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/bitmap.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 = bitmap_zalloc(max_netdevices, 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 bitmap_free(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 consume_skb(skb); 4065 fallthrough; 4066 case TC_ACT_CONSUMED: 4067 *ret = NET_XMIT_SUCCESS; 4068 return NULL; 4069 } 4070 4071 return skb; 4072 } 4073 #else 4074 static __always_inline struct sk_buff * 4075 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4076 struct net_device *orig_dev, bool *another) 4077 { 4078 return skb; 4079 } 4080 4081 static __always_inline struct sk_buff * 4082 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4083 { 4084 return skb; 4085 } 4086 #endif /* CONFIG_NET_XGRESS */ 4087 4088 #ifdef CONFIG_XPS 4089 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4090 struct xps_dev_maps *dev_maps, unsigned int tci) 4091 { 4092 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4093 struct xps_map *map; 4094 int queue_index = -1; 4095 4096 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4097 return queue_index; 4098 4099 tci *= dev_maps->num_tc; 4100 tci += tc; 4101 4102 map = rcu_dereference(dev_maps->attr_map[tci]); 4103 if (map) { 4104 if (map->len == 1) 4105 queue_index = map->queues[0]; 4106 else 4107 queue_index = map->queues[reciprocal_scale( 4108 skb_get_hash(skb), map->len)]; 4109 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4110 queue_index = -1; 4111 } 4112 return queue_index; 4113 } 4114 #endif 4115 4116 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4117 struct sk_buff *skb) 4118 { 4119 #ifdef CONFIG_XPS 4120 struct xps_dev_maps *dev_maps; 4121 struct sock *sk = skb->sk; 4122 int queue_index = -1; 4123 4124 if (!static_key_false(&xps_needed)) 4125 return -1; 4126 4127 rcu_read_lock(); 4128 if (!static_key_false(&xps_rxqs_needed)) 4129 goto get_cpus_map; 4130 4131 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4132 if (dev_maps) { 4133 int tci = sk_rx_queue_get(sk); 4134 4135 if (tci >= 0) 4136 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4137 tci); 4138 } 4139 4140 get_cpus_map: 4141 if (queue_index < 0) { 4142 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4143 if (dev_maps) { 4144 unsigned int tci = skb->sender_cpu - 1; 4145 4146 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4147 tci); 4148 } 4149 } 4150 rcu_read_unlock(); 4151 4152 return queue_index; 4153 #else 4154 return -1; 4155 #endif 4156 } 4157 4158 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4159 struct net_device *sb_dev) 4160 { 4161 return 0; 4162 } 4163 EXPORT_SYMBOL(dev_pick_tx_zero); 4164 4165 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4166 struct net_device *sb_dev) 4167 { 4168 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4169 } 4170 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4171 4172 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4173 struct net_device *sb_dev) 4174 { 4175 struct sock *sk = skb->sk; 4176 int queue_index = sk_tx_queue_get(sk); 4177 4178 sb_dev = sb_dev ? : dev; 4179 4180 if (queue_index < 0 || skb->ooo_okay || 4181 queue_index >= dev->real_num_tx_queues) { 4182 int new_index = get_xps_queue(dev, sb_dev, skb); 4183 4184 if (new_index < 0) 4185 new_index = skb_tx_hash(dev, sb_dev, skb); 4186 4187 if (queue_index != new_index && sk && 4188 sk_fullsock(sk) && 4189 rcu_access_pointer(sk->sk_dst_cache)) 4190 sk_tx_queue_set(sk, new_index); 4191 4192 queue_index = new_index; 4193 } 4194 4195 return queue_index; 4196 } 4197 EXPORT_SYMBOL(netdev_pick_tx); 4198 4199 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4200 struct sk_buff *skb, 4201 struct net_device *sb_dev) 4202 { 4203 int queue_index = 0; 4204 4205 #ifdef CONFIG_XPS 4206 u32 sender_cpu = skb->sender_cpu - 1; 4207 4208 if (sender_cpu >= (u32)NR_CPUS) 4209 skb->sender_cpu = raw_smp_processor_id() + 1; 4210 #endif 4211 4212 if (dev->real_num_tx_queues != 1) { 4213 const struct net_device_ops *ops = dev->netdev_ops; 4214 4215 if (ops->ndo_select_queue) 4216 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4217 else 4218 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4219 4220 queue_index = netdev_cap_txqueue(dev, queue_index); 4221 } 4222 4223 skb_set_queue_mapping(skb, queue_index); 4224 return netdev_get_tx_queue(dev, queue_index); 4225 } 4226 4227 /** 4228 * __dev_queue_xmit() - transmit a buffer 4229 * @skb: buffer to transmit 4230 * @sb_dev: suboordinate device used for L2 forwarding offload 4231 * 4232 * Queue a buffer for transmission to a network device. The caller must 4233 * have set the device and priority and built the buffer before calling 4234 * this function. The function can be called from an interrupt. 4235 * 4236 * When calling this method, interrupts MUST be enabled. This is because 4237 * the BH enable code must have IRQs enabled so that it will not deadlock. 4238 * 4239 * Regardless of the return value, the skb is consumed, so it is currently 4240 * difficult to retry a send to this method. (You can bump the ref count 4241 * before sending to hold a reference for retry if you are careful.) 4242 * 4243 * Return: 4244 * * 0 - buffer successfully transmitted 4245 * * positive qdisc return code - NET_XMIT_DROP etc. 4246 * * negative errno - other errors 4247 */ 4248 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4249 { 4250 struct net_device *dev = skb->dev; 4251 struct netdev_queue *txq = NULL; 4252 struct Qdisc *q; 4253 int rc = -ENOMEM; 4254 bool again = false; 4255 4256 skb_reset_mac_header(skb); 4257 skb_assert_len(skb); 4258 4259 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4260 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4261 4262 /* Disable soft irqs for various locks below. Also 4263 * stops preemption for RCU. 4264 */ 4265 rcu_read_lock_bh(); 4266 4267 skb_update_prio(skb); 4268 4269 qdisc_pkt_len_init(skb); 4270 tcx_set_ingress(skb, false); 4271 #ifdef CONFIG_NET_EGRESS 4272 if (static_branch_unlikely(&egress_needed_key)) { 4273 if (nf_hook_egress_active()) { 4274 skb = nf_hook_egress(skb, &rc, dev); 4275 if (!skb) 4276 goto out; 4277 } 4278 4279 netdev_xmit_skip_txqueue(false); 4280 4281 nf_skip_egress(skb, true); 4282 skb = sch_handle_egress(skb, &rc, dev); 4283 if (!skb) 4284 goto out; 4285 nf_skip_egress(skb, false); 4286 4287 if (netdev_xmit_txqueue_skipped()) 4288 txq = netdev_tx_queue_mapping(dev, skb); 4289 } 4290 #endif 4291 /* If device/qdisc don't need skb->dst, release it right now while 4292 * its hot in this cpu cache. 4293 */ 4294 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4295 skb_dst_drop(skb); 4296 else 4297 skb_dst_force(skb); 4298 4299 if (!txq) 4300 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4301 4302 q = rcu_dereference_bh(txq->qdisc); 4303 4304 trace_net_dev_queue(skb); 4305 if (q->enqueue) { 4306 rc = __dev_xmit_skb(skb, q, dev, txq); 4307 goto out; 4308 } 4309 4310 /* The device has no queue. Common case for software devices: 4311 * loopback, all the sorts of tunnels... 4312 4313 * Really, it is unlikely that netif_tx_lock protection is necessary 4314 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4315 * counters.) 4316 * However, it is possible, that they rely on protection 4317 * made by us here. 4318 4319 * Check this and shot the lock. It is not prone from deadlocks. 4320 *Either shot noqueue qdisc, it is even simpler 8) 4321 */ 4322 if (dev->flags & IFF_UP) { 4323 int cpu = smp_processor_id(); /* ok because BHs are off */ 4324 4325 /* Other cpus might concurrently change txq->xmit_lock_owner 4326 * to -1 or to their cpu id, but not to our id. 4327 */ 4328 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4329 if (dev_xmit_recursion()) 4330 goto recursion_alert; 4331 4332 skb = validate_xmit_skb(skb, dev, &again); 4333 if (!skb) 4334 goto out; 4335 4336 HARD_TX_LOCK(dev, txq, cpu); 4337 4338 if (!netif_xmit_stopped(txq)) { 4339 dev_xmit_recursion_inc(); 4340 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4341 dev_xmit_recursion_dec(); 4342 if (dev_xmit_complete(rc)) { 4343 HARD_TX_UNLOCK(dev, txq); 4344 goto out; 4345 } 4346 } 4347 HARD_TX_UNLOCK(dev, txq); 4348 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4349 dev->name); 4350 } else { 4351 /* Recursion is detected! It is possible, 4352 * unfortunately 4353 */ 4354 recursion_alert: 4355 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4356 dev->name); 4357 } 4358 } 4359 4360 rc = -ENETDOWN; 4361 rcu_read_unlock_bh(); 4362 4363 dev_core_stats_tx_dropped_inc(dev); 4364 kfree_skb_list(skb); 4365 return rc; 4366 out: 4367 rcu_read_unlock_bh(); 4368 return rc; 4369 } 4370 EXPORT_SYMBOL(__dev_queue_xmit); 4371 4372 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4373 { 4374 struct net_device *dev = skb->dev; 4375 struct sk_buff *orig_skb = skb; 4376 struct netdev_queue *txq; 4377 int ret = NETDEV_TX_BUSY; 4378 bool again = false; 4379 4380 if (unlikely(!netif_running(dev) || 4381 !netif_carrier_ok(dev))) 4382 goto drop; 4383 4384 skb = validate_xmit_skb_list(skb, dev, &again); 4385 if (skb != orig_skb) 4386 goto drop; 4387 4388 skb_set_queue_mapping(skb, queue_id); 4389 txq = skb_get_tx_queue(dev, skb); 4390 4391 local_bh_disable(); 4392 4393 dev_xmit_recursion_inc(); 4394 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4395 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4396 ret = netdev_start_xmit(skb, dev, txq, false); 4397 HARD_TX_UNLOCK(dev, txq); 4398 dev_xmit_recursion_dec(); 4399 4400 local_bh_enable(); 4401 return ret; 4402 drop: 4403 dev_core_stats_tx_dropped_inc(dev); 4404 kfree_skb_list(skb); 4405 return NET_XMIT_DROP; 4406 } 4407 EXPORT_SYMBOL(__dev_direct_xmit); 4408 4409 /************************************************************************* 4410 * Receiver routines 4411 *************************************************************************/ 4412 4413 int netdev_max_backlog __read_mostly = 1000; 4414 EXPORT_SYMBOL(netdev_max_backlog); 4415 4416 int netdev_tstamp_prequeue __read_mostly = 1; 4417 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4418 int netdev_budget __read_mostly = 300; 4419 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4420 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4421 int weight_p __read_mostly = 64; /* old backlog weight */ 4422 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4423 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4424 int dev_rx_weight __read_mostly = 64; 4425 int dev_tx_weight __read_mostly = 64; 4426 4427 /* Called with irq disabled */ 4428 static inline void ____napi_schedule(struct softnet_data *sd, 4429 struct napi_struct *napi) 4430 { 4431 struct task_struct *thread; 4432 4433 lockdep_assert_irqs_disabled(); 4434 4435 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4436 /* Paired with smp_mb__before_atomic() in 4437 * napi_enable()/dev_set_threaded(). 4438 * Use READ_ONCE() to guarantee a complete 4439 * read on napi->thread. Only call 4440 * wake_up_process() when it's not NULL. 4441 */ 4442 thread = READ_ONCE(napi->thread); 4443 if (thread) { 4444 /* Avoid doing set_bit() if the thread is in 4445 * INTERRUPTIBLE state, cause napi_thread_wait() 4446 * makes sure to proceed with napi polling 4447 * if the thread is explicitly woken from here. 4448 */ 4449 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4450 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4451 wake_up_process(thread); 4452 return; 4453 } 4454 } 4455 4456 list_add_tail(&napi->poll_list, &sd->poll_list); 4457 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4458 /* If not called from net_rx_action() 4459 * we have to raise NET_RX_SOFTIRQ. 4460 */ 4461 if (!sd->in_net_rx_action) 4462 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4463 } 4464 4465 #ifdef CONFIG_RPS 4466 4467 /* One global table that all flow-based protocols share. */ 4468 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4469 EXPORT_SYMBOL(rps_sock_flow_table); 4470 u32 rps_cpu_mask __read_mostly; 4471 EXPORT_SYMBOL(rps_cpu_mask); 4472 4473 struct static_key_false rps_needed __read_mostly; 4474 EXPORT_SYMBOL(rps_needed); 4475 struct static_key_false rfs_needed __read_mostly; 4476 EXPORT_SYMBOL(rfs_needed); 4477 4478 static struct rps_dev_flow * 4479 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4480 struct rps_dev_flow *rflow, u16 next_cpu) 4481 { 4482 if (next_cpu < nr_cpu_ids) { 4483 #ifdef CONFIG_RFS_ACCEL 4484 struct netdev_rx_queue *rxqueue; 4485 struct rps_dev_flow_table *flow_table; 4486 struct rps_dev_flow *old_rflow; 4487 u32 flow_id; 4488 u16 rxq_index; 4489 int rc; 4490 4491 /* Should we steer this flow to a different hardware queue? */ 4492 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4493 !(dev->features & NETIF_F_NTUPLE)) 4494 goto out; 4495 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4496 if (rxq_index == skb_get_rx_queue(skb)) 4497 goto out; 4498 4499 rxqueue = dev->_rx + rxq_index; 4500 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4501 if (!flow_table) 4502 goto out; 4503 flow_id = skb_get_hash(skb) & flow_table->mask; 4504 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4505 rxq_index, flow_id); 4506 if (rc < 0) 4507 goto out; 4508 old_rflow = rflow; 4509 rflow = &flow_table->flows[flow_id]; 4510 rflow->filter = rc; 4511 if (old_rflow->filter == rflow->filter) 4512 old_rflow->filter = RPS_NO_FILTER; 4513 out: 4514 #endif 4515 rflow->last_qtail = 4516 per_cpu(softnet_data, next_cpu).input_queue_head; 4517 } 4518 4519 rflow->cpu = next_cpu; 4520 return rflow; 4521 } 4522 4523 /* 4524 * get_rps_cpu is called from netif_receive_skb and returns the target 4525 * CPU from the RPS map of the receiving queue for a given skb. 4526 * rcu_read_lock must be held on entry. 4527 */ 4528 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4529 struct rps_dev_flow **rflowp) 4530 { 4531 const struct rps_sock_flow_table *sock_flow_table; 4532 struct netdev_rx_queue *rxqueue = dev->_rx; 4533 struct rps_dev_flow_table *flow_table; 4534 struct rps_map *map; 4535 int cpu = -1; 4536 u32 tcpu; 4537 u32 hash; 4538 4539 if (skb_rx_queue_recorded(skb)) { 4540 u16 index = skb_get_rx_queue(skb); 4541 4542 if (unlikely(index >= dev->real_num_rx_queues)) { 4543 WARN_ONCE(dev->real_num_rx_queues > 1, 4544 "%s received packet on queue %u, but number " 4545 "of RX queues is %u\n", 4546 dev->name, index, dev->real_num_rx_queues); 4547 goto done; 4548 } 4549 rxqueue += index; 4550 } 4551 4552 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4553 4554 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4555 map = rcu_dereference(rxqueue->rps_map); 4556 if (!flow_table && !map) 4557 goto done; 4558 4559 skb_reset_network_header(skb); 4560 hash = skb_get_hash(skb); 4561 if (!hash) 4562 goto done; 4563 4564 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4565 if (flow_table && sock_flow_table) { 4566 struct rps_dev_flow *rflow; 4567 u32 next_cpu; 4568 u32 ident; 4569 4570 /* First check into global flow table if there is a match. 4571 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4572 */ 4573 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4574 if ((ident ^ hash) & ~rps_cpu_mask) 4575 goto try_rps; 4576 4577 next_cpu = ident & rps_cpu_mask; 4578 4579 /* OK, now we know there is a match, 4580 * we can look at the local (per receive queue) flow table 4581 */ 4582 rflow = &flow_table->flows[hash & flow_table->mask]; 4583 tcpu = rflow->cpu; 4584 4585 /* 4586 * If the desired CPU (where last recvmsg was done) is 4587 * different from current CPU (one in the rx-queue flow 4588 * table entry), switch if one of the following holds: 4589 * - Current CPU is unset (>= nr_cpu_ids). 4590 * - Current CPU is offline. 4591 * - The current CPU's queue tail has advanced beyond the 4592 * last packet that was enqueued using this table entry. 4593 * This guarantees that all previous packets for the flow 4594 * have been dequeued, thus preserving in order delivery. 4595 */ 4596 if (unlikely(tcpu != next_cpu) && 4597 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4598 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4599 rflow->last_qtail)) >= 0)) { 4600 tcpu = next_cpu; 4601 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4602 } 4603 4604 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4605 *rflowp = rflow; 4606 cpu = tcpu; 4607 goto done; 4608 } 4609 } 4610 4611 try_rps: 4612 4613 if (map) { 4614 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4615 if (cpu_online(tcpu)) { 4616 cpu = tcpu; 4617 goto done; 4618 } 4619 } 4620 4621 done: 4622 return cpu; 4623 } 4624 4625 #ifdef CONFIG_RFS_ACCEL 4626 4627 /** 4628 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4629 * @dev: Device on which the filter was set 4630 * @rxq_index: RX queue index 4631 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4632 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4633 * 4634 * Drivers that implement ndo_rx_flow_steer() should periodically call 4635 * this function for each installed filter and remove the filters for 4636 * which it returns %true. 4637 */ 4638 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4639 u32 flow_id, u16 filter_id) 4640 { 4641 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4642 struct rps_dev_flow_table *flow_table; 4643 struct rps_dev_flow *rflow; 4644 bool expire = true; 4645 unsigned int cpu; 4646 4647 rcu_read_lock(); 4648 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4649 if (flow_table && flow_id <= flow_table->mask) { 4650 rflow = &flow_table->flows[flow_id]; 4651 cpu = READ_ONCE(rflow->cpu); 4652 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4653 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4654 rflow->last_qtail) < 4655 (int)(10 * flow_table->mask))) 4656 expire = false; 4657 } 4658 rcu_read_unlock(); 4659 return expire; 4660 } 4661 EXPORT_SYMBOL(rps_may_expire_flow); 4662 4663 #endif /* CONFIG_RFS_ACCEL */ 4664 4665 /* Called from hardirq (IPI) context */ 4666 static void rps_trigger_softirq(void *data) 4667 { 4668 struct softnet_data *sd = data; 4669 4670 ____napi_schedule(sd, &sd->backlog); 4671 sd->received_rps++; 4672 } 4673 4674 #endif /* CONFIG_RPS */ 4675 4676 /* Called from hardirq (IPI) context */ 4677 static void trigger_rx_softirq(void *data) 4678 { 4679 struct softnet_data *sd = data; 4680 4681 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4682 smp_store_release(&sd->defer_ipi_scheduled, 0); 4683 } 4684 4685 /* 4686 * After we queued a packet into sd->input_pkt_queue, 4687 * we need to make sure this queue is serviced soon. 4688 * 4689 * - If this is another cpu queue, link it to our rps_ipi_list, 4690 * and make sure we will process rps_ipi_list from net_rx_action(). 4691 * 4692 * - If this is our own queue, NAPI schedule our backlog. 4693 * Note that this also raises NET_RX_SOFTIRQ. 4694 */ 4695 static void napi_schedule_rps(struct softnet_data *sd) 4696 { 4697 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4698 4699 #ifdef CONFIG_RPS 4700 if (sd != mysd) { 4701 sd->rps_ipi_next = mysd->rps_ipi_list; 4702 mysd->rps_ipi_list = sd; 4703 4704 /* If not called from net_rx_action() or napi_threaded_poll() 4705 * we have to raise NET_RX_SOFTIRQ. 4706 */ 4707 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4708 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4709 return; 4710 } 4711 #endif /* CONFIG_RPS */ 4712 __napi_schedule_irqoff(&mysd->backlog); 4713 } 4714 4715 #ifdef CONFIG_NET_FLOW_LIMIT 4716 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4717 #endif 4718 4719 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4720 { 4721 #ifdef CONFIG_NET_FLOW_LIMIT 4722 struct sd_flow_limit *fl; 4723 struct softnet_data *sd; 4724 unsigned int old_flow, new_flow; 4725 4726 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4727 return false; 4728 4729 sd = this_cpu_ptr(&softnet_data); 4730 4731 rcu_read_lock(); 4732 fl = rcu_dereference(sd->flow_limit); 4733 if (fl) { 4734 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4735 old_flow = fl->history[fl->history_head]; 4736 fl->history[fl->history_head] = new_flow; 4737 4738 fl->history_head++; 4739 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4740 4741 if (likely(fl->buckets[old_flow])) 4742 fl->buckets[old_flow]--; 4743 4744 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4745 fl->count++; 4746 rcu_read_unlock(); 4747 return true; 4748 } 4749 } 4750 rcu_read_unlock(); 4751 #endif 4752 return false; 4753 } 4754 4755 /* 4756 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4757 * queue (may be a remote CPU queue). 4758 */ 4759 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4760 unsigned int *qtail) 4761 { 4762 enum skb_drop_reason reason; 4763 struct softnet_data *sd; 4764 unsigned long flags; 4765 unsigned int qlen; 4766 4767 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4768 sd = &per_cpu(softnet_data, cpu); 4769 4770 rps_lock_irqsave(sd, &flags); 4771 if (!netif_running(skb->dev)) 4772 goto drop; 4773 qlen = skb_queue_len(&sd->input_pkt_queue); 4774 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4775 if (qlen) { 4776 enqueue: 4777 __skb_queue_tail(&sd->input_pkt_queue, skb); 4778 input_queue_tail_incr_save(sd, qtail); 4779 rps_unlock_irq_restore(sd, &flags); 4780 return NET_RX_SUCCESS; 4781 } 4782 4783 /* Schedule NAPI for backlog device 4784 * We can use non atomic operation since we own the queue lock 4785 */ 4786 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4787 napi_schedule_rps(sd); 4788 goto enqueue; 4789 } 4790 reason = SKB_DROP_REASON_CPU_BACKLOG; 4791 4792 drop: 4793 sd->dropped++; 4794 rps_unlock_irq_restore(sd, &flags); 4795 4796 dev_core_stats_rx_dropped_inc(skb->dev); 4797 kfree_skb_reason(skb, reason); 4798 return NET_RX_DROP; 4799 } 4800 4801 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4802 { 4803 struct net_device *dev = skb->dev; 4804 struct netdev_rx_queue *rxqueue; 4805 4806 rxqueue = dev->_rx; 4807 4808 if (skb_rx_queue_recorded(skb)) { 4809 u16 index = skb_get_rx_queue(skb); 4810 4811 if (unlikely(index >= dev->real_num_rx_queues)) { 4812 WARN_ONCE(dev->real_num_rx_queues > 1, 4813 "%s received packet on queue %u, but number " 4814 "of RX queues is %u\n", 4815 dev->name, index, dev->real_num_rx_queues); 4816 4817 return rxqueue; /* Return first rxqueue */ 4818 } 4819 rxqueue += index; 4820 } 4821 return rxqueue; 4822 } 4823 4824 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4825 struct bpf_prog *xdp_prog) 4826 { 4827 void *orig_data, *orig_data_end, *hard_start; 4828 struct netdev_rx_queue *rxqueue; 4829 bool orig_bcast, orig_host; 4830 u32 mac_len, frame_sz; 4831 __be16 orig_eth_type; 4832 struct ethhdr *eth; 4833 u32 metalen, act; 4834 int off; 4835 4836 /* The XDP program wants to see the packet starting at the MAC 4837 * header. 4838 */ 4839 mac_len = skb->data - skb_mac_header(skb); 4840 hard_start = skb->data - skb_headroom(skb); 4841 4842 /* SKB "head" area always have tailroom for skb_shared_info */ 4843 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4844 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4845 4846 rxqueue = netif_get_rxqueue(skb); 4847 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4848 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4849 skb_headlen(skb) + mac_len, true); 4850 4851 orig_data_end = xdp->data_end; 4852 orig_data = xdp->data; 4853 eth = (struct ethhdr *)xdp->data; 4854 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4855 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4856 orig_eth_type = eth->h_proto; 4857 4858 act = bpf_prog_run_xdp(xdp_prog, xdp); 4859 4860 /* check if bpf_xdp_adjust_head was used */ 4861 off = xdp->data - orig_data; 4862 if (off) { 4863 if (off > 0) 4864 __skb_pull(skb, off); 4865 else if (off < 0) 4866 __skb_push(skb, -off); 4867 4868 skb->mac_header += off; 4869 skb_reset_network_header(skb); 4870 } 4871 4872 /* check if bpf_xdp_adjust_tail was used */ 4873 off = xdp->data_end - orig_data_end; 4874 if (off != 0) { 4875 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4876 skb->len += off; /* positive on grow, negative on shrink */ 4877 } 4878 4879 /* check if XDP changed eth hdr such SKB needs update */ 4880 eth = (struct ethhdr *)xdp->data; 4881 if ((orig_eth_type != eth->h_proto) || 4882 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4883 skb->dev->dev_addr)) || 4884 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4885 __skb_push(skb, ETH_HLEN); 4886 skb->pkt_type = PACKET_HOST; 4887 skb->protocol = eth_type_trans(skb, skb->dev); 4888 } 4889 4890 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4891 * before calling us again on redirect path. We do not call do_redirect 4892 * as we leave that up to the caller. 4893 * 4894 * Caller is responsible for managing lifetime of skb (i.e. calling 4895 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4896 */ 4897 switch (act) { 4898 case XDP_REDIRECT: 4899 case XDP_TX: 4900 __skb_push(skb, mac_len); 4901 break; 4902 case XDP_PASS: 4903 metalen = xdp->data - xdp->data_meta; 4904 if (metalen) 4905 skb_metadata_set(skb, metalen); 4906 break; 4907 } 4908 4909 return act; 4910 } 4911 4912 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4913 struct xdp_buff *xdp, 4914 struct bpf_prog *xdp_prog) 4915 { 4916 u32 act = XDP_DROP; 4917 4918 /* Reinjected packets coming from act_mirred or similar should 4919 * not get XDP generic processing. 4920 */ 4921 if (skb_is_redirected(skb)) 4922 return XDP_PASS; 4923 4924 /* XDP packets must be linear and must have sufficient headroom 4925 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4926 * native XDP provides, thus we need to do it here as well. 4927 */ 4928 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4929 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4930 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4931 int troom = skb->tail + skb->data_len - skb->end; 4932 4933 /* In case we have to go down the path and also linearize, 4934 * then lets do the pskb_expand_head() work just once here. 4935 */ 4936 if (pskb_expand_head(skb, 4937 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4938 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4939 goto do_drop; 4940 if (skb_linearize(skb)) 4941 goto do_drop; 4942 } 4943 4944 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4945 switch (act) { 4946 case XDP_REDIRECT: 4947 case XDP_TX: 4948 case XDP_PASS: 4949 break; 4950 default: 4951 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4952 fallthrough; 4953 case XDP_ABORTED: 4954 trace_xdp_exception(skb->dev, xdp_prog, act); 4955 fallthrough; 4956 case XDP_DROP: 4957 do_drop: 4958 kfree_skb(skb); 4959 break; 4960 } 4961 4962 return act; 4963 } 4964 4965 /* When doing generic XDP we have to bypass the qdisc layer and the 4966 * network taps in order to match in-driver-XDP behavior. This also means 4967 * that XDP packets are able to starve other packets going through a qdisc, 4968 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4969 * queues, so they do not have this starvation issue. 4970 */ 4971 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4972 { 4973 struct net_device *dev = skb->dev; 4974 struct netdev_queue *txq; 4975 bool free_skb = true; 4976 int cpu, rc; 4977 4978 txq = netdev_core_pick_tx(dev, skb, NULL); 4979 cpu = smp_processor_id(); 4980 HARD_TX_LOCK(dev, txq, cpu); 4981 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4982 rc = netdev_start_xmit(skb, dev, txq, 0); 4983 if (dev_xmit_complete(rc)) 4984 free_skb = false; 4985 } 4986 HARD_TX_UNLOCK(dev, txq); 4987 if (free_skb) { 4988 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4989 dev_core_stats_tx_dropped_inc(dev); 4990 kfree_skb(skb); 4991 } 4992 } 4993 4994 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4995 4996 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4997 { 4998 if (xdp_prog) { 4999 struct xdp_buff xdp; 5000 u32 act; 5001 int err; 5002 5003 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5004 if (act != XDP_PASS) { 5005 switch (act) { 5006 case XDP_REDIRECT: 5007 err = xdp_do_generic_redirect(skb->dev, skb, 5008 &xdp, xdp_prog); 5009 if (err) 5010 goto out_redir; 5011 break; 5012 case XDP_TX: 5013 generic_xdp_tx(skb, xdp_prog); 5014 break; 5015 } 5016 return XDP_DROP; 5017 } 5018 } 5019 return XDP_PASS; 5020 out_redir: 5021 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5022 return XDP_DROP; 5023 } 5024 EXPORT_SYMBOL_GPL(do_xdp_generic); 5025 5026 static int netif_rx_internal(struct sk_buff *skb) 5027 { 5028 int ret; 5029 5030 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5031 5032 trace_netif_rx(skb); 5033 5034 #ifdef CONFIG_RPS 5035 if (static_branch_unlikely(&rps_needed)) { 5036 struct rps_dev_flow voidflow, *rflow = &voidflow; 5037 int cpu; 5038 5039 rcu_read_lock(); 5040 5041 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5042 if (cpu < 0) 5043 cpu = smp_processor_id(); 5044 5045 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5046 5047 rcu_read_unlock(); 5048 } else 5049 #endif 5050 { 5051 unsigned int qtail; 5052 5053 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5054 } 5055 return ret; 5056 } 5057 5058 /** 5059 * __netif_rx - Slightly optimized version of netif_rx 5060 * @skb: buffer to post 5061 * 5062 * This behaves as netif_rx except that it does not disable bottom halves. 5063 * As a result this function may only be invoked from the interrupt context 5064 * (either hard or soft interrupt). 5065 */ 5066 int __netif_rx(struct sk_buff *skb) 5067 { 5068 int ret; 5069 5070 lockdep_assert_once(hardirq_count() | softirq_count()); 5071 5072 trace_netif_rx_entry(skb); 5073 ret = netif_rx_internal(skb); 5074 trace_netif_rx_exit(ret); 5075 return ret; 5076 } 5077 EXPORT_SYMBOL(__netif_rx); 5078 5079 /** 5080 * netif_rx - post buffer to the network code 5081 * @skb: buffer to post 5082 * 5083 * This function receives a packet from a device driver and queues it for 5084 * the upper (protocol) levels to process via the backlog NAPI device. It 5085 * always succeeds. The buffer may be dropped during processing for 5086 * congestion control or by the protocol layers. 5087 * The network buffer is passed via the backlog NAPI device. Modern NIC 5088 * driver should use NAPI and GRO. 5089 * This function can used from interrupt and from process context. The 5090 * caller from process context must not disable interrupts before invoking 5091 * this function. 5092 * 5093 * return values: 5094 * NET_RX_SUCCESS (no congestion) 5095 * NET_RX_DROP (packet was dropped) 5096 * 5097 */ 5098 int netif_rx(struct sk_buff *skb) 5099 { 5100 bool need_bh_off = !(hardirq_count() | softirq_count()); 5101 int ret; 5102 5103 if (need_bh_off) 5104 local_bh_disable(); 5105 trace_netif_rx_entry(skb); 5106 ret = netif_rx_internal(skb); 5107 trace_netif_rx_exit(ret); 5108 if (need_bh_off) 5109 local_bh_enable(); 5110 return ret; 5111 } 5112 EXPORT_SYMBOL(netif_rx); 5113 5114 static __latent_entropy void net_tx_action(struct softirq_action *h) 5115 { 5116 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5117 5118 if (sd->completion_queue) { 5119 struct sk_buff *clist; 5120 5121 local_irq_disable(); 5122 clist = sd->completion_queue; 5123 sd->completion_queue = NULL; 5124 local_irq_enable(); 5125 5126 while (clist) { 5127 struct sk_buff *skb = clist; 5128 5129 clist = clist->next; 5130 5131 WARN_ON(refcount_read(&skb->users)); 5132 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5133 trace_consume_skb(skb, net_tx_action); 5134 else 5135 trace_kfree_skb(skb, net_tx_action, 5136 get_kfree_skb_cb(skb)->reason); 5137 5138 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5139 __kfree_skb(skb); 5140 else 5141 __napi_kfree_skb(skb, 5142 get_kfree_skb_cb(skb)->reason); 5143 } 5144 } 5145 5146 if (sd->output_queue) { 5147 struct Qdisc *head; 5148 5149 local_irq_disable(); 5150 head = sd->output_queue; 5151 sd->output_queue = NULL; 5152 sd->output_queue_tailp = &sd->output_queue; 5153 local_irq_enable(); 5154 5155 rcu_read_lock(); 5156 5157 while (head) { 5158 struct Qdisc *q = head; 5159 spinlock_t *root_lock = NULL; 5160 5161 head = head->next_sched; 5162 5163 /* We need to make sure head->next_sched is read 5164 * before clearing __QDISC_STATE_SCHED 5165 */ 5166 smp_mb__before_atomic(); 5167 5168 if (!(q->flags & TCQ_F_NOLOCK)) { 5169 root_lock = qdisc_lock(q); 5170 spin_lock(root_lock); 5171 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5172 &q->state))) { 5173 /* There is a synchronize_net() between 5174 * STATE_DEACTIVATED flag being set and 5175 * qdisc_reset()/some_qdisc_is_busy() in 5176 * dev_deactivate(), so we can safely bail out 5177 * early here to avoid data race between 5178 * qdisc_deactivate() and some_qdisc_is_busy() 5179 * for lockless qdisc. 5180 */ 5181 clear_bit(__QDISC_STATE_SCHED, &q->state); 5182 continue; 5183 } 5184 5185 clear_bit(__QDISC_STATE_SCHED, &q->state); 5186 qdisc_run(q); 5187 if (root_lock) 5188 spin_unlock(root_lock); 5189 } 5190 5191 rcu_read_unlock(); 5192 } 5193 5194 xfrm_dev_backlog(sd); 5195 } 5196 5197 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5198 /* This hook is defined here for ATM LANE */ 5199 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5200 unsigned char *addr) __read_mostly; 5201 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5202 #endif 5203 5204 /** 5205 * netdev_is_rx_handler_busy - check if receive handler is registered 5206 * @dev: device to check 5207 * 5208 * Check if a receive handler is already registered for a given device. 5209 * Return true if there one. 5210 * 5211 * The caller must hold the rtnl_mutex. 5212 */ 5213 bool netdev_is_rx_handler_busy(struct net_device *dev) 5214 { 5215 ASSERT_RTNL(); 5216 return dev && rtnl_dereference(dev->rx_handler); 5217 } 5218 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5219 5220 /** 5221 * netdev_rx_handler_register - register receive handler 5222 * @dev: device to register a handler for 5223 * @rx_handler: receive handler to register 5224 * @rx_handler_data: data pointer that is used by rx handler 5225 * 5226 * Register a receive handler for a device. This handler will then be 5227 * called from __netif_receive_skb. A negative errno code is returned 5228 * on a failure. 5229 * 5230 * The caller must hold the rtnl_mutex. 5231 * 5232 * For a general description of rx_handler, see enum rx_handler_result. 5233 */ 5234 int netdev_rx_handler_register(struct net_device *dev, 5235 rx_handler_func_t *rx_handler, 5236 void *rx_handler_data) 5237 { 5238 if (netdev_is_rx_handler_busy(dev)) 5239 return -EBUSY; 5240 5241 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5242 return -EINVAL; 5243 5244 /* Note: rx_handler_data must be set before rx_handler */ 5245 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5246 rcu_assign_pointer(dev->rx_handler, rx_handler); 5247 5248 return 0; 5249 } 5250 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5251 5252 /** 5253 * netdev_rx_handler_unregister - unregister receive handler 5254 * @dev: device to unregister a handler from 5255 * 5256 * Unregister a receive handler from a device. 5257 * 5258 * The caller must hold the rtnl_mutex. 5259 */ 5260 void netdev_rx_handler_unregister(struct net_device *dev) 5261 { 5262 5263 ASSERT_RTNL(); 5264 RCU_INIT_POINTER(dev->rx_handler, NULL); 5265 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5266 * section has a guarantee to see a non NULL rx_handler_data 5267 * as well. 5268 */ 5269 synchronize_net(); 5270 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5271 } 5272 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5273 5274 /* 5275 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5276 * the special handling of PFMEMALLOC skbs. 5277 */ 5278 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5279 { 5280 switch (skb->protocol) { 5281 case htons(ETH_P_ARP): 5282 case htons(ETH_P_IP): 5283 case htons(ETH_P_IPV6): 5284 case htons(ETH_P_8021Q): 5285 case htons(ETH_P_8021AD): 5286 return true; 5287 default: 5288 return false; 5289 } 5290 } 5291 5292 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5293 int *ret, struct net_device *orig_dev) 5294 { 5295 if (nf_hook_ingress_active(skb)) { 5296 int ingress_retval; 5297 5298 if (*pt_prev) { 5299 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5300 *pt_prev = NULL; 5301 } 5302 5303 rcu_read_lock(); 5304 ingress_retval = nf_hook_ingress(skb); 5305 rcu_read_unlock(); 5306 return ingress_retval; 5307 } 5308 return 0; 5309 } 5310 5311 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5312 struct packet_type **ppt_prev) 5313 { 5314 struct packet_type *ptype, *pt_prev; 5315 rx_handler_func_t *rx_handler; 5316 struct sk_buff *skb = *pskb; 5317 struct net_device *orig_dev; 5318 bool deliver_exact = false; 5319 int ret = NET_RX_DROP; 5320 __be16 type; 5321 5322 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5323 5324 trace_netif_receive_skb(skb); 5325 5326 orig_dev = skb->dev; 5327 5328 skb_reset_network_header(skb); 5329 if (!skb_transport_header_was_set(skb)) 5330 skb_reset_transport_header(skb); 5331 skb_reset_mac_len(skb); 5332 5333 pt_prev = NULL; 5334 5335 another_round: 5336 skb->skb_iif = skb->dev->ifindex; 5337 5338 __this_cpu_inc(softnet_data.processed); 5339 5340 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5341 int ret2; 5342 5343 migrate_disable(); 5344 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5345 migrate_enable(); 5346 5347 if (ret2 != XDP_PASS) { 5348 ret = NET_RX_DROP; 5349 goto out; 5350 } 5351 } 5352 5353 if (eth_type_vlan(skb->protocol)) { 5354 skb = skb_vlan_untag(skb); 5355 if (unlikely(!skb)) 5356 goto out; 5357 } 5358 5359 if (skb_skip_tc_classify(skb)) 5360 goto skip_classify; 5361 5362 if (pfmemalloc) 5363 goto skip_taps; 5364 5365 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5366 if (pt_prev) 5367 ret = deliver_skb(skb, pt_prev, orig_dev); 5368 pt_prev = ptype; 5369 } 5370 5371 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5372 if (pt_prev) 5373 ret = deliver_skb(skb, pt_prev, orig_dev); 5374 pt_prev = ptype; 5375 } 5376 5377 skip_taps: 5378 #ifdef CONFIG_NET_INGRESS 5379 if (static_branch_unlikely(&ingress_needed_key)) { 5380 bool another = false; 5381 5382 nf_skip_egress(skb, true); 5383 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5384 &another); 5385 if (another) 5386 goto another_round; 5387 if (!skb) 5388 goto out; 5389 5390 nf_skip_egress(skb, false); 5391 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5392 goto out; 5393 } 5394 #endif 5395 skb_reset_redirect(skb); 5396 skip_classify: 5397 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5398 goto drop; 5399 5400 if (skb_vlan_tag_present(skb)) { 5401 if (pt_prev) { 5402 ret = deliver_skb(skb, pt_prev, orig_dev); 5403 pt_prev = NULL; 5404 } 5405 if (vlan_do_receive(&skb)) 5406 goto another_round; 5407 else if (unlikely(!skb)) 5408 goto out; 5409 } 5410 5411 rx_handler = rcu_dereference(skb->dev->rx_handler); 5412 if (rx_handler) { 5413 if (pt_prev) { 5414 ret = deliver_skb(skb, pt_prev, orig_dev); 5415 pt_prev = NULL; 5416 } 5417 switch (rx_handler(&skb)) { 5418 case RX_HANDLER_CONSUMED: 5419 ret = NET_RX_SUCCESS; 5420 goto out; 5421 case RX_HANDLER_ANOTHER: 5422 goto another_round; 5423 case RX_HANDLER_EXACT: 5424 deliver_exact = true; 5425 break; 5426 case RX_HANDLER_PASS: 5427 break; 5428 default: 5429 BUG(); 5430 } 5431 } 5432 5433 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5434 check_vlan_id: 5435 if (skb_vlan_tag_get_id(skb)) { 5436 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5437 * find vlan device. 5438 */ 5439 skb->pkt_type = PACKET_OTHERHOST; 5440 } else if (eth_type_vlan(skb->protocol)) { 5441 /* Outer header is 802.1P with vlan 0, inner header is 5442 * 802.1Q or 802.1AD and vlan_do_receive() above could 5443 * not find vlan dev for vlan id 0. 5444 */ 5445 __vlan_hwaccel_clear_tag(skb); 5446 skb = skb_vlan_untag(skb); 5447 if (unlikely(!skb)) 5448 goto out; 5449 if (vlan_do_receive(&skb)) 5450 /* After stripping off 802.1P header with vlan 0 5451 * vlan dev is found for inner header. 5452 */ 5453 goto another_round; 5454 else if (unlikely(!skb)) 5455 goto out; 5456 else 5457 /* We have stripped outer 802.1P vlan 0 header. 5458 * But could not find vlan dev. 5459 * check again for vlan id to set OTHERHOST. 5460 */ 5461 goto check_vlan_id; 5462 } 5463 /* Note: we might in the future use prio bits 5464 * and set skb->priority like in vlan_do_receive() 5465 * For the time being, just ignore Priority Code Point 5466 */ 5467 __vlan_hwaccel_clear_tag(skb); 5468 } 5469 5470 type = skb->protocol; 5471 5472 /* deliver only exact match when indicated */ 5473 if (likely(!deliver_exact)) { 5474 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5475 &ptype_base[ntohs(type) & 5476 PTYPE_HASH_MASK]); 5477 } 5478 5479 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5480 &orig_dev->ptype_specific); 5481 5482 if (unlikely(skb->dev != orig_dev)) { 5483 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5484 &skb->dev->ptype_specific); 5485 } 5486 5487 if (pt_prev) { 5488 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5489 goto drop; 5490 *ppt_prev = pt_prev; 5491 } else { 5492 drop: 5493 if (!deliver_exact) 5494 dev_core_stats_rx_dropped_inc(skb->dev); 5495 else 5496 dev_core_stats_rx_nohandler_inc(skb->dev); 5497 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5498 /* Jamal, now you will not able to escape explaining 5499 * me how you were going to use this. :-) 5500 */ 5501 ret = NET_RX_DROP; 5502 } 5503 5504 out: 5505 /* The invariant here is that if *ppt_prev is not NULL 5506 * then skb should also be non-NULL. 5507 * 5508 * Apparently *ppt_prev assignment above holds this invariant due to 5509 * skb dereferencing near it. 5510 */ 5511 *pskb = skb; 5512 return ret; 5513 } 5514 5515 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5516 { 5517 struct net_device *orig_dev = skb->dev; 5518 struct packet_type *pt_prev = NULL; 5519 int ret; 5520 5521 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5522 if (pt_prev) 5523 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5524 skb->dev, pt_prev, orig_dev); 5525 return ret; 5526 } 5527 5528 /** 5529 * netif_receive_skb_core - special purpose version of netif_receive_skb 5530 * @skb: buffer to process 5531 * 5532 * More direct receive version of netif_receive_skb(). It should 5533 * only be used by callers that have a need to skip RPS and Generic XDP. 5534 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5535 * 5536 * This function may only be called from softirq context and interrupts 5537 * should be enabled. 5538 * 5539 * Return values (usually ignored): 5540 * NET_RX_SUCCESS: no congestion 5541 * NET_RX_DROP: packet was dropped 5542 */ 5543 int netif_receive_skb_core(struct sk_buff *skb) 5544 { 5545 int ret; 5546 5547 rcu_read_lock(); 5548 ret = __netif_receive_skb_one_core(skb, false); 5549 rcu_read_unlock(); 5550 5551 return ret; 5552 } 5553 EXPORT_SYMBOL(netif_receive_skb_core); 5554 5555 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5556 struct packet_type *pt_prev, 5557 struct net_device *orig_dev) 5558 { 5559 struct sk_buff *skb, *next; 5560 5561 if (!pt_prev) 5562 return; 5563 if (list_empty(head)) 5564 return; 5565 if (pt_prev->list_func != NULL) 5566 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5567 ip_list_rcv, head, pt_prev, orig_dev); 5568 else 5569 list_for_each_entry_safe(skb, next, head, list) { 5570 skb_list_del_init(skb); 5571 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5572 } 5573 } 5574 5575 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5576 { 5577 /* Fast-path assumptions: 5578 * - There is no RX handler. 5579 * - Only one packet_type matches. 5580 * If either of these fails, we will end up doing some per-packet 5581 * processing in-line, then handling the 'last ptype' for the whole 5582 * sublist. This can't cause out-of-order delivery to any single ptype, 5583 * because the 'last ptype' must be constant across the sublist, and all 5584 * other ptypes are handled per-packet. 5585 */ 5586 /* Current (common) ptype of sublist */ 5587 struct packet_type *pt_curr = NULL; 5588 /* Current (common) orig_dev of sublist */ 5589 struct net_device *od_curr = NULL; 5590 struct list_head sublist; 5591 struct sk_buff *skb, *next; 5592 5593 INIT_LIST_HEAD(&sublist); 5594 list_for_each_entry_safe(skb, next, head, list) { 5595 struct net_device *orig_dev = skb->dev; 5596 struct packet_type *pt_prev = NULL; 5597 5598 skb_list_del_init(skb); 5599 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5600 if (!pt_prev) 5601 continue; 5602 if (pt_curr != pt_prev || od_curr != orig_dev) { 5603 /* dispatch old sublist */ 5604 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5605 /* start new sublist */ 5606 INIT_LIST_HEAD(&sublist); 5607 pt_curr = pt_prev; 5608 od_curr = orig_dev; 5609 } 5610 list_add_tail(&skb->list, &sublist); 5611 } 5612 5613 /* dispatch final sublist */ 5614 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5615 } 5616 5617 static int __netif_receive_skb(struct sk_buff *skb) 5618 { 5619 int ret; 5620 5621 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5622 unsigned int noreclaim_flag; 5623 5624 /* 5625 * PFMEMALLOC skbs are special, they should 5626 * - be delivered to SOCK_MEMALLOC sockets only 5627 * - stay away from userspace 5628 * - have bounded memory usage 5629 * 5630 * Use PF_MEMALLOC as this saves us from propagating the allocation 5631 * context down to all allocation sites. 5632 */ 5633 noreclaim_flag = memalloc_noreclaim_save(); 5634 ret = __netif_receive_skb_one_core(skb, true); 5635 memalloc_noreclaim_restore(noreclaim_flag); 5636 } else 5637 ret = __netif_receive_skb_one_core(skb, false); 5638 5639 return ret; 5640 } 5641 5642 static void __netif_receive_skb_list(struct list_head *head) 5643 { 5644 unsigned long noreclaim_flag = 0; 5645 struct sk_buff *skb, *next; 5646 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5647 5648 list_for_each_entry_safe(skb, next, head, list) { 5649 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5650 struct list_head sublist; 5651 5652 /* Handle the previous sublist */ 5653 list_cut_before(&sublist, head, &skb->list); 5654 if (!list_empty(&sublist)) 5655 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5656 pfmemalloc = !pfmemalloc; 5657 /* See comments in __netif_receive_skb */ 5658 if (pfmemalloc) 5659 noreclaim_flag = memalloc_noreclaim_save(); 5660 else 5661 memalloc_noreclaim_restore(noreclaim_flag); 5662 } 5663 } 5664 /* Handle the remaining sublist */ 5665 if (!list_empty(head)) 5666 __netif_receive_skb_list_core(head, pfmemalloc); 5667 /* Restore pflags */ 5668 if (pfmemalloc) 5669 memalloc_noreclaim_restore(noreclaim_flag); 5670 } 5671 5672 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5673 { 5674 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5675 struct bpf_prog *new = xdp->prog; 5676 int ret = 0; 5677 5678 switch (xdp->command) { 5679 case XDP_SETUP_PROG: 5680 rcu_assign_pointer(dev->xdp_prog, new); 5681 if (old) 5682 bpf_prog_put(old); 5683 5684 if (old && !new) { 5685 static_branch_dec(&generic_xdp_needed_key); 5686 } else if (new && !old) { 5687 static_branch_inc(&generic_xdp_needed_key); 5688 dev_disable_lro(dev); 5689 dev_disable_gro_hw(dev); 5690 } 5691 break; 5692 5693 default: 5694 ret = -EINVAL; 5695 break; 5696 } 5697 5698 return ret; 5699 } 5700 5701 static int netif_receive_skb_internal(struct sk_buff *skb) 5702 { 5703 int ret; 5704 5705 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5706 5707 if (skb_defer_rx_timestamp(skb)) 5708 return NET_RX_SUCCESS; 5709 5710 rcu_read_lock(); 5711 #ifdef CONFIG_RPS 5712 if (static_branch_unlikely(&rps_needed)) { 5713 struct rps_dev_flow voidflow, *rflow = &voidflow; 5714 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5715 5716 if (cpu >= 0) { 5717 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5718 rcu_read_unlock(); 5719 return ret; 5720 } 5721 } 5722 #endif 5723 ret = __netif_receive_skb(skb); 5724 rcu_read_unlock(); 5725 return ret; 5726 } 5727 5728 void netif_receive_skb_list_internal(struct list_head *head) 5729 { 5730 struct sk_buff *skb, *next; 5731 struct list_head sublist; 5732 5733 INIT_LIST_HEAD(&sublist); 5734 list_for_each_entry_safe(skb, next, head, list) { 5735 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5736 skb_list_del_init(skb); 5737 if (!skb_defer_rx_timestamp(skb)) 5738 list_add_tail(&skb->list, &sublist); 5739 } 5740 list_splice_init(&sublist, head); 5741 5742 rcu_read_lock(); 5743 #ifdef CONFIG_RPS 5744 if (static_branch_unlikely(&rps_needed)) { 5745 list_for_each_entry_safe(skb, next, head, list) { 5746 struct rps_dev_flow voidflow, *rflow = &voidflow; 5747 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5748 5749 if (cpu >= 0) { 5750 /* Will be handled, remove from list */ 5751 skb_list_del_init(skb); 5752 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5753 } 5754 } 5755 } 5756 #endif 5757 __netif_receive_skb_list(head); 5758 rcu_read_unlock(); 5759 } 5760 5761 /** 5762 * netif_receive_skb - process receive buffer from network 5763 * @skb: buffer to process 5764 * 5765 * netif_receive_skb() is the main receive data processing function. 5766 * It always succeeds. The buffer may be dropped during processing 5767 * for congestion control or by the protocol layers. 5768 * 5769 * This function may only be called from softirq context and interrupts 5770 * should be enabled. 5771 * 5772 * Return values (usually ignored): 5773 * NET_RX_SUCCESS: no congestion 5774 * NET_RX_DROP: packet was dropped 5775 */ 5776 int netif_receive_skb(struct sk_buff *skb) 5777 { 5778 int ret; 5779 5780 trace_netif_receive_skb_entry(skb); 5781 5782 ret = netif_receive_skb_internal(skb); 5783 trace_netif_receive_skb_exit(ret); 5784 5785 return ret; 5786 } 5787 EXPORT_SYMBOL(netif_receive_skb); 5788 5789 /** 5790 * netif_receive_skb_list - process many receive buffers from network 5791 * @head: list of skbs to process. 5792 * 5793 * Since return value of netif_receive_skb() is normally ignored, and 5794 * wouldn't be meaningful for a list, this function returns void. 5795 * 5796 * This function may only be called from softirq context and interrupts 5797 * should be enabled. 5798 */ 5799 void netif_receive_skb_list(struct list_head *head) 5800 { 5801 struct sk_buff *skb; 5802 5803 if (list_empty(head)) 5804 return; 5805 if (trace_netif_receive_skb_list_entry_enabled()) { 5806 list_for_each_entry(skb, head, list) 5807 trace_netif_receive_skb_list_entry(skb); 5808 } 5809 netif_receive_skb_list_internal(head); 5810 trace_netif_receive_skb_list_exit(0); 5811 } 5812 EXPORT_SYMBOL(netif_receive_skb_list); 5813 5814 static DEFINE_PER_CPU(struct work_struct, flush_works); 5815 5816 /* Network device is going away, flush any packets still pending */ 5817 static void flush_backlog(struct work_struct *work) 5818 { 5819 struct sk_buff *skb, *tmp; 5820 struct softnet_data *sd; 5821 5822 local_bh_disable(); 5823 sd = this_cpu_ptr(&softnet_data); 5824 5825 rps_lock_irq_disable(sd); 5826 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5827 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5828 __skb_unlink(skb, &sd->input_pkt_queue); 5829 dev_kfree_skb_irq(skb); 5830 input_queue_head_incr(sd); 5831 } 5832 } 5833 rps_unlock_irq_enable(sd); 5834 5835 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5836 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5837 __skb_unlink(skb, &sd->process_queue); 5838 kfree_skb(skb); 5839 input_queue_head_incr(sd); 5840 } 5841 } 5842 local_bh_enable(); 5843 } 5844 5845 static bool flush_required(int cpu) 5846 { 5847 #if IS_ENABLED(CONFIG_RPS) 5848 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5849 bool do_flush; 5850 5851 rps_lock_irq_disable(sd); 5852 5853 /* as insertion into process_queue happens with the rps lock held, 5854 * process_queue access may race only with dequeue 5855 */ 5856 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5857 !skb_queue_empty_lockless(&sd->process_queue); 5858 rps_unlock_irq_enable(sd); 5859 5860 return do_flush; 5861 #endif 5862 /* without RPS we can't safely check input_pkt_queue: during a 5863 * concurrent remote skb_queue_splice() we can detect as empty both 5864 * input_pkt_queue and process_queue even if the latter could end-up 5865 * containing a lot of packets. 5866 */ 5867 return true; 5868 } 5869 5870 static void flush_all_backlogs(void) 5871 { 5872 static cpumask_t flush_cpus; 5873 unsigned int cpu; 5874 5875 /* since we are under rtnl lock protection we can use static data 5876 * for the cpumask and avoid allocating on stack the possibly 5877 * large mask 5878 */ 5879 ASSERT_RTNL(); 5880 5881 cpus_read_lock(); 5882 5883 cpumask_clear(&flush_cpus); 5884 for_each_online_cpu(cpu) { 5885 if (flush_required(cpu)) { 5886 queue_work_on(cpu, system_highpri_wq, 5887 per_cpu_ptr(&flush_works, cpu)); 5888 cpumask_set_cpu(cpu, &flush_cpus); 5889 } 5890 } 5891 5892 /* we can have in flight packet[s] on the cpus we are not flushing, 5893 * synchronize_net() in unregister_netdevice_many() will take care of 5894 * them 5895 */ 5896 for_each_cpu(cpu, &flush_cpus) 5897 flush_work(per_cpu_ptr(&flush_works, cpu)); 5898 5899 cpus_read_unlock(); 5900 } 5901 5902 static void net_rps_send_ipi(struct softnet_data *remsd) 5903 { 5904 #ifdef CONFIG_RPS 5905 while (remsd) { 5906 struct softnet_data *next = remsd->rps_ipi_next; 5907 5908 if (cpu_online(remsd->cpu)) 5909 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5910 remsd = next; 5911 } 5912 #endif 5913 } 5914 5915 /* 5916 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5917 * Note: called with local irq disabled, but exits with local irq enabled. 5918 */ 5919 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5920 { 5921 #ifdef CONFIG_RPS 5922 struct softnet_data *remsd = sd->rps_ipi_list; 5923 5924 if (remsd) { 5925 sd->rps_ipi_list = NULL; 5926 5927 local_irq_enable(); 5928 5929 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5930 net_rps_send_ipi(remsd); 5931 } else 5932 #endif 5933 local_irq_enable(); 5934 } 5935 5936 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5937 { 5938 #ifdef CONFIG_RPS 5939 return sd->rps_ipi_list != NULL; 5940 #else 5941 return false; 5942 #endif 5943 } 5944 5945 static int process_backlog(struct napi_struct *napi, int quota) 5946 { 5947 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5948 bool again = true; 5949 int work = 0; 5950 5951 /* Check if we have pending ipi, its better to send them now, 5952 * not waiting net_rx_action() end. 5953 */ 5954 if (sd_has_rps_ipi_waiting(sd)) { 5955 local_irq_disable(); 5956 net_rps_action_and_irq_enable(sd); 5957 } 5958 5959 napi->weight = READ_ONCE(dev_rx_weight); 5960 while (again) { 5961 struct sk_buff *skb; 5962 5963 while ((skb = __skb_dequeue(&sd->process_queue))) { 5964 rcu_read_lock(); 5965 __netif_receive_skb(skb); 5966 rcu_read_unlock(); 5967 input_queue_head_incr(sd); 5968 if (++work >= quota) 5969 return work; 5970 5971 } 5972 5973 rps_lock_irq_disable(sd); 5974 if (skb_queue_empty(&sd->input_pkt_queue)) { 5975 /* 5976 * Inline a custom version of __napi_complete(). 5977 * only current cpu owns and manipulates this napi, 5978 * and NAPI_STATE_SCHED is the only possible flag set 5979 * on backlog. 5980 * We can use a plain write instead of clear_bit(), 5981 * and we dont need an smp_mb() memory barrier. 5982 */ 5983 napi->state = 0; 5984 again = false; 5985 } else { 5986 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5987 &sd->process_queue); 5988 } 5989 rps_unlock_irq_enable(sd); 5990 } 5991 5992 return work; 5993 } 5994 5995 /** 5996 * __napi_schedule - schedule for receive 5997 * @n: entry to schedule 5998 * 5999 * The entry's receive function will be scheduled to run. 6000 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6001 */ 6002 void __napi_schedule(struct napi_struct *n) 6003 { 6004 unsigned long flags; 6005 6006 local_irq_save(flags); 6007 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6008 local_irq_restore(flags); 6009 } 6010 EXPORT_SYMBOL(__napi_schedule); 6011 6012 /** 6013 * napi_schedule_prep - check if napi can be scheduled 6014 * @n: napi context 6015 * 6016 * Test if NAPI routine is already running, and if not mark 6017 * it as running. This is used as a condition variable to 6018 * insure only one NAPI poll instance runs. We also make 6019 * sure there is no pending NAPI disable. 6020 */ 6021 bool napi_schedule_prep(struct napi_struct *n) 6022 { 6023 unsigned long new, val = READ_ONCE(n->state); 6024 6025 do { 6026 if (unlikely(val & NAPIF_STATE_DISABLE)) 6027 return false; 6028 new = val | NAPIF_STATE_SCHED; 6029 6030 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6031 * This was suggested by Alexander Duyck, as compiler 6032 * emits better code than : 6033 * if (val & NAPIF_STATE_SCHED) 6034 * new |= NAPIF_STATE_MISSED; 6035 */ 6036 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6037 NAPIF_STATE_MISSED; 6038 } while (!try_cmpxchg(&n->state, &val, new)); 6039 6040 return !(val & NAPIF_STATE_SCHED); 6041 } 6042 EXPORT_SYMBOL(napi_schedule_prep); 6043 6044 /** 6045 * __napi_schedule_irqoff - schedule for receive 6046 * @n: entry to schedule 6047 * 6048 * Variant of __napi_schedule() assuming hard irqs are masked. 6049 * 6050 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6051 * because the interrupt disabled assumption might not be true 6052 * due to force-threaded interrupts and spinlock substitution. 6053 */ 6054 void __napi_schedule_irqoff(struct napi_struct *n) 6055 { 6056 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6057 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6058 else 6059 __napi_schedule(n); 6060 } 6061 EXPORT_SYMBOL(__napi_schedule_irqoff); 6062 6063 bool napi_complete_done(struct napi_struct *n, int work_done) 6064 { 6065 unsigned long flags, val, new, timeout = 0; 6066 bool ret = true; 6067 6068 /* 6069 * 1) Don't let napi dequeue from the cpu poll list 6070 * just in case its running on a different cpu. 6071 * 2) If we are busy polling, do nothing here, we have 6072 * the guarantee we will be called later. 6073 */ 6074 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6075 NAPIF_STATE_IN_BUSY_POLL))) 6076 return false; 6077 6078 if (work_done) { 6079 if (n->gro_bitmask) 6080 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6081 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6082 } 6083 if (n->defer_hard_irqs_count > 0) { 6084 n->defer_hard_irqs_count--; 6085 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6086 if (timeout) 6087 ret = false; 6088 } 6089 if (n->gro_bitmask) { 6090 /* When the NAPI instance uses a timeout and keeps postponing 6091 * it, we need to bound somehow the time packets are kept in 6092 * the GRO layer 6093 */ 6094 napi_gro_flush(n, !!timeout); 6095 } 6096 6097 gro_normal_list(n); 6098 6099 if (unlikely(!list_empty(&n->poll_list))) { 6100 /* If n->poll_list is not empty, we need to mask irqs */ 6101 local_irq_save(flags); 6102 list_del_init(&n->poll_list); 6103 local_irq_restore(flags); 6104 } 6105 WRITE_ONCE(n->list_owner, -1); 6106 6107 val = READ_ONCE(n->state); 6108 do { 6109 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6110 6111 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6112 NAPIF_STATE_SCHED_THREADED | 6113 NAPIF_STATE_PREFER_BUSY_POLL); 6114 6115 /* If STATE_MISSED was set, leave STATE_SCHED set, 6116 * because we will call napi->poll() one more time. 6117 * This C code was suggested by Alexander Duyck to help gcc. 6118 */ 6119 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6120 NAPIF_STATE_SCHED; 6121 } while (!try_cmpxchg(&n->state, &val, new)); 6122 6123 if (unlikely(val & NAPIF_STATE_MISSED)) { 6124 __napi_schedule(n); 6125 return false; 6126 } 6127 6128 if (timeout) 6129 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6130 HRTIMER_MODE_REL_PINNED); 6131 return ret; 6132 } 6133 EXPORT_SYMBOL(napi_complete_done); 6134 6135 /* must be called under rcu_read_lock(), as we dont take a reference */ 6136 static struct napi_struct *napi_by_id(unsigned int napi_id) 6137 { 6138 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6139 struct napi_struct *napi; 6140 6141 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6142 if (napi->napi_id == napi_id) 6143 return napi; 6144 6145 return NULL; 6146 } 6147 6148 #if defined(CONFIG_NET_RX_BUSY_POLL) 6149 6150 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6151 { 6152 if (!skip_schedule) { 6153 gro_normal_list(napi); 6154 __napi_schedule(napi); 6155 return; 6156 } 6157 6158 if (napi->gro_bitmask) { 6159 /* flush too old packets 6160 * If HZ < 1000, flush all packets. 6161 */ 6162 napi_gro_flush(napi, HZ >= 1000); 6163 } 6164 6165 gro_normal_list(napi); 6166 clear_bit(NAPI_STATE_SCHED, &napi->state); 6167 } 6168 6169 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6170 u16 budget) 6171 { 6172 bool skip_schedule = false; 6173 unsigned long timeout; 6174 int rc; 6175 6176 /* Busy polling means there is a high chance device driver hard irq 6177 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6178 * set in napi_schedule_prep(). 6179 * Since we are about to call napi->poll() once more, we can safely 6180 * clear NAPI_STATE_MISSED. 6181 * 6182 * Note: x86 could use a single "lock and ..." instruction 6183 * to perform these two clear_bit() 6184 */ 6185 clear_bit(NAPI_STATE_MISSED, &napi->state); 6186 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6187 6188 local_bh_disable(); 6189 6190 if (prefer_busy_poll) { 6191 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6192 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6193 if (napi->defer_hard_irqs_count && timeout) { 6194 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6195 skip_schedule = true; 6196 } 6197 } 6198 6199 /* All we really want here is to re-enable device interrupts. 6200 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6201 */ 6202 rc = napi->poll(napi, budget); 6203 /* We can't gro_normal_list() here, because napi->poll() might have 6204 * rearmed the napi (napi_complete_done()) in which case it could 6205 * already be running on another CPU. 6206 */ 6207 trace_napi_poll(napi, rc, budget); 6208 netpoll_poll_unlock(have_poll_lock); 6209 if (rc == budget) 6210 __busy_poll_stop(napi, skip_schedule); 6211 local_bh_enable(); 6212 } 6213 6214 void napi_busy_loop(unsigned int napi_id, 6215 bool (*loop_end)(void *, unsigned long), 6216 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6217 { 6218 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6219 int (*napi_poll)(struct napi_struct *napi, int budget); 6220 void *have_poll_lock = NULL; 6221 struct napi_struct *napi; 6222 6223 restart: 6224 napi_poll = NULL; 6225 6226 rcu_read_lock(); 6227 6228 napi = napi_by_id(napi_id); 6229 if (!napi) 6230 goto out; 6231 6232 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6233 preempt_disable(); 6234 for (;;) { 6235 int work = 0; 6236 6237 local_bh_disable(); 6238 if (!napi_poll) { 6239 unsigned long val = READ_ONCE(napi->state); 6240 6241 /* If multiple threads are competing for this napi, 6242 * we avoid dirtying napi->state as much as we can. 6243 */ 6244 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6245 NAPIF_STATE_IN_BUSY_POLL)) { 6246 if (prefer_busy_poll) 6247 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6248 goto count; 6249 } 6250 if (cmpxchg(&napi->state, val, 6251 val | NAPIF_STATE_IN_BUSY_POLL | 6252 NAPIF_STATE_SCHED) != val) { 6253 if (prefer_busy_poll) 6254 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6255 goto count; 6256 } 6257 have_poll_lock = netpoll_poll_lock(napi); 6258 napi_poll = napi->poll; 6259 } 6260 work = napi_poll(napi, budget); 6261 trace_napi_poll(napi, work, budget); 6262 gro_normal_list(napi); 6263 count: 6264 if (work > 0) 6265 __NET_ADD_STATS(dev_net(napi->dev), 6266 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6267 local_bh_enable(); 6268 6269 if (!loop_end || loop_end(loop_end_arg, start_time)) 6270 break; 6271 6272 if (unlikely(need_resched())) { 6273 if (napi_poll) 6274 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6275 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6276 preempt_enable(); 6277 rcu_read_unlock(); 6278 cond_resched(); 6279 if (loop_end(loop_end_arg, start_time)) 6280 return; 6281 goto restart; 6282 } 6283 cpu_relax(); 6284 } 6285 if (napi_poll) 6286 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6287 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6288 preempt_enable(); 6289 out: 6290 rcu_read_unlock(); 6291 } 6292 EXPORT_SYMBOL(napi_busy_loop); 6293 6294 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6295 6296 static void napi_hash_add(struct napi_struct *napi) 6297 { 6298 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6299 return; 6300 6301 spin_lock(&napi_hash_lock); 6302 6303 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6304 do { 6305 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6306 napi_gen_id = MIN_NAPI_ID; 6307 } while (napi_by_id(napi_gen_id)); 6308 napi->napi_id = napi_gen_id; 6309 6310 hlist_add_head_rcu(&napi->napi_hash_node, 6311 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6312 6313 spin_unlock(&napi_hash_lock); 6314 } 6315 6316 /* Warning : caller is responsible to make sure rcu grace period 6317 * is respected before freeing memory containing @napi 6318 */ 6319 static void napi_hash_del(struct napi_struct *napi) 6320 { 6321 spin_lock(&napi_hash_lock); 6322 6323 hlist_del_init_rcu(&napi->napi_hash_node); 6324 6325 spin_unlock(&napi_hash_lock); 6326 } 6327 6328 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6329 { 6330 struct napi_struct *napi; 6331 6332 napi = container_of(timer, struct napi_struct, timer); 6333 6334 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6335 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6336 */ 6337 if (!napi_disable_pending(napi) && 6338 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6339 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6340 __napi_schedule_irqoff(napi); 6341 } 6342 6343 return HRTIMER_NORESTART; 6344 } 6345 6346 static void init_gro_hash(struct napi_struct *napi) 6347 { 6348 int i; 6349 6350 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6351 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6352 napi->gro_hash[i].count = 0; 6353 } 6354 napi->gro_bitmask = 0; 6355 } 6356 6357 int dev_set_threaded(struct net_device *dev, bool threaded) 6358 { 6359 struct napi_struct *napi; 6360 int err = 0; 6361 6362 if (dev->threaded == threaded) 6363 return 0; 6364 6365 if (threaded) { 6366 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6367 if (!napi->thread) { 6368 err = napi_kthread_create(napi); 6369 if (err) { 6370 threaded = false; 6371 break; 6372 } 6373 } 6374 } 6375 } 6376 6377 dev->threaded = threaded; 6378 6379 /* Make sure kthread is created before THREADED bit 6380 * is set. 6381 */ 6382 smp_mb__before_atomic(); 6383 6384 /* Setting/unsetting threaded mode on a napi might not immediately 6385 * take effect, if the current napi instance is actively being 6386 * polled. In this case, the switch between threaded mode and 6387 * softirq mode will happen in the next round of napi_schedule(). 6388 * This should not cause hiccups/stalls to the live traffic. 6389 */ 6390 list_for_each_entry(napi, &dev->napi_list, dev_list) 6391 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6392 6393 return err; 6394 } 6395 EXPORT_SYMBOL(dev_set_threaded); 6396 6397 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6398 int (*poll)(struct napi_struct *, int), int weight) 6399 { 6400 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6401 return; 6402 6403 INIT_LIST_HEAD(&napi->poll_list); 6404 INIT_HLIST_NODE(&napi->napi_hash_node); 6405 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6406 napi->timer.function = napi_watchdog; 6407 init_gro_hash(napi); 6408 napi->skb = NULL; 6409 INIT_LIST_HEAD(&napi->rx_list); 6410 napi->rx_count = 0; 6411 napi->poll = poll; 6412 if (weight > NAPI_POLL_WEIGHT) 6413 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6414 weight); 6415 napi->weight = weight; 6416 napi->dev = dev; 6417 #ifdef CONFIG_NETPOLL 6418 napi->poll_owner = -1; 6419 #endif 6420 napi->list_owner = -1; 6421 set_bit(NAPI_STATE_SCHED, &napi->state); 6422 set_bit(NAPI_STATE_NPSVC, &napi->state); 6423 list_add_rcu(&napi->dev_list, &dev->napi_list); 6424 napi_hash_add(napi); 6425 napi_get_frags_check(napi); 6426 /* Create kthread for this napi if dev->threaded is set. 6427 * Clear dev->threaded if kthread creation failed so that 6428 * threaded mode will not be enabled in napi_enable(). 6429 */ 6430 if (dev->threaded && napi_kthread_create(napi)) 6431 dev->threaded = 0; 6432 } 6433 EXPORT_SYMBOL(netif_napi_add_weight); 6434 6435 void napi_disable(struct napi_struct *n) 6436 { 6437 unsigned long val, new; 6438 6439 might_sleep(); 6440 set_bit(NAPI_STATE_DISABLE, &n->state); 6441 6442 val = READ_ONCE(n->state); 6443 do { 6444 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6445 usleep_range(20, 200); 6446 val = READ_ONCE(n->state); 6447 } 6448 6449 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6450 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6451 } while (!try_cmpxchg(&n->state, &val, new)); 6452 6453 hrtimer_cancel(&n->timer); 6454 6455 clear_bit(NAPI_STATE_DISABLE, &n->state); 6456 } 6457 EXPORT_SYMBOL(napi_disable); 6458 6459 /** 6460 * napi_enable - enable NAPI scheduling 6461 * @n: NAPI context 6462 * 6463 * Resume NAPI from being scheduled on this context. 6464 * Must be paired with napi_disable. 6465 */ 6466 void napi_enable(struct napi_struct *n) 6467 { 6468 unsigned long new, val = READ_ONCE(n->state); 6469 6470 do { 6471 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6472 6473 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6474 if (n->dev->threaded && n->thread) 6475 new |= NAPIF_STATE_THREADED; 6476 } while (!try_cmpxchg(&n->state, &val, new)); 6477 } 6478 EXPORT_SYMBOL(napi_enable); 6479 6480 static void flush_gro_hash(struct napi_struct *napi) 6481 { 6482 int i; 6483 6484 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6485 struct sk_buff *skb, *n; 6486 6487 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6488 kfree_skb(skb); 6489 napi->gro_hash[i].count = 0; 6490 } 6491 } 6492 6493 /* Must be called in process context */ 6494 void __netif_napi_del(struct napi_struct *napi) 6495 { 6496 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6497 return; 6498 6499 napi_hash_del(napi); 6500 list_del_rcu(&napi->dev_list); 6501 napi_free_frags(napi); 6502 6503 flush_gro_hash(napi); 6504 napi->gro_bitmask = 0; 6505 6506 if (napi->thread) { 6507 kthread_stop(napi->thread); 6508 napi->thread = NULL; 6509 } 6510 } 6511 EXPORT_SYMBOL(__netif_napi_del); 6512 6513 static int __napi_poll(struct napi_struct *n, bool *repoll) 6514 { 6515 int work, weight; 6516 6517 weight = n->weight; 6518 6519 /* This NAPI_STATE_SCHED test is for avoiding a race 6520 * with netpoll's poll_napi(). Only the entity which 6521 * obtains the lock and sees NAPI_STATE_SCHED set will 6522 * actually make the ->poll() call. Therefore we avoid 6523 * accidentally calling ->poll() when NAPI is not scheduled. 6524 */ 6525 work = 0; 6526 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6527 work = n->poll(n, weight); 6528 trace_napi_poll(n, work, weight); 6529 } 6530 6531 if (unlikely(work > weight)) 6532 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6533 n->poll, work, weight); 6534 6535 if (likely(work < weight)) 6536 return work; 6537 6538 /* Drivers must not modify the NAPI state if they 6539 * consume the entire weight. In such cases this code 6540 * still "owns" the NAPI instance and therefore can 6541 * move the instance around on the list at-will. 6542 */ 6543 if (unlikely(napi_disable_pending(n))) { 6544 napi_complete(n); 6545 return work; 6546 } 6547 6548 /* The NAPI context has more processing work, but busy-polling 6549 * is preferred. Exit early. 6550 */ 6551 if (napi_prefer_busy_poll(n)) { 6552 if (napi_complete_done(n, work)) { 6553 /* If timeout is not set, we need to make sure 6554 * that the NAPI is re-scheduled. 6555 */ 6556 napi_schedule(n); 6557 } 6558 return work; 6559 } 6560 6561 if (n->gro_bitmask) { 6562 /* flush too old packets 6563 * If HZ < 1000, flush all packets. 6564 */ 6565 napi_gro_flush(n, HZ >= 1000); 6566 } 6567 6568 gro_normal_list(n); 6569 6570 /* Some drivers may have called napi_schedule 6571 * prior to exhausting their budget. 6572 */ 6573 if (unlikely(!list_empty(&n->poll_list))) { 6574 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6575 n->dev ? n->dev->name : "backlog"); 6576 return work; 6577 } 6578 6579 *repoll = true; 6580 6581 return work; 6582 } 6583 6584 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6585 { 6586 bool do_repoll = false; 6587 void *have; 6588 int work; 6589 6590 list_del_init(&n->poll_list); 6591 6592 have = netpoll_poll_lock(n); 6593 6594 work = __napi_poll(n, &do_repoll); 6595 6596 if (do_repoll) 6597 list_add_tail(&n->poll_list, repoll); 6598 6599 netpoll_poll_unlock(have); 6600 6601 return work; 6602 } 6603 6604 static int napi_thread_wait(struct napi_struct *napi) 6605 { 6606 bool woken = false; 6607 6608 set_current_state(TASK_INTERRUPTIBLE); 6609 6610 while (!kthread_should_stop()) { 6611 /* Testing SCHED_THREADED bit here to make sure the current 6612 * kthread owns this napi and could poll on this napi. 6613 * Testing SCHED bit is not enough because SCHED bit might be 6614 * set by some other busy poll thread or by napi_disable(). 6615 */ 6616 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6617 WARN_ON(!list_empty(&napi->poll_list)); 6618 __set_current_state(TASK_RUNNING); 6619 return 0; 6620 } 6621 6622 schedule(); 6623 /* woken being true indicates this thread owns this napi. */ 6624 woken = true; 6625 set_current_state(TASK_INTERRUPTIBLE); 6626 } 6627 __set_current_state(TASK_RUNNING); 6628 6629 return -1; 6630 } 6631 6632 static void skb_defer_free_flush(struct softnet_data *sd) 6633 { 6634 struct sk_buff *skb, *next; 6635 6636 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6637 if (!READ_ONCE(sd->defer_list)) 6638 return; 6639 6640 spin_lock(&sd->defer_lock); 6641 skb = sd->defer_list; 6642 sd->defer_list = NULL; 6643 sd->defer_count = 0; 6644 spin_unlock(&sd->defer_lock); 6645 6646 while (skb != NULL) { 6647 next = skb->next; 6648 napi_consume_skb(skb, 1); 6649 skb = next; 6650 } 6651 } 6652 6653 static int napi_threaded_poll(void *data) 6654 { 6655 struct napi_struct *napi = data; 6656 struct softnet_data *sd; 6657 void *have; 6658 6659 while (!napi_thread_wait(napi)) { 6660 for (;;) { 6661 bool repoll = false; 6662 6663 local_bh_disable(); 6664 sd = this_cpu_ptr(&softnet_data); 6665 sd->in_napi_threaded_poll = true; 6666 6667 have = netpoll_poll_lock(napi); 6668 __napi_poll(napi, &repoll); 6669 netpoll_poll_unlock(have); 6670 6671 sd->in_napi_threaded_poll = false; 6672 barrier(); 6673 6674 if (sd_has_rps_ipi_waiting(sd)) { 6675 local_irq_disable(); 6676 net_rps_action_and_irq_enable(sd); 6677 } 6678 skb_defer_free_flush(sd); 6679 local_bh_enable(); 6680 6681 if (!repoll) 6682 break; 6683 6684 cond_resched(); 6685 } 6686 } 6687 return 0; 6688 } 6689 6690 static __latent_entropy void net_rx_action(struct softirq_action *h) 6691 { 6692 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6693 unsigned long time_limit = jiffies + 6694 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6695 int budget = READ_ONCE(netdev_budget); 6696 LIST_HEAD(list); 6697 LIST_HEAD(repoll); 6698 6699 start: 6700 sd->in_net_rx_action = true; 6701 local_irq_disable(); 6702 list_splice_init(&sd->poll_list, &list); 6703 local_irq_enable(); 6704 6705 for (;;) { 6706 struct napi_struct *n; 6707 6708 skb_defer_free_flush(sd); 6709 6710 if (list_empty(&list)) { 6711 if (list_empty(&repoll)) { 6712 sd->in_net_rx_action = false; 6713 barrier(); 6714 /* We need to check if ____napi_schedule() 6715 * had refilled poll_list while 6716 * sd->in_net_rx_action was true. 6717 */ 6718 if (!list_empty(&sd->poll_list)) 6719 goto start; 6720 if (!sd_has_rps_ipi_waiting(sd)) 6721 goto end; 6722 } 6723 break; 6724 } 6725 6726 n = list_first_entry(&list, struct napi_struct, poll_list); 6727 budget -= napi_poll(n, &repoll); 6728 6729 /* If softirq window is exhausted then punt. 6730 * Allow this to run for 2 jiffies since which will allow 6731 * an average latency of 1.5/HZ. 6732 */ 6733 if (unlikely(budget <= 0 || 6734 time_after_eq(jiffies, time_limit))) { 6735 sd->time_squeeze++; 6736 break; 6737 } 6738 } 6739 6740 local_irq_disable(); 6741 6742 list_splice_tail_init(&sd->poll_list, &list); 6743 list_splice_tail(&repoll, &list); 6744 list_splice(&list, &sd->poll_list); 6745 if (!list_empty(&sd->poll_list)) 6746 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6747 else 6748 sd->in_net_rx_action = false; 6749 6750 net_rps_action_and_irq_enable(sd); 6751 end:; 6752 } 6753 6754 struct netdev_adjacent { 6755 struct net_device *dev; 6756 netdevice_tracker dev_tracker; 6757 6758 /* upper master flag, there can only be one master device per list */ 6759 bool master; 6760 6761 /* lookup ignore flag */ 6762 bool ignore; 6763 6764 /* counter for the number of times this device was added to us */ 6765 u16 ref_nr; 6766 6767 /* private field for the users */ 6768 void *private; 6769 6770 struct list_head list; 6771 struct rcu_head rcu; 6772 }; 6773 6774 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6775 struct list_head *adj_list) 6776 { 6777 struct netdev_adjacent *adj; 6778 6779 list_for_each_entry(adj, adj_list, list) { 6780 if (adj->dev == adj_dev) 6781 return adj; 6782 } 6783 return NULL; 6784 } 6785 6786 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6787 struct netdev_nested_priv *priv) 6788 { 6789 struct net_device *dev = (struct net_device *)priv->data; 6790 6791 return upper_dev == dev; 6792 } 6793 6794 /** 6795 * netdev_has_upper_dev - Check if device is linked to an upper device 6796 * @dev: device 6797 * @upper_dev: upper device to check 6798 * 6799 * Find out if a device is linked to specified upper device and return true 6800 * in case it is. Note that this checks only immediate upper device, 6801 * not through a complete stack of devices. The caller must hold the RTNL lock. 6802 */ 6803 bool netdev_has_upper_dev(struct net_device *dev, 6804 struct net_device *upper_dev) 6805 { 6806 struct netdev_nested_priv priv = { 6807 .data = (void *)upper_dev, 6808 }; 6809 6810 ASSERT_RTNL(); 6811 6812 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6813 &priv); 6814 } 6815 EXPORT_SYMBOL(netdev_has_upper_dev); 6816 6817 /** 6818 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6819 * @dev: device 6820 * @upper_dev: upper device to check 6821 * 6822 * Find out if a device is linked to specified upper device and return true 6823 * in case it is. Note that this checks the entire upper device chain. 6824 * The caller must hold rcu lock. 6825 */ 6826 6827 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6828 struct net_device *upper_dev) 6829 { 6830 struct netdev_nested_priv priv = { 6831 .data = (void *)upper_dev, 6832 }; 6833 6834 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6835 &priv); 6836 } 6837 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6838 6839 /** 6840 * netdev_has_any_upper_dev - Check if device is linked to some device 6841 * @dev: device 6842 * 6843 * Find out if a device is linked to an upper device and return true in case 6844 * it is. The caller must hold the RTNL lock. 6845 */ 6846 bool netdev_has_any_upper_dev(struct net_device *dev) 6847 { 6848 ASSERT_RTNL(); 6849 6850 return !list_empty(&dev->adj_list.upper); 6851 } 6852 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6853 6854 /** 6855 * netdev_master_upper_dev_get - Get master upper device 6856 * @dev: device 6857 * 6858 * Find a master upper device and return pointer to it or NULL in case 6859 * it's not there. The caller must hold the RTNL lock. 6860 */ 6861 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6862 { 6863 struct netdev_adjacent *upper; 6864 6865 ASSERT_RTNL(); 6866 6867 if (list_empty(&dev->adj_list.upper)) 6868 return NULL; 6869 6870 upper = list_first_entry(&dev->adj_list.upper, 6871 struct netdev_adjacent, list); 6872 if (likely(upper->master)) 6873 return upper->dev; 6874 return NULL; 6875 } 6876 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6877 6878 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6879 { 6880 struct netdev_adjacent *upper; 6881 6882 ASSERT_RTNL(); 6883 6884 if (list_empty(&dev->adj_list.upper)) 6885 return NULL; 6886 6887 upper = list_first_entry(&dev->adj_list.upper, 6888 struct netdev_adjacent, list); 6889 if (likely(upper->master) && !upper->ignore) 6890 return upper->dev; 6891 return NULL; 6892 } 6893 6894 /** 6895 * netdev_has_any_lower_dev - Check if device is linked to some device 6896 * @dev: device 6897 * 6898 * Find out if a device is linked to a lower device and return true in case 6899 * it is. The caller must hold the RTNL lock. 6900 */ 6901 static bool netdev_has_any_lower_dev(struct net_device *dev) 6902 { 6903 ASSERT_RTNL(); 6904 6905 return !list_empty(&dev->adj_list.lower); 6906 } 6907 6908 void *netdev_adjacent_get_private(struct list_head *adj_list) 6909 { 6910 struct netdev_adjacent *adj; 6911 6912 adj = list_entry(adj_list, struct netdev_adjacent, list); 6913 6914 return adj->private; 6915 } 6916 EXPORT_SYMBOL(netdev_adjacent_get_private); 6917 6918 /** 6919 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6920 * @dev: device 6921 * @iter: list_head ** of the current position 6922 * 6923 * Gets the next device from the dev's upper list, starting from iter 6924 * position. The caller must hold RCU read lock. 6925 */ 6926 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6927 struct list_head **iter) 6928 { 6929 struct netdev_adjacent *upper; 6930 6931 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6932 6933 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6934 6935 if (&upper->list == &dev->adj_list.upper) 6936 return NULL; 6937 6938 *iter = &upper->list; 6939 6940 return upper->dev; 6941 } 6942 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6943 6944 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6945 struct list_head **iter, 6946 bool *ignore) 6947 { 6948 struct netdev_adjacent *upper; 6949 6950 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6951 6952 if (&upper->list == &dev->adj_list.upper) 6953 return NULL; 6954 6955 *iter = &upper->list; 6956 *ignore = upper->ignore; 6957 6958 return upper->dev; 6959 } 6960 6961 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6962 struct list_head **iter) 6963 { 6964 struct netdev_adjacent *upper; 6965 6966 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6967 6968 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6969 6970 if (&upper->list == &dev->adj_list.upper) 6971 return NULL; 6972 6973 *iter = &upper->list; 6974 6975 return upper->dev; 6976 } 6977 6978 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6979 int (*fn)(struct net_device *dev, 6980 struct netdev_nested_priv *priv), 6981 struct netdev_nested_priv *priv) 6982 { 6983 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6984 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6985 int ret, cur = 0; 6986 bool ignore; 6987 6988 now = dev; 6989 iter = &dev->adj_list.upper; 6990 6991 while (1) { 6992 if (now != dev) { 6993 ret = fn(now, priv); 6994 if (ret) 6995 return ret; 6996 } 6997 6998 next = NULL; 6999 while (1) { 7000 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7001 if (!udev) 7002 break; 7003 if (ignore) 7004 continue; 7005 7006 next = udev; 7007 niter = &udev->adj_list.upper; 7008 dev_stack[cur] = now; 7009 iter_stack[cur++] = iter; 7010 break; 7011 } 7012 7013 if (!next) { 7014 if (!cur) 7015 return 0; 7016 next = dev_stack[--cur]; 7017 niter = iter_stack[cur]; 7018 } 7019 7020 now = next; 7021 iter = niter; 7022 } 7023 7024 return 0; 7025 } 7026 7027 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7028 int (*fn)(struct net_device *dev, 7029 struct netdev_nested_priv *priv), 7030 struct netdev_nested_priv *priv) 7031 { 7032 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7033 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7034 int ret, cur = 0; 7035 7036 now = dev; 7037 iter = &dev->adj_list.upper; 7038 7039 while (1) { 7040 if (now != dev) { 7041 ret = fn(now, priv); 7042 if (ret) 7043 return ret; 7044 } 7045 7046 next = NULL; 7047 while (1) { 7048 udev = netdev_next_upper_dev_rcu(now, &iter); 7049 if (!udev) 7050 break; 7051 7052 next = udev; 7053 niter = &udev->adj_list.upper; 7054 dev_stack[cur] = now; 7055 iter_stack[cur++] = iter; 7056 break; 7057 } 7058 7059 if (!next) { 7060 if (!cur) 7061 return 0; 7062 next = dev_stack[--cur]; 7063 niter = iter_stack[cur]; 7064 } 7065 7066 now = next; 7067 iter = niter; 7068 } 7069 7070 return 0; 7071 } 7072 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7073 7074 static bool __netdev_has_upper_dev(struct net_device *dev, 7075 struct net_device *upper_dev) 7076 { 7077 struct netdev_nested_priv priv = { 7078 .flags = 0, 7079 .data = (void *)upper_dev, 7080 }; 7081 7082 ASSERT_RTNL(); 7083 7084 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7085 &priv); 7086 } 7087 7088 /** 7089 * netdev_lower_get_next_private - Get the next ->private from the 7090 * lower neighbour list 7091 * @dev: device 7092 * @iter: list_head ** of the current position 7093 * 7094 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7095 * list, starting from iter position. The caller must hold either hold the 7096 * RTNL lock or its own locking that guarantees that the neighbour lower 7097 * list will remain unchanged. 7098 */ 7099 void *netdev_lower_get_next_private(struct net_device *dev, 7100 struct list_head **iter) 7101 { 7102 struct netdev_adjacent *lower; 7103 7104 lower = list_entry(*iter, struct netdev_adjacent, list); 7105 7106 if (&lower->list == &dev->adj_list.lower) 7107 return NULL; 7108 7109 *iter = lower->list.next; 7110 7111 return lower->private; 7112 } 7113 EXPORT_SYMBOL(netdev_lower_get_next_private); 7114 7115 /** 7116 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7117 * lower neighbour list, RCU 7118 * variant 7119 * @dev: device 7120 * @iter: list_head ** of the current position 7121 * 7122 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7123 * list, starting from iter position. The caller must hold RCU read lock. 7124 */ 7125 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7126 struct list_head **iter) 7127 { 7128 struct netdev_adjacent *lower; 7129 7130 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7131 7132 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7133 7134 if (&lower->list == &dev->adj_list.lower) 7135 return NULL; 7136 7137 *iter = &lower->list; 7138 7139 return lower->private; 7140 } 7141 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7142 7143 /** 7144 * netdev_lower_get_next - Get the next device from the lower neighbour 7145 * list 7146 * @dev: device 7147 * @iter: list_head ** of the current position 7148 * 7149 * Gets the next netdev_adjacent from the dev's lower neighbour 7150 * list, starting from iter position. The caller must hold RTNL lock or 7151 * its own locking that guarantees that the neighbour lower 7152 * list will remain unchanged. 7153 */ 7154 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7155 { 7156 struct netdev_adjacent *lower; 7157 7158 lower = list_entry(*iter, struct netdev_adjacent, list); 7159 7160 if (&lower->list == &dev->adj_list.lower) 7161 return NULL; 7162 7163 *iter = lower->list.next; 7164 7165 return lower->dev; 7166 } 7167 EXPORT_SYMBOL(netdev_lower_get_next); 7168 7169 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7170 struct list_head **iter) 7171 { 7172 struct netdev_adjacent *lower; 7173 7174 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7175 7176 if (&lower->list == &dev->adj_list.lower) 7177 return NULL; 7178 7179 *iter = &lower->list; 7180 7181 return lower->dev; 7182 } 7183 7184 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7185 struct list_head **iter, 7186 bool *ignore) 7187 { 7188 struct netdev_adjacent *lower; 7189 7190 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7191 7192 if (&lower->list == &dev->adj_list.lower) 7193 return NULL; 7194 7195 *iter = &lower->list; 7196 *ignore = lower->ignore; 7197 7198 return lower->dev; 7199 } 7200 7201 int netdev_walk_all_lower_dev(struct net_device *dev, 7202 int (*fn)(struct net_device *dev, 7203 struct netdev_nested_priv *priv), 7204 struct netdev_nested_priv *priv) 7205 { 7206 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7207 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7208 int ret, cur = 0; 7209 7210 now = dev; 7211 iter = &dev->adj_list.lower; 7212 7213 while (1) { 7214 if (now != dev) { 7215 ret = fn(now, priv); 7216 if (ret) 7217 return ret; 7218 } 7219 7220 next = NULL; 7221 while (1) { 7222 ldev = netdev_next_lower_dev(now, &iter); 7223 if (!ldev) 7224 break; 7225 7226 next = ldev; 7227 niter = &ldev->adj_list.lower; 7228 dev_stack[cur] = now; 7229 iter_stack[cur++] = iter; 7230 break; 7231 } 7232 7233 if (!next) { 7234 if (!cur) 7235 return 0; 7236 next = dev_stack[--cur]; 7237 niter = iter_stack[cur]; 7238 } 7239 7240 now = next; 7241 iter = niter; 7242 } 7243 7244 return 0; 7245 } 7246 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7247 7248 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7249 int (*fn)(struct net_device *dev, 7250 struct netdev_nested_priv *priv), 7251 struct netdev_nested_priv *priv) 7252 { 7253 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7254 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7255 int ret, cur = 0; 7256 bool ignore; 7257 7258 now = dev; 7259 iter = &dev->adj_list.lower; 7260 7261 while (1) { 7262 if (now != dev) { 7263 ret = fn(now, priv); 7264 if (ret) 7265 return ret; 7266 } 7267 7268 next = NULL; 7269 while (1) { 7270 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7271 if (!ldev) 7272 break; 7273 if (ignore) 7274 continue; 7275 7276 next = ldev; 7277 niter = &ldev->adj_list.lower; 7278 dev_stack[cur] = now; 7279 iter_stack[cur++] = iter; 7280 break; 7281 } 7282 7283 if (!next) { 7284 if (!cur) 7285 return 0; 7286 next = dev_stack[--cur]; 7287 niter = iter_stack[cur]; 7288 } 7289 7290 now = next; 7291 iter = niter; 7292 } 7293 7294 return 0; 7295 } 7296 7297 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7298 struct list_head **iter) 7299 { 7300 struct netdev_adjacent *lower; 7301 7302 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7303 if (&lower->list == &dev->adj_list.lower) 7304 return NULL; 7305 7306 *iter = &lower->list; 7307 7308 return lower->dev; 7309 } 7310 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7311 7312 static u8 __netdev_upper_depth(struct net_device *dev) 7313 { 7314 struct net_device *udev; 7315 struct list_head *iter; 7316 u8 max_depth = 0; 7317 bool ignore; 7318 7319 for (iter = &dev->adj_list.upper, 7320 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7321 udev; 7322 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7323 if (ignore) 7324 continue; 7325 if (max_depth < udev->upper_level) 7326 max_depth = udev->upper_level; 7327 } 7328 7329 return max_depth; 7330 } 7331 7332 static u8 __netdev_lower_depth(struct net_device *dev) 7333 { 7334 struct net_device *ldev; 7335 struct list_head *iter; 7336 u8 max_depth = 0; 7337 bool ignore; 7338 7339 for (iter = &dev->adj_list.lower, 7340 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7341 ldev; 7342 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7343 if (ignore) 7344 continue; 7345 if (max_depth < ldev->lower_level) 7346 max_depth = ldev->lower_level; 7347 } 7348 7349 return max_depth; 7350 } 7351 7352 static int __netdev_update_upper_level(struct net_device *dev, 7353 struct netdev_nested_priv *__unused) 7354 { 7355 dev->upper_level = __netdev_upper_depth(dev) + 1; 7356 return 0; 7357 } 7358 7359 #ifdef CONFIG_LOCKDEP 7360 static LIST_HEAD(net_unlink_list); 7361 7362 static void net_unlink_todo(struct net_device *dev) 7363 { 7364 if (list_empty(&dev->unlink_list)) 7365 list_add_tail(&dev->unlink_list, &net_unlink_list); 7366 } 7367 #endif 7368 7369 static int __netdev_update_lower_level(struct net_device *dev, 7370 struct netdev_nested_priv *priv) 7371 { 7372 dev->lower_level = __netdev_lower_depth(dev) + 1; 7373 7374 #ifdef CONFIG_LOCKDEP 7375 if (!priv) 7376 return 0; 7377 7378 if (priv->flags & NESTED_SYNC_IMM) 7379 dev->nested_level = dev->lower_level - 1; 7380 if (priv->flags & NESTED_SYNC_TODO) 7381 net_unlink_todo(dev); 7382 #endif 7383 return 0; 7384 } 7385 7386 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7387 int (*fn)(struct net_device *dev, 7388 struct netdev_nested_priv *priv), 7389 struct netdev_nested_priv *priv) 7390 { 7391 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7392 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7393 int ret, cur = 0; 7394 7395 now = dev; 7396 iter = &dev->adj_list.lower; 7397 7398 while (1) { 7399 if (now != dev) { 7400 ret = fn(now, priv); 7401 if (ret) 7402 return ret; 7403 } 7404 7405 next = NULL; 7406 while (1) { 7407 ldev = netdev_next_lower_dev_rcu(now, &iter); 7408 if (!ldev) 7409 break; 7410 7411 next = ldev; 7412 niter = &ldev->adj_list.lower; 7413 dev_stack[cur] = now; 7414 iter_stack[cur++] = iter; 7415 break; 7416 } 7417 7418 if (!next) { 7419 if (!cur) 7420 return 0; 7421 next = dev_stack[--cur]; 7422 niter = iter_stack[cur]; 7423 } 7424 7425 now = next; 7426 iter = niter; 7427 } 7428 7429 return 0; 7430 } 7431 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7432 7433 /** 7434 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7435 * lower neighbour list, RCU 7436 * variant 7437 * @dev: device 7438 * 7439 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7440 * list. The caller must hold RCU read lock. 7441 */ 7442 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7443 { 7444 struct netdev_adjacent *lower; 7445 7446 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7447 struct netdev_adjacent, list); 7448 if (lower) 7449 return lower->private; 7450 return NULL; 7451 } 7452 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7453 7454 /** 7455 * netdev_master_upper_dev_get_rcu - Get master upper device 7456 * @dev: device 7457 * 7458 * Find a master upper device and return pointer to it or NULL in case 7459 * it's not there. The caller must hold the RCU read lock. 7460 */ 7461 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7462 { 7463 struct netdev_adjacent *upper; 7464 7465 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7466 struct netdev_adjacent, list); 7467 if (upper && likely(upper->master)) 7468 return upper->dev; 7469 return NULL; 7470 } 7471 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7472 7473 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7474 struct net_device *adj_dev, 7475 struct list_head *dev_list) 7476 { 7477 char linkname[IFNAMSIZ+7]; 7478 7479 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7480 "upper_%s" : "lower_%s", adj_dev->name); 7481 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7482 linkname); 7483 } 7484 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7485 char *name, 7486 struct list_head *dev_list) 7487 { 7488 char linkname[IFNAMSIZ+7]; 7489 7490 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7491 "upper_%s" : "lower_%s", name); 7492 sysfs_remove_link(&(dev->dev.kobj), linkname); 7493 } 7494 7495 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7496 struct net_device *adj_dev, 7497 struct list_head *dev_list) 7498 { 7499 return (dev_list == &dev->adj_list.upper || 7500 dev_list == &dev->adj_list.lower) && 7501 net_eq(dev_net(dev), dev_net(adj_dev)); 7502 } 7503 7504 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7505 struct net_device *adj_dev, 7506 struct list_head *dev_list, 7507 void *private, bool master) 7508 { 7509 struct netdev_adjacent *adj; 7510 int ret; 7511 7512 adj = __netdev_find_adj(adj_dev, dev_list); 7513 7514 if (adj) { 7515 adj->ref_nr += 1; 7516 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7517 dev->name, adj_dev->name, adj->ref_nr); 7518 7519 return 0; 7520 } 7521 7522 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7523 if (!adj) 7524 return -ENOMEM; 7525 7526 adj->dev = adj_dev; 7527 adj->master = master; 7528 adj->ref_nr = 1; 7529 adj->private = private; 7530 adj->ignore = false; 7531 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7532 7533 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7534 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7535 7536 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7537 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7538 if (ret) 7539 goto free_adj; 7540 } 7541 7542 /* Ensure that master link is always the first item in list. */ 7543 if (master) { 7544 ret = sysfs_create_link(&(dev->dev.kobj), 7545 &(adj_dev->dev.kobj), "master"); 7546 if (ret) 7547 goto remove_symlinks; 7548 7549 list_add_rcu(&adj->list, dev_list); 7550 } else { 7551 list_add_tail_rcu(&adj->list, dev_list); 7552 } 7553 7554 return 0; 7555 7556 remove_symlinks: 7557 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7558 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7559 free_adj: 7560 netdev_put(adj_dev, &adj->dev_tracker); 7561 kfree(adj); 7562 7563 return ret; 7564 } 7565 7566 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7567 struct net_device *adj_dev, 7568 u16 ref_nr, 7569 struct list_head *dev_list) 7570 { 7571 struct netdev_adjacent *adj; 7572 7573 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7574 dev->name, adj_dev->name, ref_nr); 7575 7576 adj = __netdev_find_adj(adj_dev, dev_list); 7577 7578 if (!adj) { 7579 pr_err("Adjacency does not exist for device %s from %s\n", 7580 dev->name, adj_dev->name); 7581 WARN_ON(1); 7582 return; 7583 } 7584 7585 if (adj->ref_nr > ref_nr) { 7586 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7587 dev->name, adj_dev->name, ref_nr, 7588 adj->ref_nr - ref_nr); 7589 adj->ref_nr -= ref_nr; 7590 return; 7591 } 7592 7593 if (adj->master) 7594 sysfs_remove_link(&(dev->dev.kobj), "master"); 7595 7596 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7597 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7598 7599 list_del_rcu(&adj->list); 7600 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7601 adj_dev->name, dev->name, adj_dev->name); 7602 netdev_put(adj_dev, &adj->dev_tracker); 7603 kfree_rcu(adj, rcu); 7604 } 7605 7606 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7607 struct net_device *upper_dev, 7608 struct list_head *up_list, 7609 struct list_head *down_list, 7610 void *private, bool master) 7611 { 7612 int ret; 7613 7614 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7615 private, master); 7616 if (ret) 7617 return ret; 7618 7619 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7620 private, false); 7621 if (ret) { 7622 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7623 return ret; 7624 } 7625 7626 return 0; 7627 } 7628 7629 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7630 struct net_device *upper_dev, 7631 u16 ref_nr, 7632 struct list_head *up_list, 7633 struct list_head *down_list) 7634 { 7635 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7636 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7637 } 7638 7639 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7640 struct net_device *upper_dev, 7641 void *private, bool master) 7642 { 7643 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7644 &dev->adj_list.upper, 7645 &upper_dev->adj_list.lower, 7646 private, master); 7647 } 7648 7649 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7650 struct net_device *upper_dev) 7651 { 7652 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7653 &dev->adj_list.upper, 7654 &upper_dev->adj_list.lower); 7655 } 7656 7657 static int __netdev_upper_dev_link(struct net_device *dev, 7658 struct net_device *upper_dev, bool master, 7659 void *upper_priv, void *upper_info, 7660 struct netdev_nested_priv *priv, 7661 struct netlink_ext_ack *extack) 7662 { 7663 struct netdev_notifier_changeupper_info changeupper_info = { 7664 .info = { 7665 .dev = dev, 7666 .extack = extack, 7667 }, 7668 .upper_dev = upper_dev, 7669 .master = master, 7670 .linking = true, 7671 .upper_info = upper_info, 7672 }; 7673 struct net_device *master_dev; 7674 int ret = 0; 7675 7676 ASSERT_RTNL(); 7677 7678 if (dev == upper_dev) 7679 return -EBUSY; 7680 7681 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7682 if (__netdev_has_upper_dev(upper_dev, dev)) 7683 return -EBUSY; 7684 7685 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7686 return -EMLINK; 7687 7688 if (!master) { 7689 if (__netdev_has_upper_dev(dev, upper_dev)) 7690 return -EEXIST; 7691 } else { 7692 master_dev = __netdev_master_upper_dev_get(dev); 7693 if (master_dev) 7694 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7695 } 7696 7697 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7698 &changeupper_info.info); 7699 ret = notifier_to_errno(ret); 7700 if (ret) 7701 return ret; 7702 7703 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7704 master); 7705 if (ret) 7706 return ret; 7707 7708 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7709 &changeupper_info.info); 7710 ret = notifier_to_errno(ret); 7711 if (ret) 7712 goto rollback; 7713 7714 __netdev_update_upper_level(dev, NULL); 7715 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7716 7717 __netdev_update_lower_level(upper_dev, priv); 7718 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7719 priv); 7720 7721 return 0; 7722 7723 rollback: 7724 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7725 7726 return ret; 7727 } 7728 7729 /** 7730 * netdev_upper_dev_link - Add a link to the upper device 7731 * @dev: device 7732 * @upper_dev: new upper device 7733 * @extack: netlink extended ack 7734 * 7735 * Adds a link to device which is upper to this one. The caller must hold 7736 * the RTNL lock. On a failure a negative errno code is returned. 7737 * On success the reference counts are adjusted and the function 7738 * returns zero. 7739 */ 7740 int netdev_upper_dev_link(struct net_device *dev, 7741 struct net_device *upper_dev, 7742 struct netlink_ext_ack *extack) 7743 { 7744 struct netdev_nested_priv priv = { 7745 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7746 .data = NULL, 7747 }; 7748 7749 return __netdev_upper_dev_link(dev, upper_dev, false, 7750 NULL, NULL, &priv, extack); 7751 } 7752 EXPORT_SYMBOL(netdev_upper_dev_link); 7753 7754 /** 7755 * netdev_master_upper_dev_link - Add a master link to the upper device 7756 * @dev: device 7757 * @upper_dev: new upper device 7758 * @upper_priv: upper device private 7759 * @upper_info: upper info to be passed down via notifier 7760 * @extack: netlink extended ack 7761 * 7762 * Adds a link to device which is upper to this one. In this case, only 7763 * one master upper device can be linked, although other non-master devices 7764 * might be linked as well. The caller must hold the RTNL lock. 7765 * On a failure a negative errno code is returned. On success the reference 7766 * counts are adjusted and the function returns zero. 7767 */ 7768 int netdev_master_upper_dev_link(struct net_device *dev, 7769 struct net_device *upper_dev, 7770 void *upper_priv, void *upper_info, 7771 struct netlink_ext_ack *extack) 7772 { 7773 struct netdev_nested_priv priv = { 7774 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7775 .data = NULL, 7776 }; 7777 7778 return __netdev_upper_dev_link(dev, upper_dev, true, 7779 upper_priv, upper_info, &priv, extack); 7780 } 7781 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7782 7783 static void __netdev_upper_dev_unlink(struct net_device *dev, 7784 struct net_device *upper_dev, 7785 struct netdev_nested_priv *priv) 7786 { 7787 struct netdev_notifier_changeupper_info changeupper_info = { 7788 .info = { 7789 .dev = dev, 7790 }, 7791 .upper_dev = upper_dev, 7792 .linking = false, 7793 }; 7794 7795 ASSERT_RTNL(); 7796 7797 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7798 7799 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7800 &changeupper_info.info); 7801 7802 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7803 7804 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7805 &changeupper_info.info); 7806 7807 __netdev_update_upper_level(dev, NULL); 7808 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7809 7810 __netdev_update_lower_level(upper_dev, priv); 7811 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7812 priv); 7813 } 7814 7815 /** 7816 * netdev_upper_dev_unlink - Removes a link to upper device 7817 * @dev: device 7818 * @upper_dev: new upper device 7819 * 7820 * Removes a link to device which is upper to this one. The caller must hold 7821 * the RTNL lock. 7822 */ 7823 void netdev_upper_dev_unlink(struct net_device *dev, 7824 struct net_device *upper_dev) 7825 { 7826 struct netdev_nested_priv priv = { 7827 .flags = NESTED_SYNC_TODO, 7828 .data = NULL, 7829 }; 7830 7831 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7832 } 7833 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7834 7835 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7836 struct net_device *lower_dev, 7837 bool val) 7838 { 7839 struct netdev_adjacent *adj; 7840 7841 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7842 if (adj) 7843 adj->ignore = val; 7844 7845 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7846 if (adj) 7847 adj->ignore = val; 7848 } 7849 7850 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7851 struct net_device *lower_dev) 7852 { 7853 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7854 } 7855 7856 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7857 struct net_device *lower_dev) 7858 { 7859 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7860 } 7861 7862 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7863 struct net_device *new_dev, 7864 struct net_device *dev, 7865 struct netlink_ext_ack *extack) 7866 { 7867 struct netdev_nested_priv priv = { 7868 .flags = 0, 7869 .data = NULL, 7870 }; 7871 int err; 7872 7873 if (!new_dev) 7874 return 0; 7875 7876 if (old_dev && new_dev != old_dev) 7877 netdev_adjacent_dev_disable(dev, old_dev); 7878 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7879 extack); 7880 if (err) { 7881 if (old_dev && new_dev != old_dev) 7882 netdev_adjacent_dev_enable(dev, old_dev); 7883 return err; 7884 } 7885 7886 return 0; 7887 } 7888 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7889 7890 void netdev_adjacent_change_commit(struct net_device *old_dev, 7891 struct net_device *new_dev, 7892 struct net_device *dev) 7893 { 7894 struct netdev_nested_priv priv = { 7895 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7896 .data = NULL, 7897 }; 7898 7899 if (!new_dev || !old_dev) 7900 return; 7901 7902 if (new_dev == old_dev) 7903 return; 7904 7905 netdev_adjacent_dev_enable(dev, old_dev); 7906 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7907 } 7908 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7909 7910 void netdev_adjacent_change_abort(struct net_device *old_dev, 7911 struct net_device *new_dev, 7912 struct net_device *dev) 7913 { 7914 struct netdev_nested_priv priv = { 7915 .flags = 0, 7916 .data = NULL, 7917 }; 7918 7919 if (!new_dev) 7920 return; 7921 7922 if (old_dev && new_dev != old_dev) 7923 netdev_adjacent_dev_enable(dev, old_dev); 7924 7925 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7926 } 7927 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7928 7929 /** 7930 * netdev_bonding_info_change - Dispatch event about slave change 7931 * @dev: device 7932 * @bonding_info: info to dispatch 7933 * 7934 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7935 * The caller must hold the RTNL lock. 7936 */ 7937 void netdev_bonding_info_change(struct net_device *dev, 7938 struct netdev_bonding_info *bonding_info) 7939 { 7940 struct netdev_notifier_bonding_info info = { 7941 .info.dev = dev, 7942 }; 7943 7944 memcpy(&info.bonding_info, bonding_info, 7945 sizeof(struct netdev_bonding_info)); 7946 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7947 &info.info); 7948 } 7949 EXPORT_SYMBOL(netdev_bonding_info_change); 7950 7951 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7952 struct netlink_ext_ack *extack) 7953 { 7954 struct netdev_notifier_offload_xstats_info info = { 7955 .info.dev = dev, 7956 .info.extack = extack, 7957 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7958 }; 7959 int err; 7960 int rc; 7961 7962 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7963 GFP_KERNEL); 7964 if (!dev->offload_xstats_l3) 7965 return -ENOMEM; 7966 7967 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7968 NETDEV_OFFLOAD_XSTATS_DISABLE, 7969 &info.info); 7970 err = notifier_to_errno(rc); 7971 if (err) 7972 goto free_stats; 7973 7974 return 0; 7975 7976 free_stats: 7977 kfree(dev->offload_xstats_l3); 7978 dev->offload_xstats_l3 = NULL; 7979 return err; 7980 } 7981 7982 int netdev_offload_xstats_enable(struct net_device *dev, 7983 enum netdev_offload_xstats_type type, 7984 struct netlink_ext_ack *extack) 7985 { 7986 ASSERT_RTNL(); 7987 7988 if (netdev_offload_xstats_enabled(dev, type)) 7989 return -EALREADY; 7990 7991 switch (type) { 7992 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7993 return netdev_offload_xstats_enable_l3(dev, extack); 7994 } 7995 7996 WARN_ON(1); 7997 return -EINVAL; 7998 } 7999 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8000 8001 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8002 { 8003 struct netdev_notifier_offload_xstats_info info = { 8004 .info.dev = dev, 8005 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8006 }; 8007 8008 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8009 &info.info); 8010 kfree(dev->offload_xstats_l3); 8011 dev->offload_xstats_l3 = NULL; 8012 } 8013 8014 int netdev_offload_xstats_disable(struct net_device *dev, 8015 enum netdev_offload_xstats_type type) 8016 { 8017 ASSERT_RTNL(); 8018 8019 if (!netdev_offload_xstats_enabled(dev, type)) 8020 return -EALREADY; 8021 8022 switch (type) { 8023 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8024 netdev_offload_xstats_disable_l3(dev); 8025 return 0; 8026 } 8027 8028 WARN_ON(1); 8029 return -EINVAL; 8030 } 8031 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8032 8033 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8034 { 8035 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8036 } 8037 8038 static struct rtnl_hw_stats64 * 8039 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8040 enum netdev_offload_xstats_type type) 8041 { 8042 switch (type) { 8043 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8044 return dev->offload_xstats_l3; 8045 } 8046 8047 WARN_ON(1); 8048 return NULL; 8049 } 8050 8051 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8052 enum netdev_offload_xstats_type type) 8053 { 8054 ASSERT_RTNL(); 8055 8056 return netdev_offload_xstats_get_ptr(dev, type); 8057 } 8058 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8059 8060 struct netdev_notifier_offload_xstats_ru { 8061 bool used; 8062 }; 8063 8064 struct netdev_notifier_offload_xstats_rd { 8065 struct rtnl_hw_stats64 stats; 8066 bool used; 8067 }; 8068 8069 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8070 const struct rtnl_hw_stats64 *src) 8071 { 8072 dest->rx_packets += src->rx_packets; 8073 dest->tx_packets += src->tx_packets; 8074 dest->rx_bytes += src->rx_bytes; 8075 dest->tx_bytes += src->tx_bytes; 8076 dest->rx_errors += src->rx_errors; 8077 dest->tx_errors += src->tx_errors; 8078 dest->rx_dropped += src->rx_dropped; 8079 dest->tx_dropped += src->tx_dropped; 8080 dest->multicast += src->multicast; 8081 } 8082 8083 static int netdev_offload_xstats_get_used(struct net_device *dev, 8084 enum netdev_offload_xstats_type type, 8085 bool *p_used, 8086 struct netlink_ext_ack *extack) 8087 { 8088 struct netdev_notifier_offload_xstats_ru report_used = {}; 8089 struct netdev_notifier_offload_xstats_info info = { 8090 .info.dev = dev, 8091 .info.extack = extack, 8092 .type = type, 8093 .report_used = &report_used, 8094 }; 8095 int rc; 8096 8097 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8098 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8099 &info.info); 8100 *p_used = report_used.used; 8101 return notifier_to_errno(rc); 8102 } 8103 8104 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8105 enum netdev_offload_xstats_type type, 8106 struct rtnl_hw_stats64 *p_stats, 8107 bool *p_used, 8108 struct netlink_ext_ack *extack) 8109 { 8110 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8111 struct netdev_notifier_offload_xstats_info info = { 8112 .info.dev = dev, 8113 .info.extack = extack, 8114 .type = type, 8115 .report_delta = &report_delta, 8116 }; 8117 struct rtnl_hw_stats64 *stats; 8118 int rc; 8119 8120 stats = netdev_offload_xstats_get_ptr(dev, type); 8121 if (WARN_ON(!stats)) 8122 return -EINVAL; 8123 8124 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8125 &info.info); 8126 8127 /* Cache whatever we got, even if there was an error, otherwise the 8128 * successful stats retrievals would get lost. 8129 */ 8130 netdev_hw_stats64_add(stats, &report_delta.stats); 8131 8132 if (p_stats) 8133 *p_stats = *stats; 8134 *p_used = report_delta.used; 8135 8136 return notifier_to_errno(rc); 8137 } 8138 8139 int netdev_offload_xstats_get(struct net_device *dev, 8140 enum netdev_offload_xstats_type type, 8141 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8142 struct netlink_ext_ack *extack) 8143 { 8144 ASSERT_RTNL(); 8145 8146 if (p_stats) 8147 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8148 p_used, extack); 8149 else 8150 return netdev_offload_xstats_get_used(dev, type, p_used, 8151 extack); 8152 } 8153 EXPORT_SYMBOL(netdev_offload_xstats_get); 8154 8155 void 8156 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8157 const struct rtnl_hw_stats64 *stats) 8158 { 8159 report_delta->used = true; 8160 netdev_hw_stats64_add(&report_delta->stats, stats); 8161 } 8162 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8163 8164 void 8165 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8166 { 8167 report_used->used = true; 8168 } 8169 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8170 8171 void netdev_offload_xstats_push_delta(struct net_device *dev, 8172 enum netdev_offload_xstats_type type, 8173 const struct rtnl_hw_stats64 *p_stats) 8174 { 8175 struct rtnl_hw_stats64 *stats; 8176 8177 ASSERT_RTNL(); 8178 8179 stats = netdev_offload_xstats_get_ptr(dev, type); 8180 if (WARN_ON(!stats)) 8181 return; 8182 8183 netdev_hw_stats64_add(stats, p_stats); 8184 } 8185 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8186 8187 /** 8188 * netdev_get_xmit_slave - Get the xmit slave of master device 8189 * @dev: device 8190 * @skb: The packet 8191 * @all_slaves: assume all the slaves are active 8192 * 8193 * The reference counters are not incremented so the caller must be 8194 * careful with locks. The caller must hold RCU lock. 8195 * %NULL is returned if no slave is found. 8196 */ 8197 8198 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8199 struct sk_buff *skb, 8200 bool all_slaves) 8201 { 8202 const struct net_device_ops *ops = dev->netdev_ops; 8203 8204 if (!ops->ndo_get_xmit_slave) 8205 return NULL; 8206 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8207 } 8208 EXPORT_SYMBOL(netdev_get_xmit_slave); 8209 8210 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8211 struct sock *sk) 8212 { 8213 const struct net_device_ops *ops = dev->netdev_ops; 8214 8215 if (!ops->ndo_sk_get_lower_dev) 8216 return NULL; 8217 return ops->ndo_sk_get_lower_dev(dev, sk); 8218 } 8219 8220 /** 8221 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8222 * @dev: device 8223 * @sk: the socket 8224 * 8225 * %NULL is returned if no lower device is found. 8226 */ 8227 8228 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8229 struct sock *sk) 8230 { 8231 struct net_device *lower; 8232 8233 lower = netdev_sk_get_lower_dev(dev, sk); 8234 while (lower) { 8235 dev = lower; 8236 lower = netdev_sk_get_lower_dev(dev, sk); 8237 } 8238 8239 return dev; 8240 } 8241 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8242 8243 static void netdev_adjacent_add_links(struct net_device *dev) 8244 { 8245 struct netdev_adjacent *iter; 8246 8247 struct net *net = dev_net(dev); 8248 8249 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8250 if (!net_eq(net, dev_net(iter->dev))) 8251 continue; 8252 netdev_adjacent_sysfs_add(iter->dev, dev, 8253 &iter->dev->adj_list.lower); 8254 netdev_adjacent_sysfs_add(dev, iter->dev, 8255 &dev->adj_list.upper); 8256 } 8257 8258 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8259 if (!net_eq(net, dev_net(iter->dev))) 8260 continue; 8261 netdev_adjacent_sysfs_add(iter->dev, dev, 8262 &iter->dev->adj_list.upper); 8263 netdev_adjacent_sysfs_add(dev, iter->dev, 8264 &dev->adj_list.lower); 8265 } 8266 } 8267 8268 static void netdev_adjacent_del_links(struct net_device *dev) 8269 { 8270 struct netdev_adjacent *iter; 8271 8272 struct net *net = dev_net(dev); 8273 8274 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8275 if (!net_eq(net, dev_net(iter->dev))) 8276 continue; 8277 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8278 &iter->dev->adj_list.lower); 8279 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8280 &dev->adj_list.upper); 8281 } 8282 8283 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8284 if (!net_eq(net, dev_net(iter->dev))) 8285 continue; 8286 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8287 &iter->dev->adj_list.upper); 8288 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8289 &dev->adj_list.lower); 8290 } 8291 } 8292 8293 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8294 { 8295 struct netdev_adjacent *iter; 8296 8297 struct net *net = dev_net(dev); 8298 8299 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8300 if (!net_eq(net, dev_net(iter->dev))) 8301 continue; 8302 netdev_adjacent_sysfs_del(iter->dev, oldname, 8303 &iter->dev->adj_list.lower); 8304 netdev_adjacent_sysfs_add(iter->dev, dev, 8305 &iter->dev->adj_list.lower); 8306 } 8307 8308 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8309 if (!net_eq(net, dev_net(iter->dev))) 8310 continue; 8311 netdev_adjacent_sysfs_del(iter->dev, oldname, 8312 &iter->dev->adj_list.upper); 8313 netdev_adjacent_sysfs_add(iter->dev, dev, 8314 &iter->dev->adj_list.upper); 8315 } 8316 } 8317 8318 void *netdev_lower_dev_get_private(struct net_device *dev, 8319 struct net_device *lower_dev) 8320 { 8321 struct netdev_adjacent *lower; 8322 8323 if (!lower_dev) 8324 return NULL; 8325 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8326 if (!lower) 8327 return NULL; 8328 8329 return lower->private; 8330 } 8331 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8332 8333 8334 /** 8335 * netdev_lower_state_changed - Dispatch event about lower device state change 8336 * @lower_dev: device 8337 * @lower_state_info: state to dispatch 8338 * 8339 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8340 * The caller must hold the RTNL lock. 8341 */ 8342 void netdev_lower_state_changed(struct net_device *lower_dev, 8343 void *lower_state_info) 8344 { 8345 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8346 .info.dev = lower_dev, 8347 }; 8348 8349 ASSERT_RTNL(); 8350 changelowerstate_info.lower_state_info = lower_state_info; 8351 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8352 &changelowerstate_info.info); 8353 } 8354 EXPORT_SYMBOL(netdev_lower_state_changed); 8355 8356 static void dev_change_rx_flags(struct net_device *dev, int flags) 8357 { 8358 const struct net_device_ops *ops = dev->netdev_ops; 8359 8360 if (ops->ndo_change_rx_flags) 8361 ops->ndo_change_rx_flags(dev, flags); 8362 } 8363 8364 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8365 { 8366 unsigned int old_flags = dev->flags; 8367 kuid_t uid; 8368 kgid_t gid; 8369 8370 ASSERT_RTNL(); 8371 8372 dev->flags |= IFF_PROMISC; 8373 dev->promiscuity += inc; 8374 if (dev->promiscuity == 0) { 8375 /* 8376 * Avoid overflow. 8377 * If inc causes overflow, untouch promisc and return error. 8378 */ 8379 if (inc < 0) 8380 dev->flags &= ~IFF_PROMISC; 8381 else { 8382 dev->promiscuity -= inc; 8383 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8384 return -EOVERFLOW; 8385 } 8386 } 8387 if (dev->flags != old_flags) { 8388 netdev_info(dev, "%s promiscuous mode\n", 8389 dev->flags & IFF_PROMISC ? "entered" : "left"); 8390 if (audit_enabled) { 8391 current_uid_gid(&uid, &gid); 8392 audit_log(audit_context(), GFP_ATOMIC, 8393 AUDIT_ANOM_PROMISCUOUS, 8394 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8395 dev->name, (dev->flags & IFF_PROMISC), 8396 (old_flags & IFF_PROMISC), 8397 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8398 from_kuid(&init_user_ns, uid), 8399 from_kgid(&init_user_ns, gid), 8400 audit_get_sessionid(current)); 8401 } 8402 8403 dev_change_rx_flags(dev, IFF_PROMISC); 8404 } 8405 if (notify) 8406 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8407 return 0; 8408 } 8409 8410 /** 8411 * dev_set_promiscuity - update promiscuity count on a device 8412 * @dev: device 8413 * @inc: modifier 8414 * 8415 * Add or remove promiscuity from a device. While the count in the device 8416 * remains above zero the interface remains promiscuous. Once it hits zero 8417 * the device reverts back to normal filtering operation. A negative inc 8418 * value is used to drop promiscuity on the device. 8419 * Return 0 if successful or a negative errno code on error. 8420 */ 8421 int dev_set_promiscuity(struct net_device *dev, int inc) 8422 { 8423 unsigned int old_flags = dev->flags; 8424 int err; 8425 8426 err = __dev_set_promiscuity(dev, inc, true); 8427 if (err < 0) 8428 return err; 8429 if (dev->flags != old_flags) 8430 dev_set_rx_mode(dev); 8431 return err; 8432 } 8433 EXPORT_SYMBOL(dev_set_promiscuity); 8434 8435 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8436 { 8437 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8438 8439 ASSERT_RTNL(); 8440 8441 dev->flags |= IFF_ALLMULTI; 8442 dev->allmulti += inc; 8443 if (dev->allmulti == 0) { 8444 /* 8445 * Avoid overflow. 8446 * If inc causes overflow, untouch allmulti and return error. 8447 */ 8448 if (inc < 0) 8449 dev->flags &= ~IFF_ALLMULTI; 8450 else { 8451 dev->allmulti -= inc; 8452 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8453 return -EOVERFLOW; 8454 } 8455 } 8456 if (dev->flags ^ old_flags) { 8457 netdev_info(dev, "%s allmulticast mode\n", 8458 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8459 dev_change_rx_flags(dev, IFF_ALLMULTI); 8460 dev_set_rx_mode(dev); 8461 if (notify) 8462 __dev_notify_flags(dev, old_flags, 8463 dev->gflags ^ old_gflags, 0, NULL); 8464 } 8465 return 0; 8466 } 8467 8468 /** 8469 * dev_set_allmulti - update allmulti count on a device 8470 * @dev: device 8471 * @inc: modifier 8472 * 8473 * Add or remove reception of all multicast frames to a device. While the 8474 * count in the device remains above zero the interface remains listening 8475 * to all interfaces. Once it hits zero the device reverts back to normal 8476 * filtering operation. A negative @inc value is used to drop the counter 8477 * when releasing a resource needing all multicasts. 8478 * Return 0 if successful or a negative errno code on error. 8479 */ 8480 8481 int dev_set_allmulti(struct net_device *dev, int inc) 8482 { 8483 return __dev_set_allmulti(dev, inc, true); 8484 } 8485 EXPORT_SYMBOL(dev_set_allmulti); 8486 8487 /* 8488 * Upload unicast and multicast address lists to device and 8489 * configure RX filtering. When the device doesn't support unicast 8490 * filtering it is put in promiscuous mode while unicast addresses 8491 * are present. 8492 */ 8493 void __dev_set_rx_mode(struct net_device *dev) 8494 { 8495 const struct net_device_ops *ops = dev->netdev_ops; 8496 8497 /* dev_open will call this function so the list will stay sane. */ 8498 if (!(dev->flags&IFF_UP)) 8499 return; 8500 8501 if (!netif_device_present(dev)) 8502 return; 8503 8504 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8505 /* Unicast addresses changes may only happen under the rtnl, 8506 * therefore calling __dev_set_promiscuity here is safe. 8507 */ 8508 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8509 __dev_set_promiscuity(dev, 1, false); 8510 dev->uc_promisc = true; 8511 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8512 __dev_set_promiscuity(dev, -1, false); 8513 dev->uc_promisc = false; 8514 } 8515 } 8516 8517 if (ops->ndo_set_rx_mode) 8518 ops->ndo_set_rx_mode(dev); 8519 } 8520 8521 void dev_set_rx_mode(struct net_device *dev) 8522 { 8523 netif_addr_lock_bh(dev); 8524 __dev_set_rx_mode(dev); 8525 netif_addr_unlock_bh(dev); 8526 } 8527 8528 /** 8529 * dev_get_flags - get flags reported to userspace 8530 * @dev: device 8531 * 8532 * Get the combination of flag bits exported through APIs to userspace. 8533 */ 8534 unsigned int dev_get_flags(const struct net_device *dev) 8535 { 8536 unsigned int flags; 8537 8538 flags = (dev->flags & ~(IFF_PROMISC | 8539 IFF_ALLMULTI | 8540 IFF_RUNNING | 8541 IFF_LOWER_UP | 8542 IFF_DORMANT)) | 8543 (dev->gflags & (IFF_PROMISC | 8544 IFF_ALLMULTI)); 8545 8546 if (netif_running(dev)) { 8547 if (netif_oper_up(dev)) 8548 flags |= IFF_RUNNING; 8549 if (netif_carrier_ok(dev)) 8550 flags |= IFF_LOWER_UP; 8551 if (netif_dormant(dev)) 8552 flags |= IFF_DORMANT; 8553 } 8554 8555 return flags; 8556 } 8557 EXPORT_SYMBOL(dev_get_flags); 8558 8559 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8560 struct netlink_ext_ack *extack) 8561 { 8562 unsigned int old_flags = dev->flags; 8563 int ret; 8564 8565 ASSERT_RTNL(); 8566 8567 /* 8568 * Set the flags on our device. 8569 */ 8570 8571 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8572 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8573 IFF_AUTOMEDIA)) | 8574 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8575 IFF_ALLMULTI)); 8576 8577 /* 8578 * Load in the correct multicast list now the flags have changed. 8579 */ 8580 8581 if ((old_flags ^ flags) & IFF_MULTICAST) 8582 dev_change_rx_flags(dev, IFF_MULTICAST); 8583 8584 dev_set_rx_mode(dev); 8585 8586 /* 8587 * Have we downed the interface. We handle IFF_UP ourselves 8588 * according to user attempts to set it, rather than blindly 8589 * setting it. 8590 */ 8591 8592 ret = 0; 8593 if ((old_flags ^ flags) & IFF_UP) { 8594 if (old_flags & IFF_UP) 8595 __dev_close(dev); 8596 else 8597 ret = __dev_open(dev, extack); 8598 } 8599 8600 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8601 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8602 unsigned int old_flags = dev->flags; 8603 8604 dev->gflags ^= IFF_PROMISC; 8605 8606 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8607 if (dev->flags != old_flags) 8608 dev_set_rx_mode(dev); 8609 } 8610 8611 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8612 * is important. Some (broken) drivers set IFF_PROMISC, when 8613 * IFF_ALLMULTI is requested not asking us and not reporting. 8614 */ 8615 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8616 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8617 8618 dev->gflags ^= IFF_ALLMULTI; 8619 __dev_set_allmulti(dev, inc, false); 8620 } 8621 8622 return ret; 8623 } 8624 8625 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8626 unsigned int gchanges, u32 portid, 8627 const struct nlmsghdr *nlh) 8628 { 8629 unsigned int changes = dev->flags ^ old_flags; 8630 8631 if (gchanges) 8632 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8633 8634 if (changes & IFF_UP) { 8635 if (dev->flags & IFF_UP) 8636 call_netdevice_notifiers(NETDEV_UP, dev); 8637 else 8638 call_netdevice_notifiers(NETDEV_DOWN, dev); 8639 } 8640 8641 if (dev->flags & IFF_UP && 8642 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8643 struct netdev_notifier_change_info change_info = { 8644 .info = { 8645 .dev = dev, 8646 }, 8647 .flags_changed = changes, 8648 }; 8649 8650 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8651 } 8652 } 8653 8654 /** 8655 * dev_change_flags - change device settings 8656 * @dev: device 8657 * @flags: device state flags 8658 * @extack: netlink extended ack 8659 * 8660 * Change settings on device based state flags. The flags are 8661 * in the userspace exported format. 8662 */ 8663 int dev_change_flags(struct net_device *dev, unsigned int flags, 8664 struct netlink_ext_ack *extack) 8665 { 8666 int ret; 8667 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8668 8669 ret = __dev_change_flags(dev, flags, extack); 8670 if (ret < 0) 8671 return ret; 8672 8673 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8674 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8675 return ret; 8676 } 8677 EXPORT_SYMBOL(dev_change_flags); 8678 8679 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8680 { 8681 const struct net_device_ops *ops = dev->netdev_ops; 8682 8683 if (ops->ndo_change_mtu) 8684 return ops->ndo_change_mtu(dev, new_mtu); 8685 8686 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8687 WRITE_ONCE(dev->mtu, new_mtu); 8688 return 0; 8689 } 8690 EXPORT_SYMBOL(__dev_set_mtu); 8691 8692 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8693 struct netlink_ext_ack *extack) 8694 { 8695 /* MTU must be positive, and in range */ 8696 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8697 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8698 return -EINVAL; 8699 } 8700 8701 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8702 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8703 return -EINVAL; 8704 } 8705 return 0; 8706 } 8707 8708 /** 8709 * dev_set_mtu_ext - Change maximum transfer unit 8710 * @dev: device 8711 * @new_mtu: new transfer unit 8712 * @extack: netlink extended ack 8713 * 8714 * Change the maximum transfer size of the network device. 8715 */ 8716 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8717 struct netlink_ext_ack *extack) 8718 { 8719 int err, orig_mtu; 8720 8721 if (new_mtu == dev->mtu) 8722 return 0; 8723 8724 err = dev_validate_mtu(dev, new_mtu, extack); 8725 if (err) 8726 return err; 8727 8728 if (!netif_device_present(dev)) 8729 return -ENODEV; 8730 8731 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8732 err = notifier_to_errno(err); 8733 if (err) 8734 return err; 8735 8736 orig_mtu = dev->mtu; 8737 err = __dev_set_mtu(dev, new_mtu); 8738 8739 if (!err) { 8740 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8741 orig_mtu); 8742 err = notifier_to_errno(err); 8743 if (err) { 8744 /* setting mtu back and notifying everyone again, 8745 * so that they have a chance to revert changes. 8746 */ 8747 __dev_set_mtu(dev, orig_mtu); 8748 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8749 new_mtu); 8750 } 8751 } 8752 return err; 8753 } 8754 8755 int dev_set_mtu(struct net_device *dev, int new_mtu) 8756 { 8757 struct netlink_ext_ack extack; 8758 int err; 8759 8760 memset(&extack, 0, sizeof(extack)); 8761 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8762 if (err && extack._msg) 8763 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8764 return err; 8765 } 8766 EXPORT_SYMBOL(dev_set_mtu); 8767 8768 /** 8769 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8770 * @dev: device 8771 * @new_len: new tx queue length 8772 */ 8773 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8774 { 8775 unsigned int orig_len = dev->tx_queue_len; 8776 int res; 8777 8778 if (new_len != (unsigned int)new_len) 8779 return -ERANGE; 8780 8781 if (new_len != orig_len) { 8782 dev->tx_queue_len = new_len; 8783 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8784 res = notifier_to_errno(res); 8785 if (res) 8786 goto err_rollback; 8787 res = dev_qdisc_change_tx_queue_len(dev); 8788 if (res) 8789 goto err_rollback; 8790 } 8791 8792 return 0; 8793 8794 err_rollback: 8795 netdev_err(dev, "refused to change device tx_queue_len\n"); 8796 dev->tx_queue_len = orig_len; 8797 return res; 8798 } 8799 8800 /** 8801 * dev_set_group - Change group this device belongs to 8802 * @dev: device 8803 * @new_group: group this device should belong to 8804 */ 8805 void dev_set_group(struct net_device *dev, int new_group) 8806 { 8807 dev->group = new_group; 8808 } 8809 8810 /** 8811 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8812 * @dev: device 8813 * @addr: new address 8814 * @extack: netlink extended ack 8815 */ 8816 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8817 struct netlink_ext_ack *extack) 8818 { 8819 struct netdev_notifier_pre_changeaddr_info info = { 8820 .info.dev = dev, 8821 .info.extack = extack, 8822 .dev_addr = addr, 8823 }; 8824 int rc; 8825 8826 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8827 return notifier_to_errno(rc); 8828 } 8829 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8830 8831 /** 8832 * dev_set_mac_address - Change Media Access Control Address 8833 * @dev: device 8834 * @sa: new address 8835 * @extack: netlink extended ack 8836 * 8837 * Change the hardware (MAC) address of the device 8838 */ 8839 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8840 struct netlink_ext_ack *extack) 8841 { 8842 const struct net_device_ops *ops = dev->netdev_ops; 8843 int err; 8844 8845 if (!ops->ndo_set_mac_address) 8846 return -EOPNOTSUPP; 8847 if (sa->sa_family != dev->type) 8848 return -EINVAL; 8849 if (!netif_device_present(dev)) 8850 return -ENODEV; 8851 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8852 if (err) 8853 return err; 8854 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8855 err = ops->ndo_set_mac_address(dev, sa); 8856 if (err) 8857 return err; 8858 } 8859 dev->addr_assign_type = NET_ADDR_SET; 8860 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8861 add_device_randomness(dev->dev_addr, dev->addr_len); 8862 return 0; 8863 } 8864 EXPORT_SYMBOL(dev_set_mac_address); 8865 8866 static DECLARE_RWSEM(dev_addr_sem); 8867 8868 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8869 struct netlink_ext_ack *extack) 8870 { 8871 int ret; 8872 8873 down_write(&dev_addr_sem); 8874 ret = dev_set_mac_address(dev, sa, extack); 8875 up_write(&dev_addr_sem); 8876 return ret; 8877 } 8878 EXPORT_SYMBOL(dev_set_mac_address_user); 8879 8880 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8881 { 8882 size_t size = sizeof(sa->sa_data_min); 8883 struct net_device *dev; 8884 int ret = 0; 8885 8886 down_read(&dev_addr_sem); 8887 rcu_read_lock(); 8888 8889 dev = dev_get_by_name_rcu(net, dev_name); 8890 if (!dev) { 8891 ret = -ENODEV; 8892 goto unlock; 8893 } 8894 if (!dev->addr_len) 8895 memset(sa->sa_data, 0, size); 8896 else 8897 memcpy(sa->sa_data, dev->dev_addr, 8898 min_t(size_t, size, dev->addr_len)); 8899 sa->sa_family = dev->type; 8900 8901 unlock: 8902 rcu_read_unlock(); 8903 up_read(&dev_addr_sem); 8904 return ret; 8905 } 8906 EXPORT_SYMBOL(dev_get_mac_address); 8907 8908 /** 8909 * dev_change_carrier - Change device carrier 8910 * @dev: device 8911 * @new_carrier: new value 8912 * 8913 * Change device carrier 8914 */ 8915 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8916 { 8917 const struct net_device_ops *ops = dev->netdev_ops; 8918 8919 if (!ops->ndo_change_carrier) 8920 return -EOPNOTSUPP; 8921 if (!netif_device_present(dev)) 8922 return -ENODEV; 8923 return ops->ndo_change_carrier(dev, new_carrier); 8924 } 8925 8926 /** 8927 * dev_get_phys_port_id - Get device physical port ID 8928 * @dev: device 8929 * @ppid: port ID 8930 * 8931 * Get device physical port ID 8932 */ 8933 int dev_get_phys_port_id(struct net_device *dev, 8934 struct netdev_phys_item_id *ppid) 8935 { 8936 const struct net_device_ops *ops = dev->netdev_ops; 8937 8938 if (!ops->ndo_get_phys_port_id) 8939 return -EOPNOTSUPP; 8940 return ops->ndo_get_phys_port_id(dev, ppid); 8941 } 8942 8943 /** 8944 * dev_get_phys_port_name - Get device physical port name 8945 * @dev: device 8946 * @name: port name 8947 * @len: limit of bytes to copy to name 8948 * 8949 * Get device physical port name 8950 */ 8951 int dev_get_phys_port_name(struct net_device *dev, 8952 char *name, size_t len) 8953 { 8954 const struct net_device_ops *ops = dev->netdev_ops; 8955 int err; 8956 8957 if (ops->ndo_get_phys_port_name) { 8958 err = ops->ndo_get_phys_port_name(dev, name, len); 8959 if (err != -EOPNOTSUPP) 8960 return err; 8961 } 8962 return devlink_compat_phys_port_name_get(dev, name, len); 8963 } 8964 8965 /** 8966 * dev_get_port_parent_id - Get the device's port parent identifier 8967 * @dev: network device 8968 * @ppid: pointer to a storage for the port's parent identifier 8969 * @recurse: allow/disallow recursion to lower devices 8970 * 8971 * Get the devices's port parent identifier 8972 */ 8973 int dev_get_port_parent_id(struct net_device *dev, 8974 struct netdev_phys_item_id *ppid, 8975 bool recurse) 8976 { 8977 const struct net_device_ops *ops = dev->netdev_ops; 8978 struct netdev_phys_item_id first = { }; 8979 struct net_device *lower_dev; 8980 struct list_head *iter; 8981 int err; 8982 8983 if (ops->ndo_get_port_parent_id) { 8984 err = ops->ndo_get_port_parent_id(dev, ppid); 8985 if (err != -EOPNOTSUPP) 8986 return err; 8987 } 8988 8989 err = devlink_compat_switch_id_get(dev, ppid); 8990 if (!recurse || err != -EOPNOTSUPP) 8991 return err; 8992 8993 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8994 err = dev_get_port_parent_id(lower_dev, ppid, true); 8995 if (err) 8996 break; 8997 if (!first.id_len) 8998 first = *ppid; 8999 else if (memcmp(&first, ppid, sizeof(*ppid))) 9000 return -EOPNOTSUPP; 9001 } 9002 9003 return err; 9004 } 9005 EXPORT_SYMBOL(dev_get_port_parent_id); 9006 9007 /** 9008 * netdev_port_same_parent_id - Indicate if two network devices have 9009 * the same port parent identifier 9010 * @a: first network device 9011 * @b: second network device 9012 */ 9013 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9014 { 9015 struct netdev_phys_item_id a_id = { }; 9016 struct netdev_phys_item_id b_id = { }; 9017 9018 if (dev_get_port_parent_id(a, &a_id, true) || 9019 dev_get_port_parent_id(b, &b_id, true)) 9020 return false; 9021 9022 return netdev_phys_item_id_same(&a_id, &b_id); 9023 } 9024 EXPORT_SYMBOL(netdev_port_same_parent_id); 9025 9026 /** 9027 * dev_change_proto_down - set carrier according to proto_down. 9028 * 9029 * @dev: device 9030 * @proto_down: new value 9031 */ 9032 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9033 { 9034 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9035 return -EOPNOTSUPP; 9036 if (!netif_device_present(dev)) 9037 return -ENODEV; 9038 if (proto_down) 9039 netif_carrier_off(dev); 9040 else 9041 netif_carrier_on(dev); 9042 dev->proto_down = proto_down; 9043 return 0; 9044 } 9045 9046 /** 9047 * dev_change_proto_down_reason - proto down reason 9048 * 9049 * @dev: device 9050 * @mask: proto down mask 9051 * @value: proto down value 9052 */ 9053 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9054 u32 value) 9055 { 9056 int b; 9057 9058 if (!mask) { 9059 dev->proto_down_reason = value; 9060 } else { 9061 for_each_set_bit(b, &mask, 32) { 9062 if (value & (1 << b)) 9063 dev->proto_down_reason |= BIT(b); 9064 else 9065 dev->proto_down_reason &= ~BIT(b); 9066 } 9067 } 9068 } 9069 9070 struct bpf_xdp_link { 9071 struct bpf_link link; 9072 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9073 int flags; 9074 }; 9075 9076 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9077 { 9078 if (flags & XDP_FLAGS_HW_MODE) 9079 return XDP_MODE_HW; 9080 if (flags & XDP_FLAGS_DRV_MODE) 9081 return XDP_MODE_DRV; 9082 if (flags & XDP_FLAGS_SKB_MODE) 9083 return XDP_MODE_SKB; 9084 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9085 } 9086 9087 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9088 { 9089 switch (mode) { 9090 case XDP_MODE_SKB: 9091 return generic_xdp_install; 9092 case XDP_MODE_DRV: 9093 case XDP_MODE_HW: 9094 return dev->netdev_ops->ndo_bpf; 9095 default: 9096 return NULL; 9097 } 9098 } 9099 9100 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9101 enum bpf_xdp_mode mode) 9102 { 9103 return dev->xdp_state[mode].link; 9104 } 9105 9106 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9107 enum bpf_xdp_mode mode) 9108 { 9109 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9110 9111 if (link) 9112 return link->link.prog; 9113 return dev->xdp_state[mode].prog; 9114 } 9115 9116 u8 dev_xdp_prog_count(struct net_device *dev) 9117 { 9118 u8 count = 0; 9119 int i; 9120 9121 for (i = 0; i < __MAX_XDP_MODE; i++) 9122 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9123 count++; 9124 return count; 9125 } 9126 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9127 9128 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9129 { 9130 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9131 9132 return prog ? prog->aux->id : 0; 9133 } 9134 9135 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9136 struct bpf_xdp_link *link) 9137 { 9138 dev->xdp_state[mode].link = link; 9139 dev->xdp_state[mode].prog = NULL; 9140 } 9141 9142 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9143 struct bpf_prog *prog) 9144 { 9145 dev->xdp_state[mode].link = NULL; 9146 dev->xdp_state[mode].prog = prog; 9147 } 9148 9149 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9150 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9151 u32 flags, struct bpf_prog *prog) 9152 { 9153 struct netdev_bpf xdp; 9154 int err; 9155 9156 memset(&xdp, 0, sizeof(xdp)); 9157 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9158 xdp.extack = extack; 9159 xdp.flags = flags; 9160 xdp.prog = prog; 9161 9162 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9163 * "moved" into driver), so they don't increment it on their own, but 9164 * they do decrement refcnt when program is detached or replaced. 9165 * Given net_device also owns link/prog, we need to bump refcnt here 9166 * to prevent drivers from underflowing it. 9167 */ 9168 if (prog) 9169 bpf_prog_inc(prog); 9170 err = bpf_op(dev, &xdp); 9171 if (err) { 9172 if (prog) 9173 bpf_prog_put(prog); 9174 return err; 9175 } 9176 9177 if (mode != XDP_MODE_HW) 9178 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9179 9180 return 0; 9181 } 9182 9183 static void dev_xdp_uninstall(struct net_device *dev) 9184 { 9185 struct bpf_xdp_link *link; 9186 struct bpf_prog *prog; 9187 enum bpf_xdp_mode mode; 9188 bpf_op_t bpf_op; 9189 9190 ASSERT_RTNL(); 9191 9192 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9193 prog = dev_xdp_prog(dev, mode); 9194 if (!prog) 9195 continue; 9196 9197 bpf_op = dev_xdp_bpf_op(dev, mode); 9198 if (!bpf_op) 9199 continue; 9200 9201 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9202 9203 /* auto-detach link from net device */ 9204 link = dev_xdp_link(dev, mode); 9205 if (link) 9206 link->dev = NULL; 9207 else 9208 bpf_prog_put(prog); 9209 9210 dev_xdp_set_link(dev, mode, NULL); 9211 } 9212 } 9213 9214 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9215 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9216 struct bpf_prog *old_prog, u32 flags) 9217 { 9218 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9219 struct bpf_prog *cur_prog; 9220 struct net_device *upper; 9221 struct list_head *iter; 9222 enum bpf_xdp_mode mode; 9223 bpf_op_t bpf_op; 9224 int err; 9225 9226 ASSERT_RTNL(); 9227 9228 /* either link or prog attachment, never both */ 9229 if (link && (new_prog || old_prog)) 9230 return -EINVAL; 9231 /* link supports only XDP mode flags */ 9232 if (link && (flags & ~XDP_FLAGS_MODES)) { 9233 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9234 return -EINVAL; 9235 } 9236 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9237 if (num_modes > 1) { 9238 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9239 return -EINVAL; 9240 } 9241 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9242 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9243 NL_SET_ERR_MSG(extack, 9244 "More than one program loaded, unset mode is ambiguous"); 9245 return -EINVAL; 9246 } 9247 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9248 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9249 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9250 return -EINVAL; 9251 } 9252 9253 mode = dev_xdp_mode(dev, flags); 9254 /* can't replace attached link */ 9255 if (dev_xdp_link(dev, mode)) { 9256 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9257 return -EBUSY; 9258 } 9259 9260 /* don't allow if an upper device already has a program */ 9261 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9262 if (dev_xdp_prog_count(upper) > 0) { 9263 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9264 return -EEXIST; 9265 } 9266 } 9267 9268 cur_prog = dev_xdp_prog(dev, mode); 9269 /* can't replace attached prog with link */ 9270 if (link && cur_prog) { 9271 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9272 return -EBUSY; 9273 } 9274 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9275 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9276 return -EEXIST; 9277 } 9278 9279 /* put effective new program into new_prog */ 9280 if (link) 9281 new_prog = link->link.prog; 9282 9283 if (new_prog) { 9284 bool offload = mode == XDP_MODE_HW; 9285 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9286 ? XDP_MODE_DRV : XDP_MODE_SKB; 9287 9288 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9289 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9290 return -EBUSY; 9291 } 9292 if (!offload && dev_xdp_prog(dev, other_mode)) { 9293 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9294 return -EEXIST; 9295 } 9296 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9297 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9298 return -EINVAL; 9299 } 9300 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9301 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9302 return -EINVAL; 9303 } 9304 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9305 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9306 return -EINVAL; 9307 } 9308 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9309 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9310 return -EINVAL; 9311 } 9312 } 9313 9314 /* don't call drivers if the effective program didn't change */ 9315 if (new_prog != cur_prog) { 9316 bpf_op = dev_xdp_bpf_op(dev, mode); 9317 if (!bpf_op) { 9318 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9319 return -EOPNOTSUPP; 9320 } 9321 9322 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9323 if (err) 9324 return err; 9325 } 9326 9327 if (link) 9328 dev_xdp_set_link(dev, mode, link); 9329 else 9330 dev_xdp_set_prog(dev, mode, new_prog); 9331 if (cur_prog) 9332 bpf_prog_put(cur_prog); 9333 9334 return 0; 9335 } 9336 9337 static int dev_xdp_attach_link(struct net_device *dev, 9338 struct netlink_ext_ack *extack, 9339 struct bpf_xdp_link *link) 9340 { 9341 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9342 } 9343 9344 static int dev_xdp_detach_link(struct net_device *dev, 9345 struct netlink_ext_ack *extack, 9346 struct bpf_xdp_link *link) 9347 { 9348 enum bpf_xdp_mode mode; 9349 bpf_op_t bpf_op; 9350 9351 ASSERT_RTNL(); 9352 9353 mode = dev_xdp_mode(dev, link->flags); 9354 if (dev_xdp_link(dev, mode) != link) 9355 return -EINVAL; 9356 9357 bpf_op = dev_xdp_bpf_op(dev, mode); 9358 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9359 dev_xdp_set_link(dev, mode, NULL); 9360 return 0; 9361 } 9362 9363 static void bpf_xdp_link_release(struct bpf_link *link) 9364 { 9365 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9366 9367 rtnl_lock(); 9368 9369 /* if racing with net_device's tear down, xdp_link->dev might be 9370 * already NULL, in which case link was already auto-detached 9371 */ 9372 if (xdp_link->dev) { 9373 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9374 xdp_link->dev = NULL; 9375 } 9376 9377 rtnl_unlock(); 9378 } 9379 9380 static int bpf_xdp_link_detach(struct bpf_link *link) 9381 { 9382 bpf_xdp_link_release(link); 9383 return 0; 9384 } 9385 9386 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9387 { 9388 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9389 9390 kfree(xdp_link); 9391 } 9392 9393 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9394 struct seq_file *seq) 9395 { 9396 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9397 u32 ifindex = 0; 9398 9399 rtnl_lock(); 9400 if (xdp_link->dev) 9401 ifindex = xdp_link->dev->ifindex; 9402 rtnl_unlock(); 9403 9404 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9405 } 9406 9407 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9408 struct bpf_link_info *info) 9409 { 9410 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9411 u32 ifindex = 0; 9412 9413 rtnl_lock(); 9414 if (xdp_link->dev) 9415 ifindex = xdp_link->dev->ifindex; 9416 rtnl_unlock(); 9417 9418 info->xdp.ifindex = ifindex; 9419 return 0; 9420 } 9421 9422 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9423 struct bpf_prog *old_prog) 9424 { 9425 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9426 enum bpf_xdp_mode mode; 9427 bpf_op_t bpf_op; 9428 int err = 0; 9429 9430 rtnl_lock(); 9431 9432 /* link might have been auto-released already, so fail */ 9433 if (!xdp_link->dev) { 9434 err = -ENOLINK; 9435 goto out_unlock; 9436 } 9437 9438 if (old_prog && link->prog != old_prog) { 9439 err = -EPERM; 9440 goto out_unlock; 9441 } 9442 old_prog = link->prog; 9443 if (old_prog->type != new_prog->type || 9444 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9445 err = -EINVAL; 9446 goto out_unlock; 9447 } 9448 9449 if (old_prog == new_prog) { 9450 /* no-op, don't disturb drivers */ 9451 bpf_prog_put(new_prog); 9452 goto out_unlock; 9453 } 9454 9455 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9456 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9457 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9458 xdp_link->flags, new_prog); 9459 if (err) 9460 goto out_unlock; 9461 9462 old_prog = xchg(&link->prog, new_prog); 9463 bpf_prog_put(old_prog); 9464 9465 out_unlock: 9466 rtnl_unlock(); 9467 return err; 9468 } 9469 9470 static const struct bpf_link_ops bpf_xdp_link_lops = { 9471 .release = bpf_xdp_link_release, 9472 .dealloc = bpf_xdp_link_dealloc, 9473 .detach = bpf_xdp_link_detach, 9474 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9475 .fill_link_info = bpf_xdp_link_fill_link_info, 9476 .update_prog = bpf_xdp_link_update, 9477 }; 9478 9479 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9480 { 9481 struct net *net = current->nsproxy->net_ns; 9482 struct bpf_link_primer link_primer; 9483 struct netlink_ext_ack extack = {}; 9484 struct bpf_xdp_link *link; 9485 struct net_device *dev; 9486 int err, fd; 9487 9488 rtnl_lock(); 9489 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9490 if (!dev) { 9491 rtnl_unlock(); 9492 return -EINVAL; 9493 } 9494 9495 link = kzalloc(sizeof(*link), GFP_USER); 9496 if (!link) { 9497 err = -ENOMEM; 9498 goto unlock; 9499 } 9500 9501 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9502 link->dev = dev; 9503 link->flags = attr->link_create.flags; 9504 9505 err = bpf_link_prime(&link->link, &link_primer); 9506 if (err) { 9507 kfree(link); 9508 goto unlock; 9509 } 9510 9511 err = dev_xdp_attach_link(dev, &extack, link); 9512 rtnl_unlock(); 9513 9514 if (err) { 9515 link->dev = NULL; 9516 bpf_link_cleanup(&link_primer); 9517 trace_bpf_xdp_link_attach_failed(extack._msg); 9518 goto out_put_dev; 9519 } 9520 9521 fd = bpf_link_settle(&link_primer); 9522 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9523 dev_put(dev); 9524 return fd; 9525 9526 unlock: 9527 rtnl_unlock(); 9528 9529 out_put_dev: 9530 dev_put(dev); 9531 return err; 9532 } 9533 9534 /** 9535 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9536 * @dev: device 9537 * @extack: netlink extended ack 9538 * @fd: new program fd or negative value to clear 9539 * @expected_fd: old program fd that userspace expects to replace or clear 9540 * @flags: xdp-related flags 9541 * 9542 * Set or clear a bpf program for a device 9543 */ 9544 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9545 int fd, int expected_fd, u32 flags) 9546 { 9547 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9548 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9549 int err; 9550 9551 ASSERT_RTNL(); 9552 9553 if (fd >= 0) { 9554 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9555 mode != XDP_MODE_SKB); 9556 if (IS_ERR(new_prog)) 9557 return PTR_ERR(new_prog); 9558 } 9559 9560 if (expected_fd >= 0) { 9561 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9562 mode != XDP_MODE_SKB); 9563 if (IS_ERR(old_prog)) { 9564 err = PTR_ERR(old_prog); 9565 old_prog = NULL; 9566 goto err_out; 9567 } 9568 } 9569 9570 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9571 9572 err_out: 9573 if (err && new_prog) 9574 bpf_prog_put(new_prog); 9575 if (old_prog) 9576 bpf_prog_put(old_prog); 9577 return err; 9578 } 9579 9580 /** 9581 * dev_index_reserve() - allocate an ifindex in a namespace 9582 * @net: the applicable net namespace 9583 * @ifindex: requested ifindex, pass %0 to get one allocated 9584 * 9585 * Allocate a ifindex for a new device. Caller must either use the ifindex 9586 * to store the device (via list_netdevice()) or call dev_index_release() 9587 * to give the index up. 9588 * 9589 * Return: a suitable unique value for a new device interface number or -errno. 9590 */ 9591 static int dev_index_reserve(struct net *net, u32 ifindex) 9592 { 9593 int err; 9594 9595 if (ifindex > INT_MAX) { 9596 DEBUG_NET_WARN_ON_ONCE(1); 9597 return -EINVAL; 9598 } 9599 9600 if (!ifindex) 9601 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9602 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9603 else 9604 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9605 if (err < 0) 9606 return err; 9607 9608 return ifindex; 9609 } 9610 9611 static void dev_index_release(struct net *net, int ifindex) 9612 { 9613 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9614 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9615 } 9616 9617 /* Delayed registration/unregisteration */ 9618 LIST_HEAD(net_todo_list); 9619 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9620 9621 static void net_set_todo(struct net_device *dev) 9622 { 9623 list_add_tail(&dev->todo_list, &net_todo_list); 9624 atomic_inc(&dev_net(dev)->dev_unreg_count); 9625 } 9626 9627 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9628 struct net_device *upper, netdev_features_t features) 9629 { 9630 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9631 netdev_features_t feature; 9632 int feature_bit; 9633 9634 for_each_netdev_feature(upper_disables, feature_bit) { 9635 feature = __NETIF_F_BIT(feature_bit); 9636 if (!(upper->wanted_features & feature) 9637 && (features & feature)) { 9638 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9639 &feature, upper->name); 9640 features &= ~feature; 9641 } 9642 } 9643 9644 return features; 9645 } 9646 9647 static void netdev_sync_lower_features(struct net_device *upper, 9648 struct net_device *lower, netdev_features_t features) 9649 { 9650 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9651 netdev_features_t feature; 9652 int feature_bit; 9653 9654 for_each_netdev_feature(upper_disables, feature_bit) { 9655 feature = __NETIF_F_BIT(feature_bit); 9656 if (!(features & feature) && (lower->features & feature)) { 9657 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9658 &feature, lower->name); 9659 lower->wanted_features &= ~feature; 9660 __netdev_update_features(lower); 9661 9662 if (unlikely(lower->features & feature)) 9663 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9664 &feature, lower->name); 9665 else 9666 netdev_features_change(lower); 9667 } 9668 } 9669 } 9670 9671 static netdev_features_t netdev_fix_features(struct net_device *dev, 9672 netdev_features_t features) 9673 { 9674 /* Fix illegal checksum combinations */ 9675 if ((features & NETIF_F_HW_CSUM) && 9676 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9677 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9678 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9679 } 9680 9681 /* TSO requires that SG is present as well. */ 9682 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9683 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9684 features &= ~NETIF_F_ALL_TSO; 9685 } 9686 9687 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9688 !(features & NETIF_F_IP_CSUM)) { 9689 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9690 features &= ~NETIF_F_TSO; 9691 features &= ~NETIF_F_TSO_ECN; 9692 } 9693 9694 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9695 !(features & NETIF_F_IPV6_CSUM)) { 9696 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9697 features &= ~NETIF_F_TSO6; 9698 } 9699 9700 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9701 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9702 features &= ~NETIF_F_TSO_MANGLEID; 9703 9704 /* TSO ECN requires that TSO is present as well. */ 9705 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9706 features &= ~NETIF_F_TSO_ECN; 9707 9708 /* Software GSO depends on SG. */ 9709 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9710 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9711 features &= ~NETIF_F_GSO; 9712 } 9713 9714 /* GSO partial features require GSO partial be set */ 9715 if ((features & dev->gso_partial_features) && 9716 !(features & NETIF_F_GSO_PARTIAL)) { 9717 netdev_dbg(dev, 9718 "Dropping partially supported GSO features since no GSO partial.\n"); 9719 features &= ~dev->gso_partial_features; 9720 } 9721 9722 if (!(features & NETIF_F_RXCSUM)) { 9723 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9724 * successfully merged by hardware must also have the 9725 * checksum verified by hardware. If the user does not 9726 * want to enable RXCSUM, logically, we should disable GRO_HW. 9727 */ 9728 if (features & NETIF_F_GRO_HW) { 9729 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9730 features &= ~NETIF_F_GRO_HW; 9731 } 9732 } 9733 9734 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9735 if (features & NETIF_F_RXFCS) { 9736 if (features & NETIF_F_LRO) { 9737 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9738 features &= ~NETIF_F_LRO; 9739 } 9740 9741 if (features & NETIF_F_GRO_HW) { 9742 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9743 features &= ~NETIF_F_GRO_HW; 9744 } 9745 } 9746 9747 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9748 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9749 features &= ~NETIF_F_LRO; 9750 } 9751 9752 if (features & NETIF_F_HW_TLS_TX) { 9753 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9754 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9755 bool hw_csum = features & NETIF_F_HW_CSUM; 9756 9757 if (!ip_csum && !hw_csum) { 9758 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9759 features &= ~NETIF_F_HW_TLS_TX; 9760 } 9761 } 9762 9763 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9764 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9765 features &= ~NETIF_F_HW_TLS_RX; 9766 } 9767 9768 return features; 9769 } 9770 9771 int __netdev_update_features(struct net_device *dev) 9772 { 9773 struct net_device *upper, *lower; 9774 netdev_features_t features; 9775 struct list_head *iter; 9776 int err = -1; 9777 9778 ASSERT_RTNL(); 9779 9780 features = netdev_get_wanted_features(dev); 9781 9782 if (dev->netdev_ops->ndo_fix_features) 9783 features = dev->netdev_ops->ndo_fix_features(dev, features); 9784 9785 /* driver might be less strict about feature dependencies */ 9786 features = netdev_fix_features(dev, features); 9787 9788 /* some features can't be enabled if they're off on an upper device */ 9789 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9790 features = netdev_sync_upper_features(dev, upper, features); 9791 9792 if (dev->features == features) 9793 goto sync_lower; 9794 9795 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9796 &dev->features, &features); 9797 9798 if (dev->netdev_ops->ndo_set_features) 9799 err = dev->netdev_ops->ndo_set_features(dev, features); 9800 else 9801 err = 0; 9802 9803 if (unlikely(err < 0)) { 9804 netdev_err(dev, 9805 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9806 err, &features, &dev->features); 9807 /* return non-0 since some features might have changed and 9808 * it's better to fire a spurious notification than miss it 9809 */ 9810 return -1; 9811 } 9812 9813 sync_lower: 9814 /* some features must be disabled on lower devices when disabled 9815 * on an upper device (think: bonding master or bridge) 9816 */ 9817 netdev_for_each_lower_dev(dev, lower, iter) 9818 netdev_sync_lower_features(dev, lower, features); 9819 9820 if (!err) { 9821 netdev_features_t diff = features ^ dev->features; 9822 9823 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9824 /* udp_tunnel_{get,drop}_rx_info both need 9825 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9826 * device, or they won't do anything. 9827 * Thus we need to update dev->features 9828 * *before* calling udp_tunnel_get_rx_info, 9829 * but *after* calling udp_tunnel_drop_rx_info. 9830 */ 9831 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9832 dev->features = features; 9833 udp_tunnel_get_rx_info(dev); 9834 } else { 9835 udp_tunnel_drop_rx_info(dev); 9836 } 9837 } 9838 9839 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9840 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9841 dev->features = features; 9842 err |= vlan_get_rx_ctag_filter_info(dev); 9843 } else { 9844 vlan_drop_rx_ctag_filter_info(dev); 9845 } 9846 } 9847 9848 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9849 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9850 dev->features = features; 9851 err |= vlan_get_rx_stag_filter_info(dev); 9852 } else { 9853 vlan_drop_rx_stag_filter_info(dev); 9854 } 9855 } 9856 9857 dev->features = features; 9858 } 9859 9860 return err < 0 ? 0 : 1; 9861 } 9862 9863 /** 9864 * netdev_update_features - recalculate device features 9865 * @dev: the device to check 9866 * 9867 * Recalculate dev->features set and send notifications if it 9868 * has changed. Should be called after driver or hardware dependent 9869 * conditions might have changed that influence the features. 9870 */ 9871 void netdev_update_features(struct net_device *dev) 9872 { 9873 if (__netdev_update_features(dev)) 9874 netdev_features_change(dev); 9875 } 9876 EXPORT_SYMBOL(netdev_update_features); 9877 9878 /** 9879 * netdev_change_features - recalculate device features 9880 * @dev: the device to check 9881 * 9882 * Recalculate dev->features set and send notifications even 9883 * if they have not changed. Should be called instead of 9884 * netdev_update_features() if also dev->vlan_features might 9885 * have changed to allow the changes to be propagated to stacked 9886 * VLAN devices. 9887 */ 9888 void netdev_change_features(struct net_device *dev) 9889 { 9890 __netdev_update_features(dev); 9891 netdev_features_change(dev); 9892 } 9893 EXPORT_SYMBOL(netdev_change_features); 9894 9895 /** 9896 * netif_stacked_transfer_operstate - transfer operstate 9897 * @rootdev: the root or lower level device to transfer state from 9898 * @dev: the device to transfer operstate to 9899 * 9900 * Transfer operational state from root to device. This is normally 9901 * called when a stacking relationship exists between the root 9902 * device and the device(a leaf device). 9903 */ 9904 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9905 struct net_device *dev) 9906 { 9907 if (rootdev->operstate == IF_OPER_DORMANT) 9908 netif_dormant_on(dev); 9909 else 9910 netif_dormant_off(dev); 9911 9912 if (rootdev->operstate == IF_OPER_TESTING) 9913 netif_testing_on(dev); 9914 else 9915 netif_testing_off(dev); 9916 9917 if (netif_carrier_ok(rootdev)) 9918 netif_carrier_on(dev); 9919 else 9920 netif_carrier_off(dev); 9921 } 9922 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9923 9924 static int netif_alloc_rx_queues(struct net_device *dev) 9925 { 9926 unsigned int i, count = dev->num_rx_queues; 9927 struct netdev_rx_queue *rx; 9928 size_t sz = count * sizeof(*rx); 9929 int err = 0; 9930 9931 BUG_ON(count < 1); 9932 9933 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9934 if (!rx) 9935 return -ENOMEM; 9936 9937 dev->_rx = rx; 9938 9939 for (i = 0; i < count; i++) { 9940 rx[i].dev = dev; 9941 9942 /* XDP RX-queue setup */ 9943 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9944 if (err < 0) 9945 goto err_rxq_info; 9946 } 9947 return 0; 9948 9949 err_rxq_info: 9950 /* Rollback successful reg's and free other resources */ 9951 while (i--) 9952 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9953 kvfree(dev->_rx); 9954 dev->_rx = NULL; 9955 return err; 9956 } 9957 9958 static void netif_free_rx_queues(struct net_device *dev) 9959 { 9960 unsigned int i, count = dev->num_rx_queues; 9961 9962 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9963 if (!dev->_rx) 9964 return; 9965 9966 for (i = 0; i < count; i++) 9967 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9968 9969 kvfree(dev->_rx); 9970 } 9971 9972 static void netdev_init_one_queue(struct net_device *dev, 9973 struct netdev_queue *queue, void *_unused) 9974 { 9975 /* Initialize queue lock */ 9976 spin_lock_init(&queue->_xmit_lock); 9977 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9978 queue->xmit_lock_owner = -1; 9979 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9980 queue->dev = dev; 9981 #ifdef CONFIG_BQL 9982 dql_init(&queue->dql, HZ); 9983 #endif 9984 } 9985 9986 static void netif_free_tx_queues(struct net_device *dev) 9987 { 9988 kvfree(dev->_tx); 9989 } 9990 9991 static int netif_alloc_netdev_queues(struct net_device *dev) 9992 { 9993 unsigned int count = dev->num_tx_queues; 9994 struct netdev_queue *tx; 9995 size_t sz = count * sizeof(*tx); 9996 9997 if (count < 1 || count > 0xffff) 9998 return -EINVAL; 9999 10000 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10001 if (!tx) 10002 return -ENOMEM; 10003 10004 dev->_tx = tx; 10005 10006 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10007 spin_lock_init(&dev->tx_global_lock); 10008 10009 return 0; 10010 } 10011 10012 void netif_tx_stop_all_queues(struct net_device *dev) 10013 { 10014 unsigned int i; 10015 10016 for (i = 0; i < dev->num_tx_queues; i++) { 10017 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10018 10019 netif_tx_stop_queue(txq); 10020 } 10021 } 10022 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10023 10024 /** 10025 * register_netdevice() - register a network device 10026 * @dev: device to register 10027 * 10028 * Take a prepared network device structure and make it externally accessible. 10029 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10030 * Callers must hold the rtnl lock - you may want register_netdev() 10031 * instead of this. 10032 */ 10033 int register_netdevice(struct net_device *dev) 10034 { 10035 int ret; 10036 struct net *net = dev_net(dev); 10037 10038 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10039 NETDEV_FEATURE_COUNT); 10040 BUG_ON(dev_boot_phase); 10041 ASSERT_RTNL(); 10042 10043 might_sleep(); 10044 10045 /* When net_device's are persistent, this will be fatal. */ 10046 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10047 BUG_ON(!net); 10048 10049 ret = ethtool_check_ops(dev->ethtool_ops); 10050 if (ret) 10051 return ret; 10052 10053 spin_lock_init(&dev->addr_list_lock); 10054 netdev_set_addr_lockdep_class(dev); 10055 10056 ret = dev_get_valid_name(net, dev, dev->name); 10057 if (ret < 0) 10058 goto out; 10059 10060 ret = -ENOMEM; 10061 dev->name_node = netdev_name_node_head_alloc(dev); 10062 if (!dev->name_node) 10063 goto out; 10064 10065 /* Init, if this function is available */ 10066 if (dev->netdev_ops->ndo_init) { 10067 ret = dev->netdev_ops->ndo_init(dev); 10068 if (ret) { 10069 if (ret > 0) 10070 ret = -EIO; 10071 goto err_free_name; 10072 } 10073 } 10074 10075 if (((dev->hw_features | dev->features) & 10076 NETIF_F_HW_VLAN_CTAG_FILTER) && 10077 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10078 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10079 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10080 ret = -EINVAL; 10081 goto err_uninit; 10082 } 10083 10084 ret = dev_index_reserve(net, dev->ifindex); 10085 if (ret < 0) 10086 goto err_uninit; 10087 dev->ifindex = ret; 10088 10089 /* Transfer changeable features to wanted_features and enable 10090 * software offloads (GSO and GRO). 10091 */ 10092 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10093 dev->features |= NETIF_F_SOFT_FEATURES; 10094 10095 if (dev->udp_tunnel_nic_info) { 10096 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10097 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10098 } 10099 10100 dev->wanted_features = dev->features & dev->hw_features; 10101 10102 if (!(dev->flags & IFF_LOOPBACK)) 10103 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10104 10105 /* If IPv4 TCP segmentation offload is supported we should also 10106 * allow the device to enable segmenting the frame with the option 10107 * of ignoring a static IP ID value. This doesn't enable the 10108 * feature itself but allows the user to enable it later. 10109 */ 10110 if (dev->hw_features & NETIF_F_TSO) 10111 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10112 if (dev->vlan_features & NETIF_F_TSO) 10113 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10114 if (dev->mpls_features & NETIF_F_TSO) 10115 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10116 if (dev->hw_enc_features & NETIF_F_TSO) 10117 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10118 10119 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10120 */ 10121 dev->vlan_features |= NETIF_F_HIGHDMA; 10122 10123 /* Make NETIF_F_SG inheritable to tunnel devices. 10124 */ 10125 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10126 10127 /* Make NETIF_F_SG inheritable to MPLS. 10128 */ 10129 dev->mpls_features |= NETIF_F_SG; 10130 10131 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10132 ret = notifier_to_errno(ret); 10133 if (ret) 10134 goto err_ifindex_release; 10135 10136 ret = netdev_register_kobject(dev); 10137 write_lock(&dev_base_lock); 10138 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10139 write_unlock(&dev_base_lock); 10140 if (ret) 10141 goto err_uninit_notify; 10142 10143 __netdev_update_features(dev); 10144 10145 /* 10146 * Default initial state at registry is that the 10147 * device is present. 10148 */ 10149 10150 set_bit(__LINK_STATE_PRESENT, &dev->state); 10151 10152 linkwatch_init_dev(dev); 10153 10154 dev_init_scheduler(dev); 10155 10156 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10157 list_netdevice(dev); 10158 10159 add_device_randomness(dev->dev_addr, dev->addr_len); 10160 10161 /* If the device has permanent device address, driver should 10162 * set dev_addr and also addr_assign_type should be set to 10163 * NET_ADDR_PERM (default value). 10164 */ 10165 if (dev->addr_assign_type == NET_ADDR_PERM) 10166 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10167 10168 /* Notify protocols, that a new device appeared. */ 10169 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10170 ret = notifier_to_errno(ret); 10171 if (ret) { 10172 /* Expect explicit free_netdev() on failure */ 10173 dev->needs_free_netdev = false; 10174 unregister_netdevice_queue(dev, NULL); 10175 goto out; 10176 } 10177 /* 10178 * Prevent userspace races by waiting until the network 10179 * device is fully setup before sending notifications. 10180 */ 10181 if (!dev->rtnl_link_ops || 10182 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10183 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10184 10185 out: 10186 return ret; 10187 10188 err_uninit_notify: 10189 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10190 err_ifindex_release: 10191 dev_index_release(net, dev->ifindex); 10192 err_uninit: 10193 if (dev->netdev_ops->ndo_uninit) 10194 dev->netdev_ops->ndo_uninit(dev); 10195 if (dev->priv_destructor) 10196 dev->priv_destructor(dev); 10197 err_free_name: 10198 netdev_name_node_free(dev->name_node); 10199 goto out; 10200 } 10201 EXPORT_SYMBOL(register_netdevice); 10202 10203 /** 10204 * init_dummy_netdev - init a dummy network device for NAPI 10205 * @dev: device to init 10206 * 10207 * This takes a network device structure and initialize the minimum 10208 * amount of fields so it can be used to schedule NAPI polls without 10209 * registering a full blown interface. This is to be used by drivers 10210 * that need to tie several hardware interfaces to a single NAPI 10211 * poll scheduler due to HW limitations. 10212 */ 10213 int init_dummy_netdev(struct net_device *dev) 10214 { 10215 /* Clear everything. Note we don't initialize spinlocks 10216 * are they aren't supposed to be taken by any of the 10217 * NAPI code and this dummy netdev is supposed to be 10218 * only ever used for NAPI polls 10219 */ 10220 memset(dev, 0, sizeof(struct net_device)); 10221 10222 /* make sure we BUG if trying to hit standard 10223 * register/unregister code path 10224 */ 10225 dev->reg_state = NETREG_DUMMY; 10226 10227 /* NAPI wants this */ 10228 INIT_LIST_HEAD(&dev->napi_list); 10229 10230 /* a dummy interface is started by default */ 10231 set_bit(__LINK_STATE_PRESENT, &dev->state); 10232 set_bit(__LINK_STATE_START, &dev->state); 10233 10234 /* napi_busy_loop stats accounting wants this */ 10235 dev_net_set(dev, &init_net); 10236 10237 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10238 * because users of this 'device' dont need to change 10239 * its refcount. 10240 */ 10241 10242 return 0; 10243 } 10244 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10245 10246 10247 /** 10248 * register_netdev - register a network device 10249 * @dev: device to register 10250 * 10251 * Take a completed network device structure and add it to the kernel 10252 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10253 * chain. 0 is returned on success. A negative errno code is returned 10254 * on a failure to set up the device, or if the name is a duplicate. 10255 * 10256 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10257 * and expands the device name if you passed a format string to 10258 * alloc_netdev. 10259 */ 10260 int register_netdev(struct net_device *dev) 10261 { 10262 int err; 10263 10264 if (rtnl_lock_killable()) 10265 return -EINTR; 10266 err = register_netdevice(dev); 10267 rtnl_unlock(); 10268 return err; 10269 } 10270 EXPORT_SYMBOL(register_netdev); 10271 10272 int netdev_refcnt_read(const struct net_device *dev) 10273 { 10274 #ifdef CONFIG_PCPU_DEV_REFCNT 10275 int i, refcnt = 0; 10276 10277 for_each_possible_cpu(i) 10278 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10279 return refcnt; 10280 #else 10281 return refcount_read(&dev->dev_refcnt); 10282 #endif 10283 } 10284 EXPORT_SYMBOL(netdev_refcnt_read); 10285 10286 int netdev_unregister_timeout_secs __read_mostly = 10; 10287 10288 #define WAIT_REFS_MIN_MSECS 1 10289 #define WAIT_REFS_MAX_MSECS 250 10290 /** 10291 * netdev_wait_allrefs_any - wait until all references are gone. 10292 * @list: list of net_devices to wait on 10293 * 10294 * This is called when unregistering network devices. 10295 * 10296 * Any protocol or device that holds a reference should register 10297 * for netdevice notification, and cleanup and put back the 10298 * reference if they receive an UNREGISTER event. 10299 * We can get stuck here if buggy protocols don't correctly 10300 * call dev_put. 10301 */ 10302 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10303 { 10304 unsigned long rebroadcast_time, warning_time; 10305 struct net_device *dev; 10306 int wait = 0; 10307 10308 rebroadcast_time = warning_time = jiffies; 10309 10310 list_for_each_entry(dev, list, todo_list) 10311 if (netdev_refcnt_read(dev) == 1) 10312 return dev; 10313 10314 while (true) { 10315 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10316 rtnl_lock(); 10317 10318 /* Rebroadcast unregister notification */ 10319 list_for_each_entry(dev, list, todo_list) 10320 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10321 10322 __rtnl_unlock(); 10323 rcu_barrier(); 10324 rtnl_lock(); 10325 10326 list_for_each_entry(dev, list, todo_list) 10327 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10328 &dev->state)) { 10329 /* We must not have linkwatch events 10330 * pending on unregister. If this 10331 * happens, we simply run the queue 10332 * unscheduled, resulting in a noop 10333 * for this device. 10334 */ 10335 linkwatch_run_queue(); 10336 break; 10337 } 10338 10339 __rtnl_unlock(); 10340 10341 rebroadcast_time = jiffies; 10342 } 10343 10344 if (!wait) { 10345 rcu_barrier(); 10346 wait = WAIT_REFS_MIN_MSECS; 10347 } else { 10348 msleep(wait); 10349 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10350 } 10351 10352 list_for_each_entry(dev, list, todo_list) 10353 if (netdev_refcnt_read(dev) == 1) 10354 return dev; 10355 10356 if (time_after(jiffies, warning_time + 10357 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10358 list_for_each_entry(dev, list, todo_list) { 10359 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10360 dev->name, netdev_refcnt_read(dev)); 10361 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10362 } 10363 10364 warning_time = jiffies; 10365 } 10366 } 10367 } 10368 10369 /* The sequence is: 10370 * 10371 * rtnl_lock(); 10372 * ... 10373 * register_netdevice(x1); 10374 * register_netdevice(x2); 10375 * ... 10376 * unregister_netdevice(y1); 10377 * unregister_netdevice(y2); 10378 * ... 10379 * rtnl_unlock(); 10380 * free_netdev(y1); 10381 * free_netdev(y2); 10382 * 10383 * We are invoked by rtnl_unlock(). 10384 * This allows us to deal with problems: 10385 * 1) We can delete sysfs objects which invoke hotplug 10386 * without deadlocking with linkwatch via keventd. 10387 * 2) Since we run with the RTNL semaphore not held, we can sleep 10388 * safely in order to wait for the netdev refcnt to drop to zero. 10389 * 10390 * We must not return until all unregister events added during 10391 * the interval the lock was held have been completed. 10392 */ 10393 void netdev_run_todo(void) 10394 { 10395 struct net_device *dev, *tmp; 10396 struct list_head list; 10397 #ifdef CONFIG_LOCKDEP 10398 struct list_head unlink_list; 10399 10400 list_replace_init(&net_unlink_list, &unlink_list); 10401 10402 while (!list_empty(&unlink_list)) { 10403 struct net_device *dev = list_first_entry(&unlink_list, 10404 struct net_device, 10405 unlink_list); 10406 list_del_init(&dev->unlink_list); 10407 dev->nested_level = dev->lower_level - 1; 10408 } 10409 #endif 10410 10411 /* Snapshot list, allow later requests */ 10412 list_replace_init(&net_todo_list, &list); 10413 10414 __rtnl_unlock(); 10415 10416 /* Wait for rcu callbacks to finish before next phase */ 10417 if (!list_empty(&list)) 10418 rcu_barrier(); 10419 10420 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10421 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10422 netdev_WARN(dev, "run_todo but not unregistering\n"); 10423 list_del(&dev->todo_list); 10424 continue; 10425 } 10426 10427 write_lock(&dev_base_lock); 10428 dev->reg_state = NETREG_UNREGISTERED; 10429 write_unlock(&dev_base_lock); 10430 linkwatch_forget_dev(dev); 10431 } 10432 10433 while (!list_empty(&list)) { 10434 dev = netdev_wait_allrefs_any(&list); 10435 list_del(&dev->todo_list); 10436 10437 /* paranoia */ 10438 BUG_ON(netdev_refcnt_read(dev) != 1); 10439 BUG_ON(!list_empty(&dev->ptype_all)); 10440 BUG_ON(!list_empty(&dev->ptype_specific)); 10441 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10442 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10443 10444 if (dev->priv_destructor) 10445 dev->priv_destructor(dev); 10446 if (dev->needs_free_netdev) 10447 free_netdev(dev); 10448 10449 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10450 wake_up(&netdev_unregistering_wq); 10451 10452 /* Free network device */ 10453 kobject_put(&dev->dev.kobj); 10454 } 10455 } 10456 10457 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10458 * all the same fields in the same order as net_device_stats, with only 10459 * the type differing, but rtnl_link_stats64 may have additional fields 10460 * at the end for newer counters. 10461 */ 10462 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10463 const struct net_device_stats *netdev_stats) 10464 { 10465 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10466 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10467 u64 *dst = (u64 *)stats64; 10468 10469 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10470 for (i = 0; i < n; i++) 10471 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10472 /* zero out counters that only exist in rtnl_link_stats64 */ 10473 memset((char *)stats64 + n * sizeof(u64), 0, 10474 sizeof(*stats64) - n * sizeof(u64)); 10475 } 10476 EXPORT_SYMBOL(netdev_stats_to_stats64); 10477 10478 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10479 { 10480 struct net_device_core_stats __percpu *p; 10481 10482 p = alloc_percpu_gfp(struct net_device_core_stats, 10483 GFP_ATOMIC | __GFP_NOWARN); 10484 10485 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10486 free_percpu(p); 10487 10488 /* This READ_ONCE() pairs with the cmpxchg() above */ 10489 return READ_ONCE(dev->core_stats); 10490 } 10491 EXPORT_SYMBOL(netdev_core_stats_alloc); 10492 10493 /** 10494 * dev_get_stats - get network device statistics 10495 * @dev: device to get statistics from 10496 * @storage: place to store stats 10497 * 10498 * Get network statistics from device. Return @storage. 10499 * The device driver may provide its own method by setting 10500 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10501 * otherwise the internal statistics structure is used. 10502 */ 10503 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10504 struct rtnl_link_stats64 *storage) 10505 { 10506 const struct net_device_ops *ops = dev->netdev_ops; 10507 const struct net_device_core_stats __percpu *p; 10508 10509 if (ops->ndo_get_stats64) { 10510 memset(storage, 0, sizeof(*storage)); 10511 ops->ndo_get_stats64(dev, storage); 10512 } else if (ops->ndo_get_stats) { 10513 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10514 } else { 10515 netdev_stats_to_stats64(storage, &dev->stats); 10516 } 10517 10518 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10519 p = READ_ONCE(dev->core_stats); 10520 if (p) { 10521 const struct net_device_core_stats *core_stats; 10522 int i; 10523 10524 for_each_possible_cpu(i) { 10525 core_stats = per_cpu_ptr(p, i); 10526 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10527 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10528 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10529 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10530 } 10531 } 10532 return storage; 10533 } 10534 EXPORT_SYMBOL(dev_get_stats); 10535 10536 /** 10537 * dev_fetch_sw_netstats - get per-cpu network device statistics 10538 * @s: place to store stats 10539 * @netstats: per-cpu network stats to read from 10540 * 10541 * Read per-cpu network statistics and populate the related fields in @s. 10542 */ 10543 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10544 const struct pcpu_sw_netstats __percpu *netstats) 10545 { 10546 int cpu; 10547 10548 for_each_possible_cpu(cpu) { 10549 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10550 const struct pcpu_sw_netstats *stats; 10551 unsigned int start; 10552 10553 stats = per_cpu_ptr(netstats, cpu); 10554 do { 10555 start = u64_stats_fetch_begin(&stats->syncp); 10556 rx_packets = u64_stats_read(&stats->rx_packets); 10557 rx_bytes = u64_stats_read(&stats->rx_bytes); 10558 tx_packets = u64_stats_read(&stats->tx_packets); 10559 tx_bytes = u64_stats_read(&stats->tx_bytes); 10560 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10561 10562 s->rx_packets += rx_packets; 10563 s->rx_bytes += rx_bytes; 10564 s->tx_packets += tx_packets; 10565 s->tx_bytes += tx_bytes; 10566 } 10567 } 10568 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10569 10570 /** 10571 * dev_get_tstats64 - ndo_get_stats64 implementation 10572 * @dev: device to get statistics from 10573 * @s: place to store stats 10574 * 10575 * Populate @s from dev->stats and dev->tstats. Can be used as 10576 * ndo_get_stats64() callback. 10577 */ 10578 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10579 { 10580 netdev_stats_to_stats64(s, &dev->stats); 10581 dev_fetch_sw_netstats(s, dev->tstats); 10582 } 10583 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10584 10585 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10586 { 10587 struct netdev_queue *queue = dev_ingress_queue(dev); 10588 10589 #ifdef CONFIG_NET_CLS_ACT 10590 if (queue) 10591 return queue; 10592 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10593 if (!queue) 10594 return NULL; 10595 netdev_init_one_queue(dev, queue, NULL); 10596 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10597 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10598 rcu_assign_pointer(dev->ingress_queue, queue); 10599 #endif 10600 return queue; 10601 } 10602 10603 static const struct ethtool_ops default_ethtool_ops; 10604 10605 void netdev_set_default_ethtool_ops(struct net_device *dev, 10606 const struct ethtool_ops *ops) 10607 { 10608 if (dev->ethtool_ops == &default_ethtool_ops) 10609 dev->ethtool_ops = ops; 10610 } 10611 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10612 10613 /** 10614 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10615 * @dev: netdev to enable the IRQ coalescing on 10616 * 10617 * Sets a conservative default for SW IRQ coalescing. Users can use 10618 * sysfs attributes to override the default values. 10619 */ 10620 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10621 { 10622 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10623 10624 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10625 dev->gro_flush_timeout = 20000; 10626 dev->napi_defer_hard_irqs = 1; 10627 } 10628 } 10629 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10630 10631 void netdev_freemem(struct net_device *dev) 10632 { 10633 char *addr = (char *)dev - dev->padded; 10634 10635 kvfree(addr); 10636 } 10637 10638 /** 10639 * alloc_netdev_mqs - allocate network device 10640 * @sizeof_priv: size of private data to allocate space for 10641 * @name: device name format string 10642 * @name_assign_type: origin of device name 10643 * @setup: callback to initialize device 10644 * @txqs: the number of TX subqueues to allocate 10645 * @rxqs: the number of RX subqueues to allocate 10646 * 10647 * Allocates a struct net_device with private data area for driver use 10648 * and performs basic initialization. Also allocates subqueue structs 10649 * for each queue on the device. 10650 */ 10651 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10652 unsigned char name_assign_type, 10653 void (*setup)(struct net_device *), 10654 unsigned int txqs, unsigned int rxqs) 10655 { 10656 struct net_device *dev; 10657 unsigned int alloc_size; 10658 struct net_device *p; 10659 10660 BUG_ON(strlen(name) >= sizeof(dev->name)); 10661 10662 if (txqs < 1) { 10663 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10664 return NULL; 10665 } 10666 10667 if (rxqs < 1) { 10668 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10669 return NULL; 10670 } 10671 10672 alloc_size = sizeof(struct net_device); 10673 if (sizeof_priv) { 10674 /* ensure 32-byte alignment of private area */ 10675 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10676 alloc_size += sizeof_priv; 10677 } 10678 /* ensure 32-byte alignment of whole construct */ 10679 alloc_size += NETDEV_ALIGN - 1; 10680 10681 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10682 if (!p) 10683 return NULL; 10684 10685 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10686 dev->padded = (char *)dev - (char *)p; 10687 10688 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10689 #ifdef CONFIG_PCPU_DEV_REFCNT 10690 dev->pcpu_refcnt = alloc_percpu(int); 10691 if (!dev->pcpu_refcnt) 10692 goto free_dev; 10693 __dev_hold(dev); 10694 #else 10695 refcount_set(&dev->dev_refcnt, 1); 10696 #endif 10697 10698 if (dev_addr_init(dev)) 10699 goto free_pcpu; 10700 10701 dev_mc_init(dev); 10702 dev_uc_init(dev); 10703 10704 dev_net_set(dev, &init_net); 10705 10706 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10707 dev->xdp_zc_max_segs = 1; 10708 dev->gso_max_segs = GSO_MAX_SEGS; 10709 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10710 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10711 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10712 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10713 dev->tso_max_segs = TSO_MAX_SEGS; 10714 dev->upper_level = 1; 10715 dev->lower_level = 1; 10716 #ifdef CONFIG_LOCKDEP 10717 dev->nested_level = 0; 10718 INIT_LIST_HEAD(&dev->unlink_list); 10719 #endif 10720 10721 INIT_LIST_HEAD(&dev->napi_list); 10722 INIT_LIST_HEAD(&dev->unreg_list); 10723 INIT_LIST_HEAD(&dev->close_list); 10724 INIT_LIST_HEAD(&dev->link_watch_list); 10725 INIT_LIST_HEAD(&dev->adj_list.upper); 10726 INIT_LIST_HEAD(&dev->adj_list.lower); 10727 INIT_LIST_HEAD(&dev->ptype_all); 10728 INIT_LIST_HEAD(&dev->ptype_specific); 10729 INIT_LIST_HEAD(&dev->net_notifier_list); 10730 #ifdef CONFIG_NET_SCHED 10731 hash_init(dev->qdisc_hash); 10732 #endif 10733 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10734 setup(dev); 10735 10736 if (!dev->tx_queue_len) { 10737 dev->priv_flags |= IFF_NO_QUEUE; 10738 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10739 } 10740 10741 dev->num_tx_queues = txqs; 10742 dev->real_num_tx_queues = txqs; 10743 if (netif_alloc_netdev_queues(dev)) 10744 goto free_all; 10745 10746 dev->num_rx_queues = rxqs; 10747 dev->real_num_rx_queues = rxqs; 10748 if (netif_alloc_rx_queues(dev)) 10749 goto free_all; 10750 10751 strcpy(dev->name, name); 10752 dev->name_assign_type = name_assign_type; 10753 dev->group = INIT_NETDEV_GROUP; 10754 if (!dev->ethtool_ops) 10755 dev->ethtool_ops = &default_ethtool_ops; 10756 10757 nf_hook_netdev_init(dev); 10758 10759 return dev; 10760 10761 free_all: 10762 free_netdev(dev); 10763 return NULL; 10764 10765 free_pcpu: 10766 #ifdef CONFIG_PCPU_DEV_REFCNT 10767 free_percpu(dev->pcpu_refcnt); 10768 free_dev: 10769 #endif 10770 netdev_freemem(dev); 10771 return NULL; 10772 } 10773 EXPORT_SYMBOL(alloc_netdev_mqs); 10774 10775 /** 10776 * free_netdev - free network device 10777 * @dev: device 10778 * 10779 * This function does the last stage of destroying an allocated device 10780 * interface. The reference to the device object is released. If this 10781 * is the last reference then it will be freed.Must be called in process 10782 * context. 10783 */ 10784 void free_netdev(struct net_device *dev) 10785 { 10786 struct napi_struct *p, *n; 10787 10788 might_sleep(); 10789 10790 /* When called immediately after register_netdevice() failed the unwind 10791 * handling may still be dismantling the device. Handle that case by 10792 * deferring the free. 10793 */ 10794 if (dev->reg_state == NETREG_UNREGISTERING) { 10795 ASSERT_RTNL(); 10796 dev->needs_free_netdev = true; 10797 return; 10798 } 10799 10800 netif_free_tx_queues(dev); 10801 netif_free_rx_queues(dev); 10802 10803 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10804 10805 /* Flush device addresses */ 10806 dev_addr_flush(dev); 10807 10808 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10809 netif_napi_del(p); 10810 10811 ref_tracker_dir_exit(&dev->refcnt_tracker); 10812 #ifdef CONFIG_PCPU_DEV_REFCNT 10813 free_percpu(dev->pcpu_refcnt); 10814 dev->pcpu_refcnt = NULL; 10815 #endif 10816 free_percpu(dev->core_stats); 10817 dev->core_stats = NULL; 10818 free_percpu(dev->xdp_bulkq); 10819 dev->xdp_bulkq = NULL; 10820 10821 /* Compatibility with error handling in drivers */ 10822 if (dev->reg_state == NETREG_UNINITIALIZED) { 10823 netdev_freemem(dev); 10824 return; 10825 } 10826 10827 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10828 dev->reg_state = NETREG_RELEASED; 10829 10830 /* will free via device release */ 10831 put_device(&dev->dev); 10832 } 10833 EXPORT_SYMBOL(free_netdev); 10834 10835 /** 10836 * synchronize_net - Synchronize with packet receive processing 10837 * 10838 * Wait for packets currently being received to be done. 10839 * Does not block later packets from starting. 10840 */ 10841 void synchronize_net(void) 10842 { 10843 might_sleep(); 10844 if (rtnl_is_locked()) 10845 synchronize_rcu_expedited(); 10846 else 10847 synchronize_rcu(); 10848 } 10849 EXPORT_SYMBOL(synchronize_net); 10850 10851 /** 10852 * unregister_netdevice_queue - remove device from the kernel 10853 * @dev: device 10854 * @head: list 10855 * 10856 * This function shuts down a device interface and removes it 10857 * from the kernel tables. 10858 * If head not NULL, device is queued to be unregistered later. 10859 * 10860 * Callers must hold the rtnl semaphore. You may want 10861 * unregister_netdev() instead of this. 10862 */ 10863 10864 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10865 { 10866 ASSERT_RTNL(); 10867 10868 if (head) { 10869 list_move_tail(&dev->unreg_list, head); 10870 } else { 10871 LIST_HEAD(single); 10872 10873 list_add(&dev->unreg_list, &single); 10874 unregister_netdevice_many(&single); 10875 } 10876 } 10877 EXPORT_SYMBOL(unregister_netdevice_queue); 10878 10879 void unregister_netdevice_many_notify(struct list_head *head, 10880 u32 portid, const struct nlmsghdr *nlh) 10881 { 10882 struct net_device *dev, *tmp; 10883 LIST_HEAD(close_head); 10884 10885 BUG_ON(dev_boot_phase); 10886 ASSERT_RTNL(); 10887 10888 if (list_empty(head)) 10889 return; 10890 10891 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10892 /* Some devices call without registering 10893 * for initialization unwind. Remove those 10894 * devices and proceed with the remaining. 10895 */ 10896 if (dev->reg_state == NETREG_UNINITIALIZED) { 10897 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10898 dev->name, dev); 10899 10900 WARN_ON(1); 10901 list_del(&dev->unreg_list); 10902 continue; 10903 } 10904 dev->dismantle = true; 10905 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10906 } 10907 10908 /* If device is running, close it first. */ 10909 list_for_each_entry(dev, head, unreg_list) 10910 list_add_tail(&dev->close_list, &close_head); 10911 dev_close_many(&close_head, true); 10912 10913 list_for_each_entry(dev, head, unreg_list) { 10914 /* And unlink it from device chain. */ 10915 write_lock(&dev_base_lock); 10916 unlist_netdevice(dev, false); 10917 dev->reg_state = NETREG_UNREGISTERING; 10918 write_unlock(&dev_base_lock); 10919 } 10920 flush_all_backlogs(); 10921 10922 synchronize_net(); 10923 10924 list_for_each_entry(dev, head, unreg_list) { 10925 struct sk_buff *skb = NULL; 10926 10927 /* Shutdown queueing discipline. */ 10928 dev_shutdown(dev); 10929 dev_tcx_uninstall(dev); 10930 dev_xdp_uninstall(dev); 10931 bpf_dev_bound_netdev_unregister(dev); 10932 10933 netdev_offload_xstats_disable_all(dev); 10934 10935 /* Notify protocols, that we are about to destroy 10936 * this device. They should clean all the things. 10937 */ 10938 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10939 10940 if (!dev->rtnl_link_ops || 10941 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10942 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10943 GFP_KERNEL, NULL, 0, 10944 portid, nlh); 10945 10946 /* 10947 * Flush the unicast and multicast chains 10948 */ 10949 dev_uc_flush(dev); 10950 dev_mc_flush(dev); 10951 10952 netdev_name_node_alt_flush(dev); 10953 netdev_name_node_free(dev->name_node); 10954 10955 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10956 10957 if (dev->netdev_ops->ndo_uninit) 10958 dev->netdev_ops->ndo_uninit(dev); 10959 10960 if (skb) 10961 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 10962 10963 /* Notifier chain MUST detach us all upper devices. */ 10964 WARN_ON(netdev_has_any_upper_dev(dev)); 10965 WARN_ON(netdev_has_any_lower_dev(dev)); 10966 10967 /* Remove entries from kobject tree */ 10968 netdev_unregister_kobject(dev); 10969 #ifdef CONFIG_XPS 10970 /* Remove XPS queueing entries */ 10971 netif_reset_xps_queues_gt(dev, 0); 10972 #endif 10973 } 10974 10975 synchronize_net(); 10976 10977 list_for_each_entry(dev, head, unreg_list) { 10978 netdev_put(dev, &dev->dev_registered_tracker); 10979 net_set_todo(dev); 10980 } 10981 10982 list_del(head); 10983 } 10984 10985 /** 10986 * unregister_netdevice_many - unregister many devices 10987 * @head: list of devices 10988 * 10989 * Note: As most callers use a stack allocated list_head, 10990 * we force a list_del() to make sure stack wont be corrupted later. 10991 */ 10992 void unregister_netdevice_many(struct list_head *head) 10993 { 10994 unregister_netdevice_many_notify(head, 0, NULL); 10995 } 10996 EXPORT_SYMBOL(unregister_netdevice_many); 10997 10998 /** 10999 * unregister_netdev - remove device from the kernel 11000 * @dev: device 11001 * 11002 * This function shuts down a device interface and removes it 11003 * from the kernel tables. 11004 * 11005 * This is just a wrapper for unregister_netdevice that takes 11006 * the rtnl semaphore. In general you want to use this and not 11007 * unregister_netdevice. 11008 */ 11009 void unregister_netdev(struct net_device *dev) 11010 { 11011 rtnl_lock(); 11012 unregister_netdevice(dev); 11013 rtnl_unlock(); 11014 } 11015 EXPORT_SYMBOL(unregister_netdev); 11016 11017 /** 11018 * __dev_change_net_namespace - move device to different nethost namespace 11019 * @dev: device 11020 * @net: network namespace 11021 * @pat: If not NULL name pattern to try if the current device name 11022 * is already taken in the destination network namespace. 11023 * @new_ifindex: If not zero, specifies device index in the target 11024 * namespace. 11025 * 11026 * This function shuts down a device interface and moves it 11027 * to a new network namespace. On success 0 is returned, on 11028 * a failure a netagive errno code is returned. 11029 * 11030 * Callers must hold the rtnl semaphore. 11031 */ 11032 11033 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11034 const char *pat, int new_ifindex) 11035 { 11036 struct net *net_old = dev_net(dev); 11037 int err, new_nsid; 11038 11039 ASSERT_RTNL(); 11040 11041 /* Don't allow namespace local devices to be moved. */ 11042 err = -EINVAL; 11043 if (dev->features & NETIF_F_NETNS_LOCAL) 11044 goto out; 11045 11046 /* Ensure the device has been registrered */ 11047 if (dev->reg_state != NETREG_REGISTERED) 11048 goto out; 11049 11050 /* Get out if there is nothing todo */ 11051 err = 0; 11052 if (net_eq(net_old, net)) 11053 goto out; 11054 11055 /* Pick the destination device name, and ensure 11056 * we can use it in the destination network namespace. 11057 */ 11058 err = -EEXIST; 11059 if (netdev_name_in_use(net, dev->name)) { 11060 /* We get here if we can't use the current device name */ 11061 if (!pat) 11062 goto out; 11063 err = dev_get_valid_name(net, dev, pat); 11064 if (err < 0) 11065 goto out; 11066 } 11067 11068 /* Check that new_ifindex isn't used yet. */ 11069 if (new_ifindex) { 11070 err = dev_index_reserve(net, new_ifindex); 11071 if (err < 0) 11072 goto out; 11073 } else { 11074 /* If there is an ifindex conflict assign a new one */ 11075 err = dev_index_reserve(net, dev->ifindex); 11076 if (err == -EBUSY) 11077 err = dev_index_reserve(net, 0); 11078 if (err < 0) 11079 goto out; 11080 new_ifindex = err; 11081 } 11082 11083 /* 11084 * And now a mini version of register_netdevice unregister_netdevice. 11085 */ 11086 11087 /* If device is running close it first. */ 11088 dev_close(dev); 11089 11090 /* And unlink it from device chain */ 11091 unlist_netdevice(dev, true); 11092 11093 synchronize_net(); 11094 11095 /* Shutdown queueing discipline. */ 11096 dev_shutdown(dev); 11097 11098 /* Notify protocols, that we are about to destroy 11099 * this device. They should clean all the things. 11100 * 11101 * Note that dev->reg_state stays at NETREG_REGISTERED. 11102 * This is wanted because this way 8021q and macvlan know 11103 * the device is just moving and can keep their slaves up. 11104 */ 11105 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11106 rcu_barrier(); 11107 11108 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11109 11110 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11111 new_ifindex); 11112 11113 /* 11114 * Flush the unicast and multicast chains 11115 */ 11116 dev_uc_flush(dev); 11117 dev_mc_flush(dev); 11118 11119 /* Send a netdev-removed uevent to the old namespace */ 11120 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11121 netdev_adjacent_del_links(dev); 11122 11123 /* Move per-net netdevice notifiers that are following the netdevice */ 11124 move_netdevice_notifiers_dev_net(dev, net); 11125 11126 /* Actually switch the network namespace */ 11127 dev_net_set(dev, net); 11128 dev->ifindex = new_ifindex; 11129 11130 /* Send a netdev-add uevent to the new namespace */ 11131 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11132 netdev_adjacent_add_links(dev); 11133 11134 /* Fixup kobjects */ 11135 err = device_rename(&dev->dev, dev->name); 11136 WARN_ON(err); 11137 11138 /* Adapt owner in case owning user namespace of target network 11139 * namespace is different from the original one. 11140 */ 11141 err = netdev_change_owner(dev, net_old, net); 11142 WARN_ON(err); 11143 11144 /* Add the device back in the hashes */ 11145 list_netdevice(dev); 11146 11147 /* Notify protocols, that a new device appeared. */ 11148 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11149 11150 /* 11151 * Prevent userspace races by waiting until the network 11152 * device is fully setup before sending notifications. 11153 */ 11154 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11155 11156 synchronize_net(); 11157 err = 0; 11158 out: 11159 return err; 11160 } 11161 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11162 11163 static int dev_cpu_dead(unsigned int oldcpu) 11164 { 11165 struct sk_buff **list_skb; 11166 struct sk_buff *skb; 11167 unsigned int cpu; 11168 struct softnet_data *sd, *oldsd, *remsd = NULL; 11169 11170 local_irq_disable(); 11171 cpu = smp_processor_id(); 11172 sd = &per_cpu(softnet_data, cpu); 11173 oldsd = &per_cpu(softnet_data, oldcpu); 11174 11175 /* Find end of our completion_queue. */ 11176 list_skb = &sd->completion_queue; 11177 while (*list_skb) 11178 list_skb = &(*list_skb)->next; 11179 /* Append completion queue from offline CPU. */ 11180 *list_skb = oldsd->completion_queue; 11181 oldsd->completion_queue = NULL; 11182 11183 /* Append output queue from offline CPU. */ 11184 if (oldsd->output_queue) { 11185 *sd->output_queue_tailp = oldsd->output_queue; 11186 sd->output_queue_tailp = oldsd->output_queue_tailp; 11187 oldsd->output_queue = NULL; 11188 oldsd->output_queue_tailp = &oldsd->output_queue; 11189 } 11190 /* Append NAPI poll list from offline CPU, with one exception : 11191 * process_backlog() must be called by cpu owning percpu backlog. 11192 * We properly handle process_queue & input_pkt_queue later. 11193 */ 11194 while (!list_empty(&oldsd->poll_list)) { 11195 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11196 struct napi_struct, 11197 poll_list); 11198 11199 list_del_init(&napi->poll_list); 11200 if (napi->poll == process_backlog) 11201 napi->state = 0; 11202 else 11203 ____napi_schedule(sd, napi); 11204 } 11205 11206 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11207 local_irq_enable(); 11208 11209 #ifdef CONFIG_RPS 11210 remsd = oldsd->rps_ipi_list; 11211 oldsd->rps_ipi_list = NULL; 11212 #endif 11213 /* send out pending IPI's on offline CPU */ 11214 net_rps_send_ipi(remsd); 11215 11216 /* Process offline CPU's input_pkt_queue */ 11217 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11218 netif_rx(skb); 11219 input_queue_head_incr(oldsd); 11220 } 11221 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11222 netif_rx(skb); 11223 input_queue_head_incr(oldsd); 11224 } 11225 11226 return 0; 11227 } 11228 11229 /** 11230 * netdev_increment_features - increment feature set by one 11231 * @all: current feature set 11232 * @one: new feature set 11233 * @mask: mask feature set 11234 * 11235 * Computes a new feature set after adding a device with feature set 11236 * @one to the master device with current feature set @all. Will not 11237 * enable anything that is off in @mask. Returns the new feature set. 11238 */ 11239 netdev_features_t netdev_increment_features(netdev_features_t all, 11240 netdev_features_t one, netdev_features_t mask) 11241 { 11242 if (mask & NETIF_F_HW_CSUM) 11243 mask |= NETIF_F_CSUM_MASK; 11244 mask |= NETIF_F_VLAN_CHALLENGED; 11245 11246 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11247 all &= one | ~NETIF_F_ALL_FOR_ALL; 11248 11249 /* If one device supports hw checksumming, set for all. */ 11250 if (all & NETIF_F_HW_CSUM) 11251 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11252 11253 return all; 11254 } 11255 EXPORT_SYMBOL(netdev_increment_features); 11256 11257 static struct hlist_head * __net_init netdev_create_hash(void) 11258 { 11259 int i; 11260 struct hlist_head *hash; 11261 11262 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11263 if (hash != NULL) 11264 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11265 INIT_HLIST_HEAD(&hash[i]); 11266 11267 return hash; 11268 } 11269 11270 /* Initialize per network namespace state */ 11271 static int __net_init netdev_init(struct net *net) 11272 { 11273 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11274 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11275 11276 INIT_LIST_HEAD(&net->dev_base_head); 11277 11278 net->dev_name_head = netdev_create_hash(); 11279 if (net->dev_name_head == NULL) 11280 goto err_name; 11281 11282 net->dev_index_head = netdev_create_hash(); 11283 if (net->dev_index_head == NULL) 11284 goto err_idx; 11285 11286 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11287 11288 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11289 11290 return 0; 11291 11292 err_idx: 11293 kfree(net->dev_name_head); 11294 err_name: 11295 return -ENOMEM; 11296 } 11297 11298 /** 11299 * netdev_drivername - network driver for the device 11300 * @dev: network device 11301 * 11302 * Determine network driver for device. 11303 */ 11304 const char *netdev_drivername(const struct net_device *dev) 11305 { 11306 const struct device_driver *driver; 11307 const struct device *parent; 11308 const char *empty = ""; 11309 11310 parent = dev->dev.parent; 11311 if (!parent) 11312 return empty; 11313 11314 driver = parent->driver; 11315 if (driver && driver->name) 11316 return driver->name; 11317 return empty; 11318 } 11319 11320 static void __netdev_printk(const char *level, const struct net_device *dev, 11321 struct va_format *vaf) 11322 { 11323 if (dev && dev->dev.parent) { 11324 dev_printk_emit(level[1] - '0', 11325 dev->dev.parent, 11326 "%s %s %s%s: %pV", 11327 dev_driver_string(dev->dev.parent), 11328 dev_name(dev->dev.parent), 11329 netdev_name(dev), netdev_reg_state(dev), 11330 vaf); 11331 } else if (dev) { 11332 printk("%s%s%s: %pV", 11333 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11334 } else { 11335 printk("%s(NULL net_device): %pV", level, vaf); 11336 } 11337 } 11338 11339 void netdev_printk(const char *level, const struct net_device *dev, 11340 const char *format, ...) 11341 { 11342 struct va_format vaf; 11343 va_list args; 11344 11345 va_start(args, format); 11346 11347 vaf.fmt = format; 11348 vaf.va = &args; 11349 11350 __netdev_printk(level, dev, &vaf); 11351 11352 va_end(args); 11353 } 11354 EXPORT_SYMBOL(netdev_printk); 11355 11356 #define define_netdev_printk_level(func, level) \ 11357 void func(const struct net_device *dev, const char *fmt, ...) \ 11358 { \ 11359 struct va_format vaf; \ 11360 va_list args; \ 11361 \ 11362 va_start(args, fmt); \ 11363 \ 11364 vaf.fmt = fmt; \ 11365 vaf.va = &args; \ 11366 \ 11367 __netdev_printk(level, dev, &vaf); \ 11368 \ 11369 va_end(args); \ 11370 } \ 11371 EXPORT_SYMBOL(func); 11372 11373 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11374 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11375 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11376 define_netdev_printk_level(netdev_err, KERN_ERR); 11377 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11378 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11379 define_netdev_printk_level(netdev_info, KERN_INFO); 11380 11381 static void __net_exit netdev_exit(struct net *net) 11382 { 11383 kfree(net->dev_name_head); 11384 kfree(net->dev_index_head); 11385 xa_destroy(&net->dev_by_index); 11386 if (net != &init_net) 11387 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11388 } 11389 11390 static struct pernet_operations __net_initdata netdev_net_ops = { 11391 .init = netdev_init, 11392 .exit = netdev_exit, 11393 }; 11394 11395 static void __net_exit default_device_exit_net(struct net *net) 11396 { 11397 struct net_device *dev, *aux; 11398 /* 11399 * Push all migratable network devices back to the 11400 * initial network namespace 11401 */ 11402 ASSERT_RTNL(); 11403 for_each_netdev_safe(net, dev, aux) { 11404 int err; 11405 char fb_name[IFNAMSIZ]; 11406 11407 /* Ignore unmoveable devices (i.e. loopback) */ 11408 if (dev->features & NETIF_F_NETNS_LOCAL) 11409 continue; 11410 11411 /* Leave virtual devices for the generic cleanup */ 11412 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11413 continue; 11414 11415 /* Push remaining network devices to init_net */ 11416 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11417 if (netdev_name_in_use(&init_net, fb_name)) 11418 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11419 err = dev_change_net_namespace(dev, &init_net, fb_name); 11420 if (err) { 11421 pr_emerg("%s: failed to move %s to init_net: %d\n", 11422 __func__, dev->name, err); 11423 BUG(); 11424 } 11425 } 11426 } 11427 11428 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11429 { 11430 /* At exit all network devices most be removed from a network 11431 * namespace. Do this in the reverse order of registration. 11432 * Do this across as many network namespaces as possible to 11433 * improve batching efficiency. 11434 */ 11435 struct net_device *dev; 11436 struct net *net; 11437 LIST_HEAD(dev_kill_list); 11438 11439 rtnl_lock(); 11440 list_for_each_entry(net, net_list, exit_list) { 11441 default_device_exit_net(net); 11442 cond_resched(); 11443 } 11444 11445 list_for_each_entry(net, net_list, exit_list) { 11446 for_each_netdev_reverse(net, dev) { 11447 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11448 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11449 else 11450 unregister_netdevice_queue(dev, &dev_kill_list); 11451 } 11452 } 11453 unregister_netdevice_many(&dev_kill_list); 11454 rtnl_unlock(); 11455 } 11456 11457 static struct pernet_operations __net_initdata default_device_ops = { 11458 .exit_batch = default_device_exit_batch, 11459 }; 11460 11461 /* 11462 * Initialize the DEV module. At boot time this walks the device list and 11463 * unhooks any devices that fail to initialise (normally hardware not 11464 * present) and leaves us with a valid list of present and active devices. 11465 * 11466 */ 11467 11468 /* 11469 * This is called single threaded during boot, so no need 11470 * to take the rtnl semaphore. 11471 */ 11472 static int __init net_dev_init(void) 11473 { 11474 int i, rc = -ENOMEM; 11475 11476 BUG_ON(!dev_boot_phase); 11477 11478 if (dev_proc_init()) 11479 goto out; 11480 11481 if (netdev_kobject_init()) 11482 goto out; 11483 11484 INIT_LIST_HEAD(&ptype_all); 11485 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11486 INIT_LIST_HEAD(&ptype_base[i]); 11487 11488 if (register_pernet_subsys(&netdev_net_ops)) 11489 goto out; 11490 11491 /* 11492 * Initialise the packet receive queues. 11493 */ 11494 11495 for_each_possible_cpu(i) { 11496 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11497 struct softnet_data *sd = &per_cpu(softnet_data, i); 11498 11499 INIT_WORK(flush, flush_backlog); 11500 11501 skb_queue_head_init(&sd->input_pkt_queue); 11502 skb_queue_head_init(&sd->process_queue); 11503 #ifdef CONFIG_XFRM_OFFLOAD 11504 skb_queue_head_init(&sd->xfrm_backlog); 11505 #endif 11506 INIT_LIST_HEAD(&sd->poll_list); 11507 sd->output_queue_tailp = &sd->output_queue; 11508 #ifdef CONFIG_RPS 11509 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11510 sd->cpu = i; 11511 #endif 11512 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11513 spin_lock_init(&sd->defer_lock); 11514 11515 init_gro_hash(&sd->backlog); 11516 sd->backlog.poll = process_backlog; 11517 sd->backlog.weight = weight_p; 11518 } 11519 11520 dev_boot_phase = 0; 11521 11522 /* The loopback device is special if any other network devices 11523 * is present in a network namespace the loopback device must 11524 * be present. Since we now dynamically allocate and free the 11525 * loopback device ensure this invariant is maintained by 11526 * keeping the loopback device as the first device on the 11527 * list of network devices. Ensuring the loopback devices 11528 * is the first device that appears and the last network device 11529 * that disappears. 11530 */ 11531 if (register_pernet_device(&loopback_net_ops)) 11532 goto out; 11533 11534 if (register_pernet_device(&default_device_ops)) 11535 goto out; 11536 11537 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11538 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11539 11540 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11541 NULL, dev_cpu_dead); 11542 WARN_ON(rc < 0); 11543 rc = 0; 11544 out: 11545 return rc; 11546 } 11547 11548 subsys_initcall(net_dev_init); 11549