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 /* must be called under rcu_read_lock(), as we dont take a reference */ 757 static struct napi_struct *napi_by_id(unsigned int napi_id) 758 { 759 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 760 struct napi_struct *napi; 761 762 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 763 if (napi->napi_id == napi_id) 764 return napi; 765 766 return NULL; 767 } 768 769 /* must be called under rcu_read_lock(), as we dont take a reference */ 770 struct napi_struct *netdev_napi_by_id(struct net *net, unsigned int napi_id) 771 { 772 struct napi_struct *napi; 773 774 napi = napi_by_id(napi_id); 775 if (!napi) 776 return NULL; 777 778 if (WARN_ON_ONCE(!napi->dev)) 779 return NULL; 780 if (!net_eq(net, dev_net(napi->dev))) 781 return NULL; 782 783 return napi; 784 } 785 786 /** 787 * __dev_get_by_name - find a device by its name 788 * @net: the applicable net namespace 789 * @name: name to find 790 * 791 * Find an interface by name. Must be called under RTNL semaphore. 792 * If the name is found a pointer to the device is returned. 793 * If the name is not found then %NULL is returned. The 794 * reference counters are not incremented so the caller must be 795 * careful with locks. 796 */ 797 798 struct net_device *__dev_get_by_name(struct net *net, const char *name) 799 { 800 struct netdev_name_node *node_name; 801 802 node_name = netdev_name_node_lookup(net, name); 803 return node_name ? node_name->dev : NULL; 804 } 805 EXPORT_SYMBOL(__dev_get_by_name); 806 807 /** 808 * dev_get_by_name_rcu - find a device by its name 809 * @net: the applicable net namespace 810 * @name: name to find 811 * 812 * Find an interface by name. 813 * If the name is found a pointer to the device is returned. 814 * If the name is not found then %NULL is returned. 815 * The reference counters are not incremented so the caller must be 816 * careful with locks. The caller must hold RCU lock. 817 */ 818 819 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 820 { 821 struct netdev_name_node *node_name; 822 823 node_name = netdev_name_node_lookup_rcu(net, name); 824 return node_name ? node_name->dev : NULL; 825 } 826 EXPORT_SYMBOL(dev_get_by_name_rcu); 827 828 /* Deprecated for new users, call netdev_get_by_name() instead */ 829 struct net_device *dev_get_by_name(struct net *net, const char *name) 830 { 831 struct net_device *dev; 832 833 rcu_read_lock(); 834 dev = dev_get_by_name_rcu(net, name); 835 dev_hold(dev); 836 rcu_read_unlock(); 837 return dev; 838 } 839 EXPORT_SYMBOL(dev_get_by_name); 840 841 /** 842 * netdev_get_by_name() - find a device by its name 843 * @net: the applicable net namespace 844 * @name: name to find 845 * @tracker: tracking object for the acquired reference 846 * @gfp: allocation flags for the tracker 847 * 848 * Find an interface by name. This can be called from any 849 * context and does its own locking. The returned handle has 850 * the usage count incremented and the caller must use netdev_put() to 851 * release it when it is no longer needed. %NULL is returned if no 852 * matching device is found. 853 */ 854 struct net_device *netdev_get_by_name(struct net *net, const char *name, 855 netdevice_tracker *tracker, gfp_t gfp) 856 { 857 struct net_device *dev; 858 859 dev = dev_get_by_name(net, name); 860 if (dev) 861 netdev_tracker_alloc(dev, tracker, gfp); 862 return dev; 863 } 864 EXPORT_SYMBOL(netdev_get_by_name); 865 866 /** 867 * __dev_get_by_index - find a device by its ifindex 868 * @net: the applicable net namespace 869 * @ifindex: index of device 870 * 871 * Search for an interface by index. Returns %NULL if the device 872 * is not found or a pointer to the device. The device has not 873 * had its reference counter increased so the caller must be careful 874 * about locking. The caller must hold the RTNL semaphore. 875 */ 876 877 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 878 { 879 struct net_device *dev; 880 struct hlist_head *head = dev_index_hash(net, ifindex); 881 882 hlist_for_each_entry(dev, head, index_hlist) 883 if (dev->ifindex == ifindex) 884 return dev; 885 886 return NULL; 887 } 888 EXPORT_SYMBOL(__dev_get_by_index); 889 890 /** 891 * dev_get_by_index_rcu - find a device by its ifindex 892 * @net: the applicable net namespace 893 * @ifindex: index of device 894 * 895 * Search for an interface by index. Returns %NULL if the device 896 * is not found or a pointer to the device. The device has not 897 * had its reference counter increased so the caller must be careful 898 * about locking. The caller must hold RCU lock. 899 */ 900 901 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 902 { 903 struct net_device *dev; 904 struct hlist_head *head = dev_index_hash(net, ifindex); 905 906 hlist_for_each_entry_rcu(dev, head, index_hlist) 907 if (dev->ifindex == ifindex) 908 return dev; 909 910 return NULL; 911 } 912 EXPORT_SYMBOL(dev_get_by_index_rcu); 913 914 /* Deprecated for new users, call netdev_get_by_index() instead */ 915 struct net_device *dev_get_by_index(struct net *net, int ifindex) 916 { 917 struct net_device *dev; 918 919 rcu_read_lock(); 920 dev = dev_get_by_index_rcu(net, ifindex); 921 dev_hold(dev); 922 rcu_read_unlock(); 923 return dev; 924 } 925 EXPORT_SYMBOL(dev_get_by_index); 926 927 /** 928 * netdev_get_by_index() - find a device by its ifindex 929 * @net: the applicable net namespace 930 * @ifindex: index of device 931 * @tracker: tracking object for the acquired reference 932 * @gfp: allocation flags for the tracker 933 * 934 * Search for an interface by index. Returns NULL if the device 935 * is not found or a pointer to the device. The device returned has 936 * had a reference added and the pointer is safe until the user calls 937 * netdev_put() to indicate they have finished with it. 938 */ 939 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 940 netdevice_tracker *tracker, gfp_t gfp) 941 { 942 struct net_device *dev; 943 944 dev = dev_get_by_index(net, ifindex); 945 if (dev) 946 netdev_tracker_alloc(dev, tracker, gfp); 947 return dev; 948 } 949 EXPORT_SYMBOL(netdev_get_by_index); 950 951 /** 952 * dev_get_by_napi_id - find a device by napi_id 953 * @napi_id: ID of the NAPI struct 954 * 955 * Search for an interface by NAPI ID. Returns %NULL if the device 956 * is not found or a pointer to the device. The device has not had 957 * its reference counter increased so the caller must be careful 958 * about locking. The caller must hold RCU lock. 959 */ 960 961 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 962 { 963 struct napi_struct *napi; 964 965 WARN_ON_ONCE(!rcu_read_lock_held()); 966 967 if (napi_id < MIN_NAPI_ID) 968 return NULL; 969 970 napi = napi_by_id(napi_id); 971 972 return napi ? napi->dev : NULL; 973 } 974 EXPORT_SYMBOL(dev_get_by_napi_id); 975 976 static DEFINE_SEQLOCK(netdev_rename_lock); 977 978 void netdev_copy_name(struct net_device *dev, char *name) 979 { 980 unsigned int seq; 981 982 do { 983 seq = read_seqbegin(&netdev_rename_lock); 984 strscpy(name, dev->name, IFNAMSIZ); 985 } while (read_seqretry(&netdev_rename_lock, seq)); 986 } 987 988 /** 989 * netdev_get_name - get a netdevice name, knowing its ifindex. 990 * @net: network namespace 991 * @name: a pointer to the buffer where the name will be stored. 992 * @ifindex: the ifindex of the interface to get the name from. 993 */ 994 int netdev_get_name(struct net *net, char *name, int ifindex) 995 { 996 struct net_device *dev; 997 int ret; 998 999 rcu_read_lock(); 1000 1001 dev = dev_get_by_index_rcu(net, ifindex); 1002 if (!dev) { 1003 ret = -ENODEV; 1004 goto out; 1005 } 1006 1007 netdev_copy_name(dev, name); 1008 1009 ret = 0; 1010 out: 1011 rcu_read_unlock(); 1012 return ret; 1013 } 1014 1015 /** 1016 * dev_getbyhwaddr_rcu - find a device by its hardware address 1017 * @net: the applicable net namespace 1018 * @type: media type of device 1019 * @ha: hardware address 1020 * 1021 * Search for an interface by MAC address. Returns NULL if the device 1022 * is not found or a pointer to the device. 1023 * The caller must hold RCU or RTNL. 1024 * The returned device has not had its ref count increased 1025 * and the caller must therefore be careful about locking 1026 * 1027 */ 1028 1029 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1030 const char *ha) 1031 { 1032 struct net_device *dev; 1033 1034 for_each_netdev_rcu(net, dev) 1035 if (dev->type == type && 1036 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1037 return dev; 1038 1039 return NULL; 1040 } 1041 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1042 1043 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1044 { 1045 struct net_device *dev, *ret = NULL; 1046 1047 rcu_read_lock(); 1048 for_each_netdev_rcu(net, dev) 1049 if (dev->type == type) { 1050 dev_hold(dev); 1051 ret = dev; 1052 break; 1053 } 1054 rcu_read_unlock(); 1055 return ret; 1056 } 1057 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1058 1059 /** 1060 * __dev_get_by_flags - find any device with given flags 1061 * @net: the applicable net namespace 1062 * @if_flags: IFF_* values 1063 * @mask: bitmask of bits in if_flags to check 1064 * 1065 * Search for any interface with the given flags. Returns NULL if a device 1066 * is not found or a pointer to the device. Must be called inside 1067 * rtnl_lock(), and result refcount is unchanged. 1068 */ 1069 1070 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1071 unsigned short mask) 1072 { 1073 struct net_device *dev, *ret; 1074 1075 ASSERT_RTNL(); 1076 1077 ret = NULL; 1078 for_each_netdev(net, dev) { 1079 if (((dev->flags ^ if_flags) & mask) == 0) { 1080 ret = dev; 1081 break; 1082 } 1083 } 1084 return ret; 1085 } 1086 EXPORT_SYMBOL(__dev_get_by_flags); 1087 1088 /** 1089 * dev_valid_name - check if name is okay for network device 1090 * @name: name string 1091 * 1092 * Network device names need to be valid file names to 1093 * allow sysfs to work. We also disallow any kind of 1094 * whitespace. 1095 */ 1096 bool dev_valid_name(const char *name) 1097 { 1098 if (*name == '\0') 1099 return false; 1100 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1101 return false; 1102 if (!strcmp(name, ".") || !strcmp(name, "..")) 1103 return false; 1104 1105 while (*name) { 1106 if (*name == '/' || *name == ':' || isspace(*name)) 1107 return false; 1108 name++; 1109 } 1110 return true; 1111 } 1112 EXPORT_SYMBOL(dev_valid_name); 1113 1114 /** 1115 * __dev_alloc_name - allocate a name for a device 1116 * @net: network namespace to allocate the device name in 1117 * @name: name format string 1118 * @res: result name string 1119 * 1120 * Passed a format string - eg "lt%d" it will try and find a suitable 1121 * id. It scans list of devices to build up a free map, then chooses 1122 * the first empty slot. The caller must hold the dev_base or rtnl lock 1123 * while allocating the name and adding the device in order to avoid 1124 * duplicates. 1125 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1126 * Returns the number of the unit assigned or a negative errno code. 1127 */ 1128 1129 static int __dev_alloc_name(struct net *net, const char *name, char *res) 1130 { 1131 int i = 0; 1132 const char *p; 1133 const int max_netdevices = 8*PAGE_SIZE; 1134 unsigned long *inuse; 1135 struct net_device *d; 1136 char buf[IFNAMSIZ]; 1137 1138 /* Verify the string as this thing may have come from the user. 1139 * There must be one "%d" and no other "%" characters. 1140 */ 1141 p = strchr(name, '%'); 1142 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1143 return -EINVAL; 1144 1145 /* Use one page as a bit array of possible slots */ 1146 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1147 if (!inuse) 1148 return -ENOMEM; 1149 1150 for_each_netdev(net, d) { 1151 struct netdev_name_node *name_node; 1152 1153 netdev_for_each_altname(d, name_node) { 1154 if (!sscanf(name_node->name, name, &i)) 1155 continue; 1156 if (i < 0 || i >= max_netdevices) 1157 continue; 1158 1159 /* avoid cases where sscanf is not exact inverse of printf */ 1160 snprintf(buf, IFNAMSIZ, name, i); 1161 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1162 __set_bit(i, inuse); 1163 } 1164 if (!sscanf(d->name, name, &i)) 1165 continue; 1166 if (i < 0 || i >= max_netdevices) 1167 continue; 1168 1169 /* avoid cases where sscanf is not exact inverse of printf */ 1170 snprintf(buf, IFNAMSIZ, name, i); 1171 if (!strncmp(buf, d->name, IFNAMSIZ)) 1172 __set_bit(i, inuse); 1173 } 1174 1175 i = find_first_zero_bit(inuse, max_netdevices); 1176 bitmap_free(inuse); 1177 if (i == max_netdevices) 1178 return -ENFILE; 1179 1180 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1181 strscpy(buf, name, IFNAMSIZ); 1182 snprintf(res, IFNAMSIZ, buf, i); 1183 return i; 1184 } 1185 1186 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1187 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1188 const char *want_name, char *out_name, 1189 int dup_errno) 1190 { 1191 if (!dev_valid_name(want_name)) 1192 return -EINVAL; 1193 1194 if (strchr(want_name, '%')) 1195 return __dev_alloc_name(net, want_name, out_name); 1196 1197 if (netdev_name_in_use(net, want_name)) 1198 return -dup_errno; 1199 if (out_name != want_name) 1200 strscpy(out_name, want_name, IFNAMSIZ); 1201 return 0; 1202 } 1203 1204 /** 1205 * dev_alloc_name - allocate a name for a device 1206 * @dev: device 1207 * @name: name format string 1208 * 1209 * Passed a format string - eg "lt%d" it will try and find a suitable 1210 * id. It scans list of devices to build up a free map, then chooses 1211 * the first empty slot. The caller must hold the dev_base or rtnl lock 1212 * while allocating the name and adding the device in order to avoid 1213 * duplicates. 1214 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1215 * Returns the number of the unit assigned or a negative errno code. 1216 */ 1217 1218 int dev_alloc_name(struct net_device *dev, const char *name) 1219 { 1220 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1221 } 1222 EXPORT_SYMBOL(dev_alloc_name); 1223 1224 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1225 const char *name) 1226 { 1227 int ret; 1228 1229 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1230 return ret < 0 ? ret : 0; 1231 } 1232 1233 /** 1234 * dev_change_name - change name of a device 1235 * @dev: device 1236 * @newname: name (or format string) must be at least IFNAMSIZ 1237 * 1238 * Change name of a device, can pass format strings "eth%d". 1239 * for wildcarding. 1240 */ 1241 int dev_change_name(struct net_device *dev, const char *newname) 1242 { 1243 unsigned char old_assign_type; 1244 char oldname[IFNAMSIZ]; 1245 int err = 0; 1246 int ret; 1247 struct net *net; 1248 1249 ASSERT_RTNL(); 1250 BUG_ON(!dev_net(dev)); 1251 1252 net = dev_net(dev); 1253 1254 down_write(&devnet_rename_sem); 1255 1256 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1257 up_write(&devnet_rename_sem); 1258 return 0; 1259 } 1260 1261 memcpy(oldname, dev->name, IFNAMSIZ); 1262 1263 write_seqlock_bh(&netdev_rename_lock); 1264 err = dev_get_valid_name(net, dev, newname); 1265 write_sequnlock_bh(&netdev_rename_lock); 1266 1267 if (err < 0) { 1268 up_write(&devnet_rename_sem); 1269 return err; 1270 } 1271 1272 if (oldname[0] && !strchr(oldname, '%')) 1273 netdev_info(dev, "renamed from %s%s\n", oldname, 1274 dev->flags & IFF_UP ? " (while UP)" : ""); 1275 1276 old_assign_type = dev->name_assign_type; 1277 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); 1278 1279 rollback: 1280 ret = device_rename(&dev->dev, dev->name); 1281 if (ret) { 1282 memcpy(dev->name, oldname, IFNAMSIZ); 1283 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1284 up_write(&devnet_rename_sem); 1285 return ret; 1286 } 1287 1288 up_write(&devnet_rename_sem); 1289 1290 netdev_adjacent_rename_links(dev, oldname); 1291 1292 netdev_name_node_del(dev->name_node); 1293 1294 synchronize_net(); 1295 1296 netdev_name_node_add(net, dev->name_node); 1297 1298 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1299 ret = notifier_to_errno(ret); 1300 1301 if (ret) { 1302 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1303 if (err >= 0) { 1304 err = ret; 1305 down_write(&devnet_rename_sem); 1306 write_seqlock_bh(&netdev_rename_lock); 1307 memcpy(dev->name, oldname, IFNAMSIZ); 1308 write_sequnlock_bh(&netdev_rename_lock); 1309 memcpy(oldname, newname, IFNAMSIZ); 1310 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1311 old_assign_type = NET_NAME_RENAMED; 1312 goto rollback; 1313 } else { 1314 netdev_err(dev, "name change rollback failed: %d\n", 1315 ret); 1316 } 1317 } 1318 1319 return err; 1320 } 1321 1322 /** 1323 * dev_set_alias - change ifalias of a device 1324 * @dev: device 1325 * @alias: name up to IFALIASZ 1326 * @len: limit of bytes to copy from info 1327 * 1328 * Set ifalias for a device, 1329 */ 1330 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1331 { 1332 struct dev_ifalias *new_alias = NULL; 1333 1334 if (len >= IFALIASZ) 1335 return -EINVAL; 1336 1337 if (len) { 1338 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1339 if (!new_alias) 1340 return -ENOMEM; 1341 1342 memcpy(new_alias->ifalias, alias, len); 1343 new_alias->ifalias[len] = 0; 1344 } 1345 1346 mutex_lock(&ifalias_mutex); 1347 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1348 mutex_is_locked(&ifalias_mutex)); 1349 mutex_unlock(&ifalias_mutex); 1350 1351 if (new_alias) 1352 kfree_rcu(new_alias, rcuhead); 1353 1354 return len; 1355 } 1356 EXPORT_SYMBOL(dev_set_alias); 1357 1358 /** 1359 * dev_get_alias - get ifalias of a device 1360 * @dev: device 1361 * @name: buffer to store name of ifalias 1362 * @len: size of buffer 1363 * 1364 * get ifalias for a device. Caller must make sure dev cannot go 1365 * away, e.g. rcu read lock or own a reference count to device. 1366 */ 1367 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1368 { 1369 const struct dev_ifalias *alias; 1370 int ret = 0; 1371 1372 rcu_read_lock(); 1373 alias = rcu_dereference(dev->ifalias); 1374 if (alias) 1375 ret = snprintf(name, len, "%s", alias->ifalias); 1376 rcu_read_unlock(); 1377 1378 return ret; 1379 } 1380 1381 /** 1382 * netdev_features_change - device changes features 1383 * @dev: device to cause notification 1384 * 1385 * Called to indicate a device has changed features. 1386 */ 1387 void netdev_features_change(struct net_device *dev) 1388 { 1389 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1390 } 1391 EXPORT_SYMBOL(netdev_features_change); 1392 1393 /** 1394 * netdev_state_change - device changes state 1395 * @dev: device to cause notification 1396 * 1397 * Called to indicate a device has changed state. This function calls 1398 * the notifier chains for netdev_chain and sends a NEWLINK message 1399 * to the routing socket. 1400 */ 1401 void netdev_state_change(struct net_device *dev) 1402 { 1403 if (dev->flags & IFF_UP) { 1404 struct netdev_notifier_change_info change_info = { 1405 .info.dev = dev, 1406 }; 1407 1408 call_netdevice_notifiers_info(NETDEV_CHANGE, 1409 &change_info.info); 1410 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1411 } 1412 } 1413 EXPORT_SYMBOL(netdev_state_change); 1414 1415 /** 1416 * __netdev_notify_peers - notify network peers about existence of @dev, 1417 * to be called when rtnl lock is already held. 1418 * @dev: network device 1419 * 1420 * Generate traffic such that interested network peers are aware of 1421 * @dev, such as by generating a gratuitous ARP. This may be used when 1422 * a device wants to inform the rest of the network about some sort of 1423 * reconfiguration such as a failover event or virtual machine 1424 * migration. 1425 */ 1426 void __netdev_notify_peers(struct net_device *dev) 1427 { 1428 ASSERT_RTNL(); 1429 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1430 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1431 } 1432 EXPORT_SYMBOL(__netdev_notify_peers); 1433 1434 /** 1435 * netdev_notify_peers - notify network peers about existence of @dev 1436 * @dev: network device 1437 * 1438 * Generate traffic such that interested network peers are aware of 1439 * @dev, such as by generating a gratuitous ARP. This may be used when 1440 * a device wants to inform the rest of the network about some sort of 1441 * reconfiguration such as a failover event or virtual machine 1442 * migration. 1443 */ 1444 void netdev_notify_peers(struct net_device *dev) 1445 { 1446 rtnl_lock(); 1447 __netdev_notify_peers(dev); 1448 rtnl_unlock(); 1449 } 1450 EXPORT_SYMBOL(netdev_notify_peers); 1451 1452 static int napi_threaded_poll(void *data); 1453 1454 static int napi_kthread_create(struct napi_struct *n) 1455 { 1456 int err = 0; 1457 1458 /* Create and wake up the kthread once to put it in 1459 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1460 * warning and work with loadavg. 1461 */ 1462 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1463 n->dev->name, n->napi_id); 1464 if (IS_ERR(n->thread)) { 1465 err = PTR_ERR(n->thread); 1466 pr_err("kthread_run failed with err %d\n", err); 1467 n->thread = NULL; 1468 } 1469 1470 return err; 1471 } 1472 1473 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1474 { 1475 const struct net_device_ops *ops = dev->netdev_ops; 1476 int ret; 1477 1478 ASSERT_RTNL(); 1479 dev_addr_check(dev); 1480 1481 if (!netif_device_present(dev)) { 1482 /* may be detached because parent is runtime-suspended */ 1483 if (dev->dev.parent) 1484 pm_runtime_resume(dev->dev.parent); 1485 if (!netif_device_present(dev)) 1486 return -ENODEV; 1487 } 1488 1489 /* Block netpoll from trying to do any rx path servicing. 1490 * If we don't do this there is a chance ndo_poll_controller 1491 * or ndo_poll may be running while we open the device 1492 */ 1493 netpoll_poll_disable(dev); 1494 1495 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1496 ret = notifier_to_errno(ret); 1497 if (ret) 1498 return ret; 1499 1500 set_bit(__LINK_STATE_START, &dev->state); 1501 1502 if (ops->ndo_validate_addr) 1503 ret = ops->ndo_validate_addr(dev); 1504 1505 if (!ret && ops->ndo_open) 1506 ret = ops->ndo_open(dev); 1507 1508 netpoll_poll_enable(dev); 1509 1510 if (ret) 1511 clear_bit(__LINK_STATE_START, &dev->state); 1512 else { 1513 dev->flags |= IFF_UP; 1514 dev_set_rx_mode(dev); 1515 dev_activate(dev); 1516 add_device_randomness(dev->dev_addr, dev->addr_len); 1517 } 1518 1519 return ret; 1520 } 1521 1522 /** 1523 * dev_open - prepare an interface for use. 1524 * @dev: device to open 1525 * @extack: netlink extended ack 1526 * 1527 * Takes a device from down to up state. The device's private open 1528 * function is invoked and then the multicast lists are loaded. Finally 1529 * the device is moved into the up state and a %NETDEV_UP message is 1530 * sent to the netdev notifier chain. 1531 * 1532 * Calling this function on an active interface is a nop. On a failure 1533 * a negative errno code is returned. 1534 */ 1535 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1536 { 1537 int ret; 1538 1539 if (dev->flags & IFF_UP) 1540 return 0; 1541 1542 ret = __dev_open(dev, extack); 1543 if (ret < 0) 1544 return ret; 1545 1546 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1547 call_netdevice_notifiers(NETDEV_UP, dev); 1548 1549 return ret; 1550 } 1551 EXPORT_SYMBOL(dev_open); 1552 1553 static void __dev_close_many(struct list_head *head) 1554 { 1555 struct net_device *dev; 1556 1557 ASSERT_RTNL(); 1558 might_sleep(); 1559 1560 list_for_each_entry(dev, head, close_list) { 1561 /* Temporarily disable netpoll until the interface is down */ 1562 netpoll_poll_disable(dev); 1563 1564 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1565 1566 clear_bit(__LINK_STATE_START, &dev->state); 1567 1568 /* Synchronize to scheduled poll. We cannot touch poll list, it 1569 * can be even on different cpu. So just clear netif_running(). 1570 * 1571 * dev->stop() will invoke napi_disable() on all of it's 1572 * napi_struct instances on this device. 1573 */ 1574 smp_mb__after_atomic(); /* Commit netif_running(). */ 1575 } 1576 1577 dev_deactivate_many(head); 1578 1579 list_for_each_entry(dev, head, close_list) { 1580 const struct net_device_ops *ops = dev->netdev_ops; 1581 1582 /* 1583 * Call the device specific close. This cannot fail. 1584 * Only if device is UP 1585 * 1586 * We allow it to be called even after a DETACH hot-plug 1587 * event. 1588 */ 1589 if (ops->ndo_stop) 1590 ops->ndo_stop(dev); 1591 1592 dev->flags &= ~IFF_UP; 1593 netpoll_poll_enable(dev); 1594 } 1595 } 1596 1597 static void __dev_close(struct net_device *dev) 1598 { 1599 LIST_HEAD(single); 1600 1601 list_add(&dev->close_list, &single); 1602 __dev_close_many(&single); 1603 list_del(&single); 1604 } 1605 1606 void dev_close_many(struct list_head *head, bool unlink) 1607 { 1608 struct net_device *dev, *tmp; 1609 1610 /* Remove the devices that don't need to be closed */ 1611 list_for_each_entry_safe(dev, tmp, head, close_list) 1612 if (!(dev->flags & IFF_UP)) 1613 list_del_init(&dev->close_list); 1614 1615 __dev_close_many(head); 1616 1617 list_for_each_entry_safe(dev, tmp, head, close_list) { 1618 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1619 call_netdevice_notifiers(NETDEV_DOWN, dev); 1620 if (unlink) 1621 list_del_init(&dev->close_list); 1622 } 1623 } 1624 EXPORT_SYMBOL(dev_close_many); 1625 1626 /** 1627 * dev_close - shutdown an interface. 1628 * @dev: device to shutdown 1629 * 1630 * This function moves an active device into down state. A 1631 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1632 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1633 * chain. 1634 */ 1635 void dev_close(struct net_device *dev) 1636 { 1637 if (dev->flags & IFF_UP) { 1638 LIST_HEAD(single); 1639 1640 list_add(&dev->close_list, &single); 1641 dev_close_many(&single, true); 1642 list_del(&single); 1643 } 1644 } 1645 EXPORT_SYMBOL(dev_close); 1646 1647 1648 /** 1649 * dev_disable_lro - disable Large Receive Offload on a device 1650 * @dev: device 1651 * 1652 * Disable Large Receive Offload (LRO) on a net device. Must be 1653 * called under RTNL. This is needed if received packets may be 1654 * forwarded to another interface. 1655 */ 1656 void dev_disable_lro(struct net_device *dev) 1657 { 1658 struct net_device *lower_dev; 1659 struct list_head *iter; 1660 1661 dev->wanted_features &= ~NETIF_F_LRO; 1662 netdev_update_features(dev); 1663 1664 if (unlikely(dev->features & NETIF_F_LRO)) 1665 netdev_WARN(dev, "failed to disable LRO!\n"); 1666 1667 netdev_for_each_lower_dev(dev, lower_dev, iter) 1668 dev_disable_lro(lower_dev); 1669 } 1670 EXPORT_SYMBOL(dev_disable_lro); 1671 1672 /** 1673 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1674 * @dev: device 1675 * 1676 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1677 * called under RTNL. This is needed if Generic XDP is installed on 1678 * the device. 1679 */ 1680 static void dev_disable_gro_hw(struct net_device *dev) 1681 { 1682 dev->wanted_features &= ~NETIF_F_GRO_HW; 1683 netdev_update_features(dev); 1684 1685 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1686 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1687 } 1688 1689 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1690 { 1691 #define N(val) \ 1692 case NETDEV_##val: \ 1693 return "NETDEV_" __stringify(val); 1694 switch (cmd) { 1695 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1696 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1697 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1698 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1699 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1700 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1701 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1702 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1703 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1704 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1705 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1706 N(XDP_FEAT_CHANGE) 1707 } 1708 #undef N 1709 return "UNKNOWN_NETDEV_EVENT"; 1710 } 1711 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1712 1713 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1714 struct net_device *dev) 1715 { 1716 struct netdev_notifier_info info = { 1717 .dev = dev, 1718 }; 1719 1720 return nb->notifier_call(nb, val, &info); 1721 } 1722 1723 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1724 struct net_device *dev) 1725 { 1726 int err; 1727 1728 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1729 err = notifier_to_errno(err); 1730 if (err) 1731 return err; 1732 1733 if (!(dev->flags & IFF_UP)) 1734 return 0; 1735 1736 call_netdevice_notifier(nb, NETDEV_UP, dev); 1737 return 0; 1738 } 1739 1740 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1741 struct net_device *dev) 1742 { 1743 if (dev->flags & IFF_UP) { 1744 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1745 dev); 1746 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1747 } 1748 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1749 } 1750 1751 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1752 struct net *net) 1753 { 1754 struct net_device *dev; 1755 int err; 1756 1757 for_each_netdev(net, dev) { 1758 err = call_netdevice_register_notifiers(nb, dev); 1759 if (err) 1760 goto rollback; 1761 } 1762 return 0; 1763 1764 rollback: 1765 for_each_netdev_continue_reverse(net, dev) 1766 call_netdevice_unregister_notifiers(nb, dev); 1767 return err; 1768 } 1769 1770 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1771 struct net *net) 1772 { 1773 struct net_device *dev; 1774 1775 for_each_netdev(net, dev) 1776 call_netdevice_unregister_notifiers(nb, dev); 1777 } 1778 1779 static int dev_boot_phase = 1; 1780 1781 /** 1782 * register_netdevice_notifier - register a network notifier block 1783 * @nb: notifier 1784 * 1785 * Register a notifier to be called when network device events occur. 1786 * The notifier passed is linked into the kernel structures and must 1787 * not be reused until it has been unregistered. A negative errno code 1788 * is returned on a failure. 1789 * 1790 * When registered all registration and up events are replayed 1791 * to the new notifier to allow device to have a race free 1792 * view of the network device list. 1793 */ 1794 1795 int register_netdevice_notifier(struct notifier_block *nb) 1796 { 1797 struct net *net; 1798 int err; 1799 1800 /* Close race with setup_net() and cleanup_net() */ 1801 down_write(&pernet_ops_rwsem); 1802 1803 /* When RTNL is removed, we need protection for netdev_chain. */ 1804 rtnl_lock(); 1805 1806 err = raw_notifier_chain_register(&netdev_chain, nb); 1807 if (err) 1808 goto unlock; 1809 if (dev_boot_phase) 1810 goto unlock; 1811 for_each_net(net) { 1812 __rtnl_net_lock(net); 1813 err = call_netdevice_register_net_notifiers(nb, net); 1814 __rtnl_net_unlock(net); 1815 if (err) 1816 goto rollback; 1817 } 1818 1819 unlock: 1820 rtnl_unlock(); 1821 up_write(&pernet_ops_rwsem); 1822 return err; 1823 1824 rollback: 1825 for_each_net_continue_reverse(net) { 1826 __rtnl_net_lock(net); 1827 call_netdevice_unregister_net_notifiers(nb, net); 1828 __rtnl_net_unlock(net); 1829 } 1830 1831 raw_notifier_chain_unregister(&netdev_chain, nb); 1832 goto unlock; 1833 } 1834 EXPORT_SYMBOL(register_netdevice_notifier); 1835 1836 /** 1837 * unregister_netdevice_notifier - unregister a network notifier block 1838 * @nb: notifier 1839 * 1840 * Unregister a notifier previously registered by 1841 * register_netdevice_notifier(). The notifier is unlinked into the 1842 * kernel structures and may then be reused. A negative errno code 1843 * is returned on a failure. 1844 * 1845 * After unregistering unregister and down device events are synthesized 1846 * for all devices on the device list to the removed notifier to remove 1847 * the need for special case cleanup code. 1848 */ 1849 1850 int unregister_netdevice_notifier(struct notifier_block *nb) 1851 { 1852 struct net *net; 1853 int err; 1854 1855 /* Close race with setup_net() and cleanup_net() */ 1856 down_write(&pernet_ops_rwsem); 1857 rtnl_lock(); 1858 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1859 if (err) 1860 goto unlock; 1861 1862 for_each_net(net) { 1863 __rtnl_net_lock(net); 1864 call_netdevice_unregister_net_notifiers(nb, net); 1865 __rtnl_net_unlock(net); 1866 } 1867 1868 unlock: 1869 rtnl_unlock(); 1870 up_write(&pernet_ops_rwsem); 1871 return err; 1872 } 1873 EXPORT_SYMBOL(unregister_netdevice_notifier); 1874 1875 static int __register_netdevice_notifier_net(struct net *net, 1876 struct notifier_block *nb, 1877 bool ignore_call_fail) 1878 { 1879 int err; 1880 1881 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1882 if (err) 1883 return err; 1884 if (dev_boot_phase) 1885 return 0; 1886 1887 err = call_netdevice_register_net_notifiers(nb, net); 1888 if (err && !ignore_call_fail) 1889 goto chain_unregister; 1890 1891 return 0; 1892 1893 chain_unregister: 1894 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1895 return err; 1896 } 1897 1898 static int __unregister_netdevice_notifier_net(struct net *net, 1899 struct notifier_block *nb) 1900 { 1901 int err; 1902 1903 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1904 if (err) 1905 return err; 1906 1907 call_netdevice_unregister_net_notifiers(nb, net); 1908 return 0; 1909 } 1910 1911 /** 1912 * register_netdevice_notifier_net - register a per-netns network notifier block 1913 * @net: network namespace 1914 * @nb: notifier 1915 * 1916 * Register a notifier to be called when network device events occur. 1917 * The notifier passed is linked into the kernel structures and must 1918 * not be reused until it has been unregistered. A negative errno code 1919 * is returned on a failure. 1920 * 1921 * When registered all registration and up events are replayed 1922 * to the new notifier to allow device to have a race free 1923 * view of the network device list. 1924 */ 1925 1926 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1927 { 1928 int err; 1929 1930 rtnl_net_lock(net); 1931 err = __register_netdevice_notifier_net(net, nb, false); 1932 rtnl_net_unlock(net); 1933 1934 return err; 1935 } 1936 EXPORT_SYMBOL(register_netdevice_notifier_net); 1937 1938 /** 1939 * unregister_netdevice_notifier_net - unregister a per-netns 1940 * network notifier block 1941 * @net: network namespace 1942 * @nb: notifier 1943 * 1944 * Unregister a notifier previously registered by 1945 * register_netdevice_notifier_net(). The notifier is unlinked from the 1946 * kernel structures and may then be reused. A negative errno code 1947 * is returned on a failure. 1948 * 1949 * After unregistering unregister and down device events are synthesized 1950 * for all devices on the device list to the removed notifier to remove 1951 * the need for special case cleanup code. 1952 */ 1953 1954 int unregister_netdevice_notifier_net(struct net *net, 1955 struct notifier_block *nb) 1956 { 1957 int err; 1958 1959 rtnl_net_lock(net); 1960 err = __unregister_netdevice_notifier_net(net, nb); 1961 rtnl_net_unlock(net); 1962 1963 return err; 1964 } 1965 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1966 1967 static void __move_netdevice_notifier_net(struct net *src_net, 1968 struct net *dst_net, 1969 struct notifier_block *nb) 1970 { 1971 __unregister_netdevice_notifier_net(src_net, nb); 1972 __register_netdevice_notifier_net(dst_net, nb, true); 1973 } 1974 1975 int register_netdevice_notifier_dev_net(struct net_device *dev, 1976 struct notifier_block *nb, 1977 struct netdev_net_notifier *nn) 1978 { 1979 struct net *net = dev_net(dev); 1980 int err; 1981 1982 rtnl_net_lock(net); 1983 err = __register_netdevice_notifier_net(net, nb, false); 1984 if (!err) { 1985 nn->nb = nb; 1986 list_add(&nn->list, &dev->net_notifier_list); 1987 } 1988 rtnl_net_unlock(net); 1989 1990 return err; 1991 } 1992 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1993 1994 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1995 struct notifier_block *nb, 1996 struct netdev_net_notifier *nn) 1997 { 1998 struct net *net = dev_net(dev); 1999 int err; 2000 2001 rtnl_net_lock(net); 2002 list_del(&nn->list); 2003 err = __unregister_netdevice_notifier_net(net, nb); 2004 rtnl_net_unlock(net); 2005 2006 return err; 2007 } 2008 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2009 2010 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2011 struct net *net) 2012 { 2013 struct netdev_net_notifier *nn; 2014 2015 list_for_each_entry(nn, &dev->net_notifier_list, list) 2016 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 2017 } 2018 2019 /** 2020 * call_netdevice_notifiers_info - call all network notifier blocks 2021 * @val: value passed unmodified to notifier function 2022 * @info: notifier information data 2023 * 2024 * Call all network notifier blocks. Parameters and return value 2025 * are as for raw_notifier_call_chain(). 2026 */ 2027 2028 int call_netdevice_notifiers_info(unsigned long val, 2029 struct netdev_notifier_info *info) 2030 { 2031 struct net *net = dev_net(info->dev); 2032 int ret; 2033 2034 ASSERT_RTNL(); 2035 2036 /* Run per-netns notifier block chain first, then run the global one. 2037 * Hopefully, one day, the global one is going to be removed after 2038 * all notifier block registrators get converted to be per-netns. 2039 */ 2040 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2041 if (ret & NOTIFY_STOP_MASK) 2042 return ret; 2043 return raw_notifier_call_chain(&netdev_chain, val, info); 2044 } 2045 2046 /** 2047 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2048 * for and rollback on error 2049 * @val_up: value passed unmodified to notifier function 2050 * @val_down: value passed unmodified to the notifier function when 2051 * recovering from an error on @val_up 2052 * @info: notifier information data 2053 * 2054 * Call all per-netns network notifier blocks, but not notifier blocks on 2055 * the global notifier chain. Parameters and return value are as for 2056 * raw_notifier_call_chain_robust(). 2057 */ 2058 2059 static int 2060 call_netdevice_notifiers_info_robust(unsigned long val_up, 2061 unsigned long val_down, 2062 struct netdev_notifier_info *info) 2063 { 2064 struct net *net = dev_net(info->dev); 2065 2066 ASSERT_RTNL(); 2067 2068 return raw_notifier_call_chain_robust(&net->netdev_chain, 2069 val_up, val_down, info); 2070 } 2071 2072 static int call_netdevice_notifiers_extack(unsigned long val, 2073 struct net_device *dev, 2074 struct netlink_ext_ack *extack) 2075 { 2076 struct netdev_notifier_info info = { 2077 .dev = dev, 2078 .extack = extack, 2079 }; 2080 2081 return call_netdevice_notifiers_info(val, &info); 2082 } 2083 2084 /** 2085 * call_netdevice_notifiers - call all network notifier blocks 2086 * @val: value passed unmodified to notifier function 2087 * @dev: net_device pointer passed unmodified to notifier function 2088 * 2089 * Call all network notifier blocks. Parameters and return value 2090 * are as for raw_notifier_call_chain(). 2091 */ 2092 2093 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2094 { 2095 return call_netdevice_notifiers_extack(val, dev, NULL); 2096 } 2097 EXPORT_SYMBOL(call_netdevice_notifiers); 2098 2099 /** 2100 * call_netdevice_notifiers_mtu - call all network notifier blocks 2101 * @val: value passed unmodified to notifier function 2102 * @dev: net_device pointer passed unmodified to notifier function 2103 * @arg: additional u32 argument passed to the notifier function 2104 * 2105 * Call all network notifier blocks. Parameters and return value 2106 * are as for raw_notifier_call_chain(). 2107 */ 2108 static int call_netdevice_notifiers_mtu(unsigned long val, 2109 struct net_device *dev, u32 arg) 2110 { 2111 struct netdev_notifier_info_ext info = { 2112 .info.dev = dev, 2113 .ext.mtu = arg, 2114 }; 2115 2116 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2117 2118 return call_netdevice_notifiers_info(val, &info.info); 2119 } 2120 2121 #ifdef CONFIG_NET_INGRESS 2122 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2123 2124 void net_inc_ingress_queue(void) 2125 { 2126 static_branch_inc(&ingress_needed_key); 2127 } 2128 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2129 2130 void net_dec_ingress_queue(void) 2131 { 2132 static_branch_dec(&ingress_needed_key); 2133 } 2134 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2135 #endif 2136 2137 #ifdef CONFIG_NET_EGRESS 2138 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2139 2140 void net_inc_egress_queue(void) 2141 { 2142 static_branch_inc(&egress_needed_key); 2143 } 2144 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2145 2146 void net_dec_egress_queue(void) 2147 { 2148 static_branch_dec(&egress_needed_key); 2149 } 2150 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2151 #endif 2152 2153 #ifdef CONFIG_NET_CLS_ACT 2154 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key); 2155 EXPORT_SYMBOL(tcf_bypass_check_needed_key); 2156 #endif 2157 2158 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2159 EXPORT_SYMBOL(netstamp_needed_key); 2160 #ifdef CONFIG_JUMP_LABEL 2161 static atomic_t netstamp_needed_deferred; 2162 static atomic_t netstamp_wanted; 2163 static void netstamp_clear(struct work_struct *work) 2164 { 2165 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2166 int wanted; 2167 2168 wanted = atomic_add_return(deferred, &netstamp_wanted); 2169 if (wanted > 0) 2170 static_branch_enable(&netstamp_needed_key); 2171 else 2172 static_branch_disable(&netstamp_needed_key); 2173 } 2174 static DECLARE_WORK(netstamp_work, netstamp_clear); 2175 #endif 2176 2177 void net_enable_timestamp(void) 2178 { 2179 #ifdef CONFIG_JUMP_LABEL 2180 int wanted = atomic_read(&netstamp_wanted); 2181 2182 while (wanted > 0) { 2183 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2184 return; 2185 } 2186 atomic_inc(&netstamp_needed_deferred); 2187 schedule_work(&netstamp_work); 2188 #else 2189 static_branch_inc(&netstamp_needed_key); 2190 #endif 2191 } 2192 EXPORT_SYMBOL(net_enable_timestamp); 2193 2194 void net_disable_timestamp(void) 2195 { 2196 #ifdef CONFIG_JUMP_LABEL 2197 int wanted = atomic_read(&netstamp_wanted); 2198 2199 while (wanted > 1) { 2200 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2201 return; 2202 } 2203 atomic_dec(&netstamp_needed_deferred); 2204 schedule_work(&netstamp_work); 2205 #else 2206 static_branch_dec(&netstamp_needed_key); 2207 #endif 2208 } 2209 EXPORT_SYMBOL(net_disable_timestamp); 2210 2211 static inline void net_timestamp_set(struct sk_buff *skb) 2212 { 2213 skb->tstamp = 0; 2214 skb->tstamp_type = SKB_CLOCK_REALTIME; 2215 if (static_branch_unlikely(&netstamp_needed_key)) 2216 skb->tstamp = ktime_get_real(); 2217 } 2218 2219 #define net_timestamp_check(COND, SKB) \ 2220 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2221 if ((COND) && !(SKB)->tstamp) \ 2222 (SKB)->tstamp = ktime_get_real(); \ 2223 } \ 2224 2225 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2226 { 2227 return __is_skb_forwardable(dev, skb, true); 2228 } 2229 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2230 2231 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2232 bool check_mtu) 2233 { 2234 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2235 2236 if (likely(!ret)) { 2237 skb->protocol = eth_type_trans(skb, dev); 2238 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2239 } 2240 2241 return ret; 2242 } 2243 2244 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2245 { 2246 return __dev_forward_skb2(dev, skb, true); 2247 } 2248 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2249 2250 /** 2251 * dev_forward_skb - loopback an skb to another netif 2252 * 2253 * @dev: destination network device 2254 * @skb: buffer to forward 2255 * 2256 * return values: 2257 * NET_RX_SUCCESS (no congestion) 2258 * NET_RX_DROP (packet was dropped, but freed) 2259 * 2260 * dev_forward_skb can be used for injecting an skb from the 2261 * start_xmit function of one device into the receive queue 2262 * of another device. 2263 * 2264 * The receiving device may be in another namespace, so 2265 * we have to clear all information in the skb that could 2266 * impact namespace isolation. 2267 */ 2268 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2269 { 2270 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2271 } 2272 EXPORT_SYMBOL_GPL(dev_forward_skb); 2273 2274 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2275 { 2276 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2277 } 2278 2279 static inline int deliver_skb(struct sk_buff *skb, 2280 struct packet_type *pt_prev, 2281 struct net_device *orig_dev) 2282 { 2283 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2284 return -ENOMEM; 2285 refcount_inc(&skb->users); 2286 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2287 } 2288 2289 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2290 struct packet_type **pt, 2291 struct net_device *orig_dev, 2292 __be16 type, 2293 struct list_head *ptype_list) 2294 { 2295 struct packet_type *ptype, *pt_prev = *pt; 2296 2297 list_for_each_entry_rcu(ptype, ptype_list, list) { 2298 if (ptype->type != type) 2299 continue; 2300 if (pt_prev) 2301 deliver_skb(skb, pt_prev, orig_dev); 2302 pt_prev = ptype; 2303 } 2304 *pt = pt_prev; 2305 } 2306 2307 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2308 { 2309 if (!ptype->af_packet_priv || !skb->sk) 2310 return false; 2311 2312 if (ptype->id_match) 2313 return ptype->id_match(ptype, skb->sk); 2314 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2315 return true; 2316 2317 return false; 2318 } 2319 2320 /** 2321 * dev_nit_active - return true if any network interface taps are in use 2322 * 2323 * @dev: network device to check for the presence of taps 2324 */ 2325 bool dev_nit_active(struct net_device *dev) 2326 { 2327 return !list_empty(&net_hotdata.ptype_all) || 2328 !list_empty(&dev->ptype_all); 2329 } 2330 EXPORT_SYMBOL_GPL(dev_nit_active); 2331 2332 /* 2333 * Support routine. Sends outgoing frames to any network 2334 * taps currently in use. 2335 */ 2336 2337 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2338 { 2339 struct list_head *ptype_list = &net_hotdata.ptype_all; 2340 struct packet_type *ptype, *pt_prev = NULL; 2341 struct sk_buff *skb2 = NULL; 2342 2343 rcu_read_lock(); 2344 again: 2345 list_for_each_entry_rcu(ptype, ptype_list, list) { 2346 if (READ_ONCE(ptype->ignore_outgoing)) 2347 continue; 2348 2349 /* Never send packets back to the socket 2350 * they originated from - MvS (miquels@drinkel.ow.org) 2351 */ 2352 if (skb_loop_sk(ptype, skb)) 2353 continue; 2354 2355 if (pt_prev) { 2356 deliver_skb(skb2, pt_prev, skb->dev); 2357 pt_prev = ptype; 2358 continue; 2359 } 2360 2361 /* need to clone skb, done only once */ 2362 skb2 = skb_clone(skb, GFP_ATOMIC); 2363 if (!skb2) 2364 goto out_unlock; 2365 2366 net_timestamp_set(skb2); 2367 2368 /* skb->nh should be correctly 2369 * set by sender, so that the second statement is 2370 * just protection against buggy protocols. 2371 */ 2372 skb_reset_mac_header(skb2); 2373 2374 if (skb_network_header(skb2) < skb2->data || 2375 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2376 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2377 ntohs(skb2->protocol), 2378 dev->name); 2379 skb_reset_network_header(skb2); 2380 } 2381 2382 skb2->transport_header = skb2->network_header; 2383 skb2->pkt_type = PACKET_OUTGOING; 2384 pt_prev = ptype; 2385 } 2386 2387 if (ptype_list == &net_hotdata.ptype_all) { 2388 ptype_list = &dev->ptype_all; 2389 goto again; 2390 } 2391 out_unlock: 2392 if (pt_prev) { 2393 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2394 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2395 else 2396 kfree_skb(skb2); 2397 } 2398 rcu_read_unlock(); 2399 } 2400 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2401 2402 /** 2403 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2404 * @dev: Network device 2405 * @txq: number of queues available 2406 * 2407 * If real_num_tx_queues is changed the tc mappings may no longer be 2408 * valid. To resolve this verify the tc mapping remains valid and if 2409 * not NULL the mapping. With no priorities mapping to this 2410 * offset/count pair it will no longer be used. In the worst case TC0 2411 * is invalid nothing can be done so disable priority mappings. If is 2412 * expected that drivers will fix this mapping if they can before 2413 * calling netif_set_real_num_tx_queues. 2414 */ 2415 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2416 { 2417 int i; 2418 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2419 2420 /* If TC0 is invalidated disable TC mapping */ 2421 if (tc->offset + tc->count > txq) { 2422 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2423 dev->num_tc = 0; 2424 return; 2425 } 2426 2427 /* Invalidated prio to tc mappings set to TC0 */ 2428 for (i = 1; i < TC_BITMASK + 1; i++) { 2429 int q = netdev_get_prio_tc_map(dev, i); 2430 2431 tc = &dev->tc_to_txq[q]; 2432 if (tc->offset + tc->count > txq) { 2433 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", 2434 i, q); 2435 netdev_set_prio_tc_map(dev, i, 0); 2436 } 2437 } 2438 } 2439 2440 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2441 { 2442 if (dev->num_tc) { 2443 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2444 int i; 2445 2446 /* walk through the TCs and see if it falls into any of them */ 2447 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2448 if ((txq - tc->offset) < tc->count) 2449 return i; 2450 } 2451 2452 /* didn't find it, just return -1 to indicate no match */ 2453 return -1; 2454 } 2455 2456 return 0; 2457 } 2458 EXPORT_SYMBOL(netdev_txq_to_tc); 2459 2460 #ifdef CONFIG_XPS 2461 static struct static_key xps_needed __read_mostly; 2462 static struct static_key xps_rxqs_needed __read_mostly; 2463 static DEFINE_MUTEX(xps_map_mutex); 2464 #define xmap_dereference(P) \ 2465 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2466 2467 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2468 struct xps_dev_maps *old_maps, int tci, u16 index) 2469 { 2470 struct xps_map *map = NULL; 2471 int pos; 2472 2473 map = xmap_dereference(dev_maps->attr_map[tci]); 2474 if (!map) 2475 return false; 2476 2477 for (pos = map->len; pos--;) { 2478 if (map->queues[pos] != index) 2479 continue; 2480 2481 if (map->len > 1) { 2482 map->queues[pos] = map->queues[--map->len]; 2483 break; 2484 } 2485 2486 if (old_maps) 2487 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2488 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2489 kfree_rcu(map, rcu); 2490 return false; 2491 } 2492 2493 return true; 2494 } 2495 2496 static bool remove_xps_queue_cpu(struct net_device *dev, 2497 struct xps_dev_maps *dev_maps, 2498 int cpu, u16 offset, u16 count) 2499 { 2500 int num_tc = dev_maps->num_tc; 2501 bool active = false; 2502 int tci; 2503 2504 for (tci = cpu * num_tc; num_tc--; tci++) { 2505 int i, j; 2506 2507 for (i = count, j = offset; i--; j++) { 2508 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2509 break; 2510 } 2511 2512 active |= i < 0; 2513 } 2514 2515 return active; 2516 } 2517 2518 static void reset_xps_maps(struct net_device *dev, 2519 struct xps_dev_maps *dev_maps, 2520 enum xps_map_type type) 2521 { 2522 static_key_slow_dec_cpuslocked(&xps_needed); 2523 if (type == XPS_RXQS) 2524 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2525 2526 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2527 2528 kfree_rcu(dev_maps, rcu); 2529 } 2530 2531 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2532 u16 offset, u16 count) 2533 { 2534 struct xps_dev_maps *dev_maps; 2535 bool active = false; 2536 int i, j; 2537 2538 dev_maps = xmap_dereference(dev->xps_maps[type]); 2539 if (!dev_maps) 2540 return; 2541 2542 for (j = 0; j < dev_maps->nr_ids; j++) 2543 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2544 if (!active) 2545 reset_xps_maps(dev, dev_maps, type); 2546 2547 if (type == XPS_CPUS) { 2548 for (i = offset + (count - 1); count--; i--) 2549 netdev_queue_numa_node_write( 2550 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2551 } 2552 } 2553 2554 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2555 u16 count) 2556 { 2557 if (!static_key_false(&xps_needed)) 2558 return; 2559 2560 cpus_read_lock(); 2561 mutex_lock(&xps_map_mutex); 2562 2563 if (static_key_false(&xps_rxqs_needed)) 2564 clean_xps_maps(dev, XPS_RXQS, offset, count); 2565 2566 clean_xps_maps(dev, XPS_CPUS, offset, count); 2567 2568 mutex_unlock(&xps_map_mutex); 2569 cpus_read_unlock(); 2570 } 2571 2572 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2573 { 2574 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2575 } 2576 2577 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2578 u16 index, bool is_rxqs_map) 2579 { 2580 struct xps_map *new_map; 2581 int alloc_len = XPS_MIN_MAP_ALLOC; 2582 int i, pos; 2583 2584 for (pos = 0; map && pos < map->len; pos++) { 2585 if (map->queues[pos] != index) 2586 continue; 2587 return map; 2588 } 2589 2590 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2591 if (map) { 2592 if (pos < map->alloc_len) 2593 return map; 2594 2595 alloc_len = map->alloc_len * 2; 2596 } 2597 2598 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2599 * map 2600 */ 2601 if (is_rxqs_map) 2602 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2603 else 2604 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2605 cpu_to_node(attr_index)); 2606 if (!new_map) 2607 return NULL; 2608 2609 for (i = 0; i < pos; i++) 2610 new_map->queues[i] = map->queues[i]; 2611 new_map->alloc_len = alloc_len; 2612 new_map->len = pos; 2613 2614 return new_map; 2615 } 2616 2617 /* Copy xps maps at a given index */ 2618 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2619 struct xps_dev_maps *new_dev_maps, int index, 2620 int tc, bool skip_tc) 2621 { 2622 int i, tci = index * dev_maps->num_tc; 2623 struct xps_map *map; 2624 2625 /* copy maps belonging to foreign traffic classes */ 2626 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2627 if (i == tc && skip_tc) 2628 continue; 2629 2630 /* fill in the new device map from the old device map */ 2631 map = xmap_dereference(dev_maps->attr_map[tci]); 2632 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2633 } 2634 } 2635 2636 /* Must be called under cpus_read_lock */ 2637 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2638 u16 index, enum xps_map_type type) 2639 { 2640 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2641 const unsigned long *online_mask = NULL; 2642 bool active = false, copy = false; 2643 int i, j, tci, numa_node_id = -2; 2644 int maps_sz, num_tc = 1, tc = 0; 2645 struct xps_map *map, *new_map; 2646 unsigned int nr_ids; 2647 2648 WARN_ON_ONCE(index >= dev->num_tx_queues); 2649 2650 if (dev->num_tc) { 2651 /* Do not allow XPS on subordinate device directly */ 2652 num_tc = dev->num_tc; 2653 if (num_tc < 0) 2654 return -EINVAL; 2655 2656 /* If queue belongs to subordinate dev use its map */ 2657 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2658 2659 tc = netdev_txq_to_tc(dev, index); 2660 if (tc < 0) 2661 return -EINVAL; 2662 } 2663 2664 mutex_lock(&xps_map_mutex); 2665 2666 dev_maps = xmap_dereference(dev->xps_maps[type]); 2667 if (type == XPS_RXQS) { 2668 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2669 nr_ids = dev->num_rx_queues; 2670 } else { 2671 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2672 if (num_possible_cpus() > 1) 2673 online_mask = cpumask_bits(cpu_online_mask); 2674 nr_ids = nr_cpu_ids; 2675 } 2676 2677 if (maps_sz < L1_CACHE_BYTES) 2678 maps_sz = L1_CACHE_BYTES; 2679 2680 /* The old dev_maps could be larger or smaller than the one we're 2681 * setting up now, as dev->num_tc or nr_ids could have been updated in 2682 * between. We could try to be smart, but let's be safe instead and only 2683 * copy foreign traffic classes if the two map sizes match. 2684 */ 2685 if (dev_maps && 2686 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2687 copy = true; 2688 2689 /* allocate memory for queue storage */ 2690 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2691 j < nr_ids;) { 2692 if (!new_dev_maps) { 2693 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2694 if (!new_dev_maps) { 2695 mutex_unlock(&xps_map_mutex); 2696 return -ENOMEM; 2697 } 2698 2699 new_dev_maps->nr_ids = nr_ids; 2700 new_dev_maps->num_tc = num_tc; 2701 } 2702 2703 tci = j * num_tc + tc; 2704 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2705 2706 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2707 if (!map) 2708 goto error; 2709 2710 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2711 } 2712 2713 if (!new_dev_maps) 2714 goto out_no_new_maps; 2715 2716 if (!dev_maps) { 2717 /* Increment static keys at most once per type */ 2718 static_key_slow_inc_cpuslocked(&xps_needed); 2719 if (type == XPS_RXQS) 2720 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2721 } 2722 2723 for (j = 0; j < nr_ids; j++) { 2724 bool skip_tc = false; 2725 2726 tci = j * num_tc + tc; 2727 if (netif_attr_test_mask(j, mask, nr_ids) && 2728 netif_attr_test_online(j, online_mask, nr_ids)) { 2729 /* add tx-queue to CPU/rx-queue maps */ 2730 int pos = 0; 2731 2732 skip_tc = true; 2733 2734 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2735 while ((pos < map->len) && (map->queues[pos] != index)) 2736 pos++; 2737 2738 if (pos == map->len) 2739 map->queues[map->len++] = index; 2740 #ifdef CONFIG_NUMA 2741 if (type == XPS_CPUS) { 2742 if (numa_node_id == -2) 2743 numa_node_id = cpu_to_node(j); 2744 else if (numa_node_id != cpu_to_node(j)) 2745 numa_node_id = -1; 2746 } 2747 #endif 2748 } 2749 2750 if (copy) 2751 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2752 skip_tc); 2753 } 2754 2755 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2756 2757 /* Cleanup old maps */ 2758 if (!dev_maps) 2759 goto out_no_old_maps; 2760 2761 for (j = 0; j < dev_maps->nr_ids; j++) { 2762 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2763 map = xmap_dereference(dev_maps->attr_map[tci]); 2764 if (!map) 2765 continue; 2766 2767 if (copy) { 2768 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2769 if (map == new_map) 2770 continue; 2771 } 2772 2773 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2774 kfree_rcu(map, rcu); 2775 } 2776 } 2777 2778 old_dev_maps = dev_maps; 2779 2780 out_no_old_maps: 2781 dev_maps = new_dev_maps; 2782 active = true; 2783 2784 out_no_new_maps: 2785 if (type == XPS_CPUS) 2786 /* update Tx queue numa node */ 2787 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2788 (numa_node_id >= 0) ? 2789 numa_node_id : NUMA_NO_NODE); 2790 2791 if (!dev_maps) 2792 goto out_no_maps; 2793 2794 /* removes tx-queue from unused CPUs/rx-queues */ 2795 for (j = 0; j < dev_maps->nr_ids; j++) { 2796 tci = j * dev_maps->num_tc; 2797 2798 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2799 if (i == tc && 2800 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2801 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2802 continue; 2803 2804 active |= remove_xps_queue(dev_maps, 2805 copy ? old_dev_maps : NULL, 2806 tci, index); 2807 } 2808 } 2809 2810 if (old_dev_maps) 2811 kfree_rcu(old_dev_maps, rcu); 2812 2813 /* free map if not active */ 2814 if (!active) 2815 reset_xps_maps(dev, dev_maps, type); 2816 2817 out_no_maps: 2818 mutex_unlock(&xps_map_mutex); 2819 2820 return 0; 2821 error: 2822 /* remove any maps that we added */ 2823 for (j = 0; j < nr_ids; j++) { 2824 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2825 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2826 map = copy ? 2827 xmap_dereference(dev_maps->attr_map[tci]) : 2828 NULL; 2829 if (new_map && new_map != map) 2830 kfree(new_map); 2831 } 2832 } 2833 2834 mutex_unlock(&xps_map_mutex); 2835 2836 kfree(new_dev_maps); 2837 return -ENOMEM; 2838 } 2839 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2840 2841 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2842 u16 index) 2843 { 2844 int ret; 2845 2846 cpus_read_lock(); 2847 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2848 cpus_read_unlock(); 2849 2850 return ret; 2851 } 2852 EXPORT_SYMBOL(netif_set_xps_queue); 2853 2854 #endif 2855 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2856 { 2857 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2858 2859 /* Unbind any subordinate channels */ 2860 while (txq-- != &dev->_tx[0]) { 2861 if (txq->sb_dev) 2862 netdev_unbind_sb_channel(dev, txq->sb_dev); 2863 } 2864 } 2865 2866 void netdev_reset_tc(struct net_device *dev) 2867 { 2868 #ifdef CONFIG_XPS 2869 netif_reset_xps_queues_gt(dev, 0); 2870 #endif 2871 netdev_unbind_all_sb_channels(dev); 2872 2873 /* Reset TC configuration of device */ 2874 dev->num_tc = 0; 2875 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2876 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2877 } 2878 EXPORT_SYMBOL(netdev_reset_tc); 2879 2880 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2881 { 2882 if (tc >= dev->num_tc) 2883 return -EINVAL; 2884 2885 #ifdef CONFIG_XPS 2886 netif_reset_xps_queues(dev, offset, count); 2887 #endif 2888 dev->tc_to_txq[tc].count = count; 2889 dev->tc_to_txq[tc].offset = offset; 2890 return 0; 2891 } 2892 EXPORT_SYMBOL(netdev_set_tc_queue); 2893 2894 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2895 { 2896 if (num_tc > TC_MAX_QUEUE) 2897 return -EINVAL; 2898 2899 #ifdef CONFIG_XPS 2900 netif_reset_xps_queues_gt(dev, 0); 2901 #endif 2902 netdev_unbind_all_sb_channels(dev); 2903 2904 dev->num_tc = num_tc; 2905 return 0; 2906 } 2907 EXPORT_SYMBOL(netdev_set_num_tc); 2908 2909 void netdev_unbind_sb_channel(struct net_device *dev, 2910 struct net_device *sb_dev) 2911 { 2912 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2913 2914 #ifdef CONFIG_XPS 2915 netif_reset_xps_queues_gt(sb_dev, 0); 2916 #endif 2917 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2918 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2919 2920 while (txq-- != &dev->_tx[0]) { 2921 if (txq->sb_dev == sb_dev) 2922 txq->sb_dev = NULL; 2923 } 2924 } 2925 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2926 2927 int netdev_bind_sb_channel_queue(struct net_device *dev, 2928 struct net_device *sb_dev, 2929 u8 tc, u16 count, u16 offset) 2930 { 2931 /* Make certain the sb_dev and dev are already configured */ 2932 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2933 return -EINVAL; 2934 2935 /* We cannot hand out queues we don't have */ 2936 if ((offset + count) > dev->real_num_tx_queues) 2937 return -EINVAL; 2938 2939 /* Record the mapping */ 2940 sb_dev->tc_to_txq[tc].count = count; 2941 sb_dev->tc_to_txq[tc].offset = offset; 2942 2943 /* Provide a way for Tx queue to find the tc_to_txq map or 2944 * XPS map for itself. 2945 */ 2946 while (count--) 2947 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2948 2949 return 0; 2950 } 2951 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2952 2953 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2954 { 2955 /* Do not use a multiqueue device to represent a subordinate channel */ 2956 if (netif_is_multiqueue(dev)) 2957 return -ENODEV; 2958 2959 /* We allow channels 1 - 32767 to be used for subordinate channels. 2960 * Channel 0 is meant to be "native" mode and used only to represent 2961 * the main root device. We allow writing 0 to reset the device back 2962 * to normal mode after being used as a subordinate channel. 2963 */ 2964 if (channel > S16_MAX) 2965 return -EINVAL; 2966 2967 dev->num_tc = -channel; 2968 2969 return 0; 2970 } 2971 EXPORT_SYMBOL(netdev_set_sb_channel); 2972 2973 /* 2974 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2975 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2976 */ 2977 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2978 { 2979 bool disabling; 2980 int rc; 2981 2982 disabling = txq < dev->real_num_tx_queues; 2983 2984 if (txq < 1 || txq > dev->num_tx_queues) 2985 return -EINVAL; 2986 2987 if (dev->reg_state == NETREG_REGISTERED || 2988 dev->reg_state == NETREG_UNREGISTERING) { 2989 ASSERT_RTNL(); 2990 2991 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2992 txq); 2993 if (rc) 2994 return rc; 2995 2996 if (dev->num_tc) 2997 netif_setup_tc(dev, txq); 2998 2999 net_shaper_set_real_num_tx_queues(dev, txq); 3000 3001 dev_qdisc_change_real_num_tx(dev, txq); 3002 3003 dev->real_num_tx_queues = txq; 3004 3005 if (disabling) { 3006 synchronize_net(); 3007 qdisc_reset_all_tx_gt(dev, txq); 3008 #ifdef CONFIG_XPS 3009 netif_reset_xps_queues_gt(dev, txq); 3010 #endif 3011 } 3012 } else { 3013 dev->real_num_tx_queues = txq; 3014 } 3015 3016 return 0; 3017 } 3018 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3019 3020 #ifdef CONFIG_SYSFS 3021 /** 3022 * netif_set_real_num_rx_queues - set actual number of RX queues used 3023 * @dev: Network device 3024 * @rxq: Actual number of RX queues 3025 * 3026 * This must be called either with the rtnl_lock held or before 3027 * registration of the net device. Returns 0 on success, or a 3028 * negative error code. If called before registration, it always 3029 * succeeds. 3030 */ 3031 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3032 { 3033 int rc; 3034 3035 if (rxq < 1 || rxq > dev->num_rx_queues) 3036 return -EINVAL; 3037 3038 if (dev->reg_state == NETREG_REGISTERED) { 3039 ASSERT_RTNL(); 3040 3041 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3042 rxq); 3043 if (rc) 3044 return rc; 3045 } 3046 3047 dev->real_num_rx_queues = rxq; 3048 return 0; 3049 } 3050 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3051 #endif 3052 3053 /** 3054 * netif_set_real_num_queues - set actual number of RX and TX queues used 3055 * @dev: Network device 3056 * @txq: Actual number of TX queues 3057 * @rxq: Actual number of RX queues 3058 * 3059 * Set the real number of both TX and RX queues. 3060 * Does nothing if the number of queues is already correct. 3061 */ 3062 int netif_set_real_num_queues(struct net_device *dev, 3063 unsigned int txq, unsigned int rxq) 3064 { 3065 unsigned int old_rxq = dev->real_num_rx_queues; 3066 int err; 3067 3068 if (txq < 1 || txq > dev->num_tx_queues || 3069 rxq < 1 || rxq > dev->num_rx_queues) 3070 return -EINVAL; 3071 3072 /* Start from increases, so the error path only does decreases - 3073 * decreases can't fail. 3074 */ 3075 if (rxq > dev->real_num_rx_queues) { 3076 err = netif_set_real_num_rx_queues(dev, rxq); 3077 if (err) 3078 return err; 3079 } 3080 if (txq > dev->real_num_tx_queues) { 3081 err = netif_set_real_num_tx_queues(dev, txq); 3082 if (err) 3083 goto undo_rx; 3084 } 3085 if (rxq < dev->real_num_rx_queues) 3086 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3087 if (txq < dev->real_num_tx_queues) 3088 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3089 3090 return 0; 3091 undo_rx: 3092 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3093 return err; 3094 } 3095 EXPORT_SYMBOL(netif_set_real_num_queues); 3096 3097 /** 3098 * netif_set_tso_max_size() - set the max size of TSO frames supported 3099 * @dev: netdev to update 3100 * @size: max skb->len of a TSO frame 3101 * 3102 * Set the limit on the size of TSO super-frames the device can handle. 3103 * Unless explicitly set the stack will assume the value of 3104 * %GSO_LEGACY_MAX_SIZE. 3105 */ 3106 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3107 { 3108 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3109 if (size < READ_ONCE(dev->gso_max_size)) 3110 netif_set_gso_max_size(dev, size); 3111 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3112 netif_set_gso_ipv4_max_size(dev, size); 3113 } 3114 EXPORT_SYMBOL(netif_set_tso_max_size); 3115 3116 /** 3117 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3118 * @dev: netdev to update 3119 * @segs: max number of TCP segments 3120 * 3121 * Set the limit on the number of TCP segments the device can generate from 3122 * a single TSO super-frame. 3123 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3124 */ 3125 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3126 { 3127 dev->tso_max_segs = segs; 3128 if (segs < READ_ONCE(dev->gso_max_segs)) 3129 netif_set_gso_max_segs(dev, segs); 3130 } 3131 EXPORT_SYMBOL(netif_set_tso_max_segs); 3132 3133 /** 3134 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3135 * @to: netdev to update 3136 * @from: netdev from which to copy the limits 3137 */ 3138 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3139 { 3140 netif_set_tso_max_size(to, from->tso_max_size); 3141 netif_set_tso_max_segs(to, from->tso_max_segs); 3142 } 3143 EXPORT_SYMBOL(netif_inherit_tso_max); 3144 3145 /** 3146 * netif_get_num_default_rss_queues - default number of RSS queues 3147 * 3148 * Default value is the number of physical cores if there are only 1 or 2, or 3149 * divided by 2 if there are more. 3150 */ 3151 int netif_get_num_default_rss_queues(void) 3152 { 3153 cpumask_var_t cpus; 3154 int cpu, count = 0; 3155 3156 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3157 return 1; 3158 3159 cpumask_copy(cpus, cpu_online_mask); 3160 for_each_cpu(cpu, cpus) { 3161 ++count; 3162 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3163 } 3164 free_cpumask_var(cpus); 3165 3166 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3167 } 3168 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3169 3170 static void __netif_reschedule(struct Qdisc *q) 3171 { 3172 struct softnet_data *sd; 3173 unsigned long flags; 3174 3175 local_irq_save(flags); 3176 sd = this_cpu_ptr(&softnet_data); 3177 q->next_sched = NULL; 3178 *sd->output_queue_tailp = q; 3179 sd->output_queue_tailp = &q->next_sched; 3180 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3181 local_irq_restore(flags); 3182 } 3183 3184 void __netif_schedule(struct Qdisc *q) 3185 { 3186 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3187 __netif_reschedule(q); 3188 } 3189 EXPORT_SYMBOL(__netif_schedule); 3190 3191 struct dev_kfree_skb_cb { 3192 enum skb_drop_reason reason; 3193 }; 3194 3195 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3196 { 3197 return (struct dev_kfree_skb_cb *)skb->cb; 3198 } 3199 3200 void netif_schedule_queue(struct netdev_queue *txq) 3201 { 3202 rcu_read_lock(); 3203 if (!netif_xmit_stopped(txq)) { 3204 struct Qdisc *q = rcu_dereference(txq->qdisc); 3205 3206 __netif_schedule(q); 3207 } 3208 rcu_read_unlock(); 3209 } 3210 EXPORT_SYMBOL(netif_schedule_queue); 3211 3212 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3213 { 3214 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3215 struct Qdisc *q; 3216 3217 rcu_read_lock(); 3218 q = rcu_dereference(dev_queue->qdisc); 3219 __netif_schedule(q); 3220 rcu_read_unlock(); 3221 } 3222 } 3223 EXPORT_SYMBOL(netif_tx_wake_queue); 3224 3225 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3226 { 3227 unsigned long flags; 3228 3229 if (unlikely(!skb)) 3230 return; 3231 3232 if (likely(refcount_read(&skb->users) == 1)) { 3233 smp_rmb(); 3234 refcount_set(&skb->users, 0); 3235 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3236 return; 3237 } 3238 get_kfree_skb_cb(skb)->reason = reason; 3239 local_irq_save(flags); 3240 skb->next = __this_cpu_read(softnet_data.completion_queue); 3241 __this_cpu_write(softnet_data.completion_queue, skb); 3242 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3243 local_irq_restore(flags); 3244 } 3245 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3246 3247 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3248 { 3249 if (in_hardirq() || irqs_disabled()) 3250 dev_kfree_skb_irq_reason(skb, reason); 3251 else 3252 kfree_skb_reason(skb, reason); 3253 } 3254 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3255 3256 3257 /** 3258 * netif_device_detach - mark device as removed 3259 * @dev: network device 3260 * 3261 * Mark device as removed from system and therefore no longer available. 3262 */ 3263 void netif_device_detach(struct net_device *dev) 3264 { 3265 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3266 netif_running(dev)) { 3267 netif_tx_stop_all_queues(dev); 3268 } 3269 } 3270 EXPORT_SYMBOL(netif_device_detach); 3271 3272 /** 3273 * netif_device_attach - mark device as attached 3274 * @dev: network device 3275 * 3276 * Mark device as attached from system and restart if needed. 3277 */ 3278 void netif_device_attach(struct net_device *dev) 3279 { 3280 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3281 netif_running(dev)) { 3282 netif_tx_wake_all_queues(dev); 3283 netdev_watchdog_up(dev); 3284 } 3285 } 3286 EXPORT_SYMBOL(netif_device_attach); 3287 3288 /* 3289 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3290 * to be used as a distribution range. 3291 */ 3292 static u16 skb_tx_hash(const struct net_device *dev, 3293 const struct net_device *sb_dev, 3294 struct sk_buff *skb) 3295 { 3296 u32 hash; 3297 u16 qoffset = 0; 3298 u16 qcount = dev->real_num_tx_queues; 3299 3300 if (dev->num_tc) { 3301 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3302 3303 qoffset = sb_dev->tc_to_txq[tc].offset; 3304 qcount = sb_dev->tc_to_txq[tc].count; 3305 if (unlikely(!qcount)) { 3306 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3307 sb_dev->name, qoffset, tc); 3308 qoffset = 0; 3309 qcount = dev->real_num_tx_queues; 3310 } 3311 } 3312 3313 if (skb_rx_queue_recorded(skb)) { 3314 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3315 hash = skb_get_rx_queue(skb); 3316 if (hash >= qoffset) 3317 hash -= qoffset; 3318 while (unlikely(hash >= qcount)) 3319 hash -= qcount; 3320 return hash + qoffset; 3321 } 3322 3323 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3324 } 3325 3326 void skb_warn_bad_offload(const struct sk_buff *skb) 3327 { 3328 static const netdev_features_t null_features; 3329 struct net_device *dev = skb->dev; 3330 const char *name = ""; 3331 3332 if (!net_ratelimit()) 3333 return; 3334 3335 if (dev) { 3336 if (dev->dev.parent) 3337 name = dev_driver_string(dev->dev.parent); 3338 else 3339 name = netdev_name(dev); 3340 } 3341 skb_dump(KERN_WARNING, skb, false); 3342 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3343 name, dev ? &dev->features : &null_features, 3344 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3345 } 3346 3347 /* 3348 * Invalidate hardware checksum when packet is to be mangled, and 3349 * complete checksum manually on outgoing path. 3350 */ 3351 int skb_checksum_help(struct sk_buff *skb) 3352 { 3353 __wsum csum; 3354 int ret = 0, offset; 3355 3356 if (skb->ip_summed == CHECKSUM_COMPLETE) 3357 goto out_set_summed; 3358 3359 if (unlikely(skb_is_gso(skb))) { 3360 skb_warn_bad_offload(skb); 3361 return -EINVAL; 3362 } 3363 3364 if (!skb_frags_readable(skb)) { 3365 return -EFAULT; 3366 } 3367 3368 /* Before computing a checksum, we should make sure no frag could 3369 * be modified by an external entity : checksum could be wrong. 3370 */ 3371 if (skb_has_shared_frag(skb)) { 3372 ret = __skb_linearize(skb); 3373 if (ret) 3374 goto out; 3375 } 3376 3377 offset = skb_checksum_start_offset(skb); 3378 ret = -EINVAL; 3379 if (unlikely(offset >= skb_headlen(skb))) { 3380 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3381 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3382 offset, skb_headlen(skb)); 3383 goto out; 3384 } 3385 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3386 3387 offset += skb->csum_offset; 3388 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3389 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3390 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3391 offset + sizeof(__sum16), skb_headlen(skb)); 3392 goto out; 3393 } 3394 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3395 if (ret) 3396 goto out; 3397 3398 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3399 out_set_summed: 3400 skb->ip_summed = CHECKSUM_NONE; 3401 out: 3402 return ret; 3403 } 3404 EXPORT_SYMBOL(skb_checksum_help); 3405 3406 int skb_crc32c_csum_help(struct sk_buff *skb) 3407 { 3408 __le32 crc32c_csum; 3409 int ret = 0, offset, start; 3410 3411 if (skb->ip_summed != CHECKSUM_PARTIAL) 3412 goto out; 3413 3414 if (unlikely(skb_is_gso(skb))) 3415 goto out; 3416 3417 /* Before computing a checksum, we should make sure no frag could 3418 * be modified by an external entity : checksum could be wrong. 3419 */ 3420 if (unlikely(skb_has_shared_frag(skb))) { 3421 ret = __skb_linearize(skb); 3422 if (ret) 3423 goto out; 3424 } 3425 start = skb_checksum_start_offset(skb); 3426 offset = start + offsetof(struct sctphdr, checksum); 3427 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3428 ret = -EINVAL; 3429 goto out; 3430 } 3431 3432 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3433 if (ret) 3434 goto out; 3435 3436 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3437 skb->len - start, ~(__u32)0, 3438 crc32c_csum_stub)); 3439 *(__le32 *)(skb->data + offset) = crc32c_csum; 3440 skb_reset_csum_not_inet(skb); 3441 out: 3442 return ret; 3443 } 3444 EXPORT_SYMBOL(skb_crc32c_csum_help); 3445 3446 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3447 { 3448 __be16 type = skb->protocol; 3449 3450 /* Tunnel gso handlers can set protocol to ethernet. */ 3451 if (type == htons(ETH_P_TEB)) { 3452 struct ethhdr *eth; 3453 3454 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3455 return 0; 3456 3457 eth = (struct ethhdr *)skb->data; 3458 type = eth->h_proto; 3459 } 3460 3461 return vlan_get_protocol_and_depth(skb, type, depth); 3462 } 3463 3464 3465 /* Take action when hardware reception checksum errors are detected. */ 3466 #ifdef CONFIG_BUG 3467 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3468 { 3469 netdev_err(dev, "hw csum failure\n"); 3470 skb_dump(KERN_ERR, skb, true); 3471 dump_stack(); 3472 } 3473 3474 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3475 { 3476 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3477 } 3478 EXPORT_SYMBOL(netdev_rx_csum_fault); 3479 #endif 3480 3481 /* XXX: check that highmem exists at all on the given machine. */ 3482 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3483 { 3484 #ifdef CONFIG_HIGHMEM 3485 int i; 3486 3487 if (!(dev->features & NETIF_F_HIGHDMA)) { 3488 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3489 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3490 struct page *page = skb_frag_page(frag); 3491 3492 if (page && PageHighMem(page)) 3493 return 1; 3494 } 3495 } 3496 #endif 3497 return 0; 3498 } 3499 3500 /* If MPLS offload request, verify we are testing hardware MPLS features 3501 * instead of standard features for the netdev. 3502 */ 3503 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3504 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3505 netdev_features_t features, 3506 __be16 type) 3507 { 3508 if (eth_p_mpls(type)) 3509 features &= skb->dev->mpls_features; 3510 3511 return features; 3512 } 3513 #else 3514 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3515 netdev_features_t features, 3516 __be16 type) 3517 { 3518 return features; 3519 } 3520 #endif 3521 3522 static netdev_features_t harmonize_features(struct sk_buff *skb, 3523 netdev_features_t features) 3524 { 3525 __be16 type; 3526 3527 type = skb_network_protocol(skb, NULL); 3528 features = net_mpls_features(skb, features, type); 3529 3530 if (skb->ip_summed != CHECKSUM_NONE && 3531 !can_checksum_protocol(features, type)) { 3532 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3533 } 3534 if (illegal_highdma(skb->dev, skb)) 3535 features &= ~NETIF_F_SG; 3536 3537 return features; 3538 } 3539 3540 netdev_features_t passthru_features_check(struct sk_buff *skb, 3541 struct net_device *dev, 3542 netdev_features_t features) 3543 { 3544 return features; 3545 } 3546 EXPORT_SYMBOL(passthru_features_check); 3547 3548 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3549 struct net_device *dev, 3550 netdev_features_t features) 3551 { 3552 return vlan_features_check(skb, features); 3553 } 3554 3555 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3556 struct net_device *dev, 3557 netdev_features_t features) 3558 { 3559 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3560 3561 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3562 return features & ~NETIF_F_GSO_MASK; 3563 3564 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb))) 3565 return features & ~NETIF_F_GSO_MASK; 3566 3567 if (!skb_shinfo(skb)->gso_type) { 3568 skb_warn_bad_offload(skb); 3569 return features & ~NETIF_F_GSO_MASK; 3570 } 3571 3572 /* Support for GSO partial features requires software 3573 * intervention before we can actually process the packets 3574 * so we need to strip support for any partial features now 3575 * and we can pull them back in after we have partially 3576 * segmented the frame. 3577 */ 3578 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3579 features &= ~dev->gso_partial_features; 3580 3581 /* Make sure to clear the IPv4 ID mangling feature if the 3582 * IPv4 header has the potential to be fragmented. 3583 */ 3584 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3585 struct iphdr *iph = skb->encapsulation ? 3586 inner_ip_hdr(skb) : ip_hdr(skb); 3587 3588 if (!(iph->frag_off & htons(IP_DF))) 3589 features &= ~NETIF_F_TSO_MANGLEID; 3590 } 3591 3592 return features; 3593 } 3594 3595 netdev_features_t netif_skb_features(struct sk_buff *skb) 3596 { 3597 struct net_device *dev = skb->dev; 3598 netdev_features_t features = dev->features; 3599 3600 if (skb_is_gso(skb)) 3601 features = gso_features_check(skb, dev, features); 3602 3603 /* If encapsulation offload request, verify we are testing 3604 * hardware encapsulation features instead of standard 3605 * features for the netdev 3606 */ 3607 if (skb->encapsulation) 3608 features &= dev->hw_enc_features; 3609 3610 if (skb_vlan_tagged(skb)) 3611 features = netdev_intersect_features(features, 3612 dev->vlan_features | 3613 NETIF_F_HW_VLAN_CTAG_TX | 3614 NETIF_F_HW_VLAN_STAG_TX); 3615 3616 if (dev->netdev_ops->ndo_features_check) 3617 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3618 features); 3619 else 3620 features &= dflt_features_check(skb, dev, features); 3621 3622 return harmonize_features(skb, features); 3623 } 3624 EXPORT_SYMBOL(netif_skb_features); 3625 3626 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3627 struct netdev_queue *txq, bool more) 3628 { 3629 unsigned int len; 3630 int rc; 3631 3632 if (dev_nit_active(dev)) 3633 dev_queue_xmit_nit(skb, dev); 3634 3635 len = skb->len; 3636 trace_net_dev_start_xmit(skb, dev); 3637 rc = netdev_start_xmit(skb, dev, txq, more); 3638 trace_net_dev_xmit(skb, rc, dev, len); 3639 3640 return rc; 3641 } 3642 3643 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3644 struct netdev_queue *txq, int *ret) 3645 { 3646 struct sk_buff *skb = first; 3647 int rc = NETDEV_TX_OK; 3648 3649 while (skb) { 3650 struct sk_buff *next = skb->next; 3651 3652 skb_mark_not_on_list(skb); 3653 rc = xmit_one(skb, dev, txq, next != NULL); 3654 if (unlikely(!dev_xmit_complete(rc))) { 3655 skb->next = next; 3656 goto out; 3657 } 3658 3659 skb = next; 3660 if (netif_tx_queue_stopped(txq) && skb) { 3661 rc = NETDEV_TX_BUSY; 3662 break; 3663 } 3664 } 3665 3666 out: 3667 *ret = rc; 3668 return skb; 3669 } 3670 3671 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3672 netdev_features_t features) 3673 { 3674 if (skb_vlan_tag_present(skb) && 3675 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3676 skb = __vlan_hwaccel_push_inside(skb); 3677 return skb; 3678 } 3679 3680 int skb_csum_hwoffload_help(struct sk_buff *skb, 3681 const netdev_features_t features) 3682 { 3683 if (unlikely(skb_csum_is_sctp(skb))) 3684 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3685 skb_crc32c_csum_help(skb); 3686 3687 if (features & NETIF_F_HW_CSUM) 3688 return 0; 3689 3690 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3691 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) && 3692 skb_network_header_len(skb) != sizeof(struct ipv6hdr) && 3693 !ipv6_has_hopopt_jumbo(skb)) 3694 goto sw_checksum; 3695 3696 switch (skb->csum_offset) { 3697 case offsetof(struct tcphdr, check): 3698 case offsetof(struct udphdr, check): 3699 return 0; 3700 } 3701 } 3702 3703 sw_checksum: 3704 return skb_checksum_help(skb); 3705 } 3706 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3707 3708 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3709 { 3710 netdev_features_t features; 3711 3712 features = netif_skb_features(skb); 3713 skb = validate_xmit_vlan(skb, features); 3714 if (unlikely(!skb)) 3715 goto out_null; 3716 3717 skb = sk_validate_xmit_skb(skb, dev); 3718 if (unlikely(!skb)) 3719 goto out_null; 3720 3721 if (netif_needs_gso(skb, features)) { 3722 struct sk_buff *segs; 3723 3724 segs = skb_gso_segment(skb, features); 3725 if (IS_ERR(segs)) { 3726 goto out_kfree_skb; 3727 } else if (segs) { 3728 consume_skb(skb); 3729 skb = segs; 3730 } 3731 } else { 3732 if (skb_needs_linearize(skb, features) && 3733 __skb_linearize(skb)) 3734 goto out_kfree_skb; 3735 3736 /* If packet is not checksummed and device does not 3737 * support checksumming for this protocol, complete 3738 * checksumming here. 3739 */ 3740 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3741 if (skb->encapsulation) 3742 skb_set_inner_transport_header(skb, 3743 skb_checksum_start_offset(skb)); 3744 else 3745 skb_set_transport_header(skb, 3746 skb_checksum_start_offset(skb)); 3747 if (skb_csum_hwoffload_help(skb, features)) 3748 goto out_kfree_skb; 3749 } 3750 } 3751 3752 skb = validate_xmit_xfrm(skb, features, again); 3753 3754 return skb; 3755 3756 out_kfree_skb: 3757 kfree_skb(skb); 3758 out_null: 3759 dev_core_stats_tx_dropped_inc(dev); 3760 return NULL; 3761 } 3762 3763 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3764 { 3765 struct sk_buff *next, *head = NULL, *tail; 3766 3767 for (; skb != NULL; skb = next) { 3768 next = skb->next; 3769 skb_mark_not_on_list(skb); 3770 3771 /* in case skb won't be segmented, point to itself */ 3772 skb->prev = skb; 3773 3774 skb = validate_xmit_skb(skb, dev, again); 3775 if (!skb) 3776 continue; 3777 3778 if (!head) 3779 head = skb; 3780 else 3781 tail->next = skb; 3782 /* If skb was segmented, skb->prev points to 3783 * the last segment. If not, it still contains skb. 3784 */ 3785 tail = skb->prev; 3786 } 3787 return head; 3788 } 3789 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3790 3791 static void qdisc_pkt_len_init(struct sk_buff *skb) 3792 { 3793 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3794 3795 qdisc_skb_cb(skb)->pkt_len = skb->len; 3796 3797 /* To get more precise estimation of bytes sent on wire, 3798 * we add to pkt_len the headers size of all segments 3799 */ 3800 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3801 u16 gso_segs = shinfo->gso_segs; 3802 unsigned int hdr_len; 3803 3804 /* mac layer + network layer */ 3805 hdr_len = skb_transport_offset(skb); 3806 3807 /* + transport layer */ 3808 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3809 const struct tcphdr *th; 3810 struct tcphdr _tcphdr; 3811 3812 th = skb_header_pointer(skb, hdr_len, 3813 sizeof(_tcphdr), &_tcphdr); 3814 if (likely(th)) 3815 hdr_len += __tcp_hdrlen(th); 3816 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 3817 struct udphdr _udphdr; 3818 3819 if (skb_header_pointer(skb, hdr_len, 3820 sizeof(_udphdr), &_udphdr)) 3821 hdr_len += sizeof(struct udphdr); 3822 } 3823 3824 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) { 3825 int payload = skb->len - hdr_len; 3826 3827 /* Malicious packet. */ 3828 if (payload <= 0) 3829 return; 3830 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size); 3831 } 3832 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3833 } 3834 } 3835 3836 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3837 struct sk_buff **to_free, 3838 struct netdev_queue *txq) 3839 { 3840 int rc; 3841 3842 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3843 if (rc == NET_XMIT_SUCCESS) 3844 trace_qdisc_enqueue(q, txq, skb); 3845 return rc; 3846 } 3847 3848 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3849 struct net_device *dev, 3850 struct netdev_queue *txq) 3851 { 3852 spinlock_t *root_lock = qdisc_lock(q); 3853 struct sk_buff *to_free = NULL; 3854 bool contended; 3855 int rc; 3856 3857 qdisc_calculate_pkt_len(skb, q); 3858 3859 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3860 3861 if (q->flags & TCQ_F_NOLOCK) { 3862 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3863 qdisc_run_begin(q)) { 3864 /* Retest nolock_qdisc_is_empty() within the protection 3865 * of q->seqlock to protect from racing with requeuing. 3866 */ 3867 if (unlikely(!nolock_qdisc_is_empty(q))) { 3868 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3869 __qdisc_run(q); 3870 qdisc_run_end(q); 3871 3872 goto no_lock_out; 3873 } 3874 3875 qdisc_bstats_cpu_update(q, skb); 3876 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3877 !nolock_qdisc_is_empty(q)) 3878 __qdisc_run(q); 3879 3880 qdisc_run_end(q); 3881 return NET_XMIT_SUCCESS; 3882 } 3883 3884 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3885 qdisc_run(q); 3886 3887 no_lock_out: 3888 if (unlikely(to_free)) 3889 kfree_skb_list_reason(to_free, 3890 tcf_get_drop_reason(to_free)); 3891 return rc; 3892 } 3893 3894 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) { 3895 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP); 3896 return NET_XMIT_DROP; 3897 } 3898 /* 3899 * Heuristic to force contended enqueues to serialize on a 3900 * separate lock before trying to get qdisc main lock. 3901 * This permits qdisc->running owner to get the lock more 3902 * often and dequeue packets faster. 3903 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3904 * and then other tasks will only enqueue packets. The packets will be 3905 * sent after the qdisc owner is scheduled again. To prevent this 3906 * scenario the task always serialize on the lock. 3907 */ 3908 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3909 if (unlikely(contended)) 3910 spin_lock(&q->busylock); 3911 3912 spin_lock(root_lock); 3913 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3914 __qdisc_drop(skb, &to_free); 3915 rc = NET_XMIT_DROP; 3916 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3917 qdisc_run_begin(q)) { 3918 /* 3919 * This is a work-conserving queue; there are no old skbs 3920 * waiting to be sent out; and the qdisc is not running - 3921 * xmit the skb directly. 3922 */ 3923 3924 qdisc_bstats_update(q, skb); 3925 3926 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3927 if (unlikely(contended)) { 3928 spin_unlock(&q->busylock); 3929 contended = false; 3930 } 3931 __qdisc_run(q); 3932 } 3933 3934 qdisc_run_end(q); 3935 rc = NET_XMIT_SUCCESS; 3936 } else { 3937 WRITE_ONCE(q->owner, smp_processor_id()); 3938 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3939 WRITE_ONCE(q->owner, -1); 3940 if (qdisc_run_begin(q)) { 3941 if (unlikely(contended)) { 3942 spin_unlock(&q->busylock); 3943 contended = false; 3944 } 3945 __qdisc_run(q); 3946 qdisc_run_end(q); 3947 } 3948 } 3949 spin_unlock(root_lock); 3950 if (unlikely(to_free)) 3951 kfree_skb_list_reason(to_free, 3952 tcf_get_drop_reason(to_free)); 3953 if (unlikely(contended)) 3954 spin_unlock(&q->busylock); 3955 return rc; 3956 } 3957 3958 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3959 static void skb_update_prio(struct sk_buff *skb) 3960 { 3961 const struct netprio_map *map; 3962 const struct sock *sk; 3963 unsigned int prioidx; 3964 3965 if (skb->priority) 3966 return; 3967 map = rcu_dereference_bh(skb->dev->priomap); 3968 if (!map) 3969 return; 3970 sk = skb_to_full_sk(skb); 3971 if (!sk) 3972 return; 3973 3974 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3975 3976 if (prioidx < map->priomap_len) 3977 skb->priority = map->priomap[prioidx]; 3978 } 3979 #else 3980 #define skb_update_prio(skb) 3981 #endif 3982 3983 /** 3984 * dev_loopback_xmit - loop back @skb 3985 * @net: network namespace this loopback is happening in 3986 * @sk: sk needed to be a netfilter okfn 3987 * @skb: buffer to transmit 3988 */ 3989 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3990 { 3991 skb_reset_mac_header(skb); 3992 __skb_pull(skb, skb_network_offset(skb)); 3993 skb->pkt_type = PACKET_LOOPBACK; 3994 if (skb->ip_summed == CHECKSUM_NONE) 3995 skb->ip_summed = CHECKSUM_UNNECESSARY; 3996 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3997 skb_dst_force(skb); 3998 netif_rx(skb); 3999 return 0; 4000 } 4001 EXPORT_SYMBOL(dev_loopback_xmit); 4002 4003 #ifdef CONFIG_NET_EGRESS 4004 static struct netdev_queue * 4005 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 4006 { 4007 int qm = skb_get_queue_mapping(skb); 4008 4009 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 4010 } 4011 4012 #ifndef CONFIG_PREEMPT_RT 4013 static bool netdev_xmit_txqueue_skipped(void) 4014 { 4015 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 4016 } 4017 4018 void netdev_xmit_skip_txqueue(bool skip) 4019 { 4020 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 4021 } 4022 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4023 4024 #else 4025 static bool netdev_xmit_txqueue_skipped(void) 4026 { 4027 return current->net_xmit.skip_txqueue; 4028 } 4029 4030 void netdev_xmit_skip_txqueue(bool skip) 4031 { 4032 current->net_xmit.skip_txqueue = skip; 4033 } 4034 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4035 #endif 4036 #endif /* CONFIG_NET_EGRESS */ 4037 4038 #ifdef CONFIG_NET_XGRESS 4039 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 4040 enum skb_drop_reason *drop_reason) 4041 { 4042 int ret = TC_ACT_UNSPEC; 4043 #ifdef CONFIG_NET_CLS_ACT 4044 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 4045 struct tcf_result res; 4046 4047 if (!miniq) 4048 return ret; 4049 4050 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) { 4051 if (tcf_block_bypass_sw(miniq->block)) 4052 return ret; 4053 } 4054 4055 tc_skb_cb(skb)->mru = 0; 4056 tc_skb_cb(skb)->post_ct = false; 4057 tcf_set_drop_reason(skb, *drop_reason); 4058 4059 mini_qdisc_bstats_cpu_update(miniq, skb); 4060 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 4061 /* Only tcf related quirks below. */ 4062 switch (ret) { 4063 case TC_ACT_SHOT: 4064 *drop_reason = tcf_get_drop_reason(skb); 4065 mini_qdisc_qstats_cpu_drop(miniq); 4066 break; 4067 case TC_ACT_OK: 4068 case TC_ACT_RECLASSIFY: 4069 skb->tc_index = TC_H_MIN(res.classid); 4070 break; 4071 } 4072 #endif /* CONFIG_NET_CLS_ACT */ 4073 return ret; 4074 } 4075 4076 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 4077 4078 void tcx_inc(void) 4079 { 4080 static_branch_inc(&tcx_needed_key); 4081 } 4082 4083 void tcx_dec(void) 4084 { 4085 static_branch_dec(&tcx_needed_key); 4086 } 4087 4088 static __always_inline enum tcx_action_base 4089 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 4090 const bool needs_mac) 4091 { 4092 const struct bpf_mprog_fp *fp; 4093 const struct bpf_prog *prog; 4094 int ret = TCX_NEXT; 4095 4096 if (needs_mac) 4097 __skb_push(skb, skb->mac_len); 4098 bpf_mprog_foreach_prog(entry, fp, prog) { 4099 bpf_compute_data_pointers(skb); 4100 ret = bpf_prog_run(prog, skb); 4101 if (ret != TCX_NEXT) 4102 break; 4103 } 4104 if (needs_mac) 4105 __skb_pull(skb, skb->mac_len); 4106 return tcx_action_code(skb, ret); 4107 } 4108 4109 static __always_inline struct sk_buff * 4110 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4111 struct net_device *orig_dev, bool *another) 4112 { 4113 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4114 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 4115 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4116 int sch_ret; 4117 4118 if (!entry) 4119 return skb; 4120 4121 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4122 if (*pt_prev) { 4123 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4124 *pt_prev = NULL; 4125 } 4126 4127 qdisc_skb_cb(skb)->pkt_len = skb->len; 4128 tcx_set_ingress(skb, true); 4129 4130 if (static_branch_unlikely(&tcx_needed_key)) { 4131 sch_ret = tcx_run(entry, skb, true); 4132 if (sch_ret != TC_ACT_UNSPEC) 4133 goto ingress_verdict; 4134 } 4135 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4136 ingress_verdict: 4137 switch (sch_ret) { 4138 case TC_ACT_REDIRECT: 4139 /* skb_mac_header check was done by BPF, so we can safely 4140 * push the L2 header back before redirecting to another 4141 * netdev. 4142 */ 4143 __skb_push(skb, skb->mac_len); 4144 if (skb_do_redirect(skb) == -EAGAIN) { 4145 __skb_pull(skb, skb->mac_len); 4146 *another = true; 4147 break; 4148 } 4149 *ret = NET_RX_SUCCESS; 4150 bpf_net_ctx_clear(bpf_net_ctx); 4151 return NULL; 4152 case TC_ACT_SHOT: 4153 kfree_skb_reason(skb, drop_reason); 4154 *ret = NET_RX_DROP; 4155 bpf_net_ctx_clear(bpf_net_ctx); 4156 return NULL; 4157 /* used by tc_run */ 4158 case TC_ACT_STOLEN: 4159 case TC_ACT_QUEUED: 4160 case TC_ACT_TRAP: 4161 consume_skb(skb); 4162 fallthrough; 4163 case TC_ACT_CONSUMED: 4164 *ret = NET_RX_SUCCESS; 4165 bpf_net_ctx_clear(bpf_net_ctx); 4166 return NULL; 4167 } 4168 bpf_net_ctx_clear(bpf_net_ctx); 4169 4170 return skb; 4171 } 4172 4173 static __always_inline struct sk_buff * 4174 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4175 { 4176 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4177 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4178 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4179 int sch_ret; 4180 4181 if (!entry) 4182 return skb; 4183 4184 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4185 4186 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4187 * already set by the caller. 4188 */ 4189 if (static_branch_unlikely(&tcx_needed_key)) { 4190 sch_ret = tcx_run(entry, skb, false); 4191 if (sch_ret != TC_ACT_UNSPEC) 4192 goto egress_verdict; 4193 } 4194 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4195 egress_verdict: 4196 switch (sch_ret) { 4197 case TC_ACT_REDIRECT: 4198 /* No need to push/pop skb's mac_header here on egress! */ 4199 skb_do_redirect(skb); 4200 *ret = NET_XMIT_SUCCESS; 4201 bpf_net_ctx_clear(bpf_net_ctx); 4202 return NULL; 4203 case TC_ACT_SHOT: 4204 kfree_skb_reason(skb, drop_reason); 4205 *ret = NET_XMIT_DROP; 4206 bpf_net_ctx_clear(bpf_net_ctx); 4207 return NULL; 4208 /* used by tc_run */ 4209 case TC_ACT_STOLEN: 4210 case TC_ACT_QUEUED: 4211 case TC_ACT_TRAP: 4212 consume_skb(skb); 4213 fallthrough; 4214 case TC_ACT_CONSUMED: 4215 *ret = NET_XMIT_SUCCESS; 4216 bpf_net_ctx_clear(bpf_net_ctx); 4217 return NULL; 4218 } 4219 bpf_net_ctx_clear(bpf_net_ctx); 4220 4221 return skb; 4222 } 4223 #else 4224 static __always_inline struct sk_buff * 4225 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4226 struct net_device *orig_dev, bool *another) 4227 { 4228 return skb; 4229 } 4230 4231 static __always_inline struct sk_buff * 4232 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4233 { 4234 return skb; 4235 } 4236 #endif /* CONFIG_NET_XGRESS */ 4237 4238 #ifdef CONFIG_XPS 4239 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4240 struct xps_dev_maps *dev_maps, unsigned int tci) 4241 { 4242 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4243 struct xps_map *map; 4244 int queue_index = -1; 4245 4246 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4247 return queue_index; 4248 4249 tci *= dev_maps->num_tc; 4250 tci += tc; 4251 4252 map = rcu_dereference(dev_maps->attr_map[tci]); 4253 if (map) { 4254 if (map->len == 1) 4255 queue_index = map->queues[0]; 4256 else 4257 queue_index = map->queues[reciprocal_scale( 4258 skb_get_hash(skb), map->len)]; 4259 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4260 queue_index = -1; 4261 } 4262 return queue_index; 4263 } 4264 #endif 4265 4266 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4267 struct sk_buff *skb) 4268 { 4269 #ifdef CONFIG_XPS 4270 struct xps_dev_maps *dev_maps; 4271 struct sock *sk = skb->sk; 4272 int queue_index = -1; 4273 4274 if (!static_key_false(&xps_needed)) 4275 return -1; 4276 4277 rcu_read_lock(); 4278 if (!static_key_false(&xps_rxqs_needed)) 4279 goto get_cpus_map; 4280 4281 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4282 if (dev_maps) { 4283 int tci = sk_rx_queue_get(sk); 4284 4285 if (tci >= 0) 4286 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4287 tci); 4288 } 4289 4290 get_cpus_map: 4291 if (queue_index < 0) { 4292 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4293 if (dev_maps) { 4294 unsigned int tci = skb->sender_cpu - 1; 4295 4296 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4297 tci); 4298 } 4299 } 4300 rcu_read_unlock(); 4301 4302 return queue_index; 4303 #else 4304 return -1; 4305 #endif 4306 } 4307 4308 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4309 struct net_device *sb_dev) 4310 { 4311 return 0; 4312 } 4313 EXPORT_SYMBOL(dev_pick_tx_zero); 4314 4315 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4316 struct net_device *sb_dev) 4317 { 4318 struct sock *sk = skb->sk; 4319 int queue_index = sk_tx_queue_get(sk); 4320 4321 sb_dev = sb_dev ? : dev; 4322 4323 if (queue_index < 0 || skb->ooo_okay || 4324 queue_index >= dev->real_num_tx_queues) { 4325 int new_index = get_xps_queue(dev, sb_dev, skb); 4326 4327 if (new_index < 0) 4328 new_index = skb_tx_hash(dev, sb_dev, skb); 4329 4330 if (queue_index != new_index && sk && 4331 sk_fullsock(sk) && 4332 rcu_access_pointer(sk->sk_dst_cache)) 4333 sk_tx_queue_set(sk, new_index); 4334 4335 queue_index = new_index; 4336 } 4337 4338 return queue_index; 4339 } 4340 EXPORT_SYMBOL(netdev_pick_tx); 4341 4342 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4343 struct sk_buff *skb, 4344 struct net_device *sb_dev) 4345 { 4346 int queue_index = 0; 4347 4348 #ifdef CONFIG_XPS 4349 u32 sender_cpu = skb->sender_cpu - 1; 4350 4351 if (sender_cpu >= (u32)NR_CPUS) 4352 skb->sender_cpu = raw_smp_processor_id() + 1; 4353 #endif 4354 4355 if (dev->real_num_tx_queues != 1) { 4356 const struct net_device_ops *ops = dev->netdev_ops; 4357 4358 if (ops->ndo_select_queue) 4359 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4360 else 4361 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4362 4363 queue_index = netdev_cap_txqueue(dev, queue_index); 4364 } 4365 4366 skb_set_queue_mapping(skb, queue_index); 4367 return netdev_get_tx_queue(dev, queue_index); 4368 } 4369 4370 /** 4371 * __dev_queue_xmit() - transmit a buffer 4372 * @skb: buffer to transmit 4373 * @sb_dev: suboordinate device used for L2 forwarding offload 4374 * 4375 * Queue a buffer for transmission to a network device. The caller must 4376 * have set the device and priority and built the buffer before calling 4377 * this function. The function can be called from an interrupt. 4378 * 4379 * When calling this method, interrupts MUST be enabled. This is because 4380 * the BH enable code must have IRQs enabled so that it will not deadlock. 4381 * 4382 * Regardless of the return value, the skb is consumed, so it is currently 4383 * difficult to retry a send to this method. (You can bump the ref count 4384 * before sending to hold a reference for retry if you are careful.) 4385 * 4386 * Return: 4387 * * 0 - buffer successfully transmitted 4388 * * positive qdisc return code - NET_XMIT_DROP etc. 4389 * * negative errno - other errors 4390 */ 4391 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4392 { 4393 struct net_device *dev = skb->dev; 4394 struct netdev_queue *txq = NULL; 4395 struct Qdisc *q; 4396 int rc = -ENOMEM; 4397 bool again = false; 4398 4399 skb_reset_mac_header(skb); 4400 skb_assert_len(skb); 4401 4402 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4403 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4404 4405 /* Disable soft irqs for various locks below. Also 4406 * stops preemption for RCU. 4407 */ 4408 rcu_read_lock_bh(); 4409 4410 skb_update_prio(skb); 4411 4412 qdisc_pkt_len_init(skb); 4413 tcx_set_ingress(skb, false); 4414 #ifdef CONFIG_NET_EGRESS 4415 if (static_branch_unlikely(&egress_needed_key)) { 4416 if (nf_hook_egress_active()) { 4417 skb = nf_hook_egress(skb, &rc, dev); 4418 if (!skb) 4419 goto out; 4420 } 4421 4422 netdev_xmit_skip_txqueue(false); 4423 4424 nf_skip_egress(skb, true); 4425 skb = sch_handle_egress(skb, &rc, dev); 4426 if (!skb) 4427 goto out; 4428 nf_skip_egress(skb, false); 4429 4430 if (netdev_xmit_txqueue_skipped()) 4431 txq = netdev_tx_queue_mapping(dev, skb); 4432 } 4433 #endif 4434 /* If device/qdisc don't need skb->dst, release it right now while 4435 * its hot in this cpu cache. 4436 */ 4437 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4438 skb_dst_drop(skb); 4439 else 4440 skb_dst_force(skb); 4441 4442 if (!txq) 4443 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4444 4445 q = rcu_dereference_bh(txq->qdisc); 4446 4447 trace_net_dev_queue(skb); 4448 if (q->enqueue) { 4449 rc = __dev_xmit_skb(skb, q, dev, txq); 4450 goto out; 4451 } 4452 4453 /* The device has no queue. Common case for software devices: 4454 * loopback, all the sorts of tunnels... 4455 4456 * Really, it is unlikely that netif_tx_lock protection is necessary 4457 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4458 * counters.) 4459 * However, it is possible, that they rely on protection 4460 * made by us here. 4461 4462 * Check this and shot the lock. It is not prone from deadlocks. 4463 *Either shot noqueue qdisc, it is even simpler 8) 4464 */ 4465 if (dev->flags & IFF_UP) { 4466 int cpu = smp_processor_id(); /* ok because BHs are off */ 4467 4468 /* Other cpus might concurrently change txq->xmit_lock_owner 4469 * to -1 or to their cpu id, but not to our id. 4470 */ 4471 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4472 if (dev_xmit_recursion()) 4473 goto recursion_alert; 4474 4475 skb = validate_xmit_skb(skb, dev, &again); 4476 if (!skb) 4477 goto out; 4478 4479 HARD_TX_LOCK(dev, txq, cpu); 4480 4481 if (!netif_xmit_stopped(txq)) { 4482 dev_xmit_recursion_inc(); 4483 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4484 dev_xmit_recursion_dec(); 4485 if (dev_xmit_complete(rc)) { 4486 HARD_TX_UNLOCK(dev, txq); 4487 goto out; 4488 } 4489 } 4490 HARD_TX_UNLOCK(dev, txq); 4491 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4492 dev->name); 4493 } else { 4494 /* Recursion is detected! It is possible, 4495 * unfortunately 4496 */ 4497 recursion_alert: 4498 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4499 dev->name); 4500 } 4501 } 4502 4503 rc = -ENETDOWN; 4504 rcu_read_unlock_bh(); 4505 4506 dev_core_stats_tx_dropped_inc(dev); 4507 kfree_skb_list(skb); 4508 return rc; 4509 out: 4510 rcu_read_unlock_bh(); 4511 return rc; 4512 } 4513 EXPORT_SYMBOL(__dev_queue_xmit); 4514 4515 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4516 { 4517 struct net_device *dev = skb->dev; 4518 struct sk_buff *orig_skb = skb; 4519 struct netdev_queue *txq; 4520 int ret = NETDEV_TX_BUSY; 4521 bool again = false; 4522 4523 if (unlikely(!netif_running(dev) || 4524 !netif_carrier_ok(dev))) 4525 goto drop; 4526 4527 skb = validate_xmit_skb_list(skb, dev, &again); 4528 if (skb != orig_skb) 4529 goto drop; 4530 4531 skb_set_queue_mapping(skb, queue_id); 4532 txq = skb_get_tx_queue(dev, skb); 4533 4534 local_bh_disable(); 4535 4536 dev_xmit_recursion_inc(); 4537 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4538 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4539 ret = netdev_start_xmit(skb, dev, txq, false); 4540 HARD_TX_UNLOCK(dev, txq); 4541 dev_xmit_recursion_dec(); 4542 4543 local_bh_enable(); 4544 return ret; 4545 drop: 4546 dev_core_stats_tx_dropped_inc(dev); 4547 kfree_skb_list(skb); 4548 return NET_XMIT_DROP; 4549 } 4550 EXPORT_SYMBOL(__dev_direct_xmit); 4551 4552 /************************************************************************* 4553 * Receiver routines 4554 *************************************************************************/ 4555 static DEFINE_PER_CPU(struct task_struct *, backlog_napi); 4556 4557 int weight_p __read_mostly = 64; /* old backlog weight */ 4558 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4559 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4560 4561 /* Called with irq disabled */ 4562 static inline void ____napi_schedule(struct softnet_data *sd, 4563 struct napi_struct *napi) 4564 { 4565 struct task_struct *thread; 4566 4567 lockdep_assert_irqs_disabled(); 4568 4569 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4570 /* Paired with smp_mb__before_atomic() in 4571 * napi_enable()/dev_set_threaded(). 4572 * Use READ_ONCE() to guarantee a complete 4573 * read on napi->thread. Only call 4574 * wake_up_process() when it's not NULL. 4575 */ 4576 thread = READ_ONCE(napi->thread); 4577 if (thread) { 4578 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi)) 4579 goto use_local_napi; 4580 4581 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4582 wake_up_process(thread); 4583 return; 4584 } 4585 } 4586 4587 use_local_napi: 4588 list_add_tail(&napi->poll_list, &sd->poll_list); 4589 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4590 /* If not called from net_rx_action() 4591 * we have to raise NET_RX_SOFTIRQ. 4592 */ 4593 if (!sd->in_net_rx_action) 4594 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4595 } 4596 4597 #ifdef CONFIG_RPS 4598 4599 struct static_key_false rps_needed __read_mostly; 4600 EXPORT_SYMBOL(rps_needed); 4601 struct static_key_false rfs_needed __read_mostly; 4602 EXPORT_SYMBOL(rfs_needed); 4603 4604 static struct rps_dev_flow * 4605 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4606 struct rps_dev_flow *rflow, u16 next_cpu) 4607 { 4608 if (next_cpu < nr_cpu_ids) { 4609 u32 head; 4610 #ifdef CONFIG_RFS_ACCEL 4611 struct netdev_rx_queue *rxqueue; 4612 struct rps_dev_flow_table *flow_table; 4613 struct rps_dev_flow *old_rflow; 4614 u16 rxq_index; 4615 u32 flow_id; 4616 int rc; 4617 4618 /* Should we steer this flow to a different hardware queue? */ 4619 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4620 !(dev->features & NETIF_F_NTUPLE)) 4621 goto out; 4622 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4623 if (rxq_index == skb_get_rx_queue(skb)) 4624 goto out; 4625 4626 rxqueue = dev->_rx + rxq_index; 4627 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4628 if (!flow_table) 4629 goto out; 4630 flow_id = skb_get_hash(skb) & flow_table->mask; 4631 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4632 rxq_index, flow_id); 4633 if (rc < 0) 4634 goto out; 4635 old_rflow = rflow; 4636 rflow = &flow_table->flows[flow_id]; 4637 WRITE_ONCE(rflow->filter, rc); 4638 if (old_rflow->filter == rc) 4639 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER); 4640 out: 4641 #endif 4642 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head); 4643 rps_input_queue_tail_save(&rflow->last_qtail, head); 4644 } 4645 4646 WRITE_ONCE(rflow->cpu, next_cpu); 4647 return rflow; 4648 } 4649 4650 /* 4651 * get_rps_cpu is called from netif_receive_skb and returns the target 4652 * CPU from the RPS map of the receiving queue for a given skb. 4653 * rcu_read_lock must be held on entry. 4654 */ 4655 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4656 struct rps_dev_flow **rflowp) 4657 { 4658 const struct rps_sock_flow_table *sock_flow_table; 4659 struct netdev_rx_queue *rxqueue = dev->_rx; 4660 struct rps_dev_flow_table *flow_table; 4661 struct rps_map *map; 4662 int cpu = -1; 4663 u32 tcpu; 4664 u32 hash; 4665 4666 if (skb_rx_queue_recorded(skb)) { 4667 u16 index = skb_get_rx_queue(skb); 4668 4669 if (unlikely(index >= dev->real_num_rx_queues)) { 4670 WARN_ONCE(dev->real_num_rx_queues > 1, 4671 "%s received packet on queue %u, but number " 4672 "of RX queues is %u\n", 4673 dev->name, index, dev->real_num_rx_queues); 4674 goto done; 4675 } 4676 rxqueue += index; 4677 } 4678 4679 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4680 4681 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4682 map = rcu_dereference(rxqueue->rps_map); 4683 if (!flow_table && !map) 4684 goto done; 4685 4686 skb_reset_network_header(skb); 4687 hash = skb_get_hash(skb); 4688 if (!hash) 4689 goto done; 4690 4691 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table); 4692 if (flow_table && sock_flow_table) { 4693 struct rps_dev_flow *rflow; 4694 u32 next_cpu; 4695 u32 ident; 4696 4697 /* First check into global flow table if there is a match. 4698 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4699 */ 4700 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4701 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask) 4702 goto try_rps; 4703 4704 next_cpu = ident & net_hotdata.rps_cpu_mask; 4705 4706 /* OK, now we know there is a match, 4707 * we can look at the local (per receive queue) flow table 4708 */ 4709 rflow = &flow_table->flows[hash & flow_table->mask]; 4710 tcpu = rflow->cpu; 4711 4712 /* 4713 * If the desired CPU (where last recvmsg was done) is 4714 * different from current CPU (one in the rx-queue flow 4715 * table entry), switch if one of the following holds: 4716 * - Current CPU is unset (>= nr_cpu_ids). 4717 * - Current CPU is offline. 4718 * - The current CPU's queue tail has advanced beyond the 4719 * last packet that was enqueued using this table entry. 4720 * This guarantees that all previous packets for the flow 4721 * have been dequeued, thus preserving in order delivery. 4722 */ 4723 if (unlikely(tcpu != next_cpu) && 4724 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4725 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) - 4726 rflow->last_qtail)) >= 0)) { 4727 tcpu = next_cpu; 4728 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4729 } 4730 4731 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4732 *rflowp = rflow; 4733 cpu = tcpu; 4734 goto done; 4735 } 4736 } 4737 4738 try_rps: 4739 4740 if (map) { 4741 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4742 if (cpu_online(tcpu)) { 4743 cpu = tcpu; 4744 goto done; 4745 } 4746 } 4747 4748 done: 4749 return cpu; 4750 } 4751 4752 #ifdef CONFIG_RFS_ACCEL 4753 4754 /** 4755 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4756 * @dev: Device on which the filter was set 4757 * @rxq_index: RX queue index 4758 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4759 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4760 * 4761 * Drivers that implement ndo_rx_flow_steer() should periodically call 4762 * this function for each installed filter and remove the filters for 4763 * which it returns %true. 4764 */ 4765 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4766 u32 flow_id, u16 filter_id) 4767 { 4768 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4769 struct rps_dev_flow_table *flow_table; 4770 struct rps_dev_flow *rflow; 4771 bool expire = true; 4772 unsigned int cpu; 4773 4774 rcu_read_lock(); 4775 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4776 if (flow_table && flow_id <= flow_table->mask) { 4777 rflow = &flow_table->flows[flow_id]; 4778 cpu = READ_ONCE(rflow->cpu); 4779 if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids && 4780 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) - 4781 READ_ONCE(rflow->last_qtail)) < 4782 (int)(10 * flow_table->mask))) 4783 expire = false; 4784 } 4785 rcu_read_unlock(); 4786 return expire; 4787 } 4788 EXPORT_SYMBOL(rps_may_expire_flow); 4789 4790 #endif /* CONFIG_RFS_ACCEL */ 4791 4792 /* Called from hardirq (IPI) context */ 4793 static void rps_trigger_softirq(void *data) 4794 { 4795 struct softnet_data *sd = data; 4796 4797 ____napi_schedule(sd, &sd->backlog); 4798 sd->received_rps++; 4799 } 4800 4801 #endif /* CONFIG_RPS */ 4802 4803 /* Called from hardirq (IPI) context */ 4804 static void trigger_rx_softirq(void *data) 4805 { 4806 struct softnet_data *sd = data; 4807 4808 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4809 smp_store_release(&sd->defer_ipi_scheduled, 0); 4810 } 4811 4812 /* 4813 * After we queued a packet into sd->input_pkt_queue, 4814 * we need to make sure this queue is serviced soon. 4815 * 4816 * - If this is another cpu queue, link it to our rps_ipi_list, 4817 * and make sure we will process rps_ipi_list from net_rx_action(). 4818 * 4819 * - If this is our own queue, NAPI schedule our backlog. 4820 * Note that this also raises NET_RX_SOFTIRQ. 4821 */ 4822 static void napi_schedule_rps(struct softnet_data *sd) 4823 { 4824 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4825 4826 #ifdef CONFIG_RPS 4827 if (sd != mysd) { 4828 if (use_backlog_threads()) { 4829 __napi_schedule_irqoff(&sd->backlog); 4830 return; 4831 } 4832 4833 sd->rps_ipi_next = mysd->rps_ipi_list; 4834 mysd->rps_ipi_list = sd; 4835 4836 /* If not called from net_rx_action() or napi_threaded_poll() 4837 * we have to raise NET_RX_SOFTIRQ. 4838 */ 4839 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4840 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4841 return; 4842 } 4843 #endif /* CONFIG_RPS */ 4844 __napi_schedule_irqoff(&mysd->backlog); 4845 } 4846 4847 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu) 4848 { 4849 unsigned long flags; 4850 4851 if (use_backlog_threads()) { 4852 backlog_lock_irq_save(sd, &flags); 4853 4854 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4855 __napi_schedule_irqoff(&sd->backlog); 4856 4857 backlog_unlock_irq_restore(sd, &flags); 4858 4859 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) { 4860 smp_call_function_single_async(cpu, &sd->defer_csd); 4861 } 4862 } 4863 4864 #ifdef CONFIG_NET_FLOW_LIMIT 4865 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4866 #endif 4867 4868 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4869 { 4870 #ifdef CONFIG_NET_FLOW_LIMIT 4871 struct sd_flow_limit *fl; 4872 struct softnet_data *sd; 4873 unsigned int old_flow, new_flow; 4874 4875 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1)) 4876 return false; 4877 4878 sd = this_cpu_ptr(&softnet_data); 4879 4880 rcu_read_lock(); 4881 fl = rcu_dereference(sd->flow_limit); 4882 if (fl) { 4883 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4884 old_flow = fl->history[fl->history_head]; 4885 fl->history[fl->history_head] = new_flow; 4886 4887 fl->history_head++; 4888 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4889 4890 if (likely(fl->buckets[old_flow])) 4891 fl->buckets[old_flow]--; 4892 4893 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4894 fl->count++; 4895 rcu_read_unlock(); 4896 return true; 4897 } 4898 } 4899 rcu_read_unlock(); 4900 #endif 4901 return false; 4902 } 4903 4904 /* 4905 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4906 * queue (may be a remote CPU queue). 4907 */ 4908 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4909 unsigned int *qtail) 4910 { 4911 enum skb_drop_reason reason; 4912 struct softnet_data *sd; 4913 unsigned long flags; 4914 unsigned int qlen; 4915 int max_backlog; 4916 u32 tail; 4917 4918 reason = SKB_DROP_REASON_DEV_READY; 4919 if (!netif_running(skb->dev)) 4920 goto bad_dev; 4921 4922 reason = SKB_DROP_REASON_CPU_BACKLOG; 4923 sd = &per_cpu(softnet_data, cpu); 4924 4925 qlen = skb_queue_len_lockless(&sd->input_pkt_queue); 4926 max_backlog = READ_ONCE(net_hotdata.max_backlog); 4927 if (unlikely(qlen > max_backlog)) 4928 goto cpu_backlog_drop; 4929 backlog_lock_irq_save(sd, &flags); 4930 qlen = skb_queue_len(&sd->input_pkt_queue); 4931 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) { 4932 if (!qlen) { 4933 /* Schedule NAPI for backlog device. We can use 4934 * non atomic operation as we own the queue lock. 4935 */ 4936 if (!__test_and_set_bit(NAPI_STATE_SCHED, 4937 &sd->backlog.state)) 4938 napi_schedule_rps(sd); 4939 } 4940 __skb_queue_tail(&sd->input_pkt_queue, skb); 4941 tail = rps_input_queue_tail_incr(sd); 4942 backlog_unlock_irq_restore(sd, &flags); 4943 4944 /* save the tail outside of the critical section */ 4945 rps_input_queue_tail_save(qtail, tail); 4946 return NET_RX_SUCCESS; 4947 } 4948 4949 backlog_unlock_irq_restore(sd, &flags); 4950 4951 cpu_backlog_drop: 4952 atomic_inc(&sd->dropped); 4953 bad_dev: 4954 dev_core_stats_rx_dropped_inc(skb->dev); 4955 kfree_skb_reason(skb, reason); 4956 return NET_RX_DROP; 4957 } 4958 4959 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4960 { 4961 struct net_device *dev = skb->dev; 4962 struct netdev_rx_queue *rxqueue; 4963 4964 rxqueue = dev->_rx; 4965 4966 if (skb_rx_queue_recorded(skb)) { 4967 u16 index = skb_get_rx_queue(skb); 4968 4969 if (unlikely(index >= dev->real_num_rx_queues)) { 4970 WARN_ONCE(dev->real_num_rx_queues > 1, 4971 "%s received packet on queue %u, but number " 4972 "of RX queues is %u\n", 4973 dev->name, index, dev->real_num_rx_queues); 4974 4975 return rxqueue; /* Return first rxqueue */ 4976 } 4977 rxqueue += index; 4978 } 4979 return rxqueue; 4980 } 4981 4982 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4983 const struct bpf_prog *xdp_prog) 4984 { 4985 void *orig_data, *orig_data_end, *hard_start; 4986 struct netdev_rx_queue *rxqueue; 4987 bool orig_bcast, orig_host; 4988 u32 mac_len, frame_sz; 4989 __be16 orig_eth_type; 4990 struct ethhdr *eth; 4991 u32 metalen, act; 4992 int off; 4993 4994 /* The XDP program wants to see the packet starting at the MAC 4995 * header. 4996 */ 4997 mac_len = skb->data - skb_mac_header(skb); 4998 hard_start = skb->data - skb_headroom(skb); 4999 5000 /* SKB "head" area always have tailroom for skb_shared_info */ 5001 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 5002 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5003 5004 rxqueue = netif_get_rxqueue(skb); 5005 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 5006 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 5007 skb_headlen(skb) + mac_len, true); 5008 if (skb_is_nonlinear(skb)) { 5009 skb_shinfo(skb)->xdp_frags_size = skb->data_len; 5010 xdp_buff_set_frags_flag(xdp); 5011 } else { 5012 xdp_buff_clear_frags_flag(xdp); 5013 } 5014 5015 orig_data_end = xdp->data_end; 5016 orig_data = xdp->data; 5017 eth = (struct ethhdr *)xdp->data; 5018 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 5019 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 5020 orig_eth_type = eth->h_proto; 5021 5022 act = bpf_prog_run_xdp(xdp_prog, xdp); 5023 5024 /* check if bpf_xdp_adjust_head was used */ 5025 off = xdp->data - orig_data; 5026 if (off) { 5027 if (off > 0) 5028 __skb_pull(skb, off); 5029 else if (off < 0) 5030 __skb_push(skb, -off); 5031 5032 skb->mac_header += off; 5033 skb_reset_network_header(skb); 5034 } 5035 5036 /* check if bpf_xdp_adjust_tail was used */ 5037 off = xdp->data_end - orig_data_end; 5038 if (off != 0) { 5039 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 5040 skb->len += off; /* positive on grow, negative on shrink */ 5041 } 5042 5043 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers 5044 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. 5045 */ 5046 if (xdp_buff_has_frags(xdp)) 5047 skb->data_len = skb_shinfo(skb)->xdp_frags_size; 5048 else 5049 skb->data_len = 0; 5050 5051 /* check if XDP changed eth hdr such SKB needs update */ 5052 eth = (struct ethhdr *)xdp->data; 5053 if ((orig_eth_type != eth->h_proto) || 5054 (orig_host != ether_addr_equal_64bits(eth->h_dest, 5055 skb->dev->dev_addr)) || 5056 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 5057 __skb_push(skb, ETH_HLEN); 5058 skb->pkt_type = PACKET_HOST; 5059 skb->protocol = eth_type_trans(skb, skb->dev); 5060 } 5061 5062 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 5063 * before calling us again on redirect path. We do not call do_redirect 5064 * as we leave that up to the caller. 5065 * 5066 * Caller is responsible for managing lifetime of skb (i.e. calling 5067 * kfree_skb in response to actions it cannot handle/XDP_DROP). 5068 */ 5069 switch (act) { 5070 case XDP_REDIRECT: 5071 case XDP_TX: 5072 __skb_push(skb, mac_len); 5073 break; 5074 case XDP_PASS: 5075 metalen = xdp->data - xdp->data_meta; 5076 if (metalen) 5077 skb_metadata_set(skb, metalen); 5078 break; 5079 } 5080 5081 return act; 5082 } 5083 5084 static int 5085 netif_skb_check_for_xdp(struct sk_buff **pskb, const struct bpf_prog *prog) 5086 { 5087 struct sk_buff *skb = *pskb; 5088 int err, hroom, troom; 5089 5090 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog)) 5091 return 0; 5092 5093 /* In case we have to go down the path and also linearize, 5094 * then lets do the pskb_expand_head() work just once here. 5095 */ 5096 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 5097 troom = skb->tail + skb->data_len - skb->end; 5098 err = pskb_expand_head(skb, 5099 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 5100 troom > 0 ? troom + 128 : 0, GFP_ATOMIC); 5101 if (err) 5102 return err; 5103 5104 return skb_linearize(skb); 5105 } 5106 5107 static u32 netif_receive_generic_xdp(struct sk_buff **pskb, 5108 struct xdp_buff *xdp, 5109 const struct bpf_prog *xdp_prog) 5110 { 5111 struct sk_buff *skb = *pskb; 5112 u32 mac_len, act = XDP_DROP; 5113 5114 /* Reinjected packets coming from act_mirred or similar should 5115 * not get XDP generic processing. 5116 */ 5117 if (skb_is_redirected(skb)) 5118 return XDP_PASS; 5119 5120 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM 5121 * bytes. This is the guarantee that also native XDP provides, 5122 * thus we need to do it here as well. 5123 */ 5124 mac_len = skb->data - skb_mac_header(skb); 5125 __skb_push(skb, mac_len); 5126 5127 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 5128 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 5129 if (netif_skb_check_for_xdp(pskb, xdp_prog)) 5130 goto do_drop; 5131 } 5132 5133 __skb_pull(*pskb, mac_len); 5134 5135 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); 5136 switch (act) { 5137 case XDP_REDIRECT: 5138 case XDP_TX: 5139 case XDP_PASS: 5140 break; 5141 default: 5142 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); 5143 fallthrough; 5144 case XDP_ABORTED: 5145 trace_xdp_exception((*pskb)->dev, xdp_prog, act); 5146 fallthrough; 5147 case XDP_DROP: 5148 do_drop: 5149 kfree_skb(*pskb); 5150 break; 5151 } 5152 5153 return act; 5154 } 5155 5156 /* When doing generic XDP we have to bypass the qdisc layer and the 5157 * network taps in order to match in-driver-XDP behavior. This also means 5158 * that XDP packets are able to starve other packets going through a qdisc, 5159 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5160 * queues, so they do not have this starvation issue. 5161 */ 5162 void generic_xdp_tx(struct sk_buff *skb, const struct bpf_prog *xdp_prog) 5163 { 5164 struct net_device *dev = skb->dev; 5165 struct netdev_queue *txq; 5166 bool free_skb = true; 5167 int cpu, rc; 5168 5169 txq = netdev_core_pick_tx(dev, skb, NULL); 5170 cpu = smp_processor_id(); 5171 HARD_TX_LOCK(dev, txq, cpu); 5172 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5173 rc = netdev_start_xmit(skb, dev, txq, 0); 5174 if (dev_xmit_complete(rc)) 5175 free_skb = false; 5176 } 5177 HARD_TX_UNLOCK(dev, txq); 5178 if (free_skb) { 5179 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5180 dev_core_stats_tx_dropped_inc(dev); 5181 kfree_skb(skb); 5182 } 5183 } 5184 5185 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5186 5187 int do_xdp_generic(const struct bpf_prog *xdp_prog, struct sk_buff **pskb) 5188 { 5189 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 5190 5191 if (xdp_prog) { 5192 struct xdp_buff xdp; 5193 u32 act; 5194 int err; 5195 5196 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 5197 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); 5198 if (act != XDP_PASS) { 5199 switch (act) { 5200 case XDP_REDIRECT: 5201 err = xdp_do_generic_redirect((*pskb)->dev, *pskb, 5202 &xdp, xdp_prog); 5203 if (err) 5204 goto out_redir; 5205 break; 5206 case XDP_TX: 5207 generic_xdp_tx(*pskb, xdp_prog); 5208 break; 5209 } 5210 bpf_net_ctx_clear(bpf_net_ctx); 5211 return XDP_DROP; 5212 } 5213 bpf_net_ctx_clear(bpf_net_ctx); 5214 } 5215 return XDP_PASS; 5216 out_redir: 5217 bpf_net_ctx_clear(bpf_net_ctx); 5218 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); 5219 return XDP_DROP; 5220 } 5221 EXPORT_SYMBOL_GPL(do_xdp_generic); 5222 5223 static int netif_rx_internal(struct sk_buff *skb) 5224 { 5225 int ret; 5226 5227 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5228 5229 trace_netif_rx(skb); 5230 5231 #ifdef CONFIG_RPS 5232 if (static_branch_unlikely(&rps_needed)) { 5233 struct rps_dev_flow voidflow, *rflow = &voidflow; 5234 int cpu; 5235 5236 rcu_read_lock(); 5237 5238 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5239 if (cpu < 0) 5240 cpu = smp_processor_id(); 5241 5242 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5243 5244 rcu_read_unlock(); 5245 } else 5246 #endif 5247 { 5248 unsigned int qtail; 5249 5250 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5251 } 5252 return ret; 5253 } 5254 5255 /** 5256 * __netif_rx - Slightly optimized version of netif_rx 5257 * @skb: buffer to post 5258 * 5259 * This behaves as netif_rx except that it does not disable bottom halves. 5260 * As a result this function may only be invoked from the interrupt context 5261 * (either hard or soft interrupt). 5262 */ 5263 int __netif_rx(struct sk_buff *skb) 5264 { 5265 int ret; 5266 5267 lockdep_assert_once(hardirq_count() | softirq_count()); 5268 5269 trace_netif_rx_entry(skb); 5270 ret = netif_rx_internal(skb); 5271 trace_netif_rx_exit(ret); 5272 return ret; 5273 } 5274 EXPORT_SYMBOL(__netif_rx); 5275 5276 /** 5277 * netif_rx - post buffer to the network code 5278 * @skb: buffer to post 5279 * 5280 * This function receives a packet from a device driver and queues it for 5281 * the upper (protocol) levels to process via the backlog NAPI device. It 5282 * always succeeds. The buffer may be dropped during processing for 5283 * congestion control or by the protocol layers. 5284 * The network buffer is passed via the backlog NAPI device. Modern NIC 5285 * driver should use NAPI and GRO. 5286 * This function can used from interrupt and from process context. The 5287 * caller from process context must not disable interrupts before invoking 5288 * this function. 5289 * 5290 * return values: 5291 * NET_RX_SUCCESS (no congestion) 5292 * NET_RX_DROP (packet was dropped) 5293 * 5294 */ 5295 int netif_rx(struct sk_buff *skb) 5296 { 5297 bool need_bh_off = !(hardirq_count() | softirq_count()); 5298 int ret; 5299 5300 if (need_bh_off) 5301 local_bh_disable(); 5302 trace_netif_rx_entry(skb); 5303 ret = netif_rx_internal(skb); 5304 trace_netif_rx_exit(ret); 5305 if (need_bh_off) 5306 local_bh_enable(); 5307 return ret; 5308 } 5309 EXPORT_SYMBOL(netif_rx); 5310 5311 static __latent_entropy void net_tx_action(void) 5312 { 5313 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5314 5315 if (sd->completion_queue) { 5316 struct sk_buff *clist; 5317 5318 local_irq_disable(); 5319 clist = sd->completion_queue; 5320 sd->completion_queue = NULL; 5321 local_irq_enable(); 5322 5323 while (clist) { 5324 struct sk_buff *skb = clist; 5325 5326 clist = clist->next; 5327 5328 WARN_ON(refcount_read(&skb->users)); 5329 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5330 trace_consume_skb(skb, net_tx_action); 5331 else 5332 trace_kfree_skb(skb, net_tx_action, 5333 get_kfree_skb_cb(skb)->reason, NULL); 5334 5335 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5336 __kfree_skb(skb); 5337 else 5338 __napi_kfree_skb(skb, 5339 get_kfree_skb_cb(skb)->reason); 5340 } 5341 } 5342 5343 if (sd->output_queue) { 5344 struct Qdisc *head; 5345 5346 local_irq_disable(); 5347 head = sd->output_queue; 5348 sd->output_queue = NULL; 5349 sd->output_queue_tailp = &sd->output_queue; 5350 local_irq_enable(); 5351 5352 rcu_read_lock(); 5353 5354 while (head) { 5355 struct Qdisc *q = head; 5356 spinlock_t *root_lock = NULL; 5357 5358 head = head->next_sched; 5359 5360 /* We need to make sure head->next_sched is read 5361 * before clearing __QDISC_STATE_SCHED 5362 */ 5363 smp_mb__before_atomic(); 5364 5365 if (!(q->flags & TCQ_F_NOLOCK)) { 5366 root_lock = qdisc_lock(q); 5367 spin_lock(root_lock); 5368 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5369 &q->state))) { 5370 /* There is a synchronize_net() between 5371 * STATE_DEACTIVATED flag being set and 5372 * qdisc_reset()/some_qdisc_is_busy() in 5373 * dev_deactivate(), so we can safely bail out 5374 * early here to avoid data race between 5375 * qdisc_deactivate() and some_qdisc_is_busy() 5376 * for lockless qdisc. 5377 */ 5378 clear_bit(__QDISC_STATE_SCHED, &q->state); 5379 continue; 5380 } 5381 5382 clear_bit(__QDISC_STATE_SCHED, &q->state); 5383 qdisc_run(q); 5384 if (root_lock) 5385 spin_unlock(root_lock); 5386 } 5387 5388 rcu_read_unlock(); 5389 } 5390 5391 xfrm_dev_backlog(sd); 5392 } 5393 5394 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5395 /* This hook is defined here for ATM LANE */ 5396 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5397 unsigned char *addr) __read_mostly; 5398 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5399 #endif 5400 5401 /** 5402 * netdev_is_rx_handler_busy - check if receive handler is registered 5403 * @dev: device to check 5404 * 5405 * Check if a receive handler is already registered for a given device. 5406 * Return true if there one. 5407 * 5408 * The caller must hold the rtnl_mutex. 5409 */ 5410 bool netdev_is_rx_handler_busy(struct net_device *dev) 5411 { 5412 ASSERT_RTNL(); 5413 return dev && rtnl_dereference(dev->rx_handler); 5414 } 5415 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5416 5417 /** 5418 * netdev_rx_handler_register - register receive handler 5419 * @dev: device to register a handler for 5420 * @rx_handler: receive handler to register 5421 * @rx_handler_data: data pointer that is used by rx handler 5422 * 5423 * Register a receive handler for a device. This handler will then be 5424 * called from __netif_receive_skb. A negative errno code is returned 5425 * on a failure. 5426 * 5427 * The caller must hold the rtnl_mutex. 5428 * 5429 * For a general description of rx_handler, see enum rx_handler_result. 5430 */ 5431 int netdev_rx_handler_register(struct net_device *dev, 5432 rx_handler_func_t *rx_handler, 5433 void *rx_handler_data) 5434 { 5435 if (netdev_is_rx_handler_busy(dev)) 5436 return -EBUSY; 5437 5438 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5439 return -EINVAL; 5440 5441 /* Note: rx_handler_data must be set before rx_handler */ 5442 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5443 rcu_assign_pointer(dev->rx_handler, rx_handler); 5444 5445 return 0; 5446 } 5447 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5448 5449 /** 5450 * netdev_rx_handler_unregister - unregister receive handler 5451 * @dev: device to unregister a handler from 5452 * 5453 * Unregister a receive handler from a device. 5454 * 5455 * The caller must hold the rtnl_mutex. 5456 */ 5457 void netdev_rx_handler_unregister(struct net_device *dev) 5458 { 5459 5460 ASSERT_RTNL(); 5461 RCU_INIT_POINTER(dev->rx_handler, NULL); 5462 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5463 * section has a guarantee to see a non NULL rx_handler_data 5464 * as well. 5465 */ 5466 synchronize_net(); 5467 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5468 } 5469 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5470 5471 /* 5472 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5473 * the special handling of PFMEMALLOC skbs. 5474 */ 5475 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5476 { 5477 switch (skb->protocol) { 5478 case htons(ETH_P_ARP): 5479 case htons(ETH_P_IP): 5480 case htons(ETH_P_IPV6): 5481 case htons(ETH_P_8021Q): 5482 case htons(ETH_P_8021AD): 5483 return true; 5484 default: 5485 return false; 5486 } 5487 } 5488 5489 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5490 int *ret, struct net_device *orig_dev) 5491 { 5492 if (nf_hook_ingress_active(skb)) { 5493 int ingress_retval; 5494 5495 if (*pt_prev) { 5496 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5497 *pt_prev = NULL; 5498 } 5499 5500 rcu_read_lock(); 5501 ingress_retval = nf_hook_ingress(skb); 5502 rcu_read_unlock(); 5503 return ingress_retval; 5504 } 5505 return 0; 5506 } 5507 5508 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5509 struct packet_type **ppt_prev) 5510 { 5511 struct packet_type *ptype, *pt_prev; 5512 rx_handler_func_t *rx_handler; 5513 struct sk_buff *skb = *pskb; 5514 struct net_device *orig_dev; 5515 bool deliver_exact = false; 5516 int ret = NET_RX_DROP; 5517 __be16 type; 5518 5519 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5520 5521 trace_netif_receive_skb(skb); 5522 5523 orig_dev = skb->dev; 5524 5525 skb_reset_network_header(skb); 5526 #if !defined(CONFIG_DEBUG_NET) 5527 /* We plan to no longer reset the transport header here. 5528 * Give some time to fuzzers and dev build to catch bugs 5529 * in network stacks. 5530 */ 5531 if (!skb_transport_header_was_set(skb)) 5532 skb_reset_transport_header(skb); 5533 #endif 5534 skb_reset_mac_len(skb); 5535 5536 pt_prev = NULL; 5537 5538 another_round: 5539 skb->skb_iif = skb->dev->ifindex; 5540 5541 __this_cpu_inc(softnet_data.processed); 5542 5543 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5544 int ret2; 5545 5546 migrate_disable(); 5547 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), 5548 &skb); 5549 migrate_enable(); 5550 5551 if (ret2 != XDP_PASS) { 5552 ret = NET_RX_DROP; 5553 goto out; 5554 } 5555 } 5556 5557 if (eth_type_vlan(skb->protocol)) { 5558 skb = skb_vlan_untag(skb); 5559 if (unlikely(!skb)) 5560 goto out; 5561 } 5562 5563 if (skb_skip_tc_classify(skb)) 5564 goto skip_classify; 5565 5566 if (pfmemalloc) 5567 goto skip_taps; 5568 5569 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) { 5570 if (pt_prev) 5571 ret = deliver_skb(skb, pt_prev, orig_dev); 5572 pt_prev = ptype; 5573 } 5574 5575 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5576 if (pt_prev) 5577 ret = deliver_skb(skb, pt_prev, orig_dev); 5578 pt_prev = ptype; 5579 } 5580 5581 skip_taps: 5582 #ifdef CONFIG_NET_INGRESS 5583 if (static_branch_unlikely(&ingress_needed_key)) { 5584 bool another = false; 5585 5586 nf_skip_egress(skb, true); 5587 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5588 &another); 5589 if (another) 5590 goto another_round; 5591 if (!skb) 5592 goto out; 5593 5594 nf_skip_egress(skb, false); 5595 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5596 goto out; 5597 } 5598 #endif 5599 skb_reset_redirect(skb); 5600 skip_classify: 5601 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5602 goto drop; 5603 5604 if (skb_vlan_tag_present(skb)) { 5605 if (pt_prev) { 5606 ret = deliver_skb(skb, pt_prev, orig_dev); 5607 pt_prev = NULL; 5608 } 5609 if (vlan_do_receive(&skb)) 5610 goto another_round; 5611 else if (unlikely(!skb)) 5612 goto out; 5613 } 5614 5615 rx_handler = rcu_dereference(skb->dev->rx_handler); 5616 if (rx_handler) { 5617 if (pt_prev) { 5618 ret = deliver_skb(skb, pt_prev, orig_dev); 5619 pt_prev = NULL; 5620 } 5621 switch (rx_handler(&skb)) { 5622 case RX_HANDLER_CONSUMED: 5623 ret = NET_RX_SUCCESS; 5624 goto out; 5625 case RX_HANDLER_ANOTHER: 5626 goto another_round; 5627 case RX_HANDLER_EXACT: 5628 deliver_exact = true; 5629 break; 5630 case RX_HANDLER_PASS: 5631 break; 5632 default: 5633 BUG(); 5634 } 5635 } 5636 5637 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5638 check_vlan_id: 5639 if (skb_vlan_tag_get_id(skb)) { 5640 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5641 * find vlan device. 5642 */ 5643 skb->pkt_type = PACKET_OTHERHOST; 5644 } else if (eth_type_vlan(skb->protocol)) { 5645 /* Outer header is 802.1P with vlan 0, inner header is 5646 * 802.1Q or 802.1AD and vlan_do_receive() above could 5647 * not find vlan dev for vlan id 0. 5648 */ 5649 __vlan_hwaccel_clear_tag(skb); 5650 skb = skb_vlan_untag(skb); 5651 if (unlikely(!skb)) 5652 goto out; 5653 if (vlan_do_receive(&skb)) 5654 /* After stripping off 802.1P header with vlan 0 5655 * vlan dev is found for inner header. 5656 */ 5657 goto another_round; 5658 else if (unlikely(!skb)) 5659 goto out; 5660 else 5661 /* We have stripped outer 802.1P vlan 0 header. 5662 * But could not find vlan dev. 5663 * check again for vlan id to set OTHERHOST. 5664 */ 5665 goto check_vlan_id; 5666 } 5667 /* Note: we might in the future use prio bits 5668 * and set skb->priority like in vlan_do_receive() 5669 * For the time being, just ignore Priority Code Point 5670 */ 5671 __vlan_hwaccel_clear_tag(skb); 5672 } 5673 5674 type = skb->protocol; 5675 5676 /* deliver only exact match when indicated */ 5677 if (likely(!deliver_exact)) { 5678 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5679 &ptype_base[ntohs(type) & 5680 PTYPE_HASH_MASK]); 5681 } 5682 5683 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5684 &orig_dev->ptype_specific); 5685 5686 if (unlikely(skb->dev != orig_dev)) { 5687 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5688 &skb->dev->ptype_specific); 5689 } 5690 5691 if (pt_prev) { 5692 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5693 goto drop; 5694 *ppt_prev = pt_prev; 5695 } else { 5696 drop: 5697 if (!deliver_exact) 5698 dev_core_stats_rx_dropped_inc(skb->dev); 5699 else 5700 dev_core_stats_rx_nohandler_inc(skb->dev); 5701 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5702 /* Jamal, now you will not able to escape explaining 5703 * me how you were going to use this. :-) 5704 */ 5705 ret = NET_RX_DROP; 5706 } 5707 5708 out: 5709 /* The invariant here is that if *ppt_prev is not NULL 5710 * then skb should also be non-NULL. 5711 * 5712 * Apparently *ppt_prev assignment above holds this invariant due to 5713 * skb dereferencing near it. 5714 */ 5715 *pskb = skb; 5716 return ret; 5717 } 5718 5719 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5720 { 5721 struct net_device *orig_dev = skb->dev; 5722 struct packet_type *pt_prev = NULL; 5723 int ret; 5724 5725 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5726 if (pt_prev) 5727 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5728 skb->dev, pt_prev, orig_dev); 5729 return ret; 5730 } 5731 5732 /** 5733 * netif_receive_skb_core - special purpose version of netif_receive_skb 5734 * @skb: buffer to process 5735 * 5736 * More direct receive version of netif_receive_skb(). It should 5737 * only be used by callers that have a need to skip RPS and Generic XDP. 5738 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5739 * 5740 * This function may only be called from softirq context and interrupts 5741 * should be enabled. 5742 * 5743 * Return values (usually ignored): 5744 * NET_RX_SUCCESS: no congestion 5745 * NET_RX_DROP: packet was dropped 5746 */ 5747 int netif_receive_skb_core(struct sk_buff *skb) 5748 { 5749 int ret; 5750 5751 rcu_read_lock(); 5752 ret = __netif_receive_skb_one_core(skb, false); 5753 rcu_read_unlock(); 5754 5755 return ret; 5756 } 5757 EXPORT_SYMBOL(netif_receive_skb_core); 5758 5759 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5760 struct packet_type *pt_prev, 5761 struct net_device *orig_dev) 5762 { 5763 struct sk_buff *skb, *next; 5764 5765 if (!pt_prev) 5766 return; 5767 if (list_empty(head)) 5768 return; 5769 if (pt_prev->list_func != NULL) 5770 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5771 ip_list_rcv, head, pt_prev, orig_dev); 5772 else 5773 list_for_each_entry_safe(skb, next, head, list) { 5774 skb_list_del_init(skb); 5775 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5776 } 5777 } 5778 5779 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5780 { 5781 /* Fast-path assumptions: 5782 * - There is no RX handler. 5783 * - Only one packet_type matches. 5784 * If either of these fails, we will end up doing some per-packet 5785 * processing in-line, then handling the 'last ptype' for the whole 5786 * sublist. This can't cause out-of-order delivery to any single ptype, 5787 * because the 'last ptype' must be constant across the sublist, and all 5788 * other ptypes are handled per-packet. 5789 */ 5790 /* Current (common) ptype of sublist */ 5791 struct packet_type *pt_curr = NULL; 5792 /* Current (common) orig_dev of sublist */ 5793 struct net_device *od_curr = NULL; 5794 struct sk_buff *skb, *next; 5795 LIST_HEAD(sublist); 5796 5797 list_for_each_entry_safe(skb, next, head, list) { 5798 struct net_device *orig_dev = skb->dev; 5799 struct packet_type *pt_prev = NULL; 5800 5801 skb_list_del_init(skb); 5802 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5803 if (!pt_prev) 5804 continue; 5805 if (pt_curr != pt_prev || od_curr != orig_dev) { 5806 /* dispatch old sublist */ 5807 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5808 /* start new sublist */ 5809 INIT_LIST_HEAD(&sublist); 5810 pt_curr = pt_prev; 5811 od_curr = orig_dev; 5812 } 5813 list_add_tail(&skb->list, &sublist); 5814 } 5815 5816 /* dispatch final sublist */ 5817 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5818 } 5819 5820 static int __netif_receive_skb(struct sk_buff *skb) 5821 { 5822 int ret; 5823 5824 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5825 unsigned int noreclaim_flag; 5826 5827 /* 5828 * PFMEMALLOC skbs are special, they should 5829 * - be delivered to SOCK_MEMALLOC sockets only 5830 * - stay away from userspace 5831 * - have bounded memory usage 5832 * 5833 * Use PF_MEMALLOC as this saves us from propagating the allocation 5834 * context down to all allocation sites. 5835 */ 5836 noreclaim_flag = memalloc_noreclaim_save(); 5837 ret = __netif_receive_skb_one_core(skb, true); 5838 memalloc_noreclaim_restore(noreclaim_flag); 5839 } else 5840 ret = __netif_receive_skb_one_core(skb, false); 5841 5842 return ret; 5843 } 5844 5845 static void __netif_receive_skb_list(struct list_head *head) 5846 { 5847 unsigned long noreclaim_flag = 0; 5848 struct sk_buff *skb, *next; 5849 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5850 5851 list_for_each_entry_safe(skb, next, head, list) { 5852 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5853 struct list_head sublist; 5854 5855 /* Handle the previous sublist */ 5856 list_cut_before(&sublist, head, &skb->list); 5857 if (!list_empty(&sublist)) 5858 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5859 pfmemalloc = !pfmemalloc; 5860 /* See comments in __netif_receive_skb */ 5861 if (pfmemalloc) 5862 noreclaim_flag = memalloc_noreclaim_save(); 5863 else 5864 memalloc_noreclaim_restore(noreclaim_flag); 5865 } 5866 } 5867 /* Handle the remaining sublist */ 5868 if (!list_empty(head)) 5869 __netif_receive_skb_list_core(head, pfmemalloc); 5870 /* Restore pflags */ 5871 if (pfmemalloc) 5872 memalloc_noreclaim_restore(noreclaim_flag); 5873 } 5874 5875 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5876 { 5877 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5878 struct bpf_prog *new = xdp->prog; 5879 int ret = 0; 5880 5881 switch (xdp->command) { 5882 case XDP_SETUP_PROG: 5883 rcu_assign_pointer(dev->xdp_prog, new); 5884 if (old) 5885 bpf_prog_put(old); 5886 5887 if (old && !new) { 5888 static_branch_dec(&generic_xdp_needed_key); 5889 } else if (new && !old) { 5890 static_branch_inc(&generic_xdp_needed_key); 5891 dev_disable_lro(dev); 5892 dev_disable_gro_hw(dev); 5893 } 5894 break; 5895 5896 default: 5897 ret = -EINVAL; 5898 break; 5899 } 5900 5901 return ret; 5902 } 5903 5904 static int netif_receive_skb_internal(struct sk_buff *skb) 5905 { 5906 int ret; 5907 5908 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5909 5910 if (skb_defer_rx_timestamp(skb)) 5911 return NET_RX_SUCCESS; 5912 5913 rcu_read_lock(); 5914 #ifdef CONFIG_RPS 5915 if (static_branch_unlikely(&rps_needed)) { 5916 struct rps_dev_flow voidflow, *rflow = &voidflow; 5917 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5918 5919 if (cpu >= 0) { 5920 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5921 rcu_read_unlock(); 5922 return ret; 5923 } 5924 } 5925 #endif 5926 ret = __netif_receive_skb(skb); 5927 rcu_read_unlock(); 5928 return ret; 5929 } 5930 5931 void netif_receive_skb_list_internal(struct list_head *head) 5932 { 5933 struct sk_buff *skb, *next; 5934 LIST_HEAD(sublist); 5935 5936 list_for_each_entry_safe(skb, next, head, list) { 5937 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), 5938 skb); 5939 skb_list_del_init(skb); 5940 if (!skb_defer_rx_timestamp(skb)) 5941 list_add_tail(&skb->list, &sublist); 5942 } 5943 list_splice_init(&sublist, head); 5944 5945 rcu_read_lock(); 5946 #ifdef CONFIG_RPS 5947 if (static_branch_unlikely(&rps_needed)) { 5948 list_for_each_entry_safe(skb, next, head, list) { 5949 struct rps_dev_flow voidflow, *rflow = &voidflow; 5950 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5951 5952 if (cpu >= 0) { 5953 /* Will be handled, remove from list */ 5954 skb_list_del_init(skb); 5955 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5956 } 5957 } 5958 } 5959 #endif 5960 __netif_receive_skb_list(head); 5961 rcu_read_unlock(); 5962 } 5963 5964 /** 5965 * netif_receive_skb - process receive buffer from network 5966 * @skb: buffer to process 5967 * 5968 * netif_receive_skb() is the main receive data processing function. 5969 * It always succeeds. The buffer may be dropped during processing 5970 * for congestion control or by the protocol layers. 5971 * 5972 * This function may only be called from softirq context and interrupts 5973 * should be enabled. 5974 * 5975 * Return values (usually ignored): 5976 * NET_RX_SUCCESS: no congestion 5977 * NET_RX_DROP: packet was dropped 5978 */ 5979 int netif_receive_skb(struct sk_buff *skb) 5980 { 5981 int ret; 5982 5983 trace_netif_receive_skb_entry(skb); 5984 5985 ret = netif_receive_skb_internal(skb); 5986 trace_netif_receive_skb_exit(ret); 5987 5988 return ret; 5989 } 5990 EXPORT_SYMBOL(netif_receive_skb); 5991 5992 /** 5993 * netif_receive_skb_list - process many receive buffers from network 5994 * @head: list of skbs to process. 5995 * 5996 * Since return value of netif_receive_skb() is normally ignored, and 5997 * wouldn't be meaningful for a list, this function returns void. 5998 * 5999 * This function may only be called from softirq context and interrupts 6000 * should be enabled. 6001 */ 6002 void netif_receive_skb_list(struct list_head *head) 6003 { 6004 struct sk_buff *skb; 6005 6006 if (list_empty(head)) 6007 return; 6008 if (trace_netif_receive_skb_list_entry_enabled()) { 6009 list_for_each_entry(skb, head, list) 6010 trace_netif_receive_skb_list_entry(skb); 6011 } 6012 netif_receive_skb_list_internal(head); 6013 trace_netif_receive_skb_list_exit(0); 6014 } 6015 EXPORT_SYMBOL(netif_receive_skb_list); 6016 6017 static DEFINE_PER_CPU(struct work_struct, flush_works); 6018 6019 /* Network device is going away, flush any packets still pending */ 6020 static void flush_backlog(struct work_struct *work) 6021 { 6022 struct sk_buff *skb, *tmp; 6023 struct softnet_data *sd; 6024 6025 local_bh_disable(); 6026 sd = this_cpu_ptr(&softnet_data); 6027 6028 backlog_lock_irq_disable(sd); 6029 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 6030 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 6031 __skb_unlink(skb, &sd->input_pkt_queue); 6032 dev_kfree_skb_irq(skb); 6033 rps_input_queue_head_incr(sd); 6034 } 6035 } 6036 backlog_unlock_irq_enable(sd); 6037 6038 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6039 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 6040 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 6041 __skb_unlink(skb, &sd->process_queue); 6042 kfree_skb(skb); 6043 rps_input_queue_head_incr(sd); 6044 } 6045 } 6046 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6047 local_bh_enable(); 6048 } 6049 6050 static bool flush_required(int cpu) 6051 { 6052 #if IS_ENABLED(CONFIG_RPS) 6053 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 6054 bool do_flush; 6055 6056 backlog_lock_irq_disable(sd); 6057 6058 /* as insertion into process_queue happens with the rps lock held, 6059 * process_queue access may race only with dequeue 6060 */ 6061 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 6062 !skb_queue_empty_lockless(&sd->process_queue); 6063 backlog_unlock_irq_enable(sd); 6064 6065 return do_flush; 6066 #endif 6067 /* without RPS we can't safely check input_pkt_queue: during a 6068 * concurrent remote skb_queue_splice() we can detect as empty both 6069 * input_pkt_queue and process_queue even if the latter could end-up 6070 * containing a lot of packets. 6071 */ 6072 return true; 6073 } 6074 6075 static void flush_all_backlogs(void) 6076 { 6077 static cpumask_t flush_cpus; 6078 unsigned int cpu; 6079 6080 /* since we are under rtnl lock protection we can use static data 6081 * for the cpumask and avoid allocating on stack the possibly 6082 * large mask 6083 */ 6084 ASSERT_RTNL(); 6085 6086 cpus_read_lock(); 6087 6088 cpumask_clear(&flush_cpus); 6089 for_each_online_cpu(cpu) { 6090 if (flush_required(cpu)) { 6091 queue_work_on(cpu, system_highpri_wq, 6092 per_cpu_ptr(&flush_works, cpu)); 6093 cpumask_set_cpu(cpu, &flush_cpus); 6094 } 6095 } 6096 6097 /* we can have in flight packet[s] on the cpus we are not flushing, 6098 * synchronize_net() in unregister_netdevice_many() will take care of 6099 * them 6100 */ 6101 for_each_cpu(cpu, &flush_cpus) 6102 flush_work(per_cpu_ptr(&flush_works, cpu)); 6103 6104 cpus_read_unlock(); 6105 } 6106 6107 static void net_rps_send_ipi(struct softnet_data *remsd) 6108 { 6109 #ifdef CONFIG_RPS 6110 while (remsd) { 6111 struct softnet_data *next = remsd->rps_ipi_next; 6112 6113 if (cpu_online(remsd->cpu)) 6114 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6115 remsd = next; 6116 } 6117 #endif 6118 } 6119 6120 /* 6121 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6122 * Note: called with local irq disabled, but exits with local irq enabled. 6123 */ 6124 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6125 { 6126 #ifdef CONFIG_RPS 6127 struct softnet_data *remsd = sd->rps_ipi_list; 6128 6129 if (!use_backlog_threads() && remsd) { 6130 sd->rps_ipi_list = NULL; 6131 6132 local_irq_enable(); 6133 6134 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6135 net_rps_send_ipi(remsd); 6136 } else 6137 #endif 6138 local_irq_enable(); 6139 } 6140 6141 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6142 { 6143 #ifdef CONFIG_RPS 6144 return !use_backlog_threads() && sd->rps_ipi_list; 6145 #else 6146 return false; 6147 #endif 6148 } 6149 6150 static int process_backlog(struct napi_struct *napi, int quota) 6151 { 6152 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6153 bool again = true; 6154 int work = 0; 6155 6156 /* Check if we have pending ipi, its better to send them now, 6157 * not waiting net_rx_action() end. 6158 */ 6159 if (sd_has_rps_ipi_waiting(sd)) { 6160 local_irq_disable(); 6161 net_rps_action_and_irq_enable(sd); 6162 } 6163 6164 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight); 6165 while (again) { 6166 struct sk_buff *skb; 6167 6168 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6169 while ((skb = __skb_dequeue(&sd->process_queue))) { 6170 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6171 rcu_read_lock(); 6172 __netif_receive_skb(skb); 6173 rcu_read_unlock(); 6174 if (++work >= quota) { 6175 rps_input_queue_head_add(sd, work); 6176 return work; 6177 } 6178 6179 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6180 } 6181 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6182 6183 backlog_lock_irq_disable(sd); 6184 if (skb_queue_empty(&sd->input_pkt_queue)) { 6185 /* 6186 * Inline a custom version of __napi_complete(). 6187 * only current cpu owns and manipulates this napi, 6188 * and NAPI_STATE_SCHED is the only possible flag set 6189 * on backlog. 6190 * We can use a plain write instead of clear_bit(), 6191 * and we dont need an smp_mb() memory barrier. 6192 */ 6193 napi->state &= NAPIF_STATE_THREADED; 6194 again = false; 6195 } else { 6196 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6197 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6198 &sd->process_queue); 6199 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6200 } 6201 backlog_unlock_irq_enable(sd); 6202 } 6203 6204 if (work) 6205 rps_input_queue_head_add(sd, work); 6206 return work; 6207 } 6208 6209 /** 6210 * __napi_schedule - schedule for receive 6211 * @n: entry to schedule 6212 * 6213 * The entry's receive function will be scheduled to run. 6214 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6215 */ 6216 void __napi_schedule(struct napi_struct *n) 6217 { 6218 unsigned long flags; 6219 6220 local_irq_save(flags); 6221 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6222 local_irq_restore(flags); 6223 } 6224 EXPORT_SYMBOL(__napi_schedule); 6225 6226 /** 6227 * napi_schedule_prep - check if napi can be scheduled 6228 * @n: napi context 6229 * 6230 * Test if NAPI routine is already running, and if not mark 6231 * it as running. This is used as a condition variable to 6232 * insure only one NAPI poll instance runs. We also make 6233 * sure there is no pending NAPI disable. 6234 */ 6235 bool napi_schedule_prep(struct napi_struct *n) 6236 { 6237 unsigned long new, val = READ_ONCE(n->state); 6238 6239 do { 6240 if (unlikely(val & NAPIF_STATE_DISABLE)) 6241 return false; 6242 new = val | NAPIF_STATE_SCHED; 6243 6244 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6245 * This was suggested by Alexander Duyck, as compiler 6246 * emits better code than : 6247 * if (val & NAPIF_STATE_SCHED) 6248 * new |= NAPIF_STATE_MISSED; 6249 */ 6250 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6251 NAPIF_STATE_MISSED; 6252 } while (!try_cmpxchg(&n->state, &val, new)); 6253 6254 return !(val & NAPIF_STATE_SCHED); 6255 } 6256 EXPORT_SYMBOL(napi_schedule_prep); 6257 6258 /** 6259 * __napi_schedule_irqoff - schedule for receive 6260 * @n: entry to schedule 6261 * 6262 * Variant of __napi_schedule() assuming hard irqs are masked. 6263 * 6264 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6265 * because the interrupt disabled assumption might not be true 6266 * due to force-threaded interrupts and spinlock substitution. 6267 */ 6268 void __napi_schedule_irqoff(struct napi_struct *n) 6269 { 6270 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6271 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6272 else 6273 __napi_schedule(n); 6274 } 6275 EXPORT_SYMBOL(__napi_schedule_irqoff); 6276 6277 bool napi_complete_done(struct napi_struct *n, int work_done) 6278 { 6279 unsigned long flags, val, new, timeout = 0; 6280 bool ret = true; 6281 6282 /* 6283 * 1) Don't let napi dequeue from the cpu poll list 6284 * just in case its running on a different cpu. 6285 * 2) If we are busy polling, do nothing here, we have 6286 * the guarantee we will be called later. 6287 */ 6288 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6289 NAPIF_STATE_IN_BUSY_POLL))) 6290 return false; 6291 6292 if (work_done) { 6293 if (n->gro_bitmask) 6294 timeout = napi_get_gro_flush_timeout(n); 6295 n->defer_hard_irqs_count = napi_get_defer_hard_irqs(n); 6296 } 6297 if (n->defer_hard_irqs_count > 0) { 6298 n->defer_hard_irqs_count--; 6299 timeout = napi_get_gro_flush_timeout(n); 6300 if (timeout) 6301 ret = false; 6302 } 6303 if (n->gro_bitmask) { 6304 /* When the NAPI instance uses a timeout and keeps postponing 6305 * it, we need to bound somehow the time packets are kept in 6306 * the GRO layer 6307 */ 6308 napi_gro_flush(n, !!timeout); 6309 } 6310 6311 gro_normal_list(n); 6312 6313 if (unlikely(!list_empty(&n->poll_list))) { 6314 /* If n->poll_list is not empty, we need to mask irqs */ 6315 local_irq_save(flags); 6316 list_del_init(&n->poll_list); 6317 local_irq_restore(flags); 6318 } 6319 WRITE_ONCE(n->list_owner, -1); 6320 6321 val = READ_ONCE(n->state); 6322 do { 6323 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6324 6325 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6326 NAPIF_STATE_SCHED_THREADED | 6327 NAPIF_STATE_PREFER_BUSY_POLL); 6328 6329 /* If STATE_MISSED was set, leave STATE_SCHED set, 6330 * because we will call napi->poll() one more time. 6331 * This C code was suggested by Alexander Duyck to help gcc. 6332 */ 6333 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6334 NAPIF_STATE_SCHED; 6335 } while (!try_cmpxchg(&n->state, &val, new)); 6336 6337 if (unlikely(val & NAPIF_STATE_MISSED)) { 6338 __napi_schedule(n); 6339 return false; 6340 } 6341 6342 if (timeout) 6343 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6344 HRTIMER_MODE_REL_PINNED); 6345 return ret; 6346 } 6347 EXPORT_SYMBOL(napi_complete_done); 6348 6349 static void skb_defer_free_flush(struct softnet_data *sd) 6350 { 6351 struct sk_buff *skb, *next; 6352 6353 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6354 if (!READ_ONCE(sd->defer_list)) 6355 return; 6356 6357 spin_lock(&sd->defer_lock); 6358 skb = sd->defer_list; 6359 sd->defer_list = NULL; 6360 sd->defer_count = 0; 6361 spin_unlock(&sd->defer_lock); 6362 6363 while (skb != NULL) { 6364 next = skb->next; 6365 napi_consume_skb(skb, 1); 6366 skb = next; 6367 } 6368 } 6369 6370 #if defined(CONFIG_NET_RX_BUSY_POLL) 6371 6372 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6373 { 6374 if (!skip_schedule) { 6375 gro_normal_list(napi); 6376 __napi_schedule(napi); 6377 return; 6378 } 6379 6380 if (napi->gro_bitmask) { 6381 /* flush too old packets 6382 * If HZ < 1000, flush all packets. 6383 */ 6384 napi_gro_flush(napi, HZ >= 1000); 6385 } 6386 6387 gro_normal_list(napi); 6388 clear_bit(NAPI_STATE_SCHED, &napi->state); 6389 } 6390 6391 enum { 6392 NAPI_F_PREFER_BUSY_POLL = 1, 6393 NAPI_F_END_ON_RESCHED = 2, 6394 }; 6395 6396 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, 6397 unsigned flags, u16 budget) 6398 { 6399 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6400 bool skip_schedule = false; 6401 unsigned long timeout; 6402 int rc; 6403 6404 /* Busy polling means there is a high chance device driver hard irq 6405 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6406 * set in napi_schedule_prep(). 6407 * Since we are about to call napi->poll() once more, we can safely 6408 * clear NAPI_STATE_MISSED. 6409 * 6410 * Note: x86 could use a single "lock and ..." instruction 6411 * to perform these two clear_bit() 6412 */ 6413 clear_bit(NAPI_STATE_MISSED, &napi->state); 6414 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6415 6416 local_bh_disable(); 6417 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6418 6419 if (flags & NAPI_F_PREFER_BUSY_POLL) { 6420 napi->defer_hard_irqs_count = napi_get_defer_hard_irqs(napi); 6421 timeout = napi_get_gro_flush_timeout(napi); 6422 if (napi->defer_hard_irqs_count && timeout) { 6423 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6424 skip_schedule = true; 6425 } 6426 } 6427 6428 /* All we really want here is to re-enable device interrupts. 6429 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6430 */ 6431 rc = napi->poll(napi, budget); 6432 /* We can't gro_normal_list() here, because napi->poll() might have 6433 * rearmed the napi (napi_complete_done()) in which case it could 6434 * already be running on another CPU. 6435 */ 6436 trace_napi_poll(napi, rc, budget); 6437 netpoll_poll_unlock(have_poll_lock); 6438 if (rc == budget) 6439 __busy_poll_stop(napi, skip_schedule); 6440 bpf_net_ctx_clear(bpf_net_ctx); 6441 local_bh_enable(); 6442 } 6443 6444 static void __napi_busy_loop(unsigned int napi_id, 6445 bool (*loop_end)(void *, unsigned long), 6446 void *loop_end_arg, unsigned flags, u16 budget) 6447 { 6448 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6449 int (*napi_poll)(struct napi_struct *napi, int budget); 6450 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6451 void *have_poll_lock = NULL; 6452 struct napi_struct *napi; 6453 6454 WARN_ON_ONCE(!rcu_read_lock_held()); 6455 6456 restart: 6457 napi_poll = NULL; 6458 6459 napi = napi_by_id(napi_id); 6460 if (!napi) 6461 return; 6462 6463 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6464 preempt_disable(); 6465 for (;;) { 6466 int work = 0; 6467 6468 local_bh_disable(); 6469 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6470 if (!napi_poll) { 6471 unsigned long val = READ_ONCE(napi->state); 6472 6473 /* If multiple threads are competing for this napi, 6474 * we avoid dirtying napi->state as much as we can. 6475 */ 6476 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6477 NAPIF_STATE_IN_BUSY_POLL)) { 6478 if (flags & NAPI_F_PREFER_BUSY_POLL) 6479 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6480 goto count; 6481 } 6482 if (cmpxchg(&napi->state, val, 6483 val | NAPIF_STATE_IN_BUSY_POLL | 6484 NAPIF_STATE_SCHED) != val) { 6485 if (flags & NAPI_F_PREFER_BUSY_POLL) 6486 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6487 goto count; 6488 } 6489 have_poll_lock = netpoll_poll_lock(napi); 6490 napi_poll = napi->poll; 6491 } 6492 work = napi_poll(napi, budget); 6493 trace_napi_poll(napi, work, budget); 6494 gro_normal_list(napi); 6495 count: 6496 if (work > 0) 6497 __NET_ADD_STATS(dev_net(napi->dev), 6498 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6499 skb_defer_free_flush(this_cpu_ptr(&softnet_data)); 6500 bpf_net_ctx_clear(bpf_net_ctx); 6501 local_bh_enable(); 6502 6503 if (!loop_end || loop_end(loop_end_arg, start_time)) 6504 break; 6505 6506 if (unlikely(need_resched())) { 6507 if (flags & NAPI_F_END_ON_RESCHED) 6508 break; 6509 if (napi_poll) 6510 busy_poll_stop(napi, have_poll_lock, flags, budget); 6511 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6512 preempt_enable(); 6513 rcu_read_unlock(); 6514 cond_resched(); 6515 rcu_read_lock(); 6516 if (loop_end(loop_end_arg, start_time)) 6517 return; 6518 goto restart; 6519 } 6520 cpu_relax(); 6521 } 6522 if (napi_poll) 6523 busy_poll_stop(napi, have_poll_lock, flags, budget); 6524 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6525 preempt_enable(); 6526 } 6527 6528 void napi_busy_loop_rcu(unsigned int napi_id, 6529 bool (*loop_end)(void *, unsigned long), 6530 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6531 { 6532 unsigned flags = NAPI_F_END_ON_RESCHED; 6533 6534 if (prefer_busy_poll) 6535 flags |= NAPI_F_PREFER_BUSY_POLL; 6536 6537 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6538 } 6539 6540 void napi_busy_loop(unsigned int napi_id, 6541 bool (*loop_end)(void *, unsigned long), 6542 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6543 { 6544 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; 6545 6546 rcu_read_lock(); 6547 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6548 rcu_read_unlock(); 6549 } 6550 EXPORT_SYMBOL(napi_busy_loop); 6551 6552 void napi_suspend_irqs(unsigned int napi_id) 6553 { 6554 struct napi_struct *napi; 6555 6556 rcu_read_lock(); 6557 napi = napi_by_id(napi_id); 6558 if (napi) { 6559 unsigned long timeout = napi_get_irq_suspend_timeout(napi); 6560 6561 if (timeout) 6562 hrtimer_start(&napi->timer, ns_to_ktime(timeout), 6563 HRTIMER_MODE_REL_PINNED); 6564 } 6565 rcu_read_unlock(); 6566 } 6567 6568 void napi_resume_irqs(unsigned int napi_id) 6569 { 6570 struct napi_struct *napi; 6571 6572 rcu_read_lock(); 6573 napi = napi_by_id(napi_id); 6574 if (napi) { 6575 /* If irq_suspend_timeout is set to 0 between the call to 6576 * napi_suspend_irqs and now, the original value still 6577 * determines the safety timeout as intended and napi_watchdog 6578 * will resume irq processing. 6579 */ 6580 if (napi_get_irq_suspend_timeout(napi)) { 6581 local_bh_disable(); 6582 napi_schedule(napi); 6583 local_bh_enable(); 6584 } 6585 } 6586 rcu_read_unlock(); 6587 } 6588 6589 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6590 6591 static void __napi_hash_add_with_id(struct napi_struct *napi, 6592 unsigned int napi_id) 6593 { 6594 napi->napi_id = napi_id; 6595 hlist_add_head_rcu(&napi->napi_hash_node, 6596 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6597 } 6598 6599 static void napi_hash_add_with_id(struct napi_struct *napi, 6600 unsigned int napi_id) 6601 { 6602 unsigned long flags; 6603 6604 spin_lock_irqsave(&napi_hash_lock, flags); 6605 WARN_ON_ONCE(napi_by_id(napi_id)); 6606 __napi_hash_add_with_id(napi, napi_id); 6607 spin_unlock_irqrestore(&napi_hash_lock, flags); 6608 } 6609 6610 static void napi_hash_add(struct napi_struct *napi) 6611 { 6612 unsigned long flags; 6613 6614 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6615 return; 6616 6617 spin_lock_irqsave(&napi_hash_lock, flags); 6618 6619 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6620 do { 6621 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6622 napi_gen_id = MIN_NAPI_ID; 6623 } while (napi_by_id(napi_gen_id)); 6624 6625 __napi_hash_add_with_id(napi, napi_gen_id); 6626 6627 spin_unlock_irqrestore(&napi_hash_lock, flags); 6628 } 6629 6630 /* Warning : caller is responsible to make sure rcu grace period 6631 * is respected before freeing memory containing @napi 6632 */ 6633 static void napi_hash_del(struct napi_struct *napi) 6634 { 6635 unsigned long flags; 6636 6637 spin_lock_irqsave(&napi_hash_lock, flags); 6638 6639 hlist_del_init_rcu(&napi->napi_hash_node); 6640 6641 spin_unlock_irqrestore(&napi_hash_lock, flags); 6642 } 6643 6644 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6645 { 6646 struct napi_struct *napi; 6647 6648 napi = container_of(timer, struct napi_struct, timer); 6649 6650 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6651 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6652 */ 6653 if (!napi_disable_pending(napi) && 6654 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6655 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6656 __napi_schedule_irqoff(napi); 6657 } 6658 6659 return HRTIMER_NORESTART; 6660 } 6661 6662 static void init_gro_hash(struct napi_struct *napi) 6663 { 6664 int i; 6665 6666 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6667 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6668 napi->gro_hash[i].count = 0; 6669 } 6670 napi->gro_bitmask = 0; 6671 } 6672 6673 int dev_set_threaded(struct net_device *dev, bool threaded) 6674 { 6675 struct napi_struct *napi; 6676 int err = 0; 6677 6678 if (dev->threaded == threaded) 6679 return 0; 6680 6681 if (threaded) { 6682 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6683 if (!napi->thread) { 6684 err = napi_kthread_create(napi); 6685 if (err) { 6686 threaded = false; 6687 break; 6688 } 6689 } 6690 } 6691 } 6692 6693 WRITE_ONCE(dev->threaded, threaded); 6694 6695 /* Make sure kthread is created before THREADED bit 6696 * is set. 6697 */ 6698 smp_mb__before_atomic(); 6699 6700 /* Setting/unsetting threaded mode on a napi might not immediately 6701 * take effect, if the current napi instance is actively being 6702 * polled. In this case, the switch between threaded mode and 6703 * softirq mode will happen in the next round of napi_schedule(). 6704 * This should not cause hiccups/stalls to the live traffic. 6705 */ 6706 list_for_each_entry(napi, &dev->napi_list, dev_list) 6707 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6708 6709 return err; 6710 } 6711 EXPORT_SYMBOL(dev_set_threaded); 6712 6713 /** 6714 * netif_queue_set_napi - Associate queue with the napi 6715 * @dev: device to which NAPI and queue belong 6716 * @queue_index: Index of queue 6717 * @type: queue type as RX or TX 6718 * @napi: NAPI context, pass NULL to clear previously set NAPI 6719 * 6720 * Set queue with its corresponding napi context. This should be done after 6721 * registering the NAPI handler for the queue-vector and the queues have been 6722 * mapped to the corresponding interrupt vector. 6723 */ 6724 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6725 enum netdev_queue_type type, struct napi_struct *napi) 6726 { 6727 struct netdev_rx_queue *rxq; 6728 struct netdev_queue *txq; 6729 6730 if (WARN_ON_ONCE(napi && !napi->dev)) 6731 return; 6732 if (dev->reg_state >= NETREG_REGISTERED) 6733 ASSERT_RTNL(); 6734 6735 switch (type) { 6736 case NETDEV_QUEUE_TYPE_RX: 6737 rxq = __netif_get_rx_queue(dev, queue_index); 6738 rxq->napi = napi; 6739 return; 6740 case NETDEV_QUEUE_TYPE_TX: 6741 txq = netdev_get_tx_queue(dev, queue_index); 6742 txq->napi = napi; 6743 return; 6744 default: 6745 return; 6746 } 6747 } 6748 EXPORT_SYMBOL(netif_queue_set_napi); 6749 6750 static void napi_restore_config(struct napi_struct *n) 6751 { 6752 n->defer_hard_irqs = n->config->defer_hard_irqs; 6753 n->gro_flush_timeout = n->config->gro_flush_timeout; 6754 n->irq_suspend_timeout = n->config->irq_suspend_timeout; 6755 /* a NAPI ID might be stored in the config, if so use it. if not, use 6756 * napi_hash_add to generate one for us. 6757 */ 6758 if (n->config->napi_id) { 6759 napi_hash_add_with_id(n, n->config->napi_id); 6760 } else { 6761 napi_hash_add(n); 6762 n->config->napi_id = n->napi_id; 6763 } 6764 } 6765 6766 static void napi_save_config(struct napi_struct *n) 6767 { 6768 n->config->defer_hard_irqs = n->defer_hard_irqs; 6769 n->config->gro_flush_timeout = n->gro_flush_timeout; 6770 n->config->irq_suspend_timeout = n->irq_suspend_timeout; 6771 napi_hash_del(n); 6772 } 6773 6774 /* Netlink wants the NAPI list to be sorted by ID, if adding a NAPI which will 6775 * inherit an existing ID try to insert it at the right position. 6776 */ 6777 static void 6778 netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) 6779 { 6780 unsigned int new_id, pos_id; 6781 struct list_head *higher; 6782 struct napi_struct *pos; 6783 6784 new_id = UINT_MAX; 6785 if (napi->config && napi->config->napi_id) 6786 new_id = napi->config->napi_id; 6787 6788 higher = &dev->napi_list; 6789 list_for_each_entry(pos, &dev->napi_list, dev_list) { 6790 if (pos->napi_id >= MIN_NAPI_ID) 6791 pos_id = pos->napi_id; 6792 else if (pos->config) 6793 pos_id = pos->config->napi_id; 6794 else 6795 pos_id = UINT_MAX; 6796 6797 if (pos_id <= new_id) 6798 break; 6799 higher = &pos->dev_list; 6800 } 6801 list_add_rcu(&napi->dev_list, higher); /* adds after higher */ 6802 } 6803 6804 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6805 int (*poll)(struct napi_struct *, int), int weight) 6806 { 6807 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6808 return; 6809 6810 INIT_LIST_HEAD(&napi->poll_list); 6811 INIT_HLIST_NODE(&napi->napi_hash_node); 6812 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6813 napi->timer.function = napi_watchdog; 6814 init_gro_hash(napi); 6815 napi->skb = NULL; 6816 INIT_LIST_HEAD(&napi->rx_list); 6817 napi->rx_count = 0; 6818 napi->poll = poll; 6819 if (weight > NAPI_POLL_WEIGHT) 6820 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6821 weight); 6822 napi->weight = weight; 6823 napi->dev = dev; 6824 #ifdef CONFIG_NETPOLL 6825 napi->poll_owner = -1; 6826 #endif 6827 napi->list_owner = -1; 6828 set_bit(NAPI_STATE_SCHED, &napi->state); 6829 set_bit(NAPI_STATE_NPSVC, &napi->state); 6830 netif_napi_dev_list_add(dev, napi); 6831 6832 /* default settings from sysfs are applied to all NAPIs. any per-NAPI 6833 * configuration will be loaded in napi_enable 6834 */ 6835 napi_set_defer_hard_irqs(napi, READ_ONCE(dev->napi_defer_hard_irqs)); 6836 napi_set_gro_flush_timeout(napi, READ_ONCE(dev->gro_flush_timeout)); 6837 6838 napi_get_frags_check(napi); 6839 /* Create kthread for this napi if dev->threaded is set. 6840 * Clear dev->threaded if kthread creation failed so that 6841 * threaded mode will not be enabled in napi_enable(). 6842 */ 6843 if (dev->threaded && napi_kthread_create(napi)) 6844 dev->threaded = false; 6845 netif_napi_set_irq(napi, -1); 6846 } 6847 EXPORT_SYMBOL(netif_napi_add_weight); 6848 6849 void napi_disable(struct napi_struct *n) 6850 { 6851 unsigned long val, new; 6852 6853 might_sleep(); 6854 set_bit(NAPI_STATE_DISABLE, &n->state); 6855 6856 val = READ_ONCE(n->state); 6857 do { 6858 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6859 usleep_range(20, 200); 6860 val = READ_ONCE(n->state); 6861 } 6862 6863 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6864 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6865 } while (!try_cmpxchg(&n->state, &val, new)); 6866 6867 hrtimer_cancel(&n->timer); 6868 6869 if (n->config) 6870 napi_save_config(n); 6871 else 6872 napi_hash_del(n); 6873 6874 clear_bit(NAPI_STATE_DISABLE, &n->state); 6875 } 6876 EXPORT_SYMBOL(napi_disable); 6877 6878 /** 6879 * napi_enable - enable NAPI scheduling 6880 * @n: NAPI context 6881 * 6882 * Resume NAPI from being scheduled on this context. 6883 * Must be paired with napi_disable. 6884 */ 6885 void napi_enable(struct napi_struct *n) 6886 { 6887 unsigned long new, val = READ_ONCE(n->state); 6888 6889 if (n->config) 6890 napi_restore_config(n); 6891 else 6892 napi_hash_add(n); 6893 6894 do { 6895 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6896 6897 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6898 if (n->dev->threaded && n->thread) 6899 new |= NAPIF_STATE_THREADED; 6900 } while (!try_cmpxchg(&n->state, &val, new)); 6901 } 6902 EXPORT_SYMBOL(napi_enable); 6903 6904 static void flush_gro_hash(struct napi_struct *napi) 6905 { 6906 int i; 6907 6908 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6909 struct sk_buff *skb, *n; 6910 6911 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6912 kfree_skb(skb); 6913 napi->gro_hash[i].count = 0; 6914 } 6915 } 6916 6917 /* Must be called in process context */ 6918 void __netif_napi_del(struct napi_struct *napi) 6919 { 6920 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6921 return; 6922 6923 if (napi->config) { 6924 napi->index = -1; 6925 napi->config = NULL; 6926 } 6927 6928 list_del_rcu(&napi->dev_list); 6929 napi_free_frags(napi); 6930 6931 flush_gro_hash(napi); 6932 napi->gro_bitmask = 0; 6933 6934 if (napi->thread) { 6935 kthread_stop(napi->thread); 6936 napi->thread = NULL; 6937 } 6938 } 6939 EXPORT_SYMBOL(__netif_napi_del); 6940 6941 static int __napi_poll(struct napi_struct *n, bool *repoll) 6942 { 6943 int work, weight; 6944 6945 weight = n->weight; 6946 6947 /* This NAPI_STATE_SCHED test is for avoiding a race 6948 * with netpoll's poll_napi(). Only the entity which 6949 * obtains the lock and sees NAPI_STATE_SCHED set will 6950 * actually make the ->poll() call. Therefore we avoid 6951 * accidentally calling ->poll() when NAPI is not scheduled. 6952 */ 6953 work = 0; 6954 if (napi_is_scheduled(n)) { 6955 work = n->poll(n, weight); 6956 trace_napi_poll(n, work, weight); 6957 6958 xdp_do_check_flushed(n); 6959 } 6960 6961 if (unlikely(work > weight)) 6962 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6963 n->poll, work, weight); 6964 6965 if (likely(work < weight)) 6966 return work; 6967 6968 /* Drivers must not modify the NAPI state if they 6969 * consume the entire weight. In such cases this code 6970 * still "owns" the NAPI instance and therefore can 6971 * move the instance around on the list at-will. 6972 */ 6973 if (unlikely(napi_disable_pending(n))) { 6974 napi_complete(n); 6975 return work; 6976 } 6977 6978 /* The NAPI context has more processing work, but busy-polling 6979 * is preferred. Exit early. 6980 */ 6981 if (napi_prefer_busy_poll(n)) { 6982 if (napi_complete_done(n, work)) { 6983 /* If timeout is not set, we need to make sure 6984 * that the NAPI is re-scheduled. 6985 */ 6986 napi_schedule(n); 6987 } 6988 return work; 6989 } 6990 6991 if (n->gro_bitmask) { 6992 /* flush too old packets 6993 * If HZ < 1000, flush all packets. 6994 */ 6995 napi_gro_flush(n, HZ >= 1000); 6996 } 6997 6998 gro_normal_list(n); 6999 7000 /* Some drivers may have called napi_schedule 7001 * prior to exhausting their budget. 7002 */ 7003 if (unlikely(!list_empty(&n->poll_list))) { 7004 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7005 n->dev ? n->dev->name : "backlog"); 7006 return work; 7007 } 7008 7009 *repoll = true; 7010 7011 return work; 7012 } 7013 7014 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7015 { 7016 bool do_repoll = false; 7017 void *have; 7018 int work; 7019 7020 list_del_init(&n->poll_list); 7021 7022 have = netpoll_poll_lock(n); 7023 7024 work = __napi_poll(n, &do_repoll); 7025 7026 if (do_repoll) 7027 list_add_tail(&n->poll_list, repoll); 7028 7029 netpoll_poll_unlock(have); 7030 7031 return work; 7032 } 7033 7034 static int napi_thread_wait(struct napi_struct *napi) 7035 { 7036 set_current_state(TASK_INTERRUPTIBLE); 7037 7038 while (!kthread_should_stop()) { 7039 /* Testing SCHED_THREADED bit here to make sure the current 7040 * kthread owns this napi and could poll on this napi. 7041 * Testing SCHED bit is not enough because SCHED bit might be 7042 * set by some other busy poll thread or by napi_disable(). 7043 */ 7044 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) { 7045 WARN_ON(!list_empty(&napi->poll_list)); 7046 __set_current_state(TASK_RUNNING); 7047 return 0; 7048 } 7049 7050 schedule(); 7051 set_current_state(TASK_INTERRUPTIBLE); 7052 } 7053 __set_current_state(TASK_RUNNING); 7054 7055 return -1; 7056 } 7057 7058 static void napi_threaded_poll_loop(struct napi_struct *napi) 7059 { 7060 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7061 struct softnet_data *sd; 7062 unsigned long last_qs = jiffies; 7063 7064 for (;;) { 7065 bool repoll = false; 7066 void *have; 7067 7068 local_bh_disable(); 7069 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7070 7071 sd = this_cpu_ptr(&softnet_data); 7072 sd->in_napi_threaded_poll = true; 7073 7074 have = netpoll_poll_lock(napi); 7075 __napi_poll(napi, &repoll); 7076 netpoll_poll_unlock(have); 7077 7078 sd->in_napi_threaded_poll = false; 7079 barrier(); 7080 7081 if (sd_has_rps_ipi_waiting(sd)) { 7082 local_irq_disable(); 7083 net_rps_action_and_irq_enable(sd); 7084 } 7085 skb_defer_free_flush(sd); 7086 bpf_net_ctx_clear(bpf_net_ctx); 7087 local_bh_enable(); 7088 7089 if (!repoll) 7090 break; 7091 7092 rcu_softirq_qs_periodic(last_qs); 7093 cond_resched(); 7094 } 7095 } 7096 7097 static int napi_threaded_poll(void *data) 7098 { 7099 struct napi_struct *napi = data; 7100 7101 while (!napi_thread_wait(napi)) 7102 napi_threaded_poll_loop(napi); 7103 7104 return 0; 7105 } 7106 7107 static __latent_entropy void net_rx_action(void) 7108 { 7109 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7110 unsigned long time_limit = jiffies + 7111 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs)); 7112 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7113 int budget = READ_ONCE(net_hotdata.netdev_budget); 7114 LIST_HEAD(list); 7115 LIST_HEAD(repoll); 7116 7117 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7118 start: 7119 sd->in_net_rx_action = true; 7120 local_irq_disable(); 7121 list_splice_init(&sd->poll_list, &list); 7122 local_irq_enable(); 7123 7124 for (;;) { 7125 struct napi_struct *n; 7126 7127 skb_defer_free_flush(sd); 7128 7129 if (list_empty(&list)) { 7130 if (list_empty(&repoll)) { 7131 sd->in_net_rx_action = false; 7132 barrier(); 7133 /* We need to check if ____napi_schedule() 7134 * had refilled poll_list while 7135 * sd->in_net_rx_action was true. 7136 */ 7137 if (!list_empty(&sd->poll_list)) 7138 goto start; 7139 if (!sd_has_rps_ipi_waiting(sd)) 7140 goto end; 7141 } 7142 break; 7143 } 7144 7145 n = list_first_entry(&list, struct napi_struct, poll_list); 7146 budget -= napi_poll(n, &repoll); 7147 7148 /* If softirq window is exhausted then punt. 7149 * Allow this to run for 2 jiffies since which will allow 7150 * an average latency of 1.5/HZ. 7151 */ 7152 if (unlikely(budget <= 0 || 7153 time_after_eq(jiffies, time_limit))) { 7154 sd->time_squeeze++; 7155 break; 7156 } 7157 } 7158 7159 local_irq_disable(); 7160 7161 list_splice_tail_init(&sd->poll_list, &list); 7162 list_splice_tail(&repoll, &list); 7163 list_splice(&list, &sd->poll_list); 7164 if (!list_empty(&sd->poll_list)) 7165 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7166 else 7167 sd->in_net_rx_action = false; 7168 7169 net_rps_action_and_irq_enable(sd); 7170 end: 7171 bpf_net_ctx_clear(bpf_net_ctx); 7172 } 7173 7174 struct netdev_adjacent { 7175 struct net_device *dev; 7176 netdevice_tracker dev_tracker; 7177 7178 /* upper master flag, there can only be one master device per list */ 7179 bool master; 7180 7181 /* lookup ignore flag */ 7182 bool ignore; 7183 7184 /* counter for the number of times this device was added to us */ 7185 u16 ref_nr; 7186 7187 /* private field for the users */ 7188 void *private; 7189 7190 struct list_head list; 7191 struct rcu_head rcu; 7192 }; 7193 7194 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7195 struct list_head *adj_list) 7196 { 7197 struct netdev_adjacent *adj; 7198 7199 list_for_each_entry(adj, adj_list, list) { 7200 if (adj->dev == adj_dev) 7201 return adj; 7202 } 7203 return NULL; 7204 } 7205 7206 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7207 struct netdev_nested_priv *priv) 7208 { 7209 struct net_device *dev = (struct net_device *)priv->data; 7210 7211 return upper_dev == dev; 7212 } 7213 7214 /** 7215 * netdev_has_upper_dev - Check if device is linked to an upper device 7216 * @dev: device 7217 * @upper_dev: upper device to check 7218 * 7219 * Find out if a device is linked to specified upper device and return true 7220 * in case it is. Note that this checks only immediate upper device, 7221 * not through a complete stack of devices. The caller must hold the RTNL lock. 7222 */ 7223 bool netdev_has_upper_dev(struct net_device *dev, 7224 struct net_device *upper_dev) 7225 { 7226 struct netdev_nested_priv priv = { 7227 .data = (void *)upper_dev, 7228 }; 7229 7230 ASSERT_RTNL(); 7231 7232 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7233 &priv); 7234 } 7235 EXPORT_SYMBOL(netdev_has_upper_dev); 7236 7237 /** 7238 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7239 * @dev: device 7240 * @upper_dev: upper device to check 7241 * 7242 * Find out if a device is linked to specified upper device and return true 7243 * in case it is. Note that this checks the entire upper device chain. 7244 * The caller must hold rcu lock. 7245 */ 7246 7247 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7248 struct net_device *upper_dev) 7249 { 7250 struct netdev_nested_priv priv = { 7251 .data = (void *)upper_dev, 7252 }; 7253 7254 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7255 &priv); 7256 } 7257 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7258 7259 /** 7260 * netdev_has_any_upper_dev - Check if device is linked to some device 7261 * @dev: device 7262 * 7263 * Find out if a device is linked to an upper device and return true in case 7264 * it is. The caller must hold the RTNL lock. 7265 */ 7266 bool netdev_has_any_upper_dev(struct net_device *dev) 7267 { 7268 ASSERT_RTNL(); 7269 7270 return !list_empty(&dev->adj_list.upper); 7271 } 7272 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7273 7274 /** 7275 * netdev_master_upper_dev_get - Get master upper device 7276 * @dev: device 7277 * 7278 * Find a master upper device and return pointer to it or NULL in case 7279 * it's not there. The caller must hold the RTNL lock. 7280 */ 7281 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7282 { 7283 struct netdev_adjacent *upper; 7284 7285 ASSERT_RTNL(); 7286 7287 if (list_empty(&dev->adj_list.upper)) 7288 return NULL; 7289 7290 upper = list_first_entry(&dev->adj_list.upper, 7291 struct netdev_adjacent, list); 7292 if (likely(upper->master)) 7293 return upper->dev; 7294 return NULL; 7295 } 7296 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7297 7298 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7299 { 7300 struct netdev_adjacent *upper; 7301 7302 ASSERT_RTNL(); 7303 7304 if (list_empty(&dev->adj_list.upper)) 7305 return NULL; 7306 7307 upper = list_first_entry(&dev->adj_list.upper, 7308 struct netdev_adjacent, list); 7309 if (likely(upper->master) && !upper->ignore) 7310 return upper->dev; 7311 return NULL; 7312 } 7313 7314 /** 7315 * netdev_has_any_lower_dev - Check if device is linked to some device 7316 * @dev: device 7317 * 7318 * Find out if a device is linked to a lower device and return true in case 7319 * it is. The caller must hold the RTNL lock. 7320 */ 7321 static bool netdev_has_any_lower_dev(struct net_device *dev) 7322 { 7323 ASSERT_RTNL(); 7324 7325 return !list_empty(&dev->adj_list.lower); 7326 } 7327 7328 void *netdev_adjacent_get_private(struct list_head *adj_list) 7329 { 7330 struct netdev_adjacent *adj; 7331 7332 adj = list_entry(adj_list, struct netdev_adjacent, list); 7333 7334 return adj->private; 7335 } 7336 EXPORT_SYMBOL(netdev_adjacent_get_private); 7337 7338 /** 7339 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7340 * @dev: device 7341 * @iter: list_head ** of the current position 7342 * 7343 * Gets the next device from the dev's upper list, starting from iter 7344 * position. The caller must hold RCU read lock. 7345 */ 7346 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7347 struct list_head **iter) 7348 { 7349 struct netdev_adjacent *upper; 7350 7351 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7352 7353 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7354 7355 if (&upper->list == &dev->adj_list.upper) 7356 return NULL; 7357 7358 *iter = &upper->list; 7359 7360 return upper->dev; 7361 } 7362 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7363 7364 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7365 struct list_head **iter, 7366 bool *ignore) 7367 { 7368 struct netdev_adjacent *upper; 7369 7370 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7371 7372 if (&upper->list == &dev->adj_list.upper) 7373 return NULL; 7374 7375 *iter = &upper->list; 7376 *ignore = upper->ignore; 7377 7378 return upper->dev; 7379 } 7380 7381 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7382 struct list_head **iter) 7383 { 7384 struct netdev_adjacent *upper; 7385 7386 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7387 7388 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7389 7390 if (&upper->list == &dev->adj_list.upper) 7391 return NULL; 7392 7393 *iter = &upper->list; 7394 7395 return upper->dev; 7396 } 7397 7398 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7399 int (*fn)(struct net_device *dev, 7400 struct netdev_nested_priv *priv), 7401 struct netdev_nested_priv *priv) 7402 { 7403 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7404 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7405 int ret, cur = 0; 7406 bool ignore; 7407 7408 now = dev; 7409 iter = &dev->adj_list.upper; 7410 7411 while (1) { 7412 if (now != dev) { 7413 ret = fn(now, priv); 7414 if (ret) 7415 return ret; 7416 } 7417 7418 next = NULL; 7419 while (1) { 7420 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7421 if (!udev) 7422 break; 7423 if (ignore) 7424 continue; 7425 7426 next = udev; 7427 niter = &udev->adj_list.upper; 7428 dev_stack[cur] = now; 7429 iter_stack[cur++] = iter; 7430 break; 7431 } 7432 7433 if (!next) { 7434 if (!cur) 7435 return 0; 7436 next = dev_stack[--cur]; 7437 niter = iter_stack[cur]; 7438 } 7439 7440 now = next; 7441 iter = niter; 7442 } 7443 7444 return 0; 7445 } 7446 7447 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7448 int (*fn)(struct net_device *dev, 7449 struct netdev_nested_priv *priv), 7450 struct netdev_nested_priv *priv) 7451 { 7452 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7453 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7454 int ret, cur = 0; 7455 7456 now = dev; 7457 iter = &dev->adj_list.upper; 7458 7459 while (1) { 7460 if (now != dev) { 7461 ret = fn(now, priv); 7462 if (ret) 7463 return ret; 7464 } 7465 7466 next = NULL; 7467 while (1) { 7468 udev = netdev_next_upper_dev_rcu(now, &iter); 7469 if (!udev) 7470 break; 7471 7472 next = udev; 7473 niter = &udev->adj_list.upper; 7474 dev_stack[cur] = now; 7475 iter_stack[cur++] = iter; 7476 break; 7477 } 7478 7479 if (!next) { 7480 if (!cur) 7481 return 0; 7482 next = dev_stack[--cur]; 7483 niter = iter_stack[cur]; 7484 } 7485 7486 now = next; 7487 iter = niter; 7488 } 7489 7490 return 0; 7491 } 7492 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7493 7494 static bool __netdev_has_upper_dev(struct net_device *dev, 7495 struct net_device *upper_dev) 7496 { 7497 struct netdev_nested_priv priv = { 7498 .flags = 0, 7499 .data = (void *)upper_dev, 7500 }; 7501 7502 ASSERT_RTNL(); 7503 7504 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7505 &priv); 7506 } 7507 7508 /** 7509 * netdev_lower_get_next_private - Get the next ->private from the 7510 * lower neighbour list 7511 * @dev: device 7512 * @iter: list_head ** of the current position 7513 * 7514 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7515 * list, starting from iter position. The caller must hold either hold the 7516 * RTNL lock or its own locking that guarantees that the neighbour lower 7517 * list will remain unchanged. 7518 */ 7519 void *netdev_lower_get_next_private(struct net_device *dev, 7520 struct list_head **iter) 7521 { 7522 struct netdev_adjacent *lower; 7523 7524 lower = list_entry(*iter, struct netdev_adjacent, list); 7525 7526 if (&lower->list == &dev->adj_list.lower) 7527 return NULL; 7528 7529 *iter = lower->list.next; 7530 7531 return lower->private; 7532 } 7533 EXPORT_SYMBOL(netdev_lower_get_next_private); 7534 7535 /** 7536 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7537 * lower neighbour list, RCU 7538 * variant 7539 * @dev: device 7540 * @iter: list_head ** of the current position 7541 * 7542 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7543 * list, starting from iter position. The caller must hold RCU read lock. 7544 */ 7545 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7546 struct list_head **iter) 7547 { 7548 struct netdev_adjacent *lower; 7549 7550 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7551 7552 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7553 7554 if (&lower->list == &dev->adj_list.lower) 7555 return NULL; 7556 7557 *iter = &lower->list; 7558 7559 return lower->private; 7560 } 7561 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7562 7563 /** 7564 * netdev_lower_get_next - Get the next device from the lower neighbour 7565 * list 7566 * @dev: device 7567 * @iter: list_head ** of the current position 7568 * 7569 * Gets the next netdev_adjacent from the dev's lower neighbour 7570 * list, starting from iter position. The caller must hold RTNL lock or 7571 * its own locking that guarantees that the neighbour lower 7572 * list will remain unchanged. 7573 */ 7574 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7575 { 7576 struct netdev_adjacent *lower; 7577 7578 lower = list_entry(*iter, struct netdev_adjacent, list); 7579 7580 if (&lower->list == &dev->adj_list.lower) 7581 return NULL; 7582 7583 *iter = lower->list.next; 7584 7585 return lower->dev; 7586 } 7587 EXPORT_SYMBOL(netdev_lower_get_next); 7588 7589 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7590 struct list_head **iter) 7591 { 7592 struct netdev_adjacent *lower; 7593 7594 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7595 7596 if (&lower->list == &dev->adj_list.lower) 7597 return NULL; 7598 7599 *iter = &lower->list; 7600 7601 return lower->dev; 7602 } 7603 7604 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7605 struct list_head **iter, 7606 bool *ignore) 7607 { 7608 struct netdev_adjacent *lower; 7609 7610 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7611 7612 if (&lower->list == &dev->adj_list.lower) 7613 return NULL; 7614 7615 *iter = &lower->list; 7616 *ignore = lower->ignore; 7617 7618 return lower->dev; 7619 } 7620 7621 int netdev_walk_all_lower_dev(struct net_device *dev, 7622 int (*fn)(struct net_device *dev, 7623 struct netdev_nested_priv *priv), 7624 struct netdev_nested_priv *priv) 7625 { 7626 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7627 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7628 int ret, cur = 0; 7629 7630 now = dev; 7631 iter = &dev->adj_list.lower; 7632 7633 while (1) { 7634 if (now != dev) { 7635 ret = fn(now, priv); 7636 if (ret) 7637 return ret; 7638 } 7639 7640 next = NULL; 7641 while (1) { 7642 ldev = netdev_next_lower_dev(now, &iter); 7643 if (!ldev) 7644 break; 7645 7646 next = ldev; 7647 niter = &ldev->adj_list.lower; 7648 dev_stack[cur] = now; 7649 iter_stack[cur++] = iter; 7650 break; 7651 } 7652 7653 if (!next) { 7654 if (!cur) 7655 return 0; 7656 next = dev_stack[--cur]; 7657 niter = iter_stack[cur]; 7658 } 7659 7660 now = next; 7661 iter = niter; 7662 } 7663 7664 return 0; 7665 } 7666 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7667 7668 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7669 int (*fn)(struct net_device *dev, 7670 struct netdev_nested_priv *priv), 7671 struct netdev_nested_priv *priv) 7672 { 7673 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7674 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7675 int ret, cur = 0; 7676 bool ignore; 7677 7678 now = dev; 7679 iter = &dev->adj_list.lower; 7680 7681 while (1) { 7682 if (now != dev) { 7683 ret = fn(now, priv); 7684 if (ret) 7685 return ret; 7686 } 7687 7688 next = NULL; 7689 while (1) { 7690 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7691 if (!ldev) 7692 break; 7693 if (ignore) 7694 continue; 7695 7696 next = ldev; 7697 niter = &ldev->adj_list.lower; 7698 dev_stack[cur] = now; 7699 iter_stack[cur++] = iter; 7700 break; 7701 } 7702 7703 if (!next) { 7704 if (!cur) 7705 return 0; 7706 next = dev_stack[--cur]; 7707 niter = iter_stack[cur]; 7708 } 7709 7710 now = next; 7711 iter = niter; 7712 } 7713 7714 return 0; 7715 } 7716 7717 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7718 struct list_head **iter) 7719 { 7720 struct netdev_adjacent *lower; 7721 7722 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7723 if (&lower->list == &dev->adj_list.lower) 7724 return NULL; 7725 7726 *iter = &lower->list; 7727 7728 return lower->dev; 7729 } 7730 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7731 7732 static u8 __netdev_upper_depth(struct net_device *dev) 7733 { 7734 struct net_device *udev; 7735 struct list_head *iter; 7736 u8 max_depth = 0; 7737 bool ignore; 7738 7739 for (iter = &dev->adj_list.upper, 7740 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7741 udev; 7742 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7743 if (ignore) 7744 continue; 7745 if (max_depth < udev->upper_level) 7746 max_depth = udev->upper_level; 7747 } 7748 7749 return max_depth; 7750 } 7751 7752 static u8 __netdev_lower_depth(struct net_device *dev) 7753 { 7754 struct net_device *ldev; 7755 struct list_head *iter; 7756 u8 max_depth = 0; 7757 bool ignore; 7758 7759 for (iter = &dev->adj_list.lower, 7760 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7761 ldev; 7762 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7763 if (ignore) 7764 continue; 7765 if (max_depth < ldev->lower_level) 7766 max_depth = ldev->lower_level; 7767 } 7768 7769 return max_depth; 7770 } 7771 7772 static int __netdev_update_upper_level(struct net_device *dev, 7773 struct netdev_nested_priv *__unused) 7774 { 7775 dev->upper_level = __netdev_upper_depth(dev) + 1; 7776 return 0; 7777 } 7778 7779 #ifdef CONFIG_LOCKDEP 7780 static LIST_HEAD(net_unlink_list); 7781 7782 static void net_unlink_todo(struct net_device *dev) 7783 { 7784 if (list_empty(&dev->unlink_list)) 7785 list_add_tail(&dev->unlink_list, &net_unlink_list); 7786 } 7787 #endif 7788 7789 static int __netdev_update_lower_level(struct net_device *dev, 7790 struct netdev_nested_priv *priv) 7791 { 7792 dev->lower_level = __netdev_lower_depth(dev) + 1; 7793 7794 #ifdef CONFIG_LOCKDEP 7795 if (!priv) 7796 return 0; 7797 7798 if (priv->flags & NESTED_SYNC_IMM) 7799 dev->nested_level = dev->lower_level - 1; 7800 if (priv->flags & NESTED_SYNC_TODO) 7801 net_unlink_todo(dev); 7802 #endif 7803 return 0; 7804 } 7805 7806 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7807 int (*fn)(struct net_device *dev, 7808 struct netdev_nested_priv *priv), 7809 struct netdev_nested_priv *priv) 7810 { 7811 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7812 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7813 int ret, cur = 0; 7814 7815 now = dev; 7816 iter = &dev->adj_list.lower; 7817 7818 while (1) { 7819 if (now != dev) { 7820 ret = fn(now, priv); 7821 if (ret) 7822 return ret; 7823 } 7824 7825 next = NULL; 7826 while (1) { 7827 ldev = netdev_next_lower_dev_rcu(now, &iter); 7828 if (!ldev) 7829 break; 7830 7831 next = ldev; 7832 niter = &ldev->adj_list.lower; 7833 dev_stack[cur] = now; 7834 iter_stack[cur++] = iter; 7835 break; 7836 } 7837 7838 if (!next) { 7839 if (!cur) 7840 return 0; 7841 next = dev_stack[--cur]; 7842 niter = iter_stack[cur]; 7843 } 7844 7845 now = next; 7846 iter = niter; 7847 } 7848 7849 return 0; 7850 } 7851 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7852 7853 /** 7854 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7855 * lower neighbour list, RCU 7856 * variant 7857 * @dev: device 7858 * 7859 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7860 * list. The caller must hold RCU read lock. 7861 */ 7862 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7863 { 7864 struct netdev_adjacent *lower; 7865 7866 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7867 struct netdev_adjacent, list); 7868 if (lower) 7869 return lower->private; 7870 return NULL; 7871 } 7872 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7873 7874 /** 7875 * netdev_master_upper_dev_get_rcu - Get master upper device 7876 * @dev: device 7877 * 7878 * Find a master upper device and return pointer to it or NULL in case 7879 * it's not there. The caller must hold the RCU read lock. 7880 */ 7881 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7882 { 7883 struct netdev_adjacent *upper; 7884 7885 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7886 struct netdev_adjacent, list); 7887 if (upper && likely(upper->master)) 7888 return upper->dev; 7889 return NULL; 7890 } 7891 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7892 7893 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7894 struct net_device *adj_dev, 7895 struct list_head *dev_list) 7896 { 7897 char linkname[IFNAMSIZ+7]; 7898 7899 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7900 "upper_%s" : "lower_%s", adj_dev->name); 7901 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7902 linkname); 7903 } 7904 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7905 char *name, 7906 struct list_head *dev_list) 7907 { 7908 char linkname[IFNAMSIZ+7]; 7909 7910 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7911 "upper_%s" : "lower_%s", name); 7912 sysfs_remove_link(&(dev->dev.kobj), linkname); 7913 } 7914 7915 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7916 struct net_device *adj_dev, 7917 struct list_head *dev_list) 7918 { 7919 return (dev_list == &dev->adj_list.upper || 7920 dev_list == &dev->adj_list.lower) && 7921 net_eq(dev_net(dev), dev_net(adj_dev)); 7922 } 7923 7924 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7925 struct net_device *adj_dev, 7926 struct list_head *dev_list, 7927 void *private, bool master) 7928 { 7929 struct netdev_adjacent *adj; 7930 int ret; 7931 7932 adj = __netdev_find_adj(adj_dev, dev_list); 7933 7934 if (adj) { 7935 adj->ref_nr += 1; 7936 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7937 dev->name, adj_dev->name, adj->ref_nr); 7938 7939 return 0; 7940 } 7941 7942 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7943 if (!adj) 7944 return -ENOMEM; 7945 7946 adj->dev = adj_dev; 7947 adj->master = master; 7948 adj->ref_nr = 1; 7949 adj->private = private; 7950 adj->ignore = false; 7951 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7952 7953 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7954 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7955 7956 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7957 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7958 if (ret) 7959 goto free_adj; 7960 } 7961 7962 /* Ensure that master link is always the first item in list. */ 7963 if (master) { 7964 ret = sysfs_create_link(&(dev->dev.kobj), 7965 &(adj_dev->dev.kobj), "master"); 7966 if (ret) 7967 goto remove_symlinks; 7968 7969 list_add_rcu(&adj->list, dev_list); 7970 } else { 7971 list_add_tail_rcu(&adj->list, dev_list); 7972 } 7973 7974 return 0; 7975 7976 remove_symlinks: 7977 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7978 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7979 free_adj: 7980 netdev_put(adj_dev, &adj->dev_tracker); 7981 kfree(adj); 7982 7983 return ret; 7984 } 7985 7986 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7987 struct net_device *adj_dev, 7988 u16 ref_nr, 7989 struct list_head *dev_list) 7990 { 7991 struct netdev_adjacent *adj; 7992 7993 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7994 dev->name, adj_dev->name, ref_nr); 7995 7996 adj = __netdev_find_adj(adj_dev, dev_list); 7997 7998 if (!adj) { 7999 pr_err("Adjacency does not exist for device %s from %s\n", 8000 dev->name, adj_dev->name); 8001 WARN_ON(1); 8002 return; 8003 } 8004 8005 if (adj->ref_nr > ref_nr) { 8006 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8007 dev->name, adj_dev->name, ref_nr, 8008 adj->ref_nr - ref_nr); 8009 adj->ref_nr -= ref_nr; 8010 return; 8011 } 8012 8013 if (adj->master) 8014 sysfs_remove_link(&(dev->dev.kobj), "master"); 8015 8016 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8017 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8018 8019 list_del_rcu(&adj->list); 8020 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8021 adj_dev->name, dev->name, adj_dev->name); 8022 netdev_put(adj_dev, &adj->dev_tracker); 8023 kfree_rcu(adj, rcu); 8024 } 8025 8026 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8027 struct net_device *upper_dev, 8028 struct list_head *up_list, 8029 struct list_head *down_list, 8030 void *private, bool master) 8031 { 8032 int ret; 8033 8034 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8035 private, master); 8036 if (ret) 8037 return ret; 8038 8039 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8040 private, false); 8041 if (ret) { 8042 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8043 return ret; 8044 } 8045 8046 return 0; 8047 } 8048 8049 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8050 struct net_device *upper_dev, 8051 u16 ref_nr, 8052 struct list_head *up_list, 8053 struct list_head *down_list) 8054 { 8055 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8056 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8057 } 8058 8059 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8060 struct net_device *upper_dev, 8061 void *private, bool master) 8062 { 8063 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8064 &dev->adj_list.upper, 8065 &upper_dev->adj_list.lower, 8066 private, master); 8067 } 8068 8069 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8070 struct net_device *upper_dev) 8071 { 8072 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8073 &dev->adj_list.upper, 8074 &upper_dev->adj_list.lower); 8075 } 8076 8077 static int __netdev_upper_dev_link(struct net_device *dev, 8078 struct net_device *upper_dev, bool master, 8079 void *upper_priv, void *upper_info, 8080 struct netdev_nested_priv *priv, 8081 struct netlink_ext_ack *extack) 8082 { 8083 struct netdev_notifier_changeupper_info changeupper_info = { 8084 .info = { 8085 .dev = dev, 8086 .extack = extack, 8087 }, 8088 .upper_dev = upper_dev, 8089 .master = master, 8090 .linking = true, 8091 .upper_info = upper_info, 8092 }; 8093 struct net_device *master_dev; 8094 int ret = 0; 8095 8096 ASSERT_RTNL(); 8097 8098 if (dev == upper_dev) 8099 return -EBUSY; 8100 8101 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8102 if (__netdev_has_upper_dev(upper_dev, dev)) 8103 return -EBUSY; 8104 8105 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8106 return -EMLINK; 8107 8108 if (!master) { 8109 if (__netdev_has_upper_dev(dev, upper_dev)) 8110 return -EEXIST; 8111 } else { 8112 master_dev = __netdev_master_upper_dev_get(dev); 8113 if (master_dev) 8114 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8115 } 8116 8117 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8118 &changeupper_info.info); 8119 ret = notifier_to_errno(ret); 8120 if (ret) 8121 return ret; 8122 8123 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8124 master); 8125 if (ret) 8126 return ret; 8127 8128 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8129 &changeupper_info.info); 8130 ret = notifier_to_errno(ret); 8131 if (ret) 8132 goto rollback; 8133 8134 __netdev_update_upper_level(dev, NULL); 8135 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8136 8137 __netdev_update_lower_level(upper_dev, priv); 8138 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8139 priv); 8140 8141 return 0; 8142 8143 rollback: 8144 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8145 8146 return ret; 8147 } 8148 8149 /** 8150 * netdev_upper_dev_link - Add a link to the upper device 8151 * @dev: device 8152 * @upper_dev: new upper device 8153 * @extack: netlink extended ack 8154 * 8155 * Adds a link to device which is upper to this one. The caller must hold 8156 * the RTNL lock. On a failure a negative errno code is returned. 8157 * On success the reference counts are adjusted and the function 8158 * returns zero. 8159 */ 8160 int netdev_upper_dev_link(struct net_device *dev, 8161 struct net_device *upper_dev, 8162 struct netlink_ext_ack *extack) 8163 { 8164 struct netdev_nested_priv priv = { 8165 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8166 .data = NULL, 8167 }; 8168 8169 return __netdev_upper_dev_link(dev, upper_dev, false, 8170 NULL, NULL, &priv, extack); 8171 } 8172 EXPORT_SYMBOL(netdev_upper_dev_link); 8173 8174 /** 8175 * netdev_master_upper_dev_link - Add a master link to the upper device 8176 * @dev: device 8177 * @upper_dev: new upper device 8178 * @upper_priv: upper device private 8179 * @upper_info: upper info to be passed down via notifier 8180 * @extack: netlink extended ack 8181 * 8182 * Adds a link to device which is upper to this one. In this case, only 8183 * one master upper device can be linked, although other non-master devices 8184 * might be linked as well. The caller must hold the RTNL lock. 8185 * On a failure a negative errno code is returned. On success the reference 8186 * counts are adjusted and the function returns zero. 8187 */ 8188 int netdev_master_upper_dev_link(struct net_device *dev, 8189 struct net_device *upper_dev, 8190 void *upper_priv, void *upper_info, 8191 struct netlink_ext_ack *extack) 8192 { 8193 struct netdev_nested_priv priv = { 8194 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8195 .data = NULL, 8196 }; 8197 8198 return __netdev_upper_dev_link(dev, upper_dev, true, 8199 upper_priv, upper_info, &priv, extack); 8200 } 8201 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8202 8203 static void __netdev_upper_dev_unlink(struct net_device *dev, 8204 struct net_device *upper_dev, 8205 struct netdev_nested_priv *priv) 8206 { 8207 struct netdev_notifier_changeupper_info changeupper_info = { 8208 .info = { 8209 .dev = dev, 8210 }, 8211 .upper_dev = upper_dev, 8212 .linking = false, 8213 }; 8214 8215 ASSERT_RTNL(); 8216 8217 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8218 8219 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8220 &changeupper_info.info); 8221 8222 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8223 8224 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8225 &changeupper_info.info); 8226 8227 __netdev_update_upper_level(dev, NULL); 8228 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8229 8230 __netdev_update_lower_level(upper_dev, priv); 8231 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8232 priv); 8233 } 8234 8235 /** 8236 * netdev_upper_dev_unlink - Removes a link to upper device 8237 * @dev: device 8238 * @upper_dev: new upper device 8239 * 8240 * Removes a link to device which is upper to this one. The caller must hold 8241 * the RTNL lock. 8242 */ 8243 void netdev_upper_dev_unlink(struct net_device *dev, 8244 struct net_device *upper_dev) 8245 { 8246 struct netdev_nested_priv priv = { 8247 .flags = NESTED_SYNC_TODO, 8248 .data = NULL, 8249 }; 8250 8251 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8252 } 8253 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8254 8255 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8256 struct net_device *lower_dev, 8257 bool val) 8258 { 8259 struct netdev_adjacent *adj; 8260 8261 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8262 if (adj) 8263 adj->ignore = val; 8264 8265 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8266 if (adj) 8267 adj->ignore = val; 8268 } 8269 8270 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8271 struct net_device *lower_dev) 8272 { 8273 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8274 } 8275 8276 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8277 struct net_device *lower_dev) 8278 { 8279 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8280 } 8281 8282 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8283 struct net_device *new_dev, 8284 struct net_device *dev, 8285 struct netlink_ext_ack *extack) 8286 { 8287 struct netdev_nested_priv priv = { 8288 .flags = 0, 8289 .data = NULL, 8290 }; 8291 int err; 8292 8293 if (!new_dev) 8294 return 0; 8295 8296 if (old_dev && new_dev != old_dev) 8297 netdev_adjacent_dev_disable(dev, old_dev); 8298 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8299 extack); 8300 if (err) { 8301 if (old_dev && new_dev != old_dev) 8302 netdev_adjacent_dev_enable(dev, old_dev); 8303 return err; 8304 } 8305 8306 return 0; 8307 } 8308 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8309 8310 void netdev_adjacent_change_commit(struct net_device *old_dev, 8311 struct net_device *new_dev, 8312 struct net_device *dev) 8313 { 8314 struct netdev_nested_priv priv = { 8315 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8316 .data = NULL, 8317 }; 8318 8319 if (!new_dev || !old_dev) 8320 return; 8321 8322 if (new_dev == old_dev) 8323 return; 8324 8325 netdev_adjacent_dev_enable(dev, old_dev); 8326 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8327 } 8328 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8329 8330 void netdev_adjacent_change_abort(struct net_device *old_dev, 8331 struct net_device *new_dev, 8332 struct net_device *dev) 8333 { 8334 struct netdev_nested_priv priv = { 8335 .flags = 0, 8336 .data = NULL, 8337 }; 8338 8339 if (!new_dev) 8340 return; 8341 8342 if (old_dev && new_dev != old_dev) 8343 netdev_adjacent_dev_enable(dev, old_dev); 8344 8345 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8346 } 8347 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8348 8349 /** 8350 * netdev_bonding_info_change - Dispatch event about slave change 8351 * @dev: device 8352 * @bonding_info: info to dispatch 8353 * 8354 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8355 * The caller must hold the RTNL lock. 8356 */ 8357 void netdev_bonding_info_change(struct net_device *dev, 8358 struct netdev_bonding_info *bonding_info) 8359 { 8360 struct netdev_notifier_bonding_info info = { 8361 .info.dev = dev, 8362 }; 8363 8364 memcpy(&info.bonding_info, bonding_info, 8365 sizeof(struct netdev_bonding_info)); 8366 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8367 &info.info); 8368 } 8369 EXPORT_SYMBOL(netdev_bonding_info_change); 8370 8371 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8372 struct netlink_ext_ack *extack) 8373 { 8374 struct netdev_notifier_offload_xstats_info info = { 8375 .info.dev = dev, 8376 .info.extack = extack, 8377 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8378 }; 8379 int err; 8380 int rc; 8381 8382 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8383 GFP_KERNEL); 8384 if (!dev->offload_xstats_l3) 8385 return -ENOMEM; 8386 8387 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8388 NETDEV_OFFLOAD_XSTATS_DISABLE, 8389 &info.info); 8390 err = notifier_to_errno(rc); 8391 if (err) 8392 goto free_stats; 8393 8394 return 0; 8395 8396 free_stats: 8397 kfree(dev->offload_xstats_l3); 8398 dev->offload_xstats_l3 = NULL; 8399 return err; 8400 } 8401 8402 int netdev_offload_xstats_enable(struct net_device *dev, 8403 enum netdev_offload_xstats_type type, 8404 struct netlink_ext_ack *extack) 8405 { 8406 ASSERT_RTNL(); 8407 8408 if (netdev_offload_xstats_enabled(dev, type)) 8409 return -EALREADY; 8410 8411 switch (type) { 8412 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8413 return netdev_offload_xstats_enable_l3(dev, extack); 8414 } 8415 8416 WARN_ON(1); 8417 return -EINVAL; 8418 } 8419 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8420 8421 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8422 { 8423 struct netdev_notifier_offload_xstats_info info = { 8424 .info.dev = dev, 8425 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8426 }; 8427 8428 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8429 &info.info); 8430 kfree(dev->offload_xstats_l3); 8431 dev->offload_xstats_l3 = NULL; 8432 } 8433 8434 int netdev_offload_xstats_disable(struct net_device *dev, 8435 enum netdev_offload_xstats_type type) 8436 { 8437 ASSERT_RTNL(); 8438 8439 if (!netdev_offload_xstats_enabled(dev, type)) 8440 return -EALREADY; 8441 8442 switch (type) { 8443 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8444 netdev_offload_xstats_disable_l3(dev); 8445 return 0; 8446 } 8447 8448 WARN_ON(1); 8449 return -EINVAL; 8450 } 8451 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8452 8453 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8454 { 8455 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8456 } 8457 8458 static struct rtnl_hw_stats64 * 8459 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8460 enum netdev_offload_xstats_type type) 8461 { 8462 switch (type) { 8463 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8464 return dev->offload_xstats_l3; 8465 } 8466 8467 WARN_ON(1); 8468 return NULL; 8469 } 8470 8471 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8472 enum netdev_offload_xstats_type type) 8473 { 8474 ASSERT_RTNL(); 8475 8476 return netdev_offload_xstats_get_ptr(dev, type); 8477 } 8478 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8479 8480 struct netdev_notifier_offload_xstats_ru { 8481 bool used; 8482 }; 8483 8484 struct netdev_notifier_offload_xstats_rd { 8485 struct rtnl_hw_stats64 stats; 8486 bool used; 8487 }; 8488 8489 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8490 const struct rtnl_hw_stats64 *src) 8491 { 8492 dest->rx_packets += src->rx_packets; 8493 dest->tx_packets += src->tx_packets; 8494 dest->rx_bytes += src->rx_bytes; 8495 dest->tx_bytes += src->tx_bytes; 8496 dest->rx_errors += src->rx_errors; 8497 dest->tx_errors += src->tx_errors; 8498 dest->rx_dropped += src->rx_dropped; 8499 dest->tx_dropped += src->tx_dropped; 8500 dest->multicast += src->multicast; 8501 } 8502 8503 static int netdev_offload_xstats_get_used(struct net_device *dev, 8504 enum netdev_offload_xstats_type type, 8505 bool *p_used, 8506 struct netlink_ext_ack *extack) 8507 { 8508 struct netdev_notifier_offload_xstats_ru report_used = {}; 8509 struct netdev_notifier_offload_xstats_info info = { 8510 .info.dev = dev, 8511 .info.extack = extack, 8512 .type = type, 8513 .report_used = &report_used, 8514 }; 8515 int rc; 8516 8517 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8518 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8519 &info.info); 8520 *p_used = report_used.used; 8521 return notifier_to_errno(rc); 8522 } 8523 8524 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8525 enum netdev_offload_xstats_type type, 8526 struct rtnl_hw_stats64 *p_stats, 8527 bool *p_used, 8528 struct netlink_ext_ack *extack) 8529 { 8530 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8531 struct netdev_notifier_offload_xstats_info info = { 8532 .info.dev = dev, 8533 .info.extack = extack, 8534 .type = type, 8535 .report_delta = &report_delta, 8536 }; 8537 struct rtnl_hw_stats64 *stats; 8538 int rc; 8539 8540 stats = netdev_offload_xstats_get_ptr(dev, type); 8541 if (WARN_ON(!stats)) 8542 return -EINVAL; 8543 8544 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8545 &info.info); 8546 8547 /* Cache whatever we got, even if there was an error, otherwise the 8548 * successful stats retrievals would get lost. 8549 */ 8550 netdev_hw_stats64_add(stats, &report_delta.stats); 8551 8552 if (p_stats) 8553 *p_stats = *stats; 8554 *p_used = report_delta.used; 8555 8556 return notifier_to_errno(rc); 8557 } 8558 8559 int netdev_offload_xstats_get(struct net_device *dev, 8560 enum netdev_offload_xstats_type type, 8561 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8562 struct netlink_ext_ack *extack) 8563 { 8564 ASSERT_RTNL(); 8565 8566 if (p_stats) 8567 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8568 p_used, extack); 8569 else 8570 return netdev_offload_xstats_get_used(dev, type, p_used, 8571 extack); 8572 } 8573 EXPORT_SYMBOL(netdev_offload_xstats_get); 8574 8575 void 8576 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8577 const struct rtnl_hw_stats64 *stats) 8578 { 8579 report_delta->used = true; 8580 netdev_hw_stats64_add(&report_delta->stats, stats); 8581 } 8582 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8583 8584 void 8585 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8586 { 8587 report_used->used = true; 8588 } 8589 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8590 8591 void netdev_offload_xstats_push_delta(struct net_device *dev, 8592 enum netdev_offload_xstats_type type, 8593 const struct rtnl_hw_stats64 *p_stats) 8594 { 8595 struct rtnl_hw_stats64 *stats; 8596 8597 ASSERT_RTNL(); 8598 8599 stats = netdev_offload_xstats_get_ptr(dev, type); 8600 if (WARN_ON(!stats)) 8601 return; 8602 8603 netdev_hw_stats64_add(stats, p_stats); 8604 } 8605 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8606 8607 /** 8608 * netdev_get_xmit_slave - Get the xmit slave of master device 8609 * @dev: device 8610 * @skb: The packet 8611 * @all_slaves: assume all the slaves are active 8612 * 8613 * The reference counters are not incremented so the caller must be 8614 * careful with locks. The caller must hold RCU lock. 8615 * %NULL is returned if no slave is found. 8616 */ 8617 8618 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8619 struct sk_buff *skb, 8620 bool all_slaves) 8621 { 8622 const struct net_device_ops *ops = dev->netdev_ops; 8623 8624 if (!ops->ndo_get_xmit_slave) 8625 return NULL; 8626 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8627 } 8628 EXPORT_SYMBOL(netdev_get_xmit_slave); 8629 8630 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8631 struct sock *sk) 8632 { 8633 const struct net_device_ops *ops = dev->netdev_ops; 8634 8635 if (!ops->ndo_sk_get_lower_dev) 8636 return NULL; 8637 return ops->ndo_sk_get_lower_dev(dev, sk); 8638 } 8639 8640 /** 8641 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8642 * @dev: device 8643 * @sk: the socket 8644 * 8645 * %NULL is returned if no lower device is found. 8646 */ 8647 8648 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8649 struct sock *sk) 8650 { 8651 struct net_device *lower; 8652 8653 lower = netdev_sk_get_lower_dev(dev, sk); 8654 while (lower) { 8655 dev = lower; 8656 lower = netdev_sk_get_lower_dev(dev, sk); 8657 } 8658 8659 return dev; 8660 } 8661 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8662 8663 static void netdev_adjacent_add_links(struct net_device *dev) 8664 { 8665 struct netdev_adjacent *iter; 8666 8667 struct net *net = dev_net(dev); 8668 8669 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8670 if (!net_eq(net, dev_net(iter->dev))) 8671 continue; 8672 netdev_adjacent_sysfs_add(iter->dev, dev, 8673 &iter->dev->adj_list.lower); 8674 netdev_adjacent_sysfs_add(dev, iter->dev, 8675 &dev->adj_list.upper); 8676 } 8677 8678 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8679 if (!net_eq(net, dev_net(iter->dev))) 8680 continue; 8681 netdev_adjacent_sysfs_add(iter->dev, dev, 8682 &iter->dev->adj_list.upper); 8683 netdev_adjacent_sysfs_add(dev, iter->dev, 8684 &dev->adj_list.lower); 8685 } 8686 } 8687 8688 static void netdev_adjacent_del_links(struct net_device *dev) 8689 { 8690 struct netdev_adjacent *iter; 8691 8692 struct net *net = dev_net(dev); 8693 8694 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8695 if (!net_eq(net, dev_net(iter->dev))) 8696 continue; 8697 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8698 &iter->dev->adj_list.lower); 8699 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8700 &dev->adj_list.upper); 8701 } 8702 8703 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8704 if (!net_eq(net, dev_net(iter->dev))) 8705 continue; 8706 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8707 &iter->dev->adj_list.upper); 8708 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8709 &dev->adj_list.lower); 8710 } 8711 } 8712 8713 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8714 { 8715 struct netdev_adjacent *iter; 8716 8717 struct net *net = dev_net(dev); 8718 8719 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8720 if (!net_eq(net, dev_net(iter->dev))) 8721 continue; 8722 netdev_adjacent_sysfs_del(iter->dev, oldname, 8723 &iter->dev->adj_list.lower); 8724 netdev_adjacent_sysfs_add(iter->dev, dev, 8725 &iter->dev->adj_list.lower); 8726 } 8727 8728 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8729 if (!net_eq(net, dev_net(iter->dev))) 8730 continue; 8731 netdev_adjacent_sysfs_del(iter->dev, oldname, 8732 &iter->dev->adj_list.upper); 8733 netdev_adjacent_sysfs_add(iter->dev, dev, 8734 &iter->dev->adj_list.upper); 8735 } 8736 } 8737 8738 void *netdev_lower_dev_get_private(struct net_device *dev, 8739 struct net_device *lower_dev) 8740 { 8741 struct netdev_adjacent *lower; 8742 8743 if (!lower_dev) 8744 return NULL; 8745 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8746 if (!lower) 8747 return NULL; 8748 8749 return lower->private; 8750 } 8751 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8752 8753 8754 /** 8755 * netdev_lower_state_changed - Dispatch event about lower device state change 8756 * @lower_dev: device 8757 * @lower_state_info: state to dispatch 8758 * 8759 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8760 * The caller must hold the RTNL lock. 8761 */ 8762 void netdev_lower_state_changed(struct net_device *lower_dev, 8763 void *lower_state_info) 8764 { 8765 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8766 .info.dev = lower_dev, 8767 }; 8768 8769 ASSERT_RTNL(); 8770 changelowerstate_info.lower_state_info = lower_state_info; 8771 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8772 &changelowerstate_info.info); 8773 } 8774 EXPORT_SYMBOL(netdev_lower_state_changed); 8775 8776 static void dev_change_rx_flags(struct net_device *dev, int flags) 8777 { 8778 const struct net_device_ops *ops = dev->netdev_ops; 8779 8780 if (ops->ndo_change_rx_flags) 8781 ops->ndo_change_rx_flags(dev, flags); 8782 } 8783 8784 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8785 { 8786 unsigned int old_flags = dev->flags; 8787 unsigned int promiscuity, flags; 8788 kuid_t uid; 8789 kgid_t gid; 8790 8791 ASSERT_RTNL(); 8792 8793 promiscuity = dev->promiscuity + inc; 8794 if (promiscuity == 0) { 8795 /* 8796 * Avoid overflow. 8797 * If inc causes overflow, untouch promisc and return error. 8798 */ 8799 if (unlikely(inc > 0)) { 8800 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8801 return -EOVERFLOW; 8802 } 8803 flags = old_flags & ~IFF_PROMISC; 8804 } else { 8805 flags = old_flags | IFF_PROMISC; 8806 } 8807 WRITE_ONCE(dev->promiscuity, promiscuity); 8808 if (flags != old_flags) { 8809 WRITE_ONCE(dev->flags, flags); 8810 netdev_info(dev, "%s promiscuous mode\n", 8811 dev->flags & IFF_PROMISC ? "entered" : "left"); 8812 if (audit_enabled) { 8813 current_uid_gid(&uid, &gid); 8814 audit_log(audit_context(), GFP_ATOMIC, 8815 AUDIT_ANOM_PROMISCUOUS, 8816 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8817 dev->name, (dev->flags & IFF_PROMISC), 8818 (old_flags & IFF_PROMISC), 8819 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8820 from_kuid(&init_user_ns, uid), 8821 from_kgid(&init_user_ns, gid), 8822 audit_get_sessionid(current)); 8823 } 8824 8825 dev_change_rx_flags(dev, IFF_PROMISC); 8826 } 8827 if (notify) 8828 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8829 return 0; 8830 } 8831 8832 /** 8833 * dev_set_promiscuity - update promiscuity count on a device 8834 * @dev: device 8835 * @inc: modifier 8836 * 8837 * Add or remove promiscuity from a device. While the count in the device 8838 * remains above zero the interface remains promiscuous. Once it hits zero 8839 * the device reverts back to normal filtering operation. A negative inc 8840 * value is used to drop promiscuity on the device. 8841 * Return 0 if successful or a negative errno code on error. 8842 */ 8843 int dev_set_promiscuity(struct net_device *dev, int inc) 8844 { 8845 unsigned int old_flags = dev->flags; 8846 int err; 8847 8848 err = __dev_set_promiscuity(dev, inc, true); 8849 if (err < 0) 8850 return err; 8851 if (dev->flags != old_flags) 8852 dev_set_rx_mode(dev); 8853 return err; 8854 } 8855 EXPORT_SYMBOL(dev_set_promiscuity); 8856 8857 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8858 { 8859 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8860 unsigned int allmulti, flags; 8861 8862 ASSERT_RTNL(); 8863 8864 allmulti = dev->allmulti + inc; 8865 if (allmulti == 0) { 8866 /* 8867 * Avoid overflow. 8868 * If inc causes overflow, untouch allmulti and return error. 8869 */ 8870 if (unlikely(inc > 0)) { 8871 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8872 return -EOVERFLOW; 8873 } 8874 flags = old_flags & ~IFF_ALLMULTI; 8875 } else { 8876 flags = old_flags | IFF_ALLMULTI; 8877 } 8878 WRITE_ONCE(dev->allmulti, allmulti); 8879 if (flags != old_flags) { 8880 WRITE_ONCE(dev->flags, flags); 8881 netdev_info(dev, "%s allmulticast mode\n", 8882 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8883 dev_change_rx_flags(dev, IFF_ALLMULTI); 8884 dev_set_rx_mode(dev); 8885 if (notify) 8886 __dev_notify_flags(dev, old_flags, 8887 dev->gflags ^ old_gflags, 0, NULL); 8888 } 8889 return 0; 8890 } 8891 8892 /** 8893 * dev_set_allmulti - update allmulti count on a device 8894 * @dev: device 8895 * @inc: modifier 8896 * 8897 * Add or remove reception of all multicast frames to a device. While the 8898 * count in the device remains above zero the interface remains listening 8899 * to all interfaces. Once it hits zero the device reverts back to normal 8900 * filtering operation. A negative @inc value is used to drop the counter 8901 * when releasing a resource needing all multicasts. 8902 * Return 0 if successful or a negative errno code on error. 8903 */ 8904 8905 int dev_set_allmulti(struct net_device *dev, int inc) 8906 { 8907 return __dev_set_allmulti(dev, inc, true); 8908 } 8909 EXPORT_SYMBOL(dev_set_allmulti); 8910 8911 /* 8912 * Upload unicast and multicast address lists to device and 8913 * configure RX filtering. When the device doesn't support unicast 8914 * filtering it is put in promiscuous mode while unicast addresses 8915 * are present. 8916 */ 8917 void __dev_set_rx_mode(struct net_device *dev) 8918 { 8919 const struct net_device_ops *ops = dev->netdev_ops; 8920 8921 /* dev_open will call this function so the list will stay sane. */ 8922 if (!(dev->flags&IFF_UP)) 8923 return; 8924 8925 if (!netif_device_present(dev)) 8926 return; 8927 8928 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8929 /* Unicast addresses changes may only happen under the rtnl, 8930 * therefore calling __dev_set_promiscuity here is safe. 8931 */ 8932 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8933 __dev_set_promiscuity(dev, 1, false); 8934 dev->uc_promisc = true; 8935 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8936 __dev_set_promiscuity(dev, -1, false); 8937 dev->uc_promisc = false; 8938 } 8939 } 8940 8941 if (ops->ndo_set_rx_mode) 8942 ops->ndo_set_rx_mode(dev); 8943 } 8944 8945 void dev_set_rx_mode(struct net_device *dev) 8946 { 8947 netif_addr_lock_bh(dev); 8948 __dev_set_rx_mode(dev); 8949 netif_addr_unlock_bh(dev); 8950 } 8951 8952 /** 8953 * dev_get_flags - get flags reported to userspace 8954 * @dev: device 8955 * 8956 * Get the combination of flag bits exported through APIs to userspace. 8957 */ 8958 unsigned int dev_get_flags(const struct net_device *dev) 8959 { 8960 unsigned int flags; 8961 8962 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | 8963 IFF_ALLMULTI | 8964 IFF_RUNNING | 8965 IFF_LOWER_UP | 8966 IFF_DORMANT)) | 8967 (READ_ONCE(dev->gflags) & (IFF_PROMISC | 8968 IFF_ALLMULTI)); 8969 8970 if (netif_running(dev)) { 8971 if (netif_oper_up(dev)) 8972 flags |= IFF_RUNNING; 8973 if (netif_carrier_ok(dev)) 8974 flags |= IFF_LOWER_UP; 8975 if (netif_dormant(dev)) 8976 flags |= IFF_DORMANT; 8977 } 8978 8979 return flags; 8980 } 8981 EXPORT_SYMBOL(dev_get_flags); 8982 8983 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8984 struct netlink_ext_ack *extack) 8985 { 8986 unsigned int old_flags = dev->flags; 8987 int ret; 8988 8989 ASSERT_RTNL(); 8990 8991 /* 8992 * Set the flags on our device. 8993 */ 8994 8995 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8996 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8997 IFF_AUTOMEDIA)) | 8998 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8999 IFF_ALLMULTI)); 9000 9001 /* 9002 * Load in the correct multicast list now the flags have changed. 9003 */ 9004 9005 if ((old_flags ^ flags) & IFF_MULTICAST) 9006 dev_change_rx_flags(dev, IFF_MULTICAST); 9007 9008 dev_set_rx_mode(dev); 9009 9010 /* 9011 * Have we downed the interface. We handle IFF_UP ourselves 9012 * according to user attempts to set it, rather than blindly 9013 * setting it. 9014 */ 9015 9016 ret = 0; 9017 if ((old_flags ^ flags) & IFF_UP) { 9018 if (old_flags & IFF_UP) 9019 __dev_close(dev); 9020 else 9021 ret = __dev_open(dev, extack); 9022 } 9023 9024 if ((flags ^ dev->gflags) & IFF_PROMISC) { 9025 int inc = (flags & IFF_PROMISC) ? 1 : -1; 9026 unsigned int old_flags = dev->flags; 9027 9028 dev->gflags ^= IFF_PROMISC; 9029 9030 if (__dev_set_promiscuity(dev, inc, false) >= 0) 9031 if (dev->flags != old_flags) 9032 dev_set_rx_mode(dev); 9033 } 9034 9035 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 9036 * is important. Some (broken) drivers set IFF_PROMISC, when 9037 * IFF_ALLMULTI is requested not asking us and not reporting. 9038 */ 9039 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 9040 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 9041 9042 dev->gflags ^= IFF_ALLMULTI; 9043 __dev_set_allmulti(dev, inc, false); 9044 } 9045 9046 return ret; 9047 } 9048 9049 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 9050 unsigned int gchanges, u32 portid, 9051 const struct nlmsghdr *nlh) 9052 { 9053 unsigned int changes = dev->flags ^ old_flags; 9054 9055 if (gchanges) 9056 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 9057 9058 if (changes & IFF_UP) { 9059 if (dev->flags & IFF_UP) 9060 call_netdevice_notifiers(NETDEV_UP, dev); 9061 else 9062 call_netdevice_notifiers(NETDEV_DOWN, dev); 9063 } 9064 9065 if (dev->flags & IFF_UP && 9066 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 9067 struct netdev_notifier_change_info change_info = { 9068 .info = { 9069 .dev = dev, 9070 }, 9071 .flags_changed = changes, 9072 }; 9073 9074 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 9075 } 9076 } 9077 9078 /** 9079 * dev_change_flags - change device settings 9080 * @dev: device 9081 * @flags: device state flags 9082 * @extack: netlink extended ack 9083 * 9084 * Change settings on device based state flags. The flags are 9085 * in the userspace exported format. 9086 */ 9087 int dev_change_flags(struct net_device *dev, unsigned int flags, 9088 struct netlink_ext_ack *extack) 9089 { 9090 int ret; 9091 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 9092 9093 ret = __dev_change_flags(dev, flags, extack); 9094 if (ret < 0) 9095 return ret; 9096 9097 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 9098 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 9099 return ret; 9100 } 9101 EXPORT_SYMBOL(dev_change_flags); 9102 9103 int __dev_set_mtu(struct net_device *dev, int new_mtu) 9104 { 9105 const struct net_device_ops *ops = dev->netdev_ops; 9106 9107 if (ops->ndo_change_mtu) 9108 return ops->ndo_change_mtu(dev, new_mtu); 9109 9110 /* Pairs with all the lockless reads of dev->mtu in the stack */ 9111 WRITE_ONCE(dev->mtu, new_mtu); 9112 return 0; 9113 } 9114 EXPORT_SYMBOL(__dev_set_mtu); 9115 9116 int dev_validate_mtu(struct net_device *dev, int new_mtu, 9117 struct netlink_ext_ack *extack) 9118 { 9119 /* MTU must be positive, and in range */ 9120 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 9121 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 9122 return -EINVAL; 9123 } 9124 9125 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 9126 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 9127 return -EINVAL; 9128 } 9129 return 0; 9130 } 9131 9132 /** 9133 * dev_set_mtu_ext - Change maximum transfer unit 9134 * @dev: device 9135 * @new_mtu: new transfer unit 9136 * @extack: netlink extended ack 9137 * 9138 * Change the maximum transfer size of the network device. 9139 */ 9140 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 9141 struct netlink_ext_ack *extack) 9142 { 9143 int err, orig_mtu; 9144 9145 if (new_mtu == dev->mtu) 9146 return 0; 9147 9148 err = dev_validate_mtu(dev, new_mtu, extack); 9149 if (err) 9150 return err; 9151 9152 if (!netif_device_present(dev)) 9153 return -ENODEV; 9154 9155 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 9156 err = notifier_to_errno(err); 9157 if (err) 9158 return err; 9159 9160 orig_mtu = dev->mtu; 9161 err = __dev_set_mtu(dev, new_mtu); 9162 9163 if (!err) { 9164 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9165 orig_mtu); 9166 err = notifier_to_errno(err); 9167 if (err) { 9168 /* setting mtu back and notifying everyone again, 9169 * so that they have a chance to revert changes. 9170 */ 9171 __dev_set_mtu(dev, orig_mtu); 9172 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9173 new_mtu); 9174 } 9175 } 9176 return err; 9177 } 9178 9179 int dev_set_mtu(struct net_device *dev, int new_mtu) 9180 { 9181 struct netlink_ext_ack extack; 9182 int err; 9183 9184 memset(&extack, 0, sizeof(extack)); 9185 err = dev_set_mtu_ext(dev, new_mtu, &extack); 9186 if (err && extack._msg) 9187 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 9188 return err; 9189 } 9190 EXPORT_SYMBOL(dev_set_mtu); 9191 9192 /** 9193 * dev_change_tx_queue_len - Change TX queue length of a netdevice 9194 * @dev: device 9195 * @new_len: new tx queue length 9196 */ 9197 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 9198 { 9199 unsigned int orig_len = dev->tx_queue_len; 9200 int res; 9201 9202 if (new_len != (unsigned int)new_len) 9203 return -ERANGE; 9204 9205 if (new_len != orig_len) { 9206 WRITE_ONCE(dev->tx_queue_len, new_len); 9207 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 9208 res = notifier_to_errno(res); 9209 if (res) 9210 goto err_rollback; 9211 res = dev_qdisc_change_tx_queue_len(dev); 9212 if (res) 9213 goto err_rollback; 9214 } 9215 9216 return 0; 9217 9218 err_rollback: 9219 netdev_err(dev, "refused to change device tx_queue_len\n"); 9220 WRITE_ONCE(dev->tx_queue_len, orig_len); 9221 return res; 9222 } 9223 9224 /** 9225 * dev_set_group - Change group this device belongs to 9226 * @dev: device 9227 * @new_group: group this device should belong to 9228 */ 9229 void dev_set_group(struct net_device *dev, int new_group) 9230 { 9231 dev->group = new_group; 9232 } 9233 9234 /** 9235 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 9236 * @dev: device 9237 * @addr: new address 9238 * @extack: netlink extended ack 9239 */ 9240 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9241 struct netlink_ext_ack *extack) 9242 { 9243 struct netdev_notifier_pre_changeaddr_info info = { 9244 .info.dev = dev, 9245 .info.extack = extack, 9246 .dev_addr = addr, 9247 }; 9248 int rc; 9249 9250 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9251 return notifier_to_errno(rc); 9252 } 9253 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9254 9255 /** 9256 * dev_set_mac_address - Change Media Access Control Address 9257 * @dev: device 9258 * @sa: new address 9259 * @extack: netlink extended ack 9260 * 9261 * Change the hardware (MAC) address of the device 9262 */ 9263 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9264 struct netlink_ext_ack *extack) 9265 { 9266 const struct net_device_ops *ops = dev->netdev_ops; 9267 int err; 9268 9269 if (!ops->ndo_set_mac_address) 9270 return -EOPNOTSUPP; 9271 if (sa->sa_family != dev->type) 9272 return -EINVAL; 9273 if (!netif_device_present(dev)) 9274 return -ENODEV; 9275 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9276 if (err) 9277 return err; 9278 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 9279 err = ops->ndo_set_mac_address(dev, sa); 9280 if (err) 9281 return err; 9282 } 9283 dev->addr_assign_type = NET_ADDR_SET; 9284 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9285 add_device_randomness(dev->dev_addr, dev->addr_len); 9286 return 0; 9287 } 9288 EXPORT_SYMBOL(dev_set_mac_address); 9289 9290 DECLARE_RWSEM(dev_addr_sem); 9291 9292 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9293 struct netlink_ext_ack *extack) 9294 { 9295 int ret; 9296 9297 down_write(&dev_addr_sem); 9298 ret = dev_set_mac_address(dev, sa, extack); 9299 up_write(&dev_addr_sem); 9300 return ret; 9301 } 9302 EXPORT_SYMBOL(dev_set_mac_address_user); 9303 9304 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9305 { 9306 size_t size = sizeof(sa->sa_data_min); 9307 struct net_device *dev; 9308 int ret = 0; 9309 9310 down_read(&dev_addr_sem); 9311 rcu_read_lock(); 9312 9313 dev = dev_get_by_name_rcu(net, dev_name); 9314 if (!dev) { 9315 ret = -ENODEV; 9316 goto unlock; 9317 } 9318 if (!dev->addr_len) 9319 memset(sa->sa_data, 0, size); 9320 else 9321 memcpy(sa->sa_data, dev->dev_addr, 9322 min_t(size_t, size, dev->addr_len)); 9323 sa->sa_family = dev->type; 9324 9325 unlock: 9326 rcu_read_unlock(); 9327 up_read(&dev_addr_sem); 9328 return ret; 9329 } 9330 EXPORT_SYMBOL(dev_get_mac_address); 9331 9332 /** 9333 * dev_change_carrier - Change device carrier 9334 * @dev: device 9335 * @new_carrier: new value 9336 * 9337 * Change device carrier 9338 */ 9339 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9340 { 9341 const struct net_device_ops *ops = dev->netdev_ops; 9342 9343 if (!ops->ndo_change_carrier) 9344 return -EOPNOTSUPP; 9345 if (!netif_device_present(dev)) 9346 return -ENODEV; 9347 return ops->ndo_change_carrier(dev, new_carrier); 9348 } 9349 9350 /** 9351 * dev_get_phys_port_id - Get device physical port ID 9352 * @dev: device 9353 * @ppid: port ID 9354 * 9355 * Get device physical port ID 9356 */ 9357 int dev_get_phys_port_id(struct net_device *dev, 9358 struct netdev_phys_item_id *ppid) 9359 { 9360 const struct net_device_ops *ops = dev->netdev_ops; 9361 9362 if (!ops->ndo_get_phys_port_id) 9363 return -EOPNOTSUPP; 9364 return ops->ndo_get_phys_port_id(dev, ppid); 9365 } 9366 9367 /** 9368 * dev_get_phys_port_name - Get device physical port name 9369 * @dev: device 9370 * @name: port name 9371 * @len: limit of bytes to copy to name 9372 * 9373 * Get device physical port name 9374 */ 9375 int dev_get_phys_port_name(struct net_device *dev, 9376 char *name, size_t len) 9377 { 9378 const struct net_device_ops *ops = dev->netdev_ops; 9379 int err; 9380 9381 if (ops->ndo_get_phys_port_name) { 9382 err = ops->ndo_get_phys_port_name(dev, name, len); 9383 if (err != -EOPNOTSUPP) 9384 return err; 9385 } 9386 return devlink_compat_phys_port_name_get(dev, name, len); 9387 } 9388 9389 /** 9390 * dev_get_port_parent_id - Get the device's port parent identifier 9391 * @dev: network device 9392 * @ppid: pointer to a storage for the port's parent identifier 9393 * @recurse: allow/disallow recursion to lower devices 9394 * 9395 * Get the devices's port parent identifier 9396 */ 9397 int dev_get_port_parent_id(struct net_device *dev, 9398 struct netdev_phys_item_id *ppid, 9399 bool recurse) 9400 { 9401 const struct net_device_ops *ops = dev->netdev_ops; 9402 struct netdev_phys_item_id first = { }; 9403 struct net_device *lower_dev; 9404 struct list_head *iter; 9405 int err; 9406 9407 if (ops->ndo_get_port_parent_id) { 9408 err = ops->ndo_get_port_parent_id(dev, ppid); 9409 if (err != -EOPNOTSUPP) 9410 return err; 9411 } 9412 9413 err = devlink_compat_switch_id_get(dev, ppid); 9414 if (!recurse || err != -EOPNOTSUPP) 9415 return err; 9416 9417 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9418 err = dev_get_port_parent_id(lower_dev, ppid, true); 9419 if (err) 9420 break; 9421 if (!first.id_len) 9422 first = *ppid; 9423 else if (memcmp(&first, ppid, sizeof(*ppid))) 9424 return -EOPNOTSUPP; 9425 } 9426 9427 return err; 9428 } 9429 EXPORT_SYMBOL(dev_get_port_parent_id); 9430 9431 /** 9432 * netdev_port_same_parent_id - Indicate if two network devices have 9433 * the same port parent identifier 9434 * @a: first network device 9435 * @b: second network device 9436 */ 9437 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9438 { 9439 struct netdev_phys_item_id a_id = { }; 9440 struct netdev_phys_item_id b_id = { }; 9441 9442 if (dev_get_port_parent_id(a, &a_id, true) || 9443 dev_get_port_parent_id(b, &b_id, true)) 9444 return false; 9445 9446 return netdev_phys_item_id_same(&a_id, &b_id); 9447 } 9448 EXPORT_SYMBOL(netdev_port_same_parent_id); 9449 9450 /** 9451 * dev_change_proto_down - set carrier according to proto_down. 9452 * 9453 * @dev: device 9454 * @proto_down: new value 9455 */ 9456 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9457 { 9458 if (!dev->change_proto_down) 9459 return -EOPNOTSUPP; 9460 if (!netif_device_present(dev)) 9461 return -ENODEV; 9462 if (proto_down) 9463 netif_carrier_off(dev); 9464 else 9465 netif_carrier_on(dev); 9466 WRITE_ONCE(dev->proto_down, proto_down); 9467 return 0; 9468 } 9469 9470 /** 9471 * dev_change_proto_down_reason - proto down reason 9472 * 9473 * @dev: device 9474 * @mask: proto down mask 9475 * @value: proto down value 9476 */ 9477 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9478 u32 value) 9479 { 9480 u32 proto_down_reason; 9481 int b; 9482 9483 if (!mask) { 9484 proto_down_reason = value; 9485 } else { 9486 proto_down_reason = dev->proto_down_reason; 9487 for_each_set_bit(b, &mask, 32) { 9488 if (value & (1 << b)) 9489 proto_down_reason |= BIT(b); 9490 else 9491 proto_down_reason &= ~BIT(b); 9492 } 9493 } 9494 WRITE_ONCE(dev->proto_down_reason, proto_down_reason); 9495 } 9496 9497 struct bpf_xdp_link { 9498 struct bpf_link link; 9499 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9500 int flags; 9501 }; 9502 9503 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9504 { 9505 if (flags & XDP_FLAGS_HW_MODE) 9506 return XDP_MODE_HW; 9507 if (flags & XDP_FLAGS_DRV_MODE) 9508 return XDP_MODE_DRV; 9509 if (flags & XDP_FLAGS_SKB_MODE) 9510 return XDP_MODE_SKB; 9511 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9512 } 9513 9514 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9515 { 9516 switch (mode) { 9517 case XDP_MODE_SKB: 9518 return generic_xdp_install; 9519 case XDP_MODE_DRV: 9520 case XDP_MODE_HW: 9521 return dev->netdev_ops->ndo_bpf; 9522 default: 9523 return NULL; 9524 } 9525 } 9526 9527 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9528 enum bpf_xdp_mode mode) 9529 { 9530 return dev->xdp_state[mode].link; 9531 } 9532 9533 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9534 enum bpf_xdp_mode mode) 9535 { 9536 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9537 9538 if (link) 9539 return link->link.prog; 9540 return dev->xdp_state[mode].prog; 9541 } 9542 9543 u8 dev_xdp_prog_count(struct net_device *dev) 9544 { 9545 u8 count = 0; 9546 int i; 9547 9548 for (i = 0; i < __MAX_XDP_MODE; i++) 9549 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9550 count++; 9551 return count; 9552 } 9553 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9554 9555 int dev_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf) 9556 { 9557 if (!dev->netdev_ops->ndo_bpf) 9558 return -EOPNOTSUPP; 9559 9560 if (dev_get_min_mp_channel_count(dev)) { 9561 NL_SET_ERR_MSG(bpf->extack, "unable to propagate XDP to device using memory provider"); 9562 return -EBUSY; 9563 } 9564 9565 return dev->netdev_ops->ndo_bpf(dev, bpf); 9566 } 9567 EXPORT_SYMBOL_GPL(dev_xdp_propagate); 9568 9569 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9570 { 9571 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9572 9573 return prog ? prog->aux->id : 0; 9574 } 9575 9576 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9577 struct bpf_xdp_link *link) 9578 { 9579 dev->xdp_state[mode].link = link; 9580 dev->xdp_state[mode].prog = NULL; 9581 } 9582 9583 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9584 struct bpf_prog *prog) 9585 { 9586 dev->xdp_state[mode].link = NULL; 9587 dev->xdp_state[mode].prog = prog; 9588 } 9589 9590 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9591 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9592 u32 flags, struct bpf_prog *prog) 9593 { 9594 struct netdev_bpf xdp; 9595 int err; 9596 9597 if (dev_get_min_mp_channel_count(dev)) { 9598 NL_SET_ERR_MSG(extack, "unable to install XDP to device using memory provider"); 9599 return -EBUSY; 9600 } 9601 9602 memset(&xdp, 0, sizeof(xdp)); 9603 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9604 xdp.extack = extack; 9605 xdp.flags = flags; 9606 xdp.prog = prog; 9607 9608 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9609 * "moved" into driver), so they don't increment it on their own, but 9610 * they do decrement refcnt when program is detached or replaced. 9611 * Given net_device also owns link/prog, we need to bump refcnt here 9612 * to prevent drivers from underflowing it. 9613 */ 9614 if (prog) 9615 bpf_prog_inc(prog); 9616 err = bpf_op(dev, &xdp); 9617 if (err) { 9618 if (prog) 9619 bpf_prog_put(prog); 9620 return err; 9621 } 9622 9623 if (mode != XDP_MODE_HW) 9624 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9625 9626 return 0; 9627 } 9628 9629 static void dev_xdp_uninstall(struct net_device *dev) 9630 { 9631 struct bpf_xdp_link *link; 9632 struct bpf_prog *prog; 9633 enum bpf_xdp_mode mode; 9634 bpf_op_t bpf_op; 9635 9636 ASSERT_RTNL(); 9637 9638 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9639 prog = dev_xdp_prog(dev, mode); 9640 if (!prog) 9641 continue; 9642 9643 bpf_op = dev_xdp_bpf_op(dev, mode); 9644 if (!bpf_op) 9645 continue; 9646 9647 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9648 9649 /* auto-detach link from net device */ 9650 link = dev_xdp_link(dev, mode); 9651 if (link) 9652 link->dev = NULL; 9653 else 9654 bpf_prog_put(prog); 9655 9656 dev_xdp_set_link(dev, mode, NULL); 9657 } 9658 } 9659 9660 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9661 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9662 struct bpf_prog *old_prog, u32 flags) 9663 { 9664 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9665 struct bpf_prog *cur_prog; 9666 struct net_device *upper; 9667 struct list_head *iter; 9668 enum bpf_xdp_mode mode; 9669 bpf_op_t bpf_op; 9670 int err; 9671 9672 ASSERT_RTNL(); 9673 9674 /* either link or prog attachment, never both */ 9675 if (link && (new_prog || old_prog)) 9676 return -EINVAL; 9677 /* link supports only XDP mode flags */ 9678 if (link && (flags & ~XDP_FLAGS_MODES)) { 9679 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9680 return -EINVAL; 9681 } 9682 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9683 if (num_modes > 1) { 9684 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9685 return -EINVAL; 9686 } 9687 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9688 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9689 NL_SET_ERR_MSG(extack, 9690 "More than one program loaded, unset mode is ambiguous"); 9691 return -EINVAL; 9692 } 9693 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9694 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9695 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9696 return -EINVAL; 9697 } 9698 9699 mode = dev_xdp_mode(dev, flags); 9700 /* can't replace attached link */ 9701 if (dev_xdp_link(dev, mode)) { 9702 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9703 return -EBUSY; 9704 } 9705 9706 /* don't allow if an upper device already has a program */ 9707 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9708 if (dev_xdp_prog_count(upper) > 0) { 9709 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9710 return -EEXIST; 9711 } 9712 } 9713 9714 cur_prog = dev_xdp_prog(dev, mode); 9715 /* can't replace attached prog with link */ 9716 if (link && cur_prog) { 9717 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9718 return -EBUSY; 9719 } 9720 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9721 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9722 return -EEXIST; 9723 } 9724 9725 /* put effective new program into new_prog */ 9726 if (link) 9727 new_prog = link->link.prog; 9728 9729 if (new_prog) { 9730 bool offload = mode == XDP_MODE_HW; 9731 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9732 ? XDP_MODE_DRV : XDP_MODE_SKB; 9733 9734 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9735 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9736 return -EBUSY; 9737 } 9738 if (!offload && dev_xdp_prog(dev, other_mode)) { 9739 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9740 return -EEXIST; 9741 } 9742 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9743 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9744 return -EINVAL; 9745 } 9746 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9747 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9748 return -EINVAL; 9749 } 9750 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9751 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9752 return -EINVAL; 9753 } 9754 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9755 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9756 return -EINVAL; 9757 } 9758 } 9759 9760 /* don't call drivers if the effective program didn't change */ 9761 if (new_prog != cur_prog) { 9762 bpf_op = dev_xdp_bpf_op(dev, mode); 9763 if (!bpf_op) { 9764 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9765 return -EOPNOTSUPP; 9766 } 9767 9768 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9769 if (err) 9770 return err; 9771 } 9772 9773 if (link) 9774 dev_xdp_set_link(dev, mode, link); 9775 else 9776 dev_xdp_set_prog(dev, mode, new_prog); 9777 if (cur_prog) 9778 bpf_prog_put(cur_prog); 9779 9780 return 0; 9781 } 9782 9783 static int dev_xdp_attach_link(struct net_device *dev, 9784 struct netlink_ext_ack *extack, 9785 struct bpf_xdp_link *link) 9786 { 9787 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9788 } 9789 9790 static int dev_xdp_detach_link(struct net_device *dev, 9791 struct netlink_ext_ack *extack, 9792 struct bpf_xdp_link *link) 9793 { 9794 enum bpf_xdp_mode mode; 9795 bpf_op_t bpf_op; 9796 9797 ASSERT_RTNL(); 9798 9799 mode = dev_xdp_mode(dev, link->flags); 9800 if (dev_xdp_link(dev, mode) != link) 9801 return -EINVAL; 9802 9803 bpf_op = dev_xdp_bpf_op(dev, mode); 9804 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9805 dev_xdp_set_link(dev, mode, NULL); 9806 return 0; 9807 } 9808 9809 static void bpf_xdp_link_release(struct bpf_link *link) 9810 { 9811 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9812 9813 rtnl_lock(); 9814 9815 /* if racing with net_device's tear down, xdp_link->dev might be 9816 * already NULL, in which case link was already auto-detached 9817 */ 9818 if (xdp_link->dev) { 9819 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9820 xdp_link->dev = NULL; 9821 } 9822 9823 rtnl_unlock(); 9824 } 9825 9826 static int bpf_xdp_link_detach(struct bpf_link *link) 9827 { 9828 bpf_xdp_link_release(link); 9829 return 0; 9830 } 9831 9832 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9833 { 9834 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9835 9836 kfree(xdp_link); 9837 } 9838 9839 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9840 struct seq_file *seq) 9841 { 9842 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9843 u32 ifindex = 0; 9844 9845 rtnl_lock(); 9846 if (xdp_link->dev) 9847 ifindex = xdp_link->dev->ifindex; 9848 rtnl_unlock(); 9849 9850 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9851 } 9852 9853 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9854 struct bpf_link_info *info) 9855 { 9856 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9857 u32 ifindex = 0; 9858 9859 rtnl_lock(); 9860 if (xdp_link->dev) 9861 ifindex = xdp_link->dev->ifindex; 9862 rtnl_unlock(); 9863 9864 info->xdp.ifindex = ifindex; 9865 return 0; 9866 } 9867 9868 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9869 struct bpf_prog *old_prog) 9870 { 9871 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9872 enum bpf_xdp_mode mode; 9873 bpf_op_t bpf_op; 9874 int err = 0; 9875 9876 rtnl_lock(); 9877 9878 /* link might have been auto-released already, so fail */ 9879 if (!xdp_link->dev) { 9880 err = -ENOLINK; 9881 goto out_unlock; 9882 } 9883 9884 if (old_prog && link->prog != old_prog) { 9885 err = -EPERM; 9886 goto out_unlock; 9887 } 9888 old_prog = link->prog; 9889 if (old_prog->type != new_prog->type || 9890 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9891 err = -EINVAL; 9892 goto out_unlock; 9893 } 9894 9895 if (old_prog == new_prog) { 9896 /* no-op, don't disturb drivers */ 9897 bpf_prog_put(new_prog); 9898 goto out_unlock; 9899 } 9900 9901 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9902 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9903 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9904 xdp_link->flags, new_prog); 9905 if (err) 9906 goto out_unlock; 9907 9908 old_prog = xchg(&link->prog, new_prog); 9909 bpf_prog_put(old_prog); 9910 9911 out_unlock: 9912 rtnl_unlock(); 9913 return err; 9914 } 9915 9916 static const struct bpf_link_ops bpf_xdp_link_lops = { 9917 .release = bpf_xdp_link_release, 9918 .dealloc = bpf_xdp_link_dealloc, 9919 .detach = bpf_xdp_link_detach, 9920 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9921 .fill_link_info = bpf_xdp_link_fill_link_info, 9922 .update_prog = bpf_xdp_link_update, 9923 }; 9924 9925 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9926 { 9927 struct net *net = current->nsproxy->net_ns; 9928 struct bpf_link_primer link_primer; 9929 struct netlink_ext_ack extack = {}; 9930 struct bpf_xdp_link *link; 9931 struct net_device *dev; 9932 int err, fd; 9933 9934 rtnl_lock(); 9935 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9936 if (!dev) { 9937 rtnl_unlock(); 9938 return -EINVAL; 9939 } 9940 9941 link = kzalloc(sizeof(*link), GFP_USER); 9942 if (!link) { 9943 err = -ENOMEM; 9944 goto unlock; 9945 } 9946 9947 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9948 link->dev = dev; 9949 link->flags = attr->link_create.flags; 9950 9951 err = bpf_link_prime(&link->link, &link_primer); 9952 if (err) { 9953 kfree(link); 9954 goto unlock; 9955 } 9956 9957 err = dev_xdp_attach_link(dev, &extack, link); 9958 rtnl_unlock(); 9959 9960 if (err) { 9961 link->dev = NULL; 9962 bpf_link_cleanup(&link_primer); 9963 trace_bpf_xdp_link_attach_failed(extack._msg); 9964 goto out_put_dev; 9965 } 9966 9967 fd = bpf_link_settle(&link_primer); 9968 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9969 dev_put(dev); 9970 return fd; 9971 9972 unlock: 9973 rtnl_unlock(); 9974 9975 out_put_dev: 9976 dev_put(dev); 9977 return err; 9978 } 9979 9980 /** 9981 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9982 * @dev: device 9983 * @extack: netlink extended ack 9984 * @fd: new program fd or negative value to clear 9985 * @expected_fd: old program fd that userspace expects to replace or clear 9986 * @flags: xdp-related flags 9987 * 9988 * Set or clear a bpf program for a device 9989 */ 9990 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9991 int fd, int expected_fd, u32 flags) 9992 { 9993 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9994 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9995 int err; 9996 9997 ASSERT_RTNL(); 9998 9999 if (fd >= 0) { 10000 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 10001 mode != XDP_MODE_SKB); 10002 if (IS_ERR(new_prog)) 10003 return PTR_ERR(new_prog); 10004 } 10005 10006 if (expected_fd >= 0) { 10007 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 10008 mode != XDP_MODE_SKB); 10009 if (IS_ERR(old_prog)) { 10010 err = PTR_ERR(old_prog); 10011 old_prog = NULL; 10012 goto err_out; 10013 } 10014 } 10015 10016 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 10017 10018 err_out: 10019 if (err && new_prog) 10020 bpf_prog_put(new_prog); 10021 if (old_prog) 10022 bpf_prog_put(old_prog); 10023 return err; 10024 } 10025 10026 u32 dev_get_min_mp_channel_count(const struct net_device *dev) 10027 { 10028 int i; 10029 10030 ASSERT_RTNL(); 10031 10032 for (i = dev->real_num_rx_queues - 1; i >= 0; i--) 10033 if (dev->_rx[i].mp_params.mp_priv) 10034 /* The channel count is the idx plus 1. */ 10035 return i + 1; 10036 10037 return 0; 10038 } 10039 10040 /** 10041 * dev_index_reserve() - allocate an ifindex in a namespace 10042 * @net: the applicable net namespace 10043 * @ifindex: requested ifindex, pass %0 to get one allocated 10044 * 10045 * Allocate a ifindex for a new device. Caller must either use the ifindex 10046 * to store the device (via list_netdevice()) or call dev_index_release() 10047 * to give the index up. 10048 * 10049 * Return: a suitable unique value for a new device interface number or -errno. 10050 */ 10051 static int dev_index_reserve(struct net *net, u32 ifindex) 10052 { 10053 int err; 10054 10055 if (ifindex > INT_MAX) { 10056 DEBUG_NET_WARN_ON_ONCE(1); 10057 return -EINVAL; 10058 } 10059 10060 if (!ifindex) 10061 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 10062 xa_limit_31b, &net->ifindex, GFP_KERNEL); 10063 else 10064 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 10065 if (err < 0) 10066 return err; 10067 10068 return ifindex; 10069 } 10070 10071 static void dev_index_release(struct net *net, int ifindex) 10072 { 10073 /* Expect only unused indexes, unlist_netdevice() removes the used */ 10074 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 10075 } 10076 10077 /* Delayed registration/unregisteration */ 10078 LIST_HEAD(net_todo_list); 10079 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 10080 atomic_t dev_unreg_count = ATOMIC_INIT(0); 10081 10082 static void net_set_todo(struct net_device *dev) 10083 { 10084 list_add_tail(&dev->todo_list, &net_todo_list); 10085 } 10086 10087 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 10088 struct net_device *upper, netdev_features_t features) 10089 { 10090 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10091 netdev_features_t feature; 10092 int feature_bit; 10093 10094 for_each_netdev_feature(upper_disables, feature_bit) { 10095 feature = __NETIF_F_BIT(feature_bit); 10096 if (!(upper->wanted_features & feature) 10097 && (features & feature)) { 10098 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 10099 &feature, upper->name); 10100 features &= ~feature; 10101 } 10102 } 10103 10104 return features; 10105 } 10106 10107 static void netdev_sync_lower_features(struct net_device *upper, 10108 struct net_device *lower, netdev_features_t features) 10109 { 10110 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10111 netdev_features_t feature; 10112 int feature_bit; 10113 10114 for_each_netdev_feature(upper_disables, feature_bit) { 10115 feature = __NETIF_F_BIT(feature_bit); 10116 if (!(features & feature) && (lower->features & feature)) { 10117 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 10118 &feature, lower->name); 10119 lower->wanted_features &= ~feature; 10120 __netdev_update_features(lower); 10121 10122 if (unlikely(lower->features & feature)) 10123 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 10124 &feature, lower->name); 10125 else 10126 netdev_features_change(lower); 10127 } 10128 } 10129 } 10130 10131 static bool netdev_has_ip_or_hw_csum(netdev_features_t features) 10132 { 10133 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 10134 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask; 10135 bool hw_csum = features & NETIF_F_HW_CSUM; 10136 10137 return ip_csum || hw_csum; 10138 } 10139 10140 static netdev_features_t netdev_fix_features(struct net_device *dev, 10141 netdev_features_t features) 10142 { 10143 /* Fix illegal checksum combinations */ 10144 if ((features & NETIF_F_HW_CSUM) && 10145 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 10146 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 10147 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 10148 } 10149 10150 /* TSO requires that SG is present as well. */ 10151 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 10152 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 10153 features &= ~NETIF_F_ALL_TSO; 10154 } 10155 10156 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 10157 !(features & NETIF_F_IP_CSUM)) { 10158 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 10159 features &= ~NETIF_F_TSO; 10160 features &= ~NETIF_F_TSO_ECN; 10161 } 10162 10163 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 10164 !(features & NETIF_F_IPV6_CSUM)) { 10165 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 10166 features &= ~NETIF_F_TSO6; 10167 } 10168 10169 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 10170 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 10171 features &= ~NETIF_F_TSO_MANGLEID; 10172 10173 /* TSO ECN requires that TSO is present as well. */ 10174 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 10175 features &= ~NETIF_F_TSO_ECN; 10176 10177 /* Software GSO depends on SG. */ 10178 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 10179 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 10180 features &= ~NETIF_F_GSO; 10181 } 10182 10183 /* GSO partial features require GSO partial be set */ 10184 if ((features & dev->gso_partial_features) && 10185 !(features & NETIF_F_GSO_PARTIAL)) { 10186 netdev_dbg(dev, 10187 "Dropping partially supported GSO features since no GSO partial.\n"); 10188 features &= ~dev->gso_partial_features; 10189 } 10190 10191 if (!(features & NETIF_F_RXCSUM)) { 10192 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 10193 * successfully merged by hardware must also have the 10194 * checksum verified by hardware. If the user does not 10195 * want to enable RXCSUM, logically, we should disable GRO_HW. 10196 */ 10197 if (features & NETIF_F_GRO_HW) { 10198 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 10199 features &= ~NETIF_F_GRO_HW; 10200 } 10201 } 10202 10203 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 10204 if (features & NETIF_F_RXFCS) { 10205 if (features & NETIF_F_LRO) { 10206 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 10207 features &= ~NETIF_F_LRO; 10208 } 10209 10210 if (features & NETIF_F_GRO_HW) { 10211 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 10212 features &= ~NETIF_F_GRO_HW; 10213 } 10214 } 10215 10216 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 10217 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 10218 features &= ~NETIF_F_LRO; 10219 } 10220 10221 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) { 10222 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 10223 features &= ~NETIF_F_HW_TLS_TX; 10224 } 10225 10226 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 10227 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 10228 features &= ~NETIF_F_HW_TLS_RX; 10229 } 10230 10231 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) { 10232 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n"); 10233 features &= ~NETIF_F_GSO_UDP_L4; 10234 } 10235 10236 return features; 10237 } 10238 10239 int __netdev_update_features(struct net_device *dev) 10240 { 10241 struct net_device *upper, *lower; 10242 netdev_features_t features; 10243 struct list_head *iter; 10244 int err = -1; 10245 10246 ASSERT_RTNL(); 10247 10248 features = netdev_get_wanted_features(dev); 10249 10250 if (dev->netdev_ops->ndo_fix_features) 10251 features = dev->netdev_ops->ndo_fix_features(dev, features); 10252 10253 /* driver might be less strict about feature dependencies */ 10254 features = netdev_fix_features(dev, features); 10255 10256 /* some features can't be enabled if they're off on an upper device */ 10257 netdev_for_each_upper_dev_rcu(dev, upper, iter) 10258 features = netdev_sync_upper_features(dev, upper, features); 10259 10260 if (dev->features == features) 10261 goto sync_lower; 10262 10263 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 10264 &dev->features, &features); 10265 10266 if (dev->netdev_ops->ndo_set_features) 10267 err = dev->netdev_ops->ndo_set_features(dev, features); 10268 else 10269 err = 0; 10270 10271 if (unlikely(err < 0)) { 10272 netdev_err(dev, 10273 "set_features() failed (%d); wanted %pNF, left %pNF\n", 10274 err, &features, &dev->features); 10275 /* return non-0 since some features might have changed and 10276 * it's better to fire a spurious notification than miss it 10277 */ 10278 return -1; 10279 } 10280 10281 sync_lower: 10282 /* some features must be disabled on lower devices when disabled 10283 * on an upper device (think: bonding master or bridge) 10284 */ 10285 netdev_for_each_lower_dev(dev, lower, iter) 10286 netdev_sync_lower_features(dev, lower, features); 10287 10288 if (!err) { 10289 netdev_features_t diff = features ^ dev->features; 10290 10291 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10292 /* udp_tunnel_{get,drop}_rx_info both need 10293 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10294 * device, or they won't do anything. 10295 * Thus we need to update dev->features 10296 * *before* calling udp_tunnel_get_rx_info, 10297 * but *after* calling udp_tunnel_drop_rx_info. 10298 */ 10299 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10300 dev->features = features; 10301 udp_tunnel_get_rx_info(dev); 10302 } else { 10303 udp_tunnel_drop_rx_info(dev); 10304 } 10305 } 10306 10307 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10308 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10309 dev->features = features; 10310 err |= vlan_get_rx_ctag_filter_info(dev); 10311 } else { 10312 vlan_drop_rx_ctag_filter_info(dev); 10313 } 10314 } 10315 10316 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10317 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10318 dev->features = features; 10319 err |= vlan_get_rx_stag_filter_info(dev); 10320 } else { 10321 vlan_drop_rx_stag_filter_info(dev); 10322 } 10323 } 10324 10325 dev->features = features; 10326 } 10327 10328 return err < 0 ? 0 : 1; 10329 } 10330 10331 /** 10332 * netdev_update_features - recalculate device features 10333 * @dev: the device to check 10334 * 10335 * Recalculate dev->features set and send notifications if it 10336 * has changed. Should be called after driver or hardware dependent 10337 * conditions might have changed that influence the features. 10338 */ 10339 void netdev_update_features(struct net_device *dev) 10340 { 10341 if (__netdev_update_features(dev)) 10342 netdev_features_change(dev); 10343 } 10344 EXPORT_SYMBOL(netdev_update_features); 10345 10346 /** 10347 * netdev_change_features - recalculate device features 10348 * @dev: the device to check 10349 * 10350 * Recalculate dev->features set and send notifications even 10351 * if they have not changed. Should be called instead of 10352 * netdev_update_features() if also dev->vlan_features might 10353 * have changed to allow the changes to be propagated to stacked 10354 * VLAN devices. 10355 */ 10356 void netdev_change_features(struct net_device *dev) 10357 { 10358 __netdev_update_features(dev); 10359 netdev_features_change(dev); 10360 } 10361 EXPORT_SYMBOL(netdev_change_features); 10362 10363 /** 10364 * netif_stacked_transfer_operstate - transfer operstate 10365 * @rootdev: the root or lower level device to transfer state from 10366 * @dev: the device to transfer operstate to 10367 * 10368 * Transfer operational state from root to device. This is normally 10369 * called when a stacking relationship exists between the root 10370 * device and the device(a leaf device). 10371 */ 10372 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10373 struct net_device *dev) 10374 { 10375 if (rootdev->operstate == IF_OPER_DORMANT) 10376 netif_dormant_on(dev); 10377 else 10378 netif_dormant_off(dev); 10379 10380 if (rootdev->operstate == IF_OPER_TESTING) 10381 netif_testing_on(dev); 10382 else 10383 netif_testing_off(dev); 10384 10385 if (netif_carrier_ok(rootdev)) 10386 netif_carrier_on(dev); 10387 else 10388 netif_carrier_off(dev); 10389 } 10390 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10391 10392 static int netif_alloc_rx_queues(struct net_device *dev) 10393 { 10394 unsigned int i, count = dev->num_rx_queues; 10395 struct netdev_rx_queue *rx; 10396 size_t sz = count * sizeof(*rx); 10397 int err = 0; 10398 10399 BUG_ON(count < 1); 10400 10401 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10402 if (!rx) 10403 return -ENOMEM; 10404 10405 dev->_rx = rx; 10406 10407 for (i = 0; i < count; i++) { 10408 rx[i].dev = dev; 10409 10410 /* XDP RX-queue setup */ 10411 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10412 if (err < 0) 10413 goto err_rxq_info; 10414 } 10415 return 0; 10416 10417 err_rxq_info: 10418 /* Rollback successful reg's and free other resources */ 10419 while (i--) 10420 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10421 kvfree(dev->_rx); 10422 dev->_rx = NULL; 10423 return err; 10424 } 10425 10426 static void netif_free_rx_queues(struct net_device *dev) 10427 { 10428 unsigned int i, count = dev->num_rx_queues; 10429 10430 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10431 if (!dev->_rx) 10432 return; 10433 10434 for (i = 0; i < count; i++) 10435 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10436 10437 kvfree(dev->_rx); 10438 } 10439 10440 static void netdev_init_one_queue(struct net_device *dev, 10441 struct netdev_queue *queue, void *_unused) 10442 { 10443 /* Initialize queue lock */ 10444 spin_lock_init(&queue->_xmit_lock); 10445 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10446 queue->xmit_lock_owner = -1; 10447 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10448 queue->dev = dev; 10449 #ifdef CONFIG_BQL 10450 dql_init(&queue->dql, HZ); 10451 #endif 10452 } 10453 10454 static void netif_free_tx_queues(struct net_device *dev) 10455 { 10456 kvfree(dev->_tx); 10457 } 10458 10459 static int netif_alloc_netdev_queues(struct net_device *dev) 10460 { 10461 unsigned int count = dev->num_tx_queues; 10462 struct netdev_queue *tx; 10463 size_t sz = count * sizeof(*tx); 10464 10465 if (count < 1 || count > 0xffff) 10466 return -EINVAL; 10467 10468 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10469 if (!tx) 10470 return -ENOMEM; 10471 10472 dev->_tx = tx; 10473 10474 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10475 spin_lock_init(&dev->tx_global_lock); 10476 10477 return 0; 10478 } 10479 10480 void netif_tx_stop_all_queues(struct net_device *dev) 10481 { 10482 unsigned int i; 10483 10484 for (i = 0; i < dev->num_tx_queues; i++) { 10485 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10486 10487 netif_tx_stop_queue(txq); 10488 } 10489 } 10490 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10491 10492 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10493 { 10494 void __percpu *v; 10495 10496 /* Drivers implementing ndo_get_peer_dev must support tstat 10497 * accounting, so that skb_do_redirect() can bump the dev's 10498 * RX stats upon network namespace switch. 10499 */ 10500 if (dev->netdev_ops->ndo_get_peer_dev && 10501 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10502 return -EOPNOTSUPP; 10503 10504 switch (dev->pcpu_stat_type) { 10505 case NETDEV_PCPU_STAT_NONE: 10506 return 0; 10507 case NETDEV_PCPU_STAT_LSTATS: 10508 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10509 break; 10510 case NETDEV_PCPU_STAT_TSTATS: 10511 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10512 break; 10513 case NETDEV_PCPU_STAT_DSTATS: 10514 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10515 break; 10516 default: 10517 return -EINVAL; 10518 } 10519 10520 return v ? 0 : -ENOMEM; 10521 } 10522 10523 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10524 { 10525 switch (dev->pcpu_stat_type) { 10526 case NETDEV_PCPU_STAT_NONE: 10527 return; 10528 case NETDEV_PCPU_STAT_LSTATS: 10529 free_percpu(dev->lstats); 10530 break; 10531 case NETDEV_PCPU_STAT_TSTATS: 10532 free_percpu(dev->tstats); 10533 break; 10534 case NETDEV_PCPU_STAT_DSTATS: 10535 free_percpu(dev->dstats); 10536 break; 10537 } 10538 } 10539 10540 static void netdev_free_phy_link_topology(struct net_device *dev) 10541 { 10542 struct phy_link_topology *topo = dev->link_topo; 10543 10544 if (IS_ENABLED(CONFIG_PHYLIB) && topo) { 10545 xa_destroy(&topo->phys); 10546 kfree(topo); 10547 dev->link_topo = NULL; 10548 } 10549 } 10550 10551 /** 10552 * register_netdevice() - register a network device 10553 * @dev: device to register 10554 * 10555 * Take a prepared network device structure and make it externally accessible. 10556 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10557 * Callers must hold the rtnl lock - you may want register_netdev() 10558 * instead of this. 10559 */ 10560 int register_netdevice(struct net_device *dev) 10561 { 10562 int ret; 10563 struct net *net = dev_net(dev); 10564 10565 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10566 NETDEV_FEATURE_COUNT); 10567 BUG_ON(dev_boot_phase); 10568 ASSERT_RTNL(); 10569 10570 might_sleep(); 10571 10572 /* When net_device's are persistent, this will be fatal. */ 10573 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10574 BUG_ON(!net); 10575 10576 ret = ethtool_check_ops(dev->ethtool_ops); 10577 if (ret) 10578 return ret; 10579 10580 /* rss ctx ID 0 is reserved for the default context, start from 1 */ 10581 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1); 10582 mutex_init(&dev->ethtool->rss_lock); 10583 10584 spin_lock_init(&dev->addr_list_lock); 10585 netdev_set_addr_lockdep_class(dev); 10586 10587 ret = dev_get_valid_name(net, dev, dev->name); 10588 if (ret < 0) 10589 goto out; 10590 10591 ret = -ENOMEM; 10592 dev->name_node = netdev_name_node_head_alloc(dev); 10593 if (!dev->name_node) 10594 goto out; 10595 10596 /* Init, if this function is available */ 10597 if (dev->netdev_ops->ndo_init) { 10598 ret = dev->netdev_ops->ndo_init(dev); 10599 if (ret) { 10600 if (ret > 0) 10601 ret = -EIO; 10602 goto err_free_name; 10603 } 10604 } 10605 10606 if (((dev->hw_features | dev->features) & 10607 NETIF_F_HW_VLAN_CTAG_FILTER) && 10608 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10609 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10610 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10611 ret = -EINVAL; 10612 goto err_uninit; 10613 } 10614 10615 ret = netdev_do_alloc_pcpu_stats(dev); 10616 if (ret) 10617 goto err_uninit; 10618 10619 ret = dev_index_reserve(net, dev->ifindex); 10620 if (ret < 0) 10621 goto err_free_pcpu; 10622 dev->ifindex = ret; 10623 10624 /* Transfer changeable features to wanted_features and enable 10625 * software offloads (GSO and GRO). 10626 */ 10627 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10628 dev->features |= NETIF_F_SOFT_FEATURES; 10629 10630 if (dev->udp_tunnel_nic_info) { 10631 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10632 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10633 } 10634 10635 dev->wanted_features = dev->features & dev->hw_features; 10636 10637 if (!(dev->flags & IFF_LOOPBACK)) 10638 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10639 10640 /* If IPv4 TCP segmentation offload is supported we should also 10641 * allow the device to enable segmenting the frame with the option 10642 * of ignoring a static IP ID value. This doesn't enable the 10643 * feature itself but allows the user to enable it later. 10644 */ 10645 if (dev->hw_features & NETIF_F_TSO) 10646 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10647 if (dev->vlan_features & NETIF_F_TSO) 10648 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10649 if (dev->mpls_features & NETIF_F_TSO) 10650 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10651 if (dev->hw_enc_features & NETIF_F_TSO) 10652 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10653 10654 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10655 */ 10656 dev->vlan_features |= NETIF_F_HIGHDMA; 10657 10658 /* Make NETIF_F_SG inheritable to tunnel devices. 10659 */ 10660 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10661 10662 /* Make NETIF_F_SG inheritable to MPLS. 10663 */ 10664 dev->mpls_features |= NETIF_F_SG; 10665 10666 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10667 ret = notifier_to_errno(ret); 10668 if (ret) 10669 goto err_ifindex_release; 10670 10671 ret = netdev_register_kobject(dev); 10672 10673 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); 10674 10675 if (ret) 10676 goto err_uninit_notify; 10677 10678 __netdev_update_features(dev); 10679 10680 /* 10681 * Default initial state at registry is that the 10682 * device is present. 10683 */ 10684 10685 set_bit(__LINK_STATE_PRESENT, &dev->state); 10686 10687 linkwatch_init_dev(dev); 10688 10689 dev_init_scheduler(dev); 10690 10691 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10692 list_netdevice(dev); 10693 10694 add_device_randomness(dev->dev_addr, dev->addr_len); 10695 10696 /* If the device has permanent device address, driver should 10697 * set dev_addr and also addr_assign_type should be set to 10698 * NET_ADDR_PERM (default value). 10699 */ 10700 if (dev->addr_assign_type == NET_ADDR_PERM) 10701 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10702 10703 /* Notify protocols, that a new device appeared. */ 10704 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10705 ret = notifier_to_errno(ret); 10706 if (ret) { 10707 /* Expect explicit free_netdev() on failure */ 10708 dev->needs_free_netdev = false; 10709 unregister_netdevice_queue(dev, NULL); 10710 goto out; 10711 } 10712 /* 10713 * Prevent userspace races by waiting until the network 10714 * device is fully setup before sending notifications. 10715 */ 10716 if (!dev->rtnl_link_ops || 10717 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10718 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10719 10720 out: 10721 return ret; 10722 10723 err_uninit_notify: 10724 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10725 err_ifindex_release: 10726 dev_index_release(net, dev->ifindex); 10727 err_free_pcpu: 10728 netdev_do_free_pcpu_stats(dev); 10729 err_uninit: 10730 if (dev->netdev_ops->ndo_uninit) 10731 dev->netdev_ops->ndo_uninit(dev); 10732 if (dev->priv_destructor) 10733 dev->priv_destructor(dev); 10734 err_free_name: 10735 netdev_name_node_free(dev->name_node); 10736 goto out; 10737 } 10738 EXPORT_SYMBOL(register_netdevice); 10739 10740 /* Initialize the core of a dummy net device. 10741 * This is useful if you are calling this function after alloc_netdev(), 10742 * since it does not memset the net_device fields. 10743 */ 10744 static void init_dummy_netdev_core(struct net_device *dev) 10745 { 10746 /* make sure we BUG if trying to hit standard 10747 * register/unregister code path 10748 */ 10749 dev->reg_state = NETREG_DUMMY; 10750 10751 /* NAPI wants this */ 10752 INIT_LIST_HEAD(&dev->napi_list); 10753 10754 /* a dummy interface is started by default */ 10755 set_bit(__LINK_STATE_PRESENT, &dev->state); 10756 set_bit(__LINK_STATE_START, &dev->state); 10757 10758 /* napi_busy_loop stats accounting wants this */ 10759 dev_net_set(dev, &init_net); 10760 10761 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10762 * because users of this 'device' dont need to change 10763 * its refcount. 10764 */ 10765 } 10766 10767 /** 10768 * init_dummy_netdev - init a dummy network device for NAPI 10769 * @dev: device to init 10770 * 10771 * This takes a network device structure and initializes the minimum 10772 * amount of fields so it can be used to schedule NAPI polls without 10773 * registering a full blown interface. This is to be used by drivers 10774 * that need to tie several hardware interfaces to a single NAPI 10775 * poll scheduler due to HW limitations. 10776 */ 10777 void init_dummy_netdev(struct net_device *dev) 10778 { 10779 /* Clear everything. Note we don't initialize spinlocks 10780 * as they aren't supposed to be taken by any of the 10781 * NAPI code and this dummy netdev is supposed to be 10782 * only ever used for NAPI polls 10783 */ 10784 memset(dev, 0, sizeof(struct net_device)); 10785 init_dummy_netdev_core(dev); 10786 } 10787 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10788 10789 /** 10790 * register_netdev - register a network device 10791 * @dev: device to register 10792 * 10793 * Take a completed network device structure and add it to the kernel 10794 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10795 * chain. 0 is returned on success. A negative errno code is returned 10796 * on a failure to set up the device, or if the name is a duplicate. 10797 * 10798 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10799 * and expands the device name if you passed a format string to 10800 * alloc_netdev. 10801 */ 10802 int register_netdev(struct net_device *dev) 10803 { 10804 struct net *net = dev_net(dev); 10805 int err; 10806 10807 if (rtnl_net_lock_killable(net)) 10808 return -EINTR; 10809 10810 err = register_netdevice(dev); 10811 10812 rtnl_net_unlock(net); 10813 10814 return err; 10815 } 10816 EXPORT_SYMBOL(register_netdev); 10817 10818 int netdev_refcnt_read(const struct net_device *dev) 10819 { 10820 #ifdef CONFIG_PCPU_DEV_REFCNT 10821 int i, refcnt = 0; 10822 10823 for_each_possible_cpu(i) 10824 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10825 return refcnt; 10826 #else 10827 return refcount_read(&dev->dev_refcnt); 10828 #endif 10829 } 10830 EXPORT_SYMBOL(netdev_refcnt_read); 10831 10832 int netdev_unregister_timeout_secs __read_mostly = 10; 10833 10834 #define WAIT_REFS_MIN_MSECS 1 10835 #define WAIT_REFS_MAX_MSECS 250 10836 /** 10837 * netdev_wait_allrefs_any - wait until all references are gone. 10838 * @list: list of net_devices to wait on 10839 * 10840 * This is called when unregistering network devices. 10841 * 10842 * Any protocol or device that holds a reference should register 10843 * for netdevice notification, and cleanup and put back the 10844 * reference if they receive an UNREGISTER event. 10845 * We can get stuck here if buggy protocols don't correctly 10846 * call dev_put. 10847 */ 10848 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10849 { 10850 unsigned long rebroadcast_time, warning_time; 10851 struct net_device *dev; 10852 int wait = 0; 10853 10854 rebroadcast_time = warning_time = jiffies; 10855 10856 list_for_each_entry(dev, list, todo_list) 10857 if (netdev_refcnt_read(dev) == 1) 10858 return dev; 10859 10860 while (true) { 10861 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10862 rtnl_lock(); 10863 10864 /* Rebroadcast unregister notification */ 10865 list_for_each_entry(dev, list, todo_list) 10866 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10867 10868 __rtnl_unlock(); 10869 rcu_barrier(); 10870 rtnl_lock(); 10871 10872 list_for_each_entry(dev, list, todo_list) 10873 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10874 &dev->state)) { 10875 /* We must not have linkwatch events 10876 * pending on unregister. If this 10877 * happens, we simply run the queue 10878 * unscheduled, resulting in a noop 10879 * for this device. 10880 */ 10881 linkwatch_run_queue(); 10882 break; 10883 } 10884 10885 __rtnl_unlock(); 10886 10887 rebroadcast_time = jiffies; 10888 } 10889 10890 rcu_barrier(); 10891 10892 if (!wait) { 10893 wait = WAIT_REFS_MIN_MSECS; 10894 } else { 10895 msleep(wait); 10896 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10897 } 10898 10899 list_for_each_entry(dev, list, todo_list) 10900 if (netdev_refcnt_read(dev) == 1) 10901 return dev; 10902 10903 if (time_after(jiffies, warning_time + 10904 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10905 list_for_each_entry(dev, list, todo_list) { 10906 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10907 dev->name, netdev_refcnt_read(dev)); 10908 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10909 } 10910 10911 warning_time = jiffies; 10912 } 10913 } 10914 } 10915 10916 /* The sequence is: 10917 * 10918 * rtnl_lock(); 10919 * ... 10920 * register_netdevice(x1); 10921 * register_netdevice(x2); 10922 * ... 10923 * unregister_netdevice(y1); 10924 * unregister_netdevice(y2); 10925 * ... 10926 * rtnl_unlock(); 10927 * free_netdev(y1); 10928 * free_netdev(y2); 10929 * 10930 * We are invoked by rtnl_unlock(). 10931 * This allows us to deal with problems: 10932 * 1) We can delete sysfs objects which invoke hotplug 10933 * without deadlocking with linkwatch via keventd. 10934 * 2) Since we run with the RTNL semaphore not held, we can sleep 10935 * safely in order to wait for the netdev refcnt to drop to zero. 10936 * 10937 * We must not return until all unregister events added during 10938 * the interval the lock was held have been completed. 10939 */ 10940 void netdev_run_todo(void) 10941 { 10942 struct net_device *dev, *tmp; 10943 struct list_head list; 10944 int cnt; 10945 #ifdef CONFIG_LOCKDEP 10946 struct list_head unlink_list; 10947 10948 list_replace_init(&net_unlink_list, &unlink_list); 10949 10950 while (!list_empty(&unlink_list)) { 10951 struct net_device *dev = list_first_entry(&unlink_list, 10952 struct net_device, 10953 unlink_list); 10954 list_del_init(&dev->unlink_list); 10955 dev->nested_level = dev->lower_level - 1; 10956 } 10957 #endif 10958 10959 /* Snapshot list, allow later requests */ 10960 list_replace_init(&net_todo_list, &list); 10961 10962 __rtnl_unlock(); 10963 10964 /* Wait for rcu callbacks to finish before next phase */ 10965 if (!list_empty(&list)) 10966 rcu_barrier(); 10967 10968 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10969 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10970 netdev_WARN(dev, "run_todo but not unregistering\n"); 10971 list_del(&dev->todo_list); 10972 continue; 10973 } 10974 10975 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); 10976 linkwatch_sync_dev(dev); 10977 } 10978 10979 cnt = 0; 10980 while (!list_empty(&list)) { 10981 dev = netdev_wait_allrefs_any(&list); 10982 list_del(&dev->todo_list); 10983 10984 /* paranoia */ 10985 BUG_ON(netdev_refcnt_read(dev) != 1); 10986 BUG_ON(!list_empty(&dev->ptype_all)); 10987 BUG_ON(!list_empty(&dev->ptype_specific)); 10988 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10989 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10990 10991 netdev_do_free_pcpu_stats(dev); 10992 if (dev->priv_destructor) 10993 dev->priv_destructor(dev); 10994 if (dev->needs_free_netdev) 10995 free_netdev(dev); 10996 10997 cnt++; 10998 10999 /* Free network device */ 11000 kobject_put(&dev->dev.kobj); 11001 } 11002 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) 11003 wake_up(&netdev_unregistering_wq); 11004 } 11005 11006 /* Collate per-cpu network dstats statistics 11007 * 11008 * Read per-cpu network statistics from dev->dstats and populate the related 11009 * fields in @s. 11010 */ 11011 static void dev_fetch_dstats(struct rtnl_link_stats64 *s, 11012 const struct pcpu_dstats __percpu *dstats) 11013 { 11014 int cpu; 11015 11016 for_each_possible_cpu(cpu) { 11017 u64 rx_packets, rx_bytes, rx_drops; 11018 u64 tx_packets, tx_bytes, tx_drops; 11019 const struct pcpu_dstats *stats; 11020 unsigned int start; 11021 11022 stats = per_cpu_ptr(dstats, cpu); 11023 do { 11024 start = u64_stats_fetch_begin(&stats->syncp); 11025 rx_packets = u64_stats_read(&stats->rx_packets); 11026 rx_bytes = u64_stats_read(&stats->rx_bytes); 11027 rx_drops = u64_stats_read(&stats->rx_drops); 11028 tx_packets = u64_stats_read(&stats->tx_packets); 11029 tx_bytes = u64_stats_read(&stats->tx_bytes); 11030 tx_drops = u64_stats_read(&stats->tx_drops); 11031 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11032 11033 s->rx_packets += rx_packets; 11034 s->rx_bytes += rx_bytes; 11035 s->rx_dropped += rx_drops; 11036 s->tx_packets += tx_packets; 11037 s->tx_bytes += tx_bytes; 11038 s->tx_dropped += tx_drops; 11039 } 11040 } 11041 11042 /* ndo_get_stats64 implementation for dtstats-based accounting. 11043 * 11044 * Populate @s from dev->stats and dev->dstats. This is used internally by the 11045 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection. 11046 */ 11047 static void dev_get_dstats64(const struct net_device *dev, 11048 struct rtnl_link_stats64 *s) 11049 { 11050 netdev_stats_to_stats64(s, &dev->stats); 11051 dev_fetch_dstats(s, dev->dstats); 11052 } 11053 11054 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 11055 * all the same fields in the same order as net_device_stats, with only 11056 * the type differing, but rtnl_link_stats64 may have additional fields 11057 * at the end for newer counters. 11058 */ 11059 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 11060 const struct net_device_stats *netdev_stats) 11061 { 11062 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 11063 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 11064 u64 *dst = (u64 *)stats64; 11065 11066 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 11067 for (i = 0; i < n; i++) 11068 dst[i] = (unsigned long)atomic_long_read(&src[i]); 11069 /* zero out counters that only exist in rtnl_link_stats64 */ 11070 memset((char *)stats64 + n * sizeof(u64), 0, 11071 sizeof(*stats64) - n * sizeof(u64)); 11072 } 11073 EXPORT_SYMBOL(netdev_stats_to_stats64); 11074 11075 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 11076 struct net_device *dev) 11077 { 11078 struct net_device_core_stats __percpu *p; 11079 11080 p = alloc_percpu_gfp(struct net_device_core_stats, 11081 GFP_ATOMIC | __GFP_NOWARN); 11082 11083 if (p && cmpxchg(&dev->core_stats, NULL, p)) 11084 free_percpu(p); 11085 11086 /* This READ_ONCE() pairs with the cmpxchg() above */ 11087 return READ_ONCE(dev->core_stats); 11088 } 11089 11090 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 11091 { 11092 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11093 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 11094 unsigned long __percpu *field; 11095 11096 if (unlikely(!p)) { 11097 p = netdev_core_stats_alloc(dev); 11098 if (!p) 11099 return; 11100 } 11101 11102 field = (unsigned long __percpu *)((void __percpu *)p + offset); 11103 this_cpu_inc(*field); 11104 } 11105 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 11106 11107 /** 11108 * dev_get_stats - get network device statistics 11109 * @dev: device to get statistics from 11110 * @storage: place to store stats 11111 * 11112 * Get network statistics from device. Return @storage. 11113 * The device driver may provide its own method by setting 11114 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 11115 * otherwise the internal statistics structure is used. 11116 */ 11117 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 11118 struct rtnl_link_stats64 *storage) 11119 { 11120 const struct net_device_ops *ops = dev->netdev_ops; 11121 const struct net_device_core_stats __percpu *p; 11122 11123 if (ops->ndo_get_stats64) { 11124 memset(storage, 0, sizeof(*storage)); 11125 ops->ndo_get_stats64(dev, storage); 11126 } else if (ops->ndo_get_stats) { 11127 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 11128 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { 11129 dev_get_tstats64(dev, storage); 11130 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) { 11131 dev_get_dstats64(dev, storage); 11132 } else { 11133 netdev_stats_to_stats64(storage, &dev->stats); 11134 } 11135 11136 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11137 p = READ_ONCE(dev->core_stats); 11138 if (p) { 11139 const struct net_device_core_stats *core_stats; 11140 int i; 11141 11142 for_each_possible_cpu(i) { 11143 core_stats = per_cpu_ptr(p, i); 11144 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 11145 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 11146 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 11147 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 11148 } 11149 } 11150 return storage; 11151 } 11152 EXPORT_SYMBOL(dev_get_stats); 11153 11154 /** 11155 * dev_fetch_sw_netstats - get per-cpu network device statistics 11156 * @s: place to store stats 11157 * @netstats: per-cpu network stats to read from 11158 * 11159 * Read per-cpu network statistics and populate the related fields in @s. 11160 */ 11161 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 11162 const struct pcpu_sw_netstats __percpu *netstats) 11163 { 11164 int cpu; 11165 11166 for_each_possible_cpu(cpu) { 11167 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 11168 const struct pcpu_sw_netstats *stats; 11169 unsigned int start; 11170 11171 stats = per_cpu_ptr(netstats, cpu); 11172 do { 11173 start = u64_stats_fetch_begin(&stats->syncp); 11174 rx_packets = u64_stats_read(&stats->rx_packets); 11175 rx_bytes = u64_stats_read(&stats->rx_bytes); 11176 tx_packets = u64_stats_read(&stats->tx_packets); 11177 tx_bytes = u64_stats_read(&stats->tx_bytes); 11178 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11179 11180 s->rx_packets += rx_packets; 11181 s->rx_bytes += rx_bytes; 11182 s->tx_packets += tx_packets; 11183 s->tx_bytes += tx_bytes; 11184 } 11185 } 11186 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 11187 11188 /** 11189 * dev_get_tstats64 - ndo_get_stats64 implementation 11190 * @dev: device to get statistics from 11191 * @s: place to store stats 11192 * 11193 * Populate @s from dev->stats and dev->tstats. Can be used as 11194 * ndo_get_stats64() callback. 11195 */ 11196 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 11197 { 11198 netdev_stats_to_stats64(s, &dev->stats); 11199 dev_fetch_sw_netstats(s, dev->tstats); 11200 } 11201 EXPORT_SYMBOL_GPL(dev_get_tstats64); 11202 11203 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 11204 { 11205 struct netdev_queue *queue = dev_ingress_queue(dev); 11206 11207 #ifdef CONFIG_NET_CLS_ACT 11208 if (queue) 11209 return queue; 11210 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 11211 if (!queue) 11212 return NULL; 11213 netdev_init_one_queue(dev, queue, NULL); 11214 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 11215 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 11216 rcu_assign_pointer(dev->ingress_queue, queue); 11217 #endif 11218 return queue; 11219 } 11220 11221 static const struct ethtool_ops default_ethtool_ops; 11222 11223 void netdev_set_default_ethtool_ops(struct net_device *dev, 11224 const struct ethtool_ops *ops) 11225 { 11226 if (dev->ethtool_ops == &default_ethtool_ops) 11227 dev->ethtool_ops = ops; 11228 } 11229 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 11230 11231 /** 11232 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 11233 * @dev: netdev to enable the IRQ coalescing on 11234 * 11235 * Sets a conservative default for SW IRQ coalescing. Users can use 11236 * sysfs attributes to override the default values. 11237 */ 11238 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 11239 { 11240 WARN_ON(dev->reg_state == NETREG_REGISTERED); 11241 11242 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 11243 netdev_set_gro_flush_timeout(dev, 20000); 11244 netdev_set_defer_hard_irqs(dev, 1); 11245 } 11246 } 11247 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 11248 11249 /** 11250 * alloc_netdev_mqs - allocate network device 11251 * @sizeof_priv: size of private data to allocate space for 11252 * @name: device name format string 11253 * @name_assign_type: origin of device name 11254 * @setup: callback to initialize device 11255 * @txqs: the number of TX subqueues to allocate 11256 * @rxqs: the number of RX subqueues to allocate 11257 * 11258 * Allocates a struct net_device with private data area for driver use 11259 * and performs basic initialization. Also allocates subqueue structs 11260 * for each queue on the device. 11261 */ 11262 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 11263 unsigned char name_assign_type, 11264 void (*setup)(struct net_device *), 11265 unsigned int txqs, unsigned int rxqs) 11266 { 11267 struct net_device *dev; 11268 size_t napi_config_sz; 11269 unsigned int maxqs; 11270 11271 BUG_ON(strlen(name) >= sizeof(dev->name)); 11272 11273 if (txqs < 1) { 11274 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 11275 return NULL; 11276 } 11277 11278 if (rxqs < 1) { 11279 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 11280 return NULL; 11281 } 11282 11283 maxqs = max(txqs, rxqs); 11284 11285 dev = kvzalloc(struct_size(dev, priv, sizeof_priv), 11286 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 11287 if (!dev) 11288 return NULL; 11289 11290 dev->priv_len = sizeof_priv; 11291 11292 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 11293 #ifdef CONFIG_PCPU_DEV_REFCNT 11294 dev->pcpu_refcnt = alloc_percpu(int); 11295 if (!dev->pcpu_refcnt) 11296 goto free_dev; 11297 __dev_hold(dev); 11298 #else 11299 refcount_set(&dev->dev_refcnt, 1); 11300 #endif 11301 11302 if (dev_addr_init(dev)) 11303 goto free_pcpu; 11304 11305 dev_mc_init(dev); 11306 dev_uc_init(dev); 11307 11308 dev_net_set(dev, &init_net); 11309 11310 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 11311 dev->xdp_zc_max_segs = 1; 11312 dev->gso_max_segs = GSO_MAX_SEGS; 11313 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 11314 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 11315 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 11316 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 11317 dev->tso_max_segs = TSO_MAX_SEGS; 11318 dev->upper_level = 1; 11319 dev->lower_level = 1; 11320 #ifdef CONFIG_LOCKDEP 11321 dev->nested_level = 0; 11322 INIT_LIST_HEAD(&dev->unlink_list); 11323 #endif 11324 11325 INIT_LIST_HEAD(&dev->napi_list); 11326 INIT_LIST_HEAD(&dev->unreg_list); 11327 INIT_LIST_HEAD(&dev->close_list); 11328 INIT_LIST_HEAD(&dev->link_watch_list); 11329 INIT_LIST_HEAD(&dev->adj_list.upper); 11330 INIT_LIST_HEAD(&dev->adj_list.lower); 11331 INIT_LIST_HEAD(&dev->ptype_all); 11332 INIT_LIST_HEAD(&dev->ptype_specific); 11333 INIT_LIST_HEAD(&dev->net_notifier_list); 11334 #ifdef CONFIG_NET_SCHED 11335 hash_init(dev->qdisc_hash); 11336 #endif 11337 11338 mutex_init(&dev->lock); 11339 11340 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 11341 setup(dev); 11342 11343 if (!dev->tx_queue_len) { 11344 dev->priv_flags |= IFF_NO_QUEUE; 11345 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 11346 } 11347 11348 dev->num_tx_queues = txqs; 11349 dev->real_num_tx_queues = txqs; 11350 if (netif_alloc_netdev_queues(dev)) 11351 goto free_all; 11352 11353 dev->num_rx_queues = rxqs; 11354 dev->real_num_rx_queues = rxqs; 11355 if (netif_alloc_rx_queues(dev)) 11356 goto free_all; 11357 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT); 11358 if (!dev->ethtool) 11359 goto free_all; 11360 11361 napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config)); 11362 dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT); 11363 if (!dev->napi_config) 11364 goto free_all; 11365 11366 strscpy(dev->name, name); 11367 dev->name_assign_type = name_assign_type; 11368 dev->group = INIT_NETDEV_GROUP; 11369 if (!dev->ethtool_ops) 11370 dev->ethtool_ops = &default_ethtool_ops; 11371 11372 nf_hook_netdev_init(dev); 11373 11374 return dev; 11375 11376 free_all: 11377 free_netdev(dev); 11378 return NULL; 11379 11380 free_pcpu: 11381 #ifdef CONFIG_PCPU_DEV_REFCNT 11382 free_percpu(dev->pcpu_refcnt); 11383 free_dev: 11384 #endif 11385 kvfree(dev); 11386 return NULL; 11387 } 11388 EXPORT_SYMBOL(alloc_netdev_mqs); 11389 11390 /** 11391 * free_netdev - free network device 11392 * @dev: device 11393 * 11394 * This function does the last stage of destroying an allocated device 11395 * interface. The reference to the device object is released. If this 11396 * is the last reference then it will be freed.Must be called in process 11397 * context. 11398 */ 11399 void free_netdev(struct net_device *dev) 11400 { 11401 struct napi_struct *p, *n; 11402 11403 might_sleep(); 11404 11405 /* When called immediately after register_netdevice() failed the unwind 11406 * handling may still be dismantling the device. Handle that case by 11407 * deferring the free. 11408 */ 11409 if (dev->reg_state == NETREG_UNREGISTERING) { 11410 ASSERT_RTNL(); 11411 dev->needs_free_netdev = true; 11412 return; 11413 } 11414 11415 mutex_destroy(&dev->lock); 11416 11417 kfree(dev->ethtool); 11418 netif_free_tx_queues(dev); 11419 netif_free_rx_queues(dev); 11420 11421 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 11422 11423 /* Flush device addresses */ 11424 dev_addr_flush(dev); 11425 11426 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 11427 netif_napi_del(p); 11428 11429 kvfree(dev->napi_config); 11430 11431 ref_tracker_dir_exit(&dev->refcnt_tracker); 11432 #ifdef CONFIG_PCPU_DEV_REFCNT 11433 free_percpu(dev->pcpu_refcnt); 11434 dev->pcpu_refcnt = NULL; 11435 #endif 11436 free_percpu(dev->core_stats); 11437 dev->core_stats = NULL; 11438 free_percpu(dev->xdp_bulkq); 11439 dev->xdp_bulkq = NULL; 11440 11441 netdev_free_phy_link_topology(dev); 11442 11443 /* Compatibility with error handling in drivers */ 11444 if (dev->reg_state == NETREG_UNINITIALIZED || 11445 dev->reg_state == NETREG_DUMMY) { 11446 kvfree(dev); 11447 return; 11448 } 11449 11450 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 11451 WRITE_ONCE(dev->reg_state, NETREG_RELEASED); 11452 11453 /* will free via device release */ 11454 put_device(&dev->dev); 11455 } 11456 EXPORT_SYMBOL(free_netdev); 11457 11458 /** 11459 * alloc_netdev_dummy - Allocate and initialize a dummy net device. 11460 * @sizeof_priv: size of private data to allocate space for 11461 * 11462 * Return: the allocated net_device on success, NULL otherwise 11463 */ 11464 struct net_device *alloc_netdev_dummy(int sizeof_priv) 11465 { 11466 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN, 11467 init_dummy_netdev_core); 11468 } 11469 EXPORT_SYMBOL_GPL(alloc_netdev_dummy); 11470 11471 /** 11472 * synchronize_net - Synchronize with packet receive processing 11473 * 11474 * Wait for packets currently being received to be done. 11475 * Does not block later packets from starting. 11476 */ 11477 void synchronize_net(void) 11478 { 11479 might_sleep(); 11480 if (rtnl_is_locked()) 11481 synchronize_rcu_expedited(); 11482 else 11483 synchronize_rcu(); 11484 } 11485 EXPORT_SYMBOL(synchronize_net); 11486 11487 static void netdev_rss_contexts_free(struct net_device *dev) 11488 { 11489 struct ethtool_rxfh_context *ctx; 11490 unsigned long context; 11491 11492 mutex_lock(&dev->ethtool->rss_lock); 11493 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) { 11494 struct ethtool_rxfh_param rxfh; 11495 11496 rxfh.indir = ethtool_rxfh_context_indir(ctx); 11497 rxfh.key = ethtool_rxfh_context_key(ctx); 11498 rxfh.hfunc = ctx->hfunc; 11499 rxfh.input_xfrm = ctx->input_xfrm; 11500 rxfh.rss_context = context; 11501 rxfh.rss_delete = true; 11502 11503 xa_erase(&dev->ethtool->rss_ctx, context); 11504 if (dev->ethtool_ops->create_rxfh_context) 11505 dev->ethtool_ops->remove_rxfh_context(dev, ctx, 11506 context, NULL); 11507 else 11508 dev->ethtool_ops->set_rxfh(dev, &rxfh, NULL); 11509 kfree(ctx); 11510 } 11511 xa_destroy(&dev->ethtool->rss_ctx); 11512 mutex_unlock(&dev->ethtool->rss_lock); 11513 } 11514 11515 /** 11516 * unregister_netdevice_queue - remove device from the kernel 11517 * @dev: device 11518 * @head: list 11519 * 11520 * This function shuts down a device interface and removes it 11521 * from the kernel tables. 11522 * If head not NULL, device is queued to be unregistered later. 11523 * 11524 * Callers must hold the rtnl semaphore. You may want 11525 * unregister_netdev() instead of this. 11526 */ 11527 11528 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11529 { 11530 ASSERT_RTNL(); 11531 11532 if (head) { 11533 list_move_tail(&dev->unreg_list, head); 11534 } else { 11535 LIST_HEAD(single); 11536 11537 list_add(&dev->unreg_list, &single); 11538 unregister_netdevice_many(&single); 11539 } 11540 } 11541 EXPORT_SYMBOL(unregister_netdevice_queue); 11542 11543 void unregister_netdevice_many_notify(struct list_head *head, 11544 u32 portid, const struct nlmsghdr *nlh) 11545 { 11546 struct net_device *dev, *tmp; 11547 LIST_HEAD(close_head); 11548 int cnt = 0; 11549 11550 BUG_ON(dev_boot_phase); 11551 ASSERT_RTNL(); 11552 11553 if (list_empty(head)) 11554 return; 11555 11556 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11557 /* Some devices call without registering 11558 * for initialization unwind. Remove those 11559 * devices and proceed with the remaining. 11560 */ 11561 if (dev->reg_state == NETREG_UNINITIALIZED) { 11562 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11563 dev->name, dev); 11564 11565 WARN_ON(1); 11566 list_del(&dev->unreg_list); 11567 continue; 11568 } 11569 dev->dismantle = true; 11570 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11571 } 11572 11573 /* If device is running, close it first. */ 11574 list_for_each_entry(dev, head, unreg_list) 11575 list_add_tail(&dev->close_list, &close_head); 11576 dev_close_many(&close_head, true); 11577 11578 list_for_each_entry(dev, head, unreg_list) { 11579 /* And unlink it from device chain. */ 11580 unlist_netdevice(dev); 11581 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); 11582 } 11583 flush_all_backlogs(); 11584 11585 synchronize_net(); 11586 11587 list_for_each_entry(dev, head, unreg_list) { 11588 struct sk_buff *skb = NULL; 11589 11590 /* Shutdown queueing discipline. */ 11591 dev_shutdown(dev); 11592 dev_tcx_uninstall(dev); 11593 dev_xdp_uninstall(dev); 11594 bpf_dev_bound_netdev_unregister(dev); 11595 dev_dmabuf_uninstall(dev); 11596 11597 netdev_offload_xstats_disable_all(dev); 11598 11599 /* Notify protocols, that we are about to destroy 11600 * this device. They should clean all the things. 11601 */ 11602 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11603 11604 if (!dev->rtnl_link_ops || 11605 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11606 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11607 GFP_KERNEL, NULL, 0, 11608 portid, nlh); 11609 11610 /* 11611 * Flush the unicast and multicast chains 11612 */ 11613 dev_uc_flush(dev); 11614 dev_mc_flush(dev); 11615 11616 netdev_name_node_alt_flush(dev); 11617 netdev_name_node_free(dev->name_node); 11618 11619 netdev_rss_contexts_free(dev); 11620 11621 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11622 11623 if (dev->netdev_ops->ndo_uninit) 11624 dev->netdev_ops->ndo_uninit(dev); 11625 11626 mutex_destroy(&dev->ethtool->rss_lock); 11627 11628 net_shaper_flush_netdev(dev); 11629 11630 if (skb) 11631 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11632 11633 /* Notifier chain MUST detach us all upper devices. */ 11634 WARN_ON(netdev_has_any_upper_dev(dev)); 11635 WARN_ON(netdev_has_any_lower_dev(dev)); 11636 11637 /* Remove entries from kobject tree */ 11638 netdev_unregister_kobject(dev); 11639 #ifdef CONFIG_XPS 11640 /* Remove XPS queueing entries */ 11641 netif_reset_xps_queues_gt(dev, 0); 11642 #endif 11643 } 11644 11645 synchronize_net(); 11646 11647 list_for_each_entry(dev, head, unreg_list) { 11648 netdev_put(dev, &dev->dev_registered_tracker); 11649 net_set_todo(dev); 11650 cnt++; 11651 } 11652 atomic_add(cnt, &dev_unreg_count); 11653 11654 list_del(head); 11655 } 11656 11657 /** 11658 * unregister_netdevice_many - unregister many devices 11659 * @head: list of devices 11660 * 11661 * Note: As most callers use a stack allocated list_head, 11662 * we force a list_del() to make sure stack won't be corrupted later. 11663 */ 11664 void unregister_netdevice_many(struct list_head *head) 11665 { 11666 unregister_netdevice_many_notify(head, 0, NULL); 11667 } 11668 EXPORT_SYMBOL(unregister_netdevice_many); 11669 11670 /** 11671 * unregister_netdev - remove device from the kernel 11672 * @dev: device 11673 * 11674 * This function shuts down a device interface and removes it 11675 * from the kernel tables. 11676 * 11677 * This is just a wrapper for unregister_netdevice that takes 11678 * the rtnl semaphore. In general you want to use this and not 11679 * unregister_netdevice. 11680 */ 11681 void unregister_netdev(struct net_device *dev) 11682 { 11683 struct net *net = dev_net(dev); 11684 11685 rtnl_net_lock(net); 11686 unregister_netdevice(dev); 11687 rtnl_net_unlock(net); 11688 } 11689 EXPORT_SYMBOL(unregister_netdev); 11690 11691 /** 11692 * __dev_change_net_namespace - move device to different nethost namespace 11693 * @dev: device 11694 * @net: network namespace 11695 * @pat: If not NULL name pattern to try if the current device name 11696 * is already taken in the destination network namespace. 11697 * @new_ifindex: If not zero, specifies device index in the target 11698 * namespace. 11699 * 11700 * This function shuts down a device interface and moves it 11701 * to a new network namespace. On success 0 is returned, on 11702 * a failure a netagive errno code is returned. 11703 * 11704 * Callers must hold the rtnl semaphore. 11705 */ 11706 11707 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11708 const char *pat, int new_ifindex) 11709 { 11710 struct netdev_name_node *name_node; 11711 struct net *net_old = dev_net(dev); 11712 char new_name[IFNAMSIZ] = {}; 11713 int err, new_nsid; 11714 11715 ASSERT_RTNL(); 11716 11717 /* Don't allow namespace local devices to be moved. */ 11718 err = -EINVAL; 11719 if (dev->netns_local) 11720 goto out; 11721 11722 /* Ensure the device has been registered */ 11723 if (dev->reg_state != NETREG_REGISTERED) 11724 goto out; 11725 11726 /* Get out if there is nothing todo */ 11727 err = 0; 11728 if (net_eq(net_old, net)) 11729 goto out; 11730 11731 /* Pick the destination device name, and ensure 11732 * we can use it in the destination network namespace. 11733 */ 11734 err = -EEXIST; 11735 if (netdev_name_in_use(net, dev->name)) { 11736 /* We get here if we can't use the current device name */ 11737 if (!pat) 11738 goto out; 11739 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11740 if (err < 0) 11741 goto out; 11742 } 11743 /* Check that none of the altnames conflicts. */ 11744 err = -EEXIST; 11745 netdev_for_each_altname(dev, name_node) 11746 if (netdev_name_in_use(net, name_node->name)) 11747 goto out; 11748 11749 /* Check that new_ifindex isn't used yet. */ 11750 if (new_ifindex) { 11751 err = dev_index_reserve(net, new_ifindex); 11752 if (err < 0) 11753 goto out; 11754 } else { 11755 /* If there is an ifindex conflict assign a new one */ 11756 err = dev_index_reserve(net, dev->ifindex); 11757 if (err == -EBUSY) 11758 err = dev_index_reserve(net, 0); 11759 if (err < 0) 11760 goto out; 11761 new_ifindex = err; 11762 } 11763 11764 /* 11765 * And now a mini version of register_netdevice unregister_netdevice. 11766 */ 11767 11768 /* If device is running close it first. */ 11769 dev_close(dev); 11770 11771 /* And unlink it from device chain */ 11772 unlist_netdevice(dev); 11773 11774 synchronize_net(); 11775 11776 /* Shutdown queueing discipline. */ 11777 dev_shutdown(dev); 11778 11779 /* Notify protocols, that we are about to destroy 11780 * this device. They should clean all the things. 11781 * 11782 * Note that dev->reg_state stays at NETREG_REGISTERED. 11783 * This is wanted because this way 8021q and macvlan know 11784 * the device is just moving and can keep their slaves up. 11785 */ 11786 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11787 rcu_barrier(); 11788 11789 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11790 11791 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11792 new_ifindex); 11793 11794 /* 11795 * Flush the unicast and multicast chains 11796 */ 11797 dev_uc_flush(dev); 11798 dev_mc_flush(dev); 11799 11800 /* Send a netdev-removed uevent to the old namespace */ 11801 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11802 netdev_adjacent_del_links(dev); 11803 11804 /* Move per-net netdevice notifiers that are following the netdevice */ 11805 move_netdevice_notifiers_dev_net(dev, net); 11806 11807 /* Actually switch the network namespace */ 11808 dev_net_set(dev, net); 11809 dev->ifindex = new_ifindex; 11810 11811 if (new_name[0]) { 11812 /* Rename the netdev to prepared name */ 11813 write_seqlock_bh(&netdev_rename_lock); 11814 strscpy(dev->name, new_name, IFNAMSIZ); 11815 write_sequnlock_bh(&netdev_rename_lock); 11816 } 11817 11818 /* Fixup kobjects */ 11819 dev_set_uevent_suppress(&dev->dev, 1); 11820 err = device_rename(&dev->dev, dev->name); 11821 dev_set_uevent_suppress(&dev->dev, 0); 11822 WARN_ON(err); 11823 11824 /* Send a netdev-add uevent to the new namespace */ 11825 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11826 netdev_adjacent_add_links(dev); 11827 11828 /* Adapt owner in case owning user namespace of target network 11829 * namespace is different from the original one. 11830 */ 11831 err = netdev_change_owner(dev, net_old, net); 11832 WARN_ON(err); 11833 11834 /* Add the device back in the hashes */ 11835 list_netdevice(dev); 11836 11837 /* Notify protocols, that a new device appeared. */ 11838 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11839 11840 /* 11841 * Prevent userspace races by waiting until the network 11842 * device is fully setup before sending notifications. 11843 */ 11844 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11845 11846 synchronize_net(); 11847 err = 0; 11848 out: 11849 return err; 11850 } 11851 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11852 11853 static int dev_cpu_dead(unsigned int oldcpu) 11854 { 11855 struct sk_buff **list_skb; 11856 struct sk_buff *skb; 11857 unsigned int cpu; 11858 struct softnet_data *sd, *oldsd, *remsd = NULL; 11859 11860 local_irq_disable(); 11861 cpu = smp_processor_id(); 11862 sd = &per_cpu(softnet_data, cpu); 11863 oldsd = &per_cpu(softnet_data, oldcpu); 11864 11865 /* Find end of our completion_queue. */ 11866 list_skb = &sd->completion_queue; 11867 while (*list_skb) 11868 list_skb = &(*list_skb)->next; 11869 /* Append completion queue from offline CPU. */ 11870 *list_skb = oldsd->completion_queue; 11871 oldsd->completion_queue = NULL; 11872 11873 /* Append output queue from offline CPU. */ 11874 if (oldsd->output_queue) { 11875 *sd->output_queue_tailp = oldsd->output_queue; 11876 sd->output_queue_tailp = oldsd->output_queue_tailp; 11877 oldsd->output_queue = NULL; 11878 oldsd->output_queue_tailp = &oldsd->output_queue; 11879 } 11880 /* Append NAPI poll list from offline CPU, with one exception : 11881 * process_backlog() must be called by cpu owning percpu backlog. 11882 * We properly handle process_queue & input_pkt_queue later. 11883 */ 11884 while (!list_empty(&oldsd->poll_list)) { 11885 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11886 struct napi_struct, 11887 poll_list); 11888 11889 list_del_init(&napi->poll_list); 11890 if (napi->poll == process_backlog) 11891 napi->state &= NAPIF_STATE_THREADED; 11892 else 11893 ____napi_schedule(sd, napi); 11894 } 11895 11896 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11897 local_irq_enable(); 11898 11899 if (!use_backlog_threads()) { 11900 #ifdef CONFIG_RPS 11901 remsd = oldsd->rps_ipi_list; 11902 oldsd->rps_ipi_list = NULL; 11903 #endif 11904 /* send out pending IPI's on offline CPU */ 11905 net_rps_send_ipi(remsd); 11906 } 11907 11908 /* Process offline CPU's input_pkt_queue */ 11909 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11910 netif_rx(skb); 11911 rps_input_queue_head_incr(oldsd); 11912 } 11913 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11914 netif_rx(skb); 11915 rps_input_queue_head_incr(oldsd); 11916 } 11917 11918 return 0; 11919 } 11920 11921 /** 11922 * netdev_increment_features - increment feature set by one 11923 * @all: current feature set 11924 * @one: new feature set 11925 * @mask: mask feature set 11926 * 11927 * Computes a new feature set after adding a device with feature set 11928 * @one to the master device with current feature set @all. Will not 11929 * enable anything that is off in @mask. Returns the new feature set. 11930 */ 11931 netdev_features_t netdev_increment_features(netdev_features_t all, 11932 netdev_features_t one, netdev_features_t mask) 11933 { 11934 if (mask & NETIF_F_HW_CSUM) 11935 mask |= NETIF_F_CSUM_MASK; 11936 mask |= NETIF_F_VLAN_CHALLENGED; 11937 11938 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11939 all &= one | ~NETIF_F_ALL_FOR_ALL; 11940 11941 /* If one device supports hw checksumming, set for all. */ 11942 if (all & NETIF_F_HW_CSUM) 11943 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11944 11945 return all; 11946 } 11947 EXPORT_SYMBOL(netdev_increment_features); 11948 11949 static struct hlist_head * __net_init netdev_create_hash(void) 11950 { 11951 int i; 11952 struct hlist_head *hash; 11953 11954 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11955 if (hash != NULL) 11956 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11957 INIT_HLIST_HEAD(&hash[i]); 11958 11959 return hash; 11960 } 11961 11962 /* Initialize per network namespace state */ 11963 static int __net_init netdev_init(struct net *net) 11964 { 11965 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11966 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11967 11968 INIT_LIST_HEAD(&net->dev_base_head); 11969 11970 net->dev_name_head = netdev_create_hash(); 11971 if (net->dev_name_head == NULL) 11972 goto err_name; 11973 11974 net->dev_index_head = netdev_create_hash(); 11975 if (net->dev_index_head == NULL) 11976 goto err_idx; 11977 11978 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11979 11980 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11981 11982 return 0; 11983 11984 err_idx: 11985 kfree(net->dev_name_head); 11986 err_name: 11987 return -ENOMEM; 11988 } 11989 11990 /** 11991 * netdev_drivername - network driver for the device 11992 * @dev: network device 11993 * 11994 * Determine network driver for device. 11995 */ 11996 const char *netdev_drivername(const struct net_device *dev) 11997 { 11998 const struct device_driver *driver; 11999 const struct device *parent; 12000 const char *empty = ""; 12001 12002 parent = dev->dev.parent; 12003 if (!parent) 12004 return empty; 12005 12006 driver = parent->driver; 12007 if (driver && driver->name) 12008 return driver->name; 12009 return empty; 12010 } 12011 12012 static void __netdev_printk(const char *level, const struct net_device *dev, 12013 struct va_format *vaf) 12014 { 12015 if (dev && dev->dev.parent) { 12016 dev_printk_emit(level[1] - '0', 12017 dev->dev.parent, 12018 "%s %s %s%s: %pV", 12019 dev_driver_string(dev->dev.parent), 12020 dev_name(dev->dev.parent), 12021 netdev_name(dev), netdev_reg_state(dev), 12022 vaf); 12023 } else if (dev) { 12024 printk("%s%s%s: %pV", 12025 level, netdev_name(dev), netdev_reg_state(dev), vaf); 12026 } else { 12027 printk("%s(NULL net_device): %pV", level, vaf); 12028 } 12029 } 12030 12031 void netdev_printk(const char *level, const struct net_device *dev, 12032 const char *format, ...) 12033 { 12034 struct va_format vaf; 12035 va_list args; 12036 12037 va_start(args, format); 12038 12039 vaf.fmt = format; 12040 vaf.va = &args; 12041 12042 __netdev_printk(level, dev, &vaf); 12043 12044 va_end(args); 12045 } 12046 EXPORT_SYMBOL(netdev_printk); 12047 12048 #define define_netdev_printk_level(func, level) \ 12049 void func(const struct net_device *dev, const char *fmt, ...) \ 12050 { \ 12051 struct va_format vaf; \ 12052 va_list args; \ 12053 \ 12054 va_start(args, fmt); \ 12055 \ 12056 vaf.fmt = fmt; \ 12057 vaf.va = &args; \ 12058 \ 12059 __netdev_printk(level, dev, &vaf); \ 12060 \ 12061 va_end(args); \ 12062 } \ 12063 EXPORT_SYMBOL(func); 12064 12065 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 12066 define_netdev_printk_level(netdev_alert, KERN_ALERT); 12067 define_netdev_printk_level(netdev_crit, KERN_CRIT); 12068 define_netdev_printk_level(netdev_err, KERN_ERR); 12069 define_netdev_printk_level(netdev_warn, KERN_WARNING); 12070 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 12071 define_netdev_printk_level(netdev_info, KERN_INFO); 12072 12073 static void __net_exit netdev_exit(struct net *net) 12074 { 12075 kfree(net->dev_name_head); 12076 kfree(net->dev_index_head); 12077 xa_destroy(&net->dev_by_index); 12078 if (net != &init_net) 12079 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 12080 } 12081 12082 static struct pernet_operations __net_initdata netdev_net_ops = { 12083 .init = netdev_init, 12084 .exit = netdev_exit, 12085 }; 12086 12087 static void __net_exit default_device_exit_net(struct net *net) 12088 { 12089 struct netdev_name_node *name_node, *tmp; 12090 struct net_device *dev, *aux; 12091 /* 12092 * Push all migratable network devices back to the 12093 * initial network namespace 12094 */ 12095 ASSERT_RTNL(); 12096 for_each_netdev_safe(net, dev, aux) { 12097 int err; 12098 char fb_name[IFNAMSIZ]; 12099 12100 /* Ignore unmoveable devices (i.e. loopback) */ 12101 if (dev->netns_local) 12102 continue; 12103 12104 /* Leave virtual devices for the generic cleanup */ 12105 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 12106 continue; 12107 12108 /* Push remaining network devices to init_net */ 12109 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 12110 if (netdev_name_in_use(&init_net, fb_name)) 12111 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 12112 12113 netdev_for_each_altname_safe(dev, name_node, tmp) 12114 if (netdev_name_in_use(&init_net, name_node->name)) 12115 __netdev_name_node_alt_destroy(name_node); 12116 12117 err = dev_change_net_namespace(dev, &init_net, fb_name); 12118 if (err) { 12119 pr_emerg("%s: failed to move %s to init_net: %d\n", 12120 __func__, dev->name, err); 12121 BUG(); 12122 } 12123 } 12124 } 12125 12126 static void __net_exit default_device_exit_batch(struct list_head *net_list) 12127 { 12128 /* At exit all network devices most be removed from a network 12129 * namespace. Do this in the reverse order of registration. 12130 * Do this across as many network namespaces as possible to 12131 * improve batching efficiency. 12132 */ 12133 struct net_device *dev; 12134 struct net *net; 12135 LIST_HEAD(dev_kill_list); 12136 12137 rtnl_lock(); 12138 list_for_each_entry(net, net_list, exit_list) { 12139 default_device_exit_net(net); 12140 cond_resched(); 12141 } 12142 12143 list_for_each_entry(net, net_list, exit_list) { 12144 for_each_netdev_reverse(net, dev) { 12145 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 12146 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 12147 else 12148 unregister_netdevice_queue(dev, &dev_kill_list); 12149 } 12150 } 12151 unregister_netdevice_many(&dev_kill_list); 12152 rtnl_unlock(); 12153 } 12154 12155 static struct pernet_operations __net_initdata default_device_ops = { 12156 .exit_batch = default_device_exit_batch, 12157 }; 12158 12159 static void __init net_dev_struct_check(void) 12160 { 12161 /* TX read-mostly hotpath */ 12162 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags_fast); 12163 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 12164 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 12165 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 12166 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 12167 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 12168 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 12169 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 12170 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 12171 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 12172 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 12173 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 12174 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 12175 #ifdef CONFIG_XPS 12176 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 12177 #endif 12178 #ifdef CONFIG_NETFILTER_EGRESS 12179 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 12180 #endif 12181 #ifdef CONFIG_NET_XGRESS 12182 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 12183 #endif 12184 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 12185 12186 /* TXRX read-mostly hotpath */ 12187 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); 12188 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state); 12189 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 12190 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 12191 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 12192 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 12193 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46); 12194 12195 /* RX read-mostly hotpath */ 12196 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 12197 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 12198 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 12199 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 12200 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 12201 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 12202 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 12203 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 12204 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 12205 #ifdef CONFIG_NETPOLL 12206 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 12207 #endif 12208 #ifdef CONFIG_NET_XGRESS 12209 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 12210 #endif 12211 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 92); 12212 } 12213 12214 /* 12215 * Initialize the DEV module. At boot time this walks the device list and 12216 * unhooks any devices that fail to initialise (normally hardware not 12217 * present) and leaves us with a valid list of present and active devices. 12218 * 12219 */ 12220 12221 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ 12222 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) 12223 12224 static int net_page_pool_create(int cpuid) 12225 { 12226 #if IS_ENABLED(CONFIG_PAGE_POOL) 12227 struct page_pool_params page_pool_params = { 12228 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, 12229 .flags = PP_FLAG_SYSTEM_POOL, 12230 .nid = cpu_to_mem(cpuid), 12231 }; 12232 struct page_pool *pp_ptr; 12233 int err; 12234 12235 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); 12236 if (IS_ERR(pp_ptr)) 12237 return -ENOMEM; 12238 12239 err = xdp_reg_page_pool(pp_ptr); 12240 if (err) { 12241 page_pool_destroy(pp_ptr); 12242 return err; 12243 } 12244 12245 per_cpu(system_page_pool, cpuid) = pp_ptr; 12246 #endif 12247 return 0; 12248 } 12249 12250 static int backlog_napi_should_run(unsigned int cpu) 12251 { 12252 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12253 struct napi_struct *napi = &sd->backlog; 12254 12255 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 12256 } 12257 12258 static void run_backlog_napi(unsigned int cpu) 12259 { 12260 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12261 12262 napi_threaded_poll_loop(&sd->backlog); 12263 } 12264 12265 static void backlog_napi_setup(unsigned int cpu) 12266 { 12267 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12268 struct napi_struct *napi = &sd->backlog; 12269 12270 napi->thread = this_cpu_read(backlog_napi); 12271 set_bit(NAPI_STATE_THREADED, &napi->state); 12272 } 12273 12274 static struct smp_hotplug_thread backlog_threads = { 12275 .store = &backlog_napi, 12276 .thread_should_run = backlog_napi_should_run, 12277 .thread_fn = run_backlog_napi, 12278 .thread_comm = "backlog_napi/%u", 12279 .setup = backlog_napi_setup, 12280 }; 12281 12282 /* 12283 * This is called single threaded during boot, so no need 12284 * to take the rtnl semaphore. 12285 */ 12286 static int __init net_dev_init(void) 12287 { 12288 int i, rc = -ENOMEM; 12289 12290 BUG_ON(!dev_boot_phase); 12291 12292 net_dev_struct_check(); 12293 12294 if (dev_proc_init()) 12295 goto out; 12296 12297 if (netdev_kobject_init()) 12298 goto out; 12299 12300 for (i = 0; i < PTYPE_HASH_SIZE; i++) 12301 INIT_LIST_HEAD(&ptype_base[i]); 12302 12303 if (register_pernet_subsys(&netdev_net_ops)) 12304 goto out; 12305 12306 /* 12307 * Initialise the packet receive queues. 12308 */ 12309 12310 for_each_possible_cpu(i) { 12311 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 12312 struct softnet_data *sd = &per_cpu(softnet_data, i); 12313 12314 INIT_WORK(flush, flush_backlog); 12315 12316 skb_queue_head_init(&sd->input_pkt_queue); 12317 skb_queue_head_init(&sd->process_queue); 12318 #ifdef CONFIG_XFRM_OFFLOAD 12319 skb_queue_head_init(&sd->xfrm_backlog); 12320 #endif 12321 INIT_LIST_HEAD(&sd->poll_list); 12322 sd->output_queue_tailp = &sd->output_queue; 12323 #ifdef CONFIG_RPS 12324 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 12325 sd->cpu = i; 12326 #endif 12327 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 12328 spin_lock_init(&sd->defer_lock); 12329 12330 init_gro_hash(&sd->backlog); 12331 sd->backlog.poll = process_backlog; 12332 sd->backlog.weight = weight_p; 12333 INIT_LIST_HEAD(&sd->backlog.poll_list); 12334 12335 if (net_page_pool_create(i)) 12336 goto out; 12337 } 12338 if (use_backlog_threads()) 12339 smpboot_register_percpu_thread(&backlog_threads); 12340 12341 dev_boot_phase = 0; 12342 12343 /* The loopback device is special if any other network devices 12344 * is present in a network namespace the loopback device must 12345 * be present. Since we now dynamically allocate and free the 12346 * loopback device ensure this invariant is maintained by 12347 * keeping the loopback device as the first device on the 12348 * list of network devices. Ensuring the loopback devices 12349 * is the first device that appears and the last network device 12350 * that disappears. 12351 */ 12352 if (register_pernet_device(&loopback_net_ops)) 12353 goto out; 12354 12355 if (register_pernet_device(&default_device_ops)) 12356 goto out; 12357 12358 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 12359 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 12360 12361 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 12362 NULL, dev_cpu_dead); 12363 WARN_ON(rc < 0); 12364 rc = 0; 12365 12366 /* avoid static key IPIs to isolated CPUs */ 12367 if (housekeeping_enabled(HK_TYPE_MISC)) 12368 net_enable_timestamp(); 12369 out: 12370 if (rc < 0) { 12371 for_each_possible_cpu(i) { 12372 struct page_pool *pp_ptr; 12373 12374 pp_ptr = per_cpu(system_page_pool, i); 12375 if (!pp_ptr) 12376 continue; 12377 12378 xdp_unreg_page_pool(pp_ptr); 12379 page_pool_destroy(pp_ptr); 12380 per_cpu(system_page_pool, i) = NULL; 12381 } 12382 } 12383 12384 return rc; 12385 } 12386 12387 subsys_initcall(net_dev_init); 12388