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