FILE
dev.c
File preview with syntax highlighting.
linux-6.19.14/net/core/dev.c
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/isolation.h> 81#include <linux/sched/mm.h> 82#include <linux/smpboot.h> 83#include <linux/mutex.h> 84#include <linux/rwsem.h> 85#include <linux/string.h> 86#include <linux/mm.h> 87#include <linux/socket.h> 88#include <linux/sockios.h> 89#include <linux/errno.h> 90#include <linux/interrupt.h> 91#include <linux/if_ether.h> 92#include <linux/netdevice.h> 93#include <linux/etherdevice.h> 94#include <linux/ethtool.h> 95#include <linux/ethtool_netlink.h> 96#include <linux/skbuff.h> 97#include <linux/kthread.h> 98#include <linux/bpf.h> 99#include <linux/bpf_trace.h> 100#include <net/net_namespace.h> 101#include <net/sock.h> 102#include <net/busy_poll.h> 103#include <linux/rtnetlink.h> 104#include <linux/stat.h> 105#include <net/dsa.h> 106#include <net/dst.h> 107#include <net/dst_metadata.h> 108#include <net/gro.h> 109#include <net/netdev_queues.h> 110#include <net/pkt_sched.h> 111#include <net/pkt_cls.h> 112#include <net/checksum.h> 113#include <net/xfrm.h> 114#include <net/tcx.h> 115#include <linux/highmem.h> 116#include <linux/init.h> 117#include <linux/module.h> 118#include <linux/netpoll.h> 119#include <linux/rcupdate.h> 120#include <linux/delay.h> 121#include <net/iw_handler.h> 122#include <asm/current.h> 123#include <linux/audit.h> 124#include <linux/dmaengine.h> 125#include <linux/err.h> 126#include <linux/ctype.h> 127#include <linux/if_arp.h> 128#include <linux/if_vlan.h> 129#include <linux/ip.h> 130#include <net/ip.h> 131#include <net/mpls.h> 132#include <linux/ipv6.h> 133#include <linux/in.h> 134#include <linux/jhash.h> 135#include <linux/random.h> 136#include <trace/events/napi.h> 137#include <trace/events/net.h> 138#include <trace/events/skb.h> 139#include <trace/events/qdisc.h> 140#include <trace/events/xdp.h> 141#include <linux/inetdevice.h> 142#include <linux/cpu_rmap.h> 143#include <linux/static_key.h> 144#include <linux/hashtable.h> 145#include <linux/vmalloc.h> 146#include <linux/if_macvlan.h> 147#include <linux/errqueue.h> 148#include <linux/hrtimer.h> 149#include <linux/netfilter_netdev.h> 150#include <linux/crash_dump.h> 151#include <linux/sctp.h> 152#include <net/udp_tunnel.h> 153#include <linux/net_namespace.h> 154#include <linux/indirect_call_wrapper.h> 155#include <net/devlink.h> 156#include <linux/pm_runtime.h> 157#include <linux/prandom.h> 158#include <linux/once_lite.h> 159#include <net/netdev_lock.h> 160#include <net/netdev_rx_queue.h> 161#include <net/page_pool/types.h> 162#include <net/page_pool/helpers.h> 163#include <net/page_pool/memory_provider.h> 164#include <net/rps.h> 165#include <linux/phy_link_topology.h> 166 167#include "dev.h" 168#include "devmem.h" 169#include "net-sysfs.h" 170 171static DEFINE_SPINLOCK(ptype_lock); 172struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 173 174static int netif_rx_internal(struct sk_buff *skb); 175static int call_netdevice_notifiers_extack(unsigned long val, 176 struct net_device *dev, 177 struct netlink_ext_ack *extack); 178 179static DEFINE_MUTEX(ifalias_mutex); 180 181/* protects napi_hash addition/deletion and napi_gen_id */ 182static DEFINE_SPINLOCK(napi_hash_lock); 183 184static unsigned int napi_gen_id = NR_CPUS; 185static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 186 187static inline void dev_base_seq_inc(struct net *net) 188{ 189 unsigned int val = net->dev_base_seq + 1; 190 191 WRITE_ONCE(net->dev_base_seq, val ?: 1); 192} 193 194static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 195{ 196 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 197 198 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 199} 200 201static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 202{ 203 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 204} 205 206#ifndef CONFIG_PREEMPT_RT 207 208static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key); 209 210static int __init setup_backlog_napi_threads(char *arg) 211{ 212 static_branch_enable(&use_backlog_threads_key); 213 return 0; 214} 215early_param("thread_backlog_napi", setup_backlog_napi_threads); 216 217static bool use_backlog_threads(void) 218{ 219 return static_branch_unlikely(&use_backlog_threads_key); 220} 221 222#else 223 224static bool use_backlog_threads(void) 225{ 226 return true; 227} 228 229#endif 230 231static inline void backlog_lock_irq_save(struct softnet_data *sd, 232 unsigned long *flags) 233{ 234 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 235 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 236 else 237 local_irq_save(*flags); 238} 239 240static inline void backlog_lock_irq_disable(struct softnet_data *sd) 241{ 242 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 243 spin_lock_irq(&sd->input_pkt_queue.lock); 244 else 245 local_irq_disable(); 246} 247 248static inline void backlog_unlock_irq_restore(struct softnet_data *sd, 249 unsigned long *flags) 250{ 251 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 252 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 253 else 254 local_irq_restore(*flags); 255} 256 257static inline void backlog_unlock_irq_enable(struct softnet_data *sd) 258{ 259 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 260 spin_unlock_irq(&sd->input_pkt_queue.lock); 261 else 262 local_irq_enable(); 263} 264 265static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 266 const char *name) 267{ 268 struct netdev_name_node *name_node; 269 270 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 271 if (!name_node) 272 return NULL; 273 INIT_HLIST_NODE(&name_node->hlist); 274 name_node->dev = dev; 275 name_node->name = name; 276 return name_node; 277} 278 279static struct netdev_name_node * 280netdev_name_node_head_alloc(struct net_device *dev) 281{ 282 struct netdev_name_node *name_node; 283 284 name_node = netdev_name_node_alloc(dev, dev->name); 285 if (!name_node) 286 return NULL; 287 INIT_LIST_HEAD(&name_node->list); 288 return name_node; 289} 290 291static void netdev_name_node_free(struct netdev_name_node *name_node) 292{ 293 kfree(name_node); 294} 295 296static void netdev_name_node_add(struct net *net, 297 struct netdev_name_node *name_node) 298{ 299 hlist_add_head_rcu(&name_node->hlist, 300 dev_name_hash(net, name_node->name)); 301} 302 303static void netdev_name_node_del(struct netdev_name_node *name_node) 304{ 305 hlist_del_rcu(&name_node->hlist); 306} 307 308static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 309 const char *name) 310{ 311 struct hlist_head *head = dev_name_hash(net, name); 312 struct netdev_name_node *name_node; 313 314 hlist_for_each_entry(name_node, head, hlist) 315 if (!strcmp(name_node->name, name)) 316 return name_node; 317 return NULL; 318} 319 320static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 321 const char *name) 322{ 323 struct hlist_head *head = dev_name_hash(net, name); 324 struct netdev_name_node *name_node; 325 326 hlist_for_each_entry_rcu(name_node, head, hlist) 327 if (!strcmp(name_node->name, name)) 328 return name_node; 329 return NULL; 330} 331 332bool netdev_name_in_use(struct net *net, const char *name) 333{ 334 return netdev_name_node_lookup(net, name); 335} 336EXPORT_SYMBOL(netdev_name_in_use); 337 338int netdev_name_node_alt_create(struct net_device *dev, const char *name) 339{ 340 struct netdev_name_node *name_node; 341 struct net *net = dev_net(dev); 342 343 name_node = netdev_name_node_lookup(net, name); 344 if (name_node) 345 return -EEXIST; 346 name_node = netdev_name_node_alloc(dev, name); 347 if (!name_node) 348 return -ENOMEM; 349 netdev_name_node_add(net, name_node); 350 /* The node that holds dev->name acts as a head of per-device list. */ 351 list_add_tail_rcu(&name_node->list, &dev->name_node->list); 352 353 return 0; 354} 355 356static void netdev_name_node_alt_free(struct rcu_head *head) 357{ 358 struct netdev_name_node *name_node = 359 container_of(head, struct netdev_name_node, rcu); 360 361 kfree(name_node->name); 362 netdev_name_node_free(name_node); 363} 364 365static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 366{ 367 netdev_name_node_del(name_node); 368 list_del(&name_node->list); 369 call_rcu(&name_node->rcu, netdev_name_node_alt_free); 370} 371 372int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 373{ 374 struct netdev_name_node *name_node; 375 struct net *net = dev_net(dev); 376 377 name_node = netdev_name_node_lookup(net, name); 378 if (!name_node) 379 return -ENOENT; 380 /* lookup might have found our primary name or a name belonging 381 * to another device. 382 */ 383 if (name_node == dev->name_node || name_node->dev != dev) 384 return -EINVAL; 385 386 __netdev_name_node_alt_destroy(name_node); 387 return 0; 388} 389 390static void netdev_name_node_alt_flush(struct net_device *dev) 391{ 392 struct netdev_name_node *name_node, *tmp; 393 394 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) { 395 list_del(&name_node->list); 396 netdev_name_node_alt_free(&name_node->rcu); 397 } 398} 399 400/* Device list insertion */ 401static void list_netdevice(struct net_device *dev) 402{ 403 struct netdev_name_node *name_node; 404 struct net *net = dev_net(dev); 405 406 ASSERT_RTNL(); 407 408 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 409 netdev_name_node_add(net, dev->name_node); 410 hlist_add_head_rcu(&dev->index_hlist, 411 dev_index_hash(net, dev->ifindex)); 412 413 netdev_for_each_altname(dev, name_node) 414 netdev_name_node_add(net, name_node); 415 416 /* We reserved the ifindex, this can't fail */ 417 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 418 419 dev_base_seq_inc(net); 420} 421 422/* Device list removal 423 * caller must respect a RCU grace period before freeing/reusing dev 424 */ 425static void unlist_netdevice(struct net_device *dev) 426{ 427 struct netdev_name_node *name_node; 428 struct net *net = dev_net(dev); 429 430 ASSERT_RTNL(); 431 432 xa_erase(&net->dev_by_index, dev->ifindex); 433 434 netdev_for_each_altname(dev, name_node) 435 netdev_name_node_del(name_node); 436 437 /* Unlink dev from the device chain */ 438 list_del_rcu(&dev->dev_list); 439 netdev_name_node_del(dev->name_node); 440 hlist_del_rcu(&dev->index_hlist); 441 442 dev_base_seq_inc(dev_net(dev)); 443} 444 445/* 446 * Our notifier list 447 */ 448 449static RAW_NOTIFIER_HEAD(netdev_chain); 450 451/* 452 * Device drivers call our routines to queue packets here. We empty the 453 * queue in the local softnet handler. 454 */ 455 456DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = { 457 .process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock), 458}; 459EXPORT_PER_CPU_SYMBOL(softnet_data); 460 461/* Page_pool has a lockless array/stack to alloc/recycle pages. 462 * PP consumers must pay attention to run APIs in the appropriate context 463 * (e.g. NAPI context). 464 */ 465DEFINE_PER_CPU(struct page_pool_bh, system_page_pool) = { 466 .bh_lock = INIT_LOCAL_LOCK(bh_lock), 467}; 468 469#ifdef CONFIG_LOCKDEP 470/* 471 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 472 * according to dev->type 473 */ 474static const unsigned short netdev_lock_type[] = { 475 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 476 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 477 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 478 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 479 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 480 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 481 ARPHRD_CAN, ARPHRD_MCTP, 482 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 483 ARPHRD_RAWHDLC, ARPHRD_RAWIP, 484 ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 485 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 486 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 487 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 488 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 489 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 490 ARPHRD_IEEE80211_RADIOTAP, 491 ARPHRD_IEEE802154, ARPHRD_IEEE802154_MONITOR, 492 ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 493 ARPHRD_CAIF, ARPHRD_IP6GRE, ARPHRD_NETLINK, ARPHRD_6LOWPAN, 494 ARPHRD_VSOCKMON, 495 ARPHRD_VOID, ARPHRD_NONE}; 496 497static const char *const netdev_lock_name[] = { 498 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 499 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 500 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 501 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 502 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 503 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 504 "_xmit_CAN", "_xmit_MCTP", 505 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 506 "_xmit_RAWHDLC", "_xmit_RAWIP", 507 "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 508 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 509 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 510 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 511 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 512 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 513 "_xmit_IEEE80211_RADIOTAP", 514 "_xmit_IEEE802154", "_xmit_IEEE802154_MONITOR", 515 "_xmit_PHONET", "_xmit_PHONET_PIPE", 516 "_xmit_CAIF", "_xmit_IP6GRE", "_xmit_NETLINK", "_xmit_6LOWPAN", 517 "_xmit_VSOCKMON", 518 "_xmit_VOID", "_xmit_NONE"}; 519 520static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 521static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 522 523static inline unsigned short netdev_lock_pos(unsigned short dev_type) 524{ 525 int i; 526 527 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 528 if (netdev_lock_type[i] == dev_type) 529 return i; 530 /* the last key is used by default */ 531 WARN_ONCE(1, "netdev_lock_pos() could not find dev_type=%u\n", dev_type); 532 return ARRAY_SIZE(netdev_lock_type) - 1; 533} 534 535static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 536 unsigned short dev_type) 537{ 538 int i; 539 540 i = netdev_lock_pos(dev_type); 541 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 542 netdev_lock_name[i]); 543} 544 545static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 546{ 547 int i; 548 549 i = netdev_lock_pos(dev->type); 550 lockdep_set_class_and_name(&dev->addr_list_lock, 551 &netdev_addr_lock_key[i], 552 netdev_lock_name[i]); 553} 554#else 555static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 556 unsigned short dev_type) 557{ 558} 559 560static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 561{ 562} 563#endif 564 565/******************************************************************************* 566 * 567 * Protocol management and registration routines 568 * 569 *******************************************************************************/ 570 571 572/* 573 * Add a protocol ID to the list. Now that the input handler is 574 * smarter we can dispense with all the messy stuff that used to be 575 * here. 576 * 577 * BEWARE!!! Protocol handlers, mangling input packets, 578 * MUST BE last in hash buckets and checking protocol handlers 579 * MUST start from promiscuous ptype_all chain in net_bh. 580 * It is true now, do not change it. 581 * Explanation follows: if protocol handler, mangling packet, will 582 * be the first on list, it is not able to sense, that packet 583 * is cloned and should be copied-on-write, so that it will 584 * change it and subsequent readers will get broken packet. 585 * --ANK (980803) 586 */ 587 588static inline struct list_head *ptype_head(const struct packet_type *pt) 589{ 590 if (pt->type == htons(ETH_P_ALL)) { 591 if (!pt->af_packet_net && !pt->dev) 592 return NULL; 593 594 return pt->dev ? &pt->dev->ptype_all : 595 &pt->af_packet_net->ptype_all; 596 } 597 598 if (pt->dev) 599 return &pt->dev->ptype_specific; 600 601 return pt->af_packet_net ? &pt->af_packet_net->ptype_specific : 602 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 603} 604 605/** 606 * dev_add_pack - add packet handler 607 * @pt: packet type declaration 608 * 609 * Add a protocol handler to the networking stack. The passed &packet_type 610 * is linked into kernel lists and may not be freed until it has been 611 * removed from the kernel lists. 612 * 613 * This call does not sleep therefore it can not 614 * guarantee all CPU's that are in middle of receiving packets 615 * will see the new packet type (until the next received packet). 616 */ 617 618void dev_add_pack(struct packet_type *pt) 619{ 620 struct list_head *head = ptype_head(pt); 621 622 if (WARN_ON_ONCE(!head)) 623 return; 624 625 spin_lock(&ptype_lock); 626 list_add_rcu(&pt->list, head); 627 spin_unlock(&ptype_lock); 628} 629EXPORT_SYMBOL(dev_add_pack); 630 631/** 632 * __dev_remove_pack - remove packet handler 633 * @pt: packet type declaration 634 * 635 * Remove a protocol handler that was previously added to the kernel 636 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 637 * from the kernel lists and can be freed or reused once this function 638 * returns. 639 * 640 * The packet type might still be in use by receivers 641 * and must not be freed until after all the CPU's have gone 642 * through a quiescent state. 643 */ 644void __dev_remove_pack(struct packet_type *pt) 645{ 646 struct list_head *head = ptype_head(pt); 647 struct packet_type *pt1; 648 649 if (!head) 650 return; 651 652 spin_lock(&ptype_lock); 653 654 list_for_each_entry(pt1, head, list) { 655 if (pt == pt1) { 656 list_del_rcu(&pt->list); 657 goto out; 658 } 659 } 660 661 pr_warn("dev_remove_pack: %p not found\n", pt); 662out: 663 spin_unlock(&ptype_lock); 664} 665EXPORT_SYMBOL(__dev_remove_pack); 666 667/** 668 * dev_remove_pack - remove packet handler 669 * @pt: packet type declaration 670 * 671 * Remove a protocol handler that was previously added to the kernel 672 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 673 * from the kernel lists and can be freed or reused once this function 674 * returns. 675 * 676 * This call sleeps to guarantee that no CPU is looking at the packet 677 * type after return. 678 */ 679void dev_remove_pack(struct packet_type *pt) 680{ 681 __dev_remove_pack(pt); 682 683 synchronize_net(); 684} 685EXPORT_SYMBOL(dev_remove_pack); 686 687 688/******************************************************************************* 689 * 690 * Device Interface Subroutines 691 * 692 *******************************************************************************/ 693 694/** 695 * dev_get_iflink - get 'iflink' value of a interface 696 * @dev: targeted interface 697 * 698 * Indicates the ifindex the interface is linked to. 699 * Physical interfaces have the same 'ifindex' and 'iflink' values. 700 */ 701 702int dev_get_iflink(const struct net_device *dev) 703{ 704 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 705 return dev->netdev_ops->ndo_get_iflink(dev); 706 707 return READ_ONCE(dev->ifindex); 708} 709EXPORT_SYMBOL(dev_get_iflink); 710 711/** 712 * dev_fill_metadata_dst - Retrieve tunnel egress information. 713 * @dev: targeted interface 714 * @skb: The packet. 715 * 716 * For better visibility of tunnel traffic OVS needs to retrieve 717 * egress tunnel information for a packet. Following API allows 718 * user to get this info. 719 */ 720int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 721{ 722 struct ip_tunnel_info *info; 723 724 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 725 return -EINVAL; 726 727 info = skb_tunnel_info_unclone(skb); 728 if (!info) 729 return -ENOMEM; 730 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 731 return -EINVAL; 732 733 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 734} 735EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 736 737static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 738{ 739 int k = stack->num_paths++; 740 741 if (k >= NET_DEVICE_PATH_STACK_MAX) 742 return NULL; 743 744 return &stack->path[k]; 745} 746 747int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 748 struct net_device_path_stack *stack) 749{ 750 const struct net_device *last_dev; 751 struct net_device_path_ctx ctx = { 752 .dev = dev, 753 }; 754 struct net_device_path *path; 755 int ret = 0; 756 757 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 758 stack->num_paths = 0; 759 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 760 last_dev = ctx.dev; 761 path = dev_fwd_path(stack); 762 if (!path) 763 return -1; 764 765 memset(path, 0, sizeof(struct net_device_path)); 766 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 767 if (ret < 0) 768 return -1; 769 770 if (WARN_ON_ONCE(last_dev == ctx.dev)) 771 return -1; 772 } 773 774 if (!ctx.dev) 775 return ret; 776 777 path = dev_fwd_path(stack); 778 if (!path) 779 return -1; 780 path->type = DEV_PATH_ETHERNET; 781 path->dev = ctx.dev; 782 783 return ret; 784} 785EXPORT_SYMBOL_GPL(dev_fill_forward_path); 786 787/* must be called under rcu_read_lock(), as we dont take a reference */ 788static struct napi_struct *napi_by_id(unsigned int napi_id) 789{ 790 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 791 struct napi_struct *napi; 792 793 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 794 if (napi->napi_id == napi_id) 795 return napi; 796 797 return NULL; 798} 799 800/* must be called under rcu_read_lock(), as we dont take a reference */ 801static struct napi_struct * 802netdev_napi_by_id(struct net *net, unsigned int napi_id) 803{ 804 struct napi_struct *napi; 805 806 napi = napi_by_id(napi_id); 807 if (!napi) 808 return NULL; 809 810 if (WARN_ON_ONCE(!napi->dev)) 811 return NULL; 812 if (!net_eq(net, dev_net(napi->dev))) 813 return NULL; 814 815 return napi; 816} 817 818/** 819 * netdev_napi_by_id_lock() - find a device by NAPI ID and lock it 820 * @net: the applicable net namespace 821 * @napi_id: ID of a NAPI of a target device 822 * 823 * Find a NAPI instance with @napi_id. Lock its device. 824 * The device must be in %NETREG_REGISTERED state for lookup to succeed. 825 * netdev_unlock() must be called to release it. 826 * 827 * Return: pointer to NAPI, its device with lock held, NULL if not found. 828 */ 829struct napi_struct * 830netdev_napi_by_id_lock(struct net *net, unsigned int napi_id) 831{ 832 struct napi_struct *napi; 833 struct net_device *dev; 834 835 rcu_read_lock(); 836 napi = netdev_napi_by_id(net, napi_id); 837 if (!napi || READ_ONCE(napi->dev->reg_state) != NETREG_REGISTERED) { 838 rcu_read_unlock(); 839 return NULL; 840 } 841 842 dev = napi->dev; 843 dev_hold(dev); 844 rcu_read_unlock(); 845 846 dev = __netdev_put_lock(dev, net); 847 if (!dev) 848 return NULL; 849 850 rcu_read_lock(); 851 napi = netdev_napi_by_id(net, napi_id); 852 if (napi && napi->dev != dev) 853 napi = NULL; 854 rcu_read_unlock(); 855 856 if (!napi) 857 netdev_unlock(dev); 858 return napi; 859} 860 861/** 862 * __dev_get_by_name - find a device by its name 863 * @net: the applicable net namespace 864 * @name: name to find 865 * 866 * Find an interface by name. Must be called under RTNL semaphore. 867 * If the name is found a pointer to the device is returned. 868 * If the name is not found then %NULL is returned. The 869 * reference counters are not incremented so the caller must be 870 * careful with locks. 871 */ 872 873struct net_device *__dev_get_by_name(struct net *net, const char *name) 874{ 875 struct netdev_name_node *node_name; 876 877 node_name = netdev_name_node_lookup(net, name); 878 return node_name ? node_name->dev : NULL; 879} 880EXPORT_SYMBOL(__dev_get_by_name); 881 882/** 883 * dev_get_by_name_rcu - find a device by its name 884 * @net: the applicable net namespace 885 * @name: name to find 886 * 887 * Find an interface by name. 888 * If the name is found a pointer to the device is returned. 889 * If the name is not found then %NULL is returned. 890 * The reference counters are not incremented so the caller must be 891 * careful with locks. The caller must hold RCU lock. 892 */ 893 894struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 895{ 896 struct netdev_name_node *node_name; 897 898 node_name = netdev_name_node_lookup_rcu(net, name); 899 return node_name ? node_name->dev : NULL; 900} 901EXPORT_SYMBOL(dev_get_by_name_rcu); 902 903/* Deprecated for new users, call netdev_get_by_name() instead */ 904struct net_device *dev_get_by_name(struct net *net, const char *name) 905{ 906 struct net_device *dev; 907 908 rcu_read_lock(); 909 dev = dev_get_by_name_rcu(net, name); 910 dev_hold(dev); 911 rcu_read_unlock(); 912 return dev; 913} 914EXPORT_SYMBOL(dev_get_by_name); 915 916/** 917 * netdev_get_by_name() - find a device by its name 918 * @net: the applicable net namespace 919 * @name: name to find 920 * @tracker: tracking object for the acquired reference 921 * @gfp: allocation flags for the tracker 922 * 923 * Find an interface by name. This can be called from any 924 * context and does its own locking. The returned handle has 925 * the usage count incremented and the caller must use netdev_put() to 926 * release it when it is no longer needed. %NULL is returned if no 927 * matching device is found. 928 */ 929struct net_device *netdev_get_by_name(struct net *net, const char *name, 930 netdevice_tracker *tracker, gfp_t gfp) 931{ 932 struct net_device *dev; 933 934 dev = dev_get_by_name(net, name); 935 if (dev) 936 netdev_tracker_alloc(dev, tracker, gfp); 937 return dev; 938} 939EXPORT_SYMBOL(netdev_get_by_name); 940 941/** 942 * __dev_get_by_index - find a device by its ifindex 943 * @net: the applicable net namespace 944 * @ifindex: index of device 945 * 946 * Search for an interface by index. Returns %NULL if the device 947 * is not found or a pointer to the device. The device has not 948 * had its reference counter increased so the caller must be careful 949 * about locking. The caller must hold the RTNL semaphore. 950 */ 951 952struct net_device *__dev_get_by_index(struct net *net, int ifindex) 953{ 954 struct net_device *dev; 955 struct hlist_head *head = dev_index_hash(net, ifindex); 956 957 hlist_for_each_entry(dev, head, index_hlist) 958 if (dev->ifindex == ifindex) 959 return dev; 960 961 return NULL; 962} 963EXPORT_SYMBOL(__dev_get_by_index); 964 965/** 966 * dev_get_by_index_rcu - find a device by its ifindex 967 * @net: the applicable net namespace 968 * @ifindex: index of device 969 * 970 * Search for an interface by index. Returns %NULL if the device 971 * is not found or a pointer to the device. The device has not 972 * had its reference counter increased so the caller must be careful 973 * about locking. The caller must hold RCU lock. 974 */ 975 976struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 977{ 978 struct net_device *dev; 979 struct hlist_head *head = dev_index_hash(net, ifindex); 980 981 hlist_for_each_entry_rcu(dev, head, index_hlist) 982 if (dev->ifindex == ifindex) 983 return dev; 984 985 return NULL; 986} 987EXPORT_SYMBOL(dev_get_by_index_rcu); 988 989/* Deprecated for new users, call netdev_get_by_index() instead */ 990struct net_device *dev_get_by_index(struct net *net, int ifindex) 991{ 992 struct net_device *dev; 993 994 rcu_read_lock(); 995 dev = dev_get_by_index_rcu(net, ifindex); 996 dev_hold(dev); 997 rcu_read_unlock(); 998 return dev; 999} 1000EXPORT_SYMBOL(dev_get_by_index); 1001 1002/** 1003 * netdev_get_by_index() - find a device by its ifindex 1004 * @net: the applicable net namespace 1005 * @ifindex: index of device 1006 * @tracker: tracking object for the acquired reference 1007 * @gfp: allocation flags for the tracker 1008 * 1009 * Search for an interface by index. Returns NULL if the device 1010 * is not found or a pointer to the device. The device returned has 1011 * had a reference added and the pointer is safe until the user calls 1012 * netdev_put() to indicate they have finished with it. 1013 */ 1014struct net_device *netdev_get_by_index(struct net *net, int ifindex, 1015 netdevice_tracker *tracker, gfp_t gfp) 1016{ 1017 struct net_device *dev; 1018 1019 dev = dev_get_by_index(net, ifindex); 1020 if (dev) 1021 netdev_tracker_alloc(dev, tracker, gfp); 1022 return dev; 1023} 1024EXPORT_SYMBOL(netdev_get_by_index); 1025 1026/** 1027 * dev_get_by_napi_id - find a device by napi_id 1028 * @napi_id: ID of the NAPI struct 1029 * 1030 * Search for an interface by NAPI ID. Returns %NULL if the device 1031 * is not found or a pointer to the device. The device has not had 1032 * its reference counter increased so the caller must be careful 1033 * about locking. The caller must hold RCU lock. 1034 */ 1035struct net_device *dev_get_by_napi_id(unsigned int napi_id) 1036{ 1037 struct napi_struct *napi; 1038 1039 WARN_ON_ONCE(!rcu_read_lock_held()); 1040 1041 if (!napi_id_valid(napi_id)) 1042 return NULL; 1043 1044 napi = napi_by_id(napi_id); 1045 1046 return napi ? napi->dev : NULL; 1047} 1048 1049/* Release the held reference on the net_device, and if the net_device 1050 * is still registered try to lock the instance lock. If device is being 1051 * unregistered NULL will be returned (but the reference has been released, 1052 * either way!) 1053 * 1054 * This helper is intended for locking net_device after it has been looked up 1055 * using a lockless lookup helper. Lock prevents the instance from going away. 1056 */ 1057struct net_device *__netdev_put_lock(struct net_device *dev, struct net *net) 1058{ 1059 netdev_lock(dev); 1060 if (dev->reg_state > NETREG_REGISTERED || 1061 dev->moving_ns || !net_eq(dev_net(dev), net)) { 1062 netdev_unlock(dev); 1063 dev_put(dev); 1064 return NULL; 1065 } 1066 dev_put(dev); 1067 return dev; 1068} 1069 1070static struct net_device * 1071__netdev_put_lock_ops_compat(struct net_device *dev, struct net *net) 1072{ 1073 netdev_lock_ops_compat(dev); 1074 if (dev->reg_state > NETREG_REGISTERED || 1075 dev->moving_ns || !net_eq(dev_net(dev), net)) { 1076 netdev_unlock_ops_compat(dev); 1077 dev_put(dev); 1078 return NULL; 1079 } 1080 dev_put(dev); 1081 return dev; 1082} 1083 1084/** 1085 * netdev_get_by_index_lock() - find a device by its ifindex 1086 * @net: the applicable net namespace 1087 * @ifindex: index of device 1088 * 1089 * Search for an interface by index. If a valid device 1090 * with @ifindex is found it will be returned with netdev->lock held. 1091 * netdev_unlock() must be called to release it. 1092 * 1093 * Return: pointer to a device with lock held, NULL if not found. 1094 */ 1095struct net_device *netdev_get_by_index_lock(struct net *net, int ifindex) 1096{ 1097 struct net_device *dev; 1098 1099 dev = dev_get_by_index(net, ifindex); 1100 if (!dev) 1101 return NULL; 1102 1103 return __netdev_put_lock(dev, net); 1104} 1105 1106struct net_device * 1107netdev_get_by_index_lock_ops_compat(struct net *net, int ifindex) 1108{ 1109 struct net_device *dev; 1110 1111 dev = dev_get_by_index(net, ifindex); 1112 if (!dev) 1113 return NULL; 1114 1115 return __netdev_put_lock_ops_compat(dev, net); 1116} 1117 1118struct net_device * 1119netdev_xa_find_lock(struct net *net, struct net_device *dev, 1120 unsigned long *index) 1121{ 1122 if (dev) 1123 netdev_unlock(dev); 1124 1125 do { 1126 rcu_read_lock(); 1127 dev = xa_find(&net->dev_by_index, index, ULONG_MAX, XA_PRESENT); 1128 if (!dev) { 1129 rcu_read_unlock(); 1130 return NULL; 1131 } 1132 dev_hold(dev); 1133 rcu_read_unlock(); 1134 1135 dev = __netdev_put_lock(dev, net); 1136 if (dev) 1137 return dev; 1138 1139 (*index)++; 1140 } while (true); 1141} 1142 1143struct net_device * 1144netdev_xa_find_lock_ops_compat(struct net *net, struct net_device *dev, 1145 unsigned long *index) 1146{ 1147 if (dev) 1148 netdev_unlock_ops_compat(dev); 1149 1150 do { 1151 rcu_read_lock(); 1152 dev = xa_find(&net->dev_by_index, index, ULONG_MAX, XA_PRESENT); 1153 if (!dev) { 1154 rcu_read_unlock(); 1155 return NULL; 1156 } 1157 dev_hold(dev); 1158 rcu_read_unlock(); 1159 1160 dev = __netdev_put_lock_ops_compat(dev, net); 1161 if (dev) 1162 return dev; 1163 1164 (*index)++; 1165 } while (true); 1166} 1167 1168static DEFINE_SEQLOCK(netdev_rename_lock); 1169 1170void netdev_copy_name(struct net_device *dev, char *name) 1171{ 1172 unsigned int seq; 1173 1174 do { 1175 seq = read_seqbegin(&netdev_rename_lock); 1176 strscpy(name, dev->name, IFNAMSIZ); 1177 } while (read_seqretry(&netdev_rename_lock, seq)); 1178} 1179EXPORT_IPV6_MOD_GPL(netdev_copy_name); 1180 1181/** 1182 * netdev_get_name - get a netdevice name, knowing its ifindex. 1183 * @net: network namespace 1184 * @name: a pointer to the buffer where the name will be stored. 1185 * @ifindex: the ifindex of the interface to get the name from. 1186 */ 1187int netdev_get_name(struct net *net, char *name, int ifindex) 1188{ 1189 struct net_device *dev; 1190 int ret; 1191 1192 rcu_read_lock(); 1193 1194 dev = dev_get_by_index_rcu(net, ifindex); 1195 if (!dev) { 1196 ret = -ENODEV; 1197 goto out; 1198 } 1199 1200 netdev_copy_name(dev, name); 1201 1202 ret = 0; 1203out: 1204 rcu_read_unlock(); 1205 return ret; 1206} 1207 1208static bool dev_addr_cmp(struct net_device *dev, unsigned short type, 1209 const char *ha) 1210{ 1211 return dev->type == type && !memcmp(dev->dev_addr, ha, dev->addr_len); 1212} 1213 1214/** 1215 * dev_getbyhwaddr_rcu - find a device by its hardware address 1216 * @net: the applicable net namespace 1217 * @type: media type of device 1218 * @ha: hardware address 1219 * 1220 * Search for an interface by MAC address. Returns NULL if the device 1221 * is not found or a pointer to the device. 1222 * The caller must hold RCU. 1223 * The returned device has not had its ref count increased 1224 * and the caller must therefore be careful about locking 1225 * 1226 */ 1227 1228struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1229 const char *ha) 1230{ 1231 struct net_device *dev; 1232 1233 for_each_netdev_rcu(net, dev) 1234 if (dev_addr_cmp(dev, type, ha)) 1235 return dev; 1236 1237 return NULL; 1238} 1239EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1240 1241/** 1242 * dev_getbyhwaddr() - find a device by its hardware address 1243 * @net: the applicable net namespace 1244 * @type: media type of device 1245 * @ha: hardware address 1246 * 1247 * Similar to dev_getbyhwaddr_rcu(), but the owner needs to hold 1248 * rtnl_lock. 1249 * 1250 * Context: rtnl_lock() must be held. 1251 * Return: pointer to the net_device, or NULL if not found 1252 */ 1253struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, 1254 const char *ha) 1255{ 1256 struct net_device *dev; 1257 1258 ASSERT_RTNL(); 1259 for_each_netdev(net, dev) 1260 if (dev_addr_cmp(dev, type, ha)) 1261 return dev; 1262 1263 return NULL; 1264} 1265EXPORT_SYMBOL(dev_getbyhwaddr); 1266 1267struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1268{ 1269 struct net_device *dev, *ret = NULL; 1270 1271 rcu_read_lock(); 1272 for_each_netdev_rcu(net, dev) 1273 if (dev->type == type) { 1274 dev_hold(dev); 1275 ret = dev; 1276 break; 1277 } 1278 rcu_read_unlock(); 1279 return ret; 1280} 1281EXPORT_SYMBOL(dev_getfirstbyhwtype); 1282 1283/** 1284 * netdev_get_by_flags_rcu - find any device with given flags 1285 * @net: the applicable net namespace 1286 * @tracker: tracking object for the acquired reference 1287 * @if_flags: IFF_* values 1288 * @mask: bitmask of bits in if_flags to check 1289 * 1290 * Search for any interface with the given flags. 1291 * 1292 * Context: rcu_read_lock() must be held. 1293 * Returns: NULL if a device is not found or a pointer to the device. 1294 */ 1295struct net_device *netdev_get_by_flags_rcu(struct net *net, netdevice_tracker *tracker, 1296 unsigned short if_flags, unsigned short mask) 1297{ 1298 struct net_device *dev; 1299 1300 for_each_netdev_rcu(net, dev) { 1301 if (((READ_ONCE(dev->flags) ^ if_flags) & mask) == 0) { 1302 netdev_hold(dev, tracker, GFP_ATOMIC); 1303 return dev; 1304 } 1305 } 1306 1307 return NULL; 1308} 1309EXPORT_IPV6_MOD(netdev_get_by_flags_rcu); 1310 1311/** 1312 * dev_valid_name - check if name is okay for network device 1313 * @name: name string 1314 * 1315 * Network device names need to be valid file names to 1316 * allow sysfs to work. We also disallow any kind of 1317 * whitespace. 1318 */ 1319bool dev_valid_name(const char *name) 1320{ 1321 if (*name == '\0') 1322 return false; 1323 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1324 return false; 1325 if (!strcmp(name, ".") || !strcmp(name, "..")) 1326 return false; 1327 1328 while (*name) { 1329 if (*name == '/' || *name == ':' || isspace(*name)) 1330 return false; 1331 name++; 1332 } 1333 return true; 1334} 1335EXPORT_SYMBOL(dev_valid_name); 1336 1337/** 1338 * __dev_alloc_name - allocate a name for a device 1339 * @net: network namespace to allocate the device name in 1340 * @name: name format string 1341 * @res: result name string 1342 * 1343 * Passed a format string - eg "lt%d" it will try and find a suitable 1344 * id. It scans list of devices to build up a free map, then chooses 1345 * the first empty slot. The caller must hold the dev_base or rtnl lock 1346 * while allocating the name and adding the device in order to avoid 1347 * duplicates. 1348 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1349 * Returns the number of the unit assigned or a negative errno code. 1350 */ 1351 1352static int __dev_alloc_name(struct net *net, const char *name, char *res) 1353{ 1354 int i = 0; 1355 const char *p; 1356 const int max_netdevices = 8*PAGE_SIZE; 1357 unsigned long *inuse; 1358 struct net_device *d; 1359 char buf[IFNAMSIZ]; 1360 1361 /* Verify the string as this thing may have come from the user. 1362 * There must be one "%d" and no other "%" characters. 1363 */ 1364 p = strchr(name, '%'); 1365 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1366 return -EINVAL; 1367 1368 /* Use one page as a bit array of possible slots */ 1369 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1370 if (!inuse) 1371 return -ENOMEM; 1372 1373 for_each_netdev(net, d) { 1374 struct netdev_name_node *name_node; 1375 1376 netdev_for_each_altname(d, name_node) { 1377 if (!sscanf(name_node->name, name, &i)) 1378 continue; 1379 if (i < 0 || i >= max_netdevices) 1380 continue; 1381 1382 /* avoid cases where sscanf is not exact inverse of printf */ 1383 snprintf(buf, IFNAMSIZ, name, i); 1384 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1385 __set_bit(i, inuse); 1386 } 1387 if (!sscanf(d->name, name, &i)) 1388 continue; 1389 if (i < 0 || i >= max_netdevices) 1390 continue; 1391 1392 /* avoid cases where sscanf is not exact inverse of printf */ 1393 snprintf(buf, IFNAMSIZ, name, i); 1394 if (!strncmp(buf, d->name, IFNAMSIZ)) 1395 __set_bit(i, inuse); 1396 } 1397 1398 i = find_first_zero_bit(inuse, max_netdevices); 1399 bitmap_free(inuse); 1400 if (i == max_netdevices) 1401 return -ENFILE; 1402 1403 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1404 strscpy(buf, name, IFNAMSIZ); 1405 snprintf(res, IFNAMSIZ, buf, i); 1406 return i; 1407} 1408 1409/* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1410static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1411 const char *want_name, char *out_name, 1412 int dup_errno) 1413{ 1414 if (!dev_valid_name(want_name)) 1415 return -EINVAL; 1416 1417 if (strchr(want_name, '%')) 1418 return __dev_alloc_name(net, want_name, out_name); 1419 1420 if (netdev_name_in_use(net, want_name)) 1421 return -dup_errno; 1422 if (out_name != want_name) 1423 strscpy(out_name, want_name, IFNAMSIZ); 1424 return 0; 1425} 1426 1427/** 1428 * dev_alloc_name - allocate a name for a device 1429 * @dev: device 1430 * @name: name format string 1431 * 1432 * Passed a format string - eg "lt%d" it will try and find a suitable 1433 * id. It scans list of devices to build up a free map, then chooses 1434 * the first empty slot. The caller must hold the dev_base or rtnl lock 1435 * while allocating the name and adding the device in order to avoid 1436 * duplicates. 1437 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1438 * Returns the number of the unit assigned or a negative errno code. 1439 */ 1440 1441int dev_alloc_name(struct net_device *dev, const char *name) 1442{ 1443 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1444} 1445EXPORT_SYMBOL(dev_alloc_name); 1446 1447static int dev_get_valid_name(struct net *net, struct net_device *dev, 1448 const char *name) 1449{ 1450 int ret; 1451 1452 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1453 return ret < 0 ? ret : 0; 1454} 1455 1456int netif_change_name(struct net_device *dev, const char *newname) 1457{ 1458 struct net *net = dev_net(dev); 1459 unsigned char old_assign_type; 1460 char oldname[IFNAMSIZ]; 1461 int err = 0; 1462 int ret; 1463 1464 ASSERT_RTNL_NET(net); 1465 1466 if (!strncmp(newname, dev->name, IFNAMSIZ)) 1467 return 0; 1468 1469 memcpy(oldname, dev->name, IFNAMSIZ); 1470 1471 write_seqlock_bh(&netdev_rename_lock); 1472 err = dev_get_valid_name(net, dev, newname); 1473 write_sequnlock_bh(&netdev_rename_lock); 1474 1475 if (err < 0) 1476 return err; 1477 1478 if (oldname[0] && !strchr(oldname, '%')) 1479 netdev_info(dev, "renamed from %s%s\n", oldname, 1480 dev->flags & IFF_UP ? " (while UP)" : ""); 1481 1482 old_assign_type = dev->name_assign_type; 1483 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); 1484 1485rollback: 1486 ret = device_rename(&dev->dev, dev->name); 1487 if (ret) { 1488 write_seqlock_bh(&netdev_rename_lock); 1489 memcpy(dev->name, oldname, IFNAMSIZ); 1490 write_sequnlock_bh(&netdev_rename_lock); 1491 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1492 return ret; 1493 } 1494 1495 netdev_adjacent_rename_links(dev, oldname); 1496 1497 netdev_name_node_del(dev->name_node); 1498 1499 synchronize_net(); 1500 1501 netdev_name_node_add(net, dev->name_node); 1502 1503 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1504 ret = notifier_to_errno(ret); 1505 1506 if (ret) { 1507 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1508 if (err >= 0) { 1509 err = ret; 1510 write_seqlock_bh(&netdev_rename_lock); 1511 memcpy(dev->name, oldname, IFNAMSIZ); 1512 write_sequnlock_bh(&netdev_rename_lock); 1513 memcpy(oldname, newname, IFNAMSIZ); 1514 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1515 old_assign_type = NET_NAME_RENAMED; 1516 goto rollback; 1517 } else { 1518 netdev_err(dev, "name change rollback failed: %d\n", 1519 ret); 1520 } 1521 } 1522 1523 return err; 1524} 1525 1526int netif_set_alias(struct net_device *dev, const char *alias, size_t len) 1527{ 1528 struct dev_ifalias *new_alias = NULL; 1529 1530 if (len >= IFALIASZ) 1531 return -EINVAL; 1532 1533 if (len) { 1534 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1535 if (!new_alias) 1536 return -ENOMEM; 1537 1538 memcpy(new_alias->ifalias, alias, len); 1539 new_alias->ifalias[len] = 0; 1540 } 1541 1542 mutex_lock(&ifalias_mutex); 1543 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1544 mutex_is_locked(&ifalias_mutex)); 1545 mutex_unlock(&ifalias_mutex); 1546 1547 if (new_alias) 1548 kfree_rcu(new_alias, rcuhead); 1549 1550 return len; 1551} 1552 1553/** 1554 * dev_get_alias - get ifalias of a device 1555 * @dev: device 1556 * @name: buffer to store name of ifalias 1557 * @len: size of buffer 1558 * 1559 * get ifalias for a device. Caller must make sure dev cannot go 1560 * away, e.g. rcu read lock or own a reference count to device. 1561 */ 1562int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1563{ 1564 const struct dev_ifalias *alias; 1565 int ret = 0; 1566 1567 rcu_read_lock(); 1568 alias = rcu_dereference(dev->ifalias); 1569 if (alias) 1570 ret = snprintf(name, len, "%s", alias->ifalias); 1571 rcu_read_unlock(); 1572 1573 return ret; 1574} 1575 1576/** 1577 * netdev_features_change - device changes features 1578 * @dev: device to cause notification 1579 * 1580 * Called to indicate a device has changed features. 1581 */ 1582void netdev_features_change(struct net_device *dev) 1583{ 1584 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1585} 1586EXPORT_SYMBOL(netdev_features_change); 1587 1588void netif_state_change(struct net_device *dev) 1589{ 1590 netdev_ops_assert_locked_or_invisible(dev); 1591 1592 if (dev->flags & IFF_UP) { 1593 struct netdev_notifier_change_info change_info = { 1594 .info.dev = dev, 1595 }; 1596 1597 call_netdevice_notifiers_info(NETDEV_CHANGE, 1598 &change_info.info); 1599 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1600 } 1601} 1602 1603/** 1604 * __netdev_notify_peers - notify network peers about existence of @dev, 1605 * to be called when rtnl lock is already held. 1606 * @dev: network device 1607 * 1608 * Generate traffic such that interested network peers are aware of 1609 * @dev, such as by generating a gratuitous ARP. This may be used when 1610 * a device wants to inform the rest of the network about some sort of 1611 * reconfiguration such as a failover event or virtual machine 1612 * migration. 1613 */ 1614void __netdev_notify_peers(struct net_device *dev) 1615{ 1616 ASSERT_RTNL(); 1617 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1618 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1619} 1620EXPORT_SYMBOL(__netdev_notify_peers); 1621 1622/** 1623 * netdev_notify_peers - notify network peers about existence of @dev 1624 * @dev: network device 1625 * 1626 * Generate traffic such that interested network peers are aware of 1627 * @dev, such as by generating a gratuitous ARP. This may be used when 1628 * a device wants to inform the rest of the network about some sort of 1629 * reconfiguration such as a failover event or virtual machine 1630 * migration. 1631 */ 1632void netdev_notify_peers(struct net_device *dev) 1633{ 1634 rtnl_lock(); 1635 __netdev_notify_peers(dev); 1636 rtnl_unlock(); 1637} 1638EXPORT_SYMBOL(netdev_notify_peers); 1639 1640static int napi_threaded_poll(void *data); 1641 1642static int napi_kthread_create(struct napi_struct *n) 1643{ 1644 int err = 0; 1645 1646 /* Create and wake up the kthread once to put it in 1647 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1648 * warning and work with loadavg. 1649 */ 1650 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1651 n->dev->name, n->napi_id); 1652 if (IS_ERR(n->thread)) { 1653 err = PTR_ERR(n->thread); 1654 pr_err("kthread_run failed with err %d\n", err); 1655 n->thread = NULL; 1656 } 1657 1658 return err; 1659} 1660 1661static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1662{ 1663 const struct net_device_ops *ops = dev->netdev_ops; 1664 int ret; 1665 1666 ASSERT_RTNL(); 1667 dev_addr_check(dev); 1668 1669 if (!netif_device_present(dev)) { 1670 /* may be detached because parent is runtime-suspended */ 1671 if (dev->dev.parent) 1672 pm_runtime_resume(dev->dev.parent); 1673 if (!netif_device_present(dev)) 1674 return -ENODEV; 1675 } 1676 1677 /* Block netpoll from trying to do any rx path servicing. 1678 * If we don't do this there is a chance ndo_poll_controller 1679 * or ndo_poll may be running while we open the device 1680 */ 1681 netpoll_poll_disable(dev); 1682 1683 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1684 ret = notifier_to_errno(ret); 1685 if (ret) 1686 return ret; 1687 1688 set_bit(__LINK_STATE_START, &dev->state); 1689 1690 netdev_ops_assert_locked(dev); 1691 1692 if (ops->ndo_validate_addr) 1693 ret = ops->ndo_validate_addr(dev); 1694 1695 if (!ret && ops->ndo_open) 1696 ret = ops->ndo_open(dev); 1697 1698 netpoll_poll_enable(dev); 1699 1700 if (ret) 1701 clear_bit(__LINK_STATE_START, &dev->state); 1702 else { 1703 netif_set_up(dev, true); 1704 dev_set_rx_mode(dev); 1705 dev_activate(dev); 1706 add_device_randomness(dev->dev_addr, dev->addr_len); 1707 } 1708 1709 return ret; 1710} 1711 1712int netif_open(struct net_device *dev, struct netlink_ext_ack *extack) 1713{ 1714 int ret; 1715 1716 if (dev->flags & IFF_UP) 1717 return 0; 1718 1719 ret = __dev_open(dev, extack); 1720 if (ret < 0) 1721 return ret; 1722 1723 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1724 call_netdevice_notifiers(NETDEV_UP, dev); 1725 1726 return ret; 1727} 1728 1729static void __dev_close_many(struct list_head *head) 1730{ 1731 struct net_device *dev; 1732 1733 ASSERT_RTNL(); 1734 might_sleep(); 1735 1736 list_for_each_entry(dev, head, close_list) { 1737 /* Temporarily disable netpoll until the interface is down */ 1738 netpoll_poll_disable(dev); 1739 1740 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1741 1742 clear_bit(__LINK_STATE_START, &dev->state); 1743 1744 /* Synchronize to scheduled poll. We cannot touch poll list, it 1745 * can be even on different cpu. So just clear netif_running(). 1746 * 1747 * dev->stop() will invoke napi_disable() on all of it's 1748 * napi_struct instances on this device. 1749 */ 1750 smp_mb__after_atomic(); /* Commit netif_running(). */ 1751 } 1752 1753 dev_deactivate_many(head); 1754 1755 list_for_each_entry(dev, head, close_list) { 1756 const struct net_device_ops *ops = dev->netdev_ops; 1757 1758 /* 1759 * Call the device specific close. This cannot fail. 1760 * Only if device is UP 1761 * 1762 * We allow it to be called even after a DETACH hot-plug 1763 * event. 1764 */ 1765 1766 netdev_ops_assert_locked(dev); 1767 1768 if (ops->ndo_stop) 1769 ops->ndo_stop(dev); 1770 1771 netif_set_up(dev, false); 1772 netpoll_poll_enable(dev); 1773 } 1774} 1775 1776static void __dev_close(struct net_device *dev) 1777{ 1778 LIST_HEAD(single); 1779 1780 list_add(&dev->close_list, &single); 1781 __dev_close_many(&single); 1782 list_del(&single); 1783} 1784 1785void netif_close_many(struct list_head *head, bool unlink) 1786{ 1787 struct net_device *dev, *tmp; 1788 1789 /* Remove the devices that don't need to be closed */ 1790 list_for_each_entry_safe(dev, tmp, head, close_list) 1791 if (!(dev->flags & IFF_UP)) 1792 list_del_init(&dev->close_list); 1793 1794 __dev_close_many(head); 1795 1796 list_for_each_entry_safe(dev, tmp, head, close_list) { 1797 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1798 call_netdevice_notifiers(NETDEV_DOWN, dev); 1799 if (unlink) 1800 list_del_init(&dev->close_list); 1801 } 1802} 1803EXPORT_SYMBOL_NS_GPL(netif_close_many, "NETDEV_INTERNAL"); 1804 1805void netif_close(struct net_device *dev) 1806{ 1807 if (dev->flags & IFF_UP) { 1808 LIST_HEAD(single); 1809 1810 list_add(&dev->close_list, &single); 1811 netif_close_many(&single, true); 1812 list_del(&single); 1813 } 1814} 1815EXPORT_SYMBOL(netif_close); 1816 1817void netif_disable_lro(struct net_device *dev) 1818{ 1819 struct net_device *lower_dev; 1820 struct list_head *iter; 1821 1822 dev->wanted_features &= ~NETIF_F_LRO; 1823 netdev_update_features(dev); 1824 1825 if (unlikely(dev->features & NETIF_F_LRO)) 1826 netdev_WARN(dev, "failed to disable LRO!\n"); 1827 1828 netdev_for_each_lower_dev(dev, lower_dev, iter) { 1829 netdev_lock_ops(lower_dev); 1830 netif_disable_lro(lower_dev); 1831 netdev_unlock_ops(lower_dev); 1832 } 1833} 1834EXPORT_IPV6_MOD(netif_disable_lro); 1835 1836/** 1837 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1838 * @dev: device 1839 * 1840 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1841 * called under RTNL. This is needed if Generic XDP is installed on 1842 * the device. 1843 */ 1844static void dev_disable_gro_hw(struct net_device *dev) 1845{ 1846 dev->wanted_features &= ~NETIF_F_GRO_HW; 1847 netdev_update_features(dev); 1848 1849 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1850 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1851} 1852 1853const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1854{ 1855#define N(val) \ 1856 case NETDEV_##val: \ 1857 return "NETDEV_" __stringify(val); 1858 switch (cmd) { 1859 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1860 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1861 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1862 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1863 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1864 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1865 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1866 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1867 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1868 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1869 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1870 N(XDP_FEAT_CHANGE) 1871 } 1872#undef N 1873 return "UNKNOWN_NETDEV_EVENT"; 1874} 1875EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1876 1877static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1878 struct net_device *dev) 1879{ 1880 struct netdev_notifier_info info = { 1881 .dev = dev, 1882 }; 1883 1884 return nb->notifier_call(nb, val, &info); 1885} 1886 1887static int call_netdevice_register_notifiers(struct notifier_block *nb, 1888 struct net_device *dev) 1889{ 1890 int err; 1891 1892 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1893 err = notifier_to_errno(err); 1894 if (err) 1895 return err; 1896 1897 if (!(dev->flags & IFF_UP)) 1898 return 0; 1899 1900 call_netdevice_notifier(nb, NETDEV_UP, dev); 1901 return 0; 1902} 1903 1904static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1905 struct net_device *dev) 1906{ 1907 if (dev->flags & IFF_UP) { 1908 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1909 dev); 1910 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1911 } 1912 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1913} 1914 1915static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1916 struct net *net) 1917{ 1918 struct net_device *dev; 1919 int err; 1920 1921 for_each_netdev(net, dev) { 1922 netdev_lock_ops(dev); 1923 err = call_netdevice_register_notifiers(nb, dev); 1924 netdev_unlock_ops(dev); 1925 if (err) 1926 goto rollback; 1927 } 1928 return 0; 1929 1930rollback: 1931 for_each_netdev_continue_reverse(net, dev) 1932 call_netdevice_unregister_notifiers(nb, dev); 1933 return err; 1934} 1935 1936static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1937 struct net *net) 1938{ 1939 struct net_device *dev; 1940 1941 for_each_netdev(net, dev) 1942 call_netdevice_unregister_notifiers(nb, dev); 1943} 1944 1945static int dev_boot_phase = 1; 1946 1947/** 1948 * register_netdevice_notifier - register a network notifier block 1949 * @nb: notifier 1950 * 1951 * Register a notifier to be called when network device events occur. 1952 * The notifier passed is linked into the kernel structures and must 1953 * not be reused until it has been unregistered. A negative errno code 1954 * is returned on a failure. 1955 * 1956 * When registered all registration and up events are replayed 1957 * to the new notifier to allow device to have a race free 1958 * view of the network device list. 1959 */ 1960 1961int register_netdevice_notifier(struct notifier_block *nb) 1962{ 1963 struct net *net; 1964 int err; 1965 1966 /* Close race with setup_net() and cleanup_net() */ 1967 down_write(&pernet_ops_rwsem); 1968 1969 /* When RTNL is removed, we need protection for netdev_chain. */ 1970 rtnl_lock(); 1971 1972 err = raw_notifier_chain_register(&netdev_chain, nb); 1973 if (err) 1974 goto unlock; 1975 if (dev_boot_phase) 1976 goto unlock; 1977 for_each_net(net) { 1978 __rtnl_net_lock(net); 1979 err = call_netdevice_register_net_notifiers(nb, net); 1980 __rtnl_net_unlock(net); 1981 if (err) 1982 goto rollback; 1983 } 1984 1985unlock: 1986 rtnl_unlock(); 1987 up_write(&pernet_ops_rwsem); 1988 return err; 1989 1990rollback: 1991 for_each_net_continue_reverse(net) { 1992 __rtnl_net_lock(net); 1993 call_netdevice_unregister_net_notifiers(nb, net); 1994 __rtnl_net_unlock(net); 1995 } 1996 1997 raw_notifier_chain_unregister(&netdev_chain, nb); 1998 goto unlock; 1999} 2000EXPORT_SYMBOL(register_netdevice_notifier); 2001 2002/** 2003 * unregister_netdevice_notifier - unregister a network notifier block 2004 * @nb: notifier 2005 * 2006 * Unregister a notifier previously registered by 2007 * register_netdevice_notifier(). The notifier is unlinked into the 2008 * kernel structures and may then be reused. A negative errno code 2009 * is returned on a failure. 2010 * 2011 * After unregistering unregister and down device events are synthesized 2012 * for all devices on the device list to the removed notifier to remove 2013 * the need for special case cleanup code. 2014 */ 2015 2016int unregister_netdevice_notifier(struct notifier_block *nb) 2017{ 2018 struct net *net; 2019 int err; 2020 2021 /* Close race with setup_net() and cleanup_net() */ 2022 down_write(&pernet_ops_rwsem); 2023 rtnl_lock(); 2024 err = raw_notifier_chain_unregister(&netdev_chain, nb); 2025 if (err) 2026 goto unlock; 2027 2028 for_each_net(net) { 2029 __rtnl_net_lock(net); 2030 call_netdevice_unregister_net_notifiers(nb, net); 2031 __rtnl_net_unlock(net); 2032 } 2033 2034unlock: 2035 rtnl_unlock(); 2036 up_write(&pernet_ops_rwsem); 2037 return err; 2038} 2039EXPORT_SYMBOL(unregister_netdevice_notifier); 2040 2041static int __register_netdevice_notifier_net(struct net *net, 2042 struct notifier_block *nb, 2043 bool ignore_call_fail) 2044{ 2045 int err; 2046 2047 err = raw_notifier_chain_register(&net->netdev_chain, nb); 2048 if (err) 2049 return err; 2050 if (dev_boot_phase) 2051 return 0; 2052 2053 err = call_netdevice_register_net_notifiers(nb, net); 2054 if (err && !ignore_call_fail) 2055 goto chain_unregister; 2056 2057 return 0; 2058 2059chain_unregister: 2060 raw_notifier_chain_unregister(&net->netdev_chain, nb); 2061 return err; 2062} 2063 2064static int __unregister_netdevice_notifier_net(struct net *net, 2065 struct notifier_block *nb) 2066{ 2067 int err; 2068 2069 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 2070 if (err) 2071 return err; 2072 2073 call_netdevice_unregister_net_notifiers(nb, net); 2074 return 0; 2075} 2076 2077/** 2078 * register_netdevice_notifier_net - register a per-netns network notifier block 2079 * @net: network namespace 2080 * @nb: notifier 2081 * 2082 * Register a notifier to be called when network device events occur. 2083 * The notifier passed is linked into the kernel structures and must 2084 * not be reused until it has been unregistered. A negative errno code 2085 * is returned on a failure. 2086 * 2087 * When registered all registration and up events are replayed 2088 * to the new notifier to allow device to have a race free 2089 * view of the network device list. 2090 */ 2091 2092int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 2093{ 2094 int err; 2095 2096 rtnl_net_lock(net); 2097 err = __register_netdevice_notifier_net(net, nb, false); 2098 rtnl_net_unlock(net); 2099 2100 return err; 2101} 2102EXPORT_SYMBOL(register_netdevice_notifier_net); 2103 2104/** 2105 * unregister_netdevice_notifier_net - unregister a per-netns 2106 * network notifier block 2107 * @net: network namespace 2108 * @nb: notifier 2109 * 2110 * Unregister a notifier previously registered by 2111 * register_netdevice_notifier_net(). The notifier is unlinked from the 2112 * kernel structures and may then be reused. A negative errno code 2113 * is returned on a failure. 2114 * 2115 * After unregistering unregister and down device events are synthesized 2116 * for all devices on the device list to the removed notifier to remove 2117 * the need for special case cleanup code. 2118 */ 2119 2120int unregister_netdevice_notifier_net(struct net *net, 2121 struct notifier_block *nb) 2122{ 2123 int err; 2124 2125 rtnl_net_lock(net); 2126 err = __unregister_netdevice_notifier_net(net, nb); 2127 rtnl_net_unlock(net); 2128 2129 return err; 2130} 2131EXPORT_SYMBOL(unregister_netdevice_notifier_net); 2132 2133static void __move_netdevice_notifier_net(struct net *src_net, 2134 struct net *dst_net, 2135 struct notifier_block *nb) 2136{ 2137 __unregister_netdevice_notifier_net(src_net, nb); 2138 __register_netdevice_notifier_net(dst_net, nb, true); 2139} 2140 2141static void rtnl_net_dev_lock(struct net_device *dev) 2142{ 2143 bool again; 2144 2145 do { 2146 struct net *net; 2147 2148 again = false; 2149 2150 /* netns might be being dismantled. */ 2151 rcu_read_lock(); 2152 net = dev_net_rcu(dev); 2153 net_passive_inc(net); 2154 rcu_read_unlock(); 2155 2156 rtnl_net_lock(net); 2157 2158#ifdef CONFIG_NET_NS 2159 /* dev might have been moved to another netns. */ 2160 if (!net_eq(net, rcu_access_pointer(dev->nd_net.net))) { 2161 rtnl_net_unlock(net); 2162 net_passive_dec(net); 2163 again = true; 2164 } 2165#endif 2166 } while (again); 2167} 2168 2169static void rtnl_net_dev_unlock(struct net_device *dev) 2170{ 2171 struct net *net = dev_net(dev); 2172 2173 rtnl_net_unlock(net); 2174 net_passive_dec(net); 2175} 2176 2177int register_netdevice_notifier_dev_net(struct net_device *dev, 2178 struct notifier_block *nb, 2179 struct netdev_net_notifier *nn) 2180{ 2181 int err; 2182 2183 rtnl_net_dev_lock(dev); 2184 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 2185 if (!err) { 2186 nn->nb = nb; 2187 list_add(&nn->list, &dev->net_notifier_list); 2188 } 2189 rtnl_net_dev_unlock(dev); 2190 2191 return err; 2192} 2193EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 2194 2195int unregister_netdevice_notifier_dev_net(struct net_device *dev, 2196 struct notifier_block *nb, 2197 struct netdev_net_notifier *nn) 2198{ 2199 int err; 2200 2201 rtnl_net_dev_lock(dev); 2202 list_del(&nn->list); 2203 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 2204 rtnl_net_dev_unlock(dev); 2205 2206 return err; 2207} 2208EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2209 2210static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2211 struct net *net) 2212{ 2213 struct netdev_net_notifier *nn; 2214 2215 list_for_each_entry(nn, &dev->net_notifier_list, list) 2216 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 2217} 2218 2219/** 2220 * call_netdevice_notifiers_info - call all network notifier blocks 2221 * @val: value passed unmodified to notifier function 2222 * @info: notifier information data 2223 * 2224 * Call all network notifier blocks. Parameters and return value 2225 * are as for raw_notifier_call_chain(). 2226 */ 2227 2228int call_netdevice_notifiers_info(unsigned long val, 2229 struct netdev_notifier_info *info) 2230{ 2231 struct net *net = dev_net(info->dev); 2232 int ret; 2233 2234 ASSERT_RTNL(); 2235 2236 /* Run per-netns notifier block chain first, then run the global one. 2237 * Hopefully, one day, the global one is going to be removed after 2238 * all notifier block registrators get converted to be per-netns. 2239 */ 2240 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2241 if (ret & NOTIFY_STOP_MASK) 2242 return ret; 2243 return raw_notifier_call_chain(&netdev_chain, val, info); 2244} 2245 2246/** 2247 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2248 * for and rollback on error 2249 * @val_up: value passed unmodified to notifier function 2250 * @val_down: value passed unmodified to the notifier function when 2251 * recovering from an error on @val_up 2252 * @info: notifier information data 2253 * 2254 * Call all per-netns network notifier blocks, but not notifier blocks on 2255 * the global notifier chain. Parameters and return value are as for 2256 * raw_notifier_call_chain_robust(). 2257 */ 2258 2259static int 2260call_netdevice_notifiers_info_robust(unsigned long val_up, 2261 unsigned long val_down, 2262 struct netdev_notifier_info *info) 2263{ 2264 struct net *net = dev_net(info->dev); 2265 2266 ASSERT_RTNL(); 2267 2268 return raw_notifier_call_chain_robust(&net->netdev_chain, 2269 val_up, val_down, info); 2270} 2271 2272static int call_netdevice_notifiers_extack(unsigned long val, 2273 struct net_device *dev, 2274 struct netlink_ext_ack *extack) 2275{ 2276 struct netdev_notifier_info info = { 2277 .dev = dev, 2278 .extack = extack, 2279 }; 2280 2281 return call_netdevice_notifiers_info(val, &info); 2282} 2283 2284/** 2285 * call_netdevice_notifiers - call all network notifier blocks 2286 * @val: value passed unmodified to notifier function 2287 * @dev: net_device pointer passed unmodified to notifier function 2288 * 2289 * Call all network notifier blocks. Parameters and return value 2290 * are as for raw_notifier_call_chain(). 2291 */ 2292 2293int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2294{ 2295 return call_netdevice_notifiers_extack(val, dev, NULL); 2296} 2297EXPORT_SYMBOL(call_netdevice_notifiers); 2298 2299/** 2300 * call_netdevice_notifiers_mtu - call all network notifier blocks 2301 * @val: value passed unmodified to notifier function 2302 * @dev: net_device pointer passed unmodified to notifier function 2303 * @arg: additional u32 argument passed to the notifier function 2304 * 2305 * Call all network notifier blocks. Parameters and return value 2306 * are as for raw_notifier_call_chain(). 2307 */ 2308static int call_netdevice_notifiers_mtu(unsigned long val, 2309 struct net_device *dev, u32 arg) 2310{ 2311 struct netdev_notifier_info_ext info = { 2312 .info.dev = dev, 2313 .ext.mtu = arg, 2314 }; 2315 2316 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2317 2318 return call_netdevice_notifiers_info(val, &info.info); 2319} 2320 2321#ifdef CONFIG_NET_INGRESS 2322static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2323 2324void net_inc_ingress_queue(void) 2325{ 2326 static_branch_inc(&ingress_needed_key); 2327} 2328EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2329 2330void net_dec_ingress_queue(void) 2331{ 2332 static_branch_dec(&ingress_needed_key); 2333} 2334EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2335#endif 2336 2337#ifdef CONFIG_NET_EGRESS 2338static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2339 2340void net_inc_egress_queue(void) 2341{ 2342 static_branch_inc(&egress_needed_key); 2343} 2344EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2345 2346void net_dec_egress_queue(void) 2347{ 2348 static_branch_dec(&egress_needed_key); 2349} 2350EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2351#endif 2352 2353#ifdef CONFIG_NET_CLS_ACT 2354DEFINE_STATIC_KEY_FALSE(tcf_sw_enabled_key); 2355EXPORT_SYMBOL(tcf_sw_enabled_key); 2356#endif 2357 2358DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2359EXPORT_SYMBOL(netstamp_needed_key); 2360#ifdef CONFIG_JUMP_LABEL 2361static atomic_t netstamp_needed_deferred; 2362static atomic_t netstamp_wanted; 2363static void netstamp_clear(struct work_struct *work) 2364{ 2365 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2366 int wanted; 2367 2368 wanted = atomic_add_return(deferred, &netstamp_wanted); 2369 if (wanted > 0) 2370 static_branch_enable(&netstamp_needed_key); 2371 else 2372 static_branch_disable(&netstamp_needed_key); 2373} 2374static DECLARE_WORK(netstamp_work, netstamp_clear); 2375#endif 2376 2377void net_enable_timestamp(void) 2378{ 2379#ifdef CONFIG_JUMP_LABEL 2380 int wanted = atomic_read(&netstamp_wanted); 2381 2382 while (wanted > 0) { 2383 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2384 return; 2385 } 2386 atomic_inc(&netstamp_needed_deferred); 2387 schedule_work(&netstamp_work); 2388#else 2389 static_branch_inc(&netstamp_needed_key); 2390#endif 2391} 2392EXPORT_SYMBOL(net_enable_timestamp); 2393 2394void net_disable_timestamp(void) 2395{ 2396#ifdef CONFIG_JUMP_LABEL 2397 int wanted = atomic_read(&netstamp_wanted); 2398 2399 while (wanted > 1) { 2400 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2401 return; 2402 } 2403 atomic_dec(&netstamp_needed_deferred); 2404 schedule_work(&netstamp_work); 2405#else 2406 static_branch_dec(&netstamp_needed_key); 2407#endif 2408} 2409EXPORT_SYMBOL(net_disable_timestamp); 2410 2411static inline void net_timestamp_set(struct sk_buff *skb) 2412{ 2413 skb->tstamp = 0; 2414 skb->tstamp_type = SKB_CLOCK_REALTIME; 2415 if (static_branch_unlikely(&netstamp_needed_key)) 2416 skb->tstamp = ktime_get_real(); 2417} 2418 2419#define net_timestamp_check(COND, SKB) \ 2420 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2421 if ((COND) && !(SKB)->tstamp) \ 2422 (SKB)->tstamp = ktime_get_real(); \ 2423 } \ 2424 2425bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2426{ 2427 return __is_skb_forwardable(dev, skb, true); 2428} 2429EXPORT_SYMBOL_GPL(is_skb_forwardable); 2430 2431static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2432 bool check_mtu) 2433{ 2434 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2435 2436 if (likely(!ret)) { 2437 skb->protocol = eth_type_trans(skb, dev); 2438 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2439 } 2440 2441 return ret; 2442} 2443 2444int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2445{ 2446 return __dev_forward_skb2(dev, skb, true); 2447} 2448EXPORT_SYMBOL_GPL(__dev_forward_skb); 2449 2450/** 2451 * dev_forward_skb - loopback an skb to another netif 2452 * 2453 * @dev: destination network device 2454 * @skb: buffer to forward 2455 * 2456 * return values: 2457 * NET_RX_SUCCESS (no congestion) 2458 * NET_RX_DROP (packet was dropped, but freed) 2459 * 2460 * dev_forward_skb can be used for injecting an skb from the 2461 * start_xmit function of one device into the receive queue 2462 * of another device. 2463 * 2464 * The receiving device may be in another namespace, so 2465 * we have to clear all information in the skb that could 2466 * impact namespace isolation. 2467 */ 2468int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2469{ 2470 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2471} 2472EXPORT_SYMBOL_GPL(dev_forward_skb); 2473 2474int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2475{ 2476 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2477} 2478 2479static int deliver_skb(struct sk_buff *skb, 2480 struct packet_type *pt_prev, 2481 struct net_device *orig_dev) 2482{ 2483 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2484 return -ENOMEM; 2485 refcount_inc(&skb->users); 2486 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2487} 2488 2489static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2490 struct packet_type **pt, 2491 struct net_device *orig_dev, 2492 __be16 type, 2493 struct list_head *ptype_list) 2494{ 2495 struct packet_type *ptype, *pt_prev = *pt; 2496 2497 list_for_each_entry_rcu(ptype, ptype_list, list) { 2498 if (ptype->type != type) 2499 continue; 2500 if (unlikely(pt_prev)) 2501 deliver_skb(skb, pt_prev, orig_dev); 2502 pt_prev = ptype; 2503 } 2504 *pt = pt_prev; 2505} 2506 2507static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2508{ 2509 if (!ptype->af_packet_priv || !skb->sk) 2510 return false; 2511 2512 if (ptype->id_match) 2513 return ptype->id_match(ptype, skb->sk); 2514 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2515 return true; 2516 2517 return false; 2518} 2519 2520/** 2521 * dev_nit_active_rcu - return true if any network interface taps are in use 2522 * 2523 * The caller must hold the RCU lock 2524 * 2525 * @dev: network device to check for the presence of taps 2526 */ 2527bool dev_nit_active_rcu(const struct net_device *dev) 2528{ 2529 /* Callers may hold either RCU or RCU BH lock */ 2530 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 2531 2532 return !list_empty(&dev_net(dev)->ptype_all) || 2533 !list_empty(&dev->ptype_all); 2534} 2535EXPORT_SYMBOL_GPL(dev_nit_active_rcu); 2536 2537/* 2538 * Support routine. Sends outgoing frames to any network 2539 * taps currently in use. 2540 */ 2541 2542void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2543{ 2544 struct packet_type *ptype, *pt_prev = NULL; 2545 struct list_head *ptype_list; 2546 struct sk_buff *skb2 = NULL; 2547 2548 rcu_read_lock(); 2549 ptype_list = &dev_net_rcu(dev)->ptype_all; 2550again: 2551 list_for_each_entry_rcu(ptype, ptype_list, list) { 2552 if (READ_ONCE(ptype->ignore_outgoing)) 2553 continue; 2554 2555 /* Never send packets back to the socket 2556 * they originated from - MvS (miquels@drinkel.ow.org) 2557 */ 2558 if (skb_loop_sk(ptype, skb)) 2559 continue; 2560 2561 if (unlikely(pt_prev)) { 2562 deliver_skb(skb2, pt_prev, skb->dev); 2563 pt_prev = ptype; 2564 continue; 2565 } 2566 2567 /* need to clone skb, done only once */ 2568 skb2 = skb_clone(skb, GFP_ATOMIC); 2569 if (!skb2) 2570 goto out_unlock; 2571 2572 net_timestamp_set(skb2); 2573 2574 /* skb->nh should be correctly 2575 * set by sender, so that the second statement is 2576 * just protection against buggy protocols. 2577 */ 2578 skb_reset_mac_header(skb2); 2579 2580 if (skb_network_header(skb2) < skb2->data || 2581 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2582 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2583 ntohs(skb2->protocol), 2584 dev->name); 2585 skb_reset_network_header(skb2); 2586 } 2587 2588 skb2->transport_header = skb2->network_header; 2589 skb2->pkt_type = PACKET_OUTGOING; 2590 pt_prev = ptype; 2591 } 2592 2593 if (ptype_list != &dev->ptype_all) { 2594 ptype_list = &dev->ptype_all; 2595 goto again; 2596 } 2597out_unlock: 2598 if (pt_prev) { 2599 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2600 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2601 else 2602 kfree_skb(skb2); 2603 } 2604 rcu_read_unlock(); 2605} 2606EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2607 2608/** 2609 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2610 * @dev: Network device 2611 * @txq: number of queues available 2612 * 2613 * If real_num_tx_queues is changed the tc mappings may no longer be 2614 * valid. To resolve this verify the tc mapping remains valid and if 2615 * not NULL the mapping. With no priorities mapping to this 2616 * offset/count pair it will no longer be used. In the worst case TC0 2617 * is invalid nothing can be done so disable priority mappings. If is 2618 * expected that drivers will fix this mapping if they can before 2619 * calling netif_set_real_num_tx_queues. 2620 */ 2621static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2622{ 2623 int i; 2624 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2625 2626 /* If TC0 is invalidated disable TC mapping */ 2627 if (tc->offset + tc->count > txq) { 2628 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2629 dev->num_tc = 0; 2630 return; 2631 } 2632 2633 /* Invalidated prio to tc mappings set to TC0 */ 2634 for (i = 1; i < TC_BITMASK + 1; i++) { 2635 int q = netdev_get_prio_tc_map(dev, i); 2636 2637 tc = &dev->tc_to_txq[q]; 2638 if (tc->offset + tc->count > txq) { 2639 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", 2640 i, q); 2641 netdev_set_prio_tc_map(dev, i, 0); 2642 } 2643 } 2644} 2645 2646int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2647{ 2648 if (dev->num_tc) { 2649 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2650 int i; 2651 2652 /* walk through the TCs and see if it falls into any of them */ 2653 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2654 if ((txq - tc->offset) < tc->count) 2655 return i; 2656 } 2657 2658 /* didn't find it, just return -1 to indicate no match */ 2659 return -1; 2660 } 2661 2662 return 0; 2663} 2664EXPORT_SYMBOL(netdev_txq_to_tc); 2665 2666#ifdef CONFIG_XPS 2667static struct static_key xps_needed __read_mostly; 2668static struct static_key xps_rxqs_needed __read_mostly; 2669static DEFINE_MUTEX(xps_map_mutex); 2670#define xmap_dereference(P) \ 2671 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2672 2673static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2674 struct xps_dev_maps *old_maps, int tci, u16 index) 2675{ 2676 struct xps_map *map = NULL; 2677 int pos; 2678 2679 map = xmap_dereference(dev_maps->attr_map[tci]); 2680 if (!map) 2681 return false; 2682 2683 for (pos = map->len; pos--;) { 2684 if (map->queues[pos] != index) 2685 continue; 2686 2687 if (map->len > 1) { 2688 map->queues[pos] = map->queues[--map->len]; 2689 break; 2690 } 2691 2692 if (old_maps) 2693 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2694 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2695 kfree_rcu(map, rcu); 2696 return false; 2697 } 2698 2699 return true; 2700} 2701 2702static bool remove_xps_queue_cpu(struct net_device *dev, 2703 struct xps_dev_maps *dev_maps, 2704 int cpu, u16 offset, u16 count) 2705{ 2706 int num_tc = dev_maps->num_tc; 2707 bool active = false; 2708 int tci; 2709 2710 for (tci = cpu * num_tc; num_tc--; tci++) { 2711 int i, j; 2712 2713 for (i = count, j = offset; i--; j++) { 2714 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2715 break; 2716 } 2717 2718 active |= i < 0; 2719 } 2720 2721 return active; 2722} 2723 2724static void reset_xps_maps(struct net_device *dev, 2725 struct xps_dev_maps *dev_maps, 2726 enum xps_map_type type) 2727{ 2728 static_key_slow_dec_cpuslocked(&xps_needed); 2729 if (type == XPS_RXQS) 2730 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2731 2732 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2733 2734 kfree_rcu(dev_maps, rcu); 2735} 2736 2737static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2738 u16 offset, u16 count) 2739{ 2740 struct xps_dev_maps *dev_maps; 2741 bool active = false; 2742 int i, j; 2743 2744 dev_maps = xmap_dereference(dev->xps_maps[type]); 2745 if (!dev_maps) 2746 return; 2747 2748 for (j = 0; j < dev_maps->nr_ids; j++) 2749 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2750 if (!active) 2751 reset_xps_maps(dev, dev_maps, type); 2752 2753 if (type == XPS_CPUS) { 2754 for (i = offset + (count - 1); count--; i--) 2755 netdev_queue_numa_node_write( 2756 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2757 } 2758} 2759 2760static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2761 u16 count) 2762{ 2763 if (!static_key_false(&xps_needed)) 2764 return; 2765 2766 cpus_read_lock(); 2767 mutex_lock(&xps_map_mutex); 2768 2769 if (static_key_false(&xps_rxqs_needed)) 2770 clean_xps_maps(dev, XPS_RXQS, offset, count); 2771 2772 clean_xps_maps(dev, XPS_CPUS, offset, count); 2773 2774 mutex_unlock(&xps_map_mutex); 2775 cpus_read_unlock(); 2776} 2777 2778static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2779{ 2780 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2781} 2782 2783static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2784 u16 index, bool is_rxqs_map) 2785{ 2786 struct xps_map *new_map; 2787 int alloc_len = XPS_MIN_MAP_ALLOC; 2788 int i, pos; 2789 2790 for (pos = 0; map && pos < map->len; pos++) { 2791 if (map->queues[pos] != index) 2792 continue; 2793 return map; 2794 } 2795 2796 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2797 if (map) { 2798 if (pos < map->alloc_len) 2799 return map; 2800 2801 alloc_len = map->alloc_len * 2; 2802 } 2803 2804 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2805 * map 2806 */ 2807 if (is_rxqs_map) 2808 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2809 else 2810 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2811 cpu_to_node(attr_index)); 2812 if (!new_map) 2813 return NULL; 2814 2815 for (i = 0; i < pos; i++) 2816 new_map->queues[i] = map->queues[i]; 2817 new_map->alloc_len = alloc_len; 2818 new_map->len = pos; 2819 2820 return new_map; 2821} 2822 2823/* Copy xps maps at a given index */ 2824static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2825 struct xps_dev_maps *new_dev_maps, int index, 2826 int tc, bool skip_tc) 2827{ 2828 int i, tci = index * dev_maps->num_tc; 2829 struct xps_map *map; 2830 2831 /* copy maps belonging to foreign traffic classes */ 2832 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2833 if (i == tc && skip_tc) 2834 continue; 2835 2836 /* fill in the new device map from the old device map */ 2837 map = xmap_dereference(dev_maps->attr_map[tci]); 2838 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2839 } 2840} 2841 2842/* Must be called under cpus_read_lock */ 2843int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2844 u16 index, enum xps_map_type type) 2845{ 2846 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2847 const unsigned long *online_mask = NULL; 2848 bool active = false, copy = false; 2849 int i, j, tci, numa_node_id = -2; 2850 int maps_sz, num_tc = 1, tc = 0; 2851 struct xps_map *map, *new_map; 2852 unsigned int nr_ids; 2853 2854 WARN_ON_ONCE(index >= dev->num_tx_queues); 2855 2856 if (dev->num_tc) { 2857 /* Do not allow XPS on subordinate device directly */ 2858 num_tc = dev->num_tc; 2859 if (num_tc < 0) 2860 return -EINVAL; 2861 2862 /* If queue belongs to subordinate dev use its map */ 2863 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2864 2865 tc = netdev_txq_to_tc(dev, index); 2866 if (tc < 0) 2867 return -EINVAL; 2868 } 2869 2870 mutex_lock(&xps_map_mutex); 2871 2872 dev_maps = xmap_dereference(dev->xps_maps[type]); 2873 if (type == XPS_RXQS) { 2874 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2875 nr_ids = dev->num_rx_queues; 2876 } else { 2877 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2878 if (num_possible_cpus() > 1) 2879 online_mask = cpumask_bits(cpu_online_mask); 2880 nr_ids = nr_cpu_ids; 2881 } 2882 2883 if (maps_sz < L1_CACHE_BYTES) 2884 maps_sz = L1_CACHE_BYTES; 2885 2886 /* The old dev_maps could be larger or smaller than the one we're 2887 * setting up now, as dev->num_tc or nr_ids could have been updated in 2888 * between. We could try to be smart, but let's be safe instead and only 2889 * copy foreign traffic classes if the two map sizes match. 2890 */ 2891 if (dev_maps && 2892 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2893 copy = true; 2894 2895 /* allocate memory for queue storage */ 2896 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2897 j < nr_ids;) { 2898 if (!new_dev_maps) { 2899 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2900 if (!new_dev_maps) { 2901 mutex_unlock(&xps_map_mutex); 2902 return -ENOMEM; 2903 } 2904 2905 new_dev_maps->nr_ids = nr_ids; 2906 new_dev_maps->num_tc = num_tc; 2907 } 2908 2909 tci = j * num_tc + tc; 2910 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2911 2912 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2913 if (!map) 2914 goto error; 2915 2916 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2917 } 2918 2919 if (!new_dev_maps) 2920 goto out_no_new_maps; 2921 2922 if (!dev_maps) { 2923 /* Increment static keys at most once per type */ 2924 static_key_slow_inc_cpuslocked(&xps_needed); 2925 if (type == XPS_RXQS) 2926 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2927 } 2928 2929 for (j = 0; j < nr_ids; j++) { 2930 bool skip_tc = false; 2931 2932 tci = j * num_tc + tc; 2933 if (netif_attr_test_mask(j, mask, nr_ids) && 2934 netif_attr_test_online(j, online_mask, nr_ids)) { 2935 /* add tx-queue to CPU/rx-queue maps */ 2936 int pos = 0; 2937 2938 skip_tc = true; 2939 2940 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2941 while ((pos < map->len) && (map->queues[pos] != index)) 2942 pos++; 2943 2944 if (pos == map->len) 2945 map->queues[map->len++] = index; 2946#ifdef CONFIG_NUMA 2947 if (type == XPS_CPUS) { 2948 if (numa_node_id == -2) 2949 numa_node_id = cpu_to_node(j); 2950 else if (numa_node_id != cpu_to_node(j)) 2951 numa_node_id = -1; 2952 } 2953#endif 2954 } 2955 2956 if (copy) 2957 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2958 skip_tc); 2959 } 2960 2961 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2962 2963 /* Cleanup old maps */ 2964 if (!dev_maps) 2965 goto out_no_old_maps; 2966 2967 for (j = 0; j < dev_maps->nr_ids; j++) { 2968 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2969 map = xmap_dereference(dev_maps->attr_map[tci]); 2970 if (!map) 2971 continue; 2972 2973 if (copy) { 2974 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2975 if (map == new_map) 2976 continue; 2977 } 2978 2979 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2980 kfree_rcu(map, rcu); 2981 } 2982 } 2983 2984 old_dev_maps = dev_maps; 2985 2986out_no_old_maps: 2987 dev_maps = new_dev_maps; 2988 active = true; 2989 2990out_no_new_maps: 2991 if (type == XPS_CPUS) 2992 /* update Tx queue numa node */ 2993 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2994 (numa_node_id >= 0) ? 2995 numa_node_id : NUMA_NO_NODE); 2996 2997 if (!dev_maps) 2998 goto out_no_maps; 2999 3000 /* removes tx-queue from unused CPUs/rx-queues */ 3001 for (j = 0; j < dev_maps->nr_ids; j++) { 3002 tci = j * dev_maps->num_tc; 3003 3004 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 3005 if (i == tc && 3006 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 3007 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 3008 continue; 3009 3010 active |= remove_xps_queue(dev_maps, 3011 copy ? old_dev_maps : NULL, 3012 tci, index); 3013 } 3014 } 3015 3016 if (old_dev_maps) 3017 kfree_rcu(old_dev_maps, rcu); 3018 3019 /* free map if not active */ 3020 if (!active) 3021 reset_xps_maps(dev, dev_maps, type); 3022 3023out_no_maps: 3024 mutex_unlock(&xps_map_mutex); 3025 3026 return 0; 3027error: 3028 /* remove any maps that we added */ 3029 for (j = 0; j < nr_ids; j++) { 3030 for (i = num_tc, tci = j * num_tc; i--; tci++) { 3031 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 3032 map = copy ? 3033 xmap_dereference(dev_maps->attr_map[tci]) : 3034 NULL; 3035 if (new_map && new_map != map) 3036 kfree(new_map); 3037 } 3038 } 3039 3040 mutex_unlock(&xps_map_mutex); 3041 3042 kfree(new_dev_maps); 3043 return -ENOMEM; 3044} 3045EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 3046 3047int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 3048 u16 index) 3049{ 3050 int ret; 3051 3052 cpus_read_lock(); 3053 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 3054 cpus_read_unlock(); 3055 3056 return ret; 3057} 3058EXPORT_SYMBOL(netif_set_xps_queue); 3059 3060#endif 3061static void netdev_unbind_all_sb_channels(struct net_device *dev) 3062{ 3063 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 3064 3065 /* Unbind any subordinate channels */ 3066 while (txq-- != &dev->_tx[0]) { 3067 if (txq->sb_dev) 3068 netdev_unbind_sb_channel(dev, txq->sb_dev); 3069 } 3070} 3071 3072void netdev_reset_tc(struct net_device *dev) 3073{ 3074#ifdef CONFIG_XPS 3075 netif_reset_xps_queues_gt(dev, 0); 3076#endif 3077 netdev_unbind_all_sb_channels(dev); 3078 3079 /* Reset TC configuration of device */ 3080 dev->num_tc = 0; 3081 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 3082 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 3083} 3084EXPORT_SYMBOL(netdev_reset_tc); 3085 3086int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 3087{ 3088 if (tc >= dev->num_tc) 3089 return -EINVAL; 3090 3091#ifdef CONFIG_XPS 3092 netif_reset_xps_queues(dev, offset, count); 3093#endif 3094 dev->tc_to_txq[tc].count = count; 3095 dev->tc_to_txq[tc].offset = offset; 3096 return 0; 3097} 3098EXPORT_SYMBOL(netdev_set_tc_queue); 3099 3100int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 3101{ 3102 if (num_tc > TC_MAX_QUEUE) 3103 return -EINVAL; 3104 3105#ifdef CONFIG_XPS 3106 netif_reset_xps_queues_gt(dev, 0); 3107#endif 3108 netdev_unbind_all_sb_channels(dev); 3109 3110 dev->num_tc = num_tc; 3111 return 0; 3112} 3113EXPORT_SYMBOL(netdev_set_num_tc); 3114 3115void netdev_unbind_sb_channel(struct net_device *dev, 3116 struct net_device *sb_dev) 3117{ 3118 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 3119 3120#ifdef CONFIG_XPS 3121 netif_reset_xps_queues_gt(sb_dev, 0); 3122#endif 3123 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 3124 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 3125 3126 while (txq-- != &dev->_tx[0]) { 3127 if (txq->sb_dev == sb_dev) 3128 txq->sb_dev = NULL; 3129 } 3130} 3131EXPORT_SYMBOL(netdev_unbind_sb_channel); 3132 3133int netdev_bind_sb_channel_queue(struct net_device *dev, 3134 struct net_device *sb_dev, 3135 u8 tc, u16 count, u16 offset) 3136{ 3137 /* Make certain the sb_dev and dev are already configured */ 3138 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 3139 return -EINVAL; 3140 3141 /* We cannot hand out queues we don't have */ 3142 if ((offset + count) > dev->real_num_tx_queues) 3143 return -EINVAL; 3144 3145 /* Record the mapping */ 3146 sb_dev->tc_to_txq[tc].count = count; 3147 sb_dev->tc_to_txq[tc].offset = offset; 3148 3149 /* Provide a way for Tx queue to find the tc_to_txq map or 3150 * XPS map for itself. 3151 */ 3152 while (count--) 3153 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 3154 3155 return 0; 3156} 3157EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 3158 3159int netdev_set_sb_channel(struct net_device *dev, u16 channel) 3160{ 3161 /* Do not use a multiqueue device to represent a subordinate channel */ 3162 if (netif_is_multiqueue(dev)) 3163 return -ENODEV; 3164 3165 /* We allow channels 1 - 32767 to be used for subordinate channels. 3166 * Channel 0 is meant to be "native" mode and used only to represent 3167 * the main root device. We allow writing 0 to reset the device back 3168 * to normal mode after being used as a subordinate channel. 3169 */ 3170 if (channel > S16_MAX) 3171 return -EINVAL; 3172 3173 dev->num_tc = -channel; 3174 3175 return 0; 3176} 3177EXPORT_SYMBOL(netdev_set_sb_channel); 3178 3179/* 3180 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 3181 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 3182 */ 3183int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 3184{ 3185 bool disabling; 3186 int rc; 3187 3188 disabling = txq < dev->real_num_tx_queues; 3189 3190 if (txq < 1 || txq > dev->num_tx_queues) 3191 return -EINVAL; 3192 3193 if (dev->reg_state == NETREG_REGISTERED || 3194 dev->reg_state == NETREG_UNREGISTERING) { 3195 netdev_ops_assert_locked(dev); 3196 3197 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 3198 txq); 3199 if (rc) 3200 return rc; 3201 3202 if (dev->num_tc) 3203 netif_setup_tc(dev, txq); 3204 3205 net_shaper_set_real_num_tx_queues(dev, txq); 3206 3207 dev_qdisc_change_real_num_tx(dev, txq); 3208 3209 dev->real_num_tx_queues = txq; 3210 3211 if (disabling) { 3212 synchronize_net(); 3213 qdisc_reset_all_tx_gt(dev, txq); 3214#ifdef CONFIG_XPS 3215 netif_reset_xps_queues_gt(dev, txq); 3216#endif 3217 } 3218 } else { 3219 dev->real_num_tx_queues = txq; 3220 } 3221 3222 return 0; 3223} 3224EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3225 3226/** 3227 * netif_set_real_num_rx_queues - set actual number of RX queues used 3228 * @dev: Network device 3229 * @rxq: Actual number of RX queues 3230 * 3231 * This must be called either with the rtnl_lock held or before 3232 * registration of the net device. Returns 0 on success, or a 3233 * negative error code. If called before registration, it always 3234 * succeeds. 3235 */ 3236int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3237{ 3238 int rc; 3239 3240 if (rxq < 1 || rxq > dev->num_rx_queues) 3241 return -EINVAL; 3242 3243 if (dev->reg_state == NETREG_REGISTERED) { 3244 netdev_ops_assert_locked(dev); 3245 3246 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3247 rxq); 3248 if (rc) 3249 return rc; 3250 } 3251 3252 dev->real_num_rx_queues = rxq; 3253 return 0; 3254} 3255EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3256 3257/** 3258 * netif_set_real_num_queues - set actual number of RX and TX queues used 3259 * @dev: Network device 3260 * @txq: Actual number of TX queues 3261 * @rxq: Actual number of RX queues 3262 * 3263 * Set the real number of both TX and RX queues. 3264 * Does nothing if the number of queues is already correct. 3265 */ 3266int netif_set_real_num_queues(struct net_device *dev, 3267 unsigned int txq, unsigned int rxq) 3268{ 3269 unsigned int old_rxq = dev->real_num_rx_queues; 3270 int err; 3271 3272 if (txq < 1 || txq > dev->num_tx_queues || 3273 rxq < 1 || rxq > dev->num_rx_queues) 3274 return -EINVAL; 3275 3276 /* Start from increases, so the error path only does decreases - 3277 * decreases can't fail. 3278 */ 3279 if (rxq > dev->real_num_rx_queues) { 3280 err = netif_set_real_num_rx_queues(dev, rxq); 3281 if (err) 3282 return err; 3283 } 3284 if (txq > dev->real_num_tx_queues) { 3285 err = netif_set_real_num_tx_queues(dev, txq); 3286 if (err) 3287 goto undo_rx; 3288 } 3289 if (rxq < dev->real_num_rx_queues) 3290 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3291 if (txq < dev->real_num_tx_queues) 3292 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3293 3294 return 0; 3295undo_rx: 3296 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3297 return err; 3298} 3299EXPORT_SYMBOL(netif_set_real_num_queues); 3300 3301/** 3302 * netif_set_tso_max_size() - set the max size of TSO frames supported 3303 * @dev: netdev to update 3304 * @size: max skb->len of a TSO frame 3305 * 3306 * Set the limit on the size of TSO super-frames the device can handle. 3307 * Unless explicitly set the stack will assume the value of 3308 * %GSO_LEGACY_MAX_SIZE. 3309 */ 3310void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3311{ 3312 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3313 if (size < READ_ONCE(dev->gso_max_size)) 3314 netif_set_gso_max_size(dev, size); 3315 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3316 netif_set_gso_ipv4_max_size(dev, size); 3317} 3318EXPORT_SYMBOL(netif_set_tso_max_size); 3319 3320/** 3321 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3322 * @dev: netdev to update 3323 * @segs: max number of TCP segments 3324 * 3325 * Set the limit on the number of TCP segments the device can generate from 3326 * a single TSO super-frame. 3327 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3328 */ 3329void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3330{ 3331 dev->tso_max_segs = segs; 3332 if (segs < READ_ONCE(dev->gso_max_segs)) 3333 netif_set_gso_max_segs(dev, segs); 3334} 3335EXPORT_SYMBOL(netif_set_tso_max_segs); 3336 3337/** 3338 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3339 * @to: netdev to update 3340 * @from: netdev from which to copy the limits 3341 */ 3342void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3343{ 3344 netif_set_tso_max_size(to, from->tso_max_size); 3345 netif_set_tso_max_segs(to, from->tso_max_segs); 3346} 3347EXPORT_SYMBOL(netif_inherit_tso_max); 3348 3349/** 3350 * netif_get_num_default_rss_queues - default number of RSS queues 3351 * 3352 * Default value is the number of physical cores if there are only 1 or 2, or 3353 * divided by 2 if there are more. 3354 */ 3355int netif_get_num_default_rss_queues(void) 3356{ 3357 cpumask_var_t cpus; 3358 int cpu, count = 0; 3359 3360 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3361 return 1; 3362 3363 cpumask_copy(cpus, cpu_online_mask); 3364 for_each_cpu(cpu, cpus) { 3365 ++count; 3366 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3367 } 3368 free_cpumask_var(cpus); 3369 3370 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3371} 3372EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3373 3374static void __netif_reschedule(struct Qdisc *q) 3375{ 3376 struct softnet_data *sd; 3377 unsigned long flags; 3378 3379 local_irq_save(flags); 3380 sd = this_cpu_ptr(&softnet_data); 3381 q->next_sched = NULL; 3382 *sd->output_queue_tailp = q; 3383 sd->output_queue_tailp = &q->next_sched; 3384 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3385 local_irq_restore(flags); 3386} 3387 3388void __netif_schedule(struct Qdisc *q) 3389{ 3390 /* If q->defer_list is not empty, at least one thread is 3391 * in __dev_xmit_skb() before llist_del_all(&q->defer_list). 3392 * This thread will attempt to run the queue. 3393 */ 3394 if (!llist_empty(&q->defer_list)) 3395 return; 3396 3397 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3398 __netif_reschedule(q); 3399} 3400EXPORT_SYMBOL(__netif_schedule); 3401 3402struct dev_kfree_skb_cb { 3403 enum skb_drop_reason reason; 3404}; 3405 3406static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3407{ 3408 return (struct dev_kfree_skb_cb *)skb->cb; 3409} 3410 3411void netif_schedule_queue(struct netdev_queue *txq) 3412{ 3413 rcu_read_lock(); 3414 if (!netif_xmit_stopped(txq)) { 3415 struct Qdisc *q = rcu_dereference(txq->qdisc); 3416 3417 __netif_schedule(q); 3418 } 3419 rcu_read_unlock(); 3420} 3421EXPORT_SYMBOL(netif_schedule_queue); 3422 3423void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3424{ 3425 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3426 struct Qdisc *q; 3427 3428 rcu_read_lock(); 3429 q = rcu_dereference(dev_queue->qdisc); 3430 __netif_schedule(q); 3431 rcu_read_unlock(); 3432 } 3433} 3434EXPORT_SYMBOL(netif_tx_wake_queue); 3435 3436void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3437{ 3438 unsigned long flags; 3439 3440 if (unlikely(!skb)) 3441 return; 3442 3443 if (likely(refcount_read(&skb->users) == 1)) { 3444 smp_rmb(); 3445 refcount_set(&skb->users, 0); 3446 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3447 return; 3448 } 3449 get_kfree_skb_cb(skb)->reason = reason; 3450 local_irq_save(flags); 3451 skb->next = __this_cpu_read(softnet_data.completion_queue); 3452 __this_cpu_write(softnet_data.completion_queue, skb); 3453 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3454 local_irq_restore(flags); 3455} 3456EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3457 3458void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3459{ 3460 if (in_hardirq() || irqs_disabled()) 3461 dev_kfree_skb_irq_reason(skb, reason); 3462 else 3463 kfree_skb_reason(skb, reason); 3464} 3465EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3466 3467 3468/** 3469 * netif_device_detach - mark device as removed 3470 * @dev: network device 3471 * 3472 * Mark device as removed from system and therefore no longer available. 3473 */ 3474void netif_device_detach(struct net_device *dev) 3475{ 3476 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3477 netif_running(dev)) { 3478 netif_tx_stop_all_queues(dev); 3479 } 3480} 3481EXPORT_SYMBOL(netif_device_detach); 3482 3483/** 3484 * netif_device_attach - mark device as attached 3485 * @dev: network device 3486 * 3487 * Mark device as attached from system and restart if needed. 3488 */ 3489void netif_device_attach(struct net_device *dev) 3490{ 3491 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3492 netif_running(dev)) { 3493 netif_tx_wake_all_queues(dev); 3494 netdev_watchdog_up(dev); 3495 } 3496} 3497EXPORT_SYMBOL(netif_device_attach); 3498 3499/* 3500 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3501 * to be used as a distribution range. 3502 */ 3503static u16 skb_tx_hash(const struct net_device *dev, 3504 const struct net_device *sb_dev, 3505 struct sk_buff *skb) 3506{ 3507 u32 hash; 3508 u16 qoffset = 0; 3509 u16 qcount = dev->real_num_tx_queues; 3510 3511 if (dev->num_tc) { 3512 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3513 3514 qoffset = sb_dev->tc_to_txq[tc].offset; 3515 qcount = sb_dev->tc_to_txq[tc].count; 3516 if (unlikely(!qcount)) { 3517 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3518 sb_dev->name, qoffset, tc); 3519 qoffset = 0; 3520 qcount = dev->real_num_tx_queues; 3521 } 3522 } 3523 3524 if (skb_rx_queue_recorded(skb)) { 3525 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3526 hash = skb_get_rx_queue(skb); 3527 if (hash >= qoffset) 3528 hash -= qoffset; 3529 while (unlikely(hash >= qcount)) 3530 hash -= qcount; 3531 return hash + qoffset; 3532 } 3533 3534 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3535} 3536 3537void skb_warn_bad_offload(const struct sk_buff *skb) 3538{ 3539 static const netdev_features_t null_features; 3540 struct net_device *dev = skb->dev; 3541 const char *name = ""; 3542 3543 if (!net_ratelimit()) 3544 return; 3545 3546 if (dev) { 3547 if (dev->dev.parent) 3548 name = dev_driver_string(dev->dev.parent); 3549 else 3550 name = netdev_name(dev); 3551 } 3552 skb_dump(KERN_WARNING, skb, false); 3553 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3554 name, dev ? &dev->features : &null_features, 3555 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3556} 3557 3558/* 3559 * Invalidate hardware checksum when packet is to be mangled, and 3560 * complete checksum manually on outgoing path. 3561 */ 3562int skb_checksum_help(struct sk_buff *skb) 3563{ 3564 __wsum csum; 3565 int ret = 0, offset; 3566 3567 if (skb->ip_summed == CHECKSUM_COMPLETE) 3568 goto out_set_summed; 3569 3570 if (unlikely(skb_is_gso(skb))) { 3571 skb_warn_bad_offload(skb); 3572 return -EINVAL; 3573 } 3574 3575 if (!skb_frags_readable(skb)) { 3576 return -EFAULT; 3577 } 3578 3579 /* Before computing a checksum, we should make sure no frag could 3580 * be modified by an external entity : checksum could be wrong. 3581 */ 3582 if (skb_has_shared_frag(skb)) { 3583 ret = __skb_linearize(skb); 3584 if (ret) 3585 goto out; 3586 } 3587 3588 offset = skb_checksum_start_offset(skb); 3589 ret = -EINVAL; 3590 if (unlikely(offset >= skb_headlen(skb))) { 3591 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3592 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3593 offset, skb_headlen(skb)); 3594 goto out; 3595 } 3596 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3597 3598 offset += skb->csum_offset; 3599 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3600 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3601 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3602 offset + sizeof(__sum16), skb_headlen(skb)); 3603 goto out; 3604 } 3605 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3606 if (ret) 3607 goto out; 3608 3609 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3610out_set_summed: 3611 skb->ip_summed = CHECKSUM_NONE; 3612out: 3613 return ret; 3614} 3615EXPORT_SYMBOL(skb_checksum_help); 3616 3617#ifdef CONFIG_NET_CRC32C 3618int skb_crc32c_csum_help(struct sk_buff *skb) 3619{ 3620 u32 crc; 3621 int ret = 0, offset, start; 3622 3623 if (skb->ip_summed != CHECKSUM_PARTIAL) 3624 goto out; 3625 3626 if (unlikely(skb_is_gso(skb))) 3627 goto out; 3628 3629 /* Before computing a checksum, we should make sure no frag could 3630 * be modified by an external entity : checksum could be wrong. 3631 */ 3632 if (unlikely(skb_has_shared_frag(skb))) { 3633 ret = __skb_linearize(skb); 3634 if (ret) 3635 goto out; 3636 } 3637 start = skb_checksum_start_offset(skb); 3638 offset = start + offsetof(struct sctphdr, checksum); 3639 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3640 ret = -EINVAL; 3641 goto out; 3642 } 3643 3644 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3645 if (ret) 3646 goto out; 3647 3648 crc = ~skb_crc32c(skb, start, skb->len - start, ~0); 3649 *(__le32 *)(skb->data + offset) = cpu_to_le32(crc); 3650 skb_reset_csum_not_inet(skb); 3651out: 3652 return ret; 3653} 3654EXPORT_SYMBOL(skb_crc32c_csum_help); 3655#endif /* CONFIG_NET_CRC32C */ 3656 3657__be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3658{ 3659 __be16 type = skb->protocol; 3660 3661 /* Tunnel gso handlers can set protocol to ethernet. */ 3662 if (type == htons(ETH_P_TEB)) { 3663 struct ethhdr *eth; 3664 3665 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3666 return 0; 3667 3668 eth = (struct ethhdr *)skb->data; 3669 type = eth->h_proto; 3670 } 3671 3672 return vlan_get_protocol_and_depth(skb, type, depth); 3673} 3674 3675 3676/* Take action when hardware reception checksum errors are detected. */ 3677#ifdef CONFIG_BUG 3678static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3679{ 3680 netdev_err(dev, "hw csum failure\n"); 3681 skb_dump(KERN_ERR, skb, true); 3682 dump_stack(); 3683} 3684 3685void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3686{ 3687 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3688} 3689EXPORT_SYMBOL(netdev_rx_csum_fault); 3690#endif 3691 3692/* XXX: check that highmem exists at all on the given machine. */ 3693static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3694{ 3695#ifdef CONFIG_HIGHMEM 3696 int i; 3697 3698 if (!(dev->features & NETIF_F_HIGHDMA)) { 3699 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3700 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3701 struct page *page = skb_frag_page(frag); 3702 3703 if (page && PageHighMem(page)) 3704 return 1; 3705 } 3706 } 3707#endif 3708 return 0; 3709} 3710 3711/* If MPLS offload request, verify we are testing hardware MPLS features 3712 * instead of standard features for the netdev. 3713 */ 3714#if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3715static netdev_features_t net_mpls_features(struct sk_buff *skb, 3716 netdev_features_t features, 3717 __be16 type) 3718{ 3719 if (eth_p_mpls(type)) 3720 features &= skb->dev->mpls_features; 3721 3722 return features; 3723} 3724#else 3725static netdev_features_t net_mpls_features(struct sk_buff *skb, 3726 netdev_features_t features, 3727 __be16 type) 3728{ 3729 return features; 3730} 3731#endif 3732 3733static netdev_features_t harmonize_features(struct sk_buff *skb, 3734 netdev_features_t features) 3735{ 3736 __be16 type; 3737 3738 type = skb_network_protocol(skb, NULL); 3739 features = net_mpls_features(skb, features, type); 3740 3741 if (skb->ip_summed != CHECKSUM_NONE && 3742 !can_checksum_protocol(features, type)) { 3743 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3744 } 3745 if (illegal_highdma(skb->dev, skb)) 3746 features &= ~NETIF_F_SG; 3747 3748 return features; 3749} 3750 3751netdev_features_t passthru_features_check(struct sk_buff *skb, 3752 struct net_device *dev, 3753 netdev_features_t features) 3754{ 3755 return features; 3756} 3757EXPORT_SYMBOL(passthru_features_check); 3758 3759static netdev_features_t dflt_features_check(struct sk_buff *skb, 3760 struct net_device *dev, 3761 netdev_features_t features) 3762{ 3763 return vlan_features_check(skb, features); 3764} 3765 3766static bool skb_gso_has_extension_hdr(const struct sk_buff *skb) 3767{ 3768 if (!skb->encapsulation) 3769 return ((skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6 || 3770 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4 && 3771 vlan_get_protocol(skb) == htons(ETH_P_IPV6))) && 3772 skb_transport_header_was_set(skb) && 3773 skb_network_header_len(skb) != sizeof(struct ipv6hdr)); 3774 else 3775 return (!skb_inner_network_header_was_set(skb) || 3776 ((skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6 || 3777 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4 && 3778 inner_ip_hdr(skb)->version == 6)) && 3779 skb_inner_network_header_len(skb) != sizeof(struct ipv6hdr))); 3780} 3781 3782static netdev_features_t gso_features_check(const struct sk_buff *skb, 3783 struct net_device *dev, 3784 netdev_features_t features) 3785{ 3786 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3787 3788 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3789 return features & ~NETIF_F_GSO_MASK; 3790 3791 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb))) 3792 return features & ~NETIF_F_GSO_MASK; 3793 3794 if (!skb_shinfo(skb)->gso_type) { 3795 skb_warn_bad_offload(skb); 3796 return features & ~NETIF_F_GSO_MASK; 3797 } 3798 3799 /* Support for GSO partial features requires software 3800 * intervention before we can actually process the packets 3801 * so we need to strip support for any partial features now 3802 * and we can pull them back in after we have partially 3803 * segmented the frame. 3804 */ 3805 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3806 features &= ~dev->gso_partial_features; 3807 3808 /* Make sure to clear the IPv4 ID mangling feature if the IPv4 header 3809 * has the potential to be fragmented so that TSO does not generate 3810 * segments with the same ID. For encapsulated packets, the ID mangling 3811 * feature is guaranteed not to use the same ID for the outer IPv4 3812 * headers of the generated segments if the headers have the potential 3813 * to be fragmented, so there is no need to clear the IPv4 ID mangling 3814 * feature (see the section about NETIF_F_TSO_MANGLEID in 3815 * segmentation-offloads.rst). 3816 */ 3817 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3818 const struct iphdr *iph; 3819 struct iphdr _iph; 3820 int nhoff = skb->encapsulation ? 3821 skb_inner_network_offset(skb) : 3822 skb_network_offset(skb); 3823 3824 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph); 3825 3826 if (!iph || !(iph->frag_off & htons(IP_DF))) 3827 features &= ~dev->mangleid_features; 3828 } 3829 3830 /* NETIF_F_IPV6_CSUM does not support IPv6 extension headers, 3831 * so neither does TSO that depends on it. 3832 */ 3833 if (features & NETIF_F_IPV6_CSUM && 3834 skb_gso_has_extension_hdr(skb) && 3835 !ipv6_has_hopopt_jumbo(skb)) 3836 features &= ~(NETIF_F_IPV6_CSUM | NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4); 3837 3838 return features; 3839} 3840 3841netdev_features_t netif_skb_features(struct sk_buff *skb) 3842{ 3843 struct net_device *dev = skb->dev; 3844 netdev_features_t features = dev->features; 3845 3846 if (skb_is_gso(skb)) 3847 features = gso_features_check(skb, dev, features); 3848 3849 /* If encapsulation offload request, verify we are testing 3850 * hardware encapsulation features instead of standard 3851 * features for the netdev 3852 */ 3853 if (skb->encapsulation) 3854 features &= dev->hw_enc_features; 3855 3856 if (skb_vlan_tagged(skb)) 3857 features = netdev_intersect_features(features, 3858 dev->vlan_features | 3859 NETIF_F_HW_VLAN_CTAG_TX | 3860 NETIF_F_HW_VLAN_STAG_TX); 3861 3862 if (dev->netdev_ops->ndo_features_check) 3863 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3864 features); 3865 else 3866 features &= dflt_features_check(skb, dev, features); 3867 3868 return harmonize_features(skb, features); 3869} 3870EXPORT_SYMBOL(netif_skb_features); 3871 3872static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3873 struct netdev_queue *txq, bool more) 3874{ 3875 unsigned int len; 3876 int rc; 3877 3878 if (dev_nit_active_rcu(dev)) 3879 dev_queue_xmit_nit(skb, dev); 3880 3881 len = skb->len; 3882 trace_net_dev_start_xmit(skb, dev); 3883 rc = netdev_start_xmit(skb, dev, txq, more); 3884 trace_net_dev_xmit(skb, rc, dev, len); 3885 3886 return rc; 3887} 3888 3889struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3890 struct netdev_queue *txq, int *ret) 3891{ 3892 struct sk_buff *skb = first; 3893 int rc = NETDEV_TX_OK; 3894 3895 while (skb) { 3896 struct sk_buff *next = skb->next; 3897 3898 skb_mark_not_on_list(skb); 3899 rc = xmit_one(skb, dev, txq, next != NULL); 3900 if (unlikely(!dev_xmit_complete(rc))) { 3901 skb->next = next; 3902 goto out; 3903 } 3904 3905 skb = next; 3906 if (netif_tx_queue_stopped(txq) && skb) { 3907 rc = NETDEV_TX_BUSY; 3908 break; 3909 } 3910 } 3911 3912out: 3913 *ret = rc; 3914 return skb; 3915} 3916 3917static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3918 netdev_features_t features) 3919{ 3920 if (skb_vlan_tag_present(skb) && 3921 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3922 skb = __vlan_hwaccel_push_inside(skb); 3923 return skb; 3924} 3925 3926int skb_csum_hwoffload_help(struct sk_buff *skb, 3927 const netdev_features_t features) 3928{ 3929 if (unlikely(skb_csum_is_sctp(skb))) 3930 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3931 skb_crc32c_csum_help(skb); 3932 3933 if (features & NETIF_F_HW_CSUM) 3934 return 0; 3935 3936 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3937 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) && 3938 skb_network_header_len(skb) != sizeof(struct ipv6hdr) && 3939 !ipv6_has_hopopt_jumbo(skb)) 3940 goto sw_checksum; 3941 3942 switch (skb->csum_offset) { 3943 case offsetof(struct tcphdr, check): 3944 case offsetof(struct udphdr, check): 3945 return 0; 3946 } 3947 } 3948 3949sw_checksum: 3950 return skb_checksum_help(skb); 3951} 3952EXPORT_SYMBOL(skb_csum_hwoffload_help); 3953 3954/* Checks if this SKB belongs to an HW offloaded socket 3955 * and whether any SW fallbacks are required based on dev. 3956 * Check decrypted mark in case skb_orphan() cleared socket. 3957 */ 3958static struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 3959 struct net_device *dev) 3960{ 3961#ifdef CONFIG_SOCK_VALIDATE_XMIT 3962 struct sk_buff *(*sk_validate)(struct sock *sk, struct net_device *dev, 3963 struct sk_buff *skb); 3964 struct sock *sk = skb->sk; 3965 3966 sk_validate = NULL; 3967 if (sk) { 3968 if (sk_fullsock(sk)) 3969 sk_validate = sk->sk_validate_xmit_skb; 3970 else if (sk_is_inet(sk) && sk->sk_state == TCP_TIME_WAIT) 3971 sk_validate = inet_twsk(sk)->tw_validate_xmit_skb; 3972 } 3973 3974 if (sk_validate) { 3975 skb = sk_validate(sk, dev, skb); 3976 } else if (unlikely(skb_is_decrypted(skb))) { 3977 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 3978 kfree_skb(skb); 3979 skb = NULL; 3980 } 3981#endif 3982 3983 return skb; 3984} 3985 3986static struct sk_buff *validate_xmit_unreadable_skb(struct sk_buff *skb, 3987 struct net_device *dev) 3988{ 3989 struct skb_shared_info *shinfo; 3990 struct net_iov *niov; 3991 3992 if (likely(skb_frags_readable(skb))) 3993 goto out; 3994 3995 if (!dev->netmem_tx) 3996 goto out_free; 3997 3998 shinfo = skb_shinfo(skb); 3999 4000 if (shinfo->nr_frags > 0) { 4001 niov = netmem_to_net_iov(skb_frag_netmem(&shinfo->frags[0])); 4002 if (net_is_devmem_iov(niov) && 4003 READ_ONCE(net_devmem_iov_binding(niov)->dev) != dev) 4004 goto out_free; 4005 } 4006 4007out: 4008 return skb; 4009 4010out_free: 4011 kfree_skb(skb); 4012 return NULL; 4013} 4014 4015static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 4016{ 4017 netdev_features_t features; 4018 4019 skb = validate_xmit_unreadable_skb(skb, dev); 4020 if (unlikely(!skb)) 4021 goto out_null; 4022 4023 features = netif_skb_features(skb); 4024 skb = validate_xmit_vlan(skb, features); 4025 if (unlikely(!skb)) 4026 goto out_null; 4027 4028 skb = sk_validate_xmit_skb(skb, dev); 4029 if (unlikely(!skb)) 4030 goto out_null; 4031 4032 if (netif_needs_gso(skb, features)) { 4033 struct sk_buff *segs; 4034 4035 segs = skb_gso_segment(skb, features); 4036 if (IS_ERR(segs)) { 4037 goto out_kfree_skb; 4038 } else if (segs) { 4039 consume_skb(skb); 4040 skb = segs; 4041 } 4042 } else { 4043 if (skb_needs_linearize(skb, features) && 4044 __skb_linearize(skb)) 4045 goto out_kfree_skb; 4046 4047 /* If packet is not checksummed and device does not 4048 * support checksumming for this protocol, complete 4049 * checksumming here. 4050 */ 4051 if (skb->ip_summed == CHECKSUM_PARTIAL) { 4052 if (skb->encapsulation) 4053 skb_set_inner_transport_header(skb, 4054 skb_checksum_start_offset(skb)); 4055 else 4056 skb_set_transport_header(skb, 4057 skb_checksum_start_offset(skb)); 4058 if (skb_csum_hwoffload_help(skb, features)) 4059 goto out_kfree_skb; 4060 } 4061 } 4062 4063 skb = validate_xmit_xfrm(skb, features, again); 4064 4065 return skb; 4066 4067out_kfree_skb: 4068 kfree_skb(skb); 4069out_null: 4070 dev_core_stats_tx_dropped_inc(dev); 4071 return NULL; 4072} 4073 4074struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 4075{ 4076 struct sk_buff *next, *head = NULL, *tail; 4077 4078 for (; skb != NULL; skb = next) { 4079 next = skb->next; 4080 skb_mark_not_on_list(skb); 4081 4082 /* in case skb won't be segmented, point to itself */ 4083 skb->prev = skb; 4084 4085 skb = validate_xmit_skb(skb, dev, again); 4086 if (!skb) 4087 continue; 4088 4089 if (!head) 4090 head = skb; 4091 else 4092 tail->next = skb; 4093 /* If skb was segmented, skb->prev points to 4094 * the last segment. If not, it still contains skb. 4095 */ 4096 tail = skb->prev; 4097 } 4098 return head; 4099} 4100EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 4101 4102static void qdisc_pkt_len_segs_init(struct sk_buff *skb) 4103{ 4104 struct skb_shared_info *shinfo = skb_shinfo(skb); 4105 u16 gso_segs; 4106 4107 qdisc_skb_cb(skb)->pkt_len = skb->len; 4108 if (!shinfo->gso_size) { 4109 qdisc_skb_cb(skb)->pkt_segs = 1; 4110 return; 4111 } 4112 4113 qdisc_skb_cb(skb)->pkt_segs = gso_segs = shinfo->gso_segs; 4114 4115 /* To get more precise estimation of bytes sent on wire, 4116 * we add to pkt_len the headers size of all segments 4117 */ 4118 if (skb_transport_header_was_set(skb)) { 4119 unsigned int hdr_len; 4120 4121 /* mac layer + network layer */ 4122 if (!skb->encapsulation) 4123 hdr_len = skb_transport_offset(skb); 4124 else 4125 hdr_len = skb_inner_transport_offset(skb); 4126 4127 /* + transport layer */ 4128 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4129 const struct tcphdr *th; 4130 struct tcphdr _tcphdr; 4131 4132 th = skb_header_pointer(skb, hdr_len, 4133 sizeof(_tcphdr), &_tcphdr); 4134 if (likely(th)) 4135 hdr_len += __tcp_hdrlen(th); 4136 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 4137 struct udphdr _udphdr; 4138 4139 if (skb_header_pointer(skb, hdr_len, 4140 sizeof(_udphdr), &_udphdr)) 4141 hdr_len += sizeof(struct udphdr); 4142 } 4143 4144 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) { 4145 int payload = skb->len - hdr_len; 4146 4147 /* Malicious packet. */ 4148 if (payload <= 0) 4149 return; 4150 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size); 4151 shinfo->gso_segs = gso_segs; 4152 qdisc_skb_cb(skb)->pkt_segs = gso_segs; 4153 } 4154 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 4155 } 4156} 4157 4158static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 4159 struct sk_buff **to_free, 4160 struct netdev_queue *txq) 4161{ 4162 int rc; 4163 4164 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 4165 if (rc == NET_XMIT_SUCCESS) 4166 trace_qdisc_enqueue(q, txq, skb); 4167 return rc; 4168} 4169 4170static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 4171 struct net_device *dev, 4172 struct netdev_queue *txq) 4173{ 4174 struct sk_buff *next, *to_free = NULL, *to_free2 = NULL; 4175 spinlock_t *root_lock = qdisc_lock(q); 4176 struct llist_node *ll_list, *first_n; 4177 unsigned long defer_count = 0; 4178 int rc; 4179 4180 qdisc_calculate_pkt_len(skb, q); 4181 4182 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 4183 4184 if (q->flags & TCQ_F_NOLOCK) { 4185 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 4186 qdisc_run_begin(q)) { 4187 /* Retest nolock_qdisc_is_empty() within the protection 4188 * of q->seqlock to protect from racing with requeuing. 4189 */ 4190 if (unlikely(!nolock_qdisc_is_empty(q))) { 4191 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 4192 __qdisc_run(q); 4193 to_free2 = qdisc_run_end(q); 4194 4195 goto free_skbs; 4196 } 4197 4198 qdisc_bstats_cpu_update(q, skb); 4199 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 4200 !nolock_qdisc_is_empty(q)) 4201 __qdisc_run(q); 4202 4203 to_free2 = qdisc_run_end(q); 4204 rc = NET_XMIT_SUCCESS; 4205 goto free_skbs; 4206 } 4207 4208 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 4209 to_free2 = qdisc_run(q); 4210 goto free_skbs; 4211 } 4212 4213 /* Open code llist_add(&skb->ll_node, &q->defer_list) + queue limit. 4214 * In the try_cmpxchg() loop, we want to increment q->defer_count 4215 * at most once to limit the number of skbs in defer_list. 4216 * We perform the defer_count increment only if the list is not empty, 4217 * because some arches have slow atomic_long_inc_return(). 4218 */ 4219 first_n = READ_ONCE(q->defer_list.first); 4220 do { 4221 if (first_n && !defer_count) { 4222 defer_count = atomic_long_inc_return(&q->defer_count); 4223 if (unlikely(defer_count > READ_ONCE(net_hotdata.qdisc_max_burst))) { 4224 kfree_skb_reason(skb, SKB_DROP_REASON_QDISC_BURST_DROP); 4225 return NET_XMIT_DROP; 4226 } 4227 } 4228 skb->ll_node.next = first_n; 4229 } while (!try_cmpxchg(&q->defer_list.first, &first_n, &skb->ll_node)); 4230 4231 /* If defer_list was not empty, we know the cpu which queued 4232 * the first skb will process the whole list for us. 4233 */ 4234 if (first_n) 4235 return NET_XMIT_SUCCESS; 4236 4237 spin_lock(root_lock); 4238 4239 ll_list = llist_del_all(&q->defer_list); 4240 /* There is a small race because we clear defer_count not atomically 4241 * with the prior llist_del_all(). This means defer_list could grow 4242 * over qdisc_max_burst. 4243 */ 4244 atomic_long_set(&q->defer_count, 0); 4245 4246 ll_list = llist_reverse_order(ll_list); 4247 4248 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 4249 llist_for_each_entry_safe(skb, next, ll_list, ll_node) 4250 __qdisc_drop(skb, &to_free); 4251 rc = NET_XMIT_DROP; 4252 goto unlock; 4253 } 4254 if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 4255 !llist_next(ll_list) && qdisc_run_begin(q)) { 4256 /* 4257 * This is a work-conserving queue; there are no old skbs 4258 * waiting to be sent out; and the qdisc is not running - 4259 * xmit the skb directly. 4260 */ 4261 4262 DEBUG_NET_WARN_ON_ONCE(skb != llist_entry(ll_list, 4263 struct sk_buff, 4264 ll_node)); 4265 qdisc_bstats_update(q, skb); 4266 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) 4267 __qdisc_run(q); 4268 to_free2 = qdisc_run_end(q); 4269 rc = NET_XMIT_SUCCESS; 4270 } else { 4271 int count = 0; 4272 4273 llist_for_each_entry_safe(skb, next, ll_list, ll_node) { 4274 if (next) { 4275 prefetch(next); 4276 prefetch(&next->priority); 4277 skb_mark_not_on_list(skb); 4278 } 4279 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 4280 count++; 4281 } 4282 to_free2 = qdisc_run(q); 4283 if (count != 1) 4284 rc = NET_XMIT_SUCCESS; 4285 } 4286unlock: 4287 spin_unlock(root_lock); 4288 4289free_skbs: 4290 tcf_kfree_skb_list(to_free); 4291 tcf_kfree_skb_list(to_free2); 4292 return rc; 4293} 4294 4295#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 4296static void skb_update_prio(struct sk_buff *skb) 4297{ 4298 const struct netprio_map *map; 4299 const struct sock *sk; 4300 unsigned int prioidx; 4301 4302 if (skb->priority) 4303 return; 4304 map = rcu_dereference_bh(skb->dev->priomap); 4305 if (!map) 4306 return; 4307 sk = skb_to_full_sk(skb); 4308 if (!sk) 4309 return; 4310 4311 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 4312 4313 if (prioidx < map->priomap_len) 4314 skb->priority = map->priomap[prioidx]; 4315} 4316#else 4317#define skb_update_prio(skb) 4318#endif 4319 4320/** 4321 * dev_loopback_xmit - loop back @skb 4322 * @net: network namespace this loopback is happening in 4323 * @sk: sk needed to be a netfilter okfn 4324 * @skb: buffer to transmit 4325 */ 4326int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 4327{ 4328 skb_reset_mac_header(skb); 4329 __skb_pull(skb, skb_network_offset(skb)); 4330 skb->pkt_type = PACKET_LOOPBACK; 4331 if (skb->ip_summed == CHECKSUM_NONE) 4332 skb->ip_summed = CHECKSUM_UNNECESSARY; 4333 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 4334 skb_dst_force(skb); 4335 netif_rx(skb); 4336 return 0; 4337} 4338EXPORT_SYMBOL(dev_loopback_xmit); 4339 4340#ifdef CONFIG_NET_EGRESS 4341static struct netdev_queue * 4342netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 4343{ 4344 int qm = skb_get_queue_mapping(skb); 4345 4346 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 4347} 4348 4349#ifndef CONFIG_PREEMPT_RT 4350static bool netdev_xmit_txqueue_skipped(void) 4351{ 4352 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 4353} 4354 4355void netdev_xmit_skip_txqueue(bool skip) 4356{ 4357 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 4358} 4359EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4360 4361#else 4362static bool netdev_xmit_txqueue_skipped(void) 4363{ 4364 return current->net_xmit.skip_txqueue; 4365} 4366 4367void netdev_xmit_skip_txqueue(bool skip) 4368{ 4369 current->net_xmit.skip_txqueue = skip; 4370} 4371EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4372#endif 4373#endif /* CONFIG_NET_EGRESS */ 4374 4375#ifdef CONFIG_NET_XGRESS 4376static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 4377 enum skb_drop_reason *drop_reason) 4378{ 4379 int ret = TC_ACT_UNSPEC; 4380#ifdef CONFIG_NET_CLS_ACT 4381 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 4382 struct tcf_result res; 4383 4384 if (!miniq) 4385 return ret; 4386 4387 /* Global bypass */ 4388 if (!static_branch_likely(&tcf_sw_enabled_key)) 4389 return ret; 4390 4391 /* Block-wise bypass */ 4392 if (tcf_block_bypass_sw(miniq->block)) 4393 return ret; 4394 4395 tc_skb_cb(skb)->mru = 0; 4396 qdisc_skb_cb(skb)->post_ct = false; 4397 tcf_set_drop_reason(skb, *drop_reason); 4398 4399 mini_qdisc_bstats_cpu_update(miniq, skb); 4400 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 4401 /* Only tcf related quirks below. */ 4402 switch (ret) { 4403 case TC_ACT_SHOT: 4404 *drop_reason = tcf_get_drop_reason(skb); 4405 mini_qdisc_qstats_cpu_drop(miniq); 4406 break; 4407 case TC_ACT_OK: 4408 case TC_ACT_RECLASSIFY: 4409 skb->tc_index = TC_H_MIN(res.classid); 4410 break; 4411 } 4412#endif /* CONFIG_NET_CLS_ACT */ 4413 return ret; 4414} 4415 4416static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 4417 4418void tcx_inc(void) 4419{ 4420 static_branch_inc(&tcx_needed_key); 4421} 4422 4423void tcx_dec(void) 4424{ 4425 static_branch_dec(&tcx_needed_key); 4426} 4427 4428static __always_inline enum tcx_action_base 4429tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 4430 const bool needs_mac) 4431{ 4432 const struct bpf_mprog_fp *fp; 4433 const struct bpf_prog *prog; 4434 int ret = TCX_NEXT; 4435 4436 if (needs_mac) 4437 __skb_push(skb, skb->mac_len); 4438 bpf_mprog_foreach_prog(entry, fp, prog) { 4439 bpf_compute_data_pointers(skb); 4440 ret = bpf_prog_run(prog, skb); 4441 if (ret != TCX_NEXT) 4442 break; 4443 } 4444 if (needs_mac) 4445 __skb_pull(skb, skb->mac_len); 4446 return tcx_action_code(skb, ret); 4447} 4448 4449static __always_inline struct sk_buff * 4450sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4451 struct net_device *orig_dev, bool *another) 4452{ 4453 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4454 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 4455 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4456 int sch_ret; 4457 4458 if (!entry) 4459 return skb; 4460 4461 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4462 if (unlikely(*pt_prev)) { 4463 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4464 *pt_prev = NULL; 4465 } 4466 4467 qdisc_pkt_len_segs_init(skb); 4468 tcx_set_ingress(skb, true); 4469 4470 if (static_branch_unlikely(&tcx_needed_key)) { 4471 sch_ret = tcx_run(entry, skb, true); 4472 if (sch_ret != TC_ACT_UNSPEC) 4473 goto ingress_verdict; 4474 } 4475 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4476ingress_verdict: 4477 switch (sch_ret) { 4478 case TC_ACT_REDIRECT: 4479 /* skb_mac_header check was done by BPF, so we can safely 4480 * push the L2 header back before redirecting to another 4481 * netdev. 4482 */ 4483 __skb_push(skb, skb->mac_len); 4484 if (skb_do_redirect(skb) == -EAGAIN) { 4485 __skb_pull(skb, skb->mac_len); 4486 *another = true; 4487 break; 4488 } 4489 *ret = NET_RX_SUCCESS; 4490 bpf_net_ctx_clear(bpf_net_ctx); 4491 return NULL; 4492 case TC_ACT_SHOT: 4493 kfree_skb_reason(skb, drop_reason); 4494 *ret = NET_RX_DROP; 4495 bpf_net_ctx_clear(bpf_net_ctx); 4496 return NULL; 4497 /* used by tc_run */ 4498 case TC_ACT_STOLEN: 4499 case TC_ACT_QUEUED: 4500 case TC_ACT_TRAP: 4501 consume_skb(skb); 4502 fallthrough; 4503 case TC_ACT_CONSUMED: 4504 *ret = NET_RX_SUCCESS; 4505 bpf_net_ctx_clear(bpf_net_ctx); 4506 return NULL; 4507 } 4508 bpf_net_ctx_clear(bpf_net_ctx); 4509 4510 return skb; 4511} 4512 4513static __always_inline struct sk_buff * 4514sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4515{ 4516 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4517 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4518 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4519 int sch_ret; 4520 4521 if (!entry) 4522 return skb; 4523 4524 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4525 4526 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4527 * already set by the caller. 4528 */ 4529 if (static_branch_unlikely(&tcx_needed_key)) { 4530 sch_ret = tcx_run(entry, skb, false); 4531 if (sch_ret != TC_ACT_UNSPEC) 4532 goto egress_verdict; 4533 } 4534 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4535egress_verdict: 4536 switch (sch_ret) { 4537 case TC_ACT_REDIRECT: 4538 /* No need to push/pop skb's mac_header here on egress! */ 4539 skb_do_redirect(skb); 4540 *ret = NET_XMIT_SUCCESS; 4541 bpf_net_ctx_clear(bpf_net_ctx); 4542 return NULL; 4543 case TC_ACT_SHOT: 4544 kfree_skb_reason(skb, drop_reason); 4545 *ret = NET_XMIT_DROP; 4546 bpf_net_ctx_clear(bpf_net_ctx); 4547 return NULL; 4548 /* used by tc_run */ 4549 case TC_ACT_STOLEN: 4550 case TC_ACT_QUEUED: 4551 case TC_ACT_TRAP: 4552 consume_skb(skb); 4553 fallthrough; 4554 case TC_ACT_CONSUMED: 4555 *ret = NET_XMIT_SUCCESS; 4556 bpf_net_ctx_clear(bpf_net_ctx); 4557 return NULL; 4558 } 4559 bpf_net_ctx_clear(bpf_net_ctx); 4560 4561 return skb; 4562} 4563#else 4564static __always_inline struct sk_buff * 4565sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4566 struct net_device *orig_dev, bool *another) 4567{ 4568 return skb; 4569} 4570 4571static __always_inline struct sk_buff * 4572sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4573{ 4574 return skb; 4575} 4576#endif /* CONFIG_NET_XGRESS */ 4577 4578#ifdef CONFIG_XPS 4579static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4580 struct xps_dev_maps *dev_maps, unsigned int tci) 4581{ 4582 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4583 struct xps_map *map; 4584 int queue_index = -1; 4585 4586 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4587 return queue_index; 4588 4589 tci *= dev_maps->num_tc; 4590 tci += tc; 4591 4592 map = rcu_dereference(dev_maps->attr_map[tci]); 4593 if (map) { 4594 if (map->len == 1) 4595 queue_index = map->queues[0]; 4596 else 4597 queue_index = map->queues[reciprocal_scale( 4598 skb_get_hash(skb), map->len)]; 4599 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4600 queue_index = -1; 4601 } 4602 return queue_index; 4603} 4604#endif 4605 4606static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4607 struct sk_buff *skb) 4608{ 4609#ifdef CONFIG_XPS 4610 struct xps_dev_maps *dev_maps; 4611 struct sock *sk = skb->sk; 4612 int queue_index = -1; 4613 4614 if (!static_key_false(&xps_needed)) 4615 return -1; 4616 4617 rcu_read_lock(); 4618 if (!static_key_false(&xps_rxqs_needed)) 4619 goto get_cpus_map; 4620 4621 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4622 if (dev_maps) { 4623 int tci = sk_rx_queue_get(sk); 4624 4625 if (tci >= 0) 4626 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4627 tci); 4628 } 4629 4630get_cpus_map: 4631 if (queue_index < 0) { 4632 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4633 if (dev_maps) { 4634 unsigned int tci = skb->sender_cpu - 1; 4635 4636 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4637 tci); 4638 } 4639 } 4640 rcu_read_unlock(); 4641 4642 return queue_index; 4643#else 4644 return -1; 4645#endif 4646} 4647 4648u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4649 struct net_device *sb_dev) 4650{ 4651 return 0; 4652} 4653EXPORT_SYMBOL(dev_pick_tx_zero); 4654 4655int sk_tx_queue_get(const struct sock *sk) 4656{ 4657 int resel, val; 4658 4659 if (!sk) 4660 return -1; 4661 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 4662 * and sk_tx_queue_set(). 4663 */ 4664 val = READ_ONCE(sk->sk_tx_queue_mapping); 4665 4666 if (val == NO_QUEUE_MAPPING) 4667 return -1; 4668 4669 if (!sk_fullsock(sk)) 4670 return val; 4671 4672 resel = READ_ONCE(sock_net(sk)->core.sysctl_txq_reselection); 4673 if (resel && time_is_before_jiffies( 4674 READ_ONCE(sk->sk_tx_queue_mapping_jiffies) + resel)) 4675 return -1; 4676 4677 return val; 4678} 4679EXPORT_SYMBOL(sk_tx_queue_get); 4680 4681u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4682 struct net_device *sb_dev) 4683{ 4684 struct sock *sk = skb->sk; 4685 int queue_index = sk_tx_queue_get(sk); 4686 4687 sb_dev = sb_dev ? : dev; 4688 4689 if (queue_index < 0 || skb->ooo_okay || 4690 queue_index >= dev->real_num_tx_queues) { 4691 int new_index = get_xps_queue(dev, sb_dev, skb); 4692 4693 if (new_index < 0) 4694 new_index = skb_tx_hash(dev, sb_dev, skb); 4695 4696 if (sk && sk_fullsock(sk) && 4697 rcu_access_pointer(sk->sk_dst_cache)) 4698 sk_tx_queue_set(sk, new_index); 4699 4700 queue_index = new_index; 4701 } 4702 4703 return queue_index; 4704} 4705EXPORT_SYMBOL(netdev_pick_tx); 4706 4707struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4708 struct sk_buff *skb, 4709 struct net_device *sb_dev) 4710{ 4711 int queue_index = 0; 4712 4713#ifdef CONFIG_XPS 4714 u32 sender_cpu = skb->sender_cpu - 1; 4715 4716 if (sender_cpu >= (u32)NR_CPUS) 4717 skb->sender_cpu = raw_smp_processor_id() + 1; 4718#endif 4719 4720 if (dev->real_num_tx_queues != 1) { 4721 const struct net_device_ops *ops = dev->netdev_ops; 4722 4723 if (ops->ndo_select_queue) 4724 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4725 else 4726 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4727 4728 queue_index = netdev_cap_txqueue(dev, queue_index); 4729 } 4730 4731 skb_set_queue_mapping(skb, queue_index); 4732 return netdev_get_tx_queue(dev, queue_index); 4733} 4734 4735/** 4736 * __dev_queue_xmit() - transmit a buffer 4737 * @skb: buffer to transmit 4738 * @sb_dev: suboordinate device used for L2 forwarding offload 4739 * 4740 * Queue a buffer for transmission to a network device. The caller must 4741 * have set the device and priority and built the buffer before calling 4742 * this function. The function can be called from an interrupt. 4743 * 4744 * When calling this method, interrupts MUST be enabled. This is because 4745 * the BH enable code must have IRQs enabled so that it will not deadlock. 4746 * 4747 * Regardless of the return value, the skb is consumed, so it is currently 4748 * difficult to retry a send to this method. (You can bump the ref count 4749 * before sending to hold a reference for retry if you are careful.) 4750 * 4751 * Return: 4752 * * 0 - buffer successfully transmitted 4753 * * positive qdisc return code - NET_XMIT_DROP etc. 4754 * * negative errno - other errors 4755 */ 4756int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4757{ 4758 struct net_device *dev = skb->dev; 4759 struct netdev_queue *txq = NULL; 4760 struct Qdisc *q; 4761 int rc = -ENOMEM; 4762 bool again = false; 4763 4764 skb_reset_mac_header(skb); 4765 skb_assert_len(skb); 4766 4767 if (unlikely(skb_shinfo(skb)->tx_flags & 4768 (SKBTX_SCHED_TSTAMP | SKBTX_BPF))) 4769 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4770 4771 /* Disable soft irqs for various locks below. Also 4772 * stops preemption for RCU. 4773 */ 4774 rcu_read_lock_bh(); 4775 4776 skb_update_prio(skb); 4777 4778 qdisc_pkt_len_segs_init(skb); 4779 tcx_set_ingress(skb, false); 4780#ifdef CONFIG_NET_EGRESS 4781 if (static_branch_unlikely(&egress_needed_key)) { 4782 if (nf_hook_egress_active()) { 4783 skb = nf_hook_egress(skb, &rc, dev); 4784 if (!skb) 4785 goto out; 4786 } 4787 4788 netdev_xmit_skip_txqueue(false); 4789 4790 nf_skip_egress(skb, true); 4791 skb = sch_handle_egress(skb, &rc, dev); 4792 if (!skb) 4793 goto out; 4794 nf_skip_egress(skb, false); 4795 4796 if (netdev_xmit_txqueue_skipped()) 4797 txq = netdev_tx_queue_mapping(dev, skb); 4798 } 4799#endif 4800 /* If device/qdisc don't need skb->dst, release it right now while 4801 * its hot in this cpu cache. 4802 */ 4803 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4804 skb_dst_drop(skb); 4805 else 4806 skb_dst_force(skb); 4807 4808 if (!txq) 4809 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4810 4811 q = rcu_dereference_bh(txq->qdisc); 4812 4813 trace_net_dev_queue(skb); 4814 if (q->enqueue) { 4815 rc = __dev_xmit_skb(skb, q, dev, txq); 4816 goto out; 4817 } 4818 4819 /* The device has no queue. Common case for software devices: 4820 * loopback, all the sorts of tunnels... 4821 4822 * Really, it is unlikely that netif_tx_lock protection is necessary 4823 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4824 * counters.) 4825 * However, it is possible, that they rely on protection 4826 * made by us here. 4827 4828 * Check this and shot the lock. It is not prone from deadlocks. 4829 *Either shot noqueue qdisc, it is even simpler 8) 4830 */ 4831 if (dev->flags & IFF_UP) { 4832 int cpu = smp_processor_id(); /* ok because BHs are off */ 4833 4834 if (!netif_tx_owned(txq, cpu)) { 4835 bool is_list = false; 4836 4837 if (dev_xmit_recursion()) 4838 goto recursion_alert; 4839 4840 skb = validate_xmit_skb(skb, dev, &again); 4841 if (!skb) 4842 goto out; 4843 4844 HARD_TX_LOCK(dev, txq, cpu); 4845 4846 if (!netif_xmit_stopped(txq)) { 4847 is_list = !!skb->next; 4848 4849 dev_xmit_recursion_inc(); 4850 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4851 dev_xmit_recursion_dec(); 4852 4853 /* GSO segments a single SKB into 4854 * a list of frames. TCP expects error 4855 * to mean none of the data was sent. 4856 */ 4857 if (is_list) 4858 rc = NETDEV_TX_OK; 4859 } 4860 HARD_TX_UNLOCK(dev, txq); 4861 if (!skb) /* xmit completed */ 4862 goto out; 4863 4864 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4865 dev->name); 4866 /* NETDEV_TX_BUSY or queue was stopped */ 4867 if (!is_list) 4868 rc = -ENETDOWN; 4869 } else { 4870 /* Recursion is detected! It is possible, 4871 * unfortunately 4872 */ 4873recursion_alert: 4874 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4875 dev->name); 4876 rc = -ENETDOWN; 4877 } 4878 } 4879 4880 rcu_read_unlock_bh(); 4881 4882 dev_core_stats_tx_dropped_inc(dev); 4883 kfree_skb_list(skb); 4884 return rc; 4885out: 4886 rcu_read_unlock_bh(); 4887 return rc; 4888} 4889EXPORT_SYMBOL(__dev_queue_xmit); 4890 4891int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4892{ 4893 struct net_device *dev = skb->dev; 4894 struct sk_buff *orig_skb = skb; 4895 struct netdev_queue *txq; 4896 int ret = NETDEV_TX_BUSY; 4897 bool again = false; 4898 4899 if (unlikely(!netif_running(dev) || 4900 !netif_carrier_ok(dev))) 4901 goto drop; 4902 4903 skb = validate_xmit_skb_list(skb, dev, &again); 4904 if (skb != orig_skb) 4905 goto drop; 4906 4907 skb_set_queue_mapping(skb, queue_id); 4908 txq = skb_get_tx_queue(dev, skb); 4909 4910 local_bh_disable(); 4911 4912 dev_xmit_recursion_inc(); 4913 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4914 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4915 ret = netdev_start_xmit(skb, dev, txq, false); 4916 HARD_TX_UNLOCK(dev, txq); 4917 dev_xmit_recursion_dec(); 4918 4919 local_bh_enable(); 4920 return ret; 4921drop: 4922 dev_core_stats_tx_dropped_inc(dev); 4923 kfree_skb_list(skb); 4924 return NET_XMIT_DROP; 4925} 4926EXPORT_SYMBOL(__dev_direct_xmit); 4927 4928/************************************************************************* 4929 * Receiver routines 4930 *************************************************************************/ 4931static DEFINE_PER_CPU(struct task_struct *, backlog_napi); 4932 4933int weight_p __read_mostly = 64; /* old backlog weight */ 4934int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4935int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4936 4937/* Called with irq disabled */ 4938static inline void ____napi_schedule(struct softnet_data *sd, 4939 struct napi_struct *napi) 4940{ 4941 struct task_struct *thread; 4942 4943 lockdep_assert_irqs_disabled(); 4944 4945 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4946 /* Paired with smp_mb__before_atomic() in 4947 * napi_enable()/netif_set_threaded(). 4948 * Use READ_ONCE() to guarantee a complete 4949 * read on napi->thread. Only call 4950 * wake_up_process() when it's not NULL. 4951 */ 4952 thread = READ_ONCE(napi->thread); 4953 if (thread) { 4954 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi)) 4955 goto use_local_napi; 4956 4957 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4958 wake_up_process(thread); 4959 return; 4960 } 4961 } 4962 4963use_local_napi: 4964 DEBUG_NET_WARN_ON_ONCE(!list_empty(&napi->poll_list)); 4965 list_add_tail(&napi->poll_list, &sd->poll_list); 4966 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4967 /* If not called from net_rx_action() 4968 * we have to raise NET_RX_SOFTIRQ. 4969 */ 4970 if (!sd->in_net_rx_action) 4971 raise_softirq_irqoff(NET_RX_SOFTIRQ); 4972} 4973 4974#ifdef CONFIG_RPS 4975 4976struct static_key_false rps_needed __read_mostly; 4977EXPORT_SYMBOL(rps_needed); 4978struct static_key_false rfs_needed __read_mostly; 4979EXPORT_SYMBOL(rfs_needed); 4980 4981static u32 rfs_slot(u32 hash, const struct rps_dev_flow_table *flow_table) 4982{ 4983 return hash_32(hash, flow_table->log); 4984} 4985 4986#ifdef CONFIG_RFS_ACCEL 4987/** 4988 * rps_flow_is_active - check whether the flow is recently active. 4989 * @rflow: Specific flow to check activity. 4990 * @flow_table: per-queue flowtable that @rflow belongs to. 4991 * @cpu: CPU saved in @rflow. 4992 * 4993 * If the CPU has processed many packets since the flow's last activity 4994 * (beyond 10 times the table size), the flow is considered stale. 4995 * 4996 * Return: true if flow was recently active. 4997 */ 4998static bool rps_flow_is_active(struct rps_dev_flow *rflow, 4999 struct rps_dev_flow_table *flow_table, 5000 unsigned int cpu) 5001{ 5002 unsigned int flow_last_active; 5003 unsigned int sd_input_head; 5004 5005 if (cpu >= nr_cpu_ids) 5006 return false; 5007 5008 sd_input_head = READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head); 5009 flow_last_active = READ_ONCE(rflow->last_qtail); 5010 5011 return (int)(sd_input_head - flow_last_active) < 5012 (int)(10 << flow_table->log); 5013} 5014#endif 5015 5016static struct rps_dev_flow * 5017set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 5018 struct rps_dev_flow *rflow, u16 next_cpu, u32 hash) 5019{ 5020 if (next_cpu < nr_cpu_ids) { 5021 u32 head; 5022#ifdef CONFIG_RFS_ACCEL 5023 struct netdev_rx_queue *rxqueue; 5024 struct rps_dev_flow_table *flow_table; 5025 struct rps_dev_flow *old_rflow; 5026 struct rps_dev_flow *tmp_rflow; 5027 unsigned int tmp_cpu; 5028 u16 rxq_index; 5029 u32 flow_id; 5030 int rc; 5031 5032 /* Should we steer this flow to a different hardware queue? */ 5033 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 5034 !(dev->features & NETIF_F_NTUPLE)) 5035 goto out; 5036 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 5037 if (rxq_index == skb_get_rx_queue(skb)) 5038 goto out; 5039 5040 rxqueue = dev->_rx + rxq_index; 5041 flow_table = rcu_dereference(rxqueue->rps_flow_table); 5042 if (!flow_table) 5043 goto out; 5044 5045 flow_id = rfs_slot(hash, flow_table); 5046 tmp_rflow = &flow_table->flows[flow_id]; 5047 tmp_cpu = READ_ONCE(tmp_rflow->cpu); 5048 5049 if (READ_ONCE(tmp_rflow->filter) != RPS_NO_FILTER) { 5050 if (rps_flow_is_active(tmp_rflow, flow_table, 5051 tmp_cpu)) { 5052 if (hash != READ_ONCE(tmp_rflow->hash) || 5053 next_cpu == tmp_cpu) 5054 goto out; 5055 } 5056 } 5057 5058 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 5059 rxq_index, flow_id); 5060 if (rc < 0) 5061 goto out; 5062 5063 old_rflow = rflow; 5064 rflow = tmp_rflow; 5065 WRITE_ONCE(rflow->filter, rc); 5066 WRITE_ONCE(rflow->hash, hash); 5067 5068 if (old_rflow->filter == rc) 5069 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER); 5070 out: 5071#endif 5072 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head); 5073 rps_input_queue_tail_save(&rflow->last_qtail, head); 5074 } 5075 5076 WRITE_ONCE(rflow->cpu, next_cpu); 5077 return rflow; 5078} 5079 5080/* 5081 * get_rps_cpu is called from netif_receive_skb and returns the target 5082 * CPU from the RPS map of the receiving queue for a given skb. 5083 * rcu_read_lock must be held on entry. 5084 */ 5085static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 5086 struct rps_dev_flow **rflowp) 5087{ 5088 const struct rps_sock_flow_table *sock_flow_table; 5089 struct netdev_rx_queue *rxqueue = dev->_rx; 5090 struct rps_dev_flow_table *flow_table; 5091 struct rps_map *map; 5092 int cpu = -1; 5093 u32 tcpu; 5094 u32 hash; 5095 5096 if (skb_rx_queue_recorded(skb)) { 5097 u16 index = skb_get_rx_queue(skb); 5098 5099 if (unlikely(index >= dev->real_num_rx_queues)) { 5100 WARN_ONCE(dev->real_num_rx_queues > 1, 5101 "%s received packet on queue %u, but number " 5102 "of RX queues is %u\n", 5103 dev->name, index, dev->real_num_rx_queues); 5104 goto done; 5105 } 5106 rxqueue += index; 5107 } 5108 5109 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 5110 5111 flow_table = rcu_dereference(rxqueue->rps_flow_table); 5112 map = rcu_dereference(rxqueue->rps_map); 5113 if (!flow_table && !map) 5114 goto done; 5115 5116 skb_reset_network_header(skb); 5117 hash = skb_get_hash(skb); 5118 if (!hash) 5119 goto done; 5120 5121 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table); 5122 if (flow_table && sock_flow_table) { 5123 struct rps_dev_flow *rflow; 5124 u32 next_cpu; 5125 u32 ident; 5126 5127 /* First check into global flow table if there is a match. 5128 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 5129 */ 5130 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 5131 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask) 5132 goto try_rps; 5133 5134 next_cpu = ident & net_hotdata.rps_cpu_mask; 5135 5136 /* OK, now we know there is a match, 5137 * we can look at the local (per receive queue) flow table 5138 */ 5139 rflow = &flow_table->flows[rfs_slot(hash, flow_table)]; 5140 tcpu = rflow->cpu; 5141 5142 /* 5143 * If the desired CPU (where last recvmsg was done) is 5144 * different from current CPU (one in the rx-queue flow 5145 * table entry), switch if one of the following holds: 5146 * - Current CPU is unset (>= nr_cpu_ids). 5147 * - Current CPU is offline. 5148 * - The current CPU's queue tail has advanced beyond the 5149 * last packet that was enqueued using this table entry. 5150 * This guarantees that all previous packets for the flow 5151 * have been dequeued, thus preserving in order delivery. 5152 */ 5153 if (unlikely(tcpu != next_cpu) && 5154 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 5155 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) - 5156 rflow->last_qtail)) >= 0)) { 5157 tcpu = next_cpu; 5158 rflow = set_rps_cpu(dev, skb, rflow, next_cpu, hash); 5159 } 5160 5161 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 5162 *rflowp = rflow; 5163 cpu = tcpu; 5164 goto done; 5165 } 5166 } 5167 5168try_rps: 5169 5170 if (map) { 5171 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 5172 if (cpu_online(tcpu)) { 5173 cpu = tcpu; 5174 goto done; 5175 } 5176 } 5177 5178done: 5179 return cpu; 5180} 5181 5182#ifdef CONFIG_RFS_ACCEL 5183 5184/** 5185 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 5186 * @dev: Device on which the filter was set 5187 * @rxq_index: RX queue index 5188 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 5189 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 5190 * 5191 * Drivers that implement ndo_rx_flow_steer() should periodically call 5192 * this function for each installed filter and remove the filters for 5193 * which it returns %true. 5194 */ 5195bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 5196 u32 flow_id, u16 filter_id) 5197{ 5198 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 5199 struct rps_dev_flow_table *flow_table; 5200 struct rps_dev_flow *rflow; 5201 bool expire = true; 5202 5203 rcu_read_lock(); 5204 flow_table = rcu_dereference(rxqueue->rps_flow_table); 5205 if (flow_table && flow_id < (1UL << flow_table->log)) { 5206 unsigned int cpu; 5207 5208 rflow = &flow_table->flows[flow_id]; 5209 cpu = READ_ONCE(rflow->cpu); 5210 if (READ_ONCE(rflow->filter) == filter_id && 5211 rps_flow_is_active(rflow, flow_table, cpu)) 5212 expire = false; 5213 } 5214 rcu_read_unlock(); 5215 return expire; 5216} 5217EXPORT_SYMBOL(rps_may_expire_flow); 5218 5219#endif /* CONFIG_RFS_ACCEL */ 5220 5221/* Called from hardirq (IPI) context */ 5222static void rps_trigger_softirq(void *data) 5223{ 5224 struct softnet_data *sd = data; 5225 5226 ____napi_schedule(sd, &sd->backlog); 5227 /* Pairs with READ_ONCE() in softnet_seq_show() */ 5228 WRITE_ONCE(sd->received_rps, sd->received_rps + 1); 5229} 5230 5231#endif /* CONFIG_RPS */ 5232 5233/* Called from hardirq (IPI) context */ 5234static void trigger_rx_softirq(void *data) 5235{ 5236 struct softnet_data *sd = data; 5237 5238 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 5239 smp_store_release(&sd->defer_ipi_scheduled, 0); 5240} 5241 5242/* 5243 * After we queued a packet into sd->input_pkt_queue, 5244 * we need to make sure this queue is serviced soon. 5245 * 5246 * - If this is another cpu queue, link it to our rps_ipi_list, 5247 * and make sure we will process rps_ipi_list from net_rx_action(). 5248 * 5249 * - If this is our own queue, NAPI schedule our backlog. 5250 * Note that this also raises NET_RX_SOFTIRQ. 5251 */ 5252static void napi_schedule_rps(struct softnet_data *sd) 5253{ 5254 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 5255 5256#ifdef CONFIG_RPS 5257 if (sd != mysd) { 5258 if (use_backlog_threads()) { 5259 __napi_schedule_irqoff(&sd->backlog); 5260 return; 5261 } 5262 5263 sd->rps_ipi_next = mysd->rps_ipi_list; 5264 mysd->rps_ipi_list = sd; 5265 5266 /* If not called from net_rx_action() or napi_threaded_poll() 5267 * we have to raise NET_RX_SOFTIRQ. 5268 */ 5269 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 5270 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 5271 return; 5272 } 5273#endif /* CONFIG_RPS */ 5274 __napi_schedule_irqoff(&mysd->backlog); 5275} 5276 5277void kick_defer_list_purge(unsigned int cpu) 5278{ 5279 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5280 unsigned long flags; 5281 5282 if (use_backlog_threads()) { 5283 backlog_lock_irq_save(sd, &flags); 5284 5285 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 5286 __napi_schedule_irqoff(&sd->backlog); 5287 5288 backlog_unlock_irq_restore(sd, &flags); 5289 5290 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) { 5291 smp_call_function_single_async(cpu, &sd->defer_csd); 5292 } 5293} 5294 5295#ifdef CONFIG_NET_FLOW_LIMIT 5296int netdev_flow_limit_table_len __read_mostly = (1 << 12); 5297#endif 5298 5299static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen, 5300 int max_backlog) 5301{ 5302#ifdef CONFIG_NET_FLOW_LIMIT 5303 unsigned int old_flow, new_flow; 5304 const struct softnet_data *sd; 5305 struct sd_flow_limit *fl; 5306 5307 if (likely(qlen < (max_backlog >> 1))) 5308 return false; 5309 5310 sd = this_cpu_ptr(&softnet_data); 5311 5312 rcu_read_lock(); 5313 fl = rcu_dereference(sd->flow_limit); 5314 if (fl) { 5315 new_flow = hash_32(skb_get_hash(skb), fl->log_buckets); 5316 old_flow = fl->history[fl->history_head]; 5317 fl->history[fl->history_head] = new_flow; 5318 5319 fl->history_head++; 5320 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 5321 5322 if (likely(fl->buckets[old_flow])) 5323 fl->buckets[old_flow]--; 5324 5325 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 5326 /* Pairs with READ_ONCE() in softnet_seq_show() */ 5327 WRITE_ONCE(fl->count, fl->count + 1); 5328 rcu_read_unlock(); 5329 return true; 5330 } 5331 } 5332 rcu_read_unlock(); 5333#endif 5334 return false; 5335} 5336 5337/* 5338 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 5339 * queue (may be a remote CPU queue). 5340 */ 5341static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 5342 unsigned int *qtail) 5343{ 5344 enum skb_drop_reason reason; 5345 struct softnet_data *sd; 5346 unsigned long flags; 5347 unsigned int qlen; 5348 int max_backlog; 5349 u32 tail; 5350 5351 reason = SKB_DROP_REASON_DEV_READY; 5352 if (unlikely(!netif_running(skb->dev))) 5353 goto bad_dev; 5354 5355 sd = &per_cpu(softnet_data, cpu); 5356 5357 qlen = skb_queue_len_lockless(&sd->input_pkt_queue); 5358 max_backlog = READ_ONCE(net_hotdata.max_backlog); 5359 if (unlikely(qlen > max_backlog) || 5360 skb_flow_limit(skb, qlen, max_backlog)) 5361 goto cpu_backlog_drop; 5362 backlog_lock_irq_save(sd, &flags); 5363 qlen = skb_queue_len(&sd->input_pkt_queue); 5364 if (likely(qlen <= max_backlog)) { 5365 if (!qlen) { 5366 /* Schedule NAPI for backlog device. We can use 5367 * non atomic operation as we own the queue lock. 5368 */ 5369 if (!__test_and_set_bit(NAPI_STATE_SCHED, 5370 &sd->backlog.state)) 5371 napi_schedule_rps(sd); 5372 } 5373 __skb_queue_tail(&sd->input_pkt_queue, skb); 5374 tail = rps_input_queue_tail_incr(sd); 5375 backlog_unlock_irq_restore(sd, &flags); 5376 5377 /* save the tail outside of the critical section */ 5378 rps_input_queue_tail_save(qtail, tail); 5379 return NET_RX_SUCCESS; 5380 } 5381 5382 backlog_unlock_irq_restore(sd, &flags); 5383 5384cpu_backlog_drop: 5385 reason = SKB_DROP_REASON_CPU_BACKLOG; 5386 numa_drop_add(&sd->drop_counters, 1); 5387bad_dev: 5388 dev_core_stats_rx_dropped_inc(skb->dev); 5389 kfree_skb_reason(skb, reason); 5390 return NET_RX_DROP; 5391} 5392 5393static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 5394{ 5395 struct net_device *dev = skb->dev; 5396 struct netdev_rx_queue *rxqueue; 5397 5398 rxqueue = dev->_rx; 5399 5400 if (skb_rx_queue_recorded(skb)) { 5401 u16 index = skb_get_rx_queue(skb); 5402 5403 if (unlikely(index >= dev->real_num_rx_queues)) { 5404 WARN_ONCE(dev->real_num_rx_queues > 1, 5405 "%s received packet on queue %u, but number " 5406 "of RX queues is %u\n", 5407 dev->name, index, dev->real_num_rx_queues); 5408 5409 return rxqueue; /* Return first rxqueue */ 5410 } 5411 rxqueue += index; 5412 } 5413 return rxqueue; 5414} 5415 5416u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 5417 const struct bpf_prog *xdp_prog) 5418{ 5419 void *orig_data, *orig_data_end, *hard_start; 5420 struct netdev_rx_queue *rxqueue; 5421 bool orig_bcast, orig_host; 5422 u32 mac_len, frame_sz; 5423 __be16 orig_eth_type; 5424 struct ethhdr *eth; 5425 u32 metalen, act; 5426 int off; 5427 5428 /* The XDP program wants to see the packet starting at the MAC 5429 * header. 5430 */ 5431 mac_len = skb->data - skb_mac_header(skb); 5432 hard_start = skb->data - skb_headroom(skb); 5433 5434 /* SKB "head" area always have tailroom for skb_shared_info */ 5435 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 5436 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5437 5438 rxqueue = netif_get_rxqueue(skb); 5439 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 5440 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 5441 skb_headlen(skb) + mac_len, true); 5442 if (skb_is_nonlinear(skb)) { 5443 skb_shinfo(skb)->xdp_frags_size = skb->data_len; 5444 xdp_buff_set_frags_flag(xdp); 5445 } else { 5446 xdp_buff_clear_frags_flag(xdp); 5447 } 5448 5449 orig_data_end = xdp->data_end; 5450 orig_data = xdp->data; 5451 eth = (struct ethhdr *)xdp->data; 5452 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 5453 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 5454 orig_eth_type = eth->h_proto; 5455 5456 act = bpf_prog_run_xdp(xdp_prog, xdp); 5457 5458 /* check if bpf_xdp_adjust_head was used */ 5459 off = xdp->data - orig_data; 5460 if (off) { 5461 if (off > 0) 5462 __skb_pull(skb, off); 5463 else if (off < 0) 5464 __skb_push(skb, -off); 5465 5466 skb->mac_header += off; 5467 skb_reset_network_header(skb); 5468 } 5469 5470 /* check if bpf_xdp_adjust_tail was used */ 5471 off = xdp->data_end - orig_data_end; 5472 if (off != 0) { 5473 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 5474 skb->len += off; /* positive on grow, negative on shrink */ 5475 } 5476 5477 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers 5478 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. 5479 */ 5480 if (xdp_buff_has_frags(xdp)) 5481 skb->data_len = skb_shinfo(skb)->xdp_frags_size; 5482 else 5483 skb->data_len = 0; 5484 5485 /* check if XDP changed eth hdr such SKB needs update */ 5486 eth = (struct ethhdr *)xdp->data; 5487 if ((orig_eth_type != eth->h_proto) || 5488 (orig_host != ether_addr_equal_64bits(eth->h_dest, 5489 skb->dev->dev_addr)) || 5490 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 5491 __skb_push(skb, ETH_HLEN); 5492 skb->pkt_type = PACKET_HOST; 5493 skb->protocol = eth_type_trans(skb, skb->dev); 5494 } 5495 5496 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 5497 * before calling us again on redirect path. We do not call do_redirect 5498 * as we leave that up to the caller. 5499 * 5500 * Caller is responsible for managing lifetime of skb (i.e. calling 5501 * kfree_skb in response to actions it cannot handle/XDP_DROP). 5502 */ 5503 switch (act) { 5504 case XDP_REDIRECT: 5505 case XDP_TX: 5506 __skb_push(skb, mac_len); 5507 break; 5508 case XDP_PASS: 5509 metalen = xdp->data - xdp->data_meta; 5510 if (metalen) 5511 skb_metadata_set(skb, metalen); 5512 break; 5513 } 5514 5515 return act; 5516} 5517 5518static int 5519netif_skb_check_for_xdp(struct sk_buff **pskb, const struct bpf_prog *prog) 5520{ 5521 struct sk_buff *skb = *pskb; 5522 int err, hroom, troom; 5523 5524 local_lock_nested_bh(&system_page_pool.bh_lock); 5525 err = skb_cow_data_for_xdp(this_cpu_read(system_page_pool.pool), pskb, prog); 5526 local_unlock_nested_bh(&system_page_pool.bh_lock); 5527 if (!err) 5528 return 0; 5529 5530 /* In case we have to go down the path and also linearize, 5531 * then lets do the pskb_expand_head() work just once here. 5532 */ 5533 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 5534 troom = skb->tail + skb->data_len - skb->end; 5535 err = pskb_expand_head(skb, 5536 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 5537 troom > 0 ? troom + 128 : 0, GFP_ATOMIC); 5538 if (err) 5539 return err; 5540 5541 return skb_linearize(skb); 5542} 5543 5544static u32 netif_receive_generic_xdp(struct sk_buff **pskb, 5545 struct xdp_buff *xdp, 5546 const struct bpf_prog *xdp_prog) 5547{ 5548 struct sk_buff *skb = *pskb; 5549 u32 mac_len, act = XDP_DROP; 5550 5551 /* Reinjected packets coming from act_mirred or similar should 5552 * not get XDP generic processing. 5553 */ 5554 if (skb_is_redirected(skb)) 5555 return XDP_PASS; 5556 5557 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM 5558 * bytes. This is the guarantee that also native XDP provides, 5559 * thus we need to do it here as well. 5560 */ 5561 mac_len = skb->data - skb_mac_header(skb); 5562 __skb_push(skb, mac_len); 5563 5564 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 5565 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 5566 if (netif_skb_check_for_xdp(pskb, xdp_prog)) 5567 goto do_drop; 5568 } 5569 5570 __skb_pull(*pskb, mac_len); 5571 5572 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); 5573 switch (act) { 5574 case XDP_REDIRECT: 5575 case XDP_TX: 5576 case XDP_PASS: 5577 break; 5578 default: 5579 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); 5580 fallthrough; 5581 case XDP_ABORTED: 5582 trace_xdp_exception((*pskb)->dev, xdp_prog, act); 5583 fallthrough; 5584 case XDP_DROP: 5585 do_drop: 5586 kfree_skb(*pskb); 5587 break; 5588 } 5589 5590 return act; 5591} 5592 5593/* When doing generic XDP we have to bypass the qdisc layer and the 5594 * network taps in order to match in-driver-XDP behavior. This also means 5595 * that XDP packets are able to starve other packets going through a qdisc, 5596 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5597 * queues, so they do not have this starvation issue. 5598 */ 5599void generic_xdp_tx(struct sk_buff *skb, const struct bpf_prog *xdp_prog) 5600{ 5601 struct net_device *dev = skb->dev; 5602 struct netdev_queue *txq; 5603 bool free_skb = true; 5604 int cpu, rc; 5605 5606 txq = netdev_core_pick_tx(dev, skb, NULL); 5607 cpu = smp_processor_id(); 5608 HARD_TX_LOCK(dev, txq, cpu); 5609 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5610 rc = netdev_start_xmit(skb, dev, txq, 0); 5611 if (dev_xmit_complete(rc)) 5612 free_skb = false; 5613 } 5614 HARD_TX_UNLOCK(dev, txq); 5615 if (free_skb) { 5616 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5617 dev_core_stats_tx_dropped_inc(dev); 5618 kfree_skb(skb); 5619 } 5620} 5621 5622static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5623 5624int do_xdp_generic(const struct bpf_prog *xdp_prog, struct sk_buff **pskb) 5625{ 5626 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 5627 5628 if (xdp_prog) { 5629 struct xdp_buff xdp; 5630 u32 act; 5631 int err; 5632 5633 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 5634 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); 5635 if (act != XDP_PASS) { 5636 switch (act) { 5637 case XDP_REDIRECT: 5638 err = xdp_do_generic_redirect((*pskb)->dev, *pskb, 5639 &xdp, xdp_prog); 5640 if (err) 5641 goto out_redir; 5642 break; 5643 case XDP_TX: 5644 generic_xdp_tx(*pskb, xdp_prog); 5645 break; 5646 } 5647 bpf_net_ctx_clear(bpf_net_ctx); 5648 return XDP_DROP; 5649 } 5650 bpf_net_ctx_clear(bpf_net_ctx); 5651 } 5652 return XDP_PASS; 5653out_redir: 5654 bpf_net_ctx_clear(bpf_net_ctx); 5655 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); 5656 return XDP_DROP; 5657} 5658EXPORT_SYMBOL_GPL(do_xdp_generic); 5659 5660static int netif_rx_internal(struct sk_buff *skb) 5661{ 5662 int ret; 5663 5664 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5665 5666 trace_netif_rx(skb); 5667 5668#ifdef CONFIG_RPS 5669 if (static_branch_unlikely(&rps_needed)) { 5670 struct rps_dev_flow voidflow, *rflow = &voidflow; 5671 int cpu; 5672 5673 rcu_read_lock(); 5674 5675 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5676 if (cpu < 0) 5677 cpu = smp_processor_id(); 5678 5679 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5680 5681 rcu_read_unlock(); 5682 } else 5683#endif 5684 { 5685 unsigned int qtail; 5686 5687 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5688 } 5689 return ret; 5690} 5691 5692/** 5693 * __netif_rx - Slightly optimized version of netif_rx 5694 * @skb: buffer to post 5695 * 5696 * This behaves as netif_rx except that it does not disable bottom halves. 5697 * As a result this function may only be invoked from the interrupt context 5698 * (either hard or soft interrupt). 5699 */ 5700int __netif_rx(struct sk_buff *skb) 5701{ 5702 int ret; 5703 5704 lockdep_assert_once(hardirq_count() | softirq_count()); 5705 5706 trace_netif_rx_entry(skb); 5707 ret = netif_rx_internal(skb); 5708 trace_netif_rx_exit(ret); 5709 return ret; 5710} 5711EXPORT_SYMBOL(__netif_rx); 5712 5713/** 5714 * netif_rx - post buffer to the network code 5715 * @skb: buffer to post 5716 * 5717 * This function receives a packet from a device driver and queues it for 5718 * the upper (protocol) levels to process via the backlog NAPI device. It 5719 * always succeeds. The buffer may be dropped during processing for 5720 * congestion control or by the protocol layers. 5721 * The network buffer is passed via the backlog NAPI device. Modern NIC 5722 * driver should use NAPI and GRO. 5723 * This function can used from interrupt and from process context. The 5724 * caller from process context must not disable interrupts before invoking 5725 * this function. 5726 * 5727 * return values: 5728 * NET_RX_SUCCESS (no congestion) 5729 * NET_RX_DROP (packet was dropped) 5730 * 5731 */ 5732int netif_rx(struct sk_buff *skb) 5733{ 5734 bool need_bh_off = !(hardirq_count() | softirq_count()); 5735 int ret; 5736 5737 if (need_bh_off) 5738 local_bh_disable(); 5739 trace_netif_rx_entry(skb); 5740 ret = netif_rx_internal(skb); 5741 trace_netif_rx_exit(ret); 5742 if (need_bh_off) 5743 local_bh_enable(); 5744 return ret; 5745} 5746EXPORT_SYMBOL(netif_rx); 5747 5748static __latent_entropy void net_tx_action(void) 5749{ 5750 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5751 5752 if (sd->completion_queue) { 5753 struct sk_buff *clist; 5754 5755 local_irq_disable(); 5756 clist = sd->completion_queue; 5757 sd->completion_queue = NULL; 5758 local_irq_enable(); 5759 5760 while (clist) { 5761 struct sk_buff *skb = clist; 5762 5763 clist = clist->next; 5764 5765 WARN_ON(refcount_read(&skb->users)); 5766 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5767 trace_consume_skb(skb, net_tx_action); 5768 else 5769 trace_kfree_skb(skb, net_tx_action, 5770 get_kfree_skb_cb(skb)->reason, NULL); 5771 5772 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5773 __kfree_skb(skb); 5774 else 5775 __napi_kfree_skb(skb, 5776 get_kfree_skb_cb(skb)->reason); 5777 } 5778 } 5779 5780 if (sd->output_queue) { 5781 struct Qdisc *head; 5782 5783 local_irq_disable(); 5784 head = sd->output_queue; 5785 sd->output_queue = NULL; 5786 sd->output_queue_tailp = &sd->output_queue; 5787 local_irq_enable(); 5788 5789 rcu_read_lock(); 5790 5791 while (head) { 5792 spinlock_t *root_lock = NULL; 5793 struct sk_buff *to_free; 5794 struct Qdisc *q = head; 5795 5796 head = head->next_sched; 5797 5798 /* We need to make sure head->next_sched is read 5799 * before clearing __QDISC_STATE_SCHED 5800 */ 5801 smp_mb__before_atomic(); 5802 5803 if (!(q->flags & TCQ_F_NOLOCK)) { 5804 root_lock = qdisc_lock(q); 5805 spin_lock(root_lock); 5806 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5807 &q->state))) { 5808 /* There is a synchronize_net() between 5809 * STATE_DEACTIVATED flag being set and 5810 * qdisc_reset()/some_qdisc_is_busy() in 5811 * dev_deactivate(), so we can safely bail out 5812 * early here to avoid data race between 5813 * qdisc_deactivate() and some_qdisc_is_busy() 5814 * for lockless qdisc. 5815 */ 5816 clear_bit(__QDISC_STATE_SCHED, &q->state); 5817 continue; 5818 } 5819 5820 clear_bit(__QDISC_STATE_SCHED, &q->state); 5821 to_free = qdisc_run(q); 5822 if (root_lock) 5823 spin_unlock(root_lock); 5824 tcf_kfree_skb_list(to_free); 5825 } 5826 5827 rcu_read_unlock(); 5828 } 5829 5830 xfrm_dev_backlog(sd); 5831} 5832 5833#if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5834/* This hook is defined here for ATM LANE */ 5835int (*br_fdb_test_addr_hook)(struct net_device *dev, 5836 unsigned char *addr) __read_mostly; 5837EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5838#endif 5839 5840/** 5841 * netdev_is_rx_handler_busy - check if receive handler is registered 5842 * @dev: device to check 5843 * 5844 * Check if a receive handler is already registered for a given device. 5845 * Return true if there one. 5846 * 5847 * The caller must hold the rtnl_mutex. 5848 */ 5849bool netdev_is_rx_handler_busy(struct net_device *dev) 5850{ 5851 ASSERT_RTNL(); 5852 return dev && rtnl_dereference(dev->rx_handler); 5853} 5854EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5855 5856/** 5857 * netdev_rx_handler_register - register receive handler 5858 * @dev: device to register a handler for 5859 * @rx_handler: receive handler to register 5860 * @rx_handler_data: data pointer that is used by rx handler 5861 * 5862 * Register a receive handler for a device. This handler will then be 5863 * called from __netif_receive_skb. A negative errno code is returned 5864 * on a failure. 5865 * 5866 * The caller must hold the rtnl_mutex. 5867 * 5868 * For a general description of rx_handler, see enum rx_handler_result. 5869 */ 5870int netdev_rx_handler_register(struct net_device *dev, 5871 rx_handler_func_t *rx_handler, 5872 void *rx_handler_data) 5873{ 5874 if (netdev_is_rx_handler_busy(dev)) 5875 return -EBUSY; 5876 5877 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5878 return -EINVAL; 5879 5880 /* Note: rx_handler_data must be set before rx_handler */ 5881 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5882 rcu_assign_pointer(dev->rx_handler, rx_handler); 5883 5884 return 0; 5885} 5886EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5887 5888/** 5889 * netdev_rx_handler_unregister - unregister receive handler 5890 * @dev: device to unregister a handler from 5891 * 5892 * Unregister a receive handler from a device. 5893 * 5894 * The caller must hold the rtnl_mutex. 5895 */ 5896void netdev_rx_handler_unregister(struct net_device *dev) 5897{ 5898 5899 ASSERT_RTNL(); 5900 RCU_INIT_POINTER(dev->rx_handler, NULL); 5901 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5902 * section has a guarantee to see a non NULL rx_handler_data 5903 * as well. 5904 */ 5905 synchronize_net(); 5906 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5907} 5908EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5909 5910/* 5911 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5912 * the special handling of PFMEMALLOC skbs. 5913 */ 5914static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5915{ 5916 switch (skb->protocol) { 5917 case htons(ETH_P_ARP): 5918 case htons(ETH_P_IP): 5919 case htons(ETH_P_IPV6): 5920 case htons(ETH_P_8021Q): 5921 case htons(ETH_P_8021AD): 5922 return true; 5923 default: 5924 return false; 5925 } 5926} 5927 5928static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5929 int *ret, struct net_device *orig_dev) 5930{ 5931 if (nf_hook_ingress_active(skb)) { 5932 int ingress_retval; 5933 5934 if (unlikely(*pt_prev)) { 5935 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5936 *pt_prev = NULL; 5937 } 5938 5939 rcu_read_lock(); 5940 ingress_retval = nf_hook_ingress(skb); 5941 rcu_read_unlock(); 5942 return ingress_retval; 5943 } 5944 return 0; 5945} 5946 5947static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5948 struct packet_type **ppt_prev) 5949{ 5950 enum skb_drop_reason drop_reason = SKB_DROP_REASON_UNHANDLED_PROTO; 5951 struct packet_type *ptype, *pt_prev; 5952 rx_handler_func_t *rx_handler; 5953 struct sk_buff *skb = *pskb; 5954 struct net_device *orig_dev; 5955 bool deliver_exact = false; 5956 int ret = NET_RX_DROP; 5957 __be16 type; 5958 5959 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5960 5961 trace_netif_receive_skb(skb); 5962 5963 orig_dev = skb->dev; 5964 5965 skb_reset_network_header(skb); 5966#if !defined(CONFIG_DEBUG_NET) 5967 /* We plan to no longer reset the transport header here. 5968 * Give some time to fuzzers and dev build to catch bugs 5969 * in network stacks. 5970 */ 5971 if (!skb_transport_header_was_set(skb)) 5972 skb_reset_transport_header(skb); 5973#endif 5974 skb_reset_mac_len(skb); 5975 5976 pt_prev = NULL; 5977 5978another_round: 5979 skb->skb_iif = skb->dev->ifindex; 5980 5981 __this_cpu_inc(softnet_data.processed); 5982 5983 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5984 int ret2; 5985 5986 migrate_disable(); 5987 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), 5988 &skb); 5989 migrate_enable(); 5990 5991 if (ret2 != XDP_PASS) { 5992 ret = NET_RX_DROP; 5993 goto out; 5994 } 5995 } 5996 5997 if (eth_type_vlan(skb->protocol)) { 5998 skb = skb_vlan_untag(skb); 5999 if (unlikely(!skb)) 6000 goto out; 6001 } 6002 6003 if (skb_skip_tc_classify(skb)) 6004 goto skip_classify; 6005 6006 if (pfmemalloc) 6007 goto skip_taps; 6008 6009 list_for_each_entry_rcu(ptype, &dev_net_rcu(skb->dev)->ptype_all, 6010 list) { 6011 if (unlikely(pt_prev)) 6012 ret = deliver_skb(skb, pt_prev, orig_dev); 6013 pt_prev = ptype; 6014 } 6015 6016 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 6017 if (unlikely(pt_prev)) 6018 ret = deliver_skb(skb, pt_prev, orig_dev); 6019 pt_prev = ptype; 6020 } 6021 6022skip_taps: 6023#ifdef CONFIG_NET_INGRESS 6024 if (static_branch_unlikely(&ingress_needed_key)) { 6025 bool another = false; 6026 6027 nf_skip_egress(skb, true); 6028 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 6029 &another); 6030 if (another) 6031 goto another_round; 6032 if (!skb) 6033 goto out; 6034 6035 nf_skip_egress(skb, false); 6036 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 6037 goto out; 6038 } 6039#endif 6040 skb_reset_redirect(skb); 6041skip_classify: 6042 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) { 6043 drop_reason = SKB_DROP_REASON_PFMEMALLOC; 6044 goto drop; 6045 } 6046 6047 if (skb_vlan_tag_present(skb)) { 6048 if (unlikely(pt_prev)) { 6049 ret = deliver_skb(skb, pt_prev, orig_dev); 6050 pt_prev = NULL; 6051 } 6052 if (vlan_do_receive(&skb)) 6053 goto another_round; 6054 else if (unlikely(!skb)) 6055 goto out; 6056 } 6057 6058 rx_handler = rcu_dereference(skb->dev->rx_handler); 6059 if (rx_handler) { 6060 if (unlikely(pt_prev)) { 6061 ret = deliver_skb(skb, pt_prev, orig_dev); 6062 pt_prev = NULL; 6063 } 6064 switch (rx_handler(&skb)) { 6065 case RX_HANDLER_CONSUMED: 6066 ret = NET_RX_SUCCESS; 6067 goto out; 6068 case RX_HANDLER_ANOTHER: 6069 goto another_round; 6070 case RX_HANDLER_EXACT: 6071 deliver_exact = true; 6072 break; 6073 case RX_HANDLER_PASS: 6074 break; 6075 default: 6076 BUG(); 6077 } 6078 } 6079 6080 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 6081check_vlan_id: 6082 if (skb_vlan_tag_get_id(skb)) { 6083 /* Vlan id is non 0 and vlan_do_receive() above couldn't 6084 * find vlan device. 6085 */ 6086 skb->pkt_type = PACKET_OTHERHOST; 6087 } else if (eth_type_vlan(skb->protocol)) { 6088 /* Outer header is 802.1P with vlan 0, inner header is 6089 * 802.1Q or 802.1AD and vlan_do_receive() above could 6090 * not find vlan dev for vlan id 0. 6091 */ 6092 __vlan_hwaccel_clear_tag(skb); 6093 skb = skb_vlan_untag(skb); 6094 if (unlikely(!skb)) 6095 goto out; 6096 if (vlan_do_receive(&skb)) 6097 /* After stripping off 802.1P header with vlan 0 6098 * vlan dev is found for inner header. 6099 */ 6100 goto another_round; 6101 else if (unlikely(!skb)) 6102 goto out; 6103 else 6104 /* We have stripped outer 802.1P vlan 0 header. 6105 * But could not find vlan dev. 6106 * check again for vlan id to set OTHERHOST. 6107 */ 6108 goto check_vlan_id; 6109 } 6110 /* Note: we might in the future use prio bits 6111 * and set skb->priority like in vlan_do_receive() 6112 * For the time being, just ignore Priority Code Point 6113 */ 6114 __vlan_hwaccel_clear_tag(skb); 6115 } 6116 6117 type = skb->protocol; 6118 6119 /* deliver only exact match when indicated */ 6120 if (likely(!deliver_exact)) { 6121 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 6122 &ptype_base[ntohs(type) & 6123 PTYPE_HASH_MASK]); 6124 6125 /* orig_dev and skb->dev could belong to different netns; 6126 * Even in such case we need to traverse only the list 6127 * coming from skb->dev, as the ptype owner (packet socket) 6128 * will use dev_net(skb->dev) to do namespace filtering. 6129 */ 6130 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 6131 &dev_net_rcu(skb->dev)->ptype_specific); 6132 } 6133 6134 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 6135 &orig_dev->ptype_specific); 6136 6137 if (unlikely(skb->dev != orig_dev)) { 6138 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 6139 &skb->dev->ptype_specific); 6140 } 6141 6142 if (pt_prev) { 6143 *ppt_prev = pt_prev; 6144 } else { 6145drop: 6146 if (!deliver_exact) 6147 dev_core_stats_rx_dropped_inc(skb->dev); 6148 else 6149 dev_core_stats_rx_nohandler_inc(skb->dev); 6150 6151 kfree_skb_reason(skb, drop_reason); 6152 /* Jamal, now you will not able to escape explaining 6153 * me how you were going to use this. :-) 6154 */ 6155 ret = NET_RX_DROP; 6156 } 6157 6158out: 6159 /* The invariant here is that if *ppt_prev is not NULL 6160 * then skb should also be non-NULL. 6161 * 6162 * Apparently *ppt_prev assignment above holds this invariant due to 6163 * skb dereferencing near it. 6164 */ 6165 *pskb = skb; 6166 return ret; 6167} 6168 6169static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 6170{ 6171 struct net_device *orig_dev = skb->dev; 6172 struct packet_type *pt_prev = NULL; 6173 int ret; 6174 6175 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 6176 if (pt_prev) 6177 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 6178 skb->dev, pt_prev, orig_dev); 6179 return ret; 6180} 6181 6182/** 6183 * netif_receive_skb_core - special purpose version of netif_receive_skb 6184 * @skb: buffer to process 6185 * 6186 * More direct receive version of netif_receive_skb(). It should 6187 * only be used by callers that have a need to skip RPS and Generic XDP. 6188 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 6189 * 6190 * This function may only be called from softirq context and interrupts 6191 * should be enabled. 6192 * 6193 * Return values (usually ignored): 6194 * NET_RX_SUCCESS: no congestion 6195 * NET_RX_DROP: packet was dropped 6196 */ 6197int netif_receive_skb_core(struct sk_buff *skb) 6198{ 6199 int ret; 6200 6201 rcu_read_lock(); 6202 ret = __netif_receive_skb_one_core(skb, false); 6203 rcu_read_unlock(); 6204 6205 return ret; 6206} 6207EXPORT_SYMBOL(netif_receive_skb_core); 6208 6209static inline void __netif_receive_skb_list_ptype(struct list_head *head, 6210 struct packet_type *pt_prev, 6211 struct net_device *orig_dev) 6212{ 6213 struct sk_buff *skb, *next; 6214 6215 if (!pt_prev) 6216 return; 6217 if (list_empty(head)) 6218 return; 6219 if (pt_prev->list_func != NULL) 6220 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 6221 ip_list_rcv, head, pt_prev, orig_dev); 6222 else 6223 list_for_each_entry_safe(skb, next, head, list) { 6224 skb_list_del_init(skb); 6225 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 6226 } 6227} 6228 6229static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 6230{ 6231 /* Fast-path assumptions: 6232 * - There is no RX handler. 6233 * - Only one packet_type matches. 6234 * If either of these fails, we will end up doing some per-packet 6235 * processing in-line, then handling the 'last ptype' for the whole 6236 * sublist. This can't cause out-of-order delivery to any single ptype, 6237 * because the 'last ptype' must be constant across the sublist, and all 6238 * other ptypes are handled per-packet. 6239 */ 6240 /* Current (common) ptype of sublist */ 6241 struct packet_type *pt_curr = NULL; 6242 /* Current (common) orig_dev of sublist */ 6243 struct net_device *od_curr = NULL; 6244 struct sk_buff *skb, *next; 6245 LIST_HEAD(sublist); 6246 6247 list_for_each_entry_safe(skb, next, head, list) { 6248 struct net_device *orig_dev = skb->dev; 6249 struct packet_type *pt_prev = NULL; 6250 6251 skb_list_del_init(skb); 6252 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 6253 if (!pt_prev) 6254 continue; 6255 if (pt_curr != pt_prev || od_curr != orig_dev) { 6256 /* dispatch old sublist */ 6257 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 6258 /* start new sublist */ 6259 INIT_LIST_HEAD(&sublist); 6260 pt_curr = pt_prev; 6261 od_curr = orig_dev; 6262 } 6263 list_add_tail(&skb->list, &sublist); 6264 } 6265 6266 /* dispatch final sublist */ 6267 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 6268} 6269 6270static int __netif_receive_skb(struct sk_buff *skb) 6271{ 6272 int ret; 6273 6274 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 6275 unsigned int noreclaim_flag; 6276 6277 /* 6278 * PFMEMALLOC skbs are special, they should 6279 * - be delivered to SOCK_MEMALLOC sockets only 6280 * - stay away from userspace 6281 * - have bounded memory usage 6282 * 6283 * Use PF_MEMALLOC as this saves us from propagating the allocation 6284 * context down to all allocation sites. 6285 */ 6286 noreclaim_flag = memalloc_noreclaim_save(); 6287 ret = __netif_receive_skb_one_core(skb, true); 6288 memalloc_noreclaim_restore(noreclaim_flag); 6289 } else 6290 ret = __netif_receive_skb_one_core(skb, false); 6291 6292 return ret; 6293} 6294 6295static void __netif_receive_skb_list(struct list_head *head) 6296{ 6297 unsigned long noreclaim_flag = 0; 6298 struct sk_buff *skb, *next; 6299 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 6300 6301 list_for_each_entry_safe(skb, next, head, list) { 6302 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 6303 struct list_head sublist; 6304 6305 /* Handle the previous sublist */ 6306 list_cut_before(&sublist, head, &skb->list); 6307 if (!list_empty(&sublist)) 6308 __netif_receive_skb_list_core(&sublist, pfmemalloc); 6309 pfmemalloc = !pfmemalloc; 6310 /* See comments in __netif_receive_skb */ 6311 if (pfmemalloc) 6312 noreclaim_flag = memalloc_noreclaim_save(); 6313 else 6314 memalloc_noreclaim_restore(noreclaim_flag); 6315 } 6316 } 6317 /* Handle the remaining sublist */ 6318 if (!list_empty(head)) 6319 __netif_receive_skb_list_core(head, pfmemalloc); 6320 /* Restore pflags */ 6321 if (pfmemalloc) 6322 memalloc_noreclaim_restore(noreclaim_flag); 6323} 6324 6325static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 6326{ 6327 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 6328 struct bpf_prog *new = xdp->prog; 6329 int ret = 0; 6330 6331 switch (xdp->command) { 6332 case XDP_SETUP_PROG: 6333 rcu_assign_pointer(dev->xdp_prog, new); 6334 if (old) 6335 bpf_prog_put(old); 6336 6337 if (old && !new) { 6338 static_branch_dec(&generic_xdp_needed_key); 6339 } else if (new && !old) { 6340 static_branch_inc(&generic_xdp_needed_key); 6341 netif_disable_lro(dev); 6342 dev_disable_gro_hw(dev); 6343 } 6344 break; 6345 6346 default: 6347 ret = -EINVAL; 6348 break; 6349 } 6350 6351 return ret; 6352} 6353 6354static int netif_receive_skb_internal(struct sk_buff *skb) 6355{ 6356 int ret; 6357 6358 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 6359 6360 if (skb_defer_rx_timestamp(skb)) 6361 return NET_RX_SUCCESS; 6362 6363 rcu_read_lock(); 6364#ifdef CONFIG_RPS 6365 if (static_branch_unlikely(&rps_needed)) { 6366 struct rps_dev_flow voidflow, *rflow = &voidflow; 6367 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 6368 6369 if (cpu >= 0) { 6370 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 6371 rcu_read_unlock(); 6372 return ret; 6373 } 6374 } 6375#endif 6376 ret = __netif_receive_skb(skb); 6377 rcu_read_unlock(); 6378 return ret; 6379} 6380 6381void netif_receive_skb_list_internal(struct list_head *head) 6382{ 6383 struct sk_buff *skb, *next; 6384 LIST_HEAD(sublist); 6385 6386 list_for_each_entry_safe(skb, next, head, list) { 6387 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), 6388 skb); 6389 skb_list_del_init(skb); 6390 if (!skb_defer_rx_timestamp(skb)) 6391 list_add_tail(&skb->list, &sublist); 6392 } 6393 list_splice_init(&sublist, head); 6394 6395 rcu_read_lock(); 6396#ifdef CONFIG_RPS 6397 if (static_branch_unlikely(&rps_needed)) { 6398 list_for_each_entry_safe(skb, next, head, list) { 6399 struct rps_dev_flow voidflow, *rflow = &voidflow; 6400 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 6401 6402 if (cpu >= 0) { 6403 /* Will be handled, remove from list */ 6404 skb_list_del_init(skb); 6405 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 6406 } 6407 } 6408 } 6409#endif 6410 __netif_receive_skb_list(head); 6411 rcu_read_unlock(); 6412} 6413 6414/** 6415 * netif_receive_skb - process receive buffer from network 6416 * @skb: buffer to process 6417 * 6418 * netif_receive_skb() is the main receive data processing function. 6419 * It always succeeds. The buffer may be dropped during processing 6420 * for congestion control or by the protocol layers. 6421 * 6422 * This function may only be called from softirq context and interrupts 6423 * should be enabled. 6424 * 6425 * Return values (usually ignored): 6426 * NET_RX_SUCCESS: no congestion 6427 * NET_RX_DROP: packet was dropped 6428 */ 6429int netif_receive_skb(struct sk_buff *skb) 6430{ 6431 int ret; 6432 6433 trace_netif_receive_skb_entry(skb); 6434 6435 ret = netif_receive_skb_internal(skb); 6436 trace_netif_receive_skb_exit(ret); 6437 6438 return ret; 6439} 6440EXPORT_SYMBOL(netif_receive_skb); 6441 6442/** 6443 * netif_receive_skb_list - process many receive buffers from network 6444 * @head: list of skbs to process. 6445 * 6446 * Since return value of netif_receive_skb() is normally ignored, and 6447 * wouldn't be meaningful for a list, this function returns void. 6448 * 6449 * This function may only be called from softirq context and interrupts 6450 * should be enabled. 6451 */ 6452void netif_receive_skb_list(struct list_head *head) 6453{ 6454 struct sk_buff *skb; 6455 6456 if (list_empty(head)) 6457 return; 6458 if (trace_netif_receive_skb_list_entry_enabled()) { 6459 list_for_each_entry(skb, head, list) 6460 trace_netif_receive_skb_list_entry(skb); 6461 } 6462 netif_receive_skb_list_internal(head); 6463 trace_netif_receive_skb_list_exit(0); 6464} 6465EXPORT_SYMBOL(netif_receive_skb_list); 6466 6467/* Network device is going away, flush any packets still pending */ 6468static void flush_backlog(struct work_struct *work) 6469{ 6470 struct sk_buff *skb, *tmp; 6471 struct sk_buff_head list; 6472 struct softnet_data *sd; 6473 6474 __skb_queue_head_init(&list); 6475 local_bh_disable(); 6476 sd = this_cpu_ptr(&softnet_data); 6477 6478 backlog_lock_irq_disable(sd); 6479 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 6480 if (READ_ONCE(skb->dev->reg_state) == NETREG_UNREGISTERING) { 6481 __skb_unlink(skb, &sd->input_pkt_queue); 6482 __skb_queue_tail(&list, skb); 6483 rps_input_queue_head_incr(sd); 6484 } 6485 } 6486 backlog_unlock_irq_enable(sd); 6487 6488 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6489 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 6490 if (READ_ONCE(skb->dev->reg_state) == NETREG_UNREGISTERING) { 6491 __skb_unlink(skb, &sd->process_queue); 6492 __skb_queue_tail(&list, skb); 6493 rps_input_queue_head_incr(sd); 6494 } 6495 } 6496 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6497 local_bh_enable(); 6498 6499 __skb_queue_purge_reason(&list, SKB_DROP_REASON_DEV_READY); 6500} 6501 6502static bool flush_required(int cpu) 6503{ 6504#if IS_ENABLED(CONFIG_RPS) 6505 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 6506 bool do_flush; 6507 6508 backlog_lock_irq_disable(sd); 6509 6510 /* as insertion into process_queue happens with the rps lock held, 6511 * process_queue access may race only with dequeue 6512 */ 6513 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 6514 !skb_queue_empty_lockless(&sd->process_queue); 6515 backlog_unlock_irq_enable(sd); 6516 6517 return do_flush; 6518#endif 6519 /* without RPS we can't safely check input_pkt_queue: during a 6520 * concurrent remote skb_queue_splice() we can detect as empty both 6521 * input_pkt_queue and process_queue even if the latter could end-up 6522 * containing a lot of packets. 6523 */ 6524 return true; 6525} 6526 6527struct flush_backlogs { 6528 cpumask_t flush_cpus; 6529 struct work_struct w[]; 6530}; 6531 6532static struct flush_backlogs *flush_backlogs_alloc(void) 6533{ 6534 return kmalloc(struct_size_t(struct flush_backlogs, w, nr_cpu_ids), 6535 GFP_KERNEL); 6536} 6537 6538static struct flush_backlogs *flush_backlogs_fallback; 6539static DEFINE_MUTEX(flush_backlogs_mutex); 6540 6541static void flush_all_backlogs(void) 6542{ 6543 struct flush_backlogs *ptr = flush_backlogs_alloc(); 6544 unsigned int cpu; 6545 6546 if (!ptr) { 6547 mutex_lock(&flush_backlogs_mutex); 6548 ptr = flush_backlogs_fallback; 6549 } 6550 cpumask_clear(&ptr->flush_cpus); 6551 6552 cpus_read_lock(); 6553 6554 for_each_online_cpu(cpu) { 6555 if (flush_required(cpu)) { 6556 INIT_WORK(&ptr->w[cpu], flush_backlog); 6557 queue_work_on(cpu, system_highpri_wq, &ptr->w[cpu]); 6558 __cpumask_set_cpu(cpu, &ptr->flush_cpus); 6559 } 6560 } 6561 6562 /* we can have in flight packet[s] on the cpus we are not flushing, 6563 * synchronize_net() in unregister_netdevice_many() will take care of 6564 * them. 6565 */ 6566 for_each_cpu(cpu, &ptr->flush_cpus) 6567 flush_work(&ptr->w[cpu]); 6568 6569 cpus_read_unlock(); 6570 6571 if (ptr != flush_backlogs_fallback) 6572 kfree(ptr); 6573 else 6574 mutex_unlock(&flush_backlogs_mutex); 6575} 6576 6577static void net_rps_send_ipi(struct softnet_data *remsd) 6578{ 6579#ifdef CONFIG_RPS 6580 while (remsd) { 6581 struct softnet_data *next = remsd->rps_ipi_next; 6582 6583 if (cpu_online(remsd->cpu)) 6584 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6585 remsd = next; 6586 } 6587#endif 6588} 6589 6590/* 6591 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6592 * Note: called with local irq disabled, but exits with local irq enabled. 6593 */ 6594static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6595{ 6596#ifdef CONFIG_RPS 6597 struct softnet_data *remsd = sd->rps_ipi_list; 6598 6599 if (!use_backlog_threads() && remsd) { 6600 sd->rps_ipi_list = NULL; 6601 6602 local_irq_enable(); 6603 6604 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6605 net_rps_send_ipi(remsd); 6606 } else 6607#endif 6608 local_irq_enable(); 6609} 6610 6611static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6612{ 6613#ifdef CONFIG_RPS 6614 return !use_backlog_threads() && sd->rps_ipi_list; 6615#else 6616 return false; 6617#endif 6618} 6619 6620static int process_backlog(struct napi_struct *napi, int quota) 6621{ 6622 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6623 bool again = true; 6624 int work = 0; 6625 6626 /* Check if we have pending ipi, its better to send them now, 6627 * not waiting net_rx_action() end. 6628 */ 6629 if (sd_has_rps_ipi_waiting(sd)) { 6630 local_irq_disable(); 6631 net_rps_action_and_irq_enable(sd); 6632 } 6633 6634 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight); 6635 while (again) { 6636 struct sk_buff *skb; 6637 6638 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6639 while ((skb = __skb_dequeue(&sd->process_queue))) { 6640 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6641 rcu_read_lock(); 6642 __netif_receive_skb(skb); 6643 rcu_read_unlock(); 6644 if (++work >= quota) { 6645 rps_input_queue_head_add(sd, work); 6646 return work; 6647 } 6648 6649 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6650 } 6651 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6652 6653 backlog_lock_irq_disable(sd); 6654 if (skb_queue_empty(&sd->input_pkt_queue)) { 6655 /* 6656 * Inline a custom version of __napi_complete(). 6657 * only current cpu owns and manipulates this napi, 6658 * and NAPI_STATE_SCHED is the only possible flag set 6659 * on backlog. 6660 * We can use a plain write instead of clear_bit(), 6661 * and we dont need an smp_mb() memory barrier. 6662 */ 6663 napi->state &= NAPIF_STATE_THREADED; 6664 again = false; 6665 } else { 6666 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6667 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6668 &sd->process_queue); 6669 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6670 } 6671 backlog_unlock_irq_enable(sd); 6672 } 6673 6674 if (work) 6675 rps_input_queue_head_add(sd, work); 6676 return work; 6677} 6678 6679/** 6680 * __napi_schedule - schedule for receive 6681 * @n: entry to schedule 6682 * 6683 * The entry's receive function will be scheduled to run. 6684 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6685 */ 6686void __napi_schedule(struct napi_struct *n) 6687{ 6688 unsigned long flags; 6689 6690 local_irq_save(flags); 6691 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6692 local_irq_restore(flags); 6693} 6694EXPORT_SYMBOL(__napi_schedule); 6695 6696/** 6697 * napi_schedule_prep - check if napi can be scheduled 6698 * @n: napi context 6699 * 6700 * Test if NAPI routine is already running, and if not mark 6701 * it as running. This is used as a condition variable to 6702 * insure only one NAPI poll instance runs. We also make 6703 * sure there is no pending NAPI disable. 6704 */ 6705bool napi_schedule_prep(struct napi_struct *n) 6706{ 6707 unsigned long new, val = READ_ONCE(n->state); 6708 6709 do { 6710 if (unlikely(val & NAPIF_STATE_DISABLE)) 6711 return false; 6712 new = val | NAPIF_STATE_SCHED; 6713 6714 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6715 * This was suggested by Alexander Duyck, as compiler 6716 * emits better code than : 6717 * if (val & NAPIF_STATE_SCHED) 6718 * new |= NAPIF_STATE_MISSED; 6719 */ 6720 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6721 NAPIF_STATE_MISSED; 6722 } while (!try_cmpxchg(&n->state, &val, new)); 6723 6724 return !(val & NAPIF_STATE_SCHED); 6725} 6726EXPORT_SYMBOL(napi_schedule_prep); 6727 6728/** 6729 * __napi_schedule_irqoff - schedule for receive 6730 * @n: entry to schedule 6731 * 6732 * Variant of __napi_schedule() assuming hard irqs are masked. 6733 * 6734 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6735 * because the interrupt disabled assumption might not be true 6736 * due to force-threaded interrupts and spinlock substitution. 6737 */ 6738void __napi_schedule_irqoff(struct napi_struct *n) 6739{ 6740 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6741 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6742 else 6743 __napi_schedule(n); 6744} 6745EXPORT_SYMBOL(__napi_schedule_irqoff); 6746 6747bool napi_complete_done(struct napi_struct *n, int work_done) 6748{ 6749 unsigned long flags, val, new, timeout = 0; 6750 bool ret = true; 6751 6752 /* 6753 * 1) Don't let napi dequeue from the cpu poll list 6754 * just in case its running on a different cpu. 6755 * 2) If we are busy polling, do nothing here, we have 6756 * the guarantee we will be called later. 6757 */ 6758 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6759 NAPIF_STATE_IN_BUSY_POLL))) 6760 return false; 6761 6762 if (work_done) { 6763 if (n->gro.bitmask) 6764 timeout = napi_get_gro_flush_timeout(n); 6765 n->defer_hard_irqs_count = napi_get_defer_hard_irqs(n); 6766 } 6767 if (n->defer_hard_irqs_count > 0) { 6768 n->defer_hard_irqs_count--; 6769 timeout = napi_get_gro_flush_timeout(n); 6770 if (timeout) 6771 ret = false; 6772 } 6773 6774 /* 6775 * When the NAPI instance uses a timeout and keeps postponing 6776 * it, we need to bound somehow the time packets are kept in 6777 * the GRO layer. 6778 */ 6779 gro_flush_normal(&n->gro, !!timeout); 6780 6781 if (unlikely(!list_empty(&n->poll_list))) { 6782 /* If n->poll_list is not empty, we need to mask irqs */ 6783 local_irq_save(flags); 6784 list_del_init(&n->poll_list); 6785 local_irq_restore(flags); 6786 } 6787 WRITE_ONCE(n->list_owner, -1); 6788 6789 val = READ_ONCE(n->state); 6790 do { 6791 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6792 6793 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6794 NAPIF_STATE_SCHED_THREADED | 6795 NAPIF_STATE_PREFER_BUSY_POLL); 6796 6797 /* If STATE_MISSED was set, leave STATE_SCHED set, 6798 * because we will call napi->poll() one more time. 6799 * This C code was suggested by Alexander Duyck to help gcc. 6800 */ 6801 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6802 NAPIF_STATE_SCHED; 6803 } while (!try_cmpxchg(&n->state, &val, new)); 6804 6805 if (unlikely(val & NAPIF_STATE_MISSED)) { 6806 __napi_schedule(n); 6807 return false; 6808 } 6809 6810 if (timeout) 6811 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6812 HRTIMER_MODE_REL_PINNED); 6813 return ret; 6814} 6815EXPORT_SYMBOL(napi_complete_done); 6816 6817static void skb_defer_free_flush(void) 6818{ 6819 struct llist_node *free_list; 6820 struct sk_buff *skb, *next; 6821 struct skb_defer_node *sdn; 6822 int node; 6823 6824 for_each_node(node) { 6825 sdn = this_cpu_ptr(net_hotdata.skb_defer_nodes) + node; 6826 6827 if (llist_empty(&sdn->defer_list)) 6828 continue; 6829 atomic_long_set(&sdn->defer_count, 0); 6830 free_list = llist_del_all(&sdn->defer_list); 6831 6832 llist_for_each_entry_safe(skb, next, free_list, ll_node) { 6833 prefetch(next); 6834 napi_consume_skb(skb, 1); 6835 } 6836 } 6837} 6838 6839#if defined(CONFIG_NET_RX_BUSY_POLL) 6840 6841static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6842{ 6843 if (!skip_schedule) { 6844 gro_normal_list(&napi->gro); 6845 __napi_schedule(napi); 6846 return; 6847 } 6848 6849 /* Flush too old packets. If HZ < 1000, flush all packets */ 6850 gro_flush_normal(&napi->gro, HZ >= 1000); 6851 6852 clear_bit(NAPI_STATE_SCHED, &napi->state); 6853} 6854 6855enum { 6856 NAPI_F_PREFER_BUSY_POLL = 1, 6857 NAPI_F_END_ON_RESCHED = 2, 6858}; 6859 6860static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, 6861 unsigned flags, u16 budget) 6862{ 6863 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6864 bool skip_schedule = false; 6865 unsigned long timeout; 6866 int rc; 6867 6868 /* Busy polling means there is a high chance device driver hard irq 6869 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6870 * set in napi_schedule_prep(). 6871 * Since we are about to call napi->poll() once more, we can safely 6872 * clear NAPI_STATE_MISSED. 6873 * 6874 * Note: x86 could use a single "lock and ..." instruction 6875 * to perform these two clear_bit() 6876 */ 6877 clear_bit(NAPI_STATE_MISSED, &napi->state); 6878 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6879 6880 local_bh_disable(); 6881 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6882 6883 if (flags & NAPI_F_PREFER_BUSY_POLL) { 6884 napi->defer_hard_irqs_count = napi_get_defer_hard_irqs(napi); 6885 timeout = napi_get_gro_flush_timeout(napi); 6886 if (napi->defer_hard_irqs_count && timeout) { 6887 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6888 skip_schedule = true; 6889 } 6890 } 6891 6892 /* All we really want here is to re-enable device interrupts. 6893 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6894 */ 6895 rc = napi->poll(napi, budget); 6896 /* We can't gro_normal_list() here, because napi->poll() might have 6897 * rearmed the napi (napi_complete_done()) in which case it could 6898 * already be running on another CPU. 6899 */ 6900 trace_napi_poll(napi, rc, budget); 6901 netpoll_poll_unlock(have_poll_lock); 6902 if (rc == budget) 6903 __busy_poll_stop(napi, skip_schedule); 6904 bpf_net_ctx_clear(bpf_net_ctx); 6905 local_bh_enable(); 6906} 6907 6908static void __napi_busy_loop(unsigned int napi_id, 6909 bool (*loop_end)(void *, unsigned long), 6910 void *loop_end_arg, unsigned flags, u16 budget) 6911{ 6912 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6913 int (*napi_poll)(struct napi_struct *napi, int budget); 6914 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6915 void *have_poll_lock = NULL; 6916 struct napi_struct *napi; 6917 6918 WARN_ON_ONCE(!rcu_read_lock_held()); 6919 6920restart: 6921 napi_poll = NULL; 6922 6923 napi = napi_by_id(napi_id); 6924 if (!napi) 6925 return; 6926 6927 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6928 preempt_disable(); 6929 for (;;) { 6930 int work = 0; 6931 6932 local_bh_disable(); 6933 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6934 if (!napi_poll) { 6935 unsigned long val = READ_ONCE(napi->state); 6936 6937 /* If multiple threads are competing for this napi, 6938 * we avoid dirtying napi->state as much as we can. 6939 */ 6940 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6941 NAPIF_STATE_IN_BUSY_POLL)) { 6942 if (flags & NAPI_F_PREFER_BUSY_POLL) 6943 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6944 goto count; 6945 } 6946 if (cmpxchg(&napi->state, val, 6947 val | NAPIF_STATE_IN_BUSY_POLL | 6948 NAPIF_STATE_SCHED) != val) { 6949 if (flags & NAPI_F_PREFER_BUSY_POLL) 6950 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6951 goto count; 6952 } 6953 have_poll_lock = netpoll_poll_lock(napi); 6954 napi_poll = napi->poll; 6955 } 6956 work = napi_poll(napi, budget); 6957 trace_napi_poll(napi, work, budget); 6958 gro_normal_list(&napi->gro); 6959count: 6960 if (work > 0) 6961 __NET_ADD_STATS(dev_net(napi->dev), 6962 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6963 skb_defer_free_flush(); 6964 bpf_net_ctx_clear(bpf_net_ctx); 6965 local_bh_enable(); 6966 6967 if (!loop_end || loop_end(loop_end_arg, start_time)) 6968 break; 6969 6970 if (unlikely(need_resched())) { 6971 if (flags & NAPI_F_END_ON_RESCHED) 6972 break; 6973 if (napi_poll) 6974 busy_poll_stop(napi, have_poll_lock, flags, budget); 6975 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6976 preempt_enable(); 6977 rcu_read_unlock(); 6978 cond_resched(); 6979 rcu_read_lock(); 6980 if (loop_end(loop_end_arg, start_time)) 6981 return; 6982 goto restart; 6983 } 6984 cpu_relax(); 6985 } 6986 if (napi_poll) 6987 busy_poll_stop(napi, have_poll_lock, flags, budget); 6988 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6989 preempt_enable(); 6990} 6991 6992void napi_busy_loop_rcu(unsigned int napi_id, 6993 bool (*loop_end)(void *, unsigned long), 6994 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6995{ 6996 unsigned flags = NAPI_F_END_ON_RESCHED; 6997 6998 if (prefer_busy_poll) 6999 flags |= NAPI_F_PREFER_BUSY_POLL; 7000 7001 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 7002} 7003 7004void napi_busy_loop(unsigned int napi_id, 7005 bool (*loop_end)(void *, unsigned long), 7006 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 7007{ 7008 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; 7009 7010 rcu_read_lock(); 7011 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 7012 rcu_read_unlock(); 7013} 7014EXPORT_SYMBOL(napi_busy_loop); 7015 7016void napi_suspend_irqs(unsigned int napi_id) 7017{ 7018 struct napi_struct *napi; 7019 7020 rcu_read_lock(); 7021 napi = napi_by_id(napi_id); 7022 if (napi) { 7023 unsigned long timeout = napi_get_irq_suspend_timeout(napi); 7024 7025 if (timeout) 7026 hrtimer_start(&napi->timer, ns_to_ktime(timeout), 7027 HRTIMER_MODE_REL_PINNED); 7028 } 7029 rcu_read_unlock(); 7030} 7031 7032void napi_resume_irqs(unsigned int napi_id) 7033{ 7034 struct napi_struct *napi; 7035 7036 rcu_read_lock(); 7037 napi = napi_by_id(napi_id); 7038 if (napi) { 7039 /* If irq_suspend_timeout is set to 0 between the call to 7040 * napi_suspend_irqs and now, the original value still 7041 * determines the safety timeout as intended and napi_watchdog 7042 * will resume irq processing. 7043 */ 7044 if (napi_get_irq_suspend_timeout(napi)) { 7045 local_bh_disable(); 7046 napi_schedule(napi); 7047 local_bh_enable(); 7048 } 7049 } 7050 rcu_read_unlock(); 7051} 7052 7053#endif /* CONFIG_NET_RX_BUSY_POLL */ 7054 7055static void __napi_hash_add_with_id(struct napi_struct *napi, 7056 unsigned int napi_id) 7057{ 7058 napi->gro.cached_napi_id = napi_id; 7059 7060 WRITE_ONCE(napi->napi_id, napi_id); 7061 hlist_add_head_rcu(&napi->napi_hash_node, 7062 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 7063} 7064 7065static void napi_hash_add_with_id(struct napi_struct *napi, 7066 unsigned int napi_id) 7067{ 7068 unsigned long flags; 7069 7070 spin_lock_irqsave(&napi_hash_lock, flags); 7071 WARN_ON_ONCE(napi_by_id(napi_id)); 7072 __napi_hash_add_with_id(napi, napi_id); 7073 spin_unlock_irqrestore(&napi_hash_lock, flags); 7074} 7075 7076static void napi_hash_add(struct napi_struct *napi) 7077{ 7078 unsigned long flags; 7079 7080 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 7081 return; 7082 7083 spin_lock_irqsave(&napi_hash_lock, flags); 7084 7085 /* 0..NR_CPUS range is reserved for sender_cpu use */ 7086 do { 7087 if (unlikely(!napi_id_valid(++napi_gen_id))) 7088 napi_gen_id = MIN_NAPI_ID; 7089 } while (napi_by_id(napi_gen_id)); 7090 7091 __napi_hash_add_with_id(napi, napi_gen_id); 7092 7093 spin_unlock_irqrestore(&napi_hash_lock, flags); 7094} 7095 7096/* Warning : caller is responsible to make sure rcu grace period 7097 * is respected before freeing memory containing @napi 7098 */ 7099static void napi_hash_del(struct napi_struct *napi) 7100{ 7101 unsigned long flags; 7102 7103 spin_lock_irqsave(&napi_hash_lock, flags); 7104 7105 hlist_del_init_rcu(&napi->napi_hash_node); 7106 7107 spin_unlock_irqrestore(&napi_hash_lock, flags); 7108} 7109 7110static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 7111{ 7112 struct napi_struct *napi; 7113 7114 napi = container_of(timer, struct napi_struct, timer); 7115 7116 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 7117 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 7118 */ 7119 if (!napi_disable_pending(napi) && 7120 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 7121 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 7122 __napi_schedule_irqoff(napi); 7123 } 7124 7125 return HRTIMER_NORESTART; 7126} 7127 7128static void napi_stop_kthread(struct napi_struct *napi) 7129{ 7130 unsigned long val, new; 7131 7132 /* Wait until the napi STATE_THREADED is unset. */ 7133 while (true) { 7134 val = READ_ONCE(napi->state); 7135 7136 /* If napi kthread own this napi or the napi is idle, 7137 * STATE_THREADED can be unset here. 7138 */ 7139 if ((val & NAPIF_STATE_SCHED_THREADED) || 7140 !(val & NAPIF_STATE_SCHED)) { 7141 new = val & (~(NAPIF_STATE_THREADED | 7142 NAPIF_STATE_THREADED_BUSY_POLL)); 7143 } else { 7144 msleep(20); 7145 continue; 7146 } 7147 7148 if (try_cmpxchg(&napi->state, &val, new)) 7149 break; 7150 } 7151 7152 /* Once STATE_THREADED is unset, wait for SCHED_THREADED to be unset by 7153 * the kthread. 7154 */ 7155 while (true) { 7156 if (!test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) 7157 break; 7158 7159 msleep(20); 7160 } 7161 7162 kthread_stop(napi->thread); 7163 napi->thread = NULL; 7164} 7165 7166static void napi_set_threaded_state(struct napi_struct *napi, 7167 enum netdev_napi_threaded threaded_mode) 7168{ 7169 bool threaded = threaded_mode != NETDEV_NAPI_THREADED_DISABLED; 7170 bool busy_poll = threaded_mode == NETDEV_NAPI_THREADED_BUSY_POLL; 7171 7172 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 7173 assign_bit(NAPI_STATE_THREADED_BUSY_POLL, &napi->state, busy_poll); 7174} 7175 7176int napi_set_threaded(struct napi_struct *napi, 7177 enum netdev_napi_threaded threaded) 7178{ 7179 if (threaded) { 7180 if (!napi->thread) { 7181 int err = napi_kthread_create(napi); 7182 7183 if (err) 7184 return err; 7185 } 7186 } 7187 7188 if (napi->config) 7189 napi->config->threaded = threaded; 7190 7191 /* Setting/unsetting threaded mode on a napi might not immediately 7192 * take effect, if the current napi instance is actively being 7193 * polled. In this case, the switch between threaded mode and 7194 * softirq mode will happen in the next round of napi_schedule(). 7195 * This should not cause hiccups/stalls to the live traffic. 7196 */ 7197 if (!threaded && napi->thread) { 7198 napi_stop_kthread(napi); 7199 } else { 7200 /* Make sure kthread is created before THREADED bit is set. */ 7201 smp_mb__before_atomic(); 7202 napi_set_threaded_state(napi, threaded); 7203 } 7204 7205 return 0; 7206} 7207 7208int netif_set_threaded(struct net_device *dev, 7209 enum netdev_napi_threaded threaded) 7210{ 7211 struct napi_struct *napi; 7212 int i, err = 0; 7213 7214 netdev_assert_locked_or_invisible(dev); 7215 7216 if (threaded) { 7217 list_for_each_entry(napi, &dev->napi_list, dev_list) { 7218 if (!napi->thread) { 7219 err = napi_kthread_create(napi); 7220 if (err) { 7221 threaded = NETDEV_NAPI_THREADED_DISABLED; 7222 break; 7223 } 7224 } 7225 } 7226 } 7227 7228 WRITE_ONCE(dev->threaded, threaded); 7229 7230 /* The error should not occur as the kthreads are already created. */ 7231 list_for_each_entry(napi, &dev->napi_list, dev_list) 7232 WARN_ON_ONCE(napi_set_threaded(napi, threaded)); 7233 7234 /* Override the config for all NAPIs even if currently not listed */ 7235 for (i = 0; i < dev->num_napi_configs; i++) 7236 dev->napi_config[i].threaded = threaded; 7237 7238 return err; 7239} 7240 7241/** 7242 * netif_threaded_enable() - enable threaded NAPIs 7243 * @dev: net_device instance 7244 * 7245 * Enable threaded mode for the NAPI instances of the device. This may be useful 7246 * for devices where multiple NAPI instances get scheduled by a single 7247 * interrupt. Threaded NAPI allows moving the NAPI processing to cores other 7248 * than the core where IRQ is mapped. 7249 * 7250 * This function should be called before @dev is registered. 7251 */ 7252void netif_threaded_enable(struct net_device *dev) 7253{ 7254 WARN_ON_ONCE(netif_set_threaded(dev, NETDEV_NAPI_THREADED_ENABLED)); 7255} 7256EXPORT_SYMBOL(netif_threaded_enable); 7257 7258/** 7259 * netif_queue_set_napi - Associate queue with the napi 7260 * @dev: device to which NAPI and queue belong 7261 * @queue_index: Index of queue 7262 * @type: queue type as RX or TX 7263 * @napi: NAPI context, pass NULL to clear previously set NAPI 7264 * 7265 * Set queue with its corresponding napi context. This should be done after 7266 * registering the NAPI handler for the queue-vector and the queues have been 7267 * mapped to the corresponding interrupt vector. 7268 */ 7269void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 7270 enum netdev_queue_type type, struct napi_struct *napi) 7271{ 7272 struct netdev_rx_queue *rxq; 7273 struct netdev_queue *txq; 7274 7275 if (WARN_ON_ONCE(napi && !napi->dev)) 7276 return; 7277 netdev_ops_assert_locked_or_invisible(dev); 7278 7279 switch (type) { 7280 case NETDEV_QUEUE_TYPE_RX: 7281 rxq = __netif_get_rx_queue(dev, queue_index); 7282 rxq->napi = napi; 7283 return; 7284 case NETDEV_QUEUE_TYPE_TX: 7285 txq = netdev_get_tx_queue(dev, queue_index); 7286 txq->napi = napi; 7287 return; 7288 default: 7289 return; 7290 } 7291} 7292EXPORT_SYMBOL(netif_queue_set_napi); 7293 7294static void 7295netif_napi_irq_notify(struct irq_affinity_notify *notify, 7296 const cpumask_t *mask) 7297{ 7298 struct napi_struct *napi = 7299 container_of(notify, struct napi_struct, notify); 7300#ifdef CONFIG_RFS_ACCEL 7301 struct cpu_rmap *rmap = napi->dev->rx_cpu_rmap; 7302 int err; 7303#endif 7304 7305 if (napi->config && napi->dev->irq_affinity_auto) 7306 cpumask_copy(&napi->config->affinity_mask, mask); 7307 7308#ifdef CONFIG_RFS_ACCEL 7309 if (napi->dev->rx_cpu_rmap_auto) { 7310 err = cpu_rmap_update(rmap, napi->napi_rmap_idx, mask); 7311 if (err) 7312 netdev_warn(napi->dev, "RMAP update failed (%d)\n", 7313 err); 7314 } 7315#endif 7316} 7317 7318#ifdef CONFIG_RFS_ACCEL 7319static void netif_napi_affinity_release(struct kref *ref) 7320{ 7321 struct napi_struct *napi = 7322 container_of(ref, struct napi_struct, notify.kref); 7323 struct cpu_rmap *rmap = napi->dev->rx_cpu_rmap; 7324 7325 netdev_assert_locked(napi->dev); 7326 WARN_ON(test_and_clear_bit(NAPI_STATE_HAS_NOTIFIER, 7327 &napi->state)); 7328 7329 if (!napi->dev->rx_cpu_rmap_auto) 7330 return; 7331 rmap->obj[napi->napi_rmap_idx] = NULL; 7332 napi->napi_rmap_idx = -1; 7333 cpu_rmap_put(rmap); 7334} 7335 7336int netif_enable_cpu_rmap(struct net_device *dev, unsigned int num_irqs) 7337{ 7338 if (dev->rx_cpu_rmap_auto) 7339 return 0; 7340 7341 dev->rx_cpu_rmap = alloc_irq_cpu_rmap(num_irqs); 7342 if (!dev->rx_cpu_rmap) 7343 return -ENOMEM; 7344 7345 dev->rx_cpu_rmap_auto = true; 7346 return 0; 7347} 7348EXPORT_SYMBOL(netif_enable_cpu_rmap); 7349 7350static void netif_del_cpu_rmap(struct net_device *dev) 7351{ 7352 struct cpu_rmap *rmap = dev->rx_cpu_rmap; 7353 7354 if (!dev->rx_cpu_rmap_auto) 7355 return; 7356 7357 /* Free the rmap */ 7358 cpu_rmap_put(rmap); 7359 dev->rx_cpu_rmap = NULL; 7360 dev->rx_cpu_rmap_auto = false; 7361} 7362 7363#else 7364static void netif_napi_affinity_release(struct kref *ref) 7365{ 7366} 7367 7368int netif_enable_cpu_rmap(struct net_device *dev, unsigned int num_irqs) 7369{ 7370 return 0; 7371} 7372EXPORT_SYMBOL(netif_enable_cpu_rmap); 7373 7374static void netif_del_cpu_rmap(struct net_device *dev) 7375{ 7376} 7377#endif 7378 7379void netif_set_affinity_auto(struct net_device *dev) 7380{ 7381 unsigned int i, maxqs, numa; 7382 7383 maxqs = max(dev->num_tx_queues, dev->num_rx_queues); 7384 numa = dev_to_node(&dev->dev); 7385 7386 for (i = 0; i < maxqs; i++) 7387 cpumask_set_cpu(cpumask_local_spread(i, numa), 7388 &dev->napi_config[i].affinity_mask); 7389 7390 dev->irq_affinity_auto = true; 7391} 7392EXPORT_SYMBOL(netif_set_affinity_auto); 7393 7394void netif_napi_set_irq_locked(struct napi_struct *napi, int irq) 7395{ 7396 int rc; 7397 7398 netdev_assert_locked_or_invisible(napi->dev); 7399 7400 if (napi->irq == irq) 7401 return; 7402 7403 /* Remove existing resources */ 7404 if (test_and_clear_bit(NAPI_STATE_HAS_NOTIFIER, &napi->state)) 7405 irq_set_affinity_notifier(napi->irq, NULL); 7406 7407 napi->irq = irq; 7408 if (irq < 0 || 7409 (!napi->dev->rx_cpu_rmap_auto && !napi->dev->irq_affinity_auto)) 7410 return; 7411 7412 /* Abort for buggy drivers */ 7413 if (napi->dev->irq_affinity_auto && WARN_ON_ONCE(!napi->config)) 7414 return; 7415 7416#ifdef CONFIG_RFS_ACCEL 7417 if (napi->dev->rx_cpu_rmap_auto) { 7418 rc = cpu_rmap_add(napi->dev->rx_cpu_rmap, napi); 7419 if (rc < 0) 7420 return; 7421 7422 cpu_rmap_get(napi->dev->rx_cpu_rmap); 7423 napi->napi_rmap_idx = rc; 7424 } 7425#endif 7426 7427 /* Use core IRQ notifier */ 7428 napi->notify.notify = netif_napi_irq_notify; 7429 napi->notify.release = netif_napi_affinity_release; 7430 rc = irq_set_affinity_notifier(irq, &napi->notify); 7431 if (rc) { 7432 netdev_warn(napi->dev, "Unable to set IRQ notifier (%d)\n", 7433 rc); 7434 goto put_rmap; 7435 } 7436 7437 set_bit(NAPI_STATE_HAS_NOTIFIER, &napi->state); 7438 return; 7439 7440put_rmap: 7441#ifdef CONFIG_RFS_ACCEL 7442 if (napi->dev->rx_cpu_rmap_auto) { 7443 napi->dev->rx_cpu_rmap->obj[napi->napi_rmap_idx] = NULL; 7444 cpu_rmap_put(napi->dev->rx_cpu_rmap); 7445 napi->napi_rmap_idx = -1; 7446 } 7447#endif 7448 napi->notify.notify = NULL; 7449 napi->notify.release = NULL; 7450} 7451EXPORT_SYMBOL(netif_napi_set_irq_locked); 7452 7453static void napi_restore_config(struct napi_struct *n) 7454{ 7455 n->defer_hard_irqs = n->config->defer_hard_irqs; 7456 n->gro_flush_timeout = n->config->gro_flush_timeout; 7457 n->irq_suspend_timeout = n->config->irq_suspend_timeout; 7458 7459 if (n->dev->irq_affinity_auto && 7460 test_bit(NAPI_STATE_HAS_NOTIFIER, &n->state)) 7461 irq_set_affinity(n->irq, &n->config->affinity_mask); 7462 7463 /* a NAPI ID might be stored in the config, if so use it. if not, use 7464 * napi_hash_add to generate one for us. 7465 */ 7466 if (n->config->napi_id) { 7467 napi_hash_add_with_id(n, n->config->napi_id); 7468 } else { 7469 napi_hash_add(n); 7470 n->config->napi_id = n->napi_id; 7471 } 7472 7473 WARN_ON_ONCE(napi_set_threaded(n, n->config->threaded)); 7474} 7475 7476static void napi_save_config(struct napi_struct *n) 7477{ 7478 n->config->defer_hard_irqs = n->defer_hard_irqs; 7479 n->config->gro_flush_timeout = n->gro_flush_timeout; 7480 n->config->irq_suspend_timeout = n->irq_suspend_timeout; 7481 napi_hash_del(n); 7482} 7483 7484/* Netlink wants the NAPI list to be sorted by ID, if adding a NAPI which will 7485 * inherit an existing ID try to insert it at the right position. 7486 */ 7487static void 7488netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) 7489{ 7490 unsigned int new_id, pos_id; 7491 struct list_head *higher; 7492 struct napi_struct *pos; 7493 7494 new_id = UINT_MAX; 7495 if (napi->config && napi->config->napi_id) 7496 new_id = napi->config->napi_id; 7497 7498 higher = &dev->napi_list; 7499 list_for_each_entry(pos, &dev->napi_list, dev_list) { 7500 if (napi_id_valid(pos->napi_id)) 7501 pos_id = pos->napi_id; 7502 else if (pos->config) 7503 pos_id = pos->config->napi_id; 7504 else 7505 pos_id = UINT_MAX; 7506 7507 if (pos_id <= new_id) 7508 break; 7509 higher = &pos->dev_list; 7510 } 7511 list_add_rcu(&napi->dev_list, higher); /* adds after higher */ 7512} 7513 7514/* Double check that napi_get_frags() allocates skbs with 7515 * skb->head being backed by slab, not a page fragment. 7516 * This is to make sure bug fixed in 3226b158e67c 7517 * ("net: avoid 32 x truesize under-estimation for tiny skbs") 7518 * does not accidentally come back. 7519 */ 7520static void napi_get_frags_check(struct napi_struct *napi) 7521{ 7522 struct sk_buff *skb; 7523 7524 local_bh_disable(); 7525 skb = napi_get_frags(napi); 7526 WARN_ON_ONCE(skb && skb->head_frag); 7527 napi_free_frags(napi); 7528 local_bh_enable(); 7529} 7530 7531void netif_napi_add_weight_locked(struct net_device *dev, 7532 struct napi_struct *napi, 7533 int (*poll)(struct napi_struct *, int), 7534 int weight) 7535{ 7536 netdev_assert_locked(dev); 7537 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 7538 return; 7539 7540 INIT_LIST_HEAD(&napi->poll_list); 7541 INIT_HLIST_NODE(&napi->napi_hash_node); 7542 hrtimer_setup(&napi->timer, napi_watchdog, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 7543 gro_init(&napi->gro); 7544 napi->skb = NULL; 7545 napi->poll = poll; 7546 if (weight > NAPI_POLL_WEIGHT) 7547 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 7548 weight); 7549 napi->weight = weight; 7550 napi->dev = dev; 7551#ifdef CONFIG_NETPOLL 7552 napi->poll_owner = -1; 7553#endif 7554 napi->list_owner = -1; 7555 set_bit(NAPI_STATE_SCHED, &napi->state); 7556 set_bit(NAPI_STATE_NPSVC, &napi->state); 7557 netif_napi_dev_list_add(dev, napi); 7558 7559 /* default settings from sysfs are applied to all NAPIs. any per-NAPI 7560 * configuration will be loaded in napi_enable 7561 */ 7562 napi_set_defer_hard_irqs(napi, READ_ONCE(dev->napi_defer_hard_irqs)); 7563 napi_set_gro_flush_timeout(napi, READ_ONCE(dev->gro_flush_timeout)); 7564 7565 napi_get_frags_check(napi); 7566 /* Create kthread for this napi if dev->threaded is set. 7567 * Clear dev->threaded if kthread creation failed so that 7568 * threaded mode will not be enabled in napi_enable(). 7569 */ 7570 if (napi_get_threaded_config(dev, napi)) 7571 if (napi_kthread_create(napi)) 7572 dev->threaded = NETDEV_NAPI_THREADED_DISABLED; 7573 netif_napi_set_irq_locked(napi, -1); 7574} 7575EXPORT_SYMBOL(netif_napi_add_weight_locked); 7576 7577void napi_disable_locked(struct napi_struct *n) 7578{ 7579 unsigned long val, new; 7580 7581 might_sleep(); 7582 netdev_assert_locked(n->dev); 7583 7584 set_bit(NAPI_STATE_DISABLE, &n->state); 7585 7586 val = READ_ONCE(n->state); 7587 do { 7588 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 7589 usleep_range(20, 200); 7590 val = READ_ONCE(n->state); 7591 } 7592 7593 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 7594 new &= ~(NAPIF_STATE_THREADED | 7595 NAPIF_STATE_THREADED_BUSY_POLL | 7596 NAPIF_STATE_PREFER_BUSY_POLL); 7597 } while (!try_cmpxchg(&n->state, &val, new)); 7598 7599 hrtimer_cancel(&n->timer); 7600 7601 if (n->config) 7602 napi_save_config(n); 7603 else 7604 napi_hash_del(n); 7605 7606 clear_bit(NAPI_STATE_DISABLE, &n->state); 7607} 7608EXPORT_SYMBOL(napi_disable_locked); 7609 7610/** 7611 * napi_disable() - prevent NAPI from scheduling 7612 * @n: NAPI context 7613 * 7614 * Stop NAPI from being scheduled on this context. 7615 * Waits till any outstanding processing completes. 7616 * Takes netdev_lock() for associated net_device. 7617 */ 7618void napi_disable(struct napi_struct *n) 7619{ 7620 netdev_lock(n->dev); 7621 napi_disable_locked(n); 7622 netdev_unlock(n->dev); 7623} 7624EXPORT_SYMBOL(napi_disable); 7625 7626void napi_enable_locked(struct napi_struct *n) 7627{ 7628 unsigned long new, val = READ_ONCE(n->state); 7629 7630 if (n->config) 7631 napi_restore_config(n); 7632 else 7633 napi_hash_add(n); 7634 7635 do { 7636 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 7637 7638 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 7639 if (n->dev->threaded && n->thread) 7640 new |= NAPIF_STATE_THREADED; 7641 } while (!try_cmpxchg(&n->state, &val, new)); 7642} 7643EXPORT_SYMBOL(napi_enable_locked); 7644 7645/** 7646 * napi_enable() - enable NAPI scheduling 7647 * @n: NAPI context 7648 * 7649 * Enable scheduling of a NAPI instance. 7650 * Must be paired with napi_disable(). 7651 * Takes netdev_lock() for associated net_device. 7652 */ 7653void napi_enable(struct napi_struct *n) 7654{ 7655 netdev_lock(n->dev); 7656 napi_enable_locked(n); 7657 netdev_unlock(n->dev); 7658} 7659EXPORT_SYMBOL(napi_enable); 7660 7661/* Must be called in process context */ 7662void __netif_napi_del_locked(struct napi_struct *napi) 7663{ 7664 netdev_assert_locked(napi->dev); 7665 7666 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 7667 return; 7668 7669 /* Make sure NAPI is disabled (or was never enabled). */ 7670 WARN_ON(!test_bit(NAPI_STATE_SCHED, &napi->state)); 7671 7672 if (test_and_clear_bit(NAPI_STATE_HAS_NOTIFIER, &napi->state)) 7673 irq_set_affinity_notifier(napi->irq, NULL); 7674 7675 if (napi->config) { 7676 napi->index = -1; 7677 napi->config = NULL; 7678 } 7679 7680 list_del_rcu(&napi->dev_list); 7681 napi_free_frags(napi); 7682 7683 gro_cleanup(&napi->gro); 7684 7685 if (napi->thread) { 7686 kthread_stop(napi->thread); 7687 napi->thread = NULL; 7688 } 7689} 7690EXPORT_SYMBOL(__netif_napi_del_locked); 7691 7692static int __napi_poll(struct napi_struct *n, bool *repoll) 7693{ 7694 int work, weight; 7695 7696 weight = n->weight; 7697 7698 /* This NAPI_STATE_SCHED test is for avoiding a race 7699 * with netpoll's poll_napi(). Only the entity which 7700 * obtains the lock and sees NAPI_STATE_SCHED set will 7701 * actually make the ->poll() call. Therefore we avoid 7702 * accidentally calling ->poll() when NAPI is not scheduled. 7703 */ 7704 work = 0; 7705 if (napi_is_scheduled(n)) { 7706 work = n->poll(n, weight); 7707 trace_napi_poll(n, work, weight); 7708 7709 xdp_do_check_flushed(n); 7710 } 7711 7712 if (unlikely(work > weight)) 7713 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 7714 n->poll, work, weight); 7715 7716 if (likely(work < weight)) 7717 return work; 7718 7719 /* Drivers must not modify the NAPI state if they 7720 * consume the entire weight. In such cases this code 7721 * still "owns" the NAPI instance and therefore can 7722 * move the instance around on the list at-will. 7723 */ 7724 if (unlikely(napi_disable_pending(n))) { 7725 napi_complete(n); 7726 return work; 7727 } 7728 7729 /* The NAPI context has more processing work, but busy-polling 7730 * is preferred. Exit early. 7731 */ 7732 if (napi_prefer_busy_poll(n)) { 7733 if (napi_complete_done(n, work)) { 7734 /* If timeout is not set, we need to make sure 7735 * that the NAPI is re-scheduled. 7736 */ 7737 napi_schedule(n); 7738 } 7739 return work; 7740 } 7741 7742 /* Flush too old packets. If HZ < 1000, flush all packets */ 7743 gro_flush_normal(&n->gro, HZ >= 1000); 7744 7745 /* Some drivers may have called napi_schedule 7746 * prior to exhausting their budget. 7747 */ 7748 if (unlikely(!list_empty(&n->poll_list))) { 7749 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7750 n->dev ? n->dev->name : "backlog"); 7751 return work; 7752 } 7753 7754 *repoll = true; 7755 7756 return work; 7757} 7758 7759static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7760{ 7761 bool do_repoll = false; 7762 void *have; 7763 int work; 7764 7765 list_del_init(&n->poll_list); 7766 7767 have = netpoll_poll_lock(n); 7768 7769 work = __napi_poll(n, &do_repoll); 7770 7771 if (do_repoll) { 7772#if defined(CONFIG_DEBUG_NET) 7773 if (unlikely(!napi_is_scheduled(n))) 7774 pr_crit("repoll requested for device %s %ps but napi is not scheduled.\n", 7775 n->dev->name, n->poll); 7776#endif 7777 list_add_tail(&n->poll_list, repoll); 7778 } 7779 netpoll_poll_unlock(have); 7780 7781 return work; 7782} 7783 7784static int napi_thread_wait(struct napi_struct *napi) 7785{ 7786 set_current_state(TASK_INTERRUPTIBLE); 7787 7788 while (!kthread_should_stop()) { 7789 /* Testing SCHED_THREADED bit here to make sure the current 7790 * kthread owns this napi and could poll on this napi. 7791 * Testing SCHED bit is not enough because SCHED bit might be 7792 * set by some other busy poll thread or by napi_disable(). 7793 */ 7794 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) { 7795 WARN_ON(!list_empty(&napi->poll_list)); 7796 __set_current_state(TASK_RUNNING); 7797 return 0; 7798 } 7799 7800 schedule(); 7801 set_current_state(TASK_INTERRUPTIBLE); 7802 } 7803 __set_current_state(TASK_RUNNING); 7804 7805 return -1; 7806} 7807 7808static void napi_threaded_poll_loop(struct napi_struct *napi, 7809 unsigned long *busy_poll_last_qs) 7810{ 7811 unsigned long last_qs = busy_poll_last_qs ? *busy_poll_last_qs : jiffies; 7812 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7813 struct softnet_data *sd; 7814 7815 for (;;) { 7816 bool repoll = false; 7817 void *have; 7818 7819 local_bh_disable(); 7820 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7821 7822 sd = this_cpu_ptr(&softnet_data); 7823 sd->in_napi_threaded_poll = true; 7824 7825 have = netpoll_poll_lock(napi); 7826 __napi_poll(napi, &repoll); 7827 netpoll_poll_unlock(have); 7828 7829 sd->in_napi_threaded_poll = false; 7830 barrier(); 7831 7832 if (sd_has_rps_ipi_waiting(sd)) { 7833 local_irq_disable(); 7834 net_rps_action_and_irq_enable(sd); 7835 } 7836 skb_defer_free_flush(); 7837 bpf_net_ctx_clear(bpf_net_ctx); 7838 7839 /* When busy poll is enabled, the old packets are not flushed in 7840 * napi_complete_done. So flush them here. 7841 */ 7842 if (busy_poll_last_qs) 7843 gro_flush_normal(&napi->gro, HZ >= 1000); 7844 local_bh_enable(); 7845 7846 /* Call cond_resched here to avoid watchdog warnings. */ 7847 if (repoll || busy_poll_last_qs) { 7848 rcu_softirq_qs_periodic(last_qs); 7849 cond_resched(); 7850 } 7851 7852 if (!repoll) 7853 break; 7854 } 7855 7856 if (busy_poll_last_qs) 7857 *busy_poll_last_qs = last_qs; 7858} 7859 7860static int napi_threaded_poll(void *data) 7861{ 7862 struct napi_struct *napi = data; 7863 unsigned long last_qs = jiffies; 7864 bool want_busy_poll; 7865 bool in_busy_poll; 7866 unsigned long val; 7867 7868 while (!napi_thread_wait(napi)) { 7869 val = READ_ONCE(napi->state); 7870 7871 want_busy_poll = val & NAPIF_STATE_THREADED_BUSY_POLL; 7872 in_busy_poll = val & NAPIF_STATE_IN_BUSY_POLL; 7873 7874 if (unlikely(val & NAPIF_STATE_DISABLE)) 7875 want_busy_poll = false; 7876 7877 if (want_busy_poll != in_busy_poll) 7878 assign_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state, 7879 want_busy_poll); 7880 7881 napi_threaded_poll_loop(napi, want_busy_poll ? &last_qs : NULL); 7882 } 7883 7884 return 0; 7885} 7886 7887static __latent_entropy void net_rx_action(void) 7888{ 7889 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7890 unsigned long time_limit = jiffies + 7891 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs)); 7892 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7893 int budget = READ_ONCE(net_hotdata.netdev_budget); 7894 LIST_HEAD(list); 7895 LIST_HEAD(repoll); 7896 7897 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7898start: 7899 sd->in_net_rx_action = true; 7900 local_irq_disable(); 7901 list_splice_init(&sd->poll_list, &list); 7902 local_irq_enable(); 7903 7904 for (;;) { 7905 struct napi_struct *n; 7906 7907 skb_defer_free_flush(); 7908 7909 if (list_empty(&list)) { 7910 if (list_empty(&repoll)) { 7911 sd->in_net_rx_action = false; 7912 barrier(); 7913 /* We need to check if ____napi_schedule() 7914 * had refilled poll_list while 7915 * sd->in_net_rx_action was true. 7916 */ 7917 if (!list_empty(&sd->poll_list)) 7918 goto start; 7919 if (!sd_has_rps_ipi_waiting(sd)) 7920 goto end; 7921 } 7922 break; 7923 } 7924 7925 n = list_first_entry(&list, struct napi_struct, poll_list); 7926 budget -= napi_poll(n, &repoll); 7927 7928 /* If softirq window is exhausted then punt. 7929 * Allow this to run for 2 jiffies since which will allow 7930 * an average latency of 1.5/HZ. 7931 */ 7932 if (unlikely(budget <= 0 || 7933 time_after_eq(jiffies, time_limit))) { 7934 /* Pairs with READ_ONCE() in softnet_seq_show() */ 7935 WRITE_ONCE(sd->time_squeeze, sd->time_squeeze + 1); 7936 break; 7937 } 7938 } 7939 7940 local_irq_disable(); 7941 7942 list_splice_tail_init(&sd->poll_list, &list); 7943 list_splice_tail(&repoll, &list); 7944 list_splice(&list, &sd->poll_list); 7945 if (!list_empty(&sd->poll_list)) 7946 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7947 else 7948 sd->in_net_rx_action = false; 7949 7950 net_rps_action_and_irq_enable(sd); 7951end: 7952 bpf_net_ctx_clear(bpf_net_ctx); 7953} 7954 7955struct netdev_adjacent { 7956 struct net_device *dev; 7957 netdevice_tracker dev_tracker; 7958 7959 /* upper master flag, there can only be one master device per list */ 7960 bool master; 7961 7962 /* lookup ignore flag */ 7963 bool ignore; 7964 7965 /* counter for the number of times this device was added to us */ 7966 u16 ref_nr; 7967 7968 /* private field for the users */ 7969 void *private; 7970 7971 struct list_head list; 7972 struct rcu_head rcu; 7973}; 7974 7975static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7976 struct list_head *adj_list) 7977{ 7978 struct netdev_adjacent *adj; 7979 7980 list_for_each_entry(adj, adj_list, list) { 7981 if (adj->dev == adj_dev) 7982 return adj; 7983 } 7984 return NULL; 7985} 7986 7987static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7988 struct netdev_nested_priv *priv) 7989{ 7990 struct net_device *dev = (struct net_device *)priv->data; 7991 7992 return upper_dev == dev; 7993} 7994 7995/** 7996 * netdev_has_upper_dev - Check if device is linked to an upper device 7997 * @dev: device 7998 * @upper_dev: upper device to check 7999 * 8000 * Find out if a device is linked to specified upper device and return true 8001 * in case it is. Note that this checks only immediate upper device, 8002 * not through a complete stack of devices. The caller must hold the RTNL lock. 8003 */ 8004bool netdev_has_upper_dev(struct net_device *dev, 8005 struct net_device *upper_dev) 8006{ 8007 struct netdev_nested_priv priv = { 8008 .data = (void *)upper_dev, 8009 }; 8010 8011 ASSERT_RTNL(); 8012 8013 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 8014 &priv); 8015} 8016EXPORT_SYMBOL(netdev_has_upper_dev); 8017 8018/** 8019 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 8020 * @dev: device 8021 * @upper_dev: upper device to check 8022 * 8023 * Find out if a device is linked to specified upper device and return true 8024 * in case it is. Note that this checks the entire upper device chain. 8025 * The caller must hold rcu lock. 8026 */ 8027 8028bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 8029 struct net_device *upper_dev) 8030{ 8031 struct netdev_nested_priv priv = { 8032 .data = (void *)upper_dev, 8033 }; 8034 8035 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 8036 &priv); 8037} 8038EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 8039 8040/** 8041 * netdev_has_any_upper_dev - Check if device is linked to some device 8042 * @dev: device 8043 * 8044 * Find out if a device is linked to an upper device and return true in case 8045 * it is. The caller must hold the RTNL lock. 8046 */ 8047bool netdev_has_any_upper_dev(struct net_device *dev) 8048{ 8049 ASSERT_RTNL(); 8050 8051 return !list_empty(&dev->adj_list.upper); 8052} 8053EXPORT_SYMBOL(netdev_has_any_upper_dev); 8054 8055/** 8056 * netdev_master_upper_dev_get - Get master upper device 8057 * @dev: device 8058 * 8059 * Find a master upper device and return pointer to it or NULL in case 8060 * it's not there. The caller must hold the RTNL lock. 8061 */ 8062struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 8063{ 8064 struct netdev_adjacent *upper; 8065 8066 ASSERT_RTNL(); 8067 8068 if (list_empty(&dev->adj_list.upper)) 8069 return NULL; 8070 8071 upper = list_first_entry(&dev->adj_list.upper, 8072 struct netdev_adjacent, list); 8073 if (likely(upper->master)) 8074 return upper->dev; 8075 return NULL; 8076} 8077EXPORT_SYMBOL(netdev_master_upper_dev_get); 8078 8079static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 8080{ 8081 struct netdev_adjacent *upper; 8082 8083 ASSERT_RTNL(); 8084 8085 if (list_empty(&dev->adj_list.upper)) 8086 return NULL; 8087 8088 upper = list_first_entry(&dev->adj_list.upper, 8089 struct netdev_adjacent, list); 8090 if (likely(upper->master) && !upper->ignore) 8091 return upper->dev; 8092 return NULL; 8093} 8094 8095/** 8096 * netdev_has_any_lower_dev - Check if device is linked to some device 8097 * @dev: device 8098 * 8099 * Find out if a device is linked to a lower device and return true in case 8100 * it is. The caller must hold the RTNL lock. 8101 */ 8102static bool netdev_has_any_lower_dev(struct net_device *dev) 8103{ 8104 ASSERT_RTNL(); 8105 8106 return !list_empty(&dev->adj_list.lower); 8107} 8108 8109void *netdev_adjacent_get_private(struct list_head *adj_list) 8110{ 8111 struct netdev_adjacent *adj; 8112 8113 adj = list_entry(adj_list, struct netdev_adjacent, list); 8114 8115 return adj->private; 8116} 8117EXPORT_SYMBOL(netdev_adjacent_get_private); 8118 8119/** 8120 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 8121 * @dev: device 8122 * @iter: list_head ** of the current position 8123 * 8124 * Gets the next device from the dev's upper list, starting from iter 8125 * position. The caller must hold RCU read lock. 8126 */ 8127struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 8128 struct list_head **iter) 8129{ 8130 struct netdev_adjacent *upper; 8131 8132 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 8133 8134 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 8135 8136 if (&upper->list == &dev->adj_list.upper) 8137 return NULL; 8138 8139 *iter = &upper->list; 8140 8141 return upper->dev; 8142} 8143EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 8144 8145static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 8146 struct list_head **iter, 8147 bool *ignore) 8148{ 8149 struct netdev_adjacent *upper; 8150 8151 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 8152 8153 if (&upper->list == &dev->adj_list.upper) 8154 return NULL; 8155 8156 *iter = &upper->list; 8157 *ignore = upper->ignore; 8158 8159 return upper->dev; 8160} 8161 8162static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 8163 struct list_head **iter) 8164{ 8165 struct netdev_adjacent *upper; 8166 8167 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 8168 8169 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 8170 8171 if (&upper->list == &dev->adj_list.upper) 8172 return NULL; 8173 8174 *iter = &upper->list; 8175 8176 return upper->dev; 8177} 8178 8179static int __netdev_walk_all_upper_dev(struct net_device *dev, 8180 int (*fn)(struct net_device *dev, 8181 struct netdev_nested_priv *priv), 8182 struct netdev_nested_priv *priv) 8183{ 8184 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 8185 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 8186 int ret, cur = 0; 8187 bool ignore; 8188 8189 now = dev; 8190 iter = &dev->adj_list.upper; 8191 8192 while (1) { 8193 if (now != dev) { 8194 ret = fn(now, priv); 8195 if (ret) 8196 return ret; 8197 } 8198 8199 next = NULL; 8200 while (1) { 8201 udev = __netdev_next_upper_dev(now, &iter, &ignore); 8202 if (!udev) 8203 break; 8204 if (ignore) 8205 continue; 8206 8207 next = udev; 8208 niter = &udev->adj_list.upper; 8209 dev_stack[cur] = now; 8210 iter_stack[cur++] = iter; 8211 break; 8212 } 8213 8214 if (!next) { 8215 if (!cur) 8216 return 0; 8217 next = dev_stack[--cur]; 8218 niter = iter_stack[cur]; 8219 } 8220 8221 now = next; 8222 iter = niter; 8223 } 8224 8225 return 0; 8226} 8227 8228int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 8229 int (*fn)(struct net_device *dev, 8230 struct netdev_nested_priv *priv), 8231 struct netdev_nested_priv *priv) 8232{ 8233 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 8234 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 8235 int ret, cur = 0; 8236 8237 now = dev; 8238 iter = &dev->adj_list.upper; 8239 8240 while (1) { 8241 if (now != dev) { 8242 ret = fn(now, priv); 8243 if (ret) 8244 return ret; 8245 } 8246 8247 next = NULL; 8248 while (1) { 8249 udev = netdev_next_upper_dev_rcu(now, &iter); 8250 if (!udev) 8251 break; 8252 8253 next = udev; 8254 niter = &udev->adj_list.upper; 8255 dev_stack[cur] = now; 8256 iter_stack[cur++] = iter; 8257 break; 8258 } 8259 8260 if (!next) { 8261 if (!cur) 8262 return 0; 8263 next = dev_stack[--cur]; 8264 niter = iter_stack[cur]; 8265 } 8266 8267 now = next; 8268 iter = niter; 8269 } 8270 8271 return 0; 8272} 8273EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 8274 8275static bool __netdev_has_upper_dev(struct net_device *dev, 8276 struct net_device *upper_dev) 8277{ 8278 struct netdev_nested_priv priv = { 8279 .flags = 0, 8280 .data = (void *)upper_dev, 8281 }; 8282 8283 ASSERT_RTNL(); 8284 8285 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 8286 &priv); 8287} 8288 8289/** 8290 * netdev_lower_get_next_private - Get the next ->private from the 8291 * lower neighbour list 8292 * @dev: device 8293 * @iter: list_head ** of the current position 8294 * 8295 * Gets the next netdev_adjacent->private from the dev's lower neighbour 8296 * list, starting from iter position. The caller must hold either hold the 8297 * RTNL lock or its own locking that guarantees that the neighbour lower 8298 * list will remain unchanged. 8299 */ 8300void *netdev_lower_get_next_private(struct net_device *dev, 8301 struct list_head **iter) 8302{ 8303 struct netdev_adjacent *lower; 8304 8305 lower = list_entry(*iter, struct netdev_adjacent, list); 8306 8307 if (&lower->list == &dev->adj_list.lower) 8308 return NULL; 8309 8310 *iter = lower->list.next; 8311 8312 return lower->private; 8313} 8314EXPORT_SYMBOL(netdev_lower_get_next_private); 8315 8316/** 8317 * netdev_lower_get_next_private_rcu - Get the next ->private from the 8318 * lower neighbour list, RCU 8319 * variant 8320 * @dev: device 8321 * @iter: list_head ** of the current position 8322 * 8323 * Gets the next netdev_adjacent->private from the dev's lower neighbour 8324 * list, starting from iter position. The caller must hold RCU read lock. 8325 */ 8326void *netdev_lower_get_next_private_rcu(struct net_device *dev, 8327 struct list_head **iter) 8328{ 8329 struct netdev_adjacent *lower; 8330 8331 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 8332 8333 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 8334 8335 if (&lower->list == &dev->adj_list.lower) 8336 return NULL; 8337 8338 *iter = &lower->list; 8339 8340 return lower->private; 8341} 8342EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 8343 8344/** 8345 * netdev_lower_get_next - Get the next device from the lower neighbour 8346 * list 8347 * @dev: device 8348 * @iter: list_head ** of the current position 8349 * 8350 * Gets the next netdev_adjacent from the dev's lower neighbour 8351 * list, starting from iter position. The caller must hold RTNL lock or 8352 * its own locking that guarantees that the neighbour lower 8353 * list will remain unchanged. 8354 */ 8355void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 8356{ 8357 struct netdev_adjacent *lower; 8358 8359 lower = list_entry(*iter, struct netdev_adjacent, list); 8360 8361 if (&lower->list == &dev->adj_list.lower) 8362 return NULL; 8363 8364 *iter = lower->list.next; 8365 8366 return lower->dev; 8367} 8368EXPORT_SYMBOL(netdev_lower_get_next); 8369 8370static struct net_device *netdev_next_lower_dev(struct net_device *dev, 8371 struct list_head **iter) 8372{ 8373 struct netdev_adjacent *lower; 8374 8375 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 8376 8377 if (&lower->list == &dev->adj_list.lower) 8378 return NULL; 8379 8380 *iter = &lower->list; 8381 8382 return lower->dev; 8383} 8384 8385static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 8386 struct list_head **iter, 8387 bool *ignore) 8388{ 8389 struct netdev_adjacent *lower; 8390 8391 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 8392 8393 if (&lower->list == &dev->adj_list.lower) 8394 return NULL; 8395 8396 *iter = &lower->list; 8397 *ignore = lower->ignore; 8398 8399 return lower->dev; 8400} 8401 8402int netdev_walk_all_lower_dev(struct net_device *dev, 8403 int (*fn)(struct net_device *dev, 8404 struct netdev_nested_priv *priv), 8405 struct netdev_nested_priv *priv) 8406{ 8407 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 8408 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 8409 int ret, cur = 0; 8410 8411 now = dev; 8412 iter = &dev->adj_list.lower; 8413 8414 while (1) { 8415 if (now != dev) { 8416 ret = fn(now, priv); 8417 if (ret) 8418 return ret; 8419 } 8420 8421 next = NULL; 8422 while (1) { 8423 ldev = netdev_next_lower_dev(now, &iter); 8424 if (!ldev) 8425 break; 8426 8427 next = ldev; 8428 niter = &ldev->adj_list.lower; 8429 dev_stack[cur] = now; 8430 iter_stack[cur++] = iter; 8431 break; 8432 } 8433 8434 if (!next) { 8435 if (!cur) 8436 return 0; 8437 next = dev_stack[--cur]; 8438 niter = iter_stack[cur]; 8439 } 8440 8441 now = next; 8442 iter = niter; 8443 } 8444 8445 return 0; 8446} 8447EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 8448 8449static int __netdev_walk_all_lower_dev(struct net_device *dev, 8450 int (*fn)(struct net_device *dev, 8451 struct netdev_nested_priv *priv), 8452 struct netdev_nested_priv *priv) 8453{ 8454 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 8455 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 8456 int ret, cur = 0; 8457 bool ignore; 8458 8459 now = dev; 8460 iter = &dev->adj_list.lower; 8461 8462 while (1) { 8463 if (now != dev) { 8464 ret = fn(now, priv); 8465 if (ret) 8466 return ret; 8467 } 8468 8469 next = NULL; 8470 while (1) { 8471 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 8472 if (!ldev) 8473 break; 8474 if (ignore) 8475 continue; 8476 8477 next = ldev; 8478 niter = &ldev->adj_list.lower; 8479 dev_stack[cur] = now; 8480 iter_stack[cur++] = iter; 8481 break; 8482 } 8483 8484 if (!next) { 8485 if (!cur) 8486 return 0; 8487 next = dev_stack[--cur]; 8488 niter = iter_stack[cur]; 8489 } 8490 8491 now = next; 8492 iter = niter; 8493 } 8494 8495 return 0; 8496} 8497 8498struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 8499 struct list_head **iter) 8500{ 8501 struct netdev_adjacent *lower; 8502 8503 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 8504 if (&lower->list == &dev->adj_list.lower) 8505 return NULL; 8506 8507 *iter = &lower->list; 8508 8509 return lower->dev; 8510} 8511EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 8512 8513static u8 __netdev_upper_depth(struct net_device *dev) 8514{ 8515 struct net_device *udev; 8516 struct list_head *iter; 8517 u8 max_depth = 0; 8518 bool ignore; 8519 8520 for (iter = &dev->adj_list.upper, 8521 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 8522 udev; 8523 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 8524 if (ignore) 8525 continue; 8526 if (max_depth < udev->upper_level) 8527 max_depth = udev->upper_level; 8528 } 8529 8530 return max_depth; 8531} 8532 8533static u8 __netdev_lower_depth(struct net_device *dev) 8534{ 8535 struct net_device *ldev; 8536 struct list_head *iter; 8537 u8 max_depth = 0; 8538 bool ignore; 8539 8540 for (iter = &dev->adj_list.lower, 8541 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 8542 ldev; 8543 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 8544 if (ignore) 8545 continue; 8546 if (max_depth < ldev->lower_level) 8547 max_depth = ldev->lower_level; 8548 } 8549 8550 return max_depth; 8551} 8552 8553static int __netdev_update_upper_level(struct net_device *dev, 8554 struct netdev_nested_priv *__unused) 8555{ 8556 dev->upper_level = __netdev_upper_depth(dev) + 1; 8557 return 0; 8558} 8559 8560#ifdef CONFIG_LOCKDEP 8561static LIST_HEAD(net_unlink_list); 8562 8563static void net_unlink_todo(struct net_device *dev) 8564{ 8565 if (list_empty(&dev->unlink_list)) 8566 list_add_tail(&dev->unlink_list, &net_unlink_list); 8567} 8568#endif 8569 8570static int __netdev_update_lower_level(struct net_device *dev, 8571 struct netdev_nested_priv *priv) 8572{ 8573 dev->lower_level = __netdev_lower_depth(dev) + 1; 8574 8575#ifdef CONFIG_LOCKDEP 8576 if (!priv) 8577 return 0; 8578 8579 if (priv->flags & NESTED_SYNC_IMM) 8580 dev->nested_level = dev->lower_level - 1; 8581 if (priv->flags & NESTED_SYNC_TODO) 8582 net_unlink_todo(dev); 8583#endif 8584 return 0; 8585} 8586 8587int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 8588 int (*fn)(struct net_device *dev, 8589 struct netdev_nested_priv *priv), 8590 struct netdev_nested_priv *priv) 8591{ 8592 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 8593 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 8594 int ret, cur = 0; 8595 8596 now = dev; 8597 iter = &dev->adj_list.lower; 8598 8599 while (1) { 8600 if (now != dev) { 8601 ret = fn(now, priv); 8602 if (ret) 8603 return ret; 8604 } 8605 8606 next = NULL; 8607 while (1) { 8608 ldev = netdev_next_lower_dev_rcu(now, &iter); 8609 if (!ldev) 8610 break; 8611 8612 next = ldev; 8613 niter = &ldev->adj_list.lower; 8614 dev_stack[cur] = now; 8615 iter_stack[cur++] = iter; 8616 break; 8617 } 8618 8619 if (!next) { 8620 if (!cur) 8621 return 0; 8622 next = dev_stack[--cur]; 8623 niter = iter_stack[cur]; 8624 } 8625 8626 now = next; 8627 iter = niter; 8628 } 8629 8630 return 0; 8631} 8632EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 8633 8634/** 8635 * netdev_lower_get_first_private_rcu - Get the first ->private from the 8636 * lower neighbour list, RCU 8637 * variant 8638 * @dev: device 8639 * 8640 * Gets the first netdev_adjacent->private from the dev's lower neighbour 8641 * list. The caller must hold RCU read lock. 8642 */ 8643void *netdev_lower_get_first_private_rcu(struct net_device *dev) 8644{ 8645 struct netdev_adjacent *lower; 8646 8647 lower = list_first_or_null_rcu(&dev->adj_list.lower, 8648 struct netdev_adjacent, list); 8649 if (lower) 8650 return lower->private; 8651 return NULL; 8652} 8653EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 8654 8655/** 8656 * netdev_master_upper_dev_get_rcu - Get master upper device 8657 * @dev: device 8658 * 8659 * Find a master upper device and return pointer to it or NULL in case 8660 * it's not there. The caller must hold the RCU read lock. 8661 */ 8662struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 8663{ 8664 struct netdev_adjacent *upper; 8665 8666 upper = list_first_or_null_rcu(&dev->adj_list.upper, 8667 struct netdev_adjacent, list); 8668 if (upper && likely(upper->master)) 8669 return upper->dev; 8670 return NULL; 8671} 8672EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 8673 8674static int netdev_adjacent_sysfs_add(struct net_device *dev, 8675 struct net_device *adj_dev, 8676 struct list_head *dev_list) 8677{ 8678 char linkname[IFNAMSIZ+7]; 8679 8680 sprintf(linkname, dev_list == &dev->adj_list.upper ? 8681 "upper_%s" : "lower_%s", adj_dev->name); 8682 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 8683 linkname); 8684} 8685static void netdev_adjacent_sysfs_del(struct net_device *dev, 8686 char *name, 8687 struct list_head *dev_list) 8688{ 8689 char linkname[IFNAMSIZ+7]; 8690 8691 sprintf(linkname, dev_list == &dev->adj_list.upper ? 8692 "upper_%s" : "lower_%s", name); 8693 sysfs_remove_link(&(dev->dev.kobj), linkname); 8694} 8695 8696static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 8697 struct net_device *adj_dev, 8698 struct list_head *dev_list) 8699{ 8700 return (dev_list == &dev->adj_list.upper || 8701 dev_list == &dev->adj_list.lower) && 8702 net_eq(dev_net(dev), dev_net(adj_dev)); 8703} 8704 8705static int __netdev_adjacent_dev_insert(struct net_device *dev, 8706 struct net_device *adj_dev, 8707 struct list_head *dev_list, 8708 void *private, bool master) 8709{ 8710 struct netdev_adjacent *adj; 8711 int ret; 8712 8713 adj = __netdev_find_adj(adj_dev, dev_list); 8714 8715 if (adj) { 8716 adj->ref_nr += 1; 8717 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 8718 dev->name, adj_dev->name, adj->ref_nr); 8719 8720 return 0; 8721 } 8722 8723 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 8724 if (!adj) 8725 return -ENOMEM; 8726 8727 adj->dev = adj_dev; 8728 adj->master = master; 8729 adj->ref_nr = 1; 8730 adj->private = private; 8731 adj->ignore = false; 8732 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 8733 8734 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 8735 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 8736 8737 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 8738 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 8739 if (ret) 8740 goto free_adj; 8741 } 8742 8743 /* Ensure that master link is always the first item in list. */ 8744 if (master) { 8745 ret = sysfs_create_link(&(dev->dev.kobj), 8746 &(adj_dev->dev.kobj), "master"); 8747 if (ret) 8748 goto remove_symlinks; 8749 8750 list_add_rcu(&adj->list, dev_list); 8751 } else { 8752 list_add_tail_rcu(&adj->list, dev_list); 8753 } 8754 8755 return 0; 8756 8757remove_symlinks: 8758 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8759 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8760free_adj: 8761 netdev_put(adj_dev, &adj->dev_tracker); 8762 kfree(adj); 8763 8764 return ret; 8765} 8766 8767static void __netdev_adjacent_dev_remove(struct net_device *dev, 8768 struct net_device *adj_dev, 8769 u16 ref_nr, 8770 struct list_head *dev_list) 8771{ 8772 struct netdev_adjacent *adj; 8773 8774 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 8775 dev->name, adj_dev->name, ref_nr); 8776 8777 adj = __netdev_find_adj(adj_dev, dev_list); 8778 8779 if (!adj) { 8780 pr_err("Adjacency does not exist for device %s from %s\n", 8781 dev->name, adj_dev->name); 8782 WARN_ON(1); 8783 return; 8784 } 8785 8786 if (adj->ref_nr > ref_nr) { 8787 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8788 dev->name, adj_dev->name, ref_nr, 8789 adj->ref_nr - ref_nr); 8790 adj->ref_nr -= ref_nr; 8791 return; 8792 } 8793 8794 if (adj->master) 8795 sysfs_remove_link(&(dev->dev.kobj), "master"); 8796 8797 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8798 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8799 8800 list_del_rcu(&adj->list); 8801 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8802 adj_dev->name, dev->name, adj_dev->name); 8803 netdev_put(adj_dev, &adj->dev_tracker); 8804 kfree_rcu(adj, rcu); 8805} 8806 8807static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8808 struct net_device *upper_dev, 8809 struct list_head *up_list, 8810 struct list_head *down_list, 8811 void *private, bool master) 8812{ 8813 int ret; 8814 8815 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8816 private, master); 8817 if (ret) 8818 return ret; 8819 8820 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8821 private, false); 8822 if (ret) { 8823 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8824 return ret; 8825 } 8826 8827 return 0; 8828} 8829 8830static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8831 struct net_device *upper_dev, 8832 u16 ref_nr, 8833 struct list_head *up_list, 8834 struct list_head *down_list) 8835{ 8836 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8837 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8838} 8839 8840static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8841 struct net_device *upper_dev, 8842 void *private, bool master) 8843{ 8844 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8845 &dev->adj_list.upper, 8846 &upper_dev->adj_list.lower, 8847 private, master); 8848} 8849 8850static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8851 struct net_device *upper_dev) 8852{ 8853 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8854 &dev->adj_list.upper, 8855 &upper_dev->adj_list.lower); 8856} 8857 8858static int __netdev_upper_dev_link(struct net_device *dev, 8859 struct net_device *upper_dev, bool master, 8860 void *upper_priv, void *upper_info, 8861 struct netdev_nested_priv *priv, 8862 struct netlink_ext_ack *extack) 8863{ 8864 struct netdev_notifier_changeupper_info changeupper_info = { 8865 .info = { 8866 .dev = dev, 8867 .extack = extack, 8868 }, 8869 .upper_dev = upper_dev, 8870 .master = master, 8871 .linking = true, 8872 .upper_info = upper_info, 8873 }; 8874 struct net_device *master_dev; 8875 int ret = 0; 8876 8877 ASSERT_RTNL(); 8878 8879 if (dev == upper_dev) 8880 return -EBUSY; 8881 8882 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8883 if (__netdev_has_upper_dev(upper_dev, dev)) 8884 return -EBUSY; 8885 8886 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8887 return -EMLINK; 8888 8889 if (!master) { 8890 if (__netdev_has_upper_dev(dev, upper_dev)) 8891 return -EEXIST; 8892 } else { 8893 master_dev = __netdev_master_upper_dev_get(dev); 8894 if (master_dev) 8895 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8896 } 8897 8898 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8899 &changeupper_info.info); 8900 ret = notifier_to_errno(ret); 8901 if (ret) 8902 return ret; 8903 8904 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8905 master); 8906 if (ret) 8907 return ret; 8908 8909 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8910 &changeupper_info.info); 8911 ret = notifier_to_errno(ret); 8912 if (ret) 8913 goto rollback; 8914 8915 __netdev_update_upper_level(dev, NULL); 8916 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8917 8918 __netdev_update_lower_level(upper_dev, priv); 8919 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8920 priv); 8921 8922 return 0; 8923 8924rollback: 8925 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8926 8927 return ret; 8928} 8929 8930/** 8931 * netdev_upper_dev_link - Add a link to the upper device 8932 * @dev: device 8933 * @upper_dev: new upper device 8934 * @extack: netlink extended ack 8935 * 8936 * Adds a link to device which is upper to this one. The caller must hold 8937 * the RTNL lock. On a failure a negative errno code is returned. 8938 * On success the reference counts are adjusted and the function 8939 * returns zero. 8940 */ 8941int netdev_upper_dev_link(struct net_device *dev, 8942 struct net_device *upper_dev, 8943 struct netlink_ext_ack *extack) 8944{ 8945 struct netdev_nested_priv priv = { 8946 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8947 .data = NULL, 8948 }; 8949 8950 return __netdev_upper_dev_link(dev, upper_dev, false, 8951 NULL, NULL, &priv, extack); 8952} 8953EXPORT_SYMBOL(netdev_upper_dev_link); 8954 8955/** 8956 * netdev_master_upper_dev_link - Add a master link to the upper device 8957 * @dev: device 8958 * @upper_dev: new upper device 8959 * @upper_priv: upper device private 8960 * @upper_info: upper info to be passed down via notifier 8961 * @extack: netlink extended ack 8962 * 8963 * Adds a link to device which is upper to this one. In this case, only 8964 * one master upper device can be linked, although other non-master devices 8965 * might be linked as well. The caller must hold the RTNL lock. 8966 * On a failure a negative errno code is returned. On success the reference 8967 * counts are adjusted and the function returns zero. 8968 */ 8969int netdev_master_upper_dev_link(struct net_device *dev, 8970 struct net_device *upper_dev, 8971 void *upper_priv, void *upper_info, 8972 struct netlink_ext_ack *extack) 8973{ 8974 struct netdev_nested_priv priv = { 8975 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8976 .data = NULL, 8977 }; 8978 8979 return __netdev_upper_dev_link(dev, upper_dev, true, 8980 upper_priv, upper_info, &priv, extack); 8981} 8982EXPORT_SYMBOL(netdev_master_upper_dev_link); 8983 8984static void __netdev_upper_dev_unlink(struct net_device *dev, 8985 struct net_device *upper_dev, 8986 struct netdev_nested_priv *priv) 8987{ 8988 struct netdev_notifier_changeupper_info changeupper_info = { 8989 .info = { 8990 .dev = dev, 8991 }, 8992 .upper_dev = upper_dev, 8993 .linking = false, 8994 }; 8995 8996 ASSERT_RTNL(); 8997 8998 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8999 9000 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 9001 &changeupper_info.info); 9002 9003 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 9004 9005 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 9006 &changeupper_info.info); 9007 9008 __netdev_update_upper_level(dev, NULL); 9009 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 9010 9011 __netdev_update_lower_level(upper_dev, priv); 9012 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 9013 priv); 9014} 9015 9016/** 9017 * netdev_upper_dev_unlink - Removes a link to upper device 9018 * @dev: device 9019 * @upper_dev: new upper device 9020 * 9021 * Removes a link to device which is upper to this one. The caller must hold 9022 * the RTNL lock. 9023 */ 9024void netdev_upper_dev_unlink(struct net_device *dev, 9025 struct net_device *upper_dev) 9026{ 9027 struct netdev_nested_priv priv = { 9028 .flags = NESTED_SYNC_TODO, 9029 .data = NULL, 9030 }; 9031 9032 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 9033} 9034EXPORT_SYMBOL(netdev_upper_dev_unlink); 9035 9036static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 9037 struct net_device *lower_dev, 9038 bool val) 9039{ 9040 struct netdev_adjacent *adj; 9041 9042 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 9043 if (adj) 9044 adj->ignore = val; 9045 9046 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 9047 if (adj) 9048 adj->ignore = val; 9049} 9050 9051static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 9052 struct net_device *lower_dev) 9053{ 9054 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 9055} 9056 9057static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 9058 struct net_device *lower_dev) 9059{ 9060 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 9061} 9062 9063int netdev_adjacent_change_prepare(struct net_device *old_dev, 9064 struct net_device *new_dev, 9065 struct net_device *dev, 9066 struct netlink_ext_ack *extack) 9067{ 9068 struct netdev_nested_priv priv = { 9069 .flags = 0, 9070 .data = NULL, 9071 }; 9072 int err; 9073 9074 if (!new_dev) 9075 return 0; 9076 9077 if (old_dev && new_dev != old_dev) 9078 netdev_adjacent_dev_disable(dev, old_dev); 9079 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 9080 extack); 9081 if (err) { 9082 if (old_dev && new_dev != old_dev) 9083 netdev_adjacent_dev_enable(dev, old_dev); 9084 return err; 9085 } 9086 9087 return 0; 9088} 9089EXPORT_SYMBOL(netdev_adjacent_change_prepare); 9090 9091void netdev_adjacent_change_commit(struct net_device *old_dev, 9092 struct net_device *new_dev, 9093 struct net_device *dev) 9094{ 9095 struct netdev_nested_priv priv = { 9096 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 9097 .data = NULL, 9098 }; 9099 9100 if (!new_dev || !old_dev) 9101 return; 9102 9103 if (new_dev == old_dev) 9104 return; 9105 9106 netdev_adjacent_dev_enable(dev, old_dev); 9107 __netdev_upper_dev_unlink(old_dev, dev, &priv); 9108} 9109EXPORT_SYMBOL(netdev_adjacent_change_commit); 9110 9111void netdev_adjacent_change_abort(struct net_device *old_dev, 9112 struct net_device *new_dev, 9113 struct net_device *dev) 9114{ 9115 struct netdev_nested_priv priv = { 9116 .flags = 0, 9117 .data = NULL, 9118 }; 9119 9120 if (!new_dev) 9121 return; 9122 9123 if (old_dev && new_dev != old_dev) 9124 netdev_adjacent_dev_enable(dev, old_dev); 9125 9126 __netdev_upper_dev_unlink(new_dev, dev, &priv); 9127} 9128EXPORT_SYMBOL(netdev_adjacent_change_abort); 9129 9130/** 9131 * netdev_bonding_info_change - Dispatch event about slave change 9132 * @dev: device 9133 * @bonding_info: info to dispatch 9134 * 9135 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 9136 * The caller must hold the RTNL lock. 9137 */ 9138void netdev_bonding_info_change(struct net_device *dev, 9139 struct netdev_bonding_info *bonding_info) 9140{ 9141 struct netdev_notifier_bonding_info info = { 9142 .info.dev = dev, 9143 }; 9144 9145 memcpy(&info.bonding_info, bonding_info, 9146 sizeof(struct netdev_bonding_info)); 9147 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 9148 &info.info); 9149} 9150EXPORT_SYMBOL(netdev_bonding_info_change); 9151 9152static int netdev_offload_xstats_enable_l3(struct net_device *dev, 9153 struct netlink_ext_ack *extack) 9154{ 9155 struct netdev_notifier_offload_xstats_info info = { 9156 .info.dev = dev, 9157 .info.extack = extack, 9158 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 9159 }; 9160 int err; 9161 int rc; 9162 9163 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 9164 GFP_KERNEL); 9165 if (!dev->offload_xstats_l3) 9166 return -ENOMEM; 9167 9168 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 9169 NETDEV_OFFLOAD_XSTATS_DISABLE, 9170 &info.info); 9171 err = notifier_to_errno(rc); 9172 if (err) 9173 goto free_stats; 9174 9175 return 0; 9176 9177free_stats: 9178 kfree(dev->offload_xstats_l3); 9179 dev->offload_xstats_l3 = NULL; 9180 return err; 9181} 9182 9183int netdev_offload_xstats_enable(struct net_device *dev, 9184 enum netdev_offload_xstats_type type, 9185 struct netlink_ext_ack *extack) 9186{ 9187 ASSERT_RTNL(); 9188 9189 if (netdev_offload_xstats_enabled(dev, type)) 9190 return -EALREADY; 9191 9192 switch (type) { 9193 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 9194 return netdev_offload_xstats_enable_l3(dev, extack); 9195 } 9196 9197 WARN_ON(1); 9198 return -EINVAL; 9199} 9200EXPORT_SYMBOL(netdev_offload_xstats_enable); 9201 9202static void netdev_offload_xstats_disable_l3(struct net_device *dev) 9203{ 9204 struct netdev_notifier_offload_xstats_info info = { 9205 .info.dev = dev, 9206 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 9207 }; 9208 9209 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 9210 &info.info); 9211 kfree(dev->offload_xstats_l3); 9212 dev->offload_xstats_l3 = NULL; 9213} 9214 9215int netdev_offload_xstats_disable(struct net_device *dev, 9216 enum netdev_offload_xstats_type type) 9217{ 9218 ASSERT_RTNL(); 9219 9220 if (!netdev_offload_xstats_enabled(dev, type)) 9221 return -EALREADY; 9222 9223 switch (type) { 9224 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 9225 netdev_offload_xstats_disable_l3(dev); 9226 return 0; 9227 } 9228 9229 WARN_ON(1); 9230 return -EINVAL; 9231} 9232EXPORT_SYMBOL(netdev_offload_xstats_disable); 9233 9234static void netdev_offload_xstats_disable_all(struct net_device *dev) 9235{ 9236 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 9237} 9238 9239static struct rtnl_hw_stats64 * 9240netdev_offload_xstats_get_ptr(const struct net_device *dev, 9241 enum netdev_offload_xstats_type type) 9242{ 9243 switch (type) { 9244 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 9245 return dev->offload_xstats_l3; 9246 } 9247 9248 WARN_ON(1); 9249 return NULL; 9250} 9251 9252bool netdev_offload_xstats_enabled(const struct net_device *dev, 9253 enum netdev_offload_xstats_type type) 9254{ 9255 ASSERT_RTNL(); 9256 9257 return netdev_offload_xstats_get_ptr(dev, type); 9258} 9259EXPORT_SYMBOL(netdev_offload_xstats_enabled); 9260 9261struct netdev_notifier_offload_xstats_ru { 9262 bool used; 9263}; 9264 9265struct netdev_notifier_offload_xstats_rd { 9266 struct rtnl_hw_stats64 stats; 9267 bool used; 9268}; 9269 9270static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 9271 const struct rtnl_hw_stats64 *src) 9272{ 9273 dest->rx_packets += src->rx_packets; 9274 dest->tx_packets += src->tx_packets; 9275 dest->rx_bytes += src->rx_bytes; 9276 dest->tx_bytes += src->tx_bytes; 9277 dest->rx_errors += src->rx_errors; 9278 dest->tx_errors += src->tx_errors; 9279 dest->rx_dropped += src->rx_dropped; 9280 dest->tx_dropped += src->tx_dropped; 9281 dest->multicast += src->multicast; 9282} 9283 9284static int netdev_offload_xstats_get_used(struct net_device *dev, 9285 enum netdev_offload_xstats_type type, 9286 bool *p_used, 9287 struct netlink_ext_ack *extack) 9288{ 9289 struct netdev_notifier_offload_xstats_ru report_used = {}; 9290 struct netdev_notifier_offload_xstats_info info = { 9291 .info.dev = dev, 9292 .info.extack = extack, 9293 .type = type, 9294 .report_used = &report_used, 9295 }; 9296 int rc; 9297 9298 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 9299 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 9300 &info.info); 9301 *p_used = report_used.used; 9302 return notifier_to_errno(rc); 9303} 9304 9305static int netdev_offload_xstats_get_stats(struct net_device *dev, 9306 enum netdev_offload_xstats_type type, 9307 struct rtnl_hw_stats64 *p_stats, 9308 bool *p_used, 9309 struct netlink_ext_ack *extack) 9310{ 9311 struct netdev_notifier_offload_xstats_rd report_delta = {}; 9312 struct netdev_notifier_offload_xstats_info info = { 9313 .info.dev = dev, 9314 .info.extack = extack, 9315 .type = type, 9316 .report_delta = &report_delta, 9317 }; 9318 struct rtnl_hw_stats64 *stats; 9319 int rc; 9320 9321 stats = netdev_offload_xstats_get_ptr(dev, type); 9322 if (WARN_ON(!stats)) 9323 return -EINVAL; 9324 9325 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 9326 &info.info); 9327 9328 /* Cache whatever we got, even if there was an error, otherwise the 9329 * successful stats retrievals would get lost. 9330 */ 9331 netdev_hw_stats64_add(stats, &report_delta.stats); 9332 9333 if (p_stats) 9334 *p_stats = *stats; 9335 *p_used = report_delta.used; 9336 9337 return notifier_to_errno(rc); 9338} 9339 9340int netdev_offload_xstats_get(struct net_device *dev, 9341 enum netdev_offload_xstats_type type, 9342 struct rtnl_hw_stats64 *p_stats, bool *p_used, 9343 struct netlink_ext_ack *extack) 9344{ 9345 ASSERT_RTNL(); 9346 9347 if (p_stats) 9348 return netdev_offload_xstats_get_stats(dev, type, p_stats, 9349 p_used, extack); 9350 else 9351 return netdev_offload_xstats_get_used(dev, type, p_used, 9352 extack); 9353} 9354EXPORT_SYMBOL(netdev_offload_xstats_get); 9355 9356void 9357netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 9358 const struct rtnl_hw_stats64 *stats) 9359{ 9360 report_delta->used = true; 9361 netdev_hw_stats64_add(&report_delta->stats, stats); 9362} 9363EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 9364 9365void 9366netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 9367{ 9368 report_used->used = true; 9369} 9370EXPORT_SYMBOL(netdev_offload_xstats_report_used); 9371 9372void netdev_offload_xstats_push_delta(struct net_device *dev, 9373 enum netdev_offload_xstats_type type, 9374 const struct rtnl_hw_stats64 *p_stats) 9375{ 9376 struct rtnl_hw_stats64 *stats; 9377 9378 ASSERT_RTNL(); 9379 9380 stats = netdev_offload_xstats_get_ptr(dev, type); 9381 if (WARN_ON(!stats)) 9382 return; 9383 9384 netdev_hw_stats64_add(stats, p_stats); 9385} 9386EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 9387 9388/** 9389 * netdev_get_xmit_slave - Get the xmit slave of master device 9390 * @dev: device 9391 * @skb: The packet 9392 * @all_slaves: assume all the slaves are active 9393 * 9394 * The reference counters are not incremented so the caller must be 9395 * careful with locks. The caller must hold RCU lock. 9396 * %NULL is returned if no slave is found. 9397 */ 9398 9399struct net_device *netdev_get_xmit_slave(struct net_device *dev, 9400 struct sk_buff *skb, 9401 bool all_slaves) 9402{ 9403 const struct net_device_ops *ops = dev->netdev_ops; 9404 9405 if (!ops->ndo_get_xmit_slave) 9406 return NULL; 9407 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 9408} 9409EXPORT_SYMBOL(netdev_get_xmit_slave); 9410 9411static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 9412 struct sock *sk) 9413{ 9414 const struct net_device_ops *ops = dev->netdev_ops; 9415 9416 if (!ops->ndo_sk_get_lower_dev) 9417 return NULL; 9418 return ops->ndo_sk_get_lower_dev(dev, sk); 9419} 9420 9421/** 9422 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 9423 * @dev: device 9424 * @sk: the socket 9425 * 9426 * %NULL is returned if no lower device is found. 9427 */ 9428 9429struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 9430 struct sock *sk) 9431{ 9432 struct net_device *lower; 9433 9434 lower = netdev_sk_get_lower_dev(dev, sk); 9435 while (lower) { 9436 dev = lower; 9437 lower = netdev_sk_get_lower_dev(dev, sk); 9438 } 9439 9440 return dev; 9441} 9442EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 9443 9444static void netdev_adjacent_add_links(struct net_device *dev) 9445{ 9446 struct netdev_adjacent *iter; 9447 9448 struct net *net = dev_net(dev); 9449 9450 list_for_each_entry(iter, &dev->adj_list.upper, list) { 9451 if (!net_eq(net, dev_net(iter->dev))) 9452 continue; 9453 netdev_adjacent_sysfs_add(iter->dev, dev, 9454 &iter->dev->adj_list.lower); 9455 netdev_adjacent_sysfs_add(dev, iter->dev, 9456 &dev->adj_list.upper); 9457 } 9458 9459 list_for_each_entry(iter, &dev->adj_list.lower, list) { 9460 if (!net_eq(net, dev_net(iter->dev))) 9461 continue; 9462 netdev_adjacent_sysfs_add(iter->dev, dev, 9463 &iter->dev->adj_list.upper); 9464 netdev_adjacent_sysfs_add(dev, iter->dev, 9465 &dev->adj_list.lower); 9466 } 9467} 9468 9469static void netdev_adjacent_del_links(struct net_device *dev) 9470{ 9471 struct netdev_adjacent *iter; 9472 9473 struct net *net = dev_net(dev); 9474 9475 list_for_each_entry(iter, &dev->adj_list.upper, list) { 9476 if (!net_eq(net, dev_net(iter->dev))) 9477 continue; 9478 netdev_adjacent_sysfs_del(iter->dev, dev->name, 9479 &iter->dev->adj_list.lower); 9480 netdev_adjacent_sysfs_del(dev, iter->dev->name, 9481 &dev->adj_list.upper); 9482 } 9483 9484 list_for_each_entry(iter, &dev->adj_list.lower, list) { 9485 if (!net_eq(net, dev_net(iter->dev))) 9486 continue; 9487 netdev_adjacent_sysfs_del(iter->dev, dev->name, 9488 &iter->dev->adj_list.upper); 9489 netdev_adjacent_sysfs_del(dev, iter->dev->name, 9490 &dev->adj_list.lower); 9491 } 9492} 9493 9494void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 9495{ 9496 struct netdev_adjacent *iter; 9497 9498 struct net *net = dev_net(dev); 9499 9500 list_for_each_entry(iter, &dev->adj_list.upper, list) { 9501 if (!net_eq(net, dev_net(iter->dev))) 9502 continue; 9503 netdev_adjacent_sysfs_del(iter->dev, oldname, 9504 &iter->dev->adj_list.lower); 9505 netdev_adjacent_sysfs_add(iter->dev, dev, 9506 &iter->dev->adj_list.lower); 9507 } 9508 9509 list_for_each_entry(iter, &dev->adj_list.lower, list) { 9510 if (!net_eq(net, dev_net(iter->dev))) 9511 continue; 9512 netdev_adjacent_sysfs_del(iter->dev, oldname, 9513 &iter->dev->adj_list.upper); 9514 netdev_adjacent_sysfs_add(iter->dev, dev, 9515 &iter->dev->adj_list.upper); 9516 } 9517} 9518 9519void *netdev_lower_dev_get_private(struct net_device *dev, 9520 struct net_device *lower_dev) 9521{ 9522 struct netdev_adjacent *lower; 9523 9524 if (!lower_dev) 9525 return NULL; 9526 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 9527 if (!lower) 9528 return NULL; 9529 9530 return lower->private; 9531} 9532EXPORT_SYMBOL(netdev_lower_dev_get_private); 9533 9534 9535/** 9536 * netdev_lower_state_changed - Dispatch event about lower device state change 9537 * @lower_dev: device 9538 * @lower_state_info: state to dispatch 9539 * 9540 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 9541 * The caller must hold the RTNL lock. 9542 */ 9543void netdev_lower_state_changed(struct net_device *lower_dev, 9544 void *lower_state_info) 9545{ 9546 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 9547 .info.dev = lower_dev, 9548 }; 9549 9550 ASSERT_RTNL(); 9551 changelowerstate_info.lower_state_info = lower_state_info; 9552 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 9553 &changelowerstate_info.info); 9554} 9555EXPORT_SYMBOL(netdev_lower_state_changed); 9556 9557static void dev_change_rx_flags(struct net_device *dev, int flags) 9558{ 9559 const struct net_device_ops *ops = dev->netdev_ops; 9560 9561 if (ops->ndo_change_rx_flags) 9562 ops->ndo_change_rx_flags(dev, flags); 9563} 9564 9565static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 9566{ 9567 unsigned int old_flags = dev->flags; 9568 unsigned int promiscuity, flags; 9569 kuid_t uid; 9570 kgid_t gid; 9571 9572 ASSERT_RTNL(); 9573 9574 promiscuity = dev->promiscuity + inc; 9575 if (promiscuity == 0) { 9576 /* 9577 * Avoid overflow. 9578 * If inc causes overflow, untouch promisc and return error. 9579 */ 9580 if (unlikely(inc > 0)) { 9581 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 9582 return -EOVERFLOW; 9583 } 9584 flags = old_flags & ~IFF_PROMISC; 9585 } else { 9586 flags = old_flags | IFF_PROMISC; 9587 } 9588 WRITE_ONCE(dev->promiscuity, promiscuity); 9589 if (flags != old_flags) { 9590 WRITE_ONCE(dev->flags, flags); 9591 netdev_info(dev, "%s promiscuous mode\n", 9592 dev->flags & IFF_PROMISC ? "entered" : "left"); 9593 if (audit_enabled) { 9594 current_uid_gid(&uid, &gid); 9595 audit_log(audit_context(), GFP_ATOMIC, 9596 AUDIT_ANOM_PROMISCUOUS, 9597 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 9598 dev->name, (dev->flags & IFF_PROMISC), 9599 (old_flags & IFF_PROMISC), 9600 from_kuid(&init_user_ns, audit_get_loginuid(current)), 9601 from_kuid(&init_user_ns, uid), 9602 from_kgid(&init_user_ns, gid), 9603 audit_get_sessionid(current)); 9604 } 9605 9606 dev_change_rx_flags(dev, IFF_PROMISC); 9607 } 9608 if (notify) { 9609 /* The ops lock is only required to ensure consistent locking 9610 * for `NETDEV_CHANGE` notifiers. This function is sometimes 9611 * called without the lock, even for devices that are ops 9612 * locked, such as in `dev_uc_sync_multiple` when using 9613 * bonding or teaming. 9614 */ 9615 netdev_ops_assert_locked(dev); 9616 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 9617 } 9618 return 0; 9619} 9620 9621int netif_set_promiscuity(struct net_device *dev, int inc) 9622{ 9623 unsigned int old_flags = dev->flags; 9624 int err; 9625 9626 err = __dev_set_promiscuity(dev, inc, true); 9627 if (err < 0) 9628 return err; 9629 if (dev->flags != old_flags) 9630 dev_set_rx_mode(dev); 9631 return err; 9632} 9633 9634int netif_set_allmulti(struct net_device *dev, int inc, bool notify) 9635{ 9636 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 9637 unsigned int allmulti, flags; 9638 9639 ASSERT_RTNL(); 9640 9641 allmulti = dev->allmulti + inc; 9642 if (allmulti == 0) { 9643 /* 9644 * Avoid overflow. 9645 * If inc causes overflow, untouch allmulti and return error. 9646 */ 9647 if (unlikely(inc > 0)) { 9648 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 9649 return -EOVERFLOW; 9650 } 9651 flags = old_flags & ~IFF_ALLMULTI; 9652 } else { 9653 flags = old_flags | IFF_ALLMULTI; 9654 } 9655 WRITE_ONCE(dev->allmulti, allmulti); 9656 if (flags != old_flags) { 9657 WRITE_ONCE(dev->flags, flags); 9658 netdev_info(dev, "%s allmulticast mode\n", 9659 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 9660 dev_change_rx_flags(dev, IFF_ALLMULTI); 9661 dev_set_rx_mode(dev); 9662 if (notify) 9663 __dev_notify_flags(dev, old_flags, 9664 dev->gflags ^ old_gflags, 0, NULL); 9665 } 9666 return 0; 9667} 9668 9669/* 9670 * Upload unicast and multicast address lists to device and 9671 * configure RX filtering. When the device doesn't support unicast 9672 * filtering it is put in promiscuous mode while unicast addresses 9673 * are present. 9674 */ 9675void __dev_set_rx_mode(struct net_device *dev) 9676{ 9677 const struct net_device_ops *ops = dev->netdev_ops; 9678 9679 /* dev_open will call this function so the list will stay sane. */ 9680 if (!(dev->flags&IFF_UP)) 9681 return; 9682 9683 if (!netif_device_present(dev)) 9684 return; 9685 9686 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 9687 /* Unicast addresses changes may only happen under the rtnl, 9688 * therefore calling __dev_set_promiscuity here is safe. 9689 */ 9690 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 9691 __dev_set_promiscuity(dev, 1, false); 9692 dev->uc_promisc = true; 9693 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 9694 __dev_set_promiscuity(dev, -1, false); 9695 dev->uc_promisc = false; 9696 } 9697 } 9698 9699 if (ops->ndo_set_rx_mode) 9700 ops->ndo_set_rx_mode(dev); 9701} 9702 9703void dev_set_rx_mode(struct net_device *dev) 9704{ 9705 netif_addr_lock_bh(dev); 9706 __dev_set_rx_mode(dev); 9707 netif_addr_unlock_bh(dev); 9708} 9709 9710/** 9711 * netif_get_flags() - get flags reported to userspace 9712 * @dev: device 9713 * 9714 * Get the combination of flag bits exported through APIs to userspace. 9715 */ 9716unsigned int netif_get_flags(const struct net_device *dev) 9717{ 9718 unsigned int flags; 9719 9720 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | 9721 IFF_ALLMULTI | 9722 IFF_RUNNING | 9723 IFF_LOWER_UP | 9724 IFF_DORMANT)) | 9725 (READ_ONCE(dev->gflags) & (IFF_PROMISC | 9726 IFF_ALLMULTI)); 9727 9728 if (netif_running(dev)) { 9729 if (netif_oper_up(dev)) 9730 flags |= IFF_RUNNING; 9731 if (netif_carrier_ok(dev)) 9732 flags |= IFF_LOWER_UP; 9733 if (netif_dormant(dev)) 9734 flags |= IFF_DORMANT; 9735 } 9736 9737 return flags; 9738} 9739EXPORT_SYMBOL(netif_get_flags); 9740 9741int __dev_change_flags(struct net_device *dev, unsigned int flags, 9742 struct netlink_ext_ack *extack) 9743{ 9744 unsigned int old_flags = dev->flags; 9745 int ret; 9746 9747 ASSERT_RTNL(); 9748 9749 /* 9750 * Set the flags on our device. 9751 */ 9752 9753 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 9754 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 9755 IFF_AUTOMEDIA)) | 9756 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 9757 IFF_ALLMULTI)); 9758 9759 /* 9760 * Load in the correct multicast list now the flags have changed. 9761 */ 9762 9763 if ((old_flags ^ flags) & IFF_MULTICAST) 9764 dev_change_rx_flags(dev, IFF_MULTICAST); 9765 9766 dev_set_rx_mode(dev); 9767 9768 /* 9769 * Have we downed the interface. We handle IFF_UP ourselves 9770 * according to user attempts to set it, rather than blindly 9771 * setting it. 9772 */ 9773 9774 ret = 0; 9775 if ((old_flags ^ flags) & IFF_UP) { 9776 if (old_flags & IFF_UP) 9777 __dev_close(dev); 9778 else 9779 ret = __dev_open(dev, extack); 9780 } 9781 9782 if ((flags ^ dev->gflags) & IFF_PROMISC) { 9783 int inc = (flags & IFF_PROMISC) ? 1 : -1; 9784 old_flags = dev->flags; 9785 9786 dev->gflags ^= IFF_PROMISC; 9787 9788 if (__dev_set_promiscuity(dev, inc, false) >= 0) 9789 if (dev->flags != old_flags) 9790 dev_set_rx_mode(dev); 9791 } 9792 9793 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 9794 * is important. Some (broken) drivers set IFF_PROMISC, when 9795 * IFF_ALLMULTI is requested not asking us and not reporting. 9796 */ 9797 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 9798 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 9799 9800 dev->gflags ^= IFF_ALLMULTI; 9801 netif_set_allmulti(dev, inc, false); 9802 } 9803 9804 return ret; 9805} 9806 9807void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 9808 unsigned int gchanges, u32 portid, 9809 const struct nlmsghdr *nlh) 9810{ 9811 unsigned int changes = dev->flags ^ old_flags; 9812 9813 if (gchanges) 9814 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 9815 9816 if (changes & IFF_UP) { 9817 if (dev->flags & IFF_UP) 9818 call_netdevice_notifiers(NETDEV_UP, dev); 9819 else 9820 call_netdevice_notifiers(NETDEV_DOWN, dev); 9821 } 9822 9823 if (dev->flags & IFF_UP && 9824 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 9825 struct netdev_notifier_change_info change_info = { 9826 .info = { 9827 .dev = dev, 9828 }, 9829 .flags_changed = changes, 9830 }; 9831 9832 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 9833 } 9834} 9835 9836int netif_change_flags(struct net_device *dev, unsigned int flags, 9837 struct netlink_ext_ack *extack) 9838{ 9839 int ret; 9840 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 9841 9842 ret = __dev_change_flags(dev, flags, extack); 9843 if (ret < 0) 9844 return ret; 9845 9846 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 9847 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 9848 return ret; 9849} 9850 9851int __netif_set_mtu(struct net_device *dev, int new_mtu) 9852{ 9853 const struct net_device_ops *ops = dev->netdev_ops; 9854 9855 if (ops->ndo_change_mtu) 9856 return ops->ndo_change_mtu(dev, new_mtu); 9857 9858 /* Pairs with all the lockless reads of dev->mtu in the stack */ 9859 WRITE_ONCE(dev->mtu, new_mtu); 9860 return 0; 9861} 9862EXPORT_SYMBOL_NS_GPL(__netif_set_mtu, "NETDEV_INTERNAL"); 9863 9864int dev_validate_mtu(struct net_device *dev, int new_mtu, 9865 struct netlink_ext_ack *extack) 9866{ 9867 /* MTU must be positive, and in range */ 9868 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 9869 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 9870 return -EINVAL; 9871 } 9872 9873 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 9874 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 9875 return -EINVAL; 9876 } 9877 return 0; 9878} 9879 9880/** 9881 * netif_set_mtu_ext() - Change maximum transfer unit 9882 * @dev: device 9883 * @new_mtu: new transfer unit 9884 * @extack: netlink extended ack 9885 * 9886 * Change the maximum transfer size of the network device. 9887 * 9888 * Return: 0 on success, -errno on failure. 9889 */ 9890int netif_set_mtu_ext(struct net_device *dev, int new_mtu, 9891 struct netlink_ext_ack *extack) 9892{ 9893 int err, orig_mtu; 9894 9895 netdev_ops_assert_locked(dev); 9896 9897 if (new_mtu == dev->mtu) 9898 return 0; 9899 9900 err = dev_validate_mtu(dev, new_mtu, extack); 9901 if (err) 9902 return err; 9903 9904 if (!netif_device_present(dev)) 9905 return -ENODEV; 9906 9907 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 9908 err = notifier_to_errno(err); 9909 if (err) 9910 return err; 9911 9912 orig_mtu = dev->mtu; 9913 err = __netif_set_mtu(dev, new_mtu); 9914 9915 if (!err) { 9916 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9917 orig_mtu); 9918 err = notifier_to_errno(err); 9919 if (err) { 9920 /* setting mtu back and notifying everyone again, 9921 * so that they have a chance to revert changes. 9922 */ 9923 __netif_set_mtu(dev, orig_mtu); 9924 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9925 new_mtu); 9926 } 9927 } 9928 return err; 9929} 9930 9931int netif_set_mtu(struct net_device *dev, int new_mtu) 9932{ 9933 struct netlink_ext_ack extack; 9934 int err; 9935 9936 memset(&extack, 0, sizeof(extack)); 9937 err = netif_set_mtu_ext(dev, new_mtu, &extack); 9938 if (err && extack._msg) 9939 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 9940 return err; 9941} 9942EXPORT_SYMBOL(netif_set_mtu); 9943 9944int netif_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 9945{ 9946 unsigned int orig_len = dev->tx_queue_len; 9947 int res; 9948 9949 if (new_len != (unsigned int)new_len) 9950 return -ERANGE; 9951 9952 if (new_len != orig_len) { 9953 WRITE_ONCE(dev->tx_queue_len, new_len); 9954 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 9955 res = notifier_to_errno(res); 9956 if (res) 9957 goto err_rollback; 9958 res = dev_qdisc_change_tx_queue_len(dev); 9959 if (res) 9960 goto err_rollback; 9961 } 9962 9963 return 0; 9964 9965err_rollback: 9966 netdev_err(dev, "refused to change device tx_queue_len\n"); 9967 WRITE_ONCE(dev->tx_queue_len, orig_len); 9968 return res; 9969} 9970 9971void netif_set_group(struct net_device *dev, int new_group) 9972{ 9973 dev->group = new_group; 9974} 9975 9976/** 9977 * netif_pre_changeaddr_notify() - Call NETDEV_PRE_CHANGEADDR. 9978 * @dev: device 9979 * @addr: new address 9980 * @extack: netlink extended ack 9981 * 9982 * Return: 0 on success, -errno on failure. 9983 */ 9984int netif_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9985 struct netlink_ext_ack *extack) 9986{ 9987 struct netdev_notifier_pre_changeaddr_info info = { 9988 .info.dev = dev, 9989 .info.extack = extack, 9990 .dev_addr = addr, 9991 }; 9992 int rc; 9993 9994 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9995 return notifier_to_errno(rc); 9996} 9997EXPORT_SYMBOL_NS_GPL(netif_pre_changeaddr_notify, "NETDEV_INTERNAL"); 9998 9999int netif_set_mac_address(struct net_device *dev, struct sockaddr_storage *ss, 10000 struct netlink_ext_ack *extack) 10001{ 10002 const struct net_device_ops *ops = dev->netdev_ops; 10003 int err; 10004 10005 if (!ops->ndo_set_mac_address) 10006 return -EOPNOTSUPP; 10007 if (ss->ss_family != dev->type) 10008 return -EINVAL; 10009 if (!netif_device_present(dev)) 10010 return -ENODEV; 10011 err = netif_pre_changeaddr_notify(dev, ss->__data, extack); 10012 if (err) 10013 return err; 10014 if (memcmp(dev->dev_addr, ss->__data, dev->addr_len)) { 10015 err = ops->ndo_set_mac_address(dev, ss); 10016 if (err) 10017 return err; 10018 } 10019 dev->addr_assign_type = NET_ADDR_SET; 10020 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 10021 add_device_randomness(dev->dev_addr, dev->addr_len); 10022 return 0; 10023} 10024 10025DECLARE_RWSEM(dev_addr_sem); 10026 10027/* "sa" is a true struct sockaddr with limited "sa_data" member. */ 10028int netif_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 10029{ 10030 size_t size = sizeof(sa->sa_data); 10031 struct net_device *dev; 10032 int ret = 0; 10033 10034 down_read(&dev_addr_sem); 10035 rcu_read_lock(); 10036 10037 dev = dev_get_by_name_rcu(net, dev_name); 10038 if (!dev) { 10039 ret = -ENODEV; 10040 goto unlock; 10041 } 10042 if (!dev->addr_len) 10043 memset(sa->sa_data, 0, size); 10044 else 10045 memcpy(sa->sa_data, dev->dev_addr, 10046 min_t(size_t, size, dev->addr_len)); 10047 sa->sa_family = dev->type; 10048 10049unlock: 10050 rcu_read_unlock(); 10051 up_read(&dev_addr_sem); 10052 return ret; 10053} 10054EXPORT_SYMBOL_NS_GPL(netif_get_mac_address, "NETDEV_INTERNAL"); 10055 10056int netif_change_carrier(struct net_device *dev, bool new_carrier) 10057{ 10058 const struct net_device_ops *ops = dev->netdev_ops; 10059 10060 if (!ops->ndo_change_carrier) 10061 return -EOPNOTSUPP; 10062 if (!netif_device_present(dev)) 10063 return -ENODEV; 10064 return ops->ndo_change_carrier(dev, new_carrier); 10065} 10066 10067/** 10068 * dev_get_phys_port_id - Get device physical port ID 10069 * @dev: device 10070 * @ppid: port ID 10071 * 10072 * Get device physical port ID 10073 */ 10074int dev_get_phys_port_id(struct net_device *dev, 10075 struct netdev_phys_item_id *ppid) 10076{ 10077 const struct net_device_ops *ops = dev->netdev_ops; 10078 10079 if (!ops->ndo_get_phys_port_id) 10080 return -EOPNOTSUPP; 10081 return ops->ndo_get_phys_port_id(dev, ppid); 10082} 10083 10084/** 10085 * dev_get_phys_port_name - Get device physical port name 10086 * @dev: device 10087 * @name: port name 10088 * @len: limit of bytes to copy to name 10089 * 10090 * Get device physical port name 10091 */ 10092int dev_get_phys_port_name(struct net_device *dev, 10093 char *name, size_t len) 10094{ 10095 const struct net_device_ops *ops = dev->netdev_ops; 10096 int err; 10097 10098 if (ops->ndo_get_phys_port_name) { 10099 err = ops->ndo_get_phys_port_name(dev, name, len); 10100 if (err != -EOPNOTSUPP) 10101 return err; 10102 } 10103 return devlink_compat_phys_port_name_get(dev, name, len); 10104} 10105 10106/** 10107 * netif_get_port_parent_id() - Get the device's port parent identifier 10108 * @dev: network device 10109 * @ppid: pointer to a storage for the port's parent identifier 10110 * @recurse: allow/disallow recursion to lower devices 10111 * 10112 * Get the devices's port parent identifier. 10113 * 10114 * Return: 0 on success, -errno on failure. 10115 */ 10116int netif_get_port_parent_id(struct net_device *dev, 10117 struct netdev_phys_item_id *ppid, bool recurse) 10118{ 10119 const struct net_device_ops *ops = dev->netdev_ops; 10120 struct netdev_phys_item_id first = { }; 10121 struct net_device *lower_dev; 10122 struct list_head *iter; 10123 int err; 10124 10125 if (ops->ndo_get_port_parent_id) { 10126 err = ops->ndo_get_port_parent_id(dev, ppid); 10127 if (err != -EOPNOTSUPP) 10128 return err; 10129 } 10130 10131 err = devlink_compat_switch_id_get(dev, ppid); 10132 if (!recurse || err != -EOPNOTSUPP) 10133 return err; 10134 10135 netdev_for_each_lower_dev(dev, lower_dev, iter) { 10136 err = netif_get_port_parent_id(lower_dev, ppid, true); 10137 if (err) 10138 break; 10139 if (!first.id_len) 10140 first = *ppid; 10141 else if (memcmp(&first, ppid, sizeof(*ppid))) 10142 return -EOPNOTSUPP; 10143 } 10144 10145 return err; 10146} 10147EXPORT_SYMBOL(netif_get_port_parent_id); 10148 10149/** 10150 * netdev_port_same_parent_id - Indicate if two network devices have 10151 * the same port parent identifier 10152 * @a: first network device 10153 * @b: second network device 10154 */ 10155bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 10156{ 10157 struct netdev_phys_item_id a_id = { }; 10158 struct netdev_phys_item_id b_id = { }; 10159 10160 if (netif_get_port_parent_id(a, &a_id, true) || 10161 netif_get_port_parent_id(b, &b_id, true)) 10162 return false; 10163 10164 return netdev_phys_item_id_same(&a_id, &b_id); 10165} 10166EXPORT_SYMBOL(netdev_port_same_parent_id); 10167 10168int netif_change_proto_down(struct net_device *dev, bool proto_down) 10169{ 10170 if (!dev->change_proto_down) 10171 return -EOPNOTSUPP; 10172 if (!netif_device_present(dev)) 10173 return -ENODEV; 10174 if (proto_down) 10175 netif_carrier_off(dev); 10176 else 10177 netif_carrier_on(dev); 10178 WRITE_ONCE(dev->proto_down, proto_down); 10179 return 0; 10180} 10181 10182/** 10183 * netdev_change_proto_down_reason_locked - proto down reason 10184 * 10185 * @dev: device 10186 * @mask: proto down mask 10187 * @value: proto down value 10188 */ 10189void netdev_change_proto_down_reason_locked(struct net_device *dev, 10190 unsigned long mask, u32 value) 10191{ 10192 u32 proto_down_reason; 10193 int b; 10194 10195 if (!mask) { 10196 proto_down_reason = value; 10197 } else { 10198 proto_down_reason = dev->proto_down_reason; 10199 for_each_set_bit(b, &mask, 32) { 10200 if (value & (1 << b)) 10201 proto_down_reason |= BIT(b); 10202 else 10203 proto_down_reason &= ~BIT(b); 10204 } 10205 } 10206 WRITE_ONCE(dev->proto_down_reason, proto_down_reason); 10207} 10208 10209struct bpf_xdp_link { 10210 struct bpf_link link; 10211 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 10212 int flags; 10213}; 10214 10215static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 10216{ 10217 if (flags & XDP_FLAGS_HW_MODE) 10218 return XDP_MODE_HW; 10219 if (flags & XDP_FLAGS_DRV_MODE) 10220 return XDP_MODE_DRV; 10221 if (flags & XDP_FLAGS_SKB_MODE) 10222 return XDP_MODE_SKB; 10223 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 10224} 10225 10226static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 10227{ 10228 switch (mode) { 10229 case XDP_MODE_SKB: 10230 return generic_xdp_install; 10231 case XDP_MODE_DRV: 10232 case XDP_MODE_HW: 10233 return dev->netdev_ops->ndo_bpf; 10234 default: 10235 return NULL; 10236 } 10237} 10238 10239static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 10240 enum bpf_xdp_mode mode) 10241{ 10242 return dev->xdp_state[mode].link; 10243} 10244 10245static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 10246 enum bpf_xdp_mode mode) 10247{ 10248 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 10249 10250 if (link) 10251 return link->link.prog; 10252 return dev->xdp_state[mode].prog; 10253} 10254 10255u8 dev_xdp_prog_count(struct net_device *dev) 10256{ 10257 u8 count = 0; 10258 int i; 10259 10260 for (i = 0; i < __MAX_XDP_MODE; i++) 10261 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 10262 count++; 10263 return count; 10264} 10265EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 10266 10267u8 dev_xdp_sb_prog_count(struct net_device *dev) 10268{ 10269 u8 count = 0; 10270 int i; 10271 10272 for (i = 0; i < __MAX_XDP_MODE; i++) 10273 if (dev->xdp_state[i].prog && 10274 !dev->xdp_state[i].prog->aux->xdp_has_frags) 10275 count++; 10276 return count; 10277} 10278 10279int netif_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf) 10280{ 10281 if (!dev->netdev_ops->ndo_bpf) 10282 return -EOPNOTSUPP; 10283 10284 if (dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED && 10285 bpf->command == XDP_SETUP_PROG && 10286 bpf->prog && !bpf->prog->aux->xdp_has_frags) { 10287 NL_SET_ERR_MSG(bpf->extack, 10288 "unable to propagate XDP to device using tcp-data-split"); 10289 return -EBUSY; 10290 } 10291 10292 if (dev_get_min_mp_channel_count(dev)) { 10293 NL_SET_ERR_MSG(bpf->extack, "unable to propagate XDP to device using memory provider"); 10294 return -EBUSY; 10295 } 10296 10297 return dev->netdev_ops->ndo_bpf(dev, bpf); 10298} 10299EXPORT_SYMBOL_GPL(netif_xdp_propagate); 10300 10301u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 10302{ 10303 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 10304 10305 return prog ? prog->aux->id : 0; 10306} 10307 10308static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 10309 struct bpf_xdp_link *link) 10310{ 10311 dev->xdp_state[mode].link = link; 10312 dev->xdp_state[mode].prog = NULL; 10313} 10314 10315static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 10316 struct bpf_prog *prog) 10317{ 10318 dev->xdp_state[mode].link = NULL; 10319 dev->xdp_state[mode].prog = prog; 10320} 10321 10322static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 10323 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 10324 u32 flags, struct bpf_prog *prog) 10325{ 10326 struct netdev_bpf xdp; 10327 int err; 10328 10329 netdev_ops_assert_locked(dev); 10330 10331 if (dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED && 10332 prog && !prog->aux->xdp_has_frags) { 10333 NL_SET_ERR_MSG(extack, "unable to install XDP to device using tcp-data-split"); 10334 return -EBUSY; 10335 } 10336 10337 if (dev_get_min_mp_channel_count(dev)) { 10338 NL_SET_ERR_MSG(extack, "unable to install XDP to device using memory provider"); 10339 return -EBUSY; 10340 } 10341 10342 memset(&xdp, 0, sizeof(xdp)); 10343 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 10344 xdp.extack = extack; 10345 xdp.flags = flags; 10346 xdp.prog = prog; 10347 10348 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 10349 * "moved" into driver), so they don't increment it on their own, but 10350 * they do decrement refcnt when program is detached or replaced. 10351 * Given net_device also owns link/prog, we need to bump refcnt here 10352 * to prevent drivers from underflowing it. 10353 */ 10354 if (prog) 10355 bpf_prog_inc(prog); 10356 err = bpf_op(dev, &xdp); 10357 if (err) { 10358 if (prog) 10359 bpf_prog_put(prog); 10360 return err; 10361 } 10362 10363 if (mode != XDP_MODE_HW) 10364 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 10365 10366 return 0; 10367} 10368 10369static void dev_xdp_uninstall(struct net_device *dev) 10370{ 10371 struct bpf_xdp_link *link; 10372 struct bpf_prog *prog; 10373 enum bpf_xdp_mode mode; 10374 bpf_op_t bpf_op; 10375 10376 ASSERT_RTNL(); 10377 10378 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 10379 prog = dev_xdp_prog(dev, mode); 10380 if (!prog) 10381 continue; 10382 10383 bpf_op = dev_xdp_bpf_op(dev, mode); 10384 if (!bpf_op) 10385 continue; 10386 10387 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 10388 10389 /* auto-detach link from net device */ 10390 link = dev_xdp_link(dev, mode); 10391 if (link) 10392 link->dev = NULL; 10393 else 10394 bpf_prog_put(prog); 10395 10396 dev_xdp_set_link(dev, mode, NULL); 10397 } 10398} 10399 10400static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 10401 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 10402 struct bpf_prog *old_prog, u32 flags) 10403{ 10404 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 10405 struct bpf_prog *cur_prog; 10406 struct net_device *upper; 10407 struct list_head *iter; 10408 enum bpf_xdp_mode mode; 10409 bpf_op_t bpf_op; 10410 int err; 10411 10412 ASSERT_RTNL(); 10413 10414 /* either link or prog attachment, never both */ 10415 if (link && (new_prog || old_prog)) 10416 return -EINVAL; 10417 /* link supports only XDP mode flags */ 10418 if (link && (flags & ~XDP_FLAGS_MODES)) { 10419 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 10420 return -EINVAL; 10421 } 10422 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 10423 if (num_modes > 1) { 10424 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 10425 return -EINVAL; 10426 } 10427 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 10428 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 10429 NL_SET_ERR_MSG(extack, 10430 "More than one program loaded, unset mode is ambiguous"); 10431 return -EINVAL; 10432 } 10433 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 10434 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 10435 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 10436 return -EINVAL; 10437 } 10438 10439 mode = dev_xdp_mode(dev, flags); 10440 /* can't replace attached link */ 10441 if (dev_xdp_link(dev, mode)) { 10442 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 10443 return -EBUSY; 10444 } 10445 10446 /* don't allow if an upper device already has a program */ 10447 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 10448 if (dev_xdp_prog_count(upper) > 0) { 10449 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 10450 return -EEXIST; 10451 } 10452 } 10453 10454 cur_prog = dev_xdp_prog(dev, mode); 10455 /* can't replace attached prog with link */ 10456 if (link && cur_prog) { 10457 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 10458 return -EBUSY; 10459 } 10460 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 10461 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 10462 return -EEXIST; 10463 } 10464 10465 /* put effective new program into new_prog */ 10466 if (link) 10467 new_prog = link->link.prog; 10468 10469 if (new_prog) { 10470 bool offload = mode == XDP_MODE_HW; 10471 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 10472 ? XDP_MODE_DRV : XDP_MODE_SKB; 10473 10474 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 10475 NL_SET_ERR_MSG(extack, "XDP program already attached"); 10476 return -EBUSY; 10477 } 10478 if (!offload && dev_xdp_prog(dev, other_mode)) { 10479 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 10480 return -EEXIST; 10481 } 10482 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 10483 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 10484 return -EINVAL; 10485 } 10486 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 10487 NL_SET_ERR_MSG(extack, "Program bound to different device"); 10488 return -EINVAL; 10489 } 10490 if (bpf_prog_is_dev_bound(new_prog->aux) && mode == XDP_MODE_SKB) { 10491 NL_SET_ERR_MSG(extack, "Can't attach device-bound programs in generic mode"); 10492 return -EINVAL; 10493 } 10494 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 10495 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 10496 return -EINVAL; 10497 } 10498 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 10499 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 10500 return -EINVAL; 10501 } 10502 } 10503 10504 /* don't call drivers if the effective program didn't change */ 10505 if (new_prog != cur_prog) { 10506 bpf_op = dev_xdp_bpf_op(dev, mode); 10507 if (!bpf_op) { 10508 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 10509 return -EOPNOTSUPP; 10510 } 10511 10512 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 10513 if (err) 10514 return err; 10515 } 10516 10517 if (link) 10518 dev_xdp_set_link(dev, mode, link); 10519 else 10520 dev_xdp_set_prog(dev, mode, new_prog); 10521 if (cur_prog) 10522 bpf_prog_put(cur_prog); 10523 10524 return 0; 10525} 10526 10527static int dev_xdp_attach_link(struct net_device *dev, 10528 struct netlink_ext_ack *extack, 10529 struct bpf_xdp_link *link) 10530{ 10531 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 10532} 10533 10534static int dev_xdp_detach_link(struct net_device *dev, 10535 struct netlink_ext_ack *extack, 10536 struct bpf_xdp_link *link) 10537{ 10538 enum bpf_xdp_mode mode; 10539 bpf_op_t bpf_op; 10540 10541 ASSERT_RTNL(); 10542 10543 mode = dev_xdp_mode(dev, link->flags); 10544 if (dev_xdp_link(dev, mode) != link) 10545 return -EINVAL; 10546 10547 bpf_op = dev_xdp_bpf_op(dev, mode); 10548 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 10549 dev_xdp_set_link(dev, mode, NULL); 10550 return 0; 10551} 10552 10553static void bpf_xdp_link_release(struct bpf_link *link) 10554{ 10555 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 10556 10557 rtnl_lock(); 10558 10559 /* if racing with net_device's tear down, xdp_link->dev might be 10560 * already NULL, in which case link was already auto-detached 10561 */ 10562 if (xdp_link->dev) { 10563 netdev_lock_ops(xdp_link->dev); 10564 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 10565 netdev_unlock_ops(xdp_link->dev); 10566 xdp_link->dev = NULL; 10567 } 10568 10569 rtnl_unlock(); 10570} 10571 10572static int bpf_xdp_link_detach(struct bpf_link *link) 10573{ 10574 bpf_xdp_link_release(link); 10575 return 0; 10576} 10577 10578static void bpf_xdp_link_dealloc(struct bpf_link *link) 10579{ 10580 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 10581 10582 kfree(xdp_link); 10583} 10584 10585static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 10586 struct seq_file *seq) 10587{ 10588 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 10589 u32 ifindex = 0; 10590 10591 rtnl_lock(); 10592 if (xdp_link->dev) 10593 ifindex = xdp_link->dev->ifindex; 10594 rtnl_unlock(); 10595 10596 seq_printf(seq, "ifindex:\t%u\n", ifindex); 10597} 10598 10599static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 10600 struct bpf_link_info *info) 10601{ 10602 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 10603 u32 ifindex = 0; 10604 10605 rtnl_lock(); 10606 if (xdp_link->dev) 10607 ifindex = xdp_link->dev->ifindex; 10608 rtnl_unlock(); 10609 10610 info->xdp.ifindex = ifindex; 10611 return 0; 10612} 10613 10614static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 10615 struct bpf_prog *old_prog) 10616{ 10617 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 10618 enum bpf_xdp_mode mode; 10619 bpf_op_t bpf_op; 10620 int err = 0; 10621 10622 rtnl_lock(); 10623 10624 /* link might have been auto-released already, so fail */ 10625 if (!xdp_link->dev) { 10626 err = -ENOLINK; 10627 goto out_unlock; 10628 } 10629 10630 if (old_prog && link->prog != old_prog) { 10631 err = -EPERM; 10632 goto out_unlock; 10633 } 10634 old_prog = link->prog; 10635 if (old_prog->type != new_prog->type || 10636 old_prog->expected_attach_type != new_prog->expected_attach_type) { 10637 err = -EINVAL; 10638 goto out_unlock; 10639 } 10640 10641 if (old_prog == new_prog) { 10642 /* no-op, don't disturb drivers */ 10643 bpf_prog_put(new_prog); 10644 goto out_unlock; 10645 } 10646 10647 netdev_lock_ops(xdp_link->dev); 10648 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 10649 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 10650 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 10651 xdp_link->flags, new_prog); 10652 netdev_unlock_ops(xdp_link->dev); 10653 if (err) 10654 goto out_unlock; 10655 10656 old_prog = xchg(&link->prog, new_prog); 10657 bpf_prog_put(old_prog); 10658 10659out_unlock: 10660 rtnl_unlock(); 10661 return err; 10662} 10663 10664static const struct bpf_link_ops bpf_xdp_link_lops = { 10665 .release = bpf_xdp_link_release, 10666 .dealloc = bpf_xdp_link_dealloc, 10667 .detach = bpf_xdp_link_detach, 10668 .show_fdinfo = bpf_xdp_link_show_fdinfo, 10669 .fill_link_info = bpf_xdp_link_fill_link_info, 10670 .update_prog = bpf_xdp_link_update, 10671}; 10672 10673int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 10674{ 10675 struct net *net = current->nsproxy->net_ns; 10676 struct bpf_link_primer link_primer; 10677 struct netlink_ext_ack extack = {}; 10678 struct bpf_xdp_link *link; 10679 struct net_device *dev; 10680 int err, fd; 10681 10682 rtnl_lock(); 10683 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 10684 if (!dev) { 10685 rtnl_unlock(); 10686 return -EINVAL; 10687 } 10688 10689 link = kzalloc(sizeof(*link), GFP_USER); 10690 if (!link) { 10691 err = -ENOMEM; 10692 goto unlock; 10693 } 10694 10695 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog, 10696 attr->link_create.attach_type); 10697 link->dev = dev; 10698 link->flags = attr->link_create.flags; 10699 10700 err = bpf_link_prime(&link->link, &link_primer); 10701 if (err) { 10702 kfree(link); 10703 goto unlock; 10704 } 10705 10706 netdev_lock_ops(dev); 10707 err = dev_xdp_attach_link(dev, &extack, link); 10708 netdev_unlock_ops(dev); 10709 rtnl_unlock(); 10710 10711 if (err) { 10712 link->dev = NULL; 10713 bpf_link_cleanup(&link_primer); 10714 trace_bpf_xdp_link_attach_failed(extack._msg); 10715 goto out_put_dev; 10716 } 10717 10718 fd = bpf_link_settle(&link_primer); 10719 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 10720 dev_put(dev); 10721 return fd; 10722 10723unlock: 10724 rtnl_unlock(); 10725 10726out_put_dev: 10727 dev_put(dev); 10728 return err; 10729} 10730 10731/** 10732 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 10733 * @dev: device 10734 * @extack: netlink extended ack 10735 * @fd: new program fd or negative value to clear 10736 * @expected_fd: old program fd that userspace expects to replace or clear 10737 * @flags: xdp-related flags 10738 * 10739 * Set or clear a bpf program for a device 10740 */ 10741int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 10742 int fd, int expected_fd, u32 flags) 10743{ 10744 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 10745 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 10746 int err; 10747 10748 ASSERT_RTNL(); 10749 10750 if (fd >= 0) { 10751 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 10752 mode != XDP_MODE_SKB); 10753 if (IS_ERR(new_prog)) 10754 return PTR_ERR(new_prog); 10755 } 10756 10757 if (expected_fd >= 0) { 10758 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 10759 mode != XDP_MODE_SKB); 10760 if (IS_ERR(old_prog)) { 10761 err = PTR_ERR(old_prog); 10762 old_prog = NULL; 10763 goto err_out; 10764 } 10765 } 10766 10767 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 10768 10769err_out: 10770 if (err && new_prog) 10771 bpf_prog_put(new_prog); 10772 if (old_prog) 10773 bpf_prog_put(old_prog); 10774 return err; 10775} 10776 10777u32 dev_get_min_mp_channel_count(const struct net_device *dev) 10778{ 10779 int i; 10780 10781 netdev_ops_assert_locked(dev); 10782 10783 for (i = dev->real_num_rx_queues - 1; i >= 0; i--) 10784 if (dev->_rx[i].mp_params.mp_priv) 10785 /* The channel count is the idx plus 1. */ 10786 return i + 1; 10787 10788 return 0; 10789} 10790 10791/** 10792 * dev_index_reserve() - allocate an ifindex in a namespace 10793 * @net: the applicable net namespace 10794 * @ifindex: requested ifindex, pass %0 to get one allocated 10795 * 10796 * Allocate a ifindex for a new device. Caller must either use the ifindex 10797 * to store the device (via list_netdevice()) or call dev_index_release() 10798 * to give the index up. 10799 * 10800 * Return: a suitable unique value for a new device interface number or -errno. 10801 */ 10802static int dev_index_reserve(struct net *net, u32 ifindex) 10803{ 10804 int err; 10805 10806 if (ifindex > INT_MAX) { 10807 DEBUG_NET_WARN_ON_ONCE(1); 10808 return -EINVAL; 10809 } 10810 10811 if (!ifindex) 10812 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 10813 xa_limit_31b, &net->ifindex, GFP_KERNEL); 10814 else 10815 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 10816 if (err < 0) 10817 return err; 10818 10819 return ifindex; 10820} 10821 10822static void dev_index_release(struct net *net, int ifindex) 10823{ 10824 /* Expect only unused indexes, unlist_netdevice() removes the used */ 10825 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 10826} 10827 10828static bool from_cleanup_net(void) 10829{ 10830#ifdef CONFIG_NET_NS 10831 return current == READ_ONCE(cleanup_net_task); 10832#else 10833 return false; 10834#endif 10835} 10836 10837/* Delayed registration/unregisteration */ 10838LIST_HEAD(net_todo_list); 10839DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 10840atomic_t dev_unreg_count = ATOMIC_INIT(0); 10841 10842static void net_set_todo(struct net_device *dev) 10843{ 10844 list_add_tail(&dev->todo_list, &net_todo_list); 10845} 10846 10847static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 10848 struct net_device *upper, netdev_features_t features) 10849{ 10850 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10851 netdev_features_t feature; 10852 int feature_bit; 10853 10854 for_each_netdev_feature(upper_disables, feature_bit) { 10855 feature = __NETIF_F_BIT(feature_bit); 10856 if (!(upper->wanted_features & feature) 10857 && (features & feature)) { 10858 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 10859 &feature, upper->name); 10860 features &= ~feature; 10861 } 10862 } 10863 10864 return features; 10865} 10866 10867static void netdev_sync_lower_features(struct net_device *upper, 10868 struct net_device *lower, netdev_features_t features) 10869{ 10870 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10871 netdev_features_t feature; 10872 int feature_bit; 10873 10874 for_each_netdev_feature(upper_disables, feature_bit) { 10875 feature = __NETIF_F_BIT(feature_bit); 10876 if (!(features & feature) && (lower->features & feature)) { 10877 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 10878 &feature, lower->name); 10879 netdev_lock_ops(lower); 10880 lower->wanted_features &= ~feature; 10881 __netdev_update_features(lower); 10882 10883 if (unlikely(lower->features & feature)) 10884 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 10885 &feature, lower->name); 10886 else 10887 netdev_features_change(lower); 10888 netdev_unlock_ops(lower); 10889 } 10890 } 10891} 10892 10893static bool netdev_has_ip_or_hw_csum(netdev_features_t features) 10894{ 10895 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 10896 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask; 10897 bool hw_csum = features & NETIF_F_HW_CSUM; 10898 10899 return ip_csum || hw_csum; 10900} 10901 10902static netdev_features_t netdev_fix_features(struct net_device *dev, 10903 netdev_features_t features) 10904{ 10905 /* Fix illegal checksum combinations */ 10906 if ((features & NETIF_F_HW_CSUM) && 10907 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 10908 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 10909 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 10910 } 10911 10912 /* TSO requires that SG is present as well. */ 10913 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 10914 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 10915 features &= ~NETIF_F_ALL_TSO; 10916 } 10917 10918 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 10919 !(features & NETIF_F_IP_CSUM)) { 10920 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 10921 features &= ~NETIF_F_TSO; 10922 features &= ~NETIF_F_TSO_ECN; 10923 } 10924 10925 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 10926 !(features & NETIF_F_IPV6_CSUM)) { 10927 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 10928 features &= ~NETIF_F_TSO6; 10929 } 10930 10931 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 10932 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 10933 features &= ~NETIF_F_TSO_MANGLEID; 10934 10935 /* TSO ECN requires that TSO is present as well. */ 10936 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 10937 features &= ~NETIF_F_TSO_ECN; 10938 10939 /* Software GSO depends on SG. */ 10940 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 10941 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 10942 features &= ~NETIF_F_GSO; 10943 } 10944 10945 /* GSO partial features require GSO partial be set */ 10946 if ((features & dev->gso_partial_features) && 10947 !(features & NETIF_F_GSO_PARTIAL)) { 10948 netdev_dbg(dev, 10949 "Dropping partially supported GSO features since no GSO partial.\n"); 10950 features &= ~dev->gso_partial_features; 10951 } 10952 10953 if (!(features & NETIF_F_RXCSUM)) { 10954 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 10955 * successfully merged by hardware must also have the 10956 * checksum verified by hardware. If the user does not 10957 * want to enable RXCSUM, logically, we should disable GRO_HW. 10958 */ 10959 if (features & NETIF_F_GRO_HW) { 10960 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 10961 features &= ~NETIF_F_GRO_HW; 10962 } 10963 } 10964 10965 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 10966 if (features & NETIF_F_RXFCS) { 10967 if (features & NETIF_F_LRO) { 10968 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 10969 features &= ~NETIF_F_LRO; 10970 } 10971 10972 if (features & NETIF_F_GRO_HW) { 10973 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 10974 features &= ~NETIF_F_GRO_HW; 10975 } 10976 } 10977 10978 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 10979 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 10980 features &= ~NETIF_F_LRO; 10981 } 10982 10983 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) { 10984 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 10985 features &= ~NETIF_F_HW_TLS_TX; 10986 } 10987 10988 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 10989 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 10990 features &= ~NETIF_F_HW_TLS_RX; 10991 } 10992 10993 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) { 10994 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n"); 10995 features &= ~NETIF_F_GSO_UDP_L4; 10996 } 10997 10998 return features; 10999} 11000 11001int __netdev_update_features(struct net_device *dev) 11002{ 11003 struct net_device *upper, *lower; 11004 netdev_features_t features; 11005 struct list_head *iter; 11006 int err = -1; 11007 11008 ASSERT_RTNL(); 11009 netdev_ops_assert_locked(dev); 11010 11011 features = netdev_get_wanted_features(dev); 11012 11013 if (dev->netdev_ops->ndo_fix_features) 11014 features = dev->netdev_ops->ndo_fix_features(dev, features); 11015 11016 /* driver might be less strict about feature dependencies */ 11017 features = netdev_fix_features(dev, features); 11018 11019 /* some features can't be enabled if they're off on an upper device */ 11020 netdev_for_each_upper_dev_rcu(dev, upper, iter) 11021 features = netdev_sync_upper_features(dev, upper, features); 11022 11023 if (dev->features == features) 11024 goto sync_lower; 11025 11026 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 11027 &dev->features, &features); 11028 11029 if (dev->netdev_ops->ndo_set_features) 11030 err = dev->netdev_ops->ndo_set_features(dev, features); 11031 else 11032 err = 0; 11033 11034 if (unlikely(err < 0)) { 11035 netdev_err(dev, 11036 "set_features() failed (%d); wanted %pNF, left %pNF\n", 11037 err, &features, &dev->features); 11038 /* return non-0 since some features might have changed and 11039 * it's better to fire a spurious notification than miss it 11040 */ 11041 return -1; 11042 } 11043 11044sync_lower: 11045 /* some features must be disabled on lower devices when disabled 11046 * on an upper device (think: bonding master or bridge) 11047 */ 11048 netdev_for_each_lower_dev(dev, lower, iter) 11049 netdev_sync_lower_features(dev, lower, features); 11050 11051 if (!err) { 11052 netdev_features_t diff = features ^ dev->features; 11053 11054 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 11055 /* udp_tunnel_{get,drop}_rx_info both need 11056 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 11057 * device, or they won't do anything. 11058 * Thus we need to update dev->features 11059 * *before* calling udp_tunnel_get_rx_info, 11060 * but *after* calling udp_tunnel_drop_rx_info. 11061 */ 11062 udp_tunnel_nic_lock(dev); 11063 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 11064 dev->features = features; 11065 udp_tunnel_get_rx_info(dev); 11066 } else { 11067 udp_tunnel_drop_rx_info(dev); 11068 } 11069 udp_tunnel_nic_unlock(dev); 11070 } 11071 11072 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 11073 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 11074 dev->features = features; 11075 err |= vlan_get_rx_ctag_filter_info(dev); 11076 } else { 11077 vlan_drop_rx_ctag_filter_info(dev); 11078 } 11079 } 11080 11081 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 11082 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 11083 dev->features = features; 11084 err |= vlan_get_rx_stag_filter_info(dev); 11085 } else { 11086 vlan_drop_rx_stag_filter_info(dev); 11087 } 11088 } 11089 11090 dev->features = features; 11091 } 11092 11093 return err < 0 ? 0 : 1; 11094} 11095 11096/** 11097 * netdev_update_features - recalculate device features 11098 * @dev: the device to check 11099 * 11100 * Recalculate dev->features set and send notifications if it 11101 * has changed. Should be called after driver or hardware dependent 11102 * conditions might have changed that influence the features. 11103 */ 11104void netdev_update_features(struct net_device *dev) 11105{ 11106 if (__netdev_update_features(dev)) 11107 netdev_features_change(dev); 11108} 11109EXPORT_SYMBOL(netdev_update_features); 11110 11111/** 11112 * netdev_change_features - recalculate device features 11113 * @dev: the device to check 11114 * 11115 * Recalculate dev->features set and send notifications even 11116 * if they have not changed. Should be called instead of 11117 * netdev_update_features() if also dev->vlan_features might 11118 * have changed to allow the changes to be propagated to stacked 11119 * VLAN devices. 11120 */ 11121void netdev_change_features(struct net_device *dev) 11122{ 11123 __netdev_update_features(dev); 11124 netdev_features_change(dev); 11125} 11126EXPORT_SYMBOL(netdev_change_features); 11127 11128/** 11129 * netif_stacked_transfer_operstate - transfer operstate 11130 * @rootdev: the root or lower level device to transfer state from 11131 * @dev: the device to transfer operstate to 11132 * 11133 * Transfer operational state from root to device. This is normally 11134 * called when a stacking relationship exists between the root 11135 * device and the device(a leaf device). 11136 */ 11137void netif_stacked_transfer_operstate(const struct net_device *rootdev, 11138 struct net_device *dev) 11139{ 11140 if (rootdev->operstate == IF_OPER_DORMANT) 11141 netif_dormant_on(dev); 11142 else 11143 netif_dormant_off(dev); 11144 11145 if (rootdev->operstate == IF_OPER_TESTING) 11146 netif_testing_on(dev); 11147 else 11148 netif_testing_off(dev); 11149 11150 if (netif_carrier_ok(rootdev)) 11151 netif_carrier_on(dev); 11152 else 11153 netif_carrier_off(dev); 11154} 11155EXPORT_SYMBOL(netif_stacked_transfer_operstate); 11156 11157static int netif_alloc_rx_queues(struct net_device *dev) 11158{ 11159 unsigned int i, count = dev->num_rx_queues; 11160 struct netdev_rx_queue *rx; 11161 size_t sz = count * sizeof(*rx); 11162 int err = 0; 11163 11164 BUG_ON(count < 1); 11165 11166 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 11167 if (!rx) 11168 return -ENOMEM; 11169 11170 dev->_rx = rx; 11171 11172 for (i = 0; i < count; i++) { 11173 rx[i].dev = dev; 11174 11175 /* XDP RX-queue setup */ 11176 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 11177 if (err < 0) 11178 goto err_rxq_info; 11179 } 11180 return 0; 11181 11182err_rxq_info: 11183 /* Rollback successful reg's and free other resources */ 11184 while (i--) 11185 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 11186 kvfree(dev->_rx); 11187 dev->_rx = NULL; 11188 return err; 11189} 11190 11191static void netif_free_rx_queues(struct net_device *dev) 11192{ 11193 unsigned int i, count = dev->num_rx_queues; 11194 11195 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 11196 if (!dev->_rx) 11197 return; 11198 11199 for (i = 0; i < count; i++) 11200 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 11201 11202 kvfree(dev->_rx); 11203} 11204 11205static void netdev_init_one_queue(struct net_device *dev, 11206 struct netdev_queue *queue, void *_unused) 11207{ 11208 /* Initialize queue lock */ 11209 spin_lock_init(&queue->_xmit_lock); 11210 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 11211 queue->xmit_lock_owner = -1; 11212 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 11213 queue->dev = dev; 11214#ifdef CONFIG_BQL 11215 dql_init(&queue->dql, HZ); 11216#endif 11217} 11218 11219static void netif_free_tx_queues(struct net_device *dev) 11220{ 11221 kvfree(dev->_tx); 11222} 11223 11224static int netif_alloc_netdev_queues(struct net_device *dev) 11225{ 11226 unsigned int count = dev->num_tx_queues; 11227 struct netdev_queue *tx; 11228 size_t sz = count * sizeof(*tx); 11229 11230 if (count < 1 || count > 0xffff) 11231 return -EINVAL; 11232 11233 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 11234 if (!tx) 11235 return -ENOMEM; 11236 11237 dev->_tx = tx; 11238 11239 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 11240 spin_lock_init(&dev->tx_global_lock); 11241 11242 return 0; 11243} 11244 11245void netif_tx_stop_all_queues(struct net_device *dev) 11246{ 11247 unsigned int i; 11248 11249 for (i = 0; i < dev->num_tx_queues; i++) { 11250 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 11251 11252 netif_tx_stop_queue(txq); 11253 } 11254} 11255EXPORT_SYMBOL(netif_tx_stop_all_queues); 11256 11257static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 11258{ 11259 void __percpu *v; 11260 11261 /* Drivers implementing ndo_get_peer_dev must support tstat 11262 * accounting, so that skb_do_redirect() can bump the dev's 11263 * RX stats upon network namespace switch. 11264 */ 11265 if (dev->netdev_ops->ndo_get_peer_dev && 11266 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 11267 return -EOPNOTSUPP; 11268 11269 switch (dev->pcpu_stat_type) { 11270 case NETDEV_PCPU_STAT_NONE: 11271 return 0; 11272 case NETDEV_PCPU_STAT_LSTATS: 11273 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 11274 break; 11275 case NETDEV_PCPU_STAT_TSTATS: 11276 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 11277 break; 11278 case NETDEV_PCPU_STAT_DSTATS: 11279 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 11280 break; 11281 default: 11282 return -EINVAL; 11283 } 11284 11285 return v ? 0 : -ENOMEM; 11286} 11287 11288static void netdev_do_free_pcpu_stats(struct net_device *dev) 11289{ 11290 switch (dev->pcpu_stat_type) { 11291 case NETDEV_PCPU_STAT_NONE: 11292 return; 11293 case NETDEV_PCPU_STAT_LSTATS: 11294 free_percpu(dev->lstats); 11295 break; 11296 case NETDEV_PCPU_STAT_TSTATS: 11297 free_percpu(dev->tstats); 11298 break; 11299 case NETDEV_PCPU_STAT_DSTATS: 11300 free_percpu(dev->dstats); 11301 break; 11302 } 11303} 11304 11305static void netdev_free_phy_link_topology(struct net_device *dev) 11306{ 11307 struct phy_link_topology *topo = dev->link_topo; 11308 11309 if (IS_ENABLED(CONFIG_PHYLIB) && topo) { 11310 xa_destroy(&topo->phys); 11311 kfree(topo); 11312 dev->link_topo = NULL; 11313 } 11314} 11315 11316/** 11317 * register_netdevice() - register a network device 11318 * @dev: device to register 11319 * 11320 * Take a prepared network device structure and make it externally accessible. 11321 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 11322 * Callers must hold the rtnl lock - you may want register_netdev() 11323 * instead of this. 11324 */ 11325int register_netdevice(struct net_device *dev) 11326{ 11327 int ret; 11328 struct net *net = dev_net(dev); 11329 11330 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 11331 NETDEV_FEATURE_COUNT); 11332 BUG_ON(dev_boot_phase); 11333 ASSERT_RTNL(); 11334 11335 might_sleep(); 11336 11337 /* When net_device's are persistent, this will be fatal. */ 11338 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 11339 BUG_ON(!net); 11340 11341 ret = ethtool_check_ops(dev->ethtool_ops); 11342 if (ret) 11343 return ret; 11344 11345 /* rss ctx ID 0 is reserved for the default context, start from 1 */ 11346 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1); 11347 mutex_init(&dev->ethtool->rss_lock); 11348 11349 spin_lock_init(&dev->addr_list_lock); 11350 netdev_set_addr_lockdep_class(dev); 11351 11352 ret = dev_get_valid_name(net, dev, dev->name); 11353 if (ret < 0) 11354 goto out; 11355 11356 ret = -ENOMEM; 11357 dev->name_node = netdev_name_node_head_alloc(dev); 11358 if (!dev->name_node) 11359 goto out; 11360 11361 /* Init, if this function is available */ 11362 if (dev->netdev_ops->ndo_init) { 11363 ret = dev->netdev_ops->ndo_init(dev); 11364 if (ret) { 11365 if (ret > 0) 11366 ret = -EIO; 11367 goto err_free_name; 11368 } 11369 } 11370 11371 if (((dev->hw_features | dev->features) & 11372 NETIF_F_HW_VLAN_CTAG_FILTER) && 11373 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 11374 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 11375 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 11376 ret = -EINVAL; 11377 goto err_uninit; 11378 } 11379 11380 ret = netdev_do_alloc_pcpu_stats(dev); 11381 if (ret) 11382 goto err_uninit; 11383 11384 ret = dev_index_reserve(net, dev->ifindex); 11385 if (ret < 0) 11386 goto err_free_pcpu; 11387 dev->ifindex = ret; 11388 11389 /* Transfer changeable features to wanted_features and enable 11390 * software offloads (GSO and GRO). 11391 */ 11392 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 11393 dev->features |= NETIF_F_SOFT_FEATURES; 11394 11395 if (dev->udp_tunnel_nic_info) { 11396 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 11397 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 11398 } 11399 11400 dev->wanted_features = dev->features & dev->hw_features; 11401 11402 if (!(dev->flags & IFF_LOOPBACK)) 11403 dev->hw_features |= NETIF_F_NOCACHE_COPY; 11404 11405 /* If IPv4 TCP segmentation offload is supported we should also 11406 * allow the device to enable segmenting the frame with the option 11407 * of ignoring a static IP ID value. This doesn't enable the 11408 * feature itself but allows the user to enable it later. 11409 */ 11410 if (dev->hw_features & NETIF_F_TSO) 11411 dev->hw_features |= NETIF_F_TSO_MANGLEID; 11412 if (dev->vlan_features & NETIF_F_TSO) 11413 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 11414 if (dev->mpls_features & NETIF_F_TSO) 11415 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 11416 if (dev->hw_enc_features & NETIF_F_TSO) 11417 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 11418 11419 /* TSO_MANGLEID belongs in mangleid_features by definition */ 11420 dev->mangleid_features |= NETIF_F_TSO_MANGLEID; 11421 11422 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 11423 */ 11424 dev->vlan_features |= NETIF_F_HIGHDMA; 11425 11426 /* Make NETIF_F_SG inheritable to tunnel devices. 11427 */ 11428 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 11429 11430 /* Make NETIF_F_SG inheritable to MPLS. 11431 */ 11432 dev->mpls_features |= NETIF_F_SG; 11433 11434 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 11435 ret = notifier_to_errno(ret); 11436 if (ret) 11437 goto err_ifindex_release; 11438 11439 ret = netdev_register_kobject(dev); 11440 11441 netdev_lock(dev); 11442 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); 11443 netdev_unlock(dev); 11444 11445 if (ret) 11446 goto err_uninit_notify; 11447 11448 netdev_lock_ops(dev); 11449 __netdev_update_features(dev); 11450 netdev_unlock_ops(dev); 11451 11452 /* 11453 * Default initial state at registry is that the 11454 * device is present. 11455 */ 11456 11457 set_bit(__LINK_STATE_PRESENT, &dev->state); 11458 11459 linkwatch_init_dev(dev); 11460 11461 dev_init_scheduler(dev); 11462 11463 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 11464 list_netdevice(dev); 11465 11466 add_device_randomness(dev->dev_addr, dev->addr_len); 11467 11468 /* If the device has permanent device address, driver should 11469 * set dev_addr and also addr_assign_type should be set to 11470 * NET_ADDR_PERM (default value). 11471 */ 11472 if (dev->addr_assign_type == NET_ADDR_PERM) 11473 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 11474 11475 /* Notify protocols, that a new device appeared. */ 11476 netdev_lock_ops(dev); 11477 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 11478 netdev_unlock_ops(dev); 11479 ret = notifier_to_errno(ret); 11480 if (ret) { 11481 /* Expect explicit free_netdev() on failure */ 11482 dev->needs_free_netdev = false; 11483 unregister_netdevice_queue(dev, NULL); 11484 goto out; 11485 } 11486 /* 11487 * Prevent userspace races by waiting until the network 11488 * device is fully setup before sending notifications. 11489 */ 11490 if (!(dev->rtnl_link_ops && dev->rtnl_link_initializing)) 11491 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11492 11493out: 11494 return ret; 11495 11496err_uninit_notify: 11497 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11498err_ifindex_release: 11499 dev_index_release(net, dev->ifindex); 11500err_free_pcpu: 11501 netdev_do_free_pcpu_stats(dev); 11502err_uninit: 11503 if (dev->netdev_ops->ndo_uninit) 11504 dev->netdev_ops->ndo_uninit(dev); 11505 if (dev->priv_destructor) 11506 dev->priv_destructor(dev); 11507err_free_name: 11508 netdev_name_node_free(dev->name_node); 11509 goto out; 11510} 11511EXPORT_SYMBOL(register_netdevice); 11512 11513/* Initialize the core of a dummy net device. 11514 * The setup steps dummy netdevs need which normal netdevs get by going 11515 * through register_netdevice(). 11516 */ 11517static void init_dummy_netdev(struct net_device *dev) 11518{ 11519 /* make sure we BUG if trying to hit standard 11520 * register/unregister code path 11521 */ 11522 dev->reg_state = NETREG_DUMMY; 11523 11524 /* a dummy interface is started by default */ 11525 set_bit(__LINK_STATE_PRESENT, &dev->state); 11526 set_bit(__LINK_STATE_START, &dev->state); 11527 11528 /* Note : We dont allocate pcpu_refcnt for dummy devices, 11529 * because users of this 'device' dont need to change 11530 * its refcount. 11531 */ 11532} 11533 11534/** 11535 * register_netdev - register a network device 11536 * @dev: device to register 11537 * 11538 * Take a completed network device structure and add it to the kernel 11539 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 11540 * chain. 0 is returned on success. A negative errno code is returned 11541 * on a failure to set up the device, or if the name is a duplicate. 11542 * 11543 * This is a wrapper around register_netdevice that takes the rtnl semaphore 11544 * and expands the device name if you passed a format string to 11545 * alloc_netdev. 11546 */ 11547int register_netdev(struct net_device *dev) 11548{ 11549 struct net *net = dev_net(dev); 11550 int err; 11551 11552 if (rtnl_net_lock_killable(net)) 11553 return -EINTR; 11554 11555 err = register_netdevice(dev); 11556 11557 rtnl_net_unlock(net); 11558 11559 return err; 11560} 11561EXPORT_SYMBOL(register_netdev); 11562 11563int netdev_refcnt_read(const struct net_device *dev) 11564{ 11565#ifdef CONFIG_PCPU_DEV_REFCNT 11566 int i, refcnt = 0; 11567 11568 for_each_possible_cpu(i) 11569 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 11570 return refcnt; 11571#else 11572 return refcount_read(&dev->dev_refcnt); 11573#endif 11574} 11575EXPORT_SYMBOL(netdev_refcnt_read); 11576 11577int netdev_unregister_timeout_secs __read_mostly = 10; 11578 11579#define WAIT_REFS_MIN_MSECS 1 11580#define WAIT_REFS_MAX_MSECS 250 11581/** 11582 * netdev_wait_allrefs_any - wait until all references are gone. 11583 * @list: list of net_devices to wait on 11584 * 11585 * This is called when unregistering network devices. 11586 * 11587 * Any protocol or device that holds a reference should register 11588 * for netdevice notification, and cleanup and put back the 11589 * reference if they receive an UNREGISTER event. 11590 * We can get stuck here if buggy protocols don't correctly 11591 * call dev_put. 11592 */ 11593static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 11594{ 11595 unsigned long rebroadcast_time, warning_time; 11596 struct net_device *dev; 11597 int wait = 0; 11598 11599 rebroadcast_time = warning_time = jiffies; 11600 11601 list_for_each_entry(dev, list, todo_list) 11602 if (netdev_refcnt_read(dev) == 1) 11603 return dev; 11604 11605 while (true) { 11606 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 11607 rtnl_lock(); 11608 11609 /* Rebroadcast unregister notification */ 11610 list_for_each_entry(dev, list, todo_list) 11611 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11612 11613 __rtnl_unlock(); 11614 rcu_barrier(); 11615 rtnl_lock(); 11616 11617 list_for_each_entry(dev, list, todo_list) 11618 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 11619 &dev->state)) { 11620 /* We must not have linkwatch events 11621 * pending on unregister. If this 11622 * happens, we simply run the queue 11623 * unscheduled, resulting in a noop 11624 * for this device. 11625 */ 11626 linkwatch_run_queue(); 11627 break; 11628 } 11629 11630 __rtnl_unlock(); 11631 11632 rebroadcast_time = jiffies; 11633 } 11634 11635 rcu_barrier(); 11636 11637 if (!wait) { 11638 wait = WAIT_REFS_MIN_MSECS; 11639 } else { 11640 msleep(wait); 11641 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 11642 } 11643 11644 list_for_each_entry(dev, list, todo_list) 11645 if (netdev_refcnt_read(dev) == 1) 11646 return dev; 11647 11648 if (time_after(jiffies, warning_time + 11649 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 11650 list_for_each_entry(dev, list, todo_list) { 11651 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 11652 dev->name, netdev_refcnt_read(dev)); 11653 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 11654 } 11655 11656 warning_time = jiffies; 11657 } 11658 } 11659} 11660 11661/* The sequence is: 11662 * 11663 * rtnl_lock(); 11664 * ... 11665 * register_netdevice(x1); 11666 * register_netdevice(x2); 11667 * ... 11668 * unregister_netdevice(y1); 11669 * unregister_netdevice(y2); 11670 * ... 11671 * rtnl_unlock(); 11672 * free_netdev(y1); 11673 * free_netdev(y2); 11674 * 11675 * We are invoked by rtnl_unlock(). 11676 * This allows us to deal with problems: 11677 * 1) We can delete sysfs objects which invoke hotplug 11678 * without deadlocking with linkwatch via keventd. 11679 * 2) Since we run with the RTNL semaphore not held, we can sleep 11680 * safely in order to wait for the netdev refcnt to drop to zero. 11681 * 11682 * We must not return until all unregister events added during 11683 * the interval the lock was held have been completed. 11684 */ 11685void netdev_run_todo(void) 11686{ 11687 struct net_device *dev, *tmp; 11688 struct list_head list; 11689 int cnt; 11690#ifdef CONFIG_LOCKDEP 11691 struct list_head unlink_list; 11692 11693 list_replace_init(&net_unlink_list, &unlink_list); 11694 11695 while (!list_empty(&unlink_list)) { 11696 dev = list_first_entry(&unlink_list, struct net_device, 11697 unlink_list); 11698 list_del_init(&dev->unlink_list); 11699 dev->nested_level = dev->lower_level - 1; 11700 } 11701#endif 11702 11703 /* Snapshot list, allow later requests */ 11704 list_replace_init(&net_todo_list, &list); 11705 11706 __rtnl_unlock(); 11707 11708 /* Wait for rcu callbacks to finish before next phase */ 11709 if (!list_empty(&list)) 11710 rcu_barrier(); 11711 11712 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 11713 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 11714 netdev_WARN(dev, "run_todo but not unregistering\n"); 11715 list_del(&dev->todo_list); 11716 continue; 11717 } 11718 11719 netdev_lock(dev); 11720 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); 11721 netdev_unlock(dev); 11722 linkwatch_sync_dev(dev); 11723 } 11724 11725 cnt = 0; 11726 while (!list_empty(&list)) { 11727 dev = netdev_wait_allrefs_any(&list); 11728 list_del(&dev->todo_list); 11729 11730 /* paranoia */ 11731 BUG_ON(netdev_refcnt_read(dev) != 1); 11732 BUG_ON(!list_empty(&dev->ptype_all)); 11733 BUG_ON(!list_empty(&dev->ptype_specific)); 11734 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 11735 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 11736 11737 netdev_do_free_pcpu_stats(dev); 11738 if (dev->priv_destructor) 11739 dev->priv_destructor(dev); 11740 if (dev->needs_free_netdev) 11741 free_netdev(dev); 11742 11743 cnt++; 11744 11745 /* Free network device */ 11746 kobject_put(&dev->dev.kobj); 11747 } 11748 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) 11749 wake_up(&netdev_unregistering_wq); 11750} 11751 11752/* Collate per-cpu network dstats statistics 11753 * 11754 * Read per-cpu network statistics from dev->dstats and populate the related 11755 * fields in @s. 11756 */ 11757static void dev_fetch_dstats(struct rtnl_link_stats64 *s, 11758 const struct pcpu_dstats __percpu *dstats) 11759{ 11760 int cpu; 11761 11762 for_each_possible_cpu(cpu) { 11763 u64 rx_packets, rx_bytes, rx_drops; 11764 u64 tx_packets, tx_bytes, tx_drops; 11765 const struct pcpu_dstats *stats; 11766 unsigned int start; 11767 11768 stats = per_cpu_ptr(dstats, cpu); 11769 do { 11770 start = u64_stats_fetch_begin(&stats->syncp); 11771 rx_packets = u64_stats_read(&stats->rx_packets); 11772 rx_bytes = u64_stats_read(&stats->rx_bytes); 11773 rx_drops = u64_stats_read(&stats->rx_drops); 11774 tx_packets = u64_stats_read(&stats->tx_packets); 11775 tx_bytes = u64_stats_read(&stats->tx_bytes); 11776 tx_drops = u64_stats_read(&stats->tx_drops); 11777 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11778 11779 s->rx_packets += rx_packets; 11780 s->rx_bytes += rx_bytes; 11781 s->rx_dropped += rx_drops; 11782 s->tx_packets += tx_packets; 11783 s->tx_bytes += tx_bytes; 11784 s->tx_dropped += tx_drops; 11785 } 11786} 11787 11788/* ndo_get_stats64 implementation for dtstats-based accounting. 11789 * 11790 * Populate @s from dev->stats and dev->dstats. This is used internally by the 11791 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection. 11792 */ 11793static void dev_get_dstats64(const struct net_device *dev, 11794 struct rtnl_link_stats64 *s) 11795{ 11796 netdev_stats_to_stats64(s, &dev->stats); 11797 dev_fetch_dstats(s, dev->dstats); 11798} 11799 11800/* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 11801 * all the same fields in the same order as net_device_stats, with only 11802 * the type differing, but rtnl_link_stats64 may have additional fields 11803 * at the end for newer counters. 11804 */ 11805void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 11806 const struct net_device_stats *netdev_stats) 11807{ 11808 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 11809 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 11810 u64 *dst = (u64 *)stats64; 11811 11812 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 11813 for (i = 0; i < n; i++) 11814 dst[i] = (unsigned long)atomic_long_read(&src[i]); 11815 /* zero out counters that only exist in rtnl_link_stats64 */ 11816 memset((char *)stats64 + n * sizeof(u64), 0, 11817 sizeof(*stats64) - n * sizeof(u64)); 11818} 11819EXPORT_SYMBOL(netdev_stats_to_stats64); 11820 11821static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 11822 struct net_device *dev) 11823{ 11824 struct net_device_core_stats __percpu *p; 11825 11826 p = alloc_percpu_gfp(struct net_device_core_stats, 11827 GFP_ATOMIC | __GFP_NOWARN); 11828 11829 if (p && cmpxchg(&dev->core_stats, NULL, p)) 11830 free_percpu(p); 11831 11832 /* This READ_ONCE() pairs with the cmpxchg() above */ 11833 return READ_ONCE(dev->core_stats); 11834} 11835 11836noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 11837{ 11838 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11839 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 11840 unsigned long __percpu *field; 11841 11842 if (unlikely(!p)) { 11843 p = netdev_core_stats_alloc(dev); 11844 if (!p) 11845 return; 11846 } 11847 11848 field = (unsigned long __percpu *)((void __percpu *)p + offset); 11849 this_cpu_inc(*field); 11850} 11851EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 11852 11853/** 11854 * dev_get_stats - get network device statistics 11855 * @dev: device to get statistics from 11856 * @storage: place to store stats 11857 * 11858 * Get network statistics from device. Return @storage. 11859 * The device driver may provide its own method by setting 11860 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 11861 * otherwise the internal statistics structure is used. 11862 */ 11863struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 11864 struct rtnl_link_stats64 *storage) 11865{ 11866 const struct net_device_ops *ops = dev->netdev_ops; 11867 const struct net_device_core_stats __percpu *p; 11868 11869 /* 11870 * IPv{4,6} and udp tunnels share common stat helpers and use 11871 * different stat type (NETDEV_PCPU_STAT_TSTATS vs 11872 * NETDEV_PCPU_STAT_DSTATS). Ensure the accounting is consistent. 11873 */ 11874 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, rx_bytes) != 11875 offsetof(struct pcpu_dstats, rx_bytes)); 11876 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, rx_packets) != 11877 offsetof(struct pcpu_dstats, rx_packets)); 11878 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, tx_bytes) != 11879 offsetof(struct pcpu_dstats, tx_bytes)); 11880 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, tx_packets) != 11881 offsetof(struct pcpu_dstats, tx_packets)); 11882 11883 if (ops->ndo_get_stats64) { 11884 memset(storage, 0, sizeof(*storage)); 11885 ops->ndo_get_stats64(dev, storage); 11886 } else if (ops->ndo_get_stats) { 11887 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 11888 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { 11889 dev_get_tstats64(dev, storage); 11890 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) { 11891 dev_get_dstats64(dev, storage); 11892 } else { 11893 netdev_stats_to_stats64(storage, &dev->stats); 11894 } 11895 11896 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11897 p = READ_ONCE(dev->core_stats); 11898 if (p) { 11899 const struct net_device_core_stats *core_stats; 11900 int i; 11901 11902 for_each_possible_cpu(i) { 11903 core_stats = per_cpu_ptr(p, i); 11904 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 11905 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 11906 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 11907 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 11908 } 11909 } 11910 return storage; 11911} 11912EXPORT_SYMBOL(dev_get_stats); 11913 11914/** 11915 * dev_fetch_sw_netstats - get per-cpu network device statistics 11916 * @s: place to store stats 11917 * @netstats: per-cpu network stats to read from 11918 * 11919 * Read per-cpu network statistics and populate the related fields in @s. 11920 */ 11921void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 11922 const struct pcpu_sw_netstats __percpu *netstats) 11923{ 11924 int cpu; 11925 11926 for_each_possible_cpu(cpu) { 11927 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 11928 const struct pcpu_sw_netstats *stats; 11929 unsigned int start; 11930 11931 stats = per_cpu_ptr(netstats, cpu); 11932 do { 11933 start = u64_stats_fetch_begin(&stats->syncp); 11934 rx_packets = u64_stats_read(&stats->rx_packets); 11935 rx_bytes = u64_stats_read(&stats->rx_bytes); 11936 tx_packets = u64_stats_read(&stats->tx_packets); 11937 tx_bytes = u64_stats_read(&stats->tx_bytes); 11938 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11939 11940 s->rx_packets += rx_packets; 11941 s->rx_bytes += rx_bytes; 11942 s->tx_packets += tx_packets; 11943 s->tx_bytes += tx_bytes; 11944 } 11945} 11946EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 11947 11948/** 11949 * dev_get_tstats64 - ndo_get_stats64 implementation 11950 * @dev: device to get statistics from 11951 * @s: place to store stats 11952 * 11953 * Populate @s from dev->stats and dev->tstats. Can be used as 11954 * ndo_get_stats64() callback. 11955 */ 11956void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 11957{ 11958 netdev_stats_to_stats64(s, &dev->stats); 11959 dev_fetch_sw_netstats(s, dev->tstats); 11960} 11961EXPORT_SYMBOL_GPL(dev_get_tstats64); 11962 11963struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 11964{ 11965 struct netdev_queue *queue = dev_ingress_queue(dev); 11966 11967#ifdef CONFIG_NET_CLS_ACT 11968 if (queue) 11969 return queue; 11970 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 11971 if (!queue) 11972 return NULL; 11973 netdev_init_one_queue(dev, queue, NULL); 11974 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 11975 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 11976 rcu_assign_pointer(dev->ingress_queue, queue); 11977#endif 11978 return queue; 11979} 11980 11981static const struct ethtool_ops default_ethtool_ops; 11982 11983void netdev_set_default_ethtool_ops(struct net_device *dev, 11984 const struct ethtool_ops *ops) 11985{ 11986 if (dev->ethtool_ops == &default_ethtool_ops) 11987 dev->ethtool_ops = ops; 11988} 11989EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 11990 11991/** 11992 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 11993 * @dev: netdev to enable the IRQ coalescing on 11994 * 11995 * Sets a conservative default for SW IRQ coalescing. Users can use 11996 * sysfs attributes to override the default values. 11997 */ 11998void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 11999{ 12000 WARN_ON(dev->reg_state == NETREG_REGISTERED); 12001 12002 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 12003 netdev_set_gro_flush_timeout(dev, 20000); 12004 netdev_set_defer_hard_irqs(dev, 1); 12005 } 12006} 12007EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 12008 12009/** 12010 * alloc_netdev_mqs - allocate network device 12011 * @sizeof_priv: size of private data to allocate space for 12012 * @name: device name format string 12013 * @name_assign_type: origin of device name 12014 * @setup: callback to initialize device 12015 * @txqs: the number of TX subqueues to allocate 12016 * @rxqs: the number of RX subqueues to allocate 12017 * 12018 * Allocates a struct net_device with private data area for driver use 12019 * and performs basic initialization. Also allocates subqueue structs 12020 * for each queue on the device. 12021 */ 12022struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 12023 unsigned char name_assign_type, 12024 void (*setup)(struct net_device *), 12025 unsigned int txqs, unsigned int rxqs) 12026{ 12027 struct net_device *dev; 12028 size_t napi_config_sz; 12029 unsigned int maxqs; 12030 12031 BUG_ON(strlen(name) >= sizeof(dev->name)); 12032 12033 if (txqs < 1) { 12034 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 12035 return NULL; 12036 } 12037 12038 if (rxqs < 1) { 12039 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 12040 return NULL; 12041 } 12042 12043 maxqs = max(txqs, rxqs); 12044 12045 dev = kvzalloc(struct_size(dev, priv, sizeof_priv), 12046 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 12047 if (!dev) 12048 return NULL; 12049 12050 dev->priv_len = sizeof_priv; 12051 12052 ref_tracker_dir_init(&dev->refcnt_tracker, 128, "netdev"); 12053#ifdef CONFIG_PCPU_DEV_REFCNT 12054 dev->pcpu_refcnt = alloc_percpu(int); 12055 if (!dev->pcpu_refcnt) 12056 goto free_dev; 12057 __dev_hold(dev); 12058#else 12059 refcount_set(&dev->dev_refcnt, 1); 12060#endif 12061 12062 if (dev_addr_init(dev)) 12063 goto free_pcpu; 12064 12065 dev_mc_init(dev); 12066 dev_uc_init(dev); 12067 12068 dev_net_set(dev, &init_net); 12069 12070 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 12071 dev->xdp_zc_max_segs = 1; 12072 dev->gso_max_segs = GSO_MAX_SEGS; 12073 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 12074 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 12075 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 12076 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 12077 dev->tso_max_segs = TSO_MAX_SEGS; 12078 dev->upper_level = 1; 12079 dev->lower_level = 1; 12080#ifdef CONFIG_LOCKDEP 12081 dev->nested_level = 0; 12082 INIT_LIST_HEAD(&dev->unlink_list); 12083#endif 12084 12085 INIT_LIST_HEAD(&dev->napi_list); 12086 INIT_LIST_HEAD(&dev->unreg_list); 12087 INIT_LIST_HEAD(&dev->close_list); 12088 INIT_LIST_HEAD(&dev->link_watch_list); 12089 INIT_LIST_HEAD(&dev->adj_list.upper); 12090 INIT_LIST_HEAD(&dev->adj_list.lower); 12091 INIT_LIST_HEAD(&dev->ptype_all); 12092 INIT_LIST_HEAD(&dev->ptype_specific); 12093 INIT_LIST_HEAD(&dev->net_notifier_list); 12094#ifdef CONFIG_NET_SCHED 12095 hash_init(dev->qdisc_hash); 12096#endif 12097 12098 mutex_init(&dev->lock); 12099 12100 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 12101 setup(dev); 12102 12103 if (!dev->tx_queue_len) { 12104 dev->priv_flags |= IFF_NO_QUEUE; 12105 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 12106 } 12107 12108 dev->num_tx_queues = txqs; 12109 dev->real_num_tx_queues = txqs; 12110 if (netif_alloc_netdev_queues(dev)) 12111 goto free_all; 12112 12113 dev->num_rx_queues = rxqs; 12114 dev->real_num_rx_queues = rxqs; 12115 if (netif_alloc_rx_queues(dev)) 12116 goto free_all; 12117 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT); 12118 if (!dev->ethtool) 12119 goto free_all; 12120 12121 dev->cfg = kzalloc(sizeof(*dev->cfg), GFP_KERNEL_ACCOUNT); 12122 if (!dev->cfg) 12123 goto free_all; 12124 dev->cfg_pending = dev->cfg; 12125 12126 dev->num_napi_configs = maxqs; 12127 napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config)); 12128 dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT); 12129 if (!dev->napi_config) 12130 goto free_all; 12131 12132 strscpy(dev->name, name); 12133 dev->name_assign_type = name_assign_type; 12134 dev->group = INIT_NETDEV_GROUP; 12135 if (!dev->ethtool_ops) 12136 dev->ethtool_ops = &default_ethtool_ops; 12137 12138 nf_hook_netdev_init(dev); 12139 12140 return dev; 12141 12142free_all: 12143 free_netdev(dev); 12144 return NULL; 12145 12146free_pcpu: 12147#ifdef CONFIG_PCPU_DEV_REFCNT 12148 free_percpu(dev->pcpu_refcnt); 12149free_dev: 12150#endif 12151 kvfree(dev); 12152 return NULL; 12153} 12154EXPORT_SYMBOL(alloc_netdev_mqs); 12155 12156static void netdev_napi_exit(struct net_device *dev) 12157{ 12158 if (!list_empty(&dev->napi_list)) { 12159 struct napi_struct *p, *n; 12160 12161 netdev_lock(dev); 12162 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 12163 __netif_napi_del_locked(p); 12164 netdev_unlock(dev); 12165 12166 synchronize_net(); 12167 } 12168 12169 kvfree(dev->napi_config); 12170} 12171 12172/** 12173 * free_netdev - free network device 12174 * @dev: device 12175 * 12176 * This function does the last stage of destroying an allocated device 12177 * interface. The reference to the device object is released. If this 12178 * is the last reference then it will be freed.Must be called in process 12179 * context. 12180 */ 12181void free_netdev(struct net_device *dev) 12182{ 12183 might_sleep(); 12184 12185 /* When called immediately after register_netdevice() failed the unwind 12186 * handling may still be dismantling the device. Handle that case by 12187 * deferring the free. 12188 */ 12189 if (dev->reg_state == NETREG_UNREGISTERING) { 12190 ASSERT_RTNL(); 12191 dev->needs_free_netdev = true; 12192 return; 12193 } 12194 12195 WARN_ON(dev->cfg != dev->cfg_pending); 12196 kfree(dev->cfg); 12197 kfree(dev->ethtool); 12198 netif_free_tx_queues(dev); 12199 netif_free_rx_queues(dev); 12200 12201 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 12202 12203 /* Flush device addresses */ 12204 dev_addr_flush(dev); 12205 12206 netdev_napi_exit(dev); 12207 12208 netif_del_cpu_rmap(dev); 12209 12210 ref_tracker_dir_exit(&dev->refcnt_tracker); 12211#ifdef CONFIG_PCPU_DEV_REFCNT 12212 free_percpu(dev->pcpu_refcnt); 12213 dev->pcpu_refcnt = NULL; 12214#endif 12215 free_percpu(dev->core_stats); 12216 dev->core_stats = NULL; 12217 free_percpu(dev->xdp_bulkq); 12218 dev->xdp_bulkq = NULL; 12219 12220 netdev_free_phy_link_topology(dev); 12221 12222 mutex_destroy(&dev->lock); 12223 12224 /* Compatibility with error handling in drivers */ 12225 if (dev->reg_state == NETREG_UNINITIALIZED || 12226 dev->reg_state == NETREG_DUMMY) { 12227 kvfree(dev); 12228 return; 12229 } 12230 12231 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 12232 WRITE_ONCE(dev->reg_state, NETREG_RELEASED); 12233 12234 /* will free via device release */ 12235 put_device(&dev->dev); 12236} 12237EXPORT_SYMBOL(free_netdev); 12238 12239/** 12240 * alloc_netdev_dummy - Allocate and initialize a dummy net device. 12241 * @sizeof_priv: size of private data to allocate space for 12242 * 12243 * Return: the allocated net_device on success, NULL otherwise 12244 */ 12245struct net_device *alloc_netdev_dummy(int sizeof_priv) 12246{ 12247 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN, 12248 init_dummy_netdev); 12249} 12250EXPORT_SYMBOL_GPL(alloc_netdev_dummy); 12251 12252/** 12253 * synchronize_net - Synchronize with packet receive processing 12254 * 12255 * Wait for packets currently being received to be done. 12256 * Does not block later packets from starting. 12257 */ 12258void synchronize_net(void) 12259{ 12260 might_sleep(); 12261 if (from_cleanup_net() || rtnl_is_locked()) 12262 synchronize_rcu_expedited(); 12263 else 12264 synchronize_rcu(); 12265} 12266EXPORT_SYMBOL(synchronize_net); 12267 12268static void netdev_rss_contexts_free(struct net_device *dev) 12269{ 12270 struct ethtool_rxfh_context *ctx; 12271 unsigned long context; 12272 12273 mutex_lock(&dev->ethtool->rss_lock); 12274 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) { 12275 xa_erase(&dev->ethtool->rss_ctx, context); 12276 dev->ethtool_ops->remove_rxfh_context(dev, ctx, context, NULL); 12277 kfree(ctx); 12278 } 12279 xa_destroy(&dev->ethtool->rss_ctx); 12280 mutex_unlock(&dev->ethtool->rss_lock); 12281} 12282 12283/** 12284 * unregister_netdevice_queue - remove device from the kernel 12285 * @dev: device 12286 * @head: list 12287 * 12288 * This function shuts down a device interface and removes it 12289 * from the kernel tables. 12290 * If head not NULL, device is queued to be unregistered later. 12291 * 12292 * Callers must hold the rtnl semaphore. You may want 12293 * unregister_netdev() instead of this. 12294 */ 12295 12296void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 12297{ 12298 ASSERT_RTNL(); 12299 12300 if (head) { 12301 list_move_tail(&dev->unreg_list, head); 12302 } else { 12303 LIST_HEAD(single); 12304 12305 list_add(&dev->unreg_list, &single); 12306 unregister_netdevice_many(&single); 12307 } 12308} 12309EXPORT_SYMBOL(unregister_netdevice_queue); 12310 12311static void dev_memory_provider_uninstall(struct net_device *dev) 12312{ 12313 unsigned int i; 12314 12315 for (i = 0; i < dev->real_num_rx_queues; i++) { 12316 struct netdev_rx_queue *rxq = &dev->_rx[i]; 12317 struct pp_memory_provider_params *p = &rxq->mp_params; 12318 12319 if (p->mp_ops && p->mp_ops->uninstall) 12320 p->mp_ops->uninstall(rxq->mp_params.mp_priv, rxq); 12321 } 12322} 12323 12324/* devices must be UP and netdev_lock()'d */ 12325static void netif_close_many_and_unlock(struct list_head *close_head) 12326{ 12327 struct net_device *dev, *tmp; 12328 12329 netif_close_many(close_head, false); 12330 12331 /* ... now unlock them */ 12332 list_for_each_entry_safe(dev, tmp, close_head, close_list) { 12333 netdev_unlock(dev); 12334 list_del_init(&dev->close_list); 12335 } 12336} 12337 12338static void netif_close_many_and_unlock_cond(struct list_head *close_head) 12339{ 12340#ifdef CONFIG_LOCKDEP 12341 /* We can only track up to MAX_LOCK_DEPTH locks per task. 12342 * 12343 * Reserve half the available slots for additional locks possibly 12344 * taken by notifiers and (soft)irqs. 12345 */ 12346 unsigned int limit = MAX_LOCK_DEPTH / 2; 12347 12348 if (lockdep_depth(current) > limit) 12349 netif_close_many_and_unlock(close_head); 12350#endif 12351} 12352 12353void unregister_netdevice_many_notify(struct list_head *head, 12354 u32 portid, const struct nlmsghdr *nlh) 12355{ 12356 struct net_device *dev, *tmp; 12357 LIST_HEAD(close_head); 12358 int cnt = 0; 12359 12360 BUG_ON(dev_boot_phase); 12361 ASSERT_RTNL(); 12362 12363 if (list_empty(head)) 12364 return; 12365 12366 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 12367 /* Some devices call without registering 12368 * for initialization unwind. Remove those 12369 * devices and proceed with the remaining. 12370 */ 12371 if (dev->reg_state == NETREG_UNINITIALIZED) { 12372 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 12373 dev->name, dev); 12374 12375 WARN_ON(1); 12376 list_del(&dev->unreg_list); 12377 continue; 12378 } 12379 dev->dismantle = true; 12380 BUG_ON(dev->reg_state != NETREG_REGISTERED); 12381 } 12382 12383 /* If device is running, close it first. Start with ops locked... */ 12384 list_for_each_entry(dev, head, unreg_list) { 12385 if (!(dev->flags & IFF_UP)) 12386 continue; 12387 if (netdev_need_ops_lock(dev)) { 12388 list_add_tail(&dev->close_list, &close_head); 12389 netdev_lock(dev); 12390 } 12391 netif_close_many_and_unlock_cond(&close_head); 12392 } 12393 netif_close_many_and_unlock(&close_head); 12394 /* ... now go over the rest. */ 12395 list_for_each_entry(dev, head, unreg_list) { 12396 if (!netdev_need_ops_lock(dev)) 12397 list_add_tail(&dev->close_list, &close_head); 12398 } 12399 netif_close_many(&close_head, true); 12400 12401 list_for_each_entry(dev, head, unreg_list) { 12402 /* And unlink it from device chain. */ 12403 unlist_netdevice(dev); 12404 netdev_lock(dev); 12405 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); 12406 netdev_unlock(dev); 12407 } 12408 flush_all_backlogs(); 12409 12410 synchronize_net(); 12411 12412 list_for_each_entry(dev, head, unreg_list) { 12413 struct sk_buff *skb = NULL; 12414 12415 /* Shutdown queueing discipline. */ 12416 netdev_lock_ops(dev); 12417 dev_shutdown(dev); 12418 dev_tcx_uninstall(dev); 12419 dev_xdp_uninstall(dev); 12420 dev_memory_provider_uninstall(dev); 12421 netdev_unlock_ops(dev); 12422 bpf_dev_bound_netdev_unregister(dev); 12423 12424 netdev_offload_xstats_disable_all(dev); 12425 12426 /* Notify protocols, that we are about to destroy 12427 * this device. They should clean all the things. 12428 */ 12429 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 12430 12431 if (!(dev->rtnl_link_ops && dev->rtnl_link_initializing)) 12432 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 12433 GFP_KERNEL, NULL, 0, 12434 portid, nlh); 12435 12436 /* 12437 * Flush the unicast and multicast chains 12438 */ 12439 dev_uc_flush(dev); 12440 dev_mc_flush(dev); 12441 12442 netdev_name_node_alt_flush(dev); 12443 netdev_name_node_free(dev->name_node); 12444 12445 netdev_rss_contexts_free(dev); 12446 12447 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 12448 12449 if (dev->netdev_ops->ndo_uninit) 12450 dev->netdev_ops->ndo_uninit(dev); 12451 12452 mutex_destroy(&dev->ethtool->rss_lock); 12453 12454 net_shaper_flush_netdev(dev); 12455 12456 if (skb) 12457 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 12458 12459 /* Notifier chain MUST detach us all upper devices. */ 12460 WARN_ON(netdev_has_any_upper_dev(dev)); 12461 WARN_ON(netdev_has_any_lower_dev(dev)); 12462 12463 /* Remove entries from kobject tree */ 12464 netdev_unregister_kobject(dev); 12465#ifdef CONFIG_XPS 12466 /* Remove XPS queueing entries */ 12467 netif_reset_xps_queues_gt(dev, 0); 12468#endif 12469 } 12470 12471 synchronize_net(); 12472 12473 list_for_each_entry(dev, head, unreg_list) { 12474 netdev_put(dev, &dev->dev_registered_tracker); 12475 net_set_todo(dev); 12476 cnt++; 12477 } 12478 atomic_add(cnt, &dev_unreg_count); 12479 12480 list_del(head); 12481} 12482 12483/** 12484 * unregister_netdevice_many - unregister many devices 12485 * @head: list of devices 12486 * 12487 * Note: As most callers use a stack allocated list_head, 12488 * we force a list_del() to make sure stack won't be corrupted later. 12489 */ 12490void unregister_netdevice_many(struct list_head *head) 12491{ 12492 unregister_netdevice_many_notify(head, 0, NULL); 12493} 12494EXPORT_SYMBOL(unregister_netdevice_many); 12495 12496/** 12497 * unregister_netdev - remove device from the kernel 12498 * @dev: device 12499 * 12500 * This function shuts down a device interface and removes it 12501 * from the kernel tables. 12502 * 12503 * This is just a wrapper for unregister_netdevice that takes 12504 * the rtnl semaphore. In general you want to use this and not 12505 * unregister_netdevice. 12506 */ 12507void unregister_netdev(struct net_device *dev) 12508{ 12509 rtnl_net_dev_lock(dev); 12510 unregister_netdevice(dev); 12511 rtnl_net_dev_unlock(dev); 12512} 12513EXPORT_SYMBOL(unregister_netdev); 12514 12515int __dev_change_net_namespace(struct net_device *dev, struct net *net, 12516 const char *pat, int new_ifindex, 12517 struct netlink_ext_ack *extack) 12518{ 12519 struct netdev_name_node *name_node; 12520 struct net *net_old = dev_net(dev); 12521 char new_name[IFNAMSIZ] = {}; 12522 int err, new_nsid; 12523 12524 ASSERT_RTNL(); 12525 12526 /* Don't allow namespace local devices to be moved. */ 12527 err = -EINVAL; 12528 if (dev->netns_immutable) { 12529 NL_SET_ERR_MSG(extack, "The interface netns is immutable"); 12530 goto out; 12531 } 12532 12533 /* Ensure the device has been registered */ 12534 if (dev->reg_state != NETREG_REGISTERED) { 12535 NL_SET_ERR_MSG(extack, "The interface isn't registered"); 12536 goto out; 12537 } 12538 12539 /* Get out if there is nothing todo */ 12540 err = 0; 12541 if (net_eq(net_old, net)) 12542 goto out; 12543 12544 /* Pick the destination device name, and ensure 12545 * we can use it in the destination network namespace. 12546 */ 12547 err = -EEXIST; 12548 if (netdev_name_in_use(net, dev->name)) { 12549 /* We get here if we can't use the current device name */ 12550 if (!pat) { 12551 NL_SET_ERR_MSG(extack, 12552 "An interface with the same name exists in the target netns"); 12553 goto out; 12554 } 12555 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 12556 if (err < 0) { 12557 NL_SET_ERR_MSG_FMT(extack, 12558 "Unable to use '%s' for the new interface name in the target netns", 12559 pat); 12560 goto out; 12561 } 12562 } 12563 /* Check that none of the altnames conflicts. */ 12564 err = -EEXIST; 12565 netdev_for_each_altname(dev, name_node) { 12566 if (netdev_name_in_use(net, name_node->name)) { 12567 NL_SET_ERR_MSG_FMT(extack, 12568 "An interface with the altname %s exists in the target netns", 12569 name_node->name); 12570 goto out; 12571 } 12572 } 12573 12574 /* Check that new_ifindex isn't used yet. */ 12575 if (new_ifindex) { 12576 err = dev_index_reserve(net, new_ifindex); 12577 if (err < 0) { 12578 NL_SET_ERR_MSG_FMT(extack, 12579 "The ifindex %d is not available in the target netns", 12580 new_ifindex); 12581 goto out; 12582 } 12583 } else { 12584 /* If there is an ifindex conflict assign a new one */ 12585 err = dev_index_reserve(net, dev->ifindex); 12586 if (err == -EBUSY) 12587 err = dev_index_reserve(net, 0); 12588 if (err < 0) { 12589 NL_SET_ERR_MSG(extack, 12590 "Unable to allocate a new ifindex in the target netns"); 12591 goto out; 12592 } 12593 new_ifindex = err; 12594 } 12595 12596 /* 12597 * And now a mini version of register_netdevice unregister_netdevice. 12598 */ 12599 12600 netdev_lock_ops(dev); 12601 /* If device is running close it first. */ 12602 netif_close(dev); 12603 /* And unlink it from device chain */ 12604 unlist_netdevice(dev); 12605 12606 if (!netdev_need_ops_lock(dev)) 12607 netdev_lock(dev); 12608 dev->moving_ns = true; 12609 netdev_unlock(dev); 12610 12611 synchronize_net(); 12612 12613 /* Shutdown queueing discipline. */ 12614 netdev_lock_ops(dev); 12615 dev_shutdown(dev); 12616 netdev_unlock_ops(dev); 12617 12618 /* Notify protocols, that we are about to destroy 12619 * this device. They should clean all the things. 12620 * 12621 * Note that dev->reg_state stays at NETREG_REGISTERED. 12622 * This is wanted because this way 8021q and macvlan know 12623 * the device is just moving and can keep their slaves up. 12624 */ 12625 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 12626 rcu_barrier(); 12627 12628 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 12629 12630 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 12631 new_ifindex); 12632 12633 /* 12634 * Flush the unicast and multicast chains 12635 */ 12636 dev_uc_flush(dev); 12637 dev_mc_flush(dev); 12638 12639 /* Send a netdev-removed uevent to the old namespace */ 12640 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 12641 netdev_adjacent_del_links(dev); 12642 12643 /* Move per-net netdevice notifiers that are following the netdevice */ 12644 move_netdevice_notifiers_dev_net(dev, net); 12645 12646 /* Actually switch the network namespace */ 12647 netdev_lock(dev); 12648 dev_net_set(dev, net); 12649 netdev_unlock(dev); 12650 dev->ifindex = new_ifindex; 12651 12652 if (new_name[0]) { 12653 /* Rename the netdev to prepared name */ 12654 write_seqlock_bh(&netdev_rename_lock); 12655 strscpy(dev->name, new_name, IFNAMSIZ); 12656 write_sequnlock_bh(&netdev_rename_lock); 12657 } 12658 12659 /* Fixup kobjects */ 12660 dev_set_uevent_suppress(&dev->dev, 1); 12661 err = device_rename(&dev->dev, dev->name); 12662 dev_set_uevent_suppress(&dev->dev, 0); 12663 WARN_ON(err); 12664 12665 /* Send a netdev-add uevent to the new namespace */ 12666 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 12667 netdev_adjacent_add_links(dev); 12668 12669 /* Adapt owner in case owning user namespace of target network 12670 * namespace is different from the original one. 12671 */ 12672 err = netdev_change_owner(dev, net_old, net); 12673 WARN_ON(err); 12674 12675 netdev_lock(dev); 12676 dev->moving_ns = false; 12677 if (!netdev_need_ops_lock(dev)) 12678 netdev_unlock(dev); 12679 12680 /* Add the device back in the hashes */ 12681 list_netdevice(dev); 12682 /* Notify protocols, that a new device appeared. */ 12683 call_netdevice_notifiers(NETDEV_REGISTER, dev); 12684 netdev_unlock_ops(dev); 12685 12686 /* 12687 * Prevent userspace races by waiting until the network 12688 * device is fully setup before sending notifications. 12689 */ 12690 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 12691 12692 synchronize_net(); 12693 err = 0; 12694out: 12695 return err; 12696} 12697 12698static int dev_cpu_dead(unsigned int oldcpu) 12699{ 12700 struct sk_buff **list_skb; 12701 struct sk_buff *skb; 12702 unsigned int cpu; 12703 struct softnet_data *sd, *oldsd, *remsd = NULL; 12704 12705 local_irq_disable(); 12706 cpu = smp_processor_id(); 12707 sd = &per_cpu(softnet_data, cpu); 12708 oldsd = &per_cpu(softnet_data, oldcpu); 12709 12710 /* Find end of our completion_queue. */ 12711 list_skb = &sd->completion_queue; 12712 while (*list_skb) 12713 list_skb = &(*list_skb)->next; 12714 /* Append completion queue from offline CPU. */ 12715 *list_skb = oldsd->completion_queue; 12716 oldsd->completion_queue = NULL; 12717 12718 /* Append output queue from offline CPU. */ 12719 if (oldsd->output_queue) { 12720 *sd->output_queue_tailp = oldsd->output_queue; 12721 sd->output_queue_tailp = oldsd->output_queue_tailp; 12722 oldsd->output_queue = NULL; 12723 oldsd->output_queue_tailp = &oldsd->output_queue; 12724 } 12725 /* Append NAPI poll list from offline CPU, with one exception : 12726 * process_backlog() must be called by cpu owning percpu backlog. 12727 * We properly handle process_queue & input_pkt_queue later. 12728 */ 12729 while (!list_empty(&oldsd->poll_list)) { 12730 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 12731 struct napi_struct, 12732 poll_list); 12733 12734 list_del_init(&napi->poll_list); 12735 if (napi->poll == process_backlog) 12736 napi->state &= NAPIF_STATE_THREADED; 12737 else 12738 ____napi_schedule(sd, napi); 12739 } 12740 12741 raise_softirq_irqoff(NET_TX_SOFTIRQ); 12742 local_irq_enable(); 12743 12744 if (!use_backlog_threads()) { 12745#ifdef CONFIG_RPS 12746 remsd = oldsd->rps_ipi_list; 12747 oldsd->rps_ipi_list = NULL; 12748#endif 12749 /* send out pending IPI's on offline CPU */ 12750 net_rps_send_ipi(remsd); 12751 } 12752 12753 /* Process offline CPU's input_pkt_queue */ 12754 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 12755 netif_rx(skb); 12756 rps_input_queue_head_incr(oldsd); 12757 } 12758 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 12759 netif_rx(skb); 12760 rps_input_queue_head_incr(oldsd); 12761 } 12762 12763 return 0; 12764} 12765 12766/** 12767 * netdev_increment_features - increment feature set by one 12768 * @all: current feature set 12769 * @one: new feature set 12770 * @mask: mask feature set 12771 * 12772 * Computes a new feature set after adding a device with feature set 12773 * @one to the master device with current feature set @all. Will not 12774 * enable anything that is off in @mask. Returns the new feature set. 12775 */ 12776netdev_features_t netdev_increment_features(netdev_features_t all, 12777 netdev_features_t one, netdev_features_t mask) 12778{ 12779 if (mask & NETIF_F_HW_CSUM) 12780 mask |= NETIF_F_CSUM_MASK; 12781 mask |= NETIF_F_VLAN_CHALLENGED; 12782 12783 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 12784 all &= one | ~NETIF_F_ALL_FOR_ALL; 12785 12786 /* If one device supports hw checksumming, set for all. */ 12787 if (all & NETIF_F_HW_CSUM) 12788 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 12789 12790 return all; 12791} 12792EXPORT_SYMBOL(netdev_increment_features); 12793 12794/** 12795 * netdev_compute_master_upper_features - compute feature from lowers 12796 * @dev: the upper device 12797 * @update_header: whether to update upper device's header_len/headroom/tailroom 12798 * 12799 * Recompute the upper device's feature based on all lower devices. 12800 */ 12801void netdev_compute_master_upper_features(struct net_device *dev, bool update_header) 12802{ 12803 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 12804 netdev_features_t gso_partial_features = MASTER_UPPER_DEV_GSO_PARTIAL_FEATURES; 12805 netdev_features_t xfrm_features = MASTER_UPPER_DEV_XFRM_FEATURES; 12806 netdev_features_t mpls_features = MASTER_UPPER_DEV_MPLS_FEATURES; 12807 netdev_features_t vlan_features = MASTER_UPPER_DEV_VLAN_FEATURES; 12808 netdev_features_t enc_features = MASTER_UPPER_DEV_ENC_FEATURES; 12809 unsigned short max_header_len = ETH_HLEN; 12810 unsigned int tso_max_size = TSO_MAX_SIZE; 12811 unsigned short max_headroom = 0; 12812 unsigned short max_tailroom = 0; 12813 u16 tso_max_segs = TSO_MAX_SEGS; 12814 struct net_device *lower_dev; 12815 struct list_head *iter; 12816 12817 mpls_features = netdev_base_features(mpls_features); 12818 vlan_features = netdev_base_features(vlan_features); 12819 enc_features = netdev_base_features(enc_features); 12820 12821 netdev_for_each_lower_dev(dev, lower_dev, iter) { 12822 gso_partial_features = netdev_increment_features(gso_partial_features, 12823 lower_dev->gso_partial_features, 12824 MASTER_UPPER_DEV_GSO_PARTIAL_FEATURES); 12825 12826 vlan_features = netdev_increment_features(vlan_features, 12827 lower_dev->vlan_features, 12828 MASTER_UPPER_DEV_VLAN_FEATURES); 12829 12830 enc_features = netdev_increment_features(enc_features, 12831 lower_dev->hw_enc_features, 12832 MASTER_UPPER_DEV_ENC_FEATURES); 12833 12834 if (IS_ENABLED(CONFIG_XFRM_OFFLOAD)) 12835 xfrm_features = netdev_increment_features(xfrm_features, 12836 lower_dev->hw_enc_features, 12837 MASTER_UPPER_DEV_XFRM_FEATURES); 12838 12839 mpls_features = netdev_increment_features(mpls_features, 12840 lower_dev->mpls_features, 12841 MASTER_UPPER_DEV_MPLS_FEATURES); 12842 12843 dst_release_flag &= lower_dev->priv_flags; 12844 12845 if (update_header) { 12846 max_header_len = max(max_header_len, lower_dev->hard_header_len); 12847 max_headroom = max(max_headroom, lower_dev->needed_headroom); 12848 max_tailroom = max(max_tailroom, lower_dev->needed_tailroom); 12849 } 12850 12851 tso_max_size = min(tso_max_size, lower_dev->tso_max_size); 12852 tso_max_segs = min(tso_max_segs, lower_dev->tso_max_segs); 12853 } 12854 12855 dev->gso_partial_features = gso_partial_features; 12856 dev->vlan_features = vlan_features; 12857 dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 12858 NETIF_F_HW_VLAN_CTAG_TX | 12859 NETIF_F_HW_VLAN_STAG_TX; 12860 if (IS_ENABLED(CONFIG_XFRM_OFFLOAD)) 12861 dev->hw_enc_features |= xfrm_features; 12862 dev->mpls_features = mpls_features; 12863 12864 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 12865 if ((dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 12866 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 12867 dev->priv_flags |= IFF_XMIT_DST_RELEASE; 12868 12869 if (update_header) { 12870 dev->hard_header_len = max_header_len; 12871 dev->needed_headroom = max_headroom; 12872 dev->needed_tailroom = max_tailroom; 12873 } 12874 12875 netif_set_tso_max_segs(dev, tso_max_segs); 12876 netif_set_tso_max_size(dev, tso_max_size); 12877 12878 netdev_change_features(dev); 12879} 12880EXPORT_SYMBOL(netdev_compute_master_upper_features); 12881 12882static struct hlist_head * __net_init netdev_create_hash(void) 12883{ 12884 int i; 12885 struct hlist_head *hash; 12886 12887 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 12888 if (hash != NULL) 12889 for (i = 0; i < NETDEV_HASHENTRIES; i++) 12890 INIT_HLIST_HEAD(&hash[i]); 12891 12892 return hash; 12893} 12894 12895/* Initialize per network namespace state */ 12896static int __net_init netdev_init(struct net *net) 12897{ 12898 BUILD_BUG_ON(GRO_HASH_BUCKETS > 12899 BITS_PER_BYTE * sizeof_field(struct gro_node, bitmask)); 12900 12901 INIT_LIST_HEAD(&net->dev_base_head); 12902 12903 net->dev_name_head = netdev_create_hash(); 12904 if (net->dev_name_head == NULL) 12905 goto err_name; 12906 12907 net->dev_index_head = netdev_create_hash(); 12908 if (net->dev_index_head == NULL) 12909 goto err_idx; 12910 12911 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 12912 12913 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 12914 12915 return 0; 12916 12917err_idx: 12918 kfree(net->dev_name_head); 12919err_name: 12920 return -ENOMEM; 12921} 12922 12923/** 12924 * netdev_drivername - network driver for the device 12925 * @dev: network device 12926 * 12927 * Determine network driver for device. 12928 */ 12929const char *netdev_drivername(const struct net_device *dev) 12930{ 12931 const struct device_driver *driver; 12932 const struct device *parent; 12933 const char *empty = ""; 12934 12935 parent = dev->dev.parent; 12936 if (!parent) 12937 return empty; 12938 12939 driver = parent->driver; 12940 if (driver && driver->name) 12941 return driver->name; 12942 return empty; 12943} 12944 12945static void __netdev_printk(const char *level, const struct net_device *dev, 12946 struct va_format *vaf) 12947{ 12948 if (dev && dev->dev.parent) { 12949 dev_printk_emit(level[1] - '0', 12950 dev->dev.parent, 12951 "%s %s %s%s: %pV", 12952 dev_driver_string(dev->dev.parent), 12953 dev_name(dev->dev.parent), 12954 netdev_name(dev), netdev_reg_state(dev), 12955 vaf); 12956 } else if (dev) { 12957 printk("%s%s%s: %pV", 12958 level, netdev_name(dev), netdev_reg_state(dev), vaf); 12959 } else { 12960 printk("%s(NULL net_device): %pV", level, vaf); 12961 } 12962} 12963 12964void netdev_printk(const char *level, const struct net_device *dev, 12965 const char *format, ...) 12966{ 12967 struct va_format vaf; 12968 va_list args; 12969 12970 va_start(args, format); 12971 12972 vaf.fmt = format; 12973 vaf.va = &args; 12974 12975 __netdev_printk(level, dev, &vaf); 12976 12977 va_end(args); 12978} 12979EXPORT_SYMBOL(netdev_printk); 12980 12981#define define_netdev_printk_level(func, level) \ 12982void func(const struct net_device *dev, const char *fmt, ...) \ 12983{ \ 12984 struct va_format vaf; \ 12985 va_list args; \ 12986 \ 12987 va_start(args, fmt); \ 12988 \ 12989 vaf.fmt = fmt; \ 12990 vaf.va = &args; \ 12991 \ 12992 __netdev_printk(level, dev, &vaf); \ 12993 \ 12994 va_end(args); \ 12995} \ 12996EXPORT_SYMBOL(func); 12997 12998define_netdev_printk_level(netdev_emerg, KERN_EMERG); 12999define_netdev_printk_level(netdev_alert, KERN_ALERT); 13000define_netdev_printk_level(netdev_crit, KERN_CRIT); 13001define_netdev_printk_level(netdev_err, KERN_ERR); 13002define_netdev_printk_level(netdev_warn, KERN_WARNING); 13003define_netdev_printk_level(netdev_notice, KERN_NOTICE); 13004define_netdev_printk_level(netdev_info, KERN_INFO); 13005 13006static void __net_exit netdev_exit(struct net *net) 13007{ 13008 kfree(net->dev_name_head); 13009 kfree(net->dev_index_head); 13010 xa_destroy(&net->dev_by_index); 13011 if (net != &init_net) 13012 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 13013} 13014 13015static struct pernet_operations __net_initdata netdev_net_ops = { 13016 .init = netdev_init, 13017 .exit = netdev_exit, 13018}; 13019 13020static void __net_exit default_device_exit_net(struct net *net) 13021{ 13022 struct netdev_name_node *name_node, *tmp; 13023 struct net_device *dev, *aux; 13024 /* 13025 * Push all migratable network devices back to the 13026 * initial network namespace 13027 */ 13028 ASSERT_RTNL(); 13029 for_each_netdev_safe(net, dev, aux) { 13030 int err; 13031 char fb_name[IFNAMSIZ]; 13032 13033 /* Ignore unmoveable devices (i.e. loopback) */ 13034 if (dev->netns_immutable) 13035 continue; 13036 13037 /* Leave virtual devices for the generic cleanup */ 13038 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 13039 continue; 13040 13041 /* Push remaining network devices to init_net */ 13042 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 13043 if (netdev_name_in_use(&init_net, fb_name)) 13044 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 13045 13046 netdev_for_each_altname_safe(dev, name_node, tmp) 13047 if (netdev_name_in_use(&init_net, name_node->name)) 13048 __netdev_name_node_alt_destroy(name_node); 13049 13050 err = dev_change_net_namespace(dev, &init_net, fb_name); 13051 if (err) { 13052 pr_emerg("%s: failed to move %s to init_net: %d\n", 13053 __func__, dev->name, err); 13054 BUG(); 13055 } 13056 } 13057} 13058 13059static void __net_exit default_device_exit_batch(struct list_head *net_list) 13060{ 13061 /* At exit all network devices most be removed from a network 13062 * namespace. Do this in the reverse order of registration. 13063 * Do this across as many network namespaces as possible to 13064 * improve batching efficiency. 13065 */ 13066 struct net_device *dev; 13067 struct net *net; 13068 LIST_HEAD(dev_kill_list); 13069 13070 rtnl_lock(); 13071 list_for_each_entry(net, net_list, exit_list) { 13072 default_device_exit_net(net); 13073 cond_resched(); 13074 } 13075 13076 list_for_each_entry(net, net_list, exit_list) { 13077 for_each_netdev_reverse(net, dev) { 13078 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 13079 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 13080 else 13081 unregister_netdevice_queue(dev, &dev_kill_list); 13082 } 13083 } 13084 unregister_netdevice_many(&dev_kill_list); 13085 rtnl_unlock(); 13086} 13087 13088static struct pernet_operations __net_initdata default_device_ops = { 13089 .exit_batch = default_device_exit_batch, 13090}; 13091 13092static void __init net_dev_struct_check(void) 13093{ 13094 /* TX read-mostly hotpath */ 13095 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags_fast); 13096 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 13097 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 13098 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 13099 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 13100 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 13101 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 13102 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 13103 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 13104 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 13105 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 13106 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 13107 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 13108#ifdef CONFIG_XPS 13109 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 13110#endif 13111#ifdef CONFIG_NETFILTER_EGRESS 13112 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 13113#endif 13114#ifdef CONFIG_NET_XGRESS 13115 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 13116#endif 13117 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 13118 13119 /* TXRX read-mostly hotpath */ 13120 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); 13121 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state); 13122 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 13123 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 13124 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 13125 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 13126 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46); 13127 13128 /* RX read-mostly hotpath */ 13129 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 13130 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 13131 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 13132 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 13133 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 13134 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 13135 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 13136 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 13137 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 13138#ifdef CONFIG_NETPOLL 13139 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 13140#endif 13141#ifdef CONFIG_NET_XGRESS 13142 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 13143#endif 13144 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 92); 13145} 13146 13147/* 13148 * Initialize the DEV module. At boot time this walks the device list and 13149 * unhooks any devices that fail to initialise (normally hardware not 13150 * present) and leaves us with a valid list of present and active devices. 13151 * 13152 */ 13153 13154/* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ 13155#define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) 13156 13157static int net_page_pool_create(int cpuid) 13158{ 13159#if IS_ENABLED(CONFIG_PAGE_POOL) 13160 struct page_pool_params page_pool_params = { 13161 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, 13162 .flags = PP_FLAG_SYSTEM_POOL, 13163 .nid = cpu_to_mem(cpuid), 13164 }; 13165 struct page_pool *pp_ptr; 13166 int err; 13167 13168 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); 13169 if (IS_ERR(pp_ptr)) 13170 return -ENOMEM; 13171 13172 err = xdp_reg_page_pool(pp_ptr); 13173 if (err) { 13174 page_pool_destroy(pp_ptr); 13175 return err; 13176 } 13177 13178 per_cpu(system_page_pool.pool, cpuid) = pp_ptr; 13179#endif 13180 return 0; 13181} 13182 13183static int backlog_napi_should_run(unsigned int cpu) 13184{ 13185 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 13186 struct napi_struct *napi = &sd->backlog; 13187 13188 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 13189} 13190 13191static void run_backlog_napi(unsigned int cpu) 13192{ 13193 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 13194 13195 napi_threaded_poll_loop(&sd->backlog, NULL); 13196} 13197 13198static void backlog_napi_setup(unsigned int cpu) 13199{ 13200 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 13201 struct napi_struct *napi = &sd->backlog; 13202 13203 napi->thread = this_cpu_read(backlog_napi); 13204 set_bit(NAPI_STATE_THREADED, &napi->state); 13205} 13206 13207static struct smp_hotplug_thread backlog_threads = { 13208 .store = &backlog_napi, 13209 .thread_should_run = backlog_napi_should_run, 13210 .thread_fn = run_backlog_napi, 13211 .thread_comm = "backlog_napi/%u", 13212 .setup = backlog_napi_setup, 13213}; 13214 13215/* 13216 * This is called single threaded during boot, so no need 13217 * to take the rtnl semaphore. 13218 */ 13219static int __init net_dev_init(void) 13220{ 13221 int i, rc = -ENOMEM; 13222 13223 BUG_ON(!dev_boot_phase); 13224 13225 net_dev_struct_check(); 13226 13227 if (dev_proc_init()) 13228 goto out; 13229 13230 if (netdev_kobject_init()) 13231 goto out; 13232 13233 for (i = 0; i < PTYPE_HASH_SIZE; i++) 13234 INIT_LIST_HEAD(&ptype_base[i]); 13235 13236 if (register_pernet_subsys(&netdev_net_ops)) 13237 goto out; 13238 13239 /* 13240 * Initialise the packet receive queues. 13241 */ 13242 13243 flush_backlogs_fallback = flush_backlogs_alloc(); 13244 if (!flush_backlogs_fallback) 13245 goto out; 13246 13247 for_each_possible_cpu(i) { 13248 struct softnet_data *sd = &per_cpu(softnet_data, i); 13249 13250 skb_queue_head_init(&sd->input_pkt_queue); 13251 skb_queue_head_init(&sd->process_queue); 13252#ifdef CONFIG_XFRM_OFFLOAD 13253 skb_queue_head_init(&sd->xfrm_backlog); 13254#endif 13255 INIT_LIST_HEAD(&sd->poll_list); 13256 sd->output_queue_tailp = &sd->output_queue; 13257#ifdef CONFIG_RPS 13258 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 13259 sd->cpu = i; 13260#endif 13261 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 13262 13263 gro_init(&sd->backlog.gro); 13264 sd->backlog.poll = process_backlog; 13265 sd->backlog.weight = weight_p; 13266 INIT_LIST_HEAD(&sd->backlog.poll_list); 13267 13268 if (net_page_pool_create(i)) 13269 goto out; 13270 } 13271 net_hotdata.skb_defer_nodes = 13272 __alloc_percpu(sizeof(struct skb_defer_node) * nr_node_ids, 13273 __alignof__(struct skb_defer_node)); 13274 if (!net_hotdata.skb_defer_nodes) 13275 goto out; 13276 if (use_backlog_threads()) 13277 smpboot_register_percpu_thread(&backlog_threads); 13278 13279 dev_boot_phase = 0; 13280 13281 /* The loopback device is special if any other network devices 13282 * is present in a network namespace the loopback device must 13283 * be present. Since we now dynamically allocate and free the 13284 * loopback device ensure this invariant is maintained by 13285 * keeping the loopback device as the first device on the 13286 * list of network devices. Ensuring the loopback devices 13287 * is the first device that appears and the last network device 13288 * that disappears. 13289 */ 13290 if (register_pernet_device(&loopback_net_ops)) 13291 goto out; 13292 13293 if (register_pernet_device(&default_device_ops)) 13294 goto out; 13295 13296 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 13297 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 13298 13299 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 13300 NULL, dev_cpu_dead); 13301 WARN_ON(rc < 0); 13302 rc = 0; 13303 13304 /* avoid static key IPIs to isolated CPUs */ 13305 if (housekeeping_enabled(HK_TYPE_MISC)) 13306 net_enable_timestamp(); 13307out: 13308 if (rc < 0) { 13309 for_each_possible_cpu(i) { 13310 struct page_pool *pp_ptr; 13311 13312 pp_ptr = per_cpu(system_page_pool.pool, i); 13313 if (!pp_ptr) 13314 continue; 13315 13316 xdp_unreg_page_pool(pp_ptr); 13317 page_pool_destroy(pp_ptr); 13318 per_cpu(system_page_pool.pool, i) = NULL; 13319 } 13320 } 13321 13322 return rc; 13323} 13324 13325subsys_initcall(net_dev_init);
~/
Read-only file preview.