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 <net/tcx.h> 111 #include <linux/highmem.h> 112 #include <linux/init.h> 113 #include <linux/module.h> 114 #include <linux/netpoll.h> 115 #include <linux/rcupdate.h> 116 #include <linux/delay.h> 117 #include <net/iw_handler.h> 118 #include <asm/current.h> 119 #include <linux/audit.h> 120 #include <linux/dmaengine.h> 121 #include <linux/err.h> 122 #include <linux/ctype.h> 123 #include <linux/if_arp.h> 124 #include <linux/if_vlan.h> 125 #include <linux/ip.h> 126 #include <net/ip.h> 127 #include <net/mpls.h> 128 #include <linux/ipv6.h> 129 #include <linux/in.h> 130 #include <linux/jhash.h> 131 #include <linux/random.h> 132 #include <trace/events/napi.h> 133 #include <trace/events/net.h> 134 #include <trace/events/skb.h> 135 #include <trace/events/qdisc.h> 136 #include <trace/events/xdp.h> 137 #include <linux/inetdevice.h> 138 #include <linux/cpu_rmap.h> 139 #include <linux/static_key.h> 140 #include <linux/hashtable.h> 141 #include <linux/vmalloc.h> 142 #include <linux/if_macvlan.h> 143 #include <linux/errqueue.h> 144 #include <linux/hrtimer.h> 145 #include <linux/netfilter_netdev.h> 146 #include <linux/crash_dump.h> 147 #include <linux/sctp.h> 148 #include <net/udp_tunnel.h> 149 #include <linux/net_namespace.h> 150 #include <linux/indirect_call_wrapper.h> 151 #include <net/devlink.h> 152 #include <linux/pm_runtime.h> 153 #include <linux/prandom.h> 154 #include <linux/once_lite.h> 155 #include <net/netdev_rx_queue.h> 156 157 #include "dev.h" 158 #include "net-sysfs.h" 159 160 static DEFINE_SPINLOCK(ptype_lock); 161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 162 struct list_head ptype_all __read_mostly; /* Taps */ 163 164 static int netif_rx_internal(struct sk_buff *skb); 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 /* We reserved the ifindex, this can't fail */ 394 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 395 396 dev_base_seq_inc(net); 397 } 398 399 /* Device list removal 400 * caller must respect a RCU grace period before freeing/reusing dev 401 */ 402 static void unlist_netdevice(struct net_device *dev, bool lock) 403 { 404 struct net *net = dev_net(dev); 405 406 ASSERT_RTNL(); 407 408 xa_erase(&net->dev_by_index, dev->ifindex); 409 410 /* Unlink dev from the device chain */ 411 if (lock) 412 write_lock(&dev_base_lock); 413 list_del_rcu(&dev->dev_list); 414 netdev_name_node_del(dev->name_node); 415 hlist_del_rcu(&dev->index_hlist); 416 if (lock) 417 write_unlock(&dev_base_lock); 418 419 dev_base_seq_inc(dev_net(dev)); 420 } 421 422 /* 423 * Our notifier list 424 */ 425 426 static RAW_NOTIFIER_HEAD(netdev_chain); 427 428 /* 429 * Device drivers call our routines to queue packets here. We empty the 430 * queue in the local softnet handler. 431 */ 432 433 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 434 EXPORT_PER_CPU_SYMBOL(softnet_data); 435 436 #ifdef CONFIG_LOCKDEP 437 /* 438 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 439 * according to dev->type 440 */ 441 static const unsigned short netdev_lock_type[] = { 442 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 443 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 444 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 445 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 446 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 447 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 448 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 449 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 450 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 451 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 452 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 453 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 454 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 455 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 456 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 457 458 static const char *const netdev_lock_name[] = { 459 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 460 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 461 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 462 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 463 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 464 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 465 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 466 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 467 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 468 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 469 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 470 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 471 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 472 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 473 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 474 475 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 476 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 477 478 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 479 { 480 int i; 481 482 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 483 if (netdev_lock_type[i] == dev_type) 484 return i; 485 /* the last key is used by default */ 486 return ARRAY_SIZE(netdev_lock_type) - 1; 487 } 488 489 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 490 unsigned short dev_type) 491 { 492 int i; 493 494 i = netdev_lock_pos(dev_type); 495 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 496 netdev_lock_name[i]); 497 } 498 499 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 500 { 501 int i; 502 503 i = netdev_lock_pos(dev->type); 504 lockdep_set_class_and_name(&dev->addr_list_lock, 505 &netdev_addr_lock_key[i], 506 netdev_lock_name[i]); 507 } 508 #else 509 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 510 unsigned short dev_type) 511 { 512 } 513 514 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 515 { 516 } 517 #endif 518 519 /******************************************************************************* 520 * 521 * Protocol management and registration routines 522 * 523 *******************************************************************************/ 524 525 526 /* 527 * Add a protocol ID to the list. Now that the input handler is 528 * smarter we can dispense with all the messy stuff that used to be 529 * here. 530 * 531 * BEWARE!!! Protocol handlers, mangling input packets, 532 * MUST BE last in hash buckets and checking protocol handlers 533 * MUST start from promiscuous ptype_all chain in net_bh. 534 * It is true now, do not change it. 535 * Explanation follows: if protocol handler, mangling packet, will 536 * be the first on list, it is not able to sense, that packet 537 * is cloned and should be copied-on-write, so that it will 538 * change it and subsequent readers will get broken packet. 539 * --ANK (980803) 540 */ 541 542 static inline struct list_head *ptype_head(const struct packet_type *pt) 543 { 544 if (pt->type == htons(ETH_P_ALL)) 545 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 546 else 547 return pt->dev ? &pt->dev->ptype_specific : 548 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 549 } 550 551 /** 552 * dev_add_pack - add packet handler 553 * @pt: packet type declaration 554 * 555 * Add a protocol handler to the networking stack. The passed &packet_type 556 * is linked into kernel lists and may not be freed until it has been 557 * removed from the kernel lists. 558 * 559 * This call does not sleep therefore it can not 560 * guarantee all CPU's that are in middle of receiving packets 561 * will see the new packet type (until the next received packet). 562 */ 563 564 void dev_add_pack(struct packet_type *pt) 565 { 566 struct list_head *head = ptype_head(pt); 567 568 spin_lock(&ptype_lock); 569 list_add_rcu(&pt->list, head); 570 spin_unlock(&ptype_lock); 571 } 572 EXPORT_SYMBOL(dev_add_pack); 573 574 /** 575 * __dev_remove_pack - remove packet handler 576 * @pt: packet type declaration 577 * 578 * Remove a protocol handler that was previously added to the kernel 579 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 580 * from the kernel lists and can be freed or reused once this function 581 * returns. 582 * 583 * The packet type might still be in use by receivers 584 * and must not be freed until after all the CPU's have gone 585 * through a quiescent state. 586 */ 587 void __dev_remove_pack(struct packet_type *pt) 588 { 589 struct list_head *head = ptype_head(pt); 590 struct packet_type *pt1; 591 592 spin_lock(&ptype_lock); 593 594 list_for_each_entry(pt1, head, list) { 595 if (pt == pt1) { 596 list_del_rcu(&pt->list); 597 goto out; 598 } 599 } 600 601 pr_warn("dev_remove_pack: %p not found\n", pt); 602 out: 603 spin_unlock(&ptype_lock); 604 } 605 EXPORT_SYMBOL(__dev_remove_pack); 606 607 /** 608 * dev_remove_pack - remove packet handler 609 * @pt: packet type declaration 610 * 611 * Remove a protocol handler that was previously added to the kernel 612 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 613 * from the kernel lists and can be freed or reused once this function 614 * returns. 615 * 616 * This call sleeps to guarantee that no CPU is looking at the packet 617 * type after return. 618 */ 619 void dev_remove_pack(struct packet_type *pt) 620 { 621 __dev_remove_pack(pt); 622 623 synchronize_net(); 624 } 625 EXPORT_SYMBOL(dev_remove_pack); 626 627 628 /******************************************************************************* 629 * 630 * Device Interface Subroutines 631 * 632 *******************************************************************************/ 633 634 /** 635 * dev_get_iflink - get 'iflink' value of a interface 636 * @dev: targeted interface 637 * 638 * Indicates the ifindex the interface is linked to. 639 * Physical interfaces have the same 'ifindex' and 'iflink' values. 640 */ 641 642 int dev_get_iflink(const struct net_device *dev) 643 { 644 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 645 return dev->netdev_ops->ndo_get_iflink(dev); 646 647 return dev->ifindex; 648 } 649 EXPORT_SYMBOL(dev_get_iflink); 650 651 /** 652 * dev_fill_metadata_dst - Retrieve tunnel egress information. 653 * @dev: targeted interface 654 * @skb: The packet. 655 * 656 * For better visibility of tunnel traffic OVS needs to retrieve 657 * egress tunnel information for a packet. Following API allows 658 * user to get this info. 659 */ 660 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 661 { 662 struct ip_tunnel_info *info; 663 664 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 665 return -EINVAL; 666 667 info = skb_tunnel_info_unclone(skb); 668 if (!info) 669 return -ENOMEM; 670 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 671 return -EINVAL; 672 673 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 674 } 675 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 676 677 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 678 { 679 int k = stack->num_paths++; 680 681 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 682 return NULL; 683 684 return &stack->path[k]; 685 } 686 687 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 688 struct net_device_path_stack *stack) 689 { 690 const struct net_device *last_dev; 691 struct net_device_path_ctx ctx = { 692 .dev = dev, 693 }; 694 struct net_device_path *path; 695 int ret = 0; 696 697 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 698 stack->num_paths = 0; 699 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 700 last_dev = ctx.dev; 701 path = dev_fwd_path(stack); 702 if (!path) 703 return -1; 704 705 memset(path, 0, sizeof(struct net_device_path)); 706 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 707 if (ret < 0) 708 return -1; 709 710 if (WARN_ON_ONCE(last_dev == ctx.dev)) 711 return -1; 712 } 713 714 if (!ctx.dev) 715 return ret; 716 717 path = dev_fwd_path(stack); 718 if (!path) 719 return -1; 720 path->type = DEV_PATH_ETHERNET; 721 path->dev = ctx.dev; 722 723 return ret; 724 } 725 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 726 727 /** 728 * __dev_get_by_name - find a device by its name 729 * @net: the applicable net namespace 730 * @name: name to find 731 * 732 * Find an interface by name. Must be called under RTNL semaphore 733 * or @dev_base_lock. If the name is found a pointer to the device 734 * is returned. If the name is not found then %NULL is returned. The 735 * reference counters are not incremented so the caller must be 736 * careful with locks. 737 */ 738 739 struct net_device *__dev_get_by_name(struct net *net, const char *name) 740 { 741 struct netdev_name_node *node_name; 742 743 node_name = netdev_name_node_lookup(net, name); 744 return node_name ? node_name->dev : NULL; 745 } 746 EXPORT_SYMBOL(__dev_get_by_name); 747 748 /** 749 * dev_get_by_name_rcu - find a device by its name 750 * @net: the applicable net namespace 751 * @name: name to find 752 * 753 * Find an interface by name. 754 * If the name is found a pointer to the device is returned. 755 * If the name is not found then %NULL is returned. 756 * The reference counters are not incremented so the caller must be 757 * careful with locks. The caller must hold RCU lock. 758 */ 759 760 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 761 { 762 struct netdev_name_node *node_name; 763 764 node_name = netdev_name_node_lookup_rcu(net, name); 765 return node_name ? node_name->dev : NULL; 766 } 767 EXPORT_SYMBOL(dev_get_by_name_rcu); 768 769 /* Deprecated for new users, call netdev_get_by_name() instead */ 770 struct net_device *dev_get_by_name(struct net *net, const char *name) 771 { 772 struct net_device *dev; 773 774 rcu_read_lock(); 775 dev = dev_get_by_name_rcu(net, name); 776 dev_hold(dev); 777 rcu_read_unlock(); 778 return dev; 779 } 780 EXPORT_SYMBOL(dev_get_by_name); 781 782 /** 783 * netdev_get_by_name() - find a device by its name 784 * @net: the applicable net namespace 785 * @name: name to find 786 * @tracker: tracking object for the acquired reference 787 * @gfp: allocation flags for the tracker 788 * 789 * Find an interface by name. This can be called from any 790 * context and does its own locking. The returned handle has 791 * the usage count incremented and the caller must use netdev_put() to 792 * release it when it is no longer needed. %NULL is returned if no 793 * matching device is found. 794 */ 795 struct net_device *netdev_get_by_name(struct net *net, const char *name, 796 netdevice_tracker *tracker, gfp_t gfp) 797 { 798 struct net_device *dev; 799 800 dev = dev_get_by_name(net, name); 801 if (dev) 802 netdev_tracker_alloc(dev, tracker, gfp); 803 return dev; 804 } 805 EXPORT_SYMBOL(netdev_get_by_name); 806 807 /** 808 * __dev_get_by_index - find a device by its ifindex 809 * @net: the applicable net namespace 810 * @ifindex: index of device 811 * 812 * Search for an interface by index. Returns %NULL if the device 813 * is not found or a pointer to the device. The device has not 814 * had its reference counter increased so the caller must be careful 815 * about locking. The caller must hold either the RTNL semaphore 816 * or @dev_base_lock. 817 */ 818 819 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 820 { 821 struct net_device *dev; 822 struct hlist_head *head = dev_index_hash(net, ifindex); 823 824 hlist_for_each_entry(dev, head, index_hlist) 825 if (dev->ifindex == ifindex) 826 return dev; 827 828 return NULL; 829 } 830 EXPORT_SYMBOL(__dev_get_by_index); 831 832 /** 833 * dev_get_by_index_rcu - find a device by its ifindex 834 * @net: the applicable net namespace 835 * @ifindex: index of device 836 * 837 * Search for an interface by index. Returns %NULL if the device 838 * is not found or a pointer to the device. The device has not 839 * had its reference counter increased so the caller must be careful 840 * about locking. The caller must hold RCU lock. 841 */ 842 843 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 844 { 845 struct net_device *dev; 846 struct hlist_head *head = dev_index_hash(net, ifindex); 847 848 hlist_for_each_entry_rcu(dev, head, index_hlist) 849 if (dev->ifindex == ifindex) 850 return dev; 851 852 return NULL; 853 } 854 EXPORT_SYMBOL(dev_get_by_index_rcu); 855 856 /* Deprecated for new users, call netdev_get_by_index() instead */ 857 struct net_device *dev_get_by_index(struct net *net, int ifindex) 858 { 859 struct net_device *dev; 860 861 rcu_read_lock(); 862 dev = dev_get_by_index_rcu(net, ifindex); 863 dev_hold(dev); 864 rcu_read_unlock(); 865 return dev; 866 } 867 EXPORT_SYMBOL(dev_get_by_index); 868 869 /** 870 * netdev_get_by_index() - find a device by its ifindex 871 * @net: the applicable net namespace 872 * @ifindex: index of device 873 * @tracker: tracking object for the acquired reference 874 * @gfp: allocation flags for the tracker 875 * 876 * Search for an interface by index. Returns NULL if the device 877 * is not found or a pointer to the device. The device returned has 878 * had a reference added and the pointer is safe until the user calls 879 * netdev_put() to indicate they have finished with it. 880 */ 881 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 882 netdevice_tracker *tracker, gfp_t gfp) 883 { 884 struct net_device *dev; 885 886 dev = dev_get_by_index(net, ifindex); 887 if (dev) 888 netdev_tracker_alloc(dev, tracker, gfp); 889 return dev; 890 } 891 EXPORT_SYMBOL(netdev_get_by_index); 892 893 /** 894 * dev_get_by_napi_id - find a device by napi_id 895 * @napi_id: ID of the NAPI struct 896 * 897 * Search for an interface by NAPI ID. Returns %NULL if the device 898 * is not found or a pointer to the device. The device has not had 899 * its reference counter increased so the caller must be careful 900 * about locking. The caller must hold RCU lock. 901 */ 902 903 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 904 { 905 struct napi_struct *napi; 906 907 WARN_ON_ONCE(!rcu_read_lock_held()); 908 909 if (napi_id < MIN_NAPI_ID) 910 return NULL; 911 912 napi = napi_by_id(napi_id); 913 914 return napi ? napi->dev : NULL; 915 } 916 EXPORT_SYMBOL(dev_get_by_napi_id); 917 918 /** 919 * netdev_get_name - get a netdevice name, knowing its ifindex. 920 * @net: network namespace 921 * @name: a pointer to the buffer where the name will be stored. 922 * @ifindex: the ifindex of the interface to get the name from. 923 */ 924 int netdev_get_name(struct net *net, char *name, int ifindex) 925 { 926 struct net_device *dev; 927 int ret; 928 929 down_read(&devnet_rename_sem); 930 rcu_read_lock(); 931 932 dev = dev_get_by_index_rcu(net, ifindex); 933 if (!dev) { 934 ret = -ENODEV; 935 goto out; 936 } 937 938 strcpy(name, dev->name); 939 940 ret = 0; 941 out: 942 rcu_read_unlock(); 943 up_read(&devnet_rename_sem); 944 return ret; 945 } 946 947 /** 948 * dev_getbyhwaddr_rcu - find a device by its hardware address 949 * @net: the applicable net namespace 950 * @type: media type of device 951 * @ha: hardware address 952 * 953 * Search for an interface by MAC address. Returns NULL if the device 954 * is not found or a pointer to the device. 955 * The caller must hold RCU or RTNL. 956 * The returned device has not had its ref count increased 957 * and the caller must therefore be careful about locking 958 * 959 */ 960 961 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 962 const char *ha) 963 { 964 struct net_device *dev; 965 966 for_each_netdev_rcu(net, dev) 967 if (dev->type == type && 968 !memcmp(dev->dev_addr, ha, dev->addr_len)) 969 return dev; 970 971 return NULL; 972 } 973 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 974 975 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 976 { 977 struct net_device *dev, *ret = NULL; 978 979 rcu_read_lock(); 980 for_each_netdev_rcu(net, dev) 981 if (dev->type == type) { 982 dev_hold(dev); 983 ret = dev; 984 break; 985 } 986 rcu_read_unlock(); 987 return ret; 988 } 989 EXPORT_SYMBOL(dev_getfirstbyhwtype); 990 991 /** 992 * __dev_get_by_flags - find any device with given flags 993 * @net: the applicable net namespace 994 * @if_flags: IFF_* values 995 * @mask: bitmask of bits in if_flags to check 996 * 997 * Search for any interface with the given flags. Returns NULL if a device 998 * is not found or a pointer to the device. Must be called inside 999 * rtnl_lock(), and result refcount is unchanged. 1000 */ 1001 1002 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1003 unsigned short mask) 1004 { 1005 struct net_device *dev, *ret; 1006 1007 ASSERT_RTNL(); 1008 1009 ret = NULL; 1010 for_each_netdev(net, dev) { 1011 if (((dev->flags ^ if_flags) & mask) == 0) { 1012 ret = dev; 1013 break; 1014 } 1015 } 1016 return ret; 1017 } 1018 EXPORT_SYMBOL(__dev_get_by_flags); 1019 1020 /** 1021 * dev_valid_name - check if name is okay for network device 1022 * @name: name string 1023 * 1024 * Network device names need to be valid file names to 1025 * allow sysfs to work. We also disallow any kind of 1026 * whitespace. 1027 */ 1028 bool dev_valid_name(const char *name) 1029 { 1030 if (*name == '\0') 1031 return false; 1032 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1033 return false; 1034 if (!strcmp(name, ".") || !strcmp(name, "..")) 1035 return false; 1036 1037 while (*name) { 1038 if (*name == '/' || *name == ':' || isspace(*name)) 1039 return false; 1040 name++; 1041 } 1042 return true; 1043 } 1044 EXPORT_SYMBOL(dev_valid_name); 1045 1046 /** 1047 * __dev_alloc_name - allocate a name for a device 1048 * @net: network namespace to allocate the device name in 1049 * @name: name format string 1050 * @buf: scratch buffer and result name string 1051 * 1052 * Passed a format string - eg "lt%d" it will try and find a suitable 1053 * id. It scans list of devices to build up a free map, then chooses 1054 * the first empty slot. The caller must hold the dev_base or rtnl lock 1055 * while allocating the name and adding the device in order to avoid 1056 * duplicates. 1057 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1058 * Returns the number of the unit assigned or a negative errno code. 1059 */ 1060 1061 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1062 { 1063 int i = 0; 1064 const char *p; 1065 const int max_netdevices = 8*PAGE_SIZE; 1066 unsigned long *inuse; 1067 struct net_device *d; 1068 1069 if (!dev_valid_name(name)) 1070 return -EINVAL; 1071 1072 p = strchr(name, '%'); 1073 if (p) { 1074 /* 1075 * Verify the string as this thing may have come from 1076 * the user. There must be either one "%d" and no other "%" 1077 * characters. 1078 */ 1079 if (p[1] != 'd' || strchr(p + 2, '%')) 1080 return -EINVAL; 1081 1082 /* Use one page as a bit array of possible slots */ 1083 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1084 if (!inuse) 1085 return -ENOMEM; 1086 1087 for_each_netdev(net, d) { 1088 struct netdev_name_node *name_node; 1089 list_for_each_entry(name_node, &d->name_node->list, list) { 1090 if (!sscanf(name_node->name, name, &i)) 1091 continue; 1092 if (i < 0 || i >= max_netdevices) 1093 continue; 1094 1095 /* avoid cases where sscanf is not exact inverse of printf */ 1096 snprintf(buf, IFNAMSIZ, name, i); 1097 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1098 __set_bit(i, inuse); 1099 } 1100 if (!sscanf(d->name, name, &i)) 1101 continue; 1102 if (i < 0 || i >= max_netdevices) 1103 continue; 1104 1105 /* avoid cases where sscanf is not exact inverse of printf */ 1106 snprintf(buf, IFNAMSIZ, name, i); 1107 if (!strncmp(buf, d->name, IFNAMSIZ)) 1108 __set_bit(i, inuse); 1109 } 1110 1111 i = find_first_zero_bit(inuse, max_netdevices); 1112 free_page((unsigned long) inuse); 1113 } 1114 1115 snprintf(buf, IFNAMSIZ, name, i); 1116 if (!netdev_name_in_use(net, buf)) 1117 return i; 1118 1119 /* It is possible to run out of possible slots 1120 * when the name is long and there isn't enough space left 1121 * for the digits, or if all bits are used. 1122 */ 1123 return -ENFILE; 1124 } 1125 1126 static int dev_alloc_name_ns(struct net *net, 1127 struct net_device *dev, 1128 const char *name) 1129 { 1130 char buf[IFNAMSIZ]; 1131 int ret; 1132 1133 BUG_ON(!net); 1134 ret = __dev_alloc_name(net, name, buf); 1135 if (ret >= 0) 1136 strscpy(dev->name, buf, IFNAMSIZ); 1137 return ret; 1138 } 1139 1140 /** 1141 * dev_alloc_name - allocate a name for a device 1142 * @dev: device 1143 * @name: name format string 1144 * 1145 * Passed a format string - eg "lt%d" it will try and find a suitable 1146 * id. It scans list of devices to build up a free map, then chooses 1147 * the first empty slot. The caller must hold the dev_base or rtnl lock 1148 * while allocating the name and adding the device in order to avoid 1149 * duplicates. 1150 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1151 * Returns the number of the unit assigned or a negative errno code. 1152 */ 1153 1154 int dev_alloc_name(struct net_device *dev, const char *name) 1155 { 1156 return dev_alloc_name_ns(dev_net(dev), dev, name); 1157 } 1158 EXPORT_SYMBOL(dev_alloc_name); 1159 1160 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1161 const char *name) 1162 { 1163 BUG_ON(!net); 1164 1165 if (!dev_valid_name(name)) 1166 return -EINVAL; 1167 1168 if (strchr(name, '%')) 1169 return dev_alloc_name_ns(net, dev, name); 1170 else if (netdev_name_in_use(net, name)) 1171 return -EEXIST; 1172 else if (dev->name != name) 1173 strscpy(dev->name, name, IFNAMSIZ); 1174 1175 return 0; 1176 } 1177 1178 /** 1179 * dev_change_name - change name of a device 1180 * @dev: device 1181 * @newname: name (or format string) must be at least IFNAMSIZ 1182 * 1183 * Change name of a device, can pass format strings "eth%d". 1184 * for wildcarding. 1185 */ 1186 int dev_change_name(struct net_device *dev, const char *newname) 1187 { 1188 unsigned char old_assign_type; 1189 char oldname[IFNAMSIZ]; 1190 int err = 0; 1191 int ret; 1192 struct net *net; 1193 1194 ASSERT_RTNL(); 1195 BUG_ON(!dev_net(dev)); 1196 1197 net = dev_net(dev); 1198 1199 down_write(&devnet_rename_sem); 1200 1201 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1202 up_write(&devnet_rename_sem); 1203 return 0; 1204 } 1205 1206 memcpy(oldname, dev->name, IFNAMSIZ); 1207 1208 err = dev_get_valid_name(net, dev, newname); 1209 if (err < 0) { 1210 up_write(&devnet_rename_sem); 1211 return err; 1212 } 1213 1214 if (oldname[0] && !strchr(oldname, '%')) 1215 netdev_info(dev, "renamed from %s%s\n", oldname, 1216 dev->flags & IFF_UP ? " (while UP)" : ""); 1217 1218 old_assign_type = dev->name_assign_type; 1219 dev->name_assign_type = NET_NAME_RENAMED; 1220 1221 rollback: 1222 ret = device_rename(&dev->dev, dev->name); 1223 if (ret) { 1224 memcpy(dev->name, oldname, IFNAMSIZ); 1225 dev->name_assign_type = old_assign_type; 1226 up_write(&devnet_rename_sem); 1227 return ret; 1228 } 1229 1230 up_write(&devnet_rename_sem); 1231 1232 netdev_adjacent_rename_links(dev, oldname); 1233 1234 write_lock(&dev_base_lock); 1235 netdev_name_node_del(dev->name_node); 1236 write_unlock(&dev_base_lock); 1237 1238 synchronize_rcu(); 1239 1240 write_lock(&dev_base_lock); 1241 netdev_name_node_add(net, dev->name_node); 1242 write_unlock(&dev_base_lock); 1243 1244 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1245 ret = notifier_to_errno(ret); 1246 1247 if (ret) { 1248 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1249 if (err >= 0) { 1250 err = ret; 1251 down_write(&devnet_rename_sem); 1252 memcpy(dev->name, oldname, IFNAMSIZ); 1253 memcpy(oldname, newname, IFNAMSIZ); 1254 dev->name_assign_type = old_assign_type; 1255 old_assign_type = NET_NAME_RENAMED; 1256 goto rollback; 1257 } else { 1258 netdev_err(dev, "name change rollback failed: %d\n", 1259 ret); 1260 } 1261 } 1262 1263 return err; 1264 } 1265 1266 /** 1267 * dev_set_alias - change ifalias of a device 1268 * @dev: device 1269 * @alias: name up to IFALIASZ 1270 * @len: limit of bytes to copy from info 1271 * 1272 * Set ifalias for a device, 1273 */ 1274 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1275 { 1276 struct dev_ifalias *new_alias = NULL; 1277 1278 if (len >= IFALIASZ) 1279 return -EINVAL; 1280 1281 if (len) { 1282 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1283 if (!new_alias) 1284 return -ENOMEM; 1285 1286 memcpy(new_alias->ifalias, alias, len); 1287 new_alias->ifalias[len] = 0; 1288 } 1289 1290 mutex_lock(&ifalias_mutex); 1291 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1292 mutex_is_locked(&ifalias_mutex)); 1293 mutex_unlock(&ifalias_mutex); 1294 1295 if (new_alias) 1296 kfree_rcu(new_alias, rcuhead); 1297 1298 return len; 1299 } 1300 EXPORT_SYMBOL(dev_set_alias); 1301 1302 /** 1303 * dev_get_alias - get ifalias of a device 1304 * @dev: device 1305 * @name: buffer to store name of ifalias 1306 * @len: size of buffer 1307 * 1308 * get ifalias for a device. Caller must make sure dev cannot go 1309 * away, e.g. rcu read lock or own a reference count to device. 1310 */ 1311 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1312 { 1313 const struct dev_ifalias *alias; 1314 int ret = 0; 1315 1316 rcu_read_lock(); 1317 alias = rcu_dereference(dev->ifalias); 1318 if (alias) 1319 ret = snprintf(name, len, "%s", alias->ifalias); 1320 rcu_read_unlock(); 1321 1322 return ret; 1323 } 1324 1325 /** 1326 * netdev_features_change - device changes features 1327 * @dev: device to cause notification 1328 * 1329 * Called to indicate a device has changed features. 1330 */ 1331 void netdev_features_change(struct net_device *dev) 1332 { 1333 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1334 } 1335 EXPORT_SYMBOL(netdev_features_change); 1336 1337 /** 1338 * netdev_state_change - device changes state 1339 * @dev: device to cause notification 1340 * 1341 * Called to indicate a device has changed state. This function calls 1342 * the notifier chains for netdev_chain and sends a NEWLINK message 1343 * to the routing socket. 1344 */ 1345 void netdev_state_change(struct net_device *dev) 1346 { 1347 if (dev->flags & IFF_UP) { 1348 struct netdev_notifier_change_info change_info = { 1349 .info.dev = dev, 1350 }; 1351 1352 call_netdevice_notifiers_info(NETDEV_CHANGE, 1353 &change_info.info); 1354 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1355 } 1356 } 1357 EXPORT_SYMBOL(netdev_state_change); 1358 1359 /** 1360 * __netdev_notify_peers - notify network peers about existence of @dev, 1361 * to be called when rtnl lock is already held. 1362 * @dev: network device 1363 * 1364 * Generate traffic such that interested network peers are aware of 1365 * @dev, such as by generating a gratuitous ARP. This may be used when 1366 * a device wants to inform the rest of the network about some sort of 1367 * reconfiguration such as a failover event or virtual machine 1368 * migration. 1369 */ 1370 void __netdev_notify_peers(struct net_device *dev) 1371 { 1372 ASSERT_RTNL(); 1373 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1374 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1375 } 1376 EXPORT_SYMBOL(__netdev_notify_peers); 1377 1378 /** 1379 * netdev_notify_peers - notify network peers about existence of @dev 1380 * @dev: network device 1381 * 1382 * Generate traffic such that interested network peers are aware of 1383 * @dev, such as by generating a gratuitous ARP. This may be used when 1384 * a device wants to inform the rest of the network about some sort of 1385 * reconfiguration such as a failover event or virtual machine 1386 * migration. 1387 */ 1388 void netdev_notify_peers(struct net_device *dev) 1389 { 1390 rtnl_lock(); 1391 __netdev_notify_peers(dev); 1392 rtnl_unlock(); 1393 } 1394 EXPORT_SYMBOL(netdev_notify_peers); 1395 1396 static int napi_threaded_poll(void *data); 1397 1398 static int napi_kthread_create(struct napi_struct *n) 1399 { 1400 int err = 0; 1401 1402 /* Create and wake up the kthread once to put it in 1403 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1404 * warning and work with loadavg. 1405 */ 1406 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1407 n->dev->name, n->napi_id); 1408 if (IS_ERR(n->thread)) { 1409 err = PTR_ERR(n->thread); 1410 pr_err("kthread_run failed with err %d\n", err); 1411 n->thread = NULL; 1412 } 1413 1414 return err; 1415 } 1416 1417 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1418 { 1419 const struct net_device_ops *ops = dev->netdev_ops; 1420 int ret; 1421 1422 ASSERT_RTNL(); 1423 dev_addr_check(dev); 1424 1425 if (!netif_device_present(dev)) { 1426 /* may be detached because parent is runtime-suspended */ 1427 if (dev->dev.parent) 1428 pm_runtime_resume(dev->dev.parent); 1429 if (!netif_device_present(dev)) 1430 return -ENODEV; 1431 } 1432 1433 /* Block netpoll from trying to do any rx path servicing. 1434 * If we don't do this there is a chance ndo_poll_controller 1435 * or ndo_poll may be running while we open the device 1436 */ 1437 netpoll_poll_disable(dev); 1438 1439 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1440 ret = notifier_to_errno(ret); 1441 if (ret) 1442 return ret; 1443 1444 set_bit(__LINK_STATE_START, &dev->state); 1445 1446 if (ops->ndo_validate_addr) 1447 ret = ops->ndo_validate_addr(dev); 1448 1449 if (!ret && ops->ndo_open) 1450 ret = ops->ndo_open(dev); 1451 1452 netpoll_poll_enable(dev); 1453 1454 if (ret) 1455 clear_bit(__LINK_STATE_START, &dev->state); 1456 else { 1457 dev->flags |= IFF_UP; 1458 dev_set_rx_mode(dev); 1459 dev_activate(dev); 1460 add_device_randomness(dev->dev_addr, dev->addr_len); 1461 } 1462 1463 return ret; 1464 } 1465 1466 /** 1467 * dev_open - prepare an interface for use. 1468 * @dev: device to open 1469 * @extack: netlink extended ack 1470 * 1471 * Takes a device from down to up state. The device's private open 1472 * function is invoked and then the multicast lists are loaded. Finally 1473 * the device is moved into the up state and a %NETDEV_UP message is 1474 * sent to the netdev notifier chain. 1475 * 1476 * Calling this function on an active interface is a nop. On a failure 1477 * a negative errno code is returned. 1478 */ 1479 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1480 { 1481 int ret; 1482 1483 if (dev->flags & IFF_UP) 1484 return 0; 1485 1486 ret = __dev_open(dev, extack); 1487 if (ret < 0) 1488 return ret; 1489 1490 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1491 call_netdevice_notifiers(NETDEV_UP, dev); 1492 1493 return ret; 1494 } 1495 EXPORT_SYMBOL(dev_open); 1496 1497 static void __dev_close_many(struct list_head *head) 1498 { 1499 struct net_device *dev; 1500 1501 ASSERT_RTNL(); 1502 might_sleep(); 1503 1504 list_for_each_entry(dev, head, close_list) { 1505 /* Temporarily disable netpoll until the interface is down */ 1506 netpoll_poll_disable(dev); 1507 1508 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1509 1510 clear_bit(__LINK_STATE_START, &dev->state); 1511 1512 /* Synchronize to scheduled poll. We cannot touch poll list, it 1513 * can be even on different cpu. So just clear netif_running(). 1514 * 1515 * dev->stop() will invoke napi_disable() on all of it's 1516 * napi_struct instances on this device. 1517 */ 1518 smp_mb__after_atomic(); /* Commit netif_running(). */ 1519 } 1520 1521 dev_deactivate_many(head); 1522 1523 list_for_each_entry(dev, head, close_list) { 1524 const struct net_device_ops *ops = dev->netdev_ops; 1525 1526 /* 1527 * Call the device specific close. This cannot fail. 1528 * Only if device is UP 1529 * 1530 * We allow it to be called even after a DETACH hot-plug 1531 * event. 1532 */ 1533 if (ops->ndo_stop) 1534 ops->ndo_stop(dev); 1535 1536 dev->flags &= ~IFF_UP; 1537 netpoll_poll_enable(dev); 1538 } 1539 } 1540 1541 static void __dev_close(struct net_device *dev) 1542 { 1543 LIST_HEAD(single); 1544 1545 list_add(&dev->close_list, &single); 1546 __dev_close_many(&single); 1547 list_del(&single); 1548 } 1549 1550 void dev_close_many(struct list_head *head, bool unlink) 1551 { 1552 struct net_device *dev, *tmp; 1553 1554 /* Remove the devices that don't need to be closed */ 1555 list_for_each_entry_safe(dev, tmp, head, close_list) 1556 if (!(dev->flags & IFF_UP)) 1557 list_del_init(&dev->close_list); 1558 1559 __dev_close_many(head); 1560 1561 list_for_each_entry_safe(dev, tmp, head, close_list) { 1562 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1563 call_netdevice_notifiers(NETDEV_DOWN, dev); 1564 if (unlink) 1565 list_del_init(&dev->close_list); 1566 } 1567 } 1568 EXPORT_SYMBOL(dev_close_many); 1569 1570 /** 1571 * dev_close - shutdown an interface. 1572 * @dev: device to shutdown 1573 * 1574 * This function moves an active device into down state. A 1575 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1576 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1577 * chain. 1578 */ 1579 void dev_close(struct net_device *dev) 1580 { 1581 if (dev->flags & IFF_UP) { 1582 LIST_HEAD(single); 1583 1584 list_add(&dev->close_list, &single); 1585 dev_close_many(&single, true); 1586 list_del(&single); 1587 } 1588 } 1589 EXPORT_SYMBOL(dev_close); 1590 1591 1592 /** 1593 * dev_disable_lro - disable Large Receive Offload on a device 1594 * @dev: device 1595 * 1596 * Disable Large Receive Offload (LRO) on a net device. Must be 1597 * called under RTNL. This is needed if received packets may be 1598 * forwarded to another interface. 1599 */ 1600 void dev_disable_lro(struct net_device *dev) 1601 { 1602 struct net_device *lower_dev; 1603 struct list_head *iter; 1604 1605 dev->wanted_features &= ~NETIF_F_LRO; 1606 netdev_update_features(dev); 1607 1608 if (unlikely(dev->features & NETIF_F_LRO)) 1609 netdev_WARN(dev, "failed to disable LRO!\n"); 1610 1611 netdev_for_each_lower_dev(dev, lower_dev, iter) 1612 dev_disable_lro(lower_dev); 1613 } 1614 EXPORT_SYMBOL(dev_disable_lro); 1615 1616 /** 1617 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1618 * @dev: device 1619 * 1620 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1621 * called under RTNL. This is needed if Generic XDP is installed on 1622 * the device. 1623 */ 1624 static void dev_disable_gro_hw(struct net_device *dev) 1625 { 1626 dev->wanted_features &= ~NETIF_F_GRO_HW; 1627 netdev_update_features(dev); 1628 1629 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1630 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1631 } 1632 1633 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1634 { 1635 #define N(val) \ 1636 case NETDEV_##val: \ 1637 return "NETDEV_" __stringify(val); 1638 switch (cmd) { 1639 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1640 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1641 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1642 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1643 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1644 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1645 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1646 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1647 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1648 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1649 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1650 N(XDP_FEAT_CHANGE) 1651 } 1652 #undef N 1653 return "UNKNOWN_NETDEV_EVENT"; 1654 } 1655 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1656 1657 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1658 struct net_device *dev) 1659 { 1660 struct netdev_notifier_info info = { 1661 .dev = dev, 1662 }; 1663 1664 return nb->notifier_call(nb, val, &info); 1665 } 1666 1667 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1668 struct net_device *dev) 1669 { 1670 int err; 1671 1672 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1673 err = notifier_to_errno(err); 1674 if (err) 1675 return err; 1676 1677 if (!(dev->flags & IFF_UP)) 1678 return 0; 1679 1680 call_netdevice_notifier(nb, NETDEV_UP, dev); 1681 return 0; 1682 } 1683 1684 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1685 struct net_device *dev) 1686 { 1687 if (dev->flags & IFF_UP) { 1688 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1689 dev); 1690 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1691 } 1692 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1693 } 1694 1695 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1696 struct net *net) 1697 { 1698 struct net_device *dev; 1699 int err; 1700 1701 for_each_netdev(net, dev) { 1702 err = call_netdevice_register_notifiers(nb, dev); 1703 if (err) 1704 goto rollback; 1705 } 1706 return 0; 1707 1708 rollback: 1709 for_each_netdev_continue_reverse(net, dev) 1710 call_netdevice_unregister_notifiers(nb, dev); 1711 return err; 1712 } 1713 1714 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1715 struct net *net) 1716 { 1717 struct net_device *dev; 1718 1719 for_each_netdev(net, dev) 1720 call_netdevice_unregister_notifiers(nb, dev); 1721 } 1722 1723 static int dev_boot_phase = 1; 1724 1725 /** 1726 * register_netdevice_notifier - register a network notifier block 1727 * @nb: notifier 1728 * 1729 * Register a notifier to be called when network device events occur. 1730 * The notifier passed is linked into the kernel structures and must 1731 * not be reused until it has been unregistered. A negative errno code 1732 * is returned on a failure. 1733 * 1734 * When registered all registration and up events are replayed 1735 * to the new notifier to allow device to have a race free 1736 * view of the network device list. 1737 */ 1738 1739 int register_netdevice_notifier(struct notifier_block *nb) 1740 { 1741 struct net *net; 1742 int err; 1743 1744 /* Close race with setup_net() and cleanup_net() */ 1745 down_write(&pernet_ops_rwsem); 1746 rtnl_lock(); 1747 err = raw_notifier_chain_register(&netdev_chain, nb); 1748 if (err) 1749 goto unlock; 1750 if (dev_boot_phase) 1751 goto unlock; 1752 for_each_net(net) { 1753 err = call_netdevice_register_net_notifiers(nb, net); 1754 if (err) 1755 goto rollback; 1756 } 1757 1758 unlock: 1759 rtnl_unlock(); 1760 up_write(&pernet_ops_rwsem); 1761 return err; 1762 1763 rollback: 1764 for_each_net_continue_reverse(net) 1765 call_netdevice_unregister_net_notifiers(nb, net); 1766 1767 raw_notifier_chain_unregister(&netdev_chain, nb); 1768 goto unlock; 1769 } 1770 EXPORT_SYMBOL(register_netdevice_notifier); 1771 1772 /** 1773 * unregister_netdevice_notifier - unregister a network notifier block 1774 * @nb: notifier 1775 * 1776 * Unregister a notifier previously registered by 1777 * register_netdevice_notifier(). The notifier is unlinked into the 1778 * kernel structures and may then be reused. A negative errno code 1779 * is returned on a failure. 1780 * 1781 * After unregistering unregister and down device events are synthesized 1782 * for all devices on the device list to the removed notifier to remove 1783 * the need for special case cleanup code. 1784 */ 1785 1786 int unregister_netdevice_notifier(struct notifier_block *nb) 1787 { 1788 struct net *net; 1789 int err; 1790 1791 /* Close race with setup_net() and cleanup_net() */ 1792 down_write(&pernet_ops_rwsem); 1793 rtnl_lock(); 1794 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1795 if (err) 1796 goto unlock; 1797 1798 for_each_net(net) 1799 call_netdevice_unregister_net_notifiers(nb, net); 1800 1801 unlock: 1802 rtnl_unlock(); 1803 up_write(&pernet_ops_rwsem); 1804 return err; 1805 } 1806 EXPORT_SYMBOL(unregister_netdevice_notifier); 1807 1808 static int __register_netdevice_notifier_net(struct net *net, 1809 struct notifier_block *nb, 1810 bool ignore_call_fail) 1811 { 1812 int err; 1813 1814 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1815 if (err) 1816 return err; 1817 if (dev_boot_phase) 1818 return 0; 1819 1820 err = call_netdevice_register_net_notifiers(nb, net); 1821 if (err && !ignore_call_fail) 1822 goto chain_unregister; 1823 1824 return 0; 1825 1826 chain_unregister: 1827 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1828 return err; 1829 } 1830 1831 static int __unregister_netdevice_notifier_net(struct net *net, 1832 struct notifier_block *nb) 1833 { 1834 int err; 1835 1836 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1837 if (err) 1838 return err; 1839 1840 call_netdevice_unregister_net_notifiers(nb, net); 1841 return 0; 1842 } 1843 1844 /** 1845 * register_netdevice_notifier_net - register a per-netns network notifier block 1846 * @net: network namespace 1847 * @nb: notifier 1848 * 1849 * Register a notifier to be called when network device events occur. 1850 * The notifier passed is linked into the kernel structures and must 1851 * not be reused until it has been unregistered. A negative errno code 1852 * is returned on a failure. 1853 * 1854 * When registered all registration and up events are replayed 1855 * to the new notifier to allow device to have a race free 1856 * view of the network device list. 1857 */ 1858 1859 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1860 { 1861 int err; 1862 1863 rtnl_lock(); 1864 err = __register_netdevice_notifier_net(net, nb, false); 1865 rtnl_unlock(); 1866 return err; 1867 } 1868 EXPORT_SYMBOL(register_netdevice_notifier_net); 1869 1870 /** 1871 * unregister_netdevice_notifier_net - unregister a per-netns 1872 * network notifier block 1873 * @net: network namespace 1874 * @nb: notifier 1875 * 1876 * Unregister a notifier previously registered by 1877 * register_netdevice_notifier_net(). The notifier is unlinked from the 1878 * kernel structures and may then be reused. A negative errno code 1879 * is returned on a failure. 1880 * 1881 * After unregistering unregister and down device events are synthesized 1882 * for all devices on the device list to the removed notifier to remove 1883 * the need for special case cleanup code. 1884 */ 1885 1886 int unregister_netdevice_notifier_net(struct net *net, 1887 struct notifier_block *nb) 1888 { 1889 int err; 1890 1891 rtnl_lock(); 1892 err = __unregister_netdevice_notifier_net(net, nb); 1893 rtnl_unlock(); 1894 return err; 1895 } 1896 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1897 1898 static void __move_netdevice_notifier_net(struct net *src_net, 1899 struct net *dst_net, 1900 struct notifier_block *nb) 1901 { 1902 __unregister_netdevice_notifier_net(src_net, nb); 1903 __register_netdevice_notifier_net(dst_net, nb, true); 1904 } 1905 1906 int register_netdevice_notifier_dev_net(struct net_device *dev, 1907 struct notifier_block *nb, 1908 struct netdev_net_notifier *nn) 1909 { 1910 int err; 1911 1912 rtnl_lock(); 1913 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1914 if (!err) { 1915 nn->nb = nb; 1916 list_add(&nn->list, &dev->net_notifier_list); 1917 } 1918 rtnl_unlock(); 1919 return err; 1920 } 1921 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1922 1923 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1924 struct notifier_block *nb, 1925 struct netdev_net_notifier *nn) 1926 { 1927 int err; 1928 1929 rtnl_lock(); 1930 list_del(&nn->list); 1931 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1932 rtnl_unlock(); 1933 return err; 1934 } 1935 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1936 1937 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1938 struct net *net) 1939 { 1940 struct netdev_net_notifier *nn; 1941 1942 list_for_each_entry(nn, &dev->net_notifier_list, list) 1943 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1944 } 1945 1946 /** 1947 * call_netdevice_notifiers_info - call all network notifier blocks 1948 * @val: value passed unmodified to notifier function 1949 * @info: notifier information data 1950 * 1951 * Call all network notifier blocks. Parameters and return value 1952 * are as for raw_notifier_call_chain(). 1953 */ 1954 1955 int call_netdevice_notifiers_info(unsigned long val, 1956 struct netdev_notifier_info *info) 1957 { 1958 struct net *net = dev_net(info->dev); 1959 int ret; 1960 1961 ASSERT_RTNL(); 1962 1963 /* Run per-netns notifier block chain first, then run the global one. 1964 * Hopefully, one day, the global one is going to be removed after 1965 * all notifier block registrators get converted to be per-netns. 1966 */ 1967 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1968 if (ret & NOTIFY_STOP_MASK) 1969 return ret; 1970 return raw_notifier_call_chain(&netdev_chain, val, info); 1971 } 1972 1973 /** 1974 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1975 * for and rollback on error 1976 * @val_up: value passed unmodified to notifier function 1977 * @val_down: value passed unmodified to the notifier function when 1978 * recovering from an error on @val_up 1979 * @info: notifier information data 1980 * 1981 * Call all per-netns network notifier blocks, but not notifier blocks on 1982 * the global notifier chain. Parameters and return value are as for 1983 * raw_notifier_call_chain_robust(). 1984 */ 1985 1986 static int 1987 call_netdevice_notifiers_info_robust(unsigned long val_up, 1988 unsigned long val_down, 1989 struct netdev_notifier_info *info) 1990 { 1991 struct net *net = dev_net(info->dev); 1992 1993 ASSERT_RTNL(); 1994 1995 return raw_notifier_call_chain_robust(&net->netdev_chain, 1996 val_up, val_down, info); 1997 } 1998 1999 static int call_netdevice_notifiers_extack(unsigned long val, 2000 struct net_device *dev, 2001 struct netlink_ext_ack *extack) 2002 { 2003 struct netdev_notifier_info info = { 2004 .dev = dev, 2005 .extack = extack, 2006 }; 2007 2008 return call_netdevice_notifiers_info(val, &info); 2009 } 2010 2011 /** 2012 * call_netdevice_notifiers - call all network notifier blocks 2013 * @val: value passed unmodified to notifier function 2014 * @dev: net_device pointer passed unmodified to notifier function 2015 * 2016 * Call all network notifier blocks. Parameters and return value 2017 * are as for raw_notifier_call_chain(). 2018 */ 2019 2020 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2021 { 2022 return call_netdevice_notifiers_extack(val, dev, NULL); 2023 } 2024 EXPORT_SYMBOL(call_netdevice_notifiers); 2025 2026 /** 2027 * call_netdevice_notifiers_mtu - call all network notifier blocks 2028 * @val: value passed unmodified to notifier function 2029 * @dev: net_device pointer passed unmodified to notifier function 2030 * @arg: additional u32 argument passed to the notifier function 2031 * 2032 * Call all network notifier blocks. Parameters and return value 2033 * are as for raw_notifier_call_chain(). 2034 */ 2035 static int call_netdevice_notifiers_mtu(unsigned long val, 2036 struct net_device *dev, u32 arg) 2037 { 2038 struct netdev_notifier_info_ext info = { 2039 .info.dev = dev, 2040 .ext.mtu = arg, 2041 }; 2042 2043 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2044 2045 return call_netdevice_notifiers_info(val, &info.info); 2046 } 2047 2048 #ifdef CONFIG_NET_INGRESS 2049 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2050 2051 void net_inc_ingress_queue(void) 2052 { 2053 static_branch_inc(&ingress_needed_key); 2054 } 2055 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2056 2057 void net_dec_ingress_queue(void) 2058 { 2059 static_branch_dec(&ingress_needed_key); 2060 } 2061 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2062 #endif 2063 2064 #ifdef CONFIG_NET_EGRESS 2065 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2066 2067 void net_inc_egress_queue(void) 2068 { 2069 static_branch_inc(&egress_needed_key); 2070 } 2071 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2072 2073 void net_dec_egress_queue(void) 2074 { 2075 static_branch_dec(&egress_needed_key); 2076 } 2077 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2078 #endif 2079 2080 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2081 EXPORT_SYMBOL(netstamp_needed_key); 2082 #ifdef CONFIG_JUMP_LABEL 2083 static atomic_t netstamp_needed_deferred; 2084 static atomic_t netstamp_wanted; 2085 static void netstamp_clear(struct work_struct *work) 2086 { 2087 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2088 int wanted; 2089 2090 wanted = atomic_add_return(deferred, &netstamp_wanted); 2091 if (wanted > 0) 2092 static_branch_enable(&netstamp_needed_key); 2093 else 2094 static_branch_disable(&netstamp_needed_key); 2095 } 2096 static DECLARE_WORK(netstamp_work, netstamp_clear); 2097 #endif 2098 2099 void net_enable_timestamp(void) 2100 { 2101 #ifdef CONFIG_JUMP_LABEL 2102 int wanted = atomic_read(&netstamp_wanted); 2103 2104 while (wanted > 0) { 2105 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2106 return; 2107 } 2108 atomic_inc(&netstamp_needed_deferred); 2109 schedule_work(&netstamp_work); 2110 #else 2111 static_branch_inc(&netstamp_needed_key); 2112 #endif 2113 } 2114 EXPORT_SYMBOL(net_enable_timestamp); 2115 2116 void net_disable_timestamp(void) 2117 { 2118 #ifdef CONFIG_JUMP_LABEL 2119 int wanted = atomic_read(&netstamp_wanted); 2120 2121 while (wanted > 1) { 2122 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2123 return; 2124 } 2125 atomic_dec(&netstamp_needed_deferred); 2126 schedule_work(&netstamp_work); 2127 #else 2128 static_branch_dec(&netstamp_needed_key); 2129 #endif 2130 } 2131 EXPORT_SYMBOL(net_disable_timestamp); 2132 2133 static inline void net_timestamp_set(struct sk_buff *skb) 2134 { 2135 skb->tstamp = 0; 2136 skb->mono_delivery_time = 0; 2137 if (static_branch_unlikely(&netstamp_needed_key)) 2138 skb->tstamp = ktime_get_real(); 2139 } 2140 2141 #define net_timestamp_check(COND, SKB) \ 2142 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2143 if ((COND) && !(SKB)->tstamp) \ 2144 (SKB)->tstamp = ktime_get_real(); \ 2145 } \ 2146 2147 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2148 { 2149 return __is_skb_forwardable(dev, skb, true); 2150 } 2151 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2152 2153 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2154 bool check_mtu) 2155 { 2156 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2157 2158 if (likely(!ret)) { 2159 skb->protocol = eth_type_trans(skb, dev); 2160 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2161 } 2162 2163 return ret; 2164 } 2165 2166 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2167 { 2168 return __dev_forward_skb2(dev, skb, true); 2169 } 2170 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2171 2172 /** 2173 * dev_forward_skb - loopback an skb to another netif 2174 * 2175 * @dev: destination network device 2176 * @skb: buffer to forward 2177 * 2178 * return values: 2179 * NET_RX_SUCCESS (no congestion) 2180 * NET_RX_DROP (packet was dropped, but freed) 2181 * 2182 * dev_forward_skb can be used for injecting an skb from the 2183 * start_xmit function of one device into the receive queue 2184 * of another device. 2185 * 2186 * The receiving device may be in another namespace, so 2187 * we have to clear all information in the skb that could 2188 * impact namespace isolation. 2189 */ 2190 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2191 { 2192 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2193 } 2194 EXPORT_SYMBOL_GPL(dev_forward_skb); 2195 2196 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2197 { 2198 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2199 } 2200 2201 static inline int deliver_skb(struct sk_buff *skb, 2202 struct packet_type *pt_prev, 2203 struct net_device *orig_dev) 2204 { 2205 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2206 return -ENOMEM; 2207 refcount_inc(&skb->users); 2208 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2209 } 2210 2211 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2212 struct packet_type **pt, 2213 struct net_device *orig_dev, 2214 __be16 type, 2215 struct list_head *ptype_list) 2216 { 2217 struct packet_type *ptype, *pt_prev = *pt; 2218 2219 list_for_each_entry_rcu(ptype, ptype_list, list) { 2220 if (ptype->type != type) 2221 continue; 2222 if (pt_prev) 2223 deliver_skb(skb, pt_prev, orig_dev); 2224 pt_prev = ptype; 2225 } 2226 *pt = pt_prev; 2227 } 2228 2229 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2230 { 2231 if (!ptype->af_packet_priv || !skb->sk) 2232 return false; 2233 2234 if (ptype->id_match) 2235 return ptype->id_match(ptype, skb->sk); 2236 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2237 return true; 2238 2239 return false; 2240 } 2241 2242 /** 2243 * dev_nit_active - return true if any network interface taps are in use 2244 * 2245 * @dev: network device to check for the presence of taps 2246 */ 2247 bool dev_nit_active(struct net_device *dev) 2248 { 2249 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2250 } 2251 EXPORT_SYMBOL_GPL(dev_nit_active); 2252 2253 /* 2254 * Support routine. Sends outgoing frames to any network 2255 * taps currently in use. 2256 */ 2257 2258 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2259 { 2260 struct packet_type *ptype; 2261 struct sk_buff *skb2 = NULL; 2262 struct packet_type *pt_prev = NULL; 2263 struct list_head *ptype_list = &ptype_all; 2264 2265 rcu_read_lock(); 2266 again: 2267 list_for_each_entry_rcu(ptype, ptype_list, list) { 2268 if (ptype->ignore_outgoing) 2269 continue; 2270 2271 /* Never send packets back to the socket 2272 * they originated from - MvS (miquels@drinkel.ow.org) 2273 */ 2274 if (skb_loop_sk(ptype, skb)) 2275 continue; 2276 2277 if (pt_prev) { 2278 deliver_skb(skb2, pt_prev, skb->dev); 2279 pt_prev = ptype; 2280 continue; 2281 } 2282 2283 /* need to clone skb, done only once */ 2284 skb2 = skb_clone(skb, GFP_ATOMIC); 2285 if (!skb2) 2286 goto out_unlock; 2287 2288 net_timestamp_set(skb2); 2289 2290 /* skb->nh should be correctly 2291 * set by sender, so that the second statement is 2292 * just protection against buggy protocols. 2293 */ 2294 skb_reset_mac_header(skb2); 2295 2296 if (skb_network_header(skb2) < skb2->data || 2297 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2298 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2299 ntohs(skb2->protocol), 2300 dev->name); 2301 skb_reset_network_header(skb2); 2302 } 2303 2304 skb2->transport_header = skb2->network_header; 2305 skb2->pkt_type = PACKET_OUTGOING; 2306 pt_prev = ptype; 2307 } 2308 2309 if (ptype_list == &ptype_all) { 2310 ptype_list = &dev->ptype_all; 2311 goto again; 2312 } 2313 out_unlock: 2314 if (pt_prev) { 2315 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2316 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2317 else 2318 kfree_skb(skb2); 2319 } 2320 rcu_read_unlock(); 2321 } 2322 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2323 2324 /** 2325 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2326 * @dev: Network device 2327 * @txq: number of queues available 2328 * 2329 * If real_num_tx_queues is changed the tc mappings may no longer be 2330 * valid. To resolve this verify the tc mapping remains valid and if 2331 * not NULL the mapping. With no priorities mapping to this 2332 * offset/count pair it will no longer be used. In the worst case TC0 2333 * is invalid nothing can be done so disable priority mappings. If is 2334 * expected that drivers will fix this mapping if they can before 2335 * calling netif_set_real_num_tx_queues. 2336 */ 2337 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2338 { 2339 int i; 2340 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2341 2342 /* If TC0 is invalidated disable TC mapping */ 2343 if (tc->offset + tc->count > txq) { 2344 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2345 dev->num_tc = 0; 2346 return; 2347 } 2348 2349 /* Invalidated prio to tc mappings set to TC0 */ 2350 for (i = 1; i < TC_BITMASK + 1; i++) { 2351 int q = netdev_get_prio_tc_map(dev, i); 2352 2353 tc = &dev->tc_to_txq[q]; 2354 if (tc->offset + tc->count > txq) { 2355 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", 2356 i, q); 2357 netdev_set_prio_tc_map(dev, i, 0); 2358 } 2359 } 2360 } 2361 2362 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2363 { 2364 if (dev->num_tc) { 2365 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2366 int i; 2367 2368 /* walk through the TCs and see if it falls into any of them */ 2369 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2370 if ((txq - tc->offset) < tc->count) 2371 return i; 2372 } 2373 2374 /* didn't find it, just return -1 to indicate no match */ 2375 return -1; 2376 } 2377 2378 return 0; 2379 } 2380 EXPORT_SYMBOL(netdev_txq_to_tc); 2381 2382 #ifdef CONFIG_XPS 2383 static struct static_key xps_needed __read_mostly; 2384 static struct static_key xps_rxqs_needed __read_mostly; 2385 static DEFINE_MUTEX(xps_map_mutex); 2386 #define xmap_dereference(P) \ 2387 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2388 2389 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2390 struct xps_dev_maps *old_maps, int tci, u16 index) 2391 { 2392 struct xps_map *map = NULL; 2393 int pos; 2394 2395 map = xmap_dereference(dev_maps->attr_map[tci]); 2396 if (!map) 2397 return false; 2398 2399 for (pos = map->len; pos--;) { 2400 if (map->queues[pos] != index) 2401 continue; 2402 2403 if (map->len > 1) { 2404 map->queues[pos] = map->queues[--map->len]; 2405 break; 2406 } 2407 2408 if (old_maps) 2409 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2410 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2411 kfree_rcu(map, rcu); 2412 return false; 2413 } 2414 2415 return true; 2416 } 2417 2418 static bool remove_xps_queue_cpu(struct net_device *dev, 2419 struct xps_dev_maps *dev_maps, 2420 int cpu, u16 offset, u16 count) 2421 { 2422 int num_tc = dev_maps->num_tc; 2423 bool active = false; 2424 int tci; 2425 2426 for (tci = cpu * num_tc; num_tc--; tci++) { 2427 int i, j; 2428 2429 for (i = count, j = offset; i--; j++) { 2430 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2431 break; 2432 } 2433 2434 active |= i < 0; 2435 } 2436 2437 return active; 2438 } 2439 2440 static void reset_xps_maps(struct net_device *dev, 2441 struct xps_dev_maps *dev_maps, 2442 enum xps_map_type type) 2443 { 2444 static_key_slow_dec_cpuslocked(&xps_needed); 2445 if (type == XPS_RXQS) 2446 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2447 2448 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2449 2450 kfree_rcu(dev_maps, rcu); 2451 } 2452 2453 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2454 u16 offset, u16 count) 2455 { 2456 struct xps_dev_maps *dev_maps; 2457 bool active = false; 2458 int i, j; 2459 2460 dev_maps = xmap_dereference(dev->xps_maps[type]); 2461 if (!dev_maps) 2462 return; 2463 2464 for (j = 0; j < dev_maps->nr_ids; j++) 2465 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2466 if (!active) 2467 reset_xps_maps(dev, dev_maps, type); 2468 2469 if (type == XPS_CPUS) { 2470 for (i = offset + (count - 1); count--; i--) 2471 netdev_queue_numa_node_write( 2472 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2473 } 2474 } 2475 2476 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2477 u16 count) 2478 { 2479 if (!static_key_false(&xps_needed)) 2480 return; 2481 2482 cpus_read_lock(); 2483 mutex_lock(&xps_map_mutex); 2484 2485 if (static_key_false(&xps_rxqs_needed)) 2486 clean_xps_maps(dev, XPS_RXQS, offset, count); 2487 2488 clean_xps_maps(dev, XPS_CPUS, offset, count); 2489 2490 mutex_unlock(&xps_map_mutex); 2491 cpus_read_unlock(); 2492 } 2493 2494 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2495 { 2496 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2497 } 2498 2499 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2500 u16 index, bool is_rxqs_map) 2501 { 2502 struct xps_map *new_map; 2503 int alloc_len = XPS_MIN_MAP_ALLOC; 2504 int i, pos; 2505 2506 for (pos = 0; map && pos < map->len; pos++) { 2507 if (map->queues[pos] != index) 2508 continue; 2509 return map; 2510 } 2511 2512 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2513 if (map) { 2514 if (pos < map->alloc_len) 2515 return map; 2516 2517 alloc_len = map->alloc_len * 2; 2518 } 2519 2520 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2521 * map 2522 */ 2523 if (is_rxqs_map) 2524 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2525 else 2526 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2527 cpu_to_node(attr_index)); 2528 if (!new_map) 2529 return NULL; 2530 2531 for (i = 0; i < pos; i++) 2532 new_map->queues[i] = map->queues[i]; 2533 new_map->alloc_len = alloc_len; 2534 new_map->len = pos; 2535 2536 return new_map; 2537 } 2538 2539 /* Copy xps maps at a given index */ 2540 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2541 struct xps_dev_maps *new_dev_maps, int index, 2542 int tc, bool skip_tc) 2543 { 2544 int i, tci = index * dev_maps->num_tc; 2545 struct xps_map *map; 2546 2547 /* copy maps belonging to foreign traffic classes */ 2548 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2549 if (i == tc && skip_tc) 2550 continue; 2551 2552 /* fill in the new device map from the old device map */ 2553 map = xmap_dereference(dev_maps->attr_map[tci]); 2554 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2555 } 2556 } 2557 2558 /* Must be called under cpus_read_lock */ 2559 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2560 u16 index, enum xps_map_type type) 2561 { 2562 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2563 const unsigned long *online_mask = NULL; 2564 bool active = false, copy = false; 2565 int i, j, tci, numa_node_id = -2; 2566 int maps_sz, num_tc = 1, tc = 0; 2567 struct xps_map *map, *new_map; 2568 unsigned int nr_ids; 2569 2570 WARN_ON_ONCE(index >= dev->num_tx_queues); 2571 2572 if (dev->num_tc) { 2573 /* Do not allow XPS on subordinate device directly */ 2574 num_tc = dev->num_tc; 2575 if (num_tc < 0) 2576 return -EINVAL; 2577 2578 /* If queue belongs to subordinate dev use its map */ 2579 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2580 2581 tc = netdev_txq_to_tc(dev, index); 2582 if (tc < 0) 2583 return -EINVAL; 2584 } 2585 2586 mutex_lock(&xps_map_mutex); 2587 2588 dev_maps = xmap_dereference(dev->xps_maps[type]); 2589 if (type == XPS_RXQS) { 2590 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2591 nr_ids = dev->num_rx_queues; 2592 } else { 2593 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2594 if (num_possible_cpus() > 1) 2595 online_mask = cpumask_bits(cpu_online_mask); 2596 nr_ids = nr_cpu_ids; 2597 } 2598 2599 if (maps_sz < L1_CACHE_BYTES) 2600 maps_sz = L1_CACHE_BYTES; 2601 2602 /* The old dev_maps could be larger or smaller than the one we're 2603 * setting up now, as dev->num_tc or nr_ids could have been updated in 2604 * between. We could try to be smart, but let's be safe instead and only 2605 * copy foreign traffic classes if the two map sizes match. 2606 */ 2607 if (dev_maps && 2608 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2609 copy = true; 2610 2611 /* allocate memory for queue storage */ 2612 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2613 j < nr_ids;) { 2614 if (!new_dev_maps) { 2615 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2616 if (!new_dev_maps) { 2617 mutex_unlock(&xps_map_mutex); 2618 return -ENOMEM; 2619 } 2620 2621 new_dev_maps->nr_ids = nr_ids; 2622 new_dev_maps->num_tc = num_tc; 2623 } 2624 2625 tci = j * num_tc + tc; 2626 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2627 2628 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2629 if (!map) 2630 goto error; 2631 2632 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2633 } 2634 2635 if (!new_dev_maps) 2636 goto out_no_new_maps; 2637 2638 if (!dev_maps) { 2639 /* Increment static keys at most once per type */ 2640 static_key_slow_inc_cpuslocked(&xps_needed); 2641 if (type == XPS_RXQS) 2642 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2643 } 2644 2645 for (j = 0; j < nr_ids; j++) { 2646 bool skip_tc = false; 2647 2648 tci = j * num_tc + tc; 2649 if (netif_attr_test_mask(j, mask, nr_ids) && 2650 netif_attr_test_online(j, online_mask, nr_ids)) { 2651 /* add tx-queue to CPU/rx-queue maps */ 2652 int pos = 0; 2653 2654 skip_tc = true; 2655 2656 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2657 while ((pos < map->len) && (map->queues[pos] != index)) 2658 pos++; 2659 2660 if (pos == map->len) 2661 map->queues[map->len++] = index; 2662 #ifdef CONFIG_NUMA 2663 if (type == XPS_CPUS) { 2664 if (numa_node_id == -2) 2665 numa_node_id = cpu_to_node(j); 2666 else if (numa_node_id != cpu_to_node(j)) 2667 numa_node_id = -1; 2668 } 2669 #endif 2670 } 2671 2672 if (copy) 2673 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2674 skip_tc); 2675 } 2676 2677 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2678 2679 /* Cleanup old maps */ 2680 if (!dev_maps) 2681 goto out_no_old_maps; 2682 2683 for (j = 0; j < dev_maps->nr_ids; j++) { 2684 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2685 map = xmap_dereference(dev_maps->attr_map[tci]); 2686 if (!map) 2687 continue; 2688 2689 if (copy) { 2690 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2691 if (map == new_map) 2692 continue; 2693 } 2694 2695 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2696 kfree_rcu(map, rcu); 2697 } 2698 } 2699 2700 old_dev_maps = dev_maps; 2701 2702 out_no_old_maps: 2703 dev_maps = new_dev_maps; 2704 active = true; 2705 2706 out_no_new_maps: 2707 if (type == XPS_CPUS) 2708 /* update Tx queue numa node */ 2709 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2710 (numa_node_id >= 0) ? 2711 numa_node_id : NUMA_NO_NODE); 2712 2713 if (!dev_maps) 2714 goto out_no_maps; 2715 2716 /* removes tx-queue from unused CPUs/rx-queues */ 2717 for (j = 0; j < dev_maps->nr_ids; j++) { 2718 tci = j * dev_maps->num_tc; 2719 2720 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2721 if (i == tc && 2722 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2723 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2724 continue; 2725 2726 active |= remove_xps_queue(dev_maps, 2727 copy ? old_dev_maps : NULL, 2728 tci, index); 2729 } 2730 } 2731 2732 if (old_dev_maps) 2733 kfree_rcu(old_dev_maps, rcu); 2734 2735 /* free map if not active */ 2736 if (!active) 2737 reset_xps_maps(dev, dev_maps, type); 2738 2739 out_no_maps: 2740 mutex_unlock(&xps_map_mutex); 2741 2742 return 0; 2743 error: 2744 /* remove any maps that we added */ 2745 for (j = 0; j < nr_ids; j++) { 2746 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2747 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2748 map = copy ? 2749 xmap_dereference(dev_maps->attr_map[tci]) : 2750 NULL; 2751 if (new_map && new_map != map) 2752 kfree(new_map); 2753 } 2754 } 2755 2756 mutex_unlock(&xps_map_mutex); 2757 2758 kfree(new_dev_maps); 2759 return -ENOMEM; 2760 } 2761 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2762 2763 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2764 u16 index) 2765 { 2766 int ret; 2767 2768 cpus_read_lock(); 2769 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2770 cpus_read_unlock(); 2771 2772 return ret; 2773 } 2774 EXPORT_SYMBOL(netif_set_xps_queue); 2775 2776 #endif 2777 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2778 { 2779 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2780 2781 /* Unbind any subordinate channels */ 2782 while (txq-- != &dev->_tx[0]) { 2783 if (txq->sb_dev) 2784 netdev_unbind_sb_channel(dev, txq->sb_dev); 2785 } 2786 } 2787 2788 void netdev_reset_tc(struct net_device *dev) 2789 { 2790 #ifdef CONFIG_XPS 2791 netif_reset_xps_queues_gt(dev, 0); 2792 #endif 2793 netdev_unbind_all_sb_channels(dev); 2794 2795 /* Reset TC configuration of device */ 2796 dev->num_tc = 0; 2797 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2798 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2799 } 2800 EXPORT_SYMBOL(netdev_reset_tc); 2801 2802 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2803 { 2804 if (tc >= dev->num_tc) 2805 return -EINVAL; 2806 2807 #ifdef CONFIG_XPS 2808 netif_reset_xps_queues(dev, offset, count); 2809 #endif 2810 dev->tc_to_txq[tc].count = count; 2811 dev->tc_to_txq[tc].offset = offset; 2812 return 0; 2813 } 2814 EXPORT_SYMBOL(netdev_set_tc_queue); 2815 2816 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2817 { 2818 if (num_tc > TC_MAX_QUEUE) 2819 return -EINVAL; 2820 2821 #ifdef CONFIG_XPS 2822 netif_reset_xps_queues_gt(dev, 0); 2823 #endif 2824 netdev_unbind_all_sb_channels(dev); 2825 2826 dev->num_tc = num_tc; 2827 return 0; 2828 } 2829 EXPORT_SYMBOL(netdev_set_num_tc); 2830 2831 void netdev_unbind_sb_channel(struct net_device *dev, 2832 struct net_device *sb_dev) 2833 { 2834 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2835 2836 #ifdef CONFIG_XPS 2837 netif_reset_xps_queues_gt(sb_dev, 0); 2838 #endif 2839 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2840 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2841 2842 while (txq-- != &dev->_tx[0]) { 2843 if (txq->sb_dev == sb_dev) 2844 txq->sb_dev = NULL; 2845 } 2846 } 2847 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2848 2849 int netdev_bind_sb_channel_queue(struct net_device *dev, 2850 struct net_device *sb_dev, 2851 u8 tc, u16 count, u16 offset) 2852 { 2853 /* Make certain the sb_dev and dev are already configured */ 2854 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2855 return -EINVAL; 2856 2857 /* We cannot hand out queues we don't have */ 2858 if ((offset + count) > dev->real_num_tx_queues) 2859 return -EINVAL; 2860 2861 /* Record the mapping */ 2862 sb_dev->tc_to_txq[tc].count = count; 2863 sb_dev->tc_to_txq[tc].offset = offset; 2864 2865 /* Provide a way for Tx queue to find the tc_to_txq map or 2866 * XPS map for itself. 2867 */ 2868 while (count--) 2869 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2870 2871 return 0; 2872 } 2873 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2874 2875 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2876 { 2877 /* Do not use a multiqueue device to represent a subordinate channel */ 2878 if (netif_is_multiqueue(dev)) 2879 return -ENODEV; 2880 2881 /* We allow channels 1 - 32767 to be used for subordinate channels. 2882 * Channel 0 is meant to be "native" mode and used only to represent 2883 * the main root device. We allow writing 0 to reset the device back 2884 * to normal mode after being used as a subordinate channel. 2885 */ 2886 if (channel > S16_MAX) 2887 return -EINVAL; 2888 2889 dev->num_tc = -channel; 2890 2891 return 0; 2892 } 2893 EXPORT_SYMBOL(netdev_set_sb_channel); 2894 2895 /* 2896 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2897 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2898 */ 2899 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2900 { 2901 bool disabling; 2902 int rc; 2903 2904 disabling = txq < dev->real_num_tx_queues; 2905 2906 if (txq < 1 || txq > dev->num_tx_queues) 2907 return -EINVAL; 2908 2909 if (dev->reg_state == NETREG_REGISTERED || 2910 dev->reg_state == NETREG_UNREGISTERING) { 2911 ASSERT_RTNL(); 2912 2913 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2914 txq); 2915 if (rc) 2916 return rc; 2917 2918 if (dev->num_tc) 2919 netif_setup_tc(dev, txq); 2920 2921 dev_qdisc_change_real_num_tx(dev, txq); 2922 2923 dev->real_num_tx_queues = txq; 2924 2925 if (disabling) { 2926 synchronize_net(); 2927 qdisc_reset_all_tx_gt(dev, txq); 2928 #ifdef CONFIG_XPS 2929 netif_reset_xps_queues_gt(dev, txq); 2930 #endif 2931 } 2932 } else { 2933 dev->real_num_tx_queues = txq; 2934 } 2935 2936 return 0; 2937 } 2938 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2939 2940 #ifdef CONFIG_SYSFS 2941 /** 2942 * netif_set_real_num_rx_queues - set actual number of RX queues used 2943 * @dev: Network device 2944 * @rxq: Actual number of RX queues 2945 * 2946 * This must be called either with the rtnl_lock held or before 2947 * registration of the net device. Returns 0 on success, or a 2948 * negative error code. If called before registration, it always 2949 * succeeds. 2950 */ 2951 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2952 { 2953 int rc; 2954 2955 if (rxq < 1 || rxq > dev->num_rx_queues) 2956 return -EINVAL; 2957 2958 if (dev->reg_state == NETREG_REGISTERED) { 2959 ASSERT_RTNL(); 2960 2961 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2962 rxq); 2963 if (rc) 2964 return rc; 2965 } 2966 2967 dev->real_num_rx_queues = rxq; 2968 return 0; 2969 } 2970 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2971 #endif 2972 2973 /** 2974 * netif_set_real_num_queues - set actual number of RX and TX queues used 2975 * @dev: Network device 2976 * @txq: Actual number of TX queues 2977 * @rxq: Actual number of RX queues 2978 * 2979 * Set the real number of both TX and RX queues. 2980 * Does nothing if the number of queues is already correct. 2981 */ 2982 int netif_set_real_num_queues(struct net_device *dev, 2983 unsigned int txq, unsigned int rxq) 2984 { 2985 unsigned int old_rxq = dev->real_num_rx_queues; 2986 int err; 2987 2988 if (txq < 1 || txq > dev->num_tx_queues || 2989 rxq < 1 || rxq > dev->num_rx_queues) 2990 return -EINVAL; 2991 2992 /* Start from increases, so the error path only does decreases - 2993 * decreases can't fail. 2994 */ 2995 if (rxq > dev->real_num_rx_queues) { 2996 err = netif_set_real_num_rx_queues(dev, rxq); 2997 if (err) 2998 return err; 2999 } 3000 if (txq > dev->real_num_tx_queues) { 3001 err = netif_set_real_num_tx_queues(dev, txq); 3002 if (err) 3003 goto undo_rx; 3004 } 3005 if (rxq < dev->real_num_rx_queues) 3006 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3007 if (txq < dev->real_num_tx_queues) 3008 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3009 3010 return 0; 3011 undo_rx: 3012 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3013 return err; 3014 } 3015 EXPORT_SYMBOL(netif_set_real_num_queues); 3016 3017 /** 3018 * netif_set_tso_max_size() - set the max size of TSO frames supported 3019 * @dev: netdev to update 3020 * @size: max skb->len of a TSO frame 3021 * 3022 * Set the limit on the size of TSO super-frames the device can handle. 3023 * Unless explicitly set the stack will assume the value of 3024 * %GSO_LEGACY_MAX_SIZE. 3025 */ 3026 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3027 { 3028 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3029 if (size < READ_ONCE(dev->gso_max_size)) 3030 netif_set_gso_max_size(dev, size); 3031 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3032 netif_set_gso_ipv4_max_size(dev, size); 3033 } 3034 EXPORT_SYMBOL(netif_set_tso_max_size); 3035 3036 /** 3037 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3038 * @dev: netdev to update 3039 * @segs: max number of TCP segments 3040 * 3041 * Set the limit on the number of TCP segments the device can generate from 3042 * a single TSO super-frame. 3043 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3044 */ 3045 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3046 { 3047 dev->tso_max_segs = segs; 3048 if (segs < READ_ONCE(dev->gso_max_segs)) 3049 netif_set_gso_max_segs(dev, segs); 3050 } 3051 EXPORT_SYMBOL(netif_set_tso_max_segs); 3052 3053 /** 3054 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3055 * @to: netdev to update 3056 * @from: netdev from which to copy the limits 3057 */ 3058 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3059 { 3060 netif_set_tso_max_size(to, from->tso_max_size); 3061 netif_set_tso_max_segs(to, from->tso_max_segs); 3062 } 3063 EXPORT_SYMBOL(netif_inherit_tso_max); 3064 3065 /** 3066 * netif_get_num_default_rss_queues - default number of RSS queues 3067 * 3068 * Default value is the number of physical cores if there are only 1 or 2, or 3069 * divided by 2 if there are more. 3070 */ 3071 int netif_get_num_default_rss_queues(void) 3072 { 3073 cpumask_var_t cpus; 3074 int cpu, count = 0; 3075 3076 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3077 return 1; 3078 3079 cpumask_copy(cpus, cpu_online_mask); 3080 for_each_cpu(cpu, cpus) { 3081 ++count; 3082 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3083 } 3084 free_cpumask_var(cpus); 3085 3086 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3087 } 3088 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3089 3090 static void __netif_reschedule(struct Qdisc *q) 3091 { 3092 struct softnet_data *sd; 3093 unsigned long flags; 3094 3095 local_irq_save(flags); 3096 sd = this_cpu_ptr(&softnet_data); 3097 q->next_sched = NULL; 3098 *sd->output_queue_tailp = q; 3099 sd->output_queue_tailp = &q->next_sched; 3100 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3101 local_irq_restore(flags); 3102 } 3103 3104 void __netif_schedule(struct Qdisc *q) 3105 { 3106 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3107 __netif_reschedule(q); 3108 } 3109 EXPORT_SYMBOL(__netif_schedule); 3110 3111 struct dev_kfree_skb_cb { 3112 enum skb_drop_reason reason; 3113 }; 3114 3115 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3116 { 3117 return (struct dev_kfree_skb_cb *)skb->cb; 3118 } 3119 3120 void netif_schedule_queue(struct netdev_queue *txq) 3121 { 3122 rcu_read_lock(); 3123 if (!netif_xmit_stopped(txq)) { 3124 struct Qdisc *q = rcu_dereference(txq->qdisc); 3125 3126 __netif_schedule(q); 3127 } 3128 rcu_read_unlock(); 3129 } 3130 EXPORT_SYMBOL(netif_schedule_queue); 3131 3132 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3133 { 3134 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3135 struct Qdisc *q; 3136 3137 rcu_read_lock(); 3138 q = rcu_dereference(dev_queue->qdisc); 3139 __netif_schedule(q); 3140 rcu_read_unlock(); 3141 } 3142 } 3143 EXPORT_SYMBOL(netif_tx_wake_queue); 3144 3145 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3146 { 3147 unsigned long flags; 3148 3149 if (unlikely(!skb)) 3150 return; 3151 3152 if (likely(refcount_read(&skb->users) == 1)) { 3153 smp_rmb(); 3154 refcount_set(&skb->users, 0); 3155 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3156 return; 3157 } 3158 get_kfree_skb_cb(skb)->reason = reason; 3159 local_irq_save(flags); 3160 skb->next = __this_cpu_read(softnet_data.completion_queue); 3161 __this_cpu_write(softnet_data.completion_queue, skb); 3162 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3163 local_irq_restore(flags); 3164 } 3165 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3166 3167 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3168 { 3169 if (in_hardirq() || irqs_disabled()) 3170 dev_kfree_skb_irq_reason(skb, reason); 3171 else 3172 kfree_skb_reason(skb, reason); 3173 } 3174 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3175 3176 3177 /** 3178 * netif_device_detach - mark device as removed 3179 * @dev: network device 3180 * 3181 * Mark device as removed from system and therefore no longer available. 3182 */ 3183 void netif_device_detach(struct net_device *dev) 3184 { 3185 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3186 netif_running(dev)) { 3187 netif_tx_stop_all_queues(dev); 3188 } 3189 } 3190 EXPORT_SYMBOL(netif_device_detach); 3191 3192 /** 3193 * netif_device_attach - mark device as attached 3194 * @dev: network device 3195 * 3196 * Mark device as attached from system and restart if needed. 3197 */ 3198 void netif_device_attach(struct net_device *dev) 3199 { 3200 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3201 netif_running(dev)) { 3202 netif_tx_wake_all_queues(dev); 3203 __netdev_watchdog_up(dev); 3204 } 3205 } 3206 EXPORT_SYMBOL(netif_device_attach); 3207 3208 /* 3209 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3210 * to be used as a distribution range. 3211 */ 3212 static u16 skb_tx_hash(const struct net_device *dev, 3213 const struct net_device *sb_dev, 3214 struct sk_buff *skb) 3215 { 3216 u32 hash; 3217 u16 qoffset = 0; 3218 u16 qcount = dev->real_num_tx_queues; 3219 3220 if (dev->num_tc) { 3221 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3222 3223 qoffset = sb_dev->tc_to_txq[tc].offset; 3224 qcount = sb_dev->tc_to_txq[tc].count; 3225 if (unlikely(!qcount)) { 3226 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3227 sb_dev->name, qoffset, tc); 3228 qoffset = 0; 3229 qcount = dev->real_num_tx_queues; 3230 } 3231 } 3232 3233 if (skb_rx_queue_recorded(skb)) { 3234 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3235 hash = skb_get_rx_queue(skb); 3236 if (hash >= qoffset) 3237 hash -= qoffset; 3238 while (unlikely(hash >= qcount)) 3239 hash -= qcount; 3240 return hash + qoffset; 3241 } 3242 3243 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3244 } 3245 3246 void skb_warn_bad_offload(const struct sk_buff *skb) 3247 { 3248 static const netdev_features_t null_features; 3249 struct net_device *dev = skb->dev; 3250 const char *name = ""; 3251 3252 if (!net_ratelimit()) 3253 return; 3254 3255 if (dev) { 3256 if (dev->dev.parent) 3257 name = dev_driver_string(dev->dev.parent); 3258 else 3259 name = netdev_name(dev); 3260 } 3261 skb_dump(KERN_WARNING, skb, false); 3262 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3263 name, dev ? &dev->features : &null_features, 3264 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3265 } 3266 3267 /* 3268 * Invalidate hardware checksum when packet is to be mangled, and 3269 * complete checksum manually on outgoing path. 3270 */ 3271 int skb_checksum_help(struct sk_buff *skb) 3272 { 3273 __wsum csum; 3274 int ret = 0, offset; 3275 3276 if (skb->ip_summed == CHECKSUM_COMPLETE) 3277 goto out_set_summed; 3278 3279 if (unlikely(skb_is_gso(skb))) { 3280 skb_warn_bad_offload(skb); 3281 return -EINVAL; 3282 } 3283 3284 /* Before computing a checksum, we should make sure no frag could 3285 * be modified by an external entity : checksum could be wrong. 3286 */ 3287 if (skb_has_shared_frag(skb)) { 3288 ret = __skb_linearize(skb); 3289 if (ret) 3290 goto out; 3291 } 3292 3293 offset = skb_checksum_start_offset(skb); 3294 ret = -EINVAL; 3295 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3296 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3297 goto out; 3298 } 3299 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3300 3301 offset += skb->csum_offset; 3302 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3303 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3304 goto out; 3305 } 3306 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3307 if (ret) 3308 goto out; 3309 3310 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3311 out_set_summed: 3312 skb->ip_summed = CHECKSUM_NONE; 3313 out: 3314 return ret; 3315 } 3316 EXPORT_SYMBOL(skb_checksum_help); 3317 3318 int skb_crc32c_csum_help(struct sk_buff *skb) 3319 { 3320 __le32 crc32c_csum; 3321 int ret = 0, offset, start; 3322 3323 if (skb->ip_summed != CHECKSUM_PARTIAL) 3324 goto out; 3325 3326 if (unlikely(skb_is_gso(skb))) 3327 goto out; 3328 3329 /* Before computing a checksum, we should make sure no frag could 3330 * be modified by an external entity : checksum could be wrong. 3331 */ 3332 if (unlikely(skb_has_shared_frag(skb))) { 3333 ret = __skb_linearize(skb); 3334 if (ret) 3335 goto out; 3336 } 3337 start = skb_checksum_start_offset(skb); 3338 offset = start + offsetof(struct sctphdr, checksum); 3339 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3340 ret = -EINVAL; 3341 goto out; 3342 } 3343 3344 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3345 if (ret) 3346 goto out; 3347 3348 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3349 skb->len - start, ~(__u32)0, 3350 crc32c_csum_stub)); 3351 *(__le32 *)(skb->data + offset) = crc32c_csum; 3352 skb_reset_csum_not_inet(skb); 3353 out: 3354 return ret; 3355 } 3356 3357 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3358 { 3359 __be16 type = skb->protocol; 3360 3361 /* Tunnel gso handlers can set protocol to ethernet. */ 3362 if (type == htons(ETH_P_TEB)) { 3363 struct ethhdr *eth; 3364 3365 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3366 return 0; 3367 3368 eth = (struct ethhdr *)skb->data; 3369 type = eth->h_proto; 3370 } 3371 3372 return vlan_get_protocol_and_depth(skb, type, depth); 3373 } 3374 3375 3376 /* Take action when hardware reception checksum errors are detected. */ 3377 #ifdef CONFIG_BUG 3378 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3379 { 3380 netdev_err(dev, "hw csum failure\n"); 3381 skb_dump(KERN_ERR, skb, true); 3382 dump_stack(); 3383 } 3384 3385 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3386 { 3387 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3388 } 3389 EXPORT_SYMBOL(netdev_rx_csum_fault); 3390 #endif 3391 3392 /* XXX: check that highmem exists at all on the given machine. */ 3393 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3394 { 3395 #ifdef CONFIG_HIGHMEM 3396 int i; 3397 3398 if (!(dev->features & NETIF_F_HIGHDMA)) { 3399 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3400 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3401 3402 if (PageHighMem(skb_frag_page(frag))) 3403 return 1; 3404 } 3405 } 3406 #endif 3407 return 0; 3408 } 3409 3410 /* If MPLS offload request, verify we are testing hardware MPLS features 3411 * instead of standard features for the netdev. 3412 */ 3413 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3414 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3415 netdev_features_t features, 3416 __be16 type) 3417 { 3418 if (eth_p_mpls(type)) 3419 features &= skb->dev->mpls_features; 3420 3421 return features; 3422 } 3423 #else 3424 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3425 netdev_features_t features, 3426 __be16 type) 3427 { 3428 return features; 3429 } 3430 #endif 3431 3432 static netdev_features_t harmonize_features(struct sk_buff *skb, 3433 netdev_features_t features) 3434 { 3435 __be16 type; 3436 3437 type = skb_network_protocol(skb, NULL); 3438 features = net_mpls_features(skb, features, type); 3439 3440 if (skb->ip_summed != CHECKSUM_NONE && 3441 !can_checksum_protocol(features, type)) { 3442 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3443 } 3444 if (illegal_highdma(skb->dev, skb)) 3445 features &= ~NETIF_F_SG; 3446 3447 return features; 3448 } 3449 3450 netdev_features_t passthru_features_check(struct sk_buff *skb, 3451 struct net_device *dev, 3452 netdev_features_t features) 3453 { 3454 return features; 3455 } 3456 EXPORT_SYMBOL(passthru_features_check); 3457 3458 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3459 struct net_device *dev, 3460 netdev_features_t features) 3461 { 3462 return vlan_features_check(skb, features); 3463 } 3464 3465 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3466 struct net_device *dev, 3467 netdev_features_t features) 3468 { 3469 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3470 3471 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3472 return features & ~NETIF_F_GSO_MASK; 3473 3474 if (!skb_shinfo(skb)->gso_type) { 3475 skb_warn_bad_offload(skb); 3476 return features & ~NETIF_F_GSO_MASK; 3477 } 3478 3479 /* Support for GSO partial features requires software 3480 * intervention before we can actually process the packets 3481 * so we need to strip support for any partial features now 3482 * and we can pull them back in after we have partially 3483 * segmented the frame. 3484 */ 3485 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3486 features &= ~dev->gso_partial_features; 3487 3488 /* Make sure to clear the IPv4 ID mangling feature if the 3489 * IPv4 header has the potential to be fragmented. 3490 */ 3491 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3492 struct iphdr *iph = skb->encapsulation ? 3493 inner_ip_hdr(skb) : ip_hdr(skb); 3494 3495 if (!(iph->frag_off & htons(IP_DF))) 3496 features &= ~NETIF_F_TSO_MANGLEID; 3497 } 3498 3499 return features; 3500 } 3501 3502 netdev_features_t netif_skb_features(struct sk_buff *skb) 3503 { 3504 struct net_device *dev = skb->dev; 3505 netdev_features_t features = dev->features; 3506 3507 if (skb_is_gso(skb)) 3508 features = gso_features_check(skb, dev, features); 3509 3510 /* If encapsulation offload request, verify we are testing 3511 * hardware encapsulation features instead of standard 3512 * features for the netdev 3513 */ 3514 if (skb->encapsulation) 3515 features &= dev->hw_enc_features; 3516 3517 if (skb_vlan_tagged(skb)) 3518 features = netdev_intersect_features(features, 3519 dev->vlan_features | 3520 NETIF_F_HW_VLAN_CTAG_TX | 3521 NETIF_F_HW_VLAN_STAG_TX); 3522 3523 if (dev->netdev_ops->ndo_features_check) 3524 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3525 features); 3526 else 3527 features &= dflt_features_check(skb, dev, features); 3528 3529 return harmonize_features(skb, features); 3530 } 3531 EXPORT_SYMBOL(netif_skb_features); 3532 3533 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3534 struct netdev_queue *txq, bool more) 3535 { 3536 unsigned int len; 3537 int rc; 3538 3539 if (dev_nit_active(dev)) 3540 dev_queue_xmit_nit(skb, dev); 3541 3542 len = skb->len; 3543 trace_net_dev_start_xmit(skb, dev); 3544 rc = netdev_start_xmit(skb, dev, txq, more); 3545 trace_net_dev_xmit(skb, rc, dev, len); 3546 3547 return rc; 3548 } 3549 3550 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3551 struct netdev_queue *txq, int *ret) 3552 { 3553 struct sk_buff *skb = first; 3554 int rc = NETDEV_TX_OK; 3555 3556 while (skb) { 3557 struct sk_buff *next = skb->next; 3558 3559 skb_mark_not_on_list(skb); 3560 rc = xmit_one(skb, dev, txq, next != NULL); 3561 if (unlikely(!dev_xmit_complete(rc))) { 3562 skb->next = next; 3563 goto out; 3564 } 3565 3566 skb = next; 3567 if (netif_tx_queue_stopped(txq) && skb) { 3568 rc = NETDEV_TX_BUSY; 3569 break; 3570 } 3571 } 3572 3573 out: 3574 *ret = rc; 3575 return skb; 3576 } 3577 3578 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3579 netdev_features_t features) 3580 { 3581 if (skb_vlan_tag_present(skb) && 3582 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3583 skb = __vlan_hwaccel_push_inside(skb); 3584 return skb; 3585 } 3586 3587 int skb_csum_hwoffload_help(struct sk_buff *skb, 3588 const netdev_features_t features) 3589 { 3590 if (unlikely(skb_csum_is_sctp(skb))) 3591 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3592 skb_crc32c_csum_help(skb); 3593 3594 if (features & NETIF_F_HW_CSUM) 3595 return 0; 3596 3597 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3598 switch (skb->csum_offset) { 3599 case offsetof(struct tcphdr, check): 3600 case offsetof(struct udphdr, check): 3601 return 0; 3602 } 3603 } 3604 3605 return skb_checksum_help(skb); 3606 } 3607 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3608 3609 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3610 { 3611 netdev_features_t features; 3612 3613 features = netif_skb_features(skb); 3614 skb = validate_xmit_vlan(skb, features); 3615 if (unlikely(!skb)) 3616 goto out_null; 3617 3618 skb = sk_validate_xmit_skb(skb, dev); 3619 if (unlikely(!skb)) 3620 goto out_null; 3621 3622 if (netif_needs_gso(skb, features)) { 3623 struct sk_buff *segs; 3624 3625 segs = skb_gso_segment(skb, features); 3626 if (IS_ERR(segs)) { 3627 goto out_kfree_skb; 3628 } else if (segs) { 3629 consume_skb(skb); 3630 skb = segs; 3631 } 3632 } else { 3633 if (skb_needs_linearize(skb, features) && 3634 __skb_linearize(skb)) 3635 goto out_kfree_skb; 3636 3637 /* If packet is not checksummed and device does not 3638 * support checksumming for this protocol, complete 3639 * checksumming here. 3640 */ 3641 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3642 if (skb->encapsulation) 3643 skb_set_inner_transport_header(skb, 3644 skb_checksum_start_offset(skb)); 3645 else 3646 skb_set_transport_header(skb, 3647 skb_checksum_start_offset(skb)); 3648 if (skb_csum_hwoffload_help(skb, features)) 3649 goto out_kfree_skb; 3650 } 3651 } 3652 3653 skb = validate_xmit_xfrm(skb, features, again); 3654 3655 return skb; 3656 3657 out_kfree_skb: 3658 kfree_skb(skb); 3659 out_null: 3660 dev_core_stats_tx_dropped_inc(dev); 3661 return NULL; 3662 } 3663 3664 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3665 { 3666 struct sk_buff *next, *head = NULL, *tail; 3667 3668 for (; skb != NULL; skb = next) { 3669 next = skb->next; 3670 skb_mark_not_on_list(skb); 3671 3672 /* in case skb wont be segmented, point to itself */ 3673 skb->prev = skb; 3674 3675 skb = validate_xmit_skb(skb, dev, again); 3676 if (!skb) 3677 continue; 3678 3679 if (!head) 3680 head = skb; 3681 else 3682 tail->next = skb; 3683 /* If skb was segmented, skb->prev points to 3684 * the last segment. If not, it still contains skb. 3685 */ 3686 tail = skb->prev; 3687 } 3688 return head; 3689 } 3690 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3691 3692 static void qdisc_pkt_len_init(struct sk_buff *skb) 3693 { 3694 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3695 3696 qdisc_skb_cb(skb)->pkt_len = skb->len; 3697 3698 /* To get more precise estimation of bytes sent on wire, 3699 * we add to pkt_len the headers size of all segments 3700 */ 3701 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3702 u16 gso_segs = shinfo->gso_segs; 3703 unsigned int hdr_len; 3704 3705 /* mac layer + network layer */ 3706 hdr_len = skb_transport_offset(skb); 3707 3708 /* + transport layer */ 3709 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3710 const struct tcphdr *th; 3711 struct tcphdr _tcphdr; 3712 3713 th = skb_header_pointer(skb, hdr_len, 3714 sizeof(_tcphdr), &_tcphdr); 3715 if (likely(th)) 3716 hdr_len += __tcp_hdrlen(th); 3717 } else { 3718 struct udphdr _udphdr; 3719 3720 if (skb_header_pointer(skb, hdr_len, 3721 sizeof(_udphdr), &_udphdr)) 3722 hdr_len += sizeof(struct udphdr); 3723 } 3724 3725 if (shinfo->gso_type & SKB_GSO_DODGY) 3726 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3727 shinfo->gso_size); 3728 3729 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3730 } 3731 } 3732 3733 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3734 struct sk_buff **to_free, 3735 struct netdev_queue *txq) 3736 { 3737 int rc; 3738 3739 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3740 if (rc == NET_XMIT_SUCCESS) 3741 trace_qdisc_enqueue(q, txq, skb); 3742 return rc; 3743 } 3744 3745 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3746 struct net_device *dev, 3747 struct netdev_queue *txq) 3748 { 3749 spinlock_t *root_lock = qdisc_lock(q); 3750 struct sk_buff *to_free = NULL; 3751 bool contended; 3752 int rc; 3753 3754 qdisc_calculate_pkt_len(skb, q); 3755 3756 if (q->flags & TCQ_F_NOLOCK) { 3757 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3758 qdisc_run_begin(q)) { 3759 /* Retest nolock_qdisc_is_empty() within the protection 3760 * of q->seqlock to protect from racing with requeuing. 3761 */ 3762 if (unlikely(!nolock_qdisc_is_empty(q))) { 3763 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3764 __qdisc_run(q); 3765 qdisc_run_end(q); 3766 3767 goto no_lock_out; 3768 } 3769 3770 qdisc_bstats_cpu_update(q, skb); 3771 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3772 !nolock_qdisc_is_empty(q)) 3773 __qdisc_run(q); 3774 3775 qdisc_run_end(q); 3776 return NET_XMIT_SUCCESS; 3777 } 3778 3779 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3780 qdisc_run(q); 3781 3782 no_lock_out: 3783 if (unlikely(to_free)) 3784 kfree_skb_list_reason(to_free, 3785 SKB_DROP_REASON_QDISC_DROP); 3786 return rc; 3787 } 3788 3789 /* 3790 * Heuristic to force contended enqueues to serialize on a 3791 * separate lock before trying to get qdisc main lock. 3792 * This permits qdisc->running owner to get the lock more 3793 * often and dequeue packets faster. 3794 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3795 * and then other tasks will only enqueue packets. The packets will be 3796 * sent after the qdisc owner is scheduled again. To prevent this 3797 * scenario the task always serialize on the lock. 3798 */ 3799 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3800 if (unlikely(contended)) 3801 spin_lock(&q->busylock); 3802 3803 spin_lock(root_lock); 3804 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3805 __qdisc_drop(skb, &to_free); 3806 rc = NET_XMIT_DROP; 3807 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3808 qdisc_run_begin(q)) { 3809 /* 3810 * This is a work-conserving queue; there are no old skbs 3811 * waiting to be sent out; and the qdisc is not running - 3812 * xmit the skb directly. 3813 */ 3814 3815 qdisc_bstats_update(q, skb); 3816 3817 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3818 if (unlikely(contended)) { 3819 spin_unlock(&q->busylock); 3820 contended = false; 3821 } 3822 __qdisc_run(q); 3823 } 3824 3825 qdisc_run_end(q); 3826 rc = NET_XMIT_SUCCESS; 3827 } else { 3828 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3829 if (qdisc_run_begin(q)) { 3830 if (unlikely(contended)) { 3831 spin_unlock(&q->busylock); 3832 contended = false; 3833 } 3834 __qdisc_run(q); 3835 qdisc_run_end(q); 3836 } 3837 } 3838 spin_unlock(root_lock); 3839 if (unlikely(to_free)) 3840 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3841 if (unlikely(contended)) 3842 spin_unlock(&q->busylock); 3843 return rc; 3844 } 3845 3846 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3847 static void skb_update_prio(struct sk_buff *skb) 3848 { 3849 const struct netprio_map *map; 3850 const struct sock *sk; 3851 unsigned int prioidx; 3852 3853 if (skb->priority) 3854 return; 3855 map = rcu_dereference_bh(skb->dev->priomap); 3856 if (!map) 3857 return; 3858 sk = skb_to_full_sk(skb); 3859 if (!sk) 3860 return; 3861 3862 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3863 3864 if (prioidx < map->priomap_len) 3865 skb->priority = map->priomap[prioidx]; 3866 } 3867 #else 3868 #define skb_update_prio(skb) 3869 #endif 3870 3871 /** 3872 * dev_loopback_xmit - loop back @skb 3873 * @net: network namespace this loopback is happening in 3874 * @sk: sk needed to be a netfilter okfn 3875 * @skb: buffer to transmit 3876 */ 3877 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3878 { 3879 skb_reset_mac_header(skb); 3880 __skb_pull(skb, skb_network_offset(skb)); 3881 skb->pkt_type = PACKET_LOOPBACK; 3882 if (skb->ip_summed == CHECKSUM_NONE) 3883 skb->ip_summed = CHECKSUM_UNNECESSARY; 3884 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3885 skb_dst_force(skb); 3886 netif_rx(skb); 3887 return 0; 3888 } 3889 EXPORT_SYMBOL(dev_loopback_xmit); 3890 3891 #ifdef CONFIG_NET_EGRESS 3892 static struct netdev_queue * 3893 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3894 { 3895 int qm = skb_get_queue_mapping(skb); 3896 3897 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3898 } 3899 3900 static bool netdev_xmit_txqueue_skipped(void) 3901 { 3902 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3903 } 3904 3905 void netdev_xmit_skip_txqueue(bool skip) 3906 { 3907 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3908 } 3909 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3910 #endif /* CONFIG_NET_EGRESS */ 3911 3912 #ifdef CONFIG_NET_XGRESS 3913 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb) 3914 { 3915 int ret = TC_ACT_UNSPEC; 3916 #ifdef CONFIG_NET_CLS_ACT 3917 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3918 struct tcf_result res; 3919 3920 if (!miniq) 3921 return ret; 3922 3923 tc_skb_cb(skb)->mru = 0; 3924 tc_skb_cb(skb)->post_ct = false; 3925 3926 mini_qdisc_bstats_cpu_update(miniq, skb); 3927 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3928 /* Only tcf related quirks below. */ 3929 switch (ret) { 3930 case TC_ACT_SHOT: 3931 mini_qdisc_qstats_cpu_drop(miniq); 3932 break; 3933 case TC_ACT_OK: 3934 case TC_ACT_RECLASSIFY: 3935 skb->tc_index = TC_H_MIN(res.classid); 3936 break; 3937 } 3938 #endif /* CONFIG_NET_CLS_ACT */ 3939 return ret; 3940 } 3941 3942 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3943 3944 void tcx_inc(void) 3945 { 3946 static_branch_inc(&tcx_needed_key); 3947 } 3948 3949 void tcx_dec(void) 3950 { 3951 static_branch_dec(&tcx_needed_key); 3952 } 3953 3954 static __always_inline enum tcx_action_base 3955 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3956 const bool needs_mac) 3957 { 3958 const struct bpf_mprog_fp *fp; 3959 const struct bpf_prog *prog; 3960 int ret = TCX_NEXT; 3961 3962 if (needs_mac) 3963 __skb_push(skb, skb->mac_len); 3964 bpf_mprog_foreach_prog(entry, fp, prog) { 3965 bpf_compute_data_pointers(skb); 3966 ret = bpf_prog_run(prog, skb); 3967 if (ret != TCX_NEXT) 3968 break; 3969 } 3970 if (needs_mac) 3971 __skb_pull(skb, skb->mac_len); 3972 return tcx_action_code(skb, ret); 3973 } 3974 3975 static __always_inline struct sk_buff * 3976 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3977 struct net_device *orig_dev, bool *another) 3978 { 3979 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 3980 int sch_ret; 3981 3982 if (!entry) 3983 return skb; 3984 if (*pt_prev) { 3985 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3986 *pt_prev = NULL; 3987 } 3988 3989 qdisc_skb_cb(skb)->pkt_len = skb->len; 3990 tcx_set_ingress(skb, true); 3991 3992 if (static_branch_unlikely(&tcx_needed_key)) { 3993 sch_ret = tcx_run(entry, skb, true); 3994 if (sch_ret != TC_ACT_UNSPEC) 3995 goto ingress_verdict; 3996 } 3997 sch_ret = tc_run(tcx_entry(entry), skb); 3998 ingress_verdict: 3999 switch (sch_ret) { 4000 case TC_ACT_REDIRECT: 4001 /* skb_mac_header check was done by BPF, so we can safely 4002 * push the L2 header back before redirecting to another 4003 * netdev. 4004 */ 4005 __skb_push(skb, skb->mac_len); 4006 if (skb_do_redirect(skb) == -EAGAIN) { 4007 __skb_pull(skb, skb->mac_len); 4008 *another = true; 4009 break; 4010 } 4011 *ret = NET_RX_SUCCESS; 4012 return NULL; 4013 case TC_ACT_SHOT: 4014 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 4015 *ret = NET_RX_DROP; 4016 return NULL; 4017 /* used by tc_run */ 4018 case TC_ACT_STOLEN: 4019 case TC_ACT_QUEUED: 4020 case TC_ACT_TRAP: 4021 consume_skb(skb); 4022 fallthrough; 4023 case TC_ACT_CONSUMED: 4024 *ret = NET_RX_SUCCESS; 4025 return NULL; 4026 } 4027 4028 return skb; 4029 } 4030 4031 static __always_inline struct sk_buff * 4032 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4033 { 4034 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4035 int sch_ret; 4036 4037 if (!entry) 4038 return skb; 4039 4040 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4041 * already set by the caller. 4042 */ 4043 if (static_branch_unlikely(&tcx_needed_key)) { 4044 sch_ret = tcx_run(entry, skb, false); 4045 if (sch_ret != TC_ACT_UNSPEC) 4046 goto egress_verdict; 4047 } 4048 sch_ret = tc_run(tcx_entry(entry), skb); 4049 egress_verdict: 4050 switch (sch_ret) { 4051 case TC_ACT_REDIRECT: 4052 /* No need to push/pop skb's mac_header here on egress! */ 4053 skb_do_redirect(skb); 4054 *ret = NET_XMIT_SUCCESS; 4055 return NULL; 4056 case TC_ACT_SHOT: 4057 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 4058 *ret = NET_XMIT_DROP; 4059 return NULL; 4060 /* used by tc_run */ 4061 case TC_ACT_STOLEN: 4062 case TC_ACT_QUEUED: 4063 case TC_ACT_TRAP: 4064 *ret = NET_XMIT_SUCCESS; 4065 return NULL; 4066 } 4067 4068 return skb; 4069 } 4070 #else 4071 static __always_inline struct sk_buff * 4072 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4073 struct net_device *orig_dev, bool *another) 4074 { 4075 return skb; 4076 } 4077 4078 static __always_inline struct sk_buff * 4079 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4080 { 4081 return skb; 4082 } 4083 #endif /* CONFIG_NET_XGRESS */ 4084 4085 #ifdef CONFIG_XPS 4086 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4087 struct xps_dev_maps *dev_maps, unsigned int tci) 4088 { 4089 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4090 struct xps_map *map; 4091 int queue_index = -1; 4092 4093 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4094 return queue_index; 4095 4096 tci *= dev_maps->num_tc; 4097 tci += tc; 4098 4099 map = rcu_dereference(dev_maps->attr_map[tci]); 4100 if (map) { 4101 if (map->len == 1) 4102 queue_index = map->queues[0]; 4103 else 4104 queue_index = map->queues[reciprocal_scale( 4105 skb_get_hash(skb), map->len)]; 4106 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4107 queue_index = -1; 4108 } 4109 return queue_index; 4110 } 4111 #endif 4112 4113 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4114 struct sk_buff *skb) 4115 { 4116 #ifdef CONFIG_XPS 4117 struct xps_dev_maps *dev_maps; 4118 struct sock *sk = skb->sk; 4119 int queue_index = -1; 4120 4121 if (!static_key_false(&xps_needed)) 4122 return -1; 4123 4124 rcu_read_lock(); 4125 if (!static_key_false(&xps_rxqs_needed)) 4126 goto get_cpus_map; 4127 4128 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4129 if (dev_maps) { 4130 int tci = sk_rx_queue_get(sk); 4131 4132 if (tci >= 0) 4133 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4134 tci); 4135 } 4136 4137 get_cpus_map: 4138 if (queue_index < 0) { 4139 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4140 if (dev_maps) { 4141 unsigned int tci = skb->sender_cpu - 1; 4142 4143 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4144 tci); 4145 } 4146 } 4147 rcu_read_unlock(); 4148 4149 return queue_index; 4150 #else 4151 return -1; 4152 #endif 4153 } 4154 4155 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4156 struct net_device *sb_dev) 4157 { 4158 return 0; 4159 } 4160 EXPORT_SYMBOL(dev_pick_tx_zero); 4161 4162 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4163 struct net_device *sb_dev) 4164 { 4165 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4166 } 4167 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4168 4169 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4170 struct net_device *sb_dev) 4171 { 4172 struct sock *sk = skb->sk; 4173 int queue_index = sk_tx_queue_get(sk); 4174 4175 sb_dev = sb_dev ? : dev; 4176 4177 if (queue_index < 0 || skb->ooo_okay || 4178 queue_index >= dev->real_num_tx_queues) { 4179 int new_index = get_xps_queue(dev, sb_dev, skb); 4180 4181 if (new_index < 0) 4182 new_index = skb_tx_hash(dev, sb_dev, skb); 4183 4184 if (queue_index != new_index && sk && 4185 sk_fullsock(sk) && 4186 rcu_access_pointer(sk->sk_dst_cache)) 4187 sk_tx_queue_set(sk, new_index); 4188 4189 queue_index = new_index; 4190 } 4191 4192 return queue_index; 4193 } 4194 EXPORT_SYMBOL(netdev_pick_tx); 4195 4196 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4197 struct sk_buff *skb, 4198 struct net_device *sb_dev) 4199 { 4200 int queue_index = 0; 4201 4202 #ifdef CONFIG_XPS 4203 u32 sender_cpu = skb->sender_cpu - 1; 4204 4205 if (sender_cpu >= (u32)NR_CPUS) 4206 skb->sender_cpu = raw_smp_processor_id() + 1; 4207 #endif 4208 4209 if (dev->real_num_tx_queues != 1) { 4210 const struct net_device_ops *ops = dev->netdev_ops; 4211 4212 if (ops->ndo_select_queue) 4213 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4214 else 4215 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4216 4217 queue_index = netdev_cap_txqueue(dev, queue_index); 4218 } 4219 4220 skb_set_queue_mapping(skb, queue_index); 4221 return netdev_get_tx_queue(dev, queue_index); 4222 } 4223 4224 /** 4225 * __dev_queue_xmit() - transmit a buffer 4226 * @skb: buffer to transmit 4227 * @sb_dev: suboordinate device used for L2 forwarding offload 4228 * 4229 * Queue a buffer for transmission to a network device. The caller must 4230 * have set the device and priority and built the buffer before calling 4231 * this function. The function can be called from an interrupt. 4232 * 4233 * When calling this method, interrupts MUST be enabled. This is because 4234 * the BH enable code must have IRQs enabled so that it will not deadlock. 4235 * 4236 * Regardless of the return value, the skb is consumed, so it is currently 4237 * difficult to retry a send to this method. (You can bump the ref count 4238 * before sending to hold a reference for retry if you are careful.) 4239 * 4240 * Return: 4241 * * 0 - buffer successfully transmitted 4242 * * positive qdisc return code - NET_XMIT_DROP etc. 4243 * * negative errno - other errors 4244 */ 4245 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4246 { 4247 struct net_device *dev = skb->dev; 4248 struct netdev_queue *txq = NULL; 4249 struct Qdisc *q; 4250 int rc = -ENOMEM; 4251 bool again = false; 4252 4253 skb_reset_mac_header(skb); 4254 skb_assert_len(skb); 4255 4256 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4257 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4258 4259 /* Disable soft irqs for various locks below. Also 4260 * stops preemption for RCU. 4261 */ 4262 rcu_read_lock_bh(); 4263 4264 skb_update_prio(skb); 4265 4266 qdisc_pkt_len_init(skb); 4267 tcx_set_ingress(skb, false); 4268 #ifdef CONFIG_NET_EGRESS 4269 if (static_branch_unlikely(&egress_needed_key)) { 4270 if (nf_hook_egress_active()) { 4271 skb = nf_hook_egress(skb, &rc, dev); 4272 if (!skb) 4273 goto out; 4274 } 4275 4276 netdev_xmit_skip_txqueue(false); 4277 4278 nf_skip_egress(skb, true); 4279 skb = sch_handle_egress(skb, &rc, dev); 4280 if (!skb) 4281 goto out; 4282 nf_skip_egress(skb, false); 4283 4284 if (netdev_xmit_txqueue_skipped()) 4285 txq = netdev_tx_queue_mapping(dev, skb); 4286 } 4287 #endif 4288 /* If device/qdisc don't need skb->dst, release it right now while 4289 * its hot in this cpu cache. 4290 */ 4291 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4292 skb_dst_drop(skb); 4293 else 4294 skb_dst_force(skb); 4295 4296 if (!txq) 4297 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4298 4299 q = rcu_dereference_bh(txq->qdisc); 4300 4301 trace_net_dev_queue(skb); 4302 if (q->enqueue) { 4303 rc = __dev_xmit_skb(skb, q, dev, txq); 4304 goto out; 4305 } 4306 4307 /* The device has no queue. Common case for software devices: 4308 * loopback, all the sorts of tunnels... 4309 4310 * Really, it is unlikely that netif_tx_lock protection is necessary 4311 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4312 * counters.) 4313 * However, it is possible, that they rely on protection 4314 * made by us here. 4315 4316 * Check this and shot the lock. It is not prone from deadlocks. 4317 *Either shot noqueue qdisc, it is even simpler 8) 4318 */ 4319 if (dev->flags & IFF_UP) { 4320 int cpu = smp_processor_id(); /* ok because BHs are off */ 4321 4322 /* Other cpus might concurrently change txq->xmit_lock_owner 4323 * to -1 or to their cpu id, but not to our id. 4324 */ 4325 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4326 if (dev_xmit_recursion()) 4327 goto recursion_alert; 4328 4329 skb = validate_xmit_skb(skb, dev, &again); 4330 if (!skb) 4331 goto out; 4332 4333 HARD_TX_LOCK(dev, txq, cpu); 4334 4335 if (!netif_xmit_stopped(txq)) { 4336 dev_xmit_recursion_inc(); 4337 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4338 dev_xmit_recursion_dec(); 4339 if (dev_xmit_complete(rc)) { 4340 HARD_TX_UNLOCK(dev, txq); 4341 goto out; 4342 } 4343 } 4344 HARD_TX_UNLOCK(dev, txq); 4345 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4346 dev->name); 4347 } else { 4348 /* Recursion is detected! It is possible, 4349 * unfortunately 4350 */ 4351 recursion_alert: 4352 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4353 dev->name); 4354 } 4355 } 4356 4357 rc = -ENETDOWN; 4358 rcu_read_unlock_bh(); 4359 4360 dev_core_stats_tx_dropped_inc(dev); 4361 kfree_skb_list(skb); 4362 return rc; 4363 out: 4364 rcu_read_unlock_bh(); 4365 return rc; 4366 } 4367 EXPORT_SYMBOL(__dev_queue_xmit); 4368 4369 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4370 { 4371 struct net_device *dev = skb->dev; 4372 struct sk_buff *orig_skb = skb; 4373 struct netdev_queue *txq; 4374 int ret = NETDEV_TX_BUSY; 4375 bool again = false; 4376 4377 if (unlikely(!netif_running(dev) || 4378 !netif_carrier_ok(dev))) 4379 goto drop; 4380 4381 skb = validate_xmit_skb_list(skb, dev, &again); 4382 if (skb != orig_skb) 4383 goto drop; 4384 4385 skb_set_queue_mapping(skb, queue_id); 4386 txq = skb_get_tx_queue(dev, skb); 4387 4388 local_bh_disable(); 4389 4390 dev_xmit_recursion_inc(); 4391 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4392 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4393 ret = netdev_start_xmit(skb, dev, txq, false); 4394 HARD_TX_UNLOCK(dev, txq); 4395 dev_xmit_recursion_dec(); 4396 4397 local_bh_enable(); 4398 return ret; 4399 drop: 4400 dev_core_stats_tx_dropped_inc(dev); 4401 kfree_skb_list(skb); 4402 return NET_XMIT_DROP; 4403 } 4404 EXPORT_SYMBOL(__dev_direct_xmit); 4405 4406 /************************************************************************* 4407 * Receiver routines 4408 *************************************************************************/ 4409 4410 int netdev_max_backlog __read_mostly = 1000; 4411 EXPORT_SYMBOL(netdev_max_backlog); 4412 4413 int netdev_tstamp_prequeue __read_mostly = 1; 4414 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4415 int netdev_budget __read_mostly = 300; 4416 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4417 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4418 int weight_p __read_mostly = 64; /* old backlog weight */ 4419 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4420 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4421 int dev_rx_weight __read_mostly = 64; 4422 int dev_tx_weight __read_mostly = 64; 4423 4424 /* Called with irq disabled */ 4425 static inline void ____napi_schedule(struct softnet_data *sd, 4426 struct napi_struct *napi) 4427 { 4428 struct task_struct *thread; 4429 4430 lockdep_assert_irqs_disabled(); 4431 4432 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4433 /* Paired with smp_mb__before_atomic() in 4434 * napi_enable()/dev_set_threaded(). 4435 * Use READ_ONCE() to guarantee a complete 4436 * read on napi->thread. Only call 4437 * wake_up_process() when it's not NULL. 4438 */ 4439 thread = READ_ONCE(napi->thread); 4440 if (thread) { 4441 /* Avoid doing set_bit() if the thread is in 4442 * INTERRUPTIBLE state, cause napi_thread_wait() 4443 * makes sure to proceed with napi polling 4444 * if the thread is explicitly woken from here. 4445 */ 4446 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4447 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4448 wake_up_process(thread); 4449 return; 4450 } 4451 } 4452 4453 list_add_tail(&napi->poll_list, &sd->poll_list); 4454 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4455 /* If not called from net_rx_action() 4456 * we have to raise NET_RX_SOFTIRQ. 4457 */ 4458 if (!sd->in_net_rx_action) 4459 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4460 } 4461 4462 #ifdef CONFIG_RPS 4463 4464 /* One global table that all flow-based protocols share. */ 4465 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4466 EXPORT_SYMBOL(rps_sock_flow_table); 4467 u32 rps_cpu_mask __read_mostly; 4468 EXPORT_SYMBOL(rps_cpu_mask); 4469 4470 struct static_key_false rps_needed __read_mostly; 4471 EXPORT_SYMBOL(rps_needed); 4472 struct static_key_false rfs_needed __read_mostly; 4473 EXPORT_SYMBOL(rfs_needed); 4474 4475 static struct rps_dev_flow * 4476 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4477 struct rps_dev_flow *rflow, u16 next_cpu) 4478 { 4479 if (next_cpu < nr_cpu_ids) { 4480 #ifdef CONFIG_RFS_ACCEL 4481 struct netdev_rx_queue *rxqueue; 4482 struct rps_dev_flow_table *flow_table; 4483 struct rps_dev_flow *old_rflow; 4484 u32 flow_id; 4485 u16 rxq_index; 4486 int rc; 4487 4488 /* Should we steer this flow to a different hardware queue? */ 4489 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4490 !(dev->features & NETIF_F_NTUPLE)) 4491 goto out; 4492 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4493 if (rxq_index == skb_get_rx_queue(skb)) 4494 goto out; 4495 4496 rxqueue = dev->_rx + rxq_index; 4497 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4498 if (!flow_table) 4499 goto out; 4500 flow_id = skb_get_hash(skb) & flow_table->mask; 4501 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4502 rxq_index, flow_id); 4503 if (rc < 0) 4504 goto out; 4505 old_rflow = rflow; 4506 rflow = &flow_table->flows[flow_id]; 4507 rflow->filter = rc; 4508 if (old_rflow->filter == rflow->filter) 4509 old_rflow->filter = RPS_NO_FILTER; 4510 out: 4511 #endif 4512 rflow->last_qtail = 4513 per_cpu(softnet_data, next_cpu).input_queue_head; 4514 } 4515 4516 rflow->cpu = next_cpu; 4517 return rflow; 4518 } 4519 4520 /* 4521 * get_rps_cpu is called from netif_receive_skb and returns the target 4522 * CPU from the RPS map of the receiving queue for a given skb. 4523 * rcu_read_lock must be held on entry. 4524 */ 4525 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4526 struct rps_dev_flow **rflowp) 4527 { 4528 const struct rps_sock_flow_table *sock_flow_table; 4529 struct netdev_rx_queue *rxqueue = dev->_rx; 4530 struct rps_dev_flow_table *flow_table; 4531 struct rps_map *map; 4532 int cpu = -1; 4533 u32 tcpu; 4534 u32 hash; 4535 4536 if (skb_rx_queue_recorded(skb)) { 4537 u16 index = skb_get_rx_queue(skb); 4538 4539 if (unlikely(index >= dev->real_num_rx_queues)) { 4540 WARN_ONCE(dev->real_num_rx_queues > 1, 4541 "%s received packet on queue %u, but number " 4542 "of RX queues is %u\n", 4543 dev->name, index, dev->real_num_rx_queues); 4544 goto done; 4545 } 4546 rxqueue += index; 4547 } 4548 4549 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4550 4551 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4552 map = rcu_dereference(rxqueue->rps_map); 4553 if (!flow_table && !map) 4554 goto done; 4555 4556 skb_reset_network_header(skb); 4557 hash = skb_get_hash(skb); 4558 if (!hash) 4559 goto done; 4560 4561 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4562 if (flow_table && sock_flow_table) { 4563 struct rps_dev_flow *rflow; 4564 u32 next_cpu; 4565 u32 ident; 4566 4567 /* First check into global flow table if there is a match. 4568 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4569 */ 4570 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4571 if ((ident ^ hash) & ~rps_cpu_mask) 4572 goto try_rps; 4573 4574 next_cpu = ident & rps_cpu_mask; 4575 4576 /* OK, now we know there is a match, 4577 * we can look at the local (per receive queue) flow table 4578 */ 4579 rflow = &flow_table->flows[hash & flow_table->mask]; 4580 tcpu = rflow->cpu; 4581 4582 /* 4583 * If the desired CPU (where last recvmsg was done) is 4584 * different from current CPU (one in the rx-queue flow 4585 * table entry), switch if one of the following holds: 4586 * - Current CPU is unset (>= nr_cpu_ids). 4587 * - Current CPU is offline. 4588 * - The current CPU's queue tail has advanced beyond the 4589 * last packet that was enqueued using this table entry. 4590 * This guarantees that all previous packets for the flow 4591 * have been dequeued, thus preserving in order delivery. 4592 */ 4593 if (unlikely(tcpu != next_cpu) && 4594 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4595 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4596 rflow->last_qtail)) >= 0)) { 4597 tcpu = next_cpu; 4598 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4599 } 4600 4601 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4602 *rflowp = rflow; 4603 cpu = tcpu; 4604 goto done; 4605 } 4606 } 4607 4608 try_rps: 4609 4610 if (map) { 4611 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4612 if (cpu_online(tcpu)) { 4613 cpu = tcpu; 4614 goto done; 4615 } 4616 } 4617 4618 done: 4619 return cpu; 4620 } 4621 4622 #ifdef CONFIG_RFS_ACCEL 4623 4624 /** 4625 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4626 * @dev: Device on which the filter was set 4627 * @rxq_index: RX queue index 4628 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4629 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4630 * 4631 * Drivers that implement ndo_rx_flow_steer() should periodically call 4632 * this function for each installed filter and remove the filters for 4633 * which it returns %true. 4634 */ 4635 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4636 u32 flow_id, u16 filter_id) 4637 { 4638 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4639 struct rps_dev_flow_table *flow_table; 4640 struct rps_dev_flow *rflow; 4641 bool expire = true; 4642 unsigned int cpu; 4643 4644 rcu_read_lock(); 4645 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4646 if (flow_table && flow_id <= flow_table->mask) { 4647 rflow = &flow_table->flows[flow_id]; 4648 cpu = READ_ONCE(rflow->cpu); 4649 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4650 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4651 rflow->last_qtail) < 4652 (int)(10 * flow_table->mask))) 4653 expire = false; 4654 } 4655 rcu_read_unlock(); 4656 return expire; 4657 } 4658 EXPORT_SYMBOL(rps_may_expire_flow); 4659 4660 #endif /* CONFIG_RFS_ACCEL */ 4661 4662 /* Called from hardirq (IPI) context */ 4663 static void rps_trigger_softirq(void *data) 4664 { 4665 struct softnet_data *sd = data; 4666 4667 ____napi_schedule(sd, &sd->backlog); 4668 sd->received_rps++; 4669 } 4670 4671 #endif /* CONFIG_RPS */ 4672 4673 /* Called from hardirq (IPI) context */ 4674 static void trigger_rx_softirq(void *data) 4675 { 4676 struct softnet_data *sd = data; 4677 4678 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4679 smp_store_release(&sd->defer_ipi_scheduled, 0); 4680 } 4681 4682 /* 4683 * After we queued a packet into sd->input_pkt_queue, 4684 * we need to make sure this queue is serviced soon. 4685 * 4686 * - If this is another cpu queue, link it to our rps_ipi_list, 4687 * and make sure we will process rps_ipi_list from net_rx_action(). 4688 * 4689 * - If this is our own queue, NAPI schedule our backlog. 4690 * Note that this also raises NET_RX_SOFTIRQ. 4691 */ 4692 static void napi_schedule_rps(struct softnet_data *sd) 4693 { 4694 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4695 4696 #ifdef CONFIG_RPS 4697 if (sd != mysd) { 4698 sd->rps_ipi_next = mysd->rps_ipi_list; 4699 mysd->rps_ipi_list = sd; 4700 4701 /* If not called from net_rx_action() or napi_threaded_poll() 4702 * we have to raise NET_RX_SOFTIRQ. 4703 */ 4704 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4705 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4706 return; 4707 } 4708 #endif /* CONFIG_RPS */ 4709 __napi_schedule_irqoff(&mysd->backlog); 4710 } 4711 4712 #ifdef CONFIG_NET_FLOW_LIMIT 4713 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4714 #endif 4715 4716 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4717 { 4718 #ifdef CONFIG_NET_FLOW_LIMIT 4719 struct sd_flow_limit *fl; 4720 struct softnet_data *sd; 4721 unsigned int old_flow, new_flow; 4722 4723 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4724 return false; 4725 4726 sd = this_cpu_ptr(&softnet_data); 4727 4728 rcu_read_lock(); 4729 fl = rcu_dereference(sd->flow_limit); 4730 if (fl) { 4731 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4732 old_flow = fl->history[fl->history_head]; 4733 fl->history[fl->history_head] = new_flow; 4734 4735 fl->history_head++; 4736 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4737 4738 if (likely(fl->buckets[old_flow])) 4739 fl->buckets[old_flow]--; 4740 4741 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4742 fl->count++; 4743 rcu_read_unlock(); 4744 return true; 4745 } 4746 } 4747 rcu_read_unlock(); 4748 #endif 4749 return false; 4750 } 4751 4752 /* 4753 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4754 * queue (may be a remote CPU queue). 4755 */ 4756 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4757 unsigned int *qtail) 4758 { 4759 enum skb_drop_reason reason; 4760 struct softnet_data *sd; 4761 unsigned long flags; 4762 unsigned int qlen; 4763 4764 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4765 sd = &per_cpu(softnet_data, cpu); 4766 4767 rps_lock_irqsave(sd, &flags); 4768 if (!netif_running(skb->dev)) 4769 goto drop; 4770 qlen = skb_queue_len(&sd->input_pkt_queue); 4771 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4772 if (qlen) { 4773 enqueue: 4774 __skb_queue_tail(&sd->input_pkt_queue, skb); 4775 input_queue_tail_incr_save(sd, qtail); 4776 rps_unlock_irq_restore(sd, &flags); 4777 return NET_RX_SUCCESS; 4778 } 4779 4780 /* Schedule NAPI for backlog device 4781 * We can use non atomic operation since we own the queue lock 4782 */ 4783 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4784 napi_schedule_rps(sd); 4785 goto enqueue; 4786 } 4787 reason = SKB_DROP_REASON_CPU_BACKLOG; 4788 4789 drop: 4790 sd->dropped++; 4791 rps_unlock_irq_restore(sd, &flags); 4792 4793 dev_core_stats_rx_dropped_inc(skb->dev); 4794 kfree_skb_reason(skb, reason); 4795 return NET_RX_DROP; 4796 } 4797 4798 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4799 { 4800 struct net_device *dev = skb->dev; 4801 struct netdev_rx_queue *rxqueue; 4802 4803 rxqueue = dev->_rx; 4804 4805 if (skb_rx_queue_recorded(skb)) { 4806 u16 index = skb_get_rx_queue(skb); 4807 4808 if (unlikely(index >= dev->real_num_rx_queues)) { 4809 WARN_ONCE(dev->real_num_rx_queues > 1, 4810 "%s received packet on queue %u, but number " 4811 "of RX queues is %u\n", 4812 dev->name, index, dev->real_num_rx_queues); 4813 4814 return rxqueue; /* Return first rxqueue */ 4815 } 4816 rxqueue += index; 4817 } 4818 return rxqueue; 4819 } 4820 4821 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4822 struct bpf_prog *xdp_prog) 4823 { 4824 void *orig_data, *orig_data_end, *hard_start; 4825 struct netdev_rx_queue *rxqueue; 4826 bool orig_bcast, orig_host; 4827 u32 mac_len, frame_sz; 4828 __be16 orig_eth_type; 4829 struct ethhdr *eth; 4830 u32 metalen, act; 4831 int off; 4832 4833 /* The XDP program wants to see the packet starting at the MAC 4834 * header. 4835 */ 4836 mac_len = skb->data - skb_mac_header(skb); 4837 hard_start = skb->data - skb_headroom(skb); 4838 4839 /* SKB "head" area always have tailroom for skb_shared_info */ 4840 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4841 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4842 4843 rxqueue = netif_get_rxqueue(skb); 4844 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4845 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4846 skb_headlen(skb) + mac_len, true); 4847 4848 orig_data_end = xdp->data_end; 4849 orig_data = xdp->data; 4850 eth = (struct ethhdr *)xdp->data; 4851 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4852 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4853 orig_eth_type = eth->h_proto; 4854 4855 act = bpf_prog_run_xdp(xdp_prog, xdp); 4856 4857 /* check if bpf_xdp_adjust_head was used */ 4858 off = xdp->data - orig_data; 4859 if (off) { 4860 if (off > 0) 4861 __skb_pull(skb, off); 4862 else if (off < 0) 4863 __skb_push(skb, -off); 4864 4865 skb->mac_header += off; 4866 skb_reset_network_header(skb); 4867 } 4868 4869 /* check if bpf_xdp_adjust_tail was used */ 4870 off = xdp->data_end - orig_data_end; 4871 if (off != 0) { 4872 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4873 skb->len += off; /* positive on grow, negative on shrink */ 4874 } 4875 4876 /* check if XDP changed eth hdr such SKB needs update */ 4877 eth = (struct ethhdr *)xdp->data; 4878 if ((orig_eth_type != eth->h_proto) || 4879 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4880 skb->dev->dev_addr)) || 4881 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4882 __skb_push(skb, ETH_HLEN); 4883 skb->pkt_type = PACKET_HOST; 4884 skb->protocol = eth_type_trans(skb, skb->dev); 4885 } 4886 4887 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4888 * before calling us again on redirect path. We do not call do_redirect 4889 * as we leave that up to the caller. 4890 * 4891 * Caller is responsible for managing lifetime of skb (i.e. calling 4892 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4893 */ 4894 switch (act) { 4895 case XDP_REDIRECT: 4896 case XDP_TX: 4897 __skb_push(skb, mac_len); 4898 break; 4899 case XDP_PASS: 4900 metalen = xdp->data - xdp->data_meta; 4901 if (metalen) 4902 skb_metadata_set(skb, metalen); 4903 break; 4904 } 4905 4906 return act; 4907 } 4908 4909 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4910 struct xdp_buff *xdp, 4911 struct bpf_prog *xdp_prog) 4912 { 4913 u32 act = XDP_DROP; 4914 4915 /* Reinjected packets coming from act_mirred or similar should 4916 * not get XDP generic processing. 4917 */ 4918 if (skb_is_redirected(skb)) 4919 return XDP_PASS; 4920 4921 /* XDP packets must be linear and must have sufficient headroom 4922 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4923 * native XDP provides, thus we need to do it here as well. 4924 */ 4925 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4926 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4927 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4928 int troom = skb->tail + skb->data_len - skb->end; 4929 4930 /* In case we have to go down the path and also linearize, 4931 * then lets do the pskb_expand_head() work just once here. 4932 */ 4933 if (pskb_expand_head(skb, 4934 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4935 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4936 goto do_drop; 4937 if (skb_linearize(skb)) 4938 goto do_drop; 4939 } 4940 4941 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4942 switch (act) { 4943 case XDP_REDIRECT: 4944 case XDP_TX: 4945 case XDP_PASS: 4946 break; 4947 default: 4948 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4949 fallthrough; 4950 case XDP_ABORTED: 4951 trace_xdp_exception(skb->dev, xdp_prog, act); 4952 fallthrough; 4953 case XDP_DROP: 4954 do_drop: 4955 kfree_skb(skb); 4956 break; 4957 } 4958 4959 return act; 4960 } 4961 4962 /* When doing generic XDP we have to bypass the qdisc layer and the 4963 * network taps in order to match in-driver-XDP behavior. This also means 4964 * that XDP packets are able to starve other packets going through a qdisc, 4965 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4966 * queues, so they do not have this starvation issue. 4967 */ 4968 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4969 { 4970 struct net_device *dev = skb->dev; 4971 struct netdev_queue *txq; 4972 bool free_skb = true; 4973 int cpu, rc; 4974 4975 txq = netdev_core_pick_tx(dev, skb, NULL); 4976 cpu = smp_processor_id(); 4977 HARD_TX_LOCK(dev, txq, cpu); 4978 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4979 rc = netdev_start_xmit(skb, dev, txq, 0); 4980 if (dev_xmit_complete(rc)) 4981 free_skb = false; 4982 } 4983 HARD_TX_UNLOCK(dev, txq); 4984 if (free_skb) { 4985 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4986 dev_core_stats_tx_dropped_inc(dev); 4987 kfree_skb(skb); 4988 } 4989 } 4990 4991 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4992 4993 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4994 { 4995 if (xdp_prog) { 4996 struct xdp_buff xdp; 4997 u32 act; 4998 int err; 4999 5000 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5001 if (act != XDP_PASS) { 5002 switch (act) { 5003 case XDP_REDIRECT: 5004 err = xdp_do_generic_redirect(skb->dev, skb, 5005 &xdp, xdp_prog); 5006 if (err) 5007 goto out_redir; 5008 break; 5009 case XDP_TX: 5010 generic_xdp_tx(skb, xdp_prog); 5011 break; 5012 } 5013 return XDP_DROP; 5014 } 5015 } 5016 return XDP_PASS; 5017 out_redir: 5018 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5019 return XDP_DROP; 5020 } 5021 EXPORT_SYMBOL_GPL(do_xdp_generic); 5022 5023 static int netif_rx_internal(struct sk_buff *skb) 5024 { 5025 int ret; 5026 5027 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5028 5029 trace_netif_rx(skb); 5030 5031 #ifdef CONFIG_RPS 5032 if (static_branch_unlikely(&rps_needed)) { 5033 struct rps_dev_flow voidflow, *rflow = &voidflow; 5034 int cpu; 5035 5036 rcu_read_lock(); 5037 5038 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5039 if (cpu < 0) 5040 cpu = smp_processor_id(); 5041 5042 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5043 5044 rcu_read_unlock(); 5045 } else 5046 #endif 5047 { 5048 unsigned int qtail; 5049 5050 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5051 } 5052 return ret; 5053 } 5054 5055 /** 5056 * __netif_rx - Slightly optimized version of netif_rx 5057 * @skb: buffer to post 5058 * 5059 * This behaves as netif_rx except that it does not disable bottom halves. 5060 * As a result this function may only be invoked from the interrupt context 5061 * (either hard or soft interrupt). 5062 */ 5063 int __netif_rx(struct sk_buff *skb) 5064 { 5065 int ret; 5066 5067 lockdep_assert_once(hardirq_count() | softirq_count()); 5068 5069 trace_netif_rx_entry(skb); 5070 ret = netif_rx_internal(skb); 5071 trace_netif_rx_exit(ret); 5072 return ret; 5073 } 5074 EXPORT_SYMBOL(__netif_rx); 5075 5076 /** 5077 * netif_rx - post buffer to the network code 5078 * @skb: buffer to post 5079 * 5080 * This function receives a packet from a device driver and queues it for 5081 * the upper (protocol) levels to process via the backlog NAPI device. It 5082 * always succeeds. The buffer may be dropped during processing for 5083 * congestion control or by the protocol layers. 5084 * The network buffer is passed via the backlog NAPI device. Modern NIC 5085 * driver should use NAPI and GRO. 5086 * This function can used from interrupt and from process context. The 5087 * caller from process context must not disable interrupts before invoking 5088 * this function. 5089 * 5090 * return values: 5091 * NET_RX_SUCCESS (no congestion) 5092 * NET_RX_DROP (packet was dropped) 5093 * 5094 */ 5095 int netif_rx(struct sk_buff *skb) 5096 { 5097 bool need_bh_off = !(hardirq_count() | softirq_count()); 5098 int ret; 5099 5100 if (need_bh_off) 5101 local_bh_disable(); 5102 trace_netif_rx_entry(skb); 5103 ret = netif_rx_internal(skb); 5104 trace_netif_rx_exit(ret); 5105 if (need_bh_off) 5106 local_bh_enable(); 5107 return ret; 5108 } 5109 EXPORT_SYMBOL(netif_rx); 5110 5111 static __latent_entropy void net_tx_action(struct softirq_action *h) 5112 { 5113 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5114 5115 if (sd->completion_queue) { 5116 struct sk_buff *clist; 5117 5118 local_irq_disable(); 5119 clist = sd->completion_queue; 5120 sd->completion_queue = NULL; 5121 local_irq_enable(); 5122 5123 while (clist) { 5124 struct sk_buff *skb = clist; 5125 5126 clist = clist->next; 5127 5128 WARN_ON(refcount_read(&skb->users)); 5129 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5130 trace_consume_skb(skb, net_tx_action); 5131 else 5132 trace_kfree_skb(skb, net_tx_action, 5133 get_kfree_skb_cb(skb)->reason); 5134 5135 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5136 __kfree_skb(skb); 5137 else 5138 __napi_kfree_skb(skb, 5139 get_kfree_skb_cb(skb)->reason); 5140 } 5141 } 5142 5143 if (sd->output_queue) { 5144 struct Qdisc *head; 5145 5146 local_irq_disable(); 5147 head = sd->output_queue; 5148 sd->output_queue = NULL; 5149 sd->output_queue_tailp = &sd->output_queue; 5150 local_irq_enable(); 5151 5152 rcu_read_lock(); 5153 5154 while (head) { 5155 struct Qdisc *q = head; 5156 spinlock_t *root_lock = NULL; 5157 5158 head = head->next_sched; 5159 5160 /* We need to make sure head->next_sched is read 5161 * before clearing __QDISC_STATE_SCHED 5162 */ 5163 smp_mb__before_atomic(); 5164 5165 if (!(q->flags & TCQ_F_NOLOCK)) { 5166 root_lock = qdisc_lock(q); 5167 spin_lock(root_lock); 5168 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5169 &q->state))) { 5170 /* There is a synchronize_net() between 5171 * STATE_DEACTIVATED flag being set and 5172 * qdisc_reset()/some_qdisc_is_busy() in 5173 * dev_deactivate(), so we can safely bail out 5174 * early here to avoid data race between 5175 * qdisc_deactivate() and some_qdisc_is_busy() 5176 * for lockless qdisc. 5177 */ 5178 clear_bit(__QDISC_STATE_SCHED, &q->state); 5179 continue; 5180 } 5181 5182 clear_bit(__QDISC_STATE_SCHED, &q->state); 5183 qdisc_run(q); 5184 if (root_lock) 5185 spin_unlock(root_lock); 5186 } 5187 5188 rcu_read_unlock(); 5189 } 5190 5191 xfrm_dev_backlog(sd); 5192 } 5193 5194 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5195 /* This hook is defined here for ATM LANE */ 5196 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5197 unsigned char *addr) __read_mostly; 5198 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5199 #endif 5200 5201 /** 5202 * netdev_is_rx_handler_busy - check if receive handler is registered 5203 * @dev: device to check 5204 * 5205 * Check if a receive handler is already registered for a given device. 5206 * Return true if there one. 5207 * 5208 * The caller must hold the rtnl_mutex. 5209 */ 5210 bool netdev_is_rx_handler_busy(struct net_device *dev) 5211 { 5212 ASSERT_RTNL(); 5213 return dev && rtnl_dereference(dev->rx_handler); 5214 } 5215 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5216 5217 /** 5218 * netdev_rx_handler_register - register receive handler 5219 * @dev: device to register a handler for 5220 * @rx_handler: receive handler to register 5221 * @rx_handler_data: data pointer that is used by rx handler 5222 * 5223 * Register a receive handler for a device. This handler will then be 5224 * called from __netif_receive_skb. A negative errno code is returned 5225 * on a failure. 5226 * 5227 * The caller must hold the rtnl_mutex. 5228 * 5229 * For a general description of rx_handler, see enum rx_handler_result. 5230 */ 5231 int netdev_rx_handler_register(struct net_device *dev, 5232 rx_handler_func_t *rx_handler, 5233 void *rx_handler_data) 5234 { 5235 if (netdev_is_rx_handler_busy(dev)) 5236 return -EBUSY; 5237 5238 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5239 return -EINVAL; 5240 5241 /* Note: rx_handler_data must be set before rx_handler */ 5242 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5243 rcu_assign_pointer(dev->rx_handler, rx_handler); 5244 5245 return 0; 5246 } 5247 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5248 5249 /** 5250 * netdev_rx_handler_unregister - unregister receive handler 5251 * @dev: device to unregister a handler from 5252 * 5253 * Unregister a receive handler from a device. 5254 * 5255 * The caller must hold the rtnl_mutex. 5256 */ 5257 void netdev_rx_handler_unregister(struct net_device *dev) 5258 { 5259 5260 ASSERT_RTNL(); 5261 RCU_INIT_POINTER(dev->rx_handler, NULL); 5262 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5263 * section has a guarantee to see a non NULL rx_handler_data 5264 * as well. 5265 */ 5266 synchronize_net(); 5267 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5268 } 5269 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5270 5271 /* 5272 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5273 * the special handling of PFMEMALLOC skbs. 5274 */ 5275 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5276 { 5277 switch (skb->protocol) { 5278 case htons(ETH_P_ARP): 5279 case htons(ETH_P_IP): 5280 case htons(ETH_P_IPV6): 5281 case htons(ETH_P_8021Q): 5282 case htons(ETH_P_8021AD): 5283 return true; 5284 default: 5285 return false; 5286 } 5287 } 5288 5289 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5290 int *ret, struct net_device *orig_dev) 5291 { 5292 if (nf_hook_ingress_active(skb)) { 5293 int ingress_retval; 5294 5295 if (*pt_prev) { 5296 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5297 *pt_prev = NULL; 5298 } 5299 5300 rcu_read_lock(); 5301 ingress_retval = nf_hook_ingress(skb); 5302 rcu_read_unlock(); 5303 return ingress_retval; 5304 } 5305 return 0; 5306 } 5307 5308 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5309 struct packet_type **ppt_prev) 5310 { 5311 struct packet_type *ptype, *pt_prev; 5312 rx_handler_func_t *rx_handler; 5313 struct sk_buff *skb = *pskb; 5314 struct net_device *orig_dev; 5315 bool deliver_exact = false; 5316 int ret = NET_RX_DROP; 5317 __be16 type; 5318 5319 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5320 5321 trace_netif_receive_skb(skb); 5322 5323 orig_dev = skb->dev; 5324 5325 skb_reset_network_header(skb); 5326 if (!skb_transport_header_was_set(skb)) 5327 skb_reset_transport_header(skb); 5328 skb_reset_mac_len(skb); 5329 5330 pt_prev = NULL; 5331 5332 another_round: 5333 skb->skb_iif = skb->dev->ifindex; 5334 5335 __this_cpu_inc(softnet_data.processed); 5336 5337 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5338 int ret2; 5339 5340 migrate_disable(); 5341 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5342 migrate_enable(); 5343 5344 if (ret2 != XDP_PASS) { 5345 ret = NET_RX_DROP; 5346 goto out; 5347 } 5348 } 5349 5350 if (eth_type_vlan(skb->protocol)) { 5351 skb = skb_vlan_untag(skb); 5352 if (unlikely(!skb)) 5353 goto out; 5354 } 5355 5356 if (skb_skip_tc_classify(skb)) 5357 goto skip_classify; 5358 5359 if (pfmemalloc) 5360 goto skip_taps; 5361 5362 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5363 if (pt_prev) 5364 ret = deliver_skb(skb, pt_prev, orig_dev); 5365 pt_prev = ptype; 5366 } 5367 5368 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5369 if (pt_prev) 5370 ret = deliver_skb(skb, pt_prev, orig_dev); 5371 pt_prev = ptype; 5372 } 5373 5374 skip_taps: 5375 #ifdef CONFIG_NET_INGRESS 5376 if (static_branch_unlikely(&ingress_needed_key)) { 5377 bool another = false; 5378 5379 nf_skip_egress(skb, true); 5380 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5381 &another); 5382 if (another) 5383 goto another_round; 5384 if (!skb) 5385 goto out; 5386 5387 nf_skip_egress(skb, false); 5388 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5389 goto out; 5390 } 5391 #endif 5392 skb_reset_redirect(skb); 5393 skip_classify: 5394 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5395 goto drop; 5396 5397 if (skb_vlan_tag_present(skb)) { 5398 if (pt_prev) { 5399 ret = deliver_skb(skb, pt_prev, orig_dev); 5400 pt_prev = NULL; 5401 } 5402 if (vlan_do_receive(&skb)) 5403 goto another_round; 5404 else if (unlikely(!skb)) 5405 goto out; 5406 } 5407 5408 rx_handler = rcu_dereference(skb->dev->rx_handler); 5409 if (rx_handler) { 5410 if (pt_prev) { 5411 ret = deliver_skb(skb, pt_prev, orig_dev); 5412 pt_prev = NULL; 5413 } 5414 switch (rx_handler(&skb)) { 5415 case RX_HANDLER_CONSUMED: 5416 ret = NET_RX_SUCCESS; 5417 goto out; 5418 case RX_HANDLER_ANOTHER: 5419 goto another_round; 5420 case RX_HANDLER_EXACT: 5421 deliver_exact = true; 5422 break; 5423 case RX_HANDLER_PASS: 5424 break; 5425 default: 5426 BUG(); 5427 } 5428 } 5429 5430 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5431 check_vlan_id: 5432 if (skb_vlan_tag_get_id(skb)) { 5433 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5434 * find vlan device. 5435 */ 5436 skb->pkt_type = PACKET_OTHERHOST; 5437 } else if (eth_type_vlan(skb->protocol)) { 5438 /* Outer header is 802.1P with vlan 0, inner header is 5439 * 802.1Q or 802.1AD and vlan_do_receive() above could 5440 * not find vlan dev for vlan id 0. 5441 */ 5442 __vlan_hwaccel_clear_tag(skb); 5443 skb = skb_vlan_untag(skb); 5444 if (unlikely(!skb)) 5445 goto out; 5446 if (vlan_do_receive(&skb)) 5447 /* After stripping off 802.1P header with vlan 0 5448 * vlan dev is found for inner header. 5449 */ 5450 goto another_round; 5451 else if (unlikely(!skb)) 5452 goto out; 5453 else 5454 /* We have stripped outer 802.1P vlan 0 header. 5455 * But could not find vlan dev. 5456 * check again for vlan id to set OTHERHOST. 5457 */ 5458 goto check_vlan_id; 5459 } 5460 /* Note: we might in the future use prio bits 5461 * and set skb->priority like in vlan_do_receive() 5462 * For the time being, just ignore Priority Code Point 5463 */ 5464 __vlan_hwaccel_clear_tag(skb); 5465 } 5466 5467 type = skb->protocol; 5468 5469 /* deliver only exact match when indicated */ 5470 if (likely(!deliver_exact)) { 5471 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5472 &ptype_base[ntohs(type) & 5473 PTYPE_HASH_MASK]); 5474 } 5475 5476 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5477 &orig_dev->ptype_specific); 5478 5479 if (unlikely(skb->dev != orig_dev)) { 5480 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5481 &skb->dev->ptype_specific); 5482 } 5483 5484 if (pt_prev) { 5485 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5486 goto drop; 5487 *ppt_prev = pt_prev; 5488 } else { 5489 drop: 5490 if (!deliver_exact) 5491 dev_core_stats_rx_dropped_inc(skb->dev); 5492 else 5493 dev_core_stats_rx_nohandler_inc(skb->dev); 5494 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5495 /* Jamal, now you will not able to escape explaining 5496 * me how you were going to use this. :-) 5497 */ 5498 ret = NET_RX_DROP; 5499 } 5500 5501 out: 5502 /* The invariant here is that if *ppt_prev is not NULL 5503 * then skb should also be non-NULL. 5504 * 5505 * Apparently *ppt_prev assignment above holds this invariant due to 5506 * skb dereferencing near it. 5507 */ 5508 *pskb = skb; 5509 return ret; 5510 } 5511 5512 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5513 { 5514 struct net_device *orig_dev = skb->dev; 5515 struct packet_type *pt_prev = NULL; 5516 int ret; 5517 5518 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5519 if (pt_prev) 5520 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5521 skb->dev, pt_prev, orig_dev); 5522 return ret; 5523 } 5524 5525 /** 5526 * netif_receive_skb_core - special purpose version of netif_receive_skb 5527 * @skb: buffer to process 5528 * 5529 * More direct receive version of netif_receive_skb(). It should 5530 * only be used by callers that have a need to skip RPS and Generic XDP. 5531 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5532 * 5533 * This function may only be called from softirq context and interrupts 5534 * should be enabled. 5535 * 5536 * Return values (usually ignored): 5537 * NET_RX_SUCCESS: no congestion 5538 * NET_RX_DROP: packet was dropped 5539 */ 5540 int netif_receive_skb_core(struct sk_buff *skb) 5541 { 5542 int ret; 5543 5544 rcu_read_lock(); 5545 ret = __netif_receive_skb_one_core(skb, false); 5546 rcu_read_unlock(); 5547 5548 return ret; 5549 } 5550 EXPORT_SYMBOL(netif_receive_skb_core); 5551 5552 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5553 struct packet_type *pt_prev, 5554 struct net_device *orig_dev) 5555 { 5556 struct sk_buff *skb, *next; 5557 5558 if (!pt_prev) 5559 return; 5560 if (list_empty(head)) 5561 return; 5562 if (pt_prev->list_func != NULL) 5563 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5564 ip_list_rcv, head, pt_prev, orig_dev); 5565 else 5566 list_for_each_entry_safe(skb, next, head, list) { 5567 skb_list_del_init(skb); 5568 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5569 } 5570 } 5571 5572 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5573 { 5574 /* Fast-path assumptions: 5575 * - There is no RX handler. 5576 * - Only one packet_type matches. 5577 * If either of these fails, we will end up doing some per-packet 5578 * processing in-line, then handling the 'last ptype' for the whole 5579 * sublist. This can't cause out-of-order delivery to any single ptype, 5580 * because the 'last ptype' must be constant across the sublist, and all 5581 * other ptypes are handled per-packet. 5582 */ 5583 /* Current (common) ptype of sublist */ 5584 struct packet_type *pt_curr = NULL; 5585 /* Current (common) orig_dev of sublist */ 5586 struct net_device *od_curr = NULL; 5587 struct list_head sublist; 5588 struct sk_buff *skb, *next; 5589 5590 INIT_LIST_HEAD(&sublist); 5591 list_for_each_entry_safe(skb, next, head, list) { 5592 struct net_device *orig_dev = skb->dev; 5593 struct packet_type *pt_prev = NULL; 5594 5595 skb_list_del_init(skb); 5596 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5597 if (!pt_prev) 5598 continue; 5599 if (pt_curr != pt_prev || od_curr != orig_dev) { 5600 /* dispatch old sublist */ 5601 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5602 /* start new sublist */ 5603 INIT_LIST_HEAD(&sublist); 5604 pt_curr = pt_prev; 5605 od_curr = orig_dev; 5606 } 5607 list_add_tail(&skb->list, &sublist); 5608 } 5609 5610 /* dispatch final sublist */ 5611 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5612 } 5613 5614 static int __netif_receive_skb(struct sk_buff *skb) 5615 { 5616 int ret; 5617 5618 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5619 unsigned int noreclaim_flag; 5620 5621 /* 5622 * PFMEMALLOC skbs are special, they should 5623 * - be delivered to SOCK_MEMALLOC sockets only 5624 * - stay away from userspace 5625 * - have bounded memory usage 5626 * 5627 * Use PF_MEMALLOC as this saves us from propagating the allocation 5628 * context down to all allocation sites. 5629 */ 5630 noreclaim_flag = memalloc_noreclaim_save(); 5631 ret = __netif_receive_skb_one_core(skb, true); 5632 memalloc_noreclaim_restore(noreclaim_flag); 5633 } else 5634 ret = __netif_receive_skb_one_core(skb, false); 5635 5636 return ret; 5637 } 5638 5639 static void __netif_receive_skb_list(struct list_head *head) 5640 { 5641 unsigned long noreclaim_flag = 0; 5642 struct sk_buff *skb, *next; 5643 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5644 5645 list_for_each_entry_safe(skb, next, head, list) { 5646 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5647 struct list_head sublist; 5648 5649 /* Handle the previous sublist */ 5650 list_cut_before(&sublist, head, &skb->list); 5651 if (!list_empty(&sublist)) 5652 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5653 pfmemalloc = !pfmemalloc; 5654 /* See comments in __netif_receive_skb */ 5655 if (pfmemalloc) 5656 noreclaim_flag = memalloc_noreclaim_save(); 5657 else 5658 memalloc_noreclaim_restore(noreclaim_flag); 5659 } 5660 } 5661 /* Handle the remaining sublist */ 5662 if (!list_empty(head)) 5663 __netif_receive_skb_list_core(head, pfmemalloc); 5664 /* Restore pflags */ 5665 if (pfmemalloc) 5666 memalloc_noreclaim_restore(noreclaim_flag); 5667 } 5668 5669 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5670 { 5671 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5672 struct bpf_prog *new = xdp->prog; 5673 int ret = 0; 5674 5675 switch (xdp->command) { 5676 case XDP_SETUP_PROG: 5677 rcu_assign_pointer(dev->xdp_prog, new); 5678 if (old) 5679 bpf_prog_put(old); 5680 5681 if (old && !new) { 5682 static_branch_dec(&generic_xdp_needed_key); 5683 } else if (new && !old) { 5684 static_branch_inc(&generic_xdp_needed_key); 5685 dev_disable_lro(dev); 5686 dev_disable_gro_hw(dev); 5687 } 5688 break; 5689 5690 default: 5691 ret = -EINVAL; 5692 break; 5693 } 5694 5695 return ret; 5696 } 5697 5698 static int netif_receive_skb_internal(struct sk_buff *skb) 5699 { 5700 int ret; 5701 5702 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5703 5704 if (skb_defer_rx_timestamp(skb)) 5705 return NET_RX_SUCCESS; 5706 5707 rcu_read_lock(); 5708 #ifdef CONFIG_RPS 5709 if (static_branch_unlikely(&rps_needed)) { 5710 struct rps_dev_flow voidflow, *rflow = &voidflow; 5711 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5712 5713 if (cpu >= 0) { 5714 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5715 rcu_read_unlock(); 5716 return ret; 5717 } 5718 } 5719 #endif 5720 ret = __netif_receive_skb(skb); 5721 rcu_read_unlock(); 5722 return ret; 5723 } 5724 5725 void netif_receive_skb_list_internal(struct list_head *head) 5726 { 5727 struct sk_buff *skb, *next; 5728 struct list_head sublist; 5729 5730 INIT_LIST_HEAD(&sublist); 5731 list_for_each_entry_safe(skb, next, head, list) { 5732 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5733 skb_list_del_init(skb); 5734 if (!skb_defer_rx_timestamp(skb)) 5735 list_add_tail(&skb->list, &sublist); 5736 } 5737 list_splice_init(&sublist, head); 5738 5739 rcu_read_lock(); 5740 #ifdef CONFIG_RPS 5741 if (static_branch_unlikely(&rps_needed)) { 5742 list_for_each_entry_safe(skb, next, head, list) { 5743 struct rps_dev_flow voidflow, *rflow = &voidflow; 5744 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5745 5746 if (cpu >= 0) { 5747 /* Will be handled, remove from list */ 5748 skb_list_del_init(skb); 5749 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5750 } 5751 } 5752 } 5753 #endif 5754 __netif_receive_skb_list(head); 5755 rcu_read_unlock(); 5756 } 5757 5758 /** 5759 * netif_receive_skb - process receive buffer from network 5760 * @skb: buffer to process 5761 * 5762 * netif_receive_skb() is the main receive data processing function. 5763 * It always succeeds. The buffer may be dropped during processing 5764 * for congestion control or by the protocol layers. 5765 * 5766 * This function may only be called from softirq context and interrupts 5767 * should be enabled. 5768 * 5769 * Return values (usually ignored): 5770 * NET_RX_SUCCESS: no congestion 5771 * NET_RX_DROP: packet was dropped 5772 */ 5773 int netif_receive_skb(struct sk_buff *skb) 5774 { 5775 int ret; 5776 5777 trace_netif_receive_skb_entry(skb); 5778 5779 ret = netif_receive_skb_internal(skb); 5780 trace_netif_receive_skb_exit(ret); 5781 5782 return ret; 5783 } 5784 EXPORT_SYMBOL(netif_receive_skb); 5785 5786 /** 5787 * netif_receive_skb_list - process many receive buffers from network 5788 * @head: list of skbs to process. 5789 * 5790 * Since return value of netif_receive_skb() is normally ignored, and 5791 * wouldn't be meaningful for a list, this function returns void. 5792 * 5793 * This function may only be called from softirq context and interrupts 5794 * should be enabled. 5795 */ 5796 void netif_receive_skb_list(struct list_head *head) 5797 { 5798 struct sk_buff *skb; 5799 5800 if (list_empty(head)) 5801 return; 5802 if (trace_netif_receive_skb_list_entry_enabled()) { 5803 list_for_each_entry(skb, head, list) 5804 trace_netif_receive_skb_list_entry(skb); 5805 } 5806 netif_receive_skb_list_internal(head); 5807 trace_netif_receive_skb_list_exit(0); 5808 } 5809 EXPORT_SYMBOL(netif_receive_skb_list); 5810 5811 static DEFINE_PER_CPU(struct work_struct, flush_works); 5812 5813 /* Network device is going away, flush any packets still pending */ 5814 static void flush_backlog(struct work_struct *work) 5815 { 5816 struct sk_buff *skb, *tmp; 5817 struct softnet_data *sd; 5818 5819 local_bh_disable(); 5820 sd = this_cpu_ptr(&softnet_data); 5821 5822 rps_lock_irq_disable(sd); 5823 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5824 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5825 __skb_unlink(skb, &sd->input_pkt_queue); 5826 dev_kfree_skb_irq(skb); 5827 input_queue_head_incr(sd); 5828 } 5829 } 5830 rps_unlock_irq_enable(sd); 5831 5832 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5833 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5834 __skb_unlink(skb, &sd->process_queue); 5835 kfree_skb(skb); 5836 input_queue_head_incr(sd); 5837 } 5838 } 5839 local_bh_enable(); 5840 } 5841 5842 static bool flush_required(int cpu) 5843 { 5844 #if IS_ENABLED(CONFIG_RPS) 5845 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5846 bool do_flush; 5847 5848 rps_lock_irq_disable(sd); 5849 5850 /* as insertion into process_queue happens with the rps lock held, 5851 * process_queue access may race only with dequeue 5852 */ 5853 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5854 !skb_queue_empty_lockless(&sd->process_queue); 5855 rps_unlock_irq_enable(sd); 5856 5857 return do_flush; 5858 #endif 5859 /* without RPS we can't safely check input_pkt_queue: during a 5860 * concurrent remote skb_queue_splice() we can detect as empty both 5861 * input_pkt_queue and process_queue even if the latter could end-up 5862 * containing a lot of packets. 5863 */ 5864 return true; 5865 } 5866 5867 static void flush_all_backlogs(void) 5868 { 5869 static cpumask_t flush_cpus; 5870 unsigned int cpu; 5871 5872 /* since we are under rtnl lock protection we can use static data 5873 * for the cpumask and avoid allocating on stack the possibly 5874 * large mask 5875 */ 5876 ASSERT_RTNL(); 5877 5878 cpus_read_lock(); 5879 5880 cpumask_clear(&flush_cpus); 5881 for_each_online_cpu(cpu) { 5882 if (flush_required(cpu)) { 5883 queue_work_on(cpu, system_highpri_wq, 5884 per_cpu_ptr(&flush_works, cpu)); 5885 cpumask_set_cpu(cpu, &flush_cpus); 5886 } 5887 } 5888 5889 /* we can have in flight packet[s] on the cpus we are not flushing, 5890 * synchronize_net() in unregister_netdevice_many() will take care of 5891 * them 5892 */ 5893 for_each_cpu(cpu, &flush_cpus) 5894 flush_work(per_cpu_ptr(&flush_works, cpu)); 5895 5896 cpus_read_unlock(); 5897 } 5898 5899 static void net_rps_send_ipi(struct softnet_data *remsd) 5900 { 5901 #ifdef CONFIG_RPS 5902 while (remsd) { 5903 struct softnet_data *next = remsd->rps_ipi_next; 5904 5905 if (cpu_online(remsd->cpu)) 5906 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5907 remsd = next; 5908 } 5909 #endif 5910 } 5911 5912 /* 5913 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5914 * Note: called with local irq disabled, but exits with local irq enabled. 5915 */ 5916 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5917 { 5918 #ifdef CONFIG_RPS 5919 struct softnet_data *remsd = sd->rps_ipi_list; 5920 5921 if (remsd) { 5922 sd->rps_ipi_list = NULL; 5923 5924 local_irq_enable(); 5925 5926 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5927 net_rps_send_ipi(remsd); 5928 } else 5929 #endif 5930 local_irq_enable(); 5931 } 5932 5933 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5934 { 5935 #ifdef CONFIG_RPS 5936 return sd->rps_ipi_list != NULL; 5937 #else 5938 return false; 5939 #endif 5940 } 5941 5942 static int process_backlog(struct napi_struct *napi, int quota) 5943 { 5944 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5945 bool again = true; 5946 int work = 0; 5947 5948 /* Check if we have pending ipi, its better to send them now, 5949 * not waiting net_rx_action() end. 5950 */ 5951 if (sd_has_rps_ipi_waiting(sd)) { 5952 local_irq_disable(); 5953 net_rps_action_and_irq_enable(sd); 5954 } 5955 5956 napi->weight = READ_ONCE(dev_rx_weight); 5957 while (again) { 5958 struct sk_buff *skb; 5959 5960 while ((skb = __skb_dequeue(&sd->process_queue))) { 5961 rcu_read_lock(); 5962 __netif_receive_skb(skb); 5963 rcu_read_unlock(); 5964 input_queue_head_incr(sd); 5965 if (++work >= quota) 5966 return work; 5967 5968 } 5969 5970 rps_lock_irq_disable(sd); 5971 if (skb_queue_empty(&sd->input_pkt_queue)) { 5972 /* 5973 * Inline a custom version of __napi_complete(). 5974 * only current cpu owns and manipulates this napi, 5975 * and NAPI_STATE_SCHED is the only possible flag set 5976 * on backlog. 5977 * We can use a plain write instead of clear_bit(), 5978 * and we dont need an smp_mb() memory barrier. 5979 */ 5980 napi->state = 0; 5981 again = false; 5982 } else { 5983 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5984 &sd->process_queue); 5985 } 5986 rps_unlock_irq_enable(sd); 5987 } 5988 5989 return work; 5990 } 5991 5992 /** 5993 * __napi_schedule - schedule for receive 5994 * @n: entry to schedule 5995 * 5996 * The entry's receive function will be scheduled to run. 5997 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5998 */ 5999 void __napi_schedule(struct napi_struct *n) 6000 { 6001 unsigned long flags; 6002 6003 local_irq_save(flags); 6004 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6005 local_irq_restore(flags); 6006 } 6007 EXPORT_SYMBOL(__napi_schedule); 6008 6009 /** 6010 * napi_schedule_prep - check if napi can be scheduled 6011 * @n: napi context 6012 * 6013 * Test if NAPI routine is already running, and if not mark 6014 * it as running. This is used as a condition variable to 6015 * insure only one NAPI poll instance runs. We also make 6016 * sure there is no pending NAPI disable. 6017 */ 6018 bool napi_schedule_prep(struct napi_struct *n) 6019 { 6020 unsigned long new, val = READ_ONCE(n->state); 6021 6022 do { 6023 if (unlikely(val & NAPIF_STATE_DISABLE)) 6024 return false; 6025 new = val | NAPIF_STATE_SCHED; 6026 6027 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6028 * This was suggested by Alexander Duyck, as compiler 6029 * emits better code than : 6030 * if (val & NAPIF_STATE_SCHED) 6031 * new |= NAPIF_STATE_MISSED; 6032 */ 6033 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6034 NAPIF_STATE_MISSED; 6035 } while (!try_cmpxchg(&n->state, &val, new)); 6036 6037 return !(val & NAPIF_STATE_SCHED); 6038 } 6039 EXPORT_SYMBOL(napi_schedule_prep); 6040 6041 /** 6042 * __napi_schedule_irqoff - schedule for receive 6043 * @n: entry to schedule 6044 * 6045 * Variant of __napi_schedule() assuming hard irqs are masked. 6046 * 6047 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6048 * because the interrupt disabled assumption might not be true 6049 * due to force-threaded interrupts and spinlock substitution. 6050 */ 6051 void __napi_schedule_irqoff(struct napi_struct *n) 6052 { 6053 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6054 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6055 else 6056 __napi_schedule(n); 6057 } 6058 EXPORT_SYMBOL(__napi_schedule_irqoff); 6059 6060 bool napi_complete_done(struct napi_struct *n, int work_done) 6061 { 6062 unsigned long flags, val, new, timeout = 0; 6063 bool ret = true; 6064 6065 /* 6066 * 1) Don't let napi dequeue from the cpu poll list 6067 * just in case its running on a different cpu. 6068 * 2) If we are busy polling, do nothing here, we have 6069 * the guarantee we will be called later. 6070 */ 6071 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6072 NAPIF_STATE_IN_BUSY_POLL))) 6073 return false; 6074 6075 if (work_done) { 6076 if (n->gro_bitmask) 6077 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6078 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6079 } 6080 if (n->defer_hard_irqs_count > 0) { 6081 n->defer_hard_irqs_count--; 6082 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6083 if (timeout) 6084 ret = false; 6085 } 6086 if (n->gro_bitmask) { 6087 /* When the NAPI instance uses a timeout and keeps postponing 6088 * it, we need to bound somehow the time packets are kept in 6089 * the GRO layer 6090 */ 6091 napi_gro_flush(n, !!timeout); 6092 } 6093 6094 gro_normal_list(n); 6095 6096 if (unlikely(!list_empty(&n->poll_list))) { 6097 /* If n->poll_list is not empty, we need to mask irqs */ 6098 local_irq_save(flags); 6099 list_del_init(&n->poll_list); 6100 local_irq_restore(flags); 6101 } 6102 WRITE_ONCE(n->list_owner, -1); 6103 6104 val = READ_ONCE(n->state); 6105 do { 6106 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6107 6108 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6109 NAPIF_STATE_SCHED_THREADED | 6110 NAPIF_STATE_PREFER_BUSY_POLL); 6111 6112 /* If STATE_MISSED was set, leave STATE_SCHED set, 6113 * because we will call napi->poll() one more time. 6114 * This C code was suggested by Alexander Duyck to help gcc. 6115 */ 6116 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6117 NAPIF_STATE_SCHED; 6118 } while (!try_cmpxchg(&n->state, &val, new)); 6119 6120 if (unlikely(val & NAPIF_STATE_MISSED)) { 6121 __napi_schedule(n); 6122 return false; 6123 } 6124 6125 if (timeout) 6126 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6127 HRTIMER_MODE_REL_PINNED); 6128 return ret; 6129 } 6130 EXPORT_SYMBOL(napi_complete_done); 6131 6132 /* must be called under rcu_read_lock(), as we dont take a reference */ 6133 static struct napi_struct *napi_by_id(unsigned int napi_id) 6134 { 6135 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6136 struct napi_struct *napi; 6137 6138 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6139 if (napi->napi_id == napi_id) 6140 return napi; 6141 6142 return NULL; 6143 } 6144 6145 #if defined(CONFIG_NET_RX_BUSY_POLL) 6146 6147 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6148 { 6149 if (!skip_schedule) { 6150 gro_normal_list(napi); 6151 __napi_schedule(napi); 6152 return; 6153 } 6154 6155 if (napi->gro_bitmask) { 6156 /* flush too old packets 6157 * If HZ < 1000, flush all packets. 6158 */ 6159 napi_gro_flush(napi, HZ >= 1000); 6160 } 6161 6162 gro_normal_list(napi); 6163 clear_bit(NAPI_STATE_SCHED, &napi->state); 6164 } 6165 6166 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6167 u16 budget) 6168 { 6169 bool skip_schedule = false; 6170 unsigned long timeout; 6171 int rc; 6172 6173 /* Busy polling means there is a high chance device driver hard irq 6174 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6175 * set in napi_schedule_prep(). 6176 * Since we are about to call napi->poll() once more, we can safely 6177 * clear NAPI_STATE_MISSED. 6178 * 6179 * Note: x86 could use a single "lock and ..." instruction 6180 * to perform these two clear_bit() 6181 */ 6182 clear_bit(NAPI_STATE_MISSED, &napi->state); 6183 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6184 6185 local_bh_disable(); 6186 6187 if (prefer_busy_poll) { 6188 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6189 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6190 if (napi->defer_hard_irqs_count && timeout) { 6191 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6192 skip_schedule = true; 6193 } 6194 } 6195 6196 /* All we really want here is to re-enable device interrupts. 6197 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6198 */ 6199 rc = napi->poll(napi, budget); 6200 /* We can't gro_normal_list() here, because napi->poll() might have 6201 * rearmed the napi (napi_complete_done()) in which case it could 6202 * already be running on another CPU. 6203 */ 6204 trace_napi_poll(napi, rc, budget); 6205 netpoll_poll_unlock(have_poll_lock); 6206 if (rc == budget) 6207 __busy_poll_stop(napi, skip_schedule); 6208 local_bh_enable(); 6209 } 6210 6211 void napi_busy_loop(unsigned int napi_id, 6212 bool (*loop_end)(void *, unsigned long), 6213 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6214 { 6215 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6216 int (*napi_poll)(struct napi_struct *napi, int budget); 6217 void *have_poll_lock = NULL; 6218 struct napi_struct *napi; 6219 6220 restart: 6221 napi_poll = NULL; 6222 6223 rcu_read_lock(); 6224 6225 napi = napi_by_id(napi_id); 6226 if (!napi) 6227 goto out; 6228 6229 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6230 preempt_disable(); 6231 for (;;) { 6232 int work = 0; 6233 6234 local_bh_disable(); 6235 if (!napi_poll) { 6236 unsigned long val = READ_ONCE(napi->state); 6237 6238 /* If multiple threads are competing for this napi, 6239 * we avoid dirtying napi->state as much as we can. 6240 */ 6241 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6242 NAPIF_STATE_IN_BUSY_POLL)) { 6243 if (prefer_busy_poll) 6244 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6245 goto count; 6246 } 6247 if (cmpxchg(&napi->state, val, 6248 val | NAPIF_STATE_IN_BUSY_POLL | 6249 NAPIF_STATE_SCHED) != val) { 6250 if (prefer_busy_poll) 6251 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6252 goto count; 6253 } 6254 have_poll_lock = netpoll_poll_lock(napi); 6255 napi_poll = napi->poll; 6256 } 6257 work = napi_poll(napi, budget); 6258 trace_napi_poll(napi, work, budget); 6259 gro_normal_list(napi); 6260 count: 6261 if (work > 0) 6262 __NET_ADD_STATS(dev_net(napi->dev), 6263 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6264 local_bh_enable(); 6265 6266 if (!loop_end || loop_end(loop_end_arg, start_time)) 6267 break; 6268 6269 if (unlikely(need_resched())) { 6270 if (napi_poll) 6271 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6272 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6273 preempt_enable(); 6274 rcu_read_unlock(); 6275 cond_resched(); 6276 if (loop_end(loop_end_arg, start_time)) 6277 return; 6278 goto restart; 6279 } 6280 cpu_relax(); 6281 } 6282 if (napi_poll) 6283 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6284 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6285 preempt_enable(); 6286 out: 6287 rcu_read_unlock(); 6288 } 6289 EXPORT_SYMBOL(napi_busy_loop); 6290 6291 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6292 6293 static void napi_hash_add(struct napi_struct *napi) 6294 { 6295 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6296 return; 6297 6298 spin_lock(&napi_hash_lock); 6299 6300 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6301 do { 6302 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6303 napi_gen_id = MIN_NAPI_ID; 6304 } while (napi_by_id(napi_gen_id)); 6305 napi->napi_id = napi_gen_id; 6306 6307 hlist_add_head_rcu(&napi->napi_hash_node, 6308 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6309 6310 spin_unlock(&napi_hash_lock); 6311 } 6312 6313 /* Warning : caller is responsible to make sure rcu grace period 6314 * is respected before freeing memory containing @napi 6315 */ 6316 static void napi_hash_del(struct napi_struct *napi) 6317 { 6318 spin_lock(&napi_hash_lock); 6319 6320 hlist_del_init_rcu(&napi->napi_hash_node); 6321 6322 spin_unlock(&napi_hash_lock); 6323 } 6324 6325 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6326 { 6327 struct napi_struct *napi; 6328 6329 napi = container_of(timer, struct napi_struct, timer); 6330 6331 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6332 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6333 */ 6334 if (!napi_disable_pending(napi) && 6335 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6336 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6337 __napi_schedule_irqoff(napi); 6338 } 6339 6340 return HRTIMER_NORESTART; 6341 } 6342 6343 static void init_gro_hash(struct napi_struct *napi) 6344 { 6345 int i; 6346 6347 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6348 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6349 napi->gro_hash[i].count = 0; 6350 } 6351 napi->gro_bitmask = 0; 6352 } 6353 6354 int dev_set_threaded(struct net_device *dev, bool threaded) 6355 { 6356 struct napi_struct *napi; 6357 int err = 0; 6358 6359 if (dev->threaded == threaded) 6360 return 0; 6361 6362 if (threaded) { 6363 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6364 if (!napi->thread) { 6365 err = napi_kthread_create(napi); 6366 if (err) { 6367 threaded = false; 6368 break; 6369 } 6370 } 6371 } 6372 } 6373 6374 dev->threaded = threaded; 6375 6376 /* Make sure kthread is created before THREADED bit 6377 * is set. 6378 */ 6379 smp_mb__before_atomic(); 6380 6381 /* Setting/unsetting threaded mode on a napi might not immediately 6382 * take effect, if the current napi instance is actively being 6383 * polled. In this case, the switch between threaded mode and 6384 * softirq mode will happen in the next round of napi_schedule(). 6385 * This should not cause hiccups/stalls to the live traffic. 6386 */ 6387 list_for_each_entry(napi, &dev->napi_list, dev_list) 6388 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6389 6390 return err; 6391 } 6392 EXPORT_SYMBOL(dev_set_threaded); 6393 6394 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6395 int (*poll)(struct napi_struct *, int), int weight) 6396 { 6397 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6398 return; 6399 6400 INIT_LIST_HEAD(&napi->poll_list); 6401 INIT_HLIST_NODE(&napi->napi_hash_node); 6402 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6403 napi->timer.function = napi_watchdog; 6404 init_gro_hash(napi); 6405 napi->skb = NULL; 6406 INIT_LIST_HEAD(&napi->rx_list); 6407 napi->rx_count = 0; 6408 napi->poll = poll; 6409 if (weight > NAPI_POLL_WEIGHT) 6410 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6411 weight); 6412 napi->weight = weight; 6413 napi->dev = dev; 6414 #ifdef CONFIG_NETPOLL 6415 napi->poll_owner = -1; 6416 #endif 6417 napi->list_owner = -1; 6418 set_bit(NAPI_STATE_SCHED, &napi->state); 6419 set_bit(NAPI_STATE_NPSVC, &napi->state); 6420 list_add_rcu(&napi->dev_list, &dev->napi_list); 6421 napi_hash_add(napi); 6422 napi_get_frags_check(napi); 6423 /* Create kthread for this napi if dev->threaded is set. 6424 * Clear dev->threaded if kthread creation failed so that 6425 * threaded mode will not be enabled in napi_enable(). 6426 */ 6427 if (dev->threaded && napi_kthread_create(napi)) 6428 dev->threaded = 0; 6429 } 6430 EXPORT_SYMBOL(netif_napi_add_weight); 6431 6432 void napi_disable(struct napi_struct *n) 6433 { 6434 unsigned long val, new; 6435 6436 might_sleep(); 6437 set_bit(NAPI_STATE_DISABLE, &n->state); 6438 6439 val = READ_ONCE(n->state); 6440 do { 6441 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6442 usleep_range(20, 200); 6443 val = READ_ONCE(n->state); 6444 } 6445 6446 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6447 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6448 } while (!try_cmpxchg(&n->state, &val, new)); 6449 6450 hrtimer_cancel(&n->timer); 6451 6452 clear_bit(NAPI_STATE_DISABLE, &n->state); 6453 } 6454 EXPORT_SYMBOL(napi_disable); 6455 6456 /** 6457 * napi_enable - enable NAPI scheduling 6458 * @n: NAPI context 6459 * 6460 * Resume NAPI from being scheduled on this context. 6461 * Must be paired with napi_disable. 6462 */ 6463 void napi_enable(struct napi_struct *n) 6464 { 6465 unsigned long new, val = READ_ONCE(n->state); 6466 6467 do { 6468 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6469 6470 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6471 if (n->dev->threaded && n->thread) 6472 new |= NAPIF_STATE_THREADED; 6473 } while (!try_cmpxchg(&n->state, &val, new)); 6474 } 6475 EXPORT_SYMBOL(napi_enable); 6476 6477 static void flush_gro_hash(struct napi_struct *napi) 6478 { 6479 int i; 6480 6481 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6482 struct sk_buff *skb, *n; 6483 6484 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6485 kfree_skb(skb); 6486 napi->gro_hash[i].count = 0; 6487 } 6488 } 6489 6490 /* Must be called in process context */ 6491 void __netif_napi_del(struct napi_struct *napi) 6492 { 6493 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6494 return; 6495 6496 napi_hash_del(napi); 6497 list_del_rcu(&napi->dev_list); 6498 napi_free_frags(napi); 6499 6500 flush_gro_hash(napi); 6501 napi->gro_bitmask = 0; 6502 6503 if (napi->thread) { 6504 kthread_stop(napi->thread); 6505 napi->thread = NULL; 6506 } 6507 } 6508 EXPORT_SYMBOL(__netif_napi_del); 6509 6510 static int __napi_poll(struct napi_struct *n, bool *repoll) 6511 { 6512 int work, weight; 6513 6514 weight = n->weight; 6515 6516 /* This NAPI_STATE_SCHED test is for avoiding a race 6517 * with netpoll's poll_napi(). Only the entity which 6518 * obtains the lock and sees NAPI_STATE_SCHED set will 6519 * actually make the ->poll() call. Therefore we avoid 6520 * accidentally calling ->poll() when NAPI is not scheduled. 6521 */ 6522 work = 0; 6523 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6524 work = n->poll(n, weight); 6525 trace_napi_poll(n, work, weight); 6526 } 6527 6528 if (unlikely(work > weight)) 6529 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6530 n->poll, work, weight); 6531 6532 if (likely(work < weight)) 6533 return work; 6534 6535 /* Drivers must not modify the NAPI state if they 6536 * consume the entire weight. In such cases this code 6537 * still "owns" the NAPI instance and therefore can 6538 * move the instance around on the list at-will. 6539 */ 6540 if (unlikely(napi_disable_pending(n))) { 6541 napi_complete(n); 6542 return work; 6543 } 6544 6545 /* The NAPI context has more processing work, but busy-polling 6546 * is preferred. Exit early. 6547 */ 6548 if (napi_prefer_busy_poll(n)) { 6549 if (napi_complete_done(n, work)) { 6550 /* If timeout is not set, we need to make sure 6551 * that the NAPI is re-scheduled. 6552 */ 6553 napi_schedule(n); 6554 } 6555 return work; 6556 } 6557 6558 if (n->gro_bitmask) { 6559 /* flush too old packets 6560 * If HZ < 1000, flush all packets. 6561 */ 6562 napi_gro_flush(n, HZ >= 1000); 6563 } 6564 6565 gro_normal_list(n); 6566 6567 /* Some drivers may have called napi_schedule 6568 * prior to exhausting their budget. 6569 */ 6570 if (unlikely(!list_empty(&n->poll_list))) { 6571 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6572 n->dev ? n->dev->name : "backlog"); 6573 return work; 6574 } 6575 6576 *repoll = true; 6577 6578 return work; 6579 } 6580 6581 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6582 { 6583 bool do_repoll = false; 6584 void *have; 6585 int work; 6586 6587 list_del_init(&n->poll_list); 6588 6589 have = netpoll_poll_lock(n); 6590 6591 work = __napi_poll(n, &do_repoll); 6592 6593 if (do_repoll) 6594 list_add_tail(&n->poll_list, repoll); 6595 6596 netpoll_poll_unlock(have); 6597 6598 return work; 6599 } 6600 6601 static int napi_thread_wait(struct napi_struct *napi) 6602 { 6603 bool woken = false; 6604 6605 set_current_state(TASK_INTERRUPTIBLE); 6606 6607 while (!kthread_should_stop()) { 6608 /* Testing SCHED_THREADED bit here to make sure the current 6609 * kthread owns this napi and could poll on this napi. 6610 * Testing SCHED bit is not enough because SCHED bit might be 6611 * set by some other busy poll thread or by napi_disable(). 6612 */ 6613 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6614 WARN_ON(!list_empty(&napi->poll_list)); 6615 __set_current_state(TASK_RUNNING); 6616 return 0; 6617 } 6618 6619 schedule(); 6620 /* woken being true indicates this thread owns this napi. */ 6621 woken = true; 6622 set_current_state(TASK_INTERRUPTIBLE); 6623 } 6624 __set_current_state(TASK_RUNNING); 6625 6626 return -1; 6627 } 6628 6629 static void skb_defer_free_flush(struct softnet_data *sd) 6630 { 6631 struct sk_buff *skb, *next; 6632 6633 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6634 if (!READ_ONCE(sd->defer_list)) 6635 return; 6636 6637 spin_lock(&sd->defer_lock); 6638 skb = sd->defer_list; 6639 sd->defer_list = NULL; 6640 sd->defer_count = 0; 6641 spin_unlock(&sd->defer_lock); 6642 6643 while (skb != NULL) { 6644 next = skb->next; 6645 napi_consume_skb(skb, 1); 6646 skb = next; 6647 } 6648 } 6649 6650 static int napi_threaded_poll(void *data) 6651 { 6652 struct napi_struct *napi = data; 6653 struct softnet_data *sd; 6654 void *have; 6655 6656 while (!napi_thread_wait(napi)) { 6657 for (;;) { 6658 bool repoll = false; 6659 6660 local_bh_disable(); 6661 sd = this_cpu_ptr(&softnet_data); 6662 sd->in_napi_threaded_poll = true; 6663 6664 have = netpoll_poll_lock(napi); 6665 __napi_poll(napi, &repoll); 6666 netpoll_poll_unlock(have); 6667 6668 sd->in_napi_threaded_poll = false; 6669 barrier(); 6670 6671 if (sd_has_rps_ipi_waiting(sd)) { 6672 local_irq_disable(); 6673 net_rps_action_and_irq_enable(sd); 6674 } 6675 skb_defer_free_flush(sd); 6676 local_bh_enable(); 6677 6678 if (!repoll) 6679 break; 6680 6681 cond_resched(); 6682 } 6683 } 6684 return 0; 6685 } 6686 6687 static __latent_entropy void net_rx_action(struct softirq_action *h) 6688 { 6689 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6690 unsigned long time_limit = jiffies + 6691 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6692 int budget = READ_ONCE(netdev_budget); 6693 LIST_HEAD(list); 6694 LIST_HEAD(repoll); 6695 6696 start: 6697 sd->in_net_rx_action = true; 6698 local_irq_disable(); 6699 list_splice_init(&sd->poll_list, &list); 6700 local_irq_enable(); 6701 6702 for (;;) { 6703 struct napi_struct *n; 6704 6705 skb_defer_free_flush(sd); 6706 6707 if (list_empty(&list)) { 6708 if (list_empty(&repoll)) { 6709 sd->in_net_rx_action = false; 6710 barrier(); 6711 /* We need to check if ____napi_schedule() 6712 * had refilled poll_list while 6713 * sd->in_net_rx_action was true. 6714 */ 6715 if (!list_empty(&sd->poll_list)) 6716 goto start; 6717 if (!sd_has_rps_ipi_waiting(sd)) 6718 goto end; 6719 } 6720 break; 6721 } 6722 6723 n = list_first_entry(&list, struct napi_struct, poll_list); 6724 budget -= napi_poll(n, &repoll); 6725 6726 /* If softirq window is exhausted then punt. 6727 * Allow this to run for 2 jiffies since which will allow 6728 * an average latency of 1.5/HZ. 6729 */ 6730 if (unlikely(budget <= 0 || 6731 time_after_eq(jiffies, time_limit))) { 6732 sd->time_squeeze++; 6733 break; 6734 } 6735 } 6736 6737 local_irq_disable(); 6738 6739 list_splice_tail_init(&sd->poll_list, &list); 6740 list_splice_tail(&repoll, &list); 6741 list_splice(&list, &sd->poll_list); 6742 if (!list_empty(&sd->poll_list)) 6743 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6744 else 6745 sd->in_net_rx_action = false; 6746 6747 net_rps_action_and_irq_enable(sd); 6748 end:; 6749 } 6750 6751 struct netdev_adjacent { 6752 struct net_device *dev; 6753 netdevice_tracker dev_tracker; 6754 6755 /* upper master flag, there can only be one master device per list */ 6756 bool master; 6757 6758 /* lookup ignore flag */ 6759 bool ignore; 6760 6761 /* counter for the number of times this device was added to us */ 6762 u16 ref_nr; 6763 6764 /* private field for the users */ 6765 void *private; 6766 6767 struct list_head list; 6768 struct rcu_head rcu; 6769 }; 6770 6771 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6772 struct list_head *adj_list) 6773 { 6774 struct netdev_adjacent *adj; 6775 6776 list_for_each_entry(adj, adj_list, list) { 6777 if (adj->dev == adj_dev) 6778 return adj; 6779 } 6780 return NULL; 6781 } 6782 6783 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6784 struct netdev_nested_priv *priv) 6785 { 6786 struct net_device *dev = (struct net_device *)priv->data; 6787 6788 return upper_dev == dev; 6789 } 6790 6791 /** 6792 * netdev_has_upper_dev - Check if device is linked to an upper device 6793 * @dev: device 6794 * @upper_dev: upper device to check 6795 * 6796 * Find out if a device is linked to specified upper device and return true 6797 * in case it is. Note that this checks only immediate upper device, 6798 * not through a complete stack of devices. The caller must hold the RTNL lock. 6799 */ 6800 bool netdev_has_upper_dev(struct net_device *dev, 6801 struct net_device *upper_dev) 6802 { 6803 struct netdev_nested_priv priv = { 6804 .data = (void *)upper_dev, 6805 }; 6806 6807 ASSERT_RTNL(); 6808 6809 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6810 &priv); 6811 } 6812 EXPORT_SYMBOL(netdev_has_upper_dev); 6813 6814 /** 6815 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6816 * @dev: device 6817 * @upper_dev: upper device to check 6818 * 6819 * Find out if a device is linked to specified upper device and return true 6820 * in case it is. Note that this checks the entire upper device chain. 6821 * The caller must hold rcu lock. 6822 */ 6823 6824 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6825 struct net_device *upper_dev) 6826 { 6827 struct netdev_nested_priv priv = { 6828 .data = (void *)upper_dev, 6829 }; 6830 6831 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6832 &priv); 6833 } 6834 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6835 6836 /** 6837 * netdev_has_any_upper_dev - Check if device is linked to some device 6838 * @dev: device 6839 * 6840 * Find out if a device is linked to an upper device and return true in case 6841 * it is. The caller must hold the RTNL lock. 6842 */ 6843 bool netdev_has_any_upper_dev(struct net_device *dev) 6844 { 6845 ASSERT_RTNL(); 6846 6847 return !list_empty(&dev->adj_list.upper); 6848 } 6849 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6850 6851 /** 6852 * netdev_master_upper_dev_get - Get master upper device 6853 * @dev: device 6854 * 6855 * Find a master upper device and return pointer to it or NULL in case 6856 * it's not there. The caller must hold the RTNL lock. 6857 */ 6858 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6859 { 6860 struct netdev_adjacent *upper; 6861 6862 ASSERT_RTNL(); 6863 6864 if (list_empty(&dev->adj_list.upper)) 6865 return NULL; 6866 6867 upper = list_first_entry(&dev->adj_list.upper, 6868 struct netdev_adjacent, list); 6869 if (likely(upper->master)) 6870 return upper->dev; 6871 return NULL; 6872 } 6873 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6874 6875 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6876 { 6877 struct netdev_adjacent *upper; 6878 6879 ASSERT_RTNL(); 6880 6881 if (list_empty(&dev->adj_list.upper)) 6882 return NULL; 6883 6884 upper = list_first_entry(&dev->adj_list.upper, 6885 struct netdev_adjacent, list); 6886 if (likely(upper->master) && !upper->ignore) 6887 return upper->dev; 6888 return NULL; 6889 } 6890 6891 /** 6892 * netdev_has_any_lower_dev - Check if device is linked to some device 6893 * @dev: device 6894 * 6895 * Find out if a device is linked to a lower device and return true in case 6896 * it is. The caller must hold the RTNL lock. 6897 */ 6898 static bool netdev_has_any_lower_dev(struct net_device *dev) 6899 { 6900 ASSERT_RTNL(); 6901 6902 return !list_empty(&dev->adj_list.lower); 6903 } 6904 6905 void *netdev_adjacent_get_private(struct list_head *adj_list) 6906 { 6907 struct netdev_adjacent *adj; 6908 6909 adj = list_entry(adj_list, struct netdev_adjacent, list); 6910 6911 return adj->private; 6912 } 6913 EXPORT_SYMBOL(netdev_adjacent_get_private); 6914 6915 /** 6916 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6917 * @dev: device 6918 * @iter: list_head ** of the current position 6919 * 6920 * Gets the next device from the dev's upper list, starting from iter 6921 * position. The caller must hold RCU read lock. 6922 */ 6923 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6924 struct list_head **iter) 6925 { 6926 struct netdev_adjacent *upper; 6927 6928 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6929 6930 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6931 6932 if (&upper->list == &dev->adj_list.upper) 6933 return NULL; 6934 6935 *iter = &upper->list; 6936 6937 return upper->dev; 6938 } 6939 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6940 6941 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6942 struct list_head **iter, 6943 bool *ignore) 6944 { 6945 struct netdev_adjacent *upper; 6946 6947 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6948 6949 if (&upper->list == &dev->adj_list.upper) 6950 return NULL; 6951 6952 *iter = &upper->list; 6953 *ignore = upper->ignore; 6954 6955 return upper->dev; 6956 } 6957 6958 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6959 struct list_head **iter) 6960 { 6961 struct netdev_adjacent *upper; 6962 6963 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6964 6965 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6966 6967 if (&upper->list == &dev->adj_list.upper) 6968 return NULL; 6969 6970 *iter = &upper->list; 6971 6972 return upper->dev; 6973 } 6974 6975 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6976 int (*fn)(struct net_device *dev, 6977 struct netdev_nested_priv *priv), 6978 struct netdev_nested_priv *priv) 6979 { 6980 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6981 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6982 int ret, cur = 0; 6983 bool ignore; 6984 6985 now = dev; 6986 iter = &dev->adj_list.upper; 6987 6988 while (1) { 6989 if (now != dev) { 6990 ret = fn(now, priv); 6991 if (ret) 6992 return ret; 6993 } 6994 6995 next = NULL; 6996 while (1) { 6997 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6998 if (!udev) 6999 break; 7000 if (ignore) 7001 continue; 7002 7003 next = udev; 7004 niter = &udev->adj_list.upper; 7005 dev_stack[cur] = now; 7006 iter_stack[cur++] = iter; 7007 break; 7008 } 7009 7010 if (!next) { 7011 if (!cur) 7012 return 0; 7013 next = dev_stack[--cur]; 7014 niter = iter_stack[cur]; 7015 } 7016 7017 now = next; 7018 iter = niter; 7019 } 7020 7021 return 0; 7022 } 7023 7024 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7025 int (*fn)(struct net_device *dev, 7026 struct netdev_nested_priv *priv), 7027 struct netdev_nested_priv *priv) 7028 { 7029 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7030 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7031 int ret, cur = 0; 7032 7033 now = dev; 7034 iter = &dev->adj_list.upper; 7035 7036 while (1) { 7037 if (now != dev) { 7038 ret = fn(now, priv); 7039 if (ret) 7040 return ret; 7041 } 7042 7043 next = NULL; 7044 while (1) { 7045 udev = netdev_next_upper_dev_rcu(now, &iter); 7046 if (!udev) 7047 break; 7048 7049 next = udev; 7050 niter = &udev->adj_list.upper; 7051 dev_stack[cur] = now; 7052 iter_stack[cur++] = iter; 7053 break; 7054 } 7055 7056 if (!next) { 7057 if (!cur) 7058 return 0; 7059 next = dev_stack[--cur]; 7060 niter = iter_stack[cur]; 7061 } 7062 7063 now = next; 7064 iter = niter; 7065 } 7066 7067 return 0; 7068 } 7069 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7070 7071 static bool __netdev_has_upper_dev(struct net_device *dev, 7072 struct net_device *upper_dev) 7073 { 7074 struct netdev_nested_priv priv = { 7075 .flags = 0, 7076 .data = (void *)upper_dev, 7077 }; 7078 7079 ASSERT_RTNL(); 7080 7081 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7082 &priv); 7083 } 7084 7085 /** 7086 * netdev_lower_get_next_private - Get the next ->private from the 7087 * lower neighbour list 7088 * @dev: device 7089 * @iter: list_head ** of the current position 7090 * 7091 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7092 * list, starting from iter position. The caller must hold either hold the 7093 * RTNL lock or its own locking that guarantees that the neighbour lower 7094 * list will remain unchanged. 7095 */ 7096 void *netdev_lower_get_next_private(struct net_device *dev, 7097 struct list_head **iter) 7098 { 7099 struct netdev_adjacent *lower; 7100 7101 lower = list_entry(*iter, struct netdev_adjacent, list); 7102 7103 if (&lower->list == &dev->adj_list.lower) 7104 return NULL; 7105 7106 *iter = lower->list.next; 7107 7108 return lower->private; 7109 } 7110 EXPORT_SYMBOL(netdev_lower_get_next_private); 7111 7112 /** 7113 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7114 * lower neighbour list, RCU 7115 * variant 7116 * @dev: device 7117 * @iter: list_head ** of the current position 7118 * 7119 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7120 * list, starting from iter position. The caller must hold RCU read lock. 7121 */ 7122 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7123 struct list_head **iter) 7124 { 7125 struct netdev_adjacent *lower; 7126 7127 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7128 7129 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7130 7131 if (&lower->list == &dev->adj_list.lower) 7132 return NULL; 7133 7134 *iter = &lower->list; 7135 7136 return lower->private; 7137 } 7138 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7139 7140 /** 7141 * netdev_lower_get_next - Get the next device from the lower neighbour 7142 * list 7143 * @dev: device 7144 * @iter: list_head ** of the current position 7145 * 7146 * Gets the next netdev_adjacent from the dev's lower neighbour 7147 * list, starting from iter position. The caller must hold RTNL lock or 7148 * its own locking that guarantees that the neighbour lower 7149 * list will remain unchanged. 7150 */ 7151 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7152 { 7153 struct netdev_adjacent *lower; 7154 7155 lower = list_entry(*iter, struct netdev_adjacent, list); 7156 7157 if (&lower->list == &dev->adj_list.lower) 7158 return NULL; 7159 7160 *iter = lower->list.next; 7161 7162 return lower->dev; 7163 } 7164 EXPORT_SYMBOL(netdev_lower_get_next); 7165 7166 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7167 struct list_head **iter) 7168 { 7169 struct netdev_adjacent *lower; 7170 7171 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7172 7173 if (&lower->list == &dev->adj_list.lower) 7174 return NULL; 7175 7176 *iter = &lower->list; 7177 7178 return lower->dev; 7179 } 7180 7181 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7182 struct list_head **iter, 7183 bool *ignore) 7184 { 7185 struct netdev_adjacent *lower; 7186 7187 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7188 7189 if (&lower->list == &dev->adj_list.lower) 7190 return NULL; 7191 7192 *iter = &lower->list; 7193 *ignore = lower->ignore; 7194 7195 return lower->dev; 7196 } 7197 7198 int netdev_walk_all_lower_dev(struct net_device *dev, 7199 int (*fn)(struct net_device *dev, 7200 struct netdev_nested_priv *priv), 7201 struct netdev_nested_priv *priv) 7202 { 7203 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7204 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7205 int ret, cur = 0; 7206 7207 now = dev; 7208 iter = &dev->adj_list.lower; 7209 7210 while (1) { 7211 if (now != dev) { 7212 ret = fn(now, priv); 7213 if (ret) 7214 return ret; 7215 } 7216 7217 next = NULL; 7218 while (1) { 7219 ldev = netdev_next_lower_dev(now, &iter); 7220 if (!ldev) 7221 break; 7222 7223 next = ldev; 7224 niter = &ldev->adj_list.lower; 7225 dev_stack[cur] = now; 7226 iter_stack[cur++] = iter; 7227 break; 7228 } 7229 7230 if (!next) { 7231 if (!cur) 7232 return 0; 7233 next = dev_stack[--cur]; 7234 niter = iter_stack[cur]; 7235 } 7236 7237 now = next; 7238 iter = niter; 7239 } 7240 7241 return 0; 7242 } 7243 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7244 7245 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7246 int (*fn)(struct net_device *dev, 7247 struct netdev_nested_priv *priv), 7248 struct netdev_nested_priv *priv) 7249 { 7250 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7251 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7252 int ret, cur = 0; 7253 bool ignore; 7254 7255 now = dev; 7256 iter = &dev->adj_list.lower; 7257 7258 while (1) { 7259 if (now != dev) { 7260 ret = fn(now, priv); 7261 if (ret) 7262 return ret; 7263 } 7264 7265 next = NULL; 7266 while (1) { 7267 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7268 if (!ldev) 7269 break; 7270 if (ignore) 7271 continue; 7272 7273 next = ldev; 7274 niter = &ldev->adj_list.lower; 7275 dev_stack[cur] = now; 7276 iter_stack[cur++] = iter; 7277 break; 7278 } 7279 7280 if (!next) { 7281 if (!cur) 7282 return 0; 7283 next = dev_stack[--cur]; 7284 niter = iter_stack[cur]; 7285 } 7286 7287 now = next; 7288 iter = niter; 7289 } 7290 7291 return 0; 7292 } 7293 7294 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7295 struct list_head **iter) 7296 { 7297 struct netdev_adjacent *lower; 7298 7299 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7300 if (&lower->list == &dev->adj_list.lower) 7301 return NULL; 7302 7303 *iter = &lower->list; 7304 7305 return lower->dev; 7306 } 7307 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7308 7309 static u8 __netdev_upper_depth(struct net_device *dev) 7310 { 7311 struct net_device *udev; 7312 struct list_head *iter; 7313 u8 max_depth = 0; 7314 bool ignore; 7315 7316 for (iter = &dev->adj_list.upper, 7317 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7318 udev; 7319 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7320 if (ignore) 7321 continue; 7322 if (max_depth < udev->upper_level) 7323 max_depth = udev->upper_level; 7324 } 7325 7326 return max_depth; 7327 } 7328 7329 static u8 __netdev_lower_depth(struct net_device *dev) 7330 { 7331 struct net_device *ldev; 7332 struct list_head *iter; 7333 u8 max_depth = 0; 7334 bool ignore; 7335 7336 for (iter = &dev->adj_list.lower, 7337 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7338 ldev; 7339 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7340 if (ignore) 7341 continue; 7342 if (max_depth < ldev->lower_level) 7343 max_depth = ldev->lower_level; 7344 } 7345 7346 return max_depth; 7347 } 7348 7349 static int __netdev_update_upper_level(struct net_device *dev, 7350 struct netdev_nested_priv *__unused) 7351 { 7352 dev->upper_level = __netdev_upper_depth(dev) + 1; 7353 return 0; 7354 } 7355 7356 #ifdef CONFIG_LOCKDEP 7357 static LIST_HEAD(net_unlink_list); 7358 7359 static void net_unlink_todo(struct net_device *dev) 7360 { 7361 if (list_empty(&dev->unlink_list)) 7362 list_add_tail(&dev->unlink_list, &net_unlink_list); 7363 } 7364 #endif 7365 7366 static int __netdev_update_lower_level(struct net_device *dev, 7367 struct netdev_nested_priv *priv) 7368 { 7369 dev->lower_level = __netdev_lower_depth(dev) + 1; 7370 7371 #ifdef CONFIG_LOCKDEP 7372 if (!priv) 7373 return 0; 7374 7375 if (priv->flags & NESTED_SYNC_IMM) 7376 dev->nested_level = dev->lower_level - 1; 7377 if (priv->flags & NESTED_SYNC_TODO) 7378 net_unlink_todo(dev); 7379 #endif 7380 return 0; 7381 } 7382 7383 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7384 int (*fn)(struct net_device *dev, 7385 struct netdev_nested_priv *priv), 7386 struct netdev_nested_priv *priv) 7387 { 7388 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7389 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7390 int ret, cur = 0; 7391 7392 now = dev; 7393 iter = &dev->adj_list.lower; 7394 7395 while (1) { 7396 if (now != dev) { 7397 ret = fn(now, priv); 7398 if (ret) 7399 return ret; 7400 } 7401 7402 next = NULL; 7403 while (1) { 7404 ldev = netdev_next_lower_dev_rcu(now, &iter); 7405 if (!ldev) 7406 break; 7407 7408 next = ldev; 7409 niter = &ldev->adj_list.lower; 7410 dev_stack[cur] = now; 7411 iter_stack[cur++] = iter; 7412 break; 7413 } 7414 7415 if (!next) { 7416 if (!cur) 7417 return 0; 7418 next = dev_stack[--cur]; 7419 niter = iter_stack[cur]; 7420 } 7421 7422 now = next; 7423 iter = niter; 7424 } 7425 7426 return 0; 7427 } 7428 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7429 7430 /** 7431 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7432 * lower neighbour list, RCU 7433 * variant 7434 * @dev: device 7435 * 7436 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7437 * list. The caller must hold RCU read lock. 7438 */ 7439 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7440 { 7441 struct netdev_adjacent *lower; 7442 7443 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7444 struct netdev_adjacent, list); 7445 if (lower) 7446 return lower->private; 7447 return NULL; 7448 } 7449 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7450 7451 /** 7452 * netdev_master_upper_dev_get_rcu - Get master upper device 7453 * @dev: device 7454 * 7455 * Find a master upper device and return pointer to it or NULL in case 7456 * it's not there. The caller must hold the RCU read lock. 7457 */ 7458 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7459 { 7460 struct netdev_adjacent *upper; 7461 7462 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7463 struct netdev_adjacent, list); 7464 if (upper && likely(upper->master)) 7465 return upper->dev; 7466 return NULL; 7467 } 7468 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7469 7470 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7471 struct net_device *adj_dev, 7472 struct list_head *dev_list) 7473 { 7474 char linkname[IFNAMSIZ+7]; 7475 7476 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7477 "upper_%s" : "lower_%s", adj_dev->name); 7478 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7479 linkname); 7480 } 7481 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7482 char *name, 7483 struct list_head *dev_list) 7484 { 7485 char linkname[IFNAMSIZ+7]; 7486 7487 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7488 "upper_%s" : "lower_%s", name); 7489 sysfs_remove_link(&(dev->dev.kobj), linkname); 7490 } 7491 7492 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7493 struct net_device *adj_dev, 7494 struct list_head *dev_list) 7495 { 7496 return (dev_list == &dev->adj_list.upper || 7497 dev_list == &dev->adj_list.lower) && 7498 net_eq(dev_net(dev), dev_net(adj_dev)); 7499 } 7500 7501 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7502 struct net_device *adj_dev, 7503 struct list_head *dev_list, 7504 void *private, bool master) 7505 { 7506 struct netdev_adjacent *adj; 7507 int ret; 7508 7509 adj = __netdev_find_adj(adj_dev, dev_list); 7510 7511 if (adj) { 7512 adj->ref_nr += 1; 7513 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7514 dev->name, adj_dev->name, adj->ref_nr); 7515 7516 return 0; 7517 } 7518 7519 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7520 if (!adj) 7521 return -ENOMEM; 7522 7523 adj->dev = adj_dev; 7524 adj->master = master; 7525 adj->ref_nr = 1; 7526 adj->private = private; 7527 adj->ignore = false; 7528 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7529 7530 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7531 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7532 7533 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7534 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7535 if (ret) 7536 goto free_adj; 7537 } 7538 7539 /* Ensure that master link is always the first item in list. */ 7540 if (master) { 7541 ret = sysfs_create_link(&(dev->dev.kobj), 7542 &(adj_dev->dev.kobj), "master"); 7543 if (ret) 7544 goto remove_symlinks; 7545 7546 list_add_rcu(&adj->list, dev_list); 7547 } else { 7548 list_add_tail_rcu(&adj->list, dev_list); 7549 } 7550 7551 return 0; 7552 7553 remove_symlinks: 7554 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7555 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7556 free_adj: 7557 netdev_put(adj_dev, &adj->dev_tracker); 7558 kfree(adj); 7559 7560 return ret; 7561 } 7562 7563 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7564 struct net_device *adj_dev, 7565 u16 ref_nr, 7566 struct list_head *dev_list) 7567 { 7568 struct netdev_adjacent *adj; 7569 7570 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7571 dev->name, adj_dev->name, ref_nr); 7572 7573 adj = __netdev_find_adj(adj_dev, dev_list); 7574 7575 if (!adj) { 7576 pr_err("Adjacency does not exist for device %s from %s\n", 7577 dev->name, adj_dev->name); 7578 WARN_ON(1); 7579 return; 7580 } 7581 7582 if (adj->ref_nr > ref_nr) { 7583 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7584 dev->name, adj_dev->name, ref_nr, 7585 adj->ref_nr - ref_nr); 7586 adj->ref_nr -= ref_nr; 7587 return; 7588 } 7589 7590 if (adj->master) 7591 sysfs_remove_link(&(dev->dev.kobj), "master"); 7592 7593 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7594 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7595 7596 list_del_rcu(&adj->list); 7597 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7598 adj_dev->name, dev->name, adj_dev->name); 7599 netdev_put(adj_dev, &adj->dev_tracker); 7600 kfree_rcu(adj, rcu); 7601 } 7602 7603 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7604 struct net_device *upper_dev, 7605 struct list_head *up_list, 7606 struct list_head *down_list, 7607 void *private, bool master) 7608 { 7609 int ret; 7610 7611 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7612 private, master); 7613 if (ret) 7614 return ret; 7615 7616 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7617 private, false); 7618 if (ret) { 7619 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7620 return ret; 7621 } 7622 7623 return 0; 7624 } 7625 7626 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7627 struct net_device *upper_dev, 7628 u16 ref_nr, 7629 struct list_head *up_list, 7630 struct list_head *down_list) 7631 { 7632 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7633 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7634 } 7635 7636 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7637 struct net_device *upper_dev, 7638 void *private, bool master) 7639 { 7640 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7641 &dev->adj_list.upper, 7642 &upper_dev->adj_list.lower, 7643 private, master); 7644 } 7645 7646 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7647 struct net_device *upper_dev) 7648 { 7649 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7650 &dev->adj_list.upper, 7651 &upper_dev->adj_list.lower); 7652 } 7653 7654 static int __netdev_upper_dev_link(struct net_device *dev, 7655 struct net_device *upper_dev, bool master, 7656 void *upper_priv, void *upper_info, 7657 struct netdev_nested_priv *priv, 7658 struct netlink_ext_ack *extack) 7659 { 7660 struct netdev_notifier_changeupper_info changeupper_info = { 7661 .info = { 7662 .dev = dev, 7663 .extack = extack, 7664 }, 7665 .upper_dev = upper_dev, 7666 .master = master, 7667 .linking = true, 7668 .upper_info = upper_info, 7669 }; 7670 struct net_device *master_dev; 7671 int ret = 0; 7672 7673 ASSERT_RTNL(); 7674 7675 if (dev == upper_dev) 7676 return -EBUSY; 7677 7678 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7679 if (__netdev_has_upper_dev(upper_dev, dev)) 7680 return -EBUSY; 7681 7682 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7683 return -EMLINK; 7684 7685 if (!master) { 7686 if (__netdev_has_upper_dev(dev, upper_dev)) 7687 return -EEXIST; 7688 } else { 7689 master_dev = __netdev_master_upper_dev_get(dev); 7690 if (master_dev) 7691 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7692 } 7693 7694 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7695 &changeupper_info.info); 7696 ret = notifier_to_errno(ret); 7697 if (ret) 7698 return ret; 7699 7700 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7701 master); 7702 if (ret) 7703 return ret; 7704 7705 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7706 &changeupper_info.info); 7707 ret = notifier_to_errno(ret); 7708 if (ret) 7709 goto rollback; 7710 7711 __netdev_update_upper_level(dev, NULL); 7712 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7713 7714 __netdev_update_lower_level(upper_dev, priv); 7715 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7716 priv); 7717 7718 return 0; 7719 7720 rollback: 7721 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7722 7723 return ret; 7724 } 7725 7726 /** 7727 * netdev_upper_dev_link - Add a link to the upper device 7728 * @dev: device 7729 * @upper_dev: new upper device 7730 * @extack: netlink extended ack 7731 * 7732 * Adds a link to device which is upper to this one. The caller must hold 7733 * the RTNL lock. On a failure a negative errno code is returned. 7734 * On success the reference counts are adjusted and the function 7735 * returns zero. 7736 */ 7737 int netdev_upper_dev_link(struct net_device *dev, 7738 struct net_device *upper_dev, 7739 struct netlink_ext_ack *extack) 7740 { 7741 struct netdev_nested_priv priv = { 7742 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7743 .data = NULL, 7744 }; 7745 7746 return __netdev_upper_dev_link(dev, upper_dev, false, 7747 NULL, NULL, &priv, extack); 7748 } 7749 EXPORT_SYMBOL(netdev_upper_dev_link); 7750 7751 /** 7752 * netdev_master_upper_dev_link - Add a master link to the upper device 7753 * @dev: device 7754 * @upper_dev: new upper device 7755 * @upper_priv: upper device private 7756 * @upper_info: upper info to be passed down via notifier 7757 * @extack: netlink extended ack 7758 * 7759 * Adds a link to device which is upper to this one. In this case, only 7760 * one master upper device can be linked, although other non-master devices 7761 * might be linked as well. The caller must hold the RTNL lock. 7762 * On a failure a negative errno code is returned. On success the reference 7763 * counts are adjusted and the function returns zero. 7764 */ 7765 int netdev_master_upper_dev_link(struct net_device *dev, 7766 struct net_device *upper_dev, 7767 void *upper_priv, void *upper_info, 7768 struct netlink_ext_ack *extack) 7769 { 7770 struct netdev_nested_priv priv = { 7771 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7772 .data = NULL, 7773 }; 7774 7775 return __netdev_upper_dev_link(dev, upper_dev, true, 7776 upper_priv, upper_info, &priv, extack); 7777 } 7778 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7779 7780 static void __netdev_upper_dev_unlink(struct net_device *dev, 7781 struct net_device *upper_dev, 7782 struct netdev_nested_priv *priv) 7783 { 7784 struct netdev_notifier_changeupper_info changeupper_info = { 7785 .info = { 7786 .dev = dev, 7787 }, 7788 .upper_dev = upper_dev, 7789 .linking = false, 7790 }; 7791 7792 ASSERT_RTNL(); 7793 7794 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7795 7796 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7797 &changeupper_info.info); 7798 7799 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7800 7801 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7802 &changeupper_info.info); 7803 7804 __netdev_update_upper_level(dev, NULL); 7805 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7806 7807 __netdev_update_lower_level(upper_dev, priv); 7808 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7809 priv); 7810 } 7811 7812 /** 7813 * netdev_upper_dev_unlink - Removes a link to upper device 7814 * @dev: device 7815 * @upper_dev: new upper device 7816 * 7817 * Removes a link to device which is upper to this one. The caller must hold 7818 * the RTNL lock. 7819 */ 7820 void netdev_upper_dev_unlink(struct net_device *dev, 7821 struct net_device *upper_dev) 7822 { 7823 struct netdev_nested_priv priv = { 7824 .flags = NESTED_SYNC_TODO, 7825 .data = NULL, 7826 }; 7827 7828 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7829 } 7830 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7831 7832 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7833 struct net_device *lower_dev, 7834 bool val) 7835 { 7836 struct netdev_adjacent *adj; 7837 7838 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7839 if (adj) 7840 adj->ignore = val; 7841 7842 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7843 if (adj) 7844 adj->ignore = val; 7845 } 7846 7847 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7848 struct net_device *lower_dev) 7849 { 7850 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7851 } 7852 7853 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7854 struct net_device *lower_dev) 7855 { 7856 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7857 } 7858 7859 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7860 struct net_device *new_dev, 7861 struct net_device *dev, 7862 struct netlink_ext_ack *extack) 7863 { 7864 struct netdev_nested_priv priv = { 7865 .flags = 0, 7866 .data = NULL, 7867 }; 7868 int err; 7869 7870 if (!new_dev) 7871 return 0; 7872 7873 if (old_dev && new_dev != old_dev) 7874 netdev_adjacent_dev_disable(dev, old_dev); 7875 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7876 extack); 7877 if (err) { 7878 if (old_dev && new_dev != old_dev) 7879 netdev_adjacent_dev_enable(dev, old_dev); 7880 return err; 7881 } 7882 7883 return 0; 7884 } 7885 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7886 7887 void netdev_adjacent_change_commit(struct net_device *old_dev, 7888 struct net_device *new_dev, 7889 struct net_device *dev) 7890 { 7891 struct netdev_nested_priv priv = { 7892 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7893 .data = NULL, 7894 }; 7895 7896 if (!new_dev || !old_dev) 7897 return; 7898 7899 if (new_dev == old_dev) 7900 return; 7901 7902 netdev_adjacent_dev_enable(dev, old_dev); 7903 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7904 } 7905 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7906 7907 void netdev_adjacent_change_abort(struct net_device *old_dev, 7908 struct net_device *new_dev, 7909 struct net_device *dev) 7910 { 7911 struct netdev_nested_priv priv = { 7912 .flags = 0, 7913 .data = NULL, 7914 }; 7915 7916 if (!new_dev) 7917 return; 7918 7919 if (old_dev && new_dev != old_dev) 7920 netdev_adjacent_dev_enable(dev, old_dev); 7921 7922 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7923 } 7924 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7925 7926 /** 7927 * netdev_bonding_info_change - Dispatch event about slave change 7928 * @dev: device 7929 * @bonding_info: info to dispatch 7930 * 7931 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7932 * The caller must hold the RTNL lock. 7933 */ 7934 void netdev_bonding_info_change(struct net_device *dev, 7935 struct netdev_bonding_info *bonding_info) 7936 { 7937 struct netdev_notifier_bonding_info info = { 7938 .info.dev = dev, 7939 }; 7940 7941 memcpy(&info.bonding_info, bonding_info, 7942 sizeof(struct netdev_bonding_info)); 7943 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7944 &info.info); 7945 } 7946 EXPORT_SYMBOL(netdev_bonding_info_change); 7947 7948 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7949 struct netlink_ext_ack *extack) 7950 { 7951 struct netdev_notifier_offload_xstats_info info = { 7952 .info.dev = dev, 7953 .info.extack = extack, 7954 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7955 }; 7956 int err; 7957 int rc; 7958 7959 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7960 GFP_KERNEL); 7961 if (!dev->offload_xstats_l3) 7962 return -ENOMEM; 7963 7964 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7965 NETDEV_OFFLOAD_XSTATS_DISABLE, 7966 &info.info); 7967 err = notifier_to_errno(rc); 7968 if (err) 7969 goto free_stats; 7970 7971 return 0; 7972 7973 free_stats: 7974 kfree(dev->offload_xstats_l3); 7975 dev->offload_xstats_l3 = NULL; 7976 return err; 7977 } 7978 7979 int netdev_offload_xstats_enable(struct net_device *dev, 7980 enum netdev_offload_xstats_type type, 7981 struct netlink_ext_ack *extack) 7982 { 7983 ASSERT_RTNL(); 7984 7985 if (netdev_offload_xstats_enabled(dev, type)) 7986 return -EALREADY; 7987 7988 switch (type) { 7989 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7990 return netdev_offload_xstats_enable_l3(dev, extack); 7991 } 7992 7993 WARN_ON(1); 7994 return -EINVAL; 7995 } 7996 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7997 7998 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7999 { 8000 struct netdev_notifier_offload_xstats_info info = { 8001 .info.dev = dev, 8002 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8003 }; 8004 8005 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8006 &info.info); 8007 kfree(dev->offload_xstats_l3); 8008 dev->offload_xstats_l3 = NULL; 8009 } 8010 8011 int netdev_offload_xstats_disable(struct net_device *dev, 8012 enum netdev_offload_xstats_type type) 8013 { 8014 ASSERT_RTNL(); 8015 8016 if (!netdev_offload_xstats_enabled(dev, type)) 8017 return -EALREADY; 8018 8019 switch (type) { 8020 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8021 netdev_offload_xstats_disable_l3(dev); 8022 return 0; 8023 } 8024 8025 WARN_ON(1); 8026 return -EINVAL; 8027 } 8028 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8029 8030 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8031 { 8032 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8033 } 8034 8035 static struct rtnl_hw_stats64 * 8036 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8037 enum netdev_offload_xstats_type type) 8038 { 8039 switch (type) { 8040 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8041 return dev->offload_xstats_l3; 8042 } 8043 8044 WARN_ON(1); 8045 return NULL; 8046 } 8047 8048 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8049 enum netdev_offload_xstats_type type) 8050 { 8051 ASSERT_RTNL(); 8052 8053 return netdev_offload_xstats_get_ptr(dev, type); 8054 } 8055 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8056 8057 struct netdev_notifier_offload_xstats_ru { 8058 bool used; 8059 }; 8060 8061 struct netdev_notifier_offload_xstats_rd { 8062 struct rtnl_hw_stats64 stats; 8063 bool used; 8064 }; 8065 8066 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8067 const struct rtnl_hw_stats64 *src) 8068 { 8069 dest->rx_packets += src->rx_packets; 8070 dest->tx_packets += src->tx_packets; 8071 dest->rx_bytes += src->rx_bytes; 8072 dest->tx_bytes += src->tx_bytes; 8073 dest->rx_errors += src->rx_errors; 8074 dest->tx_errors += src->tx_errors; 8075 dest->rx_dropped += src->rx_dropped; 8076 dest->tx_dropped += src->tx_dropped; 8077 dest->multicast += src->multicast; 8078 } 8079 8080 static int netdev_offload_xstats_get_used(struct net_device *dev, 8081 enum netdev_offload_xstats_type type, 8082 bool *p_used, 8083 struct netlink_ext_ack *extack) 8084 { 8085 struct netdev_notifier_offload_xstats_ru report_used = {}; 8086 struct netdev_notifier_offload_xstats_info info = { 8087 .info.dev = dev, 8088 .info.extack = extack, 8089 .type = type, 8090 .report_used = &report_used, 8091 }; 8092 int rc; 8093 8094 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8095 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8096 &info.info); 8097 *p_used = report_used.used; 8098 return notifier_to_errno(rc); 8099 } 8100 8101 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8102 enum netdev_offload_xstats_type type, 8103 struct rtnl_hw_stats64 *p_stats, 8104 bool *p_used, 8105 struct netlink_ext_ack *extack) 8106 { 8107 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8108 struct netdev_notifier_offload_xstats_info info = { 8109 .info.dev = dev, 8110 .info.extack = extack, 8111 .type = type, 8112 .report_delta = &report_delta, 8113 }; 8114 struct rtnl_hw_stats64 *stats; 8115 int rc; 8116 8117 stats = netdev_offload_xstats_get_ptr(dev, type); 8118 if (WARN_ON(!stats)) 8119 return -EINVAL; 8120 8121 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8122 &info.info); 8123 8124 /* Cache whatever we got, even if there was an error, otherwise the 8125 * successful stats retrievals would get lost. 8126 */ 8127 netdev_hw_stats64_add(stats, &report_delta.stats); 8128 8129 if (p_stats) 8130 *p_stats = *stats; 8131 *p_used = report_delta.used; 8132 8133 return notifier_to_errno(rc); 8134 } 8135 8136 int netdev_offload_xstats_get(struct net_device *dev, 8137 enum netdev_offload_xstats_type type, 8138 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8139 struct netlink_ext_ack *extack) 8140 { 8141 ASSERT_RTNL(); 8142 8143 if (p_stats) 8144 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8145 p_used, extack); 8146 else 8147 return netdev_offload_xstats_get_used(dev, type, p_used, 8148 extack); 8149 } 8150 EXPORT_SYMBOL(netdev_offload_xstats_get); 8151 8152 void 8153 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8154 const struct rtnl_hw_stats64 *stats) 8155 { 8156 report_delta->used = true; 8157 netdev_hw_stats64_add(&report_delta->stats, stats); 8158 } 8159 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8160 8161 void 8162 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8163 { 8164 report_used->used = true; 8165 } 8166 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8167 8168 void netdev_offload_xstats_push_delta(struct net_device *dev, 8169 enum netdev_offload_xstats_type type, 8170 const struct rtnl_hw_stats64 *p_stats) 8171 { 8172 struct rtnl_hw_stats64 *stats; 8173 8174 ASSERT_RTNL(); 8175 8176 stats = netdev_offload_xstats_get_ptr(dev, type); 8177 if (WARN_ON(!stats)) 8178 return; 8179 8180 netdev_hw_stats64_add(stats, p_stats); 8181 } 8182 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8183 8184 /** 8185 * netdev_get_xmit_slave - Get the xmit slave of master device 8186 * @dev: device 8187 * @skb: The packet 8188 * @all_slaves: assume all the slaves are active 8189 * 8190 * The reference counters are not incremented so the caller must be 8191 * careful with locks. The caller must hold RCU lock. 8192 * %NULL is returned if no slave is found. 8193 */ 8194 8195 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8196 struct sk_buff *skb, 8197 bool all_slaves) 8198 { 8199 const struct net_device_ops *ops = dev->netdev_ops; 8200 8201 if (!ops->ndo_get_xmit_slave) 8202 return NULL; 8203 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8204 } 8205 EXPORT_SYMBOL(netdev_get_xmit_slave); 8206 8207 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8208 struct sock *sk) 8209 { 8210 const struct net_device_ops *ops = dev->netdev_ops; 8211 8212 if (!ops->ndo_sk_get_lower_dev) 8213 return NULL; 8214 return ops->ndo_sk_get_lower_dev(dev, sk); 8215 } 8216 8217 /** 8218 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8219 * @dev: device 8220 * @sk: the socket 8221 * 8222 * %NULL is returned if no lower device is found. 8223 */ 8224 8225 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8226 struct sock *sk) 8227 { 8228 struct net_device *lower; 8229 8230 lower = netdev_sk_get_lower_dev(dev, sk); 8231 while (lower) { 8232 dev = lower; 8233 lower = netdev_sk_get_lower_dev(dev, sk); 8234 } 8235 8236 return dev; 8237 } 8238 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8239 8240 static void netdev_adjacent_add_links(struct net_device *dev) 8241 { 8242 struct netdev_adjacent *iter; 8243 8244 struct net *net = dev_net(dev); 8245 8246 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8247 if (!net_eq(net, dev_net(iter->dev))) 8248 continue; 8249 netdev_adjacent_sysfs_add(iter->dev, dev, 8250 &iter->dev->adj_list.lower); 8251 netdev_adjacent_sysfs_add(dev, iter->dev, 8252 &dev->adj_list.upper); 8253 } 8254 8255 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8256 if (!net_eq(net, dev_net(iter->dev))) 8257 continue; 8258 netdev_adjacent_sysfs_add(iter->dev, dev, 8259 &iter->dev->adj_list.upper); 8260 netdev_adjacent_sysfs_add(dev, iter->dev, 8261 &dev->adj_list.lower); 8262 } 8263 } 8264 8265 static void netdev_adjacent_del_links(struct net_device *dev) 8266 { 8267 struct netdev_adjacent *iter; 8268 8269 struct net *net = dev_net(dev); 8270 8271 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8272 if (!net_eq(net, dev_net(iter->dev))) 8273 continue; 8274 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8275 &iter->dev->adj_list.lower); 8276 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8277 &dev->adj_list.upper); 8278 } 8279 8280 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8281 if (!net_eq(net, dev_net(iter->dev))) 8282 continue; 8283 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8284 &iter->dev->adj_list.upper); 8285 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8286 &dev->adj_list.lower); 8287 } 8288 } 8289 8290 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8291 { 8292 struct netdev_adjacent *iter; 8293 8294 struct net *net = dev_net(dev); 8295 8296 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8297 if (!net_eq(net, dev_net(iter->dev))) 8298 continue; 8299 netdev_adjacent_sysfs_del(iter->dev, oldname, 8300 &iter->dev->adj_list.lower); 8301 netdev_adjacent_sysfs_add(iter->dev, dev, 8302 &iter->dev->adj_list.lower); 8303 } 8304 8305 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8306 if (!net_eq(net, dev_net(iter->dev))) 8307 continue; 8308 netdev_adjacent_sysfs_del(iter->dev, oldname, 8309 &iter->dev->adj_list.upper); 8310 netdev_adjacent_sysfs_add(iter->dev, dev, 8311 &iter->dev->adj_list.upper); 8312 } 8313 } 8314 8315 void *netdev_lower_dev_get_private(struct net_device *dev, 8316 struct net_device *lower_dev) 8317 { 8318 struct netdev_adjacent *lower; 8319 8320 if (!lower_dev) 8321 return NULL; 8322 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8323 if (!lower) 8324 return NULL; 8325 8326 return lower->private; 8327 } 8328 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8329 8330 8331 /** 8332 * netdev_lower_state_changed - Dispatch event about lower device state change 8333 * @lower_dev: device 8334 * @lower_state_info: state to dispatch 8335 * 8336 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8337 * The caller must hold the RTNL lock. 8338 */ 8339 void netdev_lower_state_changed(struct net_device *lower_dev, 8340 void *lower_state_info) 8341 { 8342 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8343 .info.dev = lower_dev, 8344 }; 8345 8346 ASSERT_RTNL(); 8347 changelowerstate_info.lower_state_info = lower_state_info; 8348 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8349 &changelowerstate_info.info); 8350 } 8351 EXPORT_SYMBOL(netdev_lower_state_changed); 8352 8353 static void dev_change_rx_flags(struct net_device *dev, int flags) 8354 { 8355 const struct net_device_ops *ops = dev->netdev_ops; 8356 8357 if (ops->ndo_change_rx_flags) 8358 ops->ndo_change_rx_flags(dev, flags); 8359 } 8360 8361 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8362 { 8363 unsigned int old_flags = dev->flags; 8364 kuid_t uid; 8365 kgid_t gid; 8366 8367 ASSERT_RTNL(); 8368 8369 dev->flags |= IFF_PROMISC; 8370 dev->promiscuity += inc; 8371 if (dev->promiscuity == 0) { 8372 /* 8373 * Avoid overflow. 8374 * If inc causes overflow, untouch promisc and return error. 8375 */ 8376 if (inc < 0) 8377 dev->flags &= ~IFF_PROMISC; 8378 else { 8379 dev->promiscuity -= inc; 8380 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8381 return -EOVERFLOW; 8382 } 8383 } 8384 if (dev->flags != old_flags) { 8385 netdev_info(dev, "%s promiscuous mode\n", 8386 dev->flags & IFF_PROMISC ? "entered" : "left"); 8387 if (audit_enabled) { 8388 current_uid_gid(&uid, &gid); 8389 audit_log(audit_context(), GFP_ATOMIC, 8390 AUDIT_ANOM_PROMISCUOUS, 8391 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8392 dev->name, (dev->flags & IFF_PROMISC), 8393 (old_flags & IFF_PROMISC), 8394 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8395 from_kuid(&init_user_ns, uid), 8396 from_kgid(&init_user_ns, gid), 8397 audit_get_sessionid(current)); 8398 } 8399 8400 dev_change_rx_flags(dev, IFF_PROMISC); 8401 } 8402 if (notify) 8403 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8404 return 0; 8405 } 8406 8407 /** 8408 * dev_set_promiscuity - update promiscuity count on a device 8409 * @dev: device 8410 * @inc: modifier 8411 * 8412 * Add or remove promiscuity from a device. While the count in the device 8413 * remains above zero the interface remains promiscuous. Once it hits zero 8414 * the device reverts back to normal filtering operation. A negative inc 8415 * value is used to drop promiscuity on the device. 8416 * Return 0 if successful or a negative errno code on error. 8417 */ 8418 int dev_set_promiscuity(struct net_device *dev, int inc) 8419 { 8420 unsigned int old_flags = dev->flags; 8421 int err; 8422 8423 err = __dev_set_promiscuity(dev, inc, true); 8424 if (err < 0) 8425 return err; 8426 if (dev->flags != old_flags) 8427 dev_set_rx_mode(dev); 8428 return err; 8429 } 8430 EXPORT_SYMBOL(dev_set_promiscuity); 8431 8432 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8433 { 8434 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8435 8436 ASSERT_RTNL(); 8437 8438 dev->flags |= IFF_ALLMULTI; 8439 dev->allmulti += inc; 8440 if (dev->allmulti == 0) { 8441 /* 8442 * Avoid overflow. 8443 * If inc causes overflow, untouch allmulti and return error. 8444 */ 8445 if (inc < 0) 8446 dev->flags &= ~IFF_ALLMULTI; 8447 else { 8448 dev->allmulti -= inc; 8449 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8450 return -EOVERFLOW; 8451 } 8452 } 8453 if (dev->flags ^ old_flags) { 8454 netdev_info(dev, "%s allmulticast mode\n", 8455 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8456 dev_change_rx_flags(dev, IFF_ALLMULTI); 8457 dev_set_rx_mode(dev); 8458 if (notify) 8459 __dev_notify_flags(dev, old_flags, 8460 dev->gflags ^ old_gflags, 0, NULL); 8461 } 8462 return 0; 8463 } 8464 8465 /** 8466 * dev_set_allmulti - update allmulti count on a device 8467 * @dev: device 8468 * @inc: modifier 8469 * 8470 * Add or remove reception of all multicast frames to a device. While the 8471 * count in the device remains above zero the interface remains listening 8472 * to all interfaces. Once it hits zero the device reverts back to normal 8473 * filtering operation. A negative @inc value is used to drop the counter 8474 * when releasing a resource needing all multicasts. 8475 * Return 0 if successful or a negative errno code on error. 8476 */ 8477 8478 int dev_set_allmulti(struct net_device *dev, int inc) 8479 { 8480 return __dev_set_allmulti(dev, inc, true); 8481 } 8482 EXPORT_SYMBOL(dev_set_allmulti); 8483 8484 /* 8485 * Upload unicast and multicast address lists to device and 8486 * configure RX filtering. When the device doesn't support unicast 8487 * filtering it is put in promiscuous mode while unicast addresses 8488 * are present. 8489 */ 8490 void __dev_set_rx_mode(struct net_device *dev) 8491 { 8492 const struct net_device_ops *ops = dev->netdev_ops; 8493 8494 /* dev_open will call this function so the list will stay sane. */ 8495 if (!(dev->flags&IFF_UP)) 8496 return; 8497 8498 if (!netif_device_present(dev)) 8499 return; 8500 8501 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8502 /* Unicast addresses changes may only happen under the rtnl, 8503 * therefore calling __dev_set_promiscuity here is safe. 8504 */ 8505 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8506 __dev_set_promiscuity(dev, 1, false); 8507 dev->uc_promisc = true; 8508 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8509 __dev_set_promiscuity(dev, -1, false); 8510 dev->uc_promisc = false; 8511 } 8512 } 8513 8514 if (ops->ndo_set_rx_mode) 8515 ops->ndo_set_rx_mode(dev); 8516 } 8517 8518 void dev_set_rx_mode(struct net_device *dev) 8519 { 8520 netif_addr_lock_bh(dev); 8521 __dev_set_rx_mode(dev); 8522 netif_addr_unlock_bh(dev); 8523 } 8524 8525 /** 8526 * dev_get_flags - get flags reported to userspace 8527 * @dev: device 8528 * 8529 * Get the combination of flag bits exported through APIs to userspace. 8530 */ 8531 unsigned int dev_get_flags(const struct net_device *dev) 8532 { 8533 unsigned int flags; 8534 8535 flags = (dev->flags & ~(IFF_PROMISC | 8536 IFF_ALLMULTI | 8537 IFF_RUNNING | 8538 IFF_LOWER_UP | 8539 IFF_DORMANT)) | 8540 (dev->gflags & (IFF_PROMISC | 8541 IFF_ALLMULTI)); 8542 8543 if (netif_running(dev)) { 8544 if (netif_oper_up(dev)) 8545 flags |= IFF_RUNNING; 8546 if (netif_carrier_ok(dev)) 8547 flags |= IFF_LOWER_UP; 8548 if (netif_dormant(dev)) 8549 flags |= IFF_DORMANT; 8550 } 8551 8552 return flags; 8553 } 8554 EXPORT_SYMBOL(dev_get_flags); 8555 8556 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8557 struct netlink_ext_ack *extack) 8558 { 8559 unsigned int old_flags = dev->flags; 8560 int ret; 8561 8562 ASSERT_RTNL(); 8563 8564 /* 8565 * Set the flags on our device. 8566 */ 8567 8568 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8569 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8570 IFF_AUTOMEDIA)) | 8571 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8572 IFF_ALLMULTI)); 8573 8574 /* 8575 * Load in the correct multicast list now the flags have changed. 8576 */ 8577 8578 if ((old_flags ^ flags) & IFF_MULTICAST) 8579 dev_change_rx_flags(dev, IFF_MULTICAST); 8580 8581 dev_set_rx_mode(dev); 8582 8583 /* 8584 * Have we downed the interface. We handle IFF_UP ourselves 8585 * according to user attempts to set it, rather than blindly 8586 * setting it. 8587 */ 8588 8589 ret = 0; 8590 if ((old_flags ^ flags) & IFF_UP) { 8591 if (old_flags & IFF_UP) 8592 __dev_close(dev); 8593 else 8594 ret = __dev_open(dev, extack); 8595 } 8596 8597 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8598 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8599 unsigned int old_flags = dev->flags; 8600 8601 dev->gflags ^= IFF_PROMISC; 8602 8603 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8604 if (dev->flags != old_flags) 8605 dev_set_rx_mode(dev); 8606 } 8607 8608 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8609 * is important. Some (broken) drivers set IFF_PROMISC, when 8610 * IFF_ALLMULTI is requested not asking us and not reporting. 8611 */ 8612 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8613 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8614 8615 dev->gflags ^= IFF_ALLMULTI; 8616 __dev_set_allmulti(dev, inc, false); 8617 } 8618 8619 return ret; 8620 } 8621 8622 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8623 unsigned int gchanges, u32 portid, 8624 const struct nlmsghdr *nlh) 8625 { 8626 unsigned int changes = dev->flags ^ old_flags; 8627 8628 if (gchanges) 8629 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8630 8631 if (changes & IFF_UP) { 8632 if (dev->flags & IFF_UP) 8633 call_netdevice_notifiers(NETDEV_UP, dev); 8634 else 8635 call_netdevice_notifiers(NETDEV_DOWN, dev); 8636 } 8637 8638 if (dev->flags & IFF_UP && 8639 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8640 struct netdev_notifier_change_info change_info = { 8641 .info = { 8642 .dev = dev, 8643 }, 8644 .flags_changed = changes, 8645 }; 8646 8647 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8648 } 8649 } 8650 8651 /** 8652 * dev_change_flags - change device settings 8653 * @dev: device 8654 * @flags: device state flags 8655 * @extack: netlink extended ack 8656 * 8657 * Change settings on device based state flags. The flags are 8658 * in the userspace exported format. 8659 */ 8660 int dev_change_flags(struct net_device *dev, unsigned int flags, 8661 struct netlink_ext_ack *extack) 8662 { 8663 int ret; 8664 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8665 8666 ret = __dev_change_flags(dev, flags, extack); 8667 if (ret < 0) 8668 return ret; 8669 8670 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8671 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8672 return ret; 8673 } 8674 EXPORT_SYMBOL(dev_change_flags); 8675 8676 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8677 { 8678 const struct net_device_ops *ops = dev->netdev_ops; 8679 8680 if (ops->ndo_change_mtu) 8681 return ops->ndo_change_mtu(dev, new_mtu); 8682 8683 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8684 WRITE_ONCE(dev->mtu, new_mtu); 8685 return 0; 8686 } 8687 EXPORT_SYMBOL(__dev_set_mtu); 8688 8689 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8690 struct netlink_ext_ack *extack) 8691 { 8692 /* MTU must be positive, and in range */ 8693 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8694 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8695 return -EINVAL; 8696 } 8697 8698 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8699 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8700 return -EINVAL; 8701 } 8702 return 0; 8703 } 8704 8705 /** 8706 * dev_set_mtu_ext - Change maximum transfer unit 8707 * @dev: device 8708 * @new_mtu: new transfer unit 8709 * @extack: netlink extended ack 8710 * 8711 * Change the maximum transfer size of the network device. 8712 */ 8713 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8714 struct netlink_ext_ack *extack) 8715 { 8716 int err, orig_mtu; 8717 8718 if (new_mtu == dev->mtu) 8719 return 0; 8720 8721 err = dev_validate_mtu(dev, new_mtu, extack); 8722 if (err) 8723 return err; 8724 8725 if (!netif_device_present(dev)) 8726 return -ENODEV; 8727 8728 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8729 err = notifier_to_errno(err); 8730 if (err) 8731 return err; 8732 8733 orig_mtu = dev->mtu; 8734 err = __dev_set_mtu(dev, new_mtu); 8735 8736 if (!err) { 8737 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8738 orig_mtu); 8739 err = notifier_to_errno(err); 8740 if (err) { 8741 /* setting mtu back and notifying everyone again, 8742 * so that they have a chance to revert changes. 8743 */ 8744 __dev_set_mtu(dev, orig_mtu); 8745 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8746 new_mtu); 8747 } 8748 } 8749 return err; 8750 } 8751 8752 int dev_set_mtu(struct net_device *dev, int new_mtu) 8753 { 8754 struct netlink_ext_ack extack; 8755 int err; 8756 8757 memset(&extack, 0, sizeof(extack)); 8758 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8759 if (err && extack._msg) 8760 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8761 return err; 8762 } 8763 EXPORT_SYMBOL(dev_set_mtu); 8764 8765 /** 8766 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8767 * @dev: device 8768 * @new_len: new tx queue length 8769 */ 8770 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8771 { 8772 unsigned int orig_len = dev->tx_queue_len; 8773 int res; 8774 8775 if (new_len != (unsigned int)new_len) 8776 return -ERANGE; 8777 8778 if (new_len != orig_len) { 8779 dev->tx_queue_len = new_len; 8780 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8781 res = notifier_to_errno(res); 8782 if (res) 8783 goto err_rollback; 8784 res = dev_qdisc_change_tx_queue_len(dev); 8785 if (res) 8786 goto err_rollback; 8787 } 8788 8789 return 0; 8790 8791 err_rollback: 8792 netdev_err(dev, "refused to change device tx_queue_len\n"); 8793 dev->tx_queue_len = orig_len; 8794 return res; 8795 } 8796 8797 /** 8798 * dev_set_group - Change group this device belongs to 8799 * @dev: device 8800 * @new_group: group this device should belong to 8801 */ 8802 void dev_set_group(struct net_device *dev, int new_group) 8803 { 8804 dev->group = new_group; 8805 } 8806 8807 /** 8808 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8809 * @dev: device 8810 * @addr: new address 8811 * @extack: netlink extended ack 8812 */ 8813 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8814 struct netlink_ext_ack *extack) 8815 { 8816 struct netdev_notifier_pre_changeaddr_info info = { 8817 .info.dev = dev, 8818 .info.extack = extack, 8819 .dev_addr = addr, 8820 }; 8821 int rc; 8822 8823 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8824 return notifier_to_errno(rc); 8825 } 8826 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8827 8828 /** 8829 * dev_set_mac_address - Change Media Access Control Address 8830 * @dev: device 8831 * @sa: new address 8832 * @extack: netlink extended ack 8833 * 8834 * Change the hardware (MAC) address of the device 8835 */ 8836 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8837 struct netlink_ext_ack *extack) 8838 { 8839 const struct net_device_ops *ops = dev->netdev_ops; 8840 int err; 8841 8842 if (!ops->ndo_set_mac_address) 8843 return -EOPNOTSUPP; 8844 if (sa->sa_family != dev->type) 8845 return -EINVAL; 8846 if (!netif_device_present(dev)) 8847 return -ENODEV; 8848 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8849 if (err) 8850 return err; 8851 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8852 err = ops->ndo_set_mac_address(dev, sa); 8853 if (err) 8854 return err; 8855 } 8856 dev->addr_assign_type = NET_ADDR_SET; 8857 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8858 add_device_randomness(dev->dev_addr, dev->addr_len); 8859 return 0; 8860 } 8861 EXPORT_SYMBOL(dev_set_mac_address); 8862 8863 static DECLARE_RWSEM(dev_addr_sem); 8864 8865 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8866 struct netlink_ext_ack *extack) 8867 { 8868 int ret; 8869 8870 down_write(&dev_addr_sem); 8871 ret = dev_set_mac_address(dev, sa, extack); 8872 up_write(&dev_addr_sem); 8873 return ret; 8874 } 8875 EXPORT_SYMBOL(dev_set_mac_address_user); 8876 8877 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8878 { 8879 size_t size = sizeof(sa->sa_data_min); 8880 struct net_device *dev; 8881 int ret = 0; 8882 8883 down_read(&dev_addr_sem); 8884 rcu_read_lock(); 8885 8886 dev = dev_get_by_name_rcu(net, dev_name); 8887 if (!dev) { 8888 ret = -ENODEV; 8889 goto unlock; 8890 } 8891 if (!dev->addr_len) 8892 memset(sa->sa_data, 0, size); 8893 else 8894 memcpy(sa->sa_data, dev->dev_addr, 8895 min_t(size_t, size, dev->addr_len)); 8896 sa->sa_family = dev->type; 8897 8898 unlock: 8899 rcu_read_unlock(); 8900 up_read(&dev_addr_sem); 8901 return ret; 8902 } 8903 EXPORT_SYMBOL(dev_get_mac_address); 8904 8905 /** 8906 * dev_change_carrier - Change device carrier 8907 * @dev: device 8908 * @new_carrier: new value 8909 * 8910 * Change device carrier 8911 */ 8912 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8913 { 8914 const struct net_device_ops *ops = dev->netdev_ops; 8915 8916 if (!ops->ndo_change_carrier) 8917 return -EOPNOTSUPP; 8918 if (!netif_device_present(dev)) 8919 return -ENODEV; 8920 return ops->ndo_change_carrier(dev, new_carrier); 8921 } 8922 8923 /** 8924 * dev_get_phys_port_id - Get device physical port ID 8925 * @dev: device 8926 * @ppid: port ID 8927 * 8928 * Get device physical port ID 8929 */ 8930 int dev_get_phys_port_id(struct net_device *dev, 8931 struct netdev_phys_item_id *ppid) 8932 { 8933 const struct net_device_ops *ops = dev->netdev_ops; 8934 8935 if (!ops->ndo_get_phys_port_id) 8936 return -EOPNOTSUPP; 8937 return ops->ndo_get_phys_port_id(dev, ppid); 8938 } 8939 8940 /** 8941 * dev_get_phys_port_name - Get device physical port name 8942 * @dev: device 8943 * @name: port name 8944 * @len: limit of bytes to copy to name 8945 * 8946 * Get device physical port name 8947 */ 8948 int dev_get_phys_port_name(struct net_device *dev, 8949 char *name, size_t len) 8950 { 8951 const struct net_device_ops *ops = dev->netdev_ops; 8952 int err; 8953 8954 if (ops->ndo_get_phys_port_name) { 8955 err = ops->ndo_get_phys_port_name(dev, name, len); 8956 if (err != -EOPNOTSUPP) 8957 return err; 8958 } 8959 return devlink_compat_phys_port_name_get(dev, name, len); 8960 } 8961 8962 /** 8963 * dev_get_port_parent_id - Get the device's port parent identifier 8964 * @dev: network device 8965 * @ppid: pointer to a storage for the port's parent identifier 8966 * @recurse: allow/disallow recursion to lower devices 8967 * 8968 * Get the devices's port parent identifier 8969 */ 8970 int dev_get_port_parent_id(struct net_device *dev, 8971 struct netdev_phys_item_id *ppid, 8972 bool recurse) 8973 { 8974 const struct net_device_ops *ops = dev->netdev_ops; 8975 struct netdev_phys_item_id first = { }; 8976 struct net_device *lower_dev; 8977 struct list_head *iter; 8978 int err; 8979 8980 if (ops->ndo_get_port_parent_id) { 8981 err = ops->ndo_get_port_parent_id(dev, ppid); 8982 if (err != -EOPNOTSUPP) 8983 return err; 8984 } 8985 8986 err = devlink_compat_switch_id_get(dev, ppid); 8987 if (!recurse || err != -EOPNOTSUPP) 8988 return err; 8989 8990 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8991 err = dev_get_port_parent_id(lower_dev, ppid, true); 8992 if (err) 8993 break; 8994 if (!first.id_len) 8995 first = *ppid; 8996 else if (memcmp(&first, ppid, sizeof(*ppid))) 8997 return -EOPNOTSUPP; 8998 } 8999 9000 return err; 9001 } 9002 EXPORT_SYMBOL(dev_get_port_parent_id); 9003 9004 /** 9005 * netdev_port_same_parent_id - Indicate if two network devices have 9006 * the same port parent identifier 9007 * @a: first network device 9008 * @b: second network device 9009 */ 9010 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9011 { 9012 struct netdev_phys_item_id a_id = { }; 9013 struct netdev_phys_item_id b_id = { }; 9014 9015 if (dev_get_port_parent_id(a, &a_id, true) || 9016 dev_get_port_parent_id(b, &b_id, true)) 9017 return false; 9018 9019 return netdev_phys_item_id_same(&a_id, &b_id); 9020 } 9021 EXPORT_SYMBOL(netdev_port_same_parent_id); 9022 9023 /** 9024 * dev_change_proto_down - set carrier according to proto_down. 9025 * 9026 * @dev: device 9027 * @proto_down: new value 9028 */ 9029 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9030 { 9031 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9032 return -EOPNOTSUPP; 9033 if (!netif_device_present(dev)) 9034 return -ENODEV; 9035 if (proto_down) 9036 netif_carrier_off(dev); 9037 else 9038 netif_carrier_on(dev); 9039 dev->proto_down = proto_down; 9040 return 0; 9041 } 9042 9043 /** 9044 * dev_change_proto_down_reason - proto down reason 9045 * 9046 * @dev: device 9047 * @mask: proto down mask 9048 * @value: proto down value 9049 */ 9050 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9051 u32 value) 9052 { 9053 int b; 9054 9055 if (!mask) { 9056 dev->proto_down_reason = value; 9057 } else { 9058 for_each_set_bit(b, &mask, 32) { 9059 if (value & (1 << b)) 9060 dev->proto_down_reason |= BIT(b); 9061 else 9062 dev->proto_down_reason &= ~BIT(b); 9063 } 9064 } 9065 } 9066 9067 struct bpf_xdp_link { 9068 struct bpf_link link; 9069 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9070 int flags; 9071 }; 9072 9073 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9074 { 9075 if (flags & XDP_FLAGS_HW_MODE) 9076 return XDP_MODE_HW; 9077 if (flags & XDP_FLAGS_DRV_MODE) 9078 return XDP_MODE_DRV; 9079 if (flags & XDP_FLAGS_SKB_MODE) 9080 return XDP_MODE_SKB; 9081 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9082 } 9083 9084 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9085 { 9086 switch (mode) { 9087 case XDP_MODE_SKB: 9088 return generic_xdp_install; 9089 case XDP_MODE_DRV: 9090 case XDP_MODE_HW: 9091 return dev->netdev_ops->ndo_bpf; 9092 default: 9093 return NULL; 9094 } 9095 } 9096 9097 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9098 enum bpf_xdp_mode mode) 9099 { 9100 return dev->xdp_state[mode].link; 9101 } 9102 9103 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9104 enum bpf_xdp_mode mode) 9105 { 9106 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9107 9108 if (link) 9109 return link->link.prog; 9110 return dev->xdp_state[mode].prog; 9111 } 9112 9113 u8 dev_xdp_prog_count(struct net_device *dev) 9114 { 9115 u8 count = 0; 9116 int i; 9117 9118 for (i = 0; i < __MAX_XDP_MODE; i++) 9119 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9120 count++; 9121 return count; 9122 } 9123 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9124 9125 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9126 { 9127 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9128 9129 return prog ? prog->aux->id : 0; 9130 } 9131 9132 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9133 struct bpf_xdp_link *link) 9134 { 9135 dev->xdp_state[mode].link = link; 9136 dev->xdp_state[mode].prog = NULL; 9137 } 9138 9139 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9140 struct bpf_prog *prog) 9141 { 9142 dev->xdp_state[mode].link = NULL; 9143 dev->xdp_state[mode].prog = prog; 9144 } 9145 9146 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9147 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9148 u32 flags, struct bpf_prog *prog) 9149 { 9150 struct netdev_bpf xdp; 9151 int err; 9152 9153 memset(&xdp, 0, sizeof(xdp)); 9154 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9155 xdp.extack = extack; 9156 xdp.flags = flags; 9157 xdp.prog = prog; 9158 9159 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9160 * "moved" into driver), so they don't increment it on their own, but 9161 * they do decrement refcnt when program is detached or replaced. 9162 * Given net_device also owns link/prog, we need to bump refcnt here 9163 * to prevent drivers from underflowing it. 9164 */ 9165 if (prog) 9166 bpf_prog_inc(prog); 9167 err = bpf_op(dev, &xdp); 9168 if (err) { 9169 if (prog) 9170 bpf_prog_put(prog); 9171 return err; 9172 } 9173 9174 if (mode != XDP_MODE_HW) 9175 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9176 9177 return 0; 9178 } 9179 9180 static void dev_xdp_uninstall(struct net_device *dev) 9181 { 9182 struct bpf_xdp_link *link; 9183 struct bpf_prog *prog; 9184 enum bpf_xdp_mode mode; 9185 bpf_op_t bpf_op; 9186 9187 ASSERT_RTNL(); 9188 9189 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9190 prog = dev_xdp_prog(dev, mode); 9191 if (!prog) 9192 continue; 9193 9194 bpf_op = dev_xdp_bpf_op(dev, mode); 9195 if (!bpf_op) 9196 continue; 9197 9198 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9199 9200 /* auto-detach link from net device */ 9201 link = dev_xdp_link(dev, mode); 9202 if (link) 9203 link->dev = NULL; 9204 else 9205 bpf_prog_put(prog); 9206 9207 dev_xdp_set_link(dev, mode, NULL); 9208 } 9209 } 9210 9211 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9212 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9213 struct bpf_prog *old_prog, u32 flags) 9214 { 9215 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9216 struct bpf_prog *cur_prog; 9217 struct net_device *upper; 9218 struct list_head *iter; 9219 enum bpf_xdp_mode mode; 9220 bpf_op_t bpf_op; 9221 int err; 9222 9223 ASSERT_RTNL(); 9224 9225 /* either link or prog attachment, never both */ 9226 if (link && (new_prog || old_prog)) 9227 return -EINVAL; 9228 /* link supports only XDP mode flags */ 9229 if (link && (flags & ~XDP_FLAGS_MODES)) { 9230 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9231 return -EINVAL; 9232 } 9233 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9234 if (num_modes > 1) { 9235 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9236 return -EINVAL; 9237 } 9238 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9239 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9240 NL_SET_ERR_MSG(extack, 9241 "More than one program loaded, unset mode is ambiguous"); 9242 return -EINVAL; 9243 } 9244 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9245 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9246 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9247 return -EINVAL; 9248 } 9249 9250 mode = dev_xdp_mode(dev, flags); 9251 /* can't replace attached link */ 9252 if (dev_xdp_link(dev, mode)) { 9253 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9254 return -EBUSY; 9255 } 9256 9257 /* don't allow if an upper device already has a program */ 9258 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9259 if (dev_xdp_prog_count(upper) > 0) { 9260 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9261 return -EEXIST; 9262 } 9263 } 9264 9265 cur_prog = dev_xdp_prog(dev, mode); 9266 /* can't replace attached prog with link */ 9267 if (link && cur_prog) { 9268 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9269 return -EBUSY; 9270 } 9271 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9272 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9273 return -EEXIST; 9274 } 9275 9276 /* put effective new program into new_prog */ 9277 if (link) 9278 new_prog = link->link.prog; 9279 9280 if (new_prog) { 9281 bool offload = mode == XDP_MODE_HW; 9282 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9283 ? XDP_MODE_DRV : XDP_MODE_SKB; 9284 9285 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9286 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9287 return -EBUSY; 9288 } 9289 if (!offload && dev_xdp_prog(dev, other_mode)) { 9290 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9291 return -EEXIST; 9292 } 9293 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9294 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9295 return -EINVAL; 9296 } 9297 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9298 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9299 return -EINVAL; 9300 } 9301 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9302 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9303 return -EINVAL; 9304 } 9305 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9306 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9307 return -EINVAL; 9308 } 9309 } 9310 9311 /* don't call drivers if the effective program didn't change */ 9312 if (new_prog != cur_prog) { 9313 bpf_op = dev_xdp_bpf_op(dev, mode); 9314 if (!bpf_op) { 9315 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9316 return -EOPNOTSUPP; 9317 } 9318 9319 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9320 if (err) 9321 return err; 9322 } 9323 9324 if (link) 9325 dev_xdp_set_link(dev, mode, link); 9326 else 9327 dev_xdp_set_prog(dev, mode, new_prog); 9328 if (cur_prog) 9329 bpf_prog_put(cur_prog); 9330 9331 return 0; 9332 } 9333 9334 static int dev_xdp_attach_link(struct net_device *dev, 9335 struct netlink_ext_ack *extack, 9336 struct bpf_xdp_link *link) 9337 { 9338 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9339 } 9340 9341 static int dev_xdp_detach_link(struct net_device *dev, 9342 struct netlink_ext_ack *extack, 9343 struct bpf_xdp_link *link) 9344 { 9345 enum bpf_xdp_mode mode; 9346 bpf_op_t bpf_op; 9347 9348 ASSERT_RTNL(); 9349 9350 mode = dev_xdp_mode(dev, link->flags); 9351 if (dev_xdp_link(dev, mode) != link) 9352 return -EINVAL; 9353 9354 bpf_op = dev_xdp_bpf_op(dev, mode); 9355 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9356 dev_xdp_set_link(dev, mode, NULL); 9357 return 0; 9358 } 9359 9360 static void bpf_xdp_link_release(struct bpf_link *link) 9361 { 9362 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9363 9364 rtnl_lock(); 9365 9366 /* if racing with net_device's tear down, xdp_link->dev might be 9367 * already NULL, in which case link was already auto-detached 9368 */ 9369 if (xdp_link->dev) { 9370 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9371 xdp_link->dev = NULL; 9372 } 9373 9374 rtnl_unlock(); 9375 } 9376 9377 static int bpf_xdp_link_detach(struct bpf_link *link) 9378 { 9379 bpf_xdp_link_release(link); 9380 return 0; 9381 } 9382 9383 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9384 { 9385 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9386 9387 kfree(xdp_link); 9388 } 9389 9390 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9391 struct seq_file *seq) 9392 { 9393 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9394 u32 ifindex = 0; 9395 9396 rtnl_lock(); 9397 if (xdp_link->dev) 9398 ifindex = xdp_link->dev->ifindex; 9399 rtnl_unlock(); 9400 9401 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9402 } 9403 9404 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9405 struct bpf_link_info *info) 9406 { 9407 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9408 u32 ifindex = 0; 9409 9410 rtnl_lock(); 9411 if (xdp_link->dev) 9412 ifindex = xdp_link->dev->ifindex; 9413 rtnl_unlock(); 9414 9415 info->xdp.ifindex = ifindex; 9416 return 0; 9417 } 9418 9419 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9420 struct bpf_prog *old_prog) 9421 { 9422 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9423 enum bpf_xdp_mode mode; 9424 bpf_op_t bpf_op; 9425 int err = 0; 9426 9427 rtnl_lock(); 9428 9429 /* link might have been auto-released already, so fail */ 9430 if (!xdp_link->dev) { 9431 err = -ENOLINK; 9432 goto out_unlock; 9433 } 9434 9435 if (old_prog && link->prog != old_prog) { 9436 err = -EPERM; 9437 goto out_unlock; 9438 } 9439 old_prog = link->prog; 9440 if (old_prog->type != new_prog->type || 9441 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9442 err = -EINVAL; 9443 goto out_unlock; 9444 } 9445 9446 if (old_prog == new_prog) { 9447 /* no-op, don't disturb drivers */ 9448 bpf_prog_put(new_prog); 9449 goto out_unlock; 9450 } 9451 9452 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9453 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9454 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9455 xdp_link->flags, new_prog); 9456 if (err) 9457 goto out_unlock; 9458 9459 old_prog = xchg(&link->prog, new_prog); 9460 bpf_prog_put(old_prog); 9461 9462 out_unlock: 9463 rtnl_unlock(); 9464 return err; 9465 } 9466 9467 static const struct bpf_link_ops bpf_xdp_link_lops = { 9468 .release = bpf_xdp_link_release, 9469 .dealloc = bpf_xdp_link_dealloc, 9470 .detach = bpf_xdp_link_detach, 9471 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9472 .fill_link_info = bpf_xdp_link_fill_link_info, 9473 .update_prog = bpf_xdp_link_update, 9474 }; 9475 9476 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9477 { 9478 struct net *net = current->nsproxy->net_ns; 9479 struct bpf_link_primer link_primer; 9480 struct netlink_ext_ack extack = {}; 9481 struct bpf_xdp_link *link; 9482 struct net_device *dev; 9483 int err, fd; 9484 9485 rtnl_lock(); 9486 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9487 if (!dev) { 9488 rtnl_unlock(); 9489 return -EINVAL; 9490 } 9491 9492 link = kzalloc(sizeof(*link), GFP_USER); 9493 if (!link) { 9494 err = -ENOMEM; 9495 goto unlock; 9496 } 9497 9498 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9499 link->dev = dev; 9500 link->flags = attr->link_create.flags; 9501 9502 err = bpf_link_prime(&link->link, &link_primer); 9503 if (err) { 9504 kfree(link); 9505 goto unlock; 9506 } 9507 9508 err = dev_xdp_attach_link(dev, &extack, link); 9509 rtnl_unlock(); 9510 9511 if (err) { 9512 link->dev = NULL; 9513 bpf_link_cleanup(&link_primer); 9514 trace_bpf_xdp_link_attach_failed(extack._msg); 9515 goto out_put_dev; 9516 } 9517 9518 fd = bpf_link_settle(&link_primer); 9519 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9520 dev_put(dev); 9521 return fd; 9522 9523 unlock: 9524 rtnl_unlock(); 9525 9526 out_put_dev: 9527 dev_put(dev); 9528 return err; 9529 } 9530 9531 /** 9532 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9533 * @dev: device 9534 * @extack: netlink extended ack 9535 * @fd: new program fd or negative value to clear 9536 * @expected_fd: old program fd that userspace expects to replace or clear 9537 * @flags: xdp-related flags 9538 * 9539 * Set or clear a bpf program for a device 9540 */ 9541 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9542 int fd, int expected_fd, u32 flags) 9543 { 9544 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9545 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9546 int err; 9547 9548 ASSERT_RTNL(); 9549 9550 if (fd >= 0) { 9551 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9552 mode != XDP_MODE_SKB); 9553 if (IS_ERR(new_prog)) 9554 return PTR_ERR(new_prog); 9555 } 9556 9557 if (expected_fd >= 0) { 9558 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9559 mode != XDP_MODE_SKB); 9560 if (IS_ERR(old_prog)) { 9561 err = PTR_ERR(old_prog); 9562 old_prog = NULL; 9563 goto err_out; 9564 } 9565 } 9566 9567 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9568 9569 err_out: 9570 if (err && new_prog) 9571 bpf_prog_put(new_prog); 9572 if (old_prog) 9573 bpf_prog_put(old_prog); 9574 return err; 9575 } 9576 9577 /** 9578 * dev_index_reserve() - allocate an ifindex in a namespace 9579 * @net: the applicable net namespace 9580 * @ifindex: requested ifindex, pass %0 to get one allocated 9581 * 9582 * Allocate a ifindex for a new device. Caller must either use the ifindex 9583 * to store the device (via list_netdevice()) or call dev_index_release() 9584 * to give the index up. 9585 * 9586 * Return: a suitable unique value for a new device interface number or -errno. 9587 */ 9588 static int dev_index_reserve(struct net *net, u32 ifindex) 9589 { 9590 int err; 9591 9592 if (!ifindex) 9593 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9594 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9595 else 9596 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9597 if (err < 0) 9598 return err; 9599 9600 return ifindex; 9601 } 9602 9603 static void dev_index_release(struct net *net, int ifindex) 9604 { 9605 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9606 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9607 } 9608 9609 /* Delayed registration/unregisteration */ 9610 LIST_HEAD(net_todo_list); 9611 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9612 9613 static void net_set_todo(struct net_device *dev) 9614 { 9615 list_add_tail(&dev->todo_list, &net_todo_list); 9616 atomic_inc(&dev_net(dev)->dev_unreg_count); 9617 } 9618 9619 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9620 struct net_device *upper, netdev_features_t features) 9621 { 9622 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9623 netdev_features_t feature; 9624 int feature_bit; 9625 9626 for_each_netdev_feature(upper_disables, feature_bit) { 9627 feature = __NETIF_F_BIT(feature_bit); 9628 if (!(upper->wanted_features & feature) 9629 && (features & feature)) { 9630 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9631 &feature, upper->name); 9632 features &= ~feature; 9633 } 9634 } 9635 9636 return features; 9637 } 9638 9639 static void netdev_sync_lower_features(struct net_device *upper, 9640 struct net_device *lower, netdev_features_t features) 9641 { 9642 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9643 netdev_features_t feature; 9644 int feature_bit; 9645 9646 for_each_netdev_feature(upper_disables, feature_bit) { 9647 feature = __NETIF_F_BIT(feature_bit); 9648 if (!(features & feature) && (lower->features & feature)) { 9649 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9650 &feature, lower->name); 9651 lower->wanted_features &= ~feature; 9652 __netdev_update_features(lower); 9653 9654 if (unlikely(lower->features & feature)) 9655 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9656 &feature, lower->name); 9657 else 9658 netdev_features_change(lower); 9659 } 9660 } 9661 } 9662 9663 static netdev_features_t netdev_fix_features(struct net_device *dev, 9664 netdev_features_t features) 9665 { 9666 /* Fix illegal checksum combinations */ 9667 if ((features & NETIF_F_HW_CSUM) && 9668 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9669 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9670 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9671 } 9672 9673 /* TSO requires that SG is present as well. */ 9674 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9675 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9676 features &= ~NETIF_F_ALL_TSO; 9677 } 9678 9679 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9680 !(features & NETIF_F_IP_CSUM)) { 9681 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9682 features &= ~NETIF_F_TSO; 9683 features &= ~NETIF_F_TSO_ECN; 9684 } 9685 9686 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9687 !(features & NETIF_F_IPV6_CSUM)) { 9688 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9689 features &= ~NETIF_F_TSO6; 9690 } 9691 9692 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9693 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9694 features &= ~NETIF_F_TSO_MANGLEID; 9695 9696 /* TSO ECN requires that TSO is present as well. */ 9697 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9698 features &= ~NETIF_F_TSO_ECN; 9699 9700 /* Software GSO depends on SG. */ 9701 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9702 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9703 features &= ~NETIF_F_GSO; 9704 } 9705 9706 /* GSO partial features require GSO partial be set */ 9707 if ((features & dev->gso_partial_features) && 9708 !(features & NETIF_F_GSO_PARTIAL)) { 9709 netdev_dbg(dev, 9710 "Dropping partially supported GSO features since no GSO partial.\n"); 9711 features &= ~dev->gso_partial_features; 9712 } 9713 9714 if (!(features & NETIF_F_RXCSUM)) { 9715 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9716 * successfully merged by hardware must also have the 9717 * checksum verified by hardware. If the user does not 9718 * want to enable RXCSUM, logically, we should disable GRO_HW. 9719 */ 9720 if (features & NETIF_F_GRO_HW) { 9721 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9722 features &= ~NETIF_F_GRO_HW; 9723 } 9724 } 9725 9726 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9727 if (features & NETIF_F_RXFCS) { 9728 if (features & NETIF_F_LRO) { 9729 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9730 features &= ~NETIF_F_LRO; 9731 } 9732 9733 if (features & NETIF_F_GRO_HW) { 9734 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9735 features &= ~NETIF_F_GRO_HW; 9736 } 9737 } 9738 9739 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9740 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9741 features &= ~NETIF_F_LRO; 9742 } 9743 9744 if (features & NETIF_F_HW_TLS_TX) { 9745 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9746 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9747 bool hw_csum = features & NETIF_F_HW_CSUM; 9748 9749 if (!ip_csum && !hw_csum) { 9750 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9751 features &= ~NETIF_F_HW_TLS_TX; 9752 } 9753 } 9754 9755 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9756 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9757 features &= ~NETIF_F_HW_TLS_RX; 9758 } 9759 9760 return features; 9761 } 9762 9763 int __netdev_update_features(struct net_device *dev) 9764 { 9765 struct net_device *upper, *lower; 9766 netdev_features_t features; 9767 struct list_head *iter; 9768 int err = -1; 9769 9770 ASSERT_RTNL(); 9771 9772 features = netdev_get_wanted_features(dev); 9773 9774 if (dev->netdev_ops->ndo_fix_features) 9775 features = dev->netdev_ops->ndo_fix_features(dev, features); 9776 9777 /* driver might be less strict about feature dependencies */ 9778 features = netdev_fix_features(dev, features); 9779 9780 /* some features can't be enabled if they're off on an upper device */ 9781 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9782 features = netdev_sync_upper_features(dev, upper, features); 9783 9784 if (dev->features == features) 9785 goto sync_lower; 9786 9787 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9788 &dev->features, &features); 9789 9790 if (dev->netdev_ops->ndo_set_features) 9791 err = dev->netdev_ops->ndo_set_features(dev, features); 9792 else 9793 err = 0; 9794 9795 if (unlikely(err < 0)) { 9796 netdev_err(dev, 9797 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9798 err, &features, &dev->features); 9799 /* return non-0 since some features might have changed and 9800 * it's better to fire a spurious notification than miss it 9801 */ 9802 return -1; 9803 } 9804 9805 sync_lower: 9806 /* some features must be disabled on lower devices when disabled 9807 * on an upper device (think: bonding master or bridge) 9808 */ 9809 netdev_for_each_lower_dev(dev, lower, iter) 9810 netdev_sync_lower_features(dev, lower, features); 9811 9812 if (!err) { 9813 netdev_features_t diff = features ^ dev->features; 9814 9815 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9816 /* udp_tunnel_{get,drop}_rx_info both need 9817 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9818 * device, or they won't do anything. 9819 * Thus we need to update dev->features 9820 * *before* calling udp_tunnel_get_rx_info, 9821 * but *after* calling udp_tunnel_drop_rx_info. 9822 */ 9823 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9824 dev->features = features; 9825 udp_tunnel_get_rx_info(dev); 9826 } else { 9827 udp_tunnel_drop_rx_info(dev); 9828 } 9829 } 9830 9831 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9832 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9833 dev->features = features; 9834 err |= vlan_get_rx_ctag_filter_info(dev); 9835 } else { 9836 vlan_drop_rx_ctag_filter_info(dev); 9837 } 9838 } 9839 9840 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9841 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9842 dev->features = features; 9843 err |= vlan_get_rx_stag_filter_info(dev); 9844 } else { 9845 vlan_drop_rx_stag_filter_info(dev); 9846 } 9847 } 9848 9849 dev->features = features; 9850 } 9851 9852 return err < 0 ? 0 : 1; 9853 } 9854 9855 /** 9856 * netdev_update_features - recalculate device features 9857 * @dev: the device to check 9858 * 9859 * Recalculate dev->features set and send notifications if it 9860 * has changed. Should be called after driver or hardware dependent 9861 * conditions might have changed that influence the features. 9862 */ 9863 void netdev_update_features(struct net_device *dev) 9864 { 9865 if (__netdev_update_features(dev)) 9866 netdev_features_change(dev); 9867 } 9868 EXPORT_SYMBOL(netdev_update_features); 9869 9870 /** 9871 * netdev_change_features - recalculate device features 9872 * @dev: the device to check 9873 * 9874 * Recalculate dev->features set and send notifications even 9875 * if they have not changed. Should be called instead of 9876 * netdev_update_features() if also dev->vlan_features might 9877 * have changed to allow the changes to be propagated to stacked 9878 * VLAN devices. 9879 */ 9880 void netdev_change_features(struct net_device *dev) 9881 { 9882 __netdev_update_features(dev); 9883 netdev_features_change(dev); 9884 } 9885 EXPORT_SYMBOL(netdev_change_features); 9886 9887 /** 9888 * netif_stacked_transfer_operstate - transfer operstate 9889 * @rootdev: the root or lower level device to transfer state from 9890 * @dev: the device to transfer operstate to 9891 * 9892 * Transfer operational state from root to device. This is normally 9893 * called when a stacking relationship exists between the root 9894 * device and the device(a leaf device). 9895 */ 9896 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9897 struct net_device *dev) 9898 { 9899 if (rootdev->operstate == IF_OPER_DORMANT) 9900 netif_dormant_on(dev); 9901 else 9902 netif_dormant_off(dev); 9903 9904 if (rootdev->operstate == IF_OPER_TESTING) 9905 netif_testing_on(dev); 9906 else 9907 netif_testing_off(dev); 9908 9909 if (netif_carrier_ok(rootdev)) 9910 netif_carrier_on(dev); 9911 else 9912 netif_carrier_off(dev); 9913 } 9914 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9915 9916 static int netif_alloc_rx_queues(struct net_device *dev) 9917 { 9918 unsigned int i, count = dev->num_rx_queues; 9919 struct netdev_rx_queue *rx; 9920 size_t sz = count * sizeof(*rx); 9921 int err = 0; 9922 9923 BUG_ON(count < 1); 9924 9925 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9926 if (!rx) 9927 return -ENOMEM; 9928 9929 dev->_rx = rx; 9930 9931 for (i = 0; i < count; i++) { 9932 rx[i].dev = dev; 9933 9934 /* XDP RX-queue setup */ 9935 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9936 if (err < 0) 9937 goto err_rxq_info; 9938 } 9939 return 0; 9940 9941 err_rxq_info: 9942 /* Rollback successful reg's and free other resources */ 9943 while (i--) 9944 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9945 kvfree(dev->_rx); 9946 dev->_rx = NULL; 9947 return err; 9948 } 9949 9950 static void netif_free_rx_queues(struct net_device *dev) 9951 { 9952 unsigned int i, count = dev->num_rx_queues; 9953 9954 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9955 if (!dev->_rx) 9956 return; 9957 9958 for (i = 0; i < count; i++) 9959 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9960 9961 kvfree(dev->_rx); 9962 } 9963 9964 static void netdev_init_one_queue(struct net_device *dev, 9965 struct netdev_queue *queue, void *_unused) 9966 { 9967 /* Initialize queue lock */ 9968 spin_lock_init(&queue->_xmit_lock); 9969 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9970 queue->xmit_lock_owner = -1; 9971 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9972 queue->dev = dev; 9973 #ifdef CONFIG_BQL 9974 dql_init(&queue->dql, HZ); 9975 #endif 9976 } 9977 9978 static void netif_free_tx_queues(struct net_device *dev) 9979 { 9980 kvfree(dev->_tx); 9981 } 9982 9983 static int netif_alloc_netdev_queues(struct net_device *dev) 9984 { 9985 unsigned int count = dev->num_tx_queues; 9986 struct netdev_queue *tx; 9987 size_t sz = count * sizeof(*tx); 9988 9989 if (count < 1 || count > 0xffff) 9990 return -EINVAL; 9991 9992 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9993 if (!tx) 9994 return -ENOMEM; 9995 9996 dev->_tx = tx; 9997 9998 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9999 spin_lock_init(&dev->tx_global_lock); 10000 10001 return 0; 10002 } 10003 10004 void netif_tx_stop_all_queues(struct net_device *dev) 10005 { 10006 unsigned int i; 10007 10008 for (i = 0; i < dev->num_tx_queues; i++) { 10009 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10010 10011 netif_tx_stop_queue(txq); 10012 } 10013 } 10014 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10015 10016 /** 10017 * register_netdevice() - register a network device 10018 * @dev: device to register 10019 * 10020 * Take a prepared network device structure and make it externally accessible. 10021 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10022 * Callers must hold the rtnl lock - you may want register_netdev() 10023 * instead of this. 10024 */ 10025 int register_netdevice(struct net_device *dev) 10026 { 10027 int ret; 10028 struct net *net = dev_net(dev); 10029 10030 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10031 NETDEV_FEATURE_COUNT); 10032 BUG_ON(dev_boot_phase); 10033 ASSERT_RTNL(); 10034 10035 might_sleep(); 10036 10037 /* When net_device's are persistent, this will be fatal. */ 10038 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10039 BUG_ON(!net); 10040 10041 ret = ethtool_check_ops(dev->ethtool_ops); 10042 if (ret) 10043 return ret; 10044 10045 spin_lock_init(&dev->addr_list_lock); 10046 netdev_set_addr_lockdep_class(dev); 10047 10048 ret = dev_get_valid_name(net, dev, dev->name); 10049 if (ret < 0) 10050 goto out; 10051 10052 ret = -ENOMEM; 10053 dev->name_node = netdev_name_node_head_alloc(dev); 10054 if (!dev->name_node) 10055 goto out; 10056 10057 /* Init, if this function is available */ 10058 if (dev->netdev_ops->ndo_init) { 10059 ret = dev->netdev_ops->ndo_init(dev); 10060 if (ret) { 10061 if (ret > 0) 10062 ret = -EIO; 10063 goto err_free_name; 10064 } 10065 } 10066 10067 if (((dev->hw_features | dev->features) & 10068 NETIF_F_HW_VLAN_CTAG_FILTER) && 10069 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10070 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10071 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10072 ret = -EINVAL; 10073 goto err_uninit; 10074 } 10075 10076 ret = dev_index_reserve(net, dev->ifindex); 10077 if (ret < 0) 10078 goto err_uninit; 10079 dev->ifindex = ret; 10080 10081 /* Transfer changeable features to wanted_features and enable 10082 * software offloads (GSO and GRO). 10083 */ 10084 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10085 dev->features |= NETIF_F_SOFT_FEATURES; 10086 10087 if (dev->udp_tunnel_nic_info) { 10088 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10089 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10090 } 10091 10092 dev->wanted_features = dev->features & dev->hw_features; 10093 10094 if (!(dev->flags & IFF_LOOPBACK)) 10095 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10096 10097 /* If IPv4 TCP segmentation offload is supported we should also 10098 * allow the device to enable segmenting the frame with the option 10099 * of ignoring a static IP ID value. This doesn't enable the 10100 * feature itself but allows the user to enable it later. 10101 */ 10102 if (dev->hw_features & NETIF_F_TSO) 10103 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10104 if (dev->vlan_features & NETIF_F_TSO) 10105 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10106 if (dev->mpls_features & NETIF_F_TSO) 10107 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10108 if (dev->hw_enc_features & NETIF_F_TSO) 10109 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10110 10111 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10112 */ 10113 dev->vlan_features |= NETIF_F_HIGHDMA; 10114 10115 /* Make NETIF_F_SG inheritable to tunnel devices. 10116 */ 10117 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10118 10119 /* Make NETIF_F_SG inheritable to MPLS. 10120 */ 10121 dev->mpls_features |= NETIF_F_SG; 10122 10123 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10124 ret = notifier_to_errno(ret); 10125 if (ret) 10126 goto err_ifindex_release; 10127 10128 ret = netdev_register_kobject(dev); 10129 write_lock(&dev_base_lock); 10130 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10131 write_unlock(&dev_base_lock); 10132 if (ret) 10133 goto err_uninit_notify; 10134 10135 __netdev_update_features(dev); 10136 10137 /* 10138 * Default initial state at registry is that the 10139 * device is present. 10140 */ 10141 10142 set_bit(__LINK_STATE_PRESENT, &dev->state); 10143 10144 linkwatch_init_dev(dev); 10145 10146 dev_init_scheduler(dev); 10147 10148 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10149 list_netdevice(dev); 10150 10151 add_device_randomness(dev->dev_addr, dev->addr_len); 10152 10153 /* If the device has permanent device address, driver should 10154 * set dev_addr and also addr_assign_type should be set to 10155 * NET_ADDR_PERM (default value). 10156 */ 10157 if (dev->addr_assign_type == NET_ADDR_PERM) 10158 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10159 10160 /* Notify protocols, that a new device appeared. */ 10161 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10162 ret = notifier_to_errno(ret); 10163 if (ret) { 10164 /* Expect explicit free_netdev() on failure */ 10165 dev->needs_free_netdev = false; 10166 unregister_netdevice_queue(dev, NULL); 10167 goto out; 10168 } 10169 /* 10170 * Prevent userspace races by waiting until the network 10171 * device is fully setup before sending notifications. 10172 */ 10173 if (!dev->rtnl_link_ops || 10174 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10175 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10176 10177 out: 10178 return ret; 10179 10180 err_uninit_notify: 10181 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10182 err_ifindex_release: 10183 dev_index_release(net, dev->ifindex); 10184 err_uninit: 10185 if (dev->netdev_ops->ndo_uninit) 10186 dev->netdev_ops->ndo_uninit(dev); 10187 if (dev->priv_destructor) 10188 dev->priv_destructor(dev); 10189 err_free_name: 10190 netdev_name_node_free(dev->name_node); 10191 goto out; 10192 } 10193 EXPORT_SYMBOL(register_netdevice); 10194 10195 /** 10196 * init_dummy_netdev - init a dummy network device for NAPI 10197 * @dev: device to init 10198 * 10199 * This takes a network device structure and initialize the minimum 10200 * amount of fields so it can be used to schedule NAPI polls without 10201 * registering a full blown interface. This is to be used by drivers 10202 * that need to tie several hardware interfaces to a single NAPI 10203 * poll scheduler due to HW limitations. 10204 */ 10205 int init_dummy_netdev(struct net_device *dev) 10206 { 10207 /* Clear everything. Note we don't initialize spinlocks 10208 * are they aren't supposed to be taken by any of the 10209 * NAPI code and this dummy netdev is supposed to be 10210 * only ever used for NAPI polls 10211 */ 10212 memset(dev, 0, sizeof(struct net_device)); 10213 10214 /* make sure we BUG if trying to hit standard 10215 * register/unregister code path 10216 */ 10217 dev->reg_state = NETREG_DUMMY; 10218 10219 /* NAPI wants this */ 10220 INIT_LIST_HEAD(&dev->napi_list); 10221 10222 /* a dummy interface is started by default */ 10223 set_bit(__LINK_STATE_PRESENT, &dev->state); 10224 set_bit(__LINK_STATE_START, &dev->state); 10225 10226 /* napi_busy_loop stats accounting wants this */ 10227 dev_net_set(dev, &init_net); 10228 10229 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10230 * because users of this 'device' dont need to change 10231 * its refcount. 10232 */ 10233 10234 return 0; 10235 } 10236 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10237 10238 10239 /** 10240 * register_netdev - register a network device 10241 * @dev: device to register 10242 * 10243 * Take a completed network device structure and add it to the kernel 10244 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10245 * chain. 0 is returned on success. A negative errno code is returned 10246 * on a failure to set up the device, or if the name is a duplicate. 10247 * 10248 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10249 * and expands the device name if you passed a format string to 10250 * alloc_netdev. 10251 */ 10252 int register_netdev(struct net_device *dev) 10253 { 10254 int err; 10255 10256 if (rtnl_lock_killable()) 10257 return -EINTR; 10258 err = register_netdevice(dev); 10259 rtnl_unlock(); 10260 return err; 10261 } 10262 EXPORT_SYMBOL(register_netdev); 10263 10264 int netdev_refcnt_read(const struct net_device *dev) 10265 { 10266 #ifdef CONFIG_PCPU_DEV_REFCNT 10267 int i, refcnt = 0; 10268 10269 for_each_possible_cpu(i) 10270 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10271 return refcnt; 10272 #else 10273 return refcount_read(&dev->dev_refcnt); 10274 #endif 10275 } 10276 EXPORT_SYMBOL(netdev_refcnt_read); 10277 10278 int netdev_unregister_timeout_secs __read_mostly = 10; 10279 10280 #define WAIT_REFS_MIN_MSECS 1 10281 #define WAIT_REFS_MAX_MSECS 250 10282 /** 10283 * netdev_wait_allrefs_any - wait until all references are gone. 10284 * @list: list of net_devices to wait on 10285 * 10286 * This is called when unregistering network devices. 10287 * 10288 * Any protocol or device that holds a reference should register 10289 * for netdevice notification, and cleanup and put back the 10290 * reference if they receive an UNREGISTER event. 10291 * We can get stuck here if buggy protocols don't correctly 10292 * call dev_put. 10293 */ 10294 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10295 { 10296 unsigned long rebroadcast_time, warning_time; 10297 struct net_device *dev; 10298 int wait = 0; 10299 10300 rebroadcast_time = warning_time = jiffies; 10301 10302 list_for_each_entry(dev, list, todo_list) 10303 if (netdev_refcnt_read(dev) == 1) 10304 return dev; 10305 10306 while (true) { 10307 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10308 rtnl_lock(); 10309 10310 /* Rebroadcast unregister notification */ 10311 list_for_each_entry(dev, list, todo_list) 10312 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10313 10314 __rtnl_unlock(); 10315 rcu_barrier(); 10316 rtnl_lock(); 10317 10318 list_for_each_entry(dev, list, todo_list) 10319 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10320 &dev->state)) { 10321 /* We must not have linkwatch events 10322 * pending on unregister. If this 10323 * happens, we simply run the queue 10324 * unscheduled, resulting in a noop 10325 * for this device. 10326 */ 10327 linkwatch_run_queue(); 10328 break; 10329 } 10330 10331 __rtnl_unlock(); 10332 10333 rebroadcast_time = jiffies; 10334 } 10335 10336 if (!wait) { 10337 rcu_barrier(); 10338 wait = WAIT_REFS_MIN_MSECS; 10339 } else { 10340 msleep(wait); 10341 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10342 } 10343 10344 list_for_each_entry(dev, list, todo_list) 10345 if (netdev_refcnt_read(dev) == 1) 10346 return dev; 10347 10348 if (time_after(jiffies, warning_time + 10349 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10350 list_for_each_entry(dev, list, todo_list) { 10351 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10352 dev->name, netdev_refcnt_read(dev)); 10353 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10354 } 10355 10356 warning_time = jiffies; 10357 } 10358 } 10359 } 10360 10361 /* The sequence is: 10362 * 10363 * rtnl_lock(); 10364 * ... 10365 * register_netdevice(x1); 10366 * register_netdevice(x2); 10367 * ... 10368 * unregister_netdevice(y1); 10369 * unregister_netdevice(y2); 10370 * ... 10371 * rtnl_unlock(); 10372 * free_netdev(y1); 10373 * free_netdev(y2); 10374 * 10375 * We are invoked by rtnl_unlock(). 10376 * This allows us to deal with problems: 10377 * 1) We can delete sysfs objects which invoke hotplug 10378 * without deadlocking with linkwatch via keventd. 10379 * 2) Since we run with the RTNL semaphore not held, we can sleep 10380 * safely in order to wait for the netdev refcnt to drop to zero. 10381 * 10382 * We must not return until all unregister events added during 10383 * the interval the lock was held have been completed. 10384 */ 10385 void netdev_run_todo(void) 10386 { 10387 struct net_device *dev, *tmp; 10388 struct list_head list; 10389 #ifdef CONFIG_LOCKDEP 10390 struct list_head unlink_list; 10391 10392 list_replace_init(&net_unlink_list, &unlink_list); 10393 10394 while (!list_empty(&unlink_list)) { 10395 struct net_device *dev = list_first_entry(&unlink_list, 10396 struct net_device, 10397 unlink_list); 10398 list_del_init(&dev->unlink_list); 10399 dev->nested_level = dev->lower_level - 1; 10400 } 10401 #endif 10402 10403 /* Snapshot list, allow later requests */ 10404 list_replace_init(&net_todo_list, &list); 10405 10406 __rtnl_unlock(); 10407 10408 /* Wait for rcu callbacks to finish before next phase */ 10409 if (!list_empty(&list)) 10410 rcu_barrier(); 10411 10412 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10413 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10414 netdev_WARN(dev, "run_todo but not unregistering\n"); 10415 list_del(&dev->todo_list); 10416 continue; 10417 } 10418 10419 write_lock(&dev_base_lock); 10420 dev->reg_state = NETREG_UNREGISTERED; 10421 write_unlock(&dev_base_lock); 10422 linkwatch_forget_dev(dev); 10423 } 10424 10425 while (!list_empty(&list)) { 10426 dev = netdev_wait_allrefs_any(&list); 10427 list_del(&dev->todo_list); 10428 10429 /* paranoia */ 10430 BUG_ON(netdev_refcnt_read(dev) != 1); 10431 BUG_ON(!list_empty(&dev->ptype_all)); 10432 BUG_ON(!list_empty(&dev->ptype_specific)); 10433 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10434 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10435 10436 if (dev->priv_destructor) 10437 dev->priv_destructor(dev); 10438 if (dev->needs_free_netdev) 10439 free_netdev(dev); 10440 10441 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10442 wake_up(&netdev_unregistering_wq); 10443 10444 /* Free network device */ 10445 kobject_put(&dev->dev.kobj); 10446 } 10447 } 10448 10449 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10450 * all the same fields in the same order as net_device_stats, with only 10451 * the type differing, but rtnl_link_stats64 may have additional fields 10452 * at the end for newer counters. 10453 */ 10454 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10455 const struct net_device_stats *netdev_stats) 10456 { 10457 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10458 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10459 u64 *dst = (u64 *)stats64; 10460 10461 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10462 for (i = 0; i < n; i++) 10463 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10464 /* zero out counters that only exist in rtnl_link_stats64 */ 10465 memset((char *)stats64 + n * sizeof(u64), 0, 10466 sizeof(*stats64) - n * sizeof(u64)); 10467 } 10468 EXPORT_SYMBOL(netdev_stats_to_stats64); 10469 10470 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10471 { 10472 struct net_device_core_stats __percpu *p; 10473 10474 p = alloc_percpu_gfp(struct net_device_core_stats, 10475 GFP_ATOMIC | __GFP_NOWARN); 10476 10477 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10478 free_percpu(p); 10479 10480 /* This READ_ONCE() pairs with the cmpxchg() above */ 10481 return READ_ONCE(dev->core_stats); 10482 } 10483 EXPORT_SYMBOL(netdev_core_stats_alloc); 10484 10485 /** 10486 * dev_get_stats - get network device statistics 10487 * @dev: device to get statistics from 10488 * @storage: place to store stats 10489 * 10490 * Get network statistics from device. Return @storage. 10491 * The device driver may provide its own method by setting 10492 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10493 * otherwise the internal statistics structure is used. 10494 */ 10495 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10496 struct rtnl_link_stats64 *storage) 10497 { 10498 const struct net_device_ops *ops = dev->netdev_ops; 10499 const struct net_device_core_stats __percpu *p; 10500 10501 if (ops->ndo_get_stats64) { 10502 memset(storage, 0, sizeof(*storage)); 10503 ops->ndo_get_stats64(dev, storage); 10504 } else if (ops->ndo_get_stats) { 10505 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10506 } else { 10507 netdev_stats_to_stats64(storage, &dev->stats); 10508 } 10509 10510 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10511 p = READ_ONCE(dev->core_stats); 10512 if (p) { 10513 const struct net_device_core_stats *core_stats; 10514 int i; 10515 10516 for_each_possible_cpu(i) { 10517 core_stats = per_cpu_ptr(p, i); 10518 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10519 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10520 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10521 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10522 } 10523 } 10524 return storage; 10525 } 10526 EXPORT_SYMBOL(dev_get_stats); 10527 10528 /** 10529 * dev_fetch_sw_netstats - get per-cpu network device statistics 10530 * @s: place to store stats 10531 * @netstats: per-cpu network stats to read from 10532 * 10533 * Read per-cpu network statistics and populate the related fields in @s. 10534 */ 10535 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10536 const struct pcpu_sw_netstats __percpu *netstats) 10537 { 10538 int cpu; 10539 10540 for_each_possible_cpu(cpu) { 10541 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10542 const struct pcpu_sw_netstats *stats; 10543 unsigned int start; 10544 10545 stats = per_cpu_ptr(netstats, cpu); 10546 do { 10547 start = u64_stats_fetch_begin(&stats->syncp); 10548 rx_packets = u64_stats_read(&stats->rx_packets); 10549 rx_bytes = u64_stats_read(&stats->rx_bytes); 10550 tx_packets = u64_stats_read(&stats->tx_packets); 10551 tx_bytes = u64_stats_read(&stats->tx_bytes); 10552 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10553 10554 s->rx_packets += rx_packets; 10555 s->rx_bytes += rx_bytes; 10556 s->tx_packets += tx_packets; 10557 s->tx_bytes += tx_bytes; 10558 } 10559 } 10560 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10561 10562 /** 10563 * dev_get_tstats64 - ndo_get_stats64 implementation 10564 * @dev: device to get statistics from 10565 * @s: place to store stats 10566 * 10567 * Populate @s from dev->stats and dev->tstats. Can be used as 10568 * ndo_get_stats64() callback. 10569 */ 10570 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10571 { 10572 netdev_stats_to_stats64(s, &dev->stats); 10573 dev_fetch_sw_netstats(s, dev->tstats); 10574 } 10575 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10576 10577 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10578 { 10579 struct netdev_queue *queue = dev_ingress_queue(dev); 10580 10581 #ifdef CONFIG_NET_CLS_ACT 10582 if (queue) 10583 return queue; 10584 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10585 if (!queue) 10586 return NULL; 10587 netdev_init_one_queue(dev, queue, NULL); 10588 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10589 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10590 rcu_assign_pointer(dev->ingress_queue, queue); 10591 #endif 10592 return queue; 10593 } 10594 10595 static const struct ethtool_ops default_ethtool_ops; 10596 10597 void netdev_set_default_ethtool_ops(struct net_device *dev, 10598 const struct ethtool_ops *ops) 10599 { 10600 if (dev->ethtool_ops == &default_ethtool_ops) 10601 dev->ethtool_ops = ops; 10602 } 10603 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10604 10605 /** 10606 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10607 * @dev: netdev to enable the IRQ coalescing on 10608 * 10609 * Sets a conservative default for SW IRQ coalescing. Users can use 10610 * sysfs attributes to override the default values. 10611 */ 10612 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10613 { 10614 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10615 10616 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10617 dev->gro_flush_timeout = 20000; 10618 dev->napi_defer_hard_irqs = 1; 10619 } 10620 } 10621 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10622 10623 void netdev_freemem(struct net_device *dev) 10624 { 10625 char *addr = (char *)dev - dev->padded; 10626 10627 kvfree(addr); 10628 } 10629 10630 /** 10631 * alloc_netdev_mqs - allocate network device 10632 * @sizeof_priv: size of private data to allocate space for 10633 * @name: device name format string 10634 * @name_assign_type: origin of device name 10635 * @setup: callback to initialize device 10636 * @txqs: the number of TX subqueues to allocate 10637 * @rxqs: the number of RX subqueues to allocate 10638 * 10639 * Allocates a struct net_device with private data area for driver use 10640 * and performs basic initialization. Also allocates subqueue structs 10641 * for each queue on the device. 10642 */ 10643 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10644 unsigned char name_assign_type, 10645 void (*setup)(struct net_device *), 10646 unsigned int txqs, unsigned int rxqs) 10647 { 10648 struct net_device *dev; 10649 unsigned int alloc_size; 10650 struct net_device *p; 10651 10652 BUG_ON(strlen(name) >= sizeof(dev->name)); 10653 10654 if (txqs < 1) { 10655 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10656 return NULL; 10657 } 10658 10659 if (rxqs < 1) { 10660 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10661 return NULL; 10662 } 10663 10664 alloc_size = sizeof(struct net_device); 10665 if (sizeof_priv) { 10666 /* ensure 32-byte alignment of private area */ 10667 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10668 alloc_size += sizeof_priv; 10669 } 10670 /* ensure 32-byte alignment of whole construct */ 10671 alloc_size += NETDEV_ALIGN - 1; 10672 10673 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10674 if (!p) 10675 return NULL; 10676 10677 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10678 dev->padded = (char *)dev - (char *)p; 10679 10680 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10681 #ifdef CONFIG_PCPU_DEV_REFCNT 10682 dev->pcpu_refcnt = alloc_percpu(int); 10683 if (!dev->pcpu_refcnt) 10684 goto free_dev; 10685 __dev_hold(dev); 10686 #else 10687 refcount_set(&dev->dev_refcnt, 1); 10688 #endif 10689 10690 if (dev_addr_init(dev)) 10691 goto free_pcpu; 10692 10693 dev_mc_init(dev); 10694 dev_uc_init(dev); 10695 10696 dev_net_set(dev, &init_net); 10697 10698 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10699 dev->xdp_zc_max_segs = 1; 10700 dev->gso_max_segs = GSO_MAX_SEGS; 10701 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10702 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10703 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10704 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10705 dev->tso_max_segs = TSO_MAX_SEGS; 10706 dev->upper_level = 1; 10707 dev->lower_level = 1; 10708 #ifdef CONFIG_LOCKDEP 10709 dev->nested_level = 0; 10710 INIT_LIST_HEAD(&dev->unlink_list); 10711 #endif 10712 10713 INIT_LIST_HEAD(&dev->napi_list); 10714 INIT_LIST_HEAD(&dev->unreg_list); 10715 INIT_LIST_HEAD(&dev->close_list); 10716 INIT_LIST_HEAD(&dev->link_watch_list); 10717 INIT_LIST_HEAD(&dev->adj_list.upper); 10718 INIT_LIST_HEAD(&dev->adj_list.lower); 10719 INIT_LIST_HEAD(&dev->ptype_all); 10720 INIT_LIST_HEAD(&dev->ptype_specific); 10721 INIT_LIST_HEAD(&dev->net_notifier_list); 10722 #ifdef CONFIG_NET_SCHED 10723 hash_init(dev->qdisc_hash); 10724 #endif 10725 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10726 setup(dev); 10727 10728 if (!dev->tx_queue_len) { 10729 dev->priv_flags |= IFF_NO_QUEUE; 10730 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10731 } 10732 10733 dev->num_tx_queues = txqs; 10734 dev->real_num_tx_queues = txqs; 10735 if (netif_alloc_netdev_queues(dev)) 10736 goto free_all; 10737 10738 dev->num_rx_queues = rxqs; 10739 dev->real_num_rx_queues = rxqs; 10740 if (netif_alloc_rx_queues(dev)) 10741 goto free_all; 10742 10743 strcpy(dev->name, name); 10744 dev->name_assign_type = name_assign_type; 10745 dev->group = INIT_NETDEV_GROUP; 10746 if (!dev->ethtool_ops) 10747 dev->ethtool_ops = &default_ethtool_ops; 10748 10749 nf_hook_netdev_init(dev); 10750 10751 return dev; 10752 10753 free_all: 10754 free_netdev(dev); 10755 return NULL; 10756 10757 free_pcpu: 10758 #ifdef CONFIG_PCPU_DEV_REFCNT 10759 free_percpu(dev->pcpu_refcnt); 10760 free_dev: 10761 #endif 10762 netdev_freemem(dev); 10763 return NULL; 10764 } 10765 EXPORT_SYMBOL(alloc_netdev_mqs); 10766 10767 /** 10768 * free_netdev - free network device 10769 * @dev: device 10770 * 10771 * This function does the last stage of destroying an allocated device 10772 * interface. The reference to the device object is released. If this 10773 * is the last reference then it will be freed.Must be called in process 10774 * context. 10775 */ 10776 void free_netdev(struct net_device *dev) 10777 { 10778 struct napi_struct *p, *n; 10779 10780 might_sleep(); 10781 10782 /* When called immediately after register_netdevice() failed the unwind 10783 * handling may still be dismantling the device. Handle that case by 10784 * deferring the free. 10785 */ 10786 if (dev->reg_state == NETREG_UNREGISTERING) { 10787 ASSERT_RTNL(); 10788 dev->needs_free_netdev = true; 10789 return; 10790 } 10791 10792 netif_free_tx_queues(dev); 10793 netif_free_rx_queues(dev); 10794 10795 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10796 10797 /* Flush device addresses */ 10798 dev_addr_flush(dev); 10799 10800 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10801 netif_napi_del(p); 10802 10803 ref_tracker_dir_exit(&dev->refcnt_tracker); 10804 #ifdef CONFIG_PCPU_DEV_REFCNT 10805 free_percpu(dev->pcpu_refcnt); 10806 dev->pcpu_refcnt = NULL; 10807 #endif 10808 free_percpu(dev->core_stats); 10809 dev->core_stats = NULL; 10810 free_percpu(dev->xdp_bulkq); 10811 dev->xdp_bulkq = NULL; 10812 10813 /* Compatibility with error handling in drivers */ 10814 if (dev->reg_state == NETREG_UNINITIALIZED) { 10815 netdev_freemem(dev); 10816 return; 10817 } 10818 10819 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10820 dev->reg_state = NETREG_RELEASED; 10821 10822 /* will free via device release */ 10823 put_device(&dev->dev); 10824 } 10825 EXPORT_SYMBOL(free_netdev); 10826 10827 /** 10828 * synchronize_net - Synchronize with packet receive processing 10829 * 10830 * Wait for packets currently being received to be done. 10831 * Does not block later packets from starting. 10832 */ 10833 void synchronize_net(void) 10834 { 10835 might_sleep(); 10836 if (rtnl_is_locked()) 10837 synchronize_rcu_expedited(); 10838 else 10839 synchronize_rcu(); 10840 } 10841 EXPORT_SYMBOL(synchronize_net); 10842 10843 /** 10844 * unregister_netdevice_queue - remove device from the kernel 10845 * @dev: device 10846 * @head: list 10847 * 10848 * This function shuts down a device interface and removes it 10849 * from the kernel tables. 10850 * If head not NULL, device is queued to be unregistered later. 10851 * 10852 * Callers must hold the rtnl semaphore. You may want 10853 * unregister_netdev() instead of this. 10854 */ 10855 10856 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10857 { 10858 ASSERT_RTNL(); 10859 10860 if (head) { 10861 list_move_tail(&dev->unreg_list, head); 10862 } else { 10863 LIST_HEAD(single); 10864 10865 list_add(&dev->unreg_list, &single); 10866 unregister_netdevice_many(&single); 10867 } 10868 } 10869 EXPORT_SYMBOL(unregister_netdevice_queue); 10870 10871 void unregister_netdevice_many_notify(struct list_head *head, 10872 u32 portid, const struct nlmsghdr *nlh) 10873 { 10874 struct net_device *dev, *tmp; 10875 LIST_HEAD(close_head); 10876 10877 BUG_ON(dev_boot_phase); 10878 ASSERT_RTNL(); 10879 10880 if (list_empty(head)) 10881 return; 10882 10883 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10884 /* Some devices call without registering 10885 * for initialization unwind. Remove those 10886 * devices and proceed with the remaining. 10887 */ 10888 if (dev->reg_state == NETREG_UNINITIALIZED) { 10889 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10890 dev->name, dev); 10891 10892 WARN_ON(1); 10893 list_del(&dev->unreg_list); 10894 continue; 10895 } 10896 dev->dismantle = true; 10897 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10898 } 10899 10900 /* If device is running, close it first. */ 10901 list_for_each_entry(dev, head, unreg_list) 10902 list_add_tail(&dev->close_list, &close_head); 10903 dev_close_many(&close_head, true); 10904 10905 list_for_each_entry(dev, head, unreg_list) { 10906 /* And unlink it from device chain. */ 10907 write_lock(&dev_base_lock); 10908 unlist_netdevice(dev, false); 10909 dev->reg_state = NETREG_UNREGISTERING; 10910 write_unlock(&dev_base_lock); 10911 } 10912 flush_all_backlogs(); 10913 10914 synchronize_net(); 10915 10916 list_for_each_entry(dev, head, unreg_list) { 10917 struct sk_buff *skb = NULL; 10918 10919 /* Shutdown queueing discipline. */ 10920 dev_shutdown(dev); 10921 dev_tcx_uninstall(dev); 10922 dev_xdp_uninstall(dev); 10923 bpf_dev_bound_netdev_unregister(dev); 10924 10925 netdev_offload_xstats_disable_all(dev); 10926 10927 /* Notify protocols, that we are about to destroy 10928 * this device. They should clean all the things. 10929 */ 10930 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10931 10932 if (!dev->rtnl_link_ops || 10933 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10934 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10935 GFP_KERNEL, NULL, 0, 10936 portid, nlh); 10937 10938 /* 10939 * Flush the unicast and multicast chains 10940 */ 10941 dev_uc_flush(dev); 10942 dev_mc_flush(dev); 10943 10944 netdev_name_node_alt_flush(dev); 10945 netdev_name_node_free(dev->name_node); 10946 10947 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10948 10949 if (dev->netdev_ops->ndo_uninit) 10950 dev->netdev_ops->ndo_uninit(dev); 10951 10952 if (skb) 10953 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10954 10955 /* Notifier chain MUST detach us all upper devices. */ 10956 WARN_ON(netdev_has_any_upper_dev(dev)); 10957 WARN_ON(netdev_has_any_lower_dev(dev)); 10958 10959 /* Remove entries from kobject tree */ 10960 netdev_unregister_kobject(dev); 10961 #ifdef CONFIG_XPS 10962 /* Remove XPS queueing entries */ 10963 netif_reset_xps_queues_gt(dev, 0); 10964 #endif 10965 } 10966 10967 synchronize_net(); 10968 10969 list_for_each_entry(dev, head, unreg_list) { 10970 netdev_put(dev, &dev->dev_registered_tracker); 10971 net_set_todo(dev); 10972 } 10973 10974 list_del(head); 10975 } 10976 10977 /** 10978 * unregister_netdevice_many - unregister many devices 10979 * @head: list of devices 10980 * 10981 * Note: As most callers use a stack allocated list_head, 10982 * we force a list_del() to make sure stack wont be corrupted later. 10983 */ 10984 void unregister_netdevice_many(struct list_head *head) 10985 { 10986 unregister_netdevice_many_notify(head, 0, NULL); 10987 } 10988 EXPORT_SYMBOL(unregister_netdevice_many); 10989 10990 /** 10991 * unregister_netdev - remove device from the kernel 10992 * @dev: device 10993 * 10994 * This function shuts down a device interface and removes it 10995 * from the kernel tables. 10996 * 10997 * This is just a wrapper for unregister_netdevice that takes 10998 * the rtnl semaphore. In general you want to use this and not 10999 * unregister_netdevice. 11000 */ 11001 void unregister_netdev(struct net_device *dev) 11002 { 11003 rtnl_lock(); 11004 unregister_netdevice(dev); 11005 rtnl_unlock(); 11006 } 11007 EXPORT_SYMBOL(unregister_netdev); 11008 11009 /** 11010 * __dev_change_net_namespace - move device to different nethost namespace 11011 * @dev: device 11012 * @net: network namespace 11013 * @pat: If not NULL name pattern to try if the current device name 11014 * is already taken in the destination network namespace. 11015 * @new_ifindex: If not zero, specifies device index in the target 11016 * namespace. 11017 * 11018 * This function shuts down a device interface and moves it 11019 * to a new network namespace. On success 0 is returned, on 11020 * a failure a netagive errno code is returned. 11021 * 11022 * Callers must hold the rtnl semaphore. 11023 */ 11024 11025 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11026 const char *pat, int new_ifindex) 11027 { 11028 struct net *net_old = dev_net(dev); 11029 int err, new_nsid; 11030 11031 ASSERT_RTNL(); 11032 11033 /* Don't allow namespace local devices to be moved. */ 11034 err = -EINVAL; 11035 if (dev->features & NETIF_F_NETNS_LOCAL) 11036 goto out; 11037 11038 /* Ensure the device has been registrered */ 11039 if (dev->reg_state != NETREG_REGISTERED) 11040 goto out; 11041 11042 /* Get out if there is nothing todo */ 11043 err = 0; 11044 if (net_eq(net_old, net)) 11045 goto out; 11046 11047 /* Pick the destination device name, and ensure 11048 * we can use it in the destination network namespace. 11049 */ 11050 err = -EEXIST; 11051 if (netdev_name_in_use(net, dev->name)) { 11052 /* We get here if we can't use the current device name */ 11053 if (!pat) 11054 goto out; 11055 err = dev_get_valid_name(net, dev, pat); 11056 if (err < 0) 11057 goto out; 11058 } 11059 11060 /* Check that new_ifindex isn't used yet. */ 11061 if (new_ifindex) { 11062 err = dev_index_reserve(net, new_ifindex); 11063 if (err < 0) 11064 goto out; 11065 } else { 11066 /* If there is an ifindex conflict assign a new one */ 11067 err = dev_index_reserve(net, dev->ifindex); 11068 if (err == -EBUSY) 11069 err = dev_index_reserve(net, 0); 11070 if (err < 0) 11071 goto out; 11072 new_ifindex = err; 11073 } 11074 11075 /* 11076 * And now a mini version of register_netdevice unregister_netdevice. 11077 */ 11078 11079 /* If device is running close it first. */ 11080 dev_close(dev); 11081 11082 /* And unlink it from device chain */ 11083 unlist_netdevice(dev, true); 11084 11085 synchronize_net(); 11086 11087 /* Shutdown queueing discipline. */ 11088 dev_shutdown(dev); 11089 11090 /* Notify protocols, that we are about to destroy 11091 * this device. They should clean all the things. 11092 * 11093 * Note that dev->reg_state stays at NETREG_REGISTERED. 11094 * This is wanted because this way 8021q and macvlan know 11095 * the device is just moving and can keep their slaves up. 11096 */ 11097 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11098 rcu_barrier(); 11099 11100 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11101 11102 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11103 new_ifindex); 11104 11105 /* 11106 * Flush the unicast and multicast chains 11107 */ 11108 dev_uc_flush(dev); 11109 dev_mc_flush(dev); 11110 11111 /* Send a netdev-removed uevent to the old namespace */ 11112 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11113 netdev_adjacent_del_links(dev); 11114 11115 /* Move per-net netdevice notifiers that are following the netdevice */ 11116 move_netdevice_notifiers_dev_net(dev, net); 11117 11118 /* Actually switch the network namespace */ 11119 dev_net_set(dev, net); 11120 dev->ifindex = new_ifindex; 11121 11122 /* Send a netdev-add uevent to the new namespace */ 11123 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11124 netdev_adjacent_add_links(dev); 11125 11126 /* Fixup kobjects */ 11127 err = device_rename(&dev->dev, dev->name); 11128 WARN_ON(err); 11129 11130 /* Adapt owner in case owning user namespace of target network 11131 * namespace is different from the original one. 11132 */ 11133 err = netdev_change_owner(dev, net_old, net); 11134 WARN_ON(err); 11135 11136 /* Add the device back in the hashes */ 11137 list_netdevice(dev); 11138 11139 /* Notify protocols, that a new device appeared. */ 11140 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11141 11142 /* 11143 * Prevent userspace races by waiting until the network 11144 * device is fully setup before sending notifications. 11145 */ 11146 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11147 11148 synchronize_net(); 11149 err = 0; 11150 out: 11151 return err; 11152 } 11153 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11154 11155 static int dev_cpu_dead(unsigned int oldcpu) 11156 { 11157 struct sk_buff **list_skb; 11158 struct sk_buff *skb; 11159 unsigned int cpu; 11160 struct softnet_data *sd, *oldsd, *remsd = NULL; 11161 11162 local_irq_disable(); 11163 cpu = smp_processor_id(); 11164 sd = &per_cpu(softnet_data, cpu); 11165 oldsd = &per_cpu(softnet_data, oldcpu); 11166 11167 /* Find end of our completion_queue. */ 11168 list_skb = &sd->completion_queue; 11169 while (*list_skb) 11170 list_skb = &(*list_skb)->next; 11171 /* Append completion queue from offline CPU. */ 11172 *list_skb = oldsd->completion_queue; 11173 oldsd->completion_queue = NULL; 11174 11175 /* Append output queue from offline CPU. */ 11176 if (oldsd->output_queue) { 11177 *sd->output_queue_tailp = oldsd->output_queue; 11178 sd->output_queue_tailp = oldsd->output_queue_tailp; 11179 oldsd->output_queue = NULL; 11180 oldsd->output_queue_tailp = &oldsd->output_queue; 11181 } 11182 /* Append NAPI poll list from offline CPU, with one exception : 11183 * process_backlog() must be called by cpu owning percpu backlog. 11184 * We properly handle process_queue & input_pkt_queue later. 11185 */ 11186 while (!list_empty(&oldsd->poll_list)) { 11187 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11188 struct napi_struct, 11189 poll_list); 11190 11191 list_del_init(&napi->poll_list); 11192 if (napi->poll == process_backlog) 11193 napi->state = 0; 11194 else 11195 ____napi_schedule(sd, napi); 11196 } 11197 11198 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11199 local_irq_enable(); 11200 11201 #ifdef CONFIG_RPS 11202 remsd = oldsd->rps_ipi_list; 11203 oldsd->rps_ipi_list = NULL; 11204 #endif 11205 /* send out pending IPI's on offline CPU */ 11206 net_rps_send_ipi(remsd); 11207 11208 /* Process offline CPU's input_pkt_queue */ 11209 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11210 netif_rx(skb); 11211 input_queue_head_incr(oldsd); 11212 } 11213 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11214 netif_rx(skb); 11215 input_queue_head_incr(oldsd); 11216 } 11217 11218 return 0; 11219 } 11220 11221 /** 11222 * netdev_increment_features - increment feature set by one 11223 * @all: current feature set 11224 * @one: new feature set 11225 * @mask: mask feature set 11226 * 11227 * Computes a new feature set after adding a device with feature set 11228 * @one to the master device with current feature set @all. Will not 11229 * enable anything that is off in @mask. Returns the new feature set. 11230 */ 11231 netdev_features_t netdev_increment_features(netdev_features_t all, 11232 netdev_features_t one, netdev_features_t mask) 11233 { 11234 if (mask & NETIF_F_HW_CSUM) 11235 mask |= NETIF_F_CSUM_MASK; 11236 mask |= NETIF_F_VLAN_CHALLENGED; 11237 11238 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11239 all &= one | ~NETIF_F_ALL_FOR_ALL; 11240 11241 /* If one device supports hw checksumming, set for all. */ 11242 if (all & NETIF_F_HW_CSUM) 11243 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11244 11245 return all; 11246 } 11247 EXPORT_SYMBOL(netdev_increment_features); 11248 11249 static struct hlist_head * __net_init netdev_create_hash(void) 11250 { 11251 int i; 11252 struct hlist_head *hash; 11253 11254 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11255 if (hash != NULL) 11256 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11257 INIT_HLIST_HEAD(&hash[i]); 11258 11259 return hash; 11260 } 11261 11262 /* Initialize per network namespace state */ 11263 static int __net_init netdev_init(struct net *net) 11264 { 11265 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11266 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11267 11268 INIT_LIST_HEAD(&net->dev_base_head); 11269 11270 net->dev_name_head = netdev_create_hash(); 11271 if (net->dev_name_head == NULL) 11272 goto err_name; 11273 11274 net->dev_index_head = netdev_create_hash(); 11275 if (net->dev_index_head == NULL) 11276 goto err_idx; 11277 11278 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11279 11280 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11281 11282 return 0; 11283 11284 err_idx: 11285 kfree(net->dev_name_head); 11286 err_name: 11287 return -ENOMEM; 11288 } 11289 11290 /** 11291 * netdev_drivername - network driver for the device 11292 * @dev: network device 11293 * 11294 * Determine network driver for device. 11295 */ 11296 const char *netdev_drivername(const struct net_device *dev) 11297 { 11298 const struct device_driver *driver; 11299 const struct device *parent; 11300 const char *empty = ""; 11301 11302 parent = dev->dev.parent; 11303 if (!parent) 11304 return empty; 11305 11306 driver = parent->driver; 11307 if (driver && driver->name) 11308 return driver->name; 11309 return empty; 11310 } 11311 11312 static void __netdev_printk(const char *level, const struct net_device *dev, 11313 struct va_format *vaf) 11314 { 11315 if (dev && dev->dev.parent) { 11316 dev_printk_emit(level[1] - '0', 11317 dev->dev.parent, 11318 "%s %s %s%s: %pV", 11319 dev_driver_string(dev->dev.parent), 11320 dev_name(dev->dev.parent), 11321 netdev_name(dev), netdev_reg_state(dev), 11322 vaf); 11323 } else if (dev) { 11324 printk("%s%s%s: %pV", 11325 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11326 } else { 11327 printk("%s(NULL net_device): %pV", level, vaf); 11328 } 11329 } 11330 11331 void netdev_printk(const char *level, const struct net_device *dev, 11332 const char *format, ...) 11333 { 11334 struct va_format vaf; 11335 va_list args; 11336 11337 va_start(args, format); 11338 11339 vaf.fmt = format; 11340 vaf.va = &args; 11341 11342 __netdev_printk(level, dev, &vaf); 11343 11344 va_end(args); 11345 } 11346 EXPORT_SYMBOL(netdev_printk); 11347 11348 #define define_netdev_printk_level(func, level) \ 11349 void func(const struct net_device *dev, const char *fmt, ...) \ 11350 { \ 11351 struct va_format vaf; \ 11352 va_list args; \ 11353 \ 11354 va_start(args, fmt); \ 11355 \ 11356 vaf.fmt = fmt; \ 11357 vaf.va = &args; \ 11358 \ 11359 __netdev_printk(level, dev, &vaf); \ 11360 \ 11361 va_end(args); \ 11362 } \ 11363 EXPORT_SYMBOL(func); 11364 11365 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11366 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11367 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11368 define_netdev_printk_level(netdev_err, KERN_ERR); 11369 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11370 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11371 define_netdev_printk_level(netdev_info, KERN_INFO); 11372 11373 static void __net_exit netdev_exit(struct net *net) 11374 { 11375 kfree(net->dev_name_head); 11376 kfree(net->dev_index_head); 11377 xa_destroy(&net->dev_by_index); 11378 if (net != &init_net) 11379 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11380 } 11381 11382 static struct pernet_operations __net_initdata netdev_net_ops = { 11383 .init = netdev_init, 11384 .exit = netdev_exit, 11385 }; 11386 11387 static void __net_exit default_device_exit_net(struct net *net) 11388 { 11389 struct net_device *dev, *aux; 11390 /* 11391 * Push all migratable network devices back to the 11392 * initial network namespace 11393 */ 11394 ASSERT_RTNL(); 11395 for_each_netdev_safe(net, dev, aux) { 11396 int err; 11397 char fb_name[IFNAMSIZ]; 11398 11399 /* Ignore unmoveable devices (i.e. loopback) */ 11400 if (dev->features & NETIF_F_NETNS_LOCAL) 11401 continue; 11402 11403 /* Leave virtual devices for the generic cleanup */ 11404 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11405 continue; 11406 11407 /* Push remaining network devices to init_net */ 11408 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11409 if (netdev_name_in_use(&init_net, fb_name)) 11410 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11411 err = dev_change_net_namespace(dev, &init_net, fb_name); 11412 if (err) { 11413 pr_emerg("%s: failed to move %s to init_net: %d\n", 11414 __func__, dev->name, err); 11415 BUG(); 11416 } 11417 } 11418 } 11419 11420 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11421 { 11422 /* At exit all network devices most be removed from a network 11423 * namespace. Do this in the reverse order of registration. 11424 * Do this across as many network namespaces as possible to 11425 * improve batching efficiency. 11426 */ 11427 struct net_device *dev; 11428 struct net *net; 11429 LIST_HEAD(dev_kill_list); 11430 11431 rtnl_lock(); 11432 list_for_each_entry(net, net_list, exit_list) { 11433 default_device_exit_net(net); 11434 cond_resched(); 11435 } 11436 11437 list_for_each_entry(net, net_list, exit_list) { 11438 for_each_netdev_reverse(net, dev) { 11439 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11440 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11441 else 11442 unregister_netdevice_queue(dev, &dev_kill_list); 11443 } 11444 } 11445 unregister_netdevice_many(&dev_kill_list); 11446 rtnl_unlock(); 11447 } 11448 11449 static struct pernet_operations __net_initdata default_device_ops = { 11450 .exit_batch = default_device_exit_batch, 11451 }; 11452 11453 /* 11454 * Initialize the DEV module. At boot time this walks the device list and 11455 * unhooks any devices that fail to initialise (normally hardware not 11456 * present) and leaves us with a valid list of present and active devices. 11457 * 11458 */ 11459 11460 /* 11461 * This is called single threaded during boot, so no need 11462 * to take the rtnl semaphore. 11463 */ 11464 static int __init net_dev_init(void) 11465 { 11466 int i, rc = -ENOMEM; 11467 11468 BUG_ON(!dev_boot_phase); 11469 11470 if (dev_proc_init()) 11471 goto out; 11472 11473 if (netdev_kobject_init()) 11474 goto out; 11475 11476 INIT_LIST_HEAD(&ptype_all); 11477 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11478 INIT_LIST_HEAD(&ptype_base[i]); 11479 11480 if (register_pernet_subsys(&netdev_net_ops)) 11481 goto out; 11482 11483 /* 11484 * Initialise the packet receive queues. 11485 */ 11486 11487 for_each_possible_cpu(i) { 11488 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11489 struct softnet_data *sd = &per_cpu(softnet_data, i); 11490 11491 INIT_WORK(flush, flush_backlog); 11492 11493 skb_queue_head_init(&sd->input_pkt_queue); 11494 skb_queue_head_init(&sd->process_queue); 11495 #ifdef CONFIG_XFRM_OFFLOAD 11496 skb_queue_head_init(&sd->xfrm_backlog); 11497 #endif 11498 INIT_LIST_HEAD(&sd->poll_list); 11499 sd->output_queue_tailp = &sd->output_queue; 11500 #ifdef CONFIG_RPS 11501 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11502 sd->cpu = i; 11503 #endif 11504 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11505 spin_lock_init(&sd->defer_lock); 11506 11507 init_gro_hash(&sd->backlog); 11508 sd->backlog.poll = process_backlog; 11509 sd->backlog.weight = weight_p; 11510 } 11511 11512 dev_boot_phase = 0; 11513 11514 /* The loopback device is special if any other network devices 11515 * is present in a network namespace the loopback device must 11516 * be present. Since we now dynamically allocate and free the 11517 * loopback device ensure this invariant is maintained by 11518 * keeping the loopback device as the first device on the 11519 * list of network devices. Ensuring the loopback devices 11520 * is the first device that appears and the last network device 11521 * that disappears. 11522 */ 11523 if (register_pernet_device(&loopback_net_ops)) 11524 goto out; 11525 11526 if (register_pernet_device(&default_device_ops)) 11527 goto out; 11528 11529 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11530 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11531 11532 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11533 NULL, dev_cpu_dead); 11534 WARN_ON(rc < 0); 11535 rc = 0; 11536 out: 11537 return rc; 11538 } 11539 11540 subsys_initcall(net_dev_init); 11541