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, <[email protected]> 8 * Mark Evans, <[email protected]> 9 * 10 * Additional Authors: 11 * Florian la Roche <[email protected]> 12 * Alan Cox <[email protected]> 13 * David Hinds <[email protected]> 14 * Alexey Kuznetsov <[email protected]> 15 * Adam Sulmicki <[email protected]> 16 * Pekka Riikonen <[email protected]> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitmap.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/isolation.h> 81 #include <linux/sched/mm.h> 82 #include <linux/smpboot.h> 83 #include <linux/mutex.h> 84 #include <linux/rwsem.h> 85 #include <linux/string.h> 86 #include <linux/mm.h> 87 #include <linux/socket.h> 88 #include <linux/sockios.h> 89 #include <linux/errno.h> 90 #include <linux/interrupt.h> 91 #include <linux/if_ether.h> 92 #include <linux/netdevice.h> 93 #include <linux/etherdevice.h> 94 #include <linux/ethtool.h> 95 #include <linux/ethtool_netlink.h> 96 #include <linux/skbuff.h> 97 #include <linux/kthread.h> 98 #include <linux/bpf.h> 99 #include <linux/bpf_trace.h> 100 #include <net/net_namespace.h> 101 #include <net/sock.h> 102 #include <net/busy_poll.h> 103 #include <linux/rtnetlink.h> 104 #include <linux/stat.h> 105 #include <net/dsa.h> 106 #include <net/dst.h> 107 #include <net/dst_metadata.h> 108 #include <net/gro.h> 109 #include <net/pkt_sched.h> 110 #include <net/pkt_cls.h> 111 #include <net/checksum.h> 112 #include <net/xfrm.h> 113 #include <net/tcx.h> 114 #include <linux/highmem.h> 115 #include <linux/init.h> 116 #include <linux/module.h> 117 #include <linux/netpoll.h> 118 #include <linux/rcupdate.h> 119 #include <linux/delay.h> 120 #include <net/iw_handler.h> 121 #include <asm/current.h> 122 #include <linux/audit.h> 123 #include <linux/dmaengine.h> 124 #include <linux/err.h> 125 #include <linux/ctype.h> 126 #include <linux/if_arp.h> 127 #include <linux/if_vlan.h> 128 #include <linux/ip.h> 129 #include <net/ip.h> 130 #include <net/mpls.h> 131 #include <linux/ipv6.h> 132 #include <linux/in.h> 133 #include <linux/jhash.h> 134 #include <linux/random.h> 135 #include <trace/events/napi.h> 136 #include <trace/events/net.h> 137 #include <trace/events/skb.h> 138 #include <trace/events/qdisc.h> 139 #include <trace/events/xdp.h> 140 #include <linux/inetdevice.h> 141 #include <linux/cpu_rmap.h> 142 #include <linux/static_key.h> 143 #include <linux/hashtable.h> 144 #include <linux/vmalloc.h> 145 #include <linux/if_macvlan.h> 146 #include <linux/errqueue.h> 147 #include <linux/hrtimer.h> 148 #include <linux/netfilter_netdev.h> 149 #include <linux/crash_dump.h> 150 #include <linux/sctp.h> 151 #include <net/udp_tunnel.h> 152 #include <linux/net_namespace.h> 153 #include <linux/indirect_call_wrapper.h> 154 #include <net/devlink.h> 155 #include <linux/pm_runtime.h> 156 #include <linux/prandom.h> 157 #include <linux/once_lite.h> 158 #include <net/netdev_rx_queue.h> 159 #include <net/page_pool/types.h> 160 #include <net/page_pool/helpers.h> 161 #include <net/rps.h> 162 #include <linux/phy_link_topology.h> 163 164 #include "dev.h" 165 #include "devmem.h" 166 #include "net-sysfs.h" 167 168 static DEFINE_SPINLOCK(ptype_lock); 169 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 170 171 static int netif_rx_internal(struct sk_buff *skb); 172 static int call_netdevice_notifiers_extack(unsigned long val, 173 struct net_device *dev, 174 struct netlink_ext_ack *extack); 175 176 static DEFINE_MUTEX(ifalias_mutex); 177 178 /* protects napi_hash addition/deletion and napi_gen_id */ 179 static DEFINE_SPINLOCK(napi_hash_lock); 180 181 static unsigned int napi_gen_id = NR_CPUS; 182 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 183 184 static DECLARE_RWSEM(devnet_rename_sem); 185 186 static inline void dev_base_seq_inc(struct net *net) 187 { 188 unsigned int val = net->dev_base_seq + 1; 189 190 WRITE_ONCE(net->dev_base_seq, val ?: 1); 191 } 192 193 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 194 { 195 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 196 197 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 198 } 199 200 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 201 { 202 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 203 } 204 205 #ifndef CONFIG_PREEMPT_RT 206 207 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key); 208 209 static int __init setup_backlog_napi_threads(char *arg) 210 { 211 static_branch_enable(&use_backlog_threads_key); 212 return 0; 213 } 214 early_param("thread_backlog_napi", setup_backlog_napi_threads); 215 216 static bool use_backlog_threads(void) 217 { 218 return static_branch_unlikely(&use_backlog_threads_key); 219 } 220 221 #else 222 223 static bool use_backlog_threads(void) 224 { 225 return true; 226 } 227 228 #endif 229 230 static inline void backlog_lock_irq_save(struct softnet_data *sd, 231 unsigned long *flags) 232 { 233 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 234 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 235 else 236 local_irq_save(*flags); 237 } 238 239 static inline void backlog_lock_irq_disable(struct softnet_data *sd) 240 { 241 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 242 spin_lock_irq(&sd->input_pkt_queue.lock); 243 else 244 local_irq_disable(); 245 } 246 247 static inline void backlog_unlock_irq_restore(struct softnet_data *sd, 248 unsigned long *flags) 249 { 250 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 251 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 252 else 253 local_irq_restore(*flags); 254 } 255 256 static inline void backlog_unlock_irq_enable(struct softnet_data *sd) 257 { 258 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 259 spin_unlock_irq(&sd->input_pkt_queue.lock); 260 else 261 local_irq_enable(); 262 } 263 264 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 265 const char *name) 266 { 267 struct netdev_name_node *name_node; 268 269 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 270 if (!name_node) 271 return NULL; 272 INIT_HLIST_NODE(&name_node->hlist); 273 name_node->dev = dev; 274 name_node->name = name; 275 return name_node; 276 } 277 278 static struct netdev_name_node * 279 netdev_name_node_head_alloc(struct net_device *dev) 280 { 281 struct netdev_name_node *name_node; 282 283 name_node = netdev_name_node_alloc(dev, dev->name); 284 if (!name_node) 285 return NULL; 286 INIT_LIST_HEAD(&name_node->list); 287 return name_node; 288 } 289 290 static void netdev_name_node_free(struct netdev_name_node *name_node) 291 { 292 kfree(name_node); 293 } 294 295 static void netdev_name_node_add(struct net *net, 296 struct netdev_name_node *name_node) 297 { 298 hlist_add_head_rcu(&name_node->hlist, 299 dev_name_hash(net, name_node->name)); 300 } 301 302 static void netdev_name_node_del(struct netdev_name_node *name_node) 303 { 304 hlist_del_rcu(&name_node->hlist); 305 } 306 307 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 308 const char *name) 309 { 310 struct hlist_head *head = dev_name_hash(net, name); 311 struct netdev_name_node *name_node; 312 313 hlist_for_each_entry(name_node, head, hlist) 314 if (!strcmp(name_node->name, name)) 315 return name_node; 316 return NULL; 317 } 318 319 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 320 const char *name) 321 { 322 struct hlist_head *head = dev_name_hash(net, name); 323 struct netdev_name_node *name_node; 324 325 hlist_for_each_entry_rcu(name_node, head, hlist) 326 if (!strcmp(name_node->name, name)) 327 return name_node; 328 return NULL; 329 } 330 331 bool netdev_name_in_use(struct net *net, const char *name) 332 { 333 return netdev_name_node_lookup(net, name); 334 } 335 EXPORT_SYMBOL(netdev_name_in_use); 336 337 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 338 { 339 struct netdev_name_node *name_node; 340 struct net *net = dev_net(dev); 341 342 name_node = netdev_name_node_lookup(net, name); 343 if (name_node) 344 return -EEXIST; 345 name_node = netdev_name_node_alloc(dev, name); 346 if (!name_node) 347 return -ENOMEM; 348 netdev_name_node_add(net, name_node); 349 /* The node that holds dev->name acts as a head of per-device list. */ 350 list_add_tail_rcu(&name_node->list, &dev->name_node->list); 351 352 return 0; 353 } 354 355 static void netdev_name_node_alt_free(struct rcu_head *head) 356 { 357 struct netdev_name_node *name_node = 358 container_of(head, struct netdev_name_node, rcu); 359 360 kfree(name_node->name); 361 netdev_name_node_free(name_node); 362 } 363 364 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 365 { 366 netdev_name_node_del(name_node); 367 list_del(&name_node->list); 368 call_rcu(&name_node->rcu, netdev_name_node_alt_free); 369 } 370 371 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 372 { 373 struct netdev_name_node *name_node; 374 struct net *net = dev_net(dev); 375 376 name_node = netdev_name_node_lookup(net, name); 377 if (!name_node) 378 return -ENOENT; 379 /* lookup might have found our primary name or a name belonging 380 * to another device. 381 */ 382 if (name_node == dev->name_node || name_node->dev != dev) 383 return -EINVAL; 384 385 __netdev_name_node_alt_destroy(name_node); 386 return 0; 387 } 388 389 static void netdev_name_node_alt_flush(struct net_device *dev) 390 { 391 struct netdev_name_node *name_node, *tmp; 392 393 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) { 394 list_del(&name_node->list); 395 netdev_name_node_alt_free(&name_node->rcu); 396 } 397 } 398 399 /* Device list insertion */ 400 static void list_netdevice(struct net_device *dev) 401 { 402 struct netdev_name_node *name_node; 403 struct net *net = dev_net(dev); 404 405 ASSERT_RTNL(); 406 407 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 408 netdev_name_node_add(net, dev->name_node); 409 hlist_add_head_rcu(&dev->index_hlist, 410 dev_index_hash(net, dev->ifindex)); 411 412 netdev_for_each_altname(dev, name_node) 413 netdev_name_node_add(net, name_node); 414 415 /* We reserved the ifindex, this can't fail */ 416 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 417 418 dev_base_seq_inc(net); 419 } 420 421 /* Device list removal 422 * caller must respect a RCU grace period before freeing/reusing dev 423 */ 424 static void unlist_netdevice(struct net_device *dev) 425 { 426 struct netdev_name_node *name_node; 427 struct net *net = dev_net(dev); 428 429 ASSERT_RTNL(); 430 431 xa_erase(&net->dev_by_index, dev->ifindex); 432 433 netdev_for_each_altname(dev, name_node) 434 netdev_name_node_del(name_node); 435 436 /* Unlink dev from the device chain */ 437 list_del_rcu(&dev->dev_list); 438 netdev_name_node_del(dev->name_node); 439 hlist_del_rcu(&dev->index_hlist); 440 441 dev_base_seq_inc(dev_net(dev)); 442 } 443 444 /* 445 * Our notifier list 446 */ 447 448 static RAW_NOTIFIER_HEAD(netdev_chain); 449 450 /* 451 * Device drivers call our routines to queue packets here. We empty the 452 * queue in the local softnet handler. 453 */ 454 455 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = { 456 .process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock), 457 }; 458 EXPORT_PER_CPU_SYMBOL(softnet_data); 459 460 /* Page_pool has a lockless array/stack to alloc/recycle pages. 461 * PP consumers must pay attention to run APIs in the appropriate context 462 * (e.g. NAPI context). 463 */ 464 DEFINE_PER_CPU(struct page_pool *, system_page_pool); 465 466 #ifdef CONFIG_LOCKDEP 467 /* 468 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 469 * according to dev->type 470 */ 471 static const unsigned short netdev_lock_type[] = { 472 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 473 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 474 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 475 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 476 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 477 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 478 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 479 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 480 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 481 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 482 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 483 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 484 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 485 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 486 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 487 488 static const char *const netdev_lock_name[] = { 489 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 490 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 491 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 492 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 493 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 494 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 495 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 496 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 497 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 498 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 499 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 500 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 501 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 502 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 503 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 504 505 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 506 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 507 508 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 509 { 510 int i; 511 512 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 513 if (netdev_lock_type[i] == dev_type) 514 return i; 515 /* the last key is used by default */ 516 return ARRAY_SIZE(netdev_lock_type) - 1; 517 } 518 519 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 520 unsigned short dev_type) 521 { 522 int i; 523 524 i = netdev_lock_pos(dev_type); 525 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 526 netdev_lock_name[i]); 527 } 528 529 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 530 { 531 int i; 532 533 i = netdev_lock_pos(dev->type); 534 lockdep_set_class_and_name(&dev->addr_list_lock, 535 &netdev_addr_lock_key[i], 536 netdev_lock_name[i]); 537 } 538 #else 539 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 540 unsigned short dev_type) 541 { 542 } 543 544 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 545 { 546 } 547 #endif 548 549 /******************************************************************************* 550 * 551 * Protocol management and registration routines 552 * 553 *******************************************************************************/ 554 555 556 /* 557 * Add a protocol ID to the list. Now that the input handler is 558 * smarter we can dispense with all the messy stuff that used to be 559 * here. 560 * 561 * BEWARE!!! Protocol handlers, mangling input packets, 562 * MUST BE last in hash buckets and checking protocol handlers 563 * MUST start from promiscuous ptype_all chain in net_bh. 564 * It is true now, do not change it. 565 * Explanation follows: if protocol handler, mangling packet, will 566 * be the first on list, it is not able to sense, that packet 567 * is cloned and should be copied-on-write, so that it will 568 * change it and subsequent readers will get broken packet. 569 * --ANK (980803) 570 */ 571 572 static inline struct list_head *ptype_head(const struct packet_type *pt) 573 { 574 if (pt->type == htons(ETH_P_ALL)) 575 return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all; 576 else 577 return pt->dev ? &pt->dev->ptype_specific : 578 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 579 } 580 581 /** 582 * dev_add_pack - add packet handler 583 * @pt: packet type declaration 584 * 585 * Add a protocol handler to the networking stack. The passed &packet_type 586 * is linked into kernel lists and may not be freed until it has been 587 * removed from the kernel lists. 588 * 589 * This call does not sleep therefore it can not 590 * guarantee all CPU's that are in middle of receiving packets 591 * will see the new packet type (until the next received packet). 592 */ 593 594 void dev_add_pack(struct packet_type *pt) 595 { 596 struct list_head *head = ptype_head(pt); 597 598 spin_lock(&ptype_lock); 599 list_add_rcu(&pt->list, head); 600 spin_unlock(&ptype_lock); 601 } 602 EXPORT_SYMBOL(dev_add_pack); 603 604 /** 605 * __dev_remove_pack - remove packet handler 606 * @pt: packet type declaration 607 * 608 * Remove a protocol handler that was previously added to the kernel 609 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 610 * from the kernel lists and can be freed or reused once this function 611 * returns. 612 * 613 * The packet type might still be in use by receivers 614 * and must not be freed until after all the CPU's have gone 615 * through a quiescent state. 616 */ 617 void __dev_remove_pack(struct packet_type *pt) 618 { 619 struct list_head *head = ptype_head(pt); 620 struct packet_type *pt1; 621 622 spin_lock(&ptype_lock); 623 624 list_for_each_entry(pt1, head, list) { 625 if (pt == pt1) { 626 list_del_rcu(&pt->list); 627 goto out; 628 } 629 } 630 631 pr_warn("dev_remove_pack: %p not found\n", pt); 632 out: 633 spin_unlock(&ptype_lock); 634 } 635 EXPORT_SYMBOL(__dev_remove_pack); 636 637 /** 638 * dev_remove_pack - remove packet handler 639 * @pt: packet type declaration 640 * 641 * Remove a protocol handler that was previously added to the kernel 642 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 643 * from the kernel lists and can be freed or reused once this function 644 * returns. 645 * 646 * This call sleeps to guarantee that no CPU is looking at the packet 647 * type after return. 648 */ 649 void dev_remove_pack(struct packet_type *pt) 650 { 651 __dev_remove_pack(pt); 652 653 synchronize_net(); 654 } 655 EXPORT_SYMBOL(dev_remove_pack); 656 657 658 /******************************************************************************* 659 * 660 * Device Interface Subroutines 661 * 662 *******************************************************************************/ 663 664 /** 665 * dev_get_iflink - get 'iflink' value of a interface 666 * @dev: targeted interface 667 * 668 * Indicates the ifindex the interface is linked to. 669 * Physical interfaces have the same 'ifindex' and 'iflink' values. 670 */ 671 672 int dev_get_iflink(const struct net_device *dev) 673 { 674 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 675 return dev->netdev_ops->ndo_get_iflink(dev); 676 677 return READ_ONCE(dev->ifindex); 678 } 679 EXPORT_SYMBOL(dev_get_iflink); 680 681 /** 682 * dev_fill_metadata_dst - Retrieve tunnel egress information. 683 * @dev: targeted interface 684 * @skb: The packet. 685 * 686 * For better visibility of tunnel traffic OVS needs to retrieve 687 * egress tunnel information for a packet. Following API allows 688 * user to get this info. 689 */ 690 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 691 { 692 struct ip_tunnel_info *info; 693 694 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 695 return -EINVAL; 696 697 info = skb_tunnel_info_unclone(skb); 698 if (!info) 699 return -ENOMEM; 700 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 701 return -EINVAL; 702 703 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 704 } 705 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 706 707 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 708 { 709 int k = stack->num_paths++; 710 711 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 712 return NULL; 713 714 return &stack->path[k]; 715 } 716 717 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 718 struct net_device_path_stack *stack) 719 { 720 const struct net_device *last_dev; 721 struct net_device_path_ctx ctx = { 722 .dev = dev, 723 }; 724 struct net_device_path *path; 725 int ret = 0; 726 727 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 728 stack->num_paths = 0; 729 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 730 last_dev = ctx.dev; 731 path = dev_fwd_path(stack); 732 if (!path) 733 return -1; 734 735 memset(path, 0, sizeof(struct net_device_path)); 736 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 737 if (ret < 0) 738 return -1; 739 740 if (WARN_ON_ONCE(last_dev == ctx.dev)) 741 return -1; 742 } 743 744 if (!ctx.dev) 745 return ret; 746 747 path = dev_fwd_path(stack); 748 if (!path) 749 return -1; 750 path->type = DEV_PATH_ETHERNET; 751 path->dev = ctx.dev; 752 753 return ret; 754 } 755 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 756 757 /* must be called under rcu_read_lock(), as we dont take a reference */ 758 static struct napi_struct *napi_by_id(unsigned int napi_id) 759 { 760 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 761 struct napi_struct *napi; 762 763 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 764 if (napi->napi_id == napi_id) 765 return napi; 766 767 return NULL; 768 } 769 770 /* must be called under rcu_read_lock(), as we dont take a reference */ 771 struct napi_struct *netdev_napi_by_id(struct net *net, unsigned int napi_id) 772 { 773 struct napi_struct *napi; 774 775 napi = napi_by_id(napi_id); 776 if (!napi) 777 return NULL; 778 779 if (WARN_ON_ONCE(!napi->dev)) 780 return NULL; 781 if (!net_eq(net, dev_net(napi->dev))) 782 return NULL; 783 784 return napi; 785 } 786 787 /** 788 * __dev_get_by_name - find a device by its name 789 * @net: the applicable net namespace 790 * @name: name to find 791 * 792 * Find an interface by name. Must be called under RTNL semaphore. 793 * If the name is found a pointer to the device is returned. 794 * If the name is not found then %NULL is returned. The 795 * reference counters are not incremented so the caller must be 796 * careful with locks. 797 */ 798 799 struct net_device *__dev_get_by_name(struct net *net, const char *name) 800 { 801 struct netdev_name_node *node_name; 802 803 node_name = netdev_name_node_lookup(net, name); 804 return node_name ? node_name->dev : NULL; 805 } 806 EXPORT_SYMBOL(__dev_get_by_name); 807 808 /** 809 * dev_get_by_name_rcu - find a device by its name 810 * @net: the applicable net namespace 811 * @name: name to find 812 * 813 * Find an interface by name. 814 * If the name is found a pointer to the device is returned. 815 * If the name is not found then %NULL is returned. 816 * The reference counters are not incremented so the caller must be 817 * careful with locks. The caller must hold RCU lock. 818 */ 819 820 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 821 { 822 struct netdev_name_node *node_name; 823 824 node_name = netdev_name_node_lookup_rcu(net, name); 825 return node_name ? node_name->dev : NULL; 826 } 827 EXPORT_SYMBOL(dev_get_by_name_rcu); 828 829 /* Deprecated for new users, call netdev_get_by_name() instead */ 830 struct net_device *dev_get_by_name(struct net *net, const char *name) 831 { 832 struct net_device *dev; 833 834 rcu_read_lock(); 835 dev = dev_get_by_name_rcu(net, name); 836 dev_hold(dev); 837 rcu_read_unlock(); 838 return dev; 839 } 840 EXPORT_SYMBOL(dev_get_by_name); 841 842 /** 843 * netdev_get_by_name() - find a device by its name 844 * @net: the applicable net namespace 845 * @name: name to find 846 * @tracker: tracking object for the acquired reference 847 * @gfp: allocation flags for the tracker 848 * 849 * Find an interface by name. This can be called from any 850 * context and does its own locking. The returned handle has 851 * the usage count incremented and the caller must use netdev_put() to 852 * release it when it is no longer needed. %NULL is returned if no 853 * matching device is found. 854 */ 855 struct net_device *netdev_get_by_name(struct net *net, const char *name, 856 netdevice_tracker *tracker, gfp_t gfp) 857 { 858 struct net_device *dev; 859 860 dev = dev_get_by_name(net, name); 861 if (dev) 862 netdev_tracker_alloc(dev, tracker, gfp); 863 return dev; 864 } 865 EXPORT_SYMBOL(netdev_get_by_name); 866 867 /** 868 * __dev_get_by_index - find a device by its ifindex 869 * @net: the applicable net namespace 870 * @ifindex: index of device 871 * 872 * Search for an interface by index. Returns %NULL if the device 873 * is not found or a pointer to the device. The device has not 874 * had its reference counter increased so the caller must be careful 875 * about locking. The caller must hold the RTNL semaphore. 876 */ 877 878 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 879 { 880 struct net_device *dev; 881 struct hlist_head *head = dev_index_hash(net, ifindex); 882 883 hlist_for_each_entry(dev, head, index_hlist) 884 if (dev->ifindex == ifindex) 885 return dev; 886 887 return NULL; 888 } 889 EXPORT_SYMBOL(__dev_get_by_index); 890 891 /** 892 * dev_get_by_index_rcu - find a device by its ifindex 893 * @net: the applicable net namespace 894 * @ifindex: index of device 895 * 896 * Search for an interface by index. Returns %NULL if the device 897 * is not found or a pointer to the device. The device has not 898 * had its reference counter increased so the caller must be careful 899 * about locking. The caller must hold RCU lock. 900 */ 901 902 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 903 { 904 struct net_device *dev; 905 struct hlist_head *head = dev_index_hash(net, ifindex); 906 907 hlist_for_each_entry_rcu(dev, head, index_hlist) 908 if (dev->ifindex == ifindex) 909 return dev; 910 911 return NULL; 912 } 913 EXPORT_SYMBOL(dev_get_by_index_rcu); 914 915 /* Deprecated for new users, call netdev_get_by_index() instead */ 916 struct net_device *dev_get_by_index(struct net *net, int ifindex) 917 { 918 struct net_device *dev; 919 920 rcu_read_lock(); 921 dev = dev_get_by_index_rcu(net, ifindex); 922 dev_hold(dev); 923 rcu_read_unlock(); 924 return dev; 925 } 926 EXPORT_SYMBOL(dev_get_by_index); 927 928 /** 929 * netdev_get_by_index() - find a device by its ifindex 930 * @net: the applicable net namespace 931 * @ifindex: index of device 932 * @tracker: tracking object for the acquired reference 933 * @gfp: allocation flags for the tracker 934 * 935 * Search for an interface by index. Returns NULL if the device 936 * is not found or a pointer to the device. The device returned has 937 * had a reference added and the pointer is safe until the user calls 938 * netdev_put() to indicate they have finished with it. 939 */ 940 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 941 netdevice_tracker *tracker, gfp_t gfp) 942 { 943 struct net_device *dev; 944 945 dev = dev_get_by_index(net, ifindex); 946 if (dev) 947 netdev_tracker_alloc(dev, tracker, gfp); 948 return dev; 949 } 950 EXPORT_SYMBOL(netdev_get_by_index); 951 952 /** 953 * dev_get_by_napi_id - find a device by napi_id 954 * @napi_id: ID of the NAPI struct 955 * 956 * Search for an interface by NAPI ID. Returns %NULL if the device 957 * is not found or a pointer to the device. The device has not had 958 * its reference counter increased so the caller must be careful 959 * about locking. The caller must hold RCU lock. 960 */ 961 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 962 { 963 struct napi_struct *napi; 964 965 WARN_ON_ONCE(!rcu_read_lock_held()); 966 967 if (napi_id < MIN_NAPI_ID) 968 return NULL; 969 970 napi = napi_by_id(napi_id); 971 972 return napi ? napi->dev : NULL; 973 } 974 975 static DEFINE_SEQLOCK(netdev_rename_lock); 976 977 void netdev_copy_name(struct net_device *dev, char *name) 978 { 979 unsigned int seq; 980 981 do { 982 seq = read_seqbegin(&netdev_rename_lock); 983 strscpy(name, dev->name, IFNAMSIZ); 984 } while (read_seqretry(&netdev_rename_lock, seq)); 985 } 986 987 /** 988 * netdev_get_name - get a netdevice name, knowing its ifindex. 989 * @net: network namespace 990 * @name: a pointer to the buffer where the name will be stored. 991 * @ifindex: the ifindex of the interface to get the name from. 992 */ 993 int netdev_get_name(struct net *net, char *name, int ifindex) 994 { 995 struct net_device *dev; 996 int ret; 997 998 rcu_read_lock(); 999 1000 dev = dev_get_by_index_rcu(net, ifindex); 1001 if (!dev) { 1002 ret = -ENODEV; 1003 goto out; 1004 } 1005 1006 netdev_copy_name(dev, name); 1007 1008 ret = 0; 1009 out: 1010 rcu_read_unlock(); 1011 return ret; 1012 } 1013 1014 /** 1015 * dev_getbyhwaddr_rcu - find a device by its hardware address 1016 * @net: the applicable net namespace 1017 * @type: media type of device 1018 * @ha: hardware address 1019 * 1020 * Search for an interface by MAC address. Returns NULL if the device 1021 * is not found or a pointer to the device. 1022 * The caller must hold RCU or RTNL. 1023 * The returned device has not had its ref count increased 1024 * and the caller must therefore be careful about locking 1025 * 1026 */ 1027 1028 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1029 const char *ha) 1030 { 1031 struct net_device *dev; 1032 1033 for_each_netdev_rcu(net, dev) 1034 if (dev->type == type && 1035 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1036 return dev; 1037 1038 return NULL; 1039 } 1040 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1041 1042 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1043 { 1044 struct net_device *dev, *ret = NULL; 1045 1046 rcu_read_lock(); 1047 for_each_netdev_rcu(net, dev) 1048 if (dev->type == type) { 1049 dev_hold(dev); 1050 ret = dev; 1051 break; 1052 } 1053 rcu_read_unlock(); 1054 return ret; 1055 } 1056 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1057 1058 /** 1059 * __dev_get_by_flags - find any device with given flags 1060 * @net: the applicable net namespace 1061 * @if_flags: IFF_* values 1062 * @mask: bitmask of bits in if_flags to check 1063 * 1064 * Search for any interface with the given flags. Returns NULL if a device 1065 * is not found or a pointer to the device. Must be called inside 1066 * rtnl_lock(), and result refcount is unchanged. 1067 */ 1068 1069 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1070 unsigned short mask) 1071 { 1072 struct net_device *dev, *ret; 1073 1074 ASSERT_RTNL(); 1075 1076 ret = NULL; 1077 for_each_netdev(net, dev) { 1078 if (((dev->flags ^ if_flags) & mask) == 0) { 1079 ret = dev; 1080 break; 1081 } 1082 } 1083 return ret; 1084 } 1085 EXPORT_SYMBOL(__dev_get_by_flags); 1086 1087 /** 1088 * dev_valid_name - check if name is okay for network device 1089 * @name: name string 1090 * 1091 * Network device names need to be valid file names to 1092 * allow sysfs to work. We also disallow any kind of 1093 * whitespace. 1094 */ 1095 bool dev_valid_name(const char *name) 1096 { 1097 if (*name == '\0') 1098 return false; 1099 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1100 return false; 1101 if (!strcmp(name, ".") || !strcmp(name, "..")) 1102 return false; 1103 1104 while (*name) { 1105 if (*name == '/' || *name == ':' || isspace(*name)) 1106 return false; 1107 name++; 1108 } 1109 return true; 1110 } 1111 EXPORT_SYMBOL(dev_valid_name); 1112 1113 /** 1114 * __dev_alloc_name - allocate a name for a device 1115 * @net: network namespace to allocate the device name in 1116 * @name: name format string 1117 * @res: result name string 1118 * 1119 * Passed a format string - eg "lt%d" it will try and find a suitable 1120 * id. It scans list of devices to build up a free map, then chooses 1121 * the first empty slot. The caller must hold the dev_base or rtnl lock 1122 * while allocating the name and adding the device in order to avoid 1123 * duplicates. 1124 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1125 * Returns the number of the unit assigned or a negative errno code. 1126 */ 1127 1128 static int __dev_alloc_name(struct net *net, const char *name, char *res) 1129 { 1130 int i = 0; 1131 const char *p; 1132 const int max_netdevices = 8*PAGE_SIZE; 1133 unsigned long *inuse; 1134 struct net_device *d; 1135 char buf[IFNAMSIZ]; 1136 1137 /* Verify the string as this thing may have come from the user. 1138 * There must be one "%d" and no other "%" characters. 1139 */ 1140 p = strchr(name, '%'); 1141 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1142 return -EINVAL; 1143 1144 /* Use one page as a bit array of possible slots */ 1145 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1146 if (!inuse) 1147 return -ENOMEM; 1148 1149 for_each_netdev(net, d) { 1150 struct netdev_name_node *name_node; 1151 1152 netdev_for_each_altname(d, name_node) { 1153 if (!sscanf(name_node->name, name, &i)) 1154 continue; 1155 if (i < 0 || i >= max_netdevices) 1156 continue; 1157 1158 /* avoid cases where sscanf is not exact inverse of printf */ 1159 snprintf(buf, IFNAMSIZ, name, i); 1160 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1161 __set_bit(i, inuse); 1162 } 1163 if (!sscanf(d->name, name, &i)) 1164 continue; 1165 if (i < 0 || i >= max_netdevices) 1166 continue; 1167 1168 /* avoid cases where sscanf is not exact inverse of printf */ 1169 snprintf(buf, IFNAMSIZ, name, i); 1170 if (!strncmp(buf, d->name, IFNAMSIZ)) 1171 __set_bit(i, inuse); 1172 } 1173 1174 i = find_first_zero_bit(inuse, max_netdevices); 1175 bitmap_free(inuse); 1176 if (i == max_netdevices) 1177 return -ENFILE; 1178 1179 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1180 strscpy(buf, name, IFNAMSIZ); 1181 snprintf(res, IFNAMSIZ, buf, i); 1182 return i; 1183 } 1184 1185 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1186 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1187 const char *want_name, char *out_name, 1188 int dup_errno) 1189 { 1190 if (!dev_valid_name(want_name)) 1191 return -EINVAL; 1192 1193 if (strchr(want_name, '%')) 1194 return __dev_alloc_name(net, want_name, out_name); 1195 1196 if (netdev_name_in_use(net, want_name)) 1197 return -dup_errno; 1198 if (out_name != want_name) 1199 strscpy(out_name, want_name, IFNAMSIZ); 1200 return 0; 1201 } 1202 1203 /** 1204 * dev_alloc_name - allocate a name for a device 1205 * @dev: device 1206 * @name: name format string 1207 * 1208 * Passed a format string - eg "lt%d" it will try and find a suitable 1209 * id. It scans list of devices to build up a free map, then chooses 1210 * the first empty slot. The caller must hold the dev_base or rtnl lock 1211 * while allocating the name and adding the device in order to avoid 1212 * duplicates. 1213 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1214 * Returns the number of the unit assigned or a negative errno code. 1215 */ 1216 1217 int dev_alloc_name(struct net_device *dev, const char *name) 1218 { 1219 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1220 } 1221 EXPORT_SYMBOL(dev_alloc_name); 1222 1223 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1224 const char *name) 1225 { 1226 int ret; 1227 1228 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1229 return ret < 0 ? ret : 0; 1230 } 1231 1232 /** 1233 * dev_change_name - change name of a device 1234 * @dev: device 1235 * @newname: name (or format string) must be at least IFNAMSIZ 1236 * 1237 * Change name of a device, can pass format strings "eth%d". 1238 * for wildcarding. 1239 */ 1240 int dev_change_name(struct net_device *dev, const char *newname) 1241 { 1242 unsigned char old_assign_type; 1243 char oldname[IFNAMSIZ]; 1244 int err = 0; 1245 int ret; 1246 struct net *net; 1247 1248 ASSERT_RTNL(); 1249 BUG_ON(!dev_net(dev)); 1250 1251 net = dev_net(dev); 1252 1253 down_write(&devnet_rename_sem); 1254 1255 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1256 up_write(&devnet_rename_sem); 1257 return 0; 1258 } 1259 1260 memcpy(oldname, dev->name, IFNAMSIZ); 1261 1262 write_seqlock_bh(&netdev_rename_lock); 1263 err = dev_get_valid_name(net, dev, newname); 1264 write_sequnlock_bh(&netdev_rename_lock); 1265 1266 if (err < 0) { 1267 up_write(&devnet_rename_sem); 1268 return err; 1269 } 1270 1271 if (oldname[0] && !strchr(oldname, '%')) 1272 netdev_info(dev, "renamed from %s%s\n", oldname, 1273 dev->flags & IFF_UP ? " (while UP)" : ""); 1274 1275 old_assign_type = dev->name_assign_type; 1276 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); 1277 1278 rollback: 1279 ret = device_rename(&dev->dev, dev->name); 1280 if (ret) { 1281 memcpy(dev->name, oldname, IFNAMSIZ); 1282 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1283 up_write(&devnet_rename_sem); 1284 return ret; 1285 } 1286 1287 up_write(&devnet_rename_sem); 1288 1289 netdev_adjacent_rename_links(dev, oldname); 1290 1291 netdev_name_node_del(dev->name_node); 1292 1293 synchronize_net(); 1294 1295 netdev_name_node_add(net, dev->name_node); 1296 1297 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1298 ret = notifier_to_errno(ret); 1299 1300 if (ret) { 1301 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1302 if (err >= 0) { 1303 err = ret; 1304 down_write(&devnet_rename_sem); 1305 write_seqlock_bh(&netdev_rename_lock); 1306 memcpy(dev->name, oldname, IFNAMSIZ); 1307 write_sequnlock_bh(&netdev_rename_lock); 1308 memcpy(oldname, newname, IFNAMSIZ); 1309 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1310 old_assign_type = NET_NAME_RENAMED; 1311 goto rollback; 1312 } else { 1313 netdev_err(dev, "name change rollback failed: %d\n", 1314 ret); 1315 } 1316 } 1317 1318 return err; 1319 } 1320 1321 /** 1322 * dev_set_alias - change ifalias of a device 1323 * @dev: device 1324 * @alias: name up to IFALIASZ 1325 * @len: limit of bytes to copy from info 1326 * 1327 * Set ifalias for a device, 1328 */ 1329 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1330 { 1331 struct dev_ifalias *new_alias = NULL; 1332 1333 if (len >= IFALIASZ) 1334 return -EINVAL; 1335 1336 if (len) { 1337 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1338 if (!new_alias) 1339 return -ENOMEM; 1340 1341 memcpy(new_alias->ifalias, alias, len); 1342 new_alias->ifalias[len] = 0; 1343 } 1344 1345 mutex_lock(&ifalias_mutex); 1346 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1347 mutex_is_locked(&ifalias_mutex)); 1348 mutex_unlock(&ifalias_mutex); 1349 1350 if (new_alias) 1351 kfree_rcu(new_alias, rcuhead); 1352 1353 return len; 1354 } 1355 EXPORT_SYMBOL(dev_set_alias); 1356 1357 /** 1358 * dev_get_alias - get ifalias of a device 1359 * @dev: device 1360 * @name: buffer to store name of ifalias 1361 * @len: size of buffer 1362 * 1363 * get ifalias for a device. Caller must make sure dev cannot go 1364 * away, e.g. rcu read lock or own a reference count to device. 1365 */ 1366 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1367 { 1368 const struct dev_ifalias *alias; 1369 int ret = 0; 1370 1371 rcu_read_lock(); 1372 alias = rcu_dereference(dev->ifalias); 1373 if (alias) 1374 ret = snprintf(name, len, "%s", alias->ifalias); 1375 rcu_read_unlock(); 1376 1377 return ret; 1378 } 1379 1380 /** 1381 * netdev_features_change - device changes features 1382 * @dev: device to cause notification 1383 * 1384 * Called to indicate a device has changed features. 1385 */ 1386 void netdev_features_change(struct net_device *dev) 1387 { 1388 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1389 } 1390 EXPORT_SYMBOL(netdev_features_change); 1391 1392 /** 1393 * netdev_state_change - device changes state 1394 * @dev: device to cause notification 1395 * 1396 * Called to indicate a device has changed state. This function calls 1397 * the notifier chains for netdev_chain and sends a NEWLINK message 1398 * to the routing socket. 1399 */ 1400 void netdev_state_change(struct net_device *dev) 1401 { 1402 if (dev->flags & IFF_UP) { 1403 struct netdev_notifier_change_info change_info = { 1404 .info.dev = dev, 1405 }; 1406 1407 call_netdevice_notifiers_info(NETDEV_CHANGE, 1408 &change_info.info); 1409 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1410 } 1411 } 1412 EXPORT_SYMBOL(netdev_state_change); 1413 1414 /** 1415 * __netdev_notify_peers - notify network peers about existence of @dev, 1416 * to be called when rtnl lock is already held. 1417 * @dev: network device 1418 * 1419 * Generate traffic such that interested network peers are aware of 1420 * @dev, such as by generating a gratuitous ARP. This may be used when 1421 * a device wants to inform the rest of the network about some sort of 1422 * reconfiguration such as a failover event or virtual machine 1423 * migration. 1424 */ 1425 void __netdev_notify_peers(struct net_device *dev) 1426 { 1427 ASSERT_RTNL(); 1428 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1429 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1430 } 1431 EXPORT_SYMBOL(__netdev_notify_peers); 1432 1433 /** 1434 * netdev_notify_peers - notify network peers about existence of @dev 1435 * @dev: network device 1436 * 1437 * Generate traffic such that interested network peers are aware of 1438 * @dev, such as by generating a gratuitous ARP. This may be used when 1439 * a device wants to inform the rest of the network about some sort of 1440 * reconfiguration such as a failover event or virtual machine 1441 * migration. 1442 */ 1443 void netdev_notify_peers(struct net_device *dev) 1444 { 1445 rtnl_lock(); 1446 __netdev_notify_peers(dev); 1447 rtnl_unlock(); 1448 } 1449 EXPORT_SYMBOL(netdev_notify_peers); 1450 1451 static int napi_threaded_poll(void *data); 1452 1453 static int napi_kthread_create(struct napi_struct *n) 1454 { 1455 int err = 0; 1456 1457 /* Create and wake up the kthread once to put it in 1458 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1459 * warning and work with loadavg. 1460 */ 1461 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1462 n->dev->name, n->napi_id); 1463 if (IS_ERR(n->thread)) { 1464 err = PTR_ERR(n->thread); 1465 pr_err("kthread_run failed with err %d\n", err); 1466 n->thread = NULL; 1467 } 1468 1469 return err; 1470 } 1471 1472 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1473 { 1474 const struct net_device_ops *ops = dev->netdev_ops; 1475 int ret; 1476 1477 ASSERT_RTNL(); 1478 dev_addr_check(dev); 1479 1480 if (!netif_device_present(dev)) { 1481 /* may be detached because parent is runtime-suspended */ 1482 if (dev->dev.parent) 1483 pm_runtime_resume(dev->dev.parent); 1484 if (!netif_device_present(dev)) 1485 return -ENODEV; 1486 } 1487 1488 /* Block netpoll from trying to do any rx path servicing. 1489 * If we don't do this there is a chance ndo_poll_controller 1490 * or ndo_poll may be running while we open the device 1491 */ 1492 netpoll_poll_disable(dev); 1493 1494 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1495 ret = notifier_to_errno(ret); 1496 if (ret) 1497 return ret; 1498 1499 set_bit(__LINK_STATE_START, &dev->state); 1500 1501 if (ops->ndo_validate_addr) 1502 ret = ops->ndo_validate_addr(dev); 1503 1504 if (!ret && ops->ndo_open) 1505 ret = ops->ndo_open(dev); 1506 1507 netpoll_poll_enable(dev); 1508 1509 if (ret) 1510 clear_bit(__LINK_STATE_START, &dev->state); 1511 else { 1512 dev->flags |= IFF_UP; 1513 dev_set_rx_mode(dev); 1514 dev_activate(dev); 1515 add_device_randomness(dev->dev_addr, dev->addr_len); 1516 } 1517 1518 return ret; 1519 } 1520 1521 /** 1522 * dev_open - prepare an interface for use. 1523 * @dev: device to open 1524 * @extack: netlink extended ack 1525 * 1526 * Takes a device from down to up state. The device's private open 1527 * function is invoked and then the multicast lists are loaded. Finally 1528 * the device is moved into the up state and a %NETDEV_UP message is 1529 * sent to the netdev notifier chain. 1530 * 1531 * Calling this function on an active interface is a nop. On a failure 1532 * a negative errno code is returned. 1533 */ 1534 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1535 { 1536 int ret; 1537 1538 if (dev->flags & IFF_UP) 1539 return 0; 1540 1541 ret = __dev_open(dev, extack); 1542 if (ret < 0) 1543 return ret; 1544 1545 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1546 call_netdevice_notifiers(NETDEV_UP, dev); 1547 1548 return ret; 1549 } 1550 EXPORT_SYMBOL(dev_open); 1551 1552 static void __dev_close_many(struct list_head *head) 1553 { 1554 struct net_device *dev; 1555 1556 ASSERT_RTNL(); 1557 might_sleep(); 1558 1559 list_for_each_entry(dev, head, close_list) { 1560 /* Temporarily disable netpoll until the interface is down */ 1561 netpoll_poll_disable(dev); 1562 1563 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1564 1565 clear_bit(__LINK_STATE_START, &dev->state); 1566 1567 /* Synchronize to scheduled poll. We cannot touch poll list, it 1568 * can be even on different cpu. So just clear netif_running(). 1569 * 1570 * dev->stop() will invoke napi_disable() on all of it's 1571 * napi_struct instances on this device. 1572 */ 1573 smp_mb__after_atomic(); /* Commit netif_running(). */ 1574 } 1575 1576 dev_deactivate_many(head); 1577 1578 list_for_each_entry(dev, head, close_list) { 1579 const struct net_device_ops *ops = dev->netdev_ops; 1580 1581 /* 1582 * Call the device specific close. This cannot fail. 1583 * Only if device is UP 1584 * 1585 * We allow it to be called even after a DETACH hot-plug 1586 * event. 1587 */ 1588 if (ops->ndo_stop) 1589 ops->ndo_stop(dev); 1590 1591 dev->flags &= ~IFF_UP; 1592 netpoll_poll_enable(dev); 1593 } 1594 } 1595 1596 static void __dev_close(struct net_device *dev) 1597 { 1598 LIST_HEAD(single); 1599 1600 list_add(&dev->close_list, &single); 1601 __dev_close_many(&single); 1602 list_del(&single); 1603 } 1604 1605 void dev_close_many(struct list_head *head, bool unlink) 1606 { 1607 struct net_device *dev, *tmp; 1608 1609 /* Remove the devices that don't need to be closed */ 1610 list_for_each_entry_safe(dev, tmp, head, close_list) 1611 if (!(dev->flags & IFF_UP)) 1612 list_del_init(&dev->close_list); 1613 1614 __dev_close_many(head); 1615 1616 list_for_each_entry_safe(dev, tmp, head, close_list) { 1617 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1618 call_netdevice_notifiers(NETDEV_DOWN, dev); 1619 if (unlink) 1620 list_del_init(&dev->close_list); 1621 } 1622 } 1623 EXPORT_SYMBOL(dev_close_many); 1624 1625 /** 1626 * dev_close - shutdown an interface. 1627 * @dev: device to shutdown 1628 * 1629 * This function moves an active device into down state. A 1630 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1631 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1632 * chain. 1633 */ 1634 void dev_close(struct net_device *dev) 1635 { 1636 if (dev->flags & IFF_UP) { 1637 LIST_HEAD(single); 1638 1639 list_add(&dev->close_list, &single); 1640 dev_close_many(&single, true); 1641 list_del(&single); 1642 } 1643 } 1644 EXPORT_SYMBOL(dev_close); 1645 1646 1647 /** 1648 * dev_disable_lro - disable Large Receive Offload on a device 1649 * @dev: device 1650 * 1651 * Disable Large Receive Offload (LRO) on a net device. Must be 1652 * called under RTNL. This is needed if received packets may be 1653 * forwarded to another interface. 1654 */ 1655 void dev_disable_lro(struct net_device *dev) 1656 { 1657 struct net_device *lower_dev; 1658 struct list_head *iter; 1659 1660 dev->wanted_features &= ~NETIF_F_LRO; 1661 netdev_update_features(dev); 1662 1663 if (unlikely(dev->features & NETIF_F_LRO)) 1664 netdev_WARN(dev, "failed to disable LRO!\n"); 1665 1666 netdev_for_each_lower_dev(dev, lower_dev, iter) 1667 dev_disable_lro(lower_dev); 1668 } 1669 EXPORT_SYMBOL(dev_disable_lro); 1670 1671 /** 1672 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1673 * @dev: device 1674 * 1675 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1676 * called under RTNL. This is needed if Generic XDP is installed on 1677 * the device. 1678 */ 1679 static void dev_disable_gro_hw(struct net_device *dev) 1680 { 1681 dev->wanted_features &= ~NETIF_F_GRO_HW; 1682 netdev_update_features(dev); 1683 1684 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1685 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1686 } 1687 1688 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1689 { 1690 #define N(val) \ 1691 case NETDEV_##val: \ 1692 return "NETDEV_" __stringify(val); 1693 switch (cmd) { 1694 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1695 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1696 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1697 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1698 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1699 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1700 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1701 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1702 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1703 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1704 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1705 N(XDP_FEAT_CHANGE) 1706 } 1707 #undef N 1708 return "UNKNOWN_NETDEV_EVENT"; 1709 } 1710 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1711 1712 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1713 struct net_device *dev) 1714 { 1715 struct netdev_notifier_info info = { 1716 .dev = dev, 1717 }; 1718 1719 return nb->notifier_call(nb, val, &info); 1720 } 1721 1722 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1723 struct net_device *dev) 1724 { 1725 int err; 1726 1727 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1728 err = notifier_to_errno(err); 1729 if (err) 1730 return err; 1731 1732 if (!(dev->flags & IFF_UP)) 1733 return 0; 1734 1735 call_netdevice_notifier(nb, NETDEV_UP, dev); 1736 return 0; 1737 } 1738 1739 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1740 struct net_device *dev) 1741 { 1742 if (dev->flags & IFF_UP) { 1743 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1744 dev); 1745 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1746 } 1747 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1748 } 1749 1750 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1751 struct net *net) 1752 { 1753 struct net_device *dev; 1754 int err; 1755 1756 for_each_netdev(net, dev) { 1757 err = call_netdevice_register_notifiers(nb, dev); 1758 if (err) 1759 goto rollback; 1760 } 1761 return 0; 1762 1763 rollback: 1764 for_each_netdev_continue_reverse(net, dev) 1765 call_netdevice_unregister_notifiers(nb, dev); 1766 return err; 1767 } 1768 1769 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1770 struct net *net) 1771 { 1772 struct net_device *dev; 1773 1774 for_each_netdev(net, dev) 1775 call_netdevice_unregister_notifiers(nb, dev); 1776 } 1777 1778 static int dev_boot_phase = 1; 1779 1780 /** 1781 * register_netdevice_notifier - register a network notifier block 1782 * @nb: notifier 1783 * 1784 * Register a notifier to be called when network device events occur. 1785 * The notifier passed is linked into the kernel structures and must 1786 * not be reused until it has been unregistered. A negative errno code 1787 * is returned on a failure. 1788 * 1789 * When registered all registration and up events are replayed 1790 * to the new notifier to allow device to have a race free 1791 * view of the network device list. 1792 */ 1793 1794 int register_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 1802 /* When RTNL is removed, we need protection for netdev_chain. */ 1803 rtnl_lock(); 1804 1805 err = raw_notifier_chain_register(&netdev_chain, nb); 1806 if (err) 1807 goto unlock; 1808 if (dev_boot_phase) 1809 goto unlock; 1810 for_each_net(net) { 1811 __rtnl_net_lock(net); 1812 err = call_netdevice_register_net_notifiers(nb, net); 1813 __rtnl_net_unlock(net); 1814 if (err) 1815 goto rollback; 1816 } 1817 1818 unlock: 1819 rtnl_unlock(); 1820 up_write(&pernet_ops_rwsem); 1821 return err; 1822 1823 rollback: 1824 for_each_net_continue_reverse(net) { 1825 __rtnl_net_lock(net); 1826 call_netdevice_unregister_net_notifiers(nb, net); 1827 __rtnl_net_unlock(net); 1828 } 1829 1830 raw_notifier_chain_unregister(&netdev_chain, nb); 1831 goto unlock; 1832 } 1833 EXPORT_SYMBOL(register_netdevice_notifier); 1834 1835 /** 1836 * unregister_netdevice_notifier - unregister a network notifier block 1837 * @nb: notifier 1838 * 1839 * Unregister a notifier previously registered by 1840 * register_netdevice_notifier(). The notifier is unlinked into the 1841 * kernel structures and may then be reused. A negative errno code 1842 * is returned on a failure. 1843 * 1844 * After unregistering unregister and down device events are synthesized 1845 * for all devices on the device list to the removed notifier to remove 1846 * the need for special case cleanup code. 1847 */ 1848 1849 int unregister_netdevice_notifier(struct notifier_block *nb) 1850 { 1851 struct net *net; 1852 int err; 1853 1854 /* Close race with setup_net() and cleanup_net() */ 1855 down_write(&pernet_ops_rwsem); 1856 rtnl_lock(); 1857 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1858 if (err) 1859 goto unlock; 1860 1861 for_each_net(net) { 1862 __rtnl_net_lock(net); 1863 call_netdevice_unregister_net_notifiers(nb, net); 1864 __rtnl_net_unlock(net); 1865 } 1866 1867 unlock: 1868 rtnl_unlock(); 1869 up_write(&pernet_ops_rwsem); 1870 return err; 1871 } 1872 EXPORT_SYMBOL(unregister_netdevice_notifier); 1873 1874 static int __register_netdevice_notifier_net(struct net *net, 1875 struct notifier_block *nb, 1876 bool ignore_call_fail) 1877 { 1878 int err; 1879 1880 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1881 if (err) 1882 return err; 1883 if (dev_boot_phase) 1884 return 0; 1885 1886 err = call_netdevice_register_net_notifiers(nb, net); 1887 if (err && !ignore_call_fail) 1888 goto chain_unregister; 1889 1890 return 0; 1891 1892 chain_unregister: 1893 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1894 return err; 1895 } 1896 1897 static int __unregister_netdevice_notifier_net(struct net *net, 1898 struct notifier_block *nb) 1899 { 1900 int err; 1901 1902 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1903 if (err) 1904 return err; 1905 1906 call_netdevice_unregister_net_notifiers(nb, net); 1907 return 0; 1908 } 1909 1910 /** 1911 * register_netdevice_notifier_net - register a per-netns network notifier block 1912 * @net: network namespace 1913 * @nb: notifier 1914 * 1915 * Register a notifier to be called when network device events occur. 1916 * The notifier passed is linked into the kernel structures and must 1917 * not be reused until it has been unregistered. A negative errno code 1918 * is returned on a failure. 1919 * 1920 * When registered all registration and up events are replayed 1921 * to the new notifier to allow device to have a race free 1922 * view of the network device list. 1923 */ 1924 1925 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1926 { 1927 int err; 1928 1929 rtnl_net_lock(net); 1930 err = __register_netdevice_notifier_net(net, nb, false); 1931 rtnl_net_unlock(net); 1932 1933 return err; 1934 } 1935 EXPORT_SYMBOL(register_netdevice_notifier_net); 1936 1937 /** 1938 * unregister_netdevice_notifier_net - unregister a per-netns 1939 * network notifier block 1940 * @net: network namespace 1941 * @nb: notifier 1942 * 1943 * Unregister a notifier previously registered by 1944 * register_netdevice_notifier_net(). The notifier is unlinked from the 1945 * kernel structures and may then be reused. A negative errno code 1946 * is returned on a failure. 1947 * 1948 * After unregistering unregister and down device events are synthesized 1949 * for all devices on the device list to the removed notifier to remove 1950 * the need for special case cleanup code. 1951 */ 1952 1953 int unregister_netdevice_notifier_net(struct net *net, 1954 struct notifier_block *nb) 1955 { 1956 int err; 1957 1958 rtnl_net_lock(net); 1959 err = __unregister_netdevice_notifier_net(net, nb); 1960 rtnl_net_unlock(net); 1961 1962 return err; 1963 } 1964 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1965 1966 static void __move_netdevice_notifier_net(struct net *src_net, 1967 struct net *dst_net, 1968 struct notifier_block *nb) 1969 { 1970 __unregister_netdevice_notifier_net(src_net, nb); 1971 __register_netdevice_notifier_net(dst_net, nb, true); 1972 } 1973 1974 int register_netdevice_notifier_dev_net(struct net_device *dev, 1975 struct notifier_block *nb, 1976 struct netdev_net_notifier *nn) 1977 { 1978 struct net *net = dev_net(dev); 1979 int err; 1980 1981 rtnl_net_lock(net); 1982 err = __register_netdevice_notifier_net(net, nb, false); 1983 if (!err) { 1984 nn->nb = nb; 1985 list_add(&nn->list, &dev->net_notifier_list); 1986 } 1987 rtnl_net_unlock(net); 1988 1989 return err; 1990 } 1991 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1992 1993 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1994 struct notifier_block *nb, 1995 struct netdev_net_notifier *nn) 1996 { 1997 struct net *net = dev_net(dev); 1998 int err; 1999 2000 rtnl_net_lock(net); 2001 list_del(&nn->list); 2002 err = __unregister_netdevice_notifier_net(net, nb); 2003 rtnl_net_unlock(net); 2004 2005 return err; 2006 } 2007 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2008 2009 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2010 struct net *net) 2011 { 2012 struct netdev_net_notifier *nn; 2013 2014 list_for_each_entry(nn, &dev->net_notifier_list, list) 2015 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 2016 } 2017 2018 /** 2019 * call_netdevice_notifiers_info - call all network notifier blocks 2020 * @val: value passed unmodified to notifier function 2021 * @info: notifier information data 2022 * 2023 * Call all network notifier blocks. Parameters and return value 2024 * are as for raw_notifier_call_chain(). 2025 */ 2026 2027 int call_netdevice_notifiers_info(unsigned long val, 2028 struct netdev_notifier_info *info) 2029 { 2030 struct net *net = dev_net(info->dev); 2031 int ret; 2032 2033 ASSERT_RTNL(); 2034 2035 /* Run per-netns notifier block chain first, then run the global one. 2036 * Hopefully, one day, the global one is going to be removed after 2037 * all notifier block registrators get converted to be per-netns. 2038 */ 2039 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2040 if (ret & NOTIFY_STOP_MASK) 2041 return ret; 2042 return raw_notifier_call_chain(&netdev_chain, val, info); 2043 } 2044 2045 /** 2046 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2047 * for and rollback on error 2048 * @val_up: value passed unmodified to notifier function 2049 * @val_down: value passed unmodified to the notifier function when 2050 * recovering from an error on @val_up 2051 * @info: notifier information data 2052 * 2053 * Call all per-netns network notifier blocks, but not notifier blocks on 2054 * the global notifier chain. Parameters and return value are as for 2055 * raw_notifier_call_chain_robust(). 2056 */ 2057 2058 static int 2059 call_netdevice_notifiers_info_robust(unsigned long val_up, 2060 unsigned long val_down, 2061 struct netdev_notifier_info *info) 2062 { 2063 struct net *net = dev_net(info->dev); 2064 2065 ASSERT_RTNL(); 2066 2067 return raw_notifier_call_chain_robust(&net->netdev_chain, 2068 val_up, val_down, info); 2069 } 2070 2071 static int call_netdevice_notifiers_extack(unsigned long val, 2072 struct net_device *dev, 2073 struct netlink_ext_ack *extack) 2074 { 2075 struct netdev_notifier_info info = { 2076 .dev = dev, 2077 .extack = extack, 2078 }; 2079 2080 return call_netdevice_notifiers_info(val, &info); 2081 } 2082 2083 /** 2084 * call_netdevice_notifiers - call all network notifier blocks 2085 * @val: value passed unmodified to notifier function 2086 * @dev: net_device pointer passed unmodified to notifier function 2087 * 2088 * Call all network notifier blocks. Parameters and return value 2089 * are as for raw_notifier_call_chain(). 2090 */ 2091 2092 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2093 { 2094 return call_netdevice_notifiers_extack(val, dev, NULL); 2095 } 2096 EXPORT_SYMBOL(call_netdevice_notifiers); 2097 2098 /** 2099 * call_netdevice_notifiers_mtu - call all network notifier blocks 2100 * @val: value passed unmodified to notifier function 2101 * @dev: net_device pointer passed unmodified to notifier function 2102 * @arg: additional u32 argument passed to the notifier function 2103 * 2104 * Call all network notifier blocks. Parameters and return value 2105 * are as for raw_notifier_call_chain(). 2106 */ 2107 static int call_netdevice_notifiers_mtu(unsigned long val, 2108 struct net_device *dev, u32 arg) 2109 { 2110 struct netdev_notifier_info_ext info = { 2111 .info.dev = dev, 2112 .ext.mtu = arg, 2113 }; 2114 2115 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2116 2117 return call_netdevice_notifiers_info(val, &info.info); 2118 } 2119 2120 #ifdef CONFIG_NET_INGRESS 2121 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2122 2123 void net_inc_ingress_queue(void) 2124 { 2125 static_branch_inc(&ingress_needed_key); 2126 } 2127 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2128 2129 void net_dec_ingress_queue(void) 2130 { 2131 static_branch_dec(&ingress_needed_key); 2132 } 2133 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2134 #endif 2135 2136 #ifdef CONFIG_NET_EGRESS 2137 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2138 2139 void net_inc_egress_queue(void) 2140 { 2141 static_branch_inc(&egress_needed_key); 2142 } 2143 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2144 2145 void net_dec_egress_queue(void) 2146 { 2147 static_branch_dec(&egress_needed_key); 2148 } 2149 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2150 #endif 2151 2152 #ifdef CONFIG_NET_CLS_ACT 2153 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key); 2154 EXPORT_SYMBOL(tcf_bypass_check_needed_key); 2155 #endif 2156 2157 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2158 EXPORT_SYMBOL(netstamp_needed_key); 2159 #ifdef CONFIG_JUMP_LABEL 2160 static atomic_t netstamp_needed_deferred; 2161 static atomic_t netstamp_wanted; 2162 static void netstamp_clear(struct work_struct *work) 2163 { 2164 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2165 int wanted; 2166 2167 wanted = atomic_add_return(deferred, &netstamp_wanted); 2168 if (wanted > 0) 2169 static_branch_enable(&netstamp_needed_key); 2170 else 2171 static_branch_disable(&netstamp_needed_key); 2172 } 2173 static DECLARE_WORK(netstamp_work, netstamp_clear); 2174 #endif 2175 2176 void net_enable_timestamp(void) 2177 { 2178 #ifdef CONFIG_JUMP_LABEL 2179 int wanted = atomic_read(&netstamp_wanted); 2180 2181 while (wanted > 0) { 2182 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2183 return; 2184 } 2185 atomic_inc(&netstamp_needed_deferred); 2186 schedule_work(&netstamp_work); 2187 #else 2188 static_branch_inc(&netstamp_needed_key); 2189 #endif 2190 } 2191 EXPORT_SYMBOL(net_enable_timestamp); 2192 2193 void net_disable_timestamp(void) 2194 { 2195 #ifdef CONFIG_JUMP_LABEL 2196 int wanted = atomic_read(&netstamp_wanted); 2197 2198 while (wanted > 1) { 2199 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2200 return; 2201 } 2202 atomic_dec(&netstamp_needed_deferred); 2203 schedule_work(&netstamp_work); 2204 #else 2205 static_branch_dec(&netstamp_needed_key); 2206 #endif 2207 } 2208 EXPORT_SYMBOL(net_disable_timestamp); 2209 2210 static inline void net_timestamp_set(struct sk_buff *skb) 2211 { 2212 skb->tstamp = 0; 2213 skb->tstamp_type = SKB_CLOCK_REALTIME; 2214 if (static_branch_unlikely(&netstamp_needed_key)) 2215 skb->tstamp = ktime_get_real(); 2216 } 2217 2218 #define net_timestamp_check(COND, SKB) \ 2219 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2220 if ((COND) && !(SKB)->tstamp) \ 2221 (SKB)->tstamp = ktime_get_real(); \ 2222 } \ 2223 2224 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2225 { 2226 return __is_skb_forwardable(dev, skb, true); 2227 } 2228 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2229 2230 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2231 bool check_mtu) 2232 { 2233 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2234 2235 if (likely(!ret)) { 2236 skb->protocol = eth_type_trans(skb, dev); 2237 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2238 } 2239 2240 return ret; 2241 } 2242 2243 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2244 { 2245 return __dev_forward_skb2(dev, skb, true); 2246 } 2247 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2248 2249 /** 2250 * dev_forward_skb - loopback an skb to another netif 2251 * 2252 * @dev: destination network device 2253 * @skb: buffer to forward 2254 * 2255 * return values: 2256 * NET_RX_SUCCESS (no congestion) 2257 * NET_RX_DROP (packet was dropped, but freed) 2258 * 2259 * dev_forward_skb can be used for injecting an skb from the 2260 * start_xmit function of one device into the receive queue 2261 * of another device. 2262 * 2263 * The receiving device may be in another namespace, so 2264 * we have to clear all information in the skb that could 2265 * impact namespace isolation. 2266 */ 2267 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2268 { 2269 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2270 } 2271 EXPORT_SYMBOL_GPL(dev_forward_skb); 2272 2273 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2274 { 2275 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2276 } 2277 2278 static inline int deliver_skb(struct sk_buff *skb, 2279 struct packet_type *pt_prev, 2280 struct net_device *orig_dev) 2281 { 2282 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2283 return -ENOMEM; 2284 refcount_inc(&skb->users); 2285 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2286 } 2287 2288 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2289 struct packet_type **pt, 2290 struct net_device *orig_dev, 2291 __be16 type, 2292 struct list_head *ptype_list) 2293 { 2294 struct packet_type *ptype, *pt_prev = *pt; 2295 2296 list_for_each_entry_rcu(ptype, ptype_list, list) { 2297 if (ptype->type != type) 2298 continue; 2299 if (pt_prev) 2300 deliver_skb(skb, pt_prev, orig_dev); 2301 pt_prev = ptype; 2302 } 2303 *pt = pt_prev; 2304 } 2305 2306 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2307 { 2308 if (!ptype->af_packet_priv || !skb->sk) 2309 return false; 2310 2311 if (ptype->id_match) 2312 return ptype->id_match(ptype, skb->sk); 2313 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2314 return true; 2315 2316 return false; 2317 } 2318 2319 /** 2320 * dev_nit_active - return true if any network interface taps are in use 2321 * 2322 * @dev: network device to check for the presence of taps 2323 */ 2324 bool dev_nit_active(struct net_device *dev) 2325 { 2326 return !list_empty(&net_hotdata.ptype_all) || 2327 !list_empty(&dev->ptype_all); 2328 } 2329 EXPORT_SYMBOL_GPL(dev_nit_active); 2330 2331 /* 2332 * Support routine. Sends outgoing frames to any network 2333 * taps currently in use. 2334 */ 2335 2336 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2337 { 2338 struct list_head *ptype_list = &net_hotdata.ptype_all; 2339 struct packet_type *ptype, *pt_prev = NULL; 2340 struct sk_buff *skb2 = NULL; 2341 2342 rcu_read_lock(); 2343 again: 2344 list_for_each_entry_rcu(ptype, ptype_list, list) { 2345 if (READ_ONCE(ptype->ignore_outgoing)) 2346 continue; 2347 2348 /* Never send packets back to the socket 2349 * they originated from - MvS ([email protected]) 2350 */ 2351 if (skb_loop_sk(ptype, skb)) 2352 continue; 2353 2354 if (pt_prev) { 2355 deliver_skb(skb2, pt_prev, skb->dev); 2356 pt_prev = ptype; 2357 continue; 2358 } 2359 2360 /* need to clone skb, done only once */ 2361 skb2 = skb_clone(skb, GFP_ATOMIC); 2362 if (!skb2) 2363 goto out_unlock; 2364 2365 net_timestamp_set(skb2); 2366 2367 /* skb->nh should be correctly 2368 * set by sender, so that the second statement is 2369 * just protection against buggy protocols. 2370 */ 2371 skb_reset_mac_header(skb2); 2372 2373 if (skb_network_header(skb2) < skb2->data || 2374 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2375 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2376 ntohs(skb2->protocol), 2377 dev->name); 2378 skb_reset_network_header(skb2); 2379 } 2380 2381 skb2->transport_header = skb2->network_header; 2382 skb2->pkt_type = PACKET_OUTGOING; 2383 pt_prev = ptype; 2384 } 2385 2386 if (ptype_list == &net_hotdata.ptype_all) { 2387 ptype_list = &dev->ptype_all; 2388 goto again; 2389 } 2390 out_unlock: 2391 if (pt_prev) { 2392 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2393 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2394 else 2395 kfree_skb(skb2); 2396 } 2397 rcu_read_unlock(); 2398 } 2399 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2400 2401 /** 2402 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2403 * @dev: Network device 2404 * @txq: number of queues available 2405 * 2406 * If real_num_tx_queues is changed the tc mappings may no longer be 2407 * valid. To resolve this verify the tc mapping remains valid and if 2408 * not NULL the mapping. With no priorities mapping to this 2409 * offset/count pair it will no longer be used. In the worst case TC0 2410 * is invalid nothing can be done so disable priority mappings. If is 2411 * expected that drivers will fix this mapping if they can before 2412 * calling netif_set_real_num_tx_queues. 2413 */ 2414 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2415 { 2416 int i; 2417 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2418 2419 /* If TC0 is invalidated disable TC mapping */ 2420 if (tc->offset + tc->count > txq) { 2421 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2422 dev->num_tc = 0; 2423 return; 2424 } 2425 2426 /* Invalidated prio to tc mappings set to TC0 */ 2427 for (i = 1; i < TC_BITMASK + 1; i++) { 2428 int q = netdev_get_prio_tc_map(dev, i); 2429 2430 tc = &dev->tc_to_txq[q]; 2431 if (tc->offset + tc->count > txq) { 2432 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", 2433 i, q); 2434 netdev_set_prio_tc_map(dev, i, 0); 2435 } 2436 } 2437 } 2438 2439 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2440 { 2441 if (dev->num_tc) { 2442 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2443 int i; 2444 2445 /* walk through the TCs and see if it falls into any of them */ 2446 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2447 if ((txq - tc->offset) < tc->count) 2448 return i; 2449 } 2450 2451 /* didn't find it, just return -1 to indicate no match */ 2452 return -1; 2453 } 2454 2455 return 0; 2456 } 2457 EXPORT_SYMBOL(netdev_txq_to_tc); 2458 2459 #ifdef CONFIG_XPS 2460 static struct static_key xps_needed __read_mostly; 2461 static struct static_key xps_rxqs_needed __read_mostly; 2462 static DEFINE_MUTEX(xps_map_mutex); 2463 #define xmap_dereference(P) \ 2464 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2465 2466 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2467 struct xps_dev_maps *old_maps, int tci, u16 index) 2468 { 2469 struct xps_map *map = NULL; 2470 int pos; 2471 2472 map = xmap_dereference(dev_maps->attr_map[tci]); 2473 if (!map) 2474 return false; 2475 2476 for (pos = map->len; pos--;) { 2477 if (map->queues[pos] != index) 2478 continue; 2479 2480 if (map->len > 1) { 2481 map->queues[pos] = map->queues[--map->len]; 2482 break; 2483 } 2484 2485 if (old_maps) 2486 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2487 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2488 kfree_rcu(map, rcu); 2489 return false; 2490 } 2491 2492 return true; 2493 } 2494 2495 static bool remove_xps_queue_cpu(struct net_device *dev, 2496 struct xps_dev_maps *dev_maps, 2497 int cpu, u16 offset, u16 count) 2498 { 2499 int num_tc = dev_maps->num_tc; 2500 bool active = false; 2501 int tci; 2502 2503 for (tci = cpu * num_tc; num_tc--; tci++) { 2504 int i, j; 2505 2506 for (i = count, j = offset; i--; j++) { 2507 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2508 break; 2509 } 2510 2511 active |= i < 0; 2512 } 2513 2514 return active; 2515 } 2516 2517 static void reset_xps_maps(struct net_device *dev, 2518 struct xps_dev_maps *dev_maps, 2519 enum xps_map_type type) 2520 { 2521 static_key_slow_dec_cpuslocked(&xps_needed); 2522 if (type == XPS_RXQS) 2523 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2524 2525 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2526 2527 kfree_rcu(dev_maps, rcu); 2528 } 2529 2530 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2531 u16 offset, u16 count) 2532 { 2533 struct xps_dev_maps *dev_maps; 2534 bool active = false; 2535 int i, j; 2536 2537 dev_maps = xmap_dereference(dev->xps_maps[type]); 2538 if (!dev_maps) 2539 return; 2540 2541 for (j = 0; j < dev_maps->nr_ids; j++) 2542 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2543 if (!active) 2544 reset_xps_maps(dev, dev_maps, type); 2545 2546 if (type == XPS_CPUS) { 2547 for (i = offset + (count - 1); count--; i--) 2548 netdev_queue_numa_node_write( 2549 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2550 } 2551 } 2552 2553 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2554 u16 count) 2555 { 2556 if (!static_key_false(&xps_needed)) 2557 return; 2558 2559 cpus_read_lock(); 2560 mutex_lock(&xps_map_mutex); 2561 2562 if (static_key_false(&xps_rxqs_needed)) 2563 clean_xps_maps(dev, XPS_RXQS, offset, count); 2564 2565 clean_xps_maps(dev, XPS_CPUS, offset, count); 2566 2567 mutex_unlock(&xps_map_mutex); 2568 cpus_read_unlock(); 2569 } 2570 2571 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2572 { 2573 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2574 } 2575 2576 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2577 u16 index, bool is_rxqs_map) 2578 { 2579 struct xps_map *new_map; 2580 int alloc_len = XPS_MIN_MAP_ALLOC; 2581 int i, pos; 2582 2583 for (pos = 0; map && pos < map->len; pos++) { 2584 if (map->queues[pos] != index) 2585 continue; 2586 return map; 2587 } 2588 2589 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2590 if (map) { 2591 if (pos < map->alloc_len) 2592 return map; 2593 2594 alloc_len = map->alloc_len * 2; 2595 } 2596 2597 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2598 * map 2599 */ 2600 if (is_rxqs_map) 2601 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2602 else 2603 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2604 cpu_to_node(attr_index)); 2605 if (!new_map) 2606 return NULL; 2607 2608 for (i = 0; i < pos; i++) 2609 new_map->queues[i] = map->queues[i]; 2610 new_map->alloc_len = alloc_len; 2611 new_map->len = pos; 2612 2613 return new_map; 2614 } 2615 2616 /* Copy xps maps at a given index */ 2617 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2618 struct xps_dev_maps *new_dev_maps, int index, 2619 int tc, bool skip_tc) 2620 { 2621 int i, tci = index * dev_maps->num_tc; 2622 struct xps_map *map; 2623 2624 /* copy maps belonging to foreign traffic classes */ 2625 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2626 if (i == tc && skip_tc) 2627 continue; 2628 2629 /* fill in the new device map from the old device map */ 2630 map = xmap_dereference(dev_maps->attr_map[tci]); 2631 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2632 } 2633 } 2634 2635 /* Must be called under cpus_read_lock */ 2636 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2637 u16 index, enum xps_map_type type) 2638 { 2639 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2640 const unsigned long *online_mask = NULL; 2641 bool active = false, copy = false; 2642 int i, j, tci, numa_node_id = -2; 2643 int maps_sz, num_tc = 1, tc = 0; 2644 struct xps_map *map, *new_map; 2645 unsigned int nr_ids; 2646 2647 WARN_ON_ONCE(index >= dev->num_tx_queues); 2648 2649 if (dev->num_tc) { 2650 /* Do not allow XPS on subordinate device directly */ 2651 num_tc = dev->num_tc; 2652 if (num_tc < 0) 2653 return -EINVAL; 2654 2655 /* If queue belongs to subordinate dev use its map */ 2656 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2657 2658 tc = netdev_txq_to_tc(dev, index); 2659 if (tc < 0) 2660 return -EINVAL; 2661 } 2662 2663 mutex_lock(&xps_map_mutex); 2664 2665 dev_maps = xmap_dereference(dev->xps_maps[type]); 2666 if (type == XPS_RXQS) { 2667 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2668 nr_ids = dev->num_rx_queues; 2669 } else { 2670 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2671 if (num_possible_cpus() > 1) 2672 online_mask = cpumask_bits(cpu_online_mask); 2673 nr_ids = nr_cpu_ids; 2674 } 2675 2676 if (maps_sz < L1_CACHE_BYTES) 2677 maps_sz = L1_CACHE_BYTES; 2678 2679 /* The old dev_maps could be larger or smaller than the one we're 2680 * setting up now, as dev->num_tc or nr_ids could have been updated in 2681 * between. We could try to be smart, but let's be safe instead and only 2682 * copy foreign traffic classes if the two map sizes match. 2683 */ 2684 if (dev_maps && 2685 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2686 copy = true; 2687 2688 /* allocate memory for queue storage */ 2689 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2690 j < nr_ids;) { 2691 if (!new_dev_maps) { 2692 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2693 if (!new_dev_maps) { 2694 mutex_unlock(&xps_map_mutex); 2695 return -ENOMEM; 2696 } 2697 2698 new_dev_maps->nr_ids = nr_ids; 2699 new_dev_maps->num_tc = num_tc; 2700 } 2701 2702 tci = j * num_tc + tc; 2703 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2704 2705 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2706 if (!map) 2707 goto error; 2708 2709 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2710 } 2711 2712 if (!new_dev_maps) 2713 goto out_no_new_maps; 2714 2715 if (!dev_maps) { 2716 /* Increment static keys at most once per type */ 2717 static_key_slow_inc_cpuslocked(&xps_needed); 2718 if (type == XPS_RXQS) 2719 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2720 } 2721 2722 for (j = 0; j < nr_ids; j++) { 2723 bool skip_tc = false; 2724 2725 tci = j * num_tc + tc; 2726 if (netif_attr_test_mask(j, mask, nr_ids) && 2727 netif_attr_test_online(j, online_mask, nr_ids)) { 2728 /* add tx-queue to CPU/rx-queue maps */ 2729 int pos = 0; 2730 2731 skip_tc = true; 2732 2733 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2734 while ((pos < map->len) && (map->queues[pos] != index)) 2735 pos++; 2736 2737 if (pos == map->len) 2738 map->queues[map->len++] = index; 2739 #ifdef CONFIG_NUMA 2740 if (type == XPS_CPUS) { 2741 if (numa_node_id == -2) 2742 numa_node_id = cpu_to_node(j); 2743 else if (numa_node_id != cpu_to_node(j)) 2744 numa_node_id = -1; 2745 } 2746 #endif 2747 } 2748 2749 if (copy) 2750 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2751 skip_tc); 2752 } 2753 2754 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2755 2756 /* Cleanup old maps */ 2757 if (!dev_maps) 2758 goto out_no_old_maps; 2759 2760 for (j = 0; j < dev_maps->nr_ids; j++) { 2761 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2762 map = xmap_dereference(dev_maps->attr_map[tci]); 2763 if (!map) 2764 continue; 2765 2766 if (copy) { 2767 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2768 if (map == new_map) 2769 continue; 2770 } 2771 2772 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2773 kfree_rcu(map, rcu); 2774 } 2775 } 2776 2777 old_dev_maps = dev_maps; 2778 2779 out_no_old_maps: 2780 dev_maps = new_dev_maps; 2781 active = true; 2782 2783 out_no_new_maps: 2784 if (type == XPS_CPUS) 2785 /* update Tx queue numa node */ 2786 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2787 (numa_node_id >= 0) ? 2788 numa_node_id : NUMA_NO_NODE); 2789 2790 if (!dev_maps) 2791 goto out_no_maps; 2792 2793 /* removes tx-queue from unused CPUs/rx-queues */ 2794 for (j = 0; j < dev_maps->nr_ids; j++) { 2795 tci = j * dev_maps->num_tc; 2796 2797 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2798 if (i == tc && 2799 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2800 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2801 continue; 2802 2803 active |= remove_xps_queue(dev_maps, 2804 copy ? old_dev_maps : NULL, 2805 tci, index); 2806 } 2807 } 2808 2809 if (old_dev_maps) 2810 kfree_rcu(old_dev_maps, rcu); 2811 2812 /* free map if not active */ 2813 if (!active) 2814 reset_xps_maps(dev, dev_maps, type); 2815 2816 out_no_maps: 2817 mutex_unlock(&xps_map_mutex); 2818 2819 return 0; 2820 error: 2821 /* remove any maps that we added */ 2822 for (j = 0; j < nr_ids; j++) { 2823 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2824 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2825 map = copy ? 2826 xmap_dereference(dev_maps->attr_map[tci]) : 2827 NULL; 2828 if (new_map && new_map != map) 2829 kfree(new_map); 2830 } 2831 } 2832 2833 mutex_unlock(&xps_map_mutex); 2834 2835 kfree(new_dev_maps); 2836 return -ENOMEM; 2837 } 2838 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2839 2840 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2841 u16 index) 2842 { 2843 int ret; 2844 2845 cpus_read_lock(); 2846 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2847 cpus_read_unlock(); 2848 2849 return ret; 2850 } 2851 EXPORT_SYMBOL(netif_set_xps_queue); 2852 2853 #endif 2854 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2855 { 2856 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2857 2858 /* Unbind any subordinate channels */ 2859 while (txq-- != &dev->_tx[0]) { 2860 if (txq->sb_dev) 2861 netdev_unbind_sb_channel(dev, txq->sb_dev); 2862 } 2863 } 2864 2865 void netdev_reset_tc(struct net_device *dev) 2866 { 2867 #ifdef CONFIG_XPS 2868 netif_reset_xps_queues_gt(dev, 0); 2869 #endif 2870 netdev_unbind_all_sb_channels(dev); 2871 2872 /* Reset TC configuration of device */ 2873 dev->num_tc = 0; 2874 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2875 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2876 } 2877 EXPORT_SYMBOL(netdev_reset_tc); 2878 2879 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2880 { 2881 if (tc >= dev->num_tc) 2882 return -EINVAL; 2883 2884 #ifdef CONFIG_XPS 2885 netif_reset_xps_queues(dev, offset, count); 2886 #endif 2887 dev->tc_to_txq[tc].count = count; 2888 dev->tc_to_txq[tc].offset = offset; 2889 return 0; 2890 } 2891 EXPORT_SYMBOL(netdev_set_tc_queue); 2892 2893 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2894 { 2895 if (num_tc > TC_MAX_QUEUE) 2896 return -EINVAL; 2897 2898 #ifdef CONFIG_XPS 2899 netif_reset_xps_queues_gt(dev, 0); 2900 #endif 2901 netdev_unbind_all_sb_channels(dev); 2902 2903 dev->num_tc = num_tc; 2904 return 0; 2905 } 2906 EXPORT_SYMBOL(netdev_set_num_tc); 2907 2908 void netdev_unbind_sb_channel(struct net_device *dev, 2909 struct net_device *sb_dev) 2910 { 2911 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2912 2913 #ifdef CONFIG_XPS 2914 netif_reset_xps_queues_gt(sb_dev, 0); 2915 #endif 2916 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2917 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2918 2919 while (txq-- != &dev->_tx[0]) { 2920 if (txq->sb_dev == sb_dev) 2921 txq->sb_dev = NULL; 2922 } 2923 } 2924 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2925 2926 int netdev_bind_sb_channel_queue(struct net_device *dev, 2927 struct net_device *sb_dev, 2928 u8 tc, u16 count, u16 offset) 2929 { 2930 /* Make certain the sb_dev and dev are already configured */ 2931 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2932 return -EINVAL; 2933 2934 /* We cannot hand out queues we don't have */ 2935 if ((offset + count) > dev->real_num_tx_queues) 2936 return -EINVAL; 2937 2938 /* Record the mapping */ 2939 sb_dev->tc_to_txq[tc].count = count; 2940 sb_dev->tc_to_txq[tc].offset = offset; 2941 2942 /* Provide a way for Tx queue to find the tc_to_txq map or 2943 * XPS map for itself. 2944 */ 2945 while (count--) 2946 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2947 2948 return 0; 2949 } 2950 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2951 2952 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2953 { 2954 /* Do not use a multiqueue device to represent a subordinate channel */ 2955 if (netif_is_multiqueue(dev)) 2956 return -ENODEV; 2957 2958 /* We allow channels 1 - 32767 to be used for subordinate channels. 2959 * Channel 0 is meant to be "native" mode and used only to represent 2960 * the main root device. We allow writing 0 to reset the device back 2961 * to normal mode after being used as a subordinate channel. 2962 */ 2963 if (channel > S16_MAX) 2964 return -EINVAL; 2965 2966 dev->num_tc = -channel; 2967 2968 return 0; 2969 } 2970 EXPORT_SYMBOL(netdev_set_sb_channel); 2971 2972 /* 2973 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2974 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2975 */ 2976 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2977 { 2978 bool disabling; 2979 int rc; 2980 2981 disabling = txq < dev->real_num_tx_queues; 2982 2983 if (txq < 1 || txq > dev->num_tx_queues) 2984 return -EINVAL; 2985 2986 if (dev->reg_state == NETREG_REGISTERED || 2987 dev->reg_state == NETREG_UNREGISTERING) { 2988 ASSERT_RTNL(); 2989 2990 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2991 txq); 2992 if (rc) 2993 return rc; 2994 2995 if (dev->num_tc) 2996 netif_setup_tc(dev, txq); 2997 2998 net_shaper_set_real_num_tx_queues(dev, txq); 2999 3000 dev_qdisc_change_real_num_tx(dev, txq); 3001 3002 dev->real_num_tx_queues = txq; 3003 3004 if (disabling) { 3005 synchronize_net(); 3006 qdisc_reset_all_tx_gt(dev, txq); 3007 #ifdef CONFIG_XPS 3008 netif_reset_xps_queues_gt(dev, txq); 3009 #endif 3010 } 3011 } else { 3012 dev->real_num_tx_queues = txq; 3013 } 3014 3015 return 0; 3016 } 3017 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3018 3019 #ifdef CONFIG_SYSFS 3020 /** 3021 * netif_set_real_num_rx_queues - set actual number of RX queues used 3022 * @dev: Network device 3023 * @rxq: Actual number of RX queues 3024 * 3025 * This must be called either with the rtnl_lock held or before 3026 * registration of the net device. Returns 0 on success, or a 3027 * negative error code. If called before registration, it always 3028 * succeeds. 3029 */ 3030 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3031 { 3032 int rc; 3033 3034 if (rxq < 1 || rxq > dev->num_rx_queues) 3035 return -EINVAL; 3036 3037 if (dev->reg_state == NETREG_REGISTERED) { 3038 ASSERT_RTNL(); 3039 3040 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3041 rxq); 3042 if (rc) 3043 return rc; 3044 } 3045 3046 dev->real_num_rx_queues = rxq; 3047 return 0; 3048 } 3049 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3050 #endif 3051 3052 /** 3053 * netif_set_real_num_queues - set actual number of RX and TX queues used 3054 * @dev: Network device 3055 * @txq: Actual number of TX queues 3056 * @rxq: Actual number of RX queues 3057 * 3058 * Set the real number of both TX and RX queues. 3059 * Does nothing if the number of queues is already correct. 3060 */ 3061 int netif_set_real_num_queues(struct net_device *dev, 3062 unsigned int txq, unsigned int rxq) 3063 { 3064 unsigned int old_rxq = dev->real_num_rx_queues; 3065 int err; 3066 3067 if (txq < 1 || txq > dev->num_tx_queues || 3068 rxq < 1 || rxq > dev->num_rx_queues) 3069 return -EINVAL; 3070 3071 /* Start from increases, so the error path only does decreases - 3072 * decreases can't fail. 3073 */ 3074 if (rxq > dev->real_num_rx_queues) { 3075 err = netif_set_real_num_rx_queues(dev, rxq); 3076 if (err) 3077 return err; 3078 } 3079 if (txq > dev->real_num_tx_queues) { 3080 err = netif_set_real_num_tx_queues(dev, txq); 3081 if (err) 3082 goto undo_rx; 3083 } 3084 if (rxq < dev->real_num_rx_queues) 3085 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3086 if (txq < dev->real_num_tx_queues) 3087 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3088 3089 return 0; 3090 undo_rx: 3091 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3092 return err; 3093 } 3094 EXPORT_SYMBOL(netif_set_real_num_queues); 3095 3096 /** 3097 * netif_set_tso_max_size() - set the max size of TSO frames supported 3098 * @dev: netdev to update 3099 * @size: max skb->len of a TSO frame 3100 * 3101 * Set the limit on the size of TSO super-frames the device can handle. 3102 * Unless explicitly set the stack will assume the value of 3103 * %GSO_LEGACY_MAX_SIZE. 3104 */ 3105 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3106 { 3107 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3108 if (size < READ_ONCE(dev->gso_max_size)) 3109 netif_set_gso_max_size(dev, size); 3110 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3111 netif_set_gso_ipv4_max_size(dev, size); 3112 } 3113 EXPORT_SYMBOL(netif_set_tso_max_size); 3114 3115 /** 3116 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3117 * @dev: netdev to update 3118 * @segs: max number of TCP segments 3119 * 3120 * Set the limit on the number of TCP segments the device can generate from 3121 * a single TSO super-frame. 3122 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3123 */ 3124 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3125 { 3126 dev->tso_max_segs = segs; 3127 if (segs < READ_ONCE(dev->gso_max_segs)) 3128 netif_set_gso_max_segs(dev, segs); 3129 } 3130 EXPORT_SYMBOL(netif_set_tso_max_segs); 3131 3132 /** 3133 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3134 * @to: netdev to update 3135 * @from: netdev from which to copy the limits 3136 */ 3137 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3138 { 3139 netif_set_tso_max_size(to, from->tso_max_size); 3140 netif_set_tso_max_segs(to, from->tso_max_segs); 3141 } 3142 EXPORT_SYMBOL(netif_inherit_tso_max); 3143 3144 /** 3145 * netif_get_num_default_rss_queues - default number of RSS queues 3146 * 3147 * Default value is the number of physical cores if there are only 1 or 2, or 3148 * divided by 2 if there are more. 3149 */ 3150 int netif_get_num_default_rss_queues(void) 3151 { 3152 cpumask_var_t cpus; 3153 int cpu, count = 0; 3154 3155 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3156 return 1; 3157 3158 cpumask_copy(cpus, cpu_online_mask); 3159 for_each_cpu(cpu, cpus) { 3160 ++count; 3161 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3162 } 3163 free_cpumask_var(cpus); 3164 3165 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3166 } 3167 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3168 3169 static void __netif_reschedule(struct Qdisc *q) 3170 { 3171 struct softnet_data *sd; 3172 unsigned long flags; 3173 3174 local_irq_save(flags); 3175 sd = this_cpu_ptr(&softnet_data); 3176 q->next_sched = NULL; 3177 *sd->output_queue_tailp = q; 3178 sd->output_queue_tailp = &q->next_sched; 3179 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3180 local_irq_restore(flags); 3181 } 3182 3183 void __netif_schedule(struct Qdisc *q) 3184 { 3185 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3186 __netif_reschedule(q); 3187 } 3188 EXPORT_SYMBOL(__netif_schedule); 3189 3190 struct dev_kfree_skb_cb { 3191 enum skb_drop_reason reason; 3192 }; 3193 3194 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3195 { 3196 return (struct dev_kfree_skb_cb *)skb->cb; 3197 } 3198 3199 void netif_schedule_queue(struct netdev_queue *txq) 3200 { 3201 rcu_read_lock(); 3202 if (!netif_xmit_stopped(txq)) { 3203 struct Qdisc *q = rcu_dereference(txq->qdisc); 3204 3205 __netif_schedule(q); 3206 } 3207 rcu_read_unlock(); 3208 } 3209 EXPORT_SYMBOL(netif_schedule_queue); 3210 3211 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3212 { 3213 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3214 struct Qdisc *q; 3215 3216 rcu_read_lock(); 3217 q = rcu_dereference(dev_queue->qdisc); 3218 __netif_schedule(q); 3219 rcu_read_unlock(); 3220 } 3221 } 3222 EXPORT_SYMBOL(netif_tx_wake_queue); 3223 3224 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3225 { 3226 unsigned long flags; 3227 3228 if (unlikely(!skb)) 3229 return; 3230 3231 if (likely(refcount_read(&skb->users) == 1)) { 3232 smp_rmb(); 3233 refcount_set(&skb->users, 0); 3234 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3235 return; 3236 } 3237 get_kfree_skb_cb(skb)->reason = reason; 3238 local_irq_save(flags); 3239 skb->next = __this_cpu_read(softnet_data.completion_queue); 3240 __this_cpu_write(softnet_data.completion_queue, skb); 3241 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3242 local_irq_restore(flags); 3243 } 3244 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3245 3246 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3247 { 3248 if (in_hardirq() || irqs_disabled()) 3249 dev_kfree_skb_irq_reason(skb, reason); 3250 else 3251 kfree_skb_reason(skb, reason); 3252 } 3253 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3254 3255 3256 /** 3257 * netif_device_detach - mark device as removed 3258 * @dev: network device 3259 * 3260 * Mark device as removed from system and therefore no longer available. 3261 */ 3262 void netif_device_detach(struct net_device *dev) 3263 { 3264 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3265 netif_running(dev)) { 3266 netif_tx_stop_all_queues(dev); 3267 } 3268 } 3269 EXPORT_SYMBOL(netif_device_detach); 3270 3271 /** 3272 * netif_device_attach - mark device as attached 3273 * @dev: network device 3274 * 3275 * Mark device as attached from system and restart if needed. 3276 */ 3277 void netif_device_attach(struct net_device *dev) 3278 { 3279 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3280 netif_running(dev)) { 3281 netif_tx_wake_all_queues(dev); 3282 netdev_watchdog_up(dev); 3283 } 3284 } 3285 EXPORT_SYMBOL(netif_device_attach); 3286 3287 /* 3288 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3289 * to be used as a distribution range. 3290 */ 3291 static u16 skb_tx_hash(const struct net_device *dev, 3292 const struct net_device *sb_dev, 3293 struct sk_buff *skb) 3294 { 3295 u32 hash; 3296 u16 qoffset = 0; 3297 u16 qcount = dev->real_num_tx_queues; 3298 3299 if (dev->num_tc) { 3300 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3301 3302 qoffset = sb_dev->tc_to_txq[tc].offset; 3303 qcount = sb_dev->tc_to_txq[tc].count; 3304 if (unlikely(!qcount)) { 3305 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3306 sb_dev->name, qoffset, tc); 3307 qoffset = 0; 3308 qcount = dev->real_num_tx_queues; 3309 } 3310 } 3311 3312 if (skb_rx_queue_recorded(skb)) { 3313 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3314 hash = skb_get_rx_queue(skb); 3315 if (hash >= qoffset) 3316 hash -= qoffset; 3317 while (unlikely(hash >= qcount)) 3318 hash -= qcount; 3319 return hash + qoffset; 3320 } 3321 3322 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3323 } 3324 3325 void skb_warn_bad_offload(const struct sk_buff *skb) 3326 { 3327 static const netdev_features_t null_features; 3328 struct net_device *dev = skb->dev; 3329 const char *name = ""; 3330 3331 if (!net_ratelimit()) 3332 return; 3333 3334 if (dev) { 3335 if (dev->dev.parent) 3336 name = dev_driver_string(dev->dev.parent); 3337 else 3338 name = netdev_name(dev); 3339 } 3340 skb_dump(KERN_WARNING, skb, false); 3341 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3342 name, dev ? &dev->features : &null_features, 3343 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3344 } 3345 3346 /* 3347 * Invalidate hardware checksum when packet is to be mangled, and 3348 * complete checksum manually on outgoing path. 3349 */ 3350 int skb_checksum_help(struct sk_buff *skb) 3351 { 3352 __wsum csum; 3353 int ret = 0, offset; 3354 3355 if (skb->ip_summed == CHECKSUM_COMPLETE) 3356 goto out_set_summed; 3357 3358 if (unlikely(skb_is_gso(skb))) { 3359 skb_warn_bad_offload(skb); 3360 return -EINVAL; 3361 } 3362 3363 if (!skb_frags_readable(skb)) { 3364 return -EFAULT; 3365 } 3366 3367 /* Before computing a checksum, we should make sure no frag could 3368 * be modified by an external entity : checksum could be wrong. 3369 */ 3370 if (skb_has_shared_frag(skb)) { 3371 ret = __skb_linearize(skb); 3372 if (ret) 3373 goto out; 3374 } 3375 3376 offset = skb_checksum_start_offset(skb); 3377 ret = -EINVAL; 3378 if (unlikely(offset >= skb_headlen(skb))) { 3379 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3380 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3381 offset, skb_headlen(skb)); 3382 goto out; 3383 } 3384 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3385 3386 offset += skb->csum_offset; 3387 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3388 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3389 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3390 offset + sizeof(__sum16), skb_headlen(skb)); 3391 goto out; 3392 } 3393 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3394 if (ret) 3395 goto out; 3396 3397 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3398 out_set_summed: 3399 skb->ip_summed = CHECKSUM_NONE; 3400 out: 3401 return ret; 3402 } 3403 EXPORT_SYMBOL(skb_checksum_help); 3404 3405 int skb_crc32c_csum_help(struct sk_buff *skb) 3406 { 3407 __le32 crc32c_csum; 3408 int ret = 0, offset, start; 3409 3410 if (skb->ip_summed != CHECKSUM_PARTIAL) 3411 goto out; 3412 3413 if (unlikely(skb_is_gso(skb))) 3414 goto out; 3415 3416 /* Before computing a checksum, we should make sure no frag could 3417 * be modified by an external entity : checksum could be wrong. 3418 */ 3419 if (unlikely(skb_has_shared_frag(skb))) { 3420 ret = __skb_linearize(skb); 3421 if (ret) 3422 goto out; 3423 } 3424 start = skb_checksum_start_offset(skb); 3425 offset = start + offsetof(struct sctphdr, checksum); 3426 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3427 ret = -EINVAL; 3428 goto out; 3429 } 3430 3431 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3432 if (ret) 3433 goto out; 3434 3435 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3436 skb->len - start, ~(__u32)0, 3437 crc32c_csum_stub)); 3438 *(__le32 *)(skb->data + offset) = crc32c_csum; 3439 skb_reset_csum_not_inet(skb); 3440 out: 3441 return ret; 3442 } 3443 EXPORT_SYMBOL(skb_crc32c_csum_help); 3444 3445 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3446 { 3447 __be16 type = skb->protocol; 3448 3449 /* Tunnel gso handlers can set protocol to ethernet. */ 3450 if (type == htons(ETH_P_TEB)) { 3451 struct ethhdr *eth; 3452 3453 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3454 return 0; 3455 3456 eth = (struct ethhdr *)skb->data; 3457 type = eth->h_proto; 3458 } 3459 3460 return vlan_get_protocol_and_depth(skb, type, depth); 3461 } 3462 3463 3464 /* Take action when hardware reception checksum errors are detected. */ 3465 #ifdef CONFIG_BUG 3466 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3467 { 3468 netdev_err(dev, "hw csum failure\n"); 3469 skb_dump(KERN_ERR, skb, true); 3470 dump_stack(); 3471 } 3472 3473 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3474 { 3475 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3476 } 3477 EXPORT_SYMBOL(netdev_rx_csum_fault); 3478 #endif 3479 3480 /* XXX: check that highmem exists at all on the given machine. */ 3481 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3482 { 3483 #ifdef CONFIG_HIGHMEM 3484 int i; 3485 3486 if (!(dev->features & NETIF_F_HIGHDMA)) { 3487 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3488 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3489 struct page *page = skb_frag_page(frag); 3490 3491 if (page && PageHighMem(page)) 3492 return 1; 3493 } 3494 } 3495 #endif 3496 return 0; 3497 } 3498 3499 /* If MPLS offload request, verify we are testing hardware MPLS features 3500 * instead of standard features for the netdev. 3501 */ 3502 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3503 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3504 netdev_features_t features, 3505 __be16 type) 3506 { 3507 if (eth_p_mpls(type)) 3508 features &= skb->dev->mpls_features; 3509 3510 return features; 3511 } 3512 #else 3513 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3514 netdev_features_t features, 3515 __be16 type) 3516 { 3517 return features; 3518 } 3519 #endif 3520 3521 static netdev_features_t harmonize_features(struct sk_buff *skb, 3522 netdev_features_t features) 3523 { 3524 __be16 type; 3525 3526 type = skb_network_protocol(skb, NULL); 3527 features = net_mpls_features(skb, features, type); 3528 3529 if (skb->ip_summed != CHECKSUM_NONE && 3530 !can_checksum_protocol(features, type)) { 3531 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3532 } 3533 if (illegal_highdma(skb->dev, skb)) 3534 features &= ~NETIF_F_SG; 3535 3536 return features; 3537 } 3538 3539 netdev_features_t passthru_features_check(struct sk_buff *skb, 3540 struct net_device *dev, 3541 netdev_features_t features) 3542 { 3543 return features; 3544 } 3545 EXPORT_SYMBOL(passthru_features_check); 3546 3547 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3548 struct net_device *dev, 3549 netdev_features_t features) 3550 { 3551 return vlan_features_check(skb, features); 3552 } 3553 3554 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3555 struct net_device *dev, 3556 netdev_features_t features) 3557 { 3558 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3559 3560 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3561 return features & ~NETIF_F_GSO_MASK; 3562 3563 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb))) 3564 return features & ~NETIF_F_GSO_MASK; 3565 3566 if (!skb_shinfo(skb)->gso_type) { 3567 skb_warn_bad_offload(skb); 3568 return features & ~NETIF_F_GSO_MASK; 3569 } 3570 3571 /* Support for GSO partial features requires software 3572 * intervention before we can actually process the packets 3573 * so we need to strip support for any partial features now 3574 * and we can pull them back in after we have partially 3575 * segmented the frame. 3576 */ 3577 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3578 features &= ~dev->gso_partial_features; 3579 3580 /* Make sure to clear the IPv4 ID mangling feature if the 3581 * IPv4 header has the potential to be fragmented. 3582 */ 3583 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3584 struct iphdr *iph = skb->encapsulation ? 3585 inner_ip_hdr(skb) : ip_hdr(skb); 3586 3587 if (!(iph->frag_off & htons(IP_DF))) 3588 features &= ~NETIF_F_TSO_MANGLEID; 3589 } 3590 3591 return features; 3592 } 3593 3594 netdev_features_t netif_skb_features(struct sk_buff *skb) 3595 { 3596 struct net_device *dev = skb->dev; 3597 netdev_features_t features = dev->features; 3598 3599 if (skb_is_gso(skb)) 3600 features = gso_features_check(skb, dev, features); 3601 3602 /* If encapsulation offload request, verify we are testing 3603 * hardware encapsulation features instead of standard 3604 * features for the netdev 3605 */ 3606 if (skb->encapsulation) 3607 features &= dev->hw_enc_features; 3608 3609 if (skb_vlan_tagged(skb)) 3610 features = netdev_intersect_features(features, 3611 dev->vlan_features | 3612 NETIF_F_HW_VLAN_CTAG_TX | 3613 NETIF_F_HW_VLAN_STAG_TX); 3614 3615 if (dev->netdev_ops->ndo_features_check) 3616 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3617 features); 3618 else 3619 features &= dflt_features_check(skb, dev, features); 3620 3621 return harmonize_features(skb, features); 3622 } 3623 EXPORT_SYMBOL(netif_skb_features); 3624 3625 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3626 struct netdev_queue *txq, bool more) 3627 { 3628 unsigned int len; 3629 int rc; 3630 3631 if (dev_nit_active(dev)) 3632 dev_queue_xmit_nit(skb, dev); 3633 3634 len = skb->len; 3635 trace_net_dev_start_xmit(skb, dev); 3636 rc = netdev_start_xmit(skb, dev, txq, more); 3637 trace_net_dev_xmit(skb, rc, dev, len); 3638 3639 return rc; 3640 } 3641 3642 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3643 struct netdev_queue *txq, int *ret) 3644 { 3645 struct sk_buff *skb = first; 3646 int rc = NETDEV_TX_OK; 3647 3648 while (skb) { 3649 struct sk_buff *next = skb->next; 3650 3651 skb_mark_not_on_list(skb); 3652 rc = xmit_one(skb, dev, txq, next != NULL); 3653 if (unlikely(!dev_xmit_complete(rc))) { 3654 skb->next = next; 3655 goto out; 3656 } 3657 3658 skb = next; 3659 if (netif_tx_queue_stopped(txq) && skb) { 3660 rc = NETDEV_TX_BUSY; 3661 break; 3662 } 3663 } 3664 3665 out: 3666 *ret = rc; 3667 return skb; 3668 } 3669 3670 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3671 netdev_features_t features) 3672 { 3673 if (skb_vlan_tag_present(skb) && 3674 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3675 skb = __vlan_hwaccel_push_inside(skb); 3676 return skb; 3677 } 3678 3679 int skb_csum_hwoffload_help(struct sk_buff *skb, 3680 const netdev_features_t features) 3681 { 3682 if (unlikely(skb_csum_is_sctp(skb))) 3683 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3684 skb_crc32c_csum_help(skb); 3685 3686 if (features & NETIF_F_HW_CSUM) 3687 return 0; 3688 3689 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3690 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) && 3691 skb_network_header_len(skb) != sizeof(struct ipv6hdr) && 3692 !ipv6_has_hopopt_jumbo(skb)) 3693 goto sw_checksum; 3694 3695 switch (skb->csum_offset) { 3696 case offsetof(struct tcphdr, check): 3697 case offsetof(struct udphdr, check): 3698 return 0; 3699 } 3700 } 3701 3702 sw_checksum: 3703 return skb_checksum_help(skb); 3704 } 3705 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3706 3707 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3708 { 3709 netdev_features_t features; 3710 3711 features = netif_skb_features(skb); 3712 skb = validate_xmit_vlan(skb, features); 3713 if (unlikely(!skb)) 3714 goto out_null; 3715 3716 skb = sk_validate_xmit_skb(skb, dev); 3717 if (unlikely(!skb)) 3718 goto out_null; 3719 3720 if (netif_needs_gso(skb, features)) { 3721 struct sk_buff *segs; 3722 3723 segs = skb_gso_segment(skb, features); 3724 if (IS_ERR(segs)) { 3725 goto out_kfree_skb; 3726 } else if (segs) { 3727 consume_skb(skb); 3728 skb = segs; 3729 } 3730 } else { 3731 if (skb_needs_linearize(skb, features) && 3732 __skb_linearize(skb)) 3733 goto out_kfree_skb; 3734 3735 /* If packet is not checksummed and device does not 3736 * support checksumming for this protocol, complete 3737 * checksumming here. 3738 */ 3739 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3740 if (skb->encapsulation) 3741 skb_set_inner_transport_header(skb, 3742 skb_checksum_start_offset(skb)); 3743 else 3744 skb_set_transport_header(skb, 3745 skb_checksum_start_offset(skb)); 3746 if (skb_csum_hwoffload_help(skb, features)) 3747 goto out_kfree_skb; 3748 } 3749 } 3750 3751 skb = validate_xmit_xfrm(skb, features, again); 3752 3753 return skb; 3754 3755 out_kfree_skb: 3756 kfree_skb(skb); 3757 out_null: 3758 dev_core_stats_tx_dropped_inc(dev); 3759 return NULL; 3760 } 3761 3762 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3763 { 3764 struct sk_buff *next, *head = NULL, *tail; 3765 3766 for (; skb != NULL; skb = next) { 3767 next = skb->next; 3768 skb_mark_not_on_list(skb); 3769 3770 /* in case skb won't be segmented, point to itself */ 3771 skb->prev = skb; 3772 3773 skb = validate_xmit_skb(skb, dev, again); 3774 if (!skb) 3775 continue; 3776 3777 if (!head) 3778 head = skb; 3779 else 3780 tail->next = skb; 3781 /* If skb was segmented, skb->prev points to 3782 * the last segment. If not, it still contains skb. 3783 */ 3784 tail = skb->prev; 3785 } 3786 return head; 3787 } 3788 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3789 3790 static void qdisc_pkt_len_init(struct sk_buff *skb) 3791 { 3792 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3793 3794 qdisc_skb_cb(skb)->pkt_len = skb->len; 3795 3796 /* To get more precise estimation of bytes sent on wire, 3797 * we add to pkt_len the headers size of all segments 3798 */ 3799 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3800 u16 gso_segs = shinfo->gso_segs; 3801 unsigned int hdr_len; 3802 3803 /* mac layer + network layer */ 3804 hdr_len = skb_transport_offset(skb); 3805 3806 /* + transport layer */ 3807 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3808 const struct tcphdr *th; 3809 struct tcphdr _tcphdr; 3810 3811 th = skb_header_pointer(skb, hdr_len, 3812 sizeof(_tcphdr), &_tcphdr); 3813 if (likely(th)) 3814 hdr_len += __tcp_hdrlen(th); 3815 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 3816 struct udphdr _udphdr; 3817 3818 if (skb_header_pointer(skb, hdr_len, 3819 sizeof(_udphdr), &_udphdr)) 3820 hdr_len += sizeof(struct udphdr); 3821 } 3822 3823 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) { 3824 int payload = skb->len - hdr_len; 3825 3826 /* Malicious packet. */ 3827 if (payload <= 0) 3828 return; 3829 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size); 3830 } 3831 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3832 } 3833 } 3834 3835 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3836 struct sk_buff **to_free, 3837 struct netdev_queue *txq) 3838 { 3839 int rc; 3840 3841 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3842 if (rc == NET_XMIT_SUCCESS) 3843 trace_qdisc_enqueue(q, txq, skb); 3844 return rc; 3845 } 3846 3847 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3848 struct net_device *dev, 3849 struct netdev_queue *txq) 3850 { 3851 spinlock_t *root_lock = qdisc_lock(q); 3852 struct sk_buff *to_free = NULL; 3853 bool contended; 3854 int rc; 3855 3856 qdisc_calculate_pkt_len(skb, q); 3857 3858 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3859 3860 if (q->flags & TCQ_F_NOLOCK) { 3861 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3862 qdisc_run_begin(q)) { 3863 /* Retest nolock_qdisc_is_empty() within the protection 3864 * of q->seqlock to protect from racing with requeuing. 3865 */ 3866 if (unlikely(!nolock_qdisc_is_empty(q))) { 3867 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3868 __qdisc_run(q); 3869 qdisc_run_end(q); 3870 3871 goto no_lock_out; 3872 } 3873 3874 qdisc_bstats_cpu_update(q, skb); 3875 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3876 !nolock_qdisc_is_empty(q)) 3877 __qdisc_run(q); 3878 3879 qdisc_run_end(q); 3880 return NET_XMIT_SUCCESS; 3881 } 3882 3883 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3884 qdisc_run(q); 3885 3886 no_lock_out: 3887 if (unlikely(to_free)) 3888 kfree_skb_list_reason(to_free, 3889 tcf_get_drop_reason(to_free)); 3890 return rc; 3891 } 3892 3893 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) { 3894 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP); 3895 return NET_XMIT_DROP; 3896 } 3897 /* 3898 * Heuristic to force contended enqueues to serialize on a 3899 * separate lock before trying to get qdisc main lock. 3900 * This permits qdisc->running owner to get the lock more 3901 * often and dequeue packets faster. 3902 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3903 * and then other tasks will only enqueue packets. The packets will be 3904 * sent after the qdisc owner is scheduled again. To prevent this 3905 * scenario the task always serialize on the lock. 3906 */ 3907 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3908 if (unlikely(contended)) 3909 spin_lock(&q->busylock); 3910 3911 spin_lock(root_lock); 3912 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3913 __qdisc_drop(skb, &to_free); 3914 rc = NET_XMIT_DROP; 3915 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3916 qdisc_run_begin(q)) { 3917 /* 3918 * This is a work-conserving queue; there are no old skbs 3919 * waiting to be sent out; and the qdisc is not running - 3920 * xmit the skb directly. 3921 */ 3922 3923 qdisc_bstats_update(q, skb); 3924 3925 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3926 if (unlikely(contended)) { 3927 spin_unlock(&q->busylock); 3928 contended = false; 3929 } 3930 __qdisc_run(q); 3931 } 3932 3933 qdisc_run_end(q); 3934 rc = NET_XMIT_SUCCESS; 3935 } else { 3936 WRITE_ONCE(q->owner, smp_processor_id()); 3937 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3938 WRITE_ONCE(q->owner, -1); 3939 if (qdisc_run_begin(q)) { 3940 if (unlikely(contended)) { 3941 spin_unlock(&q->busylock); 3942 contended = false; 3943 } 3944 __qdisc_run(q); 3945 qdisc_run_end(q); 3946 } 3947 } 3948 spin_unlock(root_lock); 3949 if (unlikely(to_free)) 3950 kfree_skb_list_reason(to_free, 3951 tcf_get_drop_reason(to_free)); 3952 if (unlikely(contended)) 3953 spin_unlock(&q->busylock); 3954 return rc; 3955 } 3956 3957 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3958 static void skb_update_prio(struct sk_buff *skb) 3959 { 3960 const struct netprio_map *map; 3961 const struct sock *sk; 3962 unsigned int prioidx; 3963 3964 if (skb->priority) 3965 return; 3966 map = rcu_dereference_bh(skb->dev->priomap); 3967 if (!map) 3968 return; 3969 sk = skb_to_full_sk(skb); 3970 if (!sk) 3971 return; 3972 3973 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3974 3975 if (prioidx < map->priomap_len) 3976 skb->priority = map->priomap[prioidx]; 3977 } 3978 #else 3979 #define skb_update_prio(skb) 3980 #endif 3981 3982 /** 3983 * dev_loopback_xmit - loop back @skb 3984 * @net: network namespace this loopback is happening in 3985 * @sk: sk needed to be a netfilter okfn 3986 * @skb: buffer to transmit 3987 */ 3988 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3989 { 3990 skb_reset_mac_header(skb); 3991 __skb_pull(skb, skb_network_offset(skb)); 3992 skb->pkt_type = PACKET_LOOPBACK; 3993 if (skb->ip_summed == CHECKSUM_NONE) 3994 skb->ip_summed = CHECKSUM_UNNECESSARY; 3995 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3996 skb_dst_force(skb); 3997 netif_rx(skb); 3998 return 0; 3999 } 4000 EXPORT_SYMBOL(dev_loopback_xmit); 4001 4002 #ifdef CONFIG_NET_EGRESS 4003 static struct netdev_queue * 4004 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 4005 { 4006 int qm = skb_get_queue_mapping(skb); 4007 4008 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 4009 } 4010 4011 #ifndef CONFIG_PREEMPT_RT 4012 static bool netdev_xmit_txqueue_skipped(void) 4013 { 4014 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 4015 } 4016 4017 void netdev_xmit_skip_txqueue(bool skip) 4018 { 4019 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 4020 } 4021 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4022 4023 #else 4024 static bool netdev_xmit_txqueue_skipped(void) 4025 { 4026 return current->net_xmit.skip_txqueue; 4027 } 4028 4029 void netdev_xmit_skip_txqueue(bool skip) 4030 { 4031 current->net_xmit.skip_txqueue = skip; 4032 } 4033 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4034 #endif 4035 #endif /* CONFIG_NET_EGRESS */ 4036 4037 #ifdef CONFIG_NET_XGRESS 4038 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 4039 enum skb_drop_reason *drop_reason) 4040 { 4041 int ret = TC_ACT_UNSPEC; 4042 #ifdef CONFIG_NET_CLS_ACT 4043 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 4044 struct tcf_result res; 4045 4046 if (!miniq) 4047 return ret; 4048 4049 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) { 4050 if (tcf_block_bypass_sw(miniq->block)) 4051 return ret; 4052 } 4053 4054 tc_skb_cb(skb)->mru = 0; 4055 tc_skb_cb(skb)->post_ct = false; 4056 tcf_set_drop_reason(skb, *drop_reason); 4057 4058 mini_qdisc_bstats_cpu_update(miniq, skb); 4059 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 4060 /* Only tcf related quirks below. */ 4061 switch (ret) { 4062 case TC_ACT_SHOT: 4063 *drop_reason = tcf_get_drop_reason(skb); 4064 mini_qdisc_qstats_cpu_drop(miniq); 4065 break; 4066 case TC_ACT_OK: 4067 case TC_ACT_RECLASSIFY: 4068 skb->tc_index = TC_H_MIN(res.classid); 4069 break; 4070 } 4071 #endif /* CONFIG_NET_CLS_ACT */ 4072 return ret; 4073 } 4074 4075 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 4076 4077 void tcx_inc(void) 4078 { 4079 static_branch_inc(&tcx_needed_key); 4080 } 4081 4082 void tcx_dec(void) 4083 { 4084 static_branch_dec(&tcx_needed_key); 4085 } 4086 4087 static __always_inline enum tcx_action_base 4088 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 4089 const bool needs_mac) 4090 { 4091 const struct bpf_mprog_fp *fp; 4092 const struct bpf_prog *prog; 4093 int ret = TCX_NEXT; 4094 4095 if (needs_mac) 4096 __skb_push(skb, skb->mac_len); 4097 bpf_mprog_foreach_prog(entry, fp, prog) { 4098 bpf_compute_data_pointers(skb); 4099 ret = bpf_prog_run(prog, skb); 4100 if (ret != TCX_NEXT) 4101 break; 4102 } 4103 if (needs_mac) 4104 __skb_pull(skb, skb->mac_len); 4105 return tcx_action_code(skb, ret); 4106 } 4107 4108 static __always_inline struct sk_buff * 4109 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4110 struct net_device *orig_dev, bool *another) 4111 { 4112 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4113 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 4114 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4115 int sch_ret; 4116 4117 if (!entry) 4118 return skb; 4119 4120 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4121 if (*pt_prev) { 4122 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4123 *pt_prev = NULL; 4124 } 4125 4126 qdisc_skb_cb(skb)->pkt_len = skb->len; 4127 tcx_set_ingress(skb, true); 4128 4129 if (static_branch_unlikely(&tcx_needed_key)) { 4130 sch_ret = tcx_run(entry, skb, true); 4131 if (sch_ret != TC_ACT_UNSPEC) 4132 goto ingress_verdict; 4133 } 4134 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4135 ingress_verdict: 4136 switch (sch_ret) { 4137 case TC_ACT_REDIRECT: 4138 /* skb_mac_header check was done by BPF, so we can safely 4139 * push the L2 header back before redirecting to another 4140 * netdev. 4141 */ 4142 __skb_push(skb, skb->mac_len); 4143 if (skb_do_redirect(skb) == -EAGAIN) { 4144 __skb_pull(skb, skb->mac_len); 4145 *another = true; 4146 break; 4147 } 4148 *ret = NET_RX_SUCCESS; 4149 bpf_net_ctx_clear(bpf_net_ctx); 4150 return NULL; 4151 case TC_ACT_SHOT: 4152 kfree_skb_reason(skb, drop_reason); 4153 *ret = NET_RX_DROP; 4154 bpf_net_ctx_clear(bpf_net_ctx); 4155 return NULL; 4156 /* used by tc_run */ 4157 case TC_ACT_STOLEN: 4158 case TC_ACT_QUEUED: 4159 case TC_ACT_TRAP: 4160 consume_skb(skb); 4161 fallthrough; 4162 case TC_ACT_CONSUMED: 4163 *ret = NET_RX_SUCCESS; 4164 bpf_net_ctx_clear(bpf_net_ctx); 4165 return NULL; 4166 } 4167 bpf_net_ctx_clear(bpf_net_ctx); 4168 4169 return skb; 4170 } 4171 4172 static __always_inline struct sk_buff * 4173 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4174 { 4175 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4176 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4177 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 4178 int sch_ret; 4179 4180 if (!entry) 4181 return skb; 4182 4183 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 4184 4185 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4186 * already set by the caller. 4187 */ 4188 if (static_branch_unlikely(&tcx_needed_key)) { 4189 sch_ret = tcx_run(entry, skb, false); 4190 if (sch_ret != TC_ACT_UNSPEC) 4191 goto egress_verdict; 4192 } 4193 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4194 egress_verdict: 4195 switch (sch_ret) { 4196 case TC_ACT_REDIRECT: 4197 /* No need to push/pop skb's mac_header here on egress! */ 4198 skb_do_redirect(skb); 4199 *ret = NET_XMIT_SUCCESS; 4200 bpf_net_ctx_clear(bpf_net_ctx); 4201 return NULL; 4202 case TC_ACT_SHOT: 4203 kfree_skb_reason(skb, drop_reason); 4204 *ret = NET_XMIT_DROP; 4205 bpf_net_ctx_clear(bpf_net_ctx); 4206 return NULL; 4207 /* used by tc_run */ 4208 case TC_ACT_STOLEN: 4209 case TC_ACT_QUEUED: 4210 case TC_ACT_TRAP: 4211 consume_skb(skb); 4212 fallthrough; 4213 case TC_ACT_CONSUMED: 4214 *ret = NET_XMIT_SUCCESS; 4215 bpf_net_ctx_clear(bpf_net_ctx); 4216 return NULL; 4217 } 4218 bpf_net_ctx_clear(bpf_net_ctx); 4219 4220 return skb; 4221 } 4222 #else 4223 static __always_inline struct sk_buff * 4224 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4225 struct net_device *orig_dev, bool *another) 4226 { 4227 return skb; 4228 } 4229 4230 static __always_inline struct sk_buff * 4231 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4232 { 4233 return skb; 4234 } 4235 #endif /* CONFIG_NET_XGRESS */ 4236 4237 #ifdef CONFIG_XPS 4238 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4239 struct xps_dev_maps *dev_maps, unsigned int tci) 4240 { 4241 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4242 struct xps_map *map; 4243 int queue_index = -1; 4244 4245 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4246 return queue_index; 4247 4248 tci *= dev_maps->num_tc; 4249 tci += tc; 4250 4251 map = rcu_dereference(dev_maps->attr_map[tci]); 4252 if (map) { 4253 if (map->len == 1) 4254 queue_index = map->queues[0]; 4255 else 4256 queue_index = map->queues[reciprocal_scale( 4257 skb_get_hash(skb), map->len)]; 4258 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4259 queue_index = -1; 4260 } 4261 return queue_index; 4262 } 4263 #endif 4264 4265 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4266 struct sk_buff *skb) 4267 { 4268 #ifdef CONFIG_XPS 4269 struct xps_dev_maps *dev_maps; 4270 struct sock *sk = skb->sk; 4271 int queue_index = -1; 4272 4273 if (!static_key_false(&xps_needed)) 4274 return -1; 4275 4276 rcu_read_lock(); 4277 if (!static_key_false(&xps_rxqs_needed)) 4278 goto get_cpus_map; 4279 4280 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4281 if (dev_maps) { 4282 int tci = sk_rx_queue_get(sk); 4283 4284 if (tci >= 0) 4285 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4286 tci); 4287 } 4288 4289 get_cpus_map: 4290 if (queue_index < 0) { 4291 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4292 if (dev_maps) { 4293 unsigned int tci = skb->sender_cpu - 1; 4294 4295 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4296 tci); 4297 } 4298 } 4299 rcu_read_unlock(); 4300 4301 return queue_index; 4302 #else 4303 return -1; 4304 #endif 4305 } 4306 4307 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4308 struct net_device *sb_dev) 4309 { 4310 return 0; 4311 } 4312 EXPORT_SYMBOL(dev_pick_tx_zero); 4313 4314 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4315 struct net_device *sb_dev) 4316 { 4317 struct sock *sk = skb->sk; 4318 int queue_index = sk_tx_queue_get(sk); 4319 4320 sb_dev = sb_dev ? : dev; 4321 4322 if (queue_index < 0 || skb->ooo_okay || 4323 queue_index >= dev->real_num_tx_queues) { 4324 int new_index = get_xps_queue(dev, sb_dev, skb); 4325 4326 if (new_index < 0) 4327 new_index = skb_tx_hash(dev, sb_dev, skb); 4328 4329 if (queue_index != new_index && sk && 4330 sk_fullsock(sk) && 4331 rcu_access_pointer(sk->sk_dst_cache)) 4332 sk_tx_queue_set(sk, new_index); 4333 4334 queue_index = new_index; 4335 } 4336 4337 return queue_index; 4338 } 4339 EXPORT_SYMBOL(netdev_pick_tx); 4340 4341 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4342 struct sk_buff *skb, 4343 struct net_device *sb_dev) 4344 { 4345 int queue_index = 0; 4346 4347 #ifdef CONFIG_XPS 4348 u32 sender_cpu = skb->sender_cpu - 1; 4349 4350 if (sender_cpu >= (u32)NR_CPUS) 4351 skb->sender_cpu = raw_smp_processor_id() + 1; 4352 #endif 4353 4354 if (dev->real_num_tx_queues != 1) { 4355 const struct net_device_ops *ops = dev->netdev_ops; 4356 4357 if (ops->ndo_select_queue) 4358 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4359 else 4360 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4361 4362 queue_index = netdev_cap_txqueue(dev, queue_index); 4363 } 4364 4365 skb_set_queue_mapping(skb, queue_index); 4366 return netdev_get_tx_queue(dev, queue_index); 4367 } 4368 4369 /** 4370 * __dev_queue_xmit() - transmit a buffer 4371 * @skb: buffer to transmit 4372 * @sb_dev: suboordinate device used for L2 forwarding offload 4373 * 4374 * Queue a buffer for transmission to a network device. The caller must 4375 * have set the device and priority and built the buffer before calling 4376 * this function. The function can be called from an interrupt. 4377 * 4378 * When calling this method, interrupts MUST be enabled. This is because 4379 * the BH enable code must have IRQs enabled so that it will not deadlock. 4380 * 4381 * Regardless of the return value, the skb is consumed, so it is currently 4382 * difficult to retry a send to this method. (You can bump the ref count 4383 * before sending to hold a reference for retry if you are careful.) 4384 * 4385 * Return: 4386 * * 0 - buffer successfully transmitted 4387 * * positive qdisc return code - NET_XMIT_DROP etc. 4388 * * negative errno - other errors 4389 */ 4390 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4391 { 4392 struct net_device *dev = skb->dev; 4393 struct netdev_queue *txq = NULL; 4394 struct Qdisc *q; 4395 int rc = -ENOMEM; 4396 bool again = false; 4397 4398 skb_reset_mac_header(skb); 4399 skb_assert_len(skb); 4400 4401 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4402 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4403 4404 /* Disable soft irqs for various locks below. Also 4405 * stops preemption for RCU. 4406 */ 4407 rcu_read_lock_bh(); 4408 4409 skb_update_prio(skb); 4410 4411 qdisc_pkt_len_init(skb); 4412 tcx_set_ingress(skb, false); 4413 #ifdef CONFIG_NET_EGRESS 4414 if (static_branch_unlikely(&egress_needed_key)) { 4415 if (nf_hook_egress_active()) { 4416 skb = nf_hook_egress(skb, &rc, dev); 4417 if (!skb) 4418 goto out; 4419 } 4420 4421 netdev_xmit_skip_txqueue(false); 4422 4423 nf_skip_egress(skb, true); 4424 skb = sch_handle_egress(skb, &rc, dev); 4425 if (!skb) 4426 goto out; 4427 nf_skip_egress(skb, false); 4428 4429 if (netdev_xmit_txqueue_skipped()) 4430 txq = netdev_tx_queue_mapping(dev, skb); 4431 } 4432 #endif 4433 /* If device/qdisc don't need skb->dst, release it right now while 4434 * its hot in this cpu cache. 4435 */ 4436 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4437 skb_dst_drop(skb); 4438 else 4439 skb_dst_force(skb); 4440 4441 if (!txq) 4442 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4443 4444 q = rcu_dereference_bh(txq->qdisc); 4445 4446 trace_net_dev_queue(skb); 4447 if (q->enqueue) { 4448 rc = __dev_xmit_skb(skb, q, dev, txq); 4449 goto out; 4450 } 4451 4452 /* The device has no queue. Common case for software devices: 4453 * loopback, all the sorts of tunnels... 4454 4455 * Really, it is unlikely that netif_tx_lock protection is necessary 4456 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4457 * counters.) 4458 * However, it is possible, that they rely on protection 4459 * made by us here. 4460 4461 * Check this and shot the lock. It is not prone from deadlocks. 4462 *Either shot noqueue qdisc, it is even simpler 8) 4463 */ 4464 if (dev->flags & IFF_UP) { 4465 int cpu = smp_processor_id(); /* ok because BHs are off */ 4466 4467 /* Other cpus might concurrently change txq->xmit_lock_owner 4468 * to -1 or to their cpu id, but not to our id. 4469 */ 4470 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4471 if (dev_xmit_recursion()) 4472 goto recursion_alert; 4473 4474 skb = validate_xmit_skb(skb, dev, &again); 4475 if (!skb) 4476 goto out; 4477 4478 HARD_TX_LOCK(dev, txq, cpu); 4479 4480 if (!netif_xmit_stopped(txq)) { 4481 dev_xmit_recursion_inc(); 4482 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4483 dev_xmit_recursion_dec(); 4484 if (dev_xmit_complete(rc)) { 4485 HARD_TX_UNLOCK(dev, txq); 4486 goto out; 4487 } 4488 } 4489 HARD_TX_UNLOCK(dev, txq); 4490 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4491 dev->name); 4492 } else { 4493 /* Recursion is detected! It is possible, 4494 * unfortunately 4495 */ 4496 recursion_alert: 4497 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4498 dev->name); 4499 } 4500 } 4501 4502 rc = -ENETDOWN; 4503 rcu_read_unlock_bh(); 4504 4505 dev_core_stats_tx_dropped_inc(dev); 4506 kfree_skb_list(skb); 4507 return rc; 4508 out: 4509 rcu_read_unlock_bh(); 4510 return rc; 4511 } 4512 EXPORT_SYMBOL(__dev_queue_xmit); 4513 4514 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4515 { 4516 struct net_device *dev = skb->dev; 4517 struct sk_buff *orig_skb = skb; 4518 struct netdev_queue *txq; 4519 int ret = NETDEV_TX_BUSY; 4520 bool again = false; 4521 4522 if (unlikely(!netif_running(dev) || 4523 !netif_carrier_ok(dev))) 4524 goto drop; 4525 4526 skb = validate_xmit_skb_list(skb, dev, &again); 4527 if (skb != orig_skb) 4528 goto drop; 4529 4530 skb_set_queue_mapping(skb, queue_id); 4531 txq = skb_get_tx_queue(dev, skb); 4532 4533 local_bh_disable(); 4534 4535 dev_xmit_recursion_inc(); 4536 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4537 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4538 ret = netdev_start_xmit(skb, dev, txq, false); 4539 HARD_TX_UNLOCK(dev, txq); 4540 dev_xmit_recursion_dec(); 4541 4542 local_bh_enable(); 4543 return ret; 4544 drop: 4545 dev_core_stats_tx_dropped_inc(dev); 4546 kfree_skb_list(skb); 4547 return NET_XMIT_DROP; 4548 } 4549 EXPORT_SYMBOL(__dev_direct_xmit); 4550 4551 /************************************************************************* 4552 * Receiver routines 4553 *************************************************************************/ 4554 static DEFINE_PER_CPU(struct task_struct *, backlog_napi); 4555 4556 int weight_p __read_mostly = 64; /* old backlog weight */ 4557 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4558 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4559 4560 /* Called with irq disabled */ 4561 static inline void ____napi_schedule(struct softnet_data *sd, 4562 struct napi_struct *napi) 4563 { 4564 struct task_struct *thread; 4565 4566 lockdep_assert_irqs_disabled(); 4567 4568 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4569 /* Paired with smp_mb__before_atomic() in 4570 * napi_enable()/dev_set_threaded(). 4571 * Use READ_ONCE() to guarantee a complete 4572 * read on napi->thread. Only call 4573 * wake_up_process() when it's not NULL. 4574 */ 4575 thread = READ_ONCE(napi->thread); 4576 if (thread) { 4577 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi)) 4578 goto use_local_napi; 4579 4580 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4581 wake_up_process(thread); 4582 return; 4583 } 4584 } 4585 4586 use_local_napi: 4587 list_add_tail(&napi->poll_list, &sd->poll_list); 4588 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4589 /* If not called from net_rx_action() 4590 * we have to raise NET_RX_SOFTIRQ. 4591 */ 4592 if (!sd->in_net_rx_action) 4593 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4594 } 4595 4596 #ifdef CONFIG_RPS 4597 4598 struct static_key_false rps_needed __read_mostly; 4599 EXPORT_SYMBOL(rps_needed); 4600 struct static_key_false rfs_needed __read_mostly; 4601 EXPORT_SYMBOL(rfs_needed); 4602 4603 static struct rps_dev_flow * 4604 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4605 struct rps_dev_flow *rflow, u16 next_cpu) 4606 { 4607 if (next_cpu < nr_cpu_ids) { 4608 u32 head; 4609 #ifdef CONFIG_RFS_ACCEL 4610 struct netdev_rx_queue *rxqueue; 4611 struct rps_dev_flow_table *flow_table; 4612 struct rps_dev_flow *old_rflow; 4613 u16 rxq_index; 4614 u32 flow_id; 4615 int rc; 4616 4617 /* Should we steer this flow to a different hardware queue? */ 4618 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4619 !(dev->features & NETIF_F_NTUPLE)) 4620 goto out; 4621 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4622 if (rxq_index == skb_get_rx_queue(skb)) 4623 goto out; 4624 4625 rxqueue = dev->_rx + rxq_index; 4626 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4627 if (!flow_table) 4628 goto out; 4629 flow_id = skb_get_hash(skb) & flow_table->mask; 4630 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4631 rxq_index, flow_id); 4632 if (rc < 0) 4633 goto out; 4634 old_rflow = rflow; 4635 rflow = &flow_table->flows[flow_id]; 4636 WRITE_ONCE(rflow->filter, rc); 4637 if (old_rflow->filter == rc) 4638 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER); 4639 out: 4640 #endif 4641 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head); 4642 rps_input_queue_tail_save(&rflow->last_qtail, head); 4643 } 4644 4645 WRITE_ONCE(rflow->cpu, next_cpu); 4646 return rflow; 4647 } 4648 4649 /* 4650 * get_rps_cpu is called from netif_receive_skb and returns the target 4651 * CPU from the RPS map of the receiving queue for a given skb. 4652 * rcu_read_lock must be held on entry. 4653 */ 4654 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4655 struct rps_dev_flow **rflowp) 4656 { 4657 const struct rps_sock_flow_table *sock_flow_table; 4658 struct netdev_rx_queue *rxqueue = dev->_rx; 4659 struct rps_dev_flow_table *flow_table; 4660 struct rps_map *map; 4661 int cpu = -1; 4662 u32 tcpu; 4663 u32 hash; 4664 4665 if (skb_rx_queue_recorded(skb)) { 4666 u16 index = skb_get_rx_queue(skb); 4667 4668 if (unlikely(index >= dev->real_num_rx_queues)) { 4669 WARN_ONCE(dev->real_num_rx_queues > 1, 4670 "%s received packet on queue %u, but number " 4671 "of RX queues is %u\n", 4672 dev->name, index, dev->real_num_rx_queues); 4673 goto done; 4674 } 4675 rxqueue += index; 4676 } 4677 4678 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4679 4680 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4681 map = rcu_dereference(rxqueue->rps_map); 4682 if (!flow_table && !map) 4683 goto done; 4684 4685 skb_reset_network_header(skb); 4686 hash = skb_get_hash(skb); 4687 if (!hash) 4688 goto done; 4689 4690 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table); 4691 if (flow_table && sock_flow_table) { 4692 struct rps_dev_flow *rflow; 4693 u32 next_cpu; 4694 u32 ident; 4695 4696 /* First check into global flow table if there is a match. 4697 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4698 */ 4699 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4700 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask) 4701 goto try_rps; 4702 4703 next_cpu = ident & net_hotdata.rps_cpu_mask; 4704 4705 /* OK, now we know there is a match, 4706 * we can look at the local (per receive queue) flow table 4707 */ 4708 rflow = &flow_table->flows[hash & flow_table->mask]; 4709 tcpu = rflow->cpu; 4710 4711 /* 4712 * If the desired CPU (where last recvmsg was done) is 4713 * different from current CPU (one in the rx-queue flow 4714 * table entry), switch if one of the following holds: 4715 * - Current CPU is unset (>= nr_cpu_ids). 4716 * - Current CPU is offline. 4717 * - The current CPU's queue tail has advanced beyond the 4718 * last packet that was enqueued using this table entry. 4719 * This guarantees that all previous packets for the flow 4720 * have been dequeued, thus preserving in order delivery. 4721 */ 4722 if (unlikely(tcpu != next_cpu) && 4723 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4724 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) - 4725 rflow->last_qtail)) >= 0)) { 4726 tcpu = next_cpu; 4727 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4728 } 4729 4730 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4731 *rflowp = rflow; 4732 cpu = tcpu; 4733 goto done; 4734 } 4735 } 4736 4737 try_rps: 4738 4739 if (map) { 4740 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4741 if (cpu_online(tcpu)) { 4742 cpu = tcpu; 4743 goto done; 4744 } 4745 } 4746 4747 done: 4748 return cpu; 4749 } 4750 4751 #ifdef CONFIG_RFS_ACCEL 4752 4753 /** 4754 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4755 * @dev: Device on which the filter was set 4756 * @rxq_index: RX queue index 4757 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4758 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4759 * 4760 * Drivers that implement ndo_rx_flow_steer() should periodically call 4761 * this function for each installed filter and remove the filters for 4762 * which it returns %true. 4763 */ 4764 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4765 u32 flow_id, u16 filter_id) 4766 { 4767 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4768 struct rps_dev_flow_table *flow_table; 4769 struct rps_dev_flow *rflow; 4770 bool expire = true; 4771 unsigned int cpu; 4772 4773 rcu_read_lock(); 4774 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4775 if (flow_table && flow_id <= flow_table->mask) { 4776 rflow = &flow_table->flows[flow_id]; 4777 cpu = READ_ONCE(rflow->cpu); 4778 if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids && 4779 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) - 4780 READ_ONCE(rflow->last_qtail)) < 4781 (int)(10 * flow_table->mask))) 4782 expire = false; 4783 } 4784 rcu_read_unlock(); 4785 return expire; 4786 } 4787 EXPORT_SYMBOL(rps_may_expire_flow); 4788 4789 #endif /* CONFIG_RFS_ACCEL */ 4790 4791 /* Called from hardirq (IPI) context */ 4792 static void rps_trigger_softirq(void *data) 4793 { 4794 struct softnet_data *sd = data; 4795 4796 ____napi_schedule(sd, &sd->backlog); 4797 sd->received_rps++; 4798 } 4799 4800 #endif /* CONFIG_RPS */ 4801 4802 /* Called from hardirq (IPI) context */ 4803 static void trigger_rx_softirq(void *data) 4804 { 4805 struct softnet_data *sd = data; 4806 4807 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4808 smp_store_release(&sd->defer_ipi_scheduled, 0); 4809 } 4810 4811 /* 4812 * After we queued a packet into sd->input_pkt_queue, 4813 * we need to make sure this queue is serviced soon. 4814 * 4815 * - If this is another cpu queue, link it to our rps_ipi_list, 4816 * and make sure we will process rps_ipi_list from net_rx_action(). 4817 * 4818 * - If this is our own queue, NAPI schedule our backlog. 4819 * Note that this also raises NET_RX_SOFTIRQ. 4820 */ 4821 static void napi_schedule_rps(struct softnet_data *sd) 4822 { 4823 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4824 4825 #ifdef CONFIG_RPS 4826 if (sd != mysd) { 4827 if (use_backlog_threads()) { 4828 __napi_schedule_irqoff(&sd->backlog); 4829 return; 4830 } 4831 4832 sd->rps_ipi_next = mysd->rps_ipi_list; 4833 mysd->rps_ipi_list = sd; 4834 4835 /* If not called from net_rx_action() or napi_threaded_poll() 4836 * we have to raise NET_RX_SOFTIRQ. 4837 */ 4838 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4839 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4840 return; 4841 } 4842 #endif /* CONFIG_RPS */ 4843 __napi_schedule_irqoff(&mysd->backlog); 4844 } 4845 4846 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu) 4847 { 4848 unsigned long flags; 4849 4850 if (use_backlog_threads()) { 4851 backlog_lock_irq_save(sd, &flags); 4852 4853 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4854 __napi_schedule_irqoff(&sd->backlog); 4855 4856 backlog_unlock_irq_restore(sd, &flags); 4857 4858 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) { 4859 smp_call_function_single_async(cpu, &sd->defer_csd); 4860 } 4861 } 4862 4863 #ifdef CONFIG_NET_FLOW_LIMIT 4864 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4865 #endif 4866 4867 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4868 { 4869 #ifdef CONFIG_NET_FLOW_LIMIT 4870 struct sd_flow_limit *fl; 4871 struct softnet_data *sd; 4872 unsigned int old_flow, new_flow; 4873 4874 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1)) 4875 return false; 4876 4877 sd = this_cpu_ptr(&softnet_data); 4878 4879 rcu_read_lock(); 4880 fl = rcu_dereference(sd->flow_limit); 4881 if (fl) { 4882 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4883 old_flow = fl->history[fl->history_head]; 4884 fl->history[fl->history_head] = new_flow; 4885 4886 fl->history_head++; 4887 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4888 4889 if (likely(fl->buckets[old_flow])) 4890 fl->buckets[old_flow]--; 4891 4892 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4893 fl->count++; 4894 rcu_read_unlock(); 4895 return true; 4896 } 4897 } 4898 rcu_read_unlock(); 4899 #endif 4900 return false; 4901 } 4902 4903 /* 4904 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4905 * queue (may be a remote CPU queue). 4906 */ 4907 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4908 unsigned int *qtail) 4909 { 4910 enum skb_drop_reason reason; 4911 struct softnet_data *sd; 4912 unsigned long flags; 4913 unsigned int qlen; 4914 int max_backlog; 4915 u32 tail; 4916 4917 reason = SKB_DROP_REASON_DEV_READY; 4918 if (!netif_running(skb->dev)) 4919 goto bad_dev; 4920 4921 reason = SKB_DROP_REASON_CPU_BACKLOG; 4922 sd = &per_cpu(softnet_data, cpu); 4923 4924 qlen = skb_queue_len_lockless(&sd->input_pkt_queue); 4925 max_backlog = READ_ONCE(net_hotdata.max_backlog); 4926 if (unlikely(qlen > max_backlog)) 4927 goto cpu_backlog_drop; 4928 backlog_lock_irq_save(sd, &flags); 4929 qlen = skb_queue_len(&sd->input_pkt_queue); 4930 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) { 4931 if (!qlen) { 4932 /* Schedule NAPI for backlog device. We can use 4933 * non atomic operation as we own the queue lock. 4934 */ 4935 if (!__test_and_set_bit(NAPI_STATE_SCHED, 4936 &sd->backlog.state)) 4937 napi_schedule_rps(sd); 4938 } 4939 __skb_queue_tail(&sd->input_pkt_queue, skb); 4940 tail = rps_input_queue_tail_incr(sd); 4941 backlog_unlock_irq_restore(sd, &flags); 4942 4943 /* save the tail outside of the critical section */ 4944 rps_input_queue_tail_save(qtail, tail); 4945 return NET_RX_SUCCESS; 4946 } 4947 4948 backlog_unlock_irq_restore(sd, &flags); 4949 4950 cpu_backlog_drop: 4951 atomic_inc(&sd->dropped); 4952 bad_dev: 4953 dev_core_stats_rx_dropped_inc(skb->dev); 4954 kfree_skb_reason(skb, reason); 4955 return NET_RX_DROP; 4956 } 4957 4958 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4959 { 4960 struct net_device *dev = skb->dev; 4961 struct netdev_rx_queue *rxqueue; 4962 4963 rxqueue = dev->_rx; 4964 4965 if (skb_rx_queue_recorded(skb)) { 4966 u16 index = skb_get_rx_queue(skb); 4967 4968 if (unlikely(index >= dev->real_num_rx_queues)) { 4969 WARN_ONCE(dev->real_num_rx_queues > 1, 4970 "%s received packet on queue %u, but number " 4971 "of RX queues is %u\n", 4972 dev->name, index, dev->real_num_rx_queues); 4973 4974 return rxqueue; /* Return first rxqueue */ 4975 } 4976 rxqueue += index; 4977 } 4978 return rxqueue; 4979 } 4980 4981 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4982 const struct bpf_prog *xdp_prog) 4983 { 4984 void *orig_data, *orig_data_end, *hard_start; 4985 struct netdev_rx_queue *rxqueue; 4986 bool orig_bcast, orig_host; 4987 u32 mac_len, frame_sz; 4988 __be16 orig_eth_type; 4989 struct ethhdr *eth; 4990 u32 metalen, act; 4991 int off; 4992 4993 /* The XDP program wants to see the packet starting at the MAC 4994 * header. 4995 */ 4996 mac_len = skb->data - skb_mac_header(skb); 4997 hard_start = skb->data - skb_headroom(skb); 4998 4999 /* SKB "head" area always have tailroom for skb_shared_info */ 5000 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 5001 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5002 5003 rxqueue = netif_get_rxqueue(skb); 5004 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 5005 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 5006 skb_headlen(skb) + mac_len, true); 5007 if (skb_is_nonlinear(skb)) { 5008 skb_shinfo(skb)->xdp_frags_size = skb->data_len; 5009 xdp_buff_set_frags_flag(xdp); 5010 } else { 5011 xdp_buff_clear_frags_flag(xdp); 5012 } 5013 5014 orig_data_end = xdp->data_end; 5015 orig_data = xdp->data; 5016 eth = (struct ethhdr *)xdp->data; 5017 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 5018 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 5019 orig_eth_type = eth->h_proto; 5020 5021 act = bpf_prog_run_xdp(xdp_prog, xdp); 5022 5023 /* check if bpf_xdp_adjust_head was used */ 5024 off = xdp->data - orig_data; 5025 if (off) { 5026 if (off > 0) 5027 __skb_pull(skb, off); 5028 else if (off < 0) 5029 __skb_push(skb, -off); 5030 5031 skb->mac_header += off; 5032 skb_reset_network_header(skb); 5033 } 5034 5035 /* check if bpf_xdp_adjust_tail was used */ 5036 off = xdp->data_end - orig_data_end; 5037 if (off != 0) { 5038 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 5039 skb->len += off; /* positive on grow, negative on shrink */ 5040 } 5041 5042 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers 5043 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. 5044 */ 5045 if (xdp_buff_has_frags(xdp)) 5046 skb->data_len = skb_shinfo(skb)->xdp_frags_size; 5047 else 5048 skb->data_len = 0; 5049 5050 /* check if XDP changed eth hdr such SKB needs update */ 5051 eth = (struct ethhdr *)xdp->data; 5052 if ((orig_eth_type != eth->h_proto) || 5053 (orig_host != ether_addr_equal_64bits(eth->h_dest, 5054 skb->dev->dev_addr)) || 5055 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 5056 __skb_push(skb, ETH_HLEN); 5057 skb->pkt_type = PACKET_HOST; 5058 skb->protocol = eth_type_trans(skb, skb->dev); 5059 } 5060 5061 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 5062 * before calling us again on redirect path. We do not call do_redirect 5063 * as we leave that up to the caller. 5064 * 5065 * Caller is responsible for managing lifetime of skb (i.e. calling 5066 * kfree_skb in response to actions it cannot handle/XDP_DROP). 5067 */ 5068 switch (act) { 5069 case XDP_REDIRECT: 5070 case XDP_TX: 5071 __skb_push(skb, mac_len); 5072 break; 5073 case XDP_PASS: 5074 metalen = xdp->data - xdp->data_meta; 5075 if (metalen) 5076 skb_metadata_set(skb, metalen); 5077 break; 5078 } 5079 5080 return act; 5081 } 5082 5083 static int 5084 netif_skb_check_for_xdp(struct sk_buff **pskb, const struct bpf_prog *prog) 5085 { 5086 struct sk_buff *skb = *pskb; 5087 int err, hroom, troom; 5088 5089 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog)) 5090 return 0; 5091 5092 /* In case we have to go down the path and also linearize, 5093 * then lets do the pskb_expand_head() work just once here. 5094 */ 5095 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 5096 troom = skb->tail + skb->data_len - skb->end; 5097 err = pskb_expand_head(skb, 5098 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 5099 troom > 0 ? troom + 128 : 0, GFP_ATOMIC); 5100 if (err) 5101 return err; 5102 5103 return skb_linearize(skb); 5104 } 5105 5106 static u32 netif_receive_generic_xdp(struct sk_buff **pskb, 5107 struct xdp_buff *xdp, 5108 const struct bpf_prog *xdp_prog) 5109 { 5110 struct sk_buff *skb = *pskb; 5111 u32 mac_len, act = XDP_DROP; 5112 5113 /* Reinjected packets coming from act_mirred or similar should 5114 * not get XDP generic processing. 5115 */ 5116 if (skb_is_redirected(skb)) 5117 return XDP_PASS; 5118 5119 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM 5120 * bytes. This is the guarantee that also native XDP provides, 5121 * thus we need to do it here as well. 5122 */ 5123 mac_len = skb->data - skb_mac_header(skb); 5124 __skb_push(skb, mac_len); 5125 5126 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 5127 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 5128 if (netif_skb_check_for_xdp(pskb, xdp_prog)) 5129 goto do_drop; 5130 } 5131 5132 __skb_pull(*pskb, mac_len); 5133 5134 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); 5135 switch (act) { 5136 case XDP_REDIRECT: 5137 case XDP_TX: 5138 case XDP_PASS: 5139 break; 5140 default: 5141 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); 5142 fallthrough; 5143 case XDP_ABORTED: 5144 trace_xdp_exception((*pskb)->dev, xdp_prog, act); 5145 fallthrough; 5146 case XDP_DROP: 5147 do_drop: 5148 kfree_skb(*pskb); 5149 break; 5150 } 5151 5152 return act; 5153 } 5154 5155 /* When doing generic XDP we have to bypass the qdisc layer and the 5156 * network taps in order to match in-driver-XDP behavior. This also means 5157 * that XDP packets are able to starve other packets going through a qdisc, 5158 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5159 * queues, so they do not have this starvation issue. 5160 */ 5161 void generic_xdp_tx(struct sk_buff *skb, const struct bpf_prog *xdp_prog) 5162 { 5163 struct net_device *dev = skb->dev; 5164 struct netdev_queue *txq; 5165 bool free_skb = true; 5166 int cpu, rc; 5167 5168 txq = netdev_core_pick_tx(dev, skb, NULL); 5169 cpu = smp_processor_id(); 5170 HARD_TX_LOCK(dev, txq, cpu); 5171 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5172 rc = netdev_start_xmit(skb, dev, txq, 0); 5173 if (dev_xmit_complete(rc)) 5174 free_skb = false; 5175 } 5176 HARD_TX_UNLOCK(dev, txq); 5177 if (free_skb) { 5178 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5179 dev_core_stats_tx_dropped_inc(dev); 5180 kfree_skb(skb); 5181 } 5182 } 5183 5184 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5185 5186 int do_xdp_generic(const struct bpf_prog *xdp_prog, struct sk_buff **pskb) 5187 { 5188 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 5189 5190 if (xdp_prog) { 5191 struct xdp_buff xdp; 5192 u32 act; 5193 int err; 5194 5195 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 5196 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); 5197 if (act != XDP_PASS) { 5198 switch (act) { 5199 case XDP_REDIRECT: 5200 err = xdp_do_generic_redirect((*pskb)->dev, *pskb, 5201 &xdp, xdp_prog); 5202 if (err) 5203 goto out_redir; 5204 break; 5205 case XDP_TX: 5206 generic_xdp_tx(*pskb, xdp_prog); 5207 break; 5208 } 5209 bpf_net_ctx_clear(bpf_net_ctx); 5210 return XDP_DROP; 5211 } 5212 bpf_net_ctx_clear(bpf_net_ctx); 5213 } 5214 return XDP_PASS; 5215 out_redir: 5216 bpf_net_ctx_clear(bpf_net_ctx); 5217 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); 5218 return XDP_DROP; 5219 } 5220 EXPORT_SYMBOL_GPL(do_xdp_generic); 5221 5222 static int netif_rx_internal(struct sk_buff *skb) 5223 { 5224 int ret; 5225 5226 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5227 5228 trace_netif_rx(skb); 5229 5230 #ifdef CONFIG_RPS 5231 if (static_branch_unlikely(&rps_needed)) { 5232 struct rps_dev_flow voidflow, *rflow = &voidflow; 5233 int cpu; 5234 5235 rcu_read_lock(); 5236 5237 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5238 if (cpu < 0) 5239 cpu = smp_processor_id(); 5240 5241 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5242 5243 rcu_read_unlock(); 5244 } else 5245 #endif 5246 { 5247 unsigned int qtail; 5248 5249 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5250 } 5251 return ret; 5252 } 5253 5254 /** 5255 * __netif_rx - Slightly optimized version of netif_rx 5256 * @skb: buffer to post 5257 * 5258 * This behaves as netif_rx except that it does not disable bottom halves. 5259 * As a result this function may only be invoked from the interrupt context 5260 * (either hard or soft interrupt). 5261 */ 5262 int __netif_rx(struct sk_buff *skb) 5263 { 5264 int ret; 5265 5266 lockdep_assert_once(hardirq_count() | softirq_count()); 5267 5268 trace_netif_rx_entry(skb); 5269 ret = netif_rx_internal(skb); 5270 trace_netif_rx_exit(ret); 5271 return ret; 5272 } 5273 EXPORT_SYMBOL(__netif_rx); 5274 5275 /** 5276 * netif_rx - post buffer to the network code 5277 * @skb: buffer to post 5278 * 5279 * This function receives a packet from a device driver and queues it for 5280 * the upper (protocol) levels to process via the backlog NAPI device. It 5281 * always succeeds. The buffer may be dropped during processing for 5282 * congestion control or by the protocol layers. 5283 * The network buffer is passed via the backlog NAPI device. Modern NIC 5284 * driver should use NAPI and GRO. 5285 * This function can used from interrupt and from process context. The 5286 * caller from process context must not disable interrupts before invoking 5287 * this function. 5288 * 5289 * return values: 5290 * NET_RX_SUCCESS (no congestion) 5291 * NET_RX_DROP (packet was dropped) 5292 * 5293 */ 5294 int netif_rx(struct sk_buff *skb) 5295 { 5296 bool need_bh_off = !(hardirq_count() | softirq_count()); 5297 int ret; 5298 5299 if (need_bh_off) 5300 local_bh_disable(); 5301 trace_netif_rx_entry(skb); 5302 ret = netif_rx_internal(skb); 5303 trace_netif_rx_exit(ret); 5304 if (need_bh_off) 5305 local_bh_enable(); 5306 return ret; 5307 } 5308 EXPORT_SYMBOL(netif_rx); 5309 5310 static __latent_entropy void net_tx_action(void) 5311 { 5312 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5313 5314 if (sd->completion_queue) { 5315 struct sk_buff *clist; 5316 5317 local_irq_disable(); 5318 clist = sd->completion_queue; 5319 sd->completion_queue = NULL; 5320 local_irq_enable(); 5321 5322 while (clist) { 5323 struct sk_buff *skb = clist; 5324 5325 clist = clist->next; 5326 5327 WARN_ON(refcount_read(&skb->users)); 5328 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5329 trace_consume_skb(skb, net_tx_action); 5330 else 5331 trace_kfree_skb(skb, net_tx_action, 5332 get_kfree_skb_cb(skb)->reason, NULL); 5333 5334 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5335 __kfree_skb(skb); 5336 else 5337 __napi_kfree_skb(skb, 5338 get_kfree_skb_cb(skb)->reason); 5339 } 5340 } 5341 5342 if (sd->output_queue) { 5343 struct Qdisc *head; 5344 5345 local_irq_disable(); 5346 head = sd->output_queue; 5347 sd->output_queue = NULL; 5348 sd->output_queue_tailp = &sd->output_queue; 5349 local_irq_enable(); 5350 5351 rcu_read_lock(); 5352 5353 while (head) { 5354 struct Qdisc *q = head; 5355 spinlock_t *root_lock = NULL; 5356 5357 head = head->next_sched; 5358 5359 /* We need to make sure head->next_sched is read 5360 * before clearing __QDISC_STATE_SCHED 5361 */ 5362 smp_mb__before_atomic(); 5363 5364 if (!(q->flags & TCQ_F_NOLOCK)) { 5365 root_lock = qdisc_lock(q); 5366 spin_lock(root_lock); 5367 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5368 &q->state))) { 5369 /* There is a synchronize_net() between 5370 * STATE_DEACTIVATED flag being set and 5371 * qdisc_reset()/some_qdisc_is_busy() in 5372 * dev_deactivate(), so we can safely bail out 5373 * early here to avoid data race between 5374 * qdisc_deactivate() and some_qdisc_is_busy() 5375 * for lockless qdisc. 5376 */ 5377 clear_bit(__QDISC_STATE_SCHED, &q->state); 5378 continue; 5379 } 5380 5381 clear_bit(__QDISC_STATE_SCHED, &q->state); 5382 qdisc_run(q); 5383 if (root_lock) 5384 spin_unlock(root_lock); 5385 } 5386 5387 rcu_read_unlock(); 5388 } 5389 5390 xfrm_dev_backlog(sd); 5391 } 5392 5393 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5394 /* This hook is defined here for ATM LANE */ 5395 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5396 unsigned char *addr) __read_mostly; 5397 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5398 #endif 5399 5400 /** 5401 * netdev_is_rx_handler_busy - check if receive handler is registered 5402 * @dev: device to check 5403 * 5404 * Check if a receive handler is already registered for a given device. 5405 * Return true if there one. 5406 * 5407 * The caller must hold the rtnl_mutex. 5408 */ 5409 bool netdev_is_rx_handler_busy(struct net_device *dev) 5410 { 5411 ASSERT_RTNL(); 5412 return dev && rtnl_dereference(dev->rx_handler); 5413 } 5414 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5415 5416 /** 5417 * netdev_rx_handler_register - register receive handler 5418 * @dev: device to register a handler for 5419 * @rx_handler: receive handler to register 5420 * @rx_handler_data: data pointer that is used by rx handler 5421 * 5422 * Register a receive handler for a device. This handler will then be 5423 * called from __netif_receive_skb. A negative errno code is returned 5424 * on a failure. 5425 * 5426 * The caller must hold the rtnl_mutex. 5427 * 5428 * For a general description of rx_handler, see enum rx_handler_result. 5429 */ 5430 int netdev_rx_handler_register(struct net_device *dev, 5431 rx_handler_func_t *rx_handler, 5432 void *rx_handler_data) 5433 { 5434 if (netdev_is_rx_handler_busy(dev)) 5435 return -EBUSY; 5436 5437 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5438 return -EINVAL; 5439 5440 /* Note: rx_handler_data must be set before rx_handler */ 5441 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5442 rcu_assign_pointer(dev->rx_handler, rx_handler); 5443 5444 return 0; 5445 } 5446 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5447 5448 /** 5449 * netdev_rx_handler_unregister - unregister receive handler 5450 * @dev: device to unregister a handler from 5451 * 5452 * Unregister a receive handler from a device. 5453 * 5454 * The caller must hold the rtnl_mutex. 5455 */ 5456 void netdev_rx_handler_unregister(struct net_device *dev) 5457 { 5458 5459 ASSERT_RTNL(); 5460 RCU_INIT_POINTER(dev->rx_handler, NULL); 5461 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5462 * section has a guarantee to see a non NULL rx_handler_data 5463 * as well. 5464 */ 5465 synchronize_net(); 5466 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5467 } 5468 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5469 5470 /* 5471 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5472 * the special handling of PFMEMALLOC skbs. 5473 */ 5474 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5475 { 5476 switch (skb->protocol) { 5477 case htons(ETH_P_ARP): 5478 case htons(ETH_P_IP): 5479 case htons(ETH_P_IPV6): 5480 case htons(ETH_P_8021Q): 5481 case htons(ETH_P_8021AD): 5482 return true; 5483 default: 5484 return false; 5485 } 5486 } 5487 5488 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5489 int *ret, struct net_device *orig_dev) 5490 { 5491 if (nf_hook_ingress_active(skb)) { 5492 int ingress_retval; 5493 5494 if (*pt_prev) { 5495 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5496 *pt_prev = NULL; 5497 } 5498 5499 rcu_read_lock(); 5500 ingress_retval = nf_hook_ingress(skb); 5501 rcu_read_unlock(); 5502 return ingress_retval; 5503 } 5504 return 0; 5505 } 5506 5507 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5508 struct packet_type **ppt_prev) 5509 { 5510 struct packet_type *ptype, *pt_prev; 5511 rx_handler_func_t *rx_handler; 5512 struct sk_buff *skb = *pskb; 5513 struct net_device *orig_dev; 5514 bool deliver_exact = false; 5515 int ret = NET_RX_DROP; 5516 __be16 type; 5517 5518 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5519 5520 trace_netif_receive_skb(skb); 5521 5522 orig_dev = skb->dev; 5523 5524 skb_reset_network_header(skb); 5525 #if !defined(CONFIG_DEBUG_NET) 5526 /* We plan to no longer reset the transport header here. 5527 * Give some time to fuzzers and dev build to catch bugs 5528 * in network stacks. 5529 */ 5530 if (!skb_transport_header_was_set(skb)) 5531 skb_reset_transport_header(skb); 5532 #endif 5533 skb_reset_mac_len(skb); 5534 5535 pt_prev = NULL; 5536 5537 another_round: 5538 skb->skb_iif = skb->dev->ifindex; 5539 5540 __this_cpu_inc(softnet_data.processed); 5541 5542 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5543 int ret2; 5544 5545 migrate_disable(); 5546 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), 5547 &skb); 5548 migrate_enable(); 5549 5550 if (ret2 != XDP_PASS) { 5551 ret = NET_RX_DROP; 5552 goto out; 5553 } 5554 } 5555 5556 if (eth_type_vlan(skb->protocol)) { 5557 skb = skb_vlan_untag(skb); 5558 if (unlikely(!skb)) 5559 goto out; 5560 } 5561 5562 if (skb_skip_tc_classify(skb)) 5563 goto skip_classify; 5564 5565 if (pfmemalloc) 5566 goto skip_taps; 5567 5568 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) { 5569 if (pt_prev) 5570 ret = deliver_skb(skb, pt_prev, orig_dev); 5571 pt_prev = ptype; 5572 } 5573 5574 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5575 if (pt_prev) 5576 ret = deliver_skb(skb, pt_prev, orig_dev); 5577 pt_prev = ptype; 5578 } 5579 5580 skip_taps: 5581 #ifdef CONFIG_NET_INGRESS 5582 if (static_branch_unlikely(&ingress_needed_key)) { 5583 bool another = false; 5584 5585 nf_skip_egress(skb, true); 5586 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5587 &another); 5588 if (another) 5589 goto another_round; 5590 if (!skb) 5591 goto out; 5592 5593 nf_skip_egress(skb, false); 5594 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5595 goto out; 5596 } 5597 #endif 5598 skb_reset_redirect(skb); 5599 skip_classify: 5600 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5601 goto drop; 5602 5603 if (skb_vlan_tag_present(skb)) { 5604 if (pt_prev) { 5605 ret = deliver_skb(skb, pt_prev, orig_dev); 5606 pt_prev = NULL; 5607 } 5608 if (vlan_do_receive(&skb)) 5609 goto another_round; 5610 else if (unlikely(!skb)) 5611 goto out; 5612 } 5613 5614 rx_handler = rcu_dereference(skb->dev->rx_handler); 5615 if (rx_handler) { 5616 if (pt_prev) { 5617 ret = deliver_skb(skb, pt_prev, orig_dev); 5618 pt_prev = NULL; 5619 } 5620 switch (rx_handler(&skb)) { 5621 case RX_HANDLER_CONSUMED: 5622 ret = NET_RX_SUCCESS; 5623 goto out; 5624 case RX_HANDLER_ANOTHER: 5625 goto another_round; 5626 case RX_HANDLER_EXACT: 5627 deliver_exact = true; 5628 break; 5629 case RX_HANDLER_PASS: 5630 break; 5631 default: 5632 BUG(); 5633 } 5634 } 5635 5636 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5637 check_vlan_id: 5638 if (skb_vlan_tag_get_id(skb)) { 5639 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5640 * find vlan device. 5641 */ 5642 skb->pkt_type = PACKET_OTHERHOST; 5643 } else if (eth_type_vlan(skb->protocol)) { 5644 /* Outer header is 802.1P with vlan 0, inner header is 5645 * 802.1Q or 802.1AD and vlan_do_receive() above could 5646 * not find vlan dev for vlan id 0. 5647 */ 5648 __vlan_hwaccel_clear_tag(skb); 5649 skb = skb_vlan_untag(skb); 5650 if (unlikely(!skb)) 5651 goto out; 5652 if (vlan_do_receive(&skb)) 5653 /* After stripping off 802.1P header with vlan 0 5654 * vlan dev is found for inner header. 5655 */ 5656 goto another_round; 5657 else if (unlikely(!skb)) 5658 goto out; 5659 else 5660 /* We have stripped outer 802.1P vlan 0 header. 5661 * But could not find vlan dev. 5662 * check again for vlan id to set OTHERHOST. 5663 */ 5664 goto check_vlan_id; 5665 } 5666 /* Note: we might in the future use prio bits 5667 * and set skb->priority like in vlan_do_receive() 5668 * For the time being, just ignore Priority Code Point 5669 */ 5670 __vlan_hwaccel_clear_tag(skb); 5671 } 5672 5673 type = skb->protocol; 5674 5675 /* deliver only exact match when indicated */ 5676 if (likely(!deliver_exact)) { 5677 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5678 &ptype_base[ntohs(type) & 5679 PTYPE_HASH_MASK]); 5680 } 5681 5682 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5683 &orig_dev->ptype_specific); 5684 5685 if (unlikely(skb->dev != orig_dev)) { 5686 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5687 &skb->dev->ptype_specific); 5688 } 5689 5690 if (pt_prev) { 5691 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5692 goto drop; 5693 *ppt_prev = pt_prev; 5694 } else { 5695 drop: 5696 if (!deliver_exact) 5697 dev_core_stats_rx_dropped_inc(skb->dev); 5698 else 5699 dev_core_stats_rx_nohandler_inc(skb->dev); 5700 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5701 /* Jamal, now you will not able to escape explaining 5702 * me how you were going to use this. :-) 5703 */ 5704 ret = NET_RX_DROP; 5705 } 5706 5707 out: 5708 /* The invariant here is that if *ppt_prev is not NULL 5709 * then skb should also be non-NULL. 5710 * 5711 * Apparently *ppt_prev assignment above holds this invariant due to 5712 * skb dereferencing near it. 5713 */ 5714 *pskb = skb; 5715 return ret; 5716 } 5717 5718 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5719 { 5720 struct net_device *orig_dev = skb->dev; 5721 struct packet_type *pt_prev = NULL; 5722 int ret; 5723 5724 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5725 if (pt_prev) 5726 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5727 skb->dev, pt_prev, orig_dev); 5728 return ret; 5729 } 5730 5731 /** 5732 * netif_receive_skb_core - special purpose version of netif_receive_skb 5733 * @skb: buffer to process 5734 * 5735 * More direct receive version of netif_receive_skb(). It should 5736 * only be used by callers that have a need to skip RPS and Generic XDP. 5737 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5738 * 5739 * This function may only be called from softirq context and interrupts 5740 * should be enabled. 5741 * 5742 * Return values (usually ignored): 5743 * NET_RX_SUCCESS: no congestion 5744 * NET_RX_DROP: packet was dropped 5745 */ 5746 int netif_receive_skb_core(struct sk_buff *skb) 5747 { 5748 int ret; 5749 5750 rcu_read_lock(); 5751 ret = __netif_receive_skb_one_core(skb, false); 5752 rcu_read_unlock(); 5753 5754 return ret; 5755 } 5756 EXPORT_SYMBOL(netif_receive_skb_core); 5757 5758 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5759 struct packet_type *pt_prev, 5760 struct net_device *orig_dev) 5761 { 5762 struct sk_buff *skb, *next; 5763 5764 if (!pt_prev) 5765 return; 5766 if (list_empty(head)) 5767 return; 5768 if (pt_prev->list_func != NULL) 5769 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5770 ip_list_rcv, head, pt_prev, orig_dev); 5771 else 5772 list_for_each_entry_safe(skb, next, head, list) { 5773 skb_list_del_init(skb); 5774 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5775 } 5776 } 5777 5778 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5779 { 5780 /* Fast-path assumptions: 5781 * - There is no RX handler. 5782 * - Only one packet_type matches. 5783 * If either of these fails, we will end up doing some per-packet 5784 * processing in-line, then handling the 'last ptype' for the whole 5785 * sublist. This can't cause out-of-order delivery to any single ptype, 5786 * because the 'last ptype' must be constant across the sublist, and all 5787 * other ptypes are handled per-packet. 5788 */ 5789 /* Current (common) ptype of sublist */ 5790 struct packet_type *pt_curr = NULL; 5791 /* Current (common) orig_dev of sublist */ 5792 struct net_device *od_curr = NULL; 5793 struct sk_buff *skb, *next; 5794 LIST_HEAD(sublist); 5795 5796 list_for_each_entry_safe(skb, next, head, list) { 5797 struct net_device *orig_dev = skb->dev; 5798 struct packet_type *pt_prev = NULL; 5799 5800 skb_list_del_init(skb); 5801 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5802 if (!pt_prev) 5803 continue; 5804 if (pt_curr != pt_prev || od_curr != orig_dev) { 5805 /* dispatch old sublist */ 5806 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5807 /* start new sublist */ 5808 INIT_LIST_HEAD(&sublist); 5809 pt_curr = pt_prev; 5810 od_curr = orig_dev; 5811 } 5812 list_add_tail(&skb->list, &sublist); 5813 } 5814 5815 /* dispatch final sublist */ 5816 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5817 } 5818 5819 static int __netif_receive_skb(struct sk_buff *skb) 5820 { 5821 int ret; 5822 5823 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5824 unsigned int noreclaim_flag; 5825 5826 /* 5827 * PFMEMALLOC skbs are special, they should 5828 * - be delivered to SOCK_MEMALLOC sockets only 5829 * - stay away from userspace 5830 * - have bounded memory usage 5831 * 5832 * Use PF_MEMALLOC as this saves us from propagating the allocation 5833 * context down to all allocation sites. 5834 */ 5835 noreclaim_flag = memalloc_noreclaim_save(); 5836 ret = __netif_receive_skb_one_core(skb, true); 5837 memalloc_noreclaim_restore(noreclaim_flag); 5838 } else 5839 ret = __netif_receive_skb_one_core(skb, false); 5840 5841 return ret; 5842 } 5843 5844 static void __netif_receive_skb_list(struct list_head *head) 5845 { 5846 unsigned long noreclaim_flag = 0; 5847 struct sk_buff *skb, *next; 5848 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5849 5850 list_for_each_entry_safe(skb, next, head, list) { 5851 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5852 struct list_head sublist; 5853 5854 /* Handle the previous sublist */ 5855 list_cut_before(&sublist, head, &skb->list); 5856 if (!list_empty(&sublist)) 5857 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5858 pfmemalloc = !pfmemalloc; 5859 /* See comments in __netif_receive_skb */ 5860 if (pfmemalloc) 5861 noreclaim_flag = memalloc_noreclaim_save(); 5862 else 5863 memalloc_noreclaim_restore(noreclaim_flag); 5864 } 5865 } 5866 /* Handle the remaining sublist */ 5867 if (!list_empty(head)) 5868 __netif_receive_skb_list_core(head, pfmemalloc); 5869 /* Restore pflags */ 5870 if (pfmemalloc) 5871 memalloc_noreclaim_restore(noreclaim_flag); 5872 } 5873 5874 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5875 { 5876 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5877 struct bpf_prog *new = xdp->prog; 5878 int ret = 0; 5879 5880 switch (xdp->command) { 5881 case XDP_SETUP_PROG: 5882 rcu_assign_pointer(dev->xdp_prog, new); 5883 if (old) 5884 bpf_prog_put(old); 5885 5886 if (old && !new) { 5887 static_branch_dec(&generic_xdp_needed_key); 5888 } else if (new && !old) { 5889 static_branch_inc(&generic_xdp_needed_key); 5890 dev_disable_lro(dev); 5891 dev_disable_gro_hw(dev); 5892 } 5893 break; 5894 5895 default: 5896 ret = -EINVAL; 5897 break; 5898 } 5899 5900 return ret; 5901 } 5902 5903 static int netif_receive_skb_internal(struct sk_buff *skb) 5904 { 5905 int ret; 5906 5907 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5908 5909 if (skb_defer_rx_timestamp(skb)) 5910 return NET_RX_SUCCESS; 5911 5912 rcu_read_lock(); 5913 #ifdef CONFIG_RPS 5914 if (static_branch_unlikely(&rps_needed)) { 5915 struct rps_dev_flow voidflow, *rflow = &voidflow; 5916 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5917 5918 if (cpu >= 0) { 5919 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5920 rcu_read_unlock(); 5921 return ret; 5922 } 5923 } 5924 #endif 5925 ret = __netif_receive_skb(skb); 5926 rcu_read_unlock(); 5927 return ret; 5928 } 5929 5930 void netif_receive_skb_list_internal(struct list_head *head) 5931 { 5932 struct sk_buff *skb, *next; 5933 LIST_HEAD(sublist); 5934 5935 list_for_each_entry_safe(skb, next, head, list) { 5936 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), 5937 skb); 5938 skb_list_del_init(skb); 5939 if (!skb_defer_rx_timestamp(skb)) 5940 list_add_tail(&skb->list, &sublist); 5941 } 5942 list_splice_init(&sublist, head); 5943 5944 rcu_read_lock(); 5945 #ifdef CONFIG_RPS 5946 if (static_branch_unlikely(&rps_needed)) { 5947 list_for_each_entry_safe(skb, next, head, list) { 5948 struct rps_dev_flow voidflow, *rflow = &voidflow; 5949 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5950 5951 if (cpu >= 0) { 5952 /* Will be handled, remove from list */ 5953 skb_list_del_init(skb); 5954 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5955 } 5956 } 5957 } 5958 #endif 5959 __netif_receive_skb_list(head); 5960 rcu_read_unlock(); 5961 } 5962 5963 /** 5964 * netif_receive_skb - process receive buffer from network 5965 * @skb: buffer to process 5966 * 5967 * netif_receive_skb() is the main receive data processing function. 5968 * It always succeeds. The buffer may be dropped during processing 5969 * for congestion control or by the protocol layers. 5970 * 5971 * This function may only be called from softirq context and interrupts 5972 * should be enabled. 5973 * 5974 * Return values (usually ignored): 5975 * NET_RX_SUCCESS: no congestion 5976 * NET_RX_DROP: packet was dropped 5977 */ 5978 int netif_receive_skb(struct sk_buff *skb) 5979 { 5980 int ret; 5981 5982 trace_netif_receive_skb_entry(skb); 5983 5984 ret = netif_receive_skb_internal(skb); 5985 trace_netif_receive_skb_exit(ret); 5986 5987 return ret; 5988 } 5989 EXPORT_SYMBOL(netif_receive_skb); 5990 5991 /** 5992 * netif_receive_skb_list - process many receive buffers from network 5993 * @head: list of skbs to process. 5994 * 5995 * Since return value of netif_receive_skb() is normally ignored, and 5996 * wouldn't be meaningful for a list, this function returns void. 5997 * 5998 * This function may only be called from softirq context and interrupts 5999 * should be enabled. 6000 */ 6001 void netif_receive_skb_list(struct list_head *head) 6002 { 6003 struct sk_buff *skb; 6004 6005 if (list_empty(head)) 6006 return; 6007 if (trace_netif_receive_skb_list_entry_enabled()) { 6008 list_for_each_entry(skb, head, list) 6009 trace_netif_receive_skb_list_entry(skb); 6010 } 6011 netif_receive_skb_list_internal(head); 6012 trace_netif_receive_skb_list_exit(0); 6013 } 6014 EXPORT_SYMBOL(netif_receive_skb_list); 6015 6016 static DEFINE_PER_CPU(struct work_struct, flush_works); 6017 6018 /* Network device is going away, flush any packets still pending */ 6019 static void flush_backlog(struct work_struct *work) 6020 { 6021 struct sk_buff *skb, *tmp; 6022 struct softnet_data *sd; 6023 6024 local_bh_disable(); 6025 sd = this_cpu_ptr(&softnet_data); 6026 6027 backlog_lock_irq_disable(sd); 6028 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 6029 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 6030 __skb_unlink(skb, &sd->input_pkt_queue); 6031 dev_kfree_skb_irq(skb); 6032 rps_input_queue_head_incr(sd); 6033 } 6034 } 6035 backlog_unlock_irq_enable(sd); 6036 6037 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6038 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 6039 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 6040 __skb_unlink(skb, &sd->process_queue); 6041 kfree_skb(skb); 6042 rps_input_queue_head_incr(sd); 6043 } 6044 } 6045 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6046 local_bh_enable(); 6047 } 6048 6049 static bool flush_required(int cpu) 6050 { 6051 #if IS_ENABLED(CONFIG_RPS) 6052 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 6053 bool do_flush; 6054 6055 backlog_lock_irq_disable(sd); 6056 6057 /* as insertion into process_queue happens with the rps lock held, 6058 * process_queue access may race only with dequeue 6059 */ 6060 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 6061 !skb_queue_empty_lockless(&sd->process_queue); 6062 backlog_unlock_irq_enable(sd); 6063 6064 return do_flush; 6065 #endif 6066 /* without RPS we can't safely check input_pkt_queue: during a 6067 * concurrent remote skb_queue_splice() we can detect as empty both 6068 * input_pkt_queue and process_queue even if the latter could end-up 6069 * containing a lot of packets. 6070 */ 6071 return true; 6072 } 6073 6074 static void flush_all_backlogs(void) 6075 { 6076 static cpumask_t flush_cpus; 6077 unsigned int cpu; 6078 6079 /* since we are under rtnl lock protection we can use static data 6080 * for the cpumask and avoid allocating on stack the possibly 6081 * large mask 6082 */ 6083 ASSERT_RTNL(); 6084 6085 cpus_read_lock(); 6086 6087 cpumask_clear(&flush_cpus); 6088 for_each_online_cpu(cpu) { 6089 if (flush_required(cpu)) { 6090 queue_work_on(cpu, system_highpri_wq, 6091 per_cpu_ptr(&flush_works, cpu)); 6092 cpumask_set_cpu(cpu, &flush_cpus); 6093 } 6094 } 6095 6096 /* we can have in flight packet[s] on the cpus we are not flushing, 6097 * synchronize_net() in unregister_netdevice_many() will take care of 6098 * them 6099 */ 6100 for_each_cpu(cpu, &flush_cpus) 6101 flush_work(per_cpu_ptr(&flush_works, cpu)); 6102 6103 cpus_read_unlock(); 6104 } 6105 6106 static void net_rps_send_ipi(struct softnet_data *remsd) 6107 { 6108 #ifdef CONFIG_RPS 6109 while (remsd) { 6110 struct softnet_data *next = remsd->rps_ipi_next; 6111 6112 if (cpu_online(remsd->cpu)) 6113 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6114 remsd = next; 6115 } 6116 #endif 6117 } 6118 6119 /* 6120 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6121 * Note: called with local irq disabled, but exits with local irq enabled. 6122 */ 6123 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6124 { 6125 #ifdef CONFIG_RPS 6126 struct softnet_data *remsd = sd->rps_ipi_list; 6127 6128 if (!use_backlog_threads() && remsd) { 6129 sd->rps_ipi_list = NULL; 6130 6131 local_irq_enable(); 6132 6133 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6134 net_rps_send_ipi(remsd); 6135 } else 6136 #endif 6137 local_irq_enable(); 6138 } 6139 6140 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6141 { 6142 #ifdef CONFIG_RPS 6143 return !use_backlog_threads() && sd->rps_ipi_list; 6144 #else 6145 return false; 6146 #endif 6147 } 6148 6149 static int process_backlog(struct napi_struct *napi, int quota) 6150 { 6151 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6152 bool again = true; 6153 int work = 0; 6154 6155 /* Check if we have pending ipi, its better to send them now, 6156 * not waiting net_rx_action() end. 6157 */ 6158 if (sd_has_rps_ipi_waiting(sd)) { 6159 local_irq_disable(); 6160 net_rps_action_and_irq_enable(sd); 6161 } 6162 6163 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight); 6164 while (again) { 6165 struct sk_buff *skb; 6166 6167 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6168 while ((skb = __skb_dequeue(&sd->process_queue))) { 6169 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6170 rcu_read_lock(); 6171 __netif_receive_skb(skb); 6172 rcu_read_unlock(); 6173 if (++work >= quota) { 6174 rps_input_queue_head_add(sd, work); 6175 return work; 6176 } 6177 6178 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6179 } 6180 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6181 6182 backlog_lock_irq_disable(sd); 6183 if (skb_queue_empty(&sd->input_pkt_queue)) { 6184 /* 6185 * Inline a custom version of __napi_complete(). 6186 * only current cpu owns and manipulates this napi, 6187 * and NAPI_STATE_SCHED is the only possible flag set 6188 * on backlog. 6189 * We can use a plain write instead of clear_bit(), 6190 * and we dont need an smp_mb() memory barrier. 6191 */ 6192 napi->state &= NAPIF_STATE_THREADED; 6193 again = false; 6194 } else { 6195 local_lock_nested_bh(&softnet_data.process_queue_bh_lock); 6196 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6197 &sd->process_queue); 6198 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock); 6199 } 6200 backlog_unlock_irq_enable(sd); 6201 } 6202 6203 if (work) 6204 rps_input_queue_head_add(sd, work); 6205 return work; 6206 } 6207 6208 /** 6209 * __napi_schedule - schedule for receive 6210 * @n: entry to schedule 6211 * 6212 * The entry's receive function will be scheduled to run. 6213 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6214 */ 6215 void __napi_schedule(struct napi_struct *n) 6216 { 6217 unsigned long flags; 6218 6219 local_irq_save(flags); 6220 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6221 local_irq_restore(flags); 6222 } 6223 EXPORT_SYMBOL(__napi_schedule); 6224 6225 /** 6226 * napi_schedule_prep - check if napi can be scheduled 6227 * @n: napi context 6228 * 6229 * Test if NAPI routine is already running, and if not mark 6230 * it as running. This is used as a condition variable to 6231 * insure only one NAPI poll instance runs. We also make 6232 * sure there is no pending NAPI disable. 6233 */ 6234 bool napi_schedule_prep(struct napi_struct *n) 6235 { 6236 unsigned long new, val = READ_ONCE(n->state); 6237 6238 do { 6239 if (unlikely(val & NAPIF_STATE_DISABLE)) 6240 return false; 6241 new = val | NAPIF_STATE_SCHED; 6242 6243 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6244 * This was suggested by Alexander Duyck, as compiler 6245 * emits better code than : 6246 * if (val & NAPIF_STATE_SCHED) 6247 * new |= NAPIF_STATE_MISSED; 6248 */ 6249 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6250 NAPIF_STATE_MISSED; 6251 } while (!try_cmpxchg(&n->state, &val, new)); 6252 6253 return !(val & NAPIF_STATE_SCHED); 6254 } 6255 EXPORT_SYMBOL(napi_schedule_prep); 6256 6257 /** 6258 * __napi_schedule_irqoff - schedule for receive 6259 * @n: entry to schedule 6260 * 6261 * Variant of __napi_schedule() assuming hard irqs are masked. 6262 * 6263 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6264 * because the interrupt disabled assumption might not be true 6265 * due to force-threaded interrupts and spinlock substitution. 6266 */ 6267 void __napi_schedule_irqoff(struct napi_struct *n) 6268 { 6269 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6270 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6271 else 6272 __napi_schedule(n); 6273 } 6274 EXPORT_SYMBOL(__napi_schedule_irqoff); 6275 6276 bool napi_complete_done(struct napi_struct *n, int work_done) 6277 { 6278 unsigned long flags, val, new, timeout = 0; 6279 bool ret = true; 6280 6281 /* 6282 * 1) Don't let napi dequeue from the cpu poll list 6283 * just in case its running on a different cpu. 6284 * 2) If we are busy polling, do nothing here, we have 6285 * the guarantee we will be called later. 6286 */ 6287 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6288 NAPIF_STATE_IN_BUSY_POLL))) 6289 return false; 6290 6291 if (work_done) { 6292 if (n->gro_bitmask) 6293 timeout = napi_get_gro_flush_timeout(n); 6294 n->defer_hard_irqs_count = napi_get_defer_hard_irqs(n); 6295 } 6296 if (n->defer_hard_irqs_count > 0) { 6297 n->defer_hard_irqs_count--; 6298 timeout = napi_get_gro_flush_timeout(n); 6299 if (timeout) 6300 ret = false; 6301 } 6302 if (n->gro_bitmask) { 6303 /* When the NAPI instance uses a timeout and keeps postponing 6304 * it, we need to bound somehow the time packets are kept in 6305 * the GRO layer 6306 */ 6307 napi_gro_flush(n, !!timeout); 6308 } 6309 6310 gro_normal_list(n); 6311 6312 if (unlikely(!list_empty(&n->poll_list))) { 6313 /* If n->poll_list is not empty, we need to mask irqs */ 6314 local_irq_save(flags); 6315 list_del_init(&n->poll_list); 6316 local_irq_restore(flags); 6317 } 6318 WRITE_ONCE(n->list_owner, -1); 6319 6320 val = READ_ONCE(n->state); 6321 do { 6322 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6323 6324 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6325 NAPIF_STATE_SCHED_THREADED | 6326 NAPIF_STATE_PREFER_BUSY_POLL); 6327 6328 /* If STATE_MISSED was set, leave STATE_SCHED set, 6329 * because we will call napi->poll() one more time. 6330 * This C code was suggested by Alexander Duyck to help gcc. 6331 */ 6332 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6333 NAPIF_STATE_SCHED; 6334 } while (!try_cmpxchg(&n->state, &val, new)); 6335 6336 if (unlikely(val & NAPIF_STATE_MISSED)) { 6337 __napi_schedule(n); 6338 return false; 6339 } 6340 6341 if (timeout) 6342 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6343 HRTIMER_MODE_REL_PINNED); 6344 return ret; 6345 } 6346 EXPORT_SYMBOL(napi_complete_done); 6347 6348 static void skb_defer_free_flush(struct softnet_data *sd) 6349 { 6350 struct sk_buff *skb, *next; 6351 6352 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6353 if (!READ_ONCE(sd->defer_list)) 6354 return; 6355 6356 spin_lock(&sd->defer_lock); 6357 skb = sd->defer_list; 6358 sd->defer_list = NULL; 6359 sd->defer_count = 0; 6360 spin_unlock(&sd->defer_lock); 6361 6362 while (skb != NULL) { 6363 next = skb->next; 6364 napi_consume_skb(skb, 1); 6365 skb = next; 6366 } 6367 } 6368 6369 #if defined(CONFIG_NET_RX_BUSY_POLL) 6370 6371 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6372 { 6373 if (!skip_schedule) { 6374 gro_normal_list(napi); 6375 __napi_schedule(napi); 6376 return; 6377 } 6378 6379 if (napi->gro_bitmask) { 6380 /* flush too old packets 6381 * If HZ < 1000, flush all packets. 6382 */ 6383 napi_gro_flush(napi, HZ >= 1000); 6384 } 6385 6386 gro_normal_list(napi); 6387 clear_bit(NAPI_STATE_SCHED, &napi->state); 6388 } 6389 6390 enum { 6391 NAPI_F_PREFER_BUSY_POLL = 1, 6392 NAPI_F_END_ON_RESCHED = 2, 6393 }; 6394 6395 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, 6396 unsigned flags, u16 budget) 6397 { 6398 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6399 bool skip_schedule = false; 6400 unsigned long timeout; 6401 int rc; 6402 6403 /* Busy polling means there is a high chance device driver hard irq 6404 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6405 * set in napi_schedule_prep(). 6406 * Since we are about to call napi->poll() once more, we can safely 6407 * clear NAPI_STATE_MISSED. 6408 * 6409 * Note: x86 could use a single "lock and ..." instruction 6410 * to perform these two clear_bit() 6411 */ 6412 clear_bit(NAPI_STATE_MISSED, &napi->state); 6413 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6414 6415 local_bh_disable(); 6416 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6417 6418 if (flags & NAPI_F_PREFER_BUSY_POLL) { 6419 napi->defer_hard_irqs_count = napi_get_defer_hard_irqs(napi); 6420 timeout = napi_get_gro_flush_timeout(napi); 6421 if (napi->defer_hard_irqs_count && timeout) { 6422 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6423 skip_schedule = true; 6424 } 6425 } 6426 6427 /* All we really want here is to re-enable device interrupts. 6428 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6429 */ 6430 rc = napi->poll(napi, budget); 6431 /* We can't gro_normal_list() here, because napi->poll() might have 6432 * rearmed the napi (napi_complete_done()) in which case it could 6433 * already be running on another CPU. 6434 */ 6435 trace_napi_poll(napi, rc, budget); 6436 netpoll_poll_unlock(have_poll_lock); 6437 if (rc == budget) 6438 __busy_poll_stop(napi, skip_schedule); 6439 bpf_net_ctx_clear(bpf_net_ctx); 6440 local_bh_enable(); 6441 } 6442 6443 static void __napi_busy_loop(unsigned int napi_id, 6444 bool (*loop_end)(void *, unsigned long), 6445 void *loop_end_arg, unsigned flags, u16 budget) 6446 { 6447 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6448 int (*napi_poll)(struct napi_struct *napi, int budget); 6449 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 6450 void *have_poll_lock = NULL; 6451 struct napi_struct *napi; 6452 6453 WARN_ON_ONCE(!rcu_read_lock_held()); 6454 6455 restart: 6456 napi_poll = NULL; 6457 6458 napi = napi_by_id(napi_id); 6459 if (!napi) 6460 return; 6461 6462 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6463 preempt_disable(); 6464 for (;;) { 6465 int work = 0; 6466 6467 local_bh_disable(); 6468 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 6469 if (!napi_poll) { 6470 unsigned long val = READ_ONCE(napi->state); 6471 6472 /* If multiple threads are competing for this napi, 6473 * we avoid dirtying napi->state as much as we can. 6474 */ 6475 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6476 NAPIF_STATE_IN_BUSY_POLL)) { 6477 if (flags & NAPI_F_PREFER_BUSY_POLL) 6478 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6479 goto count; 6480 } 6481 if (cmpxchg(&napi->state, val, 6482 val | NAPIF_STATE_IN_BUSY_POLL | 6483 NAPIF_STATE_SCHED) != val) { 6484 if (flags & NAPI_F_PREFER_BUSY_POLL) 6485 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6486 goto count; 6487 } 6488 have_poll_lock = netpoll_poll_lock(napi); 6489 napi_poll = napi->poll; 6490 } 6491 work = napi_poll(napi, budget); 6492 trace_napi_poll(napi, work, budget); 6493 gro_normal_list(napi); 6494 count: 6495 if (work > 0) 6496 __NET_ADD_STATS(dev_net(napi->dev), 6497 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6498 skb_defer_free_flush(this_cpu_ptr(&softnet_data)); 6499 bpf_net_ctx_clear(bpf_net_ctx); 6500 local_bh_enable(); 6501 6502 if (!loop_end || loop_end(loop_end_arg, start_time)) 6503 break; 6504 6505 if (unlikely(need_resched())) { 6506 if (flags & NAPI_F_END_ON_RESCHED) 6507 break; 6508 if (napi_poll) 6509 busy_poll_stop(napi, have_poll_lock, flags, budget); 6510 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6511 preempt_enable(); 6512 rcu_read_unlock(); 6513 cond_resched(); 6514 rcu_read_lock(); 6515 if (loop_end(loop_end_arg, start_time)) 6516 return; 6517 goto restart; 6518 } 6519 cpu_relax(); 6520 } 6521 if (napi_poll) 6522 busy_poll_stop(napi, have_poll_lock, flags, budget); 6523 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6524 preempt_enable(); 6525 } 6526 6527 void napi_busy_loop_rcu(unsigned int napi_id, 6528 bool (*loop_end)(void *, unsigned long), 6529 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6530 { 6531 unsigned flags = NAPI_F_END_ON_RESCHED; 6532 6533 if (prefer_busy_poll) 6534 flags |= NAPI_F_PREFER_BUSY_POLL; 6535 6536 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6537 } 6538 6539 void napi_busy_loop(unsigned int napi_id, 6540 bool (*loop_end)(void *, unsigned long), 6541 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6542 { 6543 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; 6544 6545 rcu_read_lock(); 6546 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6547 rcu_read_unlock(); 6548 } 6549 EXPORT_SYMBOL(napi_busy_loop); 6550 6551 void napi_suspend_irqs(unsigned int napi_id) 6552 { 6553 struct napi_struct *napi; 6554 6555 rcu_read_lock(); 6556 napi = napi_by_id(napi_id); 6557 if (napi) { 6558 unsigned long timeout = napi_get_irq_suspend_timeout(napi); 6559 6560 if (timeout) 6561 hrtimer_start(&napi->timer, ns_to_ktime(timeout), 6562 HRTIMER_MODE_REL_PINNED); 6563 } 6564 rcu_read_unlock(); 6565 } 6566 6567 void napi_resume_irqs(unsigned int napi_id) 6568 { 6569 struct napi_struct *napi; 6570 6571 rcu_read_lock(); 6572 napi = napi_by_id(napi_id); 6573 if (napi) { 6574 /* If irq_suspend_timeout is set to 0 between the call to 6575 * napi_suspend_irqs and now, the original value still 6576 * determines the safety timeout as intended and napi_watchdog 6577 * will resume irq processing. 6578 */ 6579 if (napi_get_irq_suspend_timeout(napi)) { 6580 local_bh_disable(); 6581 napi_schedule(napi); 6582 local_bh_enable(); 6583 } 6584 } 6585 rcu_read_unlock(); 6586 } 6587 6588 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6589 6590 static void __napi_hash_add_with_id(struct napi_struct *napi, 6591 unsigned int napi_id) 6592 { 6593 napi->napi_id = napi_id; 6594 hlist_add_head_rcu(&napi->napi_hash_node, 6595 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6596 } 6597 6598 static void napi_hash_add_with_id(struct napi_struct *napi, 6599 unsigned int napi_id) 6600 { 6601 unsigned long flags; 6602 6603 spin_lock_irqsave(&napi_hash_lock, flags); 6604 WARN_ON_ONCE(napi_by_id(napi_id)); 6605 __napi_hash_add_with_id(napi, napi_id); 6606 spin_unlock_irqrestore(&napi_hash_lock, flags); 6607 } 6608 6609 static void napi_hash_add(struct napi_struct *napi) 6610 { 6611 unsigned long flags; 6612 6613 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6614 return; 6615 6616 spin_lock_irqsave(&napi_hash_lock, flags); 6617 6618 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6619 do { 6620 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6621 napi_gen_id = MIN_NAPI_ID; 6622 } while (napi_by_id(napi_gen_id)); 6623 6624 __napi_hash_add_with_id(napi, napi_gen_id); 6625 6626 spin_unlock_irqrestore(&napi_hash_lock, flags); 6627 } 6628 6629 /* Warning : caller is responsible to make sure rcu grace period 6630 * is respected before freeing memory containing @napi 6631 */ 6632 static void napi_hash_del(struct napi_struct *napi) 6633 { 6634 unsigned long flags; 6635 6636 spin_lock_irqsave(&napi_hash_lock, flags); 6637 6638 hlist_del_init_rcu(&napi->napi_hash_node); 6639 6640 spin_unlock_irqrestore(&napi_hash_lock, flags); 6641 } 6642 6643 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6644 { 6645 struct napi_struct *napi; 6646 6647 napi = container_of(timer, struct napi_struct, timer); 6648 6649 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6650 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6651 */ 6652 if (!napi_disable_pending(napi) && 6653 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6654 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6655 __napi_schedule_irqoff(napi); 6656 } 6657 6658 return HRTIMER_NORESTART; 6659 } 6660 6661 static void init_gro_hash(struct napi_struct *napi) 6662 { 6663 int i; 6664 6665 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6666 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6667 napi->gro_hash[i].count = 0; 6668 } 6669 napi->gro_bitmask = 0; 6670 } 6671 6672 int dev_set_threaded(struct net_device *dev, bool threaded) 6673 { 6674 struct napi_struct *napi; 6675 int err = 0; 6676 6677 if (dev->threaded == threaded) 6678 return 0; 6679 6680 if (threaded) { 6681 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6682 if (!napi->thread) { 6683 err = napi_kthread_create(napi); 6684 if (err) { 6685 threaded = false; 6686 break; 6687 } 6688 } 6689 } 6690 } 6691 6692 WRITE_ONCE(dev->threaded, threaded); 6693 6694 /* Make sure kthread is created before THREADED bit 6695 * is set. 6696 */ 6697 smp_mb__before_atomic(); 6698 6699 /* Setting/unsetting threaded mode on a napi might not immediately 6700 * take effect, if the current napi instance is actively being 6701 * polled. In this case, the switch between threaded mode and 6702 * softirq mode will happen in the next round of napi_schedule(). 6703 * This should not cause hiccups/stalls to the live traffic. 6704 */ 6705 list_for_each_entry(napi, &dev->napi_list, dev_list) 6706 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6707 6708 return err; 6709 } 6710 EXPORT_SYMBOL(dev_set_threaded); 6711 6712 /** 6713 * netif_queue_set_napi - Associate queue with the napi 6714 * @dev: device to which NAPI and queue belong 6715 * @queue_index: Index of queue 6716 * @type: queue type as RX or TX 6717 * @napi: NAPI context, pass NULL to clear previously set NAPI 6718 * 6719 * Set queue with its corresponding napi context. This should be done after 6720 * registering the NAPI handler for the queue-vector and the queues have been 6721 * mapped to the corresponding interrupt vector. 6722 */ 6723 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6724 enum netdev_queue_type type, struct napi_struct *napi) 6725 { 6726 struct netdev_rx_queue *rxq; 6727 struct netdev_queue *txq; 6728 6729 if (WARN_ON_ONCE(napi && !napi->dev)) 6730 return; 6731 if (dev->reg_state >= NETREG_REGISTERED) 6732 ASSERT_RTNL(); 6733 6734 switch (type) { 6735 case NETDEV_QUEUE_TYPE_RX: 6736 rxq = __netif_get_rx_queue(dev, queue_index); 6737 rxq->napi = napi; 6738 return; 6739 case NETDEV_QUEUE_TYPE_TX: 6740 txq = netdev_get_tx_queue(dev, queue_index); 6741 txq->napi = napi; 6742 return; 6743 default: 6744 return; 6745 } 6746 } 6747 EXPORT_SYMBOL(netif_queue_set_napi); 6748 6749 static void napi_restore_config(struct napi_struct *n) 6750 { 6751 n->defer_hard_irqs = n->config->defer_hard_irqs; 6752 n->gro_flush_timeout = n->config->gro_flush_timeout; 6753 n->irq_suspend_timeout = n->config->irq_suspend_timeout; 6754 /* a NAPI ID might be stored in the config, if so use it. if not, use 6755 * napi_hash_add to generate one for us. 6756 */ 6757 if (n->config->napi_id) { 6758 napi_hash_add_with_id(n, n->config->napi_id); 6759 } else { 6760 napi_hash_add(n); 6761 n->config->napi_id = n->napi_id; 6762 } 6763 } 6764 6765 static void napi_save_config(struct napi_struct *n) 6766 { 6767 n->config->defer_hard_irqs = n->defer_hard_irqs; 6768 n->config->gro_flush_timeout = n->gro_flush_timeout; 6769 n->config->irq_suspend_timeout = n->irq_suspend_timeout; 6770 napi_hash_del(n); 6771 } 6772 6773 /* Netlink wants the NAPI list to be sorted by ID, if adding a NAPI which will 6774 * inherit an existing ID try to insert it at the right position. 6775 */ 6776 static void 6777 netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) 6778 { 6779 unsigned int new_id, pos_id; 6780 struct list_head *higher; 6781 struct napi_struct *pos; 6782 6783 new_id = UINT_MAX; 6784 if (napi->config && napi->config->napi_id) 6785 new_id = napi->config->napi_id; 6786 6787 higher = &dev->napi_list; 6788 list_for_each_entry(pos, &dev->napi_list, dev_list) { 6789 if (pos->napi_id >= MIN_NAPI_ID) 6790 pos_id = pos->napi_id; 6791 else if (pos->config) 6792 pos_id = pos->config->napi_id; 6793 else 6794 pos_id = UINT_MAX; 6795 6796 if (pos_id <= new_id) 6797 break; 6798 higher = &pos->dev_list; 6799 } 6800 list_add_rcu(&napi->dev_list, higher); /* adds after higher */ 6801 } 6802 6803 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6804 int (*poll)(struct napi_struct *, int), int weight) 6805 { 6806 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6807 return; 6808 6809 INIT_LIST_HEAD(&napi->poll_list); 6810 INIT_HLIST_NODE(&napi->napi_hash_node); 6811 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6812 napi->timer.function = napi_watchdog; 6813 init_gro_hash(napi); 6814 napi->skb = NULL; 6815 INIT_LIST_HEAD(&napi->rx_list); 6816 napi->rx_count = 0; 6817 napi->poll = poll; 6818 if (weight > NAPI_POLL_WEIGHT) 6819 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6820 weight); 6821 napi->weight = weight; 6822 napi->dev = dev; 6823 #ifdef CONFIG_NETPOLL 6824 napi->poll_owner = -1; 6825 #endif 6826 napi->list_owner = -1; 6827 set_bit(NAPI_STATE_SCHED, &napi->state); 6828 set_bit(NAPI_STATE_NPSVC, &napi->state); 6829 netif_napi_dev_list_add(dev, napi); 6830 6831 /* default settings from sysfs are applied to all NAPIs. any per-NAPI 6832 * configuration will be loaded in napi_enable 6833 */ 6834 napi_set_defer_hard_irqs(napi, READ_ONCE(dev->napi_defer_hard_irqs)); 6835 napi_set_gro_flush_timeout(napi, READ_ONCE(dev->gro_flush_timeout)); 6836 6837 napi_get_frags_check(napi); 6838 /* Create kthread for this napi if dev->threaded is set. 6839 * Clear dev->threaded if kthread creation failed so that 6840 * threaded mode will not be enabled in napi_enable(). 6841 */ 6842 if (dev->threaded && napi_kthread_create(napi)) 6843 dev->threaded = false; 6844 netif_napi_set_irq(napi, -1); 6845 } 6846 EXPORT_SYMBOL(netif_napi_add_weight); 6847 6848 void napi_disable(struct napi_struct *n) 6849 { 6850 unsigned long val, new; 6851 6852 might_sleep(); 6853 set_bit(NAPI_STATE_DISABLE, &n->state); 6854 6855 val = READ_ONCE(n->state); 6856 do { 6857 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6858 usleep_range(20, 200); 6859 val = READ_ONCE(n->state); 6860 } 6861 6862 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6863 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6864 } while (!try_cmpxchg(&n->state, &val, new)); 6865 6866 hrtimer_cancel(&n->timer); 6867 6868 if (n->config) 6869 napi_save_config(n); 6870 else 6871 napi_hash_del(n); 6872 6873 clear_bit(NAPI_STATE_DISABLE, &n->state); 6874 } 6875 EXPORT_SYMBOL(napi_disable); 6876 6877 /** 6878 * napi_enable - enable NAPI scheduling 6879 * @n: NAPI context 6880 * 6881 * Resume NAPI from being scheduled on this context. 6882 * Must be paired with napi_disable. 6883 */ 6884 void napi_enable(struct napi_struct *n) 6885 { 6886 unsigned long new, val = READ_ONCE(n->state); 6887 6888 if (n->config) 6889 napi_restore_config(n); 6890 else 6891 napi_hash_add(n); 6892 6893 do { 6894 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6895 6896 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6897 if (n->dev->threaded && n->thread) 6898 new |= NAPIF_STATE_THREADED; 6899 } while (!try_cmpxchg(&n->state, &val, new)); 6900 } 6901 EXPORT_SYMBOL(napi_enable); 6902 6903 static void flush_gro_hash(struct napi_struct *napi) 6904 { 6905 int i; 6906 6907 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6908 struct sk_buff *skb, *n; 6909 6910 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6911 kfree_skb(skb); 6912 napi->gro_hash[i].count = 0; 6913 } 6914 } 6915 6916 /* Must be called in process context */ 6917 void __netif_napi_del(struct napi_struct *napi) 6918 { 6919 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6920 return; 6921 6922 if (napi->config) { 6923 napi->index = -1; 6924 napi->config = NULL; 6925 } 6926 6927 list_del_rcu(&napi->dev_list); 6928 napi_free_frags(napi); 6929 6930 flush_gro_hash(napi); 6931 napi->gro_bitmask = 0; 6932 6933 if (napi->thread) { 6934 kthread_stop(napi->thread); 6935 napi->thread = NULL; 6936 } 6937 } 6938 EXPORT_SYMBOL(__netif_napi_del); 6939 6940 static int __napi_poll(struct napi_struct *n, bool *repoll) 6941 { 6942 int work, weight; 6943 6944 weight = n->weight; 6945 6946 /* This NAPI_STATE_SCHED test is for avoiding a race 6947 * with netpoll's poll_napi(). Only the entity which 6948 * obtains the lock and sees NAPI_STATE_SCHED set will 6949 * actually make the ->poll() call. Therefore we avoid 6950 * accidentally calling ->poll() when NAPI is not scheduled. 6951 */ 6952 work = 0; 6953 if (napi_is_scheduled(n)) { 6954 work = n->poll(n, weight); 6955 trace_napi_poll(n, work, weight); 6956 6957 xdp_do_check_flushed(n); 6958 } 6959 6960 if (unlikely(work > weight)) 6961 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6962 n->poll, work, weight); 6963 6964 if (likely(work < weight)) 6965 return work; 6966 6967 /* Drivers must not modify the NAPI state if they 6968 * consume the entire weight. In such cases this code 6969 * still "owns" the NAPI instance and therefore can 6970 * move the instance around on the list at-will. 6971 */ 6972 if (unlikely(napi_disable_pending(n))) { 6973 napi_complete(n); 6974 return work; 6975 } 6976 6977 /* The NAPI context has more processing work, but busy-polling 6978 * is preferred. Exit early. 6979 */ 6980 if (napi_prefer_busy_poll(n)) { 6981 if (napi_complete_done(n, work)) { 6982 /* If timeout is not set, we need to make sure 6983 * that the NAPI is re-scheduled. 6984 */ 6985 napi_schedule(n); 6986 } 6987 return work; 6988 } 6989 6990 if (n->gro_bitmask) { 6991 /* flush too old packets 6992 * If HZ < 1000, flush all packets. 6993 */ 6994 napi_gro_flush(n, HZ >= 1000); 6995 } 6996 6997 gro_normal_list(n); 6998 6999 /* Some drivers may have called napi_schedule 7000 * prior to exhausting their budget. 7001 */ 7002 if (unlikely(!list_empty(&n->poll_list))) { 7003 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7004 n->dev ? n->dev->name : "backlog"); 7005 return work; 7006 } 7007 7008 *repoll = true; 7009 7010 return work; 7011 } 7012 7013 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7014 { 7015 bool do_repoll = false; 7016 void *have; 7017 int work; 7018 7019 list_del_init(&n->poll_list); 7020 7021 have = netpoll_poll_lock(n); 7022 7023 work = __napi_poll(n, &do_repoll); 7024 7025 if (do_repoll) 7026 list_add_tail(&n->poll_list, repoll); 7027 7028 netpoll_poll_unlock(have); 7029 7030 return work; 7031 } 7032 7033 static int napi_thread_wait(struct napi_struct *napi) 7034 { 7035 set_current_state(TASK_INTERRUPTIBLE); 7036 7037 while (!kthread_should_stop()) { 7038 /* Testing SCHED_THREADED bit here to make sure the current 7039 * kthread owns this napi and could poll on this napi. 7040 * Testing SCHED bit is not enough because SCHED bit might be 7041 * set by some other busy poll thread or by napi_disable(). 7042 */ 7043 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) { 7044 WARN_ON(!list_empty(&napi->poll_list)); 7045 __set_current_state(TASK_RUNNING); 7046 return 0; 7047 } 7048 7049 schedule(); 7050 set_current_state(TASK_INTERRUPTIBLE); 7051 } 7052 __set_current_state(TASK_RUNNING); 7053 7054 return -1; 7055 } 7056 7057 static void napi_threaded_poll_loop(struct napi_struct *napi) 7058 { 7059 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7060 struct softnet_data *sd; 7061 unsigned long last_qs = jiffies; 7062 7063 for (;;) { 7064 bool repoll = false; 7065 void *have; 7066 7067 local_bh_disable(); 7068 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7069 7070 sd = this_cpu_ptr(&softnet_data); 7071 sd->in_napi_threaded_poll = true; 7072 7073 have = netpoll_poll_lock(napi); 7074 __napi_poll(napi, &repoll); 7075 netpoll_poll_unlock(have); 7076 7077 sd->in_napi_threaded_poll = false; 7078 barrier(); 7079 7080 if (sd_has_rps_ipi_waiting(sd)) { 7081 local_irq_disable(); 7082 net_rps_action_and_irq_enable(sd); 7083 } 7084 skb_defer_free_flush(sd); 7085 bpf_net_ctx_clear(bpf_net_ctx); 7086 local_bh_enable(); 7087 7088 if (!repoll) 7089 break; 7090 7091 rcu_softirq_qs_periodic(last_qs); 7092 cond_resched(); 7093 } 7094 } 7095 7096 static int napi_threaded_poll(void *data) 7097 { 7098 struct napi_struct *napi = data; 7099 7100 while (!napi_thread_wait(napi)) 7101 napi_threaded_poll_loop(napi); 7102 7103 return 0; 7104 } 7105 7106 static __latent_entropy void net_rx_action(void) 7107 { 7108 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7109 unsigned long time_limit = jiffies + 7110 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs)); 7111 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx; 7112 int budget = READ_ONCE(net_hotdata.netdev_budget); 7113 LIST_HEAD(list); 7114 LIST_HEAD(repoll); 7115 7116 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx); 7117 start: 7118 sd->in_net_rx_action = true; 7119 local_irq_disable(); 7120 list_splice_init(&sd->poll_list, &list); 7121 local_irq_enable(); 7122 7123 for (;;) { 7124 struct napi_struct *n; 7125 7126 skb_defer_free_flush(sd); 7127 7128 if (list_empty(&list)) { 7129 if (list_empty(&repoll)) { 7130 sd->in_net_rx_action = false; 7131 barrier(); 7132 /* We need to check if ____napi_schedule() 7133 * had refilled poll_list while 7134 * sd->in_net_rx_action was true. 7135 */ 7136 if (!list_empty(&sd->poll_list)) 7137 goto start; 7138 if (!sd_has_rps_ipi_waiting(sd)) 7139 goto end; 7140 } 7141 break; 7142 } 7143 7144 n = list_first_entry(&list, struct napi_struct, poll_list); 7145 budget -= napi_poll(n, &repoll); 7146 7147 /* If softirq window is exhausted then punt. 7148 * Allow this to run for 2 jiffies since which will allow 7149 * an average latency of 1.5/HZ. 7150 */ 7151 if (unlikely(budget <= 0 || 7152 time_after_eq(jiffies, time_limit))) { 7153 sd->time_squeeze++; 7154 break; 7155 } 7156 } 7157 7158 local_irq_disable(); 7159 7160 list_splice_tail_init(&sd->poll_list, &list); 7161 list_splice_tail(&repoll, &list); 7162 list_splice(&list, &sd->poll_list); 7163 if (!list_empty(&sd->poll_list)) 7164 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7165 else 7166 sd->in_net_rx_action = false; 7167 7168 net_rps_action_and_irq_enable(sd); 7169 end: 7170 bpf_net_ctx_clear(bpf_net_ctx); 7171 } 7172 7173 struct netdev_adjacent { 7174 struct net_device *dev; 7175 netdevice_tracker dev_tracker; 7176 7177 /* upper master flag, there can only be one master device per list */ 7178 bool master; 7179 7180 /* lookup ignore flag */ 7181 bool ignore; 7182 7183 /* counter for the number of times this device was added to us */ 7184 u16 ref_nr; 7185 7186 /* private field for the users */ 7187 void *private; 7188 7189 struct list_head list; 7190 struct rcu_head rcu; 7191 }; 7192 7193 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7194 struct list_head *adj_list) 7195 { 7196 struct netdev_adjacent *adj; 7197 7198 list_for_each_entry(adj, adj_list, list) { 7199 if (adj->dev == adj_dev) 7200 return adj; 7201 } 7202 return NULL; 7203 } 7204 7205 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7206 struct netdev_nested_priv *priv) 7207 { 7208 struct net_device *dev = (struct net_device *)priv->data; 7209 7210 return upper_dev == dev; 7211 } 7212 7213 /** 7214 * netdev_has_upper_dev - Check if device is linked to an upper device 7215 * @dev: device 7216 * @upper_dev: upper device to check 7217 * 7218 * Find out if a device is linked to specified upper device and return true 7219 * in case it is. Note that this checks only immediate upper device, 7220 * not through a complete stack of devices. The caller must hold the RTNL lock. 7221 */ 7222 bool netdev_has_upper_dev(struct net_device *dev, 7223 struct net_device *upper_dev) 7224 { 7225 struct netdev_nested_priv priv = { 7226 .data = (void *)upper_dev, 7227 }; 7228 7229 ASSERT_RTNL(); 7230 7231 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7232 &priv); 7233 } 7234 EXPORT_SYMBOL(netdev_has_upper_dev); 7235 7236 /** 7237 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7238 * @dev: device 7239 * @upper_dev: upper device to check 7240 * 7241 * Find out if a device is linked to specified upper device and return true 7242 * in case it is. Note that this checks the entire upper device chain. 7243 * The caller must hold rcu lock. 7244 */ 7245 7246 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7247 struct net_device *upper_dev) 7248 { 7249 struct netdev_nested_priv priv = { 7250 .data = (void *)upper_dev, 7251 }; 7252 7253 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7254 &priv); 7255 } 7256 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7257 7258 /** 7259 * netdev_has_any_upper_dev - Check if device is linked to some device 7260 * @dev: device 7261 * 7262 * Find out if a device is linked to an upper device and return true in case 7263 * it is. The caller must hold the RTNL lock. 7264 */ 7265 bool netdev_has_any_upper_dev(struct net_device *dev) 7266 { 7267 ASSERT_RTNL(); 7268 7269 return !list_empty(&dev->adj_list.upper); 7270 } 7271 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7272 7273 /** 7274 * netdev_master_upper_dev_get - Get master upper device 7275 * @dev: device 7276 * 7277 * Find a master upper device and return pointer to it or NULL in case 7278 * it's not there. The caller must hold the RTNL lock. 7279 */ 7280 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7281 { 7282 struct netdev_adjacent *upper; 7283 7284 ASSERT_RTNL(); 7285 7286 if (list_empty(&dev->adj_list.upper)) 7287 return NULL; 7288 7289 upper = list_first_entry(&dev->adj_list.upper, 7290 struct netdev_adjacent, list); 7291 if (likely(upper->master)) 7292 return upper->dev; 7293 return NULL; 7294 } 7295 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7296 7297 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7298 { 7299 struct netdev_adjacent *upper; 7300 7301 ASSERT_RTNL(); 7302 7303 if (list_empty(&dev->adj_list.upper)) 7304 return NULL; 7305 7306 upper = list_first_entry(&dev->adj_list.upper, 7307 struct netdev_adjacent, list); 7308 if (likely(upper->master) && !upper->ignore) 7309 return upper->dev; 7310 return NULL; 7311 } 7312 7313 /** 7314 * netdev_has_any_lower_dev - Check if device is linked to some device 7315 * @dev: device 7316 * 7317 * Find out if a device is linked to a lower device and return true in case 7318 * it is. The caller must hold the RTNL lock. 7319 */ 7320 static bool netdev_has_any_lower_dev(struct net_device *dev) 7321 { 7322 ASSERT_RTNL(); 7323 7324 return !list_empty(&dev->adj_list.lower); 7325 } 7326 7327 void *netdev_adjacent_get_private(struct list_head *adj_list) 7328 { 7329 struct netdev_adjacent *adj; 7330 7331 adj = list_entry(adj_list, struct netdev_adjacent, list); 7332 7333 return adj->private; 7334 } 7335 EXPORT_SYMBOL(netdev_adjacent_get_private); 7336 7337 /** 7338 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7339 * @dev: device 7340 * @iter: list_head ** of the current position 7341 * 7342 * Gets the next device from the dev's upper list, starting from iter 7343 * position. The caller must hold RCU read lock. 7344 */ 7345 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7346 struct list_head **iter) 7347 { 7348 struct netdev_adjacent *upper; 7349 7350 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7351 7352 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7353 7354 if (&upper->list == &dev->adj_list.upper) 7355 return NULL; 7356 7357 *iter = &upper->list; 7358 7359 return upper->dev; 7360 } 7361 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7362 7363 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7364 struct list_head **iter, 7365 bool *ignore) 7366 { 7367 struct netdev_adjacent *upper; 7368 7369 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7370 7371 if (&upper->list == &dev->adj_list.upper) 7372 return NULL; 7373 7374 *iter = &upper->list; 7375 *ignore = upper->ignore; 7376 7377 return upper->dev; 7378 } 7379 7380 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7381 struct list_head **iter) 7382 { 7383 struct netdev_adjacent *upper; 7384 7385 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7386 7387 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7388 7389 if (&upper->list == &dev->adj_list.upper) 7390 return NULL; 7391 7392 *iter = &upper->list; 7393 7394 return upper->dev; 7395 } 7396 7397 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7398 int (*fn)(struct net_device *dev, 7399 struct netdev_nested_priv *priv), 7400 struct netdev_nested_priv *priv) 7401 { 7402 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7403 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7404 int ret, cur = 0; 7405 bool ignore; 7406 7407 now = dev; 7408 iter = &dev->adj_list.upper; 7409 7410 while (1) { 7411 if (now != dev) { 7412 ret = fn(now, priv); 7413 if (ret) 7414 return ret; 7415 } 7416 7417 next = NULL; 7418 while (1) { 7419 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7420 if (!udev) 7421 break; 7422 if (ignore) 7423 continue; 7424 7425 next = udev; 7426 niter = &udev->adj_list.upper; 7427 dev_stack[cur] = now; 7428 iter_stack[cur++] = iter; 7429 break; 7430 } 7431 7432 if (!next) { 7433 if (!cur) 7434 return 0; 7435 next = dev_stack[--cur]; 7436 niter = iter_stack[cur]; 7437 } 7438 7439 now = next; 7440 iter = niter; 7441 } 7442 7443 return 0; 7444 } 7445 7446 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7447 int (*fn)(struct net_device *dev, 7448 struct netdev_nested_priv *priv), 7449 struct netdev_nested_priv *priv) 7450 { 7451 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7452 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7453 int ret, cur = 0; 7454 7455 now = dev; 7456 iter = &dev->adj_list.upper; 7457 7458 while (1) { 7459 if (now != dev) { 7460 ret = fn(now, priv); 7461 if (ret) 7462 return ret; 7463 } 7464 7465 next = NULL; 7466 while (1) { 7467 udev = netdev_next_upper_dev_rcu(now, &iter); 7468 if (!udev) 7469 break; 7470 7471 next = udev; 7472 niter = &udev->adj_list.upper; 7473 dev_stack[cur] = now; 7474 iter_stack[cur++] = iter; 7475 break; 7476 } 7477 7478 if (!next) { 7479 if (!cur) 7480 return 0; 7481 next = dev_stack[--cur]; 7482 niter = iter_stack[cur]; 7483 } 7484 7485 now = next; 7486 iter = niter; 7487 } 7488 7489 return 0; 7490 } 7491 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7492 7493 static bool __netdev_has_upper_dev(struct net_device *dev, 7494 struct net_device *upper_dev) 7495 { 7496 struct netdev_nested_priv priv = { 7497 .flags = 0, 7498 .data = (void *)upper_dev, 7499 }; 7500 7501 ASSERT_RTNL(); 7502 7503 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7504 &priv); 7505 } 7506 7507 /** 7508 * netdev_lower_get_next_private - Get the next ->private from the 7509 * lower neighbour list 7510 * @dev: device 7511 * @iter: list_head ** of the current position 7512 * 7513 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7514 * list, starting from iter position. The caller must hold either hold the 7515 * RTNL lock or its own locking that guarantees that the neighbour lower 7516 * list will remain unchanged. 7517 */ 7518 void *netdev_lower_get_next_private(struct net_device *dev, 7519 struct list_head **iter) 7520 { 7521 struct netdev_adjacent *lower; 7522 7523 lower = list_entry(*iter, struct netdev_adjacent, list); 7524 7525 if (&lower->list == &dev->adj_list.lower) 7526 return NULL; 7527 7528 *iter = lower->list.next; 7529 7530 return lower->private; 7531 } 7532 EXPORT_SYMBOL(netdev_lower_get_next_private); 7533 7534 /** 7535 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7536 * lower neighbour list, RCU 7537 * variant 7538 * @dev: device 7539 * @iter: list_head ** of the current position 7540 * 7541 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7542 * list, starting from iter position. The caller must hold RCU read lock. 7543 */ 7544 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7545 struct list_head **iter) 7546 { 7547 struct netdev_adjacent *lower; 7548 7549 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7550 7551 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7552 7553 if (&lower->list == &dev->adj_list.lower) 7554 return NULL; 7555 7556 *iter = &lower->list; 7557 7558 return lower->private; 7559 } 7560 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7561 7562 /** 7563 * netdev_lower_get_next - Get the next device from the lower neighbour 7564 * list 7565 * @dev: device 7566 * @iter: list_head ** of the current position 7567 * 7568 * Gets the next netdev_adjacent from the dev's lower neighbour 7569 * list, starting from iter position. The caller must hold RTNL lock or 7570 * its own locking that guarantees that the neighbour lower 7571 * list will remain unchanged. 7572 */ 7573 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7574 { 7575 struct netdev_adjacent *lower; 7576 7577 lower = list_entry(*iter, struct netdev_adjacent, list); 7578 7579 if (&lower->list == &dev->adj_list.lower) 7580 return NULL; 7581 7582 *iter = lower->list.next; 7583 7584 return lower->dev; 7585 } 7586 EXPORT_SYMBOL(netdev_lower_get_next); 7587 7588 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7589 struct list_head **iter) 7590 { 7591 struct netdev_adjacent *lower; 7592 7593 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7594 7595 if (&lower->list == &dev->adj_list.lower) 7596 return NULL; 7597 7598 *iter = &lower->list; 7599 7600 return lower->dev; 7601 } 7602 7603 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7604 struct list_head **iter, 7605 bool *ignore) 7606 { 7607 struct netdev_adjacent *lower; 7608 7609 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7610 7611 if (&lower->list == &dev->adj_list.lower) 7612 return NULL; 7613 7614 *iter = &lower->list; 7615 *ignore = lower->ignore; 7616 7617 return lower->dev; 7618 } 7619 7620 int netdev_walk_all_lower_dev(struct net_device *dev, 7621 int (*fn)(struct net_device *dev, 7622 struct netdev_nested_priv *priv), 7623 struct netdev_nested_priv *priv) 7624 { 7625 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7626 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7627 int ret, cur = 0; 7628 7629 now = dev; 7630 iter = &dev->adj_list.lower; 7631 7632 while (1) { 7633 if (now != dev) { 7634 ret = fn(now, priv); 7635 if (ret) 7636 return ret; 7637 } 7638 7639 next = NULL; 7640 while (1) { 7641 ldev = netdev_next_lower_dev(now, &iter); 7642 if (!ldev) 7643 break; 7644 7645 next = ldev; 7646 niter = &ldev->adj_list.lower; 7647 dev_stack[cur] = now; 7648 iter_stack[cur++] = iter; 7649 break; 7650 } 7651 7652 if (!next) { 7653 if (!cur) 7654 return 0; 7655 next = dev_stack[--cur]; 7656 niter = iter_stack[cur]; 7657 } 7658 7659 now = next; 7660 iter = niter; 7661 } 7662 7663 return 0; 7664 } 7665 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7666 7667 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7668 int (*fn)(struct net_device *dev, 7669 struct netdev_nested_priv *priv), 7670 struct netdev_nested_priv *priv) 7671 { 7672 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7673 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7674 int ret, cur = 0; 7675 bool ignore; 7676 7677 now = dev; 7678 iter = &dev->adj_list.lower; 7679 7680 while (1) { 7681 if (now != dev) { 7682 ret = fn(now, priv); 7683 if (ret) 7684 return ret; 7685 } 7686 7687 next = NULL; 7688 while (1) { 7689 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7690 if (!ldev) 7691 break; 7692 if (ignore) 7693 continue; 7694 7695 next = ldev; 7696 niter = &ldev->adj_list.lower; 7697 dev_stack[cur] = now; 7698 iter_stack[cur++] = iter; 7699 break; 7700 } 7701 7702 if (!next) { 7703 if (!cur) 7704 return 0; 7705 next = dev_stack[--cur]; 7706 niter = iter_stack[cur]; 7707 } 7708 7709 now = next; 7710 iter = niter; 7711 } 7712 7713 return 0; 7714 } 7715 7716 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7717 struct list_head **iter) 7718 { 7719 struct netdev_adjacent *lower; 7720 7721 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7722 if (&lower->list == &dev->adj_list.lower) 7723 return NULL; 7724 7725 *iter = &lower->list; 7726 7727 return lower->dev; 7728 } 7729 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7730 7731 static u8 __netdev_upper_depth(struct net_device *dev) 7732 { 7733 struct net_device *udev; 7734 struct list_head *iter; 7735 u8 max_depth = 0; 7736 bool ignore; 7737 7738 for (iter = &dev->adj_list.upper, 7739 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7740 udev; 7741 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7742 if (ignore) 7743 continue; 7744 if (max_depth < udev->upper_level) 7745 max_depth = udev->upper_level; 7746 } 7747 7748 return max_depth; 7749 } 7750 7751 static u8 __netdev_lower_depth(struct net_device *dev) 7752 { 7753 struct net_device *ldev; 7754 struct list_head *iter; 7755 u8 max_depth = 0; 7756 bool ignore; 7757 7758 for (iter = &dev->adj_list.lower, 7759 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7760 ldev; 7761 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7762 if (ignore) 7763 continue; 7764 if (max_depth < ldev->lower_level) 7765 max_depth = ldev->lower_level; 7766 } 7767 7768 return max_depth; 7769 } 7770 7771 static int __netdev_update_upper_level(struct net_device *dev, 7772 struct netdev_nested_priv *__unused) 7773 { 7774 dev->upper_level = __netdev_upper_depth(dev) + 1; 7775 return 0; 7776 } 7777 7778 #ifdef CONFIG_LOCKDEP 7779 static LIST_HEAD(net_unlink_list); 7780 7781 static void net_unlink_todo(struct net_device *dev) 7782 { 7783 if (list_empty(&dev->unlink_list)) 7784 list_add_tail(&dev->unlink_list, &net_unlink_list); 7785 } 7786 #endif 7787 7788 static int __netdev_update_lower_level(struct net_device *dev, 7789 struct netdev_nested_priv *priv) 7790 { 7791 dev->lower_level = __netdev_lower_depth(dev) + 1; 7792 7793 #ifdef CONFIG_LOCKDEP 7794 if (!priv) 7795 return 0; 7796 7797 if (priv->flags & NESTED_SYNC_IMM) 7798 dev->nested_level = dev->lower_level - 1; 7799 if (priv->flags & NESTED_SYNC_TODO) 7800 net_unlink_todo(dev); 7801 #endif 7802 return 0; 7803 } 7804 7805 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7806 int (*fn)(struct net_device *dev, 7807 struct netdev_nested_priv *priv), 7808 struct netdev_nested_priv *priv) 7809 { 7810 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7811 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7812 int ret, cur = 0; 7813 7814 now = dev; 7815 iter = &dev->adj_list.lower; 7816 7817 while (1) { 7818 if (now != dev) { 7819 ret = fn(now, priv); 7820 if (ret) 7821 return ret; 7822 } 7823 7824 next = NULL; 7825 while (1) { 7826 ldev = netdev_next_lower_dev_rcu(now, &iter); 7827 if (!ldev) 7828 break; 7829 7830 next = ldev; 7831 niter = &ldev->adj_list.lower; 7832 dev_stack[cur] = now; 7833 iter_stack[cur++] = iter; 7834 break; 7835 } 7836 7837 if (!next) { 7838 if (!cur) 7839 return 0; 7840 next = dev_stack[--cur]; 7841 niter = iter_stack[cur]; 7842 } 7843 7844 now = next; 7845 iter = niter; 7846 } 7847 7848 return 0; 7849 } 7850 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7851 7852 /** 7853 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7854 * lower neighbour list, RCU 7855 * variant 7856 * @dev: device 7857 * 7858 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7859 * list. The caller must hold RCU read lock. 7860 */ 7861 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7862 { 7863 struct netdev_adjacent *lower; 7864 7865 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7866 struct netdev_adjacent, list); 7867 if (lower) 7868 return lower->private; 7869 return NULL; 7870 } 7871 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7872 7873 /** 7874 * netdev_master_upper_dev_get_rcu - Get master upper device 7875 * @dev: device 7876 * 7877 * Find a master upper device and return pointer to it or NULL in case 7878 * it's not there. The caller must hold the RCU read lock. 7879 */ 7880 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7881 { 7882 struct netdev_adjacent *upper; 7883 7884 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7885 struct netdev_adjacent, list); 7886 if (upper && likely(upper->master)) 7887 return upper->dev; 7888 return NULL; 7889 } 7890 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7891 7892 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7893 struct net_device *adj_dev, 7894 struct list_head *dev_list) 7895 { 7896 char linkname[IFNAMSIZ+7]; 7897 7898 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7899 "upper_%s" : "lower_%s", adj_dev->name); 7900 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7901 linkname); 7902 } 7903 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7904 char *name, 7905 struct list_head *dev_list) 7906 { 7907 char linkname[IFNAMSIZ+7]; 7908 7909 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7910 "upper_%s" : "lower_%s", name); 7911 sysfs_remove_link(&(dev->dev.kobj), linkname); 7912 } 7913 7914 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7915 struct net_device *adj_dev, 7916 struct list_head *dev_list) 7917 { 7918 return (dev_list == &dev->adj_list.upper || 7919 dev_list == &dev->adj_list.lower) && 7920 net_eq(dev_net(dev), dev_net(adj_dev)); 7921 } 7922 7923 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7924 struct net_device *adj_dev, 7925 struct list_head *dev_list, 7926 void *private, bool master) 7927 { 7928 struct netdev_adjacent *adj; 7929 int ret; 7930 7931 adj = __netdev_find_adj(adj_dev, dev_list); 7932 7933 if (adj) { 7934 adj->ref_nr += 1; 7935 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7936 dev->name, adj_dev->name, adj->ref_nr); 7937 7938 return 0; 7939 } 7940 7941 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7942 if (!adj) 7943 return -ENOMEM; 7944 7945 adj->dev = adj_dev; 7946 adj->master = master; 7947 adj->ref_nr = 1; 7948 adj->private = private; 7949 adj->ignore = false; 7950 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7951 7952 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7953 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7954 7955 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7956 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7957 if (ret) 7958 goto free_adj; 7959 } 7960 7961 /* Ensure that master link is always the first item in list. */ 7962 if (master) { 7963 ret = sysfs_create_link(&(dev->dev.kobj), 7964 &(adj_dev->dev.kobj), "master"); 7965 if (ret) 7966 goto remove_symlinks; 7967 7968 list_add_rcu(&adj->list, dev_list); 7969 } else { 7970 list_add_tail_rcu(&adj->list, dev_list); 7971 } 7972 7973 return 0; 7974 7975 remove_symlinks: 7976 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7977 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7978 free_adj: 7979 netdev_put(adj_dev, &adj->dev_tracker); 7980 kfree(adj); 7981 7982 return ret; 7983 } 7984 7985 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7986 struct net_device *adj_dev, 7987 u16 ref_nr, 7988 struct list_head *dev_list) 7989 { 7990 struct netdev_adjacent *adj; 7991 7992 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7993 dev->name, adj_dev->name, ref_nr); 7994 7995 adj = __netdev_find_adj(adj_dev, dev_list); 7996 7997 if (!adj) { 7998 pr_err("Adjacency does not exist for device %s from %s\n", 7999 dev->name, adj_dev->name); 8000 WARN_ON(1); 8001 return; 8002 } 8003 8004 if (adj->ref_nr > ref_nr) { 8005 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8006 dev->name, adj_dev->name, ref_nr, 8007 adj->ref_nr - ref_nr); 8008 adj->ref_nr -= ref_nr; 8009 return; 8010 } 8011 8012 if (adj->master) 8013 sysfs_remove_link(&(dev->dev.kobj), "master"); 8014 8015 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8016 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8017 8018 list_del_rcu(&adj->list); 8019 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8020 adj_dev->name, dev->name, adj_dev->name); 8021 netdev_put(adj_dev, &adj->dev_tracker); 8022 kfree_rcu(adj, rcu); 8023 } 8024 8025 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8026 struct net_device *upper_dev, 8027 struct list_head *up_list, 8028 struct list_head *down_list, 8029 void *private, bool master) 8030 { 8031 int ret; 8032 8033 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8034 private, master); 8035 if (ret) 8036 return ret; 8037 8038 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8039 private, false); 8040 if (ret) { 8041 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8042 return ret; 8043 } 8044 8045 return 0; 8046 } 8047 8048 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8049 struct net_device *upper_dev, 8050 u16 ref_nr, 8051 struct list_head *up_list, 8052 struct list_head *down_list) 8053 { 8054 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8055 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8056 } 8057 8058 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8059 struct net_device *upper_dev, 8060 void *private, bool master) 8061 { 8062 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8063 &dev->adj_list.upper, 8064 &upper_dev->adj_list.lower, 8065 private, master); 8066 } 8067 8068 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8069 struct net_device *upper_dev) 8070 { 8071 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8072 &dev->adj_list.upper, 8073 &upper_dev->adj_list.lower); 8074 } 8075 8076 static int __netdev_upper_dev_link(struct net_device *dev, 8077 struct net_device *upper_dev, bool master, 8078 void *upper_priv, void *upper_info, 8079 struct netdev_nested_priv *priv, 8080 struct netlink_ext_ack *extack) 8081 { 8082 struct netdev_notifier_changeupper_info changeupper_info = { 8083 .info = { 8084 .dev = dev, 8085 .extack = extack, 8086 }, 8087 .upper_dev = upper_dev, 8088 .master = master, 8089 .linking = true, 8090 .upper_info = upper_info, 8091 }; 8092 struct net_device *master_dev; 8093 int ret = 0; 8094 8095 ASSERT_RTNL(); 8096 8097 if (dev == upper_dev) 8098 return -EBUSY; 8099 8100 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8101 if (__netdev_has_upper_dev(upper_dev, dev)) 8102 return -EBUSY; 8103 8104 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8105 return -EMLINK; 8106 8107 if (!master) { 8108 if (__netdev_has_upper_dev(dev, upper_dev)) 8109 return -EEXIST; 8110 } else { 8111 master_dev = __netdev_master_upper_dev_get(dev); 8112 if (master_dev) 8113 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8114 } 8115 8116 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8117 &changeupper_info.info); 8118 ret = notifier_to_errno(ret); 8119 if (ret) 8120 return ret; 8121 8122 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8123 master); 8124 if (ret) 8125 return ret; 8126 8127 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8128 &changeupper_info.info); 8129 ret = notifier_to_errno(ret); 8130 if (ret) 8131 goto rollback; 8132 8133 __netdev_update_upper_level(dev, NULL); 8134 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8135 8136 __netdev_update_lower_level(upper_dev, priv); 8137 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8138 priv); 8139 8140 return 0; 8141 8142 rollback: 8143 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8144 8145 return ret; 8146 } 8147 8148 /** 8149 * netdev_upper_dev_link - Add a link to the upper device 8150 * @dev: device 8151 * @upper_dev: new upper device 8152 * @extack: netlink extended ack 8153 * 8154 * Adds a link to device which is upper to this one. The caller must hold 8155 * the RTNL lock. On a failure a negative errno code is returned. 8156 * On success the reference counts are adjusted and the function 8157 * returns zero. 8158 */ 8159 int netdev_upper_dev_link(struct net_device *dev, 8160 struct net_device *upper_dev, 8161 struct netlink_ext_ack *extack) 8162 { 8163 struct netdev_nested_priv priv = { 8164 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8165 .data = NULL, 8166 }; 8167 8168 return __netdev_upper_dev_link(dev, upper_dev, false, 8169 NULL, NULL, &priv, extack); 8170 } 8171 EXPORT_SYMBOL(netdev_upper_dev_link); 8172 8173 /** 8174 * netdev_master_upper_dev_link - Add a master link to the upper device 8175 * @dev: device 8176 * @upper_dev: new upper device 8177 * @upper_priv: upper device private 8178 * @upper_info: upper info to be passed down via notifier 8179 * @extack: netlink extended ack 8180 * 8181 * Adds a link to device which is upper to this one. In this case, only 8182 * one master upper device can be linked, although other non-master devices 8183 * might be linked as well. The caller must hold the RTNL lock. 8184 * On a failure a negative errno code is returned. On success the reference 8185 * counts are adjusted and the function returns zero. 8186 */ 8187 int netdev_master_upper_dev_link(struct net_device *dev, 8188 struct net_device *upper_dev, 8189 void *upper_priv, void *upper_info, 8190 struct netlink_ext_ack *extack) 8191 { 8192 struct netdev_nested_priv priv = { 8193 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8194 .data = NULL, 8195 }; 8196 8197 return __netdev_upper_dev_link(dev, upper_dev, true, 8198 upper_priv, upper_info, &priv, extack); 8199 } 8200 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8201 8202 static void __netdev_upper_dev_unlink(struct net_device *dev, 8203 struct net_device *upper_dev, 8204 struct netdev_nested_priv *priv) 8205 { 8206 struct netdev_notifier_changeupper_info changeupper_info = { 8207 .info = { 8208 .dev = dev, 8209 }, 8210 .upper_dev = upper_dev, 8211 .linking = false, 8212 }; 8213 8214 ASSERT_RTNL(); 8215 8216 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8217 8218 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8219 &changeupper_info.info); 8220 8221 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8222 8223 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8224 &changeupper_info.info); 8225 8226 __netdev_update_upper_level(dev, NULL); 8227 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8228 8229 __netdev_update_lower_level(upper_dev, priv); 8230 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8231 priv); 8232 } 8233 8234 /** 8235 * netdev_upper_dev_unlink - Removes a link to upper device 8236 * @dev: device 8237 * @upper_dev: new upper device 8238 * 8239 * Removes a link to device which is upper to this one. The caller must hold 8240 * the RTNL lock. 8241 */ 8242 void netdev_upper_dev_unlink(struct net_device *dev, 8243 struct net_device *upper_dev) 8244 { 8245 struct netdev_nested_priv priv = { 8246 .flags = NESTED_SYNC_TODO, 8247 .data = NULL, 8248 }; 8249 8250 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8251 } 8252 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8253 8254 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8255 struct net_device *lower_dev, 8256 bool val) 8257 { 8258 struct netdev_adjacent *adj; 8259 8260 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8261 if (adj) 8262 adj->ignore = val; 8263 8264 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8265 if (adj) 8266 adj->ignore = val; 8267 } 8268 8269 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8270 struct net_device *lower_dev) 8271 { 8272 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8273 } 8274 8275 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8276 struct net_device *lower_dev) 8277 { 8278 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8279 } 8280 8281 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8282 struct net_device *new_dev, 8283 struct net_device *dev, 8284 struct netlink_ext_ack *extack) 8285 { 8286 struct netdev_nested_priv priv = { 8287 .flags = 0, 8288 .data = NULL, 8289 }; 8290 int err; 8291 8292 if (!new_dev) 8293 return 0; 8294 8295 if (old_dev && new_dev != old_dev) 8296 netdev_adjacent_dev_disable(dev, old_dev); 8297 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8298 extack); 8299 if (err) { 8300 if (old_dev && new_dev != old_dev) 8301 netdev_adjacent_dev_enable(dev, old_dev); 8302 return err; 8303 } 8304 8305 return 0; 8306 } 8307 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8308 8309 void netdev_adjacent_change_commit(struct net_device *old_dev, 8310 struct net_device *new_dev, 8311 struct net_device *dev) 8312 { 8313 struct netdev_nested_priv priv = { 8314 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8315 .data = NULL, 8316 }; 8317 8318 if (!new_dev || !old_dev) 8319 return; 8320 8321 if (new_dev == old_dev) 8322 return; 8323 8324 netdev_adjacent_dev_enable(dev, old_dev); 8325 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8326 } 8327 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8328 8329 void netdev_adjacent_change_abort(struct net_device *old_dev, 8330 struct net_device *new_dev, 8331 struct net_device *dev) 8332 { 8333 struct netdev_nested_priv priv = { 8334 .flags = 0, 8335 .data = NULL, 8336 }; 8337 8338 if (!new_dev) 8339 return; 8340 8341 if (old_dev && new_dev != old_dev) 8342 netdev_adjacent_dev_enable(dev, old_dev); 8343 8344 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8345 } 8346 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8347 8348 /** 8349 * netdev_bonding_info_change - Dispatch event about slave change 8350 * @dev: device 8351 * @bonding_info: info to dispatch 8352 * 8353 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8354 * The caller must hold the RTNL lock. 8355 */ 8356 void netdev_bonding_info_change(struct net_device *dev, 8357 struct netdev_bonding_info *bonding_info) 8358 { 8359 struct netdev_notifier_bonding_info info = { 8360 .info.dev = dev, 8361 }; 8362 8363 memcpy(&info.bonding_info, bonding_info, 8364 sizeof(struct netdev_bonding_info)); 8365 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8366 &info.info); 8367 } 8368 EXPORT_SYMBOL(netdev_bonding_info_change); 8369 8370 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8371 struct netlink_ext_ack *extack) 8372 { 8373 struct netdev_notifier_offload_xstats_info info = { 8374 .info.dev = dev, 8375 .info.extack = extack, 8376 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8377 }; 8378 int err; 8379 int rc; 8380 8381 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8382 GFP_KERNEL); 8383 if (!dev->offload_xstats_l3) 8384 return -ENOMEM; 8385 8386 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8387 NETDEV_OFFLOAD_XSTATS_DISABLE, 8388 &info.info); 8389 err = notifier_to_errno(rc); 8390 if (err) 8391 goto free_stats; 8392 8393 return 0; 8394 8395 free_stats: 8396 kfree(dev->offload_xstats_l3); 8397 dev->offload_xstats_l3 = NULL; 8398 return err; 8399 } 8400 8401 int netdev_offload_xstats_enable(struct net_device *dev, 8402 enum netdev_offload_xstats_type type, 8403 struct netlink_ext_ack *extack) 8404 { 8405 ASSERT_RTNL(); 8406 8407 if (netdev_offload_xstats_enabled(dev, type)) 8408 return -EALREADY; 8409 8410 switch (type) { 8411 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8412 return netdev_offload_xstats_enable_l3(dev, extack); 8413 } 8414 8415 WARN_ON(1); 8416 return -EINVAL; 8417 } 8418 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8419 8420 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8421 { 8422 struct netdev_notifier_offload_xstats_info info = { 8423 .info.dev = dev, 8424 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8425 }; 8426 8427 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8428 &info.info); 8429 kfree(dev->offload_xstats_l3); 8430 dev->offload_xstats_l3 = NULL; 8431 } 8432 8433 int netdev_offload_xstats_disable(struct net_device *dev, 8434 enum netdev_offload_xstats_type type) 8435 { 8436 ASSERT_RTNL(); 8437 8438 if (!netdev_offload_xstats_enabled(dev, type)) 8439 return -EALREADY; 8440 8441 switch (type) { 8442 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8443 netdev_offload_xstats_disable_l3(dev); 8444 return 0; 8445 } 8446 8447 WARN_ON(1); 8448 return -EINVAL; 8449 } 8450 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8451 8452 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8453 { 8454 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8455 } 8456 8457 static struct rtnl_hw_stats64 * 8458 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8459 enum netdev_offload_xstats_type type) 8460 { 8461 switch (type) { 8462 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8463 return dev->offload_xstats_l3; 8464 } 8465 8466 WARN_ON(1); 8467 return NULL; 8468 } 8469 8470 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8471 enum netdev_offload_xstats_type type) 8472 { 8473 ASSERT_RTNL(); 8474 8475 return netdev_offload_xstats_get_ptr(dev, type); 8476 } 8477 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8478 8479 struct netdev_notifier_offload_xstats_ru { 8480 bool used; 8481 }; 8482 8483 struct netdev_notifier_offload_xstats_rd { 8484 struct rtnl_hw_stats64 stats; 8485 bool used; 8486 }; 8487 8488 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8489 const struct rtnl_hw_stats64 *src) 8490 { 8491 dest->rx_packets += src->rx_packets; 8492 dest->tx_packets += src->tx_packets; 8493 dest->rx_bytes += src->rx_bytes; 8494 dest->tx_bytes += src->tx_bytes; 8495 dest->rx_errors += src->rx_errors; 8496 dest->tx_errors += src->tx_errors; 8497 dest->rx_dropped += src->rx_dropped; 8498 dest->tx_dropped += src->tx_dropped; 8499 dest->multicast += src->multicast; 8500 } 8501 8502 static int netdev_offload_xstats_get_used(struct net_device *dev, 8503 enum netdev_offload_xstats_type type, 8504 bool *p_used, 8505 struct netlink_ext_ack *extack) 8506 { 8507 struct netdev_notifier_offload_xstats_ru report_used = {}; 8508 struct netdev_notifier_offload_xstats_info info = { 8509 .info.dev = dev, 8510 .info.extack = extack, 8511 .type = type, 8512 .report_used = &report_used, 8513 }; 8514 int rc; 8515 8516 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8517 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8518 &info.info); 8519 *p_used = report_used.used; 8520 return notifier_to_errno(rc); 8521 } 8522 8523 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8524 enum netdev_offload_xstats_type type, 8525 struct rtnl_hw_stats64 *p_stats, 8526 bool *p_used, 8527 struct netlink_ext_ack *extack) 8528 { 8529 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8530 struct netdev_notifier_offload_xstats_info info = { 8531 .info.dev = dev, 8532 .info.extack = extack, 8533 .type = type, 8534 .report_delta = &report_delta, 8535 }; 8536 struct rtnl_hw_stats64 *stats; 8537 int rc; 8538 8539 stats = netdev_offload_xstats_get_ptr(dev, type); 8540 if (WARN_ON(!stats)) 8541 return -EINVAL; 8542 8543 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8544 &info.info); 8545 8546 /* Cache whatever we got, even if there was an error, otherwise the 8547 * successful stats retrievals would get lost. 8548 */ 8549 netdev_hw_stats64_add(stats, &report_delta.stats); 8550 8551 if (p_stats) 8552 *p_stats = *stats; 8553 *p_used = report_delta.used; 8554 8555 return notifier_to_errno(rc); 8556 } 8557 8558 int netdev_offload_xstats_get(struct net_device *dev, 8559 enum netdev_offload_xstats_type type, 8560 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8561 struct netlink_ext_ack *extack) 8562 { 8563 ASSERT_RTNL(); 8564 8565 if (p_stats) 8566 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8567 p_used, extack); 8568 else 8569 return netdev_offload_xstats_get_used(dev, type, p_used, 8570 extack); 8571 } 8572 EXPORT_SYMBOL(netdev_offload_xstats_get); 8573 8574 void 8575 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8576 const struct rtnl_hw_stats64 *stats) 8577 { 8578 report_delta->used = true; 8579 netdev_hw_stats64_add(&report_delta->stats, stats); 8580 } 8581 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8582 8583 void 8584 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8585 { 8586 report_used->used = true; 8587 } 8588 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8589 8590 void netdev_offload_xstats_push_delta(struct net_device *dev, 8591 enum netdev_offload_xstats_type type, 8592 const struct rtnl_hw_stats64 *p_stats) 8593 { 8594 struct rtnl_hw_stats64 *stats; 8595 8596 ASSERT_RTNL(); 8597 8598 stats = netdev_offload_xstats_get_ptr(dev, type); 8599 if (WARN_ON(!stats)) 8600 return; 8601 8602 netdev_hw_stats64_add(stats, p_stats); 8603 } 8604 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8605 8606 /** 8607 * netdev_get_xmit_slave - Get the xmit slave of master device 8608 * @dev: device 8609 * @skb: The packet 8610 * @all_slaves: assume all the slaves are active 8611 * 8612 * The reference counters are not incremented so the caller must be 8613 * careful with locks. The caller must hold RCU lock. 8614 * %NULL is returned if no slave is found. 8615 */ 8616 8617 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8618 struct sk_buff *skb, 8619 bool all_slaves) 8620 { 8621 const struct net_device_ops *ops = dev->netdev_ops; 8622 8623 if (!ops->ndo_get_xmit_slave) 8624 return NULL; 8625 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8626 } 8627 EXPORT_SYMBOL(netdev_get_xmit_slave); 8628 8629 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8630 struct sock *sk) 8631 { 8632 const struct net_device_ops *ops = dev->netdev_ops; 8633 8634 if (!ops->ndo_sk_get_lower_dev) 8635 return NULL; 8636 return ops->ndo_sk_get_lower_dev(dev, sk); 8637 } 8638 8639 /** 8640 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8641 * @dev: device 8642 * @sk: the socket 8643 * 8644 * %NULL is returned if no lower device is found. 8645 */ 8646 8647 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8648 struct sock *sk) 8649 { 8650 struct net_device *lower; 8651 8652 lower = netdev_sk_get_lower_dev(dev, sk); 8653 while (lower) { 8654 dev = lower; 8655 lower = netdev_sk_get_lower_dev(dev, sk); 8656 } 8657 8658 return dev; 8659 } 8660 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8661 8662 static void netdev_adjacent_add_links(struct net_device *dev) 8663 { 8664 struct netdev_adjacent *iter; 8665 8666 struct net *net = dev_net(dev); 8667 8668 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8669 if (!net_eq(net, dev_net(iter->dev))) 8670 continue; 8671 netdev_adjacent_sysfs_add(iter->dev, dev, 8672 &iter->dev->adj_list.lower); 8673 netdev_adjacent_sysfs_add(dev, iter->dev, 8674 &dev->adj_list.upper); 8675 } 8676 8677 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8678 if (!net_eq(net, dev_net(iter->dev))) 8679 continue; 8680 netdev_adjacent_sysfs_add(iter->dev, dev, 8681 &iter->dev->adj_list.upper); 8682 netdev_adjacent_sysfs_add(dev, iter->dev, 8683 &dev->adj_list.lower); 8684 } 8685 } 8686 8687 static void netdev_adjacent_del_links(struct net_device *dev) 8688 { 8689 struct netdev_adjacent *iter; 8690 8691 struct net *net = dev_net(dev); 8692 8693 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8694 if (!net_eq(net, dev_net(iter->dev))) 8695 continue; 8696 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8697 &iter->dev->adj_list.lower); 8698 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8699 &dev->adj_list.upper); 8700 } 8701 8702 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8703 if (!net_eq(net, dev_net(iter->dev))) 8704 continue; 8705 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8706 &iter->dev->adj_list.upper); 8707 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8708 &dev->adj_list.lower); 8709 } 8710 } 8711 8712 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8713 { 8714 struct netdev_adjacent *iter; 8715 8716 struct net *net = dev_net(dev); 8717 8718 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8719 if (!net_eq(net, dev_net(iter->dev))) 8720 continue; 8721 netdev_adjacent_sysfs_del(iter->dev, oldname, 8722 &iter->dev->adj_list.lower); 8723 netdev_adjacent_sysfs_add(iter->dev, dev, 8724 &iter->dev->adj_list.lower); 8725 } 8726 8727 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8728 if (!net_eq(net, dev_net(iter->dev))) 8729 continue; 8730 netdev_adjacent_sysfs_del(iter->dev, oldname, 8731 &iter->dev->adj_list.upper); 8732 netdev_adjacent_sysfs_add(iter->dev, dev, 8733 &iter->dev->adj_list.upper); 8734 } 8735 } 8736 8737 void *netdev_lower_dev_get_private(struct net_device *dev, 8738 struct net_device *lower_dev) 8739 { 8740 struct netdev_adjacent *lower; 8741 8742 if (!lower_dev) 8743 return NULL; 8744 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8745 if (!lower) 8746 return NULL; 8747 8748 return lower->private; 8749 } 8750 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8751 8752 8753 /** 8754 * netdev_lower_state_changed - Dispatch event about lower device state change 8755 * @lower_dev: device 8756 * @lower_state_info: state to dispatch 8757 * 8758 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8759 * The caller must hold the RTNL lock. 8760 */ 8761 void netdev_lower_state_changed(struct net_device *lower_dev, 8762 void *lower_state_info) 8763 { 8764 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8765 .info.dev = lower_dev, 8766 }; 8767 8768 ASSERT_RTNL(); 8769 changelowerstate_info.lower_state_info = lower_state_info; 8770 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8771 &changelowerstate_info.info); 8772 } 8773 EXPORT_SYMBOL(netdev_lower_state_changed); 8774 8775 static void dev_change_rx_flags(struct net_device *dev, int flags) 8776 { 8777 const struct net_device_ops *ops = dev->netdev_ops; 8778 8779 if (ops->ndo_change_rx_flags) 8780 ops->ndo_change_rx_flags(dev, flags); 8781 } 8782 8783 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8784 { 8785 unsigned int old_flags = dev->flags; 8786 unsigned int promiscuity, flags; 8787 kuid_t uid; 8788 kgid_t gid; 8789 8790 ASSERT_RTNL(); 8791 8792 promiscuity = dev->promiscuity + inc; 8793 if (promiscuity == 0) { 8794 /* 8795 * Avoid overflow. 8796 * If inc causes overflow, untouch promisc and return error. 8797 */ 8798 if (unlikely(inc > 0)) { 8799 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8800 return -EOVERFLOW; 8801 } 8802 flags = old_flags & ~IFF_PROMISC; 8803 } else { 8804 flags = old_flags | IFF_PROMISC; 8805 } 8806 WRITE_ONCE(dev->promiscuity, promiscuity); 8807 if (flags != old_flags) { 8808 WRITE_ONCE(dev->flags, flags); 8809 netdev_info(dev, "%s promiscuous mode\n", 8810 dev->flags & IFF_PROMISC ? "entered" : "left"); 8811 if (audit_enabled) { 8812 current_uid_gid(&uid, &gid); 8813 audit_log(audit_context(), GFP_ATOMIC, 8814 AUDIT_ANOM_PROMISCUOUS, 8815 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8816 dev->name, (dev->flags & IFF_PROMISC), 8817 (old_flags & IFF_PROMISC), 8818 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8819 from_kuid(&init_user_ns, uid), 8820 from_kgid(&init_user_ns, gid), 8821 audit_get_sessionid(current)); 8822 } 8823 8824 dev_change_rx_flags(dev, IFF_PROMISC); 8825 } 8826 if (notify) 8827 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8828 return 0; 8829 } 8830 8831 /** 8832 * dev_set_promiscuity - update promiscuity count on a device 8833 * @dev: device 8834 * @inc: modifier 8835 * 8836 * Add or remove promiscuity from a device. While the count in the device 8837 * remains above zero the interface remains promiscuous. Once it hits zero 8838 * the device reverts back to normal filtering operation. A negative inc 8839 * value is used to drop promiscuity on the device. 8840 * Return 0 if successful or a negative errno code on error. 8841 */ 8842 int dev_set_promiscuity(struct net_device *dev, int inc) 8843 { 8844 unsigned int old_flags = dev->flags; 8845 int err; 8846 8847 err = __dev_set_promiscuity(dev, inc, true); 8848 if (err < 0) 8849 return err; 8850 if (dev->flags != old_flags) 8851 dev_set_rx_mode(dev); 8852 return err; 8853 } 8854 EXPORT_SYMBOL(dev_set_promiscuity); 8855 8856 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8857 { 8858 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8859 unsigned int allmulti, flags; 8860 8861 ASSERT_RTNL(); 8862 8863 allmulti = dev->allmulti + inc; 8864 if (allmulti == 0) { 8865 /* 8866 * Avoid overflow. 8867 * If inc causes overflow, untouch allmulti and return error. 8868 */ 8869 if (unlikely(inc > 0)) { 8870 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8871 return -EOVERFLOW; 8872 } 8873 flags = old_flags & ~IFF_ALLMULTI; 8874 } else { 8875 flags = old_flags | IFF_ALLMULTI; 8876 } 8877 WRITE_ONCE(dev->allmulti, allmulti); 8878 if (flags != old_flags) { 8879 WRITE_ONCE(dev->flags, flags); 8880 netdev_info(dev, "%s allmulticast mode\n", 8881 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8882 dev_change_rx_flags(dev, IFF_ALLMULTI); 8883 dev_set_rx_mode(dev); 8884 if (notify) 8885 __dev_notify_flags(dev, old_flags, 8886 dev->gflags ^ old_gflags, 0, NULL); 8887 } 8888 return 0; 8889 } 8890 8891 /** 8892 * dev_set_allmulti - update allmulti count on a device 8893 * @dev: device 8894 * @inc: modifier 8895 * 8896 * Add or remove reception of all multicast frames to a device. While the 8897 * count in the device remains above zero the interface remains listening 8898 * to all interfaces. Once it hits zero the device reverts back to normal 8899 * filtering operation. A negative @inc value is used to drop the counter 8900 * when releasing a resource needing all multicasts. 8901 * Return 0 if successful or a negative errno code on error. 8902 */ 8903 8904 int dev_set_allmulti(struct net_device *dev, int inc) 8905 { 8906 return __dev_set_allmulti(dev, inc, true); 8907 } 8908 EXPORT_SYMBOL(dev_set_allmulti); 8909 8910 /* 8911 * Upload unicast and multicast address lists to device and 8912 * configure RX filtering. When the device doesn't support unicast 8913 * filtering it is put in promiscuous mode while unicast addresses 8914 * are present. 8915 */ 8916 void __dev_set_rx_mode(struct net_device *dev) 8917 { 8918 const struct net_device_ops *ops = dev->netdev_ops; 8919 8920 /* dev_open will call this function so the list will stay sane. */ 8921 if (!(dev->flags&IFF_UP)) 8922 return; 8923 8924 if (!netif_device_present(dev)) 8925 return; 8926 8927 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8928 /* Unicast addresses changes may only happen under the rtnl, 8929 * therefore calling __dev_set_promiscuity here is safe. 8930 */ 8931 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8932 __dev_set_promiscuity(dev, 1, false); 8933 dev->uc_promisc = true; 8934 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8935 __dev_set_promiscuity(dev, -1, false); 8936 dev->uc_promisc = false; 8937 } 8938 } 8939 8940 if (ops->ndo_set_rx_mode) 8941 ops->ndo_set_rx_mode(dev); 8942 } 8943 8944 void dev_set_rx_mode(struct net_device *dev) 8945 { 8946 netif_addr_lock_bh(dev); 8947 __dev_set_rx_mode(dev); 8948 netif_addr_unlock_bh(dev); 8949 } 8950 8951 /** 8952 * dev_get_flags - get flags reported to userspace 8953 * @dev: device 8954 * 8955 * Get the combination of flag bits exported through APIs to userspace. 8956 */ 8957 unsigned int dev_get_flags(const struct net_device *dev) 8958 { 8959 unsigned int flags; 8960 8961 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | 8962 IFF_ALLMULTI | 8963 IFF_RUNNING | 8964 IFF_LOWER_UP | 8965 IFF_DORMANT)) | 8966 (READ_ONCE(dev->gflags) & (IFF_PROMISC | 8967 IFF_ALLMULTI)); 8968 8969 if (netif_running(dev)) { 8970 if (netif_oper_up(dev)) 8971 flags |= IFF_RUNNING; 8972 if (netif_carrier_ok(dev)) 8973 flags |= IFF_LOWER_UP; 8974 if (netif_dormant(dev)) 8975 flags |= IFF_DORMANT; 8976 } 8977 8978 return flags; 8979 } 8980 EXPORT_SYMBOL(dev_get_flags); 8981 8982 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8983 struct netlink_ext_ack *extack) 8984 { 8985 unsigned int old_flags = dev->flags; 8986 int ret; 8987 8988 ASSERT_RTNL(); 8989 8990 /* 8991 * Set the flags on our device. 8992 */ 8993 8994 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8995 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8996 IFF_AUTOMEDIA)) | 8997 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8998 IFF_ALLMULTI)); 8999 9000 /* 9001 * Load in the correct multicast list now the flags have changed. 9002 */ 9003 9004 if ((old_flags ^ flags) & IFF_MULTICAST) 9005 dev_change_rx_flags(dev, IFF_MULTICAST); 9006 9007 dev_set_rx_mode(dev); 9008 9009 /* 9010 * Have we downed the interface. We handle IFF_UP ourselves 9011 * according to user attempts to set it, rather than blindly 9012 * setting it. 9013 */ 9014 9015 ret = 0; 9016 if ((old_flags ^ flags) & IFF_UP) { 9017 if (old_flags & IFF_UP) 9018 __dev_close(dev); 9019 else 9020 ret = __dev_open(dev, extack); 9021 } 9022 9023 if ((flags ^ dev->gflags) & IFF_PROMISC) { 9024 int inc = (flags & IFF_PROMISC) ? 1 : -1; 9025 unsigned int old_flags = dev->flags; 9026 9027 dev->gflags ^= IFF_PROMISC; 9028 9029 if (__dev_set_promiscuity(dev, inc, false) >= 0) 9030 if (dev->flags != old_flags) 9031 dev_set_rx_mode(dev); 9032 } 9033 9034 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 9035 * is important. Some (broken) drivers set IFF_PROMISC, when 9036 * IFF_ALLMULTI is requested not asking us and not reporting. 9037 */ 9038 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 9039 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 9040 9041 dev->gflags ^= IFF_ALLMULTI; 9042 __dev_set_allmulti(dev, inc, false); 9043 } 9044 9045 return ret; 9046 } 9047 9048 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 9049 unsigned int gchanges, u32 portid, 9050 const struct nlmsghdr *nlh) 9051 { 9052 unsigned int changes = dev->flags ^ old_flags; 9053 9054 if (gchanges) 9055 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 9056 9057 if (changes & IFF_UP) { 9058 if (dev->flags & IFF_UP) 9059 call_netdevice_notifiers(NETDEV_UP, dev); 9060 else 9061 call_netdevice_notifiers(NETDEV_DOWN, dev); 9062 } 9063 9064 if (dev->flags & IFF_UP && 9065 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 9066 struct netdev_notifier_change_info change_info = { 9067 .info = { 9068 .dev = dev, 9069 }, 9070 .flags_changed = changes, 9071 }; 9072 9073 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 9074 } 9075 } 9076 9077 /** 9078 * dev_change_flags - change device settings 9079 * @dev: device 9080 * @flags: device state flags 9081 * @extack: netlink extended ack 9082 * 9083 * Change settings on device based state flags. The flags are 9084 * in the userspace exported format. 9085 */ 9086 int dev_change_flags(struct net_device *dev, unsigned int flags, 9087 struct netlink_ext_ack *extack) 9088 { 9089 int ret; 9090 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 9091 9092 ret = __dev_change_flags(dev, flags, extack); 9093 if (ret < 0) 9094 return ret; 9095 9096 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 9097 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 9098 return ret; 9099 } 9100 EXPORT_SYMBOL(dev_change_flags); 9101 9102 int __dev_set_mtu(struct net_device *dev, int new_mtu) 9103 { 9104 const struct net_device_ops *ops = dev->netdev_ops; 9105 9106 if (ops->ndo_change_mtu) 9107 return ops->ndo_change_mtu(dev, new_mtu); 9108 9109 /* Pairs with all the lockless reads of dev->mtu in the stack */ 9110 WRITE_ONCE(dev->mtu, new_mtu); 9111 return 0; 9112 } 9113 EXPORT_SYMBOL(__dev_set_mtu); 9114 9115 int dev_validate_mtu(struct net_device *dev, int new_mtu, 9116 struct netlink_ext_ack *extack) 9117 { 9118 /* MTU must be positive, and in range */ 9119 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 9120 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 9121 return -EINVAL; 9122 } 9123 9124 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 9125 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 9126 return -EINVAL; 9127 } 9128 return 0; 9129 } 9130 9131 /** 9132 * dev_set_mtu_ext - Change maximum transfer unit 9133 * @dev: device 9134 * @new_mtu: new transfer unit 9135 * @extack: netlink extended ack 9136 * 9137 * Change the maximum transfer size of the network device. 9138 */ 9139 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 9140 struct netlink_ext_ack *extack) 9141 { 9142 int err, orig_mtu; 9143 9144 if (new_mtu == dev->mtu) 9145 return 0; 9146 9147 err = dev_validate_mtu(dev, new_mtu, extack); 9148 if (err) 9149 return err; 9150 9151 if (!netif_device_present(dev)) 9152 return -ENODEV; 9153 9154 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 9155 err = notifier_to_errno(err); 9156 if (err) 9157 return err; 9158 9159 orig_mtu = dev->mtu; 9160 err = __dev_set_mtu(dev, new_mtu); 9161 9162 if (!err) { 9163 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9164 orig_mtu); 9165 err = notifier_to_errno(err); 9166 if (err) { 9167 /* setting mtu back and notifying everyone again, 9168 * so that they have a chance to revert changes. 9169 */ 9170 __dev_set_mtu(dev, orig_mtu); 9171 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 9172 new_mtu); 9173 } 9174 } 9175 return err; 9176 } 9177 9178 int dev_set_mtu(struct net_device *dev, int new_mtu) 9179 { 9180 struct netlink_ext_ack extack; 9181 int err; 9182 9183 memset(&extack, 0, sizeof(extack)); 9184 err = dev_set_mtu_ext(dev, new_mtu, &extack); 9185 if (err && extack._msg) 9186 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 9187 return err; 9188 } 9189 EXPORT_SYMBOL(dev_set_mtu); 9190 9191 /** 9192 * dev_change_tx_queue_len - Change TX queue length of a netdevice 9193 * @dev: device 9194 * @new_len: new tx queue length 9195 */ 9196 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 9197 { 9198 unsigned int orig_len = dev->tx_queue_len; 9199 int res; 9200 9201 if (new_len != (unsigned int)new_len) 9202 return -ERANGE; 9203 9204 if (new_len != orig_len) { 9205 WRITE_ONCE(dev->tx_queue_len, new_len); 9206 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 9207 res = notifier_to_errno(res); 9208 if (res) 9209 goto err_rollback; 9210 res = dev_qdisc_change_tx_queue_len(dev); 9211 if (res) 9212 goto err_rollback; 9213 } 9214 9215 return 0; 9216 9217 err_rollback: 9218 netdev_err(dev, "refused to change device tx_queue_len\n"); 9219 WRITE_ONCE(dev->tx_queue_len, orig_len); 9220 return res; 9221 } 9222 9223 /** 9224 * dev_set_group - Change group this device belongs to 9225 * @dev: device 9226 * @new_group: group this device should belong to 9227 */ 9228 void dev_set_group(struct net_device *dev, int new_group) 9229 { 9230 dev->group = new_group; 9231 } 9232 9233 /** 9234 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 9235 * @dev: device 9236 * @addr: new address 9237 * @extack: netlink extended ack 9238 */ 9239 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9240 struct netlink_ext_ack *extack) 9241 { 9242 struct netdev_notifier_pre_changeaddr_info info = { 9243 .info.dev = dev, 9244 .info.extack = extack, 9245 .dev_addr = addr, 9246 }; 9247 int rc; 9248 9249 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9250 return notifier_to_errno(rc); 9251 } 9252 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9253 9254 /** 9255 * dev_set_mac_address - Change Media Access Control Address 9256 * @dev: device 9257 * @sa: new address 9258 * @extack: netlink extended ack 9259 * 9260 * Change the hardware (MAC) address of the device 9261 */ 9262 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9263 struct netlink_ext_ack *extack) 9264 { 9265 const struct net_device_ops *ops = dev->netdev_ops; 9266 int err; 9267 9268 if (!ops->ndo_set_mac_address) 9269 return -EOPNOTSUPP; 9270 if (sa->sa_family != dev->type) 9271 return -EINVAL; 9272 if (!netif_device_present(dev)) 9273 return -ENODEV; 9274 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9275 if (err) 9276 return err; 9277 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 9278 err = ops->ndo_set_mac_address(dev, sa); 9279 if (err) 9280 return err; 9281 } 9282 dev->addr_assign_type = NET_ADDR_SET; 9283 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9284 add_device_randomness(dev->dev_addr, dev->addr_len); 9285 return 0; 9286 } 9287 EXPORT_SYMBOL(dev_set_mac_address); 9288 9289 DECLARE_RWSEM(dev_addr_sem); 9290 9291 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9292 struct netlink_ext_ack *extack) 9293 { 9294 int ret; 9295 9296 down_write(&dev_addr_sem); 9297 ret = dev_set_mac_address(dev, sa, extack); 9298 up_write(&dev_addr_sem); 9299 return ret; 9300 } 9301 EXPORT_SYMBOL(dev_set_mac_address_user); 9302 9303 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9304 { 9305 size_t size = sizeof(sa->sa_data_min); 9306 struct net_device *dev; 9307 int ret = 0; 9308 9309 down_read(&dev_addr_sem); 9310 rcu_read_lock(); 9311 9312 dev = dev_get_by_name_rcu(net, dev_name); 9313 if (!dev) { 9314 ret = -ENODEV; 9315 goto unlock; 9316 } 9317 if (!dev->addr_len) 9318 memset(sa->sa_data, 0, size); 9319 else 9320 memcpy(sa->sa_data, dev->dev_addr, 9321 min_t(size_t, size, dev->addr_len)); 9322 sa->sa_family = dev->type; 9323 9324 unlock: 9325 rcu_read_unlock(); 9326 up_read(&dev_addr_sem); 9327 return ret; 9328 } 9329 EXPORT_SYMBOL(dev_get_mac_address); 9330 9331 /** 9332 * dev_change_carrier - Change device carrier 9333 * @dev: device 9334 * @new_carrier: new value 9335 * 9336 * Change device carrier 9337 */ 9338 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9339 { 9340 const struct net_device_ops *ops = dev->netdev_ops; 9341 9342 if (!ops->ndo_change_carrier) 9343 return -EOPNOTSUPP; 9344 if (!netif_device_present(dev)) 9345 return -ENODEV; 9346 return ops->ndo_change_carrier(dev, new_carrier); 9347 } 9348 9349 /** 9350 * dev_get_phys_port_id - Get device physical port ID 9351 * @dev: device 9352 * @ppid: port ID 9353 * 9354 * Get device physical port ID 9355 */ 9356 int dev_get_phys_port_id(struct net_device *dev, 9357 struct netdev_phys_item_id *ppid) 9358 { 9359 const struct net_device_ops *ops = dev->netdev_ops; 9360 9361 if (!ops->ndo_get_phys_port_id) 9362 return -EOPNOTSUPP; 9363 return ops->ndo_get_phys_port_id(dev, ppid); 9364 } 9365 9366 /** 9367 * dev_get_phys_port_name - Get device physical port name 9368 * @dev: device 9369 * @name: port name 9370 * @len: limit of bytes to copy to name 9371 * 9372 * Get device physical port name 9373 */ 9374 int dev_get_phys_port_name(struct net_device *dev, 9375 char *name, size_t len) 9376 { 9377 const struct net_device_ops *ops = dev->netdev_ops; 9378 int err; 9379 9380 if (ops->ndo_get_phys_port_name) { 9381 err = ops->ndo_get_phys_port_name(dev, name, len); 9382 if (err != -EOPNOTSUPP) 9383 return err; 9384 } 9385 return devlink_compat_phys_port_name_get(dev, name, len); 9386 } 9387 9388 /** 9389 * dev_get_port_parent_id - Get the device's port parent identifier 9390 * @dev: network device 9391 * @ppid: pointer to a storage for the port's parent identifier 9392 * @recurse: allow/disallow recursion to lower devices 9393 * 9394 * Get the devices's port parent identifier 9395 */ 9396 int dev_get_port_parent_id(struct net_device *dev, 9397 struct netdev_phys_item_id *ppid, 9398 bool recurse) 9399 { 9400 const struct net_device_ops *ops = dev->netdev_ops; 9401 struct netdev_phys_item_id first = { }; 9402 struct net_device *lower_dev; 9403 struct list_head *iter; 9404 int err; 9405 9406 if (ops->ndo_get_port_parent_id) { 9407 err = ops->ndo_get_port_parent_id(dev, ppid); 9408 if (err != -EOPNOTSUPP) 9409 return err; 9410 } 9411 9412 err = devlink_compat_switch_id_get(dev, ppid); 9413 if (!recurse || err != -EOPNOTSUPP) 9414 return err; 9415 9416 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9417 err = dev_get_port_parent_id(lower_dev, ppid, true); 9418 if (err) 9419 break; 9420 if (!first.id_len) 9421 first = *ppid; 9422 else if (memcmp(&first, ppid, sizeof(*ppid))) 9423 return -EOPNOTSUPP; 9424 } 9425 9426 return err; 9427 } 9428 EXPORT_SYMBOL(dev_get_port_parent_id); 9429 9430 /** 9431 * netdev_port_same_parent_id - Indicate if two network devices have 9432 * the same port parent identifier 9433 * @a: first network device 9434 * @b: second network device 9435 */ 9436 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9437 { 9438 struct netdev_phys_item_id a_id = { }; 9439 struct netdev_phys_item_id b_id = { }; 9440 9441 if (dev_get_port_parent_id(a, &a_id, true) || 9442 dev_get_port_parent_id(b, &b_id, true)) 9443 return false; 9444 9445 return netdev_phys_item_id_same(&a_id, &b_id); 9446 } 9447 EXPORT_SYMBOL(netdev_port_same_parent_id); 9448 9449 /** 9450 * dev_change_proto_down - set carrier according to proto_down. 9451 * 9452 * @dev: device 9453 * @proto_down: new value 9454 */ 9455 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9456 { 9457 if (!dev->change_proto_down) 9458 return -EOPNOTSUPP; 9459 if (!netif_device_present(dev)) 9460 return -ENODEV; 9461 if (proto_down) 9462 netif_carrier_off(dev); 9463 else 9464 netif_carrier_on(dev); 9465 WRITE_ONCE(dev->proto_down, proto_down); 9466 return 0; 9467 } 9468 9469 /** 9470 * dev_change_proto_down_reason - proto down reason 9471 * 9472 * @dev: device 9473 * @mask: proto down mask 9474 * @value: proto down value 9475 */ 9476 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9477 u32 value) 9478 { 9479 u32 proto_down_reason; 9480 int b; 9481 9482 if (!mask) { 9483 proto_down_reason = value; 9484 } else { 9485 proto_down_reason = dev->proto_down_reason; 9486 for_each_set_bit(b, &mask, 32) { 9487 if (value & (1 << b)) 9488 proto_down_reason |= BIT(b); 9489 else 9490 proto_down_reason &= ~BIT(b); 9491 } 9492 } 9493 WRITE_ONCE(dev->proto_down_reason, proto_down_reason); 9494 } 9495 9496 struct bpf_xdp_link { 9497 struct bpf_link link; 9498 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9499 int flags; 9500 }; 9501 9502 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9503 { 9504 if (flags & XDP_FLAGS_HW_MODE) 9505 return XDP_MODE_HW; 9506 if (flags & XDP_FLAGS_DRV_MODE) 9507 return XDP_MODE_DRV; 9508 if (flags & XDP_FLAGS_SKB_MODE) 9509 return XDP_MODE_SKB; 9510 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9511 } 9512 9513 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9514 { 9515 switch (mode) { 9516 case XDP_MODE_SKB: 9517 return generic_xdp_install; 9518 case XDP_MODE_DRV: 9519 case XDP_MODE_HW: 9520 return dev->netdev_ops->ndo_bpf; 9521 default: 9522 return NULL; 9523 } 9524 } 9525 9526 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9527 enum bpf_xdp_mode mode) 9528 { 9529 return dev->xdp_state[mode].link; 9530 } 9531 9532 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9533 enum bpf_xdp_mode mode) 9534 { 9535 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9536 9537 if (link) 9538 return link->link.prog; 9539 return dev->xdp_state[mode].prog; 9540 } 9541 9542 u8 dev_xdp_prog_count(struct net_device *dev) 9543 { 9544 u8 count = 0; 9545 int i; 9546 9547 for (i = 0; i < __MAX_XDP_MODE; i++) 9548 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9549 count++; 9550 return count; 9551 } 9552 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9553 9554 u8 dev_xdp_sb_prog_count(struct net_device *dev) 9555 { 9556 u8 count = 0; 9557 int i; 9558 9559 for (i = 0; i < __MAX_XDP_MODE; i++) 9560 if (dev->xdp_state[i].prog && 9561 !dev->xdp_state[i].prog->aux->xdp_has_frags) 9562 count++; 9563 return count; 9564 } 9565 9566 int dev_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf) 9567 { 9568 if (!dev->netdev_ops->ndo_bpf) 9569 return -EOPNOTSUPP; 9570 9571 if (dev->ethtool->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED && 9572 bpf->command == XDP_SETUP_PROG && 9573 bpf->prog && !bpf->prog->aux->xdp_has_frags) { 9574 NL_SET_ERR_MSG(bpf->extack, 9575 "unable to propagate XDP to device using tcp-data-split"); 9576 return -EBUSY; 9577 } 9578 9579 if (dev_get_min_mp_channel_count(dev)) { 9580 NL_SET_ERR_MSG(bpf->extack, "unable to propagate XDP to device using memory provider"); 9581 return -EBUSY; 9582 } 9583 9584 return dev->netdev_ops->ndo_bpf(dev, bpf); 9585 } 9586 EXPORT_SYMBOL_GPL(dev_xdp_propagate); 9587 9588 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9589 { 9590 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9591 9592 return prog ? prog->aux->id : 0; 9593 } 9594 9595 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9596 struct bpf_xdp_link *link) 9597 { 9598 dev->xdp_state[mode].link = link; 9599 dev->xdp_state[mode].prog = NULL; 9600 } 9601 9602 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9603 struct bpf_prog *prog) 9604 { 9605 dev->xdp_state[mode].link = NULL; 9606 dev->xdp_state[mode].prog = prog; 9607 } 9608 9609 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9610 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9611 u32 flags, struct bpf_prog *prog) 9612 { 9613 struct netdev_bpf xdp; 9614 int err; 9615 9616 if (dev->ethtool->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED && 9617 prog && !prog->aux->xdp_has_frags) { 9618 NL_SET_ERR_MSG(extack, "unable to install XDP to device using tcp-data-split"); 9619 return -EBUSY; 9620 } 9621 9622 if (dev_get_min_mp_channel_count(dev)) { 9623 NL_SET_ERR_MSG(extack, "unable to install XDP to device using memory provider"); 9624 return -EBUSY; 9625 } 9626 9627 memset(&xdp, 0, sizeof(xdp)); 9628 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9629 xdp.extack = extack; 9630 xdp.flags = flags; 9631 xdp.prog = prog; 9632 9633 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9634 * "moved" into driver), so they don't increment it on their own, but 9635 * they do decrement refcnt when program is detached or replaced. 9636 * Given net_device also owns link/prog, we need to bump refcnt here 9637 * to prevent drivers from underflowing it. 9638 */ 9639 if (prog) 9640 bpf_prog_inc(prog); 9641 err = bpf_op(dev, &xdp); 9642 if (err) { 9643 if (prog) 9644 bpf_prog_put(prog); 9645 return err; 9646 } 9647 9648 if (mode != XDP_MODE_HW) 9649 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9650 9651 return 0; 9652 } 9653 9654 static void dev_xdp_uninstall(struct net_device *dev) 9655 { 9656 struct bpf_xdp_link *link; 9657 struct bpf_prog *prog; 9658 enum bpf_xdp_mode mode; 9659 bpf_op_t bpf_op; 9660 9661 ASSERT_RTNL(); 9662 9663 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9664 prog = dev_xdp_prog(dev, mode); 9665 if (!prog) 9666 continue; 9667 9668 bpf_op = dev_xdp_bpf_op(dev, mode); 9669 if (!bpf_op) 9670 continue; 9671 9672 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9673 9674 /* auto-detach link from net device */ 9675 link = dev_xdp_link(dev, mode); 9676 if (link) 9677 link->dev = NULL; 9678 else 9679 bpf_prog_put(prog); 9680 9681 dev_xdp_set_link(dev, mode, NULL); 9682 } 9683 } 9684 9685 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9686 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9687 struct bpf_prog *old_prog, u32 flags) 9688 { 9689 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9690 struct bpf_prog *cur_prog; 9691 struct net_device *upper; 9692 struct list_head *iter; 9693 enum bpf_xdp_mode mode; 9694 bpf_op_t bpf_op; 9695 int err; 9696 9697 ASSERT_RTNL(); 9698 9699 /* either link or prog attachment, never both */ 9700 if (link && (new_prog || old_prog)) 9701 return -EINVAL; 9702 /* link supports only XDP mode flags */ 9703 if (link && (flags & ~XDP_FLAGS_MODES)) { 9704 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9705 return -EINVAL; 9706 } 9707 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9708 if (num_modes > 1) { 9709 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9710 return -EINVAL; 9711 } 9712 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9713 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9714 NL_SET_ERR_MSG(extack, 9715 "More than one program loaded, unset mode is ambiguous"); 9716 return -EINVAL; 9717 } 9718 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9719 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9720 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9721 return -EINVAL; 9722 } 9723 9724 mode = dev_xdp_mode(dev, flags); 9725 /* can't replace attached link */ 9726 if (dev_xdp_link(dev, mode)) { 9727 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9728 return -EBUSY; 9729 } 9730 9731 /* don't allow if an upper device already has a program */ 9732 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9733 if (dev_xdp_prog_count(upper) > 0) { 9734 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9735 return -EEXIST; 9736 } 9737 } 9738 9739 cur_prog = dev_xdp_prog(dev, mode); 9740 /* can't replace attached prog with link */ 9741 if (link && cur_prog) { 9742 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9743 return -EBUSY; 9744 } 9745 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9746 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9747 return -EEXIST; 9748 } 9749 9750 /* put effective new program into new_prog */ 9751 if (link) 9752 new_prog = link->link.prog; 9753 9754 if (new_prog) { 9755 bool offload = mode == XDP_MODE_HW; 9756 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9757 ? XDP_MODE_DRV : XDP_MODE_SKB; 9758 9759 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9760 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9761 return -EBUSY; 9762 } 9763 if (!offload && dev_xdp_prog(dev, other_mode)) { 9764 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9765 return -EEXIST; 9766 } 9767 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9768 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9769 return -EINVAL; 9770 } 9771 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9772 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9773 return -EINVAL; 9774 } 9775 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9776 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9777 return -EINVAL; 9778 } 9779 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9780 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9781 return -EINVAL; 9782 } 9783 } 9784 9785 /* don't call drivers if the effective program didn't change */ 9786 if (new_prog != cur_prog) { 9787 bpf_op = dev_xdp_bpf_op(dev, mode); 9788 if (!bpf_op) { 9789 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9790 return -EOPNOTSUPP; 9791 } 9792 9793 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9794 if (err) 9795 return err; 9796 } 9797 9798 if (link) 9799 dev_xdp_set_link(dev, mode, link); 9800 else 9801 dev_xdp_set_prog(dev, mode, new_prog); 9802 if (cur_prog) 9803 bpf_prog_put(cur_prog); 9804 9805 return 0; 9806 } 9807 9808 static int dev_xdp_attach_link(struct net_device *dev, 9809 struct netlink_ext_ack *extack, 9810 struct bpf_xdp_link *link) 9811 { 9812 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9813 } 9814 9815 static int dev_xdp_detach_link(struct net_device *dev, 9816 struct netlink_ext_ack *extack, 9817 struct bpf_xdp_link *link) 9818 { 9819 enum bpf_xdp_mode mode; 9820 bpf_op_t bpf_op; 9821 9822 ASSERT_RTNL(); 9823 9824 mode = dev_xdp_mode(dev, link->flags); 9825 if (dev_xdp_link(dev, mode) != link) 9826 return -EINVAL; 9827 9828 bpf_op = dev_xdp_bpf_op(dev, mode); 9829 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9830 dev_xdp_set_link(dev, mode, NULL); 9831 return 0; 9832 } 9833 9834 static void bpf_xdp_link_release(struct bpf_link *link) 9835 { 9836 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9837 9838 rtnl_lock(); 9839 9840 /* if racing with net_device's tear down, xdp_link->dev might be 9841 * already NULL, in which case link was already auto-detached 9842 */ 9843 if (xdp_link->dev) { 9844 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9845 xdp_link->dev = NULL; 9846 } 9847 9848 rtnl_unlock(); 9849 } 9850 9851 static int bpf_xdp_link_detach(struct bpf_link *link) 9852 { 9853 bpf_xdp_link_release(link); 9854 return 0; 9855 } 9856 9857 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9858 { 9859 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9860 9861 kfree(xdp_link); 9862 } 9863 9864 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9865 struct seq_file *seq) 9866 { 9867 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9868 u32 ifindex = 0; 9869 9870 rtnl_lock(); 9871 if (xdp_link->dev) 9872 ifindex = xdp_link->dev->ifindex; 9873 rtnl_unlock(); 9874 9875 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9876 } 9877 9878 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9879 struct bpf_link_info *info) 9880 { 9881 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9882 u32 ifindex = 0; 9883 9884 rtnl_lock(); 9885 if (xdp_link->dev) 9886 ifindex = xdp_link->dev->ifindex; 9887 rtnl_unlock(); 9888 9889 info->xdp.ifindex = ifindex; 9890 return 0; 9891 } 9892 9893 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9894 struct bpf_prog *old_prog) 9895 { 9896 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9897 enum bpf_xdp_mode mode; 9898 bpf_op_t bpf_op; 9899 int err = 0; 9900 9901 rtnl_lock(); 9902 9903 /* link might have been auto-released already, so fail */ 9904 if (!xdp_link->dev) { 9905 err = -ENOLINK; 9906 goto out_unlock; 9907 } 9908 9909 if (old_prog && link->prog != old_prog) { 9910 err = -EPERM; 9911 goto out_unlock; 9912 } 9913 old_prog = link->prog; 9914 if (old_prog->type != new_prog->type || 9915 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9916 err = -EINVAL; 9917 goto out_unlock; 9918 } 9919 9920 if (old_prog == new_prog) { 9921 /* no-op, don't disturb drivers */ 9922 bpf_prog_put(new_prog); 9923 goto out_unlock; 9924 } 9925 9926 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9927 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9928 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9929 xdp_link->flags, new_prog); 9930 if (err) 9931 goto out_unlock; 9932 9933 old_prog = xchg(&link->prog, new_prog); 9934 bpf_prog_put(old_prog); 9935 9936 out_unlock: 9937 rtnl_unlock(); 9938 return err; 9939 } 9940 9941 static const struct bpf_link_ops bpf_xdp_link_lops = { 9942 .release = bpf_xdp_link_release, 9943 .dealloc = bpf_xdp_link_dealloc, 9944 .detach = bpf_xdp_link_detach, 9945 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9946 .fill_link_info = bpf_xdp_link_fill_link_info, 9947 .update_prog = bpf_xdp_link_update, 9948 }; 9949 9950 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9951 { 9952 struct net *net = current->nsproxy->net_ns; 9953 struct bpf_link_primer link_primer; 9954 struct netlink_ext_ack extack = {}; 9955 struct bpf_xdp_link *link; 9956 struct net_device *dev; 9957 int err, fd; 9958 9959 rtnl_lock(); 9960 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9961 if (!dev) { 9962 rtnl_unlock(); 9963 return -EINVAL; 9964 } 9965 9966 link = kzalloc(sizeof(*link), GFP_USER); 9967 if (!link) { 9968 err = -ENOMEM; 9969 goto unlock; 9970 } 9971 9972 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9973 link->dev = dev; 9974 link->flags = attr->link_create.flags; 9975 9976 err = bpf_link_prime(&link->link, &link_primer); 9977 if (err) { 9978 kfree(link); 9979 goto unlock; 9980 } 9981 9982 err = dev_xdp_attach_link(dev, &extack, link); 9983 rtnl_unlock(); 9984 9985 if (err) { 9986 link->dev = NULL; 9987 bpf_link_cleanup(&link_primer); 9988 trace_bpf_xdp_link_attach_failed(extack._msg); 9989 goto out_put_dev; 9990 } 9991 9992 fd = bpf_link_settle(&link_primer); 9993 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9994 dev_put(dev); 9995 return fd; 9996 9997 unlock: 9998 rtnl_unlock(); 9999 10000 out_put_dev: 10001 dev_put(dev); 10002 return err; 10003 } 10004 10005 /** 10006 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 10007 * @dev: device 10008 * @extack: netlink extended ack 10009 * @fd: new program fd or negative value to clear 10010 * @expected_fd: old program fd that userspace expects to replace or clear 10011 * @flags: xdp-related flags 10012 * 10013 * Set or clear a bpf program for a device 10014 */ 10015 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 10016 int fd, int expected_fd, u32 flags) 10017 { 10018 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 10019 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 10020 int err; 10021 10022 ASSERT_RTNL(); 10023 10024 if (fd >= 0) { 10025 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 10026 mode != XDP_MODE_SKB); 10027 if (IS_ERR(new_prog)) 10028 return PTR_ERR(new_prog); 10029 } 10030 10031 if (expected_fd >= 0) { 10032 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 10033 mode != XDP_MODE_SKB); 10034 if (IS_ERR(old_prog)) { 10035 err = PTR_ERR(old_prog); 10036 old_prog = NULL; 10037 goto err_out; 10038 } 10039 } 10040 10041 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 10042 10043 err_out: 10044 if (err && new_prog) 10045 bpf_prog_put(new_prog); 10046 if (old_prog) 10047 bpf_prog_put(old_prog); 10048 return err; 10049 } 10050 10051 u32 dev_get_min_mp_channel_count(const struct net_device *dev) 10052 { 10053 int i; 10054 10055 ASSERT_RTNL(); 10056 10057 for (i = dev->real_num_rx_queues - 1; i >= 0; i--) 10058 if (dev->_rx[i].mp_params.mp_priv) 10059 /* The channel count is the idx plus 1. */ 10060 return i + 1; 10061 10062 return 0; 10063 } 10064 10065 /** 10066 * dev_index_reserve() - allocate an ifindex in a namespace 10067 * @net: the applicable net namespace 10068 * @ifindex: requested ifindex, pass %0 to get one allocated 10069 * 10070 * Allocate a ifindex for a new device. Caller must either use the ifindex 10071 * to store the device (via list_netdevice()) or call dev_index_release() 10072 * to give the index up. 10073 * 10074 * Return: a suitable unique value for a new device interface number or -errno. 10075 */ 10076 static int dev_index_reserve(struct net *net, u32 ifindex) 10077 { 10078 int err; 10079 10080 if (ifindex > INT_MAX) { 10081 DEBUG_NET_WARN_ON_ONCE(1); 10082 return -EINVAL; 10083 } 10084 10085 if (!ifindex) 10086 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 10087 xa_limit_31b, &net->ifindex, GFP_KERNEL); 10088 else 10089 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 10090 if (err < 0) 10091 return err; 10092 10093 return ifindex; 10094 } 10095 10096 static void dev_index_release(struct net *net, int ifindex) 10097 { 10098 /* Expect only unused indexes, unlist_netdevice() removes the used */ 10099 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 10100 } 10101 10102 /* Delayed registration/unregisteration */ 10103 LIST_HEAD(net_todo_list); 10104 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 10105 atomic_t dev_unreg_count = ATOMIC_INIT(0); 10106 10107 static void net_set_todo(struct net_device *dev) 10108 { 10109 list_add_tail(&dev->todo_list, &net_todo_list); 10110 } 10111 10112 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 10113 struct net_device *upper, netdev_features_t features) 10114 { 10115 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10116 netdev_features_t feature; 10117 int feature_bit; 10118 10119 for_each_netdev_feature(upper_disables, feature_bit) { 10120 feature = __NETIF_F_BIT(feature_bit); 10121 if (!(upper->wanted_features & feature) 10122 && (features & feature)) { 10123 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 10124 &feature, upper->name); 10125 features &= ~feature; 10126 } 10127 } 10128 10129 return features; 10130 } 10131 10132 static void netdev_sync_lower_features(struct net_device *upper, 10133 struct net_device *lower, netdev_features_t features) 10134 { 10135 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 10136 netdev_features_t feature; 10137 int feature_bit; 10138 10139 for_each_netdev_feature(upper_disables, feature_bit) { 10140 feature = __NETIF_F_BIT(feature_bit); 10141 if (!(features & feature) && (lower->features & feature)) { 10142 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 10143 &feature, lower->name); 10144 lower->wanted_features &= ~feature; 10145 __netdev_update_features(lower); 10146 10147 if (unlikely(lower->features & feature)) 10148 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 10149 &feature, lower->name); 10150 else 10151 netdev_features_change(lower); 10152 } 10153 } 10154 } 10155 10156 static bool netdev_has_ip_or_hw_csum(netdev_features_t features) 10157 { 10158 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 10159 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask; 10160 bool hw_csum = features & NETIF_F_HW_CSUM; 10161 10162 return ip_csum || hw_csum; 10163 } 10164 10165 static netdev_features_t netdev_fix_features(struct net_device *dev, 10166 netdev_features_t features) 10167 { 10168 /* Fix illegal checksum combinations */ 10169 if ((features & NETIF_F_HW_CSUM) && 10170 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 10171 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 10172 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 10173 } 10174 10175 /* TSO requires that SG is present as well. */ 10176 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 10177 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 10178 features &= ~NETIF_F_ALL_TSO; 10179 } 10180 10181 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 10182 !(features & NETIF_F_IP_CSUM)) { 10183 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 10184 features &= ~NETIF_F_TSO; 10185 features &= ~NETIF_F_TSO_ECN; 10186 } 10187 10188 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 10189 !(features & NETIF_F_IPV6_CSUM)) { 10190 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 10191 features &= ~NETIF_F_TSO6; 10192 } 10193 10194 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 10195 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 10196 features &= ~NETIF_F_TSO_MANGLEID; 10197 10198 /* TSO ECN requires that TSO is present as well. */ 10199 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 10200 features &= ~NETIF_F_TSO_ECN; 10201 10202 /* Software GSO depends on SG. */ 10203 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 10204 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 10205 features &= ~NETIF_F_GSO; 10206 } 10207 10208 /* GSO partial features require GSO partial be set */ 10209 if ((features & dev->gso_partial_features) && 10210 !(features & NETIF_F_GSO_PARTIAL)) { 10211 netdev_dbg(dev, 10212 "Dropping partially supported GSO features since no GSO partial.\n"); 10213 features &= ~dev->gso_partial_features; 10214 } 10215 10216 if (!(features & NETIF_F_RXCSUM)) { 10217 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 10218 * successfully merged by hardware must also have the 10219 * checksum verified by hardware. If the user does not 10220 * want to enable RXCSUM, logically, we should disable GRO_HW. 10221 */ 10222 if (features & NETIF_F_GRO_HW) { 10223 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 10224 features &= ~NETIF_F_GRO_HW; 10225 } 10226 } 10227 10228 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 10229 if (features & NETIF_F_RXFCS) { 10230 if (features & NETIF_F_LRO) { 10231 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 10232 features &= ~NETIF_F_LRO; 10233 } 10234 10235 if (features & NETIF_F_GRO_HW) { 10236 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 10237 features &= ~NETIF_F_GRO_HW; 10238 } 10239 } 10240 10241 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 10242 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 10243 features &= ~NETIF_F_LRO; 10244 } 10245 10246 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) { 10247 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 10248 features &= ~NETIF_F_HW_TLS_TX; 10249 } 10250 10251 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 10252 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 10253 features &= ~NETIF_F_HW_TLS_RX; 10254 } 10255 10256 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) { 10257 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n"); 10258 features &= ~NETIF_F_GSO_UDP_L4; 10259 } 10260 10261 return features; 10262 } 10263 10264 int __netdev_update_features(struct net_device *dev) 10265 { 10266 struct net_device *upper, *lower; 10267 netdev_features_t features; 10268 struct list_head *iter; 10269 int err = -1; 10270 10271 ASSERT_RTNL(); 10272 10273 features = netdev_get_wanted_features(dev); 10274 10275 if (dev->netdev_ops->ndo_fix_features) 10276 features = dev->netdev_ops->ndo_fix_features(dev, features); 10277 10278 /* driver might be less strict about feature dependencies */ 10279 features = netdev_fix_features(dev, features); 10280 10281 /* some features can't be enabled if they're off on an upper device */ 10282 netdev_for_each_upper_dev_rcu(dev, upper, iter) 10283 features = netdev_sync_upper_features(dev, upper, features); 10284 10285 if (dev->features == features) 10286 goto sync_lower; 10287 10288 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 10289 &dev->features, &features); 10290 10291 if (dev->netdev_ops->ndo_set_features) 10292 err = dev->netdev_ops->ndo_set_features(dev, features); 10293 else 10294 err = 0; 10295 10296 if (unlikely(err < 0)) { 10297 netdev_err(dev, 10298 "set_features() failed (%d); wanted %pNF, left %pNF\n", 10299 err, &features, &dev->features); 10300 /* return non-0 since some features might have changed and 10301 * it's better to fire a spurious notification than miss it 10302 */ 10303 return -1; 10304 } 10305 10306 sync_lower: 10307 /* some features must be disabled on lower devices when disabled 10308 * on an upper device (think: bonding master or bridge) 10309 */ 10310 netdev_for_each_lower_dev(dev, lower, iter) 10311 netdev_sync_lower_features(dev, lower, features); 10312 10313 if (!err) { 10314 netdev_features_t diff = features ^ dev->features; 10315 10316 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10317 /* udp_tunnel_{get,drop}_rx_info both need 10318 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10319 * device, or they won't do anything. 10320 * Thus we need to update dev->features 10321 * *before* calling udp_tunnel_get_rx_info, 10322 * but *after* calling udp_tunnel_drop_rx_info. 10323 */ 10324 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10325 dev->features = features; 10326 udp_tunnel_get_rx_info(dev); 10327 } else { 10328 udp_tunnel_drop_rx_info(dev); 10329 } 10330 } 10331 10332 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10333 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10334 dev->features = features; 10335 err |= vlan_get_rx_ctag_filter_info(dev); 10336 } else { 10337 vlan_drop_rx_ctag_filter_info(dev); 10338 } 10339 } 10340 10341 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10342 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10343 dev->features = features; 10344 err |= vlan_get_rx_stag_filter_info(dev); 10345 } else { 10346 vlan_drop_rx_stag_filter_info(dev); 10347 } 10348 } 10349 10350 dev->features = features; 10351 } 10352 10353 return err < 0 ? 0 : 1; 10354 } 10355 10356 /** 10357 * netdev_update_features - recalculate device features 10358 * @dev: the device to check 10359 * 10360 * Recalculate dev->features set and send notifications if it 10361 * has changed. Should be called after driver or hardware dependent 10362 * conditions might have changed that influence the features. 10363 */ 10364 void netdev_update_features(struct net_device *dev) 10365 { 10366 if (__netdev_update_features(dev)) 10367 netdev_features_change(dev); 10368 } 10369 EXPORT_SYMBOL(netdev_update_features); 10370 10371 /** 10372 * netdev_change_features - recalculate device features 10373 * @dev: the device to check 10374 * 10375 * Recalculate dev->features set and send notifications even 10376 * if they have not changed. Should be called instead of 10377 * netdev_update_features() if also dev->vlan_features might 10378 * have changed to allow the changes to be propagated to stacked 10379 * VLAN devices. 10380 */ 10381 void netdev_change_features(struct net_device *dev) 10382 { 10383 __netdev_update_features(dev); 10384 netdev_features_change(dev); 10385 } 10386 EXPORT_SYMBOL(netdev_change_features); 10387 10388 /** 10389 * netif_stacked_transfer_operstate - transfer operstate 10390 * @rootdev: the root or lower level device to transfer state from 10391 * @dev: the device to transfer operstate to 10392 * 10393 * Transfer operational state from root to device. This is normally 10394 * called when a stacking relationship exists between the root 10395 * device and the device(a leaf device). 10396 */ 10397 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10398 struct net_device *dev) 10399 { 10400 if (rootdev->operstate == IF_OPER_DORMANT) 10401 netif_dormant_on(dev); 10402 else 10403 netif_dormant_off(dev); 10404 10405 if (rootdev->operstate == IF_OPER_TESTING) 10406 netif_testing_on(dev); 10407 else 10408 netif_testing_off(dev); 10409 10410 if (netif_carrier_ok(rootdev)) 10411 netif_carrier_on(dev); 10412 else 10413 netif_carrier_off(dev); 10414 } 10415 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10416 10417 static int netif_alloc_rx_queues(struct net_device *dev) 10418 { 10419 unsigned int i, count = dev->num_rx_queues; 10420 struct netdev_rx_queue *rx; 10421 size_t sz = count * sizeof(*rx); 10422 int err = 0; 10423 10424 BUG_ON(count < 1); 10425 10426 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10427 if (!rx) 10428 return -ENOMEM; 10429 10430 dev->_rx = rx; 10431 10432 for (i = 0; i < count; i++) { 10433 rx[i].dev = dev; 10434 10435 /* XDP RX-queue setup */ 10436 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10437 if (err < 0) 10438 goto err_rxq_info; 10439 } 10440 return 0; 10441 10442 err_rxq_info: 10443 /* Rollback successful reg's and free other resources */ 10444 while (i--) 10445 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10446 kvfree(dev->_rx); 10447 dev->_rx = NULL; 10448 return err; 10449 } 10450 10451 static void netif_free_rx_queues(struct net_device *dev) 10452 { 10453 unsigned int i, count = dev->num_rx_queues; 10454 10455 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10456 if (!dev->_rx) 10457 return; 10458 10459 for (i = 0; i < count; i++) 10460 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10461 10462 kvfree(dev->_rx); 10463 } 10464 10465 static void netdev_init_one_queue(struct net_device *dev, 10466 struct netdev_queue *queue, void *_unused) 10467 { 10468 /* Initialize queue lock */ 10469 spin_lock_init(&queue->_xmit_lock); 10470 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10471 queue->xmit_lock_owner = -1; 10472 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10473 queue->dev = dev; 10474 #ifdef CONFIG_BQL 10475 dql_init(&queue->dql, HZ); 10476 #endif 10477 } 10478 10479 static void netif_free_tx_queues(struct net_device *dev) 10480 { 10481 kvfree(dev->_tx); 10482 } 10483 10484 static int netif_alloc_netdev_queues(struct net_device *dev) 10485 { 10486 unsigned int count = dev->num_tx_queues; 10487 struct netdev_queue *tx; 10488 size_t sz = count * sizeof(*tx); 10489 10490 if (count < 1 || count > 0xffff) 10491 return -EINVAL; 10492 10493 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10494 if (!tx) 10495 return -ENOMEM; 10496 10497 dev->_tx = tx; 10498 10499 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10500 spin_lock_init(&dev->tx_global_lock); 10501 10502 return 0; 10503 } 10504 10505 void netif_tx_stop_all_queues(struct net_device *dev) 10506 { 10507 unsigned int i; 10508 10509 for (i = 0; i < dev->num_tx_queues; i++) { 10510 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10511 10512 netif_tx_stop_queue(txq); 10513 } 10514 } 10515 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10516 10517 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10518 { 10519 void __percpu *v; 10520 10521 /* Drivers implementing ndo_get_peer_dev must support tstat 10522 * accounting, so that skb_do_redirect() can bump the dev's 10523 * RX stats upon network namespace switch. 10524 */ 10525 if (dev->netdev_ops->ndo_get_peer_dev && 10526 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10527 return -EOPNOTSUPP; 10528 10529 switch (dev->pcpu_stat_type) { 10530 case NETDEV_PCPU_STAT_NONE: 10531 return 0; 10532 case NETDEV_PCPU_STAT_LSTATS: 10533 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10534 break; 10535 case NETDEV_PCPU_STAT_TSTATS: 10536 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10537 break; 10538 case NETDEV_PCPU_STAT_DSTATS: 10539 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10540 break; 10541 default: 10542 return -EINVAL; 10543 } 10544 10545 return v ? 0 : -ENOMEM; 10546 } 10547 10548 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10549 { 10550 switch (dev->pcpu_stat_type) { 10551 case NETDEV_PCPU_STAT_NONE: 10552 return; 10553 case NETDEV_PCPU_STAT_LSTATS: 10554 free_percpu(dev->lstats); 10555 break; 10556 case NETDEV_PCPU_STAT_TSTATS: 10557 free_percpu(dev->tstats); 10558 break; 10559 case NETDEV_PCPU_STAT_DSTATS: 10560 free_percpu(dev->dstats); 10561 break; 10562 } 10563 } 10564 10565 static void netdev_free_phy_link_topology(struct net_device *dev) 10566 { 10567 struct phy_link_topology *topo = dev->link_topo; 10568 10569 if (IS_ENABLED(CONFIG_PHYLIB) && topo) { 10570 xa_destroy(&topo->phys); 10571 kfree(topo); 10572 dev->link_topo = NULL; 10573 } 10574 } 10575 10576 /** 10577 * register_netdevice() - register a network device 10578 * @dev: device to register 10579 * 10580 * Take a prepared network device structure and make it externally accessible. 10581 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10582 * Callers must hold the rtnl lock - you may want register_netdev() 10583 * instead of this. 10584 */ 10585 int register_netdevice(struct net_device *dev) 10586 { 10587 int ret; 10588 struct net *net = dev_net(dev); 10589 10590 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10591 NETDEV_FEATURE_COUNT); 10592 BUG_ON(dev_boot_phase); 10593 ASSERT_RTNL(); 10594 10595 might_sleep(); 10596 10597 /* When net_device's are persistent, this will be fatal. */ 10598 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10599 BUG_ON(!net); 10600 10601 ret = ethtool_check_ops(dev->ethtool_ops); 10602 if (ret) 10603 return ret; 10604 10605 /* rss ctx ID 0 is reserved for the default context, start from 1 */ 10606 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1); 10607 mutex_init(&dev->ethtool->rss_lock); 10608 10609 spin_lock_init(&dev->addr_list_lock); 10610 netdev_set_addr_lockdep_class(dev); 10611 10612 ret = dev_get_valid_name(net, dev, dev->name); 10613 if (ret < 0) 10614 goto out; 10615 10616 ret = -ENOMEM; 10617 dev->name_node = netdev_name_node_head_alloc(dev); 10618 if (!dev->name_node) 10619 goto out; 10620 10621 /* Init, if this function is available */ 10622 if (dev->netdev_ops->ndo_init) { 10623 ret = dev->netdev_ops->ndo_init(dev); 10624 if (ret) { 10625 if (ret > 0) 10626 ret = -EIO; 10627 goto err_free_name; 10628 } 10629 } 10630 10631 if (((dev->hw_features | dev->features) & 10632 NETIF_F_HW_VLAN_CTAG_FILTER) && 10633 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10634 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10635 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10636 ret = -EINVAL; 10637 goto err_uninit; 10638 } 10639 10640 ret = netdev_do_alloc_pcpu_stats(dev); 10641 if (ret) 10642 goto err_uninit; 10643 10644 ret = dev_index_reserve(net, dev->ifindex); 10645 if (ret < 0) 10646 goto err_free_pcpu; 10647 dev->ifindex = ret; 10648 10649 /* Transfer changeable features to wanted_features and enable 10650 * software offloads (GSO and GRO). 10651 */ 10652 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10653 dev->features |= NETIF_F_SOFT_FEATURES; 10654 10655 if (dev->udp_tunnel_nic_info) { 10656 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10657 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10658 } 10659 10660 dev->wanted_features = dev->features & dev->hw_features; 10661 10662 if (!(dev->flags & IFF_LOOPBACK)) 10663 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10664 10665 /* If IPv4 TCP segmentation offload is supported we should also 10666 * allow the device to enable segmenting the frame with the option 10667 * of ignoring a static IP ID value. This doesn't enable the 10668 * feature itself but allows the user to enable it later. 10669 */ 10670 if (dev->hw_features & NETIF_F_TSO) 10671 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10672 if (dev->vlan_features & NETIF_F_TSO) 10673 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10674 if (dev->mpls_features & NETIF_F_TSO) 10675 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10676 if (dev->hw_enc_features & NETIF_F_TSO) 10677 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10678 10679 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10680 */ 10681 dev->vlan_features |= NETIF_F_HIGHDMA; 10682 10683 /* Make NETIF_F_SG inheritable to tunnel devices. 10684 */ 10685 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10686 10687 /* Make NETIF_F_SG inheritable to MPLS. 10688 */ 10689 dev->mpls_features |= NETIF_F_SG; 10690 10691 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10692 ret = notifier_to_errno(ret); 10693 if (ret) 10694 goto err_ifindex_release; 10695 10696 ret = netdev_register_kobject(dev); 10697 10698 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); 10699 10700 if (ret) 10701 goto err_uninit_notify; 10702 10703 __netdev_update_features(dev); 10704 10705 /* 10706 * Default initial state at registry is that the 10707 * device is present. 10708 */ 10709 10710 set_bit(__LINK_STATE_PRESENT, &dev->state); 10711 10712 linkwatch_init_dev(dev); 10713 10714 dev_init_scheduler(dev); 10715 10716 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10717 list_netdevice(dev); 10718 10719 add_device_randomness(dev->dev_addr, dev->addr_len); 10720 10721 /* If the device has permanent device address, driver should 10722 * set dev_addr and also addr_assign_type should be set to 10723 * NET_ADDR_PERM (default value). 10724 */ 10725 if (dev->addr_assign_type == NET_ADDR_PERM) 10726 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10727 10728 /* Notify protocols, that a new device appeared. */ 10729 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10730 ret = notifier_to_errno(ret); 10731 if (ret) { 10732 /* Expect explicit free_netdev() on failure */ 10733 dev->needs_free_netdev = false; 10734 unregister_netdevice_queue(dev, NULL); 10735 goto out; 10736 } 10737 /* 10738 * Prevent userspace races by waiting until the network 10739 * device is fully setup before sending notifications. 10740 */ 10741 if (!dev->rtnl_link_ops || 10742 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10743 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10744 10745 out: 10746 return ret; 10747 10748 err_uninit_notify: 10749 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10750 err_ifindex_release: 10751 dev_index_release(net, dev->ifindex); 10752 err_free_pcpu: 10753 netdev_do_free_pcpu_stats(dev); 10754 err_uninit: 10755 if (dev->netdev_ops->ndo_uninit) 10756 dev->netdev_ops->ndo_uninit(dev); 10757 if (dev->priv_destructor) 10758 dev->priv_destructor(dev); 10759 err_free_name: 10760 netdev_name_node_free(dev->name_node); 10761 goto out; 10762 } 10763 EXPORT_SYMBOL(register_netdevice); 10764 10765 /* Initialize the core of a dummy net device. 10766 * The setup steps dummy netdevs need which normal netdevs get by going 10767 * through register_netdevice(). 10768 */ 10769 static void init_dummy_netdev(struct net_device *dev) 10770 { 10771 /* make sure we BUG if trying to hit standard 10772 * register/unregister code path 10773 */ 10774 dev->reg_state = NETREG_DUMMY; 10775 10776 /* a dummy interface is started by default */ 10777 set_bit(__LINK_STATE_PRESENT, &dev->state); 10778 set_bit(__LINK_STATE_START, &dev->state); 10779 10780 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10781 * because users of this 'device' dont need to change 10782 * its refcount. 10783 */ 10784 } 10785 10786 /** 10787 * register_netdev - register a network device 10788 * @dev: device to register 10789 * 10790 * Take a completed network device structure and add it to the kernel 10791 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10792 * chain. 0 is returned on success. A negative errno code is returned 10793 * on a failure to set up the device, or if the name is a duplicate. 10794 * 10795 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10796 * and expands the device name if you passed a format string to 10797 * alloc_netdev. 10798 */ 10799 int register_netdev(struct net_device *dev) 10800 { 10801 struct net *net = dev_net(dev); 10802 int err; 10803 10804 if (rtnl_net_lock_killable(net)) 10805 return -EINTR; 10806 10807 err = register_netdevice(dev); 10808 10809 rtnl_net_unlock(net); 10810 10811 return err; 10812 } 10813 EXPORT_SYMBOL(register_netdev); 10814 10815 int netdev_refcnt_read(const struct net_device *dev) 10816 { 10817 #ifdef CONFIG_PCPU_DEV_REFCNT 10818 int i, refcnt = 0; 10819 10820 for_each_possible_cpu(i) 10821 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10822 return refcnt; 10823 #else 10824 return refcount_read(&dev->dev_refcnt); 10825 #endif 10826 } 10827 EXPORT_SYMBOL(netdev_refcnt_read); 10828 10829 int netdev_unregister_timeout_secs __read_mostly = 10; 10830 10831 #define WAIT_REFS_MIN_MSECS 1 10832 #define WAIT_REFS_MAX_MSECS 250 10833 /** 10834 * netdev_wait_allrefs_any - wait until all references are gone. 10835 * @list: list of net_devices to wait on 10836 * 10837 * This is called when unregistering network devices. 10838 * 10839 * Any protocol or device that holds a reference should register 10840 * for netdevice notification, and cleanup and put back the 10841 * reference if they receive an UNREGISTER event. 10842 * We can get stuck here if buggy protocols don't correctly 10843 * call dev_put. 10844 */ 10845 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10846 { 10847 unsigned long rebroadcast_time, warning_time; 10848 struct net_device *dev; 10849 int wait = 0; 10850 10851 rebroadcast_time = warning_time = jiffies; 10852 10853 list_for_each_entry(dev, list, todo_list) 10854 if (netdev_refcnt_read(dev) == 1) 10855 return dev; 10856 10857 while (true) { 10858 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10859 rtnl_lock(); 10860 10861 /* Rebroadcast unregister notification */ 10862 list_for_each_entry(dev, list, todo_list) 10863 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10864 10865 __rtnl_unlock(); 10866 rcu_barrier(); 10867 rtnl_lock(); 10868 10869 list_for_each_entry(dev, list, todo_list) 10870 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10871 &dev->state)) { 10872 /* We must not have linkwatch events 10873 * pending on unregister. If this 10874 * happens, we simply run the queue 10875 * unscheduled, resulting in a noop 10876 * for this device. 10877 */ 10878 linkwatch_run_queue(); 10879 break; 10880 } 10881 10882 __rtnl_unlock(); 10883 10884 rebroadcast_time = jiffies; 10885 } 10886 10887 rcu_barrier(); 10888 10889 if (!wait) { 10890 wait = WAIT_REFS_MIN_MSECS; 10891 } else { 10892 msleep(wait); 10893 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10894 } 10895 10896 list_for_each_entry(dev, list, todo_list) 10897 if (netdev_refcnt_read(dev) == 1) 10898 return dev; 10899 10900 if (time_after(jiffies, warning_time + 10901 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10902 list_for_each_entry(dev, list, todo_list) { 10903 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10904 dev->name, netdev_refcnt_read(dev)); 10905 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10906 } 10907 10908 warning_time = jiffies; 10909 } 10910 } 10911 } 10912 10913 /* The sequence is: 10914 * 10915 * rtnl_lock(); 10916 * ... 10917 * register_netdevice(x1); 10918 * register_netdevice(x2); 10919 * ... 10920 * unregister_netdevice(y1); 10921 * unregister_netdevice(y2); 10922 * ... 10923 * rtnl_unlock(); 10924 * free_netdev(y1); 10925 * free_netdev(y2); 10926 * 10927 * We are invoked by rtnl_unlock(). 10928 * This allows us to deal with problems: 10929 * 1) We can delete sysfs objects which invoke hotplug 10930 * without deadlocking with linkwatch via keventd. 10931 * 2) Since we run with the RTNL semaphore not held, we can sleep 10932 * safely in order to wait for the netdev refcnt to drop to zero. 10933 * 10934 * We must not return until all unregister events added during 10935 * the interval the lock was held have been completed. 10936 */ 10937 void netdev_run_todo(void) 10938 { 10939 struct net_device *dev, *tmp; 10940 struct list_head list; 10941 int cnt; 10942 #ifdef CONFIG_LOCKDEP 10943 struct list_head unlink_list; 10944 10945 list_replace_init(&net_unlink_list, &unlink_list); 10946 10947 while (!list_empty(&unlink_list)) { 10948 struct net_device *dev = list_first_entry(&unlink_list, 10949 struct net_device, 10950 unlink_list); 10951 list_del_init(&dev->unlink_list); 10952 dev->nested_level = dev->lower_level - 1; 10953 } 10954 #endif 10955 10956 /* Snapshot list, allow later requests */ 10957 list_replace_init(&net_todo_list, &list); 10958 10959 __rtnl_unlock(); 10960 10961 /* Wait for rcu callbacks to finish before next phase */ 10962 if (!list_empty(&list)) 10963 rcu_barrier(); 10964 10965 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10966 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10967 netdev_WARN(dev, "run_todo but not unregistering\n"); 10968 list_del(&dev->todo_list); 10969 continue; 10970 } 10971 10972 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); 10973 linkwatch_sync_dev(dev); 10974 } 10975 10976 cnt = 0; 10977 while (!list_empty(&list)) { 10978 dev = netdev_wait_allrefs_any(&list); 10979 list_del(&dev->todo_list); 10980 10981 /* paranoia */ 10982 BUG_ON(netdev_refcnt_read(dev) != 1); 10983 BUG_ON(!list_empty(&dev->ptype_all)); 10984 BUG_ON(!list_empty(&dev->ptype_specific)); 10985 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10986 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10987 10988 netdev_do_free_pcpu_stats(dev); 10989 if (dev->priv_destructor) 10990 dev->priv_destructor(dev); 10991 if (dev->needs_free_netdev) 10992 free_netdev(dev); 10993 10994 cnt++; 10995 10996 /* Free network device */ 10997 kobject_put(&dev->dev.kobj); 10998 } 10999 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) 11000 wake_up(&netdev_unregistering_wq); 11001 } 11002 11003 /* Collate per-cpu network dstats statistics 11004 * 11005 * Read per-cpu network statistics from dev->dstats and populate the related 11006 * fields in @s. 11007 */ 11008 static void dev_fetch_dstats(struct rtnl_link_stats64 *s, 11009 const struct pcpu_dstats __percpu *dstats) 11010 { 11011 int cpu; 11012 11013 for_each_possible_cpu(cpu) { 11014 u64 rx_packets, rx_bytes, rx_drops; 11015 u64 tx_packets, tx_bytes, tx_drops; 11016 const struct pcpu_dstats *stats; 11017 unsigned int start; 11018 11019 stats = per_cpu_ptr(dstats, cpu); 11020 do { 11021 start = u64_stats_fetch_begin(&stats->syncp); 11022 rx_packets = u64_stats_read(&stats->rx_packets); 11023 rx_bytes = u64_stats_read(&stats->rx_bytes); 11024 rx_drops = u64_stats_read(&stats->rx_drops); 11025 tx_packets = u64_stats_read(&stats->tx_packets); 11026 tx_bytes = u64_stats_read(&stats->tx_bytes); 11027 tx_drops = u64_stats_read(&stats->tx_drops); 11028 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11029 11030 s->rx_packets += rx_packets; 11031 s->rx_bytes += rx_bytes; 11032 s->rx_dropped += rx_drops; 11033 s->tx_packets += tx_packets; 11034 s->tx_bytes += tx_bytes; 11035 s->tx_dropped += tx_drops; 11036 } 11037 } 11038 11039 /* ndo_get_stats64 implementation for dtstats-based accounting. 11040 * 11041 * Populate @s from dev->stats and dev->dstats. This is used internally by the 11042 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection. 11043 */ 11044 static void dev_get_dstats64(const struct net_device *dev, 11045 struct rtnl_link_stats64 *s) 11046 { 11047 netdev_stats_to_stats64(s, &dev->stats); 11048 dev_fetch_dstats(s, dev->dstats); 11049 } 11050 11051 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 11052 * all the same fields in the same order as net_device_stats, with only 11053 * the type differing, but rtnl_link_stats64 may have additional fields 11054 * at the end for newer counters. 11055 */ 11056 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 11057 const struct net_device_stats *netdev_stats) 11058 { 11059 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 11060 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 11061 u64 *dst = (u64 *)stats64; 11062 11063 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 11064 for (i = 0; i < n; i++) 11065 dst[i] = (unsigned long)atomic_long_read(&src[i]); 11066 /* zero out counters that only exist in rtnl_link_stats64 */ 11067 memset((char *)stats64 + n * sizeof(u64), 0, 11068 sizeof(*stats64) - n * sizeof(u64)); 11069 } 11070 EXPORT_SYMBOL(netdev_stats_to_stats64); 11071 11072 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 11073 struct net_device *dev) 11074 { 11075 struct net_device_core_stats __percpu *p; 11076 11077 p = alloc_percpu_gfp(struct net_device_core_stats, 11078 GFP_ATOMIC | __GFP_NOWARN); 11079 11080 if (p && cmpxchg(&dev->core_stats, NULL, p)) 11081 free_percpu(p); 11082 11083 /* This READ_ONCE() pairs with the cmpxchg() above */ 11084 return READ_ONCE(dev->core_stats); 11085 } 11086 11087 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 11088 { 11089 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11090 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 11091 unsigned long __percpu *field; 11092 11093 if (unlikely(!p)) { 11094 p = netdev_core_stats_alloc(dev); 11095 if (!p) 11096 return; 11097 } 11098 11099 field = (unsigned long __percpu *)((void __percpu *)p + offset); 11100 this_cpu_inc(*field); 11101 } 11102 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 11103 11104 /** 11105 * dev_get_stats - get network device statistics 11106 * @dev: device to get statistics from 11107 * @storage: place to store stats 11108 * 11109 * Get network statistics from device. Return @storage. 11110 * The device driver may provide its own method by setting 11111 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 11112 * otherwise the internal statistics structure is used. 11113 */ 11114 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 11115 struct rtnl_link_stats64 *storage) 11116 { 11117 const struct net_device_ops *ops = dev->netdev_ops; 11118 const struct net_device_core_stats __percpu *p; 11119 11120 if (ops->ndo_get_stats64) { 11121 memset(storage, 0, sizeof(*storage)); 11122 ops->ndo_get_stats64(dev, storage); 11123 } else if (ops->ndo_get_stats) { 11124 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 11125 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { 11126 dev_get_tstats64(dev, storage); 11127 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) { 11128 dev_get_dstats64(dev, storage); 11129 } else { 11130 netdev_stats_to_stats64(storage, &dev->stats); 11131 } 11132 11133 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 11134 p = READ_ONCE(dev->core_stats); 11135 if (p) { 11136 const struct net_device_core_stats *core_stats; 11137 int i; 11138 11139 for_each_possible_cpu(i) { 11140 core_stats = per_cpu_ptr(p, i); 11141 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 11142 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 11143 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 11144 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 11145 } 11146 } 11147 return storage; 11148 } 11149 EXPORT_SYMBOL(dev_get_stats); 11150 11151 /** 11152 * dev_fetch_sw_netstats - get per-cpu network device statistics 11153 * @s: place to store stats 11154 * @netstats: per-cpu network stats to read from 11155 * 11156 * Read per-cpu network statistics and populate the related fields in @s. 11157 */ 11158 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 11159 const struct pcpu_sw_netstats __percpu *netstats) 11160 { 11161 int cpu; 11162 11163 for_each_possible_cpu(cpu) { 11164 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 11165 const struct pcpu_sw_netstats *stats; 11166 unsigned int start; 11167 11168 stats = per_cpu_ptr(netstats, cpu); 11169 do { 11170 start = u64_stats_fetch_begin(&stats->syncp); 11171 rx_packets = u64_stats_read(&stats->rx_packets); 11172 rx_bytes = u64_stats_read(&stats->rx_bytes); 11173 tx_packets = u64_stats_read(&stats->tx_packets); 11174 tx_bytes = u64_stats_read(&stats->tx_bytes); 11175 } while (u64_stats_fetch_retry(&stats->syncp, start)); 11176 11177 s->rx_packets += rx_packets; 11178 s->rx_bytes += rx_bytes; 11179 s->tx_packets += tx_packets; 11180 s->tx_bytes += tx_bytes; 11181 } 11182 } 11183 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 11184 11185 /** 11186 * dev_get_tstats64 - ndo_get_stats64 implementation 11187 * @dev: device to get statistics from 11188 * @s: place to store stats 11189 * 11190 * Populate @s from dev->stats and dev->tstats. Can be used as 11191 * ndo_get_stats64() callback. 11192 */ 11193 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 11194 { 11195 netdev_stats_to_stats64(s, &dev->stats); 11196 dev_fetch_sw_netstats(s, dev->tstats); 11197 } 11198 EXPORT_SYMBOL_GPL(dev_get_tstats64); 11199 11200 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 11201 { 11202 struct netdev_queue *queue = dev_ingress_queue(dev); 11203 11204 #ifdef CONFIG_NET_CLS_ACT 11205 if (queue) 11206 return queue; 11207 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 11208 if (!queue) 11209 return NULL; 11210 netdev_init_one_queue(dev, queue, NULL); 11211 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 11212 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 11213 rcu_assign_pointer(dev->ingress_queue, queue); 11214 #endif 11215 return queue; 11216 } 11217 11218 static const struct ethtool_ops default_ethtool_ops; 11219 11220 void netdev_set_default_ethtool_ops(struct net_device *dev, 11221 const struct ethtool_ops *ops) 11222 { 11223 if (dev->ethtool_ops == &default_ethtool_ops) 11224 dev->ethtool_ops = ops; 11225 } 11226 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 11227 11228 /** 11229 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 11230 * @dev: netdev to enable the IRQ coalescing on 11231 * 11232 * Sets a conservative default for SW IRQ coalescing. Users can use 11233 * sysfs attributes to override the default values. 11234 */ 11235 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 11236 { 11237 WARN_ON(dev->reg_state == NETREG_REGISTERED); 11238 11239 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 11240 netdev_set_gro_flush_timeout(dev, 20000); 11241 netdev_set_defer_hard_irqs(dev, 1); 11242 } 11243 } 11244 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 11245 11246 /** 11247 * alloc_netdev_mqs - allocate network device 11248 * @sizeof_priv: size of private data to allocate space for 11249 * @name: device name format string 11250 * @name_assign_type: origin of device name 11251 * @setup: callback to initialize device 11252 * @txqs: the number of TX subqueues to allocate 11253 * @rxqs: the number of RX subqueues to allocate 11254 * 11255 * Allocates a struct net_device with private data area for driver use 11256 * and performs basic initialization. Also allocates subqueue structs 11257 * for each queue on the device. 11258 */ 11259 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 11260 unsigned char name_assign_type, 11261 void (*setup)(struct net_device *), 11262 unsigned int txqs, unsigned int rxqs) 11263 { 11264 struct net_device *dev; 11265 size_t napi_config_sz; 11266 unsigned int maxqs; 11267 11268 BUG_ON(strlen(name) >= sizeof(dev->name)); 11269 11270 if (txqs < 1) { 11271 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 11272 return NULL; 11273 } 11274 11275 if (rxqs < 1) { 11276 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 11277 return NULL; 11278 } 11279 11280 maxqs = max(txqs, rxqs); 11281 11282 dev = kvzalloc(struct_size(dev, priv, sizeof_priv), 11283 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 11284 if (!dev) 11285 return NULL; 11286 11287 dev->priv_len = sizeof_priv; 11288 11289 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 11290 #ifdef CONFIG_PCPU_DEV_REFCNT 11291 dev->pcpu_refcnt = alloc_percpu(int); 11292 if (!dev->pcpu_refcnt) 11293 goto free_dev; 11294 __dev_hold(dev); 11295 #else 11296 refcount_set(&dev->dev_refcnt, 1); 11297 #endif 11298 11299 if (dev_addr_init(dev)) 11300 goto free_pcpu; 11301 11302 dev_mc_init(dev); 11303 dev_uc_init(dev); 11304 11305 dev_net_set(dev, &init_net); 11306 11307 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 11308 dev->xdp_zc_max_segs = 1; 11309 dev->gso_max_segs = GSO_MAX_SEGS; 11310 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 11311 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 11312 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 11313 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 11314 dev->tso_max_segs = TSO_MAX_SEGS; 11315 dev->upper_level = 1; 11316 dev->lower_level = 1; 11317 #ifdef CONFIG_LOCKDEP 11318 dev->nested_level = 0; 11319 INIT_LIST_HEAD(&dev->unlink_list); 11320 #endif 11321 11322 INIT_LIST_HEAD(&dev->napi_list); 11323 INIT_LIST_HEAD(&dev->unreg_list); 11324 INIT_LIST_HEAD(&dev->close_list); 11325 INIT_LIST_HEAD(&dev->link_watch_list); 11326 INIT_LIST_HEAD(&dev->adj_list.upper); 11327 INIT_LIST_HEAD(&dev->adj_list.lower); 11328 INIT_LIST_HEAD(&dev->ptype_all); 11329 INIT_LIST_HEAD(&dev->ptype_specific); 11330 INIT_LIST_HEAD(&dev->net_notifier_list); 11331 #ifdef CONFIG_NET_SCHED 11332 hash_init(dev->qdisc_hash); 11333 #endif 11334 11335 mutex_init(&dev->lock); 11336 11337 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 11338 setup(dev); 11339 11340 if (!dev->tx_queue_len) { 11341 dev->priv_flags |= IFF_NO_QUEUE; 11342 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 11343 } 11344 11345 dev->num_tx_queues = txqs; 11346 dev->real_num_tx_queues = txqs; 11347 if (netif_alloc_netdev_queues(dev)) 11348 goto free_all; 11349 11350 dev->num_rx_queues = rxqs; 11351 dev->real_num_rx_queues = rxqs; 11352 if (netif_alloc_rx_queues(dev)) 11353 goto free_all; 11354 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT); 11355 if (!dev->ethtool) 11356 goto free_all; 11357 11358 napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config)); 11359 dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT); 11360 if (!dev->napi_config) 11361 goto free_all; 11362 11363 strscpy(dev->name, name); 11364 dev->name_assign_type = name_assign_type; 11365 dev->group = INIT_NETDEV_GROUP; 11366 if (!dev->ethtool_ops) 11367 dev->ethtool_ops = &default_ethtool_ops; 11368 11369 nf_hook_netdev_init(dev); 11370 11371 return dev; 11372 11373 free_all: 11374 free_netdev(dev); 11375 return NULL; 11376 11377 free_pcpu: 11378 #ifdef CONFIG_PCPU_DEV_REFCNT 11379 free_percpu(dev->pcpu_refcnt); 11380 free_dev: 11381 #endif 11382 kvfree(dev); 11383 return NULL; 11384 } 11385 EXPORT_SYMBOL(alloc_netdev_mqs); 11386 11387 /** 11388 * free_netdev - free network device 11389 * @dev: device 11390 * 11391 * This function does the last stage of destroying an allocated device 11392 * interface. The reference to the device object is released. If this 11393 * is the last reference then it will be freed.Must be called in process 11394 * context. 11395 */ 11396 void free_netdev(struct net_device *dev) 11397 { 11398 struct napi_struct *p, *n; 11399 11400 might_sleep(); 11401 11402 /* When called immediately after register_netdevice() failed the unwind 11403 * handling may still be dismantling the device. Handle that case by 11404 * deferring the free. 11405 */ 11406 if (dev->reg_state == NETREG_UNREGISTERING) { 11407 ASSERT_RTNL(); 11408 dev->needs_free_netdev = true; 11409 return; 11410 } 11411 11412 mutex_destroy(&dev->lock); 11413 11414 kfree(dev->ethtool); 11415 netif_free_tx_queues(dev); 11416 netif_free_rx_queues(dev); 11417 11418 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 11419 11420 /* Flush device addresses */ 11421 dev_addr_flush(dev); 11422 11423 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 11424 netif_napi_del(p); 11425 11426 kvfree(dev->napi_config); 11427 11428 ref_tracker_dir_exit(&dev->refcnt_tracker); 11429 #ifdef CONFIG_PCPU_DEV_REFCNT 11430 free_percpu(dev->pcpu_refcnt); 11431 dev->pcpu_refcnt = NULL; 11432 #endif 11433 free_percpu(dev->core_stats); 11434 dev->core_stats = NULL; 11435 free_percpu(dev->xdp_bulkq); 11436 dev->xdp_bulkq = NULL; 11437 11438 netdev_free_phy_link_topology(dev); 11439 11440 /* Compatibility with error handling in drivers */ 11441 if (dev->reg_state == NETREG_UNINITIALIZED || 11442 dev->reg_state == NETREG_DUMMY) { 11443 kvfree(dev); 11444 return; 11445 } 11446 11447 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 11448 WRITE_ONCE(dev->reg_state, NETREG_RELEASED); 11449 11450 /* will free via device release */ 11451 put_device(&dev->dev); 11452 } 11453 EXPORT_SYMBOL(free_netdev); 11454 11455 /** 11456 * alloc_netdev_dummy - Allocate and initialize a dummy net device. 11457 * @sizeof_priv: size of private data to allocate space for 11458 * 11459 * Return: the allocated net_device on success, NULL otherwise 11460 */ 11461 struct net_device *alloc_netdev_dummy(int sizeof_priv) 11462 { 11463 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN, 11464 init_dummy_netdev); 11465 } 11466 EXPORT_SYMBOL_GPL(alloc_netdev_dummy); 11467 11468 /** 11469 * synchronize_net - Synchronize with packet receive processing 11470 * 11471 * Wait for packets currently being received to be done. 11472 * Does not block later packets from starting. 11473 */ 11474 void synchronize_net(void) 11475 { 11476 might_sleep(); 11477 if (rtnl_is_locked()) 11478 synchronize_rcu_expedited(); 11479 else 11480 synchronize_rcu(); 11481 } 11482 EXPORT_SYMBOL(synchronize_net); 11483 11484 static void netdev_rss_contexts_free(struct net_device *dev) 11485 { 11486 struct ethtool_rxfh_context *ctx; 11487 unsigned long context; 11488 11489 mutex_lock(&dev->ethtool->rss_lock); 11490 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) { 11491 struct ethtool_rxfh_param rxfh; 11492 11493 rxfh.indir = ethtool_rxfh_context_indir(ctx); 11494 rxfh.key = ethtool_rxfh_context_key(ctx); 11495 rxfh.hfunc = ctx->hfunc; 11496 rxfh.input_xfrm = ctx->input_xfrm; 11497 rxfh.rss_context = context; 11498 rxfh.rss_delete = true; 11499 11500 xa_erase(&dev->ethtool->rss_ctx, context); 11501 if (dev->ethtool_ops->create_rxfh_context) 11502 dev->ethtool_ops->remove_rxfh_context(dev, ctx, 11503 context, NULL); 11504 else 11505 dev->ethtool_ops->set_rxfh(dev, &rxfh, NULL); 11506 kfree(ctx); 11507 } 11508 xa_destroy(&dev->ethtool->rss_ctx); 11509 mutex_unlock(&dev->ethtool->rss_lock); 11510 } 11511 11512 /** 11513 * unregister_netdevice_queue - remove device from the kernel 11514 * @dev: device 11515 * @head: list 11516 * 11517 * This function shuts down a device interface and removes it 11518 * from the kernel tables. 11519 * If head not NULL, device is queued to be unregistered later. 11520 * 11521 * Callers must hold the rtnl semaphore. You may want 11522 * unregister_netdev() instead of this. 11523 */ 11524 11525 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11526 { 11527 ASSERT_RTNL(); 11528 11529 if (head) { 11530 list_move_tail(&dev->unreg_list, head); 11531 } else { 11532 LIST_HEAD(single); 11533 11534 list_add(&dev->unreg_list, &single); 11535 unregister_netdevice_many(&single); 11536 } 11537 } 11538 EXPORT_SYMBOL(unregister_netdevice_queue); 11539 11540 void unregister_netdevice_many_notify(struct list_head *head, 11541 u32 portid, const struct nlmsghdr *nlh) 11542 { 11543 struct net_device *dev, *tmp; 11544 LIST_HEAD(close_head); 11545 int cnt = 0; 11546 11547 BUG_ON(dev_boot_phase); 11548 ASSERT_RTNL(); 11549 11550 if (list_empty(head)) 11551 return; 11552 11553 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11554 /* Some devices call without registering 11555 * for initialization unwind. Remove those 11556 * devices and proceed with the remaining. 11557 */ 11558 if (dev->reg_state == NETREG_UNINITIALIZED) { 11559 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11560 dev->name, dev); 11561 11562 WARN_ON(1); 11563 list_del(&dev->unreg_list); 11564 continue; 11565 } 11566 dev->dismantle = true; 11567 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11568 } 11569 11570 /* If device is running, close it first. */ 11571 list_for_each_entry(dev, head, unreg_list) 11572 list_add_tail(&dev->close_list, &close_head); 11573 dev_close_many(&close_head, true); 11574 11575 list_for_each_entry(dev, head, unreg_list) { 11576 /* And unlink it from device chain. */ 11577 unlist_netdevice(dev); 11578 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); 11579 } 11580 flush_all_backlogs(); 11581 11582 synchronize_net(); 11583 11584 list_for_each_entry(dev, head, unreg_list) { 11585 struct sk_buff *skb = NULL; 11586 11587 /* Shutdown queueing discipline. */ 11588 dev_shutdown(dev); 11589 dev_tcx_uninstall(dev); 11590 dev_xdp_uninstall(dev); 11591 bpf_dev_bound_netdev_unregister(dev); 11592 dev_dmabuf_uninstall(dev); 11593 11594 netdev_offload_xstats_disable_all(dev); 11595 11596 /* Notify protocols, that we are about to destroy 11597 * this device. They should clean all the things. 11598 */ 11599 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11600 11601 if (!dev->rtnl_link_ops || 11602 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11603 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11604 GFP_KERNEL, NULL, 0, 11605 portid, nlh); 11606 11607 /* 11608 * Flush the unicast and multicast chains 11609 */ 11610 dev_uc_flush(dev); 11611 dev_mc_flush(dev); 11612 11613 netdev_name_node_alt_flush(dev); 11614 netdev_name_node_free(dev->name_node); 11615 11616 netdev_rss_contexts_free(dev); 11617 11618 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11619 11620 if (dev->netdev_ops->ndo_uninit) 11621 dev->netdev_ops->ndo_uninit(dev); 11622 11623 mutex_destroy(&dev->ethtool->rss_lock); 11624 11625 net_shaper_flush_netdev(dev); 11626 11627 if (skb) 11628 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11629 11630 /* Notifier chain MUST detach us all upper devices. */ 11631 WARN_ON(netdev_has_any_upper_dev(dev)); 11632 WARN_ON(netdev_has_any_lower_dev(dev)); 11633 11634 /* Remove entries from kobject tree */ 11635 netdev_unregister_kobject(dev); 11636 #ifdef CONFIG_XPS 11637 /* Remove XPS queueing entries */ 11638 netif_reset_xps_queues_gt(dev, 0); 11639 #endif 11640 } 11641 11642 synchronize_net(); 11643 11644 list_for_each_entry(dev, head, unreg_list) { 11645 netdev_put(dev, &dev->dev_registered_tracker); 11646 net_set_todo(dev); 11647 cnt++; 11648 } 11649 atomic_add(cnt, &dev_unreg_count); 11650 11651 list_del(head); 11652 } 11653 11654 /** 11655 * unregister_netdevice_many - unregister many devices 11656 * @head: list of devices 11657 * 11658 * Note: As most callers use a stack allocated list_head, 11659 * we force a list_del() to make sure stack won't be corrupted later. 11660 */ 11661 void unregister_netdevice_many(struct list_head *head) 11662 { 11663 unregister_netdevice_many_notify(head, 0, NULL); 11664 } 11665 EXPORT_SYMBOL(unregister_netdevice_many); 11666 11667 /** 11668 * unregister_netdev - remove device from the kernel 11669 * @dev: device 11670 * 11671 * This function shuts down a device interface and removes it 11672 * from the kernel tables. 11673 * 11674 * This is just a wrapper for unregister_netdevice that takes 11675 * the rtnl semaphore. In general you want to use this and not 11676 * unregister_netdevice. 11677 */ 11678 void unregister_netdev(struct net_device *dev) 11679 { 11680 struct net *net = dev_net(dev); 11681 11682 rtnl_net_lock(net); 11683 unregister_netdevice(dev); 11684 rtnl_net_unlock(net); 11685 } 11686 EXPORT_SYMBOL(unregister_netdev); 11687 11688 /** 11689 * __dev_change_net_namespace - move device to different nethost namespace 11690 * @dev: device 11691 * @net: network namespace 11692 * @pat: If not NULL name pattern to try if the current device name 11693 * is already taken in the destination network namespace. 11694 * @new_ifindex: If not zero, specifies device index in the target 11695 * namespace. 11696 * 11697 * This function shuts down a device interface and moves it 11698 * to a new network namespace. On success 0 is returned, on 11699 * a failure a netagive errno code is returned. 11700 * 11701 * Callers must hold the rtnl semaphore. 11702 */ 11703 11704 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11705 const char *pat, int new_ifindex) 11706 { 11707 struct netdev_name_node *name_node; 11708 struct net *net_old = dev_net(dev); 11709 char new_name[IFNAMSIZ] = {}; 11710 int err, new_nsid; 11711 11712 ASSERT_RTNL(); 11713 11714 /* Don't allow namespace local devices to be moved. */ 11715 err = -EINVAL; 11716 if (dev->netns_local) 11717 goto out; 11718 11719 /* Ensure the device has been registered */ 11720 if (dev->reg_state != NETREG_REGISTERED) 11721 goto out; 11722 11723 /* Get out if there is nothing todo */ 11724 err = 0; 11725 if (net_eq(net_old, net)) 11726 goto out; 11727 11728 /* Pick the destination device name, and ensure 11729 * we can use it in the destination network namespace. 11730 */ 11731 err = -EEXIST; 11732 if (netdev_name_in_use(net, dev->name)) { 11733 /* We get here if we can't use the current device name */ 11734 if (!pat) 11735 goto out; 11736 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11737 if (err < 0) 11738 goto out; 11739 } 11740 /* Check that none of the altnames conflicts. */ 11741 err = -EEXIST; 11742 netdev_for_each_altname(dev, name_node) 11743 if (netdev_name_in_use(net, name_node->name)) 11744 goto out; 11745 11746 /* Check that new_ifindex isn't used yet. */ 11747 if (new_ifindex) { 11748 err = dev_index_reserve(net, new_ifindex); 11749 if (err < 0) 11750 goto out; 11751 } else { 11752 /* If there is an ifindex conflict assign a new one */ 11753 err = dev_index_reserve(net, dev->ifindex); 11754 if (err == -EBUSY) 11755 err = dev_index_reserve(net, 0); 11756 if (err < 0) 11757 goto out; 11758 new_ifindex = err; 11759 } 11760 11761 /* 11762 * And now a mini version of register_netdevice unregister_netdevice. 11763 */ 11764 11765 /* If device is running close it first. */ 11766 dev_close(dev); 11767 11768 /* And unlink it from device chain */ 11769 unlist_netdevice(dev); 11770 11771 synchronize_net(); 11772 11773 /* Shutdown queueing discipline. */ 11774 dev_shutdown(dev); 11775 11776 /* Notify protocols, that we are about to destroy 11777 * this device. They should clean all the things. 11778 * 11779 * Note that dev->reg_state stays at NETREG_REGISTERED. 11780 * This is wanted because this way 8021q and macvlan know 11781 * the device is just moving and can keep their slaves up. 11782 */ 11783 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11784 rcu_barrier(); 11785 11786 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11787 11788 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11789 new_ifindex); 11790 11791 /* 11792 * Flush the unicast and multicast chains 11793 */ 11794 dev_uc_flush(dev); 11795 dev_mc_flush(dev); 11796 11797 /* Send a netdev-removed uevent to the old namespace */ 11798 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11799 netdev_adjacent_del_links(dev); 11800 11801 /* Move per-net netdevice notifiers that are following the netdevice */ 11802 move_netdevice_notifiers_dev_net(dev, net); 11803 11804 /* Actually switch the network namespace */ 11805 dev_net_set(dev, net); 11806 dev->ifindex = new_ifindex; 11807 11808 if (new_name[0]) { 11809 /* Rename the netdev to prepared name */ 11810 write_seqlock_bh(&netdev_rename_lock); 11811 strscpy(dev->name, new_name, IFNAMSIZ); 11812 write_sequnlock_bh(&netdev_rename_lock); 11813 } 11814 11815 /* Fixup kobjects */ 11816 dev_set_uevent_suppress(&dev->dev, 1); 11817 err = device_rename(&dev->dev, dev->name); 11818 dev_set_uevent_suppress(&dev->dev, 0); 11819 WARN_ON(err); 11820 11821 /* Send a netdev-add uevent to the new namespace */ 11822 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11823 netdev_adjacent_add_links(dev); 11824 11825 /* Adapt owner in case owning user namespace of target network 11826 * namespace is different from the original one. 11827 */ 11828 err = netdev_change_owner(dev, net_old, net); 11829 WARN_ON(err); 11830 11831 /* Add the device back in the hashes */ 11832 list_netdevice(dev); 11833 11834 /* Notify protocols, that a new device appeared. */ 11835 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11836 11837 /* 11838 * Prevent userspace races by waiting until the network 11839 * device is fully setup before sending notifications. 11840 */ 11841 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11842 11843 synchronize_net(); 11844 err = 0; 11845 out: 11846 return err; 11847 } 11848 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11849 11850 static int dev_cpu_dead(unsigned int oldcpu) 11851 { 11852 struct sk_buff **list_skb; 11853 struct sk_buff *skb; 11854 unsigned int cpu; 11855 struct softnet_data *sd, *oldsd, *remsd = NULL; 11856 11857 local_irq_disable(); 11858 cpu = smp_processor_id(); 11859 sd = &per_cpu(softnet_data, cpu); 11860 oldsd = &per_cpu(softnet_data, oldcpu); 11861 11862 /* Find end of our completion_queue. */ 11863 list_skb = &sd->completion_queue; 11864 while (*list_skb) 11865 list_skb = &(*list_skb)->next; 11866 /* Append completion queue from offline CPU. */ 11867 *list_skb = oldsd->completion_queue; 11868 oldsd->completion_queue = NULL; 11869 11870 /* Append output queue from offline CPU. */ 11871 if (oldsd->output_queue) { 11872 *sd->output_queue_tailp = oldsd->output_queue; 11873 sd->output_queue_tailp = oldsd->output_queue_tailp; 11874 oldsd->output_queue = NULL; 11875 oldsd->output_queue_tailp = &oldsd->output_queue; 11876 } 11877 /* Append NAPI poll list from offline CPU, with one exception : 11878 * process_backlog() must be called by cpu owning percpu backlog. 11879 * We properly handle process_queue & input_pkt_queue later. 11880 */ 11881 while (!list_empty(&oldsd->poll_list)) { 11882 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11883 struct napi_struct, 11884 poll_list); 11885 11886 list_del_init(&napi->poll_list); 11887 if (napi->poll == process_backlog) 11888 napi->state &= NAPIF_STATE_THREADED; 11889 else 11890 ____napi_schedule(sd, napi); 11891 } 11892 11893 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11894 local_irq_enable(); 11895 11896 if (!use_backlog_threads()) { 11897 #ifdef CONFIG_RPS 11898 remsd = oldsd->rps_ipi_list; 11899 oldsd->rps_ipi_list = NULL; 11900 #endif 11901 /* send out pending IPI's on offline CPU */ 11902 net_rps_send_ipi(remsd); 11903 } 11904 11905 /* Process offline CPU's input_pkt_queue */ 11906 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11907 netif_rx(skb); 11908 rps_input_queue_head_incr(oldsd); 11909 } 11910 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11911 netif_rx(skb); 11912 rps_input_queue_head_incr(oldsd); 11913 } 11914 11915 return 0; 11916 } 11917 11918 /** 11919 * netdev_increment_features - increment feature set by one 11920 * @all: current feature set 11921 * @one: new feature set 11922 * @mask: mask feature set 11923 * 11924 * Computes a new feature set after adding a device with feature set 11925 * @one to the master device with current feature set @all. Will not 11926 * enable anything that is off in @mask. Returns the new feature set. 11927 */ 11928 netdev_features_t netdev_increment_features(netdev_features_t all, 11929 netdev_features_t one, netdev_features_t mask) 11930 { 11931 if (mask & NETIF_F_HW_CSUM) 11932 mask |= NETIF_F_CSUM_MASK; 11933 mask |= NETIF_F_VLAN_CHALLENGED; 11934 11935 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11936 all &= one | ~NETIF_F_ALL_FOR_ALL; 11937 11938 /* If one device supports hw checksumming, set for all. */ 11939 if (all & NETIF_F_HW_CSUM) 11940 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11941 11942 return all; 11943 } 11944 EXPORT_SYMBOL(netdev_increment_features); 11945 11946 static struct hlist_head * __net_init netdev_create_hash(void) 11947 { 11948 int i; 11949 struct hlist_head *hash; 11950 11951 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11952 if (hash != NULL) 11953 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11954 INIT_HLIST_HEAD(&hash[i]); 11955 11956 return hash; 11957 } 11958 11959 /* Initialize per network namespace state */ 11960 static int __net_init netdev_init(struct net *net) 11961 { 11962 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11963 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11964 11965 INIT_LIST_HEAD(&net->dev_base_head); 11966 11967 net->dev_name_head = netdev_create_hash(); 11968 if (net->dev_name_head == NULL) 11969 goto err_name; 11970 11971 net->dev_index_head = netdev_create_hash(); 11972 if (net->dev_index_head == NULL) 11973 goto err_idx; 11974 11975 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11976 11977 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11978 11979 return 0; 11980 11981 err_idx: 11982 kfree(net->dev_name_head); 11983 err_name: 11984 return -ENOMEM; 11985 } 11986 11987 /** 11988 * netdev_drivername - network driver for the device 11989 * @dev: network device 11990 * 11991 * Determine network driver for device. 11992 */ 11993 const char *netdev_drivername(const struct net_device *dev) 11994 { 11995 const struct device_driver *driver; 11996 const struct device *parent; 11997 const char *empty = ""; 11998 11999 parent = dev->dev.parent; 12000 if (!parent) 12001 return empty; 12002 12003 driver = parent->driver; 12004 if (driver && driver->name) 12005 return driver->name; 12006 return empty; 12007 } 12008 12009 static void __netdev_printk(const char *level, const struct net_device *dev, 12010 struct va_format *vaf) 12011 { 12012 if (dev && dev->dev.parent) { 12013 dev_printk_emit(level[1] - '0', 12014 dev->dev.parent, 12015 "%s %s %s%s: %pV", 12016 dev_driver_string(dev->dev.parent), 12017 dev_name(dev->dev.parent), 12018 netdev_name(dev), netdev_reg_state(dev), 12019 vaf); 12020 } else if (dev) { 12021 printk("%s%s%s: %pV", 12022 level, netdev_name(dev), netdev_reg_state(dev), vaf); 12023 } else { 12024 printk("%s(NULL net_device): %pV", level, vaf); 12025 } 12026 } 12027 12028 void netdev_printk(const char *level, const struct net_device *dev, 12029 const char *format, ...) 12030 { 12031 struct va_format vaf; 12032 va_list args; 12033 12034 va_start(args, format); 12035 12036 vaf.fmt = format; 12037 vaf.va = &args; 12038 12039 __netdev_printk(level, dev, &vaf); 12040 12041 va_end(args); 12042 } 12043 EXPORT_SYMBOL(netdev_printk); 12044 12045 #define define_netdev_printk_level(func, level) \ 12046 void func(const struct net_device *dev, const char *fmt, ...) \ 12047 { \ 12048 struct va_format vaf; \ 12049 va_list args; \ 12050 \ 12051 va_start(args, fmt); \ 12052 \ 12053 vaf.fmt = fmt; \ 12054 vaf.va = &args; \ 12055 \ 12056 __netdev_printk(level, dev, &vaf); \ 12057 \ 12058 va_end(args); \ 12059 } \ 12060 EXPORT_SYMBOL(func); 12061 12062 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 12063 define_netdev_printk_level(netdev_alert, KERN_ALERT); 12064 define_netdev_printk_level(netdev_crit, KERN_CRIT); 12065 define_netdev_printk_level(netdev_err, KERN_ERR); 12066 define_netdev_printk_level(netdev_warn, KERN_WARNING); 12067 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 12068 define_netdev_printk_level(netdev_info, KERN_INFO); 12069 12070 static void __net_exit netdev_exit(struct net *net) 12071 { 12072 kfree(net->dev_name_head); 12073 kfree(net->dev_index_head); 12074 xa_destroy(&net->dev_by_index); 12075 if (net != &init_net) 12076 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 12077 } 12078 12079 static struct pernet_operations __net_initdata netdev_net_ops = { 12080 .init = netdev_init, 12081 .exit = netdev_exit, 12082 }; 12083 12084 static void __net_exit default_device_exit_net(struct net *net) 12085 { 12086 struct netdev_name_node *name_node, *tmp; 12087 struct net_device *dev, *aux; 12088 /* 12089 * Push all migratable network devices back to the 12090 * initial network namespace 12091 */ 12092 ASSERT_RTNL(); 12093 for_each_netdev_safe(net, dev, aux) { 12094 int err; 12095 char fb_name[IFNAMSIZ]; 12096 12097 /* Ignore unmoveable devices (i.e. loopback) */ 12098 if (dev->netns_local) 12099 continue; 12100 12101 /* Leave virtual devices for the generic cleanup */ 12102 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 12103 continue; 12104 12105 /* Push remaining network devices to init_net */ 12106 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 12107 if (netdev_name_in_use(&init_net, fb_name)) 12108 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 12109 12110 netdev_for_each_altname_safe(dev, name_node, tmp) 12111 if (netdev_name_in_use(&init_net, name_node->name)) 12112 __netdev_name_node_alt_destroy(name_node); 12113 12114 err = dev_change_net_namespace(dev, &init_net, fb_name); 12115 if (err) { 12116 pr_emerg("%s: failed to move %s to init_net: %d\n", 12117 __func__, dev->name, err); 12118 BUG(); 12119 } 12120 } 12121 } 12122 12123 static void __net_exit default_device_exit_batch(struct list_head *net_list) 12124 { 12125 /* At exit all network devices most be removed from a network 12126 * namespace. Do this in the reverse order of registration. 12127 * Do this across as many network namespaces as possible to 12128 * improve batching efficiency. 12129 */ 12130 struct net_device *dev; 12131 struct net *net; 12132 LIST_HEAD(dev_kill_list); 12133 12134 rtnl_lock(); 12135 list_for_each_entry(net, net_list, exit_list) { 12136 default_device_exit_net(net); 12137 cond_resched(); 12138 } 12139 12140 list_for_each_entry(net, net_list, exit_list) { 12141 for_each_netdev_reverse(net, dev) { 12142 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 12143 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 12144 else 12145 unregister_netdevice_queue(dev, &dev_kill_list); 12146 } 12147 } 12148 unregister_netdevice_many(&dev_kill_list); 12149 rtnl_unlock(); 12150 } 12151 12152 static struct pernet_operations __net_initdata default_device_ops = { 12153 .exit_batch = default_device_exit_batch, 12154 }; 12155 12156 static void __init net_dev_struct_check(void) 12157 { 12158 /* TX read-mostly hotpath */ 12159 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags_fast); 12160 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 12161 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 12162 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 12163 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 12164 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 12165 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 12166 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 12167 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 12168 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 12169 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 12170 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 12171 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 12172 #ifdef CONFIG_XPS 12173 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 12174 #endif 12175 #ifdef CONFIG_NETFILTER_EGRESS 12176 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 12177 #endif 12178 #ifdef CONFIG_NET_XGRESS 12179 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 12180 #endif 12181 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 12182 12183 /* TXRX read-mostly hotpath */ 12184 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); 12185 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state); 12186 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 12187 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 12188 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 12189 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 12190 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46); 12191 12192 /* RX read-mostly hotpath */ 12193 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 12194 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 12195 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 12196 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 12197 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 12198 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 12199 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 12200 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 12201 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 12202 #ifdef CONFIG_NETPOLL 12203 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 12204 #endif 12205 #ifdef CONFIG_NET_XGRESS 12206 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 12207 #endif 12208 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 92); 12209 } 12210 12211 /* 12212 * Initialize the DEV module. At boot time this walks the device list and 12213 * unhooks any devices that fail to initialise (normally hardware not 12214 * present) and leaves us with a valid list of present and active devices. 12215 * 12216 */ 12217 12218 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ 12219 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) 12220 12221 static int net_page_pool_create(int cpuid) 12222 { 12223 #if IS_ENABLED(CONFIG_PAGE_POOL) 12224 struct page_pool_params page_pool_params = { 12225 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, 12226 .flags = PP_FLAG_SYSTEM_POOL, 12227 .nid = cpu_to_mem(cpuid), 12228 }; 12229 struct page_pool *pp_ptr; 12230 int err; 12231 12232 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); 12233 if (IS_ERR(pp_ptr)) 12234 return -ENOMEM; 12235 12236 err = xdp_reg_page_pool(pp_ptr); 12237 if (err) { 12238 page_pool_destroy(pp_ptr); 12239 return err; 12240 } 12241 12242 per_cpu(system_page_pool, cpuid) = pp_ptr; 12243 #endif 12244 return 0; 12245 } 12246 12247 static int backlog_napi_should_run(unsigned int cpu) 12248 { 12249 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12250 struct napi_struct *napi = &sd->backlog; 12251 12252 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 12253 } 12254 12255 static void run_backlog_napi(unsigned int cpu) 12256 { 12257 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12258 12259 napi_threaded_poll_loop(&sd->backlog); 12260 } 12261 12262 static void backlog_napi_setup(unsigned int cpu) 12263 { 12264 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 12265 struct napi_struct *napi = &sd->backlog; 12266 12267 napi->thread = this_cpu_read(backlog_napi); 12268 set_bit(NAPI_STATE_THREADED, &napi->state); 12269 } 12270 12271 static struct smp_hotplug_thread backlog_threads = { 12272 .store = &backlog_napi, 12273 .thread_should_run = backlog_napi_should_run, 12274 .thread_fn = run_backlog_napi, 12275 .thread_comm = "backlog_napi/%u", 12276 .setup = backlog_napi_setup, 12277 }; 12278 12279 /* 12280 * This is called single threaded during boot, so no need 12281 * to take the rtnl semaphore. 12282 */ 12283 static int __init net_dev_init(void) 12284 { 12285 int i, rc = -ENOMEM; 12286 12287 BUG_ON(!dev_boot_phase); 12288 12289 net_dev_struct_check(); 12290 12291 if (dev_proc_init()) 12292 goto out; 12293 12294 if (netdev_kobject_init()) 12295 goto out; 12296 12297 for (i = 0; i < PTYPE_HASH_SIZE; i++) 12298 INIT_LIST_HEAD(&ptype_base[i]); 12299 12300 if (register_pernet_subsys(&netdev_net_ops)) 12301 goto out; 12302 12303 /* 12304 * Initialise the packet receive queues. 12305 */ 12306 12307 for_each_possible_cpu(i) { 12308 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 12309 struct softnet_data *sd = &per_cpu(softnet_data, i); 12310 12311 INIT_WORK(flush, flush_backlog); 12312 12313 skb_queue_head_init(&sd->input_pkt_queue); 12314 skb_queue_head_init(&sd->process_queue); 12315 #ifdef CONFIG_XFRM_OFFLOAD 12316 skb_queue_head_init(&sd->xfrm_backlog); 12317 #endif 12318 INIT_LIST_HEAD(&sd->poll_list); 12319 sd->output_queue_tailp = &sd->output_queue; 12320 #ifdef CONFIG_RPS 12321 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 12322 sd->cpu = i; 12323 #endif 12324 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 12325 spin_lock_init(&sd->defer_lock); 12326 12327 init_gro_hash(&sd->backlog); 12328 sd->backlog.poll = process_backlog; 12329 sd->backlog.weight = weight_p; 12330 INIT_LIST_HEAD(&sd->backlog.poll_list); 12331 12332 if (net_page_pool_create(i)) 12333 goto out; 12334 } 12335 if (use_backlog_threads()) 12336 smpboot_register_percpu_thread(&backlog_threads); 12337 12338 dev_boot_phase = 0; 12339 12340 /* The loopback device is special if any other network devices 12341 * is present in a network namespace the loopback device must 12342 * be present. Since we now dynamically allocate and free the 12343 * loopback device ensure this invariant is maintained by 12344 * keeping the loopback device as the first device on the 12345 * list of network devices. Ensuring the loopback devices 12346 * is the first device that appears and the last network device 12347 * that disappears. 12348 */ 12349 if (register_pernet_device(&loopback_net_ops)) 12350 goto out; 12351 12352 if (register_pernet_device(&default_device_ops)) 12353 goto out; 12354 12355 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 12356 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 12357 12358 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 12359 NULL, dev_cpu_dead); 12360 WARN_ON(rc < 0); 12361 rc = 0; 12362 12363 /* avoid static key IPIs to isolated CPUs */ 12364 if (housekeeping_enabled(HK_TYPE_MISC)) 12365 net_enable_timestamp(); 12366 out: 12367 if (rc < 0) { 12368 for_each_possible_cpu(i) { 12369 struct page_pool *pp_ptr; 12370 12371 pp_ptr = per_cpu(system_page_pool, i); 12372 if (!pp_ptr) 12373 continue; 12374 12375 xdp_unreg_page_pool(pp_ptr); 12376 page_pool_destroy(pp_ptr); 12377 per_cpu(system_page_pool, i) = NULL; 12378 } 12379 } 12380 12381 return rc; 12382 } 12383 12384 subsys_initcall(net_dev_init); 12385