail = 4390 per_cpu(softnet_data, next_cpu).input_queue_head; 4391 } 4392 4393 rflow->cpu = next_cpu; 4394 return rflow; 4395 } 4396 4397 /* 4398 * get_rps_cpu is called from netif_receive_skb and returns the target 4399 * CPU from the RPS map of the receiving queue for a given skb. 4400 * rcu_read_lock must be held on entry. 4401 */ 4402 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4403 struct rps_dev_flow **rflowp) 4404 { 4405 const struct rps_sock_flow_table *sock_flow_table; 4406 struct netdev_rx_queue *rxqueue = dev->_rx; 4407 struct rps_dev_flow_table *flow_table; 4408 struct rps_map *map; 4409 int cpu = -1; 4410 u32 tcpu; 4411 u32 hash; 4412 4413 if (skb_rx_queue_recorded(skb)) { 4414 u16 index = skb_get_rx_queue(skb); 4415 4416 if (unlikely(index >= dev->real_num_rx_queues)) { 4417 WARN_ONCE(dev->real_num_rx_queues > 1, 4418 "%s received packet on queue %u, but number " 4419 "of RX queues is %u\n", 4420 dev->name, index, dev->real_num_rx_queues); 4421 goto done; 4422 } 4423 rxqueue += index; 4424 } 4425 4426 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4427 4428 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4429 map = rcu_dereference(rxqueue->rps_map); 4430 if (!flow_table && !map) 4431 goto done; 4432 4433 skb_reset_network_header(skb); 4434 hash = skb_get_hash(skb); 4435 if (!hash) 4436 goto done; 4437 4438 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4439 if (flow_table && sock_flow_table) { 4440 struct rps_dev_flow *rflow; 4441 u32 next_cpu; 4442 u32 ident; 4443 4444 /* First check into global flow table if there is a match */ 4445 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4446 if ((ident ^ hash) & ~rps_cpu_mask) 4447 goto try_rps; 4448 4449 next_cpu = ident & rps_cpu_mask; 4450 4451 /* OK, now we know there is a match, 4452 * we can look at the local (per receive queue) flow table 4453 */ 4454 rflow = &flow_table->flows[hash & flow_table->mask]; 4455 tcpu = rflow->cpu; 4456 4457 /* 4458 * If the desired CPU (where last recvmsg was done) is 4459 * different from current CPU (one in the rx-queue flow 4460 * table entry), switch if one of the following holds: 4461 * - Current CPU is unset (>= nr_cpu_ids). 4462 * - Current CPU is offline. 4463 * - The current CPU's queue tail has advanced beyond the 4464 * last packet that was enqueued using this table entry. 4465 * This guarantees that all previous packets for the flow 4466 * have been dequeued, thus preserving in order delivery. 4467 */ 4468 if (unlikely(tcpu != next_cpu) && 4469 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4470 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4471 rflow->last_qtail)) >= 0)) { 4472 tcpu = next_cpu; 4473 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4474 } 4475 4476 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4477 *rflowp = rflow; 4478 cpu = tcpu; 4479 goto done; 4480 } 4481 } 4482 4483 try_rps: 4484 4485 if (map) { 4486 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4487 if (cpu_online(tcpu)) { 4488 cpu = tcpu; 4489 goto done; 4490 } 4491 } 4492 4493 done: 4494 return cpu; 4495 } 4496 4497 #ifdef CONFIG_RFS_ACCEL 4498 4499 /** 4500 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4501 * @dev: Device on which the filter was set 4502 * @rxq_index: RX queue index 4503 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4504 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4505 * 4506 * Drivers that implement ndo_rx_flow_steer() should periodically call 4507 * this function for each installed filter and remove the filters for 4508 * which it returns %true. 4509 */ 4510 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4511 u32 flow_id, u16 filter_id) 4512 { 4513 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4514 struct rps_dev_flow_table *flow_table; 4515 struct rps_dev_flow *rflow; 4516 bool expire = true; 4517 unsigned int cpu; 4518 4519 rcu_read_lock(); 4520 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4521 if (flow_table && flow_id <= flow_table->mask) { 4522 rflow = &flow_table->flows[flow_id]; 4523 cpu = READ_ONCE(rflow->cpu); 4524 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4525 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4526 rflow->last_qtail) < 4527 (int)(10 * flow_table->mask))) 4528 expire = false; 4529 } 4530 rcu_read_unlock(); 4531 return expire; 4532 } 4533 EXPORT_SYMBOL(rps_may_expire_flow); 4534 4535 #endif /* CONFIG_RFS_ACCEL */ 4536 4537 /* Called from hardirq (IPI) context */ 4538 static void rps_trigger_softirq(void *data) 4539 { 4540 struct softnet_data *sd = data; 4541 4542 ____napi_schedule(sd, &sd->backlog); 4543 sd->received_rps++; 4544 } 4545 4546 #endif /* CONFIG_RPS */ 4547 4548 /* Called from hardirq (IPI) context */ 4549 static void trigger_rx_softirq(void *data __always_unused) 4550 { 4551 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4552 } 4553 4554 /* 4555 * Check if this softnet_data structure is another cpu one 4556 * If yes, queue it to our IPI list and return 1 4557 * If no, return 0 4558 */ 4559 static int napi_schedule_rps(struct softnet_data *sd) 4560 { 4561 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4562 4563 #ifdef CONFIG_RPS 4564 if (sd != mysd) { 4565 sd->rps_ipi_next = mysd->rps_ipi_list; 4566 mysd->rps_ipi_list = sd; 4567 4568 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4569 return 1; 4570 } 4571 #endif /* CONFIG_RPS */ 4572 __napi_schedule_irqoff(&mysd->backlog); 4573 return 0; 4574 } 4575 4576 #ifdef CONFIG_NET_FLOW_LIMIT 4577 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4578 #endif 4579 4580 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4581 { 4582 #ifdef CONFIG_NET_FLOW_LIMIT 4583 struct sd_flow_limit *fl; 4584 struct softnet_data *sd; 4585 unsigned int old_flow, new_flow; 4586 4587 if (qlen < (netdev_max_backlog >> 1)) 4588 return false; 4589 4590 sd = this_cpu_ptr(&softnet_data); 4591 4592 rcu_read_lock(); 4593 fl = rcu_dereference(sd->flow_limit); 4594 if (fl) { 4595 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4596 old_flow = fl->history[fl->history_head]; 4597 fl->history[fl->history_head] = new_flow; 4598 4599 fl->history_head++; 4600 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4601 4602 if (likely(fl->buckets[old_flow])) 4603 fl->buckets[old_flow]--; 4604 4605 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4606 fl->count++; 4607 rcu_read_unlock(); 4608 return true; 4609 } 4610 } 4611 rcu_read_unlock(); 4612 #endif 4613 return false; 4614 } 4615 4616 /* 4617 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4618 * queue (may be a remote CPU queue). 4619 */ 4620 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4621 unsigned int *qtail) 4622 { 4623 enum skb_drop_reason reason; 4624 struct softnet_data *sd; 4625 unsigned long flags; 4626 unsigned int qlen; 4627 4628 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4629 sd = &per_cpu(softnet_data, cpu); 4630 4631 rps_lock_irqsave(sd, &flags); 4632 if (!netif_running(skb->dev)) 4633 goto drop; 4634 qlen = skb_queue_len(&sd->input_pkt_queue); 4635 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4636 if (qlen) { 4637 enqueue: 4638 __skb_queue_tail(&sd->input_pkt_queue, skb); 4639 input_queue_tail_incr_save(sd, qtail); 4640 rps_unlock_irq_restore(sd, &flags); 4641 return NET_RX_SUCCESS; 4642 } 4643 4644 /* Schedule NAPI for backlog device 4645 * We can use non atomic operation since we own the queue lock 4646 */ 4647 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4648 napi_schedule_rps(sd); 4649 goto enqueue; 4650 } 4651 reason = SKB_DROP_REASON_CPU_BACKLOG; 4652 4653 drop: 4654 sd->dropped++; 4655 rps_unlock_irq_restore(sd, &flags); 4656 4657 dev_core_stats_rx_dropped_inc(skb->dev); 4658 kfree_skb_reason(skb, reason); 4659 return NET_RX_DROP; 4660 } 4661 4662 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4663 { 4664 struct net_device *dev = skb->dev; 4665 struct netdev_rx_queue *rxqueue; 4666 4667 rxqueue = dev->_rx; 4668 4669 if (skb_rx_queue_recorded(skb)) { 4670 u16 index = skb_get_rx_queue(skb); 4671 4672 if (unlikely(index >= dev->real_num_rx_queues)) { 4673 WARN_ONCE(dev->real_num_rx_queues > 1, 4674 "%s received packet on queue %u, but number " 4675 "of RX queues is %u\n", 4676 dev->name, index, dev->real_num_rx_queues); 4677 4678 return rxqueue; /* Return first rxqueue */ 4679 } 4680 rxqueue += index; 4681 } 4682 return rxqueue; 4683 } 4684 4685 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4686 struct bpf_prog *xdp_prog) 4687 { 4688 void *orig_data, *orig_data_end, *hard_start; 4689 struct netdev_rx_queue *rxqueue; 4690 bool orig_bcast, orig_host; 4691 u32 mac_len, frame_sz; 4692 __be16 orig_eth_type; 4693 struct ethhdr *eth; 4694 u32 metalen, act; 4695 int off; 4696 4697 /* The XDP program wants to see the packet starting at the MAC 4698 * header. 4699 */ 4700 mac_len = skb->data - skb_mac_header(skb); 4701 hard_start = skb->data - skb_headroom(skb); 4702 4703 /* SKB "head" area always have tailroom for skb_shared_info */ 4704 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4705 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4706 4707 rxqueue = netif_get_rxqueue(skb); 4708 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4709 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4710 skb_headlen(skb) + mac_len, true); 4711 4712 orig_data_end = xdp->data_end; 4713 orig_data = xdp->data; 4714 eth = (struct ethhdr *)xdp->data; 4715 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4716 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4717 orig_eth_type = eth->h_proto; 4718 4719 act = bpf_prog_run_xdp(xdp_prog, xdp); 4720 4721 /* check if bpf_xdp_adjust_head was used */ 4722 off = xdp->data - orig_data; 4723 if (off) { 4724 if (off > 0) 4725 __skb_pull(skb, off); 4726 else if (off < 0) 4727 __skb_push(skb, -off); 4728 4729 skb->mac_header += off; 4730 skb_reset_network_header(skb); 4731 } 4732 4733 /* check if bpf_xdp_adjust_tail was used */ 4734 off = xdp->data_end - orig_data_end; 4735 if (off != 0) { 4736 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4737 skb->len += off; /* positive on grow, negative on shrink */ 4738 } 4739 4740 /* check if XDP changed eth hdr such SKB needs update */ 4741 eth = (struct ethhdr *)xdp->data; 4742 if ((orig_eth_type != eth->h_proto) || 4743 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4744 skb->dev->dev_addr)) || 4745 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4746 __skb_push(skb, ETH_HLEN); 4747 skb->pkt_type = PACKET_HOST; 4748 skb->protocol = eth_type_trans(skb, skb->dev); 4749 } 4750 4751 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4752 * before calling us again on redirect path. We do not call do_redirect 4753 * as we leave that up to the caller. 4754 * 4755 * Caller is responsible for managing lifetime of skb (i.e. calling 4756 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4757 */ 4758 switch (act) { 4759 case XDP_REDIRECT: 4760 case XDP_TX: 4761 __skb_push(skb, mac_len); 4762 break; 4763 case XDP_PASS: 4764 metalen = xdp->data - xdp->data_meta; 4765 if (metalen) 4766 skb_metadata_set(skb, metalen); 4767 break; 4768 } 4769 4770 return act; 4771 } 4772 4773 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4774 struct xdp_buff *xdp, 4775 struct bpf_prog *xdp_prog) 4776 { 4777 u32 act = XDP_DROP; 4778 4779 /* Reinjected packets coming from act_mirred or similar should 4780 * not get XDP generic processing. 4781 */ 4782 if (skb_is_redirected(skb)) 4783 return XDP_PASS; 4784 4785 /* XDP packets must be linear and must have sufficient headroom 4786 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4787 * native XDP provides, thus we need to do it here as well. 4788 */ 4789 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4790 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4791 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4792 int troom = skb->tail + skb->data_len - skb->end; 4793 4794 /* In case we have to go down the path and also linearize, 4795 * then lets do the pskb_expand_head() work just once here. 4796 */ 4797 if (pskb_expand_head(skb, 4798 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4799 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4800 goto do_drop; 4801 if (skb_linearize(skb)) 4802 goto do_drop; 4803 } 4804 4805 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4806 switch (act) { 4807 case XDP_REDIRECT: 4808 case XDP_TX: 4809 case XDP_PASS: 4810 break; 4811 default: 4812 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4813 fallthrough; 4814 case XDP_ABORTED: 4815 trace_xdp_exception(skb->dev, xdp_prog, act); 4816 fallthrough; 4817 case XDP_DROP: 4818 do_drop: 4819 kfree_skb(skb); 4820 break; 4821 } 4822 4823 return act; 4824 } 4825 4826 /* When doing generic XDP we have to bypass the qdisc layer and the 4827 * network taps in order to match in-driver-XDP behavior. 4828 */ 4829 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4830 { 4831 struct net_device *dev = skb->dev; 4832 struct netdev_queue *txq; 4833 bool free_skb = true; 4834 int cpu, rc; 4835 4836 txq = netdev_core_pick_tx(dev, skb, NULL); 4837 cpu = smp_processor_id(); 4838 HARD_TX_LOCK(dev, txq, cpu); 4839 if (!netif_xmit_stopped(txq)) { 4840 rc = netdev_start_xmit(skb, dev, txq, 0); 4841 if (dev_xmit_complete(rc)) 4842 free_skb = false; 4843 } 4844 HARD_TX_UNLOCK(dev, txq); 4845 if (free_skb) { 4846 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4847 kfree_skb(skb); 4848 } 4849 } 4850 4851 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4852 4853 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4854 { 4855 if (xdp_prog) { 4856 struct xdp_buff xdp; 4857 u32 act; 4858 int err; 4859 4860 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4861 if (act != XDP_PASS) { 4862 switch (act) { 4863 case XDP_REDIRECT: 4864 err = xdp_do_generic_redirect(skb->dev, skb, 4865 &xdp, xdp_prog); 4866 if (err) 4867 goto out_redir; 4868 break; 4869 case XDP_TX: 4870 generic_xdp_tx(skb, xdp_prog); 4871 break; 4872 } 4873 return XDP_DROP; 4874 } 4875 } 4876 return XDP_PASS; 4877 out_redir: 4878 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4879 return XDP_DROP; 4880 } 4881 EXPORT_SYMBOL_GPL(do_xdp_generic); 4882 4883 static int netif_rx_internal(struct sk_buff *skb) 4884 { 4885 int ret; 4886 4887 net_timestamp_check(netdev_tstamp_prequeue, skb); 4888 4889 trace_netif_rx(skb); 4890 4891 #ifdef CONFIG_RPS 4892 if (static_branch_unlikely(&rps_needed)) { 4893 struct rps_dev_flow voidflow, *rflow = &voidflow; 4894 int cpu; 4895 4896 rcu_read_lock(); 4897 4898 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4899 if (cpu < 0) 4900 cpu = smp_processor_id(); 4901 4902 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4903 4904 rcu_read_unlock(); 4905 } else 4906 #endif 4907 { 4908 unsigned int qtail; 4909 4910 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4911 } 4912 return ret; 4913 } 4914 4915 /** 4916 * __netif_rx - Slightly optimized version of netif_rx 4917 * @skb: buffer to post 4918 * 4919 * This behaves as netif_rx except that it does not disable bottom halves. 4920 * As a result this function may only be invoked from the interrupt context 4921 * (either hard or soft interrupt). 4922 */ 4923 int __netif_rx(struct sk_buff *skb) 4924 { 4925 int ret; 4926 4927 lockdep_assert_once(hardirq_count() | softirq_count()); 4928 4929 trace_netif_rx_entry(skb); 4930 ret = netif_rx_internal(skb); 4931 trace_netif_rx_exit(ret); 4932 return ret; 4933 } 4934 EXPORT_SYMBOL(__netif_rx); 4935 4936 /** 4937 * netif_rx - post buffer to the network code 4938 * @skb: buffer to post 4939 * 4940 * This function receives a packet from a device driver and queues it for 4941 * the upper (protocol) levels to process via the backlog NAPI device. It 4942 * always succeeds. The buffer may be dropped during processing for 4943 * congestion control or by the protocol layers. 4944 * The network buffer is passed via the backlog NAPI device. Modern NIC 4945 * driver should use NAPI and GRO. 4946 * This function can used from interrupt and from process context. The 4947 * caller from process context must not disable interrupts before invoking 4948 * this function. 4949 * 4950 * return values: 4951 * NET_RX_SUCCESS (no congestion) 4952 * NET_RX_DROP (packet was dropped) 4953 * 4954 */ 4955 int netif_rx(struct sk_buff *skb) 4956 { 4957 bool need_bh_off = !(hardirq_count() | softirq_count()); 4958 int ret; 4959 4960 if (need_bh_off) 4961 local_bh_disable(); 4962 trace_netif_rx_entry(skb); 4963 ret = netif_rx_internal(skb); 4964 trace_netif_rx_exit(ret); 4965 if (need_bh_off) 4966 local_bh_enable(); 4967 return ret; 4968 } 4969 EXPORT_SYMBOL(netif_rx); 4970 4971 static __latent_entropy void net_tx_action(struct softirq_action *h) 4972 { 4973 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4974 4975 if (sd->completion_queue) { 4976 struct sk_buff *clist; 4977 4978 local_irq_disable(); 4979 clist = sd->completion_queue; 4980 sd->completion_queue = NULL; 4981 local_irq_enable(); 4982 4983 while (clist) { 4984 struct sk_buff *skb = clist; 4985 4986 clist = clist->next; 4987 4988 WARN_ON(refcount_read(&skb->users)); 4989 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4990 trace_consume_skb(skb); 4991 else 4992 trace_kfree_skb(skb, net_tx_action, 4993 SKB_DROP_REASON_NOT_SPECIFIED); 4994 4995 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4996 __kfree_skb(skb); 4997 else 4998 __kfree_skb_defer(skb); 4999 } 5000 } 5001 5002 if (sd->output_queue) { 5003 struct Qdisc *head; 5004 5005 local_irq_disable(); 5006 head = sd->output_queue; 5007 sd->output_queue = NULL; 5008 sd->output_queue_tailp = &sd->output_queue; 5009 local_irq_enable(); 5010 5011 rcu_read_lock(); 5012 5013 while (head) { 5014 struct Qdisc *q = head; 5015 spinlock_t *root_lock = NULL; 5016 5017 head = head->next_sched; 5018 5019 /* We need to make sure head->next_sched is read 5020 * before clearing __QDISC_STATE_SCHED 5021 */ 5022 smp_mb__before_atomic(); 5023 5024 if (!(q->flags & TCQ_F_NOLOCK)) { 5025 root_lock = qdisc_lock(q); 5026 spin_lock(root_lock); 5027 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5028 &q->state))) { 5029 /* There is a synchronize_net() between 5030 * STATE_DEACTIVATED flag being set and 5031 * qdisc_reset()/some_qdisc_is_busy() in 5032 * dev_deactivate(), so we can safely bail out 5033 * early here to avoid data race between 5034 * qdisc_deactivate() and some_qdisc_is_busy() 5035 * for lockless qdisc. 5036 */ 5037 clear_bit(__QDISC_STATE_SCHED, &q->state); 5038 continue; 5039 } 5040 5041 clear_bit(__QDISC_STATE_SCHED, &q->state); 5042 qdisc_run(q); 5043 if (root_lock) 5044 spin_unlock(root_lock); 5045 } 5046 5047 rcu_read_unlock(); 5048 } 5049 5050 xfrm_dev_backlog(sd); 5051 } 5052 5053 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5054 /* This hook is defined here for ATM LANE */ 5055 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5056 unsigned char *addr) __read_mostly; 5057 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5058 #endif 5059 5060 static inline struct sk_buff * 5061 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5062 struct net_device *orig_dev, bool *another) 5063 { 5064 #ifdef CONFIG_NET_CLS_ACT 5065 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5066 struct tcf_result cl_res; 5067 5068 /* If there's at least one ingress present somewhere (so 5069 * we get here via enabled static key), remaining devices 5070 * that are not configured with an ingress qdisc will bail 5071 * out here. 5072 */ 5073 if (!miniq) 5074 return skb; 5075 5076 if (*pt_prev) { 5077 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5078 *pt_prev = NULL; 5079 } 5080 5081 qdisc_skb_cb(skb)->pkt_len = skb->len; 5082 tc_skb_cb(skb)->mru = 0; 5083 tc_skb_cb(skb)->post_ct = false; 5084 skb->tc_at_ingress = 1; 5085 mini_qdisc_bstats_cpu_update(miniq, skb); 5086 5087 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5088 case TC_ACT_OK: 5089 case TC_ACT_RECLASSIFY: 5090 skb->tc_index = TC_H_MIN(cl_res.classid); 5091 break; 5092 case TC_ACT_SHOT: 5093 mini_qdisc_qstats_cpu_drop(miniq); 5094 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5095 return NULL; 5096 case TC_ACT_STOLEN: 5097 case TC_ACT_QUEUED: 5098 case TC_ACT_TRAP: 5099 consume_skb(skb); 5100 return NULL; 5101 case TC_ACT_REDIRECT: 5102 /* skb_mac_header check was done by cls/act_bpf, so 5103 * we can safely push the L2 header back before 5104 * redirecting to another netdev 5105 */ 5106 __skb_push(skb, skb->mac_len); 5107 if (skb_do_redirect(skb) == -EAGAIN) { 5108 __skb_pull(skb, skb->mac_len); 5109 *another = true; 5110 break; 5111 } 5112 return NULL; 5113 case TC_ACT_CONSUMED: 5114 return NULL; 5115 default: 5116 break; 5117 } 5118 #endif /* CONFIG_NET_CLS_ACT */ 5119 return skb; 5120 } 5121 5122 /** 5123 * netdev_is_rx_handler_busy - check if receive handler is registered 5124 * @dev: device to check 5125 * 5126 * Check if a receive handler is already registered for a given device. 5127 * Return true if there one. 5128 * 5129 * The caller must hold the rtnl_mutex. 5130 */ 5131 bool netdev_is_rx_handler_busy(struct net_device *dev) 5132 { 5133 ASSERT_RTNL(); 5134 return dev && rtnl_dereference(dev->rx_handler); 5135 } 5136 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5137 5138 /** 5139 * netdev_rx_handler_register - register receive handler 5140 * @dev: device to register a handler for 5141 * @rx_handler: receive handler to register 5142 * @rx_handler_data: data pointer that is used by rx handler 5143 * 5144 * Register a receive handler for a device. This handler will then be 5145 * called from __netif_receive_skb. A negative errno code is returned 5146 * on a failure. 5147 * 5148 * The caller must hold the rtnl_mutex. 5149 * 5150 * For a general description of rx_handler, see enum rx_handler_result. 5151 */ 5152 int netdev_rx_handler_register(struct net_device *dev, 5153 rx_handler_func_t *rx_handler, 5154 void *rx_handler_data) 5155 { 5156 if (netdev_is_rx_handler_busy(dev)) 5157 return -EBUSY; 5158 5159 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5160 return -EINVAL; 5161 5162 /* Note: rx_handler_data must be set before rx_handler */ 5163 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5164 rcu_assign_pointer(dev->rx_handler, rx_handler); 5165 5166 return 0; 5167 } 5168 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5169 5170 /** 5171 * netdev_rx_handler_unregister - unregister receive handler 5172 * @dev: device to unregister a handler from 5173 * 5174 * Unregister a receive handler from a device. 5175 * 5176 * The caller must hold the rtnl_mutex. 5177 */ 5178 void netdev_rx_handler_unregister(struct net_device *dev) 5179 { 5180 5181 ASSERT_RTNL(); 5182 RCU_INIT_POINTER(dev->rx_handler, NULL); 5183 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5184 * section has a guarantee to see a non NULL rx_handler_data 5185 * as well. 5186 */ 5187 synchronize_net(); 5188 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5189 } 5190 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5191 5192 /* 5193 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5194 * the special handling of PFMEMALLOC skbs. 5195 */ 5196 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5197 { 5198 switch (skb->protocol) { 5199 case htons(ETH_P_ARP): 5200 case htons(ETH_P_IP): 5201 case htons(ETH_P_IPV6): 5202 case htons(ETH_P_8021Q): 5203 case htons(ETH_P_8021AD): 5204 return true; 5205 default: 5206 return false; 5207 } 5208 } 5209 5210 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5211 int *ret, struct net_device *orig_dev) 5212 { 5213 if (nf_hook_ingress_active(skb)) { 5214 int ingress_retval; 5215 5216 if (*pt_prev) { 5217 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5218 *pt_prev = NULL; 5219 } 5220 5221 rcu_read_lock(); 5222 ingress_retval = nf_hook_ingress(skb); 5223 rcu_read_unlock(); 5224 return ingress_retval; 5225 } 5226 return 0; 5227 } 5228 5229 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5230 struct packet_type **ppt_prev) 5231 { 5232 struct packet_type *ptype, *pt_prev; 5233 rx_handler_func_t *rx_handler; 5234 struct sk_buff *skb = *pskb; 5235 struct net_device *orig_dev; 5236 bool deliver_exact = false; 5237 int ret = NET_RX_DROP; 5238 __be16 type; 5239 5240 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5241 5242 trace_netif_receive_skb(skb); 5243 5244 orig_dev = skb->dev; 5245 5246 skb_reset_network_header(skb); 5247 if (!skb_transport_header_was_set(skb)) 5248 skb_reset_transport_header(skb); 5249 skb_reset_mac_len(skb); 5250 5251 pt_prev = NULL; 5252 5253 another_round: 5254 skb->skb_iif = skb->dev->ifindex; 5255 5256 __this_cpu_inc(softnet_data.processed); 5257 5258 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5259 int ret2; 5260 5261 migrate_disable(); 5262 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5263 migrate_enable(); 5264 5265 if (ret2 != XDP_PASS) { 5266 ret = NET_RX_DROP; 5267 goto out; 5268 } 5269 } 5270 5271 if (eth_type_vlan(skb->protocol)) { 5272 skb = skb_vlan_untag(skb); 5273 if (unlikely(!skb)) 5274 goto out; 5275 } 5276 5277 if (skb_skip_tc_classify(skb)) 5278 goto skip_classify; 5279 5280 if (pfmemalloc) 5281 goto skip_taps; 5282 5283 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5284 if (pt_prev) 5285 ret = deliver_skb(skb, pt_prev, orig_dev); 5286 pt_prev = ptype; 5287 } 5288 5289 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5290 if (pt_prev) 5291 ret = deliver_skb(skb, pt_prev, orig_dev); 5292 pt_prev = ptype; 5293 } 5294 5295 skip_taps: 5296 #ifdef CONFIG_NET_INGRESS 5297 if (static_branch_unlikely(&ingress_needed_key)) { 5298 bool another = false; 5299 5300 nf_skip_egress(skb, true); 5301 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5302 &another); 5303 if (another) 5304 goto another_round; 5305 if (!skb) 5306 goto out; 5307 5308 nf_skip_egress(skb, false); 5309 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5310 goto out; 5311 } 5312 #endif 5313 skb_reset_redirect(skb); 5314 skip_classify: 5315 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5316 goto drop; 5317 5318 if (skb_vlan_tag_present(skb)) { 5319 if (pt_prev) { 5320 ret = deliver_skb(skb, pt_prev, orig_dev); 5321 pt_prev = NULL; 5322 } 5323 if (vlan_do_receive(&skb)) 5324 goto another_round; 5325 else if (unlikely(!skb)) 5326 goto out; 5327 } 5328 5329 rx_handler = rcu_dereference(skb->dev->rx_handler); 5330 if (rx_handler) { 5331 if (pt_prev) { 5332 ret = deliver_skb(skb, pt_prev, orig_dev); 5333 pt_prev = NULL; 5334 } 5335 switch (rx_handler(&skb)) { 5336 case RX_HANDLER_CONSUMED: 5337 ret = NET_RX_SUCCESS; 5338 goto out; 5339 case RX_HANDLER_ANOTHER: 5340 goto another_round; 5341 case RX_HANDLER_EXACT: 5342 deliver_exact = true; 5343 break; 5344 case RX_HANDLER_PASS: 5345 break; 5346 default: 5347 BUG(); 5348 } 5349 } 5350 5351 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5352 check_vlan_id: 5353 if (skb_vlan_tag_get_id(skb)) { 5354 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5355 * find vlan device. 5356 */ 5357 skb->pkt_type = PACKET_OTHERHOST; 5358 } else if (eth_type_vlan(skb->protocol)) { 5359 /* Outer header is 802.1P with vlan 0, inner header is 5360 * 802.1Q or 802.1AD and vlan_do_receive() above could 5361 * not find vlan dev for vlan id 0. 5362 */ 5363 __vlan_hwaccel_clear_tag(skb); 5364 skb = skb_vlan_untag(skb); 5365 if (unlikely(!skb)) 5366 goto out; 5367 if (vlan_do_receive(&skb)) 5368 /* After stripping off 802.1P header with vlan 0 5369 * vlan dev is found for inner header. 5370 */ 5371 goto another_round; 5372 else if (unlikely(!skb)) 5373 goto out; 5374 else 5375 /* We have stripped outer 802.1P vlan 0 header. 5376 * But could not find vlan dev. 5377 * check again for vlan id to set OTHERHOST. 5378 */ 5379 goto check_vlan_id; 5380 } 5381 /* Note: we might in the future use prio bits 5382 * and set skb->priority like in vlan_do_receive() 5383 * For the time being, just ignore Priority Code Point 5384 */ 5385 __vlan_hwaccel_clear_tag(skb); 5386 } 5387 5388 type = skb->protocol; 5389 5390 /* deliver only exact match when indicated */ 5391 if (likely(!deliver_exact)) { 5392 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5393 &ptype_base[ntohs(type) & 5394 PTYPE_HASH_MASK]); 5395 } 5396 5397 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5398 &orig_dev->ptype_specific); 5399 5400 if (unlikely(skb->dev != orig_dev)) { 5401 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5402 &skb->dev->ptype_specific); 5403 } 5404 5405 if (pt_prev) { 5406 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5407 goto drop; 5408 *ppt_prev = pt_prev; 5409 } else { 5410 drop: 5411 if (!deliver_exact) 5412 dev_core_stats_rx_dropped_inc(skb->dev); 5413 else 5414 dev_core_stats_rx_nohandler_inc(skb->dev); 5415 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5416 /* Jamal, now you will not able to escape explaining 5417 * me how you were going to use this. :-) 5418 */ 5419 ret = NET_RX_DROP; 5420 } 5421 5422 out: 5423 /* The invariant here is that if *ppt_prev is not NULL 5424 * then skb should also be non-NULL. 5425 * 5426 * Apparently *ppt_prev assignment above holds this invariant due to 5427 * skb dereferencing near it. 5428 */ 5429 *pskb = skb; 5430 return ret; 5431 } 5432 5433 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5434 { 5435 struct net_device *orig_dev = skb->dev; 5436 struct packet_type *pt_prev = NULL; 5437 int ret; 5438 5439 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5440 if (pt_prev) 5441 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5442 skb->dev, pt_prev, orig_dev); 5443 return ret; 5444 } 5445 5446 /** 5447 * netif_receive_skb_core - special purpose version of netif_receive_skb 5448 * @skb: buffer to process 5449 * 5450 * More direct receive version of netif_receive_skb(). It should 5451 * only be used by callers that have a need to skip RPS and Generic XDP. 5452 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5453 * 5454 * This function may only be called from softirq context and interrupts 5455 * should be enabled. 5456 * 5457 * Return values (usually ignored): 5458 * NET_RX_SUCCESS: no congestion 5459 * NET_RX_DROP: packet was dropped 5460 */ 5461 int netif_receive_skb_core(struct sk_buff *skb) 5462 { 5463 int ret; 5464 5465 rcu_read_lock(); 5466 ret = __netif_receive_skb_one_core(skb, false); 5467 rcu_read_unlock(); 5468 5469 return ret; 5470 } 5471 EXPORT_SYMBOL(netif_receive_skb_core); 5472 5473 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5474 struct packet_type *pt_prev, 5475 struct net_device *orig_dev) 5476 { 5477 struct sk_buff *skb, *next; 5478 5479 if (!pt_prev) 5480 return; 5481 if (list_empty(head)) 5482 return; 5483 if (pt_prev->list_func != NULL) 5484 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5485 ip_list_rcv, head, pt_prev, orig_dev); 5486 else 5487 list_for_each_entry_safe(skb, next, head, list) { 5488 skb_list_del_init(skb); 5489 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5490 } 5491 } 5492 5493 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5494 { 5495 /* Fast-path assumptions: 5496 * - There is no RX handler. 5497 * - Only one packet_type matches. 5498 * If either of these fails, we will end up doing some per-packet 5499 * processing in-line, then handling the 'last ptype' for the whole 5500 * sublist. This can't cause out-of-order delivery to any single ptype, 5501 * because the 'last ptype' must be constant across the sublist, and all 5502 * other ptypes are handled per-packet. 5503 */ 5504 /* Current (common) ptype of sublist */ 5505 struct packet_type *pt_curr = NULL; 5506 /* Current (common) orig_dev of sublist */ 5507 struct net_device *od_curr = NULL; 5508 struct list_head sublist; 5509 struct sk_buff *skb, *next; 5510 5511 INIT_LIST_HEAD(&sublist); 5512 list_for_each_entry_safe(skb, next, head, list) { 5513 struct net_device *orig_dev = skb->dev; 5514 struct packet_type *pt_prev = NULL; 5515 5516 skb_list_del_init(skb); 5517 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5518 if (!pt_prev) 5519 continue; 5520 if (pt_curr != pt_prev || od_curr != orig_dev) { 5521 /* dispatch old sublist */ 5522 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5523 /* start new sublist */ 5524 INIT_LIST_HEAD(&sublist); 5525 pt_curr = pt_prev; 5526 od_curr = orig_dev; 5527 } 5528 list_add_tail(&skb->list, &sublist); 5529 } 5530 5531 /* dispatch final sublist */ 5532 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5533 } 5534 5535 static int __netif_receive_skb(struct sk_buff *skb) 5536 { 5537 int ret; 5538 5539 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5540 unsigned int noreclaim_flag; 5541 5542 /* 5543 * PFMEMALLOC skbs are special, they should 5544 * - be delivered to SOCK_MEMALLOC sockets only 5545 * - stay away from userspace 5546 * - have bounded memory usage 5547 * 5548 * Use PF_MEMALLOC as this saves us from propagating the allocation 5549 * context down to all allocation sites. 5550 */ 5551 noreclaim_flag = memalloc_noreclaim_save(); 5552 ret = __netif_receive_skb_one_core(skb, true); 5553 memalloc_noreclaim_restore(noreclaim_flag); 5554 } else 5555 ret = __netif_receive_skb_one_core(skb, false); 5556 5557 return ret; 5558 } 5559 5560 static void __netif_receive_skb_list(struct list_head *head) 5561 { 5562 unsigned long noreclaim_flag = 0; 5563 struct sk_buff *skb, *next; 5564 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5565 5566 list_for_each_entry_safe(skb, next, head, list) { 5567 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5568 struct list_head sublist; 5569 5570 /* Handle the previous sublist */ 5571 list_cut_before(&sublist, head, &skb->list); 5572 if (!list_empty(&sublist)) 5573 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5574 pfmemalloc = !pfmemalloc; 5575 /* See comments in __netif_receive_skb */ 5576 if (pfmemalloc) 5577 noreclaim_flag = memalloc_noreclaim_save(); 5578 else 5579 memalloc_noreclaim_restore(noreclaim_flag); 5580 } 5581 } 5582 /* Handle the remaining sublist */ 5583 if (!list_empty(head)) 5584 __netif_receive_skb_list_core(head, pfmemalloc); 5585 /* Restore pflags */ 5586 if (pfmemalloc) 5587 memalloc_noreclaim_restore(noreclaim_flag); 5588 } 5589 5590 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5591 { 5592 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5593 struct bpf_prog *new = xdp->prog; 5594 int ret = 0; 5595 5596 switch (xdp->command) { 5597 case XDP_SETUP_PROG: 5598 rcu_assign_pointer(dev->xdp_prog, new); 5599 if (old) 5600 bpf_prog_put(old); 5601 5602 if (old && !new) { 5603 static_branch_dec(&generic_xdp_needed_key); 5604 } else if (new && !old) { 5605 static_branch_inc(&generic_xdp_needed_key); 5606 dev_disable_lro(dev); 5607 dev_disable_gro_hw(dev); 5608 } 5609 break; 5610 5611 default: 5612 ret = -EINVAL; 5613 break; 5614 } 5615 5616 return ret; 5617 } 5618 5619 static int netif_receive_skb_internal(struct sk_buff *skb) 5620 { 5621 int ret; 5622 5623 net_timestamp_check(netdev_tstamp_prequeue, skb); 5624 5625 if (skb_defer_rx_timestamp(skb)) 5626 return NET_RX_SUCCESS; 5627 5628 rcu_read_lock(); 5629 #ifdef CONFIG_RPS 5630 if (static_branch_unlikely(&rps_needed)) { 5631 struct rps_dev_flow voidflow, *rflow = &voidflow; 5632 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5633 5634 if (cpu >= 0) { 5635 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5636 rcu_read_unlock(); 5637 return ret; 5638 } 5639 } 5640 #endif 5641 ret = __netif_receive_skb(skb); 5642 rcu_read_unlock(); 5643 return ret; 5644 } 5645 5646 void netif_receive_skb_list_internal(struct list_head *head) 5647 { 5648 struct sk_buff *skb, *next; 5649 struct list_head sublist; 5650 5651 INIT_LIST_HEAD(&sublist); 5652 list_for_each_entry_safe(skb, next, head, list) { 5653 net_timestamp_check(netdev_tstamp_prequeue, skb); 5654 skb_list_del_init(skb); 5655 if (!skb_defer_rx_timestamp(skb)) 5656 list_add_tail(&skb->list, &sublist); 5657 } 5658 list_splice_init(&sublist, head); 5659 5660 rcu_read_lock(); 5661 #ifdef CONFIG_RPS 5662 if (static_branch_unlikely(&rps_needed)) { 5663 list_for_each_entry_safe(skb, next, head, list) { 5664 struct rps_dev_flow voidflow, *rflow = &voidflow; 5665 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5666 5667 if (cpu >= 0) { 5668 /* Will be handled, remove from list */ 5669 skb_list_del_init(skb); 5670 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5671 } 5672 } 5673 } 5674 #endif 5675 __netif_receive_skb_list(head); 5676 rcu_read_unlock(); 5677 } 5678 5679 /** 5680 * netif_receive_skb - process receive buffer from network 5681 * @skb: buffer to process 5682 * 5683 * netif_receive_skb() is the main receive data processing function. 5684 * It always succeeds. The buffer may be dropped during processing 5685 * for congestion control or by the protocol layers. 5686 * 5687 * This function may only be called from softirq context and interrupts 5688 * should be enabled. 5689 * 5690 * Return values (usually ignored): 5691 * NET_RX_SUCCESS: no congestion 5692 * NET_RX_DROP: packet was dropped 5693 */ 5694 int netif_receive_skb(struct sk_buff *skb) 5695 { 5696 int ret; 5697 5698 trace_netif_receive_skb_entry(skb); 5699 5700 ret = netif_receive_skb_internal(skb); 5701 trace_netif_receive_skb_exit(ret); 5702 5703 return ret; 5704 } 5705 EXPORT_SYMBOL(netif_receive_skb); 5706 5707 /** 5708 * netif_receive_skb_list - process many receive buffers from network 5709 * @head: list of skbs to process. 5710 * 5711 * Since return value of netif_receive_skb() is normally ignored, and 5712 * wouldn't be meaningful for a list, this function returns void. 5713 * 5714 * This function may only be called from softirq context and interrupts 5715 * should be enabled. 5716 */ 5717 void netif_receive_skb_list(struct list_head *head) 5718 { 5719 struct sk_buff *skb; 5720 5721 if (list_empty(head)) 5722 return; 5723 if (trace_netif_receive_skb_list_entry_enabled()) { 5724 list_for_each_entry(skb, head, list) 5725 trace_netif_receive_skb_list_entry(skb); 5726 } 5727 netif_receive_skb_list_internal(head); 5728 trace_netif_receive_skb_list_exit(0); 5729 } 5730 EXPORT_SYMBOL(netif_receive_skb_list); 5731 5732 static DEFINE_PER_CPU(struct work_struct, flush_works); 5733 5734 /* Network device is going away, flush any packets still pending */ 5735 static void flush_backlog(struct work_struct *work) 5736 { 5737 struct sk_buff *skb, *tmp; 5738 struct softnet_data *sd; 5739 5740 local_bh_disable(); 5741 sd = this_cpu_ptr(&softnet_data); 5742 5743 rps_lock_irq_disable(sd); 5744 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5745 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5746 __skb_unlink(skb, &sd->input_pkt_queue); 5747 dev_kfree_skb_irq(skb); 5748 input_queue_head_incr(sd); 5749 } 5750 } 5751 rps_unlock_irq_enable(sd); 5752 5753 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5754 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5755 __skb_unlink(skb, &sd->process_queue); 5756 kfree_skb(skb); 5757 input_queue_head_incr(sd); 5758 } 5759 } 5760 local_bh_enable(); 5761 } 5762 5763 static bool flush_required(int cpu) 5764 { 5765 #if IS_ENABLED(CONFIG_RPS) 5766 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5767 bool do_flush; 5768 5769 rps_lock_irq_disable(sd); 5770 5771 /* as insertion into process_queue happens with the rps lock held, 5772 * process_queue access may race only with dequeue 5773 */ 5774 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5775 !skb_queue_empty_lockless(&sd->process_queue); 5776 rps_unlock_irq_enable(sd); 5777 5778 return do_flush; 5779 #endif 5780 /* without RPS we can't safely check input_pkt_queue: during a 5781 * concurrent remote skb_queue_splice() we can detect as empty both 5782 * input_pkt_queue and process_queue even if the latter could end-up 5783 * containing a lot of packets. 5784 */ 5785 return true; 5786 } 5787 5788 static void flush_all_backlogs(void) 5789 { 5790 static cpumask_t flush_cpus; 5791 unsigned int cpu; 5792 5793 /* since we are under rtnl lock protection we can use static data 5794 * for the cpumask and avoid allocating on stack the possibly 5795 * large mask 5796 */ 5797 ASSERT_RTNL(); 5798 5799 cpus_read_lock(); 5800 5801 cpumask_clear(&flush_cpus); 5802 for_each_online_cpu(cpu) { 5803 if (flush_required(cpu)) { 5804 queue_work_on(cpu, system_highpri_wq, 5805 per_cpu_ptr(&flush_works, cpu)); 5806 cpumask_set_cpu(cpu, &flush_cpus); 5807 } 5808 } 5809 5810 /* we can have in flight packet[s] on the cpus we are not flushing, 5811 * synchronize_net() in unregister_netdevice_many() will take care of 5812 * them 5813 */ 5814 for_each_cpu(cpu, &flush_cpus) 5815 flush_work(per_cpu_ptr(&flush_works, cpu)); 5816 5817 cpus_read_unlock(); 5818 } 5819 5820 static void net_rps_send_ipi(struct softnet_data *remsd) 5821 { 5822 #ifdef CONFIG_RPS 5823 while (remsd) { 5824 struct softnet_data *next = remsd->rps_ipi_next; 5825 5826 if (cpu_online(remsd->cpu)) 5827 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5828 remsd = next; 5829 } 5830 #endif 5831 } 5832 5833 /* 5834 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5835 * Note: called with local irq disabled, but exits with local irq enabled. 5836 */ 5837 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5838 { 5839 #ifdef CONFIG_RPS 5840 struct softnet_data *remsd = sd->rps_ipi_list; 5841 5842 if (remsd) { 5843 sd->rps_ipi_list = NULL; 5844 5845 local_irq_enable(); 5846 5847 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5848 net_rps_send_ipi(remsd); 5849 } else 5850 #endif 5851 local_irq_enable(); 5852 } 5853 5854 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5855 { 5856 #ifdef CONFIG_RPS 5857 return sd->rps_ipi_list != NULL; 5858 #else 5859 return false; 5860 #endif 5861 } 5862 5863 static int process_backlog(struct napi_struct *napi, int quota) 5864 { 5865 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5866 bool again = true; 5867 int work = 0; 5868 5869 /* Check if we have pending ipi, its better to send them now, 5870 * not waiting net_rx_action() end. 5871 */ 5872 if (sd_has_rps_ipi_waiting(sd)) { 5873 local_irq_disable(); 5874 net_rps_action_and_irq_enable(sd); 5875 } 5876 5877 napi->weight = dev_rx_weight; 5878 while (again) { 5879 struct sk_buff *skb; 5880 5881 while ((skb = __skb_dequeue(&sd->process_queue))) { 5882 rcu_read_lock(); 5883 __netif_receive_skb(skb); 5884 rcu_read_unlock(); 5885 input_queue_head_incr(sd); 5886 if (++work >= quota) 5887 return work; 5888 5889 } 5890 5891 rps_lock_irq_disable(sd); 5892 if (skb_queue_empty(&sd->input_pkt_queue)) { 5893 /* 5894 * Inline a custom version of __napi_complete(). 5895 * only current cpu owns and manipulates this napi, 5896 * and NAPI_STATE_SCHED is the only possible flag set 5897 * on backlog. 5898 * We can use a plain write instead of clear_bit(), 5899 * and we dont need an smp_mb() memory barrier. 5900 */ 5901 napi->state = 0; 5902 again = false; 5903 } else { 5904 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5905 &sd->process_queue); 5906 } 5907 rps_unlock_irq_enable(sd); 5908 } 5909 5910 return work; 5911 } 5912 5913 /** 5914 * __napi_schedule - schedule for receive 5915 * @n: entry to schedule 5916 * 5917 * The entry's receive function will be scheduled to run. 5918 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5919 */ 5920 void __napi_schedule(struct napi_struct *n) 5921 { 5922 unsigned long flags; 5923 5924 local_irq_save(flags); 5925 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5926 local_irq_restore(flags); 5927 } 5928 EXPORT_SYMBOL(__napi_schedule); 5929 5930 /** 5931 * napi_schedule_prep - check if napi can be scheduled 5932 * @n: napi context 5933 * 5934 * Test if NAPI routine is already running, and if not mark 5935 * it as running. This is used as a condition variable to 5936 * insure only one NAPI poll instance runs. We also make 5937 * sure there is no pending NAPI disable. 5938 */ 5939 bool napi_schedule_prep(struct napi_struct *n) 5940 { 5941 unsigned long val, new; 5942 5943 do { 5944 val = READ_ONCE(n->state); 5945 if (unlikely(val & NAPIF_STATE_DISABLE)) 5946 return false; 5947 new = val | NAPIF_STATE_SCHED; 5948 5949 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5950 * This was suggested by Alexander Duyck, as compiler 5951 * emits better code than : 5952 * if (val & NAPIF_STATE_SCHED) 5953 * new |= NAPIF_STATE_MISSED; 5954 */ 5955 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5956 NAPIF_STATE_MISSED; 5957 } while (cmpxchg(&n->state, val, new) != val); 5958 5959 return !(val & NAPIF_STATE_SCHED); 5960 } 5961 EXPORT_SYMBOL(napi_schedule_prep); 5962 5963 /** 5964 * __napi_schedule_irqoff - schedule for receive 5965 * @n: entry to schedule 5966 * 5967 * Variant of __napi_schedule() assuming hard irqs are masked. 5968 * 5969 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5970 * because the interrupt disabled assumption might not be true 5971 * due to force-threaded interrupts and spinlock substitution. 5972 */ 5973 void __napi_schedule_irqoff(struct napi_struct *n) 5974 { 5975 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5976 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5977 else 5978 __napi_schedule(n); 5979 } 5980 EXPORT_SYMBOL(__napi_schedule_irqoff); 5981 5982 bool napi_complete_done(struct napi_struct *n, int work_done) 5983 { 5984 unsigned long flags, val, new, timeout = 0; 5985 bool ret = true; 5986 5987 /* 5988 * 1) Don't let napi dequeue from the cpu poll list 5989 * just in case its running on a different cpu. 5990 * 2) If we are busy polling, do nothing here, we have 5991 * the guarantee we will be called later. 5992 */ 5993 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5994 NAPIF_STATE_IN_BUSY_POLL))) 5995 return false; 5996 5997 if (work_done) { 5998 if (n->gro_bitmask) 5999 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6000 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6001 } 6002 if (n->defer_hard_irqs_count > 0) { 6003 n->defer_hard_irqs_count--; 6004 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6005 if (timeout) 6006 ret = false; 6007 } 6008 if (n->gro_bitmask) { 6009 /* When the NAPI instance uses a timeout and keeps postponing 6010 * it, we need to bound somehow the time packets are kept in 6011 * the GRO layer 6012 */ 6013 napi_gro_flush(n, !!timeout); 6014 } 6015 6016 gro_normal_list(n); 6017 6018 if (unlikely(!list_empty(&n->poll_list))) { 6019 /* If n->poll_list is not empty, we need to mask irqs */ 6020 local_irq_save(flags); 6021 list_del_init(&n->poll_list); 6022 local_irq_restore(flags); 6023 } 6024 6025 do { 6026 val = READ_ONCE(n->state); 6027 6028 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6029 6030 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6031 NAPIF_STATE_SCHED_THREADED | 6032 NAPIF_STATE_PREFER_BUSY_POLL); 6033 6034 /* If STATE_MISSED was set, leave STATE_SCHED set, 6035 * because we will call napi->poll() one more time. 6036 * This C code was suggested by Alexander Duyck to help gcc. 6037 */ 6038 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6039 NAPIF_STATE_SCHED; 6040 } while (cmpxchg(&n->state, val, new) != val); 6041 6042 if (unlikely(val & NAPIF_STATE_MISSED)) { 6043 __napi_schedule(n); 6044 return false; 6045 } 6046 6047 if (timeout) 6048 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6049 HRTIMER_MODE_REL_PINNED); 6050 return ret; 6051 } 6052 EXPORT_SYMBOL(napi_complete_done); 6053 6054 /* must be called under rcu_read_lock(), as we dont take a reference */ 6055 static struct napi_struct *napi_by_id(unsigned int napi_id) 6056 { 6057 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6058 struct napi_struct *napi; 6059 6060 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6061 if (napi->napi_id == napi_id) 6062 return napi; 6063 6064 return NULL; 6065 } 6066 6067 #if defined(CONFIG_NET_RX_BUSY_POLL) 6068 6069 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6070 { 6071 if (!skip_schedule) { 6072 gro_normal_list(napi); 6073 __napi_schedule(napi); 6074 return; 6075 } 6076 6077 if (napi->gro_bitmask) { 6078 /* flush too old packets 6079 * If HZ < 1000, flush all packets. 6080 */ 6081 napi_gro_flush(napi, HZ >= 1000); 6082 } 6083 6084 gro_normal_list(napi); 6085 clear_bit(NAPI_STATE_SCHED, &napi->state); 6086 } 6087 6088 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6089 u16 budget) 6090 { 6091 bool skip_schedule = false; 6092 unsigned long timeout; 6093 int rc; 6094 6095 /* Busy polling means there is a high chance device driver hard irq 6096 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6097 * set in napi_schedule_prep(). 6098 * Since we are about to call napi->poll() once more, we can safely 6099 * clear NAPI_STATE_MISSED. 6100 * 6101 * Note: x86 could use a single "lock and ..." instruction 6102 * to perform these two clear_bit() 6103 */ 6104 clear_bit(NAPI_STATE_MISSED, &napi->state); 6105 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6106 6107 local_bh_disable(); 6108 6109 if (prefer_busy_poll) { 6110 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6111 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6112 if (napi->defer_hard_irqs_count && timeout) { 6113 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6114 skip_schedule = true; 6115 } 6116 } 6117 6118 /* All we really want here is to re-enable device interrupts. 6119 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6120 */ 6121 rc = napi->poll(napi, budget); 6122 /* We can't gro_normal_list() here, because napi->poll() might have 6123 * rearmed the napi (napi_complete_done()) in which case it could 6124 * already be running on another CPU. 6125 */ 6126 trace_napi_poll(napi, rc, budget); 6127 netpoll_poll_unlock(have_poll_lock); 6128 if (rc == budget) 6129 __busy_poll_stop(napi, skip_schedule); 6130 local_bh_enable(); 6131 } 6132 6133 void napi_busy_loop(unsigned int napi_id, 6134 bool (*loop_end)(void *, unsigned long), 6135 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6136 { 6137 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6138 int (*napi_poll)(struct napi_struct *napi, int budget); 6139 void *have_poll_lock = NULL; 6140 struct napi_struct *napi; 6141 6142 restart: 6143 napi_poll = NULL; 6144 6145 rcu_read_lock(); 6146 6147 napi = napi_by_id(napi_id); 6148 if (!napi) 6149 goto out; 6150 6151 preempt_disable(); 6152 for (;;) { 6153 int work = 0; 6154 6155 local_bh_disable(); 6156 if (!napi_poll) { 6157 unsigned long val = READ_ONCE(napi->state); 6158 6159 /* If multiple threads are competing for this napi, 6160 * we avoid dirtying napi->state as much as we can. 6161 */ 6162 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6163 NAPIF_STATE_IN_BUSY_POLL)) { 6164 if (prefer_busy_poll) 6165 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6166 goto count; 6167 } 6168 if (cmpxchg(&napi->state, val, 6169 val | NAPIF_STATE_IN_BUSY_POLL | 6170 NAPIF_STATE_SCHED) != val) { 6171 if (prefer_busy_poll) 6172 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6173 goto count; 6174 } 6175 have_poll_lock = netpoll_poll_lock(napi); 6176 napi_poll = napi->poll; 6177 } 6178 work = napi_poll(napi, budget); 6179 trace_napi_poll(napi, work, budget); 6180 gro_normal_list(napi); 6181 count: 6182 if (work > 0) 6183 __NET_ADD_STATS(dev_net(napi->dev), 6184 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6185 local_bh_enable(); 6186 6187 if (!loop_end || loop_end(loop_end_arg, start_time)) 6188 break; 6189 6190 if (unlikely(need_resched())) { 6191 if (napi_poll) 6192 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6193 preempt_enable(); 6194 rcu_read_unlock(); 6195 cond_resched(); 6196 if (loop_end(loop_end_arg, start_time)) 6197 return; 6198 goto restart; 6199 } 6200 cpu_relax(); 6201 } 6202 if (napi_poll) 6203 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6204 preempt_enable(); 6205 out: 6206 rcu_read_unlock(); 6207 } 6208 EXPORT_SYMBOL(napi_busy_loop); 6209 6210 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6211 6212 static void napi_hash_add(struct napi_struct *napi) 6213 { 6214 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6215 return; 6216 6217 spin_lock(&napi_hash_lock); 6218 6219 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6220 do { 6221 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6222 napi_gen_id = MIN_NAPI_ID; 6223 } while (napi_by_id(napi_gen_id)); 6224 napi->napi_id = napi_gen_id; 6225 6226 hlist_add_head_rcu(&napi->napi_hash_node, 6227 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6228 6229 spin_unlock(&napi_hash_lock); 6230 } 6231 6232 /* Warning : caller is responsible to make sure rcu grace period 6233 * is respected before freeing memory containing @napi 6234 */ 6235 static void napi_hash_del(struct napi_struct *napi) 6236 { 6237 spin_lock(&napi_hash_lock); 6238 6239 hlist_del_init_rcu(&napi->napi_hash_node); 6240 6241 spin_unlock(&napi_hash_lock); 6242 } 6243 6244 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6245 { 6246 struct napi_struct *napi; 6247 6248 napi = container_of(timer, struct napi_struct, timer); 6249 6250 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6251 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6252 */ 6253 if (!napi_disable_pending(napi) && 6254 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6255 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6256 __napi_schedule_irqoff(napi); 6257 } 6258 6259 return HRTIMER_NORESTART; 6260 } 6261 6262 static void init_gro_hash(struct napi_struct *napi) 6263 { 6264 int i; 6265 6266 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6267 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6268 napi->gro_hash[i].count = 0; 6269 } 6270 napi->gro_bitmask = 0; 6271 } 6272 6273 int dev_set_threaded(struct net_device *dev, bool threaded) 6274 { 6275 struct napi_struct *napi; 6276 int err = 0; 6277 6278 if (dev->threaded == threaded) 6279 return 0; 6280 6281 if (threaded) { 6282 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6283 if (!napi->thread) { 6284 err = napi_kthread_create(napi); 6285 if (err) { 6286 threaded = false; 6287 break; 6288 } 6289 } 6290 } 6291 } 6292 6293 dev->threaded = threaded; 6294 6295 /* Make sure kthread is created before THREADED bit 6296 * is set. 6297 */ 6298 smp_mb__before_atomic(); 6299 6300 /* Setting/unsetting threaded mode on a napi might not immediately 6301 * take effect, if the current napi instance is actively being 6302 * polled. In this case, the switch between threaded mode and 6303 * softirq mode will happen in the next round of napi_schedule(). 6304 * This should not cause hiccups/stalls to the live traffic. 6305 */ 6306 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6307 if (threaded) 6308 set_bit(NAPI_STATE_THREADED, &napi->state); 6309 else 6310 clear_bit(NAPI_STATE_THREADED, &napi->state); 6311 } 6312 6313 return err; 6314 } 6315 EXPORT_SYMBOL(dev_set_threaded); 6316 6317 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6318 int (*poll)(struct napi_struct *, int), int weight) 6319 { 6320 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6321 return; 6322 6323 INIT_LIST_HEAD(&napi->poll_list); 6324 INIT_HLIST_NODE(&napi->napi_hash_node); 6325 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6326 napi->timer.function = napi_watchdog; 6327 init_gro_hash(napi); 6328 napi->skb = NULL; 6329 INIT_LIST_HEAD(&napi->rx_list); 6330 napi->rx_count = 0; 6331 napi->poll = poll; 6332 if (weight > NAPI_POLL_WEIGHT) 6333 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6334 weight); 6335 napi->weight = weight; 6336 napi->dev = dev; 6337 #ifdef CONFIG_NETPOLL 6338 napi->poll_owner = -1; 6339 #endif 6340 set_bit(NAPI_STATE_SCHED, &napi->state); 6341 set_bit(NAPI_STATE_NPSVC, &napi->state); 6342 list_add_rcu(&napi->dev_list, &dev->napi_list); 6343 napi_hash_add(napi); 6344 /* Create kthread for this napi if dev->threaded is set. 6345 * Clear dev->threaded if kthread creation failed so that 6346 * threaded mode will not be enabled in napi_enable(). 6347 */ 6348 if (dev->threaded && napi_kthread_create(napi)) 6349 dev->threaded = 0; 6350 } 6351 EXPORT_SYMBOL(netif_napi_add); 6352 6353 void napi_disable(struct napi_struct *n) 6354 { 6355 unsigned long val, new; 6356 6357 might_sleep(); 6358 set_bit(NAPI_STATE_DISABLE, &n->state); 6359 6360 for ( ; ; ) { 6361 val = READ_ONCE(n->state); 6362 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6363 usleep_range(20, 200); 6364 continue; 6365 } 6366 6367 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6368 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6369 6370 if (cmpxchg(&n->state, val, new) == val) 6371 break; 6372 } 6373 6374 hrtimer_cancel(&n->timer); 6375 6376 clear_bit(NAPI_STATE_DISABLE, &n->state); 6377 } 6378 EXPORT_SYMBOL(napi_disable); 6379 6380 /** 6381 * napi_enable - enable NAPI scheduling 6382 * @n: NAPI context 6383 * 6384 * Resume NAPI from being scheduled on this context. 6385 * Must be paired with napi_disable. 6386 */ 6387 void napi_enable(struct napi_struct *n) 6388 { 6389 unsigned long val, new; 6390 6391 do { 6392 val = READ_ONCE(n->state); 6393 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6394 6395 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6396 if (n->dev->threaded && n->thread) 6397 new |= NAPIF_STATE_THREADED; 6398 } while (cmpxchg(&n->state, val, new) != val); 6399 } 6400 EXPORT_SYMBOL(napi_enable); 6401 6402 static void flush_gro_hash(struct napi_struct *napi) 6403 { 6404 int i; 6405 6406 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6407 struct sk_buff *skb, *n; 6408 6409 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6410 kfree_skb(skb); 6411 napi->gro_hash[i].count = 0; 6412 } 6413 } 6414 6415 /* Must be called in process context */ 6416 void __netif_napi_del(struct napi_struct *napi) 6417 { 6418 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6419 return; 6420 6421 napi_hash_del(napi); 6422 list_del_rcu(&napi->dev_list); 6423 napi_free_frags(napi); 6424 6425 flush_gro_hash(napi); 6426 napi->gro_bitmask = 0; 6427 6428 if (napi->thread) { 6429 kthread_stop(napi->thread); 6430 napi->thread = NULL; 6431 } 6432 } 6433 EXPORT_SYMBOL(__netif_napi_del); 6434 6435 static int __napi_poll(struct napi_struct *n, bool *repoll) 6436 { 6437 int work, weight; 6438 6439 weight = n->weight; 6440 6441 /* This NAPI_STATE_SCHED test is for avoiding a race 6442 * with netpoll's poll_napi(). Only the entity which 6443 * obtains the lock and sees NAPI_STATE_SCHED set will 6444 * actually make the ->poll() call. Therefore we avoid 6445 * accidentally calling ->poll() when NAPI is not scheduled. 6446 */ 6447 work = 0; 6448 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6449 work = n->poll(n, weight); 6450 trace_napi_poll(n, work, weight); 6451 } 6452 6453 if (unlikely(work > weight)) 6454 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6455 n->poll, work, weight); 6456 6457 if (likely(work < weight)) 6458 return work; 6459 6460 /* Drivers must not modify the NAPI state if they 6461 * consume the entire weight. In such cases this code 6462 * still "owns" the NAPI instance and therefore can 6463 * move the instance around on the list at-will. 6464 */ 6465 if (unlikely(napi_disable_pending(n))) { 6466 napi_complete(n); 6467 return work; 6468 } 6469 6470 /* The NAPI context has more processing work, but busy-polling 6471 * is preferred. Exit early. 6472 */ 6473 if (napi_prefer_busy_poll(n)) { 6474 if (napi_complete_done(n, work)) { 6475 /* If timeout is not set, we need to make sure 6476 * that the NAPI is re-scheduled. 6477 */ 6478 napi_schedule(n); 6479 } 6480 return work; 6481 } 6482 6483 if (n->gro_bitmask) { 6484 /* flush too old packets 6485 * If HZ < 1000, flush all packets. 6486 */ 6487 napi_gro_flush(n, HZ >= 1000); 6488 } 6489 6490 gro_normal_list(n); 6491 6492 /* Some drivers may have called napi_schedule 6493 * prior to exhausting their budget. 6494 */ 6495 if (unlikely(!list_empty(&n->poll_list))) { 6496 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6497 n->dev ? n->dev->name : "backlog"); 6498 return work; 6499 } 6500 6501 *repoll = true; 6502 6503 return work; 6504 } 6505 6506 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6507 { 6508 bool do_repoll = false; 6509 void *have; 6510 int work; 6511 6512 list_del_init(&n->poll_list); 6513 6514 have = netpoll_poll_lock(n); 6515 6516 work = __napi_poll(n, &do_repoll); 6517 6518 if (do_repoll) 6519 list_add_tail(&n->poll_list, repoll); 6520 6521 netpoll_poll_unlock(have); 6522 6523 return work; 6524 } 6525 6526 static int napi_thread_wait(struct napi_struct *napi) 6527 { 6528 bool woken = false; 6529 6530 set_current_state(TASK_INTERRUPTIBLE); 6531 6532 while (!kthread_should_stop()) { 6533 /* Testing SCHED_THREADED bit here to make sure the current 6534 * kthread owns this napi and could poll on this napi. 6535 * Testing SCHED bit is not enough because SCHED bit might be 6536 * set by some other busy poll thread or by napi_disable(). 6537 */ 6538 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6539 WARN_ON(!list_empty(&napi->poll_list)); 6540 __set_current_state(TASK_RUNNING); 6541 return 0; 6542 } 6543 6544 schedule(); 6545 /* woken being true indicates this thread owns this napi. */ 6546 woken = true; 6547 set_current_state(TASK_INTERRUPTIBLE); 6548 } 6549 __set_current_state(TASK_RUNNING); 6550 6551 return -1; 6552 } 6553 6554 static int napi_threaded_poll(void *data) 6555 { 6556 struct napi_struct *napi = data; 6557 void *have; 6558 6559 while (!napi_thread_wait(napi)) { 6560 for (;;) { 6561 bool repoll = false; 6562 6563 local_bh_disable(); 6564 6565 have = netpoll_poll_lock(napi); 6566 __napi_poll(napi, &repoll); 6567 netpoll_poll_unlock(have); 6568 6569 local_bh_enable(); 6570 6571 if (!repoll) 6572 break; 6573 6574 cond_resched(); 6575 } 6576 } 6577 return 0; 6578 } 6579 6580 static void skb_defer_free_flush(struct softnet_data *sd) 6581 { 6582 struct sk_buff *skb, *next; 6583 unsigned long flags; 6584 6585 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6586 if (!READ_ONCE(sd->defer_list)) 6587 return; 6588 6589 spin_lock_irqsave(&sd->defer_lock, flags); 6590 skb = sd->defer_list; 6591 sd->defer_list = NULL; 6592 sd->defer_count = 0; 6593 spin_unlock_irqrestore(&sd->defer_lock, flags); 6594 6595 while (skb != NULL) { 6596 next = skb->next; 6597 __kfree_skb(skb); 6598 skb = next; 6599 } 6600 } 6601 6602 static __latent_entropy void net_rx_action(struct softirq_action *h) 6603 { 6604 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6605 unsigned long time_limit = jiffies + 6606 usecs_to_jiffies(netdev_budget_usecs); 6607 int budget = netdev_budget; 6608 LIST_HEAD(list); 6609 LIST_HEAD(repoll); 6610 6611 local_irq_disable(); 6612 list_splice_init(&sd->poll_list, &list); 6613 local_irq_enable(); 6614 6615 for (;;) { 6616 struct napi_struct *n; 6617 6618 if (list_empty(&list)) { 6619 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6620 goto end; 6621 break; 6622 } 6623 6624 n = list_first_entry(&list, struct napi_struct, poll_list); 6625 budget -= napi_poll(n, &repoll); 6626 6627 /* If softirq window is exhausted then punt. 6628 * Allow this to run for 2 jiffies since which will allow 6629 * an average latency of 1.5/HZ. 6630 */ 6631 if (unlikely(budget <= 0 || 6632 time_after_eq(jiffies, time_limit))) { 6633 sd->time_squeeze++; 6634 break; 6635 } 6636 } 6637 6638 local_irq_disable(); 6639 6640 list_splice_tail_init(&sd->poll_list, &list); 6641 list_splice_tail(&repoll, &list); 6642 list_splice(&list, &sd->poll_list); 6643 if (!list_empty(&sd->poll_list)) 6644 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6645 6646 net_rps_action_and_irq_enable(sd); 6647 end: 6648 skb_defer_free_flush(sd); 6649 } 6650 6651 struct netdev_adjacent { 6652 struct net_device *dev; 6653 netdevice_tracker dev_tracker; 6654 6655 /* upper master flag, there can only be one master device per list */ 6656 bool master; 6657 6658 /* lookup ignore flag */ 6659 bool ignore; 6660 6661 /* counter for the number of times this device was added to us */ 6662 u16 ref_nr; 6663 6664 /* private field for the users */ 6665 void *private; 6666 6667 struct list_head list; 6668 struct rcu_head rcu; 6669 }; 6670 6671 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6672 struct list_head *adj_list) 6673 { 6674 struct netdev_adjacent *adj; 6675 6676 list_for_each_entry(adj, adj_list, list) { 6677 if (adj->dev == adj_dev) 6678 return adj; 6679 } 6680 return NULL; 6681 } 6682 6683 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6684 struct netdev_nested_priv *priv) 6685 { 6686 struct net_device *dev = (struct net_device *)priv->data; 6687 6688 return upper_dev == dev; 6689 } 6690 6691 /** 6692 * netdev_has_upper_dev - Check if device is linked to an upper device 6693 * @dev: device 6694 * @upper_dev: upper device to check 6695 * 6696 * Find out if a device is linked to specified upper device and return true 6697 * in case it is. Note that this checks only immediate upper device, 6698 * not through a complete stack of devices. The caller must hold the RTNL lock. 6699 */ 6700 bool netdev_has_upper_dev(struct net_device *dev, 6701 struct net_device *upper_dev) 6702 { 6703 struct netdev_nested_priv priv = { 6704 .data = (void *)upper_dev, 6705 }; 6706 6707 ASSERT_RTNL(); 6708 6709 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6710 &priv); 6711 } 6712 EXPORT_SYMBOL(netdev_has_upper_dev); 6713 6714 /** 6715 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6716 * @dev: device 6717 * @upper_dev: upper device to check 6718 * 6719 * Find out if a device is linked to specified upper device and return true 6720 * in case it is. Note that this checks the entire upper device chain. 6721 * The caller must hold rcu lock. 6722 */ 6723 6724 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6725 struct net_device *upper_dev) 6726 { 6727 struct netdev_nested_priv priv = { 6728 .data = (void *)upper_dev, 6729 }; 6730 6731 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6732 &priv); 6733 } 6734 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6735 6736 /** 6737 * netdev_has_any_upper_dev - Check if device is linked to some device 6738 * @dev: device 6739 * 6740 * Find out if a device is linked to an upper device and return true in case 6741 * it is. The caller must hold the RTNL lock. 6742 */ 6743 bool netdev_has_any_upper_dev(struct net_device *dev) 6744 { 6745 ASSERT_RTNL(); 6746 6747 return !list_empty(&dev->adj_list.upper); 6748 } 6749 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6750 6751 /** 6752 * netdev_master_upper_dev_get - Get master upper device 6753 * @dev: device 6754 * 6755 * Find a master upper device and return pointer to it or NULL in case 6756 * it's not there. The caller must hold the RTNL lock. 6757 */ 6758 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6759 { 6760 struct netdev_adjacent *upper; 6761 6762 ASSERT_RTNL(); 6763 6764 if (list_empty(&dev->adj_list.upper)) 6765 return NULL; 6766 6767 upper = list_first_entry(&dev->adj_list.upper, 6768 struct netdev_adjacent, list); 6769 if (likely(upper->master)) 6770 return upper->dev; 6771 return NULL; 6772 } 6773 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6774 6775 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6776 { 6777 struct netdev_adjacent *upper; 6778 6779 ASSERT_RTNL(); 6780 6781 if (list_empty(&dev->adj_list.upper)) 6782 return NULL; 6783 6784 upper = list_first_entry(&dev->adj_list.upper, 6785 struct netdev_adjacent, list); 6786 if (likely(upper->master) && !upper->ignore) 6787 return upper->dev; 6788 return NULL; 6789 } 6790 6791 /** 6792 * netdev_has_any_lower_dev - Check if device is linked to some device 6793 * @dev: device 6794 * 6795 * Find out if a device is linked to a lower device and return true in case 6796 * it is. The caller must hold the RTNL lock. 6797 */ 6798 static bool netdev_has_any_lower_dev(struct net_device *dev) 6799 { 6800 ASSERT_RTNL(); 6801 6802 return !list_empty(&dev->adj_list.lower); 6803 } 6804 6805 void *netdev_adjacent_get_private(struct list_head *adj_list) 6806 { 6807 struct netdev_adjacent *adj; 6808 6809 adj = list_entry(adj_list, struct netdev_adjacent, list); 6810 6811 return adj->private; 6812 } 6813 EXPORT_SYMBOL(netdev_adjacent_get_private); 6814 6815 /** 6816 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6817 * @dev: device 6818 * @iter: list_head ** of the current position 6819 * 6820 * Gets the next device from the dev's upper list, starting from iter 6821 * position. The caller must hold RCU read lock. 6822 */ 6823 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6824 struct list_head **iter) 6825 { 6826 struct netdev_adjacent *upper; 6827 6828 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6829 6830 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6831 6832 if (&upper->list == &dev->adj_list.upper) 6833 return NULL; 6834 6835 *iter = &upper->list; 6836 6837 return upper->dev; 6838 } 6839 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6840 6841 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6842 struct list_head **iter, 6843 bool *ignore) 6844 { 6845 struct netdev_adjacent *upper; 6846 6847 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6848 6849 if (&upper->list == &dev->adj_list.upper) 6850 return NULL; 6851 6852 *iter = &upper->list; 6853 *ignore = upper->ignore; 6854 6855 return upper->dev; 6856 } 6857 6858 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6859 struct list_head **iter) 6860 { 6861 struct netdev_adjacent *upper; 6862 6863 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6864 6865 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6866 6867 if (&upper->list == &dev->adj_list.upper) 6868 return NULL; 6869 6870 *iter = &upper->list; 6871 6872 return upper->dev; 6873 } 6874 6875 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6876 int (*fn)(struct net_device *dev, 6877 struct netdev_nested_priv *priv), 6878 struct netdev_nested_priv *priv) 6879 { 6880 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6881 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6882 int ret, cur = 0; 6883 bool ignore; 6884 6885 now = dev; 6886 iter = &dev->adj_list.upper; 6887 6888 while (1) { 6889 if (now != dev) { 6890 ret = fn(now, priv); 6891 if (ret) 6892 return ret; 6893 } 6894 6895 next = NULL; 6896 while (1) { 6897 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6898 if (!udev) 6899 break; 6900 if (ignore) 6901 continue; 6902 6903 next = udev; 6904 niter = &udev->adj_list.upper; 6905 dev_stack[cur] = now; 6906 iter_stack[cur++] = iter; 6907 break; 6908 } 6909 6910 if (!next) { 6911 if (!cur) 6912 return 0; 6913 next = dev_stack[--cur]; 6914 niter = iter_stack[cur]; 6915 } 6916 6917 now = next; 6918 iter = niter; 6919 } 6920 6921 return 0; 6922 } 6923 6924 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6925 int (*fn)(struct net_device *dev, 6926 struct netdev_nested_priv *priv), 6927 struct netdev_nested_priv *priv) 6928 { 6929 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6930 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6931 int ret, cur = 0; 6932 6933 now = dev; 6934 iter = &dev->adj_list.upper; 6935 6936 while (1) { 6937 if (now != dev) { 6938 ret = fn(now, priv); 6939 if (ret) 6940 return ret; 6941 } 6942 6943 next = NULL; 6944 while (1) { 6945 udev = netdev_next_upper_dev_rcu(now, &iter); 6946 if (!udev) 6947 break; 6948 6949 next = udev; 6950 niter = &udev->adj_list.upper; 6951 dev_stack[cur] = now; 6952 iter_stack[cur++] = iter; 6953 break; 6954 } 6955 6956 if (!next) { 6957 if (!cur) 6958 return 0; 6959 next = dev_stack[--cur]; 6960 niter = iter_stack[cur]; 6961 } 6962 6963 now = next; 6964 iter = niter; 6965 } 6966 6967 return 0; 6968 } 6969 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6970 6971 static bool __netdev_has_upper_dev(struct net_device *dev, 6972 struct net_device *upper_dev) 6973 { 6974 struct netdev_nested_priv priv = { 6975 .flags = 0, 6976 .data = (void *)upper_dev, 6977 }; 6978 6979 ASSERT_RTNL(); 6980 6981 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 6982 &priv); 6983 } 6984 6985 /** 6986 * netdev_lower_get_next_private - Get the next ->private from the 6987 * lower neighbour list 6988 * @dev: device 6989 * @iter: list_head ** of the current position 6990 * 6991 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6992 * list, starting from iter position. The caller must hold either hold the 6993 * RTNL lock or its own locking that guarantees that the neighbour lower 6994 * list will remain unchanged. 6995 */ 6996 void *netdev_lower_get_next_private(struct net_device *dev, 6997 struct list_head **iter) 6998 { 6999 struct netdev_adjacent *lower; 7000 7001 lower = list_entry(*iter, struct netdev_adjacent, list); 7002 7003 if (&lower->list == &dev->adj_list.lower) 7004 return NULL; 7005 7006 *iter = lower->list.next; 7007 7008 return lower->private; 7009 } 7010 EXPORT_SYMBOL(netdev_lower_get_next_private); 7011 7012 /** 7013 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7014 * lower neighbour list, RCU 7015 * variant 7016 * @dev: device 7017 * @iter: list_head ** of the current position 7018 * 7019 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7020 * list, starting from iter position. The caller must hold RCU read lock. 7021 */ 7022 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7023 struct list_head **iter) 7024 { 7025 struct netdev_adjacent *lower; 7026 7027 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7028 7029 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7030 7031 if (&lower->list == &dev->adj_list.lower) 7032 return NULL; 7033 7034 *iter = &lower->list; 7035 7036 return lower->private; 7037 } 7038 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7039 7040 /** 7041 * netdev_lower_get_next - Get the next device from the lower neighbour 7042 * list 7043 * @dev: device 7044 * @iter: list_head ** of the current position 7045 * 7046 * Gets the next netdev_adjacent from the dev's lower neighbour 7047 * list, starting from iter position. The caller must hold RTNL lock or 7048 * its own locking that guarantees that the neighbour lower 7049 * list will remain unchanged. 7050 */ 7051 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7052 { 7053 struct netdev_adjacent *lower; 7054 7055 lower = list_entry(*iter, struct netdev_adjacent, list); 7056 7057 if (&lower->list == &dev->adj_list.lower) 7058 return NULL; 7059 7060 *iter = lower->list.next; 7061 7062 return lower->dev; 7063 } 7064 EXPORT_SYMBOL(netdev_lower_get_next); 7065 7066 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7067 struct list_head **iter) 7068 { 7069 struct netdev_adjacent *lower; 7070 7071 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7072 7073 if (&lower->list == &dev->adj_list.lower) 7074 return NULL; 7075 7076 *iter = &lower->list; 7077 7078 return lower->dev; 7079 } 7080 7081 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7082 struct list_head **iter, 7083 bool *ignore) 7084 { 7085 struct netdev_adjacent *lower; 7086 7087 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7088 7089 if (&lower->list == &dev->adj_list.lower) 7090 return NULL; 7091 7092 *iter = &lower->list; 7093 *ignore = lower->ignore; 7094 7095 return lower->dev; 7096 } 7097 7098 int netdev_walk_all_lower_dev(struct net_device *dev, 7099 int (*fn)(struct net_device *dev, 7100 struct netdev_nested_priv *priv), 7101 struct netdev_nested_priv *priv) 7102 { 7103 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7104 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7105 int ret, cur = 0; 7106 7107 now = dev; 7108 iter = &dev->adj_list.lower; 7109 7110 while (1) { 7111 if (now != dev) { 7112 ret = fn(now, priv); 7113 if (ret) 7114 return ret; 7115 } 7116 7117 next = NULL; 7118 while (1) { 7119 ldev = netdev_next_lower_dev(now, &iter); 7120 if (!ldev) 7121 break; 7122 7123 next = ldev; 7124 niter = &ldev->adj_list.lower; 7125 dev_stack[cur] = now; 7126 iter_stack[cur++] = iter; 7127 break; 7128 } 7129 7130 if (!next) { 7131 if (!cur) 7132 return 0; 7133 next = dev_stack[--cur]; 7134 niter = iter_stack[cur]; 7135 } 7136 7137 now = next; 7138 iter = niter; 7139 } 7140 7141 return 0; 7142 } 7143 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7144 7145 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7146 int (*fn)(struct net_device *dev, 7147 struct netdev_nested_priv *priv), 7148 struct netdev_nested_priv *priv) 7149 { 7150 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7151 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7152 int ret, cur = 0; 7153 bool ignore; 7154 7155 now = dev; 7156 iter = &dev->adj_list.lower; 7157 7158 while (1) { 7159 if (now != dev) { 7160 ret = fn(now, priv); 7161 if (ret) 7162 return ret; 7163 } 7164 7165 next = NULL; 7166 while (1) { 7167 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7168 if (!ldev) 7169 break; 7170 if (ignore) 7171 continue; 7172 7173 next = ldev; 7174 niter = &ldev->adj_list.lower; 7175 dev_stack[cur] = now; 7176 iter_stack[cur++] = iter; 7177 break; 7178 } 7179 7180 if (!next) { 7181 if (!cur) 7182 return 0; 7183 next = dev_stack[--cur]; 7184 niter = iter_stack[cur]; 7185 } 7186 7187 now = next; 7188 iter = niter; 7189 } 7190 7191 return 0; 7192 } 7193 7194 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7195 struct list_head **iter) 7196 { 7197 struct netdev_adjacent *lower; 7198 7199 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7200 if (&lower->list == &dev->adj_list.lower) 7201 return NULL; 7202 7203 *iter = &lower->list; 7204 7205 return lower->dev; 7206 } 7207 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7208 7209 static u8 __netdev_upper_depth(struct net_device *dev) 7210 { 7211 struct net_device *udev; 7212 struct list_head *iter; 7213 u8 max_depth = 0; 7214 bool ignore; 7215 7216 for (iter = &dev->adj_list.upper, 7217 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7218 udev; 7219 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7220 if (ignore) 7221 continue; 7222 if (max_depth < udev->upper_level) 7223 max_depth = udev->upper_level; 7224 } 7225 7226 return max_depth; 7227 } 7228 7229 static u8 __netdev_lower_depth(struct net_device *dev) 7230 { 7231 struct net_device *ldev; 7232 struct list_head *iter; 7233 u8 max_depth = 0; 7234 bool ignore; 7235 7236 for (iter = &dev->adj_list.lower, 7237 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7238 ldev; 7239 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7240 if (ignore) 7241 continue; 7242 if (max_depth < ldev->lower_level) 7243 max_depth = ldev->lower_level; 7244 } 7245 7246 return max_depth; 7247 } 7248 7249 static int __netdev_update_upper_level(struct net_device *dev, 7250 struct netdev_nested_priv *__unused) 7251 { 7252 dev->upper_level = __netdev_upper_depth(dev) + 1; 7253 return 0; 7254 } 7255 7256 #ifdef CONFIG_LOCKDEP 7257 static LIST_HEAD(net_unlink_list); 7258 7259 static void net_unlink_todo(struct net_device *dev) 7260 { 7261 if (list_empty(&dev->unlink_list)) 7262 list_add_tail(&dev->unlink_list, &net_unlink_list); 7263 } 7264 #endif 7265 7266 static int __netdev_update_lower_level(struct net_device *dev, 7267 struct netdev_nested_priv *priv) 7268 { 7269 dev->lower_level = __netdev_lower_depth(dev) + 1; 7270 7271 #ifdef CONFIG_LOCKDEP 7272 if (!priv) 7273 return 0; 7274 7275 if (priv->flags & NESTED_SYNC_IMM) 7276 dev->nested_level = dev->lower_level - 1; 7277 if (priv->flags & NESTED_SYNC_TODO) 7278 net_unlink_todo(dev); 7279 #endif 7280 return 0; 7281 } 7282 7283 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7284 int (*fn)(struct net_device *dev, 7285 struct netdev_nested_priv *priv), 7286 struct netdev_nested_priv *priv) 7287 { 7288 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7289 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7290 int ret, cur = 0; 7291 7292 now = dev; 7293 iter = &dev->adj_list.lower; 7294 7295 while (1) { 7296 if (now != dev) { 7297 ret = fn(now, priv); 7298 if (ret) 7299 return ret; 7300 } 7301 7302 next = NULL; 7303 while (1) { 7304 ldev = netdev_next_lower_dev_rcu(now, &iter); 7305 if (!ldev) 7306 break; 7307 7308 next = ldev; 7309 niter = &ldev->adj_list.lower; 7310 dev_stack[cur] = now; 7311 iter_stack[cur++] = iter; 7312 break; 7313 } 7314 7315 if (!next) { 7316 if (!cur) 7317 return 0; 7318 next = dev_stack[--cur]; 7319 niter = iter_stack[cur]; 7320 } 7321 7322 now = next; 7323 iter = niter; 7324 } 7325 7326 return 0; 7327 } 7328 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7329 7330 /** 7331 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7332 * lower neighbour list, RCU 7333 * variant 7334 * @dev: device 7335 * 7336 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7337 * list. The caller must hold RCU read lock. 7338 */ 7339 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7340 { 7341 struct netdev_adjacent *lower; 7342 7343 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7344 struct netdev_adjacent, list); 7345 if (lower) 7346 return lower->private; 7347 return NULL; 7348 } 7349 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7350 7351 /** 7352 * netdev_master_upper_dev_get_rcu - Get master upper device 7353 * @dev: device 7354 * 7355 * Find a master upper device and return pointer to it or NULL in case 7356 * it's not there. The caller must hold the RCU read lock. 7357 */ 7358 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7359 { 7360 struct netdev_adjacent *upper; 7361 7362 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7363 struct netdev_adjacent, list); 7364 if (upper && likely(upper->master)) 7365 return upper->dev; 7366 return NULL; 7367 } 7368 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7369 7370 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7371 struct net_device *adj_dev, 7372 struct list_head *dev_list) 7373 { 7374 char linkname[IFNAMSIZ+7]; 7375 7376 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7377 "upper_%s" : "lower_%s", adj_dev->name); 7378 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7379 linkname); 7380 } 7381 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7382 char *name, 7383 struct list_head *dev_list) 7384 { 7385 char linkname[IFNAMSIZ+7]; 7386 7387 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7388 "upper_%s" : "lower_%s", name); 7389 sysfs_remove_link(&(dev->dev.kobj), linkname); 7390 } 7391 7392 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7393 struct net_device *adj_dev, 7394 struct list_head *dev_list) 7395 { 7396 return (dev_list == &dev->adj_list.upper || 7397 dev_list == &dev->adj_list.lower) && 7398 net_eq(dev_net(dev), dev_net(adj_dev)); 7399 } 7400 7401 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7402 struct net_device *adj_dev, 7403 struct list_head *dev_list, 7404 void *private, bool master) 7405 { 7406 struct netdev_adjacent *adj; 7407 int ret; 7408 7409 adj = __netdev_find_adj(adj_dev, dev_list); 7410 7411 if (adj) { 7412 adj->ref_nr += 1; 7413 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7414 dev->name, adj_dev->name, adj->ref_nr); 7415 7416 return 0; 7417 } 7418 7419 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7420 if (!adj) 7421 return -ENOMEM; 7422 7423 adj->dev = adj_dev; 7424 adj->master = master; 7425 adj->ref_nr = 1; 7426 adj->private = private; 7427 adj->ignore = false; 7428 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7429 7430 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7431 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7432 7433 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7434 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7435 if (ret) 7436 goto free_adj; 7437 } 7438 7439 /* Ensure that master link is always the first item in list. */ 7440 if (master) { 7441 ret = sysfs_create_link(&(dev->dev.kobj), 7442 &(adj_dev->dev.kobj), "master"); 7443 if (ret) 7444 goto remove_symlinks; 7445 7446 list_add_rcu(&adj->list, dev_list); 7447 } else { 7448 list_add_tail_rcu(&adj->list, dev_list); 7449 } 7450 7451 return 0; 7452 7453 remove_symlinks: 7454 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7455 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7456 free_adj: 7457 dev_put_track(adj_dev, &adj->dev_tracker); 7458 kfree(adj); 7459 7460 return ret; 7461 } 7462 7463 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7464 struct net_device *adj_dev, 7465 u16 ref_nr, 7466 struct list_head *dev_list) 7467 { 7468 struct netdev_adjacent *adj; 7469 7470 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7471 dev->name, adj_dev->name, ref_nr); 7472 7473 adj = __netdev_find_adj(adj_dev, dev_list); 7474 7475 if (!adj) { 7476 pr_err("Adjacency does not exist for device %s from %s\n", 7477 dev->name, adj_dev->name); 7478 WARN_ON(1); 7479 return; 7480 } 7481 7482 if (adj->ref_nr > ref_nr) { 7483 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7484 dev->name, adj_dev->name, ref_nr, 7485 adj->ref_nr - ref_nr); 7486 adj->ref_nr -= ref_nr; 7487 return; 7488 } 7489 7490 if (adj->master) 7491 sysfs_remove_link(&(dev->dev.kobj), "master"); 7492 7493 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7494 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7495 7496 list_del_rcu(&adj->list); 7497 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7498 adj_dev->name, dev->name, adj_dev->name); 7499 dev_put_track(adj_dev, &adj->dev_tracker); 7500 kfree_rcu(adj, rcu); 7501 } 7502 7503 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7504 struct net_device *upper_dev, 7505 struct list_head *up_list, 7506 struct list_head *down_list, 7507 void *private, bool master) 7508 { 7509 int ret; 7510 7511 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7512 private, master); 7513 if (ret) 7514 return ret; 7515 7516 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7517 private, false); 7518 if (ret) { 7519 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7520 return ret; 7521 } 7522 7523 return 0; 7524 } 7525 7526 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7527 struct net_device *upper_dev, 7528 u16 ref_nr, 7529 struct list_head *up_list, 7530 struct list_head *down_list) 7531 { 7532 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7533 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7534 } 7535 7536 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7537 struct net_device *upper_dev, 7538 void *private, bool master) 7539 { 7540 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7541 &dev->adj_list.upper, 7542 &upper_dev->adj_list.lower, 7543 private, master); 7544 } 7545 7546 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7547 struct net_device *upper_dev) 7548 { 7549 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7550 &dev->adj_list.upper, 7551 &upper_dev->adj_list.lower); 7552 } 7553 7554 static int __netdev_upper_dev_link(struct net_device *dev, 7555 struct net_device *upper_dev, bool master, 7556 void *upper_priv, void *upper_info, 7557 struct netdev_nested_priv *priv, 7558 struct netlink_ext_ack *extack) 7559 { 7560 struct netdev_notifier_changeupper_info changeupper_info = { 7561 .info = { 7562 .dev = dev, 7563 .extack = extack, 7564 }, 7565 .upper_dev = upper_dev, 7566 .master = master, 7567 .linking = true, 7568 .upper_info = upper_info, 7569 }; 7570 struct net_device *master_dev; 7571 int ret = 0; 7572 7573 ASSERT_RTNL(); 7574 7575 if (dev == upper_dev) 7576 return -EBUSY; 7577 7578 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7579 if (__netdev_has_upper_dev(upper_dev, dev)) 7580 return -EBUSY; 7581 7582 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7583 return -EMLINK; 7584 7585 if (!master) { 7586 if (__netdev_has_upper_dev(dev, upper_dev)) 7587 return -EEXIST; 7588 } else { 7589 master_dev = __netdev_master_upper_dev_get(dev); 7590 if (master_dev) 7591 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7592 } 7593 7594 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7595 &changeupper_info.info); 7596 ret = notifier_to_errno(ret); 7597 if (ret) 7598 return ret; 7599 7600 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7601 master); 7602 if (ret) 7603 return ret; 7604 7605 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7606 &changeupper_info.info); 7607 ret = notifier_to_errno(ret); 7608 if (ret) 7609 goto rollback; 7610 7611 __netdev_update_upper_level(dev, NULL); 7612 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7613 7614 __netdev_update_lower_level(upper_dev, priv); 7615 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7616 priv); 7617 7618 return 0; 7619 7620 rollback: 7621 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7622 7623 return ret; 7624 } 7625 7626 /** 7627 * netdev_upper_dev_link - Add a link to the upper device 7628 * @dev: device 7629 * @upper_dev: new upper device 7630 * @extack: netlink extended ack 7631 * 7632 * Adds a link to device which is upper to this one. The caller must hold 7633 * the RTNL lock. On a failure a negative errno code is returned. 7634 * On success the reference counts are adjusted and the function 7635 * returns zero. 7636 */ 7637 int netdev_upper_dev_link(struct net_device *dev, 7638 struct net_device *upper_dev, 7639 struct netlink_ext_ack *extack) 7640 { 7641 struct netdev_nested_priv priv = { 7642 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7643 .data = NULL, 7644 }; 7645 7646 return __netdev_upper_dev_link(dev, upper_dev, false, 7647 NULL, NULL, &priv, extack); 7648 } 7649 EXPORT_SYMBOL(netdev_upper_dev_link); 7650 7651 /** 7652 * netdev_master_upper_dev_link - Add a master link to the upper device 7653 * @dev: device 7654 * @upper_dev: new upper device 7655 * @upper_priv: upper device private 7656 * @upper_info: upper info to be passed down via notifier 7657 * @extack: netlink extended ack 7658 * 7659 * Adds a link to device which is upper to this one. In this case, only 7660 * one master upper device can be linked, although other non-master devices 7661 * might be linked as well. The caller must hold the RTNL lock. 7662 * On a failure a negative errno code is returned. On success the reference 7663 * counts are adjusted and the function returns zero. 7664 */ 7665 int netdev_master_upper_dev_link(struct net_device *dev, 7666 struct net_device *upper_dev, 7667 void *upper_priv, void *upper_info, 7668 struct netlink_ext_ack *extack) 7669 { 7670 struct netdev_nested_priv priv = { 7671 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7672 .data = NULL, 7673 }; 7674 7675 return __netdev_upper_dev_link(dev, upper_dev, true, 7676 upper_priv, upper_info, &priv, extack); 7677 } 7678 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7679 7680 static void __netdev_upper_dev_unlink(struct net_device *dev, 7681 struct net_device *upper_dev, 7682 struct netdev_nested_priv *priv) 7683 { 7684 struct netdev_notifier_changeupper_info changeupper_info = { 7685 .info = { 7686 .dev = dev, 7687 }, 7688 .upper_dev = upper_dev, 7689 .linking = false, 7690 }; 7691 7692 ASSERT_RTNL(); 7693 7694 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7695 7696 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7697 &changeupper_info.info); 7698 7699 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7700 7701 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7702 &changeupper_info.info); 7703 7704 __netdev_update_upper_level(dev, NULL); 7705 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7706 7707 __netdev_update_lower_level(upper_dev, priv); 7708 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7709 priv); 7710 } 7711 7712 /** 7713 * netdev_upper_dev_unlink - Removes a link to upper device 7714 * @dev: device 7715 * @upper_dev: new upper device 7716 * 7717 * Removes a link to device which is upper to this one. The caller must hold 7718 * the RTNL lock. 7719 */ 7720 void netdev_upper_dev_unlink(struct net_device *dev, 7721 struct net_device *upper_dev) 7722 { 7723 struct netdev_nested_priv priv = { 7724 .flags = NESTED_SYNC_TODO, 7725 .data = NULL, 7726 }; 7727 7728 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7729 } 7730 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7731 7732 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7733 struct net_device *lower_dev, 7734 bool val) 7735 { 7736 struct netdev_adjacent *adj; 7737 7738 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7739 if (adj) 7740 adj->ignore = val; 7741 7742 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7743 if (adj) 7744 adj->ignore = val; 7745 } 7746 7747 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7748 struct net_device *lower_dev) 7749 { 7750 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7751 } 7752 7753 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7754 struct net_device *lower_dev) 7755 { 7756 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7757 } 7758 7759 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7760 struct net_device *new_dev, 7761 struct net_device *dev, 7762 struct netlink_ext_ack *extack) 7763 { 7764 struct netdev_nested_priv priv = { 7765 .flags = 0, 7766 .data = NULL, 7767 }; 7768 int err; 7769 7770 if (!new_dev) 7771 return 0; 7772 7773 if (old_dev && new_dev != old_dev) 7774 netdev_adjacent_dev_disable(dev, old_dev); 7775 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7776 extack); 7777 if (err) { 7778 if (old_dev && new_dev != old_dev) 7779 netdev_adjacent_dev_enable(dev, old_dev); 7780 return err; 7781 } 7782 7783 return 0; 7784 } 7785 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7786 7787 void netdev_adjacent_change_commit(struct net_device *old_dev, 7788 struct net_device *new_dev, 7789 struct net_device *dev) 7790 { 7791 struct netdev_nested_priv priv = { 7792 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7793 .data = NULL, 7794 }; 7795 7796 if (!new_dev || !old_dev) 7797 return; 7798 7799 if (new_dev == old_dev) 7800 return; 7801 7802 netdev_adjacent_dev_enable(dev, old_dev); 7803 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7804 } 7805 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7806 7807 void netdev_adjacent_change_abort(struct net_device *old_dev, 7808 struct net_device *new_dev, 7809 struct net_device *dev) 7810 { 7811 struct netdev_nested_priv priv = { 7812 .flags = 0, 7813 .data = NULL, 7814 }; 7815 7816 if (!new_dev) 7817 return; 7818 7819 if (old_dev && new_dev != old_dev) 7820 netdev_adjacent_dev_enable(dev, old_dev); 7821 7822 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7823 } 7824 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7825 7826 /** 7827 * netdev_bonding_info_change - Dispatch event about slave change 7828 * @dev: device 7829 * @bonding_info: info to dispatch 7830 * 7831 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7832 * The caller must hold the RTNL lock. 7833 */ 7834 void netdev_bonding_info_change(struct net_device *dev, 7835 struct netdev_bonding_info *bonding_info) 7836 { 7837 struct netdev_notifier_bonding_info info = { 7838 .info.dev = dev, 7839 }; 7840 7841 memcpy(&info.bonding_info, bonding_info, 7842 sizeof(struct netdev_bonding_info)); 7843 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7844 &info.info); 7845 } 7846 EXPORT_SYMBOL(netdev_bonding_info_change); 7847 7848 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7849 struct netlink_ext_ack *extack) 7850 { 7851 struct netdev_notifier_offload_xstats_info info = { 7852 .info.dev = dev, 7853 .info.extack = extack, 7854 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7855 }; 7856 int err; 7857 int rc; 7858 7859 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7860 GFP_KERNEL); 7861 if (!dev->offload_xstats_l3) 7862 return -ENOMEM; 7863 7864 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7865 NETDEV_OFFLOAD_XSTATS_DISABLE, 7866 &info.info); 7867 err = notifier_to_errno(rc); 7868 if (err) 7869 goto free_stats; 7870 7871 return 0; 7872 7873 free_stats: 7874 kfree(dev->offload_xstats_l3); 7875 dev->offload_xstats_l3 = NULL; 7876 return err; 7877 } 7878 7879 int netdev_offload_xstats_enable(struct net_device *dev, 7880 enum netdev_offload_xstats_type type, 7881 struct netlink_ext_ack *extack) 7882 { 7883 ASSERT_RTNL(); 7884 7885 if (netdev_offload_xstats_enabled(dev, type)) 7886 return -EALREADY; 7887 7888 switch (type) { 7889 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7890 return netdev_offload_xstats_enable_l3(dev, extack); 7891 } 7892 7893 WARN_ON(1); 7894 return -EINVAL; 7895 } 7896 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7897 7898 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7899 { 7900 struct netdev_notifier_offload_xstats_info info = { 7901 .info.dev = dev, 7902 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7903 }; 7904 7905 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7906 &info.info); 7907 kfree(dev->offload_xstats_l3); 7908 dev->offload_xstats_l3 = NULL; 7909 } 7910 7911 int netdev_offload_xstats_disable(struct net_device *dev, 7912 enum netdev_offload_xstats_type type) 7913 { 7914 ASSERT_RTNL(); 7915 7916 if (!netdev_offload_xstats_enabled(dev, type)) 7917 return -EALREADY; 7918 7919 switch (type) { 7920 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7921 netdev_offload_xstats_disable_l3(dev); 7922 return 0; 7923 } 7924 7925 WARN_ON(1); 7926 return -EINVAL; 7927 } 7928 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7929 7930 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7931 { 7932 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7933 } 7934 7935 static struct rtnl_hw_stats64 * 7936 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7937 enum netdev_offload_xstats_type type) 7938 { 7939 switch (type) { 7940 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7941 return dev->offload_xstats_l3; 7942 } 7943 7944 WARN_ON(1); 7945 return NULL; 7946 } 7947 7948 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7949 enum netdev_offload_xstats_type type) 7950 { 7951 ASSERT_RTNL(); 7952 7953 return netdev_offload_xstats_get_ptr(dev, type); 7954 } 7955 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7956 7957 struct netdev_notifier_offload_xstats_ru { 7958 bool used; 7959 }; 7960 7961 struct netdev_notifier_offload_xstats_rd { 7962 struct rtnl_hw_stats64 stats; 7963 bool used; 7964 }; 7965 7966 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 7967 const struct rtnl_hw_stats64 *src) 7968 { 7969 dest->rx_packets += src->rx_packets; 7970 dest->tx_packets += src->tx_packets; 7971 dest->rx_bytes += src->rx_bytes; 7972 dest->tx_bytes += src->tx_bytes; 7973 dest->rx_errors += src->rx_errors; 7974 dest->tx_errors += src->tx_errors; 7975 dest->rx_dropped += src->rx_dropped; 7976 dest->tx_dropped += src->tx_dropped; 7977 dest->multicast += src->multicast; 7978 } 7979 7980 static int netdev_offload_xstats_get_used(struct net_device *dev, 7981 enum netdev_offload_xstats_type type, 7982 bool *p_used, 7983 struct netlink_ext_ack *extack) 7984 { 7985 struct netdev_notifier_offload_xstats_ru report_used = {}; 7986 struct netdev_notifier_offload_xstats_info info = { 7987 .info.dev = dev, 7988 .info.extack = extack, 7989 .type = type, 7990 .report_used = &report_used, 7991 }; 7992 int rc; 7993 7994 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 7995 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 7996 &info.info); 7997 *p_used = report_used.used; 7998 return notifier_to_errno(rc); 7999 } 8000 8001 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8002 enum netdev_offload_xstats_type type, 8003 struct rtnl_hw_stats64 *p_stats, 8004 bool *p_used, 8005 struct netlink_ext_ack *extack) 8006 { 8007 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8008 struct netdev_notifier_offload_xstats_info info = { 8009 .info.dev = dev, 8010 .info.extack = extack, 8011 .type = type, 8012 .report_delta = &report_delta, 8013 }; 8014 struct rtnl_hw_stats64 *stats; 8015 int rc; 8016 8017 stats = netdev_offload_xstats_get_ptr(dev, type); 8018 if (WARN_ON(!stats)) 8019 return -EINVAL; 8020 8021 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8022 &info.info); 8023 8024 /* Cache whatever we got, even if there was an error, otherwise the 8025 * successful stats retrievals would get lost. 8026 */ 8027 netdev_hw_stats64_add(stats, &report_delta.stats); 8028 8029 if (p_stats) 8030 *p_stats = *stats; 8031 *p_used = report_delta.used; 8032 8033 return notifier_to_errno(rc); 8034 } 8035 8036 int netdev_offload_xstats_get(struct net_device *dev, 8037 enum netdev_offload_xstats_type type, 8038 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8039 struct netlink_ext_ack *extack) 8040 { 8041 ASSERT_RTNL(); 8042 8043 if (p_stats) 8044 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8045 p_used, extack); 8046 else 8047 return netdev_offload_xstats_get_used(dev, type, p_used, 8048 extack); 8049 } 8050 EXPORT_SYMBOL(netdev_offload_xstats_get); 8051 8052 void 8053 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8054 const struct rtnl_hw_stats64 *stats) 8055 { 8056 report_delta->used = true; 8057 netdev_hw_stats64_add(&report_delta->stats, stats); 8058 } 8059 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8060 8061 void 8062 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8063 { 8064 report_used->used = true; 8065 } 8066 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8067 8068 void netdev_offload_xstats_push_delta(struct net_device *dev, 8069 enum netdev_offload_xstats_type type, 8070 const struct rtnl_hw_stats64 *p_stats) 8071 { 8072 struct rtnl_hw_stats64 *stats; 8073 8074 ASSERT_RTNL(); 8075 8076 stats = netdev_offload_xstats_get_ptr(dev, type); 8077 if (WARN_ON(!stats)) 8078 return; 8079 8080 netdev_hw_stats64_add(stats, p_stats); 8081 } 8082 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8083 8084 /** 8085 * netdev_get_xmit_slave - Get the xmit slave of master device 8086 * @dev: device 8087 * @skb: The packet 8088 * @all_slaves: assume all the slaves are active 8089 * 8090 * The reference counters are not incremented so the caller must be 8091 * careful with locks. The caller must hold RCU lock. 8092 * %NULL is returned if no slave is found. 8093 */ 8094 8095 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8096 struct sk_buff *skb, 8097 bool all_slaves) 8098 { 8099 const struct net_device_ops *ops = dev->netdev_ops; 8100 8101 if (!ops->ndo_get_xmit_slave) 8102 return NULL; 8103 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8104 } 8105 EXPORT_SYMBOL(netdev_get_xmit_slave); 8106 8107 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8108 struct sock *sk) 8109 { 8110 const struct net_device_ops *ops = dev->netdev_ops; 8111 8112 if (!ops->ndo_sk_get_lower_dev) 8113 return NULL; 8114 return ops->ndo_sk_get_lower_dev(dev, sk); 8115 } 8116 8117 /** 8118 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8119 * @dev: device 8120 * @sk: the socket 8121 * 8122 * %NULL is returned if no lower device is found. 8123 */ 8124 8125 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8126 struct sock *sk) 8127 { 8128 struct net_device *lower; 8129 8130 lower = netdev_sk_get_lower_dev(dev, sk); 8131 while (lower) { 8132 dev = lower; 8133 lower = netdev_sk_get_lower_dev(dev, sk); 8134 } 8135 8136 return dev; 8137 } 8138 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8139 8140 static void netdev_adjacent_add_links(struct net_device *dev) 8141 { 8142 struct netdev_adjacent *iter; 8143 8144 struct net *net = dev_net(dev); 8145 8146 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8147 if (!net_eq(net, dev_net(iter->dev))) 8148 continue; 8149 netdev_adjacent_sysfs_add(iter->dev, dev, 8150 &iter->dev->adj_list.lower); 8151 netdev_adjacent_sysfs_add(dev, iter->dev, 8152 &dev->adj_list.upper); 8153 } 8154 8155 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8156 if (!net_eq(net, dev_net(iter->dev))) 8157 continue; 8158 netdev_adjacent_sysfs_add(iter->dev, dev, 8159 &iter->dev->adj_list.upper); 8160 netdev_adjacent_sysfs_add(dev, iter->dev, 8161 &dev->adj_list.lower); 8162 } 8163 } 8164 8165 static void netdev_adjacent_del_links(struct net_device *dev) 8166 { 8167 struct netdev_adjacent *iter; 8168 8169 struct net *net = dev_net(dev); 8170 8171 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8172 if (!net_eq(net, dev_net(iter->dev))) 8173 continue; 8174 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8175 &iter->dev->adj_list.lower); 8176 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8177 &dev->adj_list.upper); 8178 } 8179 8180 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8181 if (!net_eq(net, dev_net(iter->dev))) 8182 continue; 8183 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8184 &iter->dev->adj_list.upper); 8185 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8186 &dev->adj_list.lower); 8187 } 8188 } 8189 8190 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8191 { 8192 struct netdev_adjacent *iter; 8193 8194 struct net *net = dev_net(dev); 8195 8196 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8197 if (!net_eq(net, dev_net(iter->dev))) 8198 continue; 8199 netdev_adjacent_sysfs_del(iter->dev, oldname, 8200 &iter->dev->adj_list.lower); 8201 netdev_adjacent_sysfs_add(iter->dev, dev, 8202 &iter->dev->adj_list.lower); 8203 } 8204 8205 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8206 if (!net_eq(net, dev_net(iter->dev))) 8207 continue; 8208 netdev_adjacent_sysfs_del(iter->dev, oldname, 8209 &iter->dev->adj_list.upper); 8210 netdev_adjacent_sysfs_add(iter->dev, dev, 8211 &iter->dev->adj_list.upper); 8212 } 8213 } 8214 8215 void *netdev_lower_dev_get_private(struct net_device *dev, 8216 struct net_device *lower_dev) 8217 { 8218 struct netdev_adjacent *lower; 8219 8220 if (!lower_dev) 8221 return NULL; 8222 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8223 if (!lower) 8224 return NULL; 8225 8226 return lower->private; 8227 } 8228 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8229 8230 8231 /** 8232 * netdev_lower_state_changed - Dispatch event about lower device state change 8233 * @lower_dev: device 8234 * @lower_state_info: state to dispatch 8235 * 8236 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8237 * The caller must hold the RTNL lock. 8238 */ 8239 void netdev_lower_state_changed(struct net_device *lower_dev, 8240 void *lower_state_info) 8241 { 8242 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8243 .info.dev = lower_dev, 8244 }; 8245 8246 ASSERT_RTNL(); 8247 changelowerstate_info.lower_state_info = lower_state_info; 8248 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8249 &changelowerstate_info.info); 8250 } 8251 EXPORT_SYMBOL(netdev_lower_state_changed); 8252 8253 static void dev_change_rx_flags(struct net_device *dev, int flags) 8254 { 8255 const struct net_device_ops *ops = dev->netdev_ops; 8256 8257 if (ops->ndo_change_rx_flags) 8258 ops->ndo_change_rx_flags(dev, flags); 8259 } 8260 8261 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8262 { 8263 unsigned int old_flags = dev->flags; 8264 kuid_t uid; 8265 kgid_t gid; 8266 8267 ASSERT_RTNL(); 8268 8269 dev->flags |= IFF_PROMISC; 8270 dev->promiscuity += inc; 8271 if (dev->promiscuity == 0) { 8272 /* 8273 * Avoid overflow. 8274 * If inc causes overflow, untouch promisc and return error. 8275 */ 8276 if (inc < 0) 8277 dev->flags &= ~IFF_PROMISC; 8278 else { 8279 dev->promiscuity -= inc; 8280 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8281 return -EOVERFLOW; 8282 } 8283 } 8284 if (dev->flags != old_flags) { 8285 pr_info("device %s %s promiscuous mode\n", 8286 dev->name, 8287 dev->flags & IFF_PROMISC ? "entered" : "left"); 8288 if (audit_enabled) { 8289 current_uid_gid(&uid, &gid); 8290 audit_log(audit_context(), GFP_ATOMIC, 8291 AUDIT_ANOM_PROMISCUOUS, 8292 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8293 dev->name, (dev->flags & IFF_PROMISC), 8294 (old_flags & IFF_PROMISC), 8295 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8296 from_kuid(&init_user_ns, uid), 8297 from_kgid(&init_user_ns, gid), 8298 audit_get_sessionid(current)); 8299 } 8300 8301 dev_change_rx_flags(dev, IFF_PROMISC); 8302 } 8303 if (notify) 8304 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8305 return 0; 8306 } 8307 8308 /** 8309 * dev_set_promiscuity - update promiscuity count on a device 8310 * @dev: device 8311 * @inc: modifier 8312 * 8313 * Add or remove promiscuity from a device. While the count in the device 8314 * remains above zero the interface remains promiscuous. Once it hits zero 8315 * the device reverts back to normal filtering operation. A negative inc 8316 * value is used to drop promiscuity on the device. 8317 * Return 0 if successful or a negative errno code on error. 8318 */ 8319 int dev_set_promiscuity(struct net_device *dev, int inc) 8320 { 8321 unsigned int old_flags = dev->flags; 8322 int err; 8323 8324 err = __dev_set_promiscuity(dev, inc, true); 8325 if (err < 0) 8326 return err; 8327 if (dev->flags != old_flags) 8328 dev_set_rx_mode(dev); 8329 return err; 8330 } 8331 EXPORT_SYMBOL(dev_set_promiscuity); 8332 8333 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8334 { 8335 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8336 8337 ASSERT_RTNL(); 8338 8339 dev->flags |= IFF_ALLMULTI; 8340 dev->allmulti += inc; 8341 if (dev->allmulti == 0) { 8342 /* 8343 * Avoid overflow. 8344 * If inc causes overflow, untouch allmulti and return error. 8345 */ 8346 if (inc < 0) 8347 dev->flags &= ~IFF_ALLMULTI; 8348 else { 8349 dev->allmulti -= inc; 8350 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8351 return -EOVERFLOW; 8352 } 8353 } 8354 if (dev->flags ^ old_flags) { 8355 dev_change_rx_flags(dev, IFF_ALLMULTI); 8356 dev_set_rx_mode(dev); 8357 if (notify) 8358 __dev_notify_flags(dev, old_flags, 8359 dev->gflags ^ old_gflags); 8360 } 8361 return 0; 8362 } 8363 8364 /** 8365 * dev_set_allmulti - update allmulti count on a device 8366 * @dev: device 8367 * @inc: modifier 8368 * 8369 * Add or remove reception of all multicast frames to a device. While the 8370 * count in the device remains above zero the interface remains listening 8371 * to all interfaces. Once it hits zero the device reverts back to normal 8372 * filtering operation. A negative @inc value is used to drop the counter 8373 * when releasing a resource needing all multicasts. 8374 * Return 0 if successful or a negative errno code on error. 8375 */ 8376 8377 int dev_set_allmulti(struct net_device *dev, int inc) 8378 { 8379 return __dev_set_allmulti(dev, inc, true); 8380 } 8381 EXPORT_SYMBOL(dev_set_allmulti); 8382 8383 /* 8384 * Upload unicast and multicast address lists to device and 8385 * configure RX filtering. When the device doesn't support unicast 8386 * filtering it is put in promiscuous mode while unicast addresses 8387 * are present. 8388 */ 8389 void __dev_set_rx_mode(struct net_device *dev) 8390 { 8391 const struct net_device_ops *ops = dev->netdev_ops; 8392 8393 /* dev_open will call this function so the list will stay sane. */ 8394 if (!(dev->flags&IFF_UP)) 8395 return; 8396 8397 if (!netif_device_present(dev)) 8398 return; 8399 8400 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8401 /* Unicast addresses changes may only happen under the rtnl, 8402 * therefore calling __dev_set_promiscuity here is safe. 8403 */ 8404 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8405 __dev_set_promiscuity(dev, 1, false); 8406 dev->uc_promisc = true; 8407 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8408 __dev_set_promiscuity(dev, -1, false); 8409 dev->uc_promisc = false; 8410 } 8411 } 8412 8413 if (ops->ndo_set_rx_mode) 8414 ops->ndo_set_rx_mode(dev); 8415 } 8416 8417 void dev_set_rx_mode(struct net_device *dev) 8418 { 8419 netif_addr_lock_bh(dev); 8420 __dev_set_rx_mode(dev); 8421 netif_addr_unlock_bh(dev); 8422 } 8423 8424 /** 8425 * dev_get_flags - get flags reported to userspace 8426 * @dev: device 8427 * 8428 * Get the combination of flag bits exported through APIs to userspace. 8429 */ 8430 unsigned int dev_get_flags(const struct net_device *dev) 8431 { 8432 unsigned int flags; 8433 8434 flags = (dev->flags & ~(IFF_PROMISC | 8435 IFF_ALLMULTI | 8436 IFF_RUNNING | 8437 IFF_LOWER_UP | 8438 IFF_DORMANT)) | 8439 (dev->gflags & (IFF_PROMISC | 8440 IFF_ALLMULTI)); 8441 8442 if (netif_running(dev)) { 8443 if (netif_oper_up(dev)) 8444 flags |= IFF_RUNNING; 8445 if (netif_carrier_ok(dev)) 8446 flags |= IFF_LOWER_UP; 8447 if (netif_dormant(dev)) 8448 flags |= IFF_DORMANT; 8449 } 8450 8451 return flags; 8452 } 8453 EXPORT_SYMBOL(dev_get_flags); 8454 8455 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8456 struct netlink_ext_ack *extack) 8457 { 8458 unsigned int old_flags = dev->flags; 8459 int ret; 8460 8461 ASSERT_RTNL(); 8462 8463 /* 8464 * Set the flags on our device. 8465 */ 8466 8467 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8468 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8469 IFF_AUTOMEDIA)) | 8470 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8471 IFF_ALLMULTI)); 8472 8473 /* 8474 * Load in the correct multicast list now the flags have changed. 8475 */ 8476 8477 if ((old_flags ^ flags) & IFF_MULTICAST) 8478 dev_change_rx_flags(dev, IFF_MULTICAST); 8479 8480 dev_set_rx_mode(dev); 8481 8482 /* 8483 * Have we downed the interface. We handle IFF_UP ourselves 8484 * according to user attempts to set it, rather than blindly 8485 * setting it. 8486 */ 8487 8488 ret = 0; 8489 if ((old_flags ^ flags) & IFF_UP) { 8490 if (old_flags & IFF_UP) 8491 __dev_close(dev); 8492 else 8493 ret = __dev_open(dev, extack); 8494 } 8495 8496 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8497 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8498 unsigned int old_flags = dev->flags; 8499 8500 dev->gflags ^= IFF_PROMISC; 8501 8502 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8503 if (dev->flags != old_flags) 8504 dev_set_rx_mode(dev); 8505 } 8506 8507 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8508 * is important. Some (broken) drivers set IFF_PROMISC, when 8509 * IFF_ALLMULTI is requested not asking us and not reporting. 8510 */ 8511 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8512 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8513 8514 dev->gflags ^= IFF_ALLMULTI; 8515 __dev_set_allmulti(dev, inc, false); 8516 } 8517 8518 return ret; 8519 } 8520 8521 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8522 unsigned int gchanges) 8523 { 8524 unsigned int changes = dev->flags ^ old_flags; 8525 8526 if (gchanges) 8527 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8528 8529 if (changes & IFF_UP) { 8530 if (dev->flags & IFF_UP) 8531 call_netdevice_notifiers(NETDEV_UP, dev); 8532 else 8533 call_netdevice_notifiers(NETDEV_DOWN, dev); 8534 } 8535 8536 if (dev->flags & IFF_UP && 8537 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8538 struct netdev_notifier_change_info change_info = { 8539 .info = { 8540 .dev = dev, 8541 }, 8542 .flags_changed = changes, 8543 }; 8544 8545 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8546 } 8547 } 8548 8549 /** 8550 * dev_change_flags - change device settings 8551 * @dev: device 8552 * @flags: device state flags 8553 * @extack: netlink extended ack 8554 * 8555 * Change settings on device based state flags. The flags are 8556 * in the userspace exported format. 8557 */ 8558 int dev_change_flags(struct net_device *dev, unsigned int flags, 8559 struct netlink_ext_ack *extack) 8560 { 8561 int ret; 8562 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8563 8564 ret = __dev_change_flags(dev, flags, extack); 8565 if (ret < 0) 8566 return ret; 8567 8568 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8569 __dev_notify_flags(dev, old_flags, changes); 8570 return ret; 8571 } 8572 EXPORT_SYMBOL(dev_change_flags); 8573 8574 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8575 { 8576 const struct net_device_ops *ops = dev->netdev_ops; 8577 8578 if (ops->ndo_change_mtu) 8579 return ops->ndo_change_mtu(dev, new_mtu); 8580 8581 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8582 WRITE_ONCE(dev->mtu, new_mtu); 8583 return 0; 8584 } 8585 EXPORT_SYMBOL(__dev_set_mtu); 8586 8587 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8588 struct netlink_ext_ack *extack) 8589 { 8590 /* MTU must be positive, and in range */ 8591 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8592 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8593 return -EINVAL; 8594 } 8595 8596 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8597 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8598 return -EINVAL; 8599 } 8600 return 0; 8601 } 8602 8603 /** 8604 * dev_set_mtu_ext - Change maximum transfer unit 8605 * @dev: device 8606 * @new_mtu: new transfer unit 8607 * @extack: netlink extended ack 8608 * 8609 * Change the maximum transfer size of the network device. 8610 */ 8611 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8612 struct netlink_ext_ack *extack) 8613 { 8614 int err, orig_mtu; 8615 8616 if (new_mtu == dev->mtu) 8617 return 0; 8618 8619 err = dev_validate_mtu(dev, new_mtu, extack); 8620 if (err) 8621 return err; 8622 8623 if (!netif_device_present(dev)) 8624 return -ENODEV; 8625 8626 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8627 err = notifier_to_errno(err); 8628 if (err) 8629 return err; 8630 8631 orig_mtu = dev->mtu; 8632 err = __dev_set_mtu(dev, new_mtu); 8633 8634 if (!err) { 8635 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8636 orig_mtu); 8637 err = notifier_to_errno(err); 8638 if (err) { 8639 /* setting mtu back and notifying everyone again, 8640 * so that they have a chance to revert changes. 8641 */ 8642 __dev_set_mtu(dev, orig_mtu); 8643 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8644 new_mtu); 8645 } 8646 } 8647 return err; 8648 } 8649 8650 int dev_set_mtu(struct net_device *dev, int new_mtu) 8651 { 8652 struct netlink_ext_ack extack; 8653 int err; 8654 8655 memset(&extack, 0, sizeof(extack)); 8656 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8657 if (err && extack._msg) 8658 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8659 return err; 8660 } 8661 EXPORT_SYMBOL(dev_set_mtu); 8662 8663 /** 8664 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8665 * @dev: device 8666 * @new_len: new tx queue length 8667 */ 8668 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8669 { 8670 unsigned int orig_len = dev->tx_queue_len; 8671 int res; 8672 8673 if (new_len != (unsigned int)new_len) 8674 return -ERANGE; 8675 8676 if (new_len != orig_len) { 8677 dev->tx_queue_len = new_len; 8678 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8679 res = notifier_to_errno(res); 8680 if (res) 8681 goto err_rollback; 8682 res = dev_qdisc_change_tx_queue_len(dev); 8683 if (res) 8684 goto err_rollback; 8685 } 8686 8687 return 0; 8688 8689 err_rollback: 8690 netdev_err(dev, "refused to change device tx_queue_len\n"); 8691 dev->tx_queue_len = orig_len; 8692 return res; 8693 } 8694 8695 /** 8696 * dev_set_group - Change group this device belongs to 8697 * @dev: device 8698 * @new_group: group this device should belong to 8699 */ 8700 void dev_set_group(struct net_device *dev, int new_group) 8701 { 8702 dev->group = new_group; 8703 } 8704 8705 /** 8706 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8707 * @dev: device 8708 * @addr: new address 8709 * @extack: netlink extended ack 8710 */ 8711 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8712 struct netlink_ext_ack *extack) 8713 { 8714 struct netdev_notifier_pre_changeaddr_info info = { 8715 .info.dev = dev, 8716 .info.extack = extack, 8717 .dev_addr = addr, 8718 }; 8719 int rc; 8720 8721 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8722 return notifier_to_errno(rc); 8723 } 8724 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8725 8726 /** 8727 * dev_set_mac_address - Change Media Access Control Address 8728 * @dev: device 8729 * @sa: new address 8730 * @extack: netlink extended ack 8731 * 8732 * Change the hardware (MAC) address of the device 8733 */ 8734 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8735 struct netlink_ext_ack *extack) 8736 { 8737 const struct net_device_ops *ops = dev->netdev_ops; 8738 int err; 8739 8740 if (!ops->ndo_set_mac_address) 8741 return -EOPNOTSUPP; 8742 if (sa->sa_family != dev->type) 8743 return -EINVAL; 8744 if (!netif_device_present(dev)) 8745 return -ENODEV; 8746 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8747 if (err) 8748 return err; 8749 err = ops->ndo_set_mac_address(dev, sa); 8750 if (err) 8751 return err; 8752 dev->addr_assign_type = NET_ADDR_SET; 8753 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8754 add_device_randomness(dev->dev_addr, dev->addr_len); 8755 return 0; 8756 } 8757 EXPORT_SYMBOL(dev_set_mac_address); 8758 8759 static DECLARE_RWSEM(dev_addr_sem); 8760 8761 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8762 struct netlink_ext_ack *extack) 8763 { 8764 int ret; 8765 8766 down_write(&dev_addr_sem); 8767 ret = dev_set_mac_address(dev, sa, extack); 8768 up_write(&dev_addr_sem); 8769 return ret; 8770 } 8771 EXPORT_SYMBOL(dev_set_mac_address_user); 8772 8773 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8774 { 8775 size_t size = sizeof(sa->sa_data); 8776 struct net_device *dev; 8777 int ret = 0; 8778 8779 down_read(&dev_addr_sem); 8780 rcu_read_lock(); 8781 8782 dev = dev_get_by_name_rcu(net, dev_name); 8783 if (!dev) { 8784 ret = -ENODEV; 8785 goto unlock; 8786 } 8787 if (!dev->addr_len) 8788 memset(sa->sa_data, 0, size); 8789 else 8790 memcpy(sa->sa_data, dev->dev_addr, 8791 min_t(size_t, size, dev->addr_len)); 8792 sa->sa_family = dev->type; 8793 8794 unlock: 8795 rcu_read_unlock(); 8796 up_read(&dev_addr_sem); 8797 return ret; 8798 } 8799 EXPORT_SYMBOL(dev_get_mac_address); 8800 8801 /** 8802 * dev_change_carrier - Change device carrier 8803 * @dev: device 8804 * @new_carrier: new value 8805 * 8806 * Change device carrier 8807 */ 8808 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8809 { 8810 const struct net_device_ops *ops = dev->netdev_ops; 8811 8812 if (!ops->ndo_change_carrier) 8813 return -EOPNOTSUPP; 8814 if (!netif_device_present(dev)) 8815 return -ENODEV; 8816 return ops->ndo_change_carrier(dev, new_carrier); 8817 } 8818 8819 /** 8820 * dev_get_phys_port_id - Get device physical port ID 8821 * @dev: device 8822 * @ppid: port ID 8823 * 8824 * Get device physical port ID 8825 */ 8826 int dev_get_phys_port_id(struct net_device *dev, 8827 struct netdev_phys_item_id *ppid) 8828 { 8829 const struct net_device_ops *ops = dev->netdev_ops; 8830 8831 if (!ops->ndo_get_phys_port_id) 8832 return -EOPNOTSUPP; 8833 return ops->ndo_get_phys_port_id(dev, ppid); 8834 } 8835 8836 /** 8837 * dev_get_phys_port_name - Get device physical port name 8838 * @dev: device 8839 * @name: port name 8840 * @len: limit of bytes to copy to name 8841 * 8842 * Get device physical port name 8843 */ 8844 int dev_get_phys_port_name(struct net_device *dev, 8845 char *name, size_t len) 8846 { 8847 const struct net_device_ops *ops = dev->netdev_ops; 8848 int err; 8849 8850 if (ops->ndo_get_phys_port_name) { 8851 err = ops->ndo_get_phys_port_name(dev, name, len); 8852 if (err != -EOPNOTSUPP) 8853 return err; 8854 } 8855 return devlink_compat_phys_port_name_get(dev, name, len); 8856 } 8857 8858 /** 8859 * dev_get_port_parent_id - Get the device's port parent identifier 8860 * @dev: network device 8861 * @ppid: pointer to a storage for the port's parent identifier 8862 * @recurse: allow/disallow recursion to lower devices 8863 * 8864 * Get the devices's port parent identifier 8865 */ 8866 int dev_get_port_parent_id(struct net_device *dev, 8867 struct netdev_phys_item_id *ppid, 8868 bool recurse) 8869 { 8870 const struct net_device_ops *ops = dev->netdev_ops; 8871 struct netdev_phys_item_id first = { }; 8872 struct net_device *lower_dev; 8873 struct list_head *iter; 8874 int err; 8875 8876 if (ops->ndo_get_port_parent_id) { 8877 err = ops->ndo_get_port_parent_id(dev, ppid); 8878 if (err != -EOPNOTSUPP) 8879 return err; 8880 } 8881 8882 err = devlink_compat_switch_id_get(dev, ppid); 8883 if (!recurse || err != -EOPNOTSUPP) 8884 return err; 8885 8886 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8887 err = dev_get_port_parent_id(lower_dev, ppid, true); 8888 if (err) 8889 break; 8890 if (!first.id_len) 8891 first = *ppid; 8892 else if (memcmp(&first, ppid, sizeof(*ppid))) 8893 return -EOPNOTSUPP; 8894 } 8895 8896 return err; 8897 } 8898 EXPORT_SYMBOL(dev_get_port_parent_id); 8899 8900 /** 8901 * netdev_port_same_parent_id - Indicate if two network devices have 8902 * the same port parent identifier 8903 * @a: first network device 8904 * @b: second network device 8905 */ 8906 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8907 { 8908 struct netdev_phys_item_id a_id = { }; 8909 struct netdev_phys_item_id b_id = { }; 8910 8911 if (dev_get_port_parent_id(a, &a_id, true) || 8912 dev_get_port_parent_id(b, &b_id, true)) 8913 return false; 8914 8915 return netdev_phys_item_id_same(&a_id, &b_id); 8916 } 8917 EXPORT_SYMBOL(netdev_port_same_parent_id); 8918 8919 /** 8920 * dev_change_proto_down - set carrier according to proto_down. 8921 * 8922 * @dev: device 8923 * @proto_down: new value 8924 */ 8925 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8926 { 8927 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8928 return -EOPNOTSUPP; 8929 if (!netif_device_present(dev)) 8930 return -ENODEV; 8931 if (proto_down) 8932 netif_carrier_off(dev); 8933 else 8934 netif_carrier_on(dev); 8935 dev->proto_down = proto_down; 8936 return 0; 8937 } 8938 8939 /** 8940 * dev_change_proto_down_reason - proto down reason 8941 * 8942 * @dev: device 8943 * @mask: proto down mask 8944 * @value: proto down value 8945 */ 8946 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8947 u32 value) 8948 { 8949 int b; 8950 8951 if (!mask) { 8952 dev->proto_down_reason = value; 8953 } else { 8954 for_each_set_bit(b, &mask, 32) { 8955 if (value & (1 << b)) 8956 dev->proto_down_reason |= BIT(b); 8957 else 8958 dev->proto_down_reason &= ~BIT(b); 8959 } 8960 } 8961 } 8962 8963 struct bpf_xdp_link { 8964 struct bpf_link link; 8965 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8966 int flags; 8967 }; 8968 8969 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8970 { 8971 if (flags & XDP_FLAGS_HW_MODE) 8972 return XDP_MODE_HW; 8973 if (flags & XDP_FLAGS_DRV_MODE) 8974 return XDP_MODE_DRV; 8975 if (flags & XDP_FLAGS_SKB_MODE) 8976 return XDP_MODE_SKB; 8977 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8978 } 8979 8980 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8981 { 8982 switch (mode) { 8983 case XDP_MODE_SKB: 8984 return generic_xdp_install; 8985 case XDP_MODE_DRV: 8986 case XDP_MODE_HW: 8987 return dev->netdev_ops->ndo_bpf; 8988 default: 8989 return NULL; 8990 } 8991 } 8992 8993 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8994 enum bpf_xdp_mode mode) 8995 { 8996 return dev->xdp_state[mode].link; 8997 } 8998 8999 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9000 enum bpf_xdp_mode mode) 9001 { 9002 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9003 9004 if (link) 9005 return link->link.prog; 9006 return dev->xdp_state[mode].prog; 9007 } 9008 9009 u8 dev_xdp_prog_count(struct net_device *dev) 9010 { 9011 u8 count = 0; 9012 int i; 9013 9014 for (i = 0; i < __MAX_XDP_MODE; i++) 9015 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9016 count++; 9017 return count; 9018 } 9019 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9020 9021 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9022 { 9023 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9024 9025 return prog ? prog->aux->id : 0; 9026 } 9027 9028 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9029 struct bpf_xdp_link *link) 9030 { 9031 dev->xdp_state[mode].link = link; 9032 dev->xdp_state[mode].prog = NULL; 9033 } 9034 9035 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9036 struct bpf_prog *prog) 9037 { 9038 dev->xdp_state[mode].link = NULL; 9039 dev->xdp_state[mode].prog = prog; 9040 } 9041 9042 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9043 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9044 u32 flags, struct bpf_prog *prog) 9045 { 9046 struct netdev_bpf xdp; 9047 int err; 9048 9049 memset(&xdp, 0, sizeof(xdp)); 9050 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9051 xdp.extack = extack; 9052 xdp.flags = flags; 9053 xdp.prog = prog; 9054 9055 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9056 * "moved" into driver), so they don't increment it on their own, but 9057 * they do decrement refcnt when program is detached or replaced. 9058 * Given net_device also owns link/prog, we need to bump refcnt here 9059 * to prevent drivers from underflowing it. 9060 */ 9061 if (prog) 9062 bpf_prog_inc(prog); 9063 err = bpf_op(dev, &xdp); 9064 if (err) { 9065 if (prog) 9066 bpf_prog_put(prog); 9067 return err; 9068 } 9069 9070 if (mode != XDP_MODE_HW) 9071 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9072 9073 return 0; 9074 } 9075 9076 static void dev_xdp_uninstall(struct net_device *dev) 9077 { 9078 struct bpf_xdp_link *link; 9079 struct bpf_prog *prog; 9080 enum bpf_xdp_mode mode; 9081 bpf_op_t bpf_op; 9082 9083 ASSERT_RTNL(); 9084 9085 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9086 prog = dev_xdp_prog(dev, mode); 9087 if (!prog) 9088 continue; 9089 9090 bpf_op = dev_xdp_bpf_op(dev, mode); 9091 if (!bpf_op) 9092 continue; 9093 9094 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9095 9096 /* auto-detach link from net device */ 9097 link = dev_xdp_link(dev, mode); 9098 if (link) 9099 link->dev = NULL; 9100 else 9101 bpf_prog_put(prog); 9102 9103 dev_xdp_set_link(dev, mode, NULL); 9104 } 9105 } 9106 9107 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9108 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9109 struct bpf_prog *old_prog, u32 flags) 9110 { 9111 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9112 struct bpf_prog *cur_prog; 9113 struct net_device *upper; 9114 struct list_head *iter; 9115 enum bpf_xdp_mode mode; 9116 bpf_op_t bpf_op; 9117 int err; 9118 9119 ASSERT_RTNL(); 9120 9121 /* either link or prog attachment, never both */ 9122 if (link && (new_prog || old_prog)) 9123 return -EINVAL; 9124 /* link supports only XDP mode flags */ 9125 if (link && (flags & ~XDP_FLAGS_MODES)) { 9126 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9127 return -EINVAL; 9128 } 9129 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9130 if (num_modes > 1) { 9131 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9132 return -EINVAL; 9133 } 9134 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9135 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9136 NL_SET_ERR_MSG(extack, 9137 "More than one program loaded, unset mode is ambiguous"); 9138 return -EINVAL; 9139 } 9140 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9141 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9142 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9143 return -EINVAL; 9144 } 9145 9146 mode = dev_xdp_mode(dev, flags); 9147 /* can't replace attached link */ 9148 if (dev_xdp_link(dev, mode)) { 9149 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9150 return -EBUSY; 9151 } 9152 9153 /* don't allow if an upper device already has a program */ 9154 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9155 if (dev_xdp_prog_count(upper) > 0) { 9156 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9157 return -EEXIST; 9158 } 9159 } 9160 9161 cur_prog = dev_xdp_prog(dev, mode); 9162 /* can't replace attached prog with link */ 9163 if (link && cur_prog) { 9164 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9165 return -EBUSY; 9166 } 9167 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9168 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9169 return -EEXIST; 9170 } 9171 9172 /* put effective new program into new_prog */ 9173 if (link) 9174 new_prog = link->link.prog; 9175 9176 if (new_prog) { 9177 bool offload = mode == XDP_MODE_HW; 9178 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9179 ? XDP_MODE_DRV : XDP_MODE_SKB; 9180 9181 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9182 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9183 return -EBUSY; 9184 } 9185 if (!offload && dev_xdp_prog(dev, other_mode)) { 9186 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9187 return -EEXIST; 9188 } 9189 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9190 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9191 return -EINVAL; 9192 } 9193 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9194 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9195 return -EINVAL; 9196 } 9197 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9198 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9199 return -EINVAL; 9200 } 9201 } 9202 9203 /* don't call drivers if the effective program didn't change */ 9204 if (new_prog != cur_prog) { 9205 bpf_op = dev_xdp_bpf_op(dev, mode); 9206 if (!bpf_op) { 9207 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9208 return -EOPNOTSUPP; 9209 } 9210 9211 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9212 if (err) 9213 return err; 9214 } 9215 9216 if (link) 9217 dev_xdp_set_link(dev, mode, link); 9218 else 9219 dev_xdp_set_prog(dev, mode, new_prog); 9220 if (cur_prog) 9221 bpf_prog_put(cur_prog); 9222 9223 return 0; 9224 } 9225 9226 static int dev_xdp_attach_link(struct net_device *dev, 9227 struct netlink_ext_ack *extack, 9228 struct bpf_xdp_link *link) 9229 { 9230 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9231 } 9232 9233 static int dev_xdp_detach_link(struct net_device *dev, 9234 struct netlink_ext_ack *extack, 9235 struct bpf_xdp_link *link) 9236 { 9237 enum bpf_xdp_mode mode; 9238 bpf_op_t bpf_op; 9239 9240 ASSERT_RTNL(); 9241 9242 mode = dev_xdp_mode(dev, link->flags); 9243 if (dev_xdp_link(dev, mode) != link) 9244 return -EINVAL; 9245 9246 bpf_op = dev_xdp_bpf_op(dev, mode); 9247 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9248 dev_xdp_set_link(dev, mode, NULL); 9249 return 0; 9250 } 9251 9252 static void bpf_xdp_link_release(struct bpf_link *link) 9253 { 9254 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9255 9256 rtnl_lock(); 9257 9258 /* if racing with net_device's tear down, xdp_link->dev might be 9259 * already NULL, in which case link was already auto-detached 9260 */ 9261 if (xdp_link->dev) { 9262 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9263 xdp_link->dev = NULL; 9264 } 9265 9266 rtnl_unlock(); 9267 } 9268 9269 static int bpf_xdp_link_detach(struct bpf_link *link) 9270 { 9271 bpf_xdp_link_release(link); 9272 return 0; 9273 } 9274 9275 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9276 { 9277 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9278 9279 kfree(xdp_link); 9280 } 9281 9282 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9283 struct seq_file *seq) 9284 { 9285 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9286 u32 ifindex = 0; 9287 9288 rtnl_lock(); 9289 if (xdp_link->dev) 9290 ifindex = xdp_link->dev->ifindex; 9291 rtnl_unlock(); 9292 9293 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9294 } 9295 9296 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9297 struct bpf_link_info *info) 9298 { 9299 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9300 u32 ifindex = 0; 9301 9302 rtnl_lock(); 9303 if (xdp_link->dev) 9304 ifindex = xdp_link->dev->ifindex; 9305 rtnl_unlock(); 9306 9307 info->xdp.ifindex = ifindex; 9308 return 0; 9309 } 9310 9311 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9312 struct bpf_prog *old_prog) 9313 { 9314 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9315 enum bpf_xdp_mode mode; 9316 bpf_op_t bpf_op; 9317 int err = 0; 9318 9319 rtnl_lock(); 9320 9321 /* link might have been auto-released already, so fail */ 9322 if (!xdp_link->dev) { 9323 err = -ENOLINK; 9324 goto out_unlock; 9325 } 9326 9327 if (old_prog && link->prog != old_prog) { 9328 err = -EPERM; 9329 goto out_unlock; 9330 } 9331 old_prog = link->prog; 9332 if (old_prog->type != new_prog->type || 9333 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9334 err = -EINVAL; 9335 goto out_unlock; 9336 } 9337 9338 if (old_prog == new_prog) { 9339 /* no-op, don't disturb drivers */ 9340 bpf_prog_put(new_prog); 9341 goto out_unlock; 9342 } 9343 9344 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9345 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9346 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9347 xdp_link->flags, new_prog); 9348 if (err) 9349 goto out_unlock; 9350 9351 old_prog = xchg(&link->prog, new_prog); 9352 bpf_prog_put(old_prog); 9353 9354 out_unlock: 9355 rtnl_unlock(); 9356 return err; 9357 } 9358 9359 static const struct bpf_link_ops bpf_xdp_link_lops = { 9360 .release = bpf_xdp_link_release, 9361 .dealloc = bpf_xdp_link_dealloc, 9362 .detach = bpf_xdp_link_detach, 9363 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9364 .fill_link_info = bpf_xdp_link_fill_link_info, 9365 .update_prog = bpf_xdp_link_update, 9366 }; 9367 9368 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9369 { 9370 struct net *net = current->nsproxy->net_ns; 9371 struct bpf_link_primer link_primer; 9372 struct bpf_xdp_link *link; 9373 struct net_device *dev; 9374 int err, fd; 9375 9376 rtnl_lock(); 9377 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9378 if (!dev) { 9379 rtnl_unlock(); 9380 return -EINVAL; 9381 } 9382 9383 link = kzalloc(sizeof(*link), GFP_USER); 9384 if (!link) { 9385 err = -ENOMEM; 9386 goto unlock; 9387 } 9388 9389 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9390 link->dev = dev; 9391 link->flags = attr->link_create.flags; 9392 9393 err = bpf_link_prime(&link->link, &link_primer); 9394 if (err) { 9395 kfree(link); 9396 goto unlock; 9397 } 9398 9399 err = dev_xdp_attach_link(dev, NULL, link); 9400 rtnl_unlock(); 9401 9402 if (err) { 9403 link->dev = NULL; 9404 bpf_link_cleanup(&link_primer); 9405 goto out_put_dev; 9406 } 9407 9408 fd = bpf_link_settle(&link_primer); 9409 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9410 dev_put(dev); 9411 return fd; 9412 9413 unlock: 9414 rtnl_unlock(); 9415 9416 out_put_dev: 9417 dev_put(dev); 9418 return err; 9419 } 9420 9421 /** 9422 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9423 * @dev: device 9424 * @extack: netlink extended ack 9425 * @fd: new program fd or negative value to clear 9426 * @expected_fd: old program fd that userspace expects to replace or clear 9427 * @flags: xdp-related flags 9428 * 9429 * Set or clear a bpf program for a device 9430 */ 9431 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9432 int fd, int expected_fd, u32 flags) 9433 { 9434 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9435 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9436 int err; 9437 9438 ASSERT_RTNL(); 9439 9440 if (fd >= 0) { 9441 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9442 mode != XDP_MODE_SKB); 9443 if (IS_ERR(new_prog)) 9444 return PTR_ERR(new_prog); 9445 } 9446 9447 if (expected_fd >= 0) { 9448 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9449 mode != XDP_MODE_SKB); 9450 if (IS_ERR(old_prog)) { 9451 err = PTR_ERR(old_prog); 9452 old_prog = NULL; 9453 goto err_out; 9454 } 9455 } 9456 9457 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9458 9459 err_out: 9460 if (err && new_prog) 9461 bpf_prog_put(new_prog); 9462 if (old_prog) 9463 bpf_prog_put(old_prog); 9464 return err; 9465 } 9466 9467 /** 9468 * dev_new_index - allocate an ifindex 9469 * @net: the applicable net namespace 9470 * 9471 * Returns a suitable unique value for a new device interface 9472 * number. The caller must hold the rtnl semaphore or the 9473 * dev_base_lock to be sure it remains unique. 9474 */ 9475 static int dev_new_index(struct net *net) 9476 { 9477 int ifindex = net->ifindex; 9478 9479 for (;;) { 9480 if (++ifindex <= 0) 9481 ifindex = 1; 9482 if (!__dev_get_by_index(net, ifindex)) 9483 return net->ifindex = ifindex; 9484 } 9485 } 9486 9487 /* Delayed registration/unregisteration */ 9488 LIST_HEAD(net_todo_list); 9489 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9490 9491 static void net_set_todo(struct net_device *dev) 9492 { 9493 list_add_tail(&dev->todo_list, &net_todo_list); 9494 atomic_inc(&dev_net(dev)->dev_unreg_count); 9495 } 9496 9497 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9498 struct net_device *upper, netdev_features_t features) 9499 { 9500 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9501 netdev_features_t feature; 9502 int feature_bit; 9503 9504 for_each_netdev_feature(upper_disables, feature_bit) { 9505 feature = __NETIF_F_BIT(feature_bit); 9506 if (!(upper->wanted_features & feature) 9507 && (features & feature)) { 9508 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9509 &feature, upper->name); 9510 features &= ~feature; 9511 } 9512 } 9513 9514 return features; 9515 } 9516 9517 static void netdev_sync_lower_features(struct net_device *upper, 9518 struct net_device *lower, netdev_features_t features) 9519 { 9520 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9521 netdev_features_t feature; 9522 int feature_bit; 9523 9524 for_each_netdev_feature(upper_disables, feature_bit) { 9525 feature = __NETIF_F_BIT(feature_bit); 9526 if (!(features & feature) && (lower->features & feature)) { 9527 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9528 &feature, lower->name); 9529 lower->wanted_features &= ~feature; 9530 __netdev_update_features(lower); 9531 9532 if (unlikely(lower->features & feature)) 9533 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9534 &feature, lower->name); 9535 else 9536 netdev_features_change(lower); 9537 } 9538 } 9539 } 9540 9541 static netdev_features_t netdev_fix_features(struct net_device *dev, 9542 netdev_features_t features) 9543 { 9544 /* Fix illegal checksum combinations */ 9545 if ((features & NETIF_F_HW_CSUM) && 9546 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9547 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9548 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9549 } 9550 9551 /* TSO requires that SG is present as well. */ 9552 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9553 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9554 features &= ~NETIF_F_ALL_TSO; 9555 } 9556 9557 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9558 !(features & NETIF_F_IP_CSUM)) { 9559 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9560 features &= ~NETIF_F_TSO; 9561 features &= ~NETIF_F_TSO_ECN; 9562 } 9563 9564 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9565 !(features & NETIF_F_IPV6_CSUM)) { 9566 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9567 features &= ~NETIF_F_TSO6; 9568 } 9569 9570 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9571 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9572 features &= ~NETIF_F_TSO_MANGLEID; 9573 9574 /* TSO ECN requires that TSO is present as well. */ 9575 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9576 features &= ~NETIF_F_TSO_ECN; 9577 9578 /* Software GSO depends on SG. */ 9579 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9580 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9581 features &= ~NETIF_F_GSO; 9582 } 9583 9584 /* GSO partial features require GSO partial be set */ 9585 if ((features & dev->gso_partial_features) && 9586 !(features & NETIF_F_GSO_PARTIAL)) { 9587 netdev_dbg(dev, 9588 "Dropping partially supported GSO features since no GSO partial.\n"); 9589 features &= ~dev->gso_partial_features; 9590 } 9591 9592 if (!(features & NETIF_F_RXCSUM)) { 9593 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9594 * successfully merged by hardware must also have the 9595 * checksum verified by hardware. If the user does not 9596 * want to enable RXCSUM, logically, we should disable GRO_HW. 9597 */ 9598 if (features & NETIF_F_GRO_HW) { 9599 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9600 features &= ~NETIF_F_GRO_HW; 9601 } 9602 } 9603 9604 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9605 if (features & NETIF_F_RXFCS) { 9606 if (features & NETIF_F_LRO) { 9607 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9608 features &= ~NETIF_F_LRO; 9609 } 9610 9611 if (features & NETIF_F_GRO_HW) { 9612 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9613 features &= ~NETIF_F_GRO_HW; 9614 } 9615 } 9616 9617 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9618 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9619 features &= ~NETIF_F_LRO; 9620 } 9621 9622 if (features & NETIF_F_HW_TLS_TX) { 9623 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9624 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9625 bool hw_csum = features & NETIF_F_HW_CSUM; 9626 9627 if (!ip_csum && !hw_csum) { 9628 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9629 features &= ~NETIF_F_HW_TLS_TX; 9630 } 9631 } 9632 9633 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9634 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9635 features &= ~NETIF_F_HW_TLS_RX; 9636 } 9637 9638 return features; 9639 } 9640 9641 int __netdev_update_features(struct net_device *dev) 9642 { 9643 struct net_device *upper, *lower; 9644 netdev_features_t features; 9645 struct list_head *iter; 9646 int err = -1; 9647 9648 ASSERT_RTNL(); 9649 9650 features = netdev_get_wanted_features(dev); 9651 9652 if (dev->netdev_ops->ndo_fix_features) 9653 features = dev->netdev_ops->ndo_fix_features(dev, features); 9654 9655 /* driver might be less strict about feature dependencies */ 9656 features = netdev_fix_features(dev, features); 9657 9658 /* some features can't be enabled if they're off on an upper device */ 9659 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9660 features = netdev_sync_upper_features(dev, upper, features); 9661 9662 if (dev->features == features) 9663 goto sync_lower; 9664 9665 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9666 &dev->features, &features); 9667 9668 if (dev->netdev_ops->ndo_set_features) 9669 err = dev->netdev_ops->ndo_set_features(dev, features); 9670 else 9671 err = 0; 9672 9673 if (unlikely(err < 0)) { 9674 netdev_err(dev, 9675 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9676 err, &features, &dev->features); 9677 /* return non-0 since some features might have changed and 9678 * it's better to fire a spurious notification than miss it 9679 */ 9680 return -1; 9681 } 9682 9683 sync_lower: 9684 /* some features must be disabled on lower devices when disabled 9685 * on an upper device (think: bonding master or bridge) 9686 */ 9687 netdev_for_each_lower_dev(dev, lower, iter) 9688 netdev_sync_lower_features(dev, lower, features); 9689 9690 if (!err) { 9691 netdev_features_t diff = features ^ dev->features; 9692 9693 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9694 /* udp_tunnel_{get,drop}_rx_info both need 9695 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9696 * device, or they won't do anything. 9697 * Thus we need to update dev->features 9698 * *before* calling udp_tunnel_get_rx_info, 9699 * but *after* calling udp_tunnel_drop_rx_info. 9700 */ 9701 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9702 dev->features = features; 9703 udp_tunnel_get_rx_info(dev); 9704 } else { 9705 udp_tunnel_drop_rx_info(dev); 9706 } 9707 } 9708 9709 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9710 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9711 dev->features = features; 9712 err |= vlan_get_rx_ctag_filter_info(dev); 9713 } else { 9714 vlan_drop_rx_ctag_filter_info(dev); 9715 } 9716 } 9717 9718 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9719 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9720 dev->features = features; 9721 err |= vlan_get_rx_stag_filter_info(dev); 9722 } else { 9723 vlan_drop_rx_stag_filter_info(dev); 9724 } 9725 } 9726 9727 dev->features = features; 9728 } 9729 9730 return err < 0 ? 0 : 1; 9731 } 9732 9733 /** 9734 * netdev_update_features - recalculate device features 9735 * @dev: the device to check 9736 * 9737 * Recalculate dev->features set and send notifications if it 9738 * has changed. Should be called after driver or hardware dependent 9739 * conditions might have changed that influence the features. 9740 */ 9741 void netdev_update_features(struct net_device *dev) 9742 { 9743 if (__netdev_update_features(dev)) 9744 netdev_features_change(dev); 9745 } 9746 EXPORT_SYMBOL(netdev_update_features); 9747 9748 /** 9749 * netdev_change_features - recalculate device features 9750 * @dev: the device to check 9751 * 9752 * Recalculate dev->features set and send notifications even 9753 * if they have not changed. Should be called instead of 9754 * netdev_update_features() if also dev->vlan_features might 9755 * have changed to allow the changes to be propagated to stacked 9756 * VLAN devices. 9757 */ 9758 void netdev_change_features(struct net_device *dev) 9759 { 9760 __netdev_update_features(dev); 9761 netdev_features_change(dev); 9762 } 9763 EXPORT_SYMBOL(netdev_change_features); 9764 9765 /** 9766 * netif_stacked_transfer_operstate - transfer operstate 9767 * @rootdev: the root or lower level device to transfer state from 9768 * @dev: the device to transfer operstate to 9769 * 9770 * Transfer operational state from root to device. This is normally 9771 * called when a stacking relationship exists between the root 9772 * device and the device(a leaf device). 9773 */ 9774 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9775 struct net_device *dev) 9776 { 9777 if (rootdev->operstate == IF_OPER_DORMANT) 9778 netif_dormant_on(dev); 9779 else 9780 netif_dormant_off(dev); 9781 9782 if (rootdev->operstate == IF_OPER_TESTING) 9783 netif_testing_on(dev); 9784 else 9785 netif_testing_off(dev); 9786 9787 if (netif_carrier_ok(rootdev)) 9788 netif_carrier_on(dev); 9789 else 9790 netif_carrier_off(dev); 9791 } 9792 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9793 9794 static int netif_alloc_rx_queues(struct net_device *dev) 9795 { 9796 unsigned int i, count = dev->num_rx_queues; 9797 struct netdev_rx_queue *rx; 9798 size_t sz = count * sizeof(*rx); 9799 int err = 0; 9800 9801 BUG_ON(count < 1); 9802 9803 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9804 if (!rx) 9805 return -ENOMEM; 9806 9807 dev->_rx = rx; 9808 9809 for (i = 0; i < count; i++) { 9810 rx[i].dev = dev; 9811 9812 /* XDP RX-queue setup */ 9813 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9814 if (err < 0) 9815 goto err_rxq_info; 9816 } 9817 return 0; 9818 9819 err_rxq_info: 9820 /* Rollback successful reg's and free other resources */ 9821 while (i--) 9822 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9823 kvfree(dev->_rx); 9824 dev->_rx = NULL; 9825 return err; 9826 } 9827 9828 static void netif_free_rx_queues(struct net_device *dev) 9829 { 9830 unsigned int i, count = dev->num_rx_queues; 9831 9832 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9833 if (!dev->_rx) 9834 return; 9835 9836 for (i = 0; i < count; i++) 9837 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9838 9839 kvfree(dev->_rx); 9840 } 9841 9842 static void netdev_init_one_queue(struct net_device *dev, 9843 struct netdev_queue *queue, void *_unused) 9844 { 9845 /* Initialize queue lock */ 9846 spin_lock_init(&queue->_xmit_lock); 9847 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9848 queue->xmit_lock_owner = -1; 9849 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9850 queue->dev = dev; 9851 #ifdef CONFIG_BQL 9852 dql_init(&queue->dql, HZ); 9853 #endif 9854 } 9855 9856 static void netif_free_tx_queues(struct net_device *dev) 9857 { 9858 kvfree(dev->_tx); 9859 } 9860 9861 static int netif_alloc_netdev_queues(struct net_device *dev) 9862 { 9863 unsigned int count = dev->num_tx_queues; 9864 struct netdev_queue *tx; 9865 size_t sz = count * sizeof(*tx); 9866 9867 if (count < 1 || count > 0xffff) 9868 return -EINVAL; 9869 9870 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9871 if (!tx) 9872 return -ENOMEM; 9873 9874 dev->_tx = tx; 9875 9876 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9877 spin_lock_init(&dev->tx_global_lock); 9878 9879 return 0; 9880 } 9881 9882 void netif_tx_stop_all_queues(struct net_device *dev) 9883 { 9884 unsigned int i; 9885 9886 for (i = 0; i < dev->num_tx_queues; i++) { 9887 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9888 9889 netif_tx_stop_queue(txq); 9890 } 9891 } 9892 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9893 9894 /** 9895 * register_netdevice - register a network device 9896 * @dev: device to register 9897 * 9898 * Take a completed network device structure and add it to the kernel 9899 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9900 * chain. 0 is returned on success. A negative errno code is returned 9901 * on a failure to set up the device, or if the name is a duplicate. 9902 * 9903 * Callers must hold the rtnl semaphore. You may want 9904 * register_netdev() instead of this. 9905 * 9906 * BUGS: 9907 * The locking appears insufficient to guarantee two parallel registers 9908 * will not get the same name. 9909 */ 9910 9911 int register_netdevice(struct net_device *dev) 9912 { 9913 int ret; 9914 struct net *net = dev_net(dev); 9915 9916 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9917 NETDEV_FEATURE_COUNT); 9918 BUG_ON(dev_boot_phase); 9919 ASSERT_RTNL(); 9920 9921 might_sleep(); 9922 9923 /* When net_device's are persistent, this will be fatal. */ 9924 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9925 BUG_ON(!net); 9926 9927 ret = ethtool_check_ops(dev->ethtool_ops); 9928 if (ret) 9929 return ret; 9930 9931 spin_lock_init(&dev->addr_list_lock); 9932 netdev_set_addr_lockdep_class(dev); 9933 9934 ret = dev_get_valid_name(net, dev, dev->name); 9935 if (ret < 0) 9936 goto out; 9937 9938 ret = -ENOMEM; 9939 dev->name_node = netdev_name_node_head_alloc(dev); 9940 if (!dev->name_node) 9941 goto out; 9942 9943 /* Init, if this function is available */ 9944 if (dev->netdev_ops->ndo_init) { 9945 ret = dev->netdev_ops->ndo_init(dev); 9946 if (ret) { 9947 if (ret > 0) 9948 ret = -EIO; 9949 goto err_free_name; 9950 } 9951 } 9952 9953 if (((dev->hw_features | dev->features) & 9954 NETIF_F_HW_VLAN_CTAG_FILTER) && 9955 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9956 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9957 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9958 ret = -EINVAL; 9959 goto err_uninit; 9960 } 9961 9962 ret = -EBUSY; 9963 if (!dev->ifindex) 9964 dev->ifindex = dev_new_index(net); 9965 else if (__dev_get_by_index(net, dev->ifindex)) 9966 goto err_uninit; 9967 9968 /* Transfer changeable features to wanted_features and enable 9969 * software offloads (GSO and GRO). 9970 */ 9971 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 9972 dev->features |= NETIF_F_SOFT_FEATURES; 9973 9974 if (dev->udp_tunnel_nic_info) { 9975 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9976 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9977 } 9978 9979 dev->wanted_features = dev->features & dev->hw_features; 9980 9981 if (!(dev->flags & IFF_LOOPBACK)) 9982 dev->hw_features |= NETIF_F_NOCACHE_COPY; 9983 9984 /* If IPv4 TCP segmentation offload is supported we should also 9985 * allow the device to enable segmenting the frame with the option 9986 * of ignoring a static IP ID value. This doesn't enable the 9987 * feature itself but allows the user to enable it later. 9988 */ 9989 if (dev->hw_features & NETIF_F_TSO) 9990 dev->hw_features |= NETIF_F_TSO_MANGLEID; 9991 if (dev->vlan_features & NETIF_F_TSO) 9992 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 9993 if (dev->mpls_features & NETIF_F_TSO) 9994 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 9995 if (dev->hw_enc_features & NETIF_F_TSO) 9996 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 9997 9998 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 9999 */ 10000 dev->vlan_features |= NETIF_F_HIGHDMA; 10001 10002 /* Make NETIF_F_SG inheritable to tunnel devices. 10003 */ 10004 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10005 10006 /* Make NETIF_F_SG inheritable to MPLS. 10007 */ 10008 dev->mpls_features |= NETIF_F_SG; 10009 10010 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10011 ret = notifier_to_errno(ret); 10012 if (ret) 10013 goto err_uninit; 10014 10015 ret = netdev_register_kobject(dev); 10016 if (ret) { 10017 dev->reg_state = NETREG_UNREGISTERED; 10018 goto err_uninit; 10019 } 10020 dev->reg_state = NETREG_REGISTERED; 10021 10022 __netdev_update_features(dev); 10023 10024 /* 10025 * Default initial state at registry is that the 10026 * device is present. 10027 */ 10028 10029 set_bit(__LINK_STATE_PRESENT, &dev->state); 10030 10031 linkwatch_init_dev(dev); 10032 10033 dev_init_scheduler(dev); 10034 10035 dev_hold_track(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10036 list_netdevice(dev); 10037 10038 add_device_randomness(dev->dev_addr, dev->addr_len); 10039 10040 /* If the device has permanent device address, driver should 10041 * set dev_addr and also addr_assign_type should be set to 10042 * NET_ADDR_PERM (default value). 10043 */ 10044 if (dev->addr_assign_type == NET_ADDR_PERM) 10045 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10046 10047 /* Notify protocols, that a new device appeared. */ 10048 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10049 ret = notifier_to_errno(ret); 10050 if (ret) { 10051 /* Expect explicit free_netdev() on failure */ 10052 dev->needs_free_netdev = false; 10053 unregister_netdevice_queue(dev, NULL); 10054 goto out; 10055 } 10056 /* 10057 * Prevent userspace races by waiting until the network 10058 * device is fully setup before sending notifications. 10059 */ 10060 if (!dev->rtnl_link_ops || 10061 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10062 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10063 10064 out: 10065 return ret; 10066 10067 err_uninit: 10068 if (dev->netdev_ops->ndo_uninit) 10069 dev->netdev_ops->ndo_uninit(dev); 10070 if (dev->priv_destructor) 10071 dev->priv_destructor(dev); 10072 err_free_name: 10073 netdev_name_node_free(dev->name_node); 10074 goto out; 10075 } 10076 EXPORT_SYMBOL(register_netdevice); 10077 10078 /** 10079 * init_dummy_netdev - init a dummy network device for NAPI 10080 * @dev: device to init 10081 * 10082 * This takes a network device structure and initialize the minimum 10083 * amount of fields so it can be used to schedule NAPI polls without 10084 * registering a full blown interface. This is to be used by drivers 10085 * that need to tie several hardware interfaces to a single NAPI 10086 * poll scheduler due to HW limitations. 10087 */ 10088 int init_dummy_netdev(struct net_device *dev) 10089 { 10090 /* Clear everything. Note we don't initialize spinlocks 10091 * are they aren't supposed to be taken by any of the 10092 * NAPI code and this dummy netdev is supposed to be 10093 * only ever used for NAPI polls 10094 */ 10095 memset(dev, 0, sizeof(struct net_device)); 10096 10097 /* make sure we BUG if trying to hit standard 10098 * register/unregister code path 10099 */ 10100 dev->reg_state = NETREG_DUMMY; 10101 10102 /* NAPI wants this */ 10103 INIT_LIST_HEAD(&dev->napi_list); 10104 10105 /* a dummy interface is started by default */ 10106 set_bit(__LINK_STATE_PRESENT, &dev->state); 10107 set_bit(__LINK_STATE_START, &dev->state); 10108 10109 /* napi_busy_loop stats accounting wants this */ 10110 dev_net_set(dev, &init_net); 10111 10112 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10113 * because users of this 'device' dont need to change 10114 * its refcount. 10115 */ 10116 10117 return 0; 10118 } 10119 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10120 10121 10122 /** 10123 * register_netdev - register a network device 10124 * @dev: device to register 10125 * 10126 * Take a completed network device structure and add it to the kernel 10127 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10128 * chain. 0 is returned on success. A negative errno code is returned 10129 * on a failure to set up the device, or if the name is a duplicate. 10130 * 10131 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10132 * and expands the device name if you passed a format string to 10133 * alloc_netdev. 10134 */ 10135 int register_netdev(struct net_device *dev) 10136 { 10137 int err; 10138 10139 if (rtnl_lock_killable()) 10140 return -EINTR; 10141 err = register_netdevice(dev); 10142 rtnl_unlock(); 10143 return err; 10144 } 10145 EXPORT_SYMBOL(register_netdev); 10146 10147 int netdev_refcnt_read(const struct net_device *dev) 10148 { 10149 #ifdef CONFIG_PCPU_DEV_REFCNT 10150 int i, refcnt = 0; 10151 10152 for_each_possible_cpu(i) 10153 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10154 return refcnt; 10155 #else 10156 return refcount_read(&dev->dev_refcnt); 10157 #endif 10158 } 10159 EXPORT_SYMBOL(netdev_refcnt_read); 10160 10161 int netdev_unregister_timeout_secs __read_mostly = 10; 10162 10163 #define WAIT_REFS_MIN_MSECS 1 10164 #define WAIT_REFS_MAX_MSECS 250 10165 /** 10166 * netdev_wait_allrefs_any - wait until all references are gone. 10167 * @list: list of net_devices to wait on 10168 * 10169 * This is called when unregistering network devices. 10170 * 10171 * Any protocol or device that holds a reference should register 10172 * for netdevice notification, and cleanup and put back the 10173 * reference if they receive an UNREGISTER event. 10174 * We can get stuck here if buggy protocols don't correctly 10175 * call dev_put. 10176 */ 10177 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10178 { 10179 unsigned long rebroadcast_time, warning_time; 10180 struct net_device *dev; 10181 int wait = 0; 10182 10183 rebroadcast_time = warning_time = jiffies; 10184 10185 list_for_each_entry(dev, list, todo_list) 10186 if (netdev_refcnt_read(dev) == 1) 10187 return dev; 10188 10189 while (true) { 10190 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10191 rtnl_lock(); 10192 10193 /* Rebroadcast unregister notification */ 10194 list_for_each_entry(dev, list, todo_list) 10195 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10196 10197 __rtnl_unlock(); 10198 rcu_barrier(); 10199 rtnl_lock(); 10200 10201 list_for_each_entry(dev, list, todo_list) 10202 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10203 &dev->state)) { 10204 /* We must not have linkwatch events 10205 * pending on unregister. If this 10206 * happens, we simply run the queue 10207 * unscheduled, resulting in a noop 10208 * for this device. 10209 */ 10210 linkwatch_run_queue(); 10211 break; 10212 } 10213 10214 __rtnl_unlock(); 10215 10216 rebroadcast_time = jiffies; 10217 } 10218 10219 if (!wait) { 10220 rcu_barrier(); 10221 wait = WAIT_REFS_MIN_MSECS; 10222 } else { 10223 msleep(wait); 10224 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10225 } 10226 10227 list_for_each_entry(dev, list, todo_list) 10228 if (netdev_refcnt_read(dev) == 1) 10229 return dev; 10230 10231 if (time_after(jiffies, warning_time + 10232 netdev_unregister_timeout_secs * HZ)) { 10233 list_for_each_entry(dev, list, todo_list) { 10234 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10235 dev->name, netdev_refcnt_read(dev)); 10236 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10237 } 10238 10239 warning_time = jiffies; 10240 } 10241 } 10242 } 10243 10244 /* The sequence is: 10245 * 10246 * rtnl_lock(); 10247 * ... 10248 * register_netdevice(x1); 10249 * register_netdevice(x2); 10250 * ... 10251 * unregister_netdevice(y1); 10252 * unregister_netdevice(y2); 10253 * ... 10254 * rtnl_unlock(); 10255 * free_netdev(y1); 10256 * free_netdev(y2); 10257 * 10258 * We are invoked by rtnl_unlock(). 10259 * This allows us to deal with problems: 10260 * 1) We can delete sysfs objects which invoke hotplug 10261 * without deadlocking with linkwatch via keventd. 10262 * 2) Since we run with the RTNL semaphore not held, we can sleep 10263 * safely in order to wait for the netdev refcnt to drop to zero. 10264 * 10265 * We must not return until all unregister events added during 10266 * the interval the lock was held have been completed. 10267 */ 10268 void netdev_run_todo(void) 10269 { 10270 struct net_device *dev, *tmp; 10271 struct list_head list; 10272 #ifdef CONFIG_LOCKDEP 10273 struct list_head unlink_list; 10274 10275 list_replace_init(&net_unlink_list, &unlink_list); 10276 10277 while (!list_empty(&unlink_list)) { 10278 struct net_device *dev = list_first_entry(&unlink_list, 10279 struct net_device, 10280 unlink_list); 10281 list_del_init(&dev->unlink_list); 10282 dev->nested_level = dev->lower_level - 1; 10283 } 10284 #endif 10285 10286 /* Snapshot list, allow later requests */ 10287 list_replace_init(&net_todo_list, &list); 10288 10289 __rtnl_unlock(); 10290 10291 /* Wait for rcu callbacks to finish before next phase */ 10292 if (!list_empty(&list)) 10293 rcu_barrier(); 10294 10295 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10296 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10297 netdev_WARN(dev, "run_todo but not unregistering\n"); 10298 list_del(&dev->todo_list); 10299 continue; 10300 } 10301 10302 dev->reg_state = NETREG_UNREGISTERED; 10303 linkwatch_forget_dev(dev); 10304 } 10305 10306 while (!list_empty(&list)) { 10307 dev = netdev_wait_allrefs_any(&list); 10308 list_del(&dev->todo_list); 10309 10310 /* paranoia */ 10311 BUG_ON(netdev_refcnt_read(dev) != 1); 10312 BUG_ON(!list_empty(&dev->ptype_all)); 10313 BUG_ON(!list_empty(&dev->ptype_specific)); 10314 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10315 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10316 #if IS_ENABLED(CONFIG_DECNET) 10317 WARN_ON(dev->dn_ptr); 10318 #endif 10319 if (dev->priv_destructor) 10320 dev->priv_destructor(dev); 10321 if (dev->needs_free_netdev) 10322 free_netdev(dev); 10323 10324 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10325 wake_up(&netdev_unregistering_wq); 10326 10327 /* Free network device */ 10328 kobject_put(&dev->dev.kobj); 10329 } 10330 } 10331 10332 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10333 * all the same fields in the same order as net_device_stats, with only 10334 * the type differing, but rtnl_link_stats64 may have additional fields 10335 * at the end for newer counters. 10336 */ 10337 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10338 const struct net_device_stats *netdev_stats) 10339 { 10340 #if BITS_PER_LONG == 64 10341 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10342 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10343 /* zero out counters that only exist in rtnl_link_stats64 */ 10344 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10345 sizeof(*stats64) - sizeof(*netdev_stats)); 10346 #else 10347 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10348 const unsigned long *src = (const unsigned long *)netdev_stats; 10349 u64 *dst = (u64 *)stats64; 10350 10351 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10352 for (i = 0; i < n; i++) 10353 dst[i] = src[i]; 10354 /* zero out counters that only exist in rtnl_link_stats64 */ 10355 memset((char *)stats64 + n * sizeof(u64), 0, 10356 sizeof(*stats64) - n * sizeof(u64)); 10357 #endif 10358 } 10359 EXPORT_SYMBOL(netdev_stats_to_stats64); 10360 10361 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10362 { 10363 struct net_device_core_stats __percpu *p; 10364 10365 p = alloc_percpu_gfp(struct net_device_core_stats, 10366 GFP_ATOMIC | __GFP_NOWARN); 10367 10368 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10369 free_percpu(p); 10370 10371 /* This READ_ONCE() pairs with the cmpxchg() above */ 10372 return READ_ONCE(dev->core_stats); 10373 } 10374 EXPORT_SYMBOL(netdev_core_stats_alloc); 10375 10376 /** 10377 * dev_get_stats - get network device statistics 10378 * @dev: device to get statistics from 10379 * @storage: place to store stats 10380 * 10381 * Get network statistics from device. Return @storage. 10382 * The device driver may provide its own method by setting 10383 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10384 * otherwise the internal statistics structure is used. 10385 */ 10386 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10387 struct rtnl_link_stats64 *storage) 10388 { 10389 const struct net_device_ops *ops = dev->netdev_ops; 10390 const struct net_device_core_stats __percpu *p; 10391 10392 if (ops->ndo_get_stats64) { 10393 memset(storage, 0, sizeof(*storage)); 10394 ops->ndo_get_stats64(dev, storage); 10395 } else if (ops->ndo_get_stats) { 10396 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10397 } else { 10398 netdev_stats_to_stats64(storage, &dev->stats); 10399 } 10400 10401 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10402 p = READ_ONCE(dev->core_stats); 10403 if (p) { 10404 const struct net_device_core_stats *core_stats; 10405 int i; 10406 10407 for_each_possible_cpu(i) { 10408 core_stats = per_cpu_ptr(p, i); 10409 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10410 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10411 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10412 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10413 } 10414 } 10415 return storage; 10416 } 10417 EXPORT_SYMBOL(dev_get_stats); 10418 10419 /** 10420 * dev_fetch_sw_netstats - get per-cpu network device statistics 10421 * @s: place to store stats 10422 * @netstats: per-cpu network stats to read from 10423 * 10424 * Read per-cpu network statistics and populate the related fields in @s. 10425 */ 10426 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10427 const struct pcpu_sw_netstats __percpu *netstats) 10428 { 10429 int cpu; 10430 10431 for_each_possible_cpu(cpu) { 10432 const struct pcpu_sw_netstats *stats; 10433 struct pcpu_sw_netstats tmp; 10434 unsigned int start; 10435 10436 stats = per_cpu_ptr(netstats, cpu); 10437 do { 10438 start = u64_stats_fetch_begin_irq(&stats->syncp); 10439 tmp.rx_packets = stats->rx_packets; 10440 tmp.rx_bytes = stats->rx_bytes; 10441 tmp.tx_packets = stats->tx_packets; 10442 tmp.tx_bytes = stats->tx_bytes; 10443 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10444 10445 s->rx_packets += tmp.rx_packets; 10446 s->rx_bytes += tmp.rx_bytes; 10447 s->tx_packets += tmp.tx_packets; 10448 s->tx_bytes += tmp.tx_bytes; 10449 } 10450 } 10451 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10452 10453 /** 10454 * dev_get_tstats64 - ndo_get_stats64 implementation 10455 * @dev: device to get statistics from 10456 * @s: place to store stats 10457 * 10458 * Populate @s from dev->stats and dev->tstats. Can be used as 10459 * ndo_get_stats64() callback. 10460 */ 10461 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10462 { 10463 netdev_stats_to_stats64(s, &dev->stats); 10464 dev_fetch_sw_netstats(s, dev->tstats); 10465 } 10466 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10467 10468 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10469 { 10470 struct netdev_queue *queue = dev_ingress_queue(dev); 10471 10472 #ifdef CONFIG_NET_CLS_ACT 10473 if (queue) 10474 return queue; 10475 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10476 if (!queue) 10477 return NULL; 10478 netdev_init_one_queue(dev, queue, NULL); 10479 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10480 queue->qdisc_sleeping = &noop_qdisc; 10481 rcu_assign_pointer(dev->ingress_queue, queue); 10482 #endif 10483 return queue; 10484 } 10485 10486 static const struct ethtool_ops default_ethtool_ops; 10487 10488 void netdev_set_default_ethtool_ops(struct net_device *dev, 10489 const struct ethtool_ops *ops) 10490 { 10491 if (dev->ethtool_ops == &default_ethtool_ops) 10492 dev->ethtool_ops = ops; 10493 } 10494 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10495 10496 void netdev_freemem(struct net_device *dev) 10497 { 10498 char *addr = (char *)dev - dev->padded; 10499 10500 kvfree(addr); 10501 } 10502 10503 /** 10504 * alloc_netdev_mqs - allocate network device 10505 * @sizeof_priv: size of private data to allocate space for 10506 * @name: device name format string 10507 * @name_assign_type: origin of device name 10508 * @setup: callback to initialize device 10509 * @txqs: the number of TX subqueues to allocate 10510 * @rxqs: the number of RX subqueues to allocate 10511 * 10512 * Allocates a struct net_device with private data area for driver use 10513 * and performs basic initialization. Also allocates subqueue structs 10514 * for each queue on the device. 10515 */ 10516 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10517 unsigned char name_assign_type, 10518 void (*setup)(struct net_device *), 10519 unsigned int txqs, unsigned int rxqs) 10520 { 10521 struct net_device *dev; 10522 unsigned int alloc_size; 10523 struct net_device *p; 10524 10525 BUG_ON(strlen(name) >= sizeof(dev->name)); 10526 10527 if (txqs < 1) { 10528 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10529 return NULL; 10530 } 10531 10532 if (rxqs < 1) { 10533 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10534 return NULL; 10535 } 10536 10537 alloc_size = sizeof(struct net_device); 10538 if (sizeof_priv) { 10539 /* ensure 32-byte alignment of private area */ 10540 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10541 alloc_size += sizeof_priv; 10542 } 10543 /* ensure 32-byte alignment of whole construct */ 10544 alloc_size += NETDEV_ALIGN - 1; 10545 10546 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10547 if (!p) 10548 return NULL; 10549 10550 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10551 dev->padded = (char *)dev - (char *)p; 10552 10553 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10554 #ifdef CONFIG_PCPU_DEV_REFCNT 10555 dev->pcpu_refcnt = alloc_percpu(int); 10556 if (!dev->pcpu_refcnt) 10557 goto free_dev; 10558 __dev_hold(dev); 10559 #else 10560 refcount_set(&dev->dev_refcnt, 1); 10561 #endif 10562 10563 if (dev_addr_init(dev)) 10564 goto free_pcpu; 10565 10566 dev_mc_init(dev); 10567 dev_uc_init(dev); 10568 10569 dev_net_set(dev, &init_net); 10570 10571 dev->gso_max_size = GSO_MAX_SIZE; 10572 dev->gso_max_segs = GSO_MAX_SEGS; 10573 dev->gro_max_size = GRO_MAX_SIZE; 10574 dev->upper_level = 1; 10575 dev->lower_level = 1; 10576 #ifdef CONFIG_LOCKDEP 10577 dev->nested_level = 0; 10578 INIT_LIST_HEAD(&dev->unlink_list); 10579 #endif 10580 10581 INIT_LIST_HEAD(&dev->napi_list); 10582 INIT_LIST_HEAD(&dev->unreg_list); 10583 INIT_LIST_HEAD(&dev->close_list); 10584 INIT_LIST_HEAD(&dev->link_watch_list); 10585 INIT_LIST_HEAD(&dev->adj_list.upper); 10586 INIT_LIST_HEAD(&dev->adj_list.lower); 10587 INIT_LIST_HEAD(&dev->ptype_all); 10588 INIT_LIST_HEAD(&dev->ptype_specific); 10589 INIT_LIST_HEAD(&dev->net_notifier_list); 10590 #ifdef CONFIG_NET_SCHED 10591 hash_init(dev->qdisc_hash); 10592 #endif 10593 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10594 setup(dev); 10595 10596 if (!dev->tx_queue_len) { 10597 dev->priv_flags |= IFF_NO_QUEUE; 10598 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10599 } 10600 10601 dev->num_tx_queues = txqs; 10602 dev->real_num_tx_queues = txqs; 10603 if (netif_alloc_netdev_queues(dev)) 10604 goto free_all; 10605 10606 dev->num_rx_queues = rxqs; 10607 dev->real_num_rx_queues = rxqs; 10608 if (netif_alloc_rx_queues(dev)) 10609 goto free_all; 10610 10611 strcpy(dev->name, name); 10612 dev->name_assign_type = name_assign_type; 10613 dev->group = INIT_NETDEV_GROUP; 10614 if (!dev->ethtool_ops) 10615 dev->ethtool_ops = &default_ethtool_ops; 10616 10617 nf_hook_netdev_init(dev); 10618 10619 return dev; 10620 10621 free_all: 10622 free_netdev(dev); 10623 return NULL; 10624 10625 free_pcpu: 10626 #ifdef CONFIG_PCPU_DEV_REFCNT 10627 free_percpu(dev->pcpu_refcnt); 10628 free_dev: 10629 #endif 10630 netdev_freemem(dev); 10631 return NULL; 10632 } 10633 EXPORT_SYMBOL(alloc_netdev_mqs); 10634 10635 /** 10636 * free_netdev - free network device 10637 * @dev: device 10638 * 10639 * This function does the last stage of destroying an allocated device 10640 * interface. The reference to the device object is released. If this 10641 * is the last reference then it will be freed.Must be called in process 10642 * context. 10643 */ 10644 void free_netdev(struct net_device *dev) 10645 { 10646 struct napi_struct *p, *n; 10647 10648 might_sleep(); 10649 10650 /* When called immediately after register_netdevice() failed the unwind 10651 * handling may still be dismantling the device. Handle that case by 10652 * deferring the free. 10653 */ 10654 if (dev->reg_state == NETREG_UNREGISTERING) { 10655 ASSERT_RTNL(); 10656 dev->needs_free_netdev = true; 10657 return; 10658 } 10659 10660 netif_free_tx_queues(dev); 10661 netif_free_rx_queues(dev); 10662 10663 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10664 10665 /* Flush device addresses */ 10666 dev_addr_flush(dev); 10667 10668 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10669 netif_napi_del(p); 10670 10671 ref_tracker_dir_exit(&dev->refcnt_tracker); 10672 #ifdef CONFIG_PCPU_DEV_REFCNT 10673 free_percpu(dev->pcpu_refcnt); 10674 dev->pcpu_refcnt = NULL; 10675 #endif 10676 free_percpu(dev->core_stats); 10677 dev->core_stats = NULL; 10678 free_percpu(dev->xdp_bulkq); 10679 dev->xdp_bulkq = NULL; 10680 10681 /* Compatibility with error handling in drivers */ 10682 if (dev->reg_state == NETREG_UNINITIALIZED) { 10683 netdev_freemem(dev); 10684 return; 10685 } 10686 10687 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10688 dev->reg_state = NETREG_RELEASED; 10689 10690 /* will free via device release */ 10691 put_device(&dev->dev); 10692 } 10693 EXPORT_SYMBOL(free_netdev); 10694 10695 /** 10696 * synchronize_net - Synchronize with packet receive processing 10697 * 10698 * Wait for packets currently being received to be done. 10699 * Does not block later packets from starting. 10700 */ 10701 void synchronize_net(void) 10702 { 10703 might_sleep(); 10704 if (rtnl_is_locked()) 10705 synchronize_rcu_expedited(); 10706 else 10707 synchronize_rcu(); 10708 } 10709 EXPORT_SYMBOL(synchronize_net); 10710 10711 /** 10712 * unregister_netdevice_queue - remove device from the kernel 10713 * @dev: device 10714 * @head: list 10715 * 10716 * This function shuts down a device interface and removes it 10717 * from the kernel tables. 10718 * If head not NULL, device is queued to be unregistered later. 10719 * 10720 * Callers must hold the rtnl semaphore. You may want 10721 * unregister_netdev() instead of this. 10722 */ 10723 10724 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10725 { 10726 ASSERT_RTNL(); 10727 10728 if (head) { 10729 list_move_tail(&dev->unreg_list, head); 10730 } else { 10731 LIST_HEAD(single); 10732 10733 list_add(&dev->unreg_list, &single); 10734 unregister_netdevice_many(&single); 10735 } 10736 } 10737 EXPORT_SYMBOL(unregister_netdevice_queue); 10738 10739 /** 10740 * unregister_netdevice_many - unregister many devices 10741 * @head: list of devices 10742 * 10743 * Note: As most callers use a stack allocated list_head, 10744 * we force a list_del() to make sure stack wont be corrupted later. 10745 */ 10746 void unregister_netdevice_many(struct list_head *head) 10747 { 10748 struct net_device *dev, *tmp; 10749 LIST_HEAD(close_head); 10750 10751 BUG_ON(dev_boot_phase); 10752 ASSERT_RTNL(); 10753 10754 if (list_empty(head)) 10755 return; 10756 10757 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10758 /* Some devices call without registering 10759 * for initialization unwind. Remove those 10760 * devices and proceed with the remaining. 10761 */ 10762 if (dev->reg_state == NETREG_UNINITIALIZED) { 10763 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10764 dev->name, dev); 10765 10766 WARN_ON(1); 10767 list_del(&dev->unreg_list); 10768 continue; 10769 } 10770 dev->dismantle = true; 10771 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10772 } 10773 10774 /* If device is running, close it first. */ 10775 list_for_each_entry(dev, head, unreg_list) 10776 list_add_tail(&dev->close_list, &close_head); 10777 dev_close_many(&close_head, true); 10778 10779 list_for_each_entry(dev, head, unreg_list) { 10780 /* And unlink it from device chain. */ 10781 unlist_netdevice(dev); 10782 10783 dev->reg_state = NETREG_UNREGISTERING; 10784 } 10785 flush_all_backlogs(); 10786 10787 synchronize_net(); 10788 10789 list_for_each_entry(dev, head, unreg_list) { 10790 struct sk_buff *skb = NULL; 10791 10792 /* Shutdown queueing discipline. */ 10793 dev_shutdown(dev); 10794 10795 dev_xdp_uninstall(dev); 10796 10797 netdev_offload_xstats_disable_all(dev); 10798 10799 /* Notify protocols, that we are about to destroy 10800 * this device. They should clean all the things. 10801 */ 10802 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10803 10804 if (!dev->rtnl_link_ops || 10805 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10806 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10807 GFP_KERNEL, NULL, 0); 10808 10809 /* 10810 * Flush the unicast and multicast chains 10811 */ 10812 dev_uc_flush(dev); 10813 dev_mc_flush(dev); 10814 10815 netdev_name_node_alt_flush(dev); 10816 netdev_name_node_free(dev->name_node); 10817 10818 if (dev->netdev_ops->ndo_uninit) 10819 dev->netdev_ops->ndo_uninit(dev); 10820 10821 if (skb) 10822 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10823 10824 /* Notifier chain MUST detach us all upper devices. */ 10825 WARN_ON(netdev_has_any_upper_dev(dev)); 10826 WARN_ON(netdev_has_any_lower_dev(dev)); 10827 10828 /* Remove entries from kobject tree */ 10829 netdev_unregister_kobject(dev); 10830 #ifdef CONFIG_XPS 10831 /* Remove XPS queueing entries */ 10832 netif_reset_xps_queues_gt(dev, 0); 10833 #endif 10834 } 10835 10836 synchronize_net(); 10837 10838 list_for_each_entry(dev, head, unreg_list) { 10839 dev_put_track(dev, &dev->dev_registered_tracker); 10840 net_set_todo(dev); 10841 } 10842 10843 list_del(head); 10844 } 10845 EXPORT_SYMBOL(unregister_netdevice_many); 10846 10847 /** 10848 * unregister_netdev - remove device from the kernel 10849 * @dev: device 10850 * 10851 * This function shuts down a device interface and removes it 10852 * from the kernel tables. 10853 * 10854 * This is just a wrapper for unregister_netdevice that takes 10855 * the rtnl semaphore. In general you want to use this and not 10856 * unregister_netdevice. 10857 */ 10858 void unregister_netdev(struct net_device *dev) 10859 { 10860 rtnl_lock(); 10861 unregister_netdevice(dev); 10862 rtnl_unlock(); 10863 } 10864 EXPORT_SYMBOL(unregister_netdev); 10865 10866 /** 10867 * __dev_change_net_namespace - move device to different nethost namespace 10868 * @dev: device 10869 * @net: network namespace 10870 * @pat: If not NULL name pattern to try if the current device name 10871 * is already taken in the destination network namespace. 10872 * @new_ifindex: If not zero, specifies device index in the target 10873 * namespace. 10874 * 10875 * This function shuts down a device interface and moves it 10876 * to a new network namespace. On success 0 is returned, on 10877 * a failure a netagive errno code is returned. 10878 * 10879 * Callers must hold the rtnl semaphore. 10880 */ 10881 10882 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10883 const char *pat, int new_ifindex) 10884 { 10885 struct net *net_old = dev_net(dev); 10886 int err, new_nsid; 10887 10888 ASSERT_RTNL(); 10889 10890 /* Don't allow namespace local devices to be moved. */ 10891 err = -EINVAL; 10892 if (dev->features & NETIF_F_NETNS_LOCAL) 10893 goto out; 10894 10895 /* Ensure the device has been registrered */ 10896 if (dev->reg_state != NETREG_REGISTERED) 10897 goto out; 10898 10899 /* Get out if there is nothing todo */ 10900 err = 0; 10901 if (net_eq(net_old, net)) 10902 goto out; 10903 10904 /* Pick the destination device name, and ensure 10905 * we can use it in the destination network namespace. 10906 */ 10907 err = -EEXIST; 10908 if (netdev_name_in_use(net, dev->name)) { 10909 /* We get here if we can't use the current device name */ 10910 if (!pat) 10911 goto out; 10912 err = dev_get_valid_name(net, dev, pat); 10913 if (err < 0) 10914 goto out; 10915 } 10916 10917 /* Check that new_ifindex isn't used yet. */ 10918 err = -EBUSY; 10919 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10920 goto out; 10921 10922 /* 10923 * And now a mini version of register_netdevice unregister_netdevice. 10924 */ 10925 10926 /* If device is running close it first. */ 10927 dev_close(dev); 10928 10929 /* And unlink it from device chain */ 10930 unlist_netdevice(dev); 10931 10932 synchronize_net(); 10933 10934 /* Shutdown queueing discipline. */ 10935 dev_shutdown(dev); 10936 10937 /* Notify protocols, that we are about to destroy 10938 * this device. They should clean all the things. 10939 * 10940 * Note that dev->reg_state stays at NETREG_REGISTERED. 10941 * This is wanted because this way 8021q and macvlan know 10942 * the device is just moving and can keep their slaves up. 10943 */ 10944 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10945 rcu_barrier(); 10946 10947 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10948 /* If there is an ifindex conflict assign a new one */ 10949 if (!new_ifindex) { 10950 if (__dev_get_by_index(net, dev->ifindex)) 10951 new_ifindex = dev_new_index(net); 10952 else 10953 new_ifindex = dev->ifindex; 10954 } 10955 10956 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10957 new_ifindex); 10958 10959 /* 10960 * Flush the unicast and multicast chains 10961 */ 10962 dev_uc_flush(dev); 10963 dev_mc_flush(dev); 10964 10965 /* Send a netdev-removed uevent to the old namespace */ 10966 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10967 netdev_adjacent_del_links(dev); 10968 10969 /* Move per-net netdevice notifiers that are following the netdevice */ 10970 move_netdevice_notifiers_dev_net(dev, net); 10971 10972 /* Actually switch the network namespace */ 10973 dev_net_set(dev, net); 10974 dev->ifindex = new_ifindex; 10975 10976 /* Send a netdev-add uevent to the new namespace */ 10977 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10978 netdev_adjacent_add_links(dev); 10979 10980 /* Fixup kobjects */ 10981 err = device_rename(&dev->dev, dev->name); 10982 WARN_ON(err); 10983 10984 /* Adapt owner in case owning user namespace of target network 10985 * namespace is different from the original one. 10986 */ 10987 err = netdev_change_owner(dev, net_old, net); 10988 WARN_ON(err); 10989 10990 /* Add the device back in the hashes */ 10991 list_netdevice(dev); 10992 10993 /* Notify protocols, that a new device appeared. */ 10994 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10995 10996 /* 10997 * Prevent userspace races by waiting until the network 10998 * device is fully setup before sending notifications. 10999 */ 11000 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11001 11002 synchronize_net(); 11003 err = 0; 11004 out: 11005 return err; 11006 } 11007 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11008 11009 static int dev_cpu_dead(unsigned int oldcpu) 11010 { 11011 struct sk_buff **list_skb; 11012 struct sk_buff *skb; 11013 unsigned int cpu; 11014 struct softnet_data *sd, *oldsd, *remsd = NULL; 11015 11016 local_irq_disable(); 11017 cpu = smp_processor_id(); 11018 sd = &per_cpu(softnet_data, cpu); 11019 oldsd = &per_cpu(softnet_data, oldcpu); 11020 11021 /* Find end of our completion_queue. */ 11022 list_skb = &sd->completion_queue; 11023 while (*list_skb) 11024 list_skb = &(*list_skb)->next; 11025 /* Append completion queue from offline CPU. */ 11026 *list_skb = oldsd->completion_queue; 11027 oldsd->completion_queue = NULL; 11028 11029 /* Append output queue from offline CPU. */ 11030 if (oldsd->output_queue) { 11031 *sd->output_queue_tailp = oldsd->output_queue; 11032 sd->output_queue_tailp = oldsd->output_queue_tailp; 11033 oldsd->output_queue = NULL; 11034 oldsd->output_queue_tailp = &oldsd->output_queue; 11035 } 11036 /* Append NAPI poll list from offline CPU, with one exception : 11037 * process_backlog() must be called by cpu owning percpu backlog. 11038 * We properly handle process_queue & input_pkt_queue later. 11039 */ 11040 while (!list_empty(&oldsd->poll_list)) { 11041 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11042 struct napi_struct, 11043 poll_list); 11044 11045 list_del_init(&napi->poll_list); 11046 if (napi->poll == process_backlog) 11047 napi->state = 0; 11048 else 11049 ____napi_schedule(sd, napi); 11050 } 11051 11052 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11053 local_irq_enable(); 11054 11055 #ifdef CONFIG_RPS 11056 remsd = oldsd->rps_ipi_list; 11057 oldsd->rps_ipi_list = NULL; 11058 #endif 11059 /* send out pending IPI's on offline CPU */ 11060 net_rps_send_ipi(remsd); 11061 11062 /* Process offline CPU's input_pkt_queue */ 11063 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11064 netif_rx(skb); 11065 input_queue_head_incr(oldsd); 11066 } 11067 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11068 netif_rx(skb); 11069 input_queue_head_incr(oldsd); 11070 } 11071 11072 return 0; 11073 } 11074 11075 /** 11076 * netdev_increment_features - increment feature set by one 11077 * @all: current feature set 11078 * @one: new feature set 11079 * @mask: mask feature set 11080 * 11081 * Computes a new feature set after adding a device with feature set 11082 * @one to the master device with current feature set @all. Will not 11083 * enable anything that is off in @mask. Returns the new feature set. 11084 */ 11085 netdev_features_t netdev_increment_features(netdev_features_t all, 11086 netdev_features_t one, netdev_features_t mask) 11087 { 11088 if (mask & NETIF_F_HW_CSUM) 11089 mask |= NETIF_F_CSUM_MASK; 11090 mask |= NETIF_F_VLAN_CHALLENGED; 11091 11092 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11093 all &= one | ~NETIF_F_ALL_FOR_ALL; 11094 11095 /* If one device supports hw checksumming, set for all. */ 11096 if (all & NETIF_F_HW_CSUM) 11097 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11098 11099 return all; 11100 } 11101 EXPORT_SYMBOL(netdev_increment_features); 11102 11103 static struct hlist_head * __net_init netdev_create_hash(void) 11104 { 11105 int i; 11106 struct hlist_head *hash; 11107 11108 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11109 if (hash != NULL) 11110 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11111 INIT_HLIST_HEAD(&hash[i]); 11112 11113 return hash; 11114 } 11115 11116 /* Initialize per network namespace state */ 11117 static int __net_init netdev_init(struct net *net) 11118 { 11119 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11120 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11121 11122 INIT_LIST_HEAD(&net->dev_base_head); 11123 11124 net->dev_name_head = netdev_create_hash(); 11125 if (net->dev_name_head == NULL) 11126 goto err_name; 11127 11128 net->dev_index_head = netdev_create_hash(); 11129 if (net->dev_index_head == NULL) 11130 goto err_idx; 11131 11132 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11133 11134 return 0; 11135 11136 err_idx: 11137 kfree(net->dev_name_head); 11138 err_name: 11139 return -ENOMEM; 11140 } 11141 11142 /** 11143 * netdev_drivername - network driver for the device 11144 * @dev: network device 11145 * 11146 * Determine network driver for device. 11147 */ 11148 const char *netdev_drivername(const struct net_device *dev) 11149 { 11150 const struct device_driver *driver; 11151 const struct device *parent; 11152 const char *empty = ""; 11153 11154 parent = dev->dev.parent; 11155 if (!parent) 11156 return empty; 11157 11158 driver = parent->driver; 11159 if (driver && driver->name) 11160 return driver->name; 11161 return empty; 11162 } 11163 11164 static void __netdev_printk(const char *level, const struct net_device *dev, 11165 struct va_format *vaf) 11166 { 11167 if (dev && dev->dev.parent) { 11168 dev_printk_emit(level[1] - '0', 11169 dev->dev.parent, 11170 "%s %s %s%s: %pV", 11171 dev_driver_string(dev->dev.parent), 11172 dev_name(dev->dev.parent), 11173 netdev_name(dev), netdev_reg_state(dev), 11174 vaf); 11175 } else if (dev) { 11176 printk("%s%s%s: %pV", 11177 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11178 } else { 11179 printk("%s(NULL net_device): %pV", level, vaf); 11180 } 11181 } 11182 11183 void netdev_printk(const char *level, const struct net_device *dev, 11184 const char *format, ...) 11185 { 11186 struct va_format vaf; 11187 va_list args; 11188 11189 va_start(args, format); 11190 11191 vaf.fmt = format; 11192 vaf.va = &args; 11193 11194 __netdev_printk(level, dev, &vaf); 11195 11196 va_end(args); 11197 } 11198 EXPORT_SYMBOL(netdev_printk); 11199 11200 #define define_netdev_printk_level(func, level) \ 11201 void func(const struct net_device *dev, const char *fmt, ...) \ 11202 { \ 11203 struct va_format vaf; \ 11204 va_list args; \ 11205 \ 11206 va_start(args, fmt); \ 11207 \ 11208 vaf.fmt = fmt; \ 11209 vaf.va = &args; \ 11210 \ 11211 __netdev_printk(level, dev, &vaf); \ 11212 \ 11213 va_end(args); \ 11214 } \ 11215 EXPORT_SYMBOL(func); 11216 11217 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11218 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11219 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11220 define_netdev_printk_level(netdev_err, KERN_ERR); 11221 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11222 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11223 define_netdev_printk_level(netdev_info, KERN_INFO); 11224 11225 static void __net_exit netdev_exit(struct net *net) 11226 { 11227 kfree(net->dev_name_head); 11228 kfree(net->dev_index_head); 11229 if (net != &init_net) 11230 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11231 } 11232 11233 static struct pernet_operations __net_initdata netdev_net_ops = { 11234 .init = netdev_init, 11235 .exit = netdev_exit, 11236 }; 11237 11238 static void __net_exit default_device_exit_net(struct net *net) 11239 { 11240 struct net_device *dev, *aux; 11241 /* 11242 * Push all migratable network devices back to the 11243 * initial network namespace 11244 */ 11245 ASSERT_RTNL(); 11246 for_each_netdev_safe(net, dev, aux) { 11247 int err; 11248 char fb_name[IFNAMSIZ]; 11249 11250 /* Ignore unmoveable devices (i.e. loopback) */ 11251 if (dev->features & NETIF_F_NETNS_LOCAL) 11252 continue; 11253 11254 /* Leave virtual devices for the generic cleanup */ 11255 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11256 continue; 11257 11258 /* Push remaining network devices to init_net */ 11259 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11260 if (netdev_name_in_use(&init_net, fb_name)) 11261 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11262 err = dev_change_net_namespace(dev, &init_net, fb_name); 11263 if (err) { 11264 pr_emerg("%s: failed to move %s to init_net: %d\n", 11265 __func__, dev->name, err); 11266 BUG(); 11267 } 11268 } 11269 } 11270 11271 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11272 { 11273 /* At exit all network devices most be removed from a network 11274 * namespace. Do this in the reverse order of registration. 11275 * Do this across as many network namespaces as possible to 11276 * improve batching efficiency. 11277 */ 11278 struct net_device *dev; 11279 struct net *net; 11280 LIST_HEAD(dev_kill_list); 11281 11282 rtnl_lock(); 11283 list_for_each_entry(net, net_list, exit_list) { 11284 default_device_exit_net(net); 11285 cond_resched(); 11286 } 11287 11288 list_for_each_entry(net, net_list, exit_list) { 11289 for_each_netdev_reverse(net, dev) { 11290 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11291 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11292 else 11293 unregister_netdevice_queue(dev, &dev_kill_list); 11294 } 11295 } 11296 unregister_netdevice_many(&dev_kill_list); 11297 rtnl_unlock(); 11298 } 11299 11300 static struct pernet_operations __net_initdata default_device_ops = { 11301 .exit_batch = default_device_exit_batch, 11302 }; 11303 11304 /* 11305 * Initialize the DEV module. At boot time this walks the device list and 11306 * unhooks any devices that fail to initialise (normally hardware not 11307 * present) and leaves us with a valid list of present and active devices. 11308 * 11309 */ 11310 11311 /* 11312 * This is called single threaded during boot, so no need 11313 * to take the rtnl semaphore. 11314 */ 11315 static int __init net_dev_init(void) 11316 { 11317 int i, rc = -ENOMEM; 11318 11319 BUG_ON(!dev_boot_phase); 11320 11321 if (dev_proc_init()) 11322 goto out; 11323 11324 if (netdev_kobject_init()) 11325 goto out; 11326 11327 INIT_LIST_HEAD(&ptype_all); 11328 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11329 INIT_LIST_HEAD(&ptype_base[i]); 11330 11331 if (register_pernet_subsys(&netdev_net_ops)) 11332 goto out; 11333 11334 /* 11335 * Initialise the packet receive queues. 11336 */ 11337 11338 for_each_possible_cpu(i) { 11339 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11340 struct softnet_data *sd = &per_cpu(softnet_data, i); 11341 11342 INIT_WORK(flush, flush_backlog); 11343 11344 skb_queue_head_init(&sd->input_pkt_queue); 11345 skb_queue_head_init(&sd->process_queue); 11346 #ifdef CONFIG_XFRM_OFFLOAD 11347 skb_queue_head_init(&sd->xfrm_backlog); 11348 #endif 11349 INIT_LIST_HEAD(&sd->poll_list); 11350 sd->output_queue_tailp = &sd->output_queue; 11351 #ifdef CONFIG_RPS 11352 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11353 sd->cpu = i; 11354 #endif 11355 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, NULL); 11356 spin_lock_init(&sd->defer_lock); 11357 11358 init_gro_hash(&sd->backlog); 11359 sd->backlog.poll = process_backlog; 11360 sd->backlog.weight = weight_p; 11361 } 11362 11363 dev_boot_phase = 0; 11364 11365 /* The loopback device is special if any other network devices 11366 * is present in a network namespace the loopback device must 11367 * be present. Since we now dynamically allocate and free the 11368 * loopback device ensure this invariant is maintained by 11369 * keeping the loopback device as the first device on the 11370 * list of network devices. Ensuring the loopback devices 11371 * is the first device that appears and the last network device 11372 * that disappears. 11373 */ 11374 if (register_pernet_device(&loopback_net_ops)) 11375 goto out; 11376 11377 if (register_pernet_device(&default_device_ops)) 11378 goto out; 11379 11380 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11381 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11382 11383 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11384 NULL, dev_cpu_dead); 11385 WARN_ON(rc < 0); 11386 rc = 0; 11387 out: 11388 return rc; 11389 } 11390 11391 subsys_initcall(net_dev_init); 11392