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/skbuff.h> 96 #include <linux/kthread.h> 97 #include <linux/bpf.h> 98 #include <linux/bpf_trace.h> 99 #include <net/net_namespace.h> 100 #include <net/sock.h> 101 #include <net/busy_poll.h> 102 #include <linux/rtnetlink.h> 103 #include <linux/stat.h> 104 #include <net/dsa.h> 105 #include <net/dst.h> 106 #include <net/dst_metadata.h> 107 #include <net/gro.h> 108 #include <net/pkt_sched.h> 109 #include <net/pkt_cls.h> 110 #include <net/checksum.h> 111 #include <net/xfrm.h> 112 #include <net/tcx.h> 113 #include <linux/highmem.h> 114 #include <linux/init.h> 115 #include <linux/module.h> 116 #include <linux/netpoll.h> 117 #include <linux/rcupdate.h> 118 #include <linux/delay.h> 119 #include <net/iw_handler.h> 120 #include <asm/current.h> 121 #include <linux/audit.h> 122 #include <linux/dmaengine.h> 123 #include <linux/err.h> 124 #include <linux/ctype.h> 125 #include <linux/if_arp.h> 126 #include <linux/if_vlan.h> 127 #include <linux/ip.h> 128 #include <net/ip.h> 129 #include <net/mpls.h> 130 #include <linux/ipv6.h> 131 #include <linux/in.h> 132 #include <linux/jhash.h> 133 #include <linux/random.h> 134 #include <trace/events/napi.h> 135 #include <trace/events/net.h> 136 #include <trace/events/skb.h> 137 #include <trace/events/qdisc.h> 138 #include <trace/events/xdp.h> 139 #include <linux/inetdevice.h> 140 #include <linux/cpu_rmap.h> 141 #include <linux/static_key.h> 142 #include <linux/hashtable.h> 143 #include <linux/vmalloc.h> 144 #include <linux/if_macvlan.h> 145 #include <linux/errqueue.h> 146 #include <linux/hrtimer.h> 147 #include <linux/netfilter_netdev.h> 148 #include <linux/crash_dump.h> 149 #include <linux/sctp.h> 150 #include <net/udp_tunnel.h> 151 #include <linux/net_namespace.h> 152 #include <linux/indirect_call_wrapper.h> 153 #include <net/devlink.h> 154 #include <linux/pm_runtime.h> 155 #include <linux/prandom.h> 156 #include <linux/once_lite.h> 157 #include <net/netdev_rx_queue.h> 158 #include <net/page_pool/types.h> 159 #include <net/page_pool/helpers.h> 160 #include <net/rps.h> 161 #include <linux/phy_link_topology.h> 162 163 #include "dev.h" 164 #include "devmem.h" 165 #include "net-sysfs.h" 166 167 static DEFINE_SPINLOCK(ptype_lock); 168 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 169 170 static int netif_rx_internal(struct sk_buff *skb); 171 static int call_netdevice_notifiers_extack(unsigned long val, 172 struct net_device *dev, 173 struct netlink_ext_ack *extack); 174 175 static DEFINE_MUTEX(ifalias_mutex); 176 177 /* protects napi_hash addition/deletion and napi_gen_id */ 178 static DEFINE_SPINLOCK(napi_hash_lock); 179 180 static unsigned int napi_gen_id = NR_CPUS; 181 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 182 183 static DECLARE_RWSEM(devnet_rename_sem); 184 185 static inline void dev_base_seq_inc(struct net *net) 186 { 187 unsigned int val = net->dev_base_seq + 1; 188 189 WRITE_ONCE(net->dev_base_seq, val ?: 1); 190 } 191 192 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 193 { 194 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 195 196 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 197 } 198 199 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 200 { 201 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 202 } 203 204 #ifndef CONFIG_PREEMPT_RT 205 206 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key); 207 208 static int __init setup_backlog_napi_threads(char *arg) 209 { 210 static_branch_enable(&use_backlog_threads_key); 211 return 0; 212 } 213 early_param("thread_backlog_napi", setup_backlog_napi_threads); 214 215 static bool use_backlog_threads(void) 216 { 217 return static_branch_unlikely(&use_backlog_threads_key); 218 } 219 220 #else 221 222 static bool use_backlog_threads(void) 223 { 224 return true; 225 } 226 227 #endif 228 229 static inline void backlog_lock_irq_save(struct softnet_data *sd, 230 unsigned long *flags) 231 { 232 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 233 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 234 else 235 local_irq_save(*flags); 236 } 237 238 static inline void backlog_lock_irq_disable(struct softnet_data *sd) 239 { 240 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 241 spin_lock_irq(&sd->input_pkt_queue.lock); 242 else 243 local_irq_disable(); 244 } 245 246 static inline void backlog_unlock_irq_restore(struct softnet_data *sd, 247 unsigned long *flags) 248 { 249 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 250 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 251 else 252 local_irq_restore(*flags); 253 } 254 255 static inline void backlog_unlock_irq_enable(struct softnet_data *sd) 256 { 257 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 258 spin_unlock_irq(&sd->input_pkt_queue.lock); 259 else 260 local_irq_enable(); 261 } 262 263 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 264 const char *name) 265 { 266 struct netdev_name_node *name_node; 267 268 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 269 if (!name_node) 270 return NULL; 271 INIT_HLIST_NODE(&name_node->hlist); 272 name_node->dev = dev; 273 name_node->name = name; 274 return name_node; 275 } 276 277 static struct netdev_name_node * 278 netdev_name_node_head_alloc(struct net_device *dev) 279 { 280 struct netdev_name_node *name_node; 281 282 name_node = netdev_name_node_alloc(dev, dev->name); 283 if (!name_node) 284 return NULL; 285 INIT_LIST_HEAD(&name_node->list); 286 return name_node; 287 } 288 289 static void netdev_name_node_free(struct netdev_name_node *name_node) 290 { 291 kfree(name_node); 292 } 293 294 static void netdev_name_node_add(struct net *net, 295 struct netdev_name_node *name_node) 296 { 297 hlist_add_head_rcu(&name_node->hlist, 298 dev_name_hash(net, name_node->name)); 299 } 300 301 static void netdev_name_node_del(struct netdev_name_node *name_node) 302 { 303 hlist_del_rcu(&name_node->hlist); 304 } 305 306 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 307 const char *name) 308 { 309 struct hlist_head *head = dev_name_hash(net, name); 310 struct netdev_name_node *name_node; 311 312 hlist_for_each_entry(name_node, head, hlist) 313 if (!strcmp(name_node->name, name)) 314 return name_node; 315 return NULL; 316 } 317 318 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 319 const char *name) 320 { 321 struct hlist_head *head = dev_name_hash(net, name); 322 struct netdev_name_node *name_node; 323 324 hlist_for_each_entry_rcu(name_node, head, hlist) 325 if (!strcmp(name_node->name, name)) 326 return name_node; 327 return NULL; 328 } 329 330 bool netdev_name_in_use(struct net *net, const char *name) 331 { 332 return netdev_name_node_lookup(net, name); 333 } 334 EXPORT_SYMBOL(netdev_name_in_use); 335 336 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 337 { 338 struct netdev_name_node *name_node; 339 struct net *net = dev_net(dev); 340 341 name_node = netdev_name_node_lookup(net, name); 342 if (name_node) 343 return -EEXIST; 344 name_node = netdev_name_node_alloc(dev, name); 345 if (!name_node) 346 return -ENOMEM; 347 netdev_name_node_add(net, name_node); 348 /* The node that holds dev->name acts as a head of per-device list. */ 349 list_add_tail_rcu(&name_node->list, &dev->name_node->list); 350 351 return 0; 352 } 353 354 static void netdev_name_node_alt_free(struct rcu_head *head) 355 { 356 struct netdev_name_node *name_node = 357 container_of(head, struct netdev_name_node, rcu); 358 359 kfree(name_node->name); 360 netdev_name_node_free(name_node); 361 } 362 363 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 364 { 365 netdev_name_node_del(name_node); 366 list_del(&name_node->list); 367 call_rcu(&name_node->rcu, netdev_name_node_alt_free); 368 } 369 370 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 371 { 372 struct netdev_name_node *name_node; 373 struct net *net = dev_net(dev); 374 375 name_node = netdev_name_node_lookup(net, name); 376 if (!name_node) 377 return -ENOENT; 378 /* lookup might have found our primary name or a name belonging 379 * to another device. 380 */ 381 if (name_node == dev->name_node || name_node->dev != dev) 382 return -EINVAL; 383 384 __netdev_name_node_alt_destroy(name_node); 385 return 0; 386 } 387 388 static void netdev_name_node_alt_flush(struct net_device *dev) 389 { 390 struct netdev_name_node *name_node, *tmp; 391 392 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) { 393 list_del(&name_node->list); 394 netdev_name_node_alt_free(&name_node->rcu); 395 } 396 } 397 398 /* Device list insertion */ 399 static void list_netdevice(struct net_device *dev) 400 { 401 struct netdev_name_node *name_node; 402 struct net *net = dev_net(dev); 403 404 ASSERT_RTNL(); 405 406 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 407 netdev_name_node_add(net, dev->name_node); 408 hlist_add_head_rcu(&dev->index_hlist, 409 dev_index_hash(net, dev->ifindex)); 410 411 netdev_for_each_altname(dev, name_node) 412 netdev_name_node_add(net, name_node); 413 414 /* We reserved the ifindex, this can't fail */ 415 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 416 417 dev_base_seq_inc(net); 418 } 419 420 /* Device list removal 421 * caller must respect a RCU grace period before freeing/reusing dev 422 */ 423 static void unlist_netdevice(struct net_device *dev) 424 { 425 struct netdev_name_node *name_node; 426 struct net *net = dev_net(dev); 427 428 ASSERT_RTNL(); 429 430 xa_erase(&net->dev_by_index, dev->ifindex); 431 432 netdev_for_each_altname(dev, name_node) 433 netdev_name_node_del(name_node); 434 435 /* Unlink dev from the device chain */ 436 list_del_rcu(&dev->dev_list); 437 netdev_name_node_del(dev->name_node); 438 hlist_del_rcu(&dev->index_hlist); 439 440 dev_base_seq_inc(dev_net(dev)); 441 } 442 443 /* 444 * Our notifier list 445 */ 446 447 static RAW_NOTIFIER_HEAD(netdev_chain); 448 449 /* 450 * Device drivers call our routines to queue packets here. We empty the 451 * queue in the local softnet handler. 452 */ 453 454 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = { 455 .process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock), 456 }; 457 EXPORT_PER_CPU_SYMBOL(softnet_data); 458 459 /* Page_pool has a lockless array/stack to alloc/recycle pages. 460 * PP consumers must pay attention to run APIs in the appropriate context 461 * (e.g. NAPI context). 462 */ 463 DEFINE_PER_CPU(struct page_pool *, system_page_pool); 464 465 #ifdef CONFIG_LOCKDEP 466 /* 467 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 468 * according to dev->type 469 */ 470 static const unsigned short netdev_lock_type[] = { 471 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 472 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 473 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 474 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 475 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 476 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 477 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 478 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 479 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 480 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 481 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 482 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 483 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 484 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 485 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 486 487 static const char *const netdev_lock_name[] = { 488 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 489 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 490 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 491 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 492 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 493 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 494 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 495 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 496 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 497 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 498 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 499 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 500 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 501 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 502 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 503 504 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 505 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 506 507 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 508 { 509 int i; 510 511 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 512 if (netdev_lock_type[i] == dev_type) 513 return i; 514 /* the last key is used by default */ 515 return ARRAY_SIZE(netdev_lock_type) - 1; 516 } 517 518 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 519 unsigned short dev_type) 520 { 521 int i; 522 523 i = netdev_lock_pos(dev_type); 524 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 525 netdev_lock_name[i]); 526 } 527 528 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 529 { 530 int i; 531 532 i = netdev_lock_pos(dev->type); 533 lockdep_set_class_and_name(&dev->addr_list_lock, 534 &netdev_addr_lock_key[i], 535 netdev_lock_name[i]); 536 } 537 #else 538 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 539 unsigned short dev_type) 540 { 541 } 542 543 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 544 { 545 } 546 #endif 547 548 /******************************************************************************* 549 * 550 * Protocol management and registration routines 551 * 552 *******************************************************************************/ 553 554 555 /* 556 * Add a protocol ID to the list. Now that the input handler is 557 * smarter we can dispense with all the messy stuff that used to be 558 * here. 559 * 560 * BEWARE!!! Protocol handlers, mangling input packets, 561 * MUST BE last in hash buckets and checking protocol handlers 562 * MUST start from promiscuous ptype_all chain in net_bh. 563 * It is true now, do not change it. 564 * Explanation follows: if protocol handler, mangling packet, will 565 * be the first on list, it is not able to sense, that packet 566 * is cloned and should be copied-on-write, so that it will 567 * change it and subsequent readers will get broken packet. 568 * --ANK (980803) 569 */ 570 571 static inline struct list_head *ptype_head(const struct packet_type *pt) 572 { 573 if (pt->type == htons(ETH_P_ALL)) 574 return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all; 575 else 576 return pt->dev ? &pt->dev->ptype_specific : 577 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 578 } 579 580 /** 581 * dev_add_pack - add packet handler 582 * @pt: packet type declaration 583 * 584 * Add a protocol handler to the networking stack. The passed &packet_type 585 * is linked into kernel lists and may not be freed until it has been 586 * removed from the kernel lists. 587 * 588 * This call does not sleep therefore it can not 589 * guarantee all CPU's that are in middle of receiving packets 590 * will see the new packet type (until the next received packet). 591 */ 592 593 void dev_add_pack(struct packet_type *pt) 594 { 595 struct list_head *head = ptype_head(pt); 596 597 spin_lock(&ptype_lock); 598 list_add_rcu(&pt->list, head); 599 spin_unlock(&ptype_lock); 600 } 601 EXPORT_SYMBOL(dev_add_pack); 602 603 /** 604 * __dev_remove_pack - remove packet handler 605 * @pt: packet type declaration 606 * 607 * Remove a protocol handler that was previously added to the kernel 608 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 609 * from the kernel lists and can be freed or reused once this function 610 * returns. 611 * 612 * The packet type might still be in use by receivers 613 * and must not be freed until after all the CPU's have gone 614 * through a quiescent state. 615 */ 616 void __dev_remove_pack(struct packet_type *pt) 617 { 618 struct list_head *head = ptype_head(pt); 619 struct packet_type *pt1; 620 621 spin_lock(&ptype_lock); 622 623 list_for_each_entry(pt1, head, list) { 624 if (pt == pt1) { 625 list_del_rcu(&pt->list); 626 goto out; 627 } 628 } 629 630 pr_warn("dev_remove_pack: %p not found\n", pt); 631 out: 632 spin_unlock(&ptype_lock); 633 } 634 EXPORT_SYMBOL(__dev_remove_pack); 635 636 /** 637 * dev_remove_pack - remove packet handler 638 * @pt: packet type declaration 639 * 640 * Remove a protocol handler that was previously added to the kernel 641 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 642 * from the kernel lists and can be freed or reused once this function 643 * returns. 644 * 645 * This call sleeps to guarantee that no CPU is looking at the packet 646 * type after return. 647 */ 648 void dev_remove_pack(struct packet_type *pt) 649 { 650 __dev_remove_pack(pt); 651 652 synchronize_net(); 653 } 654 EXPORT_SYMBOL(dev_remove_pack); 655 656 657 /******************************************************************************* 658 * 659 * Device Interface Subroutines 660 * 661 *******************************************************************************/ 662 663 /** 664 * dev_get_iflink - get 'iflink' value of a interface 665 * @dev: targeted interface 666 * 667 * Indicates the ifindex the interface is linked to. 668 * Physical interfaces have the same 'ifindex' and 'iflink' values. 669 */ 670 671 int dev_get_iflink(const struct net_device *dev) 672 { 673 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 674 return dev->netdev_ops->ndo_get_iflink(dev); 675 676 return READ_ONCE(dev->ifindex); 677 } 678 EXPORT_SYMBOL(dev_get_iflink); 679 680 /** 681 * dev_fill_metadata_dst - Retrieve tunnel egress information. 682 * @dev: targeted interface 683 * @skb: The packet. 684 * 685 * For better visibility of tunnel traffic OVS needs to retrieve 686 * egress tunnel information for a packet. Following API allows 687 * user to get this info. 688 */ 689 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 690 { 691 struct ip_tunnel_info *info; 692 693 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 694 return -EINVAL; 695 696 info = skb_tunnel_info_unclone(skb); 697 if (!info) 698 return -ENOMEM; 699 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 700 return -EINVAL; 701 702 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 703 } 704 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 705 706 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 707 { 708 int k = stack->num_paths++; 709 710 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 711 return NULL; 712 713 return &stack->path[k]; 714 } 715 716 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 717 struct net_device_path_stack *stack) 718 { 719 const struct net_device *last_dev; 720 struct net_device_path_ctx ctx = { 721 .dev = dev, 722 }; 723 struct net_device_path *path; 724 int ret = 0; 725 726 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 727 stack->num_paths = 0; 728 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 729 last_dev = ctx.dev; 730 path = dev_fwd_path(stack); 731 if (!path) 732 return -1; 733 734 memset(path, 0, sizeof(struct net_device_path)); 735 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 736 if (ret < 0) 737 return -1; 738 739 if (WARN_ON_ONCE(last_dev == ctx.dev)) 740 return -1; 741 } 742 743 if (!ctx.dev) 744 return ret; 745 746 path = dev_fwd_path(stack); 747 if (!path) 748 return -1; 749 path->type = DEV_PATH_ETHERNET; 750 path->dev = ctx.dev; 751 752 return ret; 753 } 754 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 755 756 /** 757 * __dev_get_by_name - find a device by its name 758 * @net: the applicable net namespace 759 * @name: name to find 760 * 761 * Find an interface by name. Must be called under RTNL semaphore. 762 * If the name is found a pointer to the device is returned. 763 * If the name is not found then %NULL is returned. The 764 * reference counters are not incremented so the caller must be 765 * careful with locks. 766 */ 767 768 struct net_device *__dev_get_by_name(struct net *net, const char *name) 769 { 770 struct netdev_name_node *node_name; 771 772 node_name = netdev_name_node_lookup(net, name); 773 return node_name ? node_name->dev : NULL; 774 } 775 EXPORT_SYMBOL(__dev_get_by_name); 776 777 /** 778 * dev_get_by_name_rcu - find a device by its name 779 * @net: the applicable net namespace 780 * @name: name to find 781 * 782 * Find an interface by name. 783 * If the name is found a pointer to the device is returned. 784 * If the name is not found then %NULL is returned. 785 * The reference counters are not incremented so the caller must be 786 * careful with locks. The caller must hold RCU lock. 787 */ 788 789 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 790 { 791 struct netdev_name_node *node_name; 792 793 node_name = netdev_name_node_lookup_rcu(net, name); 794 return node_name ? node_name->dev : NULL; 795 } 796 EXPORT_SYMBOL(dev_get_by_name_rcu); 797 798 /* Deprecated for new users, call netdev_get_by_name() instead */ 799 struct net_device *dev_get_by_name(struct net *net, const char *name) 800 { 801 struct net_device *dev; 802 803 rcu_read_lock(); 804 dev = dev_get_by_name_rcu(net, name); 805 dev_hold(dev); 806 rcu_read_unlock(); 807 return dev; 808 } 809 EXPORT_SYMBOL(dev_get_by_name); 810 811 /** 812 * netdev_get_by_name() - find a device by its name 813 * @net: the applicable net namespace 814 * @name: name to find 815 * @tracker: tracking object for the acquired reference 816 * @gfp: allocation flags for the tracker 817 * 818 * Find an interface by name. This can be called from any 819 * context and does its own locking. The returned handle has 820 * the usage count incremented and the caller must use netdev_put() to 821 * release it when it is no longer needed. %NULL is returned if no 822 * matching device is found. 823 */ 824 struct net_device *netdev_get_by_name(struct net *net, const char *name, 825 netdevice_tracker *tracker, gfp_t gfp) 826 { 827 struct net_device *dev; 828 829 dev = dev_get_by_name(net, name); 830 if (dev) 831 netdev_tracker_alloc(dev, tracker, gfp); 832 return dev; 833 } 834 EXPORT_SYMBOL(netdev_get_by_name); 835 836 /** 837 * __dev_get_by_index - find a device by its ifindex 838 * @net: the applicable net namespace 839 * @ifindex: index of device 840 * 841 * Search for an interface by index. Returns %NULL if the device 842 * is not found or a pointer to the device. The device has not 843 * had its reference counter increased so the caller must be careful 844 * about locking. The caller must hold the RTNL semaphore. 845 */ 846 847 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 848 { 849 struct net_device *dev; 850 struct hlist_head *head = dev_index_hash(net, ifindex); 851 852 hlist_for_each_entry(dev, head, index_hlist) 853 if (dev->ifindex == ifindex) 854 return dev; 855 856 return NULL; 857 } 858 EXPORT_SYMBOL(__dev_get_by_index); 859 860 /** 861 * dev_get_by_index_rcu - find a device by its ifindex 862 * @net: the applicable net namespace 863 * @ifindex: index of device 864 * 865 * Search for an interface by index. Returns %NULL if the device 866 * is not found or a pointer to the device. The device has not 867 * had its reference counter increased so the caller must be careful 868 * about locking. The caller must hold RCU lock. 869 */ 870 871 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 872 { 873 struct net_device *dev; 874 struct hlist_head *head = dev_index_hash(net, ifindex); 875 876 hlist_for_each_entry_rcu(dev, head, index_hlist) 877 if (dev->ifindex == ifindex) 878 return dev; 879 880 return NULL; 881 } 882 EXPORT_SYMBOL(dev_get_by_index_rcu); 883 884 /* Deprecated for new users, call netdev_get_by_index() instead */ 885 struct net_device *dev_get_by_index(struct net *net, int ifindex) 886 { 887 struct net_device *dev; 888 889 rcu_read_lock(); 890 dev = dev_get_by_index_rcu(net, ifindex); 891 dev_hold(dev); 892 rcu_read_unlock(); 893 return dev; 894 } 895 EXPORT_SYMBOL(dev_get_by_index); 896 897 /** 898 * netdev_get_by_index() - find a device by its ifindex 899 * @net: the applicable net namespace 900 * @ifindex: index of device 901 * @tracker: tracking object for the acquired reference 902 * @gfp: allocation flags for the tracker 903 * 904 * Search for an interface by index. Returns NULL if the device 905 * is not found or a pointer to the device. The device returned has 906 * had a reference added and the pointer is safe until the user calls 907 * netdev_put() to indicate they have finished with it. 908 */ 909 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 910 netdevice_tracker *tracker, gfp_t gfp) 911 { 912 struct net_device *dev; 913 914 dev = dev_get_by_index(net, ifindex); 915 if (dev) 916 netdev_tracker_alloc(dev, tracker, gfp); 917 return dev; 918 } 919 EXPORT_SYMBOL(netdev_get_by_index); 920 921 /** 922 * dev_get_by_napi_id - find a device by napi_id 923 * @napi_id: ID of the NAPI struct 924 * 925 * Search for an interface by NAPI ID. Returns %NULL if the device 926 * is not found or a pointer to the device. The device has not had 927 * its reference counter increased so the caller must be careful 928 * about locking. The caller must hold RCU lock. 929 */ 930 931 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 932 { 933 struct napi_struct *napi; 934 935 WARN_ON_ONCE(!rcu_read_lock_held()); 936 937 if (napi_id < MIN_NAPI_ID) 938 return NULL; 939 940 napi = napi_by_id(napi_id); 941 942 return napi ? napi->dev : NULL; 943 } 944 EXPORT_SYMBOL(dev_get_by_napi_id); 945 946 static DEFINE_SEQLOCK(netdev_rename_lock); 947 948 void netdev_copy_name(struct net_device *dev, char *name) 949 { 950 unsigned int seq; 951 952 do { 953 seq = read_seqbegin(&netdev_rename_lock); 954 strscpy(name, dev->name, IFNAMSIZ); 955 } while (read_seqretry(&netdev_rename_lock, seq)); 956 } 957 958 /** 959 * netdev_get_name - get a netdevice name, knowing its ifindex. 960 * @net: network namespace 961 * @name: a pointer to the buffer where the name will be stored. 962 * @ifindex: the ifindex of the interface to get the name from. 963 */ 964 int netdev_get_name(struct net *net, char *name, int ifindex) 965 { 966 struct net_device *dev; 967 int ret; 968 969 rcu_read_lock(); 970 971 dev = dev_get_by_index_rcu(net, ifindex); 972 if (!dev) { 973 ret = -ENODEV; 974 goto out; 975 } 976 977 netdev_copy_name(dev, name); 978 979 ret = 0; 980 out: 981 rcu_read_unlock(); 982 return ret; 983 } 984 985 /** 986 * dev_getbyhwaddr_rcu - find a device by its hardware address 987 * @net: the applicable net namespace 988 * @type: media type of device 989 * @ha: hardware address 990 * 991 * Search for an interface by MAC address. Returns NULL if the device 992 * is not found or a pointer to the device. 993 * The caller must hold RCU or RTNL. 994 * The returned device has not had its ref count increased 995 * and the caller must therefore be careful about locking 996 * 997 */ 998 999 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1000 const char *ha) 1001 { 1002 struct net_device *dev; 1003 1004 for_each_netdev_rcu(net, dev) 1005 if (dev->type == type && 1006 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1007 return dev; 1008 1009 return NULL; 1010 } 1011 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1012 1013 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1014 { 1015 struct net_device *dev, *ret = NULL; 1016 1017 rcu_read_lock(); 1018 for_each_netdev_rcu(net, dev) 1019 if (dev->type == type) { 1020 dev_hold(dev); 1021 ret = dev; 1022 break; 1023 } 1024 rcu_read_unlock(); 1025 return ret; 1026 } 1027 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1028 1029 /** 1030 * __dev_get_by_flags - find any device with given flags 1031 * @net: the applicable net namespace 1032 * @if_flags: IFF_* values 1033 * @mask: bitmask of bits in if_flags to check 1034 * 1035 * Search for any interface with the given flags. Returns NULL if a device 1036 * is not found or a pointer to the device. Must be called inside 1037 * rtnl_lock(), and result refcount is unchanged. 1038 */ 1039 1040 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1041 unsigned short mask) 1042 { 1043 struct net_device *dev, *ret; 1044 1045 ASSERT_RTNL(); 1046 1047 ret = NULL; 1048 for_each_netdev(net, dev) { 1049 if (((dev->flags ^ if_flags) & mask) == 0) { 1050 ret = dev; 1051 break; 1052 } 1053 } 1054 return ret; 1055 } 1056 EXPORT_SYMBOL(__dev_get_by_flags); 1057 1058 /** 1059 * dev_valid_name - check if name is okay for network device 1060 * @name: name string 1061 * 1062 * Network device names need to be valid file names to 1063 * allow sysfs to work. We also disallow any kind of 1064 * whitespace. 1065 */ 1066 bool dev_valid_name(const char *name) 1067 { 1068 if (*name == '\0') 1069 return false; 1070 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1071 return false; 1072 if (!strcmp(name, ".") || !strcmp(name, "..")) 1073 return false; 1074 1075 while (*name) { 1076 if (*name == '/' || *name == ':' || isspace(*name)) 1077 return false; 1078 name++; 1079 } 1080 return true; 1081 } 1082 EXPORT_SYMBOL(dev_valid_name); 1083 1084 /** 1085 * __dev_alloc_name - allocate a name for a device 1086 * @net: network namespace to allocate the device name in 1087 * @name: name format string 1088 * @res: result name string 1089 * 1090 * Passed a format string - eg "lt%d" it will try and find a suitable 1091 * id. It scans list of devices to build up a free map, then chooses 1092 * the first empty slot. The caller must hold the dev_base or rtnl lock 1093 * while allocating the name and adding the device in order to avoid 1094 * duplicates. 1095 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1096 * Returns the number of the unit assigned or a negative errno code. 1097 */ 1098 1099 static int __dev_alloc_name(struct net *net, const char *name, char *res) 1100 { 1101 int i = 0; 1102 const char *p; 1103 const int max_netdevices = 8*PAGE_SIZE; 1104 unsigned long *inuse; 1105 struct net_device *d; 1106 char buf[IFNAMSIZ]; 1107 1108 /* Verify the string as this thing may have come from the user. 1109 * There must be one "%d" and no other "%" characters. 1110 */ 1111 p = strchr(name, '%'); 1112 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1113 return -EINVAL; 1114 1115 /* Use one page as a bit array of possible slots */ 1116 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1117 if (!inuse) 1118 return -ENOMEM; 1119 1120 for_each_netdev(net, d) { 1121 struct netdev_name_node *name_node; 1122 1123 netdev_for_each_altname(d, name_node) { 1124 if (!sscanf(name_node->name, name, &i)) 1125 continue; 1126 if (i < 0 || i >= max_netdevices) 1127 continue; 1128 1129 /* avoid cases where sscanf is not exact inverse of printf */ 1130 snprintf(buf, IFNAMSIZ, name, i); 1131 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1132 __set_bit(i, inuse); 1133 } 1134 if (!sscanf(d->name, name, &i)) 1135 continue; 1136 if (i < 0 || i >= max_netdevices) 1137 continue; 1138 1139 /* avoid cases where sscanf is not exact inverse of printf */ 1140 snprintf(buf, IFNAMSIZ, name, i); 1141 if (!strncmp(buf, d->name, IFNAMSIZ)) 1142 __set_bit(i, inuse); 1143 } 1144 1145 i = find_first_zero_bit(inuse, max_netdevices); 1146 bitmap_free(inuse); 1147 if (i == max_netdevices) 1148 return -ENFILE; 1149 1150 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1151 strscpy(buf, name, IFNAMSIZ); 1152 snprintf(res, IFNAMSIZ, buf, i); 1153 return i; 1154 } 1155 1156 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1157 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1158 const char *want_name, char *out_name, 1159 int dup_errno) 1160 { 1161 if (!dev_valid_name(want_name)) 1162 return -EINVAL; 1163 1164 if (strchr(want_name, '%')) 1165 return __dev_alloc_name(net, want_name, out_name); 1166 1167 if (netdev_name_in_use(net, want_name)) 1168 return -dup_errno; 1169 if (out_name != want_name) 1170 strscpy(out_name, want_name, IFNAMSIZ); 1171 return 0; 1172 } 1173 1174 /** 1175 * dev_alloc_name - allocate a name for a device 1176 * @dev: device 1177 * @name: name format string 1178 * 1179 * Passed a format string - eg "lt%d" it will try and find a suitable 1180 * id. It scans list of devices to build up a free map, then chooses 1181 * the first empty slot. The caller must hold the dev_base or rtnl lock 1182 * while allocating the name and adding the device in order to avoid 1183 * duplicates. 1184 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1185 * Returns the number of the unit assigned or a negative errno code. 1186 */ 1187 1188 int dev_alloc_name(struct net_device *dev, const char *name) 1189 { 1190 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1191 } 1192 EXPORT_SYMBOL(dev_alloc_name); 1193 1194 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1195 const char *name) 1196 { 1197 int ret; 1198 1199 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1200 return ret < 0 ? ret : 0; 1201 } 1202 1203 /** 1204 * dev_change_name - change name of a device 1205 * @dev: device 1206 * @newname: name (or format string) must be at least IFNAMSIZ 1207 * 1208 * Change name of a device, can pass format strings "eth%d". 1209 * for wildcarding. 1210 */ 1211 int dev_change_name(struct net_device *dev, const char *newname) 1212 { 1213 unsigned char old_assign_type; 1214 char oldname[IFNAMSIZ]; 1215 int err = 0; 1216 int ret; 1217 struct net *net; 1218 1219 ASSERT_RTNL(); 1220 BUG_ON(!dev_net(dev)); 1221 1222 net = dev_net(dev); 1223 1224 down_write(&devnet_rename_sem); 1225 1226 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1227 up_write(&devnet_rename_sem); 1228 return 0; 1229 } 1230 1231 memcpy(oldname, dev->name, IFNAMSIZ); 1232 1233 write_seqlock_bh(&netdev_rename_lock); 1234 err = dev_get_valid_name(net, dev, newname); 1235 write_sequnlock_bh(&netdev_rename_lock); 1236 1237 if (err < 0) { 1238 up_write(&devnet_rename_sem); 1239 return err; 1240 } 1241 1242 if (oldname[0] && !strchr(oldname, '%')) 1243 netdev_info(dev, "renamed from %s%s\n", oldname, 1244 dev->flags & IFF_UP ? " (while UP)" : ""); 1245 1246 old_assign_type = dev->name_assign_type; 1247 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); 1248 1249 rollback: 1250 ret = device_rename(&dev->dev, dev->name); 1251 if (ret) { 1252 memcpy(dev->name, oldname, IFNAMSIZ); 1253 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1254 up_write(&devnet_rename_sem); 1255 return ret; 1256 } 1257 1258 up_write(&devnet_rename_sem); 1259 1260 netdev_adjacent_rename_links(dev, oldname); 1261 1262 netdev_name_node_del(dev->name_node); 1263 1264 synchronize_net(); 1265 1266 netdev_name_node_add(net, dev->name_node); 1267 1268 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1269 ret = notifier_to_errno(ret); 1270 1271 if (ret) { 1272 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1273 if (err >= 0) { 1274 err = ret; 1275 down_write(&devnet_rename_sem); 1276 write_seqlock_bh(&netdev_rename_lock); 1277 memcpy(dev->name, oldname, IFNAMSIZ); 1278 write_sequnlock_bh(&netdev_rename_lock); 1279 memcpy(oldname, newname, IFNAMSIZ); 1280 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1281 old_assign_type = NET_NAME_RENAMED; 1282 goto rollback; 1283 } else { 1284 netdev_err(dev, "name change rollback failed: %d\n", 1285 ret); 1286 } 1287 } 1288 1289 return err; 1290 } 1291 1292 /** 1293 * dev_set_alias - change ifalias of a device 1294 * @dev: device 1295 * @alias: name up to IFALIASZ 1296 * @len: limit of bytes to copy from info 1297 * 1298 * Set ifalias for a device, 1299 */ 1300 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1301 { 1302 struct dev_ifalias *new_alias = NULL; 1303 1304 if (len >= IFALIASZ) 1305 return -EINVAL; 1306 1307 if (len) { 1308 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1309 if (!new_alias) 1310 return -ENOMEM; 1311 1312 memcpy(new_alias->ifalias, alias, len); 1313 new_alias->ifalias[len] = 0; 1314 } 1315 1316 mutex_lock(&ifalias_mutex); 1317 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1318 mutex_is_locked(&ifalias_mutex)); 1319 mutex_unlock(&ifalias_mutex); 1320 1321 if (new_alias) 1322 kfree_rcu(new_alias, rcuhead); 1323 1324 return len; 1325 } 1326 EXPORT_SYMBOL(dev_set_alias); 1327 1328 /** 1329 * dev_get_alias - get ifalias of a device 1330 * @dev: device 1331 * @name: buffer to store name of ifalias 1332 * @len: size of buffer 1333 * 1334 * get ifalias for a device. Caller must make sure dev cannot go 1335 * away, e.g. rcu read lock or own a reference count to device. 1336 */ 1337 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1338 { 1339 const struct dev_ifalias *alias; 1340 int ret = 0; 1341 1342 rcu_read_lock(); 1343 alias = rcu_dereference(dev->ifalias); 1344 if (alias) 1345 ret = snprintf(name, len, "%s", alias->ifalias); 1346 rcu_read_unlock(); 1347 1348 return ret; 1349 } 1350 1351 /** 1352 * netdev_features_change - device changes features 1353 * @dev: device to cause notification 1354 * 1355 * Called to indicate a device has changed features. 1356 */ 1357 void netdev_features_change(struct net_device *dev) 1358 { 1359 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1360 } 1361 EXPORT_SYMBOL(netdev_features_change); 1362 1363 /** 1364 * netdev_state_change - device changes state 1365 * @dev: device to cause notification 1366 * 1367 * Called to indicate a device has changed state. This function calls 1368 * the notifier chains for netdev_chain and sends a NEWLINK message 1369 * to the routing socket. 1370 */ 1371 void netdev_state_change(struct net_device *dev) 1372 { 1373 if (dev->flags & IFF_UP) { 1374 struct netdev_notifier_change_info change_info = { 1375 .info.dev = dev, 1376 }; 1377 1378 call_netdevice_notifiers_info(NETDEV_CHANGE, 1379 &change_info.info); 1380 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1381 } 1382 } 1383 EXPORT_SYMBOL(netdev_state_change); 1384 1385 /** 1386 * __netdev_notify_peers - notify network peers about existence of @dev, 1387 * to be called when rtnl lock is already held. 1388 * @dev: network device 1389 * 1390 * Generate traffic such that interested network peers are aware of 1391 * @dev, such as by generating a gratuitous ARP. This may be used when 1392 * a device wants to inform the rest of the network about some sort of 1393 * reconfiguration such as a failover event or virtual machine 1394 * migration. 1395 */ 1396 void __netdev_notify_peers(struct net_device *dev) 1397 { 1398 ASSERT_RTNL(); 1399 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1400 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1401 } 1402 EXPORT_SYMBOL(__netdev_notify_peers); 1403 1404 /** 1405 * netdev_notify_peers - notify network peers about existence of @dev 1406 * @dev: network device 1407 * 1408 * Generate traffic such that interested network peers are aware of 1409 * @dev, such as by generating a gratuitous ARP. This may be used when 1410 * a device wants to inform the rest of the network about some sort of 1411 * reconfiguration such as a failover event or virtual machine 1412 * migration. 1413 */ 1414 void netdev_notify_peers(struct net_device *dev) 1415 { 1416 rtnl_lock(); 1417 __netdev_notify_peers(dev); 1418 rtnl_unlock(); 1419 } 1420 EXPORT_SYMBOL(netdev_notify_peers); 1421 1422 static int napi_threaded_poll(void *data); 1423 1424 static int napi_kthread_create(struct napi_struct *n) 1425 { 1426 int err = 0; 1427 1428 /* Create and wake up the kthread once to put it in 1429 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1430 * warning and work with loadavg. 1431 */ 1432 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1433 n->dev->name, n->napi_id); 1434 if (IS_ERR(n->thread)) { 1435 err = PTR_ERR(n->thread); 1436 pr_err("kthread_run failed with err %d\n", err); 1437 n->thread = NULL; 1438 } 1439 1440 return err; 1441 } 1442 1443 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1444 { 1445 const struct net_device_ops *ops = dev->netdev_ops; 1446 int ret; 1447 1448 ASSERT_RTNL(); 1449 dev_addr_check(dev); 1450 1451 if (!netif_device_present(dev)) { 1452 /* may be detached because parent is runtime-suspended */ 1453 if (dev->dev.parent) 1454 pm_runtime_resume(dev->dev.parent); 1455 if (!netif_device_present(dev)) 1456 return -ENODEV; 1457 } 1458 1459 /* Block netpoll from trying to do any rx path servicing. 1460 * If we don't do this there is a chance ndo_poll_controller 1461 * or ndo_poll may be running while we open the device 1462 */ 1463 netpoll_poll_disable(dev); 1464 1465 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1466 ret = notifier_to_errno(ret); 1467 if (ret) 1468 return ret; 1469 1470 set_bit(__LINK_STATE_START, &dev->state); 1471 1472 if (ops->ndo_validate_addr) 1473 ret = ops->ndo_validate_addr(dev); 1474 1475 if (!ret && ops->ndo_open) 1476 ret = ops->ndo_open(dev); 1477 1478 netpoll_poll_enable(dev); 1479 1480 if (ret) 1481 clear_bit(__LINK_STATE_START, &dev->state); 1482 else { 1483 dev->flags |= IFF_UP; 1484 dev_set_rx_mode(dev); 1485 dev_activate(dev); 1486 add_device_randomness(dev->dev_addr, dev->addr_len); 1487 } 1488 1489 return ret; 1490 } 1491 1492 /** 1493 * dev_open - prepare an interface for use. 1494 * @dev: device to open 1495 * @extack: netlink extended ack 1496 * 1497 * Takes a device from down to up state. The device's private open 1498 * function is invoked and then the multicast lists are loaded. Finally 1499 * the device is moved into the up state and a %NETDEV_UP message is 1500 * sent to the netdev notifier chain. 1501 * 1502 * Calling this function on an active interface is a nop. On a failure 1503 * a negative errno code is returned. 1504 */ 1505 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1506 { 1507 int ret; 1508 1509 if (dev->flags & IFF_UP) 1510 return 0; 1511 1512 ret = __dev_open(dev, extack); 1513 if (ret < 0) 1514 return ret; 1515 1516 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1517 call_netdevice_notifiers(NETDEV_UP, dev); 1518 1519 return ret; 1520 } 1521 EXPORT_SYMBOL(dev_open); 1522 1523 static void __dev_close_many(struct list_head *head) 1524 { 1525 struct net_device *dev; 1526 1527 ASSERT_RTNL(); 1528 might_sleep(); 1529 1530 list_for_each_entry(dev, head, close_list) { 1531 /* Temporarily disable netpoll until the interface is down */ 1532 netpoll_poll_disable(dev); 1533 1534 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1535 1536 clear_bit(__LINK_STATE_START, &dev->state); 1537 1538 /* Synchronize to scheduled poll. We cannot touch poll list, it 1539 * can be even on different cpu. So just clear netif_running(). 1540 * 1541 * dev->stop() will invoke napi_disable() on all of it's 1542 * napi_struct instances on this device. 1543 */ 1544 smp_mb__after_atomic(); /* Commit netif_running(). */ 1545 } 1546 1547 dev_deactivate_many(head); 1548 1549 list_for_each_entry(dev, head, close_list) { 1550 const struct net_device_ops *ops = dev->netdev_ops; 1551 1552 /* 1553 * Call the device specific close. This cannot fail. 1554 * Only if device is UP 1555 * 1556 * We allow it to be called even after a DETACH hot-plug 1557 * event. 1558 */ 1559 if (ops->ndo_stop) 1560 ops->ndo_stop(dev); 1561 1562 dev->flags &= ~IFF_UP; 1563 netpoll_poll_enable(dev); 1564 } 1565 } 1566 1567 static void __dev_close(struct net_device *dev) 1568 { 1569 LIST_HEAD(single); 1570 1571 list_add(&dev->close_list, &single); 1572 __dev_close_many(&single); 1573 list_del(&single); 1574 } 1575 1576 void dev_close_many(struct list_head *head, bool unlink) 1577 { 1578 struct net_device *dev, *tmp; 1579 1580 /* Remove the devices that don't need to be closed */ 1581 list_for_each_entry_safe(dev, tmp, head, close_list) 1582 if (!(dev->flags & IFF_UP)) 1583 list_del_init(&dev->close_list); 1584 1585 __dev_close_many(head); 1586 1587 list_for_each_entry_safe(dev, tmp, head, close_list) { 1588 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1589 call_netdevice_notifiers(NETDEV_DOWN, dev); 1590 if (unlink) 1591 list_del_init(&dev->close_list); 1592 } 1593 } 1594 EXPORT_SYMBOL(dev_close_many); 1595 1596 /** 1597 * dev_close - shutdown an interface. 1598 * @dev: device to shutdown 1599 * 1600 * This function moves an active device into down state. A 1601 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1602 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1603 * chain. 1604 */ 1605 void dev_close(struct net_device *dev) 1606 { 1607 if (dev->flags & IFF_UP) { 1608 LIST_HEAD(single); 1609 1610 list_add(&dev->close_list, &single); 1611 dev_close_many(&single, true); 1612 list_del(&single); 1613 } 1614 } 1615 EXPORT_SYMBOL(dev_close); 1616 1617 1618 /** 1619 * dev_disable_lro - disable Large Receive Offload on a device 1620 * @dev: device 1621 * 1622 * Disable Large Receive Offload (LRO) on a net device. Must be 1623 * called under RTNL. This is needed if received packets may be 1624 * forwarded to another interface. 1625 */ 1626 void dev_disable_lro(struct net_device *dev) 1627 { 1628 struct net_device *lower_dev; 1629 struct list_head *iter; 1630 1631 dev->wanted_features &= ~NETIF_F_LRO; 1632 netdev_update_features(dev); 1633 1634 if (unlikely(dev->features & NETIF_F_LRO)) 1635 netdev_WARN(dev, "failed to disable LRO!\n"); 1636 1637 netdev_for_each_lower_dev(dev, lower_dev, iter) 1638 dev_disable_lro(lower_dev); 1639 } 1640 EXPORT_SYMBOL(dev_disable_lro); 1641 1642 /** 1643 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1644 * @dev: device 1645 * 1646 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1647 * called under RTNL. This is needed if Generic XDP is installed on 1648 * the device. 1649 */ 1650 static void dev_disable_gro_hw(struct net_device *dev) 1651 { 1652 dev->wanted_features &= ~NETIF_F_GRO_HW; 1653 netdev_update_features(dev); 1654 1655 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1656 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1657 } 1658 1659 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1660 { 1661 #define N(val) \ 1662 case NETDEV_##val: \ 1663 return "NETDEV_" __stringify(val); 1664 switch (cmd) { 1665 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1666 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1667 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1668 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1669 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1670 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1671 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1672 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1673 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1674 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1675 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1676 N(XDP_FEAT_CHANGE) 1677 } 1678 #undef N 1679 return "UNKNOWN_NETDEV_EVENT"; 1680 } 1681 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1682 1683 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1684 struct net_device *dev) 1685 { 1686 struct netdev_notifier_info info = { 1687 .dev = dev, 1688 }; 1689 1690 return nb->notifier_call(nb, val, &info); 1691 } 1692 1693 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1694 struct net_device *dev) 1695 { 1696 int err; 1697 1698 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1699 err = notifier_to_errno(err); 1700 if (err) 1701 return err; 1702 1703 if (!(dev->flags & IFF_UP)) 1704 return 0; 1705 1706 call_netdevice_notifier(nb, NETDEV_UP, dev); 1707 return 0; 1708 } 1709 1710 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1711 struct net_device *dev) 1712 { 1713 if (dev->flags & IFF_UP) { 1714 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1715 dev); 1716 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1717 } 1718 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1719 } 1720 1721 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1722 struct net *net) 1723 { 1724 struct net_device *dev; 1725 int err; 1726 1727 for_each_netdev(net, dev) { 1728 err = call_netdevice_register_notifiers(nb, dev); 1729 if (err) 1730 goto rollback; 1731 } 1732 return 0; 1733 1734 rollback: 1735 for_each_netdev_continue_reverse(net, dev) 1736 call_netdevice_unregister_notifiers(nb, dev); 1737 return err; 1738 } 1739 1740 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1741 struct net *net) 1742 { 1743 struct net_device *dev; 1744 1745 for_each_netdev(net, dev) 1746 call_netdevice_unregister_notifiers(nb, dev); 1747 } 1748 1749 static int dev_boot_phase = 1; 1750 1751 /** 1752 * register_netdevice_notifier - register a network notifier block 1753 * @nb: notifier 1754 * 1755 * Register a notifier to be called when network device events occur. 1756 * The notifier passed is linked into the kernel structures and must 1757 * not be reused until it has been unregistered. A negative errno code 1758 * is returned on a failure. 1759 * 1760 * When registered all registration and up events are replayed 1761 * to the new notifier to allow device to have a race free 1762 * view of the network device list. 1763 */ 1764 1765 int register_netdevice_notifier(struct notifier_block *nb) 1766 { 1767 struct net *net; 1768 int err; 1769 1770 /* Close race with setup_net() and cleanup_net() */ 1771 down_write(&pernet_ops_rwsem); 1772 1773 /* When RTNL is removed, we need protection for netdev_chain. */ 1774 rtnl_lock(); 1775 1776 err = raw_notifier_chain_register(&netdev_chain, nb); 1777 if (err) 1778 goto unlock; 1779 if (dev_boot_phase) 1780 goto unlock; 1781 for_each_net(net) { 1782 __rtnl_net_lock(net); 1783 err = call_netdevice_register_net_notifiers(nb, net); 1784 __rtnl_net_unlock(net); 1785 if (err) 1786 goto rollback; 1787 } 1788 1789 unlock: 1790 rtnl_unlock(); 1791 up_write(&pernet_ops_rwsem); 1792 return err; 1793 1794 rollback: 1795 for_each_net_continue_reverse(net) { 1796 __rtnl_net_lock(net); 1797 call_netdevice_unregister_net_notifiers(nb, net); 1798 __rtnl_net_unlock(net); 1799 } 1800 1801 raw_notifier_chain_unregister(&netdev_chain, nb); 1802 goto unlock; 1803 } 1804 EXPORT_SYMBOL(register_netdevice_notifier); 1805 1806 /** 1807 * unregister_netdevice_notifier - unregister a network notifier block 1808 * @nb: notifier 1809 * 1810 * Unregister a notifier previously registered by 1811 * register_netdevice_notifier(). The notifier is unlinked into the 1812 * kernel structures and may then be reused. A negative errno code 1813 * is returned on a failure. 1814 * 1815 * After unregistering unregister and down device events are synthesized 1816 * for all devices on the device list to the removed notifier to remove 1817 * the need for special case cleanup code. 1818 */ 1819 1820 int unregister_netdevice_notifier(struct notifier_block *nb) 1821 { 1822 struct net *net; 1823 int err; 1824 1825 /* Close race with setup_net() and cleanup_net() */ 1826 down_write(&pernet_ops_rwsem); 1827 rtnl_lock(); 1828 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1829 if (err) 1830 goto unlock; 1831 1832 for_each_net(net) { 1833 __rtnl_net_lock(net); 1834 call_netdevice_unregister_net_notifiers(nb, net); 1835 __rtnl_net_unlock(net); 1836 } 1837 1838 unlock: 1839 rtnl_unlock(); 1840 up_write(&pernet_ops_rwsem); 1841 return err; 1842 } 1843 EXPORT_SYMBOL(unregister_netdevice_notifier); 1844 1845 static int __register_netdevice_notifier_net(struct net *net, 1846 struct notifier_block *nb, 1847 bool ignore_call_fail) 1848 { 1849 int err; 1850 1851 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1852 if (err) 1853 return err; 1854 if (dev_boot_phase) 1855 return 0; 1856 1857 err = call_netdevice_register_net_notifiers(nb, net); 1858 if (err && !ignore_call_fail) 1859 goto chain_unregister; 1860 1861 return 0; 1862 1863 chain_unregister: 1864 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1865 return err; 1866 } 1867 1868 static int __unregister_netdevice_notifier_net(struct net *net, 1869 struct notifier_block *nb) 1870 { 1871 int err; 1872 1873 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1874 if (err) 1875 return err; 1876 1877 call_netdevice_unregister_net_notifiers(nb, net); 1878 return 0; 1879 } 1880 1881 /** 1882 * register_netdevice_notifier_net - register a per-netns network notifier block 1883 * @net: network namespace 1884 * @nb: notifier 1885 * 1886 * Register a notifier to be called when network device events occur. 1887 * The notifier passed is linked into the kernel structures and must 1888 * not be reused until it has been unregistered. A negative errno code 1889 * is returned on a failure. 1890 * 1891 * When registered all registration and up events are replayed 1892 * to the new notifier to allow device to have a race free 1893 * view of the network device list. 1894 */ 1895 1896 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1897 { 1898 int err; 1899 1900 rtnl_net_lock(net); 1901 err = __register_netdevice_notifier_net(net, nb, false); 1902 rtnl_net_unlock(net); 1903 1904 return err; 1905 } 1906 EXPORT_SYMBOL(register_netdevice_notifier_net); 1907 1908 /** 1909 * unregister_netdevice_notifier_net - unregister a per-netns 1910 * network notifier block 1911 * @net: network namespace 1912 * @nb: notifier 1913 * 1914 * Unregister a notifier previously registered by 1915 * register_netdevice_notifier_net(). The notifier is unlinked from the 1916 * kernel structures and may then be reused. A negative errno code 1917 * is returned on a failure. 1918 * 1919 * After unregistering unregister and down device events are synthesized 1920 * for all devices on the device list to the removed notifier to remove 1921 * the need for special case cleanup code. 1922 */ 1923 1924 int unregister_netdevice_notifier_net(struct net *net, 1925 struct notifier_block *nb) 1926 { 1927 int err; 1928 1929 rtnl_net_lock(net); 1930 err = __unregister_netdevice_notifier_net(net, nb); 1931 rtnl_net_unlock(net); 1932 1933 return err; 1934 } 1935 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1936 1937 static void __move_netdevice_notifier_net(struct net *src_net, 1938 struct net *dst_net, 1939 struct notifier_block *nb) 1940 { 1941 __unregister_netdevice_notifier_net(src_net, nb); 1942 __register_netdevice_notifier_net(dst_net, nb, true); 1943 } 1944 1945 int register_netdevice_notifier_dev_net(struct net_device *dev, 1946 struct notifier_block *nb, 1947 struct netdev_net_notifier *nn) 1948 { 1949 struct net *net = dev_net(dev); 1950 int err; 1951 1952 rtnl_net_lock(net); 1953 err = __register_netdevice_notifier_net(net, nb, false); 1954 if (!err) { 1955 nn->nb = nb; 1956 list_add(&nn->list, &dev->net_notifier_list); 1957 } 1958 rtnl_net_unlock(net); 1959 1960 return err; 1961 } 1962 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1963 1964 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1965 struct notifier_block *nb, 1966 struct netdev_net_notifier *nn) 1967 { 1968 struct net *net = dev_net(dev); 1969 int err; 1970 1971 rtnl_net_lock(net); 1972 list_del(&nn->list); 1973 err = __unregister_netdevice_notifier_net(net, nb); 1974 rtnl_net_unlock(net); 1975 1976 return err; 1977 } 1978 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1979 1980 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1981 struct net *net) 1982 { 1983 struct netdev_net_notifier *nn; 1984 1985 list_for_each_entry(nn, &dev->net_notifier_list, list) 1986 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1987 } 1988 1989 /** 1990 * call_netdevice_notifiers_info - call all network notifier blocks 1991 * @val: value passed unmodified to notifier function 1992 * @info: notifier information data 1993 * 1994 * Call all network notifier blocks. Parameters and return value 1995 * are as for raw_notifier_call_chain(). 1996 */ 1997 1998 int call_netdevice_notifiers_info(unsigned long val, 1999 struct netdev_notifier_info *info) 2000 { 2001 struct net *net = dev_net(info->dev); 2002 int ret; 2003 2004 ASSERT_RTNL(); 2005 2006 /* Run per-netns notifier block chain first, then run the global one. 2007 * Hopefully, one day, the global one is going to be removed after 2008 * all notifier block registrators get converted to be per-netns. 2009 */ 2010 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2011 if (ret & NOTIFY_STOP_MASK) 2012 return ret; 2013 return raw_notifier_call_chain(&netdev_chain, val, info); 2014 } 2015 2016 /** 2017 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2018 * for and rollback on error 2019 * @val_up: value passed unmodified to notifier function 2020 * @val_down: value passed unmodified to the notifier function when 2021 * recovering from an error on @val_up 2022 * @info: notifier information data 2023 * 2024 * Call all per-netns network notifier blocks, but not notifier blocks on 2025 * the global notifier chain. Parameters and return value are as for 2026 * raw_notifier_call_chain_robust(). 2027 */ 2028 2029 static int 2030 call_netdevice_notifiers_info_robust(unsigned long val_up, 2031 unsigned long val_down, 2032 struct netdev_notifier_info *info) 2033 { 2034 struct net *net = dev_net(info->dev); 2035 2036 ASSERT_RTNL(); 2037 2038 return raw_notifier_call_chain_robust(&net->netdev_chain, 2039 val_up, val_down, info); 2040 } 2041 2042 static int call_netdevice_notifiers_extack(unsigned long val, 2043 struct net_device *dev, 2044 struct netlink_ext_ack *extack) 2045 { 2046 struct netdev_notifier_info info = { 2047 .dev = dev, 2048 .extack = extack, 2049 }; 2050 2051 return call_netdevice_notifiers_info(val, &info); 2052 } 2053 2054 /** 2055 * call_netdevice_notifiers - call all network notifier blocks 2056 * @val: value passed unmodified to notifier function 2057 * @dev: net_device pointer passed unmodified to notifier function 2058 * 2059 * Call all network notifier blocks. Parameters and return value 2060 * are as for raw_notifier_call_chain(). 2061 */ 2062 2063 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2064 { 2065 return call_netdevice_notifiers_extack(val, dev, NULL); 2066 } 2067 EXPORT_SYMBOL(call_netdevice_notifiers); 2068 2069 /** 2070 * call_netdevice_notifiers_mtu - call all network notifier blocks 2071 * @val: value passed unmodified to notifier function 2072 * @dev: net_device pointer passed unmodified to notifier function 2073 * @arg: additional u32 argument passed to the notifier function 2074 * 2075 * Call all network notifier blocks. Parameters and return value 2076 * are as for raw_notifier_call_chain(). 2077 */ 2078 static int call_netdevice_notifiers_mtu(unsigned long val, 2079 struct net_device *dev, u32 arg) 2080 { 2081 struct netdev_notifier_info_ext info = { 2082 .info.dev = dev, 2083 .ext.mtu = arg, 2084 }; 2085 2086 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2087 2088 return call_netdevice_notifiers_info(val, &info.info); 2089 } 2090 2091 #ifdef CONFIG_NET_INGRESS 2092 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2093 2094 void net_inc_ingress_queue(void) 2095 { 2096 static_branch_inc(&ingress_needed_key); 2097 } 2098 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2099 2100 void net_dec_ingress_queue(void) 2101 { 2102 static_branch_dec(&ingress_needed_key); 2103 } 2104 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2105 #endif 2106 2107 #ifdef CONFIG_NET_EGRESS 2108 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2109 2110 void net_inc_egress_queue(void) 2111 { 2112 static_branch_inc(&egress_needed_key); 2113 } 2114 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2115 2116 void net_dec_egress_queue(void) 2117 { 2118 static_branch_dec(&egress_needed_key); 2119 } 2120 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2121 #endif 2122 2123 #ifdef CONFIG_NET_CLS_ACT 2124 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key); 2125 EXPORT_SYMBOL(tcf_bypass_check_needed_key); 2126 #endif 2127 2128 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2129 EXPORT_SYMBOL(netstamp_needed_key); 2130 #ifdef CONFIG_JUMP_LABEL 2131 static atomic_t netstamp_needed_deferred; 2132 static atomic_t netstamp_wanted; 2133 static void netstamp_clear(struct work_struct *work) 2134 { 2135 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2136 int wanted; 2137 2138 wanted = atomic_add_return(deferred, &netstamp_wanted); 2139 if (wanted > 0) 2140 static_branch_enable(&netstamp_needed_key); 2141 else 2142 static_branch_disable(&netstamp_needed_key); 2143 } 2144 static DECLARE_WORK(netstamp_work, netstamp_clear); 2145 #endif 2146 2147 void net_enable_timestamp(void) 2148 { 2149 #ifdef CONFIG_JUMP_LABEL 2150 int wanted = atomic_read(&netstamp_wanted); 2151 2152 while (wanted > 0) { 2153 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2154 return; 2155 } 2156 atomic_inc(&netstamp_needed_deferred); 2157 schedule_work(&netstamp_work); 2158 #else 2159 static_branch_inc(&netstamp_needed_key); 2160 #endif 2161 } 2162 EXPORT_SYMBOL(net_enable_timestamp); 2163 2164 void net_disable_timestamp(void) 2165 { 2166 #ifdef CONFIG_JUMP_LABEL 2167 int wanted = atomic_read(&netstamp_wanted); 2168 2169 while (wanted > 1) { 2170 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2171 return; 2172 } 2173 atomic_dec(&netstamp_needed_deferred); 2174 schedule_work(&netstamp_work); 2175 #else 2176 static_branch_dec(&netstamp_needed_key); 2177 #endif 2178 } 2179 EXPORT_SYMBOL(net_disable_timestamp); 2180 2181 static inline void net_timestamp_set(struct sk_buff *skb) 2182 { 2183 skb->tstamp = 0; 2184 skb->tstamp_type = SKB_CLOCK_REALTIME; 2185 if (static_branch_unlikely(&netstamp_needed_key)) 2186 skb->tstamp = ktime_get_real(); 2187 } 2188 2189 #define net_timestamp_check(COND, SKB) \ 2190 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2191 if ((COND) && !(SKB)->tstamp) \ 2192 (SKB)->tstamp = ktime_get_real(); \ 2193 } \ 2194 2195 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2196 { 2197 return __is_skb_forwardable(dev, skb, true); 2198 } 2199 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2200 2201 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2202 bool check_mtu) 2203 { 2204 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2205 2206 if (likely(!ret)) { 2207 skb->protocol = eth_type_trans(skb, dev); 2208 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2209 } 2210 2211 return ret; 2212 } 2213 2214 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2215 { 2216 return __dev_forward_skb2(dev, skb, true); 2217 } 2218 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2219 2220 /** 2221 * dev_forward_skb - loopback an skb to another netif 2222 * 2223 * @dev: destination network device 2224 * @skb: buffer to forward 2225 * 2226 * return values: 2227 * NET_RX_SUCCESS (no congestion) 2228 * NET_RX_DROP (packet was dropped, but freed) 2229 * 2230 * dev_forward_skb can be used for injecting an skb from the 2231 * start_xmit function of one device into the receive queue 2232 * of another device. 2233 * 2234 * The receiving device may be in another namespace, so 2235 * we have to clear all information in the skb that could 2236 * impact namespace isolation. 2237 */ 2238 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2239 { 2240 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2241 } 2242 EXPORT_SYMBOL_GPL(dev_forward_skb); 2243 2244 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2245 { 2246 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2247 } 2248 2249 static inline int deliver_skb(struct sk_buff *skb, 2250 struct packet_type *pt_prev, 2251 struct net_device *orig_dev) 2252 { 2253 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2254 return -ENOMEM; 2255 refcount_inc(&skb->users); 2256 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2257 } 2258 2259 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2260 struct packet_type **pt, 2261 struct net_device *orig_dev, 2262 __be16 type, 2263 struct list_head *ptype_list) 2264 { 2265 struct packet_type *ptype, *pt_prev = *pt; 2266 2267 list_for_each_entry_rcu(ptype, ptype_list, list) { 2268 if (ptype->type != type) 2269 continue; 2270 if (pt_prev) 2271 deliver_skb(skb, pt_prev, orig_dev); 2272 pt_prev = ptype; 2273 } 2274 *pt = pt_prev; 2275 } 2276 2277 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2278 { 2279 if (!ptype->af_packet_priv || !skb->sk) 2280 return false; 2281 2282 if (ptype->id_match) 2283 return ptype->id_match(ptype, skb->sk); 2284 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2285 return true; 2286 2287 return false; 2288 } 2289 2290 /** 2291 * dev_nit_active - return true if any network interface taps are in use 2292 * 2293 * @dev: network device to check for the presence of taps 2294 */ 2295 bool dev_nit_active(struct net_device *dev) 2296 { 2297 return !list_empty(&net_hotdata.ptype_all) || 2298 !list_empty(&dev->ptype_all); 2299 } 2300 EXPORT_SYMBOL_GPL(dev_nit_active); 2301 2302 /* 2303 * Support routine. Sends outgoing frames to any network 2304 * taps currently in use. 2305 */ 2306 2307 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2308 { 2309 struct list_head *ptype_list = &net_hotdata.ptype_all; 2310 struct packet_type *ptype, *pt_prev = NULL; 2311 struct sk_buff *skb2 = NULL; 2312 2313 rcu_read_lock(); 2314 again: 2315 list_for_each_entry_rcu(ptype, ptype_list, list) { 2316 if (READ_ONCE(ptype->ignore_outgoing)) 2317 continue; 2318 2319 /* Never send packets back to the socket 2320 * they originated from - MvS (miquels@drinkel.ow.org) 2321 */ 2322 if (skb_loop_sk(ptype, skb)) 2323 continue; 2324 2325 if (pt_prev) { 2326 deliver_skb(skb2, pt_prev, skb->dev); 2327 pt_prev = ptype; 2328 continue; 2329 } 2330 2331 /* need to clone skb, done only once */ 2332 skb2 = skb_clone(skb, GFP_ATOMIC); 2333 if (!skb2) 2334 goto out_unlock; 2335 2336 net_timestamp_set(skb2); 2337 2338 /* skb->nh should be correctly 2339 * set by sender, so that the second statement is 2340 * just protection against buggy protocols. 2341 */ 2342 skb_reset_mac_header(skb2); 2343 2344 if (skb_network_header(skb2) < skb2->data || 2345 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2346 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2347 ntohs(skb2->protocol), 2348 dev->name); 2349 skb_reset_network_header(skb2); 2350 } 2351 2352 skb2->transport_header = skb2->network_header; 2353 skb2->pkt_type = PACKET_OUTGOING; 2354 pt_prev = ptype; 2355 } 2356 2357 if (ptype_list == &net_hotdata.ptype_all) { 2358 ptype_list = &dev->ptype_all; 2359 goto again; 2360 } 2361 out_unlock: 2362 if (pt_prev) { 2363 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2364 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2365 else 2366 kfree_skb(skb2); 2367 } 2368 rcu_read_unlock(); 2369 } 2370 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2371 2372 /** 2373 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2374 * @dev: Network device 2375 * @txq: number of queues available 2376 * 2377 * If real_num_tx_queues is changed the tc mappings may no longer be 2378 * valid. To resolve this verify the tc mapping remains valid and if 2379 * not NULL the mapping. With no priorities mapping to this 2380 * offset/count pair it will no longer be used. In the worst case TC0 2381 * is invalid nothing can be done so disable priority mappings. If is 2382 * expected that drivers will fix this mapping if they can before 2383 * calling netif_set_real_num_tx_queues. 2384 */ 2385 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2386 { 2387 int i; 2388 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2389 2390 /* If TC0 is invalidated disable TC mapping */ 2391 if (tc->offset + tc->count > txq) { 2392 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2393 dev->num_tc = 0; 2394 return; 2395 } 2396 2397 /* Invalidated prio to tc mappings set to TC0 */ 2398 for (i = 1; i < TC_BITMASK + 1; i++) { 2399 int q = netdev_get_prio_tc_map(dev, i); 2400 2401 tc = &dev->tc_to_txq[q]; 2402 if (tc->offset + tc->count > txq) { 2403 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", 2404 i, q); 2405 netdev_set_prio_tc_map(dev, i, 0); 2406 } 2407 } 2408 } 2409 2410 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2411 { 2412 if (dev->num_tc) { 2413 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2414 int i; 2415 2416 /* walk through the TCs and see if it falls into any of them */ 2417 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2418 if ((txq - tc->offset) < tc->count) 2419 return i; 2420 } 2421 2422 /* didn't find it, just return -1 to indicate no match */ 2423 return -1; 2424 } 2425 2426 return 0; 2427 } 2428 EXPORT_SYMBOL(netdev_txq_to_tc); 2429 2430 #ifdef CONFIG_XPS 2431 static struct static_key xps_needed __read_mostly; 2432 static struct static_key xps_rxqs_needed __read_mostly; 2433 static DEFINE_MUTEX(xps_map_mutex); 2434 #define xmap_dereference(P) \ 2435 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2436 2437 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2438 struct xps_dev_maps *old_maps, int tci, u16 index) 2439 { 2440 struct xps_map *map = NULL; 2441 int pos; 2442 2443 map = xmap_dereference(dev_maps->attr_map[tci]); 2444 if (!map) 2445 return false; 2446 2447 for (pos = map->len; pos--;) { 2448 if (map->queues[pos] != index) 2449 continue; 2450 2451 if (map->len > 1) { 2452 map->queues[pos] = map->queues[--map->len]; 2453 break; 2454 } 2455 2456 if (old_maps) 2457 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2458 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2459 kfree_rcu(map, rcu); 2460 return false; 2461 } 2462 2463 return true; 2464 } 2465 2466 static bool remove_xps_queue_cpu(struct net_device *dev, 2467 struct xps_dev_maps *dev_maps, 2468 int cpu, u16 offset, u16 count) 2469 { 2470 int num_tc = dev_maps->num_tc; 2471 bool active = false; 2472 int tci; 2473 2474 for (tci = cpu * num_tc; num_tc--; tci++) { 2475 int i, j; 2476 2477 for (i = count, j = offset; i--; j++) { 2478 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2479 break; 2480 } 2481 2482 active |= i < 0; 2483 } 2484 2485 return active; 2486 } 2487 2488 static void reset_xps_maps(struct net_device *dev, 2489 struct xps_dev_maps *dev_maps, 2490 enum xps_map_type type) 2491 { 2492 static_key_slow_dec_cpuslocked(&xps_needed); 2493 if (type == XPS_RXQS) 2494 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2495 2496 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2497 2498 kfree_rcu(dev_maps, rcu); 2499 } 2500 2501 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2502 u16 offset, u16 count) 2503 { 2504 struct xps_dev_maps *dev_maps; 2505 bool active = false; 2506 int i, j; 2507 2508 dev_maps = xmap_dereference(dev->xps_maps[type]); 2509 if (!dev_maps) 2510 return; 2511 2512 for (j = 0; j < dev_maps->nr_ids; j++) 2513 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2514 if (!active) 2515 reset_xps_maps(dev, dev_maps, type); 2516 2517 if (type == XPS_CPUS) { 2518 for (i = offset + (count - 1); count--; i--) 2519 netdev_queue_numa_node_write( 2520 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2521 } 2522 } 2523 2524 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2525 u16 count) 2526 { 2527 if (!static_key_false(&xps_needed)) 2528 return; 2529 2530 cpus_read_lock(); 2531 mutex_lock(&xps_map_mutex); 2532 2533 if (static_key_false(&xps_rxqs_needed)) 2534 clean_xps_maps(dev, XPS_RXQS, offset, count); 2535 2536 clean_xps_maps(dev, XPS_CPUS, offset, count); 2537 2538 mutex_unlock(&xps_map_mutex); 2539 cpus_read_unlock(); 2540 } 2541 2542 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2543 { 2544 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2545 } 2546 2547 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2548 u16 index, bool is_rxqs_map) 2549 { 2550 struct xps_map *new_map; 2551 int alloc_len = XPS_MIN_MAP_ALLOC; 2552 int i, pos; 2553 2554 for (pos = 0; map && pos < map->len; pos++) { 2555 if (map->queues[pos] != index) 2556 continue; 2557 return map; 2558 } 2559 2560 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2561 if (map) { 2562 if (pos < map->alloc_len) 2563 return map; 2564 2565 alloc_len = map->alloc_len * 2; 2566 } 2567 2568 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2569 * map 2570 */ 2571 if (is_rxqs_map) 2572 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2573 else 2574 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2575 cpu_to_node(attr_index)); 2576 if (!new_map) 2577 return NULL; 2578 2579 for (i = 0; i < pos; i++) 2580 new_map->queues[i] = map->queues[i]; 2581 new_map->alloc_len = alloc_len; 2582 new_map->len = pos; 2583 2584 return new_map; 2585 } 2586 2587 /* Copy xps maps at a given index */ 2588 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2589 struct xps_dev_maps *new_dev_maps, int index, 2590 int tc, bool skip_tc) 2591 { 2592 int i, tci = index * dev_maps->num_tc; 2593 struct xps_map *map; 2594 2595 /* copy maps belonging to foreign traffic classes */ 2596 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2597 if (i == tc && skip_tc) 2598 continue; 2599 2600 /* fill in the new device map from the old device map */ 2601 map = xmap_dereference(dev_maps->attr_map[tci]); 2602 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2603 } 2604 } 2605 2606 /* Must be called under cpus_read_lock */ 2607 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2608 u16 index, enum xps_map_type type) 2609 { 2610 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2611 const unsigned long *online_mask = NULL; 2612 bool active = false, copy = false; 2613 int i, j, tci, numa_node_id = -2; 2614 int maps_sz, num_tc = 1, tc = 0; 2615 struct xps_map *map, *new_map; 2616 unsigned int nr_ids; 2617 2618 WARN_ON_ONCE(index >= dev->num_tx_queues); 2619 2620 if (dev->num_tc) { 2621 /* Do not allow XPS on subordinate device directly */ 2622 num_tc = dev->num_tc; 2623 if (num_tc < 0) 2624 return -EINVAL; 2625 2626 /* If queue belongs to subordinate dev use its map */ 2627 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2628 2629 tc = netdev_txq_to_tc(dev, index); 2630 if (tc < 0) 2631 return -EINVAL; 2632 } 2633 2634 mutex_lock(&xps_map_mutex); 2635 2636 dev_maps = xmap_dereference(dev->xps_maps[type]); 2637 if (type == XPS_RXQS) { 2638 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2639 nr_ids = dev->num_rx_queues; 2640 } else { 2641 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2642 if (num_possible_cpus() > 1) 2643 online_mask = cpumask_bits(cpu_online_mask); 2644 nr_ids = nr_cpu_ids; 2645 } 2646 2647 if (maps_sz < L1_CACHE_BYTES) 2648 maps_sz = L1_CACHE_BYTES; 2649 2650 /* The old dev_maps could be larger or smaller than the one we're 2651 * setting up now, as dev->num_tc or nr_ids could have been updated in 2652 * between. We could try to be smart, but let's be safe instead and only 2653 * copy foreign traffic classes if the two map sizes match. 2654 */ 2655 if (dev_maps && 2656 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2657 copy = true; 2658 2659 /* allocate memory for queue storage */ 2660 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2661 j < nr_ids;) { 2662 if (!new_dev_maps) { 2663 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2664 if (!new_dev_maps) { 2665 mutex_unlock(&xps_map_mutex); 2666 return -ENOMEM; 2667 } 2668 2669 new_dev_maps->nr_ids = nr_ids; 2670 new_dev_maps->num_tc = num_tc; 2671 } 2672 2673 tci = j * num_tc + tc; 2674 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2675 2676 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2677 if (!map) 2678 goto error; 2679 2680 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2681 } 2682 2683 if (!new_dev_maps) 2684 goto out_no_new_maps; 2685 2686 if (!dev_maps) { 2687 /* Increment static keys at most once per type */ 2688 static_key_slow_inc_cpuslocked(&xps_needed); 2689 if (type == XPS_RXQS) 2690 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2691 } 2692 2693 for (j = 0; j < nr_ids; j++) { 2694 bool skip_tc = false; 2695 2696 tci = j * num_tc + tc; 2697 if (netif_attr_test_mask(j, mask, nr_ids) && 2698 netif_attr_test_online(j, online_mask, nr_ids)) { 2699 /* add tx-queue to CPU/rx-queue maps */ 2700 int pos = 0; 2701 2702 skip_tc = true; 2703 2704 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2705 while ((pos < map->len) && (map->queues[pos] != index)) 2706 pos++; 2707 2708 if (pos == map->len) 2709 map->queues[map->len++] = index; 2710 #ifdef CONFIG_NUMA 2711 if (type == XPS_CPUS) { 2712 if (numa_node_id == -2) 2713 numa_node_id = cpu_to_node(j); 2714 else if (numa_node_id != cpu_to_node(j)) 2715 numa_node_id = -1; 2716 } 2717 #endif 2718 } 2719 2720 if (copy) 2721 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2722 skip_tc); 2723 } 2724 2725 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2726 2727 /* Cleanup old maps */ 2728 if (!dev_maps) 2729 goto out_no_old_maps; 2730 2731 for (j = 0; j < dev_maps->nr_ids; j++) { 2732 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2733 map = xmap_dereference(dev_maps->attr_map[tci]); 2734 if (!map) 2735 continue; 2736 2737 if (copy) { 2738 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2739 if (map == new_map) 2740 continue; 2741 } 2742 2743 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2744 kfree_rcu(map, rcu); 2745 } 2746 } 2747 2748 old_dev_maps = dev_maps; 2749 2750 out_no_old_maps: 2751 dev_maps = new_dev_maps; 2752 active = true; 2753 2754 out_no_new_maps: 2755 if (type == XPS_CPUS) 2756 /* update Tx queue numa node */ 2757 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2758 (numa_node_id >= 0) ? 2759 numa_node_id : NUMA_NO_NODE); 2760 2761 if (!dev_maps) 2762 goto out_no_maps; 2763 2764 /* removes tx-queue from unused CPUs/rx-queues */ 2765 for (j = 0; j < dev_maps->nr_ids; j++) { 2766 tci = j * dev_maps->num_tc; 2767 2768 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2769 if (i == tc && 2770 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2771 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2772 continue; 2773 2774 active |= remove_xps_queue(dev_maps, 2775 copy ? old_dev_maps : NULL, 2776 tci, index); 2777 } 2778 } 2779 2780 if (old_dev_maps) 2781 kfree_rcu(old_dev_maps, rcu); 2782 2783 /* free map if not active */ 2784 if (!active) 2785 reset_xps_maps(dev, dev_maps, type); 2786 2787 out_no_maps: 2788 mutex_unlock(&xps_map_mutex); 2789 2790 return 0; 2791 error: 2792 /* remove any maps that we added */ 2793 for (j = 0; j < nr_ids; j++) { 2794 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2795 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2796 map = copy ? 2797 xmap_dereference(dev_maps->attr_map[tci]) : 2798 NULL; 2799 if (new_map && new_map != map) 2800 kfree(new_map); 2801 } 2802 } 2803 2804 mutex_unlock(&xps_map_mutex); 2805 2806 kfree(new_dev_maps); 2807 return -ENOMEM; 2808 } 2809 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2810 2811 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2812 u16 index) 2813 { 2814 int ret; 2815 2816 cpus_read_lock(); 2817 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2818 cpus_read_unlock(); 2819 2820 return ret; 2821 } 2822 EXPORT_SYMBOL(netif_set_xps_queue); 2823 2824 #endif 2825 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2826 { 2827 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2828 2829 /* Unbind any subordinate channels */ 2830 while (txq-- != &dev->_tx[0]) { 2831 if (txq->sb_dev) 2832 netdev_unbind_sb_channel(dev, txq->sb_dev); 2833 } 2834 } 2835 2836 void netdev_reset_tc(struct net_device *dev) 2837 { 2838 #ifdef CONFIG_XPS 2839 netif_reset_xps_queues_gt(dev, 0); 2840 #endif 2841 netdev_unbind_all_sb_channels(dev); 2842 2843 /* Reset TC configuration of device */ 2844 dev->num_tc = 0; 2845 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2846 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2847 } 2848 EXPORT_SYMBOL(netdev_reset_tc); 2849 2850 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2851 { 2852 if (tc >= dev->num_tc) 2853 return -EINVAL; 2854 2855 #ifdef CONFIG_XPS 2856 netif_reset_xps_queues(dev, offset, count); 2857 #endif 2858 dev->tc_to_txq[tc].count = count; 2859 dev->tc_to_txq[tc].offset = offset; 2860 return 0; 2861 } 2862 EXPORT_SYMBOL(netdev_set_tc_queue); 2863 2864 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2865 { 2866 if (num_tc > TC_MAX_QUEUE) 2867 return -EINVAL; 2868 2869 #ifdef CONFIG_XPS 2870 netif_reset_xps_queues_gt(dev, 0); 2871 #endif 2872 netdev_unbind_all_sb_channels(dev); 2873 2874 dev->num_tc = num_tc; 2875 return 0; 2876 } 2877 EXPORT_SYMBOL(netdev_set_num_tc); 2878 2879 void netdev_unbind_sb_channel(struct net_device *dev, 2880 struct net_device *sb_dev) 2881 { 2882 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2883 2884 #ifdef CONFIG_XPS 2885 netif_reset_xps_queues_gt(sb_dev, 0); 2886 #endif 2887 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2888 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2889 2890 while (txq-- != &dev->_tx[0]) { 2891 if (txq->sb_dev == sb_dev) 2892 txq->sb_dev = NULL; 2893 } 2894 } 2895 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2896 2897 int netdev_bind_sb_channel_queue(struct net_device *dev, 2898 struct net_device *sb_dev, 2899 u8 tc, u16 count, u16 offset) 2900 { 2901 /* Make certain the sb_dev and dev are already configured */ 2902 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2903 return -EINVAL; 2904 2905 /* We cannot hand out queues we don't have */ 2906 if ((offset + count) > dev->real_num_tx_queues) 2907 return -EINVAL; 2908 2909 /* Record the mapping */ 2910 sb_dev->tc_to_txq[tc].count = count; 2911 sb_dev->tc_to_txq[tc].offset = offset; 2912 2913 /* Provide a way for Tx queue to find the tc_to_txq map or 2914 * XPS map for itself. 2915 */ 2916 while (count--) 2917 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2918 2919 return 0; 2920 } 2921 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2922 2923 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2924 { 2925 /* Do not use a multiqueue device to represent a subordinate channel */ 2926 if (netif_is_multiqueue(dev)) 2927 return -ENODEV; 2928 2929 /* We allow channels 1 - 32767 to be used for subordinate channels. 2930 * Channel 0 is meant to be "native" mode and used only to represent 2931 * the main root device. We allow writing 0 to reset the device back 2932 * to normal mode after being used as a subordinate channel. 2933 */ 2934 if (channel > S16_MAX) 2935 return -EINVAL; 2936 2937 dev->num_tc = -channel; 2938 2939 return 0; 2940 } 2941 EXPORT_SYMBOL(netdev_set_sb_channel); 2942 2943 /* 2944 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2945 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2946 */ 2947 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2948 { 2949 bool disabling; 2950 int rc; 2951 2952 disabling = txq < dev->real_num_tx_queues; 2953 2954 if (txq < 1 || txq > dev->num_tx_queues) 2955 return -EINVAL; 2956 2957 if (dev->reg_state == NETREG_REGISTERED || 2958 dev->reg_state == NETREG_UNREGISTERING) { 2959 ASSERT_RTNL(); 2960 2961 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2962 txq); 2963 if (rc) 2964 return rc; 2965 2966 if (dev->num_tc) 2967 netif_setup_tc(dev, txq); 2968 2969 net_shaper_set_real_num_tx_queues(dev, txq); 2970 2971 dev_qdisc_change_real_num_tx(dev, txq); 2972 2973 dev->real_num_tx_queues = txq; 2974 2975 if (disabling) { 2976 synchronize_net(); 2977 qdisc_reset_all_tx_gt(dev, txq); 2978 #ifdef CONFIG_XPS 2979 netif_reset_xps_queues_gt(dev, txq); 2980 #endif 2981 } 2982 } else { 2983 dev->real_num_tx_queues = txq; 2984 } 2985 2986 return 0; 2987 } 2988 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2989 2990 #ifdef CONFIG_SYSFS 2991 /** 2992 * netif_set_real_num_rx_queues - set actual number of RX queues used 2993 * @dev: Network device 2994 * @rxq: Actual number of RX queues 2995 * 2996 * This must be called either with the rtnl_lock held or before 2997 * registration of the net device. Returns 0 on success, or a 2998 * negative error code. If called before registration, it always 2999 * succeeds. 3000 */ 3001 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3002 { 3003 int rc; 3004 3005 if (rxq < 1 || rxq > dev->num_rx_queues) 3006 return -EINVAL; 3007 3008 if (dev->reg_state == NETREG_REGISTERED) { 3009 ASSERT_RTNL(); 3010 3011 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3012 rxq); 3013 if (rc) 3014 return rc; 3015 } 3016 3017 dev->real_num_rx_queues = rxq; 3018 return 0; 3019 } 3020 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3021 #endif 3022 3023 /** 3024 * netif_set_real_num_queues - set actual number of RX and TX queues used 3025 * @dev: Network device 3026 * @txq: Actual number of TX queues 3027 * @rxq: Actual number of RX queues 3028 * 3029 * Set the real number of both TX and RX queues. 3030 * Does nothing if the number of queues is already correct. 3031 */ 3032 int netif_set_real_num_queues(struct net_device *dev, 3033 unsigned int txq, unsigned int rxq) 3034 { 3035 unsigned int old_rxq = dev->real_num_rx_queues; 3036 int err; 3037 3038 if (txq < 1 || txq > dev->num_tx_queues || 3039 rxq < 1 || rxq > dev->num_rx_queues) 3040 return -EINVAL; 3041 3042 /* Start from increases, so the error path only does decreases - 3043 * decreases can't fail. 3044 */ 3045 if (rxq > dev->real_num_rx_queues) { 3046 err = netif_set_real_num_rx_queues(dev, rxq); 3047 if (err) 3048 return err; 3049 } 3050 if (txq > dev->real_num_tx_queues) { 3051 err = netif_set_real_num_tx_queues(dev, txq); 3052 if (err) 3053 goto undo_rx; 3054 } 3055 if (rxq < dev->real_num_rx_queues) 3056 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3057 if (txq < dev->real_num_tx_queues) 3058 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3059 3060 return 0; 3061 undo_rx: 3062 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3063 return err; 3064 } 3065 EXPORT_SYMBOL(netif_set_real_num_queues); 3066 3067 /** 3068 * netif_set_tso_max_size() - set the max size of TSO frames supported 3069 * @dev: netdev to update 3070 * @size: max skb->len of a TSO frame 3071 * 3072 * Set the limit on the size of TSO super-frames the device can handle. 3073 * Unless explicitly set the stack will assume the value of 3074 * %GSO_LEGACY_MAX_SIZE. 3075 */ 3076 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3077 { 3078 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3079 if (size < READ_ONCE(dev->gso_max_size)) 3080 netif_set_gso_max_size(dev, size); 3081 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3082 netif_set_gso_ipv4_max_size(dev, size); 3083 } 3084 EXPORT_SYMBOL(netif_set_tso_max_size); 3085 3086 /** 3087 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3088 * @dev: netdev to update 3089 * @segs: max number of TCP segments 3090 * 3091 * Set the limit on the number of TCP segments the device can generate from 3092 * a single TSO super-frame. 3093 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3094 */ 3095 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3096 { 3097 dev->tso_max_segs = segs; 3098 if (segs < READ_ONCE(dev->gso_max_segs)) 3099 netif_set_gso_max_segs(dev, segs); 3100 } 3101 EXPORT_SYMBOL(netif_set_tso_max_segs); 3102 3103 /** 3104 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3105 * @to: netdev to update 3106 * @from: netdev from which to copy the limits 3107 */ 3108 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3109 { 3110 netif_set_tso_max_size(to, from->tso_max_size); 3111 netif_set_tso_max_segs(to, from->tso_max_segs); 3112 } 3113 EXPORT_SYMBOL(netif_inherit_tso_max); 3114 3115 /** 3116 * netif_get_num_default_rss_queues - default number of RSS queues 3117 * 3118 * Default value is the number of physical cores if there are only 1 or 2, or 3119 * divided by 2 if there are more. 3120 */ 3121 int netif_get_num_default_rss_queues(void) 3122 { 3123 cpumask_var_t cpus; 3124 int cpu, count = 0; 3125 3126 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3127 return 1; 3128 3129 cpumask_copy(cpus, cpu_online_mask); 3130 for_each_cpu(cpu, cpus) { 3131 ++count; 3132 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3133 } 3134 free_cpumask_var(cpus); 3135 3136 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3137 } 3138 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3139 3140 static void __netif_reschedule(struct Qdisc *q) 3141 { 3142 struct softnet_data *sd; 3143 unsigned long flags; 3144 3145 local_irq_save(flags); 3146 sd = this_cpu_ptr(&softnet_data); 3147 q->next_sched = NULL; 3148 *sd->output_queue_tailp = q; 3149 sd->output_queue_tailp = &q->next_sched; 3150 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3151 local_irq_restore(flags); 3152 } 3153 3154 void __netif_schedule(struct Qdisc *q) 3155 { 3156 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3157 __netif_reschedule(q); 3158 } 3159 EXPORT_SYMBOL(__netif_schedule); 3160 3161 struct dev_kfree_skb_cb { 3162 enum skb_drop_reason reason; 3163 }; 3164 3165 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3166 { 3167 return (struct dev_kfree_skb_cb *)skb->cb; 3168 } 3169 3170 void netif_schedule_queue(struct netdev_queue *txq) 3171 { 3172 rcu_read_lock(); 3173 if (!netif_xmit_stopped(txq)) { 3174 struct Qdisc *q = rcu_dereference(txq->qdisc); 3175 3176 __netif_schedule(q); 3177 } 3178 rcu_read_unlock(); 3179 } 3180 EXPORT_SYMBOL(netif_schedule_queue); 3181 3182 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3183 { 3184 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3185 struct Qdisc *q; 3186 3187 rcu_read_lock(); 3188 q = rcu_dereference(dev_queue->qdisc); 3189 __netif_schedule(q); 3190 rcu_read_unlock(); 3191 } 3192 } 3193 EXPORT_SYMBOL(netif_tx_wake_queue); 3194 3195 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3196 { 3197 unsigned long flags; 3198 3199 if (unlikely(!skb)) 3200 return; 3201 3202 if (likely(refcount_read(&skb->users) == 1)) { 3203 smp_rmb(); 3204 refcount_set(&skb->users, 0); 3205 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3206 return; 3207 } 3208 get_kfree_skb_cb(skb)->reason = reason; 3209 local_irq_save(flags); 3210 skb->next = __this_cpu_read(softnet_data.completion_queue); 3211 __this_cpu_write(softnet_data.completion_queue, skb); 3212 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3213 local_irq_restore(flags); 3214 } 3215 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3216 3217 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3218 { 3219 if (in_hardirq() || irqs_disabled()) 3220 dev_kfree_skb_irq_reason(skb, reason); 3221 else 3222 kfree_skb_reason(skb, reason); 3223 } 3224 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3225 3226 3227 /** 3228 * netif_device_detach - mark device as removed 3229 * @dev: network device 3230 * 3231 * Mark device as removed from system and therefore no longer available. 3232 */ 3233 void netif_device_detach(struct net_device *dev) 3234 { 3235 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3236 netif_running(dev)) { 3237 netif_tx_stop_all_queues(dev); 3238 } 3239 } 3240 EXPORT_SYMBOL(netif_device_detach); 3241 3242 /** 3243 * netif_device_attach - mark device as attached 3244 * @dev: network device 3245 * 3246 * Mark device as attached from system and restart if needed. 3247 */ 3248 void netif_device_attach(struct net_device *dev) 3249 { 3250 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3251 netif_running(dev)) { 3252 netif_tx_wake_all_queues(dev); 3253 netdev_watchdog_up(dev); 3254 } 3255 } 3256 EXPORT_SYMBOL(netif_device_attach); 3257 3258 /* 3259 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3260 * to be used as a distribution range. 3261 */ 3262 static u16 skb_tx_hash(const struct net_device *dev, 3263 const struct net_device *sb_dev, 3264 struct sk_buff *skb) 3265 { 3266 u32 hash; 3267 u16 qoffset = 0; 3268 u16 qcount = dev->real_num_tx_queues; 3269 3270 if (dev->num_tc) { 3271 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3272 3273 qoffset = sb_dev->tc_to_txq[tc].offset; 3274 qcount = sb_dev->tc_to_txq[tc].count; 3275 if (unlikely(!qcount)) { 3276 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3277 sb_dev->name, qoffset, tc); 3278 qoffset = 0; 3279 qcount = dev->real_num_tx_queues; 3280 } 3281 } 3282 3283 if (skb_rx_queue_recorded(skb)) { 3284 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3285 hash = skb_get_rx_queue(skb); 3286 if (hash >= qoffset) 3287 hash -= qoffset; 3288 while (unlikely(hash >= qcount)) 3289 hash -= qcount; 3290 return hash + qoffset; 3291 } 3292 3293 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3294 } 3295 3296 void skb_warn_bad_offload(const struct sk_buff *skb) 3297 { 3298 static const netdev_features_t null_features; 3299 struct net_device *dev = skb->dev; 3300 const char *name = ""; 3301 3302 if (!net_ratelimit()) 3303 return; 3304 3305 if (dev) { 3306 if (dev->dev.parent) 3307 name = dev_driver_string(dev->dev.parent); 3308 else 3309 name = netdev_name(dev); 3310 } 3311 skb_dump(KERN_WARNING, skb, false); 3312 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3313 name, dev ? &dev->features : &null_features, 3314 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3315 } 3316 3317 /* 3318 * Invalidate hardware checksum when packet is to be mangled, and 3319 * complete checksum manually on outgoing path. 3320 */ 3321 int skb_checksum_help(struct sk_buff *skb) 3322 { 3323 __wsum csum; 3324 int ret = 0, offset; 3325 3326 if (skb->ip_summed == CHECKSUM_COMPLETE) 3327 goto out_set_summed; 3328 3329 if (unlikely(skb_is_gso(skb))) { 3330 skb_warn_bad_offload(skb); 3331 return -EINVAL; 3332 } 3333 3334 if (!skb_frags_readable(skb)) { 3335 return -EFAULT; 3336 } 3337 3338 /* Before computing a checksum, we should make sure no frag could 3339 * be modified by an external entity : checksum could be wrong. 3340 */ 3341 if (skb_has_shared_frag(skb)) { 3342 ret = __skb_linearize(skb); 3343 if (ret) 3344 goto out; 3345 } 3346 3347 offset = skb_checksum_start_offset(skb); 3348 ret = -EINVAL; 3349 if (unlikely(offset >= skb_headlen(skb))) { 3350 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3351 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3352 offset, skb_headlen(skb)); 3353 goto out; 3354 } 3355 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3356 3357 offset += skb->csum_offset; 3358 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3359 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3360 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3361 offset + sizeof(__sum16), skb_headlen(skb)); 3362 goto out; 3363 } 3364 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3365 if (ret) 3366 goto out; 3367 3368 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3369 out_set_summed: 3370 skb->ip_summed = CHECKSUM_NONE; 3371 out: 3372 return ret; 3373 } 3374 EXPORT_SYMBOL(skb_checksum_help); 3375 3376 int skb_crc32c_csum_help(struct sk_buff *skb) 3377 { 3378 __le32 crc32c_csum; 3379 int ret = 0, offset, start; 3380 3381 if (skb->ip_summed != CHECKSUM_PARTIAL) 3382 goto out; 3383 3384 if (unlikely(skb_is_gso(skb))) 3385 goto out; 3386 3387 /* Before computing a checksum, we should make sure no frag could 3388 * be modified by an external entity : checksum could be wrong. 3389 */ 3390 if (unlikely(skb_has_shared_frag(skb))) { 3391 ret = __skb_linearize(skb); 3392 if (ret) 3393 goto out; 3394 } 3395 start = skb_checksum_start_offset(skb); 3396 offset = start + offsetof(struct sctphdr, checksum); 3397 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3398 ret = -EINVAL; 3399 goto out; 3400 } 3401 3402 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3403 if (ret) 3404 goto out; 3405 3406 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3407 skb->len - start, ~(__u32)0, 3408 crc32c_csum_stub)); 3409 *(__le32 *)(skb->data + offset) = crc32c_csum; 3410 skb_reset_csum_not_inet(skb); 3411 out: 3412 return ret; 3413 } 3414 EXPORT_SYMBOL(skb_crc32c_csum_help); 3415 3416 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3417 { 3418 __be16 type = skb->protocol; 3419 3420 /* Tunnel gso handlers can set protocol to ethernet. */ 3421 if (type == htons(ETH_P_TEB)) { 3422 struct ethhdr *eth; 3423 3424 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3425 return 0; 3426 3427 eth = (struct ethhdr *)skb->data; 3428 type = eth->h_proto; 3429 } 3430 3431 return vlan_get_protocol_and_depth(skb, type, depth); 3432 } 3433 3434 3435 /* Take action when hardware reception checksum errors are detected. */ 3436 #ifdef CONFIG_BUG 3437 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3438 { 3439 netdev_err(dev, "hw csum failure\n"); 3440 skb_dump(KERN_ERR, skb, true); 3441 dump_stack(); 3442 } 3443 3444 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3445 { 3446 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3447 } 3448 EXPORT_SYMBOL(netdev_rx_csum_fault); 3449 #endif 3450 3451 /* XXX: check that highmem exists at all on the given machine. */ 3452 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3453 { 3454 #ifdef CONFIG_HIGHMEM 3455 int i; 3456 3457 if (!(dev->features & NETIF_F_HIGHDMA)) { 3458 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3459 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3460 struct page *page = skb_frag_page(frag); 3461 3462 if (page && PageHighMem(page)) 3463 return 1; 3464 } 3465 } 3466 #endif 3467 return 0; 3468 } 3469 3470 /* If MPLS offload request, verify we are testing hardware MPLS features 3471 * instead of standard features for the netdev. 3472 */ 3473 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3474 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3475 netdev_features_t features, 3476 __be16 type) 3477 { 3478 if (eth_p_mpls(type)) 3479 features &= skb->dev->mpls_features; 3480 3481 return features; 3482 } 3483 #else 3484 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3485 netdev_features_t features, 3486 __be16 type) 3487 { 3488 return features; 3489 } 3490 #endif 3491 3492 static netdev_features_t harmonize_features(struct sk_buff *skb, 3493 netdev_features_t features) 3494 { 3495 __be16 type; 3496 3497 type = skb_network_protocol(skb, NULL); 3498 features = net_mpls_features(skb, features, type); 3499 3500 if (skb->ip_summed != CHECKSUM_NONE && 3501 !can_checksum_protocol(features, type)) { 3502 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3503 } 3504 if (illegal_highdma(skb->dev, skb)) 3505 features &= ~NETIF_F_SG; 3506 3507 return features; 3508 } 3509 3510 netdev_features_t passthru_features_check(struct sk_buff *skb, 3511 struct net_device *dev, 3512 netdev_features_t features) 3513 { 3514 return features; 3515 } 3516 EXPORT_SYMBOL(passthru_features_check); 3517 3518 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3519 struct net_device *dev, 3520 netdev_features_t features) 3521 { 3522 return vlan_features_check(skb, features); 3523 } 3524 3525 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3526 struct net_device *dev, 3527 netdev_features_t features) 3528 { 3529 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3530 3531 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3532 return features & ~NETIF_F_GSO_MASK; 3533 3534 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb))) 3535 return features & ~NETIF_F_GSO_MASK; 3536 3537 if (!skb_shinfo(skb)->gso_type) { 3538 skb_warn_bad_offload(skb); 3539 return features & ~NETIF_F_GSO_MASK; 3540 } 3541 3542 /* Support for GSO partial features requires software 3543 * intervention before we can actually process the packets 3544 * so we need to strip support for any partial features now 3545 * and we can pull them back in after we have partially 3546 * segmented the frame. 3547 */ 3548 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3549 features &= ~dev->gso_partial_features; 3550 3551 /* Make sure to clear the IPv4 ID mangling feature if the 3552 * IPv4 header has the potential to be fragmented. 3553 */ 3554 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3555 struct iphdr *iph = skb->encapsulation ? 3556 inner_ip_hdr(skb) : ip_hdr(skb); 3557 3558 if (!(iph->frag_off & htons(IP_DF))) 3559 features &= ~NETIF_F_TSO_MANGLEID; 3560 } 3561 3562 return features; 3563 } 3564 3565 netdev_features_t netif_skb_features(struct sk_buff *skb) 3566 { 3567 struct net_device *dev = skb->dev; 3568 netdev_features_t features = dev->features; 3569 3570 if (skb_is_gso(skb)) 3571 features = gso_features_check(skb, dev, features); 3572 3573 /* If encapsulation offload request, verify we are testing 3574 * hardware encapsulation features instead of standard 3575 * features for the netdev 3576 */ 3577 if (skb->encapsulation) 3578 features &= dev->hw_enc_features; 3579 3580 if (skb_vlan_tagged(skb)) 3581 features = netdev_intersect_features(features, 3582 dev->vlan_features | 3583 NETIF_F_HW_VLAN_CTAG_TX | 3584 NETIF_F_HW_VLAN_STAG_TX); 3585 3586 if (dev->netdev_ops->ndo_features_check) 3587 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3588 features); 3589 else 3590 features &= dflt_features_check(skb, dev, features); 3591 3592 return harmonize_features(skb, features); 3593 } 3594 EXPORT_SYMBOL(netif_skb_features); 3595 3596 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3597 struct netdev_queue *txq, bool more) 3598 { 3599 unsigned int len; 3600 int rc; 3601 3602 if (dev_nit_active(dev)) 3603 dev_queue_xmit_nit(skb, dev); 3604 3605 len = skb->len; 3606 trace_net_dev_start_xmit(skb, dev); 3607 rc = netdev_start_xmit(skb, dev, txq, more); 3608 trace_net_dev_xmit(skb, rc, dev, len); 3609 3610 return rc; 3611 } 3612 3613 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3614 struct netdev_queue *txq, int *ret) 3615 { 3616 struct sk_buff *skb = first; 3617 int rc = NETDEV_TX_OK; 3618 3619 while (skb) { 3620 struct sk_buff *next = skb->next; 3621 3622 skb_mark_not_on_list(skb); 3623 rc = xmit_one(skb, dev, txq, next != NULL); 3624 if (unlikely(!dev_xmit_complete(rc))) { 3625 skb->next = next; 3626 goto out; 3627 } 3628 3629 skb = next; 3630 if (netif_tx_queue_stopped(txq) && skb) { 3631 rc = NETDEV_TX_BUSY; 3632 break; 3633 } 3634 } 3635 3636 out: 3637 *ret = rc; 3638 return skb; 3639 } 3640 3641 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3642 netdev_features_t features) 3643 { 3644 if (skb_vlan_tag_present(skb) && 3645 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3646 skb = __vlan_hwaccel_push_inside(skb); 3647 return skb; 3648 } 3649 3650 int skb_csum_hwoffload_help(struct sk_buff *skb, 3651 const netdev_features_t features) 3652 { 3653 if (unlikely(skb_csum_is_sctp(skb))) 3654 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3655 skb_crc32c_csum_help(skb); 3656 3657 if (features & NETIF_F_HW_CSUM) 3658 return 0; 3659 3660 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3661 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) && 3662 skb_network_header_len(skb) != sizeof(struct ipv6hdr) && 3663 !ipv6_has_hopopt_jumbo(skb)) 3664 goto sw_checksum; 3665 3666 switch (skb->csum_offset) { 3667 case offsetof(struct tcphdr, check): 3668 case offsetof(struct udphdr, check): 3669 return 0; 3670 } 3671 } 3672 3673 sw_checksum: 3674 return skb_checksum_help(skb); 3675 } 3676 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3677 3678 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3679 { 3680 netdev_features_t features; 3681 3682 features = netif_skb_features(skb); 3683 skb = validate_xmit_vlan(skb, features); 3684 if (unlikely(!skb)) 3685 goto out_null; 3686 3687 skb = sk_validate_xmit_skb(skb, dev); 3688 if (unlikely(!skb)) 3689 goto out_null; 3690 3691 if (netif_needs_gso(skb, features)) { 3692 struct sk_buff *segs; 3693 3694 segs = skb_gso_segment(skb, features); 3695 if (IS_ERR(segs)) { 3696 goto out_kfree_skb; 3697 } else if (segs) { 3698 consume_skb(skb); 3699 skb = segs; 3700 } 3701 } else { 3702 if (skb_needs_linearize(skb, features) && 3703 __skb_linearize(skb)) 3704 goto out_kfree_skb; 3705 3706 /* If packet is not checksummed and device does not 3707 * support checksumming for this protocol, complete 3708 * checksumming here. 3709 */ 3710 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3711 if (skb->encapsulation) 3712 skb_set_inner_transport_header(skb, 3713 skb_checksum_start_offset(skb)); 3714 else 3715 skb_set_transport_header(skb, 3716 skb_checksum_start_offset(skb)); 3717 if (skb_csum_hwoffload_help(skb, features)) 3718 goto out_kfree_skb; 3719 } 3720 } 3721 3722 skb = validate_xmit_xfrm(skb, features, again); 3723 3724 return skb; 3725 3726 out_kfree_skb: 3727 kfree_skb(skb); 3728 out_null: 3729 dev_core_stats_tx_dropped_inc(dev); 3730 return NULL; 3731 } 3732 3733 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3734 { 3735 struct sk_buff *next, *head = NULL, *tail; 3736 3737 for (; skb != NULL; skb = next) { 3738 next = skb->next; 3739 skb_mark_not_on_list(skb); 3740 3741 /* in case skb won't be segmented, point to itself */ 3742 skb->prev = skb; 3743 3744 skb = validate_xmit_skb(skb, dev, again); 3745 if (!skb) 3746 continue; 3747 3748 if (!head) 3749 head = skb; 3750 else 3751 tail->next = skb; 3752 /* If skb was segmented, skb->prev points to 3753 * the last segment. If not, it still contains skb. 3754 */ 3755 tail = skb->prev; 3756 } 3757 return head; 3758 } 3759 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3760 3761 static void qdisc_pkt_len_init(struct sk_buff *skb) 3762 { 3763 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3764 3765 qdisc_skb_cb(skb)->pkt_len = skb->len; 3766 3767 /* To get more precise estimation of bytes sent on wire, 3768 * we add to pkt_len the headers size of all segments 3769 */ 3770 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3771 u16 gso_segs = shinfo->gso_segs; 3772 unsigned int hdr_len; 3773 3774 /* mac layer + network layer */ 3775 hdr_len = skb_transport_offset(skb); 3776 3777 /* + transport layer */ 3778 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3779 const struct tcphdr *th; 3780 struct tcphdr _tcphdr; 3781 3782 th = skb_header_pointer(skb, hdr_len, 3783 sizeof(_tcphdr), &_tcphdr); 3784 if (likely(th)) 3785 hdr_len += __tcp_hdrlen(th); 3786 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 3787 struct udphdr _udphdr; 3788 3789 if (skb_header_pointer(skb, hdr_len, 3790 sizeof(_udphdr), &_udphdr)) 3791 hdr_len += sizeof(struct udphdr); 3792 } 3793 3794 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) { 3795 int payload = skb->len - hdr_len; 3796 3797 /* Malicious packet. */ 3798 if (payload <= 0) 3799 return; 3800 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size); 3801 } 3802 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3803 } 3804 } 3805 3806 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3807 struct sk_buff **to_free, 3808 struct netdev_queue *txq) 3809 { 3810 int rc; 3811 3812 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3813 if (rc == NET_XMIT_SUCCESS) 3814 trace_qdisc_enqueue(q, txq, skb); 3815 return rc; 3816 } 3817 3818 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3819 struct net_device *dev, 3820 struct netdev_queue *txq) 3821 { 3822 spinlock_t *root_lock = qdisc_lock(q); 3823 struct sk_buff *to_free = NULL; 3824 bool contended; 3825 int rc; 3826 3827 qdisc_calculate_pkt_len(skb, q); 3828 3829 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3830 3831 if (q->flags & TCQ_F_NOLOCK) { 3832 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3833 qdisc_run_begin(q)) { 3834 /* Retest nolock_qdisc_is_empty() within the protection 3835 * of q->seqlock to protect from racing with requeuing. 3836 */ 3837 if (unlikely(!nolock_qdisc_is_empty(q))) { 3838 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3839 __qdisc_run(q); 3840 qdisc_run_end(q); 3841 3842 goto no_lock_out; 3843 } 3844 3845 qdisc_bstats_cpu_update(q, skb); 3846 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3847 !nolock_qdisc_is_empty(q)) 3848 __qdisc_run(q); 3849 3850 qdisc_run_end(q); 3851 return NET_XMIT_SUCCESS; 3852 } 3853 3854 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3855 qdisc_run(q); 3856 3857 no_lock_out: 3858 if (unlikely(to_free)) 3859 kfree_skb_list_reason(to_free, 3860 tcf_get_drop_reason(to_free)); 3861 return rc; 3862 } 3863 3864 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) { 3865 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP); 3866 return NET_XMIT_DROP; 3867 } 3868 /* 3869 * Heuristic to force contended enqueues to serialize on a 3870 * separate lock before trying to get qdisc main lock. 3871 * This permits qdisc->running owner to get the lock more 3872 * often and dequeue packets faster. 3873 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3874 * and then other tasks will only enqueue packets. The packets will be 3875 * sent after the qdisc owner is scheduled again. To prevent this 3876 * scenario the task always serialize on the lock. 3877 */ 3878 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3879 if (unlikely(contended)) 3880 spin_lock(&q->busylock); 3881 3882 spin_lock(root_lock); 3883 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3884 __qdisc_drop(skb, &to_free); 3885 rc = NET_XMIT_DROP; 3886 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3887 qdisc_run_begin(q)) { 3888 /* 3889 * This is a work-conserving queue; there are no old skbs 3890 * waiting to be sent out; and the qdisc is not running - 3891 * xmit the skb directly. 3892 */ 3893 3894 qdisc_bstats_update(q, skb); 3895 3896 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3897 if (unlikely(contended)) { 3898 spin_unlock(&q->busylock); 3899 contended = false; 3900 } 3901 __qdisc_run(q); 3902 } 3903 3904 qdisc_run_end(q); 3905 rc = NET_XMIT_SUCCESS; 3906 } else { 3907 WRITE_ONCE(q->owner, smp_processor_id()); 3908 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3909 WRITE_ONCE(q->owner, -1); 3910 if (qdisc_run_begin(q)) { 3911 if (unlikely(contended)) { 3912 spin_unlock(&q->busylock); 3913 contended = false; 3914 } 3915 __qdisc_run(q); 3916 qdisc_run_end(q); 3917 } 3918 } 3919 spin_unlock(root_lock); 3920 if (unlikely(to_free)) 3921 kfree_skb_list_reason(to_free, 3922 tcf_get_drop_reason(to_free)); 3923 if (unlikely(contended)) 3924 spin_unlock(&q->busylock); 3925 return rc; 3926 } 3927 3928 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3929 static void skb_update_prio(struct sk_buff *skb) 3930 { 3931 const struct netprio_map *map; 3932 const struct sock *sk; 3933 unsigned int prioidx; 3934 3935 if (skb->priority) 3936 return; 3937 map = rcu_dereference_bh(skb->dev->priomap); 3938 if (!map) 3939 return; 3940 sk = skb_to_full_sk(skb); 3941 if (!sk) 3942 return; 3943 3944 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3945 3946 if (prioidx < map->priomap_len) 3947 skb->priority = map->priomap[prioidx]; 3948 } 3949 #else 3950 #define skb_update_prio(skb) 3951 #endif 3952 3953 /** 3954 * dev_loopback_xmit - loop back @skb 3955 * @net: network namespace this loopback is happening in 3956 * @sk: sk needed to be a netfilter okfn 3957 * @skb: buffer to transmit 3958 */ 3959 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3960 { 3961 skb_reset_mac_header(skb); 3962 __skb_pull(skb, skb_network_offset(skb)); 3963 skb->pkt_type = PACKET_LOOPBACK; 3964 if (skb->ip_summed == CHECKSUM_NONE) 3965 skb->ip_summed = CHECKSUM_UNNECESSARY; 3966 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3967 skb_dst_force(skb); 3968 netif_rx(skb); 3969 return 0; 3970 } 3971 EXPORT_SYMBOL(dev_loopback_xmit); 3972 3973 #ifdef CONFIG_NET_EGRESS 3974 static struct netdev_queue * 3975 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3976 { 3977 int qm = skb_get_queue_mapping(skb); 3978 3979 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3980 } 3981 3982 #ifndef CONFIG_PREEMPT_RT 3983 static bool netdev_xmit_txqueue_skipped(void) 3984 { 3985 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3986 } 3987 3988 void netdev_xmit_skip_txqueue(bool skip) 3989 { 3990 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3991 } 3992 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3993 3994 #else 3995 static bool netdev_xmit_txqueue_skipped(void) 3996 { 3997 return current->net_xmit.skip_txqueue; 3998 } 3999 4000 void netdev_xmit_skip_txqueue(bool skip) 4001 { 4002 current->net_xmit.skip_txqueue = skip; 4003 } 4004 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4005 #endif 4006 #endif /* CONFIG_NET_EGRESS */ 4007 4008 #ifdef CONFIG_NET_XGRESS 4009 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 4010 enum skb_drop_reason *drop_reason) 4011 { 4012 int ret = TC_ACT_UNSPEC; 4013 #ifdef CONFIG_NET_CLS_ACT 4014 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 4015 struct tcf_result res; 4016 4017 if (!miniq) 4018 return ret; 4019 4020 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) { 4021 if (tcf_block_bypass_sw(miniq->block)) 4022 return ret; 4023 } 4024 4025 tc_skb_cb(skb)->mru = 0; 4026 tc_skb_cb(skb)->post_ct = false; 4027 tcf_set_drop_reason(skb, *drop_reason); 4028 4029 mini_qdisc_bstats_cpu_update(miniq, skb); 4030 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 4031 /* Only tcf related quirks below. */ 4032 switch (ret) { 4033 case TC_ACT_SHOT: 4034 *drop_reason = tcf_get_drop_reason(skb); 4035 mini_qdisc_qstats_cpu_drop(miniq); 4036 break; 4037 case TC_ACT_OK: 4038 case TC_ACT_RECLASSIFY: 4039 skb->tc_index = TC_H_MIN(res.classid); 4040 break; 4041 } 4042 #endif /* CONFIG_NET_CLS_ACT */ 4043 return ret; 4044 } 4045 4046 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 4047 4048 void tcx_inc(void) 4049 { 4050 static_branch_inc(&tcx_needed_key); 4051 } 4052 4053 void tcx_dec(void) 4054 { 4055 static_branch_dec(&tcx_needed_key); 4056 } 4057 4058 static __always_inline enum tcx_action_base 4059 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 4060 const bool needs_mac) 4061 { 4062 const struct bpf_mprog_fp *fp; 4063 const struct bpf_prog *prog; 4064 int ret = TCX_NEXT; 4065 4066 if (needs_mac) 4067 __skb_push(skb, skb->mac_len); 4068 bpf_mprog_foreach_prog(entry, fp, prog) { 4069 bpf_compute_data_pointers(skb); 4070 ret = bpf_prog_run(prog, skb); 4071 if (ret != TCX_NEXT) 4072 break; 4073 } 4074 if (needs_mac) 4075 __skb_pull(skb, skb->mac_len); 4076 return tcx_action_code(skb, ret); 4077 } 4078 4079 static __always_inline struct sk_buff * 4080 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4081 struct net_device *orig_dev, bool *another) 4082 { 4083 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4084 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 4085 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4086 int sch_ret; 4087 4088 if (!entry) 4089 return skb; 4090 4091 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4092 if (*pt_prev) { 4093 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4094 *pt_prev = NULL; 4095 } 4096 4097 qdisc_skb_cb(skb)->pkt_len = skb->len; 4098 tcx_set_ingress(skb, true); 4099 4100 if (static_branch_unlikely(&tcx_needed_key)) { 4101 sch_ret = tcx_run(entry, skb, true); 4102 if (sch_ret != TC_ACT_UNSPEC) 4103 goto ingress_verdict; 4104 } 4105 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4106 ingress_verdict: 4107 switch (sch_ret) { 4108 case TC_ACT_REDIRECT: 4109 /* skb_mac_header check was done by BPF, so we can safely 4110 * push the L2 header back before redirecting to another 4111 * netdev. 4112 */ 4113 __skb_push(skb, skb->mac_len); 4114 if (skb_do_redirect(skb) == -EAGAIN) { 4115 __skb_pull(skb, skb->mac_len); 4116 *another = true; 4117 break; 4118 } 4119 *ret = NET_RX_SUCCESS; 4120 bpf_net_ctx_clear(bpf_net_ctx); 4121 return NULL; 4122 case TC_ACT_SHOT: 4123 kfree_skb_reason(skb, drop_reason); 4124 *ret = NET_RX_DROP; 4125 bpf_net_ctx_clear(bpf_net_ctx); 4126 return NULL; 4127 /* used by tc_run */ 4128 case TC_ACT_STOLEN: 4129 case TC_ACT_QUEUED: 4130 case TC_ACT_TRAP: 4131 consume_skb(skb); 4132 fallthrough; 4133 case TC_ACT_CONSUMED: 4134 *ret = NET_RX_SUCCESS; 4135 bpf_net_ctx_clear(bpf_net_ctx); 4136 return NULL; 4137 } 4138 bpf_net_ctx_clear(bpf_net_ctx); 4139 4140 return skb; 4141 } 4142 4143 static __always_inline struct sk_buff * 4144 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4145 { 4146 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4147 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4148 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4149 int sch_ret; 4150 4151 if (!entry) 4152 return skb; 4153 4154 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4155 4156 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4157 * already set by the caller. 4158 */ 4159 if (static_branch_unlikely(&tcx_needed_key)) { 4160 sch_ret = tcx_run(entry, skb, false); 4161 if (sch_ret != TC_ACT_UNSPEC) 4162 goto egress_verdict; 4163 } 4164 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4165 egress_verdict: 4166 switch (sch_ret) { 4167 case TC_ACT_REDIRECT: 4168 /* No need to push/pop skb's mac_header here on egress! */ 4169 skb_do_redirect(skb); 4170 *ret = NET_XMIT_SUCCESS; 4171 bpf_net_ctx_clear(bpf_net_ctx); 4172 return NULL; 4173 case TC_ACT_SHOT: 4174 kfree_skb_reason(skb, drop_reason); 4175 *ret = NET_XMIT_DROP; 4176 bpf_net_ctx_clear(bpf_net_ctx); 4177 return NULL; 4178 /* used by tc_run */ 4179 case TC_ACT_STOLEN: 4180 case TC_ACT_QUEUED: 4181 case TC_ACT_TRAP: 4182 consume_skb(skb); 4183 fallthrough; 4184 case TC_ACT_CONSUMED: 4185 *ret = NET_XMIT_SUCCESS; 4186 bpf_net_ctx_clear(bpf_net_ctx); 4187 return NULL; 4188 } 4189 bpf_net_ctx_clear(bpf_net_ctx); 4190 4191 return skb; 4192 } 4193 #else 4194 static __always_inline struct sk_buff * 4195 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4196 struct net_device *orig_dev, bool *another) 4197 { 4198 return skb; 4199 } 4200 4201 static __always_inline struct sk_buff * 4202 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4203 { 4204 return skb; 4205 } 4206 #endif /* CONFIG_NET_XGRESS */ 4207 4208 #ifdef CONFIG_XPS 4209 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4210 struct xps_dev_maps *dev_maps, unsigned int tci) 4211 { 4212 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4213 struct xps_map *map; 4214 int queue_index = -1; 4215 4216 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4217 return queue_index; 4218 4219 tci *= dev_maps->num_tc; 4220 tci += tc; 4221 4222 map = rcu_dereference(dev_maps->attr_map[tci]); 4223 if (map) { 4224 if (map->len == 1) 4225 queue_index = map->queues[0]; 4226 else 4227 queue_index = map->queues[reciprocal_scale( 4228 skb_get_hash(skb), map->len)]; 4229 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4230 queue_index = -1; 4231 } 4232 return queue_index; 4233 } 4234 #endif 4235 4236 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4237 struct sk_buff *skb) 4238 { 4239 #ifdef CONFIG_XPS 4240 struct xps_dev_maps *dev_maps; 4241 struct sock *sk = skb->sk; 4242 int queue_index = -1; 4243 4244 if (!static_key_false(&xps_needed)) 4245 return -1; 4246 4247 rcu_read_lock(); 4248 if (!static_key_false(&xps_rxqs_needed)) 4249 goto get_cpus_map; 4250 4251 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4252 if (dev_maps) { 4253 int tci = sk_rx_queue_get(sk); 4254 4255 if (tci >= 0) 4256 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4257 tci); 4258 } 4259 4260 get_cpus_map: 4261 if (queue_index < 0) { 4262 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4263 if (dev_maps) { 4264 unsigned int tci = skb->sender_cpu - 1; 4265 4266 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4267 tci); 4268 } 4269 } 4270 rcu_read_unlock(); 4271 4272 return queue_index; 4273 #else 4274 return -1; 4275 #endif 4276 } 4277 4278 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4279 struct net_device *sb_dev) 4280 { 4281 return 0; 4282 } 4283 EXPORT_SYMBOL(dev_pick_tx_zero); 4284 4285 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4286 struct net_device *sb_dev) 4287 { 4288 struct sock *sk = skb->sk; 4289 int queue_index = sk_tx_queue_get(sk); 4290 4291 sb_dev = sb_dev ? : dev; 4292 4293 if (queue_index < 0 || skb->ooo_okay || 4294 queue_index >= dev->real_num_tx_queues) { 4295 int new_index = get_xps_queue(dev, sb_dev, skb); 4296 4297 if (new_index < 0) 4298 new_index = skb_tx_hash(dev, sb_dev, skb); 4299 4300 if (queue_index != new_index && sk && 4301 sk_fullsock(sk) && 4302 rcu_access_pointer(sk->sk_dst_cache)) 4303 sk_tx_queue_set(sk, new_index); 4304 4305 queue_index = new_index; 4306 } 4307 4308 return queue_index; 4309 } 4310 EXPORT_SYMBOL(netdev_pick_tx); 4311 4312 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4313 struct sk_buff *skb, 4314 struct net_device *sb_dev) 4315 { 4316 int queue_index = 0; 4317 4318 #ifdef CONFIG_XPS 4319 u32 sender_cpu = skb->sender_cpu - 1; 4320 4321 if (sender_cpu >= (u32)NR_CPUS) 4322 skb->sender_cpu = raw_smp_processor_id() + 1; 4323 #endif 4324 4325 if (dev->real_num_tx_queues != 1) { 4326 const struct net_device_ops *ops = dev->netdev_ops; 4327 4328 if (ops->ndo_select_queue) 4329 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4330 else 4331 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4332 4333 queue_index = netdev_cap_txqueue(dev, queue_index); 4334 } 4335 4336 skb_set_queue_mapping(skb, queue_index); 4337 return netdev_get_tx_queue(dev, queue_index); 4338 } 4339 4340 /** 4341 * __dev_queue_xmit() - transmit a buffer 4342 * @skb: buffer to transmit 4343 * @sb_dev: suboordinate device used for L2 forwarding offload 4344 * 4345 * Queue a buffer for transmission to a network device. The caller must 4346 * have set the device and priority and built the buffer before calling 4347 * this function. The function can be called from an interrupt. 4348 * 4349 * When calling this method, interrupts MUST be enabled. This is because 4350 * the BH enable code must have IRQs enabled so that it will not deadlock. 4351 * 4352 * Regardless of the return value, the skb is consumed, so it is currently 4353 * difficult to retry a send to this method. (You can bump the ref count 4354 * before sending to hold a reference for retry if you are careful.) 4355 * 4356 * Return: 4357 * * 0 - buffer successfully transmitted 4358 * * positive qdisc return code - NET_XMIT_DROP etc. 4359 * * negative errno - other errors 4360 */ 4361 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4362 { 4363 struct net_device *dev = skb->dev; 4364 struct netdev_queue *txq = NULL; 4365 struct Qdisc *q; 4366 int rc = -ENOMEM; 4367 bool again = false; 4368 4369 skb_reset_mac_header(skb); 4370 skb_assert_len(skb); 4371 4372 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4373 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4374 4375 /* Disable soft irqs for various locks below. Also 4376 * stops preemption for RCU. 4377 */ 4378 rcu_read_lock_bh(); 4379 4380 skb_update_prio(skb); 4381 4382 qdisc_pkt_len_init(skb); 4383 tcx_set_ingress(skb, false); 4384 #ifdef CONFIG_NET_EGRESS 4385 if (static_branch_unlikely(&egress_needed_key)) { 4386 if (nf_hook_egress_active()) { 4387 skb = nf_hook_egress(skb, &rc, dev); 4388 if (!skb) 4389 goto out; 4390 } 4391 4392 netdev_xmit_skip_txqueue(false); 4393 4394 nf_skip_egress(skb, true); 4395 skb = sch_handle_egress(skb, &rc, dev); 4396 if (!skb) 4397 goto out; 4398 nf_skip_egress(skb, false); 4399 4400 if (netdev_xmit_txqueue_skipped()) 4401 txq = netdev_tx_queue_mapping(dev, skb); 4402 } 4403 #endif 4404 /* If device/qdisc don't need skb->dst, release it right now while 4405 * its hot in this cpu cache. 4406 */ 4407 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4408 skb_dst_drop(skb); 4409 else 4410 skb_dst_force(skb); 4411 4412 if (!txq) 4413 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4414 4415 q = rcu_dereference_bh(txq->qdisc); 4416 4417 trace_net_dev_queue(skb); 4418 if (q->enqueue) { 4419 rc = __dev_xmit_skb(skb, q, dev, txq); 4420 goto out; 4421 } 4422 4423 /* The device has no queue. Common case for software devices: 4424 * loopback, all the sorts of tunnels... 4425 4426 * Really, it is unlikely that netif_tx_lock protection is necessary 4427 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4428 * counters.) 4429 * However, it is possible, that they rely on protection 4430 * made by us here. 4431 4432 * Check this and shot the lock. It is not prone from deadlocks. 4433 *Either shot noqueue qdisc, it is even simpler 8) 4434 */ 4435 if (dev->flags & IFF_UP) { 4436 int cpu = smp_processor_id(); /* ok because BHs are off */ 4437 4438 /* Other cpus might concurrently change txq->xmit_lock_owner 4439 * to -1 or to their cpu id, but not to our id. 4440 */ 4441 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4442 if (dev_xmit_recursion()) 4443 goto recursion_alert; 4444 4445 skb = validate_xmit_skb(skb, dev, &again); 4446 if (!skb) 4447 goto out; 4448 4449 HARD_TX_LOCK(dev, txq, cpu); 4450 4451 if (!netif_xmit_stopped(txq)) { 4452 dev_xmit_recursion_inc(); 4453 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4454 dev_xmit_recursion_dec(); 4455 if (dev_xmit_complete(rc)) { 4456 HARD_TX_UNLOCK(dev, txq); 4457 goto out; 4458 } 4459 } 4460 HARD_TX_UNLOCK(dev, txq); 4461 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4462 dev->name); 4463 } else { 4464 /* Recursion is detected! It is possible, 4465 * unfortunately 4466 */ 4467 recursion_alert: 4468 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4469 dev->name); 4470 } 4471 } 4472 4473 rc = -ENETDOWN; 4474 rcu_read_unlock_bh(); 4475 4476 dev_core_stats_tx_dropped_inc(dev); 4477 kfree_skb_list(skb); 4478 return rc; 4479 out: 4480 rcu_read_unlock_bh(); 4481 return rc; 4482 } 4483 EXPORT_SYMBOL(__dev_queue_xmit); 4484 4485 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4486 { 4487 struct net_device *dev = skb->dev; 4488 struct sk_buff *orig_skb = skb; 4489 struct netdev_queue *txq; 4490 int ret = NETDEV_TX_BUSY; 4491 bool again = false; 4492 4493 if (unlikely(!netif_running(dev) || 4494 !netif_carrier_ok(dev))) 4495 goto drop; 4496 4497 skb = validate_xmit_skb_list(skb, dev, &again); 4498 if (skb != orig_skb) 4499 goto drop; 4500 4501 skb_set_queue_mapping(skb, queue_id); 4502 txq = skb_get_tx_queue(dev, skb); 4503 4504 local_bh_disable(); 4505 4506 dev_xmit_recursion_inc(); 4507 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4508 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4509 ret = netdev_start_xmit(skb, dev, txq, false); 4510 HARD_TX_UNLOCK(dev, txq); 4511 dev_xmit_recursion_dec(); 4512 4513 local_bh_enable(); 4514 return ret; 4515 drop: 4516 dev_core_stats_tx_dropped_inc(dev); 4517 kfree_skb_list(skb); 4518 return NET_XMIT_DROP; 4519 } 4520 EXPORT_SYMBOL(__dev_direct_xmit); 4521 4522 /************************************************************************* 4523 * Receiver routines 4524 *************************************************************************/ 4525 static DEFINE_PER_CPU(struct task_struct *, backlog_napi); 4526 4527 int weight_p __read_mostly = 64; /* old backlog weight */ 4528 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4529 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4530 4531 /* Called with irq disabled */ 4532 static inline void ____napi_schedule(struct softnet_data *sd, 4533 struct napi_struct *napi) 4534 { 4535 struct task_struct *thread; 4536 4537 lockdep_assert_irqs_disabled(); 4538 4539 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4540 /* Paired with smp_mb__before_atomic() in 4541 * napi_enable()/dev_set_threaded(). 4542 * Use READ_ONCE() to guarantee a complete 4543 * read on napi->thread. Only call 4544 * wake_up_process() when it's not NULL. 4545 */ 4546 thread = READ_ONCE(napi->thread); 4547 if (thread) { 4548 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi)) 4549 goto use_local_napi; 4550 4551 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4552 wake_up_process(thread); 4553 return; 4554 } 4555 } 4556 4557 use_local_napi: 4558 list_add_tail(&napi->poll_list, &sd->poll_list); 4559 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4560 /* If not called from net_rx_action() 4561 * we have to raise NET_RX_SOFTIRQ. 4562 */ 4563 if (!sd->in_net_rx_action) 4564 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4565 } 4566 4567 #ifdef CONFIG_RPS 4568 4569 struct static_key_false rps_needed __read_mostly; 4570 EXPORT_SYMBOL(rps_needed); 4571 struct static_key_false rfs_needed __read_mostly; 4572 EXPORT_SYMBOL(rfs_needed); 4573 4574 static struct rps_dev_flow * 4575 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4576 struct rps_dev_flow *rflow, u16 next_cpu) 4577 { 4578 if (next_cpu < nr_cpu_ids) { 4579 u32 head; 4580 #ifdef CONFIG_RFS_ACCEL 4581 struct netdev_rx_queue *rxqueue; 4582 struct rps_dev_flow_table *flow_table; 4583 struct rps_dev_flow *old_rflow; 4584 u16 rxq_index; 4585 u32 flow_id; 4586 int rc; 4587 4588 /* Should we steer this flow to a different hardware queue? */ 4589 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4590 !(dev->features & NETIF_F_NTUPLE)) 4591 goto out; 4592 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4593 if (rxq_index == skb_get_rx_queue(skb)) 4594 goto out; 4595 4596 rxqueue = dev->_rx + rxq_index; 4597 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4598 if (!flow_table) 4599 goto out; 4600 flow_id = skb_get_hash(skb) & flow_table->mask; 4601 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4602 rxq_index, flow_id); 4603 if (rc < 0) 4604 goto out; 4605 old_rflow = rflow; 4606 rflow = &flow_table->flows[flow_id]; 4607 WRITE_ONCE(rflow->filter, rc); 4608 if (old_rflow->filter == rc) 4609 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER); 4610 out: 4611 #endif 4612 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head); 4613 rps_input_queue_tail_save(&rflow->last_qtail, head); 4614 } 4615 4616 WRITE_ONCE(rflow->cpu, next_cpu); 4617 return rflow; 4618 } 4619 4620 /* 4621 * get_rps_cpu is called from netif_receive_skb and returns the target 4622 * CPU from the RPS map of the receiving queue for a given skb. 4623 * rcu_read_lock must be held on entry. 4624 */ 4625 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4626 struct rps_dev_flow **rflowp) 4627 { 4628 const struct rps_sock_flow_table *sock_flow_table; 4629 struct netdev_rx_queue *rxqueue = dev->_rx; 4630 struct rps_dev_flow_table *flow_table; 4631 struct rps_map *map; 4632 int cpu = -1; 4633 u32 tcpu; 4634 u32 hash; 4635 4636 if (skb_rx_queue_recorded(skb)) { 4637 u16 index = skb_get_rx_queue(skb); 4638 4639 if (unlikely(index >= dev->real_num_rx_queues)) { 4640 WARN_ONCE(dev->real_num_rx_queues > 1, 4641 "%s received packet on queue %u, but number " 4642 "of RX queues is %u\n", 4643 dev->name, index, dev->real_num_rx_queues); 4644 goto done; 4645 } 4646 rxqueue += index; 4647 } 4648 4649 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4650 4651 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4652 map = rcu_dereference(rxqueue->rps_map); 4653 if (!flow_table && !map) 4654 goto done; 4655 4656 skb_reset_network_header(skb); 4657 hash = skb_get_hash(skb); 4658 if (!hash) 4659 goto done; 4660 4661 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table); 4662 if (flow_table && sock_flow_table) { 4663 struct rps_dev_flow *rflow; 4664 u32 next_cpu; 4665 u32 ident; 4666 4667 /* First check into global flow table if there is a match. 4668 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4669 */ 4670 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4671 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask) 4672 goto try_rps; 4673 4674 next_cpu = ident & net_hotdata.rps_cpu_mask; 4675 4676 /* OK, now we know there is a match, 4677 * we can look at the local (per receive queue) flow table 4678 */ 4679 rflow = &flow_table->flows[hash & flow_table->mask]; 4680 tcpu = rflow->cpu; 4681 4682 /* 4683 * If the desired CPU (where last recvmsg was done) is 4684 * different from current CPU (one in the rx-queue flow 4685 * table entry), switch if one of the following holds: 4686 * - Current CPU is unset (>= nr_cpu_ids). 4687 * - Current CPU is offline. 4688 * - The current CPU's queue tail has advanced beyond the 4689 * last packet that was enqueued using this table entry. 4690 * This guarantees that all previous packets for the flow 4691 * have been dequeued, thus preserving in order delivery. 4692 */ 4693 if (unlikely(tcpu != next_cpu) && 4694 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4695 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) - 4696 rflow->last_qtail)) >= 0)) { 4697 tcpu = next_cpu; 4698 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4699 } 4700 4701 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4702 *rflowp = rflow; 4703 cpu = tcpu; 4704 goto done; 4705 } 4706 } 4707 4708 try_rps: 4709 4710 if (map) { 4711 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4712 if (cpu_online(tcpu)) { 4713 cpu = tcpu; 4714 goto done; 4715 } 4716 } 4717 4718 done: 4719 return cpu; 4720 } 4721 4722 #ifdef CONFIG_RFS_ACCEL 4723 4724 /** 4725 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4726 * @dev: Device on which the filter was set 4727 * @rxq_index: RX queue index 4728 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4729 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4730 * 4731 * Drivers that implement ndo_rx_flow_steer() should periodically call 4732 * this function for each installed filter and remove the filters for 4733 * which it returns %true. 4734 */ 4735 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4736 u32 flow_id, u16 filter_id) 4737 { 4738 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4739 struct rps_dev_flow_table *flow_table; 4740 struct rps_dev_flow *rflow; 4741 bool expire = true; 4742 unsigned int cpu; 4743 4744 rcu_read_lock(); 4745 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4746 if (flow_table && flow_id <= flow_table->mask) { 4747 rflow = &flow_table->flows[flow_id]; 4748 cpu = READ_ONCE(rflow->cpu); 4749 if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids && 4750 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) - 4751 READ_ONCE(rflow->last_qtail)) < 4752 (int)(10 * flow_table->mask))) 4753 expire = false; 4754 } 4755 rcu_read_unlock(); 4756 return expire; 4757 } 4758 EXPORT_SYMBOL(rps_may_expire_flow); 4759 4760 #endif /* CONFIG_RFS_ACCEL */ 4761 4762 /* Called from hardirq (IPI) context */ 4763 static void rps_trigger_softirq(void *data) 4764 { 4765 struct softnet_data *sd = data; 4766 4767 ____napi_schedule(sd, &sd->backlog); 4768 sd->received_rps++; 4769 } 4770 4771 #endif /* CONFIG_RPS */ 4772 4773 /* Called from hardirq (IPI) context */ 4774 static void trigger_rx_softirq(void *data) 4775 { 4776 struct softnet_data *sd = data; 4777 4778 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4779 smp_store_release(&sd->defer_ipi_scheduled, 0); 4780 } 4781 4782 /* 4783 * After we queued a packet into sd->input_pkt_queue, 4784 * we need to make sure this queue is serviced soon. 4785 * 4786 * - If this is another cpu queue, link it to our rps_ipi_list, 4787 * and make sure we will process rps_ipi_list from net_rx_action(). 4788 * 4789 * - If this is our own queue, NAPI schedule our backlog. 4790 * Note that this also raises NET_RX_SOFTIRQ. 4791 */ 4792 static void napi_schedule_rps(struct softnet_data *sd) 4793 { 4794 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4795 4796 #ifdef CONFIG_RPS 4797 if (sd != mysd) { 4798 if (use_backlog_threads()) { 4799 __napi_schedule_irqoff(&sd->backlog); 4800 return; 4801 } 4802 4803 sd->rps_ipi_next = mysd->rps_ipi_list; 4804 mysd->rps_ipi_list = sd; 4805 4806 /* If not called from net_rx_action() or napi_threaded_poll() 4807 * we have to raise NET_RX_SOFTIRQ. 4808 */ 4809 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4810 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4811 return; 4812 } 4813 #endif /* CONFIG_RPS */ 4814 __napi_schedule_irqoff(&mysd->backlog); 4815 } 4816 4817 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu) 4818 { 4819 unsigned long flags; 4820 4821 if (use_backlog_threads()) { 4822 backlog_lock_irq_save(sd, &flags); 4823 4824 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4825 __napi_schedule_irqoff(&sd->backlog); 4826 4827 backlog_unlock_irq_restore(sd, &flags); 4828 4829 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) { 4830 smp_call_function_single_async(cpu, &sd->defer_csd); 4831 } 4832 } 4833 4834 #ifdef CONFIG_NET_FLOW_LIMIT 4835 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4836 #endif 4837 4838 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4839 { 4840 #ifdef CONFIG_NET_FLOW_LIMIT 4841 struct sd_flow_limit *fl; 4842 struct softnet_data *sd; 4843 unsigned int old_flow, new_flow; 4844 4845 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1)) 4846 return false; 4847 4848 sd = this_cpu_ptr(&softnet_data); 4849 4850 rcu_read_lock(); 4851 fl = rcu_dereference(sd->flow_limit); 4852 if (fl) { 4853 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4854 old_flow = fl->history[fl->history_head]; 4855 fl->history[fl->history_head] = new_flow; 4856 4857 fl->history_head++; 4858 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4859 4860 if (likely(fl->buckets[old_flow])) 4861 fl->buckets[old_flow]--; 4862 4863 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4864 fl->count++; 4865 rcu_read_unlock(); 4866 return true; 4867 } 4868 } 4869 rcu_read_unlock(); 4870 #endif 4871 return false; 4872 } 4873 4874 /* 4875 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4876 * queue (may be a remote CPU queue). 4877 */ 4878 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4879 unsigned int *qtail) 4880 { 4881 enum skb_drop_reason reason; 4882 struct softnet_data *sd; 4883 unsigned long flags; 4884 unsigned int qlen; 4885 int max_backlog; 4886 u32 tail; 4887 4888 reason = SKB_DROP_REASON_DEV_READY; 4889 if (!netif_running(skb->dev)) 4890 goto bad_dev; 4891 4892 reason = SKB_DROP_REASON_CPU_BACKLOG; 4893 sd = &per_cpu(softnet_data, cpu); 4894 4895 qlen = skb_queue_len_lockless(&sd->input_pkt_queue); 4896 max_backlog = READ_ONCE(net_hotdata.max_backlog); 4897 if (unlikely(qlen > max_backlog)) 4898 goto cpu_backlog_drop; 4899 backlog_lock_irq_save(sd, &flags); 4900 qlen = skb_queue_len(&sd->input_pkt_queue); 4901 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) { 4902 if (!qlen) { 4903 /* Schedule NAPI for backlog device. We can use 4904 * non atomic operation as we own the queue lock. 4905 */ 4906 if (!__test_and_set_bit(NAPI_STATE_SCHED, 4907 &sd->backlog.state)) 4908 napi_schedule_rps(sd); 4909 } 4910 __skb_queue_tail(&sd->input_pkt_queue, skb); 4911 tail = rps_input_queue_tail_incr(sd); 4912 backlog_unlock_irq_restore(sd, &flags); 4913 4914 /* save the tail outside of the critical section */ 4915 rps_input_queue_tail_save(qtail, tail); 4916 return NET_RX_SUCCESS; 4917 } 4918 4919 backlog_unlock_irq_restore(sd, &flags); 4920 4921 cpu_backlog_drop: 4922 atomic_inc(&sd->dropped); 4923 bad_dev: 4924 dev_core_stats_rx_dropped_inc(skb->dev); 4925 kfree_skb_reason(skb, reason); 4926 return NET_RX_DROP; 4927 } 4928 4929 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4930 { 4931 struct net_device *dev = skb->dev; 4932 struct netdev_rx_queue *rxqueue; 4933 4934 rxqueue = dev->_rx; 4935 4936 if (skb_rx_queue_recorded(skb)) { 4937 u16 index = skb_get_rx_queue(skb); 4938 4939 if (unlikely(index >= dev->real_num_rx_queues)) { 4940 WARN_ONCE(dev->real_num_rx_queues > 1, 4941 "%s received packet on queue %u, but number " 4942 "of RX queues is %u\n", 4943 dev->name, index, dev->real_num_rx_queues); 4944 4945 return rxqueue; /* Return first rxqueue */ 4946 } 4947 rxqueue += index; 4948 } 4949 return rxqueue; 4950 } 4951 4952 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4953 const struct bpf_prog *xdp_prog) 4954 { 4955 void *orig_data, *orig_data_end, *hard_start; 4956 struct netdev_rx_queue *rxqueue; 4957 bool orig_bcast, orig_host; 4958 u32 mac_len, frame_sz; 4959 __be16 orig_eth_type; 4960 struct ethhdr *eth; 4961 u32 metalen, act; 4962 int off; 4963 4964 /* The XDP program wants to see the packet starting at the MAC 4965 * header. 4966 */ 4967 mac_len = skb->data - skb_mac_header(skb); 4968 hard_start = skb->data - skb_headroom(skb); 4969 4970 /* SKB "head" area always have tailroom for skb_shared_info */ 4971 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4972 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4973 4974 rxqueue = netif_get_rxqueue(skb); 4975 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4976 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4977 skb_headlen(skb) + mac_len, true); 4978 if (skb_is_nonlinear(skb)) { 4979 skb_shinfo(skb)->xdp_frags_size = skb->data_len; 4980 xdp_buff_set_frags_flag(xdp); 4981 } else { 4982 xdp_buff_clear_frags_flag(xdp); 4983 } 4984 4985 orig_data_end = xdp->data_end; 4986 orig_data = xdp->data; 4987 eth = (struct ethhdr *)xdp->data; 4988 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4989 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4990 orig_eth_type = eth->h_proto; 4991 4992 act = bpf_prog_run_xdp(xdp_prog, xdp); 4993 4994 /* check if bpf_xdp_adjust_head was used */ 4995 off = xdp->data - orig_data; 4996 if (off) { 4997 if (off > 0) 4998 __skb_pull(skb, off); 4999 else if (off < 0) 5000 __skb_push(skb, -off); 5001 5002 skb->mac_header += off; 5003 skb_reset_network_header(skb); 5004 } 5005 5006 /* check if bpf_xdp_adjust_tail was used */ 5007 off = xdp->data_end - orig_data_end; 5008 if (off != 0) { 5009 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 5010 skb->len += off; /* positive on grow, negative on shrink */ 5011 } 5012 5013 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers 5014 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. 5015 */ 5016 if (xdp_buff_has_frags(xdp)) 5017 skb->data_len = skb_shinfo(skb)->xdp_frags_size; 5018 else 5019 skb->data_len = 0; 5020 5021 /* check if XDP changed eth hdr such SKB needs update */ 5022 eth = (struct ethhdr *)xdp->data; 5023 if ((orig_eth_type != eth->h_proto) || 5024 (orig_host != ether_addr_equal_64bits(eth->h_dest, 5025 skb->dev->dev_addr)) || 5026 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 5027 __skb_push(skb, ETH_HLEN); 5028 skb->pkt_type = PACKET_HOST; 5029 skb->protocol = eth_type_trans(skb, skb->dev); 5030 } 5031 5032 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 5033 * before calling us again on redirect path. We do not call do_redirect 5034 * as we leave that up to the caller. 5035 * 5036 * Caller is responsible for managing lifetime of skb (i.e. calling 5037 * kfree_skb in response to actions it cannot handle/XDP_DROP). 5038 */ 5039 switch (act) { 5040 case XDP_REDIRECT: 5041 case XDP_TX: 5042 __skb_push(skb, mac_len); 5043 break; 5044 case XDP_PASS: 5045 metalen = xdp->data - xdp->data_meta; 5046 if (metalen) 5047 skb_metadata_set(skb, metalen); 5048 break; 5049 } 5050 5051 return act; 5052 } 5053 5054 static int 5055 netif_skb_check_for_xdp(struct sk_buff **pskb, const struct bpf_prog *prog) 5056 { 5057 struct sk_buff *skb = *pskb; 5058 int err, hroom, troom; 5059 5060 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog)) 5061 return 0; 5062 5063 /* In case we have to go down the path and also linearize, 5064 * then lets do the pskb_expand_head() work just once here. 5065 */ 5066 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 5067 troom = skb->tail + skb->data_len - skb->end; 5068 err = pskb_expand_head(skb, 5069 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 5070 troom > 0 ? troom + 128 : 0, GFP_ATOMIC); 5071 if (err) 5072 return err; 5073 5074 return skb_linearize(skb); 5075 } 5076 5077 static u32 netif_receive_generic_xdp(struct sk_buff **pskb, 5078 struct xdp_buff *xdp, 5079 const struct bpf_prog *xdp_prog) 5080 { 5081 struct sk_buff *skb = *pskb; 5082 u32 mac_len, act = XDP_DROP; 5083 5084 /* Reinjected packets coming from act_mirred or similar should 5085 * not get XDP generic processing. 5086 */ 5087 if (skb_is_redirected(skb)) 5088 return XDP_PASS; 5089 5090 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM 5091 * bytes. This is the guarantee that also native XDP provides, 5092 * thus we need to do it here as well. 5093 */ 5094 mac_len = skb->data - skb_mac_header(skb); 5095 __skb_push(skb, mac_len); 5096 5097 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 5098 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 5099 if (netif_skb_check_for_xdp(pskb, xdp_prog)) 5100 goto do_drop; 5101 } 5102 5103 __skb_pull(*pskb, mac_len); 5104 5105 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); 5106 switch (act) { 5107 case XDP_REDIRECT: 5108 case XDP_TX: 5109 case XDP_PASS: 5110 break; 5111 default: 5112 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); 5113 fallthrough; 5114 case XDP_ABORTED: 5115 trace_xdp_exception((*pskb)->dev, xdp_prog, act); 5116 fallthrough; 5117 case XDP_DROP: 5118 do_drop: 5119 kfree_skb(*pskb); 5120 break; 5121 } 5122 5123 return act; 5124 } 5125 5126 /* When doing generic XDP we have to bypass the qdisc layer and the 5127 * network taps in order to match in-driver-XDP behavior. This also means 5128 * that XDP packets are able to starve other packets going through a qdisc, 5129 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5130 * queues, so they do not have this starvation issue. 5131 */ 5132 void generic_xdp_tx(struct sk_buff *skb, const struct bpf_prog *xdp_prog) 5133 { 5134 struct net_device *dev = skb->dev; 5135 struct netdev_queue *txq; 5136 bool free_skb = true; 5137 int cpu, rc; 5138 5139 txq = netdev_core_pick_tx(dev, skb, NULL); 5140 cpu = smp_processor_id(); 5141 HARD_TX_LOCK(dev, txq, cpu); 5142 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5143 rc = netdev_start_xmit(skb, dev, txq, 0); 5144 if (dev_xmit_complete(rc)) 5145 free_skb = false; 5146 } 5147 HARD_TX_UNLOCK(dev, txq); 5148 if (free_skb) { 5149 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5150 dev_core_stats_tx_dropped_inc(dev); 5151 kfree_skb(skb); 5152 } 5153 } 5154 5155 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5156 5157 int do_xdp_generic(const struct bpf_prog *xdp_prog, struct sk_buff **pskb) 5158 { 5159 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 5160 5161 if (xdp_prog) { 5162 struct xdp_buff xdp; 5163 u32 act; 5164 int err; 5165 5166 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 5167 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); 5168 if (act != XDP_PASS) { 5169 switch (act) { 5170 case XDP_REDIRECT: 5171 err = xdp_do_generic_redirect((*pskb)->dev, *pskb, 5172 &xdp, xdp_prog); 5173 if (err) 5174 goto out_redir; 5175 break; 5176 case XDP_TX: 5177 generic_xdp_tx(*pskb, xdp_prog); 5178 break; 5179 } 5180 bpf_net_ctx_clear(bpf_net_ctx); 5181 return XDP_DROP; 5182 } 5183 bpf_net_ctx_clear(bpf_net_ctx); 5184 } 5185 return XDP_PASS; 5186 out_redir: 5187 bpf_net_ctx_clear(bpf_net_ctx); 5188 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); 5189 return XDP_DROP; 5190 } 5191 EXPORT_SYMBOL_GPL(do_xdp_generic); 5192 5193 static int netif_rx_internal(struct sk_buff *skb) 5194 { 5195 int ret; 5196 5197 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5198 5199 trace_netif_rx(skb); 5200 5201 #ifdef CONFIG_RPS 5202 if (static_branch_unlikely(&rps_needed)) { 5203 struct rps_dev_flow voidflow, *rflow = &voidflow; 5204 int cpu; 5205 5206 rcu_read_lock(); 5207 5208 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5209 if (cpu < 0) 5210 cpu = smp_processor_id(); 5211 5212 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5213 5214 rcu_read_unlock(); 5215 } else 5216 #endif 5217 { 5218 unsigned int qtail; 5219 5220 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5221 } 5222 return ret; 5223 } 5224 5225 /** 5226 * __netif_rx - Slightly optimized version of netif_rx 5227 * @skb: buffer to post 5228 * 5229 * This behaves as netif_rx except that it does not disable bottom halves. 5230 * As a result this function may only be invoked from the interrupt context 5231 * (either hard or soft interrupt). 5232 */ 5233 int __netif_rx(struct sk_buff *skb) 5234 { 5235 int ret; 5236 5237 lockdep_assert_once(hardirq_count() | softirq_count()); 5238 5239 trace_netif_rx_entry(skb); 5240 ret = netif_rx_internal(skb); 5241 trace_netif_rx_exit(ret); 5242 return ret; 5243 } 5244 EXPORT_SYMBOL(__netif_rx); 5245 5246 /** 5247 * netif_rx - post buffer to the network code 5248 * @skb: buffer to post 5249 * 5250 * This function receives a packet from a device driver and queues it for 5251 * the upper (protocol) levels to process via the backlog NAPI device. It 5252 * always succeeds. The buffer may be dropped during processing for 5253 * congestion control or by the protocol layers. 5254 * The network buffer is passed via the backlog NAPI device. Modern NIC 5255 * driver should use NAPI and GRO. 5256 * This function can used from interrupt and from process context. The 5257 * caller from process context must not disable interrupts before invoking 5258 * this function. 5259 * 5260 * return values: 5261 * NET_RX_SUCCESS (no congestion) 5262 * NET_RX_DROP (packet was dropped) 5263 * 5264 */ 5265 int netif_rx(struct sk_buff *skb) 5266 { 5267 bool need_bh_off = !(hardirq_count() | softirq_count()); 5268 int ret; 5269 5270 if (need_bh_off) 5271 local_bh_disable(); 5272 trace_netif_rx_entry(skb); 5273 ret = netif_rx_internal(skb); 5274 trace_netif_rx_exit(ret); 5275 if (need_bh_off) 5276 local_bh_enable(); 5277 return ret; 5278 } 5279 EXPORT_SYMBOL(netif_rx); 5280 5281 static __latent_entropy void net_tx_action(void) 5282 { 5283 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5284 5285 if (sd->completion_queue) { 5286 struct sk_buff *clist; 5287 5288 local_irq_disable(); 5289 clist = sd->completion_queue; 5290 sd->completion_queue = NULL; 5291 local_irq_enable(); 5292 5293 while (clist) { 5294 struct sk_buff *skb = clist; 5295 5296 clist = clist->next; 5297 5298 WARN_ON(refcount_read(&skb->users)); 5299 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5300 trace_consume_skb(skb, net_tx_action); 5301 else 5302 trace_kfree_skb(skb, net_tx_action, 5303 get_kfree_skb_cb(skb)->reason, NULL); 5304 5305 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5306 __kfree_skb(skb); 5307 else 5308 __napi_kfree_skb(skb, 5309 get_kfree_skb_cb(skb)->reason); 5310 } 5311 } 5312 5313 if (sd->output_queue) { 5314 struct Qdisc *head; 5315 5316 local_irq_disable(); 5317 head = sd->output_queue; 5318 sd->output_queue = NULL; 5319 sd->output_queue_tailp = &sd->output_queue; 5320 local_irq_enable(); 5321 5322 rcu_read_lock(); 5323 5324 while (head) { 5325 struct Qdisc *q = head; 5326 spinlock_t *root_lock = NULL; 5327 5328 head = head->next_sched; 5329 5330 /* We need to make sure head->next_sched is read 5331 * before clearing __QDISC_STATE_SCHED 5332 */ 5333 smp_mb__before_atomic(); 5334 5335 if (!(q->flags & TCQ_F_NOLOCK)) { 5336 root_lock = qdisc_lock(q); 5337 spin_lock(root_lock); 5338 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5339 &q->state))) { 5340 /* There is a synchronize_net() between 5341 * STATE_DEACTIVATED flag being set and 5342 * qdisc_reset()/some_qdisc_is_busy() in 5343 * dev_deactivate(), so we can safely bail out 5344 * early here to avoid data race between 5345 * qdisc_deactivate() and some_qdisc_is_busy() 5346 * for lockless qdisc. 5347 */ 5348 clear_bit(__QDISC_STATE_SCHED, &q->state); 5349 continue; 5350 } 5351 5352 clear_bit(__QDISC_STATE_SCHED, &q->state); 5353 qdisc_run(q); 5354 if (root_lock) 5355 spin_unlock(root_lock); 5356 } 5357 5358 rcu_read_unlock(); 5359 } 5360 5361 xfrm_dev_backlog(sd); 5362 } 5363 5364 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5365 /* This hook is defined here for ATM LANE */ 5366 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5367 unsigned char *addr) __read_mostly; 5368 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5369 #endif 5370 5371 /** 5372 * netdev_is_rx_handler_busy - check if receive handler is registered 5373 * @dev: device to check 5374 * 5375 * Check if a receive handler is already registered for a given device. 5376 * Return true if there one. 5377 * 5378 * The caller must hold the rtnl_mutex. 5379 */ 5380 bool netdev_is_rx_handler_busy(struct net_device *dev) 5381 { 5382 ASSERT_RTNL(); 5383 return dev && rtnl_dereference(dev->rx_handler); 5384 } 5385 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5386 5387 /** 5388 * netdev_rx_handler_register - register receive handler 5389 * @dev: device to register a handler for 5390 * @rx_handler: receive handler to register 5391 * @rx_handler_data: data pointer that is used by rx handler 5392 * 5393 * Register a receive handler for a device. This handler will then be 5394 * called from __netif_receive_skb. A negative errno code is returned 5395 * on a failure. 5396 * 5397 * The caller must hold the rtnl_mutex. 5398 * 5399 * For a general description of rx_handler, see enum rx_handler_result. 5400 */ 5401 int netdev_rx_handler_register(struct net_device *dev, 5402 rx_handler_func_t *rx_handler, 5403 void *rx_handler_data) 5404 { 5405 if (netdev_is_rx_handler_busy(dev)) 5406 return -EBUSY; 5407 5408 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5409 return -EINVAL; 5410 5411 /* Note: rx_handler_data must be set before rx_handler */ 5412 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5413 rcu_assign_pointer(dev->rx_handler, rx_handler); 5414 5415 return 0; 5416 } 5417 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5418 5419 /** 5420 * netdev_rx_handler_unregister - unregister receive handler 5421 * @dev: device to unregister a handler from 5422 * 5423 * Unregister a receive handler from a device. 5424 * 5425 * The caller must hold the rtnl_mutex. 5426 */ 5427 void netdev_rx_handler_unregister(struct net_device *dev) 5428 { 5429 5430 ASSERT_RTNL(); 5431 RCU_INIT_POINTER(dev->rx_handler, NULL); 5432 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5433 * section has a guarantee to see a non NULL rx_handler_data 5434 * as well. 5435 */ 5436 synchronize_net(); 5437 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5438 } 5439 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5440 5441 /* 5442 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5443 * the special handling of PFMEMALLOC skbs. 5444 */ 5445 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5446 { 5447 switch (skb->protocol) { 5448 case htons(ETH_P_ARP): 5449 case htons(ETH_P_IP): 5450 case htons(ETH_P_IPV6): 5451 case htons(ETH_P_8021Q): 5452 case htons(ETH_P_8021AD): 5453 return true; 5454 default: 5455 return false; 5456 } 5457 } 5458 5459 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5460 int *ret, struct net_device *orig_dev) 5461 { 5462 if (nf_hook_ingress_active(skb)) { 5463 int ingress_retval; 5464 5465 if (*pt_prev) { 5466 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5467 *pt_prev = NULL; 5468 } 5469 5470 rcu_read_lock(); 5471 ingress_retval = nf_hook_ingress(skb); 5472 rcu_read_unlock(); 5473 return ingress_retval; 5474 } 5475 return 0; 5476 } 5477 5478 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5479 struct packet_type **ppt_prev) 5480 { 5481 struct packet_type *ptype, *pt_prev; 5482 rx_handler_func_t *rx_handler; 5483 struct sk_buff *skb = *pskb; 5484 struct net_device *orig_dev; 5485 bool deliver_exact = false; 5486 int ret = NET_RX_DROP; 5487 __be16 type; 5488 5489 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5490 5491 trace_netif_receive_skb(skb); 5492 5493 orig_dev = skb->dev; 5494 5495 skb_reset_network_header(skb); 5496 if (!skb_transport_header_was_set(skb)) 5497 skb_reset_transport_header(skb); 5498 skb_reset_mac_len(skb); 5499 5500 pt_prev = NULL; 5501 5502 another_round: 5503 skb->skb_iif = skb->dev->ifindex; 5504 5505 __this_cpu_inc(softnet_data.processed); 5506 5507 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5508 int ret2; 5509 5510 migrate_disable(); 5511 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), 5512 &skb); 5513 migrate_enable(); 5514 5515 if (ret2 != XDP_PASS) { 5516 ret = NET_RX_DROP; 5517 goto out; 5518 } 5519 } 5520 5521 if (eth_type_vlan(skb->protocol)) { 5522 skb = skb_vlan_untag(skb); 5523 if (unlikely(!skb)) 5524 goto out; 5525 } 5526 5527 if (skb_skip_tc_classify(skb)) 5528 goto skip_classify; 5529 5530 if (pfmemalloc) 5531 goto skip_taps; 5532 5533 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) { 5534 if (pt_prev) 5535 ret = deliver_skb(skb, pt_prev, orig_dev); 5536 pt_prev = ptype; 5537 } 5538 5539 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5540 if (pt_prev) 5541 ret = deliver_skb(skb, pt_prev, orig_dev); 5542 pt_prev = ptype; 5543 } 5544 5545 skip_taps: 5546 #ifdef CONFIG_NET_INGRESS 5547 if (static_branch_unlikely(&ingress_needed_key)) { 5548 bool another = false; 5549 5550 nf_skip_egress(skb, true); 5551 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5552 &another); 5553 if (another) 5554 goto another_round; 5555 if (!skb) 5556 goto out; 5557 5558 nf_skip_egress(skb, false); 5559 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5560 goto out; 5561 } 5562 #endif 5563 skb_reset_redirect(skb); 5564 skip_classify: 5565 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5566 goto drop; 5567 5568 if (skb_vlan_tag_present(skb)) { 5569 if (pt_prev) { 5570 ret = deliver_skb(skb, pt_prev, orig_dev); 5571 pt_prev = NULL; 5572 } 5573 if (vlan_do_receive(&skb)) 5574 goto another_round; 5575 else if (unlikely(!skb)) 5576 goto out; 5577 } 5578 5579 rx_handler = rcu_dereference(skb->dev->rx_handler); 5580 if (rx_handler) { 5581 if (pt_prev) { 5582 ret = deliver_skb(skb, pt_prev, orig_dev); 5583 pt_prev = NULL; 5584 } 5585 switch (rx_handler(&skb)) { 5586 case RX_HANDLER_CONSUMED: 5587 ret = NET_RX_SUCCESS; 5588 goto out; 5589 case RX_HANDLER_ANOTHER: 5590 goto another_round; 5591 case RX_HANDLER_EXACT: 5592 deliver_exact = true; 5593 break; 5594 case RX_HANDLER_PASS: 5595 break; 5596 default: 5597 BUG(); 5598 } 5599 } 5600 5601 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5602 check_vlan_id: 5603 if (skb_vlan_tag_get_id(skb)) { 5604 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5605 * find vlan device. 5606 */ 5607 skb->pkt_type = PACKET_OTHERHOST; 5608 } else if (eth_type_vlan(skb->protocol)) { 5609 /* Outer header is 802.1P with vlan 0, inner header is 5610 * 802.1Q or 802.1AD and vlan_do_receive() above could 5611 * not find vlan dev for vlan id 0. 5612 */ 5613 __vlan_hwaccel_clear_tag(skb); 5614 skb = skb_vlan_untag(skb); 5615 if (unlikely(!skb)) 5616 goto out; 5617 if (vlan_do_receive(&skb)) 5618 /* After stripping off 802.1P header with vlan 0 5619 * vlan dev is found for inner header. 5620 */ 5621 goto another_round; 5622 else if (unlikely(!skb)) 5623 goto out; 5624 else 5625 /* We have stripped outer 802.1P vlan 0 header. 5626 * But could not find vlan dev. 5627 * check again for vlan id to set OTHERHOST. 5628 */ 5629 goto check_vlan_id; 5630 } 5631 /* Note: we might in the future use prio bits 5632 * and set skb->priority like in vlan_do_receive() 5633 * For the time being, just ignore Priority Code Point 5634 */ 5635 __vlan_hwaccel_clear_tag(skb); 5636 } 5637 5638 type = skb->protocol; 5639 5640 /* deliver only exact match when indicated */ 5641 if (likely(!deliver_exact)) { 5642 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5643 &ptype_base[ntohs(type) & 5644 PTYPE_HASH_MASK]); 5645 } 5646 5647 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5648 &orig_dev->ptype_specific); 5649 5650 if (unlikely(skb->dev != orig_dev)) { 5651 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5652 &skb->dev->ptype_specific); 5653 } 5654 5655 if (pt_prev) { 5656 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5657 goto drop; 5658 *ppt_prev = pt_prev; 5659 } else { 5660 drop: 5661 if (!deliver_exact) 5662 dev_core_stats_rx_dropped_inc(skb->dev); 5663 else 5664 dev_core_stats_rx_nohandler_inc(skb->dev); 5665 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5666 /* Jamal, now you will not able to escape explaining 5667 * me how you were going to use this. :-) 5668 */ 5669 ret = NET_RX_DROP; 5670 } 5671 5672 out: 5673 /* The invariant here is that if *ppt_prev is not NULL 5674 * then skb should also be non-NULL. 5675 * 5676 * Apparently *ppt_prev assignment above holds this invariant due to 5677 * skb dereferencing near it. 5678 */ 5679 *pskb = skb; 5680 return ret; 5681 } 5682 5683 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5684 { 5685 struct net_device *orig_dev = skb->dev; 5686 struct packet_type *pt_prev = NULL; 5687 int ret; 5688 5689 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5690 if (pt_prev) 5691 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5692 skb->dev, pt_prev, orig_dev); 5693 return ret; 5694 } 5695 5696 /** 5697 * netif_receive_skb_core - special purpose version of netif_receive_skb 5698 * @skb: buffer to process 5699 * 5700 * More direct receive version of netif_receive_skb(). It should 5701 * only be used by callers that have a need to skip RPS and Generic XDP. 5702 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5703 * 5704 * This function may only be called from softirq context and interrupts 5705 * should be enabled. 5706 * 5707 * Return values (usually ignored): 5708 * NET_RX_SUCCESS: no congestion 5709 * NET_RX_DROP: packet was dropped 5710 */ 5711 int netif_receive_skb_core(struct sk_buff *skb) 5712 { 5713 int ret; 5714 5715 rcu_read_lock(); 5716 ret = __netif_receive_skb_one_core(skb, false); 5717 rcu_read_unlock(); 5718 5719 return ret; 5720 } 5721 EXPORT_SYMBOL(netif_receive_skb_core); 5722 5723 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5724 struct packet_type *pt_prev, 5725 struct net_device *orig_dev) 5726 { 5727 struct sk_buff *skb, *next; 5728 5729 if (!pt_prev) 5730 return; 5731 if (list_empty(head)) 5732 return; 5733 if (pt_prev->list_func != NULL) 5734 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5735 ip_list_rcv, head, pt_prev, orig_dev); 5736 else 5737 list_for_each_entry_safe(skb, next, head, list) { 5738 skb_list_del_init(skb); 5739 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5740 } 5741 } 5742 5743 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5744 { 5745 /* Fast-path assumptions: 5746 * - There is no RX handler. 5747 * - Only one packet_type matches. 5748 * If either of these fails, we will end up doing some per-packet 5749 * processing in-line, then handling the 'last ptype' for the whole 5750 * sublist. This can't cause out-of-order delivery to any single ptype, 5751 * because the 'last ptype' must be constant across the sublist, and all 5752 * other ptypes are handled per-packet. 5753 */ 5754 /* Current (common) ptype of sublist */ 5755 struct packet_type *pt_curr = NULL; 5756 /* Current (common) orig_dev of sublist */ 5757 struct net_device *od_curr = NULL; 5758 struct sk_buff *skb, *next; 5759 LIST_HEAD(sublist); 5760 5761 list_for_each_entry_safe(skb, next, head, list) { 5762 struct net_device *orig_dev = skb->dev; 5763 struct packet_type *pt_prev = NULL; 5764 5765 skb_list_del_init(skb); 5766 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5767 if (!pt_prev) 5768 continue; 5769 if (pt_curr != pt_prev || od_curr != orig_dev) { 5770 /* dispatch old sublist */ 5771 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5772 /* start new sublist */ 5773 INIT_LIST_HEAD(&sublist); 5774 pt_curr = pt_prev; 5775 od_curr = orig_dev; 5776 } 5777 list_add_tail(&skb->list, &sublist); 5778 } 5779 5780 /* dispatch final sublist */ 5781 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5782 } 5783 5784 static int __netif_receive_skb(struct sk_buff *skb) 5785 { 5786 int ret; 5787 5788 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5789 unsigned int noreclaim_flag; 5790 5791 /* 5792 * PFMEMALLOC skbs are special, they should 5793 * - be delivered to SOCK_MEMALLOC sockets only 5794 * - stay away from userspace 5795 * - have bounded memory usage 5796 * 5797 * Use PF_MEMALLOC as this saves us from propagating the allocation 5798 * context down to all allocation sites. 5799 */ 5800 noreclaim_flag = memalloc_noreclaim_save(); 5801 ret = __netif_receive_skb_one_core(skb, true); 5802 memalloc_noreclaim_restore(noreclaim_flag); 5803 } else 5804 ret = __netif_receive_skb_one_core(skb, false); 5805 5806 return ret; 5807 } 5808 5809 static void __netif_receive_skb_list(struct list_head *head) 5810 { 5811 unsigned long noreclaim_flag = 0; 5812 struct sk_buff *skb, *next; 5813 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5814 5815 list_for_each_entry_safe(skb, next, head, list) { 5816 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5817 struct list_head sublist; 5818 5819 /* Handle the previous sublist */ 5820 list_cut_before(&sublist, head, &skb->list); 5821 if (!list_empty(&sublist)) 5822 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5823 pfmemalloc = !pfmemalloc; 5824 /* See comments in __netif_receive_skb */ 5825 if (pfmemalloc) 5826 noreclaim_flag = memalloc_noreclaim_save(); 5827 else 5828 memalloc_noreclaim_restore(noreclaim_flag); 5829 } 5830 } 5831 /* Handle the remaining sublist */ 5832 if (!list_empty(head)) 5833 __netif_receive_skb_list_core(head, pfmemalloc); 5834 /* Restore pflags */ 5835 if (pfmemalloc) 5836 memalloc_noreclaim_restore(noreclaim_flag); 5837 } 5838 5839 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5840 { 5841 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5842 struct bpf_prog *new = xdp->prog; 5843 int ret = 0; 5844 5845 switch (xdp->command) { 5846 case XDP_SETUP_PROG: 5847 rcu_assign_pointer(dev->xdp_prog, new); 5848 if (old) 5849 bpf_prog_put(old); 5850 5851 if (old && !new) { 5852 static_branch_dec(&generic_xdp_needed_key); 5853 } else if (new && !old) { 5854 static_branch_inc(&generic_xdp_needed_key); 5855 dev_disable_lro(dev); 5856 dev_disable_gro_hw(dev); 5857 } 5858 break; 5859 5860 default: 5861 ret = -EINVAL; 5862 break; 5863 } 5864 5865 return ret; 5866 } 5867 5868 static int netif_receive_skb_internal(struct sk_buff *skb) 5869 { 5870 int ret; 5871 5872 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5873 5874 if (skb_defer_rx_timestamp(skb)) 5875 return NET_RX_SUCCESS; 5876 5877 rcu_read_lock(); 5878 #ifdef CONFIG_RPS 5879 if (static_branch_unlikely(&rps_needed)) { 5880 struct rps_dev_flow voidflow, *rflow = &voidflow; 5881 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5882 5883 if (cpu >= 0) { 5884 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5885 rcu_read_unlock(); 5886 return ret; 5887 } 5888 } 5889 #endif 5890 ret = __netif_receive_skb(skb); 5891 rcu_read_unlock(); 5892 return ret; 5893 } 5894 5895 void netif_receive_skb_list_internal(struct list_head *head) 5896 { 5897 struct sk_buff *skb, *next; 5898 LIST_HEAD(sublist); 5899 5900 list_for_each_entry_safe(skb, next, head, list) { 5901 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), 5902 skb); 5903 skb_list_del_init(skb); 5904 if (!skb_defer_rx_timestamp(skb)) 5905 list_add_tail(&skb->list, &sublist); 5906 } 5907 list_splice_init(&sublist, head); 5908 5909 rcu_read_lock(); 5910 #ifdef CONFIG_RPS 5911 if (static_branch_unlikely(&rps_needed)) { 5912 list_for_each_entry_safe(skb, next, head, list) { 5913 struct rps_dev_flow voidflow, *rflow = &voidflow; 5914 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5915 5916 if (cpu >= 0) { 5917 /* Will be handled, remove from list */ 5918 skb_list_del_init(skb); 5919 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5920 } 5921 } 5922 } 5923 #endif 5924 __netif_receive_skb_list(head); 5925 rcu_read_unlock(); 5926 } 5927 5928 /** 5929 * netif_receive_skb - process receive buffer from network 5930 * @skb: buffer to process 5931 * 5932 * netif_receive_skb() is the main receive data processing function. 5933 * It always succeeds. The buffer may be dropped during processing 5934 * for congestion control or by the protocol layers. 5935 * 5936 * This function may only be called from softirq context and interrupts 5937 * should be enabled. 5938 * 5939 * Return values (usually ignored): 5940 * NET_RX_SUCCESS: no congestion 5941 * NET_RX_DROP: packet was dropped 5942 */ 5943 int netif_receive_skb(struct sk_buff *skb) 5944 { 5945 int ret; 5946 5947 trace_netif_receive_skb_entry(skb); 5948 5949 ret = netif_receive_skb_internal(skb); 5950 trace_netif_receive_skb_exit(ret); 5951 5952 return ret; 5953 } 5954 EXPORT_SYMBOL(netif_receive_skb); 5955 5956 /** 5957 * netif_receive_skb_list - process many receive buffers from network 5958 * @head: list of skbs to process. 5959 * 5960 * Since return value of netif_receive_skb() is normally ignored, and 5961 * wouldn't be meaningful for a list, this function returns void. 5962 * 5963 * This function may only be called from softirq context and interrupts 5964 * should be enabled. 5965 */ 5966 void netif_receive_skb_list(struct list_head *head) 5967 { 5968 struct sk_buff *skb; 5969 5970 if (list_empty(head)) 5971 return; 5972 if (trace_netif_receive_skb_list_entry_enabled()) { 5973 list_for_each_entry(skb, head, list) 5974 trace_netif_receive_skb_list_entry(skb); 5975 } 5976 netif_receive_skb_list_internal(head); 5977 trace_netif_receive_skb_list_exit(0); 5978 } 5979 EXPORT_SYMBOL(netif_receive_skb_list); 5980 5981 static DEFINE_PER_CPU(struct work_struct, flush_works); 5982 5983 /* Network device is going away, flush any packets still pending */ 5984 static void flush_backlog(struct work_struct *work) 5985 { 5986 struct sk_buff *skb, *tmp; 5987 struct softnet_data *sd; 5988 5989 local_bh_disable(); 5990 sd = this_cpu_ptr(&softnet_data); 5991 5992 backlog_lock_irq_disable(sd); 5993 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5994 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5995 __skb_unlink(skb, &sd->input_pkt_queue); 5996 dev_kfree_skb_irq(skb); 5997 rps_input_queue_head_incr(sd); 5998 } 5999 } 6000 backlog_unlock_irq_enable(sd); 6001 6002 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6003 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 6004 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 6005 __skb_unlink(skb, &sd->process_queue); 6006 kfree_skb(skb); 6007 rps_input_queue_head_incr(sd); 6008 } 6009 } 6010 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6011 local_bh_enable(); 6012 } 6013 6014 static bool flush_required(int cpu) 6015 { 6016 #if IS_ENABLED(CONFIG_RPS) 6017 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 6018 bool do_flush; 6019 6020 backlog_lock_irq_disable(sd); 6021 6022 /* as insertion into process_queue happens with the rps lock held, 6023 * process_queue access may race only with dequeue 6024 */ 6025 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 6026 !skb_queue_empty_lockless(&sd->process_queue); 6027 backlog_unlock_irq_enable(sd); 6028 6029 return do_flush; 6030 #endif 6031 /* without RPS we can't safely check input_pkt_queue: during a 6032 * concurrent remote skb_queue_splice() we can detect as empty both 6033 * input_pkt_queue and process_queue even if the latter could end-up 6034 * containing a lot of packets. 6035 */ 6036 return true; 6037 } 6038 6039 static void flush_all_backlogs(void) 6040 { 6041 static cpumask_t flush_cpus; 6042 unsigned int cpu; 6043 6044 /* since we are under rtnl lock protection we can use static data 6045 * for the cpumask and avoid allocating on stack the possibly 6046 * large mask 6047 */ 6048 ASSERT_RTNL(); 6049 6050 cpus_read_lock(); 6051 6052 cpumask_clear(&flush_cpus); 6053 for_each_online_cpu(cpu) { 6054 if (flush_required(cpu)) { 6055 queue_work_on(cpu, system_highpri_wq, 6056 per_cpu_ptr(&flush_works, cpu)); 6057 cpumask_set_cpu(cpu, &flush_cpus); 6058 } 6059 } 6060 6061 /* we can have in flight packet[s] on the cpus we are not flushing, 6062 * synchronize_net() in unregister_netdevice_many() will take care of 6063 * them 6064 */ 6065 for_each_cpu(cpu, &flush_cpus) 6066 flush_work(per_cpu_ptr(&flush_works, cpu)); 6067 6068 cpus_read_unlock(); 6069 } 6070 6071 static void net_rps_send_ipi(struct softnet_data *remsd) 6072 { 6073 #ifdef CONFIG_RPS 6074 while (remsd) { 6075 struct softnet_data *next = remsd->rps_ipi_next; 6076 6077 if (cpu_online(remsd->cpu)) 6078 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6079 remsd = next; 6080 } 6081 #endif 6082 } 6083 6084 /* 6085 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6086 * Note: called with local irq disabled, but exits with local irq enabled. 6087 */ 6088 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6089 { 6090 #ifdef CONFIG_RPS 6091 struct softnet_data *remsd = sd->rps_ipi_list; 6092 6093 if (!use_backlog_threads() && remsd) { 6094 sd->rps_ipi_list = NULL; 6095 6096 local_irq_enable(); 6097 6098 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6099 net_rps_send_ipi(remsd); 6100 } else 6101 #endif 6102 local_irq_enable(); 6103 } 6104 6105 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6106 { 6107 #ifdef CONFIG_RPS 6108 return !use_backlog_threads() && sd->rps_ipi_list; 6109 #else 6110 return false; 6111 #endif 6112 } 6113 6114 static int process_backlog(struct napi_struct *napi, int quota) 6115 { 6116 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6117 bool again = true; 6118 int work = 0; 6119 6120 /* Check if we have pending ipi, its better to send them now, 6121 * not waiting net_rx_action() end. 6122 */ 6123 if (sd_has_rps_ipi_waiting(sd)) { 6124 local_irq_disable(); 6125 net_rps_action_and_irq_enable(sd); 6126 } 6127 6128 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight); 6129 while (again) { 6130 struct sk_buff *skb; 6131 6132 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6133 while ((skb = __skb_dequeue(&sd->process_queue))) { 6134 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6135 rcu_read_lock(); 6136 __netif_receive_skb(skb); 6137 rcu_read_unlock(); 6138 if (++work >= quota) { 6139 rps_input_queue_head_add(sd, work); 6140 return work; 6141 } 6142 6143 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6144 } 6145 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6146 6147 backlog_lock_irq_disable(sd); 6148 if (skb_queue_empty(&sd->input_pkt_queue)) { 6149 /* 6150 * Inline a custom version of __napi_complete(). 6151 * only current cpu owns and manipulates this napi, 6152 * and NAPI_STATE_SCHED is the only possible flag set 6153 * on backlog. 6154 * We can use a plain write instead of clear_bit(), 6155 * and we dont need an smp_mb() memory barrier. 6156 */ 6157 napi->state &= NAPIF_STATE_THREADED; 6158 again = false; 6159 } else { 6160 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6161 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6162 &sd->process_queue); 6163 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6164 } 6165 backlog_unlock_irq_enable(sd); 6166 } 6167 6168 if (work) 6169 rps_input_queue_head_add(sd, work); 6170 return work; 6171 } 6172 6173 /** 6174 * __napi_schedule - schedule for receive 6175 * @n: entry to schedule 6176 * 6177 * The entry's receive function will be scheduled to run. 6178 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6179 */ 6180 void __napi_schedule(struct napi_struct *n) 6181 { 6182 unsigned long flags; 6183 6184 local_irq_save(flags); 6185 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6186 local_irq_restore(flags); 6187 } 6188 EXPORT_SYMBOL(__napi_schedule); 6189 6190 /** 6191 * napi_schedule_prep - check if napi can be scheduled 6192 * @n: napi context 6193 * 6194 * Test if NAPI routine is already running, and if not mark 6195 * it as running. This is used as a condition variable to 6196 * insure only one NAPI poll instance runs. We also make 6197 * sure there is no pending NAPI disable. 6198 */ 6199 bool napi_schedule_prep(struct napi_struct *n) 6200 { 6201 unsigned long new, val = READ_ONCE(n->state); 6202 6203 do { 6204 if (unlikely(val & NAPIF_STATE_DISABLE)) 6205 return false; 6206 new = val | NAPIF_STATE_SCHED; 6207 6208 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6209 * This was suggested by Alexander Duyck, as compiler 6210 * emits better code than : 6211 * if (val & NAPIF_STATE_SCHED) 6212 * new |= NAPIF_STATE_MISSED; 6213 */ 6214 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6215 NAPIF_STATE_MISSED; 6216 } while (!try_cmpxchg(&n->state, &val, new)); 6217 6218 return !(val & NAPIF_STATE_SCHED); 6219 } 6220 EXPORT_SYMBOL(napi_schedule_prep); 6221 6222 /** 6223 * __napi_schedule_irqoff - schedule for receive 6224 * @n: entry to schedule 6225 * 6226 * Variant of __napi_schedule() assuming hard irqs are masked. 6227 * 6228 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6229 * because the interrupt disabled assumption might not be true 6230 * due to force-threaded interrupts and spinlock substitution. 6231 */ 6232 void __napi_schedule_irqoff(struct napi_struct *n) 6233 { 6234 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6235 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6236 else 6237 __napi_schedule(n); 6238 } 6239 EXPORT_SYMBOL(__napi_schedule_irqoff); 6240 6241 bool napi_complete_done(struct napi_struct *n, int work_done) 6242 { 6243 unsigned long flags, val, new, timeout = 0; 6244 bool ret = true; 6245 6246 /* 6247 * 1) Don't let napi dequeue from the cpu poll list 6248 * just in case its running on a different cpu. 6249 * 2) If we are busy polling, do nothing here, we have 6250 * the guarantee we will be called later. 6251 */ 6252 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6253 NAPIF_STATE_IN_BUSY_POLL))) 6254 return false; 6255 6256 if (work_done) { 6257 if (n->gro_bitmask) 6258 timeout = napi_get_gro_flush_timeout(n); 6259 n->defer_hard_irqs_count = napi_get_defer_hard_irqs(n); 6260 } 6261 if (n->defer_hard_irqs_count > 0) { 6262 n->defer_hard_irqs_count--; 6263 timeout = napi_get_gro_flush_timeout(n); 6264 if (timeout) 6265 ret = false; 6266 } 6267 if (n->gro_bitmask) { 6268 /* When the NAPI instance uses a timeout and keeps postponing 6269 * it, we need to bound somehow the time packets are kept in 6270 * the GRO layer 6271 */ 6272 napi_gro_flush(n, !!timeout); 6273 } 6274 6275 gro_normal_list(n); 6276 6277 if (unlikely(!list_empty(&n->poll_list))) { 6278 /* If n->poll_list is not empty, we need to mask irqs */ 6279 local_irq_save(flags); 6280 list_del_init(&n->poll_list); 6281 local_irq_restore(flags); 6282 } 6283 WRITE_ONCE(n->list_owner, -1); 6284 6285 val = READ_ONCE(n->state); 6286 do { 6287 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6288 6289 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6290 NAPIF_STATE_SCHED_THREADED | 6291 NAPIF_STATE_PREFER_BUSY_POLL); 6292 6293 /* If STATE_MISSED was set, leave STATE_SCHED set, 6294 * because we will call napi->poll() one more time. 6295 * This C code was suggested by Alexander Duyck to help gcc. 6296 */ 6297 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6298 NAPIF_STATE_SCHED; 6299 } while (!try_cmpxchg(&n->state, &val, new)); 6300 6301 if (unlikely(val & NAPIF_STATE_MISSED)) { 6302 __napi_schedule(n); 6303 return false; 6304 } 6305 6306 if (timeout) 6307 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6308 HRTIMER_MODE_REL_PINNED); 6309 return ret; 6310 } 6311 EXPORT_SYMBOL(napi_complete_done); 6312 6313 /* must be called under rcu_read_lock(), as we dont take a reference */ 6314 struct napi_struct *napi_by_id(unsigned int napi_id) 6315 { 6316 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6317 struct napi_struct *napi; 6318 6319 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6320 if (napi->napi_id == napi_id) 6321 return napi; 6322 6323 return NULL; 6324 } 6325 6326 static void skb_defer_free_flush(struct softnet_data *sd) 6327 { 6328 struct sk_buff *skb, *next; 6329 6330 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6331 if (!READ_ONCE(sd->defer_list)) 6332 return; 6333 6334 spin_lock(&sd->defer_lock); 6335 skb = sd->defer_list; 6336 sd->defer_list = NULL; 6337 sd->defer_count = 0; 6338 spin_unlock(&sd->defer_lock); 6339 6340 while (skb != NULL) { 6341 next = skb->next; 6342 napi_consume_skb(skb, 1); 6343 skb = next; 6344 } 6345 } 6346 6347 #if defined(CONFIG_NET_RX_BUSY_POLL) 6348 6349 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6350 { 6351 if (!skip_schedule) { 6352 gro_normal_list(napi); 6353 __napi_schedule(napi); 6354 return; 6355 } 6356 6357 if (napi->gro_bitmask) { 6358 /* flush too old packets 6359 * If HZ < 1000, flush all packets. 6360 */ 6361 napi_gro_flush(napi, HZ >= 1000); 6362 } 6363 6364 gro_normal_list(napi); 6365 clear_bit(NAPI_STATE_SCHED, &napi->state); 6366 } 6367 6368 enum { 6369 NAPI_F_PREFER_BUSY_POLL = 1, 6370 NAPI_F_END_ON_RESCHED = 2, 6371 }; 6372 6373 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, 6374 unsigned flags, u16 budget) 6375 { 6376 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6377 bool skip_schedule = false; 6378 unsigned long timeout; 6379 int rc; 6380 6381 /* Busy polling means there is a high chance device driver hard irq 6382 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6383 * set in napi_schedule_prep(). 6384 * Since we are about to call napi->poll() once more, we can safely 6385 * clear NAPI_STATE_MISSED. 6386 * 6387 * Note: x86 could use a single "lock and ..." instruction 6388 * to perform these two clear_bit() 6389 */ 6390 clear_bit(NAPI_STATE_MISSED, &napi->state); 6391 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6392 6393 local_bh_disable(); 6394 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6395 6396 if (flags & NAPI_F_PREFER_BUSY_POLL) { 6397 napi->defer_hard_irqs_count = napi_get_defer_hard_irqs(napi); 6398 timeout = napi_get_gro_flush_timeout(napi); 6399 if (napi->defer_hard_irqs_count && timeout) { 6400 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6401 skip_schedule = true; 6402 } 6403 } 6404 6405 /* All we really want here is to re-enable device interrupts. 6406 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6407 */ 6408 rc = napi->poll(napi, budget); 6409 /* We can't gro_normal_list() here, because napi->poll() might have 6410 * rearmed the napi (napi_complete_done()) in which case it could 6411 * already be running on another CPU. 6412 */ 6413 trace_napi_poll(napi, rc, budget); 6414 netpoll_poll_unlock(have_poll_lock); 6415 if (rc == budget) 6416 __busy_poll_stop(napi, skip_schedule); 6417 bpf_net_ctx_clear(bpf_net_ctx); 6418 local_bh_enable(); 6419 } 6420 6421 static void __napi_busy_loop(unsigned int napi_id, 6422 bool (*loop_end)(void *, unsigned long), 6423 void *loop_end_arg, unsigned flags, u16 budget) 6424 { 6425 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6426 int (*napi_poll)(struct napi_struct *napi, int budget); 6427 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6428 void *have_poll_lock = NULL; 6429 struct napi_struct *napi; 6430 6431 WARN_ON_ONCE(!rcu_read_lock_held()); 6432 6433 restart: 6434 napi_poll = NULL; 6435 6436 napi = napi_by_id(napi_id); 6437 if (!napi) 6438 return; 6439 6440 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6441 preempt_disable(); 6442 for (;;) { 6443 int work = 0; 6444 6445 local_bh_disable(); 6446 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6447 if (!napi_poll) { 6448 unsigned long val = READ_ONCE(napi->state); 6449 6450 /* If multiple threads are competing for this napi, 6451 * we avoid dirtying napi->state as much as we can. 6452 */ 6453 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6454 NAPIF_STATE_IN_BUSY_POLL)) { 6455 if (flags & NAPI_F_PREFER_BUSY_POLL) 6456 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6457 goto count; 6458 } 6459 if (cmpxchg(&napi->state, val, 6460 val | NAPIF_STATE_IN_BUSY_POLL | 6461 NAPIF_STATE_SCHED) != val) { 6462 if (flags & NAPI_F_PREFER_BUSY_POLL) 6463 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6464 goto count; 6465 } 6466 have_poll_lock = netpoll_poll_lock(napi); 6467 napi_poll = napi->poll; 6468 } 6469 work = napi_poll(napi, budget); 6470 trace_napi_poll(napi, work, budget); 6471 gro_normal_list(napi); 6472 count: 6473 if (work > 0) 6474 __NET_ADD_STATS(dev_net(napi->dev), 6475 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6476 skb_defer_free_flush(this_cpu_ptr(&softnet_data)); 6477 bpf_net_ctx_clear(bpf_net_ctx); 6478 local_bh_enable(); 6479 6480 if (!loop_end || loop_end(loop_end_arg, start_time)) 6481 break; 6482 6483 if (unlikely(need_resched())) { 6484 if (flags & NAPI_F_END_ON_RESCHED) 6485 break; 6486 if (napi_poll) 6487 busy_poll_stop(napi, have_poll_lock, flags, budget); 6488 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6489 preempt_enable(); 6490 rcu_read_unlock(); 6491 cond_resched(); 6492 rcu_read_lock(); 6493 if (loop_end(loop_end_arg, start_time)) 6494 return; 6495 goto restart; 6496 } 6497 cpu_relax(); 6498 } 6499 if (napi_poll) 6500 busy_poll_stop(napi, have_poll_lock, flags, budget); 6501 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6502 preempt_enable(); 6503 } 6504 6505 void napi_busy_loop_rcu(unsigned int napi_id, 6506 bool (*loop_end)(void *, unsigned long), 6507 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6508 { 6509 unsigned flags = NAPI_F_END_ON_RESCHED; 6510 6511 if (prefer_busy_poll) 6512 flags |= NAPI_F_PREFER_BUSY_POLL; 6513 6514 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6515 } 6516 6517 void napi_busy_loop(unsigned int napi_id, 6518 bool (*loop_end)(void *, unsigned long), 6519 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6520 { 6521 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; 6522 6523 rcu_read_lock(); 6524 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6525 rcu_read_unlock(); 6526 } 6527 EXPORT_SYMBOL(napi_busy_loop); 6528 6529 void napi_suspend_irqs(unsigned int napi_id) 6530 { 6531 struct napi_struct *napi; 6532 6533 rcu_read_lock(); 6534 napi = napi_by_id(napi_id); 6535 if (napi) { 6536 unsigned long timeout = napi_get_irq_suspend_timeout(napi); 6537 6538 if (timeout) 6539 hrtimer_start(&napi->timer, ns_to_ktime(timeout), 6540 HRTIMER_MODE_REL_PINNED); 6541 } 6542 rcu_read_unlock(); 6543 } 6544 6545 void napi_resume_irqs(unsigned int napi_id) 6546 { 6547 struct napi_struct *napi; 6548 6549 rcu_read_lock(); 6550 napi = napi_by_id(napi_id); 6551 if (napi) { 6552 /* If irq_suspend_timeout is set to 0 between the call to 6553 * napi_suspend_irqs and now, the original value still 6554 * determines the safety timeout as intended and napi_watchdog 6555 * will resume irq processing. 6556 */ 6557 if (napi_get_irq_suspend_timeout(napi)) { 6558 local_bh_disable(); 6559 napi_schedule(napi); 6560 local_bh_enable(); 6561 } 6562 } 6563 rcu_read_unlock(); 6564 } 6565 6566 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6567 6568 static void __napi_hash_add_with_id(struct napi_struct *napi, 6569 unsigned int napi_id) 6570 { 6571 napi->napi_id = napi_id; 6572 hlist_add_head_rcu(&napi->napi_hash_node, 6573 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6574 } 6575 6576 static void napi_hash_add_with_id(struct napi_struct *napi, 6577 unsigned int napi_id) 6578 { 6579 unsigned long flags; 6580 6581 spin_lock_irqsave(&napi_hash_lock, flags); 6582 WARN_ON_ONCE(napi_by_id(napi_id)); 6583 __napi_hash_add_with_id(napi, napi_id); 6584 spin_unlock_irqrestore(&napi_hash_lock, flags); 6585 } 6586 6587 static void napi_hash_add(struct napi_struct *napi) 6588 { 6589 unsigned long flags; 6590 6591 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6592 return; 6593 6594 spin_lock_irqsave(&napi_hash_lock, flags); 6595 6596 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6597 do { 6598 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6599 napi_gen_id = MIN_NAPI_ID; 6600 } while (napi_by_id(napi_gen_id)); 6601 6602 __napi_hash_add_with_id(napi, napi_gen_id); 6603 6604 spin_unlock_irqrestore(&napi_hash_lock, flags); 6605 } 6606 6607 /* Warning : caller is responsible to make sure rcu grace period 6608 * is respected before freeing memory containing @napi 6609 */ 6610 static void napi_hash_del(struct napi_struct *napi) 6611 { 6612 unsigned long flags; 6613 6614 spin_lock_irqsave(&napi_hash_lock, flags); 6615 6616 hlist_del_init_rcu(&napi->napi_hash_node); 6617 6618 spin_unlock_irqrestore(&napi_hash_lock, flags); 6619 } 6620 6621 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6622 { 6623 struct napi_struct *napi; 6624 6625 napi = container_of(timer, struct napi_struct, timer); 6626 6627 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6628 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6629 */ 6630 if (!napi_disable_pending(napi) && 6631 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6632 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6633 __napi_schedule_irqoff(napi); 6634 } 6635 6636 return HRTIMER_NORESTART; 6637 } 6638 6639 static void init_gro_hash(struct napi_struct *napi) 6640 { 6641 int i; 6642 6643 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6644 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6645 napi->gro_hash[i].count = 0; 6646 } 6647 napi->gro_bitmask = 0; 6648 } 6649 6650 int dev_set_threaded(struct net_device *dev, bool threaded) 6651 { 6652 struct napi_struct *napi; 6653 int err = 0; 6654 6655 if (dev->threaded == threaded) 6656 return 0; 6657 6658 if (threaded) { 6659 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6660 if (!napi->thread) { 6661 err = napi_kthread_create(napi); 6662 if (err) { 6663 threaded = false; 6664 break; 6665 } 6666 } 6667 } 6668 } 6669 6670 WRITE_ONCE(dev->threaded, threaded); 6671 6672 /* Make sure kthread is created before THREADED bit 6673 * is set. 6674 */ 6675 smp_mb__before_atomic(); 6676 6677 /* Setting/unsetting threaded mode on a napi might not immediately 6678 * take effect, if the current napi instance is actively being 6679 * polled. In this case, the switch between threaded mode and 6680 * softirq mode will happen in the next round of napi_schedule(). 6681 * This should not cause hiccups/stalls to the live traffic. 6682 */ 6683 list_for_each_entry(napi, &dev->napi_list, dev_list) 6684 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6685 6686 return err; 6687 } 6688 EXPORT_SYMBOL(dev_set_threaded); 6689 6690 /** 6691 * netif_queue_set_napi - Associate queue with the napi 6692 * @dev: device to which NAPI and queue belong 6693 * @queue_index: Index of queue 6694 * @type: queue type as RX or TX 6695 * @napi: NAPI context, pass NULL to clear previously set NAPI 6696 * 6697 * Set queue with its corresponding napi context. This should be done after 6698 * registering the NAPI handler for the queue-vector and the queues have been 6699 * mapped to the corresponding interrupt vector. 6700 */ 6701 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6702 enum netdev_queue_type type, struct napi_struct *napi) 6703 { 6704 struct netdev_rx_queue *rxq; 6705 struct netdev_queue *txq; 6706 6707 if (WARN_ON_ONCE(napi && !napi->dev)) 6708 return; 6709 if (dev->reg_state >= NETREG_REGISTERED) 6710 ASSERT_RTNL(); 6711 6712 switch (type) { 6713 case NETDEV_QUEUE_TYPE_RX: 6714 rxq = __netif_get_rx_queue(dev, queue_index); 6715 rxq->napi = napi; 6716 return; 6717 case NETDEV_QUEUE_TYPE_TX: 6718 txq = netdev_get_tx_queue(dev, queue_index); 6719 txq->napi = napi; 6720 return; 6721 default: 6722 return; 6723 } 6724 } 6725 EXPORT_SYMBOL(netif_queue_set_napi); 6726 6727 static void napi_restore_config(struct napi_struct *n) 6728 { 6729 n->defer_hard_irqs = n->config->defer_hard_irqs; 6730 n->gro_flush_timeout = n->config->gro_flush_timeout; 6731 n->irq_suspend_timeout = n->config->irq_suspend_timeout; 6732 /* a NAPI ID might be stored in the config, if so use it. if not, use 6733 * napi_hash_add to generate one for us. It will be saved to the config 6734 * in napi_disable. 6735 */ 6736 if (n->config->napi_id) 6737 napi_hash_add_with_id(n, n->config->napi_id); 6738 else 6739 napi_hash_add(n); 6740 } 6741 6742 static void napi_save_config(struct napi_struct *n) 6743 { 6744 n->config->defer_hard_irqs = n->defer_hard_irqs; 6745 n->config->gro_flush_timeout = n->gro_flush_timeout; 6746 n->config->irq_suspend_timeout = n->irq_suspend_timeout; 6747 n->config->napi_id = n->napi_id; 6748 napi_hash_del(n); 6749 } 6750 6751 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6752 int (*poll)(struct napi_struct *, int), int weight) 6753 { 6754 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6755 return; 6756 6757 INIT_LIST_HEAD(&napi->poll_list); 6758 INIT_HLIST_NODE(&napi->napi_hash_node); 6759 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6760 napi->timer.function = napi_watchdog; 6761 init_gro_hash(napi); 6762 napi->skb = NULL; 6763 INIT_LIST_HEAD(&napi->rx_list); 6764 napi->rx_count = 0; 6765 napi->poll = poll; 6766 if (weight > NAPI_POLL_WEIGHT) 6767 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6768 weight); 6769 napi->weight = weight; 6770 napi->dev = dev; 6771 #ifdef CONFIG_NETPOLL 6772 napi->poll_owner = -1; 6773 #endif 6774 napi->list_owner = -1; 6775 set_bit(NAPI_STATE_SCHED, &napi->state); 6776 set_bit(NAPI_STATE_NPSVC, &napi->state); 6777 list_add_rcu(&napi->dev_list, &dev->napi_list); 6778 6779 /* default settings from sysfs are applied to all NAPIs. any per-NAPI 6780 * configuration will be loaded in napi_enable 6781 */ 6782 napi_set_defer_hard_irqs(napi, READ_ONCE(dev->napi_defer_hard_irqs)); 6783 napi_set_gro_flush_timeout(napi, READ_ONCE(dev->gro_flush_timeout)); 6784 6785 napi_get_frags_check(napi); 6786 /* Create kthread for this napi if dev->threaded is set. 6787 * Clear dev->threaded if kthread creation failed so that 6788 * threaded mode will not be enabled in napi_enable(). 6789 */ 6790 if (dev->threaded && napi_kthread_create(napi)) 6791 dev->threaded = false; 6792 netif_napi_set_irq(napi, -1); 6793 } 6794 EXPORT_SYMBOL(netif_napi_add_weight); 6795 6796 void napi_disable(struct napi_struct *n) 6797 { 6798 unsigned long val, new; 6799 6800 might_sleep(); 6801 set_bit(NAPI_STATE_DISABLE, &n->state); 6802 6803 val = READ_ONCE(n->state); 6804 do { 6805 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6806 usleep_range(20, 200); 6807 val = READ_ONCE(n->state); 6808 } 6809 6810 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6811 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6812 } while (!try_cmpxchg(&n->state, &val, new)); 6813 6814 hrtimer_cancel(&n->timer); 6815 6816 if (n->config) 6817 napi_save_config(n); 6818 else 6819 napi_hash_del(n); 6820 6821 clear_bit(NAPI_STATE_DISABLE, &n->state); 6822 } 6823 EXPORT_SYMBOL(napi_disable); 6824 6825 /** 6826 * napi_enable - enable NAPI scheduling 6827 * @n: NAPI context 6828 * 6829 * Resume NAPI from being scheduled on this context. 6830 * Must be paired with napi_disable. 6831 */ 6832 void napi_enable(struct napi_struct *n) 6833 { 6834 unsigned long new, val = READ_ONCE(n->state); 6835 6836 if (n->config) 6837 napi_restore_config(n); 6838 else 6839 napi_hash_add(n); 6840 6841 do { 6842 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6843 6844 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6845 if (n->dev->threaded && n->thread) 6846 new |= NAPIF_STATE_THREADED; 6847 } while (!try_cmpxchg(&n->state, &val, new)); 6848 } 6849 EXPORT_SYMBOL(napi_enable); 6850 6851 static void flush_gro_hash(struct napi_struct *napi) 6852 { 6853 int i; 6854 6855 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6856 struct sk_buff *skb, *n; 6857 6858 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6859 kfree_skb(skb); 6860 napi->gro_hash[i].count = 0; 6861 } 6862 } 6863 6864 /* Must be called in process context */ 6865 void __netif_napi_del(struct napi_struct *napi) 6866 { 6867 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6868 return; 6869 6870 if (napi->config) { 6871 napi->index = -1; 6872 napi->config = NULL; 6873 } 6874 6875 list_del_rcu(&napi->dev_list); 6876 napi_free_frags(napi); 6877 6878 flush_gro_hash(napi); 6879 napi->gro_bitmask = 0; 6880 6881 if (napi->thread) { 6882 kthread_stop(napi->thread); 6883 napi->thread = NULL; 6884 } 6885 } 6886 EXPORT_SYMBOL(__netif_napi_del); 6887 6888 static int __napi_poll(struct napi_struct *n, bool *repoll) 6889 { 6890 int work, weight; 6891 6892 weight = n->weight; 6893 6894 /* This NAPI_STATE_SCHED test is for avoiding a race 6895 * with netpoll's poll_napi(). Only the entity which 6896 * obtains the lock and sees NAPI_STATE_SCHED set will 6897 * actually make the ->poll() call. Therefore we avoid 6898 * accidentally calling ->poll() when NAPI is not scheduled. 6899 */ 6900 work = 0; 6901 if (napi_is_scheduled(n)) { 6902 work = n->poll(n, weight); 6903 trace_napi_poll(n, work, weight); 6904 6905 xdp_do_check_flushed(n); 6906 } 6907 6908 if (unlikely(work > weight)) 6909 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6910 n->poll, work, weight); 6911 6912 if (likely(work < weight)) 6913 return work; 6914 6915 /* Drivers must not modify the NAPI state if they 6916 * consume the entire weight. In such cases this code 6917 * still "owns" the NAPI instance and therefore can 6918 * move the instance around on the list at-will. 6919 */ 6920 if (unlikely(napi_disable_pending(n))) { 6921 napi_complete(n); 6922 return work; 6923 } 6924 6925 /* The NAPI context has more processing work, but busy-polling 6926 * is preferred. Exit early. 6927 */ 6928 if (napi_prefer_busy_poll(n)) { 6929 if (napi_complete_done(n, work)) { 6930 /* If timeout is not set, we need to make sure 6931 * that the NAPI is re-scheduled. 6932 */ 6933 napi_schedule(n); 6934 } 6935 return work; 6936 } 6937 6938 if (n->gro_bitmask) { 6939 /* flush too old packets 6940 * If HZ < 1000, flush all packets. 6941 */ 6942 napi_gro_flush(n, HZ >= 1000); 6943 } 6944 6945 gro_normal_list(n); 6946 6947 /* Some drivers may have called napi_schedule 6948 * prior to exhausting their budget. 6949 */ 6950 if (unlikely(!list_empty(&n->poll_list))) { 6951 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6952 n->dev ? n->dev->name : "backlog"); 6953 return work; 6954 } 6955 6956 *repoll = true; 6957 6958 return work; 6959 } 6960 6961 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6962 { 6963 bool do_repoll = false; 6964 void *have; 6965 int work; 6966 6967 list_del_init(&n->poll_list); 6968 6969 have = netpoll_poll_lock(n); 6970 6971 work = __napi_poll(n, &do_repoll); 6972 6973 if (do_repoll) 6974 list_add_tail(&n->poll_list, repoll); 6975 6976 netpoll_poll_unlock(have); 6977 6978 return work; 6979 } 6980 6981 static int napi_thread_wait(struct napi_struct *napi) 6982 { 6983 set_current_state(TASK_INTERRUPTIBLE); 6984 6985 while (!kthread_should_stop()) { 6986 /* Testing SCHED_THREADED bit here to make sure the current 6987 * kthread owns this napi and could poll on this napi. 6988 * Testing SCHED bit is not enough because SCHED bit might be 6989 * set by some other busy poll thread or by napi_disable(). 6990 */ 6991 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) { 6992 WARN_ON(!list_empty(&napi->poll_list)); 6993 __set_current_state(TASK_RUNNING); 6994 return 0; 6995 } 6996 6997 schedule(); 6998 set_current_state(TASK_INTERRUPTIBLE); 6999 } 7000 __set_current_state(TASK_RUNNING); 7001 7002 return -1; 7003 } 7004 7005 static void napi_threaded_poll_loop(struct napi_struct *napi) 7006 { 7007 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7008 struct softnet_data *sd; 7009 unsigned long last_qs = jiffies; 7010 7011 for (;;) { 7012 bool repoll = false; 7013 void *have; 7014 7015 local_bh_disable(); 7016 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7017 7018 sd = this_cpu_ptr(&softnet_data); 7019 sd->in_napi_threaded_poll = true; 7020 7021 have = netpoll_poll_lock(napi); 7022 __napi_poll(napi, &repoll); 7023 netpoll_poll_unlock(have); 7024 7025 sd->in_napi_threaded_poll = false; 7026 barrier(); 7027 7028 if (sd_has_rps_ipi_waiting(sd)) { 7029 local_irq_disable(); 7030 net_rps_action_and_irq_enable(sd); 7031 } 7032 skb_defer_free_flush(sd); 7033 bpf_net_ctx_clear(bpf_net_ctx); 7034 local_bh_enable(); 7035 7036 if (!repoll) 7037 break; 7038 7039 rcu_softirq_qs_periodic(last_qs); 7040 cond_resched(); 7041 } 7042 } 7043 7044 static int napi_threaded_poll(void *data) 7045 { 7046 struct napi_struct *napi = data; 7047 7048 while (!napi_thread_wait(napi)) 7049 napi_threaded_poll_loop(napi); 7050 7051 return 0; 7052 } 7053 7054 static __latent_entropy void net_rx_action(void) 7055 { 7056 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7057 unsigned long time_limit = jiffies + 7058 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs)); 7059 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7060 int budget = READ_ONCE(net_hotdata.netdev_budget); 7061 LIST_HEAD(list); 7062 LIST_HEAD(repoll); 7063 7064 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7065 start: 7066 sd->in_net_rx_action = true; 7067 local_irq_disable(); 7068 list_splice_init(&sd->poll_list, &list); 7069 local_irq_enable(); 7070 7071 for (;;) { 7072 struct napi_struct *n; 7073 7074 skb_defer_free_flush(sd); 7075 7076 if (list_empty(&list)) { 7077 if (list_empty(&repoll)) { 7078 sd->in_net_rx_action = false; 7079 barrier(); 7080 /* We need to check if ____napi_schedule() 7081 * had refilled poll_list while 7082 * sd->in_net_rx_action was true. 7083 */ 7084 if (!list_empty(&sd->poll_list)) 7085 goto start; 7086 if (!sd_has_rps_ipi_waiting(sd)) 7087 goto end; 7088 } 7089 break; 7090 } 7091 7092 n = list_first_entry(&list, struct napi_struct, poll_list); 7093 budget -= napi_poll(n, &repoll); 7094 7095 /* If softirq window is exhausted then punt. 7096 * Allow this to run for 2 jiffies since which will allow 7097 * an average latency of 1.5/HZ. 7098 */ 7099 if (unlikely(budget <= 0 || 7100 time_after_eq(jiffies, time_limit))) { 7101 sd->time_squeeze++; 7102 break; 7103 } 7104 } 7105 7106 local_irq_disable(); 7107 7108 list_splice_tail_init(&sd->poll_list, &list); 7109 list_splice_tail(&repoll, &list); 7110 list_splice(&list, &sd->poll_list); 7111 if (!list_empty(&sd->poll_list)) 7112 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7113 else 7114 sd->in_net_rx_action = false; 7115 7116 net_rps_action_and_irq_enable(sd); 7117 end: 7118 bpf_net_ctx_clear(bpf_net_ctx); 7119 } 7120 7121 struct netdev_adjacent { 7122 struct net_device *dev; 7123 netdevice_tracker dev_tracker; 7124 7125 /* upper master flag, there can only be one master device per list */ 7126 bool master; 7127 7128 /* lookup ignore flag */ 7129 bool ignore; 7130 7131 /* counter for the number of times this device was added to us */ 7132 u16 ref_nr; 7133 7134 /* private field for the users */ 7135 void *private; 7136 7137 struct list_head list; 7138 struct rcu_head rcu; 7139 }; 7140 7141 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7142 struct list_head *adj_list) 7143 { 7144 struct netdev_adjacent *adj; 7145 7146 list_for_each_entry(adj, adj_list, list) { 7147 if (adj->dev == adj_dev) 7148 return adj; 7149 } 7150 return NULL; 7151 } 7152 7153 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7154 struct netdev_nested_priv *priv) 7155 { 7156 struct net_device *dev = (struct net_device *)priv->data; 7157 7158 return upper_dev == dev; 7159 } 7160 7161 /** 7162 * netdev_has_upper_dev - Check if device is linked to an upper device 7163 * @dev: device 7164 * @upper_dev: upper device to check 7165 * 7166 * Find out if a device is linked to specified upper device and return true 7167 * in case it is. Note that this checks only immediate upper device, 7168 * not through a complete stack of devices. The caller must hold the RTNL lock. 7169 */ 7170 bool netdev_has_upper_dev(struct net_device *dev, 7171 struct net_device *upper_dev) 7172 { 7173 struct netdev_nested_priv priv = { 7174 .data = (void *)upper_dev, 7175 }; 7176 7177 ASSERT_RTNL(); 7178 7179 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7180 &priv); 7181 } 7182 EXPORT_SYMBOL(netdev_has_upper_dev); 7183 7184 /** 7185 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7186 * @dev: device 7187 * @upper_dev: upper device to check 7188 * 7189 * Find out if a device is linked to specified upper device and return true 7190 * in case it is. Note that this checks the entire upper device chain. 7191 * The caller must hold rcu lock. 7192 */ 7193 7194 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7195 struct net_device *upper_dev) 7196 { 7197 struct netdev_nested_priv priv = { 7198 .data = (void *)upper_dev, 7199 }; 7200 7201 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7202 &priv); 7203 } 7204 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7205 7206 /** 7207 * netdev_has_any_upper_dev - Check if device is linked to some device 7208 * @dev: device 7209 * 7210 * Find out if a device is linked to an upper device and return true in case 7211 * it is. The caller must hold the RTNL lock. 7212 */ 7213 bool netdev_has_any_upper_dev(struct net_device *dev) 7214 { 7215 ASSERT_RTNL(); 7216 7217 return !list_empty(&dev->adj_list.upper); 7218 } 7219 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7220 7221 /** 7222 * netdev_master_upper_dev_get - Get master upper device 7223 * @dev: device 7224 * 7225 * Find a master upper device and return pointer to it or NULL in case 7226 * it's not there. The caller must hold the RTNL lock. 7227 */ 7228 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7229 { 7230 struct netdev_adjacent *upper; 7231 7232 ASSERT_RTNL(); 7233 7234 if (list_empty(&dev->adj_list.upper)) 7235 return NULL; 7236 7237 upper = list_first_entry(&dev->adj_list.upper, 7238 struct netdev_adjacent, list); 7239 if (likely(upper->master)) 7240 return upper->dev; 7241 return NULL; 7242 } 7243 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7244 7245 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7246 { 7247 struct netdev_adjacent *upper; 7248 7249 ASSERT_RTNL(); 7250 7251 if (list_empty(&dev->adj_list.upper)) 7252 return NULL; 7253 7254 upper = list_first_entry(&dev->adj_list.upper, 7255 struct netdev_adjacent, list); 7256 if (likely(upper->master) && !upper->ignore) 7257 return upper->dev; 7258 return NULL; 7259 } 7260 7261 /** 7262 * netdev_has_any_lower_dev - Check if device is linked to some device 7263 * @dev: device 7264 * 7265 * Find out if a device is linked to a lower device and return true in case 7266 * it is. The caller must hold the RTNL lock. 7267 */ 7268 static bool netdev_has_any_lower_dev(struct net_device *dev) 7269 { 7270 ASSERT_RTNL(); 7271 7272 return !list_empty(&dev->adj_list.lower); 7273 } 7274 7275 void *netdev_adjacent_get_private(struct list_head *adj_list) 7276 { 7277 struct netdev_adjacent *adj; 7278 7279 adj = list_entry(adj_list, struct netdev_adjacent, list); 7280 7281 return adj->private; 7282 } 7283 EXPORT_SYMBOL(netdev_adjacent_get_private); 7284 7285 /** 7286 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7287 * @dev: device 7288 * @iter: list_head ** of the current position 7289 * 7290 * Gets the next device from the dev's upper list, starting from iter 7291 * position. The caller must hold RCU read lock. 7292 */ 7293 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7294 struct list_head **iter) 7295 { 7296 struct netdev_adjacent *upper; 7297 7298 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7299 7300 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7301 7302 if (&upper->list == &dev->adj_list.upper) 7303 return NULL; 7304 7305 *iter = &upper->list; 7306 7307 return upper->dev; 7308 } 7309 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7310 7311 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7312 struct list_head **iter, 7313 bool *ignore) 7314 { 7315 struct netdev_adjacent *upper; 7316 7317 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7318 7319 if (&upper->list == &dev->adj_list.upper) 7320 return NULL; 7321 7322 *iter = &upper->list; 7323 *ignore = upper->ignore; 7324 7325 return upper->dev; 7326 } 7327 7328 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7329 struct list_head **iter) 7330 { 7331 struct netdev_adjacent *upper; 7332 7333 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7334 7335 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7336 7337 if (&upper->list == &dev->adj_list.upper) 7338 return NULL; 7339 7340 *iter = &upper->list; 7341 7342 return upper->dev; 7343 } 7344 7345 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7346 int (*fn)(struct net_device *dev, 7347 struct netdev_nested_priv *priv), 7348 struct netdev_nested_priv *priv) 7349 { 7350 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7351 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7352 int ret, cur = 0; 7353 bool ignore; 7354 7355 now = dev; 7356 iter = &dev->adj_list.upper; 7357 7358 while (1) { 7359 if (now != dev) { 7360 ret = fn(now, priv); 7361 if (ret) 7362 return ret; 7363 } 7364 7365 next = NULL; 7366 while (1) { 7367 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7368 if (!udev) 7369 break; 7370 if (ignore) 7371 continue; 7372 7373 next = udev; 7374 niter = &udev->adj_list.upper; 7375 dev_stack[cur] = now; 7376 iter_stack[cur++] = iter; 7377 break; 7378 } 7379 7380 if (!next) { 7381 if (!cur) 7382 return 0; 7383 next = dev_stack[--cur]; 7384 niter = iter_stack[cur]; 7385 } 7386 7387 now = next; 7388 iter = niter; 7389 } 7390 7391 return 0; 7392 } 7393 7394 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7395 int (*fn)(struct net_device *dev, 7396 struct netdev_nested_priv *priv), 7397 struct netdev_nested_priv *priv) 7398 { 7399 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7400 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7401 int ret, cur = 0; 7402 7403 now = dev; 7404 iter = &dev->adj_list.upper; 7405 7406 while (1) { 7407 if (now != dev) { 7408 ret = fn(now, priv); 7409 if (ret) 7410 return ret; 7411 } 7412 7413 next = NULL; 7414 while (1) { 7415 udev = netdev_next_upper_dev_rcu(now, &iter); 7416 if (!udev) 7417 break; 7418 7419 next = udev; 7420 niter = &udev->adj_list.upper; 7421 dev_stack[cur] = now; 7422 iter_stack[cur++] = iter; 7423 break; 7424 } 7425 7426 if (!next) { 7427 if (!cur) 7428 return 0; 7429 next = dev_stack[--cur]; 7430 niter = iter_stack[cur]; 7431 } 7432 7433 now = next; 7434 iter = niter; 7435 } 7436 7437 return 0; 7438 } 7439 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7440 7441 static bool __netdev_has_upper_dev(struct net_device *dev, 7442 struct net_device *upper_dev) 7443 { 7444 struct netdev_nested_priv priv = { 7445 .flags = 0, 7446 .data = (void *)upper_dev, 7447 }; 7448 7449 ASSERT_RTNL(); 7450 7451 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7452 &priv); 7453 } 7454 7455 /** 7456 * netdev_lower_get_next_private - Get the next ->private from the 7457 * lower neighbour list 7458 * @dev: device 7459 * @iter: list_head ** of the current position 7460 * 7461 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7462 * list, starting from iter position. The caller must hold either hold the 7463 * RTNL lock or its own locking that guarantees that the neighbour lower 7464 * list will remain unchanged. 7465 */ 7466 void *netdev_lower_get_next_private(struct net_device *dev, 7467 struct list_head **iter) 7468 { 7469 struct netdev_adjacent *lower; 7470 7471 lower = list_entry(*iter, struct netdev_adjacent, list); 7472 7473 if (&lower->list == &dev->adj_list.lower) 7474 return NULL; 7475 7476 *iter = lower->list.next; 7477 7478 return lower->private; 7479 } 7480 EXPORT_SYMBOL(netdev_lower_get_next_private); 7481 7482 /** 7483 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7484 * lower neighbour list, RCU 7485 * variant 7486 * @dev: device 7487 * @iter: list_head ** of the current position 7488 * 7489 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7490 * list, starting from iter position. The caller must hold RCU read lock. 7491 */ 7492 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7493 struct list_head **iter) 7494 { 7495 struct netdev_adjacent *lower; 7496 7497 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7498 7499 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7500 7501 if (&lower->list == &dev->adj_list.lower) 7502 return NULL; 7503 7504 *iter = &lower->list; 7505 7506 return lower->private; 7507 } 7508 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7509 7510 /** 7511 * netdev_lower_get_next - Get the next device from the lower neighbour 7512 * list 7513 * @dev: device 7514 * @iter: list_head ** of the current position 7515 * 7516 * Gets the next netdev_adjacent from the dev's lower neighbour 7517 * list, starting from iter position. The caller must hold RTNL lock or 7518 * its own locking that guarantees that the neighbour lower 7519 * list will remain unchanged. 7520 */ 7521 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7522 { 7523 struct netdev_adjacent *lower; 7524 7525 lower = list_entry(*iter, struct netdev_adjacent, list); 7526 7527 if (&lower->list == &dev->adj_list.lower) 7528 return NULL; 7529 7530 *iter = lower->list.next; 7531 7532 return lower->dev; 7533 } 7534 EXPORT_SYMBOL(netdev_lower_get_next); 7535 7536 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7537 struct list_head **iter) 7538 { 7539 struct netdev_adjacent *lower; 7540 7541 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7542 7543 if (&lower->list == &dev->adj_list.lower) 7544 return NULL; 7545 7546 *iter = &lower->list; 7547 7548 return lower->dev; 7549 } 7550 7551 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7552 struct list_head **iter, 7553 bool *ignore) 7554 { 7555 struct netdev_adjacent *lower; 7556 7557 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7558 7559 if (&lower->list == &dev->adj_list.lower) 7560 return NULL; 7561 7562 *iter = &lower->list; 7563 *ignore = lower->ignore; 7564 7565 return lower->dev; 7566 } 7567 7568 int netdev_walk_all_lower_dev(struct net_device *dev, 7569 int (*fn)(struct net_device *dev, 7570 struct netdev_nested_priv *priv), 7571 struct netdev_nested_priv *priv) 7572 { 7573 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7574 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7575 int ret, cur = 0; 7576 7577 now = dev; 7578 iter = &dev->adj_list.lower; 7579 7580 while (1) { 7581 if (now != dev) { 7582 ret = fn(now, priv); 7583 if (ret) 7584 return ret; 7585 } 7586 7587 next = NULL; 7588 while (1) { 7589 ldev = netdev_next_lower_dev(now, &iter); 7590 if (!ldev) 7591 break; 7592 7593 next = ldev; 7594 niter = &ldev->adj_list.lower; 7595 dev_stack[cur] = now; 7596 iter_stack[cur++] = iter; 7597 break; 7598 } 7599 7600 if (!next) { 7601 if (!cur) 7602 return 0; 7603 next = dev_stack[--cur]; 7604 niter = iter_stack[cur]; 7605 } 7606 7607 now = next; 7608 iter = niter; 7609 } 7610 7611 return 0; 7612 } 7613 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7614 7615 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7616 int (*fn)(struct net_device *dev, 7617 struct netdev_nested_priv *priv), 7618 struct netdev_nested_priv *priv) 7619 { 7620 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7621 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7622 int ret, cur = 0; 7623 bool ignore; 7624 7625 now = dev; 7626 iter = &dev->adj_list.lower; 7627 7628 while (1) { 7629 if (now != dev) { 7630 ret = fn(now, priv); 7631 if (ret) 7632 return ret; 7633 } 7634 7635 next = NULL; 7636 while (1) { 7637 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7638 if (!ldev) 7639 break; 7640 if (ignore) 7641 continue; 7642 7643 next = ldev; 7644 niter = &ldev->adj_list.lower; 7645 dev_stack[cur] = now; 7646 iter_stack[cur++] = iter; 7647 break; 7648 } 7649 7650 if (!next) { 7651 if (!cur) 7652 return 0; 7653 next = dev_stack[--cur]; 7654 niter = iter_stack[cur]; 7655 } 7656 7657 now = next; 7658 iter = niter; 7659 } 7660 7661 return 0; 7662 } 7663 7664 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7665 struct list_head **iter) 7666 { 7667 struct netdev_adjacent *lower; 7668 7669 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7670 if (&lower->list == &dev->adj_list.lower) 7671 return NULL; 7672 7673 *iter = &lower->list; 7674 7675 return lower->dev; 7676 } 7677 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7678 7679 static u8 __netdev_upper_depth(struct net_device *dev) 7680 { 7681 struct net_device *udev; 7682 struct list_head *iter; 7683 u8 max_depth = 0; 7684 bool ignore; 7685 7686 for (iter = &dev->adj_list.upper, 7687 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7688 udev; 7689 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7690 if (ignore) 7691 continue; 7692 if (max_depth < udev->upper_level) 7693 max_depth = udev->upper_level; 7694 } 7695 7696 return max_depth; 7697 } 7698 7699 static u8 __netdev_lower_depth(struct net_device *dev) 7700 { 7701 struct net_device *ldev; 7702 struct list_head *iter; 7703 u8 max_depth = 0; 7704 bool ignore; 7705 7706 for (iter = &dev->adj_list.lower, 7707 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7708 ldev; 7709 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7710 if (ignore) 7711 continue; 7712 if (max_depth < ldev->lower_level) 7713 max_depth = ldev->lower_level; 7714 } 7715 7716 return max_depth; 7717 } 7718 7719 static int __netdev_update_upper_level(struct net_device *dev, 7720 struct netdev_nested_priv *__unused) 7721 { 7722 dev->upper_level = __netdev_upper_depth(dev) + 1; 7723 return 0; 7724 } 7725 7726 #ifdef CONFIG_LOCKDEP 7727 static LIST_HEAD(net_unlink_list); 7728 7729 static void net_unlink_todo(struct net_device *dev) 7730 { 7731 if (list_empty(&dev->unlink_list)) 7732 list_add_tail(&dev->unlink_list, &net_unlink_list); 7733 } 7734 #endif 7735 7736 static int __netdev_update_lower_level(struct net_device *dev, 7737 struct netdev_nested_priv *priv) 7738 { 7739 dev->lower_level = __netdev_lower_depth(dev) + 1; 7740 7741 #ifdef CONFIG_LOCKDEP 7742 if (!priv) 7743 return 0; 7744 7745 if (priv->flags & NESTED_SYNC_IMM) 7746 dev->nested_level = dev->lower_level - 1; 7747 if (priv->flags & NESTED_SYNC_TODO) 7748 net_unlink_todo(dev); 7749 #endif 7750 return 0; 7751 } 7752 7753 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7754 int (*fn)(struct net_device *dev, 7755 struct netdev_nested_priv *priv), 7756 struct netdev_nested_priv *priv) 7757 { 7758 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7759 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7760 int ret, cur = 0; 7761 7762 now = dev; 7763 iter = &dev->adj_list.lower; 7764 7765 while (1) { 7766 if (now != dev) { 7767 ret = fn(now, priv); 7768 if (ret) 7769 return ret; 7770 } 7771 7772 next = NULL; 7773 while (1) { 7774 ldev = netdev_next_lower_dev_rcu(now, &iter); 7775 if (!ldev) 7776 break; 7777 7778 next = ldev; 7779 niter = &ldev->adj_list.lower; 7780 dev_stack[cur] = now; 7781 iter_stack[cur++] = iter; 7782 break; 7783 } 7784 7785 if (!next) { 7786 if (!cur) 7787 return 0; 7788 next = dev_stack[--cur]; 7789 niter = iter_stack[cur]; 7790 } 7791 7792 now = next; 7793 iter = niter; 7794 } 7795 7796 return 0; 7797 } 7798 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7799 7800 /** 7801 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7802 * lower neighbour list, RCU 7803 * variant 7804 * @dev: device 7805 * 7806 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7807 * list. The caller must hold RCU read lock. 7808 */ 7809 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7810 { 7811 struct netdev_adjacent *lower; 7812 7813 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7814 struct netdev_adjacent, list); 7815 if (lower) 7816 return lower->private; 7817 return NULL; 7818 } 7819 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7820 7821 /** 7822 * netdev_master_upper_dev_get_rcu - Get master upper device 7823 * @dev: device 7824 * 7825 * Find a master upper device and return pointer to it or NULL in case 7826 * it's not there. The caller must hold the RCU read lock. 7827 */ 7828 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7829 { 7830 struct netdev_adjacent *upper; 7831 7832 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7833 struct netdev_adjacent, list); 7834 if (upper && likely(upper->master)) 7835 return upper->dev; 7836 return NULL; 7837 } 7838 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7839 7840 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7841 struct net_device *adj_dev, 7842 struct list_head *dev_list) 7843 { 7844 char linkname[IFNAMSIZ+7]; 7845 7846 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7847 "upper_%s" : "lower_%s", adj_dev->name); 7848 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7849 linkname); 7850 } 7851 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7852 char *name, 7853 struct list_head *dev_list) 7854 { 7855 char linkname[IFNAMSIZ+7]; 7856 7857 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7858 "upper_%s" : "lower_%s", name); 7859 sysfs_remove_link(&(dev->dev.kobj), linkname); 7860 } 7861 7862 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7863 struct net_device *adj_dev, 7864 struct list_head *dev_list) 7865 { 7866 return (dev_list == &dev->adj_list.upper || 7867 dev_list == &dev->adj_list.lower) && 7868 net_eq(dev_net(dev), dev_net(adj_dev)); 7869 } 7870 7871 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7872 struct net_device *adj_dev, 7873 struct list_head *dev_list, 7874 void *private, bool master) 7875 { 7876 struct netdev_adjacent *adj; 7877 int ret; 7878 7879 adj = __netdev_find_adj(adj_dev, dev_list); 7880 7881 if (adj) { 7882 adj->ref_nr += 1; 7883 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7884 dev->name, adj_dev->name, adj->ref_nr); 7885 7886 return 0; 7887 } 7888 7889 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7890 if (!adj) 7891 return -ENOMEM; 7892 7893 adj->dev = adj_dev; 7894 adj->master = master; 7895 adj->ref_nr = 1; 7896 adj->private = private; 7897 adj->ignore = false; 7898 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7899 7900 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7901 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7902 7903 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7904 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7905 if (ret) 7906 goto free_adj; 7907 } 7908 7909 /* Ensure that master link is always the first item in list. */ 7910 if (master) { 7911 ret = sysfs_create_link(&(dev->dev.kobj), 7912 &(adj_dev->dev.kobj), "master"); 7913 if (ret) 7914 goto remove_symlinks; 7915 7916 list_add_rcu(&adj->list, dev_list); 7917 } else { 7918 list_add_tail_rcu(&adj->list, dev_list); 7919 } 7920 7921 return 0; 7922 7923 remove_symlinks: 7924 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7925 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7926 free_adj: 7927 netdev_put(adj_dev, &adj->dev_tracker); 7928 kfree(adj); 7929 7930 return ret; 7931 } 7932 7933 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7934 struct net_device *adj_dev, 7935 u16 ref_nr, 7936 struct list_head *dev_list) 7937 { 7938 struct netdev_adjacent *adj; 7939 7940 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7941 dev->name, adj_dev->name, ref_nr); 7942 7943 adj = __netdev_find_adj(adj_dev, dev_list); 7944 7945 if (!adj) { 7946 pr_err("Adjacency does not exist for device %s from %s\n", 7947 dev->name, adj_dev->name); 7948 WARN_ON(1); 7949 return; 7950 } 7951 7952 if (adj->ref_nr > ref_nr) { 7953 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7954 dev->name, adj_dev->name, ref_nr, 7955 adj->ref_nr - ref_nr); 7956 adj->ref_nr -= ref_nr; 7957 return; 7958 } 7959 7960 if (adj->master) 7961 sysfs_remove_link(&(dev->dev.kobj), "master"); 7962 7963 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7964 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7965 7966 list_del_rcu(&adj->list); 7967 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7968 adj_dev->name, dev->name, adj_dev->name); 7969 netdev_put(adj_dev, &adj->dev_tracker); 7970 kfree_rcu(adj, rcu); 7971 } 7972 7973 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7974 struct net_device *upper_dev, 7975 struct list_head *up_list, 7976 struct list_head *down_list, 7977 void *private, bool master) 7978 { 7979 int ret; 7980 7981 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7982 private, master); 7983 if (ret) 7984 return ret; 7985 7986 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7987 private, false); 7988 if (ret) { 7989 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7990 return ret; 7991 } 7992 7993 return 0; 7994 } 7995 7996 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7997 struct net_device *upper_dev, 7998 u16 ref_nr, 7999 struct list_head *up_list, 8000 struct list_head *down_list) 8001 { 8002 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8003 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8004 } 8005 8006 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8007 struct net_device *upper_dev, 8008 void *private, bool master) 8009 { 8010 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8011 &dev->adj_list.upper, 8012 &upper_dev->adj_list.lower, 8013 private, master); 8014 } 8015 8016 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8017 struct net_device *upper_dev) 8018 { 8019 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8020 &dev->adj_list.upper, 8021 &upper_dev->adj_list.lower); 8022 } 8023 8024 static int __netdev_upper_dev_link(struct net_device *dev, 8025 struct net_device *upper_dev, bool master, 8026 void *upper_priv, void *upper_info, 8027 struct netdev_nested_priv *priv, 8028 struct netlink_ext_ack *extack) 8029 { 8030 struct netdev_notifier_changeupper_info changeupper_info = { 8031 .info = { 8032 .dev = dev, 8033 .extack = extack, 8034 }, 8035 .upper_dev = upper_dev, 8036 .master = master, 8037 .linking = true, 8038 .upper_info = upper_info, 8039 }; 8040 struct net_device *master_dev; 8041 int ret = 0; 8042 8043 ASSERT_RTNL(); 8044 8045 if (dev == upper_dev) 8046 return -EBUSY; 8047 8048 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8049 if (__netdev_has_upper_dev(upper_dev, dev)) 8050 return -EBUSY; 8051 8052 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8053 return -EMLINK; 8054 8055 if (!master) { 8056 if (__netdev_has_upper_dev(dev, upper_dev)) 8057 return -EEXIST; 8058 } else { 8059 master_dev = __netdev_master_upper_dev_get(dev); 8060 if (master_dev) 8061 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8062 } 8063 8064 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8065 &changeupper_info.info); 8066 ret = notifier_to_errno(ret); 8067 if (ret) 8068 return ret; 8069 8070 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8071 master); 8072 if (ret) 8073 return ret; 8074 8075 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8076 &changeupper_info.info); 8077 ret = notifier_to_errno(ret); 8078 if (ret) 8079 goto rollback; 8080 8081 __netdev_update_upper_level(dev, NULL); 8082 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8083 8084 __netdev_update_lower_level(upper_dev, priv); 8085 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8086 priv); 8087 8088 return 0; 8089 8090 rollback: 8091 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8092 8093 return ret; 8094 } 8095 8096 /** 8097 * netdev_upper_dev_link - Add a link to the upper device 8098 * @dev: device 8099 * @upper_dev: new upper device 8100 * @extack: netlink extended ack 8101 * 8102 * Adds a link to device which is upper to this one. The caller must hold 8103 * the RTNL lock. On a failure a negative errno code is returned. 8104 * On success the reference counts are adjusted and the function 8105 * returns zero. 8106 */ 8107 int netdev_upper_dev_link(struct net_device *dev, 8108 struct net_device *upper_dev, 8109 struct netlink_ext_ack *extack) 8110 { 8111 struct netdev_nested_priv priv = { 8112 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8113 .data = NULL, 8114 }; 8115 8116 return __netdev_upper_dev_link(dev, upper_dev, false, 8117 NULL, NULL, &priv, extack); 8118 } 8119 EXPORT_SYMBOL(netdev_upper_dev_link); 8120 8121 /** 8122 * netdev_master_upper_dev_link - Add a master link to the upper device 8123 * @dev: device 8124 * @upper_dev: new upper device 8125 * @upper_priv: upper device private 8126 * @upper_info: upper info to be passed down via notifier 8127 * @extack: netlink extended ack 8128 * 8129 * Adds a link to device which is upper to this one. In this case, only 8130 * one master upper device can be linked, although other non-master devices 8131 * might be linked as well. The caller must hold the RTNL lock. 8132 * On a failure a negative errno code is returned. On success the reference 8133 * counts are adjusted and the function returns zero. 8134 */ 8135 int netdev_master_upper_dev_link(struct net_device *dev, 8136 struct net_device *upper_dev, 8137 void *upper_priv, void *upper_info, 8138 struct netlink_ext_ack *extack) 8139 { 8140 struct netdev_nested_priv priv = { 8141 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8142 .data = NULL, 8143 }; 8144 8145 return __netdev_upper_dev_link(dev, upper_dev, true, 8146 upper_priv, upper_info, &priv, extack); 8147 } 8148 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8149 8150 static void __netdev_upper_dev_unlink(struct net_device *dev, 8151 struct net_device *upper_dev, 8152 struct netdev_nested_priv *priv) 8153 { 8154 struct netdev_notifier_changeupper_info changeupper_info = { 8155 .info = { 8156 .dev = dev, 8157 }, 8158 .upper_dev = upper_dev, 8159 .linking = false, 8160 }; 8161 8162 ASSERT_RTNL(); 8163 8164 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8165 8166 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8167 &changeupper_info.info); 8168 8169 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8170 8171 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8172 &changeupper_info.info); 8173 8174 __netdev_update_upper_level(dev, NULL); 8175 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8176 8177 __netdev_update_lower_level(upper_dev, priv); 8178 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8179 priv); 8180 } 8181 8182 /** 8183 * netdev_upper_dev_unlink - Removes a link to upper device 8184 * @dev: device 8185 * @upper_dev: new upper device 8186 * 8187 * Removes a link to device which is upper to this one. The caller must hold 8188 * the RTNL lock. 8189 */ 8190 void netdev_upper_dev_unlink(struct net_device *dev, 8191 struct net_device *upper_dev) 8192 { 8193 struct netdev_nested_priv priv = { 8194 .flags = NESTED_SYNC_TODO, 8195 .data = NULL, 8196 }; 8197 8198 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8199 } 8200 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8201 8202 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8203 struct net_device *lower_dev, 8204 bool val) 8205 { 8206 struct netdev_adjacent *adj; 8207 8208 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8209 if (adj) 8210 adj->ignore = val; 8211 8212 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8213 if (adj) 8214 adj->ignore = val; 8215 } 8216 8217 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8218 struct net_device *lower_dev) 8219 { 8220 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8221 } 8222 8223 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8224 struct net_device *lower_dev) 8225 { 8226 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8227 } 8228 8229 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8230 struct net_device *new_dev, 8231 struct net_device *dev, 8232 struct netlink_ext_ack *extack) 8233 { 8234 struct netdev_nested_priv priv = { 8235 .flags = 0, 8236 .data = NULL, 8237 }; 8238 int err; 8239 8240 if (!new_dev) 8241 return 0; 8242 8243 if (old_dev && new_dev != old_dev) 8244 netdev_adjacent_dev_disable(dev, old_dev); 8245 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8246 extack); 8247 if (err) { 8248 if (old_dev && new_dev != old_dev) 8249 netdev_adjacent_dev_enable(dev, old_dev); 8250 return err; 8251 } 8252 8253 return 0; 8254 } 8255 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8256 8257 void netdev_adjacent_change_commit(struct net_device *old_dev, 8258 struct net_device *new_dev, 8259 struct net_device *dev) 8260 { 8261 struct netdev_nested_priv priv = { 8262 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8263 .data = NULL, 8264 }; 8265 8266 if (!new_dev || !old_dev) 8267 return; 8268 8269 if (new_dev == old_dev) 8270 return; 8271 8272 netdev_adjacent_dev_enable(dev, old_dev); 8273 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8274 } 8275 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8276 8277 void netdev_adjacent_change_abort(struct net_device *old_dev, 8278 struct net_device *new_dev, 8279 struct net_device *dev) 8280 { 8281 struct netdev_nested_priv priv = { 8282 .flags = 0, 8283 .data = NULL, 8284 }; 8285 8286 if (!new_dev) 8287 return; 8288 8289 if (old_dev && new_dev != old_dev) 8290 netdev_adjacent_dev_enable(dev, old_dev); 8291 8292 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8293 } 8294 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8295 8296 /** 8297 * netdev_bonding_info_change - Dispatch event about slave change 8298 * @dev: device 8299 * @bonding_info: info to dispatch 8300 * 8301 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8302 * The caller must hold the RTNL lock. 8303 */ 8304 void netdev_bonding_info_change(struct net_device *dev, 8305 struct netdev_bonding_info *bonding_info) 8306 { 8307 struct netdev_notifier_bonding_info info = { 8308 .info.dev = dev, 8309 }; 8310 8311 memcpy(&info.bonding_info, bonding_info, 8312 sizeof(struct netdev_bonding_info)); 8313 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8314 &info.info); 8315 } 8316 EXPORT_SYMBOL(netdev_bonding_info_change); 8317 8318 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8319 struct netlink_ext_ack *extack) 8320 { 8321 struct netdev_notifier_offload_xstats_info info = { 8322 .info.dev = dev, 8323 .info.extack = extack, 8324 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8325 }; 8326 int err; 8327 int rc; 8328 8329 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8330 GFP_KERNEL); 8331 if (!dev->offload_xstats_l3) 8332 return -ENOMEM; 8333 8334 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8335 NETDEV_OFFLOAD_XSTATS_DISABLE, 8336 &info.info); 8337 err = notifier_to_errno(rc); 8338 if (err) 8339 goto free_stats; 8340 8341 return 0; 8342 8343 free_stats: 8344 kfree(dev->offload_xstats_l3); 8345 dev->offload_xstats_l3 = NULL; 8346 return err; 8347 } 8348 8349 int netdev_offload_xstats_enable(struct net_device *dev, 8350 enum netdev_offload_xstats_type type, 8351 struct netlink_ext_ack *extack) 8352 { 8353 ASSERT_RTNL(); 8354 8355 if (netdev_offload_xstats_enabled(dev, type)) 8356 return -EALREADY; 8357 8358 switch (type) { 8359 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8360 return netdev_offload_xstats_enable_l3(dev, extack); 8361 } 8362 8363 WARN_ON(1); 8364 return -EINVAL; 8365 } 8366 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8367 8368 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8369 { 8370 struct netdev_notifier_offload_xstats_info info = { 8371 .info.dev = dev, 8372 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8373 }; 8374 8375 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8376 &info.info); 8377 kfree(dev->offload_xstats_l3); 8378 dev->offload_xstats_l3 = NULL; 8379 } 8380 8381 int netdev_offload_xstats_disable(struct net_device *dev, 8382 enum netdev_offload_xstats_type type) 8383 { 8384 ASSERT_RTNL(); 8385 8386 if (!netdev_offload_xstats_enabled(dev, type)) 8387 return -EALREADY; 8388 8389 switch (type) { 8390 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8391 netdev_offload_xstats_disable_l3(dev); 8392 return 0; 8393 } 8394 8395 WARN_ON(1); 8396 return -EINVAL; 8397 } 8398 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8399 8400 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8401 { 8402 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8403 } 8404 8405 static struct rtnl_hw_stats64 * 8406 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8407 enum netdev_offload_xstats_type type) 8408 { 8409 switch (type) { 8410 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8411 return dev->offload_xstats_l3; 8412 } 8413 8414 WARN_ON(1); 8415 return NULL; 8416 } 8417 8418 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8419 enum netdev_offload_xstats_type type) 8420 { 8421 ASSERT_RTNL(); 8422 8423 return netdev_offload_xstats_get_ptr(dev, type); 8424 } 8425 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8426 8427 struct netdev_notifier_offload_xstats_ru { 8428 bool used; 8429 }; 8430 8431 struct netdev_notifier_offload_xstats_rd { 8432 struct rtnl_hw_stats64 stats; 8433 bool used; 8434 }; 8435 8436 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8437 const struct rtnl_hw_stats64 *src) 8438 { 8439 dest->rx_packets += src->rx_packets; 8440 dest->tx_packets += src->tx_packets; 8441 dest->rx_bytes += src->rx_bytes; 8442 dest->tx_bytes += src->tx_bytes; 8443 dest->rx_errors += src->rx_errors; 8444 dest->tx_errors += src->tx_errors; 8445 dest->rx_dropped += src->rx_dropped; 8446 dest->tx_dropped += src->tx_dropped; 8447 dest->multicast += src->multicast; 8448 } 8449 8450 static int netdev_offload_xstats_get_used(struct net_device *dev, 8451 enum netdev_offload_xstats_type type, 8452 bool *p_used, 8453 struct netlink_ext_ack *extack) 8454 { 8455 struct netdev_notifier_offload_xstats_ru report_used = {}; 8456 struct netdev_notifier_offload_xstats_info info = { 8457 .info.dev = dev, 8458 .info.extack = extack, 8459 .type = type, 8460 .report_used = &report_used, 8461 }; 8462 int rc; 8463 8464 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8465 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8466 &info.info); 8467 *p_used = report_used.used; 8468 return notifier_to_errno(rc); 8469 } 8470 8471 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8472 enum netdev_offload_xstats_type type, 8473 struct rtnl_hw_stats64 *p_stats, 8474 bool *p_used, 8475 struct netlink_ext_ack *extack) 8476 { 8477 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8478 struct netdev_notifier_offload_xstats_info info = { 8479 .info.dev = dev, 8480 .info.extack = extack, 8481 .type = type, 8482 .report_delta = &report_delta, 8483 }; 8484 struct rtnl_hw_stats64 *stats; 8485 int rc; 8486 8487 stats = netdev_offload_xstats_get_ptr(dev, type); 8488 if (WARN_ON(!stats)) 8489 return -EINVAL; 8490 8491 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8492 &info.info); 8493 8494 /* Cache whatever we got, even if there was an error, otherwise the 8495 * successful stats retrievals would get lost. 8496 */ 8497 netdev_hw_stats64_add(stats, &report_delta.stats); 8498 8499 if (p_stats) 8500 *p_stats = *stats; 8501 *p_used = report_delta.used; 8502 8503 return notifier_to_errno(rc); 8504 } 8505 8506 int netdev_offload_xstats_get(struct net_device *dev, 8507 enum netdev_offload_xstats_type type, 8508 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8509 struct netlink_ext_ack *extack) 8510 { 8511 ASSERT_RTNL(); 8512 8513 if (p_stats) 8514 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8515 p_used, extack); 8516 else 8517 return netdev_offload_xstats_get_used(dev, type, p_used, 8518 extack); 8519 } 8520 EXPORT_SYMBOL(netdev_offload_xstats_get); 8521 8522 void 8523 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8524 const struct rtnl_hw_stats64 *stats) 8525 { 8526 report_delta->used = true; 8527 netdev_hw_stats64_add(&report_delta->stats, stats); 8528 } 8529 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8530 8531 void 8532 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8533 { 8534 report_used->used = true; 8535 } 8536 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8537 8538 void netdev_offload_xstats_push_delta(struct net_device *dev, 8539 enum netdev_offload_xstats_type type, 8540 const struct rtnl_hw_stats64 *p_stats) 8541 { 8542 struct rtnl_hw_stats64 *stats; 8543 8544 ASSERT_RTNL(); 8545 8546 stats = netdev_offload_xstats_get_ptr(dev, type); 8547 if (WARN_ON(!stats)) 8548 return; 8549 8550 netdev_hw_stats64_add(stats, p_stats); 8551 } 8552 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8553 8554 /** 8555 * netdev_get_xmit_slave - Get the xmit slave of master device 8556 * @dev: device 8557 * @skb: The packet 8558 * @all_slaves: assume all the slaves are active 8559 * 8560 * The reference counters are not incremented so the caller must be 8561 * careful with locks. The caller must hold RCU lock. 8562 * %NULL is returned if no slave is found. 8563 */ 8564 8565 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8566 struct sk_buff *skb, 8567 bool all_slaves) 8568 { 8569 const struct net_device_ops *ops = dev->netdev_ops; 8570 8571 if (!ops->ndo_get_xmit_slave) 8572 return NULL; 8573 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8574 } 8575 EXPORT_SYMBOL(netdev_get_xmit_slave); 8576 8577 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8578 struct sock *sk) 8579 { 8580 const struct net_device_ops *ops = dev->netdev_ops; 8581 8582 if (!ops->ndo_sk_get_lower_dev) 8583 return NULL; 8584 return ops->ndo_sk_get_lower_dev(dev, sk); 8585 } 8586 8587 /** 8588 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8589 * @dev: device 8590 * @sk: the socket 8591 * 8592 * %NULL is returned if no lower device is found. 8593 */ 8594 8595 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8596 struct sock *sk) 8597 { 8598 struct net_device *lower; 8599 8600 lower = netdev_sk_get_lower_dev(dev, sk); 8601 while (lower) { 8602 dev = lower; 8603 lower = netdev_sk_get_lower_dev(dev, sk); 8604 } 8605 8606 return dev; 8607 } 8608 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8609 8610 static void netdev_adjacent_add_links(struct net_device *dev) 8611 { 8612 struct netdev_adjacent *iter; 8613 8614 struct net *net = dev_net(dev); 8615 8616 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8617 if (!net_eq(net, dev_net(iter->dev))) 8618 continue; 8619 netdev_adjacent_sysfs_add(iter->dev, dev, 8620 &iter->dev->adj_list.lower); 8621 netdev_adjacent_sysfs_add(dev, iter->dev, 8622 &dev->adj_list.upper); 8623 } 8624 8625 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8626 if (!net_eq(net, dev_net(iter->dev))) 8627 continue; 8628 netdev_adjacent_sysfs_add(iter->dev, dev, 8629 &iter->dev->adj_list.upper); 8630 netdev_adjacent_sysfs_add(dev, iter->dev, 8631 &dev->adj_list.lower); 8632 } 8633 } 8634 8635 static void netdev_adjacent_del_links(struct net_device *dev) 8636 { 8637 struct netdev_adjacent *iter; 8638 8639 struct net *net = dev_net(dev); 8640 8641 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8642 if (!net_eq(net, dev_net(iter->dev))) 8643 continue; 8644 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8645 &iter->dev->adj_list.lower); 8646 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8647 &dev->adj_list.upper); 8648 } 8649 8650 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8651 if (!net_eq(net, dev_net(iter->dev))) 8652 continue; 8653 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8654 &iter->dev->adj_list.upper); 8655 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8656 &dev->adj_list.lower); 8657 } 8658 } 8659 8660 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8661 { 8662 struct netdev_adjacent *iter; 8663 8664 struct net *net = dev_net(dev); 8665 8666 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8667 if (!net_eq(net, dev_net(iter->dev))) 8668 continue; 8669 netdev_adjacent_sysfs_del(iter->dev, oldname, 8670 &iter->dev->adj_list.lower); 8671 netdev_adjacent_sysfs_add(iter->dev, dev, 8672 &iter->dev->adj_list.lower); 8673 } 8674 8675 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8676 if (!net_eq(net, dev_net(iter->dev))) 8677 continue; 8678 netdev_adjacent_sysfs_del(iter->dev, oldname, 8679 &iter->dev->adj_list.upper); 8680 netdev_adjacent_sysfs_add(iter->dev, dev, 8681 &iter->dev->adj_list.upper); 8682 } 8683 } 8684 8685 void *netdev_lower_dev_get_private(struct net_device *dev, 8686 struct net_device *lower_dev) 8687 { 8688 struct netdev_adjacent *lower; 8689 8690 if (!lower_dev) 8691 return NULL; 8692 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8693 if (!lower) 8694 return NULL; 8695 8696 return lower->private; 8697 } 8698 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8699 8700 8701 /** 8702 * netdev_lower_state_changed - Dispatch event about lower device state change 8703 * @lower_dev: device 8704 * @lower_state_info: state to dispatch 8705 * 8706 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8707 * The caller must hold the RTNL lock. 8708 */ 8709 void netdev_lower_state_changed(struct net_device *lower_dev, 8710 void *lower_state_info) 8711 { 8712 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8713 .info.dev = lower_dev, 8714 }; 8715 8716 ASSERT_RTNL(); 8717 changelowerstate_info.lower_state_info = lower_state_info; 8718 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8719 &changelowerstate_info.info); 8720 } 8721 EXPORT_SYMBOL(netdev_lower_state_changed); 8722 8723 static void dev_change_rx_flags(struct net_device *dev, int flags) 8724 { 8725 const struct net_device_ops *ops = dev->netdev_ops; 8726 8727 if (ops->ndo_change_rx_flags) 8728 ops->ndo_change_rx_flags(dev, flags); 8729 } 8730 8731 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8732 { 8733 unsigned int old_flags = dev->flags; 8734 unsigned int promiscuity, flags; 8735 kuid_t uid; 8736 kgid_t gid; 8737 8738 ASSERT_RTNL(); 8739 8740 promiscuity = dev->promiscuity + inc; 8741 if (promiscuity == 0) { 8742 /* 8743 * Avoid overflow. 8744 * If inc causes overflow, untouch promisc and return error. 8745 */ 8746 if (unlikely(inc > 0)) { 8747 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8748 return -EOVERFLOW; 8749 } 8750 flags = old_flags & ~IFF_PROMISC; 8751 } else { 8752 flags = old_flags | IFF_PROMISC; 8753 } 8754 WRITE_ONCE(dev->promiscuity, promiscuity); 8755 if (flags != old_flags) { 8756 WRITE_ONCE(dev->flags, flags); 8757 netdev_info(dev, "%s promiscuous mode\n", 8758 dev->flags & IFF_PROMISC ? "entered" : "left"); 8759 if (audit_enabled) { 8760 current_uid_gid(&uid, &gid); 8761 audit_log(audit_context(), GFP_ATOMIC, 8762 AUDIT_ANOM_PROMISCUOUS, 8763 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8764 dev->name, (dev->flags & IFF_PROMISC), 8765 (old_flags & IFF_PROMISC), 8766 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8767 from_kuid(&init_user_ns, uid), 8768 from_kgid(&init_user_ns, gid), 8769 audit_get_sessionid(current)); 8770 } 8771 8772 dev_change_rx_flags(dev, IFF_PROMISC); 8773 } 8774 if (notify) 8775 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8776 return 0; 8777 } 8778 8779 /** 8780 * dev_set_promiscuity - update promiscuity count on a device 8781 * @dev: device 8782 * @inc: modifier 8783 * 8784 * Add or remove promiscuity from a device. While the count in the device 8785 * remains above zero the interface remains promiscuous. Once it hits zero 8786 * the device reverts back to normal filtering operation. A negative inc 8787 * value is used to drop promiscuity on the device. 8788 * Return 0 if successful or a negative errno code on error. 8789 */ 8790 int dev_set_promiscuity(struct net_device *dev, int inc) 8791 { 8792 unsigned int old_flags = dev->flags; 8793 int err; 8794 8795 err = __dev_set_promiscuity(dev, inc, true); 8796 if (err < 0) 8797 return err; 8798 if (dev->flags != old_flags) 8799 dev_set_rx_mode(dev); 8800 return err; 8801 } 8802 EXPORT_SYMBOL(dev_set_promiscuity); 8803 8804 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8805 { 8806 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8807 unsigned int allmulti, flags; 8808 8809 ASSERT_RTNL(); 8810 8811 allmulti = dev->allmulti + inc; 8812 if (allmulti == 0) { 8813 /* 8814 * Avoid overflow. 8815 * If inc causes overflow, untouch allmulti and return error. 8816 */ 8817 if (unlikely(inc > 0)) { 8818 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8819 return -EOVERFLOW; 8820 } 8821 flags = old_flags & ~IFF_ALLMULTI; 8822 } else { 8823 flags = old_flags | IFF_ALLMULTI; 8824 } 8825 WRITE_ONCE(dev->allmulti, allmulti); 8826 if (flags != old_flags) { 8827 WRITE_ONCE(dev->flags, flags); 8828 netdev_info(dev, "%s allmulticast mode\n", 8829 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8830 dev_change_rx_flags(dev, IFF_ALLMULTI); 8831 dev_set_rx_mode(dev); 8832 if (notify) 8833 __dev_notify_flags(dev, old_flags, 8834 dev->gflags ^ old_gflags, 0, NULL); 8835 } 8836 return 0; 8837 } 8838 8839 /** 8840 * dev_set_allmulti - update allmulti count on a device 8841 * @dev: device 8842 * @inc: modifier 8843 * 8844 * Add or remove reception of all multicast frames to a device. While the 8845 * count in the device remains above zero the interface remains listening 8846 * to all interfaces. Once it hits zero the device reverts back to normal 8847 * filtering operation. A negative @inc value is used to drop the counter 8848 * when releasing a resource needing all multicasts. 8849 * Return 0 if successful or a negative errno code on error. 8850 */ 8851 8852 int dev_set_allmulti(struct net_device *dev, int inc) 8853 { 8854 return __dev_set_allmulti(dev, inc, true); 8855 } 8856 EXPORT_SYMBOL(dev_set_allmulti); 8857 8858 /* 8859 * Upload unicast and multicast address lists to device and 8860 * configure RX filtering. When the device doesn't support unicast 8861 * filtering it is put in promiscuous mode while unicast addresses 8862 * are present. 8863 */ 8864 void __dev_set_rx_mode(struct net_device *dev) 8865 { 8866 const struct net_device_ops *ops = dev->netdev_ops; 8867 8868 /* dev_open will call this function so the list will stay sane. */ 8869 if (!(dev->flags&IFF_UP)) 8870 return; 8871 8872 if (!netif_device_present(dev)) 8873 return; 8874 8875 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8876 /* Unicast addresses changes may only happen under the rtnl, 8877 * therefore calling __dev_set_promiscuity here is safe. 8878 */ 8879 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8880 __dev_set_promiscuity(dev, 1, false); 8881 dev->uc_promisc = true; 8882 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8883 __dev_set_promiscuity(dev, -1, false); 8884 dev->uc_promisc = false; 8885 } 8886 } 8887 8888 if (ops->ndo_set_rx_mode) 8889 ops->ndo_set_rx_mode(dev); 8890 } 8891 8892 void dev_set_rx_mode(struct net_device *dev) 8893 { 8894 netif_addr_lock_bh(dev); 8895 __dev_set_rx_mode(dev); 8896 netif_addr_unlock_bh(dev); 8897 } 8898 8899 /** 8900 * dev_get_flags - get flags reported to userspace 8901 * @dev: device 8902 * 8903 * Get the combination of flag bits exported through APIs to userspace. 8904 */ 8905 unsigned int dev_get_flags(const struct net_device *dev) 8906 { 8907 unsigned int flags; 8908 8909 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | 8910 IFF_ALLMULTI | 8911 IFF_RUNNING | 8912 IFF_LOWER_UP | 8913 IFF_DORMANT)) | 8914 (READ_ONCE(dev->gflags) & (IFF_PROMISC | 8915 IFF_ALLMULTI)); 8916 8917 if (netif_running(dev)) { 8918 if (netif_oper_up(dev)) 8919 flags |= IFF_RUNNING; 8920 if (netif_carrier_ok(dev)) 8921 flags |= IFF_LOWER_UP; 8922 if (netif_dormant(dev)) 8923 flags |= IFF_DORMANT; 8924 } 8925 8926 return flags; 8927 } 8928 EXPORT_SYMBOL(dev_get_flags); 8929 8930 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8931 struct netlink_ext_ack *extack) 8932 { 8933 unsigned int old_flags = dev->flags; 8934 int ret; 8935 8936 ASSERT_RTNL(); 8937 8938 /* 8939 * Set the flags on our device. 8940 */ 8941 8942 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8943 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8944 IFF_AUTOMEDIA)) | 8945 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8946 IFF_ALLMULTI)); 8947 8948 /* 8949 * Load in the correct multicast list now the flags have changed. 8950 */ 8951 8952 if ((old_flags ^ flags) & IFF_MULTICAST) 8953 dev_change_rx_flags(dev, IFF_MULTICAST); 8954 8955 dev_set_rx_mode(dev); 8956 8957 /* 8958 * Have we downed the interface. We handle IFF_UP ourselves 8959 * according to user attempts to set it, rather than blindly 8960 * setting it. 8961 */ 8962 8963 ret = 0; 8964 if ((old_flags ^ flags) & IFF_UP) { 8965 if (old_flags & IFF_UP) 8966 __dev_close(dev); 8967 else 8968 ret = __dev_open(dev, extack); 8969 } 8970 8971 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8972 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8973 unsigned int old_flags = dev->flags; 8974 8975 dev->gflags ^= IFF_PROMISC; 8976 8977 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8978 if (dev->flags != old_flags) 8979 dev_set_rx_mode(dev); 8980 } 8981 8982 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8983 * is important. Some (broken) drivers set IFF_PROMISC, when 8984 * IFF_ALLMULTI is requested not asking us and not reporting. 8985 */ 8986 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8987 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8988 8989 dev->gflags ^= IFF_ALLMULTI; 8990 __dev_set_allmulti(dev, inc, false); 8991 } 8992 8993 return ret; 8994 } 8995 8996 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8997 unsigned int gchanges, u32 portid, 8998 const struct nlmsghdr *nlh) 8999 { 9000 unsigned int changes = dev->flags ^ old_flags; 9001 9002 if (gchanges) 9003 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 9004 9005 if (changes & IFF_UP) { 9006 if (dev->flags & IFF_UP) 9007 call_netdevice_notifiers(NETDEV_UP, dev); 9008 else 9009 call_netdevice_notifiers(NETDEV_DOWN, dev); 9010 } 9011 9012 if (dev->flags & IFF_UP && 9013 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 9014 struct netdev_notifier_change_info change_info = { 9015 .info = { 9016 .dev = dev, 9017 }, 9018 .flags_changed = changes, 9019 }; 9020 9021 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 9022 } 9023 } 9024 9025 /** 9026 * dev_change_flags - change device settings 9027 * @dev: device 9028 * @flags: device state flags 9029 * @extack: netlink extended ack 9030 * 9031 * Change settings on device based state flags. The flags are 9032 * in the userspace exported format. 9033 */ 9034 int dev_change_flags(struct net_device *dev, unsigned int flags, 9035 struct netlink_ext_ack *extack) 9036 { 9037 int ret; 9038 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 9039 9040 ret = __dev_change_flags(dev, flags, extack); 9041 if (ret < 0) 9042 return ret; 9043 9044 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 9045 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 9046 return ret; 9047 } 9048 EXPORT_SYMBOL(dev_change_flags); 9049 9050 int __dev_set_mtu(struct net_device *dev, int new_mtu) 9051 { 9052 const struct net_device_ops *ops = dev->netdev_ops; 9053 9054 if (ops->ndo_change_mtu) 9055 return ops->ndo_change_mtu(dev, new_mtu); 9056 9057 /* Pairs with all the lockless reads of dev->mtu in the stack */ 9058 WRITE_ONCE(dev->mtu, new_mtu); 9059 return 0; 9060 } 9061 EXPORT_SYMBOL(__dev_set_mtu); 9062 9063 int dev_validate_mtu(struct net_device *dev, int new_mtu, 9064 struct netlink_ext_ack *extack) 9065 { 9066 /* MTU must be positive, and in range */ 9067 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 9068 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 9069 return -EINVAL; 9070 } 9071 9072 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 9073 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 9074 return -EINVAL; 9075 } 9076 return 0; 9077 } 9078 9079 /** 9080 * dev_set_mtu_ext - Change maximum transfer unit 9081 * @dev: device 9082 * @new_mtu: new transfer unit 9083 * @extack: netlink extended ack 9084 * 9085 * Change the maximum transfer size of the network device. 9086 */ 9087 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 9088 struct netlink_ext_ack *extack) 9089 { 9090 int err, orig_mtu; 9091 9092 if (new_mtu == dev->mtu) 9093 return 0; 9094 9095 err = dev_validate_mtu(dev, new_mtu, extack); 9096 if (err) 9097 return err; 9098 9099 if (!netif_device_present(dev)) 9100 return -ENODEV; 9101 9102 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 9103 err = notifier_to_errno(err); 9104 if (err) 9105 return err; 9106 9107 orig_mtu = dev->mtu; 9108 err = __dev_set_mtu(dev, new_mtu); 9109 9110 if (!err) { 9111 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9112 orig_mtu); 9113 err = notifier_to_errno(err); 9114 if (err) { 9115 /* setting mtu back and notifying everyone again, 9116 * so that they have a chance to revert changes. 9117 */ 9118 __dev_set_mtu(dev, orig_mtu); 9119 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9120 new_mtu); 9121 } 9122 } 9123 return err; 9124 } 9125 9126 int dev_set_mtu(struct net_device *dev, int new_mtu) 9127 { 9128 struct netlink_ext_ack extack; 9129 int err; 9130 9131 memset(&extack, 0, sizeof(extack)); 9132 err = dev_set_mtu_ext(dev, new_mtu, &extack); 9133 if (err && extack._msg) 9134 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 9135 return err; 9136 } 9137 EXPORT_SYMBOL(dev_set_mtu); 9138 9139 /** 9140 * dev_change_tx_queue_len - Change TX queue length of a netdevice 9141 * @dev: device 9142 * @new_len: new tx queue length 9143 */ 9144 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 9145 { 9146 unsigned int orig_len = dev->tx_queue_len; 9147 int res; 9148 9149 if (new_len != (unsigned int)new_len) 9150 return -ERANGE; 9151 9152 if (new_len != orig_len) { 9153 WRITE_ONCE(dev->tx_queue_len, new_len); 9154 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 9155 res = notifier_to_errno(res); 9156 if (res) 9157 goto err_rollback; 9158 res = dev_qdisc_change_tx_queue_len(dev); 9159 if (res) 9160 goto err_rollback; 9161 } 9162 9163 return 0; 9164 9165 err_rollback: 9166 netdev_err(dev, "refused to change device tx_queue_len\n"); 9167 WRITE_ONCE(dev->tx_queue_len, orig_len); 9168 return res; 9169 } 9170 9171 /** 9172 * dev_set_group - Change group this device belongs to 9173 * @dev: device 9174 * @new_group: group this device should belong to 9175 */ 9176 void dev_set_group(struct net_device *dev, int new_group) 9177 { 9178 dev->group = new_group; 9179 } 9180 9181 /** 9182 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 9183 * @dev: device 9184 * @addr: new address 9185 * @extack: netlink extended ack 9186 */ 9187 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9188 struct netlink_ext_ack *extack) 9189 { 9190 struct netdev_notifier_pre_changeaddr_info info = { 9191 .info.dev = dev, 9192 .info.extack = extack, 9193 .dev_addr = addr, 9194 }; 9195 int rc; 9196 9197 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9198 return notifier_to_errno(rc); 9199 } 9200 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9201 9202 /** 9203 * dev_set_mac_address - Change Media Access Control Address 9204 * @dev: device 9205 * @sa: new address 9206 * @extack: netlink extended ack 9207 * 9208 * Change the hardware (MAC) address of the device 9209 */ 9210 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9211 struct netlink_ext_ack *extack) 9212 { 9213 const struct net_device_ops *ops = dev->netdev_ops; 9214 int err; 9215 9216 if (!ops->ndo_set_mac_address) 9217 return -EOPNOTSUPP; 9218 if (sa->sa_family != dev->type) 9219 return -EINVAL; 9220 if (!netif_device_present(dev)) 9221 return -ENODEV; 9222 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9223 if (err) 9224 return err; 9225 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 9226 err = ops->ndo_set_mac_address(dev, sa); 9227 if (err) 9228 return err; 9229 } 9230 dev->addr_assign_type = NET_ADDR_SET; 9231 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9232 add_device_randomness(dev->dev_addr, dev->addr_len); 9233 return 0; 9234 } 9235 EXPORT_SYMBOL(dev_set_mac_address); 9236 9237 DECLARE_RWSEM(dev_addr_sem); 9238 9239 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9240 struct netlink_ext_ack *extack) 9241 { 9242 int ret; 9243 9244 down_write(&dev_addr_sem); 9245 ret = dev_set_mac_address(dev, sa, extack); 9246 up_write(&dev_addr_sem); 9247 return ret; 9248 } 9249 EXPORT_SYMBOL(dev_set_mac_address_user); 9250 9251 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9252 { 9253 size_t size = sizeof(sa->sa_data_min); 9254 struct net_device *dev; 9255 int ret = 0; 9256 9257 down_read(&dev_addr_sem); 9258 rcu_read_lock(); 9259 9260 dev = dev_get_by_name_rcu(net, dev_name); 9261 if (!dev) { 9262 ret = -ENODEV; 9263 goto unlock; 9264 } 9265 if (!dev->addr_len) 9266 memset(sa->sa_data, 0, size); 9267 else 9268 memcpy(sa->sa_data, dev->dev_addr, 9269 min_t(size_t, size, dev->addr_len)); 9270 sa->sa_family = dev->type; 9271 9272 unlock: 9273 rcu_read_unlock(); 9274 up_read(&dev_addr_sem); 9275 return ret; 9276 } 9277 EXPORT_SYMBOL(dev_get_mac_address); 9278 9279 /** 9280 * dev_change_carrier - Change device carrier 9281 * @dev: device 9282 * @new_carrier: new value 9283 * 9284 * Change device carrier 9285 */ 9286 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9287 { 9288 const struct net_device_ops *ops = dev->netdev_ops; 9289 9290 if (!ops->ndo_change_carrier) 9291 return -EOPNOTSUPP; 9292 if (!netif_device_present(dev)) 9293 return -ENODEV; 9294 return ops->ndo_change_carrier(dev, new_carrier); 9295 } 9296 9297 /** 9298 * dev_get_phys_port_id - Get device physical port ID 9299 * @dev: device 9300 * @ppid: port ID 9301 * 9302 * Get device physical port ID 9303 */ 9304 int dev_get_phys_port_id(struct net_device *dev, 9305 struct netdev_phys_item_id *ppid) 9306 { 9307 const struct net_device_ops *ops = dev->netdev_ops; 9308 9309 if (!ops->ndo_get_phys_port_id) 9310 return -EOPNOTSUPP; 9311 return ops->ndo_get_phys_port_id(dev, ppid); 9312 } 9313 9314 /** 9315 * dev_get_phys_port_name - Get device physical port name 9316 * @dev: device 9317 * @name: port name 9318 * @len: limit of bytes to copy to name 9319 * 9320 * Get device physical port name 9321 */ 9322 int dev_get_phys_port_name(struct net_device *dev, 9323 char *name, size_t len) 9324 { 9325 const struct net_device_ops *ops = dev->netdev_ops; 9326 int err; 9327 9328 if (ops->ndo_get_phys_port_name) { 9329 err = ops->ndo_get_phys_port_name(dev, name, len); 9330 if (err != -EOPNOTSUPP) 9331 return err; 9332 } 9333 return devlink_compat_phys_port_name_get(dev, name, len); 9334 } 9335 9336 /** 9337 * dev_get_port_parent_id - Get the device's port parent identifier 9338 * @dev: network device 9339 * @ppid: pointer to a storage for the port's parent identifier 9340 * @recurse: allow/disallow recursion to lower devices 9341 * 9342 * Get the devices's port parent identifier 9343 */ 9344 int dev_get_port_parent_id(struct net_device *dev, 9345 struct netdev_phys_item_id *ppid, 9346 bool recurse) 9347 { 9348 const struct net_device_ops *ops = dev->netdev_ops; 9349 struct netdev_phys_item_id first = { }; 9350 struct net_device *lower_dev; 9351 struct list_head *iter; 9352 int err; 9353 9354 if (ops->ndo_get_port_parent_id) { 9355 err = ops->ndo_get_port_parent_id(dev, ppid); 9356 if (err != -EOPNOTSUPP) 9357 return err; 9358 } 9359 9360 err = devlink_compat_switch_id_get(dev, ppid); 9361 if (!recurse || err != -EOPNOTSUPP) 9362 return err; 9363 9364 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9365 err = dev_get_port_parent_id(lower_dev, ppid, true); 9366 if (err) 9367 break; 9368 if (!first.id_len) 9369 first = *ppid; 9370 else if (memcmp(&first, ppid, sizeof(*ppid))) 9371 return -EOPNOTSUPP; 9372 } 9373 9374 return err; 9375 } 9376 EXPORT_SYMBOL(dev_get_port_parent_id); 9377 9378 /** 9379 * netdev_port_same_parent_id - Indicate if two network devices have 9380 * the same port parent identifier 9381 * @a: first network device 9382 * @b: second network device 9383 */ 9384 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9385 { 9386 struct netdev_phys_item_id a_id = { }; 9387 struct netdev_phys_item_id b_id = { }; 9388 9389 if (dev_get_port_parent_id(a, &a_id, true) || 9390 dev_get_port_parent_id(b, &b_id, true)) 9391 return false; 9392 9393 return netdev_phys_item_id_same(&a_id, &b_id); 9394 } 9395 EXPORT_SYMBOL(netdev_port_same_parent_id); 9396 9397 /** 9398 * dev_change_proto_down - set carrier according to proto_down. 9399 * 9400 * @dev: device 9401 * @proto_down: new value 9402 */ 9403 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9404 { 9405 if (!dev->change_proto_down) 9406 return -EOPNOTSUPP; 9407 if (!netif_device_present(dev)) 9408 return -ENODEV; 9409 if (proto_down) 9410 netif_carrier_off(dev); 9411 else 9412 netif_carrier_on(dev); 9413 WRITE_ONCE(dev->proto_down, proto_down); 9414 return 0; 9415 } 9416 9417 /** 9418 * dev_change_proto_down_reason - proto down reason 9419 * 9420 * @dev: device 9421 * @mask: proto down mask 9422 * @value: proto down value 9423 */ 9424 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9425 u32 value) 9426 { 9427 u32 proto_down_reason; 9428 int b; 9429 9430 if (!mask) { 9431 proto_down_reason = value; 9432 } else { 9433 proto_down_reason = dev->proto_down_reason; 9434 for_each_set_bit(b, &mask, 32) { 9435 if (value & (1 << b)) 9436 proto_down_reason |= BIT(b); 9437 else 9438 proto_down_reason &= ~BIT(b); 9439 } 9440 } 9441 WRITE_ONCE(dev->proto_down_reason, proto_down_reason); 9442 } 9443 9444 struct bpf_xdp_link { 9445 struct bpf_link link; 9446 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9447 int flags; 9448 }; 9449 9450 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9451 { 9452 if (flags & XDP_FLAGS_HW_MODE) 9453 return XDP_MODE_HW; 9454 if (flags & XDP_FLAGS_DRV_MODE) 9455 return XDP_MODE_DRV; 9456 if (flags & XDP_FLAGS_SKB_MODE) 9457 return XDP_MODE_SKB; 9458 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9459 } 9460 9461 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9462 { 9463 switch (mode) { 9464 case XDP_MODE_SKB: 9465 return generic_xdp_install; 9466 case XDP_MODE_DRV: 9467 case XDP_MODE_HW: 9468 return dev->netdev_ops->ndo_bpf; 9469 default: 9470 return NULL; 9471 } 9472 } 9473 9474 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9475 enum bpf_xdp_mode mode) 9476 { 9477 return dev->xdp_state[mode].link; 9478 } 9479 9480 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9481 enum bpf_xdp_mode mode) 9482 { 9483 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9484 9485 if (link) 9486 return link->link.prog; 9487 return dev->xdp_state[mode].prog; 9488 } 9489 9490 u8 dev_xdp_prog_count(struct net_device *dev) 9491 { 9492 u8 count = 0; 9493 int i; 9494 9495 for (i = 0; i < __MAX_XDP_MODE; i++) 9496 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9497 count++; 9498 return count; 9499 } 9500 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9501 9502 int dev_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf) 9503 { 9504 if (!dev->netdev_ops->ndo_bpf) 9505 return -EOPNOTSUPP; 9506 9507 if (dev_get_min_mp_channel_count(dev)) { 9508 NL_SET_ERR_MSG(bpf->extack, "unable to propagate XDP to device using memory provider"); 9509 return -EBUSY; 9510 } 9511 9512 return dev->netdev_ops->ndo_bpf(dev, bpf); 9513 } 9514 EXPORT_SYMBOL_GPL(dev_xdp_propagate); 9515 9516 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9517 { 9518 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9519 9520 return prog ? prog->aux->id : 0; 9521 } 9522 9523 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9524 struct bpf_xdp_link *link) 9525 { 9526 dev->xdp_state[mode].link = link; 9527 dev->xdp_state[mode].prog = NULL; 9528 } 9529 9530 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9531 struct bpf_prog *prog) 9532 { 9533 dev->xdp_state[mode].link = NULL; 9534 dev->xdp_state[mode].prog = prog; 9535 } 9536 9537 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9538 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9539 u32 flags, struct bpf_prog *prog) 9540 { 9541 struct netdev_bpf xdp; 9542 int err; 9543 9544 if (dev_get_min_mp_channel_count(dev)) { 9545 NL_SET_ERR_MSG(extack, "unable to install XDP to device using memory provider"); 9546 return -EBUSY; 9547 } 9548 9549 memset(&xdp, 0, sizeof(xdp)); 9550 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9551 xdp.extack = extack; 9552 xdp.flags = flags; 9553 xdp.prog = prog; 9554 9555 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9556 * "moved" into driver), so they don't increment it on their own, but 9557 * they do decrement refcnt when program is detached or replaced. 9558 * Given net_device also owns link/prog, we need to bump refcnt here 9559 * to prevent drivers from underflowing it. 9560 */ 9561 if (prog) 9562 bpf_prog_inc(prog); 9563 err = bpf_op(dev, &xdp); 9564 if (err) { 9565 if (prog) 9566 bpf_prog_put(prog); 9567 return err; 9568 } 9569 9570 if (mode != XDP_MODE_HW) 9571 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9572 9573 return 0; 9574 } 9575 9576 static void dev_xdp_uninstall(struct net_device *dev) 9577 { 9578 struct bpf_xdp_link *link; 9579 struct bpf_prog *prog; 9580 enum bpf_xdp_mode mode; 9581 bpf_op_t bpf_op; 9582 9583 ASSERT_RTNL(); 9584 9585 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9586 prog = dev_xdp_prog(dev, mode); 9587 if (!prog) 9588 continue; 9589 9590 bpf_op = dev_xdp_bpf_op(dev, mode); 9591 if (!bpf_op) 9592 continue; 9593 9594 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9595 9596 /* auto-detach link from net device */ 9597 link = dev_xdp_link(dev, mode); 9598 if (link) 9599 link->dev = NULL; 9600 else 9601 bpf_prog_put(prog); 9602 9603 dev_xdp_set_link(dev, mode, NULL); 9604 } 9605 } 9606 9607 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9608 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9609 struct bpf_prog *old_prog, u32 flags) 9610 { 9611 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9612 struct bpf_prog *cur_prog; 9613 struct net_device *upper; 9614 struct list_head *iter; 9615 enum bpf_xdp_mode mode; 9616 bpf_op_t bpf_op; 9617 int err; 9618 9619 ASSERT_RTNL(); 9620 9621 /* either link or prog attachment, never both */ 9622 if (link && (new_prog || old_prog)) 9623 return -EINVAL; 9624 /* link supports only XDP mode flags */ 9625 if (link && (flags & ~XDP_FLAGS_MODES)) { 9626 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9627 return -EINVAL; 9628 } 9629 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9630 if (num_modes > 1) { 9631 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9632 return -EINVAL; 9633 } 9634 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9635 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9636 NL_SET_ERR_MSG(extack, 9637 "More than one program loaded, unset mode is ambiguous"); 9638 return -EINVAL; 9639 } 9640 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9641 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9642 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9643 return -EINVAL; 9644 } 9645 9646 mode = dev_xdp_mode(dev, flags); 9647 /* can't replace attached link */ 9648 if (dev_xdp_link(dev, mode)) { 9649 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9650 return -EBUSY; 9651 } 9652 9653 /* don't allow if an upper device already has a program */ 9654 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9655 if (dev_xdp_prog_count(upper) > 0) { 9656 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9657 return -EEXIST; 9658 } 9659 } 9660 9661 cur_prog = dev_xdp_prog(dev, mode); 9662 /* can't replace attached prog with link */ 9663 if (link && cur_prog) { 9664 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9665 return -EBUSY; 9666 } 9667 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9668 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9669 return -EEXIST; 9670 } 9671 9672 /* put effective new program into new_prog */ 9673 if (link) 9674 new_prog = link->link.prog; 9675 9676 if (new_prog) { 9677 bool offload = mode == XDP_MODE_HW; 9678 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9679 ? XDP_MODE_DRV : XDP_MODE_SKB; 9680 9681 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9682 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9683 return -EBUSY; 9684 } 9685 if (!offload && dev_xdp_prog(dev, other_mode)) { 9686 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9687 return -EEXIST; 9688 } 9689 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9690 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9691 return -EINVAL; 9692 } 9693 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9694 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9695 return -EINVAL; 9696 } 9697 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9698 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9699 return -EINVAL; 9700 } 9701 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9702 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9703 return -EINVAL; 9704 } 9705 } 9706 9707 /* don't call drivers if the effective program didn't change */ 9708 if (new_prog != cur_prog) { 9709 bpf_op = dev_xdp_bpf_op(dev, mode); 9710 if (!bpf_op) { 9711 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9712 return -EOPNOTSUPP; 9713 } 9714 9715 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9716 if (err) 9717 return err; 9718 } 9719 9720 if (link) 9721 dev_xdp_set_link(dev, mode, link); 9722 else 9723 dev_xdp_set_prog(dev, mode, new_prog); 9724 if (cur_prog) 9725 bpf_prog_put(cur_prog); 9726 9727 return 0; 9728 } 9729 9730 static int dev_xdp_attach_link(struct net_device *dev, 9731 struct netlink_ext_ack *extack, 9732 struct bpf_xdp_link *link) 9733 { 9734 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9735 } 9736 9737 static int dev_xdp_detach_link(struct net_device *dev, 9738 struct netlink_ext_ack *extack, 9739 struct bpf_xdp_link *link) 9740 { 9741 enum bpf_xdp_mode mode; 9742 bpf_op_t bpf_op; 9743 9744 ASSERT_RTNL(); 9745 9746 mode = dev_xdp_mode(dev, link->flags); 9747 if (dev_xdp_link(dev, mode) != link) 9748 return -EINVAL; 9749 9750 bpf_op = dev_xdp_bpf_op(dev, mode); 9751 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9752 dev_xdp_set_link(dev, mode, NULL); 9753 return 0; 9754 } 9755 9756 static void bpf_xdp_link_release(struct bpf_link *link) 9757 { 9758 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9759 9760 rtnl_lock(); 9761 9762 /* if racing with net_device's tear down, xdp_link->dev might be 9763 * already NULL, in which case link was already auto-detached 9764 */ 9765 if (xdp_link->dev) { 9766 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9767 xdp_link->dev = NULL; 9768 } 9769 9770 rtnl_unlock(); 9771 } 9772 9773 static int bpf_xdp_link_detach(struct bpf_link *link) 9774 { 9775 bpf_xdp_link_release(link); 9776 return 0; 9777 } 9778 9779 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9780 { 9781 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9782 9783 kfree(xdp_link); 9784 } 9785 9786 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9787 struct seq_file *seq) 9788 { 9789 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9790 u32 ifindex = 0; 9791 9792 rtnl_lock(); 9793 if (xdp_link->dev) 9794 ifindex = xdp_link->dev->ifindex; 9795 rtnl_unlock(); 9796 9797 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9798 } 9799 9800 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9801 struct bpf_link_info *info) 9802 { 9803 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9804 u32 ifindex = 0; 9805 9806 rtnl_lock(); 9807 if (xdp_link->dev) 9808 ifindex = xdp_link->dev->ifindex; 9809 rtnl_unlock(); 9810 9811 info->xdp.ifindex = ifindex; 9812 return 0; 9813 } 9814 9815 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9816 struct bpf_prog *old_prog) 9817 { 9818 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9819 enum bpf_xdp_mode mode; 9820 bpf_op_t bpf_op; 9821 int err = 0; 9822 9823 rtnl_lock(); 9824 9825 /* link might have been auto-released already, so fail */ 9826 if (!xdp_link->dev) { 9827 err = -ENOLINK; 9828 goto out_unlock; 9829 } 9830 9831 if (old_prog && link->prog != old_prog) { 9832 err = -EPERM; 9833 goto out_unlock; 9834 } 9835 old_prog = link->prog; 9836 if (old_prog->type != new_prog->type || 9837 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9838 err = -EINVAL; 9839 goto out_unlock; 9840 } 9841 9842 if (old_prog == new_prog) { 9843 /* no-op, don't disturb drivers */ 9844 bpf_prog_put(new_prog); 9845 goto out_unlock; 9846 } 9847 9848 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9849 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9850 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9851 xdp_link->flags, new_prog); 9852 if (err) 9853 goto out_unlock; 9854 9855 old_prog = xchg(&link->prog, new_prog); 9856 bpf_prog_put(old_prog); 9857 9858 out_unlock: 9859 rtnl_unlock(); 9860 return err; 9861 } 9862 9863 static const struct bpf_link_ops bpf_xdp_link_lops = { 9864 .release = bpf_xdp_link_release, 9865 .dealloc = bpf_xdp_link_dealloc, 9866 .detach = bpf_xdp_link_detach, 9867 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9868 .fill_link_info = bpf_xdp_link_fill_link_info, 9869 .update_prog = bpf_xdp_link_update, 9870 }; 9871 9872 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9873 { 9874 struct net *net = current->nsproxy->net_ns; 9875 struct bpf_link_primer link_primer; 9876 struct netlink_ext_ack extack = {}; 9877 struct bpf_xdp_link *link; 9878 struct net_device *dev; 9879 int err, fd; 9880 9881 rtnl_lock(); 9882 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9883 if (!dev) { 9884 rtnl_unlock(); 9885 return -EINVAL; 9886 } 9887 9888 link = kzalloc(sizeof(*link), GFP_USER); 9889 if (!link) { 9890 err = -ENOMEM; 9891 goto unlock; 9892 } 9893 9894 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9895 link->dev = dev; 9896 link->flags = attr->link_create.flags; 9897 9898 err = bpf_link_prime(&link->link, &link_primer); 9899 if (err) { 9900 kfree(link); 9901 goto unlock; 9902 } 9903 9904 err = dev_xdp_attach_link(dev, &extack, link); 9905 rtnl_unlock(); 9906 9907 if (err) { 9908 link->dev = NULL; 9909 bpf_link_cleanup(&link_primer); 9910 trace_bpf_xdp_link_attach_failed(extack._msg); 9911 goto out_put_dev; 9912 } 9913 9914 fd = bpf_link_settle(&link_primer); 9915 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9916 dev_put(dev); 9917 return fd; 9918 9919 unlock: 9920 rtnl_unlock(); 9921 9922 out_put_dev: 9923 dev_put(dev); 9924 return err; 9925 } 9926 9927 /** 9928 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9929 * @dev: device 9930 * @extack: netlink extended ack 9931 * @fd: new program fd or negative value to clear 9932 * @expected_fd: old program fd that userspace expects to replace or clear 9933 * @flags: xdp-related flags 9934 * 9935 * Set or clear a bpf program for a device 9936 */ 9937 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9938 int fd, int expected_fd, u32 flags) 9939 { 9940 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9941 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9942 int err; 9943 9944 ASSERT_RTNL(); 9945 9946 if (fd >= 0) { 9947 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9948 mode != XDP_MODE_SKB); 9949 if (IS_ERR(new_prog)) 9950 return PTR_ERR(new_prog); 9951 } 9952 9953 if (expected_fd >= 0) { 9954 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9955 mode != XDP_MODE_SKB); 9956 if (IS_ERR(old_prog)) { 9957 err = PTR_ERR(old_prog); 9958 old_prog = NULL; 9959 goto err_out; 9960 } 9961 } 9962 9963 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9964 9965 err_out: 9966 if (err && new_prog) 9967 bpf_prog_put(new_prog); 9968 if (old_prog) 9969 bpf_prog_put(old_prog); 9970 return err; 9971 } 9972 9973 u32 dev_get_min_mp_channel_count(const struct net_device *dev) 9974 { 9975 int i; 9976 9977 ASSERT_RTNL(); 9978 9979 for (i = dev->real_num_rx_queues - 1; i >= 0; i--) 9980 if (dev->_rx[i].mp_params.mp_priv) 9981 /* The channel count is the idx plus 1. */ 9982 return i + 1; 9983 9984 return 0; 9985 } 9986 9987 /** 9988 * dev_index_reserve() - allocate an ifindex in a namespace 9989 * @net: the applicable net namespace 9990 * @ifindex: requested ifindex, pass %0 to get one allocated 9991 * 9992 * Allocate a ifindex for a new device. Caller must either use the ifindex 9993 * to store the device (via list_netdevice()) or call dev_index_release() 9994 * to give the index up. 9995 * 9996 * Return: a suitable unique value for a new device interface number or -errno. 9997 */ 9998 static int dev_index_reserve(struct net *net, u32 ifindex) 9999 { 10000 int err; 10001 10002 if (ifindex > INT_MAX) { 10003 DEBUG_NET_WARN_ON_ONCE(1); 10004 return -EINVAL; 10005 } 10006 10007 if (!ifindex) 10008 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 10009 xa_limit_31b, &net->ifindex, GFP_KERNEL); 10010 else 10011 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 10012 if (err < 0) 10013 return err; 10014 10015 return ifindex; 10016 } 10017 10018 static void dev_index_release(struct net *net, int ifindex) 10019 { 10020 /* Expect only unused indexes, unlist_netdevice() removes the used */ 10021 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 10022 } 10023 10024 /* Delayed registration/unregisteration */ 10025 LIST_HEAD(net_todo_list); 10026 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 10027 atomic_t dev_unreg_count = ATOMIC_INIT(0); 10028 10029 static void net_set_todo(struct net_device *dev) 10030 { 10031 list_add_tail(&dev->todo_list, &net_todo_list); 10032 } 10033 10034 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 10035 struct net_device *upper, netdev_features_t features) 10036 { 10037 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10038 netdev_features_t feature; 10039 int feature_bit; 10040 10041 for_each_netdev_feature(upper_disables, feature_bit) { 10042 feature = __NETIF_F_BIT(feature_bit); 10043 if (!(upper->wanted_features & feature) 10044 && (features & feature)) { 10045 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 10046 &feature, upper->name); 10047 features &= ~feature; 10048 } 10049 } 10050 10051 return features; 10052 } 10053 10054 static void netdev_sync_lower_features(struct net_device *upper, 10055 struct net_device *lower, netdev_features_t features) 10056 { 10057 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10058 netdev_features_t feature; 10059 int feature_bit; 10060 10061 for_each_netdev_feature(upper_disables, feature_bit) { 10062 feature = __NETIF_F_BIT(feature_bit); 10063 if (!(features & feature) && (lower->features & feature)) { 10064 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 10065 &feature, lower->name); 10066 lower->wanted_features &= ~feature; 10067 __netdev_update_features(lower); 10068 10069 if (unlikely(lower->features & feature)) 10070 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 10071 &feature, lower->name); 10072 else 10073 netdev_features_change(lower); 10074 } 10075 } 10076 } 10077 10078 static bool netdev_has_ip_or_hw_csum(netdev_features_t features) 10079 { 10080 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 10081 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask; 10082 bool hw_csum = features & NETIF_F_HW_CSUM; 10083 10084 return ip_csum || hw_csum; 10085 } 10086 10087 static netdev_features_t netdev_fix_features(struct net_device *dev, 10088 netdev_features_t features) 10089 { 10090 /* Fix illegal checksum combinations */ 10091 if ((features & NETIF_F_HW_CSUM) && 10092 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 10093 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 10094 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 10095 } 10096 10097 /* TSO requires that SG is present as well. */ 10098 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 10099 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 10100 features &= ~NETIF_F_ALL_TSO; 10101 } 10102 10103 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 10104 !(features & NETIF_F_IP_CSUM)) { 10105 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 10106 features &= ~NETIF_F_TSO; 10107 features &= ~NETIF_F_TSO_ECN; 10108 } 10109 10110 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 10111 !(features & NETIF_F_IPV6_CSUM)) { 10112 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 10113 features &= ~NETIF_F_TSO6; 10114 } 10115 10116 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 10117 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 10118 features &= ~NETIF_F_TSO_MANGLEID; 10119 10120 /* TSO ECN requires that TSO is present as well. */ 10121 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 10122 features &= ~NETIF_F_TSO_ECN; 10123 10124 /* Software GSO depends on SG. */ 10125 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 10126 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 10127 features &= ~NETIF_F_GSO; 10128 } 10129 10130 /* GSO partial features require GSO partial be set */ 10131 if ((features & dev->gso_partial_features) && 10132 !(features & NETIF_F_GSO_PARTIAL)) { 10133 netdev_dbg(dev, 10134 "Dropping partially supported GSO features since no GSO partial.\n"); 10135 features &= ~dev->gso_partial_features; 10136 } 10137 10138 if (!(features & NETIF_F_RXCSUM)) { 10139 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 10140 * successfully merged by hardware must also have the 10141 * checksum verified by hardware. If the user does not 10142 * want to enable RXCSUM, logically, we should disable GRO_HW. 10143 */ 10144 if (features & NETIF_F_GRO_HW) { 10145 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 10146 features &= ~NETIF_F_GRO_HW; 10147 } 10148 } 10149 10150 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 10151 if (features & NETIF_F_RXFCS) { 10152 if (features & NETIF_F_LRO) { 10153 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 10154 features &= ~NETIF_F_LRO; 10155 } 10156 10157 if (features & NETIF_F_GRO_HW) { 10158 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 10159 features &= ~NETIF_F_GRO_HW; 10160 } 10161 } 10162 10163 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 10164 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 10165 features &= ~NETIF_F_LRO; 10166 } 10167 10168 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) { 10169 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 10170 features &= ~NETIF_F_HW_TLS_TX; 10171 } 10172 10173 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 10174 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 10175 features &= ~NETIF_F_HW_TLS_RX; 10176 } 10177 10178 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) { 10179 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n"); 10180 features &= ~NETIF_F_GSO_UDP_L4; 10181 } 10182 10183 return features; 10184 } 10185 10186 int __netdev_update_features(struct net_device *dev) 10187 { 10188 struct net_device *upper, *lower; 10189 netdev_features_t features; 10190 struct list_head *iter; 10191 int err = -1; 10192 10193 ASSERT_RTNL(); 10194 10195 features = netdev_get_wanted_features(dev); 10196 10197 if (dev->netdev_ops->ndo_fix_features) 10198 features = dev->netdev_ops->ndo_fix_features(dev, features); 10199 10200 /* driver might be less strict about feature dependencies */ 10201 features = netdev_fix_features(dev, features); 10202 10203 /* some features can't be enabled if they're off on an upper device */ 10204 netdev_for_each_upper_dev_rcu(dev, upper, iter) 10205 features = netdev_sync_upper_features(dev, upper, features); 10206 10207 if (dev->features == features) 10208 goto sync_lower; 10209 10210 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 10211 &dev->features, &features); 10212 10213 if (dev->netdev_ops->ndo_set_features) 10214 err = dev->netdev_ops->ndo_set_features(dev, features); 10215 else 10216 err = 0; 10217 10218 if (unlikely(err < 0)) { 10219 netdev_err(dev, 10220 "set_features() failed (%d); wanted %pNF, left %pNF\n", 10221 err, &features, &dev->features); 10222 /* return non-0 since some features might have changed and 10223 * it's better to fire a spurious notification than miss it 10224 */ 10225 return -1; 10226 } 10227 10228 sync_lower: 10229 /* some features must be disabled on lower devices when disabled 10230 * on an upper device (think: bonding master or bridge) 10231 */ 10232 netdev_for_each_lower_dev(dev, lower, iter) 10233 netdev_sync_lower_features(dev, lower, features); 10234 10235 if (!err) { 10236 netdev_features_t diff = features ^ dev->features; 10237 10238 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10239 /* udp_tunnel_{get,drop}_rx_info both need 10240 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10241 * device, or they won't do anything. 10242 * Thus we need to update dev->features 10243 * *before* calling udp_tunnel_get_rx_info, 10244 * but *after* calling udp_tunnel_drop_rx_info. 10245 */ 10246 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10247 dev->features = features; 10248 udp_tunnel_get_rx_info(dev); 10249 } else { 10250 udp_tunnel_drop_rx_info(dev); 10251 } 10252 } 10253 10254 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10255 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10256 dev->features = features; 10257 err |= vlan_get_rx_ctag_filter_info(dev); 10258 } else { 10259 vlan_drop_rx_ctag_filter_info(dev); 10260 } 10261 } 10262 10263 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10264 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10265 dev->features = features; 10266 err |= vlan_get_rx_stag_filter_info(dev); 10267 } else { 10268 vlan_drop_rx_stag_filter_info(dev); 10269 } 10270 } 10271 10272 dev->features = features; 10273 } 10274 10275 return err < 0 ? 0 : 1; 10276 } 10277 10278 /** 10279 * netdev_update_features - recalculate device features 10280 * @dev: the device to check 10281 * 10282 * Recalculate dev->features set and send notifications if it 10283 * has changed. Should be called after driver or hardware dependent 10284 * conditions might have changed that influence the features. 10285 */ 10286 void netdev_update_features(struct net_device *dev) 10287 { 10288 if (__netdev_update_features(dev)) 10289 netdev_features_change(dev); 10290 } 10291 EXPORT_SYMBOL(netdev_update_features); 10292 10293 /** 10294 * netdev_change_features - recalculate device features 10295 * @dev: the device to check 10296 * 10297 * Recalculate dev->features set and send notifications even 10298 * if they have not changed. Should be called instead of 10299 * netdev_update_features() if also dev->vlan_features might 10300 * have changed to allow the changes to be propagated to stacked 10301 * VLAN devices. 10302 */ 10303 void netdev_change_features(struct net_device *dev) 10304 { 10305 __netdev_update_features(dev); 10306 netdev_features_change(dev); 10307 } 10308 EXPORT_SYMBOL(netdev_change_features); 10309 10310 /** 10311 * netif_stacked_transfer_operstate - transfer operstate 10312 * @rootdev: the root or lower level device to transfer state from 10313 * @dev: the device to transfer operstate to 10314 * 10315 * Transfer operational state from root to device. This is normally 10316 * called when a stacking relationship exists between the root 10317 * device and the device(a leaf device). 10318 */ 10319 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10320 struct net_device *dev) 10321 { 10322 if (rootdev->operstate == IF_OPER_DORMANT) 10323 netif_dormant_on(dev); 10324 else 10325 netif_dormant_off(dev); 10326 10327 if (rootdev->operstate == IF_OPER_TESTING) 10328 netif_testing_on(dev); 10329 else 10330 netif_testing_off(dev); 10331 10332 if (netif_carrier_ok(rootdev)) 10333 netif_carrier_on(dev); 10334 else 10335 netif_carrier_off(dev); 10336 } 10337 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10338 10339 static int netif_alloc_rx_queues(struct net_device *dev) 10340 { 10341 unsigned int i, count = dev->num_rx_queues; 10342 struct netdev_rx_queue *rx; 10343 size_t sz = count * sizeof(*rx); 10344 int err = 0; 10345 10346 BUG_ON(count < 1); 10347 10348 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10349 if (!rx) 10350 return -ENOMEM; 10351 10352 dev->_rx = rx; 10353 10354 for (i = 0; i < count; i++) { 10355 rx[i].dev = dev; 10356 10357 /* XDP RX-queue setup */ 10358 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10359 if (err < 0) 10360 goto err_rxq_info; 10361 } 10362 return 0; 10363 10364 err_rxq_info: 10365 /* Rollback successful reg's and free other resources */ 10366 while (i--) 10367 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10368 kvfree(dev->_rx); 10369 dev->_rx = NULL; 10370 return err; 10371 } 10372 10373 static void netif_free_rx_queues(struct net_device *dev) 10374 { 10375 unsigned int i, count = dev->num_rx_queues; 10376 10377 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10378 if (!dev->_rx) 10379 return; 10380 10381 for (i = 0; i < count; i++) 10382 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10383 10384 kvfree(dev->_rx); 10385 } 10386 10387 static void netdev_init_one_queue(struct net_device *dev, 10388 struct netdev_queue *queue, void *_unused) 10389 { 10390 /* Initialize queue lock */ 10391 spin_lock_init(&queue->_xmit_lock); 10392 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10393 queue->xmit_lock_owner = -1; 10394 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10395 queue->dev = dev; 10396 #ifdef CONFIG_BQL 10397 dql_init(&queue->dql, HZ); 10398 #endif 10399 } 10400 10401 static void netif_free_tx_queues(struct net_device *dev) 10402 { 10403 kvfree(dev->_tx); 10404 } 10405 10406 static int netif_alloc_netdev_queues(struct net_device *dev) 10407 { 10408 unsigned int count = dev->num_tx_queues; 10409 struct netdev_queue *tx; 10410 size_t sz = count * sizeof(*tx); 10411 10412 if (count < 1 || count > 0xffff) 10413 return -EINVAL; 10414 10415 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10416 if (!tx) 10417 return -ENOMEM; 10418 10419 dev->_tx = tx; 10420 10421 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10422 spin_lock_init(&dev->tx_global_lock); 10423 10424 return 0; 10425 } 10426 10427 void netif_tx_stop_all_queues(struct net_device *dev) 10428 { 10429 unsigned int i; 10430 10431 for (i = 0; i < dev->num_tx_queues; i++) { 10432 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10433 10434 netif_tx_stop_queue(txq); 10435 } 10436 } 10437 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10438 10439 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10440 { 10441 void __percpu *v; 10442 10443 /* Drivers implementing ndo_get_peer_dev must support tstat 10444 * accounting, so that skb_do_redirect() can bump the dev's 10445 * RX stats upon network namespace switch. 10446 */ 10447 if (dev->netdev_ops->ndo_get_peer_dev && 10448 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10449 return -EOPNOTSUPP; 10450 10451 switch (dev->pcpu_stat_type) { 10452 case NETDEV_PCPU_STAT_NONE: 10453 return 0; 10454 case NETDEV_PCPU_STAT_LSTATS: 10455 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10456 break; 10457 case NETDEV_PCPU_STAT_TSTATS: 10458 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10459 break; 10460 case NETDEV_PCPU_STAT_DSTATS: 10461 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10462 break; 10463 default: 10464 return -EINVAL; 10465 } 10466 10467 return v ? 0 : -ENOMEM; 10468 } 10469 10470 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10471 { 10472 switch (dev->pcpu_stat_type) { 10473 case NETDEV_PCPU_STAT_NONE: 10474 return; 10475 case NETDEV_PCPU_STAT_LSTATS: 10476 free_percpu(dev->lstats); 10477 break; 10478 case NETDEV_PCPU_STAT_TSTATS: 10479 free_percpu(dev->tstats); 10480 break; 10481 case NETDEV_PCPU_STAT_DSTATS: 10482 free_percpu(dev->dstats); 10483 break; 10484 } 10485 } 10486 10487 static void netdev_free_phy_link_topology(struct net_device *dev) 10488 { 10489 struct phy_link_topology *topo = dev->link_topo; 10490 10491 if (IS_ENABLED(CONFIG_PHYLIB) && topo) { 10492 xa_destroy(&topo->phys); 10493 kfree(topo); 10494 dev->link_topo = NULL; 10495 } 10496 } 10497 10498 /** 10499 * register_netdevice() - register a network device 10500 * @dev: device to register 10501 * 10502 * Take a prepared network device structure and make it externally accessible. 10503 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10504 * Callers must hold the rtnl lock - you may want register_netdev() 10505 * instead of this. 10506 */ 10507 int register_netdevice(struct net_device *dev) 10508 { 10509 int ret; 10510 struct net *net = dev_net(dev); 10511 10512 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10513 NETDEV_FEATURE_COUNT); 10514 BUG_ON(dev_boot_phase); 10515 ASSERT_RTNL(); 10516 10517 might_sleep(); 10518 10519 /* When net_device's are persistent, this will be fatal. */ 10520 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10521 BUG_ON(!net); 10522 10523 ret = ethtool_check_ops(dev->ethtool_ops); 10524 if (ret) 10525 return ret; 10526 10527 /* rss ctx ID 0 is reserved for the default context, start from 1 */ 10528 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1); 10529 mutex_init(&dev->ethtool->rss_lock); 10530 10531 spin_lock_init(&dev->addr_list_lock); 10532 netdev_set_addr_lockdep_class(dev); 10533 10534 ret = dev_get_valid_name(net, dev, dev->name); 10535 if (ret < 0) 10536 goto out; 10537 10538 ret = -ENOMEM; 10539 dev->name_node = netdev_name_node_head_alloc(dev); 10540 if (!dev->name_node) 10541 goto out; 10542 10543 /* Init, if this function is available */ 10544 if (dev->netdev_ops->ndo_init) { 10545 ret = dev->netdev_ops->ndo_init(dev); 10546 if (ret) { 10547 if (ret > 0) 10548 ret = -EIO; 10549 goto err_free_name; 10550 } 10551 } 10552 10553 if (((dev->hw_features | dev->features) & 10554 NETIF_F_HW_VLAN_CTAG_FILTER) && 10555 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10556 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10557 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10558 ret = -EINVAL; 10559 goto err_uninit; 10560 } 10561 10562 ret = netdev_do_alloc_pcpu_stats(dev); 10563 if (ret) 10564 goto err_uninit; 10565 10566 ret = dev_index_reserve(net, dev->ifindex); 10567 if (ret < 0) 10568 goto err_free_pcpu; 10569 dev->ifindex = ret; 10570 10571 /* Transfer changeable features to wanted_features and enable 10572 * software offloads (GSO and GRO). 10573 */ 10574 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10575 dev->features |= NETIF_F_SOFT_FEATURES; 10576 10577 if (dev->udp_tunnel_nic_info) { 10578 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10579 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10580 } 10581 10582 dev->wanted_features = dev->features & dev->hw_features; 10583 10584 if (!(dev->flags & IFF_LOOPBACK)) 10585 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10586 10587 /* If IPv4 TCP segmentation offload is supported we should also 10588 * allow the device to enable segmenting the frame with the option 10589 * of ignoring a static IP ID value. This doesn't enable the 10590 * feature itself but allows the user to enable it later. 10591 */ 10592 if (dev->hw_features & NETIF_F_TSO) 10593 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10594 if (dev->vlan_features & NETIF_F_TSO) 10595 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10596 if (dev->mpls_features & NETIF_F_TSO) 10597 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10598 if (dev->hw_enc_features & NETIF_F_TSO) 10599 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10600 10601 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10602 */ 10603 dev->vlan_features |= NETIF_F_HIGHDMA; 10604 10605 /* Make NETIF_F_SG inheritable to tunnel devices. 10606 */ 10607 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10608 10609 /* Make NETIF_F_SG inheritable to MPLS. 10610 */ 10611 dev->mpls_features |= NETIF_F_SG; 10612 10613 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10614 ret = notifier_to_errno(ret); 10615 if (ret) 10616 goto err_ifindex_release; 10617 10618 ret = netdev_register_kobject(dev); 10619 10620 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); 10621 10622 if (ret) 10623 goto err_uninit_notify; 10624 10625 __netdev_update_features(dev); 10626 10627 /* 10628 * Default initial state at registry is that the 10629 * device is present. 10630 */ 10631 10632 set_bit(__LINK_STATE_PRESENT, &dev->state); 10633 10634 linkwatch_init_dev(dev); 10635 10636 dev_init_scheduler(dev); 10637 10638 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10639 list_netdevice(dev); 10640 10641 add_device_randomness(dev->dev_addr, dev->addr_len); 10642 10643 /* If the device has permanent device address, driver should 10644 * set dev_addr and also addr_assign_type should be set to 10645 * NET_ADDR_PERM (default value). 10646 */ 10647 if (dev->addr_assign_type == NET_ADDR_PERM) 10648 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10649 10650 /* Notify protocols, that a new device appeared. */ 10651 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10652 ret = notifier_to_errno(ret); 10653 if (ret) { 10654 /* Expect explicit free_netdev() on failure */ 10655 dev->needs_free_netdev = false; 10656 unregister_netdevice_queue(dev, NULL); 10657 goto out; 10658 } 10659 /* 10660 * Prevent userspace races by waiting until the network 10661 * device is fully setup before sending notifications. 10662 */ 10663 if (!dev->rtnl_link_ops || 10664 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10665 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10666 10667 out: 10668 return ret; 10669 10670 err_uninit_notify: 10671 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10672 err_ifindex_release: 10673 dev_index_release(net, dev->ifindex); 10674 err_free_pcpu: 10675 netdev_do_free_pcpu_stats(dev); 10676 err_uninit: 10677 if (dev->netdev_ops->ndo_uninit) 10678 dev->netdev_ops->ndo_uninit(dev); 10679 if (dev->priv_destructor) 10680 dev->priv_destructor(dev); 10681 err_free_name: 10682 netdev_name_node_free(dev->name_node); 10683 goto out; 10684 } 10685 EXPORT_SYMBOL(register_netdevice); 10686 10687 /* Initialize the core of a dummy net device. 10688 * This is useful if you are calling this function after alloc_netdev(), 10689 * since it does not memset the net_device fields. 10690 */ 10691 static void init_dummy_netdev_core(struct net_device *dev) 10692 { 10693 /* make sure we BUG if trying to hit standard 10694 * register/unregister code path 10695 */ 10696 dev->reg_state = NETREG_DUMMY; 10697 10698 /* NAPI wants this */ 10699 INIT_LIST_HEAD(&dev->napi_list); 10700 10701 /* a dummy interface is started by default */ 10702 set_bit(__LINK_STATE_PRESENT, &dev->state); 10703 set_bit(__LINK_STATE_START, &dev->state); 10704 10705 /* napi_busy_loop stats accounting wants this */ 10706 dev_net_set(dev, &init_net); 10707 10708 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10709 * because users of this 'device' dont need to change 10710 * its refcount. 10711 */ 10712 } 10713 10714 /** 10715 * init_dummy_netdev - init a dummy network device for NAPI 10716 * @dev: device to init 10717 * 10718 * This takes a network device structure and initializes the minimum 10719 * amount of fields so it can be used to schedule NAPI polls without 10720 * registering a full blown interface. This is to be used by drivers 10721 * that need to tie several hardware interfaces to a single NAPI 10722 * poll scheduler due to HW limitations. 10723 */ 10724 void init_dummy_netdev(struct net_device *dev) 10725 { 10726 /* Clear everything. Note we don't initialize spinlocks 10727 * as they aren't supposed to be taken by any of the 10728 * NAPI code and this dummy netdev is supposed to be 10729 * only ever used for NAPI polls 10730 */ 10731 memset(dev, 0, sizeof(struct net_device)); 10732 init_dummy_netdev_core(dev); 10733 } 10734 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10735 10736 /** 10737 * register_netdev - register a network device 10738 * @dev: device to register 10739 * 10740 * Take a completed network device structure and add it to the kernel 10741 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10742 * chain. 0 is returned on success. A negative errno code is returned 10743 * on a failure to set up the device, or if the name is a duplicate. 10744 * 10745 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10746 * and expands the device name if you passed a format string to 10747 * alloc_netdev. 10748 */ 10749 int register_netdev(struct net_device *dev) 10750 { 10751 struct net *net = dev_net(dev); 10752 int err; 10753 10754 if (rtnl_net_lock_killable(net)) 10755 return -EINTR; 10756 10757 err = register_netdevice(dev); 10758 10759 rtnl_net_unlock(net); 10760 10761 return err; 10762 } 10763 EXPORT_SYMBOL(register_netdev); 10764 10765 int netdev_refcnt_read(const struct net_device *dev) 10766 { 10767 #ifdef CONFIG_PCPU_DEV_REFCNT 10768 int i, refcnt = 0; 10769 10770 for_each_possible_cpu(i) 10771 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10772 return refcnt; 10773 #else 10774 return refcount_read(&dev->dev_refcnt); 10775 #endif 10776 } 10777 EXPORT_SYMBOL(netdev_refcnt_read); 10778 10779 int netdev_unregister_timeout_secs __read_mostly = 10; 10780 10781 #define WAIT_REFS_MIN_MSECS 1 10782 #define WAIT_REFS_MAX_MSECS 250 10783 /** 10784 * netdev_wait_allrefs_any - wait until all references are gone. 10785 * @list: list of net_devices to wait on 10786 * 10787 * This is called when unregistering network devices. 10788 * 10789 * Any protocol or device that holds a reference should register 10790 * for netdevice notification, and cleanup and put back the 10791 * reference if they receive an UNREGISTER event. 10792 * We can get stuck here if buggy protocols don't correctly 10793 * call dev_put. 10794 */ 10795 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10796 { 10797 unsigned long rebroadcast_time, warning_time; 10798 struct net_device *dev; 10799 int wait = 0; 10800 10801 rebroadcast_time = warning_time = jiffies; 10802 10803 list_for_each_entry(dev, list, todo_list) 10804 if (netdev_refcnt_read(dev) == 1) 10805 return dev; 10806 10807 while (true) { 10808 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10809 rtnl_lock(); 10810 10811 /* Rebroadcast unregister notification */ 10812 list_for_each_entry(dev, list, todo_list) 10813 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10814 10815 __rtnl_unlock(); 10816 rcu_barrier(); 10817 rtnl_lock(); 10818 10819 list_for_each_entry(dev, list, todo_list) 10820 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10821 &dev->state)) { 10822 /* We must not have linkwatch events 10823 * pending on unregister. If this 10824 * happens, we simply run the queue 10825 * unscheduled, resulting in a noop 10826 * for this device. 10827 */ 10828 linkwatch_run_queue(); 10829 break; 10830 } 10831 10832 __rtnl_unlock(); 10833 10834 rebroadcast_time = jiffies; 10835 } 10836 10837 rcu_barrier(); 10838 10839 if (!wait) { 10840 wait = WAIT_REFS_MIN_MSECS; 10841 } else { 10842 msleep(wait); 10843 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10844 } 10845 10846 list_for_each_entry(dev, list, todo_list) 10847 if (netdev_refcnt_read(dev) == 1) 10848 return dev; 10849 10850 if (time_after(jiffies, warning_time + 10851 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10852 list_for_each_entry(dev, list, todo_list) { 10853 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10854 dev->name, netdev_refcnt_read(dev)); 10855 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10856 } 10857 10858 warning_time = jiffies; 10859 } 10860 } 10861 } 10862 10863 /* The sequence is: 10864 * 10865 * rtnl_lock(); 10866 * ... 10867 * register_netdevice(x1); 10868 * register_netdevice(x2); 10869 * ... 10870 * unregister_netdevice(y1); 10871 * unregister_netdevice(y2); 10872 * ... 10873 * rtnl_unlock(); 10874 * free_netdev(y1); 10875 * free_netdev(y2); 10876 * 10877 * We are invoked by rtnl_unlock(). 10878 * This allows us to deal with problems: 10879 * 1) We can delete sysfs objects which invoke hotplug 10880 * without deadlocking with linkwatch via keventd. 10881 * 2) Since we run with the RTNL semaphore not held, we can sleep 10882 * safely in order to wait for the netdev refcnt to drop to zero. 10883 * 10884 * We must not return until all unregister events added during 10885 * the interval the lock was held have been completed. 10886 */ 10887 void netdev_run_todo(void) 10888 { 10889 struct net_device *dev, *tmp; 10890 struct list_head list; 10891 int cnt; 10892 #ifdef CONFIG_LOCKDEP 10893 struct list_head unlink_list; 10894 10895 list_replace_init(&net_unlink_list, &unlink_list); 10896 10897 while (!list_empty(&unlink_list)) { 10898 struct net_device *dev = list_first_entry(&unlink_list, 10899 struct net_device, 10900 unlink_list); 10901 list_del_init(&dev->unlink_list); 10902 dev->nested_level = dev->lower_level - 1; 10903 } 10904 #endif 10905 10906 /* Snapshot list, allow later requests */ 10907 list_replace_init(&net_todo_list, &list); 10908 10909 __rtnl_unlock(); 10910 10911 /* Wait for rcu callbacks to finish before next phase */ 10912 if (!list_empty(&list)) 10913 rcu_barrier(); 10914 10915 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10916 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10917 netdev_WARN(dev, "run_todo but not unregistering\n"); 10918 list_del(&dev->todo_list); 10919 continue; 10920 } 10921 10922 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); 10923 linkwatch_sync_dev(dev); 10924 } 10925 10926 cnt = 0; 10927 while (!list_empty(&list)) { 10928 dev = netdev_wait_allrefs_any(&list); 10929 list_del(&dev->todo_list); 10930 10931 /* paranoia */ 10932 BUG_ON(netdev_refcnt_read(dev) != 1); 10933 BUG_ON(!list_empty(&dev->ptype_all)); 10934 BUG_ON(!list_empty(&dev->ptype_specific)); 10935 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10936 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10937 10938 netdev_do_free_pcpu_stats(dev); 10939 if (dev->priv_destructor) 10940 dev->priv_destructor(dev); 10941 if (dev->needs_free_netdev) 10942 free_netdev(dev); 10943 10944 cnt++; 10945 10946 /* Free network device */ 10947 kobject_put(&dev->dev.kobj); 10948 } 10949 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) 10950 wake_up(&netdev_unregistering_wq); 10951 } 10952 10953 /* Collate per-cpu network dstats statistics 10954 * 10955 * Read per-cpu network statistics from dev->dstats and populate the related 10956 * fields in @s. 10957 */ 10958 static void dev_fetch_dstats(struct rtnl_link_stats64 *s, 10959 const struct pcpu_dstats __percpu *dstats) 10960 { 10961 int cpu; 10962 10963 for_each_possible_cpu(cpu) { 10964 u64 rx_packets, rx_bytes, rx_drops; 10965 u64 tx_packets, tx_bytes, tx_drops; 10966 const struct pcpu_dstats *stats; 10967 unsigned int start; 10968 10969 stats = per_cpu_ptr(dstats, cpu); 10970 do { 10971 start = u64_stats_fetch_begin(&stats->syncp); 10972 rx_packets = u64_stats_read(&stats->rx_packets); 10973 rx_bytes = u64_stats_read(&stats->rx_bytes); 10974 rx_drops = u64_stats_read(&stats->rx_drops); 10975 tx_packets = u64_stats_read(&stats->tx_packets); 10976 tx_bytes = u64_stats_read(&stats->tx_bytes); 10977 tx_drops = u64_stats_read(&stats->tx_drops); 10978 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10979 10980 s->rx_packets += rx_packets; 10981 s->rx_bytes += rx_bytes; 10982 s->rx_dropped += rx_drops; 10983 s->tx_packets += tx_packets; 10984 s->tx_bytes += tx_bytes; 10985 s->tx_dropped += tx_drops; 10986 } 10987 } 10988 10989 /* ndo_get_stats64 implementation for dtstats-based accounting. 10990 * 10991 * Populate @s from dev->stats and dev->dstats. This is used internally by the 10992 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection. 10993 */ 10994 static void dev_get_dstats64(const struct net_device *dev, 10995 struct rtnl_link_stats64 *s) 10996 { 10997 netdev_stats_to_stats64(s, &dev->stats); 10998 dev_fetch_dstats(s, dev->dstats); 10999 } 11000 11001 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 11002 * all the same fields in the same order as net_device_stats, with only 11003 * the type differing, but rtnl_link_stats64 may have additional fields 11004 * at the end for newer counters. 11005 */ 11006 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 11007 const struct net_device_stats *netdev_stats) 11008 { 11009 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 11010 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 11011 u64 *dst = (u64 *)stats64; 11012 11013 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 11014 for (i = 0; i < n; i++) 11015 dst[i] = (unsigned long)atomic_long_read(&src[i]); 11016 /* zero out counters that only exist in rtnl_link_stats64 */ 11017 memset((char *)stats64 + n * sizeof(u64), 0, 11018 sizeof(*stats64) - n * sizeof(u64)); 11019 } 11020 EXPORT_SYMBOL(netdev_stats_to_stats64); 11021 11022 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 11023 struct net_device *dev) 11024 { 11025 struct net_device_core_stats __percpu *p; 11026 11027 p = alloc_percpu_gfp(struct net_device_core_stats, 11028 GFP_ATOMIC | __GFP_NOWARN); 11029 11030 if (p && cmpxchg(&dev->core_stats, NULL, p)) 11031 free_percpu(p); 11032 11033 /* This READ_ONCE() pairs with the cmpxchg() above */ 11034 return READ_ONCE(dev->core_stats); 11035 } 11036 11037 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 11038 { 11039 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11040 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 11041 unsigned long __percpu *field; 11042 11043 if (unlikely(!p)) { 11044 p = netdev_core_stats_alloc(dev); 11045 if (!p) 11046 return; 11047 } 11048 11049 field = (unsigned long __percpu *)((void __percpu *)p + offset); 11050 this_cpu_inc(*field); 11051 } 11052 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 11053 11054 /** 11055 * dev_get_stats - get network device statistics 11056 * @dev: device to get statistics from 11057 * @storage: place to store stats 11058 * 11059 * Get network statistics from device. Return @storage. 11060 * The device driver may provide its own method by setting 11061 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 11062 * otherwise the internal statistics structure is used. 11063 */ 11064 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 11065 struct rtnl_link_stats64 *storage) 11066 { 11067 const struct net_device_ops *ops = dev->netdev_ops; 11068 const struct net_device_core_stats __percpu *p; 11069 11070 if (ops->ndo_get_stats64) { 11071 memset(storage, 0, sizeof(*storage)); 11072 ops->ndo_get_stats64(dev, storage); 11073 } else if (ops->ndo_get_stats) { 11074 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 11075 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { 11076 dev_get_tstats64(dev, storage); 11077 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) { 11078 dev_get_dstats64(dev, storage); 11079 } else { 11080 netdev_stats_to_stats64(storage, &dev->stats); 11081 } 11082 11083 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11084 p = READ_ONCE(dev->core_stats); 11085 if (p) { 11086 const struct net_device_core_stats *core_stats; 11087 int i; 11088 11089 for_each_possible_cpu(i) { 11090 core_stats = per_cpu_ptr(p, i); 11091 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 11092 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 11093 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 11094 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 11095 } 11096 } 11097 return storage; 11098 } 11099 EXPORT_SYMBOL(dev_get_stats); 11100 11101 /** 11102 * dev_fetch_sw_netstats - get per-cpu network device statistics 11103 * @s: place to store stats 11104 * @netstats: per-cpu network stats to read from 11105 * 11106 * Read per-cpu network statistics and populate the related fields in @s. 11107 */ 11108 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 11109 const struct pcpu_sw_netstats __percpu *netstats) 11110 { 11111 int cpu; 11112 11113 for_each_possible_cpu(cpu) { 11114 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 11115 const struct pcpu_sw_netstats *stats; 11116 unsigned int start; 11117 11118 stats = per_cpu_ptr(netstats, cpu); 11119 do { 11120 start = u64_stats_fetch_begin(&stats->syncp); 11121 rx_packets = u64_stats_read(&stats->rx_packets); 11122 rx_bytes = u64_stats_read(&stats->rx_bytes); 11123 tx_packets = u64_stats_read(&stats->tx_packets); 11124 tx_bytes = u64_stats_read(&stats->tx_bytes); 11125 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11126 11127 s->rx_packets += rx_packets; 11128 s->rx_bytes += rx_bytes; 11129 s->tx_packets += tx_packets; 11130 s->tx_bytes += tx_bytes; 11131 } 11132 } 11133 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 11134 11135 /** 11136 * dev_get_tstats64 - ndo_get_stats64 implementation 11137 * @dev: device to get statistics from 11138 * @s: place to store stats 11139 * 11140 * Populate @s from dev->stats and dev->tstats. Can be used as 11141 * ndo_get_stats64() callback. 11142 */ 11143 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 11144 { 11145 netdev_stats_to_stats64(s, &dev->stats); 11146 dev_fetch_sw_netstats(s, dev->tstats); 11147 } 11148 EXPORT_SYMBOL_GPL(dev_get_tstats64); 11149 11150 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 11151 { 11152 struct netdev_queue *queue = dev_ingress_queue(dev); 11153 11154 #ifdef CONFIG_NET_CLS_ACT 11155 if (queue) 11156 return queue; 11157 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 11158 if (!queue) 11159 return NULL; 11160 netdev_init_one_queue(dev, queue, NULL); 11161 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 11162 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 11163 rcu_assign_pointer(dev->ingress_queue, queue); 11164 #endif 11165 return queue; 11166 } 11167 11168 static const struct ethtool_ops default_ethtool_ops; 11169 11170 void netdev_set_default_ethtool_ops(struct net_device *dev, 11171 const struct ethtool_ops *ops) 11172 { 11173 if (dev->ethtool_ops == &default_ethtool_ops) 11174 dev->ethtool_ops = ops; 11175 } 11176 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 11177 11178 /** 11179 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 11180 * @dev: netdev to enable the IRQ coalescing on 11181 * 11182 * Sets a conservative default for SW IRQ coalescing. Users can use 11183 * sysfs attributes to override the default values. 11184 */ 11185 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 11186 { 11187 WARN_ON(dev->reg_state == NETREG_REGISTERED); 11188 11189 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 11190 netdev_set_gro_flush_timeout(dev, 20000); 11191 netdev_set_defer_hard_irqs(dev, 1); 11192 } 11193 } 11194 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 11195 11196 /** 11197 * alloc_netdev_mqs - allocate network device 11198 * @sizeof_priv: size of private data to allocate space for 11199 * @name: device name format string 11200 * @name_assign_type: origin of device name 11201 * @setup: callback to initialize device 11202 * @txqs: the number of TX subqueues to allocate 11203 * @rxqs: the number of RX subqueues to allocate 11204 * 11205 * Allocates a struct net_device with private data area for driver use 11206 * and performs basic initialization. Also allocates subqueue structs 11207 * for each queue on the device. 11208 */ 11209 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 11210 unsigned char name_assign_type, 11211 void (*setup)(struct net_device *), 11212 unsigned int txqs, unsigned int rxqs) 11213 { 11214 struct net_device *dev; 11215 size_t napi_config_sz; 11216 unsigned int maxqs; 11217 11218 BUG_ON(strlen(name) >= sizeof(dev->name)); 11219 11220 if (txqs < 1) { 11221 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 11222 return NULL; 11223 } 11224 11225 if (rxqs < 1) { 11226 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 11227 return NULL; 11228 } 11229 11230 maxqs = max(txqs, rxqs); 11231 11232 dev = kvzalloc(struct_size(dev, priv, sizeof_priv), 11233 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 11234 if (!dev) 11235 return NULL; 11236 11237 dev->priv_len = sizeof_priv; 11238 11239 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 11240 #ifdef CONFIG_PCPU_DEV_REFCNT 11241 dev->pcpu_refcnt = alloc_percpu(int); 11242 if (!dev->pcpu_refcnt) 11243 goto free_dev; 11244 __dev_hold(dev); 11245 #else 11246 refcount_set(&dev->dev_refcnt, 1); 11247 #endif 11248 11249 if (dev_addr_init(dev)) 11250 goto free_pcpu; 11251 11252 dev_mc_init(dev); 11253 dev_uc_init(dev); 11254 11255 dev_net_set(dev, &init_net); 11256 11257 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 11258 dev->xdp_zc_max_segs = 1; 11259 dev->gso_max_segs = GSO_MAX_SEGS; 11260 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 11261 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 11262 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 11263 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 11264 dev->tso_max_segs = TSO_MAX_SEGS; 11265 dev->upper_level = 1; 11266 dev->lower_level = 1; 11267 #ifdef CONFIG_LOCKDEP 11268 dev->nested_level = 0; 11269 INIT_LIST_HEAD(&dev->unlink_list); 11270 #endif 11271 11272 INIT_LIST_HEAD(&dev->napi_list); 11273 INIT_LIST_HEAD(&dev->unreg_list); 11274 INIT_LIST_HEAD(&dev->close_list); 11275 INIT_LIST_HEAD(&dev->link_watch_list); 11276 INIT_LIST_HEAD(&dev->adj_list.upper); 11277 INIT_LIST_HEAD(&dev->adj_list.lower); 11278 INIT_LIST_HEAD(&dev->ptype_all); 11279 INIT_LIST_HEAD(&dev->ptype_specific); 11280 INIT_LIST_HEAD(&dev->net_notifier_list); 11281 #ifdef CONFIG_NET_SCHED 11282 hash_init(dev->qdisc_hash); 11283 #endif 11284 11285 mutex_init(&dev->lock); 11286 11287 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 11288 setup(dev); 11289 11290 if (!dev->tx_queue_len) { 11291 dev->priv_flags |= IFF_NO_QUEUE; 11292 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 11293 } 11294 11295 dev->num_tx_queues = txqs; 11296 dev->real_num_tx_queues = txqs; 11297 if (netif_alloc_netdev_queues(dev)) 11298 goto free_all; 11299 11300 dev->num_rx_queues = rxqs; 11301 dev->real_num_rx_queues = rxqs; 11302 if (netif_alloc_rx_queues(dev)) 11303 goto free_all; 11304 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT); 11305 if (!dev->ethtool) 11306 goto free_all; 11307 11308 napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config)); 11309 dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT); 11310 if (!dev->napi_config) 11311 goto free_all; 11312 11313 strscpy(dev->name, name); 11314 dev->name_assign_type = name_assign_type; 11315 dev->group = INIT_NETDEV_GROUP; 11316 if (!dev->ethtool_ops) 11317 dev->ethtool_ops = &default_ethtool_ops; 11318 11319 nf_hook_netdev_init(dev); 11320 11321 return dev; 11322 11323 free_all: 11324 free_netdev(dev); 11325 return NULL; 11326 11327 free_pcpu: 11328 #ifdef CONFIG_PCPU_DEV_REFCNT 11329 free_percpu(dev->pcpu_refcnt); 11330 free_dev: 11331 #endif 11332 kvfree(dev); 11333 return NULL; 11334 } 11335 EXPORT_SYMBOL(alloc_netdev_mqs); 11336 11337 /** 11338 * free_netdev - free network device 11339 * @dev: device 11340 * 11341 * This function does the last stage of destroying an allocated device 11342 * interface. The reference to the device object is released. If this 11343 * is the last reference then it will be freed.Must be called in process 11344 * context. 11345 */ 11346 void free_netdev(struct net_device *dev) 11347 { 11348 struct napi_struct *p, *n; 11349 11350 might_sleep(); 11351 11352 /* When called immediately after register_netdevice() failed the unwind 11353 * handling may still be dismantling the device. Handle that case by 11354 * deferring the free. 11355 */ 11356 if (dev->reg_state == NETREG_UNREGISTERING) { 11357 ASSERT_RTNL(); 11358 dev->needs_free_netdev = true; 11359 return; 11360 } 11361 11362 mutex_destroy(&dev->lock); 11363 11364 kfree(dev->ethtool); 11365 netif_free_tx_queues(dev); 11366 netif_free_rx_queues(dev); 11367 11368 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 11369 11370 /* Flush device addresses */ 11371 dev_addr_flush(dev); 11372 11373 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 11374 netif_napi_del(p); 11375 11376 kvfree(dev->napi_config); 11377 11378 ref_tracker_dir_exit(&dev->refcnt_tracker); 11379 #ifdef CONFIG_PCPU_DEV_REFCNT 11380 free_percpu(dev->pcpu_refcnt); 11381 dev->pcpu_refcnt = NULL; 11382 #endif 11383 free_percpu(dev->core_stats); 11384 dev->core_stats = NULL; 11385 free_percpu(dev->xdp_bulkq); 11386 dev->xdp_bulkq = NULL; 11387 11388 netdev_free_phy_link_topology(dev); 11389 11390 /* Compatibility with error handling in drivers */ 11391 if (dev->reg_state == NETREG_UNINITIALIZED || 11392 dev->reg_state == NETREG_DUMMY) { 11393 kvfree(dev); 11394 return; 11395 } 11396 11397 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 11398 WRITE_ONCE(dev->reg_state, NETREG_RELEASED); 11399 11400 /* will free via device release */ 11401 put_device(&dev->dev); 11402 } 11403 EXPORT_SYMBOL(free_netdev); 11404 11405 /** 11406 * alloc_netdev_dummy - Allocate and initialize a dummy net device. 11407 * @sizeof_priv: size of private data to allocate space for 11408 * 11409 * Return: the allocated net_device on success, NULL otherwise 11410 */ 11411 struct net_device *alloc_netdev_dummy(int sizeof_priv) 11412 { 11413 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN, 11414 init_dummy_netdev_core); 11415 } 11416 EXPORT_SYMBOL_GPL(alloc_netdev_dummy); 11417 11418 /** 11419 * synchronize_net - Synchronize with packet receive processing 11420 * 11421 * Wait for packets currently being received to be done. 11422 * Does not block later packets from starting. 11423 */ 11424 void synchronize_net(void) 11425 { 11426 might_sleep(); 11427 if (rtnl_is_locked()) 11428 synchronize_rcu_expedited(); 11429 else 11430 synchronize_rcu(); 11431 } 11432 EXPORT_SYMBOL(synchronize_net); 11433 11434 static void netdev_rss_contexts_free(struct net_device *dev) 11435 { 11436 struct ethtool_rxfh_context *ctx; 11437 unsigned long context; 11438 11439 mutex_lock(&dev->ethtool->rss_lock); 11440 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) { 11441 struct ethtool_rxfh_param rxfh; 11442 11443 rxfh.indir = ethtool_rxfh_context_indir(ctx); 11444 rxfh.key = ethtool_rxfh_context_key(ctx); 11445 rxfh.hfunc = ctx->hfunc; 11446 rxfh.input_xfrm = ctx->input_xfrm; 11447 rxfh.rss_context = context; 11448 rxfh.rss_delete = true; 11449 11450 xa_erase(&dev->ethtool->rss_ctx, context); 11451 if (dev->ethtool_ops->create_rxfh_context) 11452 dev->ethtool_ops->remove_rxfh_context(dev, ctx, 11453 context, NULL); 11454 else 11455 dev->ethtool_ops->set_rxfh(dev, &rxfh, NULL); 11456 kfree(ctx); 11457 } 11458 xa_destroy(&dev->ethtool->rss_ctx); 11459 mutex_unlock(&dev->ethtool->rss_lock); 11460 } 11461 11462 /** 11463 * unregister_netdevice_queue - remove device from the kernel 11464 * @dev: device 11465 * @head: list 11466 * 11467 * This function shuts down a device interface and removes it 11468 * from the kernel tables. 11469 * If head not NULL, device is queued to be unregistered later. 11470 * 11471 * Callers must hold the rtnl semaphore. You may want 11472 * unregister_netdev() instead of this. 11473 */ 11474 11475 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11476 { 11477 ASSERT_RTNL(); 11478 11479 if (head) { 11480 list_move_tail(&dev->unreg_list, head); 11481 } else { 11482 LIST_HEAD(single); 11483 11484 list_add(&dev->unreg_list, &single); 11485 unregister_netdevice_many(&single); 11486 } 11487 } 11488 EXPORT_SYMBOL(unregister_netdevice_queue); 11489 11490 void unregister_netdevice_many_notify(struct list_head *head, 11491 u32 portid, const struct nlmsghdr *nlh) 11492 { 11493 struct net_device *dev, *tmp; 11494 LIST_HEAD(close_head); 11495 int cnt = 0; 11496 11497 BUG_ON(dev_boot_phase); 11498 ASSERT_RTNL(); 11499 11500 if (list_empty(head)) 11501 return; 11502 11503 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11504 /* Some devices call without registering 11505 * for initialization unwind. Remove those 11506 * devices and proceed with the remaining. 11507 */ 11508 if (dev->reg_state == NETREG_UNINITIALIZED) { 11509 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11510 dev->name, dev); 11511 11512 WARN_ON(1); 11513 list_del(&dev->unreg_list); 11514 continue; 11515 } 11516 dev->dismantle = true; 11517 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11518 } 11519 11520 /* If device is running, close it first. */ 11521 list_for_each_entry(dev, head, unreg_list) 11522 list_add_tail(&dev->close_list, &close_head); 11523 dev_close_many(&close_head, true); 11524 11525 list_for_each_entry(dev, head, unreg_list) { 11526 /* And unlink it from device chain. */ 11527 unlist_netdevice(dev); 11528 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); 11529 } 11530 flush_all_backlogs(); 11531 11532 synchronize_net(); 11533 11534 list_for_each_entry(dev, head, unreg_list) { 11535 struct sk_buff *skb = NULL; 11536 11537 /* Shutdown queueing discipline. */ 11538 dev_shutdown(dev); 11539 dev_tcx_uninstall(dev); 11540 dev_xdp_uninstall(dev); 11541 bpf_dev_bound_netdev_unregister(dev); 11542 dev_dmabuf_uninstall(dev); 11543 11544 netdev_offload_xstats_disable_all(dev); 11545 11546 /* Notify protocols, that we are about to destroy 11547 * this device. They should clean all the things. 11548 */ 11549 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11550 11551 if (!dev->rtnl_link_ops || 11552 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11553 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11554 GFP_KERNEL, NULL, 0, 11555 portid, nlh); 11556 11557 /* 11558 * Flush the unicast and multicast chains 11559 */ 11560 dev_uc_flush(dev); 11561 dev_mc_flush(dev); 11562 11563 netdev_name_node_alt_flush(dev); 11564 netdev_name_node_free(dev->name_node); 11565 11566 netdev_rss_contexts_free(dev); 11567 11568 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11569 11570 if (dev->netdev_ops->ndo_uninit) 11571 dev->netdev_ops->ndo_uninit(dev); 11572 11573 mutex_destroy(&dev->ethtool->rss_lock); 11574 11575 net_shaper_flush_netdev(dev); 11576 11577 if (skb) 11578 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11579 11580 /* Notifier chain MUST detach us all upper devices. */ 11581 WARN_ON(netdev_has_any_upper_dev(dev)); 11582 WARN_ON(netdev_has_any_lower_dev(dev)); 11583 11584 /* Remove entries from kobject tree */ 11585 netdev_unregister_kobject(dev); 11586 #ifdef CONFIG_XPS 11587 /* Remove XPS queueing entries */ 11588 netif_reset_xps_queues_gt(dev, 0); 11589 #endif 11590 } 11591 11592 synchronize_net(); 11593 11594 list_for_each_entry(dev, head, unreg_list) { 11595 netdev_put(dev, &dev->dev_registered_tracker); 11596 net_set_todo(dev); 11597 cnt++; 11598 } 11599 atomic_add(cnt, &dev_unreg_count); 11600 11601 list_del(head); 11602 } 11603 11604 /** 11605 * unregister_netdevice_many - unregister many devices 11606 * @head: list of devices 11607 * 11608 * Note: As most callers use a stack allocated list_head, 11609 * we force a list_del() to make sure stack won't be corrupted later. 11610 */ 11611 void unregister_netdevice_many(struct list_head *head) 11612 { 11613 unregister_netdevice_many_notify(head, 0, NULL); 11614 } 11615 EXPORT_SYMBOL(unregister_netdevice_many); 11616 11617 /** 11618 * unregister_netdev - remove device from the kernel 11619 * @dev: device 11620 * 11621 * This function shuts down a device interface and removes it 11622 * from the kernel tables. 11623 * 11624 * This is just a wrapper for unregister_netdevice that takes 11625 * the rtnl semaphore. In general you want to use this and not 11626 * unregister_netdevice. 11627 */ 11628 void unregister_netdev(struct net_device *dev) 11629 { 11630 struct net *net = dev_net(dev); 11631 11632 rtnl_net_lock(net); 11633 unregister_netdevice(dev); 11634 rtnl_net_unlock(net); 11635 } 11636 EXPORT_SYMBOL(unregister_netdev); 11637 11638 /** 11639 * __dev_change_net_namespace - move device to different nethost namespace 11640 * @dev: device 11641 * @net: network namespace 11642 * @pat: If not NULL name pattern to try if the current device name 11643 * is already taken in the destination network namespace. 11644 * @new_ifindex: If not zero, specifies device index in the target 11645 * namespace. 11646 * 11647 * This function shuts down a device interface and moves it 11648 * to a new network namespace. On success 0 is returned, on 11649 * a failure a netagive errno code is returned. 11650 * 11651 * Callers must hold the rtnl semaphore. 11652 */ 11653 11654 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11655 const char *pat, int new_ifindex) 11656 { 11657 struct netdev_name_node *name_node; 11658 struct net *net_old = dev_net(dev); 11659 char new_name[IFNAMSIZ] = {}; 11660 int err, new_nsid; 11661 11662 ASSERT_RTNL(); 11663 11664 /* Don't allow namespace local devices to be moved. */ 11665 err = -EINVAL; 11666 if (dev->netns_local) 11667 goto out; 11668 11669 /* Ensure the device has been registered */ 11670 if (dev->reg_state != NETREG_REGISTERED) 11671 goto out; 11672 11673 /* Get out if there is nothing todo */ 11674 err = 0; 11675 if (net_eq(net_old, net)) 11676 goto out; 11677 11678 /* Pick the destination device name, and ensure 11679 * we can use it in the destination network namespace. 11680 */ 11681 err = -EEXIST; 11682 if (netdev_name_in_use(net, dev->name)) { 11683 /* We get here if we can't use the current device name */ 11684 if (!pat) 11685 goto out; 11686 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11687 if (err < 0) 11688 goto out; 11689 } 11690 /* Check that none of the altnames conflicts. */ 11691 err = -EEXIST; 11692 netdev_for_each_altname(dev, name_node) 11693 if (netdev_name_in_use(net, name_node->name)) 11694 goto out; 11695 11696 /* Check that new_ifindex isn't used yet. */ 11697 if (new_ifindex) { 11698 err = dev_index_reserve(net, new_ifindex); 11699 if (err < 0) 11700 goto out; 11701 } else { 11702 /* If there is an ifindex conflict assign a new one */ 11703 err = dev_index_reserve(net, dev->ifindex); 11704 if (err == -EBUSY) 11705 err = dev_index_reserve(net, 0); 11706 if (err < 0) 11707 goto out; 11708 new_ifindex = err; 11709 } 11710 11711 /* 11712 * And now a mini version of register_netdevice unregister_netdevice. 11713 */ 11714 11715 /* If device is running close it first. */ 11716 dev_close(dev); 11717 11718 /* And unlink it from device chain */ 11719 unlist_netdevice(dev); 11720 11721 synchronize_net(); 11722 11723 /* Shutdown queueing discipline. */ 11724 dev_shutdown(dev); 11725 11726 /* Notify protocols, that we are about to destroy 11727 * this device. They should clean all the things. 11728 * 11729 * Note that dev->reg_state stays at NETREG_REGISTERED. 11730 * This is wanted because this way 8021q and macvlan know 11731 * the device is just moving and can keep their slaves up. 11732 */ 11733 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11734 rcu_barrier(); 11735 11736 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11737 11738 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11739 new_ifindex); 11740 11741 /* 11742 * Flush the unicast and multicast chains 11743 */ 11744 dev_uc_flush(dev); 11745 dev_mc_flush(dev); 11746 11747 /* Send a netdev-removed uevent to the old namespace */ 11748 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11749 netdev_adjacent_del_links(dev); 11750 11751 /* Move per-net netdevice notifiers that are following the netdevice */ 11752 move_netdevice_notifiers_dev_net(dev, net); 11753 11754 /* Actually switch the network namespace */ 11755 dev_net_set(dev, net); 11756 dev->ifindex = new_ifindex; 11757 11758 if (new_name[0]) { 11759 /* Rename the netdev to prepared name */ 11760 write_seqlock_bh(&netdev_rename_lock); 11761 strscpy(dev->name, new_name, IFNAMSIZ); 11762 write_sequnlock_bh(&netdev_rename_lock); 11763 } 11764 11765 /* Fixup kobjects */ 11766 dev_set_uevent_suppress(&dev->dev, 1); 11767 err = device_rename(&dev->dev, dev->name); 11768 dev_set_uevent_suppress(&dev->dev, 0); 11769 WARN_ON(err); 11770 11771 /* Send a netdev-add uevent to the new namespace */ 11772 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11773 netdev_adjacent_add_links(dev); 11774 11775 /* Adapt owner in case owning user namespace of target network 11776 * namespace is different from the original one. 11777 */ 11778 err = netdev_change_owner(dev, net_old, net); 11779 WARN_ON(err); 11780 11781 /* Add the device back in the hashes */ 11782 list_netdevice(dev); 11783 11784 /* Notify protocols, that a new device appeared. */ 11785 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11786 11787 /* 11788 * Prevent userspace races by waiting until the network 11789 * device is fully setup before sending notifications. 11790 */ 11791 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11792 11793 synchronize_net(); 11794 err = 0; 11795 out: 11796 return err; 11797 } 11798 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11799 11800 static int dev_cpu_dead(unsigned int oldcpu) 11801 { 11802 struct sk_buff **list_skb; 11803 struct sk_buff *skb; 11804 unsigned int cpu; 11805 struct softnet_data *sd, *oldsd, *remsd = NULL; 11806 11807 local_irq_disable(); 11808 cpu = smp_processor_id(); 11809 sd = &per_cpu(softnet_data, cpu); 11810 oldsd = &per_cpu(softnet_data, oldcpu); 11811 11812 /* Find end of our completion_queue. */ 11813 list_skb = &sd->completion_queue; 11814 while (*list_skb) 11815 list_skb = &(*list_skb)->next; 11816 /* Append completion queue from offline CPU. */ 11817 *list_skb = oldsd->completion_queue; 11818 oldsd->completion_queue = NULL; 11819 11820 /* Append output queue from offline CPU. */ 11821 if (oldsd->output_queue) { 11822 *sd->output_queue_tailp = oldsd->output_queue; 11823 sd->output_queue_tailp = oldsd->output_queue_tailp; 11824 oldsd->output_queue = NULL; 11825 oldsd->output_queue_tailp = &oldsd->output_queue; 11826 } 11827 /* Append NAPI poll list from offline CPU, with one exception : 11828 * process_backlog() must be called by cpu owning percpu backlog. 11829 * We properly handle process_queue & input_pkt_queue later. 11830 */ 11831 while (!list_empty(&oldsd->poll_list)) { 11832 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11833 struct napi_struct, 11834 poll_list); 11835 11836 list_del_init(&napi->poll_list); 11837 if (napi->poll == process_backlog) 11838 napi->state &= NAPIF_STATE_THREADED; 11839 else 11840 ____napi_schedule(sd, napi); 11841 } 11842 11843 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11844 local_irq_enable(); 11845 11846 if (!use_backlog_threads()) { 11847 #ifdef CONFIG_RPS 11848 remsd = oldsd->rps_ipi_list; 11849 oldsd->rps_ipi_list = NULL; 11850 #endif 11851 /* send out pending IPI's on offline CPU */ 11852 net_rps_send_ipi(remsd); 11853 } 11854 11855 /* Process offline CPU's input_pkt_queue */ 11856 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11857 netif_rx(skb); 11858 rps_input_queue_head_incr(oldsd); 11859 } 11860 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11861 netif_rx(skb); 11862 rps_input_queue_head_incr(oldsd); 11863 } 11864 11865 return 0; 11866 } 11867 11868 /** 11869 * netdev_increment_features - increment feature set by one 11870 * @all: current feature set 11871 * @one: new feature set 11872 * @mask: mask feature set 11873 * 11874 * Computes a new feature set after adding a device with feature set 11875 * @one to the master device with current feature set @all. Will not 11876 * enable anything that is off in @mask. Returns the new feature set. 11877 */ 11878 netdev_features_t netdev_increment_features(netdev_features_t all, 11879 netdev_features_t one, netdev_features_t mask) 11880 { 11881 if (mask & NETIF_F_HW_CSUM) 11882 mask |= NETIF_F_CSUM_MASK; 11883 mask |= NETIF_F_VLAN_CHALLENGED; 11884 11885 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11886 all &= one | ~NETIF_F_ALL_FOR_ALL; 11887 11888 /* If one device supports hw checksumming, set for all. */ 11889 if (all & NETIF_F_HW_CSUM) 11890 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11891 11892 return all; 11893 } 11894 EXPORT_SYMBOL(netdev_increment_features); 11895 11896 static struct hlist_head * __net_init netdev_create_hash(void) 11897 { 11898 int i; 11899 struct hlist_head *hash; 11900 11901 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11902 if (hash != NULL) 11903 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11904 INIT_HLIST_HEAD(&hash[i]); 11905 11906 return hash; 11907 } 11908 11909 /* Initialize per network namespace state */ 11910 static int __net_init netdev_init(struct net *net) 11911 { 11912 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11913 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11914 11915 INIT_LIST_HEAD(&net->dev_base_head); 11916 11917 net->dev_name_head = netdev_create_hash(); 11918 if (net->dev_name_head == NULL) 11919 goto err_name; 11920 11921 net->dev_index_head = netdev_create_hash(); 11922 if (net->dev_index_head == NULL) 11923 goto err_idx; 11924 11925 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11926 11927 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11928 11929 return 0; 11930 11931 err_idx: 11932 kfree(net->dev_name_head); 11933 err_name: 11934 return -ENOMEM; 11935 } 11936 11937 /** 11938 * netdev_drivername - network driver for the device 11939 * @dev: network device 11940 * 11941 * Determine network driver for device. 11942 */ 11943 const char *netdev_drivername(const struct net_device *dev) 11944 { 11945 const struct device_driver *driver; 11946 const struct device *parent; 11947 const char *empty = ""; 11948 11949 parent = dev->dev.parent; 11950 if (!parent) 11951 return empty; 11952 11953 driver = parent->driver; 11954 if (driver && driver->name) 11955 return driver->name; 11956 return empty; 11957 } 11958 11959 static void __netdev_printk(const char *level, const struct net_device *dev, 11960 struct va_format *vaf) 11961 { 11962 if (dev && dev->dev.parent) { 11963 dev_printk_emit(level[1] - '0', 11964 dev->dev.parent, 11965 "%s %s %s%s: %pV", 11966 dev_driver_string(dev->dev.parent), 11967 dev_name(dev->dev.parent), 11968 netdev_name(dev), netdev_reg_state(dev), 11969 vaf); 11970 } else if (dev) { 11971 printk("%s%s%s: %pV", 11972 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11973 } else { 11974 printk("%s(NULL net_device): %pV", level, vaf); 11975 } 11976 } 11977 11978 void netdev_printk(const char *level, const struct net_device *dev, 11979 const char *format, ...) 11980 { 11981 struct va_format vaf; 11982 va_list args; 11983 11984 va_start(args, format); 11985 11986 vaf.fmt = format; 11987 vaf.va = &args; 11988 11989 __netdev_printk(level, dev, &vaf); 11990 11991 va_end(args); 11992 } 11993 EXPORT_SYMBOL(netdev_printk); 11994 11995 #define define_netdev_printk_level(func, level) \ 11996 void func(const struct net_device *dev, const char *fmt, ...) \ 11997 { \ 11998 struct va_format vaf; \ 11999 va_list args; \ 12000 \ 12001 va_start(args, fmt); \ 12002 \ 12003 vaf.fmt = fmt; \ 12004 vaf.va = &args; \ 12005 \ 12006 __netdev_printk(level, dev, &vaf); \ 12007 \ 12008 va_end(args); \ 12009 } \ 12010 EXPORT_SYMBOL(func); 12011 12012 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 12013 define_netdev_printk_level(netdev_alert, KERN_ALERT); 12014 define_netdev_printk_level(netdev_crit, KERN_CRIT); 12015 define_netdev_printk_level(netdev_err, KERN_ERR); 12016 define_netdev_printk_level(netdev_warn, KERN_WARNING); 12017 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 12018 define_netdev_printk_level(netdev_info, KERN_INFO); 12019 12020 static void __net_exit netdev_exit(struct net *net) 12021 { 12022 kfree(net->dev_name_head); 12023 kfree(net->dev_index_head); 12024 xa_destroy(&net->dev_by_index); 12025 if (net != &init_net) 12026 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 12027 } 12028 12029 static struct pernet_operations __net_initdata netdev_net_ops = { 12030 .init = netdev_init, 12031 .exit = netdev_exit, 12032 }; 12033 12034 static void __net_exit default_device_exit_net(struct net *net) 12035 { 12036 struct netdev_name_node *name_node, *tmp; 12037 struct net_device *dev, *aux; 12038 /* 12039 * Push all migratable network devices back to the 12040 * initial network namespace 12041 */ 12042 ASSERT_RTNL(); 12043 for_each_netdev_safe(net, dev, aux) { 12044 int err; 12045 char fb_name[IFNAMSIZ]; 12046 12047 /* Ignore unmoveable devices (i.e. loopback) */ 12048 if (dev->netns_local) 12049 continue; 12050 12051 /* Leave virtual devices for the generic cleanup */ 12052 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 12053 continue; 12054 12055 /* Push remaining network devices to init_net */ 12056 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 12057 if (netdev_name_in_use(&init_net, fb_name)) 12058 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 12059 12060 netdev_for_each_altname_safe(dev, name_node, tmp) 12061 if (netdev_name_in_use(&init_net, name_node->name)) 12062 __netdev_name_node_alt_destroy(name_node); 12063 12064 err = dev_change_net_namespace(dev, &init_net, fb_name); 12065 if (err) { 12066 pr_emerg("%s: failed to move %s to init_net: %d\n", 12067 __func__, dev->name, err); 12068 BUG(); 12069 } 12070 } 12071 } 12072 12073 static void __net_exit default_device_exit_batch(struct list_head *net_list) 12074 { 12075 /* At exit all network devices most be removed from a network 12076 * namespace. Do this in the reverse order of registration. 12077 * Do this across as many network namespaces as possible to 12078 * improve batching efficiency. 12079 */ 12080 struct net_device *dev; 12081 struct net *net; 12082 LIST_HEAD(dev_kill_list); 12083 12084 rtnl_lock(); 12085 list_for_each_entry(net, net_list, exit_list) { 12086 default_device_exit_net(net); 12087 cond_resched(); 12088 } 12089 12090 list_for_each_entry(net, net_list, exit_list) { 12091 for_each_netdev_reverse(net, dev) { 12092 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 12093 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 12094 else 12095 unregister_netdevice_queue(dev, &dev_kill_list); 12096 } 12097 } 12098 unregister_netdevice_many(&dev_kill_list); 12099 rtnl_unlock(); 12100 } 12101 12102 static struct pernet_operations __net_initdata default_device_ops = { 12103 .exit_batch = default_device_exit_batch, 12104 }; 12105 12106 static void __init net_dev_struct_check(void) 12107 { 12108 /* TX read-mostly hotpath */ 12109 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags_fast); 12110 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 12111 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 12112 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 12113 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 12114 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 12115 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 12116 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 12117 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 12118 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 12119 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 12120 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 12121 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 12122 #ifdef CONFIG_XPS 12123 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 12124 #endif 12125 #ifdef CONFIG_NETFILTER_EGRESS 12126 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 12127 #endif 12128 #ifdef CONFIG_NET_XGRESS 12129 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 12130 #endif 12131 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 12132 12133 /* TXRX read-mostly hotpath */ 12134 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); 12135 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state); 12136 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 12137 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 12138 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 12139 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 12140 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46); 12141 12142 /* RX read-mostly hotpath */ 12143 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 12144 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 12145 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 12146 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 12147 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 12148 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 12149 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 12150 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 12151 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 12152 #ifdef CONFIG_NETPOLL 12153 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 12154 #endif 12155 #ifdef CONFIG_NET_XGRESS 12156 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 12157 #endif 12158 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 92); 12159 } 12160 12161 /* 12162 * Initialize the DEV module. At boot time this walks the device list and 12163 * unhooks any devices that fail to initialise (normally hardware not 12164 * present) and leaves us with a valid list of present and active devices. 12165 * 12166 */ 12167 12168 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ 12169 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) 12170 12171 static int net_page_pool_create(int cpuid) 12172 { 12173 #if IS_ENABLED(CONFIG_PAGE_POOL) 12174 struct page_pool_params page_pool_params = { 12175 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, 12176 .flags = PP_FLAG_SYSTEM_POOL, 12177 .nid = cpu_to_mem(cpuid), 12178 }; 12179 struct page_pool *pp_ptr; 12180 int err; 12181 12182 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); 12183 if (IS_ERR(pp_ptr)) 12184 return -ENOMEM; 12185 12186 err = xdp_reg_page_pool(pp_ptr); 12187 if (err) { 12188 page_pool_destroy(pp_ptr); 12189 return err; 12190 } 12191 12192 per_cpu(system_page_pool, cpuid) = pp_ptr; 12193 #endif 12194 return 0; 12195 } 12196 12197 static int backlog_napi_should_run(unsigned int cpu) 12198 { 12199 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12200 struct napi_struct *napi = &sd->backlog; 12201 12202 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 12203 } 12204 12205 static void run_backlog_napi(unsigned int cpu) 12206 { 12207 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12208 12209 napi_threaded_poll_loop(&sd->backlog); 12210 } 12211 12212 static void backlog_napi_setup(unsigned int cpu) 12213 { 12214 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12215 struct napi_struct *napi = &sd->backlog; 12216 12217 napi->thread = this_cpu_read(backlog_napi); 12218 set_bit(NAPI_STATE_THREADED, &napi->state); 12219 } 12220 12221 static struct smp_hotplug_thread backlog_threads = { 12222 .store = &backlog_napi, 12223 .thread_should_run = backlog_napi_should_run, 12224 .thread_fn = run_backlog_napi, 12225 .thread_comm = "backlog_napi/%u", 12226 .setup = backlog_napi_setup, 12227 }; 12228 12229 /* 12230 * This is called single threaded during boot, so no need 12231 * to take the rtnl semaphore. 12232 */ 12233 static int __init net_dev_init(void) 12234 { 12235 int i, rc = -ENOMEM; 12236 12237 BUG_ON(!dev_boot_phase); 12238 12239 net_dev_struct_check(); 12240 12241 if (dev_proc_init()) 12242 goto out; 12243 12244 if (netdev_kobject_init()) 12245 goto out; 12246 12247 for (i = 0; i < PTYPE_HASH_SIZE; i++) 12248 INIT_LIST_HEAD(&ptype_base[i]); 12249 12250 if (register_pernet_subsys(&netdev_net_ops)) 12251 goto out; 12252 12253 /* 12254 * Initialise the packet receive queues. 12255 */ 12256 12257 for_each_possible_cpu(i) { 12258 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 12259 struct softnet_data *sd = &per_cpu(softnet_data, i); 12260 12261 INIT_WORK(flush, flush_backlog); 12262 12263 skb_queue_head_init(&sd->input_pkt_queue); 12264 skb_queue_head_init(&sd->process_queue); 12265 #ifdef CONFIG_XFRM_OFFLOAD 12266 skb_queue_head_init(&sd->xfrm_backlog); 12267 #endif 12268 INIT_LIST_HEAD(&sd->poll_list); 12269 sd->output_queue_tailp = &sd->output_queue; 12270 #ifdef CONFIG_RPS 12271 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 12272 sd->cpu = i; 12273 #endif 12274 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 12275 spin_lock_init(&sd->defer_lock); 12276 12277 init_gro_hash(&sd->backlog); 12278 sd->backlog.poll = process_backlog; 12279 sd->backlog.weight = weight_p; 12280 INIT_LIST_HEAD(&sd->backlog.poll_list); 12281 12282 if (net_page_pool_create(i)) 12283 goto out; 12284 } 12285 if (use_backlog_threads()) 12286 smpboot_register_percpu_thread(&backlog_threads); 12287 12288 dev_boot_phase = 0; 12289 12290 /* The loopback device is special if any other network devices 12291 * is present in a network namespace the loopback device must 12292 * be present. Since we now dynamically allocate and free the 12293 * loopback device ensure this invariant is maintained by 12294 * keeping the loopback device as the first device on the 12295 * list of network devices. Ensuring the loopback devices 12296 * is the first device that appears and the last network device 12297 * that disappears. 12298 */ 12299 if (register_pernet_device(&loopback_net_ops)) 12300 goto out; 12301 12302 if (register_pernet_device(&default_device_ops)) 12303 goto out; 12304 12305 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 12306 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 12307 12308 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 12309 NULL, dev_cpu_dead); 12310 WARN_ON(rc < 0); 12311 rc = 0; 12312 12313 /* avoid static key IPIs to isolated CPUs */ 12314 if (housekeeping_enabled(HK_TYPE_MISC)) 12315 net_enable_timestamp(); 12316 out: 12317 if (rc < 0) { 12318 for_each_possible_cpu(i) { 12319 struct page_pool *pp_ptr; 12320 12321 pp_ptr = per_cpu(system_page_pool, i); 12322 if (!pp_ptr) 12323 continue; 12324 12325 xdp_unreg_page_pool(pp_ptr); 12326 page_pool_destroy(pp_ptr); 12327 per_cpu(system_page_pool, i) = NULL; 12328 } 12329 } 12330 12331 return rc; 12332 } 12333 12334 subsys_initcall(net_dev_init); 12335