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/mm.h> 81 #include <linux/smpboot.h> 82 #include <linux/mutex.h> 83 #include <linux/rwsem.h> 84 #include <linux/string.h> 85 #include <linux/mm.h> 86 #include <linux/socket.h> 87 #include <linux/sockios.h> 88 #include <linux/errno.h> 89 #include <linux/interrupt.h> 90 #include <linux/if_ether.h> 91 #include <linux/netdevice.h> 92 #include <linux/etherdevice.h> 93 #include <linux/ethtool.h> 94 #include <linux/skbuff.h> 95 #include <linux/kthread.h> 96 #include <linux/bpf.h> 97 #include <linux/bpf_trace.h> 98 #include <net/net_namespace.h> 99 #include <net/sock.h> 100 #include <net/busy_poll.h> 101 #include <linux/rtnetlink.h> 102 #include <linux/stat.h> 103 #include <net/dsa.h> 104 #include <net/dst.h> 105 #include <net/dst_metadata.h> 106 #include <net/gro.h> 107 #include <net/pkt_sched.h> 108 #include <net/pkt_cls.h> 109 #include <net/checksum.h> 110 #include <net/xfrm.h> 111 #include <net/tcx.h> 112 #include <linux/highmem.h> 113 #include <linux/init.h> 114 #include <linux/module.h> 115 #include <linux/netpoll.h> 116 #include <linux/rcupdate.h> 117 #include <linux/delay.h> 118 #include <net/iw_handler.h> 119 #include <asm/current.h> 120 #include <linux/audit.h> 121 #include <linux/dmaengine.h> 122 #include <linux/err.h> 123 #include <linux/ctype.h> 124 #include <linux/if_arp.h> 125 #include <linux/if_vlan.h> 126 #include <linux/ip.h> 127 #include <net/ip.h> 128 #include <net/mpls.h> 129 #include <linux/ipv6.h> 130 #include <linux/in.h> 131 #include <linux/jhash.h> 132 #include <linux/random.h> 133 #include <trace/events/napi.h> 134 #include <trace/events/net.h> 135 #include <trace/events/skb.h> 136 #include <trace/events/qdisc.h> 137 #include <trace/events/xdp.h> 138 #include <linux/inetdevice.h> 139 #include <linux/cpu_rmap.h> 140 #include <linux/static_key.h> 141 #include <linux/hashtable.h> 142 #include <linux/vmalloc.h> 143 #include <linux/if_macvlan.h> 144 #include <linux/errqueue.h> 145 #include <linux/hrtimer.h> 146 #include <linux/netfilter_netdev.h> 147 #include <linux/crash_dump.h> 148 #include <linux/sctp.h> 149 #include <net/udp_tunnel.h> 150 #include <linux/net_namespace.h> 151 #include <linux/indirect_call_wrapper.h> 152 #include <net/devlink.h> 153 #include <linux/pm_runtime.h> 154 #include <linux/prandom.h> 155 #include <linux/once_lite.h> 156 #include <net/netdev_rx_queue.h> 157 #include <net/page_pool/types.h> 158 #include <net/page_pool/helpers.h> 159 #include <net/rps.h> 160 161 #include "dev.h" 162 #include "net-sysfs.h" 163 164 static DEFINE_SPINLOCK(ptype_lock); 165 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 166 167 static int netif_rx_internal(struct sk_buff *skb); 168 static int call_netdevice_notifiers_extack(unsigned long val, 169 struct net_device *dev, 170 struct netlink_ext_ack *extack); 171 172 static DEFINE_MUTEX(ifalias_mutex); 173 174 /* protects napi_hash addition/deletion and napi_gen_id */ 175 static DEFINE_SPINLOCK(napi_hash_lock); 176 177 static unsigned int napi_gen_id = NR_CPUS; 178 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 179 180 static DECLARE_RWSEM(devnet_rename_sem); 181 182 static inline void dev_base_seq_inc(struct net *net) 183 { 184 unsigned int val = net->dev_base_seq + 1; 185 186 WRITE_ONCE(net->dev_base_seq, val ?: 1); 187 } 188 189 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 190 { 191 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 192 193 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 194 } 195 196 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 197 { 198 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 199 } 200 201 #ifndef CONFIG_PREEMPT_RT 202 203 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key); 204 205 static int __init setup_backlog_napi_threads(char *arg) 206 { 207 static_branch_enable(&use_backlog_threads_key); 208 return 0; 209 } 210 early_param("thread_backlog_napi", setup_backlog_napi_threads); 211 212 static bool use_backlog_threads(void) 213 { 214 return static_branch_unlikely(&use_backlog_threads_key); 215 } 216 217 #else 218 219 static bool use_backlog_threads(void) 220 { 221 return true; 222 } 223 224 #endif 225 226 static inline void backlog_lock_irq_save(struct softnet_data *sd, 227 unsigned long *flags) 228 { 229 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 230 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 231 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 232 local_irq_save(*flags); 233 } 234 235 static inline void backlog_lock_irq_disable(struct softnet_data *sd) 236 { 237 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 238 spin_lock_irq(&sd->input_pkt_queue.lock); 239 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 240 local_irq_disable(); 241 } 242 243 static inline void backlog_unlock_irq_restore(struct softnet_data *sd, 244 unsigned long *flags) 245 { 246 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 247 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 248 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 249 local_irq_restore(*flags); 250 } 251 252 static inline void backlog_unlock_irq_enable(struct softnet_data *sd) 253 { 254 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads()) 255 spin_unlock_irq(&sd->input_pkt_queue.lock); 256 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 257 local_irq_enable(); 258 } 259 260 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 261 const char *name) 262 { 263 struct netdev_name_node *name_node; 264 265 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 266 if (!name_node) 267 return NULL; 268 INIT_HLIST_NODE(&name_node->hlist); 269 name_node->dev = dev; 270 name_node->name = name; 271 return name_node; 272 } 273 274 static struct netdev_name_node * 275 netdev_name_node_head_alloc(struct net_device *dev) 276 { 277 struct netdev_name_node *name_node; 278 279 name_node = netdev_name_node_alloc(dev, dev->name); 280 if (!name_node) 281 return NULL; 282 INIT_LIST_HEAD(&name_node->list); 283 return name_node; 284 } 285 286 static void netdev_name_node_free(struct netdev_name_node *name_node) 287 { 288 kfree(name_node); 289 } 290 291 static void netdev_name_node_add(struct net *net, 292 struct netdev_name_node *name_node) 293 { 294 hlist_add_head_rcu(&name_node->hlist, 295 dev_name_hash(net, name_node->name)); 296 } 297 298 static void netdev_name_node_del(struct netdev_name_node *name_node) 299 { 300 hlist_del_rcu(&name_node->hlist); 301 } 302 303 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 304 const char *name) 305 { 306 struct hlist_head *head = dev_name_hash(net, name); 307 struct netdev_name_node *name_node; 308 309 hlist_for_each_entry(name_node, head, hlist) 310 if (!strcmp(name_node->name, name)) 311 return name_node; 312 return NULL; 313 } 314 315 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 316 const char *name) 317 { 318 struct hlist_head *head = dev_name_hash(net, name); 319 struct netdev_name_node *name_node; 320 321 hlist_for_each_entry_rcu(name_node, head, hlist) 322 if (!strcmp(name_node->name, name)) 323 return name_node; 324 return NULL; 325 } 326 327 bool netdev_name_in_use(struct net *net, const char *name) 328 { 329 return netdev_name_node_lookup(net, name); 330 } 331 EXPORT_SYMBOL(netdev_name_in_use); 332 333 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 334 { 335 struct netdev_name_node *name_node; 336 struct net *net = dev_net(dev); 337 338 name_node = netdev_name_node_lookup(net, name); 339 if (name_node) 340 return -EEXIST; 341 name_node = netdev_name_node_alloc(dev, name); 342 if (!name_node) 343 return -ENOMEM; 344 netdev_name_node_add(net, name_node); 345 /* The node that holds dev->name acts as a head of per-device list. */ 346 list_add_tail_rcu(&name_node->list, &dev->name_node->list); 347 348 return 0; 349 } 350 351 static void netdev_name_node_alt_free(struct rcu_head *head) 352 { 353 struct netdev_name_node *name_node = 354 container_of(head, struct netdev_name_node, rcu); 355 356 kfree(name_node->name); 357 netdev_name_node_free(name_node); 358 } 359 360 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 361 { 362 netdev_name_node_del(name_node); 363 list_del(&name_node->list); 364 call_rcu(&name_node->rcu, netdev_name_node_alt_free); 365 } 366 367 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 368 { 369 struct netdev_name_node *name_node; 370 struct net *net = dev_net(dev); 371 372 name_node = netdev_name_node_lookup(net, name); 373 if (!name_node) 374 return -ENOENT; 375 /* lookup might have found our primary name or a name belonging 376 * to another device. 377 */ 378 if (name_node == dev->name_node || name_node->dev != dev) 379 return -EINVAL; 380 381 __netdev_name_node_alt_destroy(name_node); 382 return 0; 383 } 384 385 static void netdev_name_node_alt_flush(struct net_device *dev) 386 { 387 struct netdev_name_node *name_node, *tmp; 388 389 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) { 390 list_del(&name_node->list); 391 netdev_name_node_alt_free(&name_node->rcu); 392 } 393 } 394 395 /* Device list insertion */ 396 static void list_netdevice(struct net_device *dev) 397 { 398 struct netdev_name_node *name_node; 399 struct net *net = dev_net(dev); 400 401 ASSERT_RTNL(); 402 403 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 404 netdev_name_node_add(net, dev->name_node); 405 hlist_add_head_rcu(&dev->index_hlist, 406 dev_index_hash(net, dev->ifindex)); 407 408 netdev_for_each_altname(dev, name_node) 409 netdev_name_node_add(net, name_node); 410 411 /* We reserved the ifindex, this can't fail */ 412 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 413 414 dev_base_seq_inc(net); 415 } 416 417 /* Device list removal 418 * caller must respect a RCU grace period before freeing/reusing dev 419 */ 420 static void unlist_netdevice(struct net_device *dev) 421 { 422 struct netdev_name_node *name_node; 423 struct net *net = dev_net(dev); 424 425 ASSERT_RTNL(); 426 427 xa_erase(&net->dev_by_index, dev->ifindex); 428 429 netdev_for_each_altname(dev, name_node) 430 netdev_name_node_del(name_node); 431 432 /* Unlink dev from the device chain */ 433 list_del_rcu(&dev->dev_list); 434 netdev_name_node_del(dev->name_node); 435 hlist_del_rcu(&dev->index_hlist); 436 437 dev_base_seq_inc(dev_net(dev)); 438 } 439 440 /* 441 * Our notifier list 442 */ 443 444 static RAW_NOTIFIER_HEAD(netdev_chain); 445 446 /* 447 * Device drivers call our routines to queue packets here. We empty the 448 * queue in the local softnet handler. 449 */ 450 451 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 452 EXPORT_PER_CPU_SYMBOL(softnet_data); 453 454 /* Page_pool has a lockless array/stack to alloc/recycle pages. 455 * PP consumers must pay attention to run APIs in the appropriate context 456 * (e.g. NAPI context). 457 */ 458 static DEFINE_PER_CPU_ALIGNED(struct page_pool *, system_page_pool); 459 460 #ifdef CONFIG_LOCKDEP 461 /* 462 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 463 * according to dev->type 464 */ 465 static const unsigned short netdev_lock_type[] = { 466 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 467 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 468 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 469 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 470 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 471 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 472 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 473 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 474 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 475 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 476 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 477 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 478 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 479 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 480 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 481 482 static const char *const netdev_lock_name[] = { 483 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 484 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 485 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 486 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 487 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 488 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 489 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 490 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 491 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 492 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 493 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 494 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 495 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 496 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 497 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 498 499 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 500 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 501 502 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 503 { 504 int i; 505 506 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 507 if (netdev_lock_type[i] == dev_type) 508 return i; 509 /* the last key is used by default */ 510 return ARRAY_SIZE(netdev_lock_type) - 1; 511 } 512 513 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 514 unsigned short dev_type) 515 { 516 int i; 517 518 i = netdev_lock_pos(dev_type); 519 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 520 netdev_lock_name[i]); 521 } 522 523 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 524 { 525 int i; 526 527 i = netdev_lock_pos(dev->type); 528 lockdep_set_class_and_name(&dev->addr_list_lock, 529 &netdev_addr_lock_key[i], 530 netdev_lock_name[i]); 531 } 532 #else 533 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 534 unsigned short dev_type) 535 { 536 } 537 538 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 539 { 540 } 541 #endif 542 543 /******************************************************************************* 544 * 545 * Protocol management and registration routines 546 * 547 *******************************************************************************/ 548 549 550 /* 551 * Add a protocol ID to the list. Now that the input handler is 552 * smarter we can dispense with all the messy stuff that used to be 553 * here. 554 * 555 * BEWARE!!! Protocol handlers, mangling input packets, 556 * MUST BE last in hash buckets and checking protocol handlers 557 * MUST start from promiscuous ptype_all chain in net_bh. 558 * It is true now, do not change it. 559 * Explanation follows: if protocol handler, mangling packet, will 560 * be the first on list, it is not able to sense, that packet 561 * is cloned and should be copied-on-write, so that it will 562 * change it and subsequent readers will get broken packet. 563 * --ANK (980803) 564 */ 565 566 static inline struct list_head *ptype_head(const struct packet_type *pt) 567 { 568 if (pt->type == htons(ETH_P_ALL)) 569 return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all; 570 else 571 return pt->dev ? &pt->dev->ptype_specific : 572 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 573 } 574 575 /** 576 * dev_add_pack - add packet handler 577 * @pt: packet type declaration 578 * 579 * Add a protocol handler to the networking stack. The passed &packet_type 580 * is linked into kernel lists and may not be freed until it has been 581 * removed from the kernel lists. 582 * 583 * This call does not sleep therefore it can not 584 * guarantee all CPU's that are in middle of receiving packets 585 * will see the new packet type (until the next received packet). 586 */ 587 588 void dev_add_pack(struct packet_type *pt) 589 { 590 struct list_head *head = ptype_head(pt); 591 592 spin_lock(&ptype_lock); 593 list_add_rcu(&pt->list, head); 594 spin_unlock(&ptype_lock); 595 } 596 EXPORT_SYMBOL(dev_add_pack); 597 598 /** 599 * __dev_remove_pack - remove packet handler 600 * @pt: packet type declaration 601 * 602 * Remove a protocol handler that was previously added to the kernel 603 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 604 * from the kernel lists and can be freed or reused once this function 605 * returns. 606 * 607 * The packet type might still be in use by receivers 608 * and must not be freed until after all the CPU's have gone 609 * through a quiescent state. 610 */ 611 void __dev_remove_pack(struct packet_type *pt) 612 { 613 struct list_head *head = ptype_head(pt); 614 struct packet_type *pt1; 615 616 spin_lock(&ptype_lock); 617 618 list_for_each_entry(pt1, head, list) { 619 if (pt == pt1) { 620 list_del_rcu(&pt->list); 621 goto out; 622 } 623 } 624 625 pr_warn("dev_remove_pack: %p not found\n", pt); 626 out: 627 spin_unlock(&ptype_lock); 628 } 629 EXPORT_SYMBOL(__dev_remove_pack); 630 631 /** 632 * dev_remove_pack - remove packet handler 633 * @pt: packet type declaration 634 * 635 * Remove a protocol handler that was previously added to the kernel 636 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 637 * from the kernel lists and can be freed or reused once this function 638 * returns. 639 * 640 * This call sleeps to guarantee that no CPU is looking at the packet 641 * type after return. 642 */ 643 void dev_remove_pack(struct packet_type *pt) 644 { 645 __dev_remove_pack(pt); 646 647 synchronize_net(); 648 } 649 EXPORT_SYMBOL(dev_remove_pack); 650 651 652 /******************************************************************************* 653 * 654 * Device Interface Subroutines 655 * 656 *******************************************************************************/ 657 658 /** 659 * dev_get_iflink - get 'iflink' value of a interface 660 * @dev: targeted interface 661 * 662 * Indicates the ifindex the interface is linked to. 663 * Physical interfaces have the same 'ifindex' and 'iflink' values. 664 */ 665 666 int dev_get_iflink(const struct net_device *dev) 667 { 668 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 669 return dev->netdev_ops->ndo_get_iflink(dev); 670 671 return READ_ONCE(dev->ifindex); 672 } 673 EXPORT_SYMBOL(dev_get_iflink); 674 675 /** 676 * dev_fill_metadata_dst - Retrieve tunnel egress information. 677 * @dev: targeted interface 678 * @skb: The packet. 679 * 680 * For better visibility of tunnel traffic OVS needs to retrieve 681 * egress tunnel information for a packet. Following API allows 682 * user to get this info. 683 */ 684 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 685 { 686 struct ip_tunnel_info *info; 687 688 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 689 return -EINVAL; 690 691 info = skb_tunnel_info_unclone(skb); 692 if (!info) 693 return -ENOMEM; 694 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 695 return -EINVAL; 696 697 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 698 } 699 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 700 701 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 702 { 703 int k = stack->num_paths++; 704 705 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 706 return NULL; 707 708 return &stack->path[k]; 709 } 710 711 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 712 struct net_device_path_stack *stack) 713 { 714 const struct net_device *last_dev; 715 struct net_device_path_ctx ctx = { 716 .dev = dev, 717 }; 718 struct net_device_path *path; 719 int ret = 0; 720 721 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 722 stack->num_paths = 0; 723 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 724 last_dev = ctx.dev; 725 path = dev_fwd_path(stack); 726 if (!path) 727 return -1; 728 729 memset(path, 0, sizeof(struct net_device_path)); 730 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 731 if (ret < 0) 732 return -1; 733 734 if (WARN_ON_ONCE(last_dev == ctx.dev)) 735 return -1; 736 } 737 738 if (!ctx.dev) 739 return ret; 740 741 path = dev_fwd_path(stack); 742 if (!path) 743 return -1; 744 path->type = DEV_PATH_ETHERNET; 745 path->dev = ctx.dev; 746 747 return ret; 748 } 749 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 750 751 /** 752 * __dev_get_by_name - find a device by its name 753 * @net: the applicable net namespace 754 * @name: name to find 755 * 756 * Find an interface by name. Must be called under RTNL semaphore. 757 * If the name is found a pointer to the device is returned. 758 * If the name is not found then %NULL is returned. The 759 * reference counters are not incremented so the caller must be 760 * careful with locks. 761 */ 762 763 struct net_device *__dev_get_by_name(struct net *net, const char *name) 764 { 765 struct netdev_name_node *node_name; 766 767 node_name = netdev_name_node_lookup(net, name); 768 return node_name ? node_name->dev : NULL; 769 } 770 EXPORT_SYMBOL(__dev_get_by_name); 771 772 /** 773 * dev_get_by_name_rcu - find a device by its name 774 * @net: the applicable net namespace 775 * @name: name to find 776 * 777 * Find an interface by name. 778 * If the name is found a pointer to the device is returned. 779 * If the name is not found then %NULL is returned. 780 * The reference counters are not incremented so the caller must be 781 * careful with locks. The caller must hold RCU lock. 782 */ 783 784 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 785 { 786 struct netdev_name_node *node_name; 787 788 node_name = netdev_name_node_lookup_rcu(net, name); 789 return node_name ? node_name->dev : NULL; 790 } 791 EXPORT_SYMBOL(dev_get_by_name_rcu); 792 793 /* Deprecated for new users, call netdev_get_by_name() instead */ 794 struct net_device *dev_get_by_name(struct net *net, const char *name) 795 { 796 struct net_device *dev; 797 798 rcu_read_lock(); 799 dev = dev_get_by_name_rcu(net, name); 800 dev_hold(dev); 801 rcu_read_unlock(); 802 return dev; 803 } 804 EXPORT_SYMBOL(dev_get_by_name); 805 806 /** 807 * netdev_get_by_name() - find a device by its name 808 * @net: the applicable net namespace 809 * @name: name to find 810 * @tracker: tracking object for the acquired reference 811 * @gfp: allocation flags for the tracker 812 * 813 * Find an interface by name. This can be called from any 814 * context and does its own locking. The returned handle has 815 * the usage count incremented and the caller must use netdev_put() to 816 * release it when it is no longer needed. %NULL is returned if no 817 * matching device is found. 818 */ 819 struct net_device *netdev_get_by_name(struct net *net, const char *name, 820 netdevice_tracker *tracker, gfp_t gfp) 821 { 822 struct net_device *dev; 823 824 dev = dev_get_by_name(net, name); 825 if (dev) 826 netdev_tracker_alloc(dev, tracker, gfp); 827 return dev; 828 } 829 EXPORT_SYMBOL(netdev_get_by_name); 830 831 /** 832 * __dev_get_by_index - find a device by its ifindex 833 * @net: the applicable net namespace 834 * @ifindex: index of device 835 * 836 * Search for an interface by index. Returns %NULL if the device 837 * is not found or a pointer to the device. The device has not 838 * had its reference counter increased so the caller must be careful 839 * about locking. The caller must hold the RTNL semaphore. 840 */ 841 842 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 843 { 844 struct net_device *dev; 845 struct hlist_head *head = dev_index_hash(net, ifindex); 846 847 hlist_for_each_entry(dev, head, index_hlist) 848 if (dev->ifindex == ifindex) 849 return dev; 850 851 return NULL; 852 } 853 EXPORT_SYMBOL(__dev_get_by_index); 854 855 /** 856 * dev_get_by_index_rcu - find a device by its ifindex 857 * @net: the applicable net namespace 858 * @ifindex: index of device 859 * 860 * Search for an interface by index. Returns %NULL if the device 861 * is not found or a pointer to the device. The device has not 862 * had its reference counter increased so the caller must be careful 863 * about locking. The caller must hold RCU lock. 864 */ 865 866 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 867 { 868 struct net_device *dev; 869 struct hlist_head *head = dev_index_hash(net, ifindex); 870 871 hlist_for_each_entry_rcu(dev, head, index_hlist) 872 if (dev->ifindex == ifindex) 873 return dev; 874 875 return NULL; 876 } 877 EXPORT_SYMBOL(dev_get_by_index_rcu); 878 879 /* Deprecated for new users, call netdev_get_by_index() instead */ 880 struct net_device *dev_get_by_index(struct net *net, int ifindex) 881 { 882 struct net_device *dev; 883 884 rcu_read_lock(); 885 dev = dev_get_by_index_rcu(net, ifindex); 886 dev_hold(dev); 887 rcu_read_unlock(); 888 return dev; 889 } 890 EXPORT_SYMBOL(dev_get_by_index); 891 892 /** 893 * netdev_get_by_index() - find a device by its ifindex 894 * @net: the applicable net namespace 895 * @ifindex: index of device 896 * @tracker: tracking object for the acquired reference 897 * @gfp: allocation flags for the tracker 898 * 899 * Search for an interface by index. Returns NULL if the device 900 * is not found or a pointer to the device. The device returned has 901 * had a reference added and the pointer is safe until the user calls 902 * netdev_put() to indicate they have finished with it. 903 */ 904 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 905 netdevice_tracker *tracker, gfp_t gfp) 906 { 907 struct net_device *dev; 908 909 dev = dev_get_by_index(net, ifindex); 910 if (dev) 911 netdev_tracker_alloc(dev, tracker, gfp); 912 return dev; 913 } 914 EXPORT_SYMBOL(netdev_get_by_index); 915 916 /** 917 * dev_get_by_napi_id - find a device by napi_id 918 * @napi_id: ID of the NAPI struct 919 * 920 * Search for an interface by NAPI ID. Returns %NULL if the device 921 * is not found or a pointer to the device. The device has not had 922 * its reference counter increased so the caller must be careful 923 * about locking. The caller must hold RCU lock. 924 */ 925 926 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 927 { 928 struct napi_struct *napi; 929 930 WARN_ON_ONCE(!rcu_read_lock_held()); 931 932 if (napi_id < MIN_NAPI_ID) 933 return NULL; 934 935 napi = napi_by_id(napi_id); 936 937 return napi ? napi->dev : NULL; 938 } 939 EXPORT_SYMBOL(dev_get_by_napi_id); 940 941 /** 942 * netdev_get_name - get a netdevice name, knowing its ifindex. 943 * @net: network namespace 944 * @name: a pointer to the buffer where the name will be stored. 945 * @ifindex: the ifindex of the interface to get the name from. 946 */ 947 int netdev_get_name(struct net *net, char *name, int ifindex) 948 { 949 struct net_device *dev; 950 int ret; 951 952 down_read(&devnet_rename_sem); 953 rcu_read_lock(); 954 955 dev = dev_get_by_index_rcu(net, ifindex); 956 if (!dev) { 957 ret = -ENODEV; 958 goto out; 959 } 960 961 strcpy(name, dev->name); 962 963 ret = 0; 964 out: 965 rcu_read_unlock(); 966 up_read(&devnet_rename_sem); 967 return ret; 968 } 969 970 /** 971 * dev_getbyhwaddr_rcu - find a device by its hardware address 972 * @net: the applicable net namespace 973 * @type: media type of device 974 * @ha: hardware address 975 * 976 * Search for an interface by MAC address. Returns NULL if the device 977 * is not found or a pointer to the device. 978 * The caller must hold RCU or RTNL. 979 * The returned device has not had its ref count increased 980 * and the caller must therefore be careful about locking 981 * 982 */ 983 984 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 985 const char *ha) 986 { 987 struct net_device *dev; 988 989 for_each_netdev_rcu(net, dev) 990 if (dev->type == type && 991 !memcmp(dev->dev_addr, ha, dev->addr_len)) 992 return dev; 993 994 return NULL; 995 } 996 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 997 998 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 999 { 1000 struct net_device *dev, *ret = NULL; 1001 1002 rcu_read_lock(); 1003 for_each_netdev_rcu(net, dev) 1004 if (dev->type == type) { 1005 dev_hold(dev); 1006 ret = dev; 1007 break; 1008 } 1009 rcu_read_unlock(); 1010 return ret; 1011 } 1012 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1013 1014 /** 1015 * __dev_get_by_flags - find any device with given flags 1016 * @net: the applicable net namespace 1017 * @if_flags: IFF_* values 1018 * @mask: bitmask of bits in if_flags to check 1019 * 1020 * Search for any interface with the given flags. Returns NULL if a device 1021 * is not found or a pointer to the device. Must be called inside 1022 * rtnl_lock(), and result refcount is unchanged. 1023 */ 1024 1025 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1026 unsigned short mask) 1027 { 1028 struct net_device *dev, *ret; 1029 1030 ASSERT_RTNL(); 1031 1032 ret = NULL; 1033 for_each_netdev(net, dev) { 1034 if (((dev->flags ^ if_flags) & mask) == 0) { 1035 ret = dev; 1036 break; 1037 } 1038 } 1039 return ret; 1040 } 1041 EXPORT_SYMBOL(__dev_get_by_flags); 1042 1043 /** 1044 * dev_valid_name - check if name is okay for network device 1045 * @name: name string 1046 * 1047 * Network device names need to be valid file names to 1048 * allow sysfs to work. We also disallow any kind of 1049 * whitespace. 1050 */ 1051 bool dev_valid_name(const char *name) 1052 { 1053 if (*name == '\0') 1054 return false; 1055 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1056 return false; 1057 if (!strcmp(name, ".") || !strcmp(name, "..")) 1058 return false; 1059 1060 while (*name) { 1061 if (*name == '/' || *name == ':' || isspace(*name)) 1062 return false; 1063 name++; 1064 } 1065 return true; 1066 } 1067 EXPORT_SYMBOL(dev_valid_name); 1068 1069 /** 1070 * __dev_alloc_name - allocate a name for a device 1071 * @net: network namespace to allocate the device name in 1072 * @name: name format string 1073 * @res: result name string 1074 * 1075 * Passed a format string - eg "lt%d" it will try and find a suitable 1076 * id. It scans list of devices to build up a free map, then chooses 1077 * the first empty slot. The caller must hold the dev_base or rtnl lock 1078 * while allocating the name and adding the device in order to avoid 1079 * duplicates. 1080 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1081 * Returns the number of the unit assigned or a negative errno code. 1082 */ 1083 1084 static int __dev_alloc_name(struct net *net, const char *name, char *res) 1085 { 1086 int i = 0; 1087 const char *p; 1088 const int max_netdevices = 8*PAGE_SIZE; 1089 unsigned long *inuse; 1090 struct net_device *d; 1091 char buf[IFNAMSIZ]; 1092 1093 /* Verify the string as this thing may have come from the user. 1094 * There must be one "%d" and no other "%" characters. 1095 */ 1096 p = strchr(name, '%'); 1097 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1098 return -EINVAL; 1099 1100 /* Use one page as a bit array of possible slots */ 1101 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1102 if (!inuse) 1103 return -ENOMEM; 1104 1105 for_each_netdev(net, d) { 1106 struct netdev_name_node *name_node; 1107 1108 netdev_for_each_altname(d, name_node) { 1109 if (!sscanf(name_node->name, name, &i)) 1110 continue; 1111 if (i < 0 || i >= max_netdevices) 1112 continue; 1113 1114 /* avoid cases where sscanf is not exact inverse of printf */ 1115 snprintf(buf, IFNAMSIZ, name, i); 1116 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1117 __set_bit(i, inuse); 1118 } 1119 if (!sscanf(d->name, name, &i)) 1120 continue; 1121 if (i < 0 || i >= max_netdevices) 1122 continue; 1123 1124 /* avoid cases where sscanf is not exact inverse of printf */ 1125 snprintf(buf, IFNAMSIZ, name, i); 1126 if (!strncmp(buf, d->name, IFNAMSIZ)) 1127 __set_bit(i, inuse); 1128 } 1129 1130 i = find_first_zero_bit(inuse, max_netdevices); 1131 bitmap_free(inuse); 1132 if (i == max_netdevices) 1133 return -ENFILE; 1134 1135 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1136 strscpy(buf, name, IFNAMSIZ); 1137 snprintf(res, IFNAMSIZ, buf, i); 1138 return i; 1139 } 1140 1141 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1142 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1143 const char *want_name, char *out_name, 1144 int dup_errno) 1145 { 1146 if (!dev_valid_name(want_name)) 1147 return -EINVAL; 1148 1149 if (strchr(want_name, '%')) 1150 return __dev_alloc_name(net, want_name, out_name); 1151 1152 if (netdev_name_in_use(net, want_name)) 1153 return -dup_errno; 1154 if (out_name != want_name) 1155 strscpy(out_name, want_name, IFNAMSIZ); 1156 return 0; 1157 } 1158 1159 /** 1160 * dev_alloc_name - allocate a name for a device 1161 * @dev: device 1162 * @name: name format string 1163 * 1164 * Passed a format string - eg "lt%d" it will try and find a suitable 1165 * id. It scans list of devices to build up a free map, then chooses 1166 * the first empty slot. The caller must hold the dev_base or rtnl lock 1167 * while allocating the name and adding the device in order to avoid 1168 * duplicates. 1169 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1170 * Returns the number of the unit assigned or a negative errno code. 1171 */ 1172 1173 int dev_alloc_name(struct net_device *dev, const char *name) 1174 { 1175 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1176 } 1177 EXPORT_SYMBOL(dev_alloc_name); 1178 1179 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1180 const char *name) 1181 { 1182 int ret; 1183 1184 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1185 return ret < 0 ? ret : 0; 1186 } 1187 1188 /** 1189 * dev_change_name - change name of a device 1190 * @dev: device 1191 * @newname: name (or format string) must be at least IFNAMSIZ 1192 * 1193 * Change name of a device, can pass format strings "eth%d". 1194 * for wildcarding. 1195 */ 1196 int dev_change_name(struct net_device *dev, const char *newname) 1197 { 1198 unsigned char old_assign_type; 1199 char oldname[IFNAMSIZ]; 1200 int err = 0; 1201 int ret; 1202 struct net *net; 1203 1204 ASSERT_RTNL(); 1205 BUG_ON(!dev_net(dev)); 1206 1207 net = dev_net(dev); 1208 1209 down_write(&devnet_rename_sem); 1210 1211 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1212 up_write(&devnet_rename_sem); 1213 return 0; 1214 } 1215 1216 memcpy(oldname, dev->name, IFNAMSIZ); 1217 1218 err = dev_get_valid_name(net, dev, newname); 1219 if (err < 0) { 1220 up_write(&devnet_rename_sem); 1221 return err; 1222 } 1223 1224 if (oldname[0] && !strchr(oldname, '%')) 1225 netdev_info(dev, "renamed from %s%s\n", oldname, 1226 dev->flags & IFF_UP ? " (while UP)" : ""); 1227 1228 old_assign_type = dev->name_assign_type; 1229 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); 1230 1231 rollback: 1232 ret = device_rename(&dev->dev, dev->name); 1233 if (ret) { 1234 memcpy(dev->name, oldname, IFNAMSIZ); 1235 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1236 up_write(&devnet_rename_sem); 1237 return ret; 1238 } 1239 1240 up_write(&devnet_rename_sem); 1241 1242 netdev_adjacent_rename_links(dev, oldname); 1243 1244 netdev_name_node_del(dev->name_node); 1245 1246 synchronize_net(); 1247 1248 netdev_name_node_add(net, dev->name_node); 1249 1250 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1251 ret = notifier_to_errno(ret); 1252 1253 if (ret) { 1254 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1255 if (err >= 0) { 1256 err = ret; 1257 down_write(&devnet_rename_sem); 1258 memcpy(dev->name, oldname, IFNAMSIZ); 1259 memcpy(oldname, newname, IFNAMSIZ); 1260 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1261 old_assign_type = NET_NAME_RENAMED; 1262 goto rollback; 1263 } else { 1264 netdev_err(dev, "name change rollback failed: %d\n", 1265 ret); 1266 } 1267 } 1268 1269 return err; 1270 } 1271 1272 /** 1273 * dev_set_alias - change ifalias of a device 1274 * @dev: device 1275 * @alias: name up to IFALIASZ 1276 * @len: limit of bytes to copy from info 1277 * 1278 * Set ifalias for a device, 1279 */ 1280 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1281 { 1282 struct dev_ifalias *new_alias = NULL; 1283 1284 if (len >= IFALIASZ) 1285 return -EINVAL; 1286 1287 if (len) { 1288 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1289 if (!new_alias) 1290 return -ENOMEM; 1291 1292 memcpy(new_alias->ifalias, alias, len); 1293 new_alias->ifalias[len] = 0; 1294 } 1295 1296 mutex_lock(&ifalias_mutex); 1297 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1298 mutex_is_locked(&ifalias_mutex)); 1299 mutex_unlock(&ifalias_mutex); 1300 1301 if (new_alias) 1302 kfree_rcu(new_alias, rcuhead); 1303 1304 return len; 1305 } 1306 EXPORT_SYMBOL(dev_set_alias); 1307 1308 /** 1309 * dev_get_alias - get ifalias of a device 1310 * @dev: device 1311 * @name: buffer to store name of ifalias 1312 * @len: size of buffer 1313 * 1314 * get ifalias for a device. Caller must make sure dev cannot go 1315 * away, e.g. rcu read lock or own a reference count to device. 1316 */ 1317 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1318 { 1319 const struct dev_ifalias *alias; 1320 int ret = 0; 1321 1322 rcu_read_lock(); 1323 alias = rcu_dereference(dev->ifalias); 1324 if (alias) 1325 ret = snprintf(name, len, "%s", alias->ifalias); 1326 rcu_read_unlock(); 1327 1328 return ret; 1329 } 1330 1331 /** 1332 * netdev_features_change - device changes features 1333 * @dev: device to cause notification 1334 * 1335 * Called to indicate a device has changed features. 1336 */ 1337 void netdev_features_change(struct net_device *dev) 1338 { 1339 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1340 } 1341 EXPORT_SYMBOL(netdev_features_change); 1342 1343 /** 1344 * netdev_state_change - device changes state 1345 * @dev: device to cause notification 1346 * 1347 * Called to indicate a device has changed state. This function calls 1348 * the notifier chains for netdev_chain and sends a NEWLINK message 1349 * to the routing socket. 1350 */ 1351 void netdev_state_change(struct net_device *dev) 1352 { 1353 if (dev->flags & IFF_UP) { 1354 struct netdev_notifier_change_info change_info = { 1355 .info.dev = dev, 1356 }; 1357 1358 call_netdevice_notifiers_info(NETDEV_CHANGE, 1359 &change_info.info); 1360 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1361 } 1362 } 1363 EXPORT_SYMBOL(netdev_state_change); 1364 1365 /** 1366 * __netdev_notify_peers - notify network peers about existence of @dev, 1367 * to be called when rtnl lock is already held. 1368 * @dev: network device 1369 * 1370 * Generate traffic such that interested network peers are aware of 1371 * @dev, such as by generating a gratuitous ARP. This may be used when 1372 * a device wants to inform the rest of the network about some sort of 1373 * reconfiguration such as a failover event or virtual machine 1374 * migration. 1375 */ 1376 void __netdev_notify_peers(struct net_device *dev) 1377 { 1378 ASSERT_RTNL(); 1379 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1380 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1381 } 1382 EXPORT_SYMBOL(__netdev_notify_peers); 1383 1384 /** 1385 * netdev_notify_peers - notify network peers about existence of @dev 1386 * @dev: network device 1387 * 1388 * Generate traffic such that interested network peers are aware of 1389 * @dev, such as by generating a gratuitous ARP. This may be used when 1390 * a device wants to inform the rest of the network about some sort of 1391 * reconfiguration such as a failover event or virtual machine 1392 * migration. 1393 */ 1394 void netdev_notify_peers(struct net_device *dev) 1395 { 1396 rtnl_lock(); 1397 __netdev_notify_peers(dev); 1398 rtnl_unlock(); 1399 } 1400 EXPORT_SYMBOL(netdev_notify_peers); 1401 1402 static int napi_threaded_poll(void *data); 1403 1404 static int napi_kthread_create(struct napi_struct *n) 1405 { 1406 int err = 0; 1407 1408 /* Create and wake up the kthread once to put it in 1409 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1410 * warning and work with loadavg. 1411 */ 1412 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1413 n->dev->name, n->napi_id); 1414 if (IS_ERR(n->thread)) { 1415 err = PTR_ERR(n->thread); 1416 pr_err("kthread_run failed with err %d\n", err); 1417 n->thread = NULL; 1418 } 1419 1420 return err; 1421 } 1422 1423 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1424 { 1425 const struct net_device_ops *ops = dev->netdev_ops; 1426 int ret; 1427 1428 ASSERT_RTNL(); 1429 dev_addr_check(dev); 1430 1431 if (!netif_device_present(dev)) { 1432 /* may be detached because parent is runtime-suspended */ 1433 if (dev->dev.parent) 1434 pm_runtime_resume(dev->dev.parent); 1435 if (!netif_device_present(dev)) 1436 return -ENODEV; 1437 } 1438 1439 /* Block netpoll from trying to do any rx path servicing. 1440 * If we don't do this there is a chance ndo_poll_controller 1441 * or ndo_poll may be running while we open the device 1442 */ 1443 netpoll_poll_disable(dev); 1444 1445 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1446 ret = notifier_to_errno(ret); 1447 if (ret) 1448 return ret; 1449 1450 set_bit(__LINK_STATE_START, &dev->state); 1451 1452 if (ops->ndo_validate_addr) 1453 ret = ops->ndo_validate_addr(dev); 1454 1455 if (!ret && ops->ndo_open) 1456 ret = ops->ndo_open(dev); 1457 1458 netpoll_poll_enable(dev); 1459 1460 if (ret) 1461 clear_bit(__LINK_STATE_START, &dev->state); 1462 else { 1463 dev->flags |= IFF_UP; 1464 dev_set_rx_mode(dev); 1465 dev_activate(dev); 1466 add_device_randomness(dev->dev_addr, dev->addr_len); 1467 } 1468 1469 return ret; 1470 } 1471 1472 /** 1473 * dev_open - prepare an interface for use. 1474 * @dev: device to open 1475 * @extack: netlink extended ack 1476 * 1477 * Takes a device from down to up state. The device's private open 1478 * function is invoked and then the multicast lists are loaded. Finally 1479 * the device is moved into the up state and a %NETDEV_UP message is 1480 * sent to the netdev notifier chain. 1481 * 1482 * Calling this function on an active interface is a nop. On a failure 1483 * a negative errno code is returned. 1484 */ 1485 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1486 { 1487 int ret; 1488 1489 if (dev->flags & IFF_UP) 1490 return 0; 1491 1492 ret = __dev_open(dev, extack); 1493 if (ret < 0) 1494 return ret; 1495 1496 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1497 call_netdevice_notifiers(NETDEV_UP, dev); 1498 1499 return ret; 1500 } 1501 EXPORT_SYMBOL(dev_open); 1502 1503 static void __dev_close_many(struct list_head *head) 1504 { 1505 struct net_device *dev; 1506 1507 ASSERT_RTNL(); 1508 might_sleep(); 1509 1510 list_for_each_entry(dev, head, close_list) { 1511 /* Temporarily disable netpoll until the interface is down */ 1512 netpoll_poll_disable(dev); 1513 1514 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1515 1516 clear_bit(__LINK_STATE_START, &dev->state); 1517 1518 /* Synchronize to scheduled poll. We cannot touch poll list, it 1519 * can be even on different cpu. So just clear netif_running(). 1520 * 1521 * dev->stop() will invoke napi_disable() on all of it's 1522 * napi_struct instances on this device. 1523 */ 1524 smp_mb__after_atomic(); /* Commit netif_running(). */ 1525 } 1526 1527 dev_deactivate_many(head); 1528 1529 list_for_each_entry(dev, head, close_list) { 1530 const struct net_device_ops *ops = dev->netdev_ops; 1531 1532 /* 1533 * Call the device specific close. This cannot fail. 1534 * Only if device is UP 1535 * 1536 * We allow it to be called even after a DETACH hot-plug 1537 * event. 1538 */ 1539 if (ops->ndo_stop) 1540 ops->ndo_stop(dev); 1541 1542 dev->flags &= ~IFF_UP; 1543 netpoll_poll_enable(dev); 1544 } 1545 } 1546 1547 static void __dev_close(struct net_device *dev) 1548 { 1549 LIST_HEAD(single); 1550 1551 list_add(&dev->close_list, &single); 1552 __dev_close_many(&single); 1553 list_del(&single); 1554 } 1555 1556 void dev_close_many(struct list_head *head, bool unlink) 1557 { 1558 struct net_device *dev, *tmp; 1559 1560 /* Remove the devices that don't need to be closed */ 1561 list_for_each_entry_safe(dev, tmp, head, close_list) 1562 if (!(dev->flags & IFF_UP)) 1563 list_del_init(&dev->close_list); 1564 1565 __dev_close_many(head); 1566 1567 list_for_each_entry_safe(dev, tmp, head, close_list) { 1568 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1569 call_netdevice_notifiers(NETDEV_DOWN, dev); 1570 if (unlink) 1571 list_del_init(&dev->close_list); 1572 } 1573 } 1574 EXPORT_SYMBOL(dev_close_many); 1575 1576 /** 1577 * dev_close - shutdown an interface. 1578 * @dev: device to shutdown 1579 * 1580 * This function moves an active device into down state. A 1581 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1582 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1583 * chain. 1584 */ 1585 void dev_close(struct net_device *dev) 1586 { 1587 if (dev->flags & IFF_UP) { 1588 LIST_HEAD(single); 1589 1590 list_add(&dev->close_list, &single); 1591 dev_close_many(&single, true); 1592 list_del(&single); 1593 } 1594 } 1595 EXPORT_SYMBOL(dev_close); 1596 1597 1598 /** 1599 * dev_disable_lro - disable Large Receive Offload on a device 1600 * @dev: device 1601 * 1602 * Disable Large Receive Offload (LRO) on a net device. Must be 1603 * called under RTNL. This is needed if received packets may be 1604 * forwarded to another interface. 1605 */ 1606 void dev_disable_lro(struct net_device *dev) 1607 { 1608 struct net_device *lower_dev; 1609 struct list_head *iter; 1610 1611 dev->wanted_features &= ~NETIF_F_LRO; 1612 netdev_update_features(dev); 1613 1614 if (unlikely(dev->features & NETIF_F_LRO)) 1615 netdev_WARN(dev, "failed to disable LRO!\n"); 1616 1617 netdev_for_each_lower_dev(dev, lower_dev, iter) 1618 dev_disable_lro(lower_dev); 1619 } 1620 EXPORT_SYMBOL(dev_disable_lro); 1621 1622 /** 1623 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1624 * @dev: device 1625 * 1626 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1627 * called under RTNL. This is needed if Generic XDP is installed on 1628 * the device. 1629 */ 1630 static void dev_disable_gro_hw(struct net_device *dev) 1631 { 1632 dev->wanted_features &= ~NETIF_F_GRO_HW; 1633 netdev_update_features(dev); 1634 1635 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1636 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1637 } 1638 1639 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1640 { 1641 #define N(val) \ 1642 case NETDEV_##val: \ 1643 return "NETDEV_" __stringify(val); 1644 switch (cmd) { 1645 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1646 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1647 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1648 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1649 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1650 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1651 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1652 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1653 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1654 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1655 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1656 N(XDP_FEAT_CHANGE) 1657 } 1658 #undef N 1659 return "UNKNOWN_NETDEV_EVENT"; 1660 } 1661 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1662 1663 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1664 struct net_device *dev) 1665 { 1666 struct netdev_notifier_info info = { 1667 .dev = dev, 1668 }; 1669 1670 return nb->notifier_call(nb, val, &info); 1671 } 1672 1673 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1674 struct net_device *dev) 1675 { 1676 int err; 1677 1678 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1679 err = notifier_to_errno(err); 1680 if (err) 1681 return err; 1682 1683 if (!(dev->flags & IFF_UP)) 1684 return 0; 1685 1686 call_netdevice_notifier(nb, NETDEV_UP, dev); 1687 return 0; 1688 } 1689 1690 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1691 struct net_device *dev) 1692 { 1693 if (dev->flags & IFF_UP) { 1694 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1695 dev); 1696 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1697 } 1698 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1699 } 1700 1701 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1702 struct net *net) 1703 { 1704 struct net_device *dev; 1705 int err; 1706 1707 for_each_netdev(net, dev) { 1708 err = call_netdevice_register_notifiers(nb, dev); 1709 if (err) 1710 goto rollback; 1711 } 1712 return 0; 1713 1714 rollback: 1715 for_each_netdev_continue_reverse(net, dev) 1716 call_netdevice_unregister_notifiers(nb, dev); 1717 return err; 1718 } 1719 1720 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1721 struct net *net) 1722 { 1723 struct net_device *dev; 1724 1725 for_each_netdev(net, dev) 1726 call_netdevice_unregister_notifiers(nb, dev); 1727 } 1728 1729 static int dev_boot_phase = 1; 1730 1731 /** 1732 * register_netdevice_notifier - register a network notifier block 1733 * @nb: notifier 1734 * 1735 * Register a notifier to be called when network device events occur. 1736 * The notifier passed is linked into the kernel structures and must 1737 * not be reused until it has been unregistered. A negative errno code 1738 * is returned on a failure. 1739 * 1740 * When registered all registration and up events are replayed 1741 * to the new notifier to allow device to have a race free 1742 * view of the network device list. 1743 */ 1744 1745 int register_netdevice_notifier(struct notifier_block *nb) 1746 { 1747 struct net *net; 1748 int err; 1749 1750 /* Close race with setup_net() and cleanup_net() */ 1751 down_write(&pernet_ops_rwsem); 1752 rtnl_lock(); 1753 err = raw_notifier_chain_register(&netdev_chain, nb); 1754 if (err) 1755 goto unlock; 1756 if (dev_boot_phase) 1757 goto unlock; 1758 for_each_net(net) { 1759 err = call_netdevice_register_net_notifiers(nb, net); 1760 if (err) 1761 goto rollback; 1762 } 1763 1764 unlock: 1765 rtnl_unlock(); 1766 up_write(&pernet_ops_rwsem); 1767 return err; 1768 1769 rollback: 1770 for_each_net_continue_reverse(net) 1771 call_netdevice_unregister_net_notifiers(nb, net); 1772 1773 raw_notifier_chain_unregister(&netdev_chain, nb); 1774 goto unlock; 1775 } 1776 EXPORT_SYMBOL(register_netdevice_notifier); 1777 1778 /** 1779 * unregister_netdevice_notifier - unregister a network notifier block 1780 * @nb: notifier 1781 * 1782 * Unregister a notifier previously registered by 1783 * register_netdevice_notifier(). The notifier is unlinked into the 1784 * kernel structures and may then be reused. A negative errno code 1785 * is returned on a failure. 1786 * 1787 * After unregistering unregister and down device events are synthesized 1788 * for all devices on the device list to the removed notifier to remove 1789 * the need for special case cleanup code. 1790 */ 1791 1792 int unregister_netdevice_notifier(struct notifier_block *nb) 1793 { 1794 struct net *net; 1795 int err; 1796 1797 /* Close race with setup_net() and cleanup_net() */ 1798 down_write(&pernet_ops_rwsem); 1799 rtnl_lock(); 1800 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1801 if (err) 1802 goto unlock; 1803 1804 for_each_net(net) 1805 call_netdevice_unregister_net_notifiers(nb, net); 1806 1807 unlock: 1808 rtnl_unlock(); 1809 up_write(&pernet_ops_rwsem); 1810 return err; 1811 } 1812 EXPORT_SYMBOL(unregister_netdevice_notifier); 1813 1814 static int __register_netdevice_notifier_net(struct net *net, 1815 struct notifier_block *nb, 1816 bool ignore_call_fail) 1817 { 1818 int err; 1819 1820 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1821 if (err) 1822 return err; 1823 if (dev_boot_phase) 1824 return 0; 1825 1826 err = call_netdevice_register_net_notifiers(nb, net); 1827 if (err && !ignore_call_fail) 1828 goto chain_unregister; 1829 1830 return 0; 1831 1832 chain_unregister: 1833 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1834 return err; 1835 } 1836 1837 static int __unregister_netdevice_notifier_net(struct net *net, 1838 struct notifier_block *nb) 1839 { 1840 int err; 1841 1842 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1843 if (err) 1844 return err; 1845 1846 call_netdevice_unregister_net_notifiers(nb, net); 1847 return 0; 1848 } 1849 1850 /** 1851 * register_netdevice_notifier_net - register a per-netns network notifier block 1852 * @net: network namespace 1853 * @nb: notifier 1854 * 1855 * Register a notifier to be called when network device events occur. 1856 * The notifier passed is linked into the kernel structures and must 1857 * not be reused until it has been unregistered. A negative errno code 1858 * is returned on a failure. 1859 * 1860 * When registered all registration and up events are replayed 1861 * to the new notifier to allow device to have a race free 1862 * view of the network device list. 1863 */ 1864 1865 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1866 { 1867 int err; 1868 1869 rtnl_lock(); 1870 err = __register_netdevice_notifier_net(net, nb, false); 1871 rtnl_unlock(); 1872 return err; 1873 } 1874 EXPORT_SYMBOL(register_netdevice_notifier_net); 1875 1876 /** 1877 * unregister_netdevice_notifier_net - unregister a per-netns 1878 * network notifier block 1879 * @net: network namespace 1880 * @nb: notifier 1881 * 1882 * Unregister a notifier previously registered by 1883 * register_netdevice_notifier_net(). The notifier is unlinked from the 1884 * kernel structures and may then be reused. A negative errno code 1885 * is returned on a failure. 1886 * 1887 * After unregistering unregister and down device events are synthesized 1888 * for all devices on the device list to the removed notifier to remove 1889 * the need for special case cleanup code. 1890 */ 1891 1892 int unregister_netdevice_notifier_net(struct net *net, 1893 struct notifier_block *nb) 1894 { 1895 int err; 1896 1897 rtnl_lock(); 1898 err = __unregister_netdevice_notifier_net(net, nb); 1899 rtnl_unlock(); 1900 return err; 1901 } 1902 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1903 1904 static void __move_netdevice_notifier_net(struct net *src_net, 1905 struct net *dst_net, 1906 struct notifier_block *nb) 1907 { 1908 __unregister_netdevice_notifier_net(src_net, nb); 1909 __register_netdevice_notifier_net(dst_net, nb, true); 1910 } 1911 1912 int register_netdevice_notifier_dev_net(struct net_device *dev, 1913 struct notifier_block *nb, 1914 struct netdev_net_notifier *nn) 1915 { 1916 int err; 1917 1918 rtnl_lock(); 1919 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1920 if (!err) { 1921 nn->nb = nb; 1922 list_add(&nn->list, &dev->net_notifier_list); 1923 } 1924 rtnl_unlock(); 1925 return err; 1926 } 1927 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1928 1929 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1930 struct notifier_block *nb, 1931 struct netdev_net_notifier *nn) 1932 { 1933 int err; 1934 1935 rtnl_lock(); 1936 list_del(&nn->list); 1937 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1938 rtnl_unlock(); 1939 return err; 1940 } 1941 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1942 1943 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1944 struct net *net) 1945 { 1946 struct netdev_net_notifier *nn; 1947 1948 list_for_each_entry(nn, &dev->net_notifier_list, list) 1949 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1950 } 1951 1952 /** 1953 * call_netdevice_notifiers_info - call all network notifier blocks 1954 * @val: value passed unmodified to notifier function 1955 * @info: notifier information data 1956 * 1957 * Call all network notifier blocks. Parameters and return value 1958 * are as for raw_notifier_call_chain(). 1959 */ 1960 1961 int call_netdevice_notifiers_info(unsigned long val, 1962 struct netdev_notifier_info *info) 1963 { 1964 struct net *net = dev_net(info->dev); 1965 int ret; 1966 1967 ASSERT_RTNL(); 1968 1969 /* Run per-netns notifier block chain first, then run the global one. 1970 * Hopefully, one day, the global one is going to be removed after 1971 * all notifier block registrators get converted to be per-netns. 1972 */ 1973 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1974 if (ret & NOTIFY_STOP_MASK) 1975 return ret; 1976 return raw_notifier_call_chain(&netdev_chain, val, info); 1977 } 1978 1979 /** 1980 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1981 * for and rollback on error 1982 * @val_up: value passed unmodified to notifier function 1983 * @val_down: value passed unmodified to the notifier function when 1984 * recovering from an error on @val_up 1985 * @info: notifier information data 1986 * 1987 * Call all per-netns network notifier blocks, but not notifier blocks on 1988 * the global notifier chain. Parameters and return value are as for 1989 * raw_notifier_call_chain_robust(). 1990 */ 1991 1992 static int 1993 call_netdevice_notifiers_info_robust(unsigned long val_up, 1994 unsigned long val_down, 1995 struct netdev_notifier_info *info) 1996 { 1997 struct net *net = dev_net(info->dev); 1998 1999 ASSERT_RTNL(); 2000 2001 return raw_notifier_call_chain_robust(&net->netdev_chain, 2002 val_up, val_down, info); 2003 } 2004 2005 static int call_netdevice_notifiers_extack(unsigned long val, 2006 struct net_device *dev, 2007 struct netlink_ext_ack *extack) 2008 { 2009 struct netdev_notifier_info info = { 2010 .dev = dev, 2011 .extack = extack, 2012 }; 2013 2014 return call_netdevice_notifiers_info(val, &info); 2015 } 2016 2017 /** 2018 * call_netdevice_notifiers - call all network notifier blocks 2019 * @val: value passed unmodified to notifier function 2020 * @dev: net_device pointer passed unmodified to notifier function 2021 * 2022 * Call all network notifier blocks. Parameters and return value 2023 * are as for raw_notifier_call_chain(). 2024 */ 2025 2026 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2027 { 2028 return call_netdevice_notifiers_extack(val, dev, NULL); 2029 } 2030 EXPORT_SYMBOL(call_netdevice_notifiers); 2031 2032 /** 2033 * call_netdevice_notifiers_mtu - call all network notifier blocks 2034 * @val: value passed unmodified to notifier function 2035 * @dev: net_device pointer passed unmodified to notifier function 2036 * @arg: additional u32 argument passed to the notifier function 2037 * 2038 * Call all network notifier blocks. Parameters and return value 2039 * are as for raw_notifier_call_chain(). 2040 */ 2041 static int call_netdevice_notifiers_mtu(unsigned long val, 2042 struct net_device *dev, u32 arg) 2043 { 2044 struct netdev_notifier_info_ext info = { 2045 .info.dev = dev, 2046 .ext.mtu = arg, 2047 }; 2048 2049 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2050 2051 return call_netdevice_notifiers_info(val, &info.info); 2052 } 2053 2054 #ifdef CONFIG_NET_INGRESS 2055 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2056 2057 void net_inc_ingress_queue(void) 2058 { 2059 static_branch_inc(&ingress_needed_key); 2060 } 2061 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2062 2063 void net_dec_ingress_queue(void) 2064 { 2065 static_branch_dec(&ingress_needed_key); 2066 } 2067 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2068 #endif 2069 2070 #ifdef CONFIG_NET_EGRESS 2071 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2072 2073 void net_inc_egress_queue(void) 2074 { 2075 static_branch_inc(&egress_needed_key); 2076 } 2077 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2078 2079 void net_dec_egress_queue(void) 2080 { 2081 static_branch_dec(&egress_needed_key); 2082 } 2083 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2084 #endif 2085 2086 #ifdef CONFIG_NET_CLS_ACT 2087 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key); 2088 EXPORT_SYMBOL(tcf_bypass_check_needed_key); 2089 #endif 2090 2091 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2092 EXPORT_SYMBOL(netstamp_needed_key); 2093 #ifdef CONFIG_JUMP_LABEL 2094 static atomic_t netstamp_needed_deferred; 2095 static atomic_t netstamp_wanted; 2096 static void netstamp_clear(struct work_struct *work) 2097 { 2098 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2099 int wanted; 2100 2101 wanted = atomic_add_return(deferred, &netstamp_wanted); 2102 if (wanted > 0) 2103 static_branch_enable(&netstamp_needed_key); 2104 else 2105 static_branch_disable(&netstamp_needed_key); 2106 } 2107 static DECLARE_WORK(netstamp_work, netstamp_clear); 2108 #endif 2109 2110 void net_enable_timestamp(void) 2111 { 2112 #ifdef CONFIG_JUMP_LABEL 2113 int wanted = atomic_read(&netstamp_wanted); 2114 2115 while (wanted > 0) { 2116 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2117 return; 2118 } 2119 atomic_inc(&netstamp_needed_deferred); 2120 schedule_work(&netstamp_work); 2121 #else 2122 static_branch_inc(&netstamp_needed_key); 2123 #endif 2124 } 2125 EXPORT_SYMBOL(net_enable_timestamp); 2126 2127 void net_disable_timestamp(void) 2128 { 2129 #ifdef CONFIG_JUMP_LABEL 2130 int wanted = atomic_read(&netstamp_wanted); 2131 2132 while (wanted > 1) { 2133 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2134 return; 2135 } 2136 atomic_dec(&netstamp_needed_deferred); 2137 schedule_work(&netstamp_work); 2138 #else 2139 static_branch_dec(&netstamp_needed_key); 2140 #endif 2141 } 2142 EXPORT_SYMBOL(net_disable_timestamp); 2143 2144 static inline void net_timestamp_set(struct sk_buff *skb) 2145 { 2146 skb->tstamp = 0; 2147 skb->mono_delivery_time = 0; 2148 if (static_branch_unlikely(&netstamp_needed_key)) 2149 skb->tstamp = ktime_get_real(); 2150 } 2151 2152 #define net_timestamp_check(COND, SKB) \ 2153 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2154 if ((COND) && !(SKB)->tstamp) \ 2155 (SKB)->tstamp = ktime_get_real(); \ 2156 } \ 2157 2158 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2159 { 2160 return __is_skb_forwardable(dev, skb, true); 2161 } 2162 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2163 2164 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2165 bool check_mtu) 2166 { 2167 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2168 2169 if (likely(!ret)) { 2170 skb->protocol = eth_type_trans(skb, dev); 2171 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2172 } 2173 2174 return ret; 2175 } 2176 2177 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2178 { 2179 return __dev_forward_skb2(dev, skb, true); 2180 } 2181 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2182 2183 /** 2184 * dev_forward_skb - loopback an skb to another netif 2185 * 2186 * @dev: destination network device 2187 * @skb: buffer to forward 2188 * 2189 * return values: 2190 * NET_RX_SUCCESS (no congestion) 2191 * NET_RX_DROP (packet was dropped, but freed) 2192 * 2193 * dev_forward_skb can be used for injecting an skb from the 2194 * start_xmit function of one device into the receive queue 2195 * of another device. 2196 * 2197 * The receiving device may be in another namespace, so 2198 * we have to clear all information in the skb that could 2199 * impact namespace isolation. 2200 */ 2201 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2202 { 2203 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2204 } 2205 EXPORT_SYMBOL_GPL(dev_forward_skb); 2206 2207 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2208 { 2209 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2210 } 2211 2212 static inline int deliver_skb(struct sk_buff *skb, 2213 struct packet_type *pt_prev, 2214 struct net_device *orig_dev) 2215 { 2216 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2217 return -ENOMEM; 2218 refcount_inc(&skb->users); 2219 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2220 } 2221 2222 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2223 struct packet_type **pt, 2224 struct net_device *orig_dev, 2225 __be16 type, 2226 struct list_head *ptype_list) 2227 { 2228 struct packet_type *ptype, *pt_prev = *pt; 2229 2230 list_for_each_entry_rcu(ptype, ptype_list, list) { 2231 if (ptype->type != type) 2232 continue; 2233 if (pt_prev) 2234 deliver_skb(skb, pt_prev, orig_dev); 2235 pt_prev = ptype; 2236 } 2237 *pt = pt_prev; 2238 } 2239 2240 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2241 { 2242 if (!ptype->af_packet_priv || !skb->sk) 2243 return false; 2244 2245 if (ptype->id_match) 2246 return ptype->id_match(ptype, skb->sk); 2247 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2248 return true; 2249 2250 return false; 2251 } 2252 2253 /** 2254 * dev_nit_active - return true if any network interface taps are in use 2255 * 2256 * @dev: network device to check for the presence of taps 2257 */ 2258 bool dev_nit_active(struct net_device *dev) 2259 { 2260 return !list_empty(&net_hotdata.ptype_all) || 2261 !list_empty(&dev->ptype_all); 2262 } 2263 EXPORT_SYMBOL_GPL(dev_nit_active); 2264 2265 /* 2266 * Support routine. Sends outgoing frames to any network 2267 * taps currently in use. 2268 */ 2269 2270 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2271 { 2272 struct list_head *ptype_list = &net_hotdata.ptype_all; 2273 struct packet_type *ptype, *pt_prev = NULL; 2274 struct sk_buff *skb2 = NULL; 2275 2276 rcu_read_lock(); 2277 again: 2278 list_for_each_entry_rcu(ptype, ptype_list, list) { 2279 if (READ_ONCE(ptype->ignore_outgoing)) 2280 continue; 2281 2282 /* Never send packets back to the socket 2283 * they originated from - MvS ([email protected]) 2284 */ 2285 if (skb_loop_sk(ptype, skb)) 2286 continue; 2287 2288 if (pt_prev) { 2289 deliver_skb(skb2, pt_prev, skb->dev); 2290 pt_prev = ptype; 2291 continue; 2292 } 2293 2294 /* need to clone skb, done only once */ 2295 skb2 = skb_clone(skb, GFP_ATOMIC); 2296 if (!skb2) 2297 goto out_unlock; 2298 2299 net_timestamp_set(skb2); 2300 2301 /* skb->nh should be correctly 2302 * set by sender, so that the second statement is 2303 * just protection against buggy protocols. 2304 */ 2305 skb_reset_mac_header(skb2); 2306 2307 if (skb_network_header(skb2) < skb2->data || 2308 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2309 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2310 ntohs(skb2->protocol), 2311 dev->name); 2312 skb_reset_network_header(skb2); 2313 } 2314 2315 skb2->transport_header = skb2->network_header; 2316 skb2->pkt_type = PACKET_OUTGOING; 2317 pt_prev = ptype; 2318 } 2319 2320 if (ptype_list == &net_hotdata.ptype_all) { 2321 ptype_list = &dev->ptype_all; 2322 goto again; 2323 } 2324 out_unlock: 2325 if (pt_prev) { 2326 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2327 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2328 else 2329 kfree_skb(skb2); 2330 } 2331 rcu_read_unlock(); 2332 } 2333 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2334 2335 /** 2336 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2337 * @dev: Network device 2338 * @txq: number of queues available 2339 * 2340 * If real_num_tx_queues is changed the tc mappings may no longer be 2341 * valid. To resolve this verify the tc mapping remains valid and if 2342 * not NULL the mapping. With no priorities mapping to this 2343 * offset/count pair it will no longer be used. In the worst case TC0 2344 * is invalid nothing can be done so disable priority mappings. If is 2345 * expected that drivers will fix this mapping if they can before 2346 * calling netif_set_real_num_tx_queues. 2347 */ 2348 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2349 { 2350 int i; 2351 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2352 2353 /* If TC0 is invalidated disable TC mapping */ 2354 if (tc->offset + tc->count > txq) { 2355 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2356 dev->num_tc = 0; 2357 return; 2358 } 2359 2360 /* Invalidated prio to tc mappings set to TC0 */ 2361 for (i = 1; i < TC_BITMASK + 1; i++) { 2362 int q = netdev_get_prio_tc_map(dev, i); 2363 2364 tc = &dev->tc_to_txq[q]; 2365 if (tc->offset + tc->count > txq) { 2366 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", 2367 i, q); 2368 netdev_set_prio_tc_map(dev, i, 0); 2369 } 2370 } 2371 } 2372 2373 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2374 { 2375 if (dev->num_tc) { 2376 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2377 int i; 2378 2379 /* walk through the TCs and see if it falls into any of them */ 2380 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2381 if ((txq - tc->offset) < tc->count) 2382 return i; 2383 } 2384 2385 /* didn't find it, just return -1 to indicate no match */ 2386 return -1; 2387 } 2388 2389 return 0; 2390 } 2391 EXPORT_SYMBOL(netdev_txq_to_tc); 2392 2393 #ifdef CONFIG_XPS 2394 static struct static_key xps_needed __read_mostly; 2395 static struct static_key xps_rxqs_needed __read_mostly; 2396 static DEFINE_MUTEX(xps_map_mutex); 2397 #define xmap_dereference(P) \ 2398 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2399 2400 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2401 struct xps_dev_maps *old_maps, int tci, u16 index) 2402 { 2403 struct xps_map *map = NULL; 2404 int pos; 2405 2406 map = xmap_dereference(dev_maps->attr_map[tci]); 2407 if (!map) 2408 return false; 2409 2410 for (pos = map->len; pos--;) { 2411 if (map->queues[pos] != index) 2412 continue; 2413 2414 if (map->len > 1) { 2415 map->queues[pos] = map->queues[--map->len]; 2416 break; 2417 } 2418 2419 if (old_maps) 2420 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2421 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2422 kfree_rcu(map, rcu); 2423 return false; 2424 } 2425 2426 return true; 2427 } 2428 2429 static bool remove_xps_queue_cpu(struct net_device *dev, 2430 struct xps_dev_maps *dev_maps, 2431 int cpu, u16 offset, u16 count) 2432 { 2433 int num_tc = dev_maps->num_tc; 2434 bool active = false; 2435 int tci; 2436 2437 for (tci = cpu * num_tc; num_tc--; tci++) { 2438 int i, j; 2439 2440 for (i = count, j = offset; i--; j++) { 2441 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2442 break; 2443 } 2444 2445 active |= i < 0; 2446 } 2447 2448 return active; 2449 } 2450 2451 static void reset_xps_maps(struct net_device *dev, 2452 struct xps_dev_maps *dev_maps, 2453 enum xps_map_type type) 2454 { 2455 static_key_slow_dec_cpuslocked(&xps_needed); 2456 if (type == XPS_RXQS) 2457 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2458 2459 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2460 2461 kfree_rcu(dev_maps, rcu); 2462 } 2463 2464 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2465 u16 offset, u16 count) 2466 { 2467 struct xps_dev_maps *dev_maps; 2468 bool active = false; 2469 int i, j; 2470 2471 dev_maps = xmap_dereference(dev->xps_maps[type]); 2472 if (!dev_maps) 2473 return; 2474 2475 for (j = 0; j < dev_maps->nr_ids; j++) 2476 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2477 if (!active) 2478 reset_xps_maps(dev, dev_maps, type); 2479 2480 if (type == XPS_CPUS) { 2481 for (i = offset + (count - 1); count--; i--) 2482 netdev_queue_numa_node_write( 2483 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2484 } 2485 } 2486 2487 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2488 u16 count) 2489 { 2490 if (!static_key_false(&xps_needed)) 2491 return; 2492 2493 cpus_read_lock(); 2494 mutex_lock(&xps_map_mutex); 2495 2496 if (static_key_false(&xps_rxqs_needed)) 2497 clean_xps_maps(dev, XPS_RXQS, offset, count); 2498 2499 clean_xps_maps(dev, XPS_CPUS, offset, count); 2500 2501 mutex_unlock(&xps_map_mutex); 2502 cpus_read_unlock(); 2503 } 2504 2505 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2506 { 2507 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2508 } 2509 2510 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2511 u16 index, bool is_rxqs_map) 2512 { 2513 struct xps_map *new_map; 2514 int alloc_len = XPS_MIN_MAP_ALLOC; 2515 int i, pos; 2516 2517 for (pos = 0; map && pos < map->len; pos++) { 2518 if (map->queues[pos] != index) 2519 continue; 2520 return map; 2521 } 2522 2523 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2524 if (map) { 2525 if (pos < map->alloc_len) 2526 return map; 2527 2528 alloc_len = map->alloc_len * 2; 2529 } 2530 2531 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2532 * map 2533 */ 2534 if (is_rxqs_map) 2535 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2536 else 2537 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2538 cpu_to_node(attr_index)); 2539 if (!new_map) 2540 return NULL; 2541 2542 for (i = 0; i < pos; i++) 2543 new_map->queues[i] = map->queues[i]; 2544 new_map->alloc_len = alloc_len; 2545 new_map->len = pos; 2546 2547 return new_map; 2548 } 2549 2550 /* Copy xps maps at a given index */ 2551 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2552 struct xps_dev_maps *new_dev_maps, int index, 2553 int tc, bool skip_tc) 2554 { 2555 int i, tci = index * dev_maps->num_tc; 2556 struct xps_map *map; 2557 2558 /* copy maps belonging to foreign traffic classes */ 2559 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2560 if (i == tc && skip_tc) 2561 continue; 2562 2563 /* fill in the new device map from the old device map */ 2564 map = xmap_dereference(dev_maps->attr_map[tci]); 2565 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2566 } 2567 } 2568 2569 /* Must be called under cpus_read_lock */ 2570 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2571 u16 index, enum xps_map_type type) 2572 { 2573 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2574 const unsigned long *online_mask = NULL; 2575 bool active = false, copy = false; 2576 int i, j, tci, numa_node_id = -2; 2577 int maps_sz, num_tc = 1, tc = 0; 2578 struct xps_map *map, *new_map; 2579 unsigned int nr_ids; 2580 2581 WARN_ON_ONCE(index >= dev->num_tx_queues); 2582 2583 if (dev->num_tc) { 2584 /* Do not allow XPS on subordinate device directly */ 2585 num_tc = dev->num_tc; 2586 if (num_tc < 0) 2587 return -EINVAL; 2588 2589 /* If queue belongs to subordinate dev use its map */ 2590 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2591 2592 tc = netdev_txq_to_tc(dev, index); 2593 if (tc < 0) 2594 return -EINVAL; 2595 } 2596 2597 mutex_lock(&xps_map_mutex); 2598 2599 dev_maps = xmap_dereference(dev->xps_maps[type]); 2600 if (type == XPS_RXQS) { 2601 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2602 nr_ids = dev->num_rx_queues; 2603 } else { 2604 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2605 if (num_possible_cpus() > 1) 2606 online_mask = cpumask_bits(cpu_online_mask); 2607 nr_ids = nr_cpu_ids; 2608 } 2609 2610 if (maps_sz < L1_CACHE_BYTES) 2611 maps_sz = L1_CACHE_BYTES; 2612 2613 /* The old dev_maps could be larger or smaller than the one we're 2614 * setting up now, as dev->num_tc or nr_ids could have been updated in 2615 * between. We could try to be smart, but let's be safe instead and only 2616 * copy foreign traffic classes if the two map sizes match. 2617 */ 2618 if (dev_maps && 2619 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2620 copy = true; 2621 2622 /* allocate memory for queue storage */ 2623 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2624 j < nr_ids;) { 2625 if (!new_dev_maps) { 2626 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2627 if (!new_dev_maps) { 2628 mutex_unlock(&xps_map_mutex); 2629 return -ENOMEM; 2630 } 2631 2632 new_dev_maps->nr_ids = nr_ids; 2633 new_dev_maps->num_tc = num_tc; 2634 } 2635 2636 tci = j * num_tc + tc; 2637 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2638 2639 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2640 if (!map) 2641 goto error; 2642 2643 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2644 } 2645 2646 if (!new_dev_maps) 2647 goto out_no_new_maps; 2648 2649 if (!dev_maps) { 2650 /* Increment static keys at most once per type */ 2651 static_key_slow_inc_cpuslocked(&xps_needed); 2652 if (type == XPS_RXQS) 2653 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2654 } 2655 2656 for (j = 0; j < nr_ids; j++) { 2657 bool skip_tc = false; 2658 2659 tci = j * num_tc + tc; 2660 if (netif_attr_test_mask(j, mask, nr_ids) && 2661 netif_attr_test_online(j, online_mask, nr_ids)) { 2662 /* add tx-queue to CPU/rx-queue maps */ 2663 int pos = 0; 2664 2665 skip_tc = true; 2666 2667 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2668 while ((pos < map->len) && (map->queues[pos] != index)) 2669 pos++; 2670 2671 if (pos == map->len) 2672 map->queues[map->len++] = index; 2673 #ifdef CONFIG_NUMA 2674 if (type == XPS_CPUS) { 2675 if (numa_node_id == -2) 2676 numa_node_id = cpu_to_node(j); 2677 else if (numa_node_id != cpu_to_node(j)) 2678 numa_node_id = -1; 2679 } 2680 #endif 2681 } 2682 2683 if (copy) 2684 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2685 skip_tc); 2686 } 2687 2688 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2689 2690 /* Cleanup old maps */ 2691 if (!dev_maps) 2692 goto out_no_old_maps; 2693 2694 for (j = 0; j < dev_maps->nr_ids; j++) { 2695 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2696 map = xmap_dereference(dev_maps->attr_map[tci]); 2697 if (!map) 2698 continue; 2699 2700 if (copy) { 2701 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2702 if (map == new_map) 2703 continue; 2704 } 2705 2706 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2707 kfree_rcu(map, rcu); 2708 } 2709 } 2710 2711 old_dev_maps = dev_maps; 2712 2713 out_no_old_maps: 2714 dev_maps = new_dev_maps; 2715 active = true; 2716 2717 out_no_new_maps: 2718 if (type == XPS_CPUS) 2719 /* update Tx queue numa node */ 2720 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2721 (numa_node_id >= 0) ? 2722 numa_node_id : NUMA_NO_NODE); 2723 2724 if (!dev_maps) 2725 goto out_no_maps; 2726 2727 /* removes tx-queue from unused CPUs/rx-queues */ 2728 for (j = 0; j < dev_maps->nr_ids; j++) { 2729 tci = j * dev_maps->num_tc; 2730 2731 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2732 if (i == tc && 2733 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2734 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2735 continue; 2736 2737 active |= remove_xps_queue(dev_maps, 2738 copy ? old_dev_maps : NULL, 2739 tci, index); 2740 } 2741 } 2742 2743 if (old_dev_maps) 2744 kfree_rcu(old_dev_maps, rcu); 2745 2746 /* free map if not active */ 2747 if (!active) 2748 reset_xps_maps(dev, dev_maps, type); 2749 2750 out_no_maps: 2751 mutex_unlock(&xps_map_mutex); 2752 2753 return 0; 2754 error: 2755 /* remove any maps that we added */ 2756 for (j = 0; j < nr_ids; j++) { 2757 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2758 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2759 map = copy ? 2760 xmap_dereference(dev_maps->attr_map[tci]) : 2761 NULL; 2762 if (new_map && new_map != map) 2763 kfree(new_map); 2764 } 2765 } 2766 2767 mutex_unlock(&xps_map_mutex); 2768 2769 kfree(new_dev_maps); 2770 return -ENOMEM; 2771 } 2772 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2773 2774 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2775 u16 index) 2776 { 2777 int ret; 2778 2779 cpus_read_lock(); 2780 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2781 cpus_read_unlock(); 2782 2783 return ret; 2784 } 2785 EXPORT_SYMBOL(netif_set_xps_queue); 2786 2787 #endif 2788 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2789 { 2790 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2791 2792 /* Unbind any subordinate channels */ 2793 while (txq-- != &dev->_tx[0]) { 2794 if (txq->sb_dev) 2795 netdev_unbind_sb_channel(dev, txq->sb_dev); 2796 } 2797 } 2798 2799 void netdev_reset_tc(struct net_device *dev) 2800 { 2801 #ifdef CONFIG_XPS 2802 netif_reset_xps_queues_gt(dev, 0); 2803 #endif 2804 netdev_unbind_all_sb_channels(dev); 2805 2806 /* Reset TC configuration of device */ 2807 dev->num_tc = 0; 2808 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2809 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2810 } 2811 EXPORT_SYMBOL(netdev_reset_tc); 2812 2813 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2814 { 2815 if (tc >= dev->num_tc) 2816 return -EINVAL; 2817 2818 #ifdef CONFIG_XPS 2819 netif_reset_xps_queues(dev, offset, count); 2820 #endif 2821 dev->tc_to_txq[tc].count = count; 2822 dev->tc_to_txq[tc].offset = offset; 2823 return 0; 2824 } 2825 EXPORT_SYMBOL(netdev_set_tc_queue); 2826 2827 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2828 { 2829 if (num_tc > TC_MAX_QUEUE) 2830 return -EINVAL; 2831 2832 #ifdef CONFIG_XPS 2833 netif_reset_xps_queues_gt(dev, 0); 2834 #endif 2835 netdev_unbind_all_sb_channels(dev); 2836 2837 dev->num_tc = num_tc; 2838 return 0; 2839 } 2840 EXPORT_SYMBOL(netdev_set_num_tc); 2841 2842 void netdev_unbind_sb_channel(struct net_device *dev, 2843 struct net_device *sb_dev) 2844 { 2845 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2846 2847 #ifdef CONFIG_XPS 2848 netif_reset_xps_queues_gt(sb_dev, 0); 2849 #endif 2850 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2851 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2852 2853 while (txq-- != &dev->_tx[0]) { 2854 if (txq->sb_dev == sb_dev) 2855 txq->sb_dev = NULL; 2856 } 2857 } 2858 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2859 2860 int netdev_bind_sb_channel_queue(struct net_device *dev, 2861 struct net_device *sb_dev, 2862 u8 tc, u16 count, u16 offset) 2863 { 2864 /* Make certain the sb_dev and dev are already configured */ 2865 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2866 return -EINVAL; 2867 2868 /* We cannot hand out queues we don't have */ 2869 if ((offset + count) > dev->real_num_tx_queues) 2870 return -EINVAL; 2871 2872 /* Record the mapping */ 2873 sb_dev->tc_to_txq[tc].count = count; 2874 sb_dev->tc_to_txq[tc].offset = offset; 2875 2876 /* Provide a way for Tx queue to find the tc_to_txq map or 2877 * XPS map for itself. 2878 */ 2879 while (count--) 2880 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2881 2882 return 0; 2883 } 2884 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2885 2886 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2887 { 2888 /* Do not use a multiqueue device to represent a subordinate channel */ 2889 if (netif_is_multiqueue(dev)) 2890 return -ENODEV; 2891 2892 /* We allow channels 1 - 32767 to be used for subordinate channels. 2893 * Channel 0 is meant to be "native" mode and used only to represent 2894 * the main root device. We allow writing 0 to reset the device back 2895 * to normal mode after being used as a subordinate channel. 2896 */ 2897 if (channel > S16_MAX) 2898 return -EINVAL; 2899 2900 dev->num_tc = -channel; 2901 2902 return 0; 2903 } 2904 EXPORT_SYMBOL(netdev_set_sb_channel); 2905 2906 /* 2907 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2908 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2909 */ 2910 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2911 { 2912 bool disabling; 2913 int rc; 2914 2915 disabling = txq < dev->real_num_tx_queues; 2916 2917 if (txq < 1 || txq > dev->num_tx_queues) 2918 return -EINVAL; 2919 2920 if (dev->reg_state == NETREG_REGISTERED || 2921 dev->reg_state == NETREG_UNREGISTERING) { 2922 ASSERT_RTNL(); 2923 2924 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2925 txq); 2926 if (rc) 2927 return rc; 2928 2929 if (dev->num_tc) 2930 netif_setup_tc(dev, txq); 2931 2932 dev_qdisc_change_real_num_tx(dev, txq); 2933 2934 dev->real_num_tx_queues = txq; 2935 2936 if (disabling) { 2937 synchronize_net(); 2938 qdisc_reset_all_tx_gt(dev, txq); 2939 #ifdef CONFIG_XPS 2940 netif_reset_xps_queues_gt(dev, txq); 2941 #endif 2942 } 2943 } else { 2944 dev->real_num_tx_queues = txq; 2945 } 2946 2947 return 0; 2948 } 2949 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2950 2951 #ifdef CONFIG_SYSFS 2952 /** 2953 * netif_set_real_num_rx_queues - set actual number of RX queues used 2954 * @dev: Network device 2955 * @rxq: Actual number of RX queues 2956 * 2957 * This must be called either with the rtnl_lock held or before 2958 * registration of the net device. Returns 0 on success, or a 2959 * negative error code. If called before registration, it always 2960 * succeeds. 2961 */ 2962 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2963 { 2964 int rc; 2965 2966 if (rxq < 1 || rxq > dev->num_rx_queues) 2967 return -EINVAL; 2968 2969 if (dev->reg_state == NETREG_REGISTERED) { 2970 ASSERT_RTNL(); 2971 2972 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2973 rxq); 2974 if (rc) 2975 return rc; 2976 } 2977 2978 dev->real_num_rx_queues = rxq; 2979 return 0; 2980 } 2981 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2982 #endif 2983 2984 /** 2985 * netif_set_real_num_queues - set actual number of RX and TX queues used 2986 * @dev: Network device 2987 * @txq: Actual number of TX queues 2988 * @rxq: Actual number of RX queues 2989 * 2990 * Set the real number of both TX and RX queues. 2991 * Does nothing if the number of queues is already correct. 2992 */ 2993 int netif_set_real_num_queues(struct net_device *dev, 2994 unsigned int txq, unsigned int rxq) 2995 { 2996 unsigned int old_rxq = dev->real_num_rx_queues; 2997 int err; 2998 2999 if (txq < 1 || txq > dev->num_tx_queues || 3000 rxq < 1 || rxq > dev->num_rx_queues) 3001 return -EINVAL; 3002 3003 /* Start from increases, so the error path only does decreases - 3004 * decreases can't fail. 3005 */ 3006 if (rxq > dev->real_num_rx_queues) { 3007 err = netif_set_real_num_rx_queues(dev, rxq); 3008 if (err) 3009 return err; 3010 } 3011 if (txq > dev->real_num_tx_queues) { 3012 err = netif_set_real_num_tx_queues(dev, txq); 3013 if (err) 3014 goto undo_rx; 3015 } 3016 if (rxq < dev->real_num_rx_queues) 3017 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3018 if (txq < dev->real_num_tx_queues) 3019 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3020 3021 return 0; 3022 undo_rx: 3023 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3024 return err; 3025 } 3026 EXPORT_SYMBOL(netif_set_real_num_queues); 3027 3028 /** 3029 * netif_set_tso_max_size() - set the max size of TSO frames supported 3030 * @dev: netdev to update 3031 * @size: max skb->len of a TSO frame 3032 * 3033 * Set the limit on the size of TSO super-frames the device can handle. 3034 * Unless explicitly set the stack will assume the value of 3035 * %GSO_LEGACY_MAX_SIZE. 3036 */ 3037 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3038 { 3039 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3040 if (size < READ_ONCE(dev->gso_max_size)) 3041 netif_set_gso_max_size(dev, size); 3042 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3043 netif_set_gso_ipv4_max_size(dev, size); 3044 } 3045 EXPORT_SYMBOL(netif_set_tso_max_size); 3046 3047 /** 3048 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3049 * @dev: netdev to update 3050 * @segs: max number of TCP segments 3051 * 3052 * Set the limit on the number of TCP segments the device can generate from 3053 * a single TSO super-frame. 3054 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3055 */ 3056 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3057 { 3058 dev->tso_max_segs = segs; 3059 if (segs < READ_ONCE(dev->gso_max_segs)) 3060 netif_set_gso_max_segs(dev, segs); 3061 } 3062 EXPORT_SYMBOL(netif_set_tso_max_segs); 3063 3064 /** 3065 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3066 * @to: netdev to update 3067 * @from: netdev from which to copy the limits 3068 */ 3069 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3070 { 3071 netif_set_tso_max_size(to, from->tso_max_size); 3072 netif_set_tso_max_segs(to, from->tso_max_segs); 3073 } 3074 EXPORT_SYMBOL(netif_inherit_tso_max); 3075 3076 /** 3077 * netif_get_num_default_rss_queues - default number of RSS queues 3078 * 3079 * Default value is the number of physical cores if there are only 1 or 2, or 3080 * divided by 2 if there are more. 3081 */ 3082 int netif_get_num_default_rss_queues(void) 3083 { 3084 cpumask_var_t cpus; 3085 int cpu, count = 0; 3086 3087 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3088 return 1; 3089 3090 cpumask_copy(cpus, cpu_online_mask); 3091 for_each_cpu(cpu, cpus) { 3092 ++count; 3093 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3094 } 3095 free_cpumask_var(cpus); 3096 3097 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3098 } 3099 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3100 3101 static void __netif_reschedule(struct Qdisc *q) 3102 { 3103 struct softnet_data *sd; 3104 unsigned long flags; 3105 3106 local_irq_save(flags); 3107 sd = this_cpu_ptr(&softnet_data); 3108 q->next_sched = NULL; 3109 *sd->output_queue_tailp = q; 3110 sd->output_queue_tailp = &q->next_sched; 3111 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3112 local_irq_restore(flags); 3113 } 3114 3115 void __netif_schedule(struct Qdisc *q) 3116 { 3117 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3118 __netif_reschedule(q); 3119 } 3120 EXPORT_SYMBOL(__netif_schedule); 3121 3122 struct dev_kfree_skb_cb { 3123 enum skb_drop_reason reason; 3124 }; 3125 3126 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3127 { 3128 return (struct dev_kfree_skb_cb *)skb->cb; 3129 } 3130 3131 void netif_schedule_queue(struct netdev_queue *txq) 3132 { 3133 rcu_read_lock(); 3134 if (!netif_xmit_stopped(txq)) { 3135 struct Qdisc *q = rcu_dereference(txq->qdisc); 3136 3137 __netif_schedule(q); 3138 } 3139 rcu_read_unlock(); 3140 } 3141 EXPORT_SYMBOL(netif_schedule_queue); 3142 3143 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3144 { 3145 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3146 struct Qdisc *q; 3147 3148 rcu_read_lock(); 3149 q = rcu_dereference(dev_queue->qdisc); 3150 __netif_schedule(q); 3151 rcu_read_unlock(); 3152 } 3153 } 3154 EXPORT_SYMBOL(netif_tx_wake_queue); 3155 3156 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3157 { 3158 unsigned long flags; 3159 3160 if (unlikely(!skb)) 3161 return; 3162 3163 if (likely(refcount_read(&skb->users) == 1)) { 3164 smp_rmb(); 3165 refcount_set(&skb->users, 0); 3166 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3167 return; 3168 } 3169 get_kfree_skb_cb(skb)->reason = reason; 3170 local_irq_save(flags); 3171 skb->next = __this_cpu_read(softnet_data.completion_queue); 3172 __this_cpu_write(softnet_data.completion_queue, skb); 3173 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3174 local_irq_restore(flags); 3175 } 3176 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3177 3178 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3179 { 3180 if (in_hardirq() || irqs_disabled()) 3181 dev_kfree_skb_irq_reason(skb, reason); 3182 else 3183 kfree_skb_reason(skb, reason); 3184 } 3185 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3186 3187 3188 /** 3189 * netif_device_detach - mark device as removed 3190 * @dev: network device 3191 * 3192 * Mark device as removed from system and therefore no longer available. 3193 */ 3194 void netif_device_detach(struct net_device *dev) 3195 { 3196 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3197 netif_running(dev)) { 3198 netif_tx_stop_all_queues(dev); 3199 } 3200 } 3201 EXPORT_SYMBOL(netif_device_detach); 3202 3203 /** 3204 * netif_device_attach - mark device as attached 3205 * @dev: network device 3206 * 3207 * Mark device as attached from system and restart if needed. 3208 */ 3209 void netif_device_attach(struct net_device *dev) 3210 { 3211 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3212 netif_running(dev)) { 3213 netif_tx_wake_all_queues(dev); 3214 __netdev_watchdog_up(dev); 3215 } 3216 } 3217 EXPORT_SYMBOL(netif_device_attach); 3218 3219 /* 3220 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3221 * to be used as a distribution range. 3222 */ 3223 static u16 skb_tx_hash(const struct net_device *dev, 3224 const struct net_device *sb_dev, 3225 struct sk_buff *skb) 3226 { 3227 u32 hash; 3228 u16 qoffset = 0; 3229 u16 qcount = dev->real_num_tx_queues; 3230 3231 if (dev->num_tc) { 3232 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3233 3234 qoffset = sb_dev->tc_to_txq[tc].offset; 3235 qcount = sb_dev->tc_to_txq[tc].count; 3236 if (unlikely(!qcount)) { 3237 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3238 sb_dev->name, qoffset, tc); 3239 qoffset = 0; 3240 qcount = dev->real_num_tx_queues; 3241 } 3242 } 3243 3244 if (skb_rx_queue_recorded(skb)) { 3245 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3246 hash = skb_get_rx_queue(skb); 3247 if (hash >= qoffset) 3248 hash -= qoffset; 3249 while (unlikely(hash >= qcount)) 3250 hash -= qcount; 3251 return hash + qoffset; 3252 } 3253 3254 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3255 } 3256 3257 void skb_warn_bad_offload(const struct sk_buff *skb) 3258 { 3259 static const netdev_features_t null_features; 3260 struct net_device *dev = skb->dev; 3261 const char *name = ""; 3262 3263 if (!net_ratelimit()) 3264 return; 3265 3266 if (dev) { 3267 if (dev->dev.parent) 3268 name = dev_driver_string(dev->dev.parent); 3269 else 3270 name = netdev_name(dev); 3271 } 3272 skb_dump(KERN_WARNING, skb, false); 3273 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3274 name, dev ? &dev->features : &null_features, 3275 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3276 } 3277 3278 /* 3279 * Invalidate hardware checksum when packet is to be mangled, and 3280 * complete checksum manually on outgoing path. 3281 */ 3282 int skb_checksum_help(struct sk_buff *skb) 3283 { 3284 __wsum csum; 3285 int ret = 0, offset; 3286 3287 if (skb->ip_summed == CHECKSUM_COMPLETE) 3288 goto out_set_summed; 3289 3290 if (unlikely(skb_is_gso(skb))) { 3291 skb_warn_bad_offload(skb); 3292 return -EINVAL; 3293 } 3294 3295 /* Before computing a checksum, we should make sure no frag could 3296 * be modified by an external entity : checksum could be wrong. 3297 */ 3298 if (skb_has_shared_frag(skb)) { 3299 ret = __skb_linearize(skb); 3300 if (ret) 3301 goto out; 3302 } 3303 3304 offset = skb_checksum_start_offset(skb); 3305 ret = -EINVAL; 3306 if (unlikely(offset >= skb_headlen(skb))) { 3307 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3308 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3309 offset, skb_headlen(skb)); 3310 goto out; 3311 } 3312 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3313 3314 offset += skb->csum_offset; 3315 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3316 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3317 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3318 offset + sizeof(__sum16), skb_headlen(skb)); 3319 goto out; 3320 } 3321 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3322 if (ret) 3323 goto out; 3324 3325 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3326 out_set_summed: 3327 skb->ip_summed = CHECKSUM_NONE; 3328 out: 3329 return ret; 3330 } 3331 EXPORT_SYMBOL(skb_checksum_help); 3332 3333 int skb_crc32c_csum_help(struct sk_buff *skb) 3334 { 3335 __le32 crc32c_csum; 3336 int ret = 0, offset, start; 3337 3338 if (skb->ip_summed != CHECKSUM_PARTIAL) 3339 goto out; 3340 3341 if (unlikely(skb_is_gso(skb))) 3342 goto out; 3343 3344 /* Before computing a checksum, we should make sure no frag could 3345 * be modified by an external entity : checksum could be wrong. 3346 */ 3347 if (unlikely(skb_has_shared_frag(skb))) { 3348 ret = __skb_linearize(skb); 3349 if (ret) 3350 goto out; 3351 } 3352 start = skb_checksum_start_offset(skb); 3353 offset = start + offsetof(struct sctphdr, checksum); 3354 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3355 ret = -EINVAL; 3356 goto out; 3357 } 3358 3359 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3360 if (ret) 3361 goto out; 3362 3363 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3364 skb->len - start, ~(__u32)0, 3365 crc32c_csum_stub)); 3366 *(__le32 *)(skb->data + offset) = crc32c_csum; 3367 skb_reset_csum_not_inet(skb); 3368 out: 3369 return ret; 3370 } 3371 3372 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3373 { 3374 __be16 type = skb->protocol; 3375 3376 /* Tunnel gso handlers can set protocol to ethernet. */ 3377 if (type == htons(ETH_P_TEB)) { 3378 struct ethhdr *eth; 3379 3380 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3381 return 0; 3382 3383 eth = (struct ethhdr *)skb->data; 3384 type = eth->h_proto; 3385 } 3386 3387 return vlan_get_protocol_and_depth(skb, type, depth); 3388 } 3389 3390 3391 /* Take action when hardware reception checksum errors are detected. */ 3392 #ifdef CONFIG_BUG 3393 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3394 { 3395 netdev_err(dev, "hw csum failure\n"); 3396 skb_dump(KERN_ERR, skb, true); 3397 dump_stack(); 3398 } 3399 3400 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3401 { 3402 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3403 } 3404 EXPORT_SYMBOL(netdev_rx_csum_fault); 3405 #endif 3406 3407 /* XXX: check that highmem exists at all on the given machine. */ 3408 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3409 { 3410 #ifdef CONFIG_HIGHMEM 3411 int i; 3412 3413 if (!(dev->features & NETIF_F_HIGHDMA)) { 3414 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3415 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3416 3417 if (PageHighMem(skb_frag_page(frag))) 3418 return 1; 3419 } 3420 } 3421 #endif 3422 return 0; 3423 } 3424 3425 /* If MPLS offload request, verify we are testing hardware MPLS features 3426 * instead of standard features for the netdev. 3427 */ 3428 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3429 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3430 netdev_features_t features, 3431 __be16 type) 3432 { 3433 if (eth_p_mpls(type)) 3434 features &= skb->dev->mpls_features; 3435 3436 return features; 3437 } 3438 #else 3439 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3440 netdev_features_t features, 3441 __be16 type) 3442 { 3443 return features; 3444 } 3445 #endif 3446 3447 static netdev_features_t harmonize_features(struct sk_buff *skb, 3448 netdev_features_t features) 3449 { 3450 __be16 type; 3451 3452 type = skb_network_protocol(skb, NULL); 3453 features = net_mpls_features(skb, features, type); 3454 3455 if (skb->ip_summed != CHECKSUM_NONE && 3456 !can_checksum_protocol(features, type)) { 3457 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3458 } 3459 if (illegal_highdma(skb->dev, skb)) 3460 features &= ~NETIF_F_SG; 3461 3462 return features; 3463 } 3464 3465 netdev_features_t passthru_features_check(struct sk_buff *skb, 3466 struct net_device *dev, 3467 netdev_features_t features) 3468 { 3469 return features; 3470 } 3471 EXPORT_SYMBOL(passthru_features_check); 3472 3473 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3474 struct net_device *dev, 3475 netdev_features_t features) 3476 { 3477 return vlan_features_check(skb, features); 3478 } 3479 3480 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3481 struct net_device *dev, 3482 netdev_features_t features) 3483 { 3484 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3485 3486 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3487 return features & ~NETIF_F_GSO_MASK; 3488 3489 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size))) 3490 return features & ~NETIF_F_GSO_MASK; 3491 3492 if (!skb_shinfo(skb)->gso_type) { 3493 skb_warn_bad_offload(skb); 3494 return features & ~NETIF_F_GSO_MASK; 3495 } 3496 3497 /* Support for GSO partial features requires software 3498 * intervention before we can actually process the packets 3499 * so we need to strip support for any partial features now 3500 * and we can pull them back in after we have partially 3501 * segmented the frame. 3502 */ 3503 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3504 features &= ~dev->gso_partial_features; 3505 3506 /* Make sure to clear the IPv4 ID mangling feature if the 3507 * IPv4 header has the potential to be fragmented. 3508 */ 3509 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3510 struct iphdr *iph = skb->encapsulation ? 3511 inner_ip_hdr(skb) : ip_hdr(skb); 3512 3513 if (!(iph->frag_off & htons(IP_DF))) 3514 features &= ~NETIF_F_TSO_MANGLEID; 3515 } 3516 3517 return features; 3518 } 3519 3520 netdev_features_t netif_skb_features(struct sk_buff *skb) 3521 { 3522 struct net_device *dev = skb->dev; 3523 netdev_features_t features = dev->features; 3524 3525 if (skb_is_gso(skb)) 3526 features = gso_features_check(skb, dev, features); 3527 3528 /* If encapsulation offload request, verify we are testing 3529 * hardware encapsulation features instead of standard 3530 * features for the netdev 3531 */ 3532 if (skb->encapsulation) 3533 features &= dev->hw_enc_features; 3534 3535 if (skb_vlan_tagged(skb)) 3536 features = netdev_intersect_features(features, 3537 dev->vlan_features | 3538 NETIF_F_HW_VLAN_CTAG_TX | 3539 NETIF_F_HW_VLAN_STAG_TX); 3540 3541 if (dev->netdev_ops->ndo_features_check) 3542 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3543 features); 3544 else 3545 features &= dflt_features_check(skb, dev, features); 3546 3547 return harmonize_features(skb, features); 3548 } 3549 EXPORT_SYMBOL(netif_skb_features); 3550 3551 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3552 struct netdev_queue *txq, bool more) 3553 { 3554 unsigned int len; 3555 int rc; 3556 3557 if (dev_nit_active(dev)) 3558 dev_queue_xmit_nit(skb, dev); 3559 3560 len = skb->len; 3561 trace_net_dev_start_xmit(skb, dev); 3562 rc = netdev_start_xmit(skb, dev, txq, more); 3563 trace_net_dev_xmit(skb, rc, dev, len); 3564 3565 return rc; 3566 } 3567 3568 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3569 struct netdev_queue *txq, int *ret) 3570 { 3571 struct sk_buff *skb = first; 3572 int rc = NETDEV_TX_OK; 3573 3574 while (skb) { 3575 struct sk_buff *next = skb->next; 3576 3577 skb_mark_not_on_list(skb); 3578 rc = xmit_one(skb, dev, txq, next != NULL); 3579 if (unlikely(!dev_xmit_complete(rc))) { 3580 skb->next = next; 3581 goto out; 3582 } 3583 3584 skb = next; 3585 if (netif_tx_queue_stopped(txq) && skb) { 3586 rc = NETDEV_TX_BUSY; 3587 break; 3588 } 3589 } 3590 3591 out: 3592 *ret = rc; 3593 return skb; 3594 } 3595 3596 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3597 netdev_features_t features) 3598 { 3599 if (skb_vlan_tag_present(skb) && 3600 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3601 skb = __vlan_hwaccel_push_inside(skb); 3602 return skb; 3603 } 3604 3605 int skb_csum_hwoffload_help(struct sk_buff *skb, 3606 const netdev_features_t features) 3607 { 3608 if (unlikely(skb_csum_is_sctp(skb))) 3609 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3610 skb_crc32c_csum_help(skb); 3611 3612 if (features & NETIF_F_HW_CSUM) 3613 return 0; 3614 3615 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3616 switch (skb->csum_offset) { 3617 case offsetof(struct tcphdr, check): 3618 case offsetof(struct udphdr, check): 3619 return 0; 3620 } 3621 } 3622 3623 return skb_checksum_help(skb); 3624 } 3625 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3626 3627 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3628 { 3629 netdev_features_t features; 3630 3631 features = netif_skb_features(skb); 3632 skb = validate_xmit_vlan(skb, features); 3633 if (unlikely(!skb)) 3634 goto out_null; 3635 3636 skb = sk_validate_xmit_skb(skb, dev); 3637 if (unlikely(!skb)) 3638 goto out_null; 3639 3640 if (netif_needs_gso(skb, features)) { 3641 struct sk_buff *segs; 3642 3643 segs = skb_gso_segment(skb, features); 3644 if (IS_ERR(segs)) { 3645 goto out_kfree_skb; 3646 } else if (segs) { 3647 consume_skb(skb); 3648 skb = segs; 3649 } 3650 } else { 3651 if (skb_needs_linearize(skb, features) && 3652 __skb_linearize(skb)) 3653 goto out_kfree_skb; 3654 3655 /* If packet is not checksummed and device does not 3656 * support checksumming for this protocol, complete 3657 * checksumming here. 3658 */ 3659 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3660 if (skb->encapsulation) 3661 skb_set_inner_transport_header(skb, 3662 skb_checksum_start_offset(skb)); 3663 else 3664 skb_set_transport_header(skb, 3665 skb_checksum_start_offset(skb)); 3666 if (skb_csum_hwoffload_help(skb, features)) 3667 goto out_kfree_skb; 3668 } 3669 } 3670 3671 skb = validate_xmit_xfrm(skb, features, again); 3672 3673 return skb; 3674 3675 out_kfree_skb: 3676 kfree_skb(skb); 3677 out_null: 3678 dev_core_stats_tx_dropped_inc(dev); 3679 return NULL; 3680 } 3681 3682 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3683 { 3684 struct sk_buff *next, *head = NULL, *tail; 3685 3686 for (; skb != NULL; skb = next) { 3687 next = skb->next; 3688 skb_mark_not_on_list(skb); 3689 3690 /* in case skb wont be segmented, point to itself */ 3691 skb->prev = skb; 3692 3693 skb = validate_xmit_skb(skb, dev, again); 3694 if (!skb) 3695 continue; 3696 3697 if (!head) 3698 head = skb; 3699 else 3700 tail->next = skb; 3701 /* If skb was segmented, skb->prev points to 3702 * the last segment. If not, it still contains skb. 3703 */ 3704 tail = skb->prev; 3705 } 3706 return head; 3707 } 3708 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3709 3710 static void qdisc_pkt_len_init(struct sk_buff *skb) 3711 { 3712 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3713 3714 qdisc_skb_cb(skb)->pkt_len = skb->len; 3715 3716 /* To get more precise estimation of bytes sent on wire, 3717 * we add to pkt_len the headers size of all segments 3718 */ 3719 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3720 u16 gso_segs = shinfo->gso_segs; 3721 unsigned int hdr_len; 3722 3723 /* mac layer + network layer */ 3724 hdr_len = skb_transport_offset(skb); 3725 3726 /* + transport layer */ 3727 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3728 const struct tcphdr *th; 3729 struct tcphdr _tcphdr; 3730 3731 th = skb_header_pointer(skb, hdr_len, 3732 sizeof(_tcphdr), &_tcphdr); 3733 if (likely(th)) 3734 hdr_len += __tcp_hdrlen(th); 3735 } else { 3736 struct udphdr _udphdr; 3737 3738 if (skb_header_pointer(skb, hdr_len, 3739 sizeof(_udphdr), &_udphdr)) 3740 hdr_len += sizeof(struct udphdr); 3741 } 3742 3743 if (shinfo->gso_type & SKB_GSO_DODGY) 3744 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3745 shinfo->gso_size); 3746 3747 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3748 } 3749 } 3750 3751 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3752 struct sk_buff **to_free, 3753 struct netdev_queue *txq) 3754 { 3755 int rc; 3756 3757 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3758 if (rc == NET_XMIT_SUCCESS) 3759 trace_qdisc_enqueue(q, txq, skb); 3760 return rc; 3761 } 3762 3763 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3764 struct net_device *dev, 3765 struct netdev_queue *txq) 3766 { 3767 spinlock_t *root_lock = qdisc_lock(q); 3768 struct sk_buff *to_free = NULL; 3769 bool contended; 3770 int rc; 3771 3772 qdisc_calculate_pkt_len(skb, q); 3773 3774 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3775 3776 if (q->flags & TCQ_F_NOLOCK) { 3777 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3778 qdisc_run_begin(q)) { 3779 /* Retest nolock_qdisc_is_empty() within the protection 3780 * of q->seqlock to protect from racing with requeuing. 3781 */ 3782 if (unlikely(!nolock_qdisc_is_empty(q))) { 3783 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3784 __qdisc_run(q); 3785 qdisc_run_end(q); 3786 3787 goto no_lock_out; 3788 } 3789 3790 qdisc_bstats_cpu_update(q, skb); 3791 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3792 !nolock_qdisc_is_empty(q)) 3793 __qdisc_run(q); 3794 3795 qdisc_run_end(q); 3796 return NET_XMIT_SUCCESS; 3797 } 3798 3799 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3800 qdisc_run(q); 3801 3802 no_lock_out: 3803 if (unlikely(to_free)) 3804 kfree_skb_list_reason(to_free, 3805 tcf_get_drop_reason(to_free)); 3806 return rc; 3807 } 3808 3809 /* 3810 * Heuristic to force contended enqueues to serialize on a 3811 * separate lock before trying to get qdisc main lock. 3812 * This permits qdisc->running owner to get the lock more 3813 * often and dequeue packets faster. 3814 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3815 * and then other tasks will only enqueue packets. The packets will be 3816 * sent after the qdisc owner is scheduled again. To prevent this 3817 * scenario the task always serialize on the lock. 3818 */ 3819 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3820 if (unlikely(contended)) 3821 spin_lock(&q->busylock); 3822 3823 spin_lock(root_lock); 3824 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3825 __qdisc_drop(skb, &to_free); 3826 rc = NET_XMIT_DROP; 3827 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3828 qdisc_run_begin(q)) { 3829 /* 3830 * This is a work-conserving queue; there are no old skbs 3831 * waiting to be sent out; and the qdisc is not running - 3832 * xmit the skb directly. 3833 */ 3834 3835 qdisc_bstats_update(q, skb); 3836 3837 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3838 if (unlikely(contended)) { 3839 spin_unlock(&q->busylock); 3840 contended = false; 3841 } 3842 __qdisc_run(q); 3843 } 3844 3845 qdisc_run_end(q); 3846 rc = NET_XMIT_SUCCESS; 3847 } else { 3848 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3849 if (qdisc_run_begin(q)) { 3850 if (unlikely(contended)) { 3851 spin_unlock(&q->busylock); 3852 contended = false; 3853 } 3854 __qdisc_run(q); 3855 qdisc_run_end(q); 3856 } 3857 } 3858 spin_unlock(root_lock); 3859 if (unlikely(to_free)) 3860 kfree_skb_list_reason(to_free, 3861 tcf_get_drop_reason(to_free)); 3862 if (unlikely(contended)) 3863 spin_unlock(&q->busylock); 3864 return rc; 3865 } 3866 3867 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3868 static void skb_update_prio(struct sk_buff *skb) 3869 { 3870 const struct netprio_map *map; 3871 const struct sock *sk; 3872 unsigned int prioidx; 3873 3874 if (skb->priority) 3875 return; 3876 map = rcu_dereference_bh(skb->dev->priomap); 3877 if (!map) 3878 return; 3879 sk = skb_to_full_sk(skb); 3880 if (!sk) 3881 return; 3882 3883 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3884 3885 if (prioidx < map->priomap_len) 3886 skb->priority = map->priomap[prioidx]; 3887 } 3888 #else 3889 #define skb_update_prio(skb) 3890 #endif 3891 3892 /** 3893 * dev_loopback_xmit - loop back @skb 3894 * @net: network namespace this loopback is happening in 3895 * @sk: sk needed to be a netfilter okfn 3896 * @skb: buffer to transmit 3897 */ 3898 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3899 { 3900 skb_reset_mac_header(skb); 3901 __skb_pull(skb, skb_network_offset(skb)); 3902 skb->pkt_type = PACKET_LOOPBACK; 3903 if (skb->ip_summed == CHECKSUM_NONE) 3904 skb->ip_summed = CHECKSUM_UNNECESSARY; 3905 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3906 skb_dst_force(skb); 3907 netif_rx(skb); 3908 return 0; 3909 } 3910 EXPORT_SYMBOL(dev_loopback_xmit); 3911 3912 #ifdef CONFIG_NET_EGRESS 3913 static struct netdev_queue * 3914 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3915 { 3916 int qm = skb_get_queue_mapping(skb); 3917 3918 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3919 } 3920 3921 static bool netdev_xmit_txqueue_skipped(void) 3922 { 3923 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3924 } 3925 3926 void netdev_xmit_skip_txqueue(bool skip) 3927 { 3928 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3929 } 3930 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3931 #endif /* CONFIG_NET_EGRESS */ 3932 3933 #ifdef CONFIG_NET_XGRESS 3934 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 3935 enum skb_drop_reason *drop_reason) 3936 { 3937 int ret = TC_ACT_UNSPEC; 3938 #ifdef CONFIG_NET_CLS_ACT 3939 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3940 struct tcf_result res; 3941 3942 if (!miniq) 3943 return ret; 3944 3945 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) { 3946 if (tcf_block_bypass_sw(miniq->block)) 3947 return ret; 3948 } 3949 3950 tc_skb_cb(skb)->mru = 0; 3951 tc_skb_cb(skb)->post_ct = false; 3952 tcf_set_drop_reason(skb, *drop_reason); 3953 3954 mini_qdisc_bstats_cpu_update(miniq, skb); 3955 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3956 /* Only tcf related quirks below. */ 3957 switch (ret) { 3958 case TC_ACT_SHOT: 3959 *drop_reason = tcf_get_drop_reason(skb); 3960 mini_qdisc_qstats_cpu_drop(miniq); 3961 break; 3962 case TC_ACT_OK: 3963 case TC_ACT_RECLASSIFY: 3964 skb->tc_index = TC_H_MIN(res.classid); 3965 break; 3966 } 3967 #endif /* CONFIG_NET_CLS_ACT */ 3968 return ret; 3969 } 3970 3971 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3972 3973 void tcx_inc(void) 3974 { 3975 static_branch_inc(&tcx_needed_key); 3976 } 3977 3978 void tcx_dec(void) 3979 { 3980 static_branch_dec(&tcx_needed_key); 3981 } 3982 3983 static __always_inline enum tcx_action_base 3984 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3985 const bool needs_mac) 3986 { 3987 const struct bpf_mprog_fp *fp; 3988 const struct bpf_prog *prog; 3989 int ret = TCX_NEXT; 3990 3991 if (needs_mac) 3992 __skb_push(skb, skb->mac_len); 3993 bpf_mprog_foreach_prog(entry, fp, prog) { 3994 bpf_compute_data_pointers(skb); 3995 ret = bpf_prog_run(prog, skb); 3996 if (ret != TCX_NEXT) 3997 break; 3998 } 3999 if (needs_mac) 4000 __skb_pull(skb, skb->mac_len); 4001 return tcx_action_code(skb, ret); 4002 } 4003 4004 static __always_inline struct sk_buff * 4005 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4006 struct net_device *orig_dev, bool *another) 4007 { 4008 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4009 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 4010 int sch_ret; 4011 4012 if (!entry) 4013 return skb; 4014 if (*pt_prev) { 4015 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4016 *pt_prev = NULL; 4017 } 4018 4019 qdisc_skb_cb(skb)->pkt_len = skb->len; 4020 tcx_set_ingress(skb, true); 4021 4022 if (static_branch_unlikely(&tcx_needed_key)) { 4023 sch_ret = tcx_run(entry, skb, true); 4024 if (sch_ret != TC_ACT_UNSPEC) 4025 goto ingress_verdict; 4026 } 4027 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4028 ingress_verdict: 4029 switch (sch_ret) { 4030 case TC_ACT_REDIRECT: 4031 /* skb_mac_header check was done by BPF, so we can safely 4032 * push the L2 header back before redirecting to another 4033 * netdev. 4034 */ 4035 __skb_push(skb, skb->mac_len); 4036 if (skb_do_redirect(skb) == -EAGAIN) { 4037 __skb_pull(skb, skb->mac_len); 4038 *another = true; 4039 break; 4040 } 4041 *ret = NET_RX_SUCCESS; 4042 return NULL; 4043 case TC_ACT_SHOT: 4044 kfree_skb_reason(skb, drop_reason); 4045 *ret = NET_RX_DROP; 4046 return NULL; 4047 /* used by tc_run */ 4048 case TC_ACT_STOLEN: 4049 case TC_ACT_QUEUED: 4050 case TC_ACT_TRAP: 4051 consume_skb(skb); 4052 fallthrough; 4053 case TC_ACT_CONSUMED: 4054 *ret = NET_RX_SUCCESS; 4055 return NULL; 4056 } 4057 4058 return skb; 4059 } 4060 4061 static __always_inline struct sk_buff * 4062 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4063 { 4064 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4065 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4066 int sch_ret; 4067 4068 if (!entry) 4069 return skb; 4070 4071 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4072 * already set by the caller. 4073 */ 4074 if (static_branch_unlikely(&tcx_needed_key)) { 4075 sch_ret = tcx_run(entry, skb, false); 4076 if (sch_ret != TC_ACT_UNSPEC) 4077 goto egress_verdict; 4078 } 4079 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4080 egress_verdict: 4081 switch (sch_ret) { 4082 case TC_ACT_REDIRECT: 4083 /* No need to push/pop skb's mac_header here on egress! */ 4084 skb_do_redirect(skb); 4085 *ret = NET_XMIT_SUCCESS; 4086 return NULL; 4087 case TC_ACT_SHOT: 4088 kfree_skb_reason(skb, drop_reason); 4089 *ret = NET_XMIT_DROP; 4090 return NULL; 4091 /* used by tc_run */ 4092 case TC_ACT_STOLEN: 4093 case TC_ACT_QUEUED: 4094 case TC_ACT_TRAP: 4095 consume_skb(skb); 4096 fallthrough; 4097 case TC_ACT_CONSUMED: 4098 *ret = NET_XMIT_SUCCESS; 4099 return NULL; 4100 } 4101 4102 return skb; 4103 } 4104 #else 4105 static __always_inline struct sk_buff * 4106 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4107 struct net_device *orig_dev, bool *another) 4108 { 4109 return skb; 4110 } 4111 4112 static __always_inline struct sk_buff * 4113 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4114 { 4115 return skb; 4116 } 4117 #endif /* CONFIG_NET_XGRESS */ 4118 4119 #ifdef CONFIG_XPS 4120 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4121 struct xps_dev_maps *dev_maps, unsigned int tci) 4122 { 4123 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4124 struct xps_map *map; 4125 int queue_index = -1; 4126 4127 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4128 return queue_index; 4129 4130 tci *= dev_maps->num_tc; 4131 tci += tc; 4132 4133 map = rcu_dereference(dev_maps->attr_map[tci]); 4134 if (map) { 4135 if (map->len == 1) 4136 queue_index = map->queues[0]; 4137 else 4138 queue_index = map->queues[reciprocal_scale( 4139 skb_get_hash(skb), map->len)]; 4140 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4141 queue_index = -1; 4142 } 4143 return queue_index; 4144 } 4145 #endif 4146 4147 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4148 struct sk_buff *skb) 4149 { 4150 #ifdef CONFIG_XPS 4151 struct xps_dev_maps *dev_maps; 4152 struct sock *sk = skb->sk; 4153 int queue_index = -1; 4154 4155 if (!static_key_false(&xps_needed)) 4156 return -1; 4157 4158 rcu_read_lock(); 4159 if (!static_key_false(&xps_rxqs_needed)) 4160 goto get_cpus_map; 4161 4162 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4163 if (dev_maps) { 4164 int tci = sk_rx_queue_get(sk); 4165 4166 if (tci >= 0) 4167 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4168 tci); 4169 } 4170 4171 get_cpus_map: 4172 if (queue_index < 0) { 4173 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4174 if (dev_maps) { 4175 unsigned int tci = skb->sender_cpu - 1; 4176 4177 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4178 tci); 4179 } 4180 } 4181 rcu_read_unlock(); 4182 4183 return queue_index; 4184 #else 4185 return -1; 4186 #endif 4187 } 4188 4189 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4190 struct net_device *sb_dev) 4191 { 4192 return 0; 4193 } 4194 EXPORT_SYMBOL(dev_pick_tx_zero); 4195 4196 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4197 struct net_device *sb_dev) 4198 { 4199 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4200 } 4201 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4202 4203 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4204 struct net_device *sb_dev) 4205 { 4206 struct sock *sk = skb->sk; 4207 int queue_index = sk_tx_queue_get(sk); 4208 4209 sb_dev = sb_dev ? : dev; 4210 4211 if (queue_index < 0 || skb->ooo_okay || 4212 queue_index >= dev->real_num_tx_queues) { 4213 int new_index = get_xps_queue(dev, sb_dev, skb); 4214 4215 if (new_index < 0) 4216 new_index = skb_tx_hash(dev, sb_dev, skb); 4217 4218 if (queue_index != new_index && sk && 4219 sk_fullsock(sk) && 4220 rcu_access_pointer(sk->sk_dst_cache)) 4221 sk_tx_queue_set(sk, new_index); 4222 4223 queue_index = new_index; 4224 } 4225 4226 return queue_index; 4227 } 4228 EXPORT_SYMBOL(netdev_pick_tx); 4229 4230 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4231 struct sk_buff *skb, 4232 struct net_device *sb_dev) 4233 { 4234 int queue_index = 0; 4235 4236 #ifdef CONFIG_XPS 4237 u32 sender_cpu = skb->sender_cpu - 1; 4238 4239 if (sender_cpu >= (u32)NR_CPUS) 4240 skb->sender_cpu = raw_smp_processor_id() + 1; 4241 #endif 4242 4243 if (dev->real_num_tx_queues != 1) { 4244 const struct net_device_ops *ops = dev->netdev_ops; 4245 4246 if (ops->ndo_select_queue) 4247 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4248 else 4249 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4250 4251 queue_index = netdev_cap_txqueue(dev, queue_index); 4252 } 4253 4254 skb_set_queue_mapping(skb, queue_index); 4255 return netdev_get_tx_queue(dev, queue_index); 4256 } 4257 4258 /** 4259 * __dev_queue_xmit() - transmit a buffer 4260 * @skb: buffer to transmit 4261 * @sb_dev: suboordinate device used for L2 forwarding offload 4262 * 4263 * Queue a buffer for transmission to a network device. The caller must 4264 * have set the device and priority and built the buffer before calling 4265 * this function. The function can be called from an interrupt. 4266 * 4267 * When calling this method, interrupts MUST be enabled. This is because 4268 * the BH enable code must have IRQs enabled so that it will not deadlock. 4269 * 4270 * Regardless of the return value, the skb is consumed, so it is currently 4271 * difficult to retry a send to this method. (You can bump the ref count 4272 * before sending to hold a reference for retry if you are careful.) 4273 * 4274 * Return: 4275 * * 0 - buffer successfully transmitted 4276 * * positive qdisc return code - NET_XMIT_DROP etc. 4277 * * negative errno - other errors 4278 */ 4279 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4280 { 4281 struct net_device *dev = skb->dev; 4282 struct netdev_queue *txq = NULL; 4283 struct Qdisc *q; 4284 int rc = -ENOMEM; 4285 bool again = false; 4286 4287 skb_reset_mac_header(skb); 4288 skb_assert_len(skb); 4289 4290 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4291 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4292 4293 /* Disable soft irqs for various locks below. Also 4294 * stops preemption for RCU. 4295 */ 4296 rcu_read_lock_bh(); 4297 4298 skb_update_prio(skb); 4299 4300 qdisc_pkt_len_init(skb); 4301 tcx_set_ingress(skb, false); 4302 #ifdef CONFIG_NET_EGRESS 4303 if (static_branch_unlikely(&egress_needed_key)) { 4304 if (nf_hook_egress_active()) { 4305 skb = nf_hook_egress(skb, &rc, dev); 4306 if (!skb) 4307 goto out; 4308 } 4309 4310 netdev_xmit_skip_txqueue(false); 4311 4312 nf_skip_egress(skb, true); 4313 skb = sch_handle_egress(skb, &rc, dev); 4314 if (!skb) 4315 goto out; 4316 nf_skip_egress(skb, false); 4317 4318 if (netdev_xmit_txqueue_skipped()) 4319 txq = netdev_tx_queue_mapping(dev, skb); 4320 } 4321 #endif 4322 /* If device/qdisc don't need skb->dst, release it right now while 4323 * its hot in this cpu cache. 4324 */ 4325 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4326 skb_dst_drop(skb); 4327 else 4328 skb_dst_force(skb); 4329 4330 if (!txq) 4331 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4332 4333 q = rcu_dereference_bh(txq->qdisc); 4334 4335 trace_net_dev_queue(skb); 4336 if (q->enqueue) { 4337 rc = __dev_xmit_skb(skb, q, dev, txq); 4338 goto out; 4339 } 4340 4341 /* The device has no queue. Common case for software devices: 4342 * loopback, all the sorts of tunnels... 4343 4344 * Really, it is unlikely that netif_tx_lock protection is necessary 4345 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4346 * counters.) 4347 * However, it is possible, that they rely on protection 4348 * made by us here. 4349 4350 * Check this and shot the lock. It is not prone from deadlocks. 4351 *Either shot noqueue qdisc, it is even simpler 8) 4352 */ 4353 if (dev->flags & IFF_UP) { 4354 int cpu = smp_processor_id(); /* ok because BHs are off */ 4355 4356 /* Other cpus might concurrently change txq->xmit_lock_owner 4357 * to -1 or to their cpu id, but not to our id. 4358 */ 4359 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4360 if (dev_xmit_recursion()) 4361 goto recursion_alert; 4362 4363 skb = validate_xmit_skb(skb, dev, &again); 4364 if (!skb) 4365 goto out; 4366 4367 HARD_TX_LOCK(dev, txq, cpu); 4368 4369 if (!netif_xmit_stopped(txq)) { 4370 dev_xmit_recursion_inc(); 4371 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4372 dev_xmit_recursion_dec(); 4373 if (dev_xmit_complete(rc)) { 4374 HARD_TX_UNLOCK(dev, txq); 4375 goto out; 4376 } 4377 } 4378 HARD_TX_UNLOCK(dev, txq); 4379 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4380 dev->name); 4381 } else { 4382 /* Recursion is detected! It is possible, 4383 * unfortunately 4384 */ 4385 recursion_alert: 4386 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4387 dev->name); 4388 } 4389 } 4390 4391 rc = -ENETDOWN; 4392 rcu_read_unlock_bh(); 4393 4394 dev_core_stats_tx_dropped_inc(dev); 4395 kfree_skb_list(skb); 4396 return rc; 4397 out: 4398 rcu_read_unlock_bh(); 4399 return rc; 4400 } 4401 EXPORT_SYMBOL(__dev_queue_xmit); 4402 4403 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4404 { 4405 struct net_device *dev = skb->dev; 4406 struct sk_buff *orig_skb = skb; 4407 struct netdev_queue *txq; 4408 int ret = NETDEV_TX_BUSY; 4409 bool again = false; 4410 4411 if (unlikely(!netif_running(dev) || 4412 !netif_carrier_ok(dev))) 4413 goto drop; 4414 4415 skb = validate_xmit_skb_list(skb, dev, &again); 4416 if (skb != orig_skb) 4417 goto drop; 4418 4419 skb_set_queue_mapping(skb, queue_id); 4420 txq = skb_get_tx_queue(dev, skb); 4421 4422 local_bh_disable(); 4423 4424 dev_xmit_recursion_inc(); 4425 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4426 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4427 ret = netdev_start_xmit(skb, dev, txq, false); 4428 HARD_TX_UNLOCK(dev, txq); 4429 dev_xmit_recursion_dec(); 4430 4431 local_bh_enable(); 4432 return ret; 4433 drop: 4434 dev_core_stats_tx_dropped_inc(dev); 4435 kfree_skb_list(skb); 4436 return NET_XMIT_DROP; 4437 } 4438 EXPORT_SYMBOL(__dev_direct_xmit); 4439 4440 /************************************************************************* 4441 * Receiver routines 4442 *************************************************************************/ 4443 static DEFINE_PER_CPU(struct task_struct *, backlog_napi); 4444 4445 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4446 int weight_p __read_mostly = 64; /* old backlog weight */ 4447 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4448 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4449 4450 /* Called with irq disabled */ 4451 static inline void ____napi_schedule(struct softnet_data *sd, 4452 struct napi_struct *napi) 4453 { 4454 struct task_struct *thread; 4455 4456 lockdep_assert_irqs_disabled(); 4457 4458 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4459 /* Paired with smp_mb__before_atomic() in 4460 * napi_enable()/dev_set_threaded(). 4461 * Use READ_ONCE() to guarantee a complete 4462 * read on napi->thread. Only call 4463 * wake_up_process() when it's not NULL. 4464 */ 4465 thread = READ_ONCE(napi->thread); 4466 if (thread) { 4467 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi)) 4468 goto use_local_napi; 4469 4470 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4471 wake_up_process(thread); 4472 return; 4473 } 4474 } 4475 4476 use_local_napi: 4477 list_add_tail(&napi->poll_list, &sd->poll_list); 4478 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4479 /* If not called from net_rx_action() 4480 * we have to raise NET_RX_SOFTIRQ. 4481 */ 4482 if (!sd->in_net_rx_action) 4483 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4484 } 4485 4486 #ifdef CONFIG_RPS 4487 4488 struct static_key_false rps_needed __read_mostly; 4489 EXPORT_SYMBOL(rps_needed); 4490 struct static_key_false rfs_needed __read_mostly; 4491 EXPORT_SYMBOL(rfs_needed); 4492 4493 static struct rps_dev_flow * 4494 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4495 struct rps_dev_flow *rflow, u16 next_cpu) 4496 { 4497 if (next_cpu < nr_cpu_ids) { 4498 #ifdef CONFIG_RFS_ACCEL 4499 struct netdev_rx_queue *rxqueue; 4500 struct rps_dev_flow_table *flow_table; 4501 struct rps_dev_flow *old_rflow; 4502 u32 flow_id; 4503 u16 rxq_index; 4504 int rc; 4505 4506 /* Should we steer this flow to a different hardware queue? */ 4507 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4508 !(dev->features & NETIF_F_NTUPLE)) 4509 goto out; 4510 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4511 if (rxq_index == skb_get_rx_queue(skb)) 4512 goto out; 4513 4514 rxqueue = dev->_rx + rxq_index; 4515 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4516 if (!flow_table) 4517 goto out; 4518 flow_id = skb_get_hash(skb) & flow_table->mask; 4519 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4520 rxq_index, flow_id); 4521 if (rc < 0) 4522 goto out; 4523 old_rflow = rflow; 4524 rflow = &flow_table->flows[flow_id]; 4525 rflow->filter = rc; 4526 if (old_rflow->filter == rflow->filter) 4527 old_rflow->filter = RPS_NO_FILTER; 4528 out: 4529 #endif 4530 rflow->last_qtail = 4531 READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head); 4532 } 4533 4534 rflow->cpu = next_cpu; 4535 return rflow; 4536 } 4537 4538 /* 4539 * get_rps_cpu is called from netif_receive_skb and returns the target 4540 * CPU from the RPS map of the receiving queue for a given skb. 4541 * rcu_read_lock must be held on entry. 4542 */ 4543 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4544 struct rps_dev_flow **rflowp) 4545 { 4546 const struct rps_sock_flow_table *sock_flow_table; 4547 struct netdev_rx_queue *rxqueue = dev->_rx; 4548 struct rps_dev_flow_table *flow_table; 4549 struct rps_map *map; 4550 int cpu = -1; 4551 u32 tcpu; 4552 u32 hash; 4553 4554 if (skb_rx_queue_recorded(skb)) { 4555 u16 index = skb_get_rx_queue(skb); 4556 4557 if (unlikely(index >= dev->real_num_rx_queues)) { 4558 WARN_ONCE(dev->real_num_rx_queues > 1, 4559 "%s received packet on queue %u, but number " 4560 "of RX queues is %u\n", 4561 dev->name, index, dev->real_num_rx_queues); 4562 goto done; 4563 } 4564 rxqueue += index; 4565 } 4566 4567 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4568 4569 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4570 map = rcu_dereference(rxqueue->rps_map); 4571 if (!flow_table && !map) 4572 goto done; 4573 4574 skb_reset_network_header(skb); 4575 hash = skb_get_hash(skb); 4576 if (!hash) 4577 goto done; 4578 4579 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table); 4580 if (flow_table && sock_flow_table) { 4581 struct rps_dev_flow *rflow; 4582 u32 next_cpu; 4583 u32 ident; 4584 4585 /* First check into global flow table if there is a match. 4586 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4587 */ 4588 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4589 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask) 4590 goto try_rps; 4591 4592 next_cpu = ident & net_hotdata.rps_cpu_mask; 4593 4594 /* OK, now we know there is a match, 4595 * we can look at the local (per receive queue) flow table 4596 */ 4597 rflow = &flow_table->flows[hash & flow_table->mask]; 4598 tcpu = rflow->cpu; 4599 4600 /* 4601 * If the desired CPU (where last recvmsg was done) is 4602 * different from current CPU (one in the rx-queue flow 4603 * table entry), switch if one of the following holds: 4604 * - Current CPU is unset (>= nr_cpu_ids). 4605 * - Current CPU is offline. 4606 * - The current CPU's queue tail has advanced beyond the 4607 * last packet that was enqueued using this table entry. 4608 * This guarantees that all previous packets for the flow 4609 * have been dequeued, thus preserving in order delivery. 4610 */ 4611 if (unlikely(tcpu != next_cpu) && 4612 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4613 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) - 4614 READ_ONCE(rflow->last_qtail))) >= 0)) { 4615 tcpu = next_cpu; 4616 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4617 } 4618 4619 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4620 *rflowp = rflow; 4621 cpu = tcpu; 4622 goto done; 4623 } 4624 } 4625 4626 try_rps: 4627 4628 if (map) { 4629 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4630 if (cpu_online(tcpu)) { 4631 cpu = tcpu; 4632 goto done; 4633 } 4634 } 4635 4636 done: 4637 return cpu; 4638 } 4639 4640 #ifdef CONFIG_RFS_ACCEL 4641 4642 /** 4643 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4644 * @dev: Device on which the filter was set 4645 * @rxq_index: RX queue index 4646 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4647 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4648 * 4649 * Drivers that implement ndo_rx_flow_steer() should periodically call 4650 * this function for each installed filter and remove the filters for 4651 * which it returns %true. 4652 */ 4653 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4654 u32 flow_id, u16 filter_id) 4655 { 4656 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4657 struct rps_dev_flow_table *flow_table; 4658 struct rps_dev_flow *rflow; 4659 bool expire = true; 4660 unsigned int cpu; 4661 4662 rcu_read_lock(); 4663 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4664 if (flow_table && flow_id <= flow_table->mask) { 4665 rflow = &flow_table->flows[flow_id]; 4666 cpu = READ_ONCE(rflow->cpu); 4667 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4668 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) - 4669 READ_ONCE(rflow->last_qtail)) < 4670 (int)(10 * flow_table->mask))) 4671 expire = false; 4672 } 4673 rcu_read_unlock(); 4674 return expire; 4675 } 4676 EXPORT_SYMBOL(rps_may_expire_flow); 4677 4678 #endif /* CONFIG_RFS_ACCEL */ 4679 4680 /* Called from hardirq (IPI) context */ 4681 static void rps_trigger_softirq(void *data) 4682 { 4683 struct softnet_data *sd = data; 4684 4685 ____napi_schedule(sd, &sd->backlog); 4686 sd->received_rps++; 4687 } 4688 4689 #endif /* CONFIG_RPS */ 4690 4691 /* Called from hardirq (IPI) context */ 4692 static void trigger_rx_softirq(void *data) 4693 { 4694 struct softnet_data *sd = data; 4695 4696 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4697 smp_store_release(&sd->defer_ipi_scheduled, 0); 4698 } 4699 4700 /* 4701 * After we queued a packet into sd->input_pkt_queue, 4702 * we need to make sure this queue is serviced soon. 4703 * 4704 * - If this is another cpu queue, link it to our rps_ipi_list, 4705 * and make sure we will process rps_ipi_list from net_rx_action(). 4706 * 4707 * - If this is our own queue, NAPI schedule our backlog. 4708 * Note that this also raises NET_RX_SOFTIRQ. 4709 */ 4710 static void napi_schedule_rps(struct softnet_data *sd) 4711 { 4712 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4713 4714 #ifdef CONFIG_RPS 4715 if (sd != mysd) { 4716 if (use_backlog_threads()) { 4717 __napi_schedule_irqoff(&sd->backlog); 4718 return; 4719 } 4720 4721 sd->rps_ipi_next = mysd->rps_ipi_list; 4722 mysd->rps_ipi_list = sd; 4723 4724 /* If not called from net_rx_action() or napi_threaded_poll() 4725 * we have to raise NET_RX_SOFTIRQ. 4726 */ 4727 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4728 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4729 return; 4730 } 4731 #endif /* CONFIG_RPS */ 4732 __napi_schedule_irqoff(&mysd->backlog); 4733 } 4734 4735 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu) 4736 { 4737 unsigned long flags; 4738 4739 if (use_backlog_threads()) { 4740 backlog_lock_irq_save(sd, &flags); 4741 4742 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4743 __napi_schedule_irqoff(&sd->backlog); 4744 4745 backlog_unlock_irq_restore(sd, &flags); 4746 4747 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) { 4748 smp_call_function_single_async(cpu, &sd->defer_csd); 4749 } 4750 } 4751 4752 #ifdef CONFIG_NET_FLOW_LIMIT 4753 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4754 #endif 4755 4756 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4757 { 4758 #ifdef CONFIG_NET_FLOW_LIMIT 4759 struct sd_flow_limit *fl; 4760 struct softnet_data *sd; 4761 unsigned int old_flow, new_flow; 4762 4763 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1)) 4764 return false; 4765 4766 sd = this_cpu_ptr(&softnet_data); 4767 4768 rcu_read_lock(); 4769 fl = rcu_dereference(sd->flow_limit); 4770 if (fl) { 4771 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4772 old_flow = fl->history[fl->history_head]; 4773 fl->history[fl->history_head] = new_flow; 4774 4775 fl->history_head++; 4776 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4777 4778 if (likely(fl->buckets[old_flow])) 4779 fl->buckets[old_flow]--; 4780 4781 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4782 fl->count++; 4783 rcu_read_unlock(); 4784 return true; 4785 } 4786 } 4787 rcu_read_unlock(); 4788 #endif 4789 return false; 4790 } 4791 4792 /* 4793 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4794 * queue (may be a remote CPU queue). 4795 */ 4796 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4797 unsigned int *qtail) 4798 { 4799 enum skb_drop_reason reason; 4800 struct softnet_data *sd; 4801 unsigned long flags; 4802 unsigned int qlen; 4803 int max_backlog; 4804 u32 tail; 4805 4806 reason = SKB_DROP_REASON_DEV_READY; 4807 if (!netif_running(skb->dev)) 4808 goto bad_dev; 4809 4810 reason = SKB_DROP_REASON_CPU_BACKLOG; 4811 sd = &per_cpu(softnet_data, cpu); 4812 4813 qlen = skb_queue_len_lockless(&sd->input_pkt_queue); 4814 max_backlog = READ_ONCE(net_hotdata.max_backlog); 4815 if (unlikely(qlen > max_backlog)) 4816 goto cpu_backlog_drop; 4817 backlog_lock_irq_save(sd, &flags); 4818 qlen = skb_queue_len(&sd->input_pkt_queue); 4819 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) { 4820 if (!qlen) { 4821 /* Schedule NAPI for backlog device. We can use 4822 * non atomic operation as we own the queue lock. 4823 */ 4824 if (!__test_and_set_bit(NAPI_STATE_SCHED, 4825 &sd->backlog.state)) 4826 napi_schedule_rps(sd); 4827 } 4828 __skb_queue_tail(&sd->input_pkt_queue, skb); 4829 tail = rps_input_queue_tail_incr(sd); 4830 backlog_unlock_irq_restore(sd, &flags); 4831 4832 /* save the tail outside of the critical section */ 4833 rps_input_queue_tail_save(qtail, tail); 4834 return NET_RX_SUCCESS; 4835 } 4836 4837 backlog_unlock_irq_restore(sd, &flags); 4838 4839 cpu_backlog_drop: 4840 atomic_inc(&sd->dropped); 4841 bad_dev: 4842 dev_core_stats_rx_dropped_inc(skb->dev); 4843 kfree_skb_reason(skb, reason); 4844 return NET_RX_DROP; 4845 } 4846 4847 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4848 { 4849 struct net_device *dev = skb->dev; 4850 struct netdev_rx_queue *rxqueue; 4851 4852 rxqueue = dev->_rx; 4853 4854 if (skb_rx_queue_recorded(skb)) { 4855 u16 index = skb_get_rx_queue(skb); 4856 4857 if (unlikely(index >= dev->real_num_rx_queues)) { 4858 WARN_ONCE(dev->real_num_rx_queues > 1, 4859 "%s received packet on queue %u, but number " 4860 "of RX queues is %u\n", 4861 dev->name, index, dev->real_num_rx_queues); 4862 4863 return rxqueue; /* Return first rxqueue */ 4864 } 4865 rxqueue += index; 4866 } 4867 return rxqueue; 4868 } 4869 4870 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4871 struct bpf_prog *xdp_prog) 4872 { 4873 void *orig_data, *orig_data_end, *hard_start; 4874 struct netdev_rx_queue *rxqueue; 4875 bool orig_bcast, orig_host; 4876 u32 mac_len, frame_sz; 4877 __be16 orig_eth_type; 4878 struct ethhdr *eth; 4879 u32 metalen, act; 4880 int off; 4881 4882 /* The XDP program wants to see the packet starting at the MAC 4883 * header. 4884 */ 4885 mac_len = skb->data - skb_mac_header(skb); 4886 hard_start = skb->data - skb_headroom(skb); 4887 4888 /* SKB "head" area always have tailroom for skb_shared_info */ 4889 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4890 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4891 4892 rxqueue = netif_get_rxqueue(skb); 4893 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4894 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4895 skb_headlen(skb) + mac_len, true); 4896 if (skb_is_nonlinear(skb)) { 4897 skb_shinfo(skb)->xdp_frags_size = skb->data_len; 4898 xdp_buff_set_frags_flag(xdp); 4899 } else { 4900 xdp_buff_clear_frags_flag(xdp); 4901 } 4902 4903 orig_data_end = xdp->data_end; 4904 orig_data = xdp->data; 4905 eth = (struct ethhdr *)xdp->data; 4906 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4907 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4908 orig_eth_type = eth->h_proto; 4909 4910 act = bpf_prog_run_xdp(xdp_prog, xdp); 4911 4912 /* check if bpf_xdp_adjust_head was used */ 4913 off = xdp->data - orig_data; 4914 if (off) { 4915 if (off > 0) 4916 __skb_pull(skb, off); 4917 else if (off < 0) 4918 __skb_push(skb, -off); 4919 4920 skb->mac_header += off; 4921 skb_reset_network_header(skb); 4922 } 4923 4924 /* check if bpf_xdp_adjust_tail was used */ 4925 off = xdp->data_end - orig_data_end; 4926 if (off != 0) { 4927 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4928 skb->len += off; /* positive on grow, negative on shrink */ 4929 } 4930 4931 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers 4932 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. 4933 */ 4934 if (xdp_buff_has_frags(xdp)) 4935 skb->data_len = skb_shinfo(skb)->xdp_frags_size; 4936 else 4937 skb->data_len = 0; 4938 4939 /* check if XDP changed eth hdr such SKB needs update */ 4940 eth = (struct ethhdr *)xdp->data; 4941 if ((orig_eth_type != eth->h_proto) || 4942 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4943 skb->dev->dev_addr)) || 4944 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4945 __skb_push(skb, ETH_HLEN); 4946 skb->pkt_type = PACKET_HOST; 4947 skb->protocol = eth_type_trans(skb, skb->dev); 4948 } 4949 4950 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4951 * before calling us again on redirect path. We do not call do_redirect 4952 * as we leave that up to the caller. 4953 * 4954 * Caller is responsible for managing lifetime of skb (i.e. calling 4955 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4956 */ 4957 switch (act) { 4958 case XDP_REDIRECT: 4959 case XDP_TX: 4960 __skb_push(skb, mac_len); 4961 break; 4962 case XDP_PASS: 4963 metalen = xdp->data - xdp->data_meta; 4964 if (metalen) 4965 skb_metadata_set(skb, metalen); 4966 break; 4967 } 4968 4969 return act; 4970 } 4971 4972 static int 4973 netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog) 4974 { 4975 struct sk_buff *skb = *pskb; 4976 int err, hroom, troom; 4977 4978 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog)) 4979 return 0; 4980 4981 /* In case we have to go down the path and also linearize, 4982 * then lets do the pskb_expand_head() work just once here. 4983 */ 4984 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4985 troom = skb->tail + skb->data_len - skb->end; 4986 err = pskb_expand_head(skb, 4987 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4988 troom > 0 ? troom + 128 : 0, GFP_ATOMIC); 4989 if (err) 4990 return err; 4991 4992 return skb_linearize(skb); 4993 } 4994 4995 static u32 netif_receive_generic_xdp(struct sk_buff **pskb, 4996 struct xdp_buff *xdp, 4997 struct bpf_prog *xdp_prog) 4998 { 4999 struct sk_buff *skb = *pskb; 5000 u32 mac_len, act = XDP_DROP; 5001 5002 /* Reinjected packets coming from act_mirred or similar should 5003 * not get XDP generic processing. 5004 */ 5005 if (skb_is_redirected(skb)) 5006 return XDP_PASS; 5007 5008 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM 5009 * bytes. This is the guarantee that also native XDP provides, 5010 * thus we need to do it here as well. 5011 */ 5012 mac_len = skb->data - skb_mac_header(skb); 5013 __skb_push(skb, mac_len); 5014 5015 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 5016 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 5017 if (netif_skb_check_for_xdp(pskb, xdp_prog)) 5018 goto do_drop; 5019 } 5020 5021 __skb_pull(*pskb, mac_len); 5022 5023 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); 5024 switch (act) { 5025 case XDP_REDIRECT: 5026 case XDP_TX: 5027 case XDP_PASS: 5028 break; 5029 default: 5030 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); 5031 fallthrough; 5032 case XDP_ABORTED: 5033 trace_xdp_exception((*pskb)->dev, xdp_prog, act); 5034 fallthrough; 5035 case XDP_DROP: 5036 do_drop: 5037 kfree_skb(*pskb); 5038 break; 5039 } 5040 5041 return act; 5042 } 5043 5044 /* When doing generic XDP we have to bypass the qdisc layer and the 5045 * network taps in order to match in-driver-XDP behavior. This also means 5046 * that XDP packets are able to starve other packets going through a qdisc, 5047 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5048 * queues, so they do not have this starvation issue. 5049 */ 5050 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 5051 { 5052 struct net_device *dev = skb->dev; 5053 struct netdev_queue *txq; 5054 bool free_skb = true; 5055 int cpu, rc; 5056 5057 txq = netdev_core_pick_tx(dev, skb, NULL); 5058 cpu = smp_processor_id(); 5059 HARD_TX_LOCK(dev, txq, cpu); 5060 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5061 rc = netdev_start_xmit(skb, dev, txq, 0); 5062 if (dev_xmit_complete(rc)) 5063 free_skb = false; 5064 } 5065 HARD_TX_UNLOCK(dev, txq); 5066 if (free_skb) { 5067 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5068 dev_core_stats_tx_dropped_inc(dev); 5069 kfree_skb(skb); 5070 } 5071 } 5072 5073 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5074 5075 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb) 5076 { 5077 if (xdp_prog) { 5078 struct xdp_buff xdp; 5079 u32 act; 5080 int err; 5081 5082 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); 5083 if (act != XDP_PASS) { 5084 switch (act) { 5085 case XDP_REDIRECT: 5086 err = xdp_do_generic_redirect((*pskb)->dev, *pskb, 5087 &xdp, xdp_prog); 5088 if (err) 5089 goto out_redir; 5090 break; 5091 case XDP_TX: 5092 generic_xdp_tx(*pskb, xdp_prog); 5093 break; 5094 } 5095 return XDP_DROP; 5096 } 5097 } 5098 return XDP_PASS; 5099 out_redir: 5100 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); 5101 return XDP_DROP; 5102 } 5103 EXPORT_SYMBOL_GPL(do_xdp_generic); 5104 5105 static int netif_rx_internal(struct sk_buff *skb) 5106 { 5107 int ret; 5108 5109 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5110 5111 trace_netif_rx(skb); 5112 5113 #ifdef CONFIG_RPS 5114 if (static_branch_unlikely(&rps_needed)) { 5115 struct rps_dev_flow voidflow, *rflow = &voidflow; 5116 int cpu; 5117 5118 rcu_read_lock(); 5119 5120 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5121 if (cpu < 0) 5122 cpu = smp_processor_id(); 5123 5124 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5125 5126 rcu_read_unlock(); 5127 } else 5128 #endif 5129 { 5130 unsigned int qtail; 5131 5132 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5133 } 5134 return ret; 5135 } 5136 5137 /** 5138 * __netif_rx - Slightly optimized version of netif_rx 5139 * @skb: buffer to post 5140 * 5141 * This behaves as netif_rx except that it does not disable bottom halves. 5142 * As a result this function may only be invoked from the interrupt context 5143 * (either hard or soft interrupt). 5144 */ 5145 int __netif_rx(struct sk_buff *skb) 5146 { 5147 int ret; 5148 5149 lockdep_assert_once(hardirq_count() | softirq_count()); 5150 5151 trace_netif_rx_entry(skb); 5152 ret = netif_rx_internal(skb); 5153 trace_netif_rx_exit(ret); 5154 return ret; 5155 } 5156 EXPORT_SYMBOL(__netif_rx); 5157 5158 /** 5159 * netif_rx - post buffer to the network code 5160 * @skb: buffer to post 5161 * 5162 * This function receives a packet from a device driver and queues it for 5163 * the upper (protocol) levels to process via the backlog NAPI device. It 5164 * always succeeds. The buffer may be dropped during processing for 5165 * congestion control or by the protocol layers. 5166 * The network buffer is passed via the backlog NAPI device. Modern NIC 5167 * driver should use NAPI and GRO. 5168 * This function can used from interrupt and from process context. The 5169 * caller from process context must not disable interrupts before invoking 5170 * this function. 5171 * 5172 * return values: 5173 * NET_RX_SUCCESS (no congestion) 5174 * NET_RX_DROP (packet was dropped) 5175 * 5176 */ 5177 int netif_rx(struct sk_buff *skb) 5178 { 5179 bool need_bh_off = !(hardirq_count() | softirq_count()); 5180 int ret; 5181 5182 if (need_bh_off) 5183 local_bh_disable(); 5184 trace_netif_rx_entry(skb); 5185 ret = netif_rx_internal(skb); 5186 trace_netif_rx_exit(ret); 5187 if (need_bh_off) 5188 local_bh_enable(); 5189 return ret; 5190 } 5191 EXPORT_SYMBOL(netif_rx); 5192 5193 static __latent_entropy void net_tx_action(struct softirq_action *h) 5194 { 5195 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5196 5197 if (sd->completion_queue) { 5198 struct sk_buff *clist; 5199 5200 local_irq_disable(); 5201 clist = sd->completion_queue; 5202 sd->completion_queue = NULL; 5203 local_irq_enable(); 5204 5205 while (clist) { 5206 struct sk_buff *skb = clist; 5207 5208 clist = clist->next; 5209 5210 WARN_ON(refcount_read(&skb->users)); 5211 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5212 trace_consume_skb(skb, net_tx_action); 5213 else 5214 trace_kfree_skb(skb, net_tx_action, 5215 get_kfree_skb_cb(skb)->reason); 5216 5217 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5218 __kfree_skb(skb); 5219 else 5220 __napi_kfree_skb(skb, 5221 get_kfree_skb_cb(skb)->reason); 5222 } 5223 } 5224 5225 if (sd->output_queue) { 5226 struct Qdisc *head; 5227 5228 local_irq_disable(); 5229 head = sd->output_queue; 5230 sd->output_queue = NULL; 5231 sd->output_queue_tailp = &sd->output_queue; 5232 local_irq_enable(); 5233 5234 rcu_read_lock(); 5235 5236 while (head) { 5237 struct Qdisc *q = head; 5238 spinlock_t *root_lock = NULL; 5239 5240 head = head->next_sched; 5241 5242 /* We need to make sure head->next_sched is read 5243 * before clearing __QDISC_STATE_SCHED 5244 */ 5245 smp_mb__before_atomic(); 5246 5247 if (!(q->flags & TCQ_F_NOLOCK)) { 5248 root_lock = qdisc_lock(q); 5249 spin_lock(root_lock); 5250 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5251 &q->state))) { 5252 /* There is a synchronize_net() between 5253 * STATE_DEACTIVATED flag being set and 5254 * qdisc_reset()/some_qdisc_is_busy() in 5255 * dev_deactivate(), so we can safely bail out 5256 * early here to avoid data race between 5257 * qdisc_deactivate() and some_qdisc_is_busy() 5258 * for lockless qdisc. 5259 */ 5260 clear_bit(__QDISC_STATE_SCHED, &q->state); 5261 continue; 5262 } 5263 5264 clear_bit(__QDISC_STATE_SCHED, &q->state); 5265 qdisc_run(q); 5266 if (root_lock) 5267 spin_unlock(root_lock); 5268 } 5269 5270 rcu_read_unlock(); 5271 } 5272 5273 xfrm_dev_backlog(sd); 5274 } 5275 5276 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5277 /* This hook is defined here for ATM LANE */ 5278 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5279 unsigned char *addr) __read_mostly; 5280 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5281 #endif 5282 5283 /** 5284 * netdev_is_rx_handler_busy - check if receive handler is registered 5285 * @dev: device to check 5286 * 5287 * Check if a receive handler is already registered for a given device. 5288 * Return true if there one. 5289 * 5290 * The caller must hold the rtnl_mutex. 5291 */ 5292 bool netdev_is_rx_handler_busy(struct net_device *dev) 5293 { 5294 ASSERT_RTNL(); 5295 return dev && rtnl_dereference(dev->rx_handler); 5296 } 5297 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5298 5299 /** 5300 * netdev_rx_handler_register - register receive handler 5301 * @dev: device to register a handler for 5302 * @rx_handler: receive handler to register 5303 * @rx_handler_data: data pointer that is used by rx handler 5304 * 5305 * Register a receive handler for a device. This handler will then be 5306 * called from __netif_receive_skb. A negative errno code is returned 5307 * on a failure. 5308 * 5309 * The caller must hold the rtnl_mutex. 5310 * 5311 * For a general description of rx_handler, see enum rx_handler_result. 5312 */ 5313 int netdev_rx_handler_register(struct net_device *dev, 5314 rx_handler_func_t *rx_handler, 5315 void *rx_handler_data) 5316 { 5317 if (netdev_is_rx_handler_busy(dev)) 5318 return -EBUSY; 5319 5320 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5321 return -EINVAL; 5322 5323 /* Note: rx_handler_data must be set before rx_handler */ 5324 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5325 rcu_assign_pointer(dev->rx_handler, rx_handler); 5326 5327 return 0; 5328 } 5329 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5330 5331 /** 5332 * netdev_rx_handler_unregister - unregister receive handler 5333 * @dev: device to unregister a handler from 5334 * 5335 * Unregister a receive handler from a device. 5336 * 5337 * The caller must hold the rtnl_mutex. 5338 */ 5339 void netdev_rx_handler_unregister(struct net_device *dev) 5340 { 5341 5342 ASSERT_RTNL(); 5343 RCU_INIT_POINTER(dev->rx_handler, NULL); 5344 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5345 * section has a guarantee to see a non NULL rx_handler_data 5346 * as well. 5347 */ 5348 synchronize_net(); 5349 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5350 } 5351 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5352 5353 /* 5354 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5355 * the special handling of PFMEMALLOC skbs. 5356 */ 5357 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5358 { 5359 switch (skb->protocol) { 5360 case htons(ETH_P_ARP): 5361 case htons(ETH_P_IP): 5362 case htons(ETH_P_IPV6): 5363 case htons(ETH_P_8021Q): 5364 case htons(ETH_P_8021AD): 5365 return true; 5366 default: 5367 return false; 5368 } 5369 } 5370 5371 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5372 int *ret, struct net_device *orig_dev) 5373 { 5374 if (nf_hook_ingress_active(skb)) { 5375 int ingress_retval; 5376 5377 if (*pt_prev) { 5378 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5379 *pt_prev = NULL; 5380 } 5381 5382 rcu_read_lock(); 5383 ingress_retval = nf_hook_ingress(skb); 5384 rcu_read_unlock(); 5385 return ingress_retval; 5386 } 5387 return 0; 5388 } 5389 5390 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5391 struct packet_type **ppt_prev) 5392 { 5393 struct packet_type *ptype, *pt_prev; 5394 rx_handler_func_t *rx_handler; 5395 struct sk_buff *skb = *pskb; 5396 struct net_device *orig_dev; 5397 bool deliver_exact = false; 5398 int ret = NET_RX_DROP; 5399 __be16 type; 5400 5401 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5402 5403 trace_netif_receive_skb(skb); 5404 5405 orig_dev = skb->dev; 5406 5407 skb_reset_network_header(skb); 5408 if (!skb_transport_header_was_set(skb)) 5409 skb_reset_transport_header(skb); 5410 skb_reset_mac_len(skb); 5411 5412 pt_prev = NULL; 5413 5414 another_round: 5415 skb->skb_iif = skb->dev->ifindex; 5416 5417 __this_cpu_inc(softnet_data.processed); 5418 5419 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5420 int ret2; 5421 5422 migrate_disable(); 5423 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), 5424 &skb); 5425 migrate_enable(); 5426 5427 if (ret2 != XDP_PASS) { 5428 ret = NET_RX_DROP; 5429 goto out; 5430 } 5431 } 5432 5433 if (eth_type_vlan(skb->protocol)) { 5434 skb = skb_vlan_untag(skb); 5435 if (unlikely(!skb)) 5436 goto out; 5437 } 5438 5439 if (skb_skip_tc_classify(skb)) 5440 goto skip_classify; 5441 5442 if (pfmemalloc) 5443 goto skip_taps; 5444 5445 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) { 5446 if (pt_prev) 5447 ret = deliver_skb(skb, pt_prev, orig_dev); 5448 pt_prev = ptype; 5449 } 5450 5451 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5452 if (pt_prev) 5453 ret = deliver_skb(skb, pt_prev, orig_dev); 5454 pt_prev = ptype; 5455 } 5456 5457 skip_taps: 5458 #ifdef CONFIG_NET_INGRESS 5459 if (static_branch_unlikely(&ingress_needed_key)) { 5460 bool another = false; 5461 5462 nf_skip_egress(skb, true); 5463 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5464 &another); 5465 if (another) 5466 goto another_round; 5467 if (!skb) 5468 goto out; 5469 5470 nf_skip_egress(skb, false); 5471 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5472 goto out; 5473 } 5474 #endif 5475 skb_reset_redirect(skb); 5476 skip_classify: 5477 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5478 goto drop; 5479 5480 if (skb_vlan_tag_present(skb)) { 5481 if (pt_prev) { 5482 ret = deliver_skb(skb, pt_prev, orig_dev); 5483 pt_prev = NULL; 5484 } 5485 if (vlan_do_receive(&skb)) 5486 goto another_round; 5487 else if (unlikely(!skb)) 5488 goto out; 5489 } 5490 5491 rx_handler = rcu_dereference(skb->dev->rx_handler); 5492 if (rx_handler) { 5493 if (pt_prev) { 5494 ret = deliver_skb(skb, pt_prev, orig_dev); 5495 pt_prev = NULL; 5496 } 5497 switch (rx_handler(&skb)) { 5498 case RX_HANDLER_CONSUMED: 5499 ret = NET_RX_SUCCESS; 5500 goto out; 5501 case RX_HANDLER_ANOTHER: 5502 goto another_round; 5503 case RX_HANDLER_EXACT: 5504 deliver_exact = true; 5505 break; 5506 case RX_HANDLER_PASS: 5507 break; 5508 default: 5509 BUG(); 5510 } 5511 } 5512 5513 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5514 check_vlan_id: 5515 if (skb_vlan_tag_get_id(skb)) { 5516 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5517 * find vlan device. 5518 */ 5519 skb->pkt_type = PACKET_OTHERHOST; 5520 } else if (eth_type_vlan(skb->protocol)) { 5521 /* Outer header is 802.1P with vlan 0, inner header is 5522 * 802.1Q or 802.1AD and vlan_do_receive() above could 5523 * not find vlan dev for vlan id 0. 5524 */ 5525 __vlan_hwaccel_clear_tag(skb); 5526 skb = skb_vlan_untag(skb); 5527 if (unlikely(!skb)) 5528 goto out; 5529 if (vlan_do_receive(&skb)) 5530 /* After stripping off 802.1P header with vlan 0 5531 * vlan dev is found for inner header. 5532 */ 5533 goto another_round; 5534 else if (unlikely(!skb)) 5535 goto out; 5536 else 5537 /* We have stripped outer 802.1P vlan 0 header. 5538 * But could not find vlan dev. 5539 * check again for vlan id to set OTHERHOST. 5540 */ 5541 goto check_vlan_id; 5542 } 5543 /* Note: we might in the future use prio bits 5544 * and set skb->priority like in vlan_do_receive() 5545 * For the time being, just ignore Priority Code Point 5546 */ 5547 __vlan_hwaccel_clear_tag(skb); 5548 } 5549 5550 type = skb->protocol; 5551 5552 /* deliver only exact match when indicated */ 5553 if (likely(!deliver_exact)) { 5554 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5555 &ptype_base[ntohs(type) & 5556 PTYPE_HASH_MASK]); 5557 } 5558 5559 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5560 &orig_dev->ptype_specific); 5561 5562 if (unlikely(skb->dev != orig_dev)) { 5563 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5564 &skb->dev->ptype_specific); 5565 } 5566 5567 if (pt_prev) { 5568 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5569 goto drop; 5570 *ppt_prev = pt_prev; 5571 } else { 5572 drop: 5573 if (!deliver_exact) 5574 dev_core_stats_rx_dropped_inc(skb->dev); 5575 else 5576 dev_core_stats_rx_nohandler_inc(skb->dev); 5577 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5578 /* Jamal, now you will not able to escape explaining 5579 * me how you were going to use this. :-) 5580 */ 5581 ret = NET_RX_DROP; 5582 } 5583 5584 out: 5585 /* The invariant here is that if *ppt_prev is not NULL 5586 * then skb should also be non-NULL. 5587 * 5588 * Apparently *ppt_prev assignment above holds this invariant due to 5589 * skb dereferencing near it. 5590 */ 5591 *pskb = skb; 5592 return ret; 5593 } 5594 5595 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5596 { 5597 struct net_device *orig_dev = skb->dev; 5598 struct packet_type *pt_prev = NULL; 5599 int ret; 5600 5601 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5602 if (pt_prev) 5603 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5604 skb->dev, pt_prev, orig_dev); 5605 return ret; 5606 } 5607 5608 /** 5609 * netif_receive_skb_core - special purpose version of netif_receive_skb 5610 * @skb: buffer to process 5611 * 5612 * More direct receive version of netif_receive_skb(). It should 5613 * only be used by callers that have a need to skip RPS and Generic XDP. 5614 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5615 * 5616 * This function may only be called from softirq context and interrupts 5617 * should be enabled. 5618 * 5619 * Return values (usually ignored): 5620 * NET_RX_SUCCESS: no congestion 5621 * NET_RX_DROP: packet was dropped 5622 */ 5623 int netif_receive_skb_core(struct sk_buff *skb) 5624 { 5625 int ret; 5626 5627 rcu_read_lock(); 5628 ret = __netif_receive_skb_one_core(skb, false); 5629 rcu_read_unlock(); 5630 5631 return ret; 5632 } 5633 EXPORT_SYMBOL(netif_receive_skb_core); 5634 5635 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5636 struct packet_type *pt_prev, 5637 struct net_device *orig_dev) 5638 { 5639 struct sk_buff *skb, *next; 5640 5641 if (!pt_prev) 5642 return; 5643 if (list_empty(head)) 5644 return; 5645 if (pt_prev->list_func != NULL) 5646 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5647 ip_list_rcv, head, pt_prev, orig_dev); 5648 else 5649 list_for_each_entry_safe(skb, next, head, list) { 5650 skb_list_del_init(skb); 5651 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5652 } 5653 } 5654 5655 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5656 { 5657 /* Fast-path assumptions: 5658 * - There is no RX handler. 5659 * - Only one packet_type matches. 5660 * If either of these fails, we will end up doing some per-packet 5661 * processing in-line, then handling the 'last ptype' for the whole 5662 * sublist. This can't cause out-of-order delivery to any single ptype, 5663 * because the 'last ptype' must be constant across the sublist, and all 5664 * other ptypes are handled per-packet. 5665 */ 5666 /* Current (common) ptype of sublist */ 5667 struct packet_type *pt_curr = NULL; 5668 /* Current (common) orig_dev of sublist */ 5669 struct net_device *od_curr = NULL; 5670 struct list_head sublist; 5671 struct sk_buff *skb, *next; 5672 5673 INIT_LIST_HEAD(&sublist); 5674 list_for_each_entry_safe(skb, next, head, list) { 5675 struct net_device *orig_dev = skb->dev; 5676 struct packet_type *pt_prev = NULL; 5677 5678 skb_list_del_init(skb); 5679 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5680 if (!pt_prev) 5681 continue; 5682 if (pt_curr != pt_prev || od_curr != orig_dev) { 5683 /* dispatch old sublist */ 5684 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5685 /* start new sublist */ 5686 INIT_LIST_HEAD(&sublist); 5687 pt_curr = pt_prev; 5688 od_curr = orig_dev; 5689 } 5690 list_add_tail(&skb->list, &sublist); 5691 } 5692 5693 /* dispatch final sublist */ 5694 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5695 } 5696 5697 static int __netif_receive_skb(struct sk_buff *skb) 5698 { 5699 int ret; 5700 5701 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5702 unsigned int noreclaim_flag; 5703 5704 /* 5705 * PFMEMALLOC skbs are special, they should 5706 * - be delivered to SOCK_MEMALLOC sockets only 5707 * - stay away from userspace 5708 * - have bounded memory usage 5709 * 5710 * Use PF_MEMALLOC as this saves us from propagating the allocation 5711 * context down to all allocation sites. 5712 */ 5713 noreclaim_flag = memalloc_noreclaim_save(); 5714 ret = __netif_receive_skb_one_core(skb, true); 5715 memalloc_noreclaim_restore(noreclaim_flag); 5716 } else 5717 ret = __netif_receive_skb_one_core(skb, false); 5718 5719 return ret; 5720 } 5721 5722 static void __netif_receive_skb_list(struct list_head *head) 5723 { 5724 unsigned long noreclaim_flag = 0; 5725 struct sk_buff *skb, *next; 5726 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5727 5728 list_for_each_entry_safe(skb, next, head, list) { 5729 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5730 struct list_head sublist; 5731 5732 /* Handle the previous sublist */ 5733 list_cut_before(&sublist, head, &skb->list); 5734 if (!list_empty(&sublist)) 5735 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5736 pfmemalloc = !pfmemalloc; 5737 /* See comments in __netif_receive_skb */ 5738 if (pfmemalloc) 5739 noreclaim_flag = memalloc_noreclaim_save(); 5740 else 5741 memalloc_noreclaim_restore(noreclaim_flag); 5742 } 5743 } 5744 /* Handle the remaining sublist */ 5745 if (!list_empty(head)) 5746 __netif_receive_skb_list_core(head, pfmemalloc); 5747 /* Restore pflags */ 5748 if (pfmemalloc) 5749 memalloc_noreclaim_restore(noreclaim_flag); 5750 } 5751 5752 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5753 { 5754 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5755 struct bpf_prog *new = xdp->prog; 5756 int ret = 0; 5757 5758 switch (xdp->command) { 5759 case XDP_SETUP_PROG: 5760 rcu_assign_pointer(dev->xdp_prog, new); 5761 if (old) 5762 bpf_prog_put(old); 5763 5764 if (old && !new) { 5765 static_branch_dec(&generic_xdp_needed_key); 5766 } else if (new && !old) { 5767 static_branch_inc(&generic_xdp_needed_key); 5768 dev_disable_lro(dev); 5769 dev_disable_gro_hw(dev); 5770 } 5771 break; 5772 5773 default: 5774 ret = -EINVAL; 5775 break; 5776 } 5777 5778 return ret; 5779 } 5780 5781 static int netif_receive_skb_internal(struct sk_buff *skb) 5782 { 5783 int ret; 5784 5785 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb); 5786 5787 if (skb_defer_rx_timestamp(skb)) 5788 return NET_RX_SUCCESS; 5789 5790 rcu_read_lock(); 5791 #ifdef CONFIG_RPS 5792 if (static_branch_unlikely(&rps_needed)) { 5793 struct rps_dev_flow voidflow, *rflow = &voidflow; 5794 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5795 5796 if (cpu >= 0) { 5797 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5798 rcu_read_unlock(); 5799 return ret; 5800 } 5801 } 5802 #endif 5803 ret = __netif_receive_skb(skb); 5804 rcu_read_unlock(); 5805 return ret; 5806 } 5807 5808 void netif_receive_skb_list_internal(struct list_head *head) 5809 { 5810 struct sk_buff *skb, *next; 5811 struct list_head sublist; 5812 5813 INIT_LIST_HEAD(&sublist); 5814 list_for_each_entry_safe(skb, next, head, list) { 5815 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), 5816 skb); 5817 skb_list_del_init(skb); 5818 if (!skb_defer_rx_timestamp(skb)) 5819 list_add_tail(&skb->list, &sublist); 5820 } 5821 list_splice_init(&sublist, head); 5822 5823 rcu_read_lock(); 5824 #ifdef CONFIG_RPS 5825 if (static_branch_unlikely(&rps_needed)) { 5826 list_for_each_entry_safe(skb, next, head, list) { 5827 struct rps_dev_flow voidflow, *rflow = &voidflow; 5828 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5829 5830 if (cpu >= 0) { 5831 /* Will be handled, remove from list */ 5832 skb_list_del_init(skb); 5833 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5834 } 5835 } 5836 } 5837 #endif 5838 __netif_receive_skb_list(head); 5839 rcu_read_unlock(); 5840 } 5841 5842 /** 5843 * netif_receive_skb - process receive buffer from network 5844 * @skb: buffer to process 5845 * 5846 * netif_receive_skb() is the main receive data processing function. 5847 * It always succeeds. The buffer may be dropped during processing 5848 * for congestion control or by the protocol layers. 5849 * 5850 * This function may only be called from softirq context and interrupts 5851 * should be enabled. 5852 * 5853 * Return values (usually ignored): 5854 * NET_RX_SUCCESS: no congestion 5855 * NET_RX_DROP: packet was dropped 5856 */ 5857 int netif_receive_skb(struct sk_buff *skb) 5858 { 5859 int ret; 5860 5861 trace_netif_receive_skb_entry(skb); 5862 5863 ret = netif_receive_skb_internal(skb); 5864 trace_netif_receive_skb_exit(ret); 5865 5866 return ret; 5867 } 5868 EXPORT_SYMBOL(netif_receive_skb); 5869 5870 /** 5871 * netif_receive_skb_list - process many receive buffers from network 5872 * @head: list of skbs to process. 5873 * 5874 * Since return value of netif_receive_skb() is normally ignored, and 5875 * wouldn't be meaningful for a list, this function returns void. 5876 * 5877 * This function may only be called from softirq context and interrupts 5878 * should be enabled. 5879 */ 5880 void netif_receive_skb_list(struct list_head *head) 5881 { 5882 struct sk_buff *skb; 5883 5884 if (list_empty(head)) 5885 return; 5886 if (trace_netif_receive_skb_list_entry_enabled()) { 5887 list_for_each_entry(skb, head, list) 5888 trace_netif_receive_skb_list_entry(skb); 5889 } 5890 netif_receive_skb_list_internal(head); 5891 trace_netif_receive_skb_list_exit(0); 5892 } 5893 EXPORT_SYMBOL(netif_receive_skb_list); 5894 5895 static DEFINE_PER_CPU(struct work_struct, flush_works); 5896 5897 /* Network device is going away, flush any packets still pending */ 5898 static void flush_backlog(struct work_struct *work) 5899 { 5900 struct sk_buff *skb, *tmp; 5901 struct softnet_data *sd; 5902 5903 local_bh_disable(); 5904 sd = this_cpu_ptr(&softnet_data); 5905 5906 backlog_lock_irq_disable(sd); 5907 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5908 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5909 __skb_unlink(skb, &sd->input_pkt_queue); 5910 dev_kfree_skb_irq(skb); 5911 rps_input_queue_head_incr(sd); 5912 } 5913 } 5914 backlog_unlock_irq_enable(sd); 5915 5916 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5917 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5918 __skb_unlink(skb, &sd->process_queue); 5919 kfree_skb(skb); 5920 rps_input_queue_head_incr(sd); 5921 } 5922 } 5923 local_bh_enable(); 5924 } 5925 5926 static bool flush_required(int cpu) 5927 { 5928 #if IS_ENABLED(CONFIG_RPS) 5929 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5930 bool do_flush; 5931 5932 backlog_lock_irq_disable(sd); 5933 5934 /* as insertion into process_queue happens with the rps lock held, 5935 * process_queue access may race only with dequeue 5936 */ 5937 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5938 !skb_queue_empty_lockless(&sd->process_queue); 5939 backlog_unlock_irq_enable(sd); 5940 5941 return do_flush; 5942 #endif 5943 /* without RPS we can't safely check input_pkt_queue: during a 5944 * concurrent remote skb_queue_splice() we can detect as empty both 5945 * input_pkt_queue and process_queue even if the latter could end-up 5946 * containing a lot of packets. 5947 */ 5948 return true; 5949 } 5950 5951 static void flush_all_backlogs(void) 5952 { 5953 static cpumask_t flush_cpus; 5954 unsigned int cpu; 5955 5956 /* since we are under rtnl lock protection we can use static data 5957 * for the cpumask and avoid allocating on stack the possibly 5958 * large mask 5959 */ 5960 ASSERT_RTNL(); 5961 5962 cpus_read_lock(); 5963 5964 cpumask_clear(&flush_cpus); 5965 for_each_online_cpu(cpu) { 5966 if (flush_required(cpu)) { 5967 queue_work_on(cpu, system_highpri_wq, 5968 per_cpu_ptr(&flush_works, cpu)); 5969 cpumask_set_cpu(cpu, &flush_cpus); 5970 } 5971 } 5972 5973 /* we can have in flight packet[s] on the cpus we are not flushing, 5974 * synchronize_net() in unregister_netdevice_many() will take care of 5975 * them 5976 */ 5977 for_each_cpu(cpu, &flush_cpus) 5978 flush_work(per_cpu_ptr(&flush_works, cpu)); 5979 5980 cpus_read_unlock(); 5981 } 5982 5983 static void net_rps_send_ipi(struct softnet_data *remsd) 5984 { 5985 #ifdef CONFIG_RPS 5986 while (remsd) { 5987 struct softnet_data *next = remsd->rps_ipi_next; 5988 5989 if (cpu_online(remsd->cpu)) 5990 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5991 remsd = next; 5992 } 5993 #endif 5994 } 5995 5996 /* 5997 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5998 * Note: called with local irq disabled, but exits with local irq enabled. 5999 */ 6000 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6001 { 6002 #ifdef CONFIG_RPS 6003 struct softnet_data *remsd = sd->rps_ipi_list; 6004 6005 if (!use_backlog_threads() && remsd) { 6006 sd->rps_ipi_list = NULL; 6007 6008 local_irq_enable(); 6009 6010 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6011 net_rps_send_ipi(remsd); 6012 } else 6013 #endif 6014 local_irq_enable(); 6015 } 6016 6017 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6018 { 6019 #ifdef CONFIG_RPS 6020 return !use_backlog_threads() && sd->rps_ipi_list; 6021 #else 6022 return false; 6023 #endif 6024 } 6025 6026 static int process_backlog(struct napi_struct *napi, int quota) 6027 { 6028 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6029 bool again = true; 6030 int work = 0; 6031 6032 /* Check if we have pending ipi, its better to send them now, 6033 * not waiting net_rx_action() end. 6034 */ 6035 if (sd_has_rps_ipi_waiting(sd)) { 6036 local_irq_disable(); 6037 net_rps_action_and_irq_enable(sd); 6038 } 6039 6040 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight); 6041 while (again) { 6042 struct sk_buff *skb; 6043 6044 while ((skb = __skb_dequeue(&sd->process_queue))) { 6045 rcu_read_lock(); 6046 __netif_receive_skb(skb); 6047 rcu_read_unlock(); 6048 if (++work >= quota) { 6049 rps_input_queue_head_add(sd, work); 6050 return work; 6051 } 6052 6053 } 6054 6055 backlog_lock_irq_disable(sd); 6056 if (skb_queue_empty(&sd->input_pkt_queue)) { 6057 /* 6058 * Inline a custom version of __napi_complete(). 6059 * only current cpu owns and manipulates this napi, 6060 * and NAPI_STATE_SCHED is the only possible flag set 6061 * on backlog. 6062 * We can use a plain write instead of clear_bit(), 6063 * and we dont need an smp_mb() memory barrier. 6064 */ 6065 napi->state &= NAPIF_STATE_THREADED; 6066 again = false; 6067 } else { 6068 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6069 &sd->process_queue); 6070 } 6071 backlog_unlock_irq_enable(sd); 6072 } 6073 6074 if (work) 6075 rps_input_queue_head_add(sd, work); 6076 return work; 6077 } 6078 6079 /** 6080 * __napi_schedule - schedule for receive 6081 * @n: entry to schedule 6082 * 6083 * The entry's receive function will be scheduled to run. 6084 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6085 */ 6086 void __napi_schedule(struct napi_struct *n) 6087 { 6088 unsigned long flags; 6089 6090 local_irq_save(flags); 6091 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6092 local_irq_restore(flags); 6093 } 6094 EXPORT_SYMBOL(__napi_schedule); 6095 6096 /** 6097 * napi_schedule_prep - check if napi can be scheduled 6098 * @n: napi context 6099 * 6100 * Test if NAPI routine is already running, and if not mark 6101 * it as running. This is used as a condition variable to 6102 * insure only one NAPI poll instance runs. We also make 6103 * sure there is no pending NAPI disable. 6104 */ 6105 bool napi_schedule_prep(struct napi_struct *n) 6106 { 6107 unsigned long new, val = READ_ONCE(n->state); 6108 6109 do { 6110 if (unlikely(val & NAPIF_STATE_DISABLE)) 6111 return false; 6112 new = val | NAPIF_STATE_SCHED; 6113 6114 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6115 * This was suggested by Alexander Duyck, as compiler 6116 * emits better code than : 6117 * if (val & NAPIF_STATE_SCHED) 6118 * new |= NAPIF_STATE_MISSED; 6119 */ 6120 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6121 NAPIF_STATE_MISSED; 6122 } while (!try_cmpxchg(&n->state, &val, new)); 6123 6124 return !(val & NAPIF_STATE_SCHED); 6125 } 6126 EXPORT_SYMBOL(napi_schedule_prep); 6127 6128 /** 6129 * __napi_schedule_irqoff - schedule for receive 6130 * @n: entry to schedule 6131 * 6132 * Variant of __napi_schedule() assuming hard irqs are masked. 6133 * 6134 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6135 * because the interrupt disabled assumption might not be true 6136 * due to force-threaded interrupts and spinlock substitution. 6137 */ 6138 void __napi_schedule_irqoff(struct napi_struct *n) 6139 { 6140 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6141 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6142 else 6143 __napi_schedule(n); 6144 } 6145 EXPORT_SYMBOL(__napi_schedule_irqoff); 6146 6147 bool napi_complete_done(struct napi_struct *n, int work_done) 6148 { 6149 unsigned long flags, val, new, timeout = 0; 6150 bool ret = true; 6151 6152 /* 6153 * 1) Don't let napi dequeue from the cpu poll list 6154 * just in case its running on a different cpu. 6155 * 2) If we are busy polling, do nothing here, we have 6156 * the guarantee we will be called later. 6157 */ 6158 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6159 NAPIF_STATE_IN_BUSY_POLL))) 6160 return false; 6161 6162 if (work_done) { 6163 if (n->gro_bitmask) 6164 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6165 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6166 } 6167 if (n->defer_hard_irqs_count > 0) { 6168 n->defer_hard_irqs_count--; 6169 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6170 if (timeout) 6171 ret = false; 6172 } 6173 if (n->gro_bitmask) { 6174 /* When the NAPI instance uses a timeout and keeps postponing 6175 * it, we need to bound somehow the time packets are kept in 6176 * the GRO layer 6177 */ 6178 napi_gro_flush(n, !!timeout); 6179 } 6180 6181 gro_normal_list(n); 6182 6183 if (unlikely(!list_empty(&n->poll_list))) { 6184 /* If n->poll_list is not empty, we need to mask irqs */ 6185 local_irq_save(flags); 6186 list_del_init(&n->poll_list); 6187 local_irq_restore(flags); 6188 } 6189 WRITE_ONCE(n->list_owner, -1); 6190 6191 val = READ_ONCE(n->state); 6192 do { 6193 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6194 6195 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6196 NAPIF_STATE_SCHED_THREADED | 6197 NAPIF_STATE_PREFER_BUSY_POLL); 6198 6199 /* If STATE_MISSED was set, leave STATE_SCHED set, 6200 * because we will call napi->poll() one more time. 6201 * This C code was suggested by Alexander Duyck to help gcc. 6202 */ 6203 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6204 NAPIF_STATE_SCHED; 6205 } while (!try_cmpxchg(&n->state, &val, new)); 6206 6207 if (unlikely(val & NAPIF_STATE_MISSED)) { 6208 __napi_schedule(n); 6209 return false; 6210 } 6211 6212 if (timeout) 6213 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6214 HRTIMER_MODE_REL_PINNED); 6215 return ret; 6216 } 6217 EXPORT_SYMBOL(napi_complete_done); 6218 6219 /* must be called under rcu_read_lock(), as we dont take a reference */ 6220 struct napi_struct *napi_by_id(unsigned int napi_id) 6221 { 6222 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6223 struct napi_struct *napi; 6224 6225 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6226 if (napi->napi_id == napi_id) 6227 return napi; 6228 6229 return NULL; 6230 } 6231 6232 static void skb_defer_free_flush(struct softnet_data *sd) 6233 { 6234 struct sk_buff *skb, *next; 6235 6236 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6237 if (!READ_ONCE(sd->defer_list)) 6238 return; 6239 6240 spin_lock(&sd->defer_lock); 6241 skb = sd->defer_list; 6242 sd->defer_list = NULL; 6243 sd->defer_count = 0; 6244 spin_unlock(&sd->defer_lock); 6245 6246 while (skb != NULL) { 6247 next = skb->next; 6248 napi_consume_skb(skb, 1); 6249 skb = next; 6250 } 6251 } 6252 6253 #if defined(CONFIG_NET_RX_BUSY_POLL) 6254 6255 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6256 { 6257 if (!skip_schedule) { 6258 gro_normal_list(napi); 6259 __napi_schedule(napi); 6260 return; 6261 } 6262 6263 if (napi->gro_bitmask) { 6264 /* flush too old packets 6265 * If HZ < 1000, flush all packets. 6266 */ 6267 napi_gro_flush(napi, HZ >= 1000); 6268 } 6269 6270 gro_normal_list(napi); 6271 clear_bit(NAPI_STATE_SCHED, &napi->state); 6272 } 6273 6274 enum { 6275 NAPI_F_PREFER_BUSY_POLL = 1, 6276 NAPI_F_END_ON_RESCHED = 2, 6277 }; 6278 6279 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, 6280 unsigned flags, u16 budget) 6281 { 6282 bool skip_schedule = false; 6283 unsigned long timeout; 6284 int rc; 6285 6286 /* Busy polling means there is a high chance device driver hard irq 6287 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6288 * set in napi_schedule_prep(). 6289 * Since we are about to call napi->poll() once more, we can safely 6290 * clear NAPI_STATE_MISSED. 6291 * 6292 * Note: x86 could use a single "lock and ..." instruction 6293 * to perform these two clear_bit() 6294 */ 6295 clear_bit(NAPI_STATE_MISSED, &napi->state); 6296 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6297 6298 local_bh_disable(); 6299 6300 if (flags & NAPI_F_PREFER_BUSY_POLL) { 6301 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6302 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6303 if (napi->defer_hard_irqs_count && timeout) { 6304 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6305 skip_schedule = true; 6306 } 6307 } 6308 6309 /* All we really want here is to re-enable device interrupts. 6310 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6311 */ 6312 rc = napi->poll(napi, budget); 6313 /* We can't gro_normal_list() here, because napi->poll() might have 6314 * rearmed the napi (napi_complete_done()) in which case it could 6315 * already be running on another CPU. 6316 */ 6317 trace_napi_poll(napi, rc, budget); 6318 netpoll_poll_unlock(have_poll_lock); 6319 if (rc == budget) 6320 __busy_poll_stop(napi, skip_schedule); 6321 local_bh_enable(); 6322 } 6323 6324 static void __napi_busy_loop(unsigned int napi_id, 6325 bool (*loop_end)(void *, unsigned long), 6326 void *loop_end_arg, unsigned flags, u16 budget) 6327 { 6328 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6329 int (*napi_poll)(struct napi_struct *napi, int budget); 6330 void *have_poll_lock = NULL; 6331 struct napi_struct *napi; 6332 6333 WARN_ON_ONCE(!rcu_read_lock_held()); 6334 6335 restart: 6336 napi_poll = NULL; 6337 6338 napi = napi_by_id(napi_id); 6339 if (!napi) 6340 return; 6341 6342 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6343 preempt_disable(); 6344 for (;;) { 6345 int work = 0; 6346 6347 local_bh_disable(); 6348 if (!napi_poll) { 6349 unsigned long val = READ_ONCE(napi->state); 6350 6351 /* If multiple threads are competing for this napi, 6352 * we avoid dirtying napi->state as much as we can. 6353 */ 6354 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6355 NAPIF_STATE_IN_BUSY_POLL)) { 6356 if (flags & NAPI_F_PREFER_BUSY_POLL) 6357 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6358 goto count; 6359 } 6360 if (cmpxchg(&napi->state, val, 6361 val | NAPIF_STATE_IN_BUSY_POLL | 6362 NAPIF_STATE_SCHED) != val) { 6363 if (flags & NAPI_F_PREFER_BUSY_POLL) 6364 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6365 goto count; 6366 } 6367 have_poll_lock = netpoll_poll_lock(napi); 6368 napi_poll = napi->poll; 6369 } 6370 work = napi_poll(napi, budget); 6371 trace_napi_poll(napi, work, budget); 6372 gro_normal_list(napi); 6373 count: 6374 if (work > 0) 6375 __NET_ADD_STATS(dev_net(napi->dev), 6376 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6377 skb_defer_free_flush(this_cpu_ptr(&softnet_data)); 6378 local_bh_enable(); 6379 6380 if (!loop_end || loop_end(loop_end_arg, start_time)) 6381 break; 6382 6383 if (unlikely(need_resched())) { 6384 if (flags & NAPI_F_END_ON_RESCHED) 6385 break; 6386 if (napi_poll) 6387 busy_poll_stop(napi, have_poll_lock, flags, budget); 6388 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6389 preempt_enable(); 6390 rcu_read_unlock(); 6391 cond_resched(); 6392 rcu_read_lock(); 6393 if (loop_end(loop_end_arg, start_time)) 6394 return; 6395 goto restart; 6396 } 6397 cpu_relax(); 6398 } 6399 if (napi_poll) 6400 busy_poll_stop(napi, have_poll_lock, flags, budget); 6401 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6402 preempt_enable(); 6403 } 6404 6405 void napi_busy_loop_rcu(unsigned int napi_id, 6406 bool (*loop_end)(void *, unsigned long), 6407 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6408 { 6409 unsigned flags = NAPI_F_END_ON_RESCHED; 6410 6411 if (prefer_busy_poll) 6412 flags |= NAPI_F_PREFER_BUSY_POLL; 6413 6414 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6415 } 6416 6417 void napi_busy_loop(unsigned int napi_id, 6418 bool (*loop_end)(void *, unsigned long), 6419 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6420 { 6421 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; 6422 6423 rcu_read_lock(); 6424 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6425 rcu_read_unlock(); 6426 } 6427 EXPORT_SYMBOL(napi_busy_loop); 6428 6429 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6430 6431 static void napi_hash_add(struct napi_struct *napi) 6432 { 6433 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6434 return; 6435 6436 spin_lock(&napi_hash_lock); 6437 6438 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6439 do { 6440 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6441 napi_gen_id = MIN_NAPI_ID; 6442 } while (napi_by_id(napi_gen_id)); 6443 napi->napi_id = napi_gen_id; 6444 6445 hlist_add_head_rcu(&napi->napi_hash_node, 6446 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6447 6448 spin_unlock(&napi_hash_lock); 6449 } 6450 6451 /* Warning : caller is responsible to make sure rcu grace period 6452 * is respected before freeing memory containing @napi 6453 */ 6454 static void napi_hash_del(struct napi_struct *napi) 6455 { 6456 spin_lock(&napi_hash_lock); 6457 6458 hlist_del_init_rcu(&napi->napi_hash_node); 6459 6460 spin_unlock(&napi_hash_lock); 6461 } 6462 6463 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6464 { 6465 struct napi_struct *napi; 6466 6467 napi = container_of(timer, struct napi_struct, timer); 6468 6469 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6470 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6471 */ 6472 if (!napi_disable_pending(napi) && 6473 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6474 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6475 __napi_schedule_irqoff(napi); 6476 } 6477 6478 return HRTIMER_NORESTART; 6479 } 6480 6481 static void init_gro_hash(struct napi_struct *napi) 6482 { 6483 int i; 6484 6485 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6486 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6487 napi->gro_hash[i].count = 0; 6488 } 6489 napi->gro_bitmask = 0; 6490 } 6491 6492 int dev_set_threaded(struct net_device *dev, bool threaded) 6493 { 6494 struct napi_struct *napi; 6495 int err = 0; 6496 6497 if (dev->threaded == threaded) 6498 return 0; 6499 6500 if (threaded) { 6501 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6502 if (!napi->thread) { 6503 err = napi_kthread_create(napi); 6504 if (err) { 6505 threaded = false; 6506 break; 6507 } 6508 } 6509 } 6510 } 6511 6512 dev->threaded = threaded; 6513 6514 /* Make sure kthread is created before THREADED bit 6515 * is set. 6516 */ 6517 smp_mb__before_atomic(); 6518 6519 /* Setting/unsetting threaded mode on a napi might not immediately 6520 * take effect, if the current napi instance is actively being 6521 * polled. In this case, the switch between threaded mode and 6522 * softirq mode will happen in the next round of napi_schedule(). 6523 * This should not cause hiccups/stalls to the live traffic. 6524 */ 6525 list_for_each_entry(napi, &dev->napi_list, dev_list) 6526 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6527 6528 return err; 6529 } 6530 EXPORT_SYMBOL(dev_set_threaded); 6531 6532 /** 6533 * netif_queue_set_napi - Associate queue with the napi 6534 * @dev: device to which NAPI and queue belong 6535 * @queue_index: Index of queue 6536 * @type: queue type as RX or TX 6537 * @napi: NAPI context, pass NULL to clear previously set NAPI 6538 * 6539 * Set queue with its corresponding napi context. This should be done after 6540 * registering the NAPI handler for the queue-vector and the queues have been 6541 * mapped to the corresponding interrupt vector. 6542 */ 6543 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6544 enum netdev_queue_type type, struct napi_struct *napi) 6545 { 6546 struct netdev_rx_queue *rxq; 6547 struct netdev_queue *txq; 6548 6549 if (WARN_ON_ONCE(napi && !napi->dev)) 6550 return; 6551 if (dev->reg_state >= NETREG_REGISTERED) 6552 ASSERT_RTNL(); 6553 6554 switch (type) { 6555 case NETDEV_QUEUE_TYPE_RX: 6556 rxq = __netif_get_rx_queue(dev, queue_index); 6557 rxq->napi = napi; 6558 return; 6559 case NETDEV_QUEUE_TYPE_TX: 6560 txq = netdev_get_tx_queue(dev, queue_index); 6561 txq->napi = napi; 6562 return; 6563 default: 6564 return; 6565 } 6566 } 6567 EXPORT_SYMBOL(netif_queue_set_napi); 6568 6569 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6570 int (*poll)(struct napi_struct *, int), int weight) 6571 { 6572 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6573 return; 6574 6575 INIT_LIST_HEAD(&napi->poll_list); 6576 INIT_HLIST_NODE(&napi->napi_hash_node); 6577 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6578 napi->timer.function = napi_watchdog; 6579 init_gro_hash(napi); 6580 napi->skb = NULL; 6581 INIT_LIST_HEAD(&napi->rx_list); 6582 napi->rx_count = 0; 6583 napi->poll = poll; 6584 if (weight > NAPI_POLL_WEIGHT) 6585 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6586 weight); 6587 napi->weight = weight; 6588 napi->dev = dev; 6589 #ifdef CONFIG_NETPOLL 6590 napi->poll_owner = -1; 6591 #endif 6592 napi->list_owner = -1; 6593 set_bit(NAPI_STATE_SCHED, &napi->state); 6594 set_bit(NAPI_STATE_NPSVC, &napi->state); 6595 list_add_rcu(&napi->dev_list, &dev->napi_list); 6596 napi_hash_add(napi); 6597 napi_get_frags_check(napi); 6598 /* Create kthread for this napi if dev->threaded is set. 6599 * Clear dev->threaded if kthread creation failed so that 6600 * threaded mode will not be enabled in napi_enable(). 6601 */ 6602 if (dev->threaded && napi_kthread_create(napi)) 6603 dev->threaded = 0; 6604 netif_napi_set_irq(napi, -1); 6605 } 6606 EXPORT_SYMBOL(netif_napi_add_weight); 6607 6608 void napi_disable(struct napi_struct *n) 6609 { 6610 unsigned long val, new; 6611 6612 might_sleep(); 6613 set_bit(NAPI_STATE_DISABLE, &n->state); 6614 6615 val = READ_ONCE(n->state); 6616 do { 6617 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6618 usleep_range(20, 200); 6619 val = READ_ONCE(n->state); 6620 } 6621 6622 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6623 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6624 } while (!try_cmpxchg(&n->state, &val, new)); 6625 6626 hrtimer_cancel(&n->timer); 6627 6628 clear_bit(NAPI_STATE_DISABLE, &n->state); 6629 } 6630 EXPORT_SYMBOL(napi_disable); 6631 6632 /** 6633 * napi_enable - enable NAPI scheduling 6634 * @n: NAPI context 6635 * 6636 * Resume NAPI from being scheduled on this context. 6637 * Must be paired with napi_disable. 6638 */ 6639 void napi_enable(struct napi_struct *n) 6640 { 6641 unsigned long new, val = READ_ONCE(n->state); 6642 6643 do { 6644 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6645 6646 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6647 if (n->dev->threaded && n->thread) 6648 new |= NAPIF_STATE_THREADED; 6649 } while (!try_cmpxchg(&n->state, &val, new)); 6650 } 6651 EXPORT_SYMBOL(napi_enable); 6652 6653 static void flush_gro_hash(struct napi_struct *napi) 6654 { 6655 int i; 6656 6657 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6658 struct sk_buff *skb, *n; 6659 6660 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6661 kfree_skb(skb); 6662 napi->gro_hash[i].count = 0; 6663 } 6664 } 6665 6666 /* Must be called in process context */ 6667 void __netif_napi_del(struct napi_struct *napi) 6668 { 6669 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6670 return; 6671 6672 napi_hash_del(napi); 6673 list_del_rcu(&napi->dev_list); 6674 napi_free_frags(napi); 6675 6676 flush_gro_hash(napi); 6677 napi->gro_bitmask = 0; 6678 6679 if (napi->thread) { 6680 kthread_stop(napi->thread); 6681 napi->thread = NULL; 6682 } 6683 } 6684 EXPORT_SYMBOL(__netif_napi_del); 6685 6686 static int __napi_poll(struct napi_struct *n, bool *repoll) 6687 { 6688 int work, weight; 6689 6690 weight = n->weight; 6691 6692 /* This NAPI_STATE_SCHED test is for avoiding a race 6693 * with netpoll's poll_napi(). Only the entity which 6694 * obtains the lock and sees NAPI_STATE_SCHED set will 6695 * actually make the ->poll() call. Therefore we avoid 6696 * accidentally calling ->poll() when NAPI is not scheduled. 6697 */ 6698 work = 0; 6699 if (napi_is_scheduled(n)) { 6700 work = n->poll(n, weight); 6701 trace_napi_poll(n, work, weight); 6702 6703 xdp_do_check_flushed(n); 6704 } 6705 6706 if (unlikely(work > weight)) 6707 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6708 n->poll, work, weight); 6709 6710 if (likely(work < weight)) 6711 return work; 6712 6713 /* Drivers must not modify the NAPI state if they 6714 * consume the entire weight. In such cases this code 6715 * still "owns" the NAPI instance and therefore can 6716 * move the instance around on the list at-will. 6717 */ 6718 if (unlikely(napi_disable_pending(n))) { 6719 napi_complete(n); 6720 return work; 6721 } 6722 6723 /* The NAPI context has more processing work, but busy-polling 6724 * is preferred. Exit early. 6725 */ 6726 if (napi_prefer_busy_poll(n)) { 6727 if (napi_complete_done(n, work)) { 6728 /* If timeout is not set, we need to make sure 6729 * that the NAPI is re-scheduled. 6730 */ 6731 napi_schedule(n); 6732 } 6733 return work; 6734 } 6735 6736 if (n->gro_bitmask) { 6737 /* flush too old packets 6738 * If HZ < 1000, flush all packets. 6739 */ 6740 napi_gro_flush(n, HZ >= 1000); 6741 } 6742 6743 gro_normal_list(n); 6744 6745 /* Some drivers may have called napi_schedule 6746 * prior to exhausting their budget. 6747 */ 6748 if (unlikely(!list_empty(&n->poll_list))) { 6749 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6750 n->dev ? n->dev->name : "backlog"); 6751 return work; 6752 } 6753 6754 *repoll = true; 6755 6756 return work; 6757 } 6758 6759 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6760 { 6761 bool do_repoll = false; 6762 void *have; 6763 int work; 6764 6765 list_del_init(&n->poll_list); 6766 6767 have = netpoll_poll_lock(n); 6768 6769 work = __napi_poll(n, &do_repoll); 6770 6771 if (do_repoll) 6772 list_add_tail(&n->poll_list, repoll); 6773 6774 netpoll_poll_unlock(have); 6775 6776 return work; 6777 } 6778 6779 static int napi_thread_wait(struct napi_struct *napi) 6780 { 6781 set_current_state(TASK_INTERRUPTIBLE); 6782 6783 while (!kthread_should_stop()) { 6784 /* Testing SCHED_THREADED bit here to make sure the current 6785 * kthread owns this napi and could poll on this napi. 6786 * Testing SCHED bit is not enough because SCHED bit might be 6787 * set by some other busy poll thread or by napi_disable(). 6788 */ 6789 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) { 6790 WARN_ON(!list_empty(&napi->poll_list)); 6791 __set_current_state(TASK_RUNNING); 6792 return 0; 6793 } 6794 6795 schedule(); 6796 set_current_state(TASK_INTERRUPTIBLE); 6797 } 6798 __set_current_state(TASK_RUNNING); 6799 6800 return -1; 6801 } 6802 6803 static void napi_threaded_poll_loop(struct napi_struct *napi) 6804 { 6805 struct softnet_data *sd; 6806 unsigned long last_qs = jiffies; 6807 6808 for (;;) { 6809 bool repoll = false; 6810 void *have; 6811 6812 local_bh_disable(); 6813 sd = this_cpu_ptr(&softnet_data); 6814 sd->in_napi_threaded_poll = true; 6815 6816 have = netpoll_poll_lock(napi); 6817 __napi_poll(napi, &repoll); 6818 netpoll_poll_unlock(have); 6819 6820 sd->in_napi_threaded_poll = false; 6821 barrier(); 6822 6823 if (sd_has_rps_ipi_waiting(sd)) { 6824 local_irq_disable(); 6825 net_rps_action_and_irq_enable(sd); 6826 } 6827 skb_defer_free_flush(sd); 6828 local_bh_enable(); 6829 6830 if (!repoll) 6831 break; 6832 6833 rcu_softirq_qs_periodic(last_qs); 6834 cond_resched(); 6835 } 6836 } 6837 6838 static int napi_threaded_poll(void *data) 6839 { 6840 struct napi_struct *napi = data; 6841 6842 while (!napi_thread_wait(napi)) 6843 napi_threaded_poll_loop(napi); 6844 6845 return 0; 6846 } 6847 6848 static __latent_entropy void net_rx_action(struct softirq_action *h) 6849 { 6850 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6851 unsigned long time_limit = jiffies + 6852 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs)); 6853 int budget = READ_ONCE(net_hotdata.netdev_budget); 6854 LIST_HEAD(list); 6855 LIST_HEAD(repoll); 6856 6857 start: 6858 sd->in_net_rx_action = true; 6859 local_irq_disable(); 6860 list_splice_init(&sd->poll_list, &list); 6861 local_irq_enable(); 6862 6863 for (;;) { 6864 struct napi_struct *n; 6865 6866 skb_defer_free_flush(sd); 6867 6868 if (list_empty(&list)) { 6869 if (list_empty(&repoll)) { 6870 sd->in_net_rx_action = false; 6871 barrier(); 6872 /* We need to check if ____napi_schedule() 6873 * had refilled poll_list while 6874 * sd->in_net_rx_action was true. 6875 */ 6876 if (!list_empty(&sd->poll_list)) 6877 goto start; 6878 if (!sd_has_rps_ipi_waiting(sd)) 6879 goto end; 6880 } 6881 break; 6882 } 6883 6884 n = list_first_entry(&list, struct napi_struct, poll_list); 6885 budget -= napi_poll(n, &repoll); 6886 6887 /* If softirq window is exhausted then punt. 6888 * Allow this to run for 2 jiffies since which will allow 6889 * an average latency of 1.5/HZ. 6890 */ 6891 if (unlikely(budget <= 0 || 6892 time_after_eq(jiffies, time_limit))) { 6893 sd->time_squeeze++; 6894 break; 6895 } 6896 } 6897 6898 local_irq_disable(); 6899 6900 list_splice_tail_init(&sd->poll_list, &list); 6901 list_splice_tail(&repoll, &list); 6902 list_splice(&list, &sd->poll_list); 6903 if (!list_empty(&sd->poll_list)) 6904 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6905 else 6906 sd->in_net_rx_action = false; 6907 6908 net_rps_action_and_irq_enable(sd); 6909 end:; 6910 } 6911 6912 struct netdev_adjacent { 6913 struct net_device *dev; 6914 netdevice_tracker dev_tracker; 6915 6916 /* upper master flag, there can only be one master device per list */ 6917 bool master; 6918 6919 /* lookup ignore flag */ 6920 bool ignore; 6921 6922 /* counter for the number of times this device was added to us */ 6923 u16 ref_nr; 6924 6925 /* private field for the users */ 6926 void *private; 6927 6928 struct list_head list; 6929 struct rcu_head rcu; 6930 }; 6931 6932 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6933 struct list_head *adj_list) 6934 { 6935 struct netdev_adjacent *adj; 6936 6937 list_for_each_entry(adj, adj_list, list) { 6938 if (adj->dev == adj_dev) 6939 return adj; 6940 } 6941 return NULL; 6942 } 6943 6944 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6945 struct netdev_nested_priv *priv) 6946 { 6947 struct net_device *dev = (struct net_device *)priv->data; 6948 6949 return upper_dev == dev; 6950 } 6951 6952 /** 6953 * netdev_has_upper_dev - Check if device is linked to an upper device 6954 * @dev: device 6955 * @upper_dev: upper device to check 6956 * 6957 * Find out if a device is linked to specified upper device and return true 6958 * in case it is. Note that this checks only immediate upper device, 6959 * not through a complete stack of devices. The caller must hold the RTNL lock. 6960 */ 6961 bool netdev_has_upper_dev(struct net_device *dev, 6962 struct net_device *upper_dev) 6963 { 6964 struct netdev_nested_priv priv = { 6965 .data = (void *)upper_dev, 6966 }; 6967 6968 ASSERT_RTNL(); 6969 6970 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6971 &priv); 6972 } 6973 EXPORT_SYMBOL(netdev_has_upper_dev); 6974 6975 /** 6976 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6977 * @dev: device 6978 * @upper_dev: upper device to check 6979 * 6980 * Find out if a device is linked to specified upper device and return true 6981 * in case it is. Note that this checks the entire upper device chain. 6982 * The caller must hold rcu lock. 6983 */ 6984 6985 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6986 struct net_device *upper_dev) 6987 { 6988 struct netdev_nested_priv priv = { 6989 .data = (void *)upper_dev, 6990 }; 6991 6992 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6993 &priv); 6994 } 6995 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6996 6997 /** 6998 * netdev_has_any_upper_dev - Check if device is linked to some device 6999 * @dev: device 7000 * 7001 * Find out if a device is linked to an upper device and return true in case 7002 * it is. The caller must hold the RTNL lock. 7003 */ 7004 bool netdev_has_any_upper_dev(struct net_device *dev) 7005 { 7006 ASSERT_RTNL(); 7007 7008 return !list_empty(&dev->adj_list.upper); 7009 } 7010 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7011 7012 /** 7013 * netdev_master_upper_dev_get - Get master upper device 7014 * @dev: device 7015 * 7016 * Find a master upper device and return pointer to it or NULL in case 7017 * it's not there. The caller must hold the RTNL lock. 7018 */ 7019 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7020 { 7021 struct netdev_adjacent *upper; 7022 7023 ASSERT_RTNL(); 7024 7025 if (list_empty(&dev->adj_list.upper)) 7026 return NULL; 7027 7028 upper = list_first_entry(&dev->adj_list.upper, 7029 struct netdev_adjacent, list); 7030 if (likely(upper->master)) 7031 return upper->dev; 7032 return NULL; 7033 } 7034 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7035 7036 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7037 { 7038 struct netdev_adjacent *upper; 7039 7040 ASSERT_RTNL(); 7041 7042 if (list_empty(&dev->adj_list.upper)) 7043 return NULL; 7044 7045 upper = list_first_entry(&dev->adj_list.upper, 7046 struct netdev_adjacent, list); 7047 if (likely(upper->master) && !upper->ignore) 7048 return upper->dev; 7049 return NULL; 7050 } 7051 7052 /** 7053 * netdev_has_any_lower_dev - Check if device is linked to some device 7054 * @dev: device 7055 * 7056 * Find out if a device is linked to a lower device and return true in case 7057 * it is. The caller must hold the RTNL lock. 7058 */ 7059 static bool netdev_has_any_lower_dev(struct net_device *dev) 7060 { 7061 ASSERT_RTNL(); 7062 7063 return !list_empty(&dev->adj_list.lower); 7064 } 7065 7066 void *netdev_adjacent_get_private(struct list_head *adj_list) 7067 { 7068 struct netdev_adjacent *adj; 7069 7070 adj = list_entry(adj_list, struct netdev_adjacent, list); 7071 7072 return adj->private; 7073 } 7074 EXPORT_SYMBOL(netdev_adjacent_get_private); 7075 7076 /** 7077 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7078 * @dev: device 7079 * @iter: list_head ** of the current position 7080 * 7081 * Gets the next device from the dev's upper list, starting from iter 7082 * position. The caller must hold RCU read lock. 7083 */ 7084 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7085 struct list_head **iter) 7086 { 7087 struct netdev_adjacent *upper; 7088 7089 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7090 7091 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7092 7093 if (&upper->list == &dev->adj_list.upper) 7094 return NULL; 7095 7096 *iter = &upper->list; 7097 7098 return upper->dev; 7099 } 7100 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7101 7102 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7103 struct list_head **iter, 7104 bool *ignore) 7105 { 7106 struct netdev_adjacent *upper; 7107 7108 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7109 7110 if (&upper->list == &dev->adj_list.upper) 7111 return NULL; 7112 7113 *iter = &upper->list; 7114 *ignore = upper->ignore; 7115 7116 return upper->dev; 7117 } 7118 7119 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7120 struct list_head **iter) 7121 { 7122 struct netdev_adjacent *upper; 7123 7124 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7125 7126 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7127 7128 if (&upper->list == &dev->adj_list.upper) 7129 return NULL; 7130 7131 *iter = &upper->list; 7132 7133 return upper->dev; 7134 } 7135 7136 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7137 int (*fn)(struct net_device *dev, 7138 struct netdev_nested_priv *priv), 7139 struct netdev_nested_priv *priv) 7140 { 7141 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7142 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7143 int ret, cur = 0; 7144 bool ignore; 7145 7146 now = dev; 7147 iter = &dev->adj_list.upper; 7148 7149 while (1) { 7150 if (now != dev) { 7151 ret = fn(now, priv); 7152 if (ret) 7153 return ret; 7154 } 7155 7156 next = NULL; 7157 while (1) { 7158 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7159 if (!udev) 7160 break; 7161 if (ignore) 7162 continue; 7163 7164 next = udev; 7165 niter = &udev->adj_list.upper; 7166 dev_stack[cur] = now; 7167 iter_stack[cur++] = iter; 7168 break; 7169 } 7170 7171 if (!next) { 7172 if (!cur) 7173 return 0; 7174 next = dev_stack[--cur]; 7175 niter = iter_stack[cur]; 7176 } 7177 7178 now = next; 7179 iter = niter; 7180 } 7181 7182 return 0; 7183 } 7184 7185 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7186 int (*fn)(struct net_device *dev, 7187 struct netdev_nested_priv *priv), 7188 struct netdev_nested_priv *priv) 7189 { 7190 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7191 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7192 int ret, cur = 0; 7193 7194 now = dev; 7195 iter = &dev->adj_list.upper; 7196 7197 while (1) { 7198 if (now != dev) { 7199 ret = fn(now, priv); 7200 if (ret) 7201 return ret; 7202 } 7203 7204 next = NULL; 7205 while (1) { 7206 udev = netdev_next_upper_dev_rcu(now, &iter); 7207 if (!udev) 7208 break; 7209 7210 next = udev; 7211 niter = &udev->adj_list.upper; 7212 dev_stack[cur] = now; 7213 iter_stack[cur++] = iter; 7214 break; 7215 } 7216 7217 if (!next) { 7218 if (!cur) 7219 return 0; 7220 next = dev_stack[--cur]; 7221 niter = iter_stack[cur]; 7222 } 7223 7224 now = next; 7225 iter = niter; 7226 } 7227 7228 return 0; 7229 } 7230 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7231 7232 static bool __netdev_has_upper_dev(struct net_device *dev, 7233 struct net_device *upper_dev) 7234 { 7235 struct netdev_nested_priv priv = { 7236 .flags = 0, 7237 .data = (void *)upper_dev, 7238 }; 7239 7240 ASSERT_RTNL(); 7241 7242 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7243 &priv); 7244 } 7245 7246 /** 7247 * netdev_lower_get_next_private - Get the next ->private from the 7248 * lower neighbour list 7249 * @dev: device 7250 * @iter: list_head ** of the current position 7251 * 7252 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7253 * list, starting from iter position. The caller must hold either hold the 7254 * RTNL lock or its own locking that guarantees that the neighbour lower 7255 * list will remain unchanged. 7256 */ 7257 void *netdev_lower_get_next_private(struct net_device *dev, 7258 struct list_head **iter) 7259 { 7260 struct netdev_adjacent *lower; 7261 7262 lower = list_entry(*iter, struct netdev_adjacent, list); 7263 7264 if (&lower->list == &dev->adj_list.lower) 7265 return NULL; 7266 7267 *iter = lower->list.next; 7268 7269 return lower->private; 7270 } 7271 EXPORT_SYMBOL(netdev_lower_get_next_private); 7272 7273 /** 7274 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7275 * lower neighbour list, RCU 7276 * variant 7277 * @dev: device 7278 * @iter: list_head ** of the current position 7279 * 7280 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7281 * list, starting from iter position. The caller must hold RCU read lock. 7282 */ 7283 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7284 struct list_head **iter) 7285 { 7286 struct netdev_adjacent *lower; 7287 7288 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7289 7290 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7291 7292 if (&lower->list == &dev->adj_list.lower) 7293 return NULL; 7294 7295 *iter = &lower->list; 7296 7297 return lower->private; 7298 } 7299 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7300 7301 /** 7302 * netdev_lower_get_next - Get the next device from the lower neighbour 7303 * list 7304 * @dev: device 7305 * @iter: list_head ** of the current position 7306 * 7307 * Gets the next netdev_adjacent from the dev's lower neighbour 7308 * list, starting from iter position. The caller must hold RTNL lock or 7309 * its own locking that guarantees that the neighbour lower 7310 * list will remain unchanged. 7311 */ 7312 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7313 { 7314 struct netdev_adjacent *lower; 7315 7316 lower = list_entry(*iter, struct netdev_adjacent, list); 7317 7318 if (&lower->list == &dev->adj_list.lower) 7319 return NULL; 7320 7321 *iter = lower->list.next; 7322 7323 return lower->dev; 7324 } 7325 EXPORT_SYMBOL(netdev_lower_get_next); 7326 7327 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7328 struct list_head **iter) 7329 { 7330 struct netdev_adjacent *lower; 7331 7332 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7333 7334 if (&lower->list == &dev->adj_list.lower) 7335 return NULL; 7336 7337 *iter = &lower->list; 7338 7339 return lower->dev; 7340 } 7341 7342 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7343 struct list_head **iter, 7344 bool *ignore) 7345 { 7346 struct netdev_adjacent *lower; 7347 7348 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7349 7350 if (&lower->list == &dev->adj_list.lower) 7351 return NULL; 7352 7353 *iter = &lower->list; 7354 *ignore = lower->ignore; 7355 7356 return lower->dev; 7357 } 7358 7359 int netdev_walk_all_lower_dev(struct net_device *dev, 7360 int (*fn)(struct net_device *dev, 7361 struct netdev_nested_priv *priv), 7362 struct netdev_nested_priv *priv) 7363 { 7364 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7365 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7366 int ret, cur = 0; 7367 7368 now = dev; 7369 iter = &dev->adj_list.lower; 7370 7371 while (1) { 7372 if (now != dev) { 7373 ret = fn(now, priv); 7374 if (ret) 7375 return ret; 7376 } 7377 7378 next = NULL; 7379 while (1) { 7380 ldev = netdev_next_lower_dev(now, &iter); 7381 if (!ldev) 7382 break; 7383 7384 next = ldev; 7385 niter = &ldev->adj_list.lower; 7386 dev_stack[cur] = now; 7387 iter_stack[cur++] = iter; 7388 break; 7389 } 7390 7391 if (!next) { 7392 if (!cur) 7393 return 0; 7394 next = dev_stack[--cur]; 7395 niter = iter_stack[cur]; 7396 } 7397 7398 now = next; 7399 iter = niter; 7400 } 7401 7402 return 0; 7403 } 7404 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7405 7406 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7407 int (*fn)(struct net_device *dev, 7408 struct netdev_nested_priv *priv), 7409 struct netdev_nested_priv *priv) 7410 { 7411 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7412 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7413 int ret, cur = 0; 7414 bool ignore; 7415 7416 now = dev; 7417 iter = &dev->adj_list.lower; 7418 7419 while (1) { 7420 if (now != dev) { 7421 ret = fn(now, priv); 7422 if (ret) 7423 return ret; 7424 } 7425 7426 next = NULL; 7427 while (1) { 7428 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7429 if (!ldev) 7430 break; 7431 if (ignore) 7432 continue; 7433 7434 next = ldev; 7435 niter = &ldev->adj_list.lower; 7436 dev_stack[cur] = now; 7437 iter_stack[cur++] = iter; 7438 break; 7439 } 7440 7441 if (!next) { 7442 if (!cur) 7443 return 0; 7444 next = dev_stack[--cur]; 7445 niter = iter_stack[cur]; 7446 } 7447 7448 now = next; 7449 iter = niter; 7450 } 7451 7452 return 0; 7453 } 7454 7455 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7456 struct list_head **iter) 7457 { 7458 struct netdev_adjacent *lower; 7459 7460 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7461 if (&lower->list == &dev->adj_list.lower) 7462 return NULL; 7463 7464 *iter = &lower->list; 7465 7466 return lower->dev; 7467 } 7468 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7469 7470 static u8 __netdev_upper_depth(struct net_device *dev) 7471 { 7472 struct net_device *udev; 7473 struct list_head *iter; 7474 u8 max_depth = 0; 7475 bool ignore; 7476 7477 for (iter = &dev->adj_list.upper, 7478 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7479 udev; 7480 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7481 if (ignore) 7482 continue; 7483 if (max_depth < udev->upper_level) 7484 max_depth = udev->upper_level; 7485 } 7486 7487 return max_depth; 7488 } 7489 7490 static u8 __netdev_lower_depth(struct net_device *dev) 7491 { 7492 struct net_device *ldev; 7493 struct list_head *iter; 7494 u8 max_depth = 0; 7495 bool ignore; 7496 7497 for (iter = &dev->adj_list.lower, 7498 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7499 ldev; 7500 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7501 if (ignore) 7502 continue; 7503 if (max_depth < ldev->lower_level) 7504 max_depth = ldev->lower_level; 7505 } 7506 7507 return max_depth; 7508 } 7509 7510 static int __netdev_update_upper_level(struct net_device *dev, 7511 struct netdev_nested_priv *__unused) 7512 { 7513 dev->upper_level = __netdev_upper_depth(dev) + 1; 7514 return 0; 7515 } 7516 7517 #ifdef CONFIG_LOCKDEP 7518 static LIST_HEAD(net_unlink_list); 7519 7520 static void net_unlink_todo(struct net_device *dev) 7521 { 7522 if (list_empty(&dev->unlink_list)) 7523 list_add_tail(&dev->unlink_list, &net_unlink_list); 7524 } 7525 #endif 7526 7527 static int __netdev_update_lower_level(struct net_device *dev, 7528 struct netdev_nested_priv *priv) 7529 { 7530 dev->lower_level = __netdev_lower_depth(dev) + 1; 7531 7532 #ifdef CONFIG_LOCKDEP 7533 if (!priv) 7534 return 0; 7535 7536 if (priv->flags & NESTED_SYNC_IMM) 7537 dev->nested_level = dev->lower_level - 1; 7538 if (priv->flags & NESTED_SYNC_TODO) 7539 net_unlink_todo(dev); 7540 #endif 7541 return 0; 7542 } 7543 7544 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7545 int (*fn)(struct net_device *dev, 7546 struct netdev_nested_priv *priv), 7547 struct netdev_nested_priv *priv) 7548 { 7549 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7550 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7551 int ret, cur = 0; 7552 7553 now = dev; 7554 iter = &dev->adj_list.lower; 7555 7556 while (1) { 7557 if (now != dev) { 7558 ret = fn(now, priv); 7559 if (ret) 7560 return ret; 7561 } 7562 7563 next = NULL; 7564 while (1) { 7565 ldev = netdev_next_lower_dev_rcu(now, &iter); 7566 if (!ldev) 7567 break; 7568 7569 next = ldev; 7570 niter = &ldev->adj_list.lower; 7571 dev_stack[cur] = now; 7572 iter_stack[cur++] = iter; 7573 break; 7574 } 7575 7576 if (!next) { 7577 if (!cur) 7578 return 0; 7579 next = dev_stack[--cur]; 7580 niter = iter_stack[cur]; 7581 } 7582 7583 now = next; 7584 iter = niter; 7585 } 7586 7587 return 0; 7588 } 7589 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7590 7591 /** 7592 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7593 * lower neighbour list, RCU 7594 * variant 7595 * @dev: device 7596 * 7597 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7598 * list. The caller must hold RCU read lock. 7599 */ 7600 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7601 { 7602 struct netdev_adjacent *lower; 7603 7604 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7605 struct netdev_adjacent, list); 7606 if (lower) 7607 return lower->private; 7608 return NULL; 7609 } 7610 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7611 7612 /** 7613 * netdev_master_upper_dev_get_rcu - Get master upper device 7614 * @dev: device 7615 * 7616 * Find a master upper device and return pointer to it or NULL in case 7617 * it's not there. The caller must hold the RCU read lock. 7618 */ 7619 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7620 { 7621 struct netdev_adjacent *upper; 7622 7623 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7624 struct netdev_adjacent, list); 7625 if (upper && likely(upper->master)) 7626 return upper->dev; 7627 return NULL; 7628 } 7629 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7630 7631 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7632 struct net_device *adj_dev, 7633 struct list_head *dev_list) 7634 { 7635 char linkname[IFNAMSIZ+7]; 7636 7637 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7638 "upper_%s" : "lower_%s", adj_dev->name); 7639 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7640 linkname); 7641 } 7642 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7643 char *name, 7644 struct list_head *dev_list) 7645 { 7646 char linkname[IFNAMSIZ+7]; 7647 7648 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7649 "upper_%s" : "lower_%s", name); 7650 sysfs_remove_link(&(dev->dev.kobj), linkname); 7651 } 7652 7653 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7654 struct net_device *adj_dev, 7655 struct list_head *dev_list) 7656 { 7657 return (dev_list == &dev->adj_list.upper || 7658 dev_list == &dev->adj_list.lower) && 7659 net_eq(dev_net(dev), dev_net(adj_dev)); 7660 } 7661 7662 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7663 struct net_device *adj_dev, 7664 struct list_head *dev_list, 7665 void *private, bool master) 7666 { 7667 struct netdev_adjacent *adj; 7668 int ret; 7669 7670 adj = __netdev_find_adj(adj_dev, dev_list); 7671 7672 if (adj) { 7673 adj->ref_nr += 1; 7674 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7675 dev->name, adj_dev->name, adj->ref_nr); 7676 7677 return 0; 7678 } 7679 7680 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7681 if (!adj) 7682 return -ENOMEM; 7683 7684 adj->dev = adj_dev; 7685 adj->master = master; 7686 adj->ref_nr = 1; 7687 adj->private = private; 7688 adj->ignore = false; 7689 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7690 7691 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7692 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7693 7694 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7695 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7696 if (ret) 7697 goto free_adj; 7698 } 7699 7700 /* Ensure that master link is always the first item in list. */ 7701 if (master) { 7702 ret = sysfs_create_link(&(dev->dev.kobj), 7703 &(adj_dev->dev.kobj), "master"); 7704 if (ret) 7705 goto remove_symlinks; 7706 7707 list_add_rcu(&adj->list, dev_list); 7708 } else { 7709 list_add_tail_rcu(&adj->list, dev_list); 7710 } 7711 7712 return 0; 7713 7714 remove_symlinks: 7715 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7716 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7717 free_adj: 7718 netdev_put(adj_dev, &adj->dev_tracker); 7719 kfree(adj); 7720 7721 return ret; 7722 } 7723 7724 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7725 struct net_device *adj_dev, 7726 u16 ref_nr, 7727 struct list_head *dev_list) 7728 { 7729 struct netdev_adjacent *adj; 7730 7731 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7732 dev->name, adj_dev->name, ref_nr); 7733 7734 adj = __netdev_find_adj(adj_dev, dev_list); 7735 7736 if (!adj) { 7737 pr_err("Adjacency does not exist for device %s from %s\n", 7738 dev->name, adj_dev->name); 7739 WARN_ON(1); 7740 return; 7741 } 7742 7743 if (adj->ref_nr > ref_nr) { 7744 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7745 dev->name, adj_dev->name, ref_nr, 7746 adj->ref_nr - ref_nr); 7747 adj->ref_nr -= ref_nr; 7748 return; 7749 } 7750 7751 if (adj->master) 7752 sysfs_remove_link(&(dev->dev.kobj), "master"); 7753 7754 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7755 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7756 7757 list_del_rcu(&adj->list); 7758 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7759 adj_dev->name, dev->name, adj_dev->name); 7760 netdev_put(adj_dev, &adj->dev_tracker); 7761 kfree_rcu(adj, rcu); 7762 } 7763 7764 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7765 struct net_device *upper_dev, 7766 struct list_head *up_list, 7767 struct list_head *down_list, 7768 void *private, bool master) 7769 { 7770 int ret; 7771 7772 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7773 private, master); 7774 if (ret) 7775 return ret; 7776 7777 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7778 private, false); 7779 if (ret) { 7780 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7781 return ret; 7782 } 7783 7784 return 0; 7785 } 7786 7787 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7788 struct net_device *upper_dev, 7789 u16 ref_nr, 7790 struct list_head *up_list, 7791 struct list_head *down_list) 7792 { 7793 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7794 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7795 } 7796 7797 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7798 struct net_device *upper_dev, 7799 void *private, bool master) 7800 { 7801 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7802 &dev->adj_list.upper, 7803 &upper_dev->adj_list.lower, 7804 private, master); 7805 } 7806 7807 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7808 struct net_device *upper_dev) 7809 { 7810 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7811 &dev->adj_list.upper, 7812 &upper_dev->adj_list.lower); 7813 } 7814 7815 static int __netdev_upper_dev_link(struct net_device *dev, 7816 struct net_device *upper_dev, bool master, 7817 void *upper_priv, void *upper_info, 7818 struct netdev_nested_priv *priv, 7819 struct netlink_ext_ack *extack) 7820 { 7821 struct netdev_notifier_changeupper_info changeupper_info = { 7822 .info = { 7823 .dev = dev, 7824 .extack = extack, 7825 }, 7826 .upper_dev = upper_dev, 7827 .master = master, 7828 .linking = true, 7829 .upper_info = upper_info, 7830 }; 7831 struct net_device *master_dev; 7832 int ret = 0; 7833 7834 ASSERT_RTNL(); 7835 7836 if (dev == upper_dev) 7837 return -EBUSY; 7838 7839 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7840 if (__netdev_has_upper_dev(upper_dev, dev)) 7841 return -EBUSY; 7842 7843 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7844 return -EMLINK; 7845 7846 if (!master) { 7847 if (__netdev_has_upper_dev(dev, upper_dev)) 7848 return -EEXIST; 7849 } else { 7850 master_dev = __netdev_master_upper_dev_get(dev); 7851 if (master_dev) 7852 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7853 } 7854 7855 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7856 &changeupper_info.info); 7857 ret = notifier_to_errno(ret); 7858 if (ret) 7859 return ret; 7860 7861 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7862 master); 7863 if (ret) 7864 return ret; 7865 7866 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7867 &changeupper_info.info); 7868 ret = notifier_to_errno(ret); 7869 if (ret) 7870 goto rollback; 7871 7872 __netdev_update_upper_level(dev, NULL); 7873 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7874 7875 __netdev_update_lower_level(upper_dev, priv); 7876 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7877 priv); 7878 7879 return 0; 7880 7881 rollback: 7882 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7883 7884 return ret; 7885 } 7886 7887 /** 7888 * netdev_upper_dev_link - Add a link to the upper device 7889 * @dev: device 7890 * @upper_dev: new upper device 7891 * @extack: netlink extended ack 7892 * 7893 * Adds a link to device which is upper to this one. The caller must hold 7894 * the RTNL lock. On a failure a negative errno code is returned. 7895 * On success the reference counts are adjusted and the function 7896 * returns zero. 7897 */ 7898 int netdev_upper_dev_link(struct net_device *dev, 7899 struct net_device *upper_dev, 7900 struct netlink_ext_ack *extack) 7901 { 7902 struct netdev_nested_priv priv = { 7903 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7904 .data = NULL, 7905 }; 7906 7907 return __netdev_upper_dev_link(dev, upper_dev, false, 7908 NULL, NULL, &priv, extack); 7909 } 7910 EXPORT_SYMBOL(netdev_upper_dev_link); 7911 7912 /** 7913 * netdev_master_upper_dev_link - Add a master link to the upper device 7914 * @dev: device 7915 * @upper_dev: new upper device 7916 * @upper_priv: upper device private 7917 * @upper_info: upper info to be passed down via notifier 7918 * @extack: netlink extended ack 7919 * 7920 * Adds a link to device which is upper to this one. In this case, only 7921 * one master upper device can be linked, although other non-master devices 7922 * might be linked as well. The caller must hold the RTNL lock. 7923 * On a failure a negative errno code is returned. On success the reference 7924 * counts are adjusted and the function returns zero. 7925 */ 7926 int netdev_master_upper_dev_link(struct net_device *dev, 7927 struct net_device *upper_dev, 7928 void *upper_priv, void *upper_info, 7929 struct netlink_ext_ack *extack) 7930 { 7931 struct netdev_nested_priv priv = { 7932 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7933 .data = NULL, 7934 }; 7935 7936 return __netdev_upper_dev_link(dev, upper_dev, true, 7937 upper_priv, upper_info, &priv, extack); 7938 } 7939 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7940 7941 static void __netdev_upper_dev_unlink(struct net_device *dev, 7942 struct net_device *upper_dev, 7943 struct netdev_nested_priv *priv) 7944 { 7945 struct netdev_notifier_changeupper_info changeupper_info = { 7946 .info = { 7947 .dev = dev, 7948 }, 7949 .upper_dev = upper_dev, 7950 .linking = false, 7951 }; 7952 7953 ASSERT_RTNL(); 7954 7955 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7956 7957 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7958 &changeupper_info.info); 7959 7960 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7961 7962 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7963 &changeupper_info.info); 7964 7965 __netdev_update_upper_level(dev, NULL); 7966 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7967 7968 __netdev_update_lower_level(upper_dev, priv); 7969 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7970 priv); 7971 } 7972 7973 /** 7974 * netdev_upper_dev_unlink - Removes a link to upper device 7975 * @dev: device 7976 * @upper_dev: new upper device 7977 * 7978 * Removes a link to device which is upper to this one. The caller must hold 7979 * the RTNL lock. 7980 */ 7981 void netdev_upper_dev_unlink(struct net_device *dev, 7982 struct net_device *upper_dev) 7983 { 7984 struct netdev_nested_priv priv = { 7985 .flags = NESTED_SYNC_TODO, 7986 .data = NULL, 7987 }; 7988 7989 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7990 } 7991 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7992 7993 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7994 struct net_device *lower_dev, 7995 bool val) 7996 { 7997 struct netdev_adjacent *adj; 7998 7999 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8000 if (adj) 8001 adj->ignore = val; 8002 8003 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8004 if (adj) 8005 adj->ignore = val; 8006 } 8007 8008 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8009 struct net_device *lower_dev) 8010 { 8011 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8012 } 8013 8014 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8015 struct net_device *lower_dev) 8016 { 8017 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8018 } 8019 8020 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8021 struct net_device *new_dev, 8022 struct net_device *dev, 8023 struct netlink_ext_ack *extack) 8024 { 8025 struct netdev_nested_priv priv = { 8026 .flags = 0, 8027 .data = NULL, 8028 }; 8029 int err; 8030 8031 if (!new_dev) 8032 return 0; 8033 8034 if (old_dev && new_dev != old_dev) 8035 netdev_adjacent_dev_disable(dev, old_dev); 8036 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8037 extack); 8038 if (err) { 8039 if (old_dev && new_dev != old_dev) 8040 netdev_adjacent_dev_enable(dev, old_dev); 8041 return err; 8042 } 8043 8044 return 0; 8045 } 8046 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8047 8048 void netdev_adjacent_change_commit(struct net_device *old_dev, 8049 struct net_device *new_dev, 8050 struct net_device *dev) 8051 { 8052 struct netdev_nested_priv priv = { 8053 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8054 .data = NULL, 8055 }; 8056 8057 if (!new_dev || !old_dev) 8058 return; 8059 8060 if (new_dev == old_dev) 8061 return; 8062 8063 netdev_adjacent_dev_enable(dev, old_dev); 8064 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8065 } 8066 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8067 8068 void netdev_adjacent_change_abort(struct net_device *old_dev, 8069 struct net_device *new_dev, 8070 struct net_device *dev) 8071 { 8072 struct netdev_nested_priv priv = { 8073 .flags = 0, 8074 .data = NULL, 8075 }; 8076 8077 if (!new_dev) 8078 return; 8079 8080 if (old_dev && new_dev != old_dev) 8081 netdev_adjacent_dev_enable(dev, old_dev); 8082 8083 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8084 } 8085 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8086 8087 /** 8088 * netdev_bonding_info_change - Dispatch event about slave change 8089 * @dev: device 8090 * @bonding_info: info to dispatch 8091 * 8092 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8093 * The caller must hold the RTNL lock. 8094 */ 8095 void netdev_bonding_info_change(struct net_device *dev, 8096 struct netdev_bonding_info *bonding_info) 8097 { 8098 struct netdev_notifier_bonding_info info = { 8099 .info.dev = dev, 8100 }; 8101 8102 memcpy(&info.bonding_info, bonding_info, 8103 sizeof(struct netdev_bonding_info)); 8104 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8105 &info.info); 8106 } 8107 EXPORT_SYMBOL(netdev_bonding_info_change); 8108 8109 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8110 struct netlink_ext_ack *extack) 8111 { 8112 struct netdev_notifier_offload_xstats_info info = { 8113 .info.dev = dev, 8114 .info.extack = extack, 8115 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8116 }; 8117 int err; 8118 int rc; 8119 8120 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8121 GFP_KERNEL); 8122 if (!dev->offload_xstats_l3) 8123 return -ENOMEM; 8124 8125 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8126 NETDEV_OFFLOAD_XSTATS_DISABLE, 8127 &info.info); 8128 err = notifier_to_errno(rc); 8129 if (err) 8130 goto free_stats; 8131 8132 return 0; 8133 8134 free_stats: 8135 kfree(dev->offload_xstats_l3); 8136 dev->offload_xstats_l3 = NULL; 8137 return err; 8138 } 8139 8140 int netdev_offload_xstats_enable(struct net_device *dev, 8141 enum netdev_offload_xstats_type type, 8142 struct netlink_ext_ack *extack) 8143 { 8144 ASSERT_RTNL(); 8145 8146 if (netdev_offload_xstats_enabled(dev, type)) 8147 return -EALREADY; 8148 8149 switch (type) { 8150 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8151 return netdev_offload_xstats_enable_l3(dev, extack); 8152 } 8153 8154 WARN_ON(1); 8155 return -EINVAL; 8156 } 8157 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8158 8159 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8160 { 8161 struct netdev_notifier_offload_xstats_info info = { 8162 .info.dev = dev, 8163 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8164 }; 8165 8166 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8167 &info.info); 8168 kfree(dev->offload_xstats_l3); 8169 dev->offload_xstats_l3 = NULL; 8170 } 8171 8172 int netdev_offload_xstats_disable(struct net_device *dev, 8173 enum netdev_offload_xstats_type type) 8174 { 8175 ASSERT_RTNL(); 8176 8177 if (!netdev_offload_xstats_enabled(dev, type)) 8178 return -EALREADY; 8179 8180 switch (type) { 8181 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8182 netdev_offload_xstats_disable_l3(dev); 8183 return 0; 8184 } 8185 8186 WARN_ON(1); 8187 return -EINVAL; 8188 } 8189 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8190 8191 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8192 { 8193 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8194 } 8195 8196 static struct rtnl_hw_stats64 * 8197 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8198 enum netdev_offload_xstats_type type) 8199 { 8200 switch (type) { 8201 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8202 return dev->offload_xstats_l3; 8203 } 8204 8205 WARN_ON(1); 8206 return NULL; 8207 } 8208 8209 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8210 enum netdev_offload_xstats_type type) 8211 { 8212 ASSERT_RTNL(); 8213 8214 return netdev_offload_xstats_get_ptr(dev, type); 8215 } 8216 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8217 8218 struct netdev_notifier_offload_xstats_ru { 8219 bool used; 8220 }; 8221 8222 struct netdev_notifier_offload_xstats_rd { 8223 struct rtnl_hw_stats64 stats; 8224 bool used; 8225 }; 8226 8227 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8228 const struct rtnl_hw_stats64 *src) 8229 { 8230 dest->rx_packets += src->rx_packets; 8231 dest->tx_packets += src->tx_packets; 8232 dest->rx_bytes += src->rx_bytes; 8233 dest->tx_bytes += src->tx_bytes; 8234 dest->rx_errors += src->rx_errors; 8235 dest->tx_errors += src->tx_errors; 8236 dest->rx_dropped += src->rx_dropped; 8237 dest->tx_dropped += src->tx_dropped; 8238 dest->multicast += src->multicast; 8239 } 8240 8241 static int netdev_offload_xstats_get_used(struct net_device *dev, 8242 enum netdev_offload_xstats_type type, 8243 bool *p_used, 8244 struct netlink_ext_ack *extack) 8245 { 8246 struct netdev_notifier_offload_xstats_ru report_used = {}; 8247 struct netdev_notifier_offload_xstats_info info = { 8248 .info.dev = dev, 8249 .info.extack = extack, 8250 .type = type, 8251 .report_used = &report_used, 8252 }; 8253 int rc; 8254 8255 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8256 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8257 &info.info); 8258 *p_used = report_used.used; 8259 return notifier_to_errno(rc); 8260 } 8261 8262 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8263 enum netdev_offload_xstats_type type, 8264 struct rtnl_hw_stats64 *p_stats, 8265 bool *p_used, 8266 struct netlink_ext_ack *extack) 8267 { 8268 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8269 struct netdev_notifier_offload_xstats_info info = { 8270 .info.dev = dev, 8271 .info.extack = extack, 8272 .type = type, 8273 .report_delta = &report_delta, 8274 }; 8275 struct rtnl_hw_stats64 *stats; 8276 int rc; 8277 8278 stats = netdev_offload_xstats_get_ptr(dev, type); 8279 if (WARN_ON(!stats)) 8280 return -EINVAL; 8281 8282 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8283 &info.info); 8284 8285 /* Cache whatever we got, even if there was an error, otherwise the 8286 * successful stats retrievals would get lost. 8287 */ 8288 netdev_hw_stats64_add(stats, &report_delta.stats); 8289 8290 if (p_stats) 8291 *p_stats = *stats; 8292 *p_used = report_delta.used; 8293 8294 return notifier_to_errno(rc); 8295 } 8296 8297 int netdev_offload_xstats_get(struct net_device *dev, 8298 enum netdev_offload_xstats_type type, 8299 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8300 struct netlink_ext_ack *extack) 8301 { 8302 ASSERT_RTNL(); 8303 8304 if (p_stats) 8305 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8306 p_used, extack); 8307 else 8308 return netdev_offload_xstats_get_used(dev, type, p_used, 8309 extack); 8310 } 8311 EXPORT_SYMBOL(netdev_offload_xstats_get); 8312 8313 void 8314 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8315 const struct rtnl_hw_stats64 *stats) 8316 { 8317 report_delta->used = true; 8318 netdev_hw_stats64_add(&report_delta->stats, stats); 8319 } 8320 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8321 8322 void 8323 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8324 { 8325 report_used->used = true; 8326 } 8327 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8328 8329 void netdev_offload_xstats_push_delta(struct net_device *dev, 8330 enum netdev_offload_xstats_type type, 8331 const struct rtnl_hw_stats64 *p_stats) 8332 { 8333 struct rtnl_hw_stats64 *stats; 8334 8335 ASSERT_RTNL(); 8336 8337 stats = netdev_offload_xstats_get_ptr(dev, type); 8338 if (WARN_ON(!stats)) 8339 return; 8340 8341 netdev_hw_stats64_add(stats, p_stats); 8342 } 8343 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8344 8345 /** 8346 * netdev_get_xmit_slave - Get the xmit slave of master device 8347 * @dev: device 8348 * @skb: The packet 8349 * @all_slaves: assume all the slaves are active 8350 * 8351 * The reference counters are not incremented so the caller must be 8352 * careful with locks. The caller must hold RCU lock. 8353 * %NULL is returned if no slave is found. 8354 */ 8355 8356 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8357 struct sk_buff *skb, 8358 bool all_slaves) 8359 { 8360 const struct net_device_ops *ops = dev->netdev_ops; 8361 8362 if (!ops->ndo_get_xmit_slave) 8363 return NULL; 8364 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8365 } 8366 EXPORT_SYMBOL(netdev_get_xmit_slave); 8367 8368 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8369 struct sock *sk) 8370 { 8371 const struct net_device_ops *ops = dev->netdev_ops; 8372 8373 if (!ops->ndo_sk_get_lower_dev) 8374 return NULL; 8375 return ops->ndo_sk_get_lower_dev(dev, sk); 8376 } 8377 8378 /** 8379 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8380 * @dev: device 8381 * @sk: the socket 8382 * 8383 * %NULL is returned if no lower device is found. 8384 */ 8385 8386 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8387 struct sock *sk) 8388 { 8389 struct net_device *lower; 8390 8391 lower = netdev_sk_get_lower_dev(dev, sk); 8392 while (lower) { 8393 dev = lower; 8394 lower = netdev_sk_get_lower_dev(dev, sk); 8395 } 8396 8397 return dev; 8398 } 8399 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8400 8401 static void netdev_adjacent_add_links(struct net_device *dev) 8402 { 8403 struct netdev_adjacent *iter; 8404 8405 struct net *net = dev_net(dev); 8406 8407 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8408 if (!net_eq(net, dev_net(iter->dev))) 8409 continue; 8410 netdev_adjacent_sysfs_add(iter->dev, dev, 8411 &iter->dev->adj_list.lower); 8412 netdev_adjacent_sysfs_add(dev, iter->dev, 8413 &dev->adj_list.upper); 8414 } 8415 8416 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8417 if (!net_eq(net, dev_net(iter->dev))) 8418 continue; 8419 netdev_adjacent_sysfs_add(iter->dev, dev, 8420 &iter->dev->adj_list.upper); 8421 netdev_adjacent_sysfs_add(dev, iter->dev, 8422 &dev->adj_list.lower); 8423 } 8424 } 8425 8426 static void netdev_adjacent_del_links(struct net_device *dev) 8427 { 8428 struct netdev_adjacent *iter; 8429 8430 struct net *net = dev_net(dev); 8431 8432 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8433 if (!net_eq(net, dev_net(iter->dev))) 8434 continue; 8435 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8436 &iter->dev->adj_list.lower); 8437 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8438 &dev->adj_list.upper); 8439 } 8440 8441 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8442 if (!net_eq(net, dev_net(iter->dev))) 8443 continue; 8444 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8445 &iter->dev->adj_list.upper); 8446 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8447 &dev->adj_list.lower); 8448 } 8449 } 8450 8451 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8452 { 8453 struct netdev_adjacent *iter; 8454 8455 struct net *net = dev_net(dev); 8456 8457 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8458 if (!net_eq(net, dev_net(iter->dev))) 8459 continue; 8460 netdev_adjacent_sysfs_del(iter->dev, oldname, 8461 &iter->dev->adj_list.lower); 8462 netdev_adjacent_sysfs_add(iter->dev, dev, 8463 &iter->dev->adj_list.lower); 8464 } 8465 8466 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8467 if (!net_eq(net, dev_net(iter->dev))) 8468 continue; 8469 netdev_adjacent_sysfs_del(iter->dev, oldname, 8470 &iter->dev->adj_list.upper); 8471 netdev_adjacent_sysfs_add(iter->dev, dev, 8472 &iter->dev->adj_list.upper); 8473 } 8474 } 8475 8476 void *netdev_lower_dev_get_private(struct net_device *dev, 8477 struct net_device *lower_dev) 8478 { 8479 struct netdev_adjacent *lower; 8480 8481 if (!lower_dev) 8482 return NULL; 8483 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8484 if (!lower) 8485 return NULL; 8486 8487 return lower->private; 8488 } 8489 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8490 8491 8492 /** 8493 * netdev_lower_state_changed - Dispatch event about lower device state change 8494 * @lower_dev: device 8495 * @lower_state_info: state to dispatch 8496 * 8497 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8498 * The caller must hold the RTNL lock. 8499 */ 8500 void netdev_lower_state_changed(struct net_device *lower_dev, 8501 void *lower_state_info) 8502 { 8503 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8504 .info.dev = lower_dev, 8505 }; 8506 8507 ASSERT_RTNL(); 8508 changelowerstate_info.lower_state_info = lower_state_info; 8509 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8510 &changelowerstate_info.info); 8511 } 8512 EXPORT_SYMBOL(netdev_lower_state_changed); 8513 8514 static void dev_change_rx_flags(struct net_device *dev, int flags) 8515 { 8516 const struct net_device_ops *ops = dev->netdev_ops; 8517 8518 if (ops->ndo_change_rx_flags) 8519 ops->ndo_change_rx_flags(dev, flags); 8520 } 8521 8522 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8523 { 8524 unsigned int old_flags = dev->flags; 8525 kuid_t uid; 8526 kgid_t gid; 8527 8528 ASSERT_RTNL(); 8529 8530 dev->flags |= IFF_PROMISC; 8531 dev->promiscuity += inc; 8532 if (dev->promiscuity == 0) { 8533 /* 8534 * Avoid overflow. 8535 * If inc causes overflow, untouch promisc and return error. 8536 */ 8537 if (inc < 0) 8538 dev->flags &= ~IFF_PROMISC; 8539 else { 8540 dev->promiscuity -= inc; 8541 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8542 return -EOVERFLOW; 8543 } 8544 } 8545 if (dev->flags != old_flags) { 8546 netdev_info(dev, "%s promiscuous mode\n", 8547 dev->flags & IFF_PROMISC ? "entered" : "left"); 8548 if (audit_enabled) { 8549 current_uid_gid(&uid, &gid); 8550 audit_log(audit_context(), GFP_ATOMIC, 8551 AUDIT_ANOM_PROMISCUOUS, 8552 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8553 dev->name, (dev->flags & IFF_PROMISC), 8554 (old_flags & IFF_PROMISC), 8555 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8556 from_kuid(&init_user_ns, uid), 8557 from_kgid(&init_user_ns, gid), 8558 audit_get_sessionid(current)); 8559 } 8560 8561 dev_change_rx_flags(dev, IFF_PROMISC); 8562 } 8563 if (notify) 8564 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8565 return 0; 8566 } 8567 8568 /** 8569 * dev_set_promiscuity - update promiscuity count on a device 8570 * @dev: device 8571 * @inc: modifier 8572 * 8573 * Add or remove promiscuity from a device. While the count in the device 8574 * remains above zero the interface remains promiscuous. Once it hits zero 8575 * the device reverts back to normal filtering operation. A negative inc 8576 * value is used to drop promiscuity on the device. 8577 * Return 0 if successful or a negative errno code on error. 8578 */ 8579 int dev_set_promiscuity(struct net_device *dev, int inc) 8580 { 8581 unsigned int old_flags = dev->flags; 8582 int err; 8583 8584 err = __dev_set_promiscuity(dev, inc, true); 8585 if (err < 0) 8586 return err; 8587 if (dev->flags != old_flags) 8588 dev_set_rx_mode(dev); 8589 return err; 8590 } 8591 EXPORT_SYMBOL(dev_set_promiscuity); 8592 8593 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8594 { 8595 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8596 8597 ASSERT_RTNL(); 8598 8599 dev->flags |= IFF_ALLMULTI; 8600 dev->allmulti += inc; 8601 if (dev->allmulti == 0) { 8602 /* 8603 * Avoid overflow. 8604 * If inc causes overflow, untouch allmulti and return error. 8605 */ 8606 if (inc < 0) 8607 dev->flags &= ~IFF_ALLMULTI; 8608 else { 8609 dev->allmulti -= inc; 8610 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8611 return -EOVERFLOW; 8612 } 8613 } 8614 if (dev->flags ^ old_flags) { 8615 netdev_info(dev, "%s allmulticast mode\n", 8616 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8617 dev_change_rx_flags(dev, IFF_ALLMULTI); 8618 dev_set_rx_mode(dev); 8619 if (notify) 8620 __dev_notify_flags(dev, old_flags, 8621 dev->gflags ^ old_gflags, 0, NULL); 8622 } 8623 return 0; 8624 } 8625 8626 /** 8627 * dev_set_allmulti - update allmulti count on a device 8628 * @dev: device 8629 * @inc: modifier 8630 * 8631 * Add or remove reception of all multicast frames to a device. While the 8632 * count in the device remains above zero the interface remains listening 8633 * to all interfaces. Once it hits zero the device reverts back to normal 8634 * filtering operation. A negative @inc value is used to drop the counter 8635 * when releasing a resource needing all multicasts. 8636 * Return 0 if successful or a negative errno code on error. 8637 */ 8638 8639 int dev_set_allmulti(struct net_device *dev, int inc) 8640 { 8641 return __dev_set_allmulti(dev, inc, true); 8642 } 8643 EXPORT_SYMBOL(dev_set_allmulti); 8644 8645 /* 8646 * Upload unicast and multicast address lists to device and 8647 * configure RX filtering. When the device doesn't support unicast 8648 * filtering it is put in promiscuous mode while unicast addresses 8649 * are present. 8650 */ 8651 void __dev_set_rx_mode(struct net_device *dev) 8652 { 8653 const struct net_device_ops *ops = dev->netdev_ops; 8654 8655 /* dev_open will call this function so the list will stay sane. */ 8656 if (!(dev->flags&IFF_UP)) 8657 return; 8658 8659 if (!netif_device_present(dev)) 8660 return; 8661 8662 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8663 /* Unicast addresses changes may only happen under the rtnl, 8664 * therefore calling __dev_set_promiscuity here is safe. 8665 */ 8666 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8667 __dev_set_promiscuity(dev, 1, false); 8668 dev->uc_promisc = true; 8669 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8670 __dev_set_promiscuity(dev, -1, false); 8671 dev->uc_promisc = false; 8672 } 8673 } 8674 8675 if (ops->ndo_set_rx_mode) 8676 ops->ndo_set_rx_mode(dev); 8677 } 8678 8679 void dev_set_rx_mode(struct net_device *dev) 8680 { 8681 netif_addr_lock_bh(dev); 8682 __dev_set_rx_mode(dev); 8683 netif_addr_unlock_bh(dev); 8684 } 8685 8686 /** 8687 * dev_get_flags - get flags reported to userspace 8688 * @dev: device 8689 * 8690 * Get the combination of flag bits exported through APIs to userspace. 8691 */ 8692 unsigned int dev_get_flags(const struct net_device *dev) 8693 { 8694 unsigned int flags; 8695 8696 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | 8697 IFF_ALLMULTI | 8698 IFF_RUNNING | 8699 IFF_LOWER_UP | 8700 IFF_DORMANT)) | 8701 (READ_ONCE(dev->gflags) & (IFF_PROMISC | 8702 IFF_ALLMULTI)); 8703 8704 if (netif_running(dev)) { 8705 if (netif_oper_up(dev)) 8706 flags |= IFF_RUNNING; 8707 if (netif_carrier_ok(dev)) 8708 flags |= IFF_LOWER_UP; 8709 if (netif_dormant(dev)) 8710 flags |= IFF_DORMANT; 8711 } 8712 8713 return flags; 8714 } 8715 EXPORT_SYMBOL(dev_get_flags); 8716 8717 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8718 struct netlink_ext_ack *extack) 8719 { 8720 unsigned int old_flags = dev->flags; 8721 int ret; 8722 8723 ASSERT_RTNL(); 8724 8725 /* 8726 * Set the flags on our device. 8727 */ 8728 8729 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8730 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8731 IFF_AUTOMEDIA)) | 8732 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8733 IFF_ALLMULTI)); 8734 8735 /* 8736 * Load in the correct multicast list now the flags have changed. 8737 */ 8738 8739 if ((old_flags ^ flags) & IFF_MULTICAST) 8740 dev_change_rx_flags(dev, IFF_MULTICAST); 8741 8742 dev_set_rx_mode(dev); 8743 8744 /* 8745 * Have we downed the interface. We handle IFF_UP ourselves 8746 * according to user attempts to set it, rather than blindly 8747 * setting it. 8748 */ 8749 8750 ret = 0; 8751 if ((old_flags ^ flags) & IFF_UP) { 8752 if (old_flags & IFF_UP) 8753 __dev_close(dev); 8754 else 8755 ret = __dev_open(dev, extack); 8756 } 8757 8758 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8759 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8760 unsigned int old_flags = dev->flags; 8761 8762 dev->gflags ^= IFF_PROMISC; 8763 8764 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8765 if (dev->flags != old_flags) 8766 dev_set_rx_mode(dev); 8767 } 8768 8769 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8770 * is important. Some (broken) drivers set IFF_PROMISC, when 8771 * IFF_ALLMULTI is requested not asking us and not reporting. 8772 */ 8773 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8774 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8775 8776 dev->gflags ^= IFF_ALLMULTI; 8777 __dev_set_allmulti(dev, inc, false); 8778 } 8779 8780 return ret; 8781 } 8782 8783 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8784 unsigned int gchanges, u32 portid, 8785 const struct nlmsghdr *nlh) 8786 { 8787 unsigned int changes = dev->flags ^ old_flags; 8788 8789 if (gchanges) 8790 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8791 8792 if (changes & IFF_UP) { 8793 if (dev->flags & IFF_UP) 8794 call_netdevice_notifiers(NETDEV_UP, dev); 8795 else 8796 call_netdevice_notifiers(NETDEV_DOWN, dev); 8797 } 8798 8799 if (dev->flags & IFF_UP && 8800 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8801 struct netdev_notifier_change_info change_info = { 8802 .info = { 8803 .dev = dev, 8804 }, 8805 .flags_changed = changes, 8806 }; 8807 8808 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8809 } 8810 } 8811 8812 /** 8813 * dev_change_flags - change device settings 8814 * @dev: device 8815 * @flags: device state flags 8816 * @extack: netlink extended ack 8817 * 8818 * Change settings on device based state flags. The flags are 8819 * in the userspace exported format. 8820 */ 8821 int dev_change_flags(struct net_device *dev, unsigned int flags, 8822 struct netlink_ext_ack *extack) 8823 { 8824 int ret; 8825 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8826 8827 ret = __dev_change_flags(dev, flags, extack); 8828 if (ret < 0) 8829 return ret; 8830 8831 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8832 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8833 return ret; 8834 } 8835 EXPORT_SYMBOL(dev_change_flags); 8836 8837 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8838 { 8839 const struct net_device_ops *ops = dev->netdev_ops; 8840 8841 if (ops->ndo_change_mtu) 8842 return ops->ndo_change_mtu(dev, new_mtu); 8843 8844 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8845 WRITE_ONCE(dev->mtu, new_mtu); 8846 return 0; 8847 } 8848 EXPORT_SYMBOL(__dev_set_mtu); 8849 8850 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8851 struct netlink_ext_ack *extack) 8852 { 8853 /* MTU must be positive, and in range */ 8854 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8855 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8856 return -EINVAL; 8857 } 8858 8859 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8860 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8861 return -EINVAL; 8862 } 8863 return 0; 8864 } 8865 8866 /** 8867 * dev_set_mtu_ext - Change maximum transfer unit 8868 * @dev: device 8869 * @new_mtu: new transfer unit 8870 * @extack: netlink extended ack 8871 * 8872 * Change the maximum transfer size of the network device. 8873 */ 8874 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8875 struct netlink_ext_ack *extack) 8876 { 8877 int err, orig_mtu; 8878 8879 if (new_mtu == dev->mtu) 8880 return 0; 8881 8882 err = dev_validate_mtu(dev, new_mtu, extack); 8883 if (err) 8884 return err; 8885 8886 if (!netif_device_present(dev)) 8887 return -ENODEV; 8888 8889 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8890 err = notifier_to_errno(err); 8891 if (err) 8892 return err; 8893 8894 orig_mtu = dev->mtu; 8895 err = __dev_set_mtu(dev, new_mtu); 8896 8897 if (!err) { 8898 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8899 orig_mtu); 8900 err = notifier_to_errno(err); 8901 if (err) { 8902 /* setting mtu back and notifying everyone again, 8903 * so that they have a chance to revert changes. 8904 */ 8905 __dev_set_mtu(dev, orig_mtu); 8906 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8907 new_mtu); 8908 } 8909 } 8910 return err; 8911 } 8912 8913 int dev_set_mtu(struct net_device *dev, int new_mtu) 8914 { 8915 struct netlink_ext_ack extack; 8916 int err; 8917 8918 memset(&extack, 0, sizeof(extack)); 8919 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8920 if (err && extack._msg) 8921 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8922 return err; 8923 } 8924 EXPORT_SYMBOL(dev_set_mtu); 8925 8926 /** 8927 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8928 * @dev: device 8929 * @new_len: new tx queue length 8930 */ 8931 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8932 { 8933 unsigned int orig_len = dev->tx_queue_len; 8934 int res; 8935 8936 if (new_len != (unsigned int)new_len) 8937 return -ERANGE; 8938 8939 if (new_len != orig_len) { 8940 dev->tx_queue_len = new_len; 8941 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8942 res = notifier_to_errno(res); 8943 if (res) 8944 goto err_rollback; 8945 res = dev_qdisc_change_tx_queue_len(dev); 8946 if (res) 8947 goto err_rollback; 8948 } 8949 8950 return 0; 8951 8952 err_rollback: 8953 netdev_err(dev, "refused to change device tx_queue_len\n"); 8954 dev->tx_queue_len = orig_len; 8955 return res; 8956 } 8957 8958 /** 8959 * dev_set_group - Change group this device belongs to 8960 * @dev: device 8961 * @new_group: group this device should belong to 8962 */ 8963 void dev_set_group(struct net_device *dev, int new_group) 8964 { 8965 dev->group = new_group; 8966 } 8967 8968 /** 8969 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8970 * @dev: device 8971 * @addr: new address 8972 * @extack: netlink extended ack 8973 */ 8974 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8975 struct netlink_ext_ack *extack) 8976 { 8977 struct netdev_notifier_pre_changeaddr_info info = { 8978 .info.dev = dev, 8979 .info.extack = extack, 8980 .dev_addr = addr, 8981 }; 8982 int rc; 8983 8984 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8985 return notifier_to_errno(rc); 8986 } 8987 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8988 8989 /** 8990 * dev_set_mac_address - Change Media Access Control Address 8991 * @dev: device 8992 * @sa: new address 8993 * @extack: netlink extended ack 8994 * 8995 * Change the hardware (MAC) address of the device 8996 */ 8997 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8998 struct netlink_ext_ack *extack) 8999 { 9000 const struct net_device_ops *ops = dev->netdev_ops; 9001 int err; 9002 9003 if (!ops->ndo_set_mac_address) 9004 return -EOPNOTSUPP; 9005 if (sa->sa_family != dev->type) 9006 return -EINVAL; 9007 if (!netif_device_present(dev)) 9008 return -ENODEV; 9009 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9010 if (err) 9011 return err; 9012 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 9013 err = ops->ndo_set_mac_address(dev, sa); 9014 if (err) 9015 return err; 9016 } 9017 dev->addr_assign_type = NET_ADDR_SET; 9018 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9019 add_device_randomness(dev->dev_addr, dev->addr_len); 9020 return 0; 9021 } 9022 EXPORT_SYMBOL(dev_set_mac_address); 9023 9024 DECLARE_RWSEM(dev_addr_sem); 9025 9026 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9027 struct netlink_ext_ack *extack) 9028 { 9029 int ret; 9030 9031 down_write(&dev_addr_sem); 9032 ret = dev_set_mac_address(dev, sa, extack); 9033 up_write(&dev_addr_sem); 9034 return ret; 9035 } 9036 EXPORT_SYMBOL(dev_set_mac_address_user); 9037 9038 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9039 { 9040 size_t size = sizeof(sa->sa_data_min); 9041 struct net_device *dev; 9042 int ret = 0; 9043 9044 down_read(&dev_addr_sem); 9045 rcu_read_lock(); 9046 9047 dev = dev_get_by_name_rcu(net, dev_name); 9048 if (!dev) { 9049 ret = -ENODEV; 9050 goto unlock; 9051 } 9052 if (!dev->addr_len) 9053 memset(sa->sa_data, 0, size); 9054 else 9055 memcpy(sa->sa_data, dev->dev_addr, 9056 min_t(size_t, size, dev->addr_len)); 9057 sa->sa_family = dev->type; 9058 9059 unlock: 9060 rcu_read_unlock(); 9061 up_read(&dev_addr_sem); 9062 return ret; 9063 } 9064 EXPORT_SYMBOL(dev_get_mac_address); 9065 9066 /** 9067 * dev_change_carrier - Change device carrier 9068 * @dev: device 9069 * @new_carrier: new value 9070 * 9071 * Change device carrier 9072 */ 9073 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9074 { 9075 const struct net_device_ops *ops = dev->netdev_ops; 9076 9077 if (!ops->ndo_change_carrier) 9078 return -EOPNOTSUPP; 9079 if (!netif_device_present(dev)) 9080 return -ENODEV; 9081 return ops->ndo_change_carrier(dev, new_carrier); 9082 } 9083 9084 /** 9085 * dev_get_phys_port_id - Get device physical port ID 9086 * @dev: device 9087 * @ppid: port ID 9088 * 9089 * Get device physical port ID 9090 */ 9091 int dev_get_phys_port_id(struct net_device *dev, 9092 struct netdev_phys_item_id *ppid) 9093 { 9094 const struct net_device_ops *ops = dev->netdev_ops; 9095 9096 if (!ops->ndo_get_phys_port_id) 9097 return -EOPNOTSUPP; 9098 return ops->ndo_get_phys_port_id(dev, ppid); 9099 } 9100 9101 /** 9102 * dev_get_phys_port_name - Get device physical port name 9103 * @dev: device 9104 * @name: port name 9105 * @len: limit of bytes to copy to name 9106 * 9107 * Get device physical port name 9108 */ 9109 int dev_get_phys_port_name(struct net_device *dev, 9110 char *name, size_t len) 9111 { 9112 const struct net_device_ops *ops = dev->netdev_ops; 9113 int err; 9114 9115 if (ops->ndo_get_phys_port_name) { 9116 err = ops->ndo_get_phys_port_name(dev, name, len); 9117 if (err != -EOPNOTSUPP) 9118 return err; 9119 } 9120 return devlink_compat_phys_port_name_get(dev, name, len); 9121 } 9122 9123 /** 9124 * dev_get_port_parent_id - Get the device's port parent identifier 9125 * @dev: network device 9126 * @ppid: pointer to a storage for the port's parent identifier 9127 * @recurse: allow/disallow recursion to lower devices 9128 * 9129 * Get the devices's port parent identifier 9130 */ 9131 int dev_get_port_parent_id(struct net_device *dev, 9132 struct netdev_phys_item_id *ppid, 9133 bool recurse) 9134 { 9135 const struct net_device_ops *ops = dev->netdev_ops; 9136 struct netdev_phys_item_id first = { }; 9137 struct net_device *lower_dev; 9138 struct list_head *iter; 9139 int err; 9140 9141 if (ops->ndo_get_port_parent_id) { 9142 err = ops->ndo_get_port_parent_id(dev, ppid); 9143 if (err != -EOPNOTSUPP) 9144 return err; 9145 } 9146 9147 err = devlink_compat_switch_id_get(dev, ppid); 9148 if (!recurse || err != -EOPNOTSUPP) 9149 return err; 9150 9151 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9152 err = dev_get_port_parent_id(lower_dev, ppid, true); 9153 if (err) 9154 break; 9155 if (!first.id_len) 9156 first = *ppid; 9157 else if (memcmp(&first, ppid, sizeof(*ppid))) 9158 return -EOPNOTSUPP; 9159 } 9160 9161 return err; 9162 } 9163 EXPORT_SYMBOL(dev_get_port_parent_id); 9164 9165 /** 9166 * netdev_port_same_parent_id - Indicate if two network devices have 9167 * the same port parent identifier 9168 * @a: first network device 9169 * @b: second network device 9170 */ 9171 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9172 { 9173 struct netdev_phys_item_id a_id = { }; 9174 struct netdev_phys_item_id b_id = { }; 9175 9176 if (dev_get_port_parent_id(a, &a_id, true) || 9177 dev_get_port_parent_id(b, &b_id, true)) 9178 return false; 9179 9180 return netdev_phys_item_id_same(&a_id, &b_id); 9181 } 9182 EXPORT_SYMBOL(netdev_port_same_parent_id); 9183 9184 /** 9185 * dev_change_proto_down - set carrier according to proto_down. 9186 * 9187 * @dev: device 9188 * @proto_down: new value 9189 */ 9190 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9191 { 9192 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9193 return -EOPNOTSUPP; 9194 if (!netif_device_present(dev)) 9195 return -ENODEV; 9196 if (proto_down) 9197 netif_carrier_off(dev); 9198 else 9199 netif_carrier_on(dev); 9200 dev->proto_down = proto_down; 9201 return 0; 9202 } 9203 9204 /** 9205 * dev_change_proto_down_reason - proto down reason 9206 * 9207 * @dev: device 9208 * @mask: proto down mask 9209 * @value: proto down value 9210 */ 9211 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9212 u32 value) 9213 { 9214 int b; 9215 9216 if (!mask) { 9217 dev->proto_down_reason = value; 9218 } else { 9219 for_each_set_bit(b, &mask, 32) { 9220 if (value & (1 << b)) 9221 dev->proto_down_reason |= BIT(b); 9222 else 9223 dev->proto_down_reason &= ~BIT(b); 9224 } 9225 } 9226 } 9227 9228 struct bpf_xdp_link { 9229 struct bpf_link link; 9230 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9231 int flags; 9232 }; 9233 9234 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9235 { 9236 if (flags & XDP_FLAGS_HW_MODE) 9237 return XDP_MODE_HW; 9238 if (flags & XDP_FLAGS_DRV_MODE) 9239 return XDP_MODE_DRV; 9240 if (flags & XDP_FLAGS_SKB_MODE) 9241 return XDP_MODE_SKB; 9242 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9243 } 9244 9245 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9246 { 9247 switch (mode) { 9248 case XDP_MODE_SKB: 9249 return generic_xdp_install; 9250 case XDP_MODE_DRV: 9251 case XDP_MODE_HW: 9252 return dev->netdev_ops->ndo_bpf; 9253 default: 9254 return NULL; 9255 } 9256 } 9257 9258 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9259 enum bpf_xdp_mode mode) 9260 { 9261 return dev->xdp_state[mode].link; 9262 } 9263 9264 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9265 enum bpf_xdp_mode mode) 9266 { 9267 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9268 9269 if (link) 9270 return link->link.prog; 9271 return dev->xdp_state[mode].prog; 9272 } 9273 9274 u8 dev_xdp_prog_count(struct net_device *dev) 9275 { 9276 u8 count = 0; 9277 int i; 9278 9279 for (i = 0; i < __MAX_XDP_MODE; i++) 9280 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9281 count++; 9282 return count; 9283 } 9284 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9285 9286 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9287 { 9288 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9289 9290 return prog ? prog->aux->id : 0; 9291 } 9292 9293 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9294 struct bpf_xdp_link *link) 9295 { 9296 dev->xdp_state[mode].link = link; 9297 dev->xdp_state[mode].prog = NULL; 9298 } 9299 9300 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9301 struct bpf_prog *prog) 9302 { 9303 dev->xdp_state[mode].link = NULL; 9304 dev->xdp_state[mode].prog = prog; 9305 } 9306 9307 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9308 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9309 u32 flags, struct bpf_prog *prog) 9310 { 9311 struct netdev_bpf xdp; 9312 int err; 9313 9314 memset(&xdp, 0, sizeof(xdp)); 9315 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9316 xdp.extack = extack; 9317 xdp.flags = flags; 9318 xdp.prog = prog; 9319 9320 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9321 * "moved" into driver), so they don't increment it on their own, but 9322 * they do decrement refcnt when program is detached or replaced. 9323 * Given net_device also owns link/prog, we need to bump refcnt here 9324 * to prevent drivers from underflowing it. 9325 */ 9326 if (prog) 9327 bpf_prog_inc(prog); 9328 err = bpf_op(dev, &xdp); 9329 if (err) { 9330 if (prog) 9331 bpf_prog_put(prog); 9332 return err; 9333 } 9334 9335 if (mode != XDP_MODE_HW) 9336 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9337 9338 return 0; 9339 } 9340 9341 static void dev_xdp_uninstall(struct net_device *dev) 9342 { 9343 struct bpf_xdp_link *link; 9344 struct bpf_prog *prog; 9345 enum bpf_xdp_mode mode; 9346 bpf_op_t bpf_op; 9347 9348 ASSERT_RTNL(); 9349 9350 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9351 prog = dev_xdp_prog(dev, mode); 9352 if (!prog) 9353 continue; 9354 9355 bpf_op = dev_xdp_bpf_op(dev, mode); 9356 if (!bpf_op) 9357 continue; 9358 9359 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9360 9361 /* auto-detach link from net device */ 9362 link = dev_xdp_link(dev, mode); 9363 if (link) 9364 link->dev = NULL; 9365 else 9366 bpf_prog_put(prog); 9367 9368 dev_xdp_set_link(dev, mode, NULL); 9369 } 9370 } 9371 9372 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9373 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9374 struct bpf_prog *old_prog, u32 flags) 9375 { 9376 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9377 struct bpf_prog *cur_prog; 9378 struct net_device *upper; 9379 struct list_head *iter; 9380 enum bpf_xdp_mode mode; 9381 bpf_op_t bpf_op; 9382 int err; 9383 9384 ASSERT_RTNL(); 9385 9386 /* either link or prog attachment, never both */ 9387 if (link && (new_prog || old_prog)) 9388 return -EINVAL; 9389 /* link supports only XDP mode flags */ 9390 if (link && (flags & ~XDP_FLAGS_MODES)) { 9391 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9392 return -EINVAL; 9393 } 9394 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9395 if (num_modes > 1) { 9396 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9397 return -EINVAL; 9398 } 9399 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9400 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9401 NL_SET_ERR_MSG(extack, 9402 "More than one program loaded, unset mode is ambiguous"); 9403 return -EINVAL; 9404 } 9405 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9406 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9407 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9408 return -EINVAL; 9409 } 9410 9411 mode = dev_xdp_mode(dev, flags); 9412 /* can't replace attached link */ 9413 if (dev_xdp_link(dev, mode)) { 9414 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9415 return -EBUSY; 9416 } 9417 9418 /* don't allow if an upper device already has a program */ 9419 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9420 if (dev_xdp_prog_count(upper) > 0) { 9421 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9422 return -EEXIST; 9423 } 9424 } 9425 9426 cur_prog = dev_xdp_prog(dev, mode); 9427 /* can't replace attached prog with link */ 9428 if (link && cur_prog) { 9429 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9430 return -EBUSY; 9431 } 9432 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9433 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9434 return -EEXIST; 9435 } 9436 9437 /* put effective new program into new_prog */ 9438 if (link) 9439 new_prog = link->link.prog; 9440 9441 if (new_prog) { 9442 bool offload = mode == XDP_MODE_HW; 9443 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9444 ? XDP_MODE_DRV : XDP_MODE_SKB; 9445 9446 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9447 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9448 return -EBUSY; 9449 } 9450 if (!offload && dev_xdp_prog(dev, other_mode)) { 9451 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9452 return -EEXIST; 9453 } 9454 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9455 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9456 return -EINVAL; 9457 } 9458 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9459 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9460 return -EINVAL; 9461 } 9462 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9463 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9464 return -EINVAL; 9465 } 9466 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9467 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9468 return -EINVAL; 9469 } 9470 } 9471 9472 /* don't call drivers if the effective program didn't change */ 9473 if (new_prog != cur_prog) { 9474 bpf_op = dev_xdp_bpf_op(dev, mode); 9475 if (!bpf_op) { 9476 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9477 return -EOPNOTSUPP; 9478 } 9479 9480 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9481 if (err) 9482 return err; 9483 } 9484 9485 if (link) 9486 dev_xdp_set_link(dev, mode, link); 9487 else 9488 dev_xdp_set_prog(dev, mode, new_prog); 9489 if (cur_prog) 9490 bpf_prog_put(cur_prog); 9491 9492 return 0; 9493 } 9494 9495 static int dev_xdp_attach_link(struct net_device *dev, 9496 struct netlink_ext_ack *extack, 9497 struct bpf_xdp_link *link) 9498 { 9499 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9500 } 9501 9502 static int dev_xdp_detach_link(struct net_device *dev, 9503 struct netlink_ext_ack *extack, 9504 struct bpf_xdp_link *link) 9505 { 9506 enum bpf_xdp_mode mode; 9507 bpf_op_t bpf_op; 9508 9509 ASSERT_RTNL(); 9510 9511 mode = dev_xdp_mode(dev, link->flags); 9512 if (dev_xdp_link(dev, mode) != link) 9513 return -EINVAL; 9514 9515 bpf_op = dev_xdp_bpf_op(dev, mode); 9516 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9517 dev_xdp_set_link(dev, mode, NULL); 9518 return 0; 9519 } 9520 9521 static void bpf_xdp_link_release(struct bpf_link *link) 9522 { 9523 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9524 9525 rtnl_lock(); 9526 9527 /* if racing with net_device's tear down, xdp_link->dev might be 9528 * already NULL, in which case link was already auto-detached 9529 */ 9530 if (xdp_link->dev) { 9531 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9532 xdp_link->dev = NULL; 9533 } 9534 9535 rtnl_unlock(); 9536 } 9537 9538 static int bpf_xdp_link_detach(struct bpf_link *link) 9539 { 9540 bpf_xdp_link_release(link); 9541 return 0; 9542 } 9543 9544 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9545 { 9546 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9547 9548 kfree(xdp_link); 9549 } 9550 9551 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9552 struct seq_file *seq) 9553 { 9554 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9555 u32 ifindex = 0; 9556 9557 rtnl_lock(); 9558 if (xdp_link->dev) 9559 ifindex = xdp_link->dev->ifindex; 9560 rtnl_unlock(); 9561 9562 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9563 } 9564 9565 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9566 struct bpf_link_info *info) 9567 { 9568 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9569 u32 ifindex = 0; 9570 9571 rtnl_lock(); 9572 if (xdp_link->dev) 9573 ifindex = xdp_link->dev->ifindex; 9574 rtnl_unlock(); 9575 9576 info->xdp.ifindex = ifindex; 9577 return 0; 9578 } 9579 9580 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9581 struct bpf_prog *old_prog) 9582 { 9583 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9584 enum bpf_xdp_mode mode; 9585 bpf_op_t bpf_op; 9586 int err = 0; 9587 9588 rtnl_lock(); 9589 9590 /* link might have been auto-released already, so fail */ 9591 if (!xdp_link->dev) { 9592 err = -ENOLINK; 9593 goto out_unlock; 9594 } 9595 9596 if (old_prog && link->prog != old_prog) { 9597 err = -EPERM; 9598 goto out_unlock; 9599 } 9600 old_prog = link->prog; 9601 if (old_prog->type != new_prog->type || 9602 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9603 err = -EINVAL; 9604 goto out_unlock; 9605 } 9606 9607 if (old_prog == new_prog) { 9608 /* no-op, don't disturb drivers */ 9609 bpf_prog_put(new_prog); 9610 goto out_unlock; 9611 } 9612 9613 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9614 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9615 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9616 xdp_link->flags, new_prog); 9617 if (err) 9618 goto out_unlock; 9619 9620 old_prog = xchg(&link->prog, new_prog); 9621 bpf_prog_put(old_prog); 9622 9623 out_unlock: 9624 rtnl_unlock(); 9625 return err; 9626 } 9627 9628 static const struct bpf_link_ops bpf_xdp_link_lops = { 9629 .release = bpf_xdp_link_release, 9630 .dealloc = bpf_xdp_link_dealloc, 9631 .detach = bpf_xdp_link_detach, 9632 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9633 .fill_link_info = bpf_xdp_link_fill_link_info, 9634 .update_prog = bpf_xdp_link_update, 9635 }; 9636 9637 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9638 { 9639 struct net *net = current->nsproxy->net_ns; 9640 struct bpf_link_primer link_primer; 9641 struct netlink_ext_ack extack = {}; 9642 struct bpf_xdp_link *link; 9643 struct net_device *dev; 9644 int err, fd; 9645 9646 rtnl_lock(); 9647 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9648 if (!dev) { 9649 rtnl_unlock(); 9650 return -EINVAL; 9651 } 9652 9653 link = kzalloc(sizeof(*link), GFP_USER); 9654 if (!link) { 9655 err = -ENOMEM; 9656 goto unlock; 9657 } 9658 9659 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9660 link->dev = dev; 9661 link->flags = attr->link_create.flags; 9662 9663 err = bpf_link_prime(&link->link, &link_primer); 9664 if (err) { 9665 kfree(link); 9666 goto unlock; 9667 } 9668 9669 err = dev_xdp_attach_link(dev, &extack, link); 9670 rtnl_unlock(); 9671 9672 if (err) { 9673 link->dev = NULL; 9674 bpf_link_cleanup(&link_primer); 9675 trace_bpf_xdp_link_attach_failed(extack._msg); 9676 goto out_put_dev; 9677 } 9678 9679 fd = bpf_link_settle(&link_primer); 9680 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9681 dev_put(dev); 9682 return fd; 9683 9684 unlock: 9685 rtnl_unlock(); 9686 9687 out_put_dev: 9688 dev_put(dev); 9689 return err; 9690 } 9691 9692 /** 9693 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9694 * @dev: device 9695 * @extack: netlink extended ack 9696 * @fd: new program fd or negative value to clear 9697 * @expected_fd: old program fd that userspace expects to replace or clear 9698 * @flags: xdp-related flags 9699 * 9700 * Set or clear a bpf program for a device 9701 */ 9702 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9703 int fd, int expected_fd, u32 flags) 9704 { 9705 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9706 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9707 int err; 9708 9709 ASSERT_RTNL(); 9710 9711 if (fd >= 0) { 9712 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9713 mode != XDP_MODE_SKB); 9714 if (IS_ERR(new_prog)) 9715 return PTR_ERR(new_prog); 9716 } 9717 9718 if (expected_fd >= 0) { 9719 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9720 mode != XDP_MODE_SKB); 9721 if (IS_ERR(old_prog)) { 9722 err = PTR_ERR(old_prog); 9723 old_prog = NULL; 9724 goto err_out; 9725 } 9726 } 9727 9728 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9729 9730 err_out: 9731 if (err && new_prog) 9732 bpf_prog_put(new_prog); 9733 if (old_prog) 9734 bpf_prog_put(old_prog); 9735 return err; 9736 } 9737 9738 /** 9739 * dev_index_reserve() - allocate an ifindex in a namespace 9740 * @net: the applicable net namespace 9741 * @ifindex: requested ifindex, pass %0 to get one allocated 9742 * 9743 * Allocate a ifindex for a new device. Caller must either use the ifindex 9744 * to store the device (via list_netdevice()) or call dev_index_release() 9745 * to give the index up. 9746 * 9747 * Return: a suitable unique value for a new device interface number or -errno. 9748 */ 9749 static int dev_index_reserve(struct net *net, u32 ifindex) 9750 { 9751 int err; 9752 9753 if (ifindex > INT_MAX) { 9754 DEBUG_NET_WARN_ON_ONCE(1); 9755 return -EINVAL; 9756 } 9757 9758 if (!ifindex) 9759 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9760 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9761 else 9762 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9763 if (err < 0) 9764 return err; 9765 9766 return ifindex; 9767 } 9768 9769 static void dev_index_release(struct net *net, int ifindex) 9770 { 9771 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9772 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9773 } 9774 9775 /* Delayed registration/unregisteration */ 9776 LIST_HEAD(net_todo_list); 9777 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9778 atomic_t dev_unreg_count = ATOMIC_INIT(0); 9779 9780 static void net_set_todo(struct net_device *dev) 9781 { 9782 list_add_tail(&dev->todo_list, &net_todo_list); 9783 } 9784 9785 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9786 struct net_device *upper, netdev_features_t features) 9787 { 9788 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9789 netdev_features_t feature; 9790 int feature_bit; 9791 9792 for_each_netdev_feature(upper_disables, feature_bit) { 9793 feature = __NETIF_F_BIT(feature_bit); 9794 if (!(upper->wanted_features & feature) 9795 && (features & feature)) { 9796 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9797 &feature, upper->name); 9798 features &= ~feature; 9799 } 9800 } 9801 9802 return features; 9803 } 9804 9805 static void netdev_sync_lower_features(struct net_device *upper, 9806 struct net_device *lower, netdev_features_t features) 9807 { 9808 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9809 netdev_features_t feature; 9810 int feature_bit; 9811 9812 for_each_netdev_feature(upper_disables, feature_bit) { 9813 feature = __NETIF_F_BIT(feature_bit); 9814 if (!(features & feature) && (lower->features & feature)) { 9815 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9816 &feature, lower->name); 9817 lower->wanted_features &= ~feature; 9818 __netdev_update_features(lower); 9819 9820 if (unlikely(lower->features & feature)) 9821 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9822 &feature, lower->name); 9823 else 9824 netdev_features_change(lower); 9825 } 9826 } 9827 } 9828 9829 static netdev_features_t netdev_fix_features(struct net_device *dev, 9830 netdev_features_t features) 9831 { 9832 /* Fix illegal checksum combinations */ 9833 if ((features & NETIF_F_HW_CSUM) && 9834 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9835 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9836 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9837 } 9838 9839 /* TSO requires that SG is present as well. */ 9840 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9841 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9842 features &= ~NETIF_F_ALL_TSO; 9843 } 9844 9845 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9846 !(features & NETIF_F_IP_CSUM)) { 9847 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9848 features &= ~NETIF_F_TSO; 9849 features &= ~NETIF_F_TSO_ECN; 9850 } 9851 9852 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9853 !(features & NETIF_F_IPV6_CSUM)) { 9854 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9855 features &= ~NETIF_F_TSO6; 9856 } 9857 9858 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9859 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9860 features &= ~NETIF_F_TSO_MANGLEID; 9861 9862 /* TSO ECN requires that TSO is present as well. */ 9863 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9864 features &= ~NETIF_F_TSO_ECN; 9865 9866 /* Software GSO depends on SG. */ 9867 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9868 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9869 features &= ~NETIF_F_GSO; 9870 } 9871 9872 /* GSO partial features require GSO partial be set */ 9873 if ((features & dev->gso_partial_features) && 9874 !(features & NETIF_F_GSO_PARTIAL)) { 9875 netdev_dbg(dev, 9876 "Dropping partially supported GSO features since no GSO partial.\n"); 9877 features &= ~dev->gso_partial_features; 9878 } 9879 9880 if (!(features & NETIF_F_RXCSUM)) { 9881 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9882 * successfully merged by hardware must also have the 9883 * checksum verified by hardware. If the user does not 9884 * want to enable RXCSUM, logically, we should disable GRO_HW. 9885 */ 9886 if (features & NETIF_F_GRO_HW) { 9887 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9888 features &= ~NETIF_F_GRO_HW; 9889 } 9890 } 9891 9892 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9893 if (features & NETIF_F_RXFCS) { 9894 if (features & NETIF_F_LRO) { 9895 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9896 features &= ~NETIF_F_LRO; 9897 } 9898 9899 if (features & NETIF_F_GRO_HW) { 9900 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9901 features &= ~NETIF_F_GRO_HW; 9902 } 9903 } 9904 9905 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9906 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9907 features &= ~NETIF_F_LRO; 9908 } 9909 9910 if (features & NETIF_F_HW_TLS_TX) { 9911 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9912 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9913 bool hw_csum = features & NETIF_F_HW_CSUM; 9914 9915 if (!ip_csum && !hw_csum) { 9916 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9917 features &= ~NETIF_F_HW_TLS_TX; 9918 } 9919 } 9920 9921 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9922 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9923 features &= ~NETIF_F_HW_TLS_RX; 9924 } 9925 9926 return features; 9927 } 9928 9929 int __netdev_update_features(struct net_device *dev) 9930 { 9931 struct net_device *upper, *lower; 9932 netdev_features_t features; 9933 struct list_head *iter; 9934 int err = -1; 9935 9936 ASSERT_RTNL(); 9937 9938 features = netdev_get_wanted_features(dev); 9939 9940 if (dev->netdev_ops->ndo_fix_features) 9941 features = dev->netdev_ops->ndo_fix_features(dev, features); 9942 9943 /* driver might be less strict about feature dependencies */ 9944 features = netdev_fix_features(dev, features); 9945 9946 /* some features can't be enabled if they're off on an upper device */ 9947 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9948 features = netdev_sync_upper_features(dev, upper, features); 9949 9950 if (dev->features == features) 9951 goto sync_lower; 9952 9953 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9954 &dev->features, &features); 9955 9956 if (dev->netdev_ops->ndo_set_features) 9957 err = dev->netdev_ops->ndo_set_features(dev, features); 9958 else 9959 err = 0; 9960 9961 if (unlikely(err < 0)) { 9962 netdev_err(dev, 9963 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9964 err, &features, &dev->features); 9965 /* return non-0 since some features might have changed and 9966 * it's better to fire a spurious notification than miss it 9967 */ 9968 return -1; 9969 } 9970 9971 sync_lower: 9972 /* some features must be disabled on lower devices when disabled 9973 * on an upper device (think: bonding master or bridge) 9974 */ 9975 netdev_for_each_lower_dev(dev, lower, iter) 9976 netdev_sync_lower_features(dev, lower, features); 9977 9978 if (!err) { 9979 netdev_features_t diff = features ^ dev->features; 9980 9981 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9982 /* udp_tunnel_{get,drop}_rx_info both need 9983 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9984 * device, or they won't do anything. 9985 * Thus we need to update dev->features 9986 * *before* calling udp_tunnel_get_rx_info, 9987 * but *after* calling udp_tunnel_drop_rx_info. 9988 */ 9989 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9990 dev->features = features; 9991 udp_tunnel_get_rx_info(dev); 9992 } else { 9993 udp_tunnel_drop_rx_info(dev); 9994 } 9995 } 9996 9997 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9998 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9999 dev->features = features; 10000 err |= vlan_get_rx_ctag_filter_info(dev); 10001 } else { 10002 vlan_drop_rx_ctag_filter_info(dev); 10003 } 10004 } 10005 10006 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10007 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10008 dev->features = features; 10009 err |= vlan_get_rx_stag_filter_info(dev); 10010 } else { 10011 vlan_drop_rx_stag_filter_info(dev); 10012 } 10013 } 10014 10015 dev->features = features; 10016 } 10017 10018 return err < 0 ? 0 : 1; 10019 } 10020 10021 /** 10022 * netdev_update_features - recalculate device features 10023 * @dev: the device to check 10024 * 10025 * Recalculate dev->features set and send notifications if it 10026 * has changed. Should be called after driver or hardware dependent 10027 * conditions might have changed that influence the features. 10028 */ 10029 void netdev_update_features(struct net_device *dev) 10030 { 10031 if (__netdev_update_features(dev)) 10032 netdev_features_change(dev); 10033 } 10034 EXPORT_SYMBOL(netdev_update_features); 10035 10036 /** 10037 * netdev_change_features - recalculate device features 10038 * @dev: the device to check 10039 * 10040 * Recalculate dev->features set and send notifications even 10041 * if they have not changed. Should be called instead of 10042 * netdev_update_features() if also dev->vlan_features might 10043 * have changed to allow the changes to be propagated to stacked 10044 * VLAN devices. 10045 */ 10046 void netdev_change_features(struct net_device *dev) 10047 { 10048 __netdev_update_features(dev); 10049 netdev_features_change(dev); 10050 } 10051 EXPORT_SYMBOL(netdev_change_features); 10052 10053 /** 10054 * netif_stacked_transfer_operstate - transfer operstate 10055 * @rootdev: the root or lower level device to transfer state from 10056 * @dev: the device to transfer operstate to 10057 * 10058 * Transfer operational state from root to device. This is normally 10059 * called when a stacking relationship exists between the root 10060 * device and the device(a leaf device). 10061 */ 10062 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10063 struct net_device *dev) 10064 { 10065 if (rootdev->operstate == IF_OPER_DORMANT) 10066 netif_dormant_on(dev); 10067 else 10068 netif_dormant_off(dev); 10069 10070 if (rootdev->operstate == IF_OPER_TESTING) 10071 netif_testing_on(dev); 10072 else 10073 netif_testing_off(dev); 10074 10075 if (netif_carrier_ok(rootdev)) 10076 netif_carrier_on(dev); 10077 else 10078 netif_carrier_off(dev); 10079 } 10080 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10081 10082 static int netif_alloc_rx_queues(struct net_device *dev) 10083 { 10084 unsigned int i, count = dev->num_rx_queues; 10085 struct netdev_rx_queue *rx; 10086 size_t sz = count * sizeof(*rx); 10087 int err = 0; 10088 10089 BUG_ON(count < 1); 10090 10091 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10092 if (!rx) 10093 return -ENOMEM; 10094 10095 dev->_rx = rx; 10096 10097 for (i = 0; i < count; i++) { 10098 rx[i].dev = dev; 10099 10100 /* XDP RX-queue setup */ 10101 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10102 if (err < 0) 10103 goto err_rxq_info; 10104 } 10105 return 0; 10106 10107 err_rxq_info: 10108 /* Rollback successful reg's and free other resources */ 10109 while (i--) 10110 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10111 kvfree(dev->_rx); 10112 dev->_rx = NULL; 10113 return err; 10114 } 10115 10116 static void netif_free_rx_queues(struct net_device *dev) 10117 { 10118 unsigned int i, count = dev->num_rx_queues; 10119 10120 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10121 if (!dev->_rx) 10122 return; 10123 10124 for (i = 0; i < count; i++) 10125 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10126 10127 kvfree(dev->_rx); 10128 } 10129 10130 static void netdev_init_one_queue(struct net_device *dev, 10131 struct netdev_queue *queue, void *_unused) 10132 { 10133 /* Initialize queue lock */ 10134 spin_lock_init(&queue->_xmit_lock); 10135 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10136 queue->xmit_lock_owner = -1; 10137 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10138 queue->dev = dev; 10139 #ifdef CONFIG_BQL 10140 dql_init(&queue->dql, HZ); 10141 #endif 10142 } 10143 10144 static void netif_free_tx_queues(struct net_device *dev) 10145 { 10146 kvfree(dev->_tx); 10147 } 10148 10149 static int netif_alloc_netdev_queues(struct net_device *dev) 10150 { 10151 unsigned int count = dev->num_tx_queues; 10152 struct netdev_queue *tx; 10153 size_t sz = count * sizeof(*tx); 10154 10155 if (count < 1 || count > 0xffff) 10156 return -EINVAL; 10157 10158 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10159 if (!tx) 10160 return -ENOMEM; 10161 10162 dev->_tx = tx; 10163 10164 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10165 spin_lock_init(&dev->tx_global_lock); 10166 10167 return 0; 10168 } 10169 10170 void netif_tx_stop_all_queues(struct net_device *dev) 10171 { 10172 unsigned int i; 10173 10174 for (i = 0; i < dev->num_tx_queues; i++) { 10175 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10176 10177 netif_tx_stop_queue(txq); 10178 } 10179 } 10180 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10181 10182 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10183 { 10184 void __percpu *v; 10185 10186 /* Drivers implementing ndo_get_peer_dev must support tstat 10187 * accounting, so that skb_do_redirect() can bump the dev's 10188 * RX stats upon network namespace switch. 10189 */ 10190 if (dev->netdev_ops->ndo_get_peer_dev && 10191 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10192 return -EOPNOTSUPP; 10193 10194 switch (dev->pcpu_stat_type) { 10195 case NETDEV_PCPU_STAT_NONE: 10196 return 0; 10197 case NETDEV_PCPU_STAT_LSTATS: 10198 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10199 break; 10200 case NETDEV_PCPU_STAT_TSTATS: 10201 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10202 break; 10203 case NETDEV_PCPU_STAT_DSTATS: 10204 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10205 break; 10206 default: 10207 return -EINVAL; 10208 } 10209 10210 return v ? 0 : -ENOMEM; 10211 } 10212 10213 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10214 { 10215 switch (dev->pcpu_stat_type) { 10216 case NETDEV_PCPU_STAT_NONE: 10217 return; 10218 case NETDEV_PCPU_STAT_LSTATS: 10219 free_percpu(dev->lstats); 10220 break; 10221 case NETDEV_PCPU_STAT_TSTATS: 10222 free_percpu(dev->tstats); 10223 break; 10224 case NETDEV_PCPU_STAT_DSTATS: 10225 free_percpu(dev->dstats); 10226 break; 10227 } 10228 } 10229 10230 /** 10231 * register_netdevice() - register a network device 10232 * @dev: device to register 10233 * 10234 * Take a prepared network device structure and make it externally accessible. 10235 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10236 * Callers must hold the rtnl lock - you may want register_netdev() 10237 * instead of this. 10238 */ 10239 int register_netdevice(struct net_device *dev) 10240 { 10241 int ret; 10242 struct net *net = dev_net(dev); 10243 10244 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10245 NETDEV_FEATURE_COUNT); 10246 BUG_ON(dev_boot_phase); 10247 ASSERT_RTNL(); 10248 10249 might_sleep(); 10250 10251 /* When net_device's are persistent, this will be fatal. */ 10252 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10253 BUG_ON(!net); 10254 10255 ret = ethtool_check_ops(dev->ethtool_ops); 10256 if (ret) 10257 return ret; 10258 10259 spin_lock_init(&dev->addr_list_lock); 10260 netdev_set_addr_lockdep_class(dev); 10261 10262 ret = dev_get_valid_name(net, dev, dev->name); 10263 if (ret < 0) 10264 goto out; 10265 10266 ret = -ENOMEM; 10267 dev->name_node = netdev_name_node_head_alloc(dev); 10268 if (!dev->name_node) 10269 goto out; 10270 10271 /* Init, if this function is available */ 10272 if (dev->netdev_ops->ndo_init) { 10273 ret = dev->netdev_ops->ndo_init(dev); 10274 if (ret) { 10275 if (ret > 0) 10276 ret = -EIO; 10277 goto err_free_name; 10278 } 10279 } 10280 10281 if (((dev->hw_features | dev->features) & 10282 NETIF_F_HW_VLAN_CTAG_FILTER) && 10283 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10284 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10285 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10286 ret = -EINVAL; 10287 goto err_uninit; 10288 } 10289 10290 ret = netdev_do_alloc_pcpu_stats(dev); 10291 if (ret) 10292 goto err_uninit; 10293 10294 ret = dev_index_reserve(net, dev->ifindex); 10295 if (ret < 0) 10296 goto err_free_pcpu; 10297 dev->ifindex = ret; 10298 10299 /* Transfer changeable features to wanted_features and enable 10300 * software offloads (GSO and GRO). 10301 */ 10302 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10303 dev->features |= NETIF_F_SOFT_FEATURES; 10304 10305 if (dev->udp_tunnel_nic_info) { 10306 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10307 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10308 } 10309 10310 dev->wanted_features = dev->features & dev->hw_features; 10311 10312 if (!(dev->flags & IFF_LOOPBACK)) 10313 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10314 10315 /* If IPv4 TCP segmentation offload is supported we should also 10316 * allow the device to enable segmenting the frame with the option 10317 * of ignoring a static IP ID value. This doesn't enable the 10318 * feature itself but allows the user to enable it later. 10319 */ 10320 if (dev->hw_features & NETIF_F_TSO) 10321 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10322 if (dev->vlan_features & NETIF_F_TSO) 10323 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10324 if (dev->mpls_features & NETIF_F_TSO) 10325 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10326 if (dev->hw_enc_features & NETIF_F_TSO) 10327 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10328 10329 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10330 */ 10331 dev->vlan_features |= NETIF_F_HIGHDMA; 10332 10333 /* Make NETIF_F_SG inheritable to tunnel devices. 10334 */ 10335 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10336 10337 /* Make NETIF_F_SG inheritable to MPLS. 10338 */ 10339 dev->mpls_features |= NETIF_F_SG; 10340 10341 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10342 ret = notifier_to_errno(ret); 10343 if (ret) 10344 goto err_ifindex_release; 10345 10346 ret = netdev_register_kobject(dev); 10347 10348 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); 10349 10350 if (ret) 10351 goto err_uninit_notify; 10352 10353 __netdev_update_features(dev); 10354 10355 /* 10356 * Default initial state at registry is that the 10357 * device is present. 10358 */ 10359 10360 set_bit(__LINK_STATE_PRESENT, &dev->state); 10361 10362 linkwatch_init_dev(dev); 10363 10364 dev_init_scheduler(dev); 10365 10366 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10367 list_netdevice(dev); 10368 10369 add_device_randomness(dev->dev_addr, dev->addr_len); 10370 10371 /* If the device has permanent device address, driver should 10372 * set dev_addr and also addr_assign_type should be set to 10373 * NET_ADDR_PERM (default value). 10374 */ 10375 if (dev->addr_assign_type == NET_ADDR_PERM) 10376 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10377 10378 /* Notify protocols, that a new device appeared. */ 10379 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10380 ret = notifier_to_errno(ret); 10381 if (ret) { 10382 /* Expect explicit free_netdev() on failure */ 10383 dev->needs_free_netdev = false; 10384 unregister_netdevice_queue(dev, NULL); 10385 goto out; 10386 } 10387 /* 10388 * Prevent userspace races by waiting until the network 10389 * device is fully setup before sending notifications. 10390 */ 10391 if (!dev->rtnl_link_ops || 10392 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10393 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10394 10395 out: 10396 return ret; 10397 10398 err_uninit_notify: 10399 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10400 err_ifindex_release: 10401 dev_index_release(net, dev->ifindex); 10402 err_free_pcpu: 10403 netdev_do_free_pcpu_stats(dev); 10404 err_uninit: 10405 if (dev->netdev_ops->ndo_uninit) 10406 dev->netdev_ops->ndo_uninit(dev); 10407 if (dev->priv_destructor) 10408 dev->priv_destructor(dev); 10409 err_free_name: 10410 netdev_name_node_free(dev->name_node); 10411 goto out; 10412 } 10413 EXPORT_SYMBOL(register_netdevice); 10414 10415 /** 10416 * init_dummy_netdev - init a dummy network device for NAPI 10417 * @dev: device to init 10418 * 10419 * This takes a network device structure and initialize the minimum 10420 * amount of fields so it can be used to schedule NAPI polls without 10421 * registering a full blown interface. This is to be used by drivers 10422 * that need to tie several hardware interfaces to a single NAPI 10423 * poll scheduler due to HW limitations. 10424 */ 10425 void init_dummy_netdev(struct net_device *dev) 10426 { 10427 /* Clear everything. Note we don't initialize spinlocks 10428 * are they aren't supposed to be taken by any of the 10429 * NAPI code and this dummy netdev is supposed to be 10430 * only ever used for NAPI polls 10431 */ 10432 memset(dev, 0, sizeof(struct net_device)); 10433 10434 /* make sure we BUG if trying to hit standard 10435 * register/unregister code path 10436 */ 10437 dev->reg_state = NETREG_DUMMY; 10438 10439 /* NAPI wants this */ 10440 INIT_LIST_HEAD(&dev->napi_list); 10441 10442 /* a dummy interface is started by default */ 10443 set_bit(__LINK_STATE_PRESENT, &dev->state); 10444 set_bit(__LINK_STATE_START, &dev->state); 10445 10446 /* napi_busy_loop stats accounting wants this */ 10447 dev_net_set(dev, &init_net); 10448 10449 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10450 * because users of this 'device' dont need to change 10451 * its refcount. 10452 */ 10453 } 10454 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10455 10456 10457 /** 10458 * register_netdev - register a network device 10459 * @dev: device to register 10460 * 10461 * Take a completed network device structure and add it to the kernel 10462 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10463 * chain. 0 is returned on success. A negative errno code is returned 10464 * on a failure to set up the device, or if the name is a duplicate. 10465 * 10466 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10467 * and expands the device name if you passed a format string to 10468 * alloc_netdev. 10469 */ 10470 int register_netdev(struct net_device *dev) 10471 { 10472 int err; 10473 10474 if (rtnl_lock_killable()) 10475 return -EINTR; 10476 err = register_netdevice(dev); 10477 rtnl_unlock(); 10478 return err; 10479 } 10480 EXPORT_SYMBOL(register_netdev); 10481 10482 int netdev_refcnt_read(const struct net_device *dev) 10483 { 10484 #ifdef CONFIG_PCPU_DEV_REFCNT 10485 int i, refcnt = 0; 10486 10487 for_each_possible_cpu(i) 10488 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10489 return refcnt; 10490 #else 10491 return refcount_read(&dev->dev_refcnt); 10492 #endif 10493 } 10494 EXPORT_SYMBOL(netdev_refcnt_read); 10495 10496 int netdev_unregister_timeout_secs __read_mostly = 10; 10497 10498 #define WAIT_REFS_MIN_MSECS 1 10499 #define WAIT_REFS_MAX_MSECS 250 10500 /** 10501 * netdev_wait_allrefs_any - wait until all references are gone. 10502 * @list: list of net_devices to wait on 10503 * 10504 * This is called when unregistering network devices. 10505 * 10506 * Any protocol or device that holds a reference should register 10507 * for netdevice notification, and cleanup and put back the 10508 * reference if they receive an UNREGISTER event. 10509 * We can get stuck here if buggy protocols don't correctly 10510 * call dev_put. 10511 */ 10512 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10513 { 10514 unsigned long rebroadcast_time, warning_time; 10515 struct net_device *dev; 10516 int wait = 0; 10517 10518 rebroadcast_time = warning_time = jiffies; 10519 10520 list_for_each_entry(dev, list, todo_list) 10521 if (netdev_refcnt_read(dev) == 1) 10522 return dev; 10523 10524 while (true) { 10525 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10526 rtnl_lock(); 10527 10528 /* Rebroadcast unregister notification */ 10529 list_for_each_entry(dev, list, todo_list) 10530 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10531 10532 __rtnl_unlock(); 10533 rcu_barrier(); 10534 rtnl_lock(); 10535 10536 list_for_each_entry(dev, list, todo_list) 10537 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10538 &dev->state)) { 10539 /* We must not have linkwatch events 10540 * pending on unregister. If this 10541 * happens, we simply run the queue 10542 * unscheduled, resulting in a noop 10543 * for this device. 10544 */ 10545 linkwatch_run_queue(); 10546 break; 10547 } 10548 10549 __rtnl_unlock(); 10550 10551 rebroadcast_time = jiffies; 10552 } 10553 10554 if (!wait) { 10555 rcu_barrier(); 10556 wait = WAIT_REFS_MIN_MSECS; 10557 } else { 10558 msleep(wait); 10559 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10560 } 10561 10562 list_for_each_entry(dev, list, todo_list) 10563 if (netdev_refcnt_read(dev) == 1) 10564 return dev; 10565 10566 if (time_after(jiffies, warning_time + 10567 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10568 list_for_each_entry(dev, list, todo_list) { 10569 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10570 dev->name, netdev_refcnt_read(dev)); 10571 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10572 } 10573 10574 warning_time = jiffies; 10575 } 10576 } 10577 } 10578 10579 /* The sequence is: 10580 * 10581 * rtnl_lock(); 10582 * ... 10583 * register_netdevice(x1); 10584 * register_netdevice(x2); 10585 * ... 10586 * unregister_netdevice(y1); 10587 * unregister_netdevice(y2); 10588 * ... 10589 * rtnl_unlock(); 10590 * free_netdev(y1); 10591 * free_netdev(y2); 10592 * 10593 * We are invoked by rtnl_unlock(). 10594 * This allows us to deal with problems: 10595 * 1) We can delete sysfs objects which invoke hotplug 10596 * without deadlocking with linkwatch via keventd. 10597 * 2) Since we run with the RTNL semaphore not held, we can sleep 10598 * safely in order to wait for the netdev refcnt to drop to zero. 10599 * 10600 * We must not return until all unregister events added during 10601 * the interval the lock was held have been completed. 10602 */ 10603 void netdev_run_todo(void) 10604 { 10605 struct net_device *dev, *tmp; 10606 struct list_head list; 10607 int cnt; 10608 #ifdef CONFIG_LOCKDEP 10609 struct list_head unlink_list; 10610 10611 list_replace_init(&net_unlink_list, &unlink_list); 10612 10613 while (!list_empty(&unlink_list)) { 10614 struct net_device *dev = list_first_entry(&unlink_list, 10615 struct net_device, 10616 unlink_list); 10617 list_del_init(&dev->unlink_list); 10618 dev->nested_level = dev->lower_level - 1; 10619 } 10620 #endif 10621 10622 /* Snapshot list, allow later requests */ 10623 list_replace_init(&net_todo_list, &list); 10624 10625 __rtnl_unlock(); 10626 10627 /* Wait for rcu callbacks to finish before next phase */ 10628 if (!list_empty(&list)) 10629 rcu_barrier(); 10630 10631 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10632 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10633 netdev_WARN(dev, "run_todo but not unregistering\n"); 10634 list_del(&dev->todo_list); 10635 continue; 10636 } 10637 10638 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); 10639 linkwatch_sync_dev(dev); 10640 } 10641 10642 cnt = 0; 10643 while (!list_empty(&list)) { 10644 dev = netdev_wait_allrefs_any(&list); 10645 list_del(&dev->todo_list); 10646 10647 /* paranoia */ 10648 BUG_ON(netdev_refcnt_read(dev) != 1); 10649 BUG_ON(!list_empty(&dev->ptype_all)); 10650 BUG_ON(!list_empty(&dev->ptype_specific)); 10651 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10652 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10653 10654 netdev_do_free_pcpu_stats(dev); 10655 if (dev->priv_destructor) 10656 dev->priv_destructor(dev); 10657 if (dev->needs_free_netdev) 10658 free_netdev(dev); 10659 10660 cnt++; 10661 10662 /* Free network device */ 10663 kobject_put(&dev->dev.kobj); 10664 } 10665 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) 10666 wake_up(&netdev_unregistering_wq); 10667 } 10668 10669 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10670 * all the same fields in the same order as net_device_stats, with only 10671 * the type differing, but rtnl_link_stats64 may have additional fields 10672 * at the end for newer counters. 10673 */ 10674 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10675 const struct net_device_stats *netdev_stats) 10676 { 10677 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10678 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10679 u64 *dst = (u64 *)stats64; 10680 10681 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10682 for (i = 0; i < n; i++) 10683 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10684 /* zero out counters that only exist in rtnl_link_stats64 */ 10685 memset((char *)stats64 + n * sizeof(u64), 0, 10686 sizeof(*stats64) - n * sizeof(u64)); 10687 } 10688 EXPORT_SYMBOL(netdev_stats_to_stats64); 10689 10690 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 10691 struct net_device *dev) 10692 { 10693 struct net_device_core_stats __percpu *p; 10694 10695 p = alloc_percpu_gfp(struct net_device_core_stats, 10696 GFP_ATOMIC | __GFP_NOWARN); 10697 10698 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10699 free_percpu(p); 10700 10701 /* This READ_ONCE() pairs with the cmpxchg() above */ 10702 return READ_ONCE(dev->core_stats); 10703 } 10704 10705 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 10706 { 10707 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10708 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 10709 unsigned long __percpu *field; 10710 10711 if (unlikely(!p)) { 10712 p = netdev_core_stats_alloc(dev); 10713 if (!p) 10714 return; 10715 } 10716 10717 field = (__force unsigned long __percpu *)((__force void *)p + offset); 10718 this_cpu_inc(*field); 10719 } 10720 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 10721 10722 /** 10723 * dev_get_stats - get network device statistics 10724 * @dev: device to get statistics from 10725 * @storage: place to store stats 10726 * 10727 * Get network statistics from device. Return @storage. 10728 * The device driver may provide its own method by setting 10729 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10730 * otherwise the internal statistics structure is used. 10731 */ 10732 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10733 struct rtnl_link_stats64 *storage) 10734 { 10735 const struct net_device_ops *ops = dev->netdev_ops; 10736 const struct net_device_core_stats __percpu *p; 10737 10738 if (ops->ndo_get_stats64) { 10739 memset(storage, 0, sizeof(*storage)); 10740 ops->ndo_get_stats64(dev, storage); 10741 } else if (ops->ndo_get_stats) { 10742 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10743 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { 10744 dev_get_tstats64(dev, storage); 10745 } else { 10746 netdev_stats_to_stats64(storage, &dev->stats); 10747 } 10748 10749 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10750 p = READ_ONCE(dev->core_stats); 10751 if (p) { 10752 const struct net_device_core_stats *core_stats; 10753 int i; 10754 10755 for_each_possible_cpu(i) { 10756 core_stats = per_cpu_ptr(p, i); 10757 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10758 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10759 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10760 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10761 } 10762 } 10763 return storage; 10764 } 10765 EXPORT_SYMBOL(dev_get_stats); 10766 10767 /** 10768 * dev_fetch_sw_netstats - get per-cpu network device statistics 10769 * @s: place to store stats 10770 * @netstats: per-cpu network stats to read from 10771 * 10772 * Read per-cpu network statistics and populate the related fields in @s. 10773 */ 10774 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10775 const struct pcpu_sw_netstats __percpu *netstats) 10776 { 10777 int cpu; 10778 10779 for_each_possible_cpu(cpu) { 10780 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10781 const struct pcpu_sw_netstats *stats; 10782 unsigned int start; 10783 10784 stats = per_cpu_ptr(netstats, cpu); 10785 do { 10786 start = u64_stats_fetch_begin(&stats->syncp); 10787 rx_packets = u64_stats_read(&stats->rx_packets); 10788 rx_bytes = u64_stats_read(&stats->rx_bytes); 10789 tx_packets = u64_stats_read(&stats->tx_packets); 10790 tx_bytes = u64_stats_read(&stats->tx_bytes); 10791 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10792 10793 s->rx_packets += rx_packets; 10794 s->rx_bytes += rx_bytes; 10795 s->tx_packets += tx_packets; 10796 s->tx_bytes += tx_bytes; 10797 } 10798 } 10799 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10800 10801 /** 10802 * dev_get_tstats64 - ndo_get_stats64 implementation 10803 * @dev: device to get statistics from 10804 * @s: place to store stats 10805 * 10806 * Populate @s from dev->stats and dev->tstats. Can be used as 10807 * ndo_get_stats64() callback. 10808 */ 10809 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10810 { 10811 netdev_stats_to_stats64(s, &dev->stats); 10812 dev_fetch_sw_netstats(s, dev->tstats); 10813 } 10814 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10815 10816 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10817 { 10818 struct netdev_queue *queue = dev_ingress_queue(dev); 10819 10820 #ifdef CONFIG_NET_CLS_ACT 10821 if (queue) 10822 return queue; 10823 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10824 if (!queue) 10825 return NULL; 10826 netdev_init_one_queue(dev, queue, NULL); 10827 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10828 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10829 rcu_assign_pointer(dev->ingress_queue, queue); 10830 #endif 10831 return queue; 10832 } 10833 10834 static const struct ethtool_ops default_ethtool_ops; 10835 10836 void netdev_set_default_ethtool_ops(struct net_device *dev, 10837 const struct ethtool_ops *ops) 10838 { 10839 if (dev->ethtool_ops == &default_ethtool_ops) 10840 dev->ethtool_ops = ops; 10841 } 10842 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10843 10844 /** 10845 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10846 * @dev: netdev to enable the IRQ coalescing on 10847 * 10848 * Sets a conservative default for SW IRQ coalescing. Users can use 10849 * sysfs attributes to override the default values. 10850 */ 10851 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10852 { 10853 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10854 10855 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10856 dev->gro_flush_timeout = 20000; 10857 dev->napi_defer_hard_irqs = 1; 10858 } 10859 } 10860 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10861 10862 void netdev_freemem(struct net_device *dev) 10863 { 10864 char *addr = (char *)dev - dev->padded; 10865 10866 kvfree(addr); 10867 } 10868 10869 /** 10870 * alloc_netdev_mqs - allocate network device 10871 * @sizeof_priv: size of private data to allocate space for 10872 * @name: device name format string 10873 * @name_assign_type: origin of device name 10874 * @setup: callback to initialize device 10875 * @txqs: the number of TX subqueues to allocate 10876 * @rxqs: the number of RX subqueues to allocate 10877 * 10878 * Allocates a struct net_device with private data area for driver use 10879 * and performs basic initialization. Also allocates subqueue structs 10880 * for each queue on the device. 10881 */ 10882 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10883 unsigned char name_assign_type, 10884 void (*setup)(struct net_device *), 10885 unsigned int txqs, unsigned int rxqs) 10886 { 10887 struct net_device *dev; 10888 unsigned int alloc_size; 10889 struct net_device *p; 10890 10891 BUG_ON(strlen(name) >= sizeof(dev->name)); 10892 10893 if (txqs < 1) { 10894 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10895 return NULL; 10896 } 10897 10898 if (rxqs < 1) { 10899 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10900 return NULL; 10901 } 10902 10903 alloc_size = sizeof(struct net_device); 10904 if (sizeof_priv) { 10905 /* ensure 32-byte alignment of private area */ 10906 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10907 alloc_size += sizeof_priv; 10908 } 10909 /* ensure 32-byte alignment of whole construct */ 10910 alloc_size += NETDEV_ALIGN - 1; 10911 10912 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10913 if (!p) 10914 return NULL; 10915 10916 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10917 dev->padded = (char *)dev - (char *)p; 10918 10919 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10920 #ifdef CONFIG_PCPU_DEV_REFCNT 10921 dev->pcpu_refcnt = alloc_percpu(int); 10922 if (!dev->pcpu_refcnt) 10923 goto free_dev; 10924 __dev_hold(dev); 10925 #else 10926 refcount_set(&dev->dev_refcnt, 1); 10927 #endif 10928 10929 if (dev_addr_init(dev)) 10930 goto free_pcpu; 10931 10932 dev_mc_init(dev); 10933 dev_uc_init(dev); 10934 10935 dev_net_set(dev, &init_net); 10936 10937 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10938 dev->xdp_zc_max_segs = 1; 10939 dev->gso_max_segs = GSO_MAX_SEGS; 10940 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10941 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10942 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10943 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10944 dev->tso_max_segs = TSO_MAX_SEGS; 10945 dev->upper_level = 1; 10946 dev->lower_level = 1; 10947 #ifdef CONFIG_LOCKDEP 10948 dev->nested_level = 0; 10949 INIT_LIST_HEAD(&dev->unlink_list); 10950 #endif 10951 10952 INIT_LIST_HEAD(&dev->napi_list); 10953 INIT_LIST_HEAD(&dev->unreg_list); 10954 INIT_LIST_HEAD(&dev->close_list); 10955 INIT_LIST_HEAD(&dev->link_watch_list); 10956 INIT_LIST_HEAD(&dev->adj_list.upper); 10957 INIT_LIST_HEAD(&dev->adj_list.lower); 10958 INIT_LIST_HEAD(&dev->ptype_all); 10959 INIT_LIST_HEAD(&dev->ptype_specific); 10960 INIT_LIST_HEAD(&dev->net_notifier_list); 10961 #ifdef CONFIG_NET_SCHED 10962 hash_init(dev->qdisc_hash); 10963 #endif 10964 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10965 setup(dev); 10966 10967 if (!dev->tx_queue_len) { 10968 dev->priv_flags |= IFF_NO_QUEUE; 10969 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10970 } 10971 10972 dev->num_tx_queues = txqs; 10973 dev->real_num_tx_queues = txqs; 10974 if (netif_alloc_netdev_queues(dev)) 10975 goto free_all; 10976 10977 dev->num_rx_queues = rxqs; 10978 dev->real_num_rx_queues = rxqs; 10979 if (netif_alloc_rx_queues(dev)) 10980 goto free_all; 10981 10982 strcpy(dev->name, name); 10983 dev->name_assign_type = name_assign_type; 10984 dev->group = INIT_NETDEV_GROUP; 10985 if (!dev->ethtool_ops) 10986 dev->ethtool_ops = &default_ethtool_ops; 10987 10988 nf_hook_netdev_init(dev); 10989 10990 return dev; 10991 10992 free_all: 10993 free_netdev(dev); 10994 return NULL; 10995 10996 free_pcpu: 10997 #ifdef CONFIG_PCPU_DEV_REFCNT 10998 free_percpu(dev->pcpu_refcnt); 10999 free_dev: 11000 #endif 11001 netdev_freemem(dev); 11002 return NULL; 11003 } 11004 EXPORT_SYMBOL(alloc_netdev_mqs); 11005 11006 /** 11007 * free_netdev - free network device 11008 * @dev: device 11009 * 11010 * This function does the last stage of destroying an allocated device 11011 * interface. The reference to the device object is released. If this 11012 * is the last reference then it will be freed.Must be called in process 11013 * context. 11014 */ 11015 void free_netdev(struct net_device *dev) 11016 { 11017 struct napi_struct *p, *n; 11018 11019 might_sleep(); 11020 11021 /* When called immediately after register_netdevice() failed the unwind 11022 * handling may still be dismantling the device. Handle that case by 11023 * deferring the free. 11024 */ 11025 if (dev->reg_state == NETREG_UNREGISTERING) { 11026 ASSERT_RTNL(); 11027 dev->needs_free_netdev = true; 11028 return; 11029 } 11030 11031 netif_free_tx_queues(dev); 11032 netif_free_rx_queues(dev); 11033 11034 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 11035 11036 /* Flush device addresses */ 11037 dev_addr_flush(dev); 11038 11039 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 11040 netif_napi_del(p); 11041 11042 ref_tracker_dir_exit(&dev->refcnt_tracker); 11043 #ifdef CONFIG_PCPU_DEV_REFCNT 11044 free_percpu(dev->pcpu_refcnt); 11045 dev->pcpu_refcnt = NULL; 11046 #endif 11047 free_percpu(dev->core_stats); 11048 dev->core_stats = NULL; 11049 free_percpu(dev->xdp_bulkq); 11050 dev->xdp_bulkq = NULL; 11051 11052 /* Compatibility with error handling in drivers */ 11053 if (dev->reg_state == NETREG_UNINITIALIZED) { 11054 netdev_freemem(dev); 11055 return; 11056 } 11057 11058 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 11059 WRITE_ONCE(dev->reg_state, NETREG_RELEASED); 11060 11061 /* will free via device release */ 11062 put_device(&dev->dev); 11063 } 11064 EXPORT_SYMBOL(free_netdev); 11065 11066 /** 11067 * synchronize_net - Synchronize with packet receive processing 11068 * 11069 * Wait for packets currently being received to be done. 11070 * Does not block later packets from starting. 11071 */ 11072 void synchronize_net(void) 11073 { 11074 might_sleep(); 11075 if (rtnl_is_locked()) 11076 synchronize_rcu_expedited(); 11077 else 11078 synchronize_rcu(); 11079 } 11080 EXPORT_SYMBOL(synchronize_net); 11081 11082 /** 11083 * unregister_netdevice_queue - remove device from the kernel 11084 * @dev: device 11085 * @head: list 11086 * 11087 * This function shuts down a device interface and removes it 11088 * from the kernel tables. 11089 * If head not NULL, device is queued to be unregistered later. 11090 * 11091 * Callers must hold the rtnl semaphore. You may want 11092 * unregister_netdev() instead of this. 11093 */ 11094 11095 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11096 { 11097 ASSERT_RTNL(); 11098 11099 if (head) { 11100 list_move_tail(&dev->unreg_list, head); 11101 } else { 11102 LIST_HEAD(single); 11103 11104 list_add(&dev->unreg_list, &single); 11105 unregister_netdevice_many(&single); 11106 } 11107 } 11108 EXPORT_SYMBOL(unregister_netdevice_queue); 11109 11110 void unregister_netdevice_many_notify(struct list_head *head, 11111 u32 portid, const struct nlmsghdr *nlh) 11112 { 11113 struct net_device *dev, *tmp; 11114 LIST_HEAD(close_head); 11115 int cnt = 0; 11116 11117 BUG_ON(dev_boot_phase); 11118 ASSERT_RTNL(); 11119 11120 if (list_empty(head)) 11121 return; 11122 11123 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11124 /* Some devices call without registering 11125 * for initialization unwind. Remove those 11126 * devices and proceed with the remaining. 11127 */ 11128 if (dev->reg_state == NETREG_UNINITIALIZED) { 11129 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11130 dev->name, dev); 11131 11132 WARN_ON(1); 11133 list_del(&dev->unreg_list); 11134 continue; 11135 } 11136 dev->dismantle = true; 11137 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11138 } 11139 11140 /* If device is running, close it first. */ 11141 list_for_each_entry(dev, head, unreg_list) 11142 list_add_tail(&dev->close_list, &close_head); 11143 dev_close_many(&close_head, true); 11144 11145 list_for_each_entry(dev, head, unreg_list) { 11146 /* And unlink it from device chain. */ 11147 unlist_netdevice(dev); 11148 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); 11149 } 11150 flush_all_backlogs(); 11151 11152 synchronize_net(); 11153 11154 list_for_each_entry(dev, head, unreg_list) { 11155 struct sk_buff *skb = NULL; 11156 11157 /* Shutdown queueing discipline. */ 11158 dev_shutdown(dev); 11159 dev_tcx_uninstall(dev); 11160 dev_xdp_uninstall(dev); 11161 bpf_dev_bound_netdev_unregister(dev); 11162 11163 netdev_offload_xstats_disable_all(dev); 11164 11165 /* Notify protocols, that we are about to destroy 11166 * this device. They should clean all the things. 11167 */ 11168 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11169 11170 if (!dev->rtnl_link_ops || 11171 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11172 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11173 GFP_KERNEL, NULL, 0, 11174 portid, nlh); 11175 11176 /* 11177 * Flush the unicast and multicast chains 11178 */ 11179 dev_uc_flush(dev); 11180 dev_mc_flush(dev); 11181 11182 netdev_name_node_alt_flush(dev); 11183 netdev_name_node_free(dev->name_node); 11184 11185 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11186 11187 if (dev->netdev_ops->ndo_uninit) 11188 dev->netdev_ops->ndo_uninit(dev); 11189 11190 if (skb) 11191 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11192 11193 /* Notifier chain MUST detach us all upper devices. */ 11194 WARN_ON(netdev_has_any_upper_dev(dev)); 11195 WARN_ON(netdev_has_any_lower_dev(dev)); 11196 11197 /* Remove entries from kobject tree */ 11198 netdev_unregister_kobject(dev); 11199 #ifdef CONFIG_XPS 11200 /* Remove XPS queueing entries */ 11201 netif_reset_xps_queues_gt(dev, 0); 11202 #endif 11203 } 11204 11205 synchronize_net(); 11206 11207 list_for_each_entry(dev, head, unreg_list) { 11208 netdev_put(dev, &dev->dev_registered_tracker); 11209 net_set_todo(dev); 11210 cnt++; 11211 } 11212 atomic_add(cnt, &dev_unreg_count); 11213 11214 list_del(head); 11215 } 11216 11217 /** 11218 * unregister_netdevice_many - unregister many devices 11219 * @head: list of devices 11220 * 11221 * Note: As most callers use a stack allocated list_head, 11222 * we force a list_del() to make sure stack wont be corrupted later. 11223 */ 11224 void unregister_netdevice_many(struct list_head *head) 11225 { 11226 unregister_netdevice_many_notify(head, 0, NULL); 11227 } 11228 EXPORT_SYMBOL(unregister_netdevice_many); 11229 11230 /** 11231 * unregister_netdev - remove device from the kernel 11232 * @dev: device 11233 * 11234 * This function shuts down a device interface and removes it 11235 * from the kernel tables. 11236 * 11237 * This is just a wrapper for unregister_netdevice that takes 11238 * the rtnl semaphore. In general you want to use this and not 11239 * unregister_netdevice. 11240 */ 11241 void unregister_netdev(struct net_device *dev) 11242 { 11243 rtnl_lock(); 11244 unregister_netdevice(dev); 11245 rtnl_unlock(); 11246 } 11247 EXPORT_SYMBOL(unregister_netdev); 11248 11249 /** 11250 * __dev_change_net_namespace - move device to different nethost namespace 11251 * @dev: device 11252 * @net: network namespace 11253 * @pat: If not NULL name pattern to try if the current device name 11254 * is already taken in the destination network namespace. 11255 * @new_ifindex: If not zero, specifies device index in the target 11256 * namespace. 11257 * 11258 * This function shuts down a device interface and moves it 11259 * to a new network namespace. On success 0 is returned, on 11260 * a failure a netagive errno code is returned. 11261 * 11262 * Callers must hold the rtnl semaphore. 11263 */ 11264 11265 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11266 const char *pat, int new_ifindex) 11267 { 11268 struct netdev_name_node *name_node; 11269 struct net *net_old = dev_net(dev); 11270 char new_name[IFNAMSIZ] = {}; 11271 int err, new_nsid; 11272 11273 ASSERT_RTNL(); 11274 11275 /* Don't allow namespace local devices to be moved. */ 11276 err = -EINVAL; 11277 if (dev->features & NETIF_F_NETNS_LOCAL) 11278 goto out; 11279 11280 /* Ensure the device has been registrered */ 11281 if (dev->reg_state != NETREG_REGISTERED) 11282 goto out; 11283 11284 /* Get out if there is nothing todo */ 11285 err = 0; 11286 if (net_eq(net_old, net)) 11287 goto out; 11288 11289 /* Pick the destination device name, and ensure 11290 * we can use it in the destination network namespace. 11291 */ 11292 err = -EEXIST; 11293 if (netdev_name_in_use(net, dev->name)) { 11294 /* We get here if we can't use the current device name */ 11295 if (!pat) 11296 goto out; 11297 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11298 if (err < 0) 11299 goto out; 11300 } 11301 /* Check that none of the altnames conflicts. */ 11302 err = -EEXIST; 11303 netdev_for_each_altname(dev, name_node) 11304 if (netdev_name_in_use(net, name_node->name)) 11305 goto out; 11306 11307 /* Check that new_ifindex isn't used yet. */ 11308 if (new_ifindex) { 11309 err = dev_index_reserve(net, new_ifindex); 11310 if (err < 0) 11311 goto out; 11312 } else { 11313 /* If there is an ifindex conflict assign a new one */ 11314 err = dev_index_reserve(net, dev->ifindex); 11315 if (err == -EBUSY) 11316 err = dev_index_reserve(net, 0); 11317 if (err < 0) 11318 goto out; 11319 new_ifindex = err; 11320 } 11321 11322 /* 11323 * And now a mini version of register_netdevice unregister_netdevice. 11324 */ 11325 11326 /* If device is running close it first. */ 11327 dev_close(dev); 11328 11329 /* And unlink it from device chain */ 11330 unlist_netdevice(dev); 11331 11332 synchronize_net(); 11333 11334 /* Shutdown queueing discipline. */ 11335 dev_shutdown(dev); 11336 11337 /* Notify protocols, that we are about to destroy 11338 * this device. They should clean all the things. 11339 * 11340 * Note that dev->reg_state stays at NETREG_REGISTERED. 11341 * This is wanted because this way 8021q and macvlan know 11342 * the device is just moving and can keep their slaves up. 11343 */ 11344 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11345 rcu_barrier(); 11346 11347 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11348 11349 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11350 new_ifindex); 11351 11352 /* 11353 * Flush the unicast and multicast chains 11354 */ 11355 dev_uc_flush(dev); 11356 dev_mc_flush(dev); 11357 11358 /* Send a netdev-removed uevent to the old namespace */ 11359 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11360 netdev_adjacent_del_links(dev); 11361 11362 /* Move per-net netdevice notifiers that are following the netdevice */ 11363 move_netdevice_notifiers_dev_net(dev, net); 11364 11365 /* Actually switch the network namespace */ 11366 dev_net_set(dev, net); 11367 dev->ifindex = new_ifindex; 11368 11369 if (new_name[0]) /* Rename the netdev to prepared name */ 11370 strscpy(dev->name, new_name, IFNAMSIZ); 11371 11372 /* Fixup kobjects */ 11373 dev_set_uevent_suppress(&dev->dev, 1); 11374 err = device_rename(&dev->dev, dev->name); 11375 dev_set_uevent_suppress(&dev->dev, 0); 11376 WARN_ON(err); 11377 11378 /* Send a netdev-add uevent to the new namespace */ 11379 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11380 netdev_adjacent_add_links(dev); 11381 11382 /* Adapt owner in case owning user namespace of target network 11383 * namespace is different from the original one. 11384 */ 11385 err = netdev_change_owner(dev, net_old, net); 11386 WARN_ON(err); 11387 11388 /* Add the device back in the hashes */ 11389 list_netdevice(dev); 11390 11391 /* Notify protocols, that a new device appeared. */ 11392 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11393 11394 /* 11395 * Prevent userspace races by waiting until the network 11396 * device is fully setup before sending notifications. 11397 */ 11398 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11399 11400 synchronize_net(); 11401 err = 0; 11402 out: 11403 return err; 11404 } 11405 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11406 11407 static int dev_cpu_dead(unsigned int oldcpu) 11408 { 11409 struct sk_buff **list_skb; 11410 struct sk_buff *skb; 11411 unsigned int cpu; 11412 struct softnet_data *sd, *oldsd, *remsd = NULL; 11413 11414 local_irq_disable(); 11415 cpu = smp_processor_id(); 11416 sd = &per_cpu(softnet_data, cpu); 11417 oldsd = &per_cpu(softnet_data, oldcpu); 11418 11419 /* Find end of our completion_queue. */ 11420 list_skb = &sd->completion_queue; 11421 while (*list_skb) 11422 list_skb = &(*list_skb)->next; 11423 /* Append completion queue from offline CPU. */ 11424 *list_skb = oldsd->completion_queue; 11425 oldsd->completion_queue = NULL; 11426 11427 /* Append output queue from offline CPU. */ 11428 if (oldsd->output_queue) { 11429 *sd->output_queue_tailp = oldsd->output_queue; 11430 sd->output_queue_tailp = oldsd->output_queue_tailp; 11431 oldsd->output_queue = NULL; 11432 oldsd->output_queue_tailp = &oldsd->output_queue; 11433 } 11434 /* Append NAPI poll list from offline CPU, with one exception : 11435 * process_backlog() must be called by cpu owning percpu backlog. 11436 * We properly handle process_queue & input_pkt_queue later. 11437 */ 11438 while (!list_empty(&oldsd->poll_list)) { 11439 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11440 struct napi_struct, 11441 poll_list); 11442 11443 list_del_init(&napi->poll_list); 11444 if (napi->poll == process_backlog) 11445 napi->state &= NAPIF_STATE_THREADED; 11446 else 11447 ____napi_schedule(sd, napi); 11448 } 11449 11450 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11451 local_irq_enable(); 11452 11453 if (!use_backlog_threads()) { 11454 #ifdef CONFIG_RPS 11455 remsd = oldsd->rps_ipi_list; 11456 oldsd->rps_ipi_list = NULL; 11457 #endif 11458 /* send out pending IPI's on offline CPU */ 11459 net_rps_send_ipi(remsd); 11460 } 11461 11462 /* Process offline CPU's input_pkt_queue */ 11463 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11464 netif_rx(skb); 11465 rps_input_queue_head_incr(oldsd); 11466 } 11467 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11468 netif_rx(skb); 11469 rps_input_queue_head_incr(oldsd); 11470 } 11471 11472 return 0; 11473 } 11474 11475 /** 11476 * netdev_increment_features - increment feature set by one 11477 * @all: current feature set 11478 * @one: new feature set 11479 * @mask: mask feature set 11480 * 11481 * Computes a new feature set after adding a device with feature set 11482 * @one to the master device with current feature set @all. Will not 11483 * enable anything that is off in @mask. Returns the new feature set. 11484 */ 11485 netdev_features_t netdev_increment_features(netdev_features_t all, 11486 netdev_features_t one, netdev_features_t mask) 11487 { 11488 if (mask & NETIF_F_HW_CSUM) 11489 mask |= NETIF_F_CSUM_MASK; 11490 mask |= NETIF_F_VLAN_CHALLENGED; 11491 11492 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11493 all &= one | ~NETIF_F_ALL_FOR_ALL; 11494 11495 /* If one device supports hw checksumming, set for all. */ 11496 if (all & NETIF_F_HW_CSUM) 11497 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11498 11499 return all; 11500 } 11501 EXPORT_SYMBOL(netdev_increment_features); 11502 11503 static struct hlist_head * __net_init netdev_create_hash(void) 11504 { 11505 int i; 11506 struct hlist_head *hash; 11507 11508 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11509 if (hash != NULL) 11510 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11511 INIT_HLIST_HEAD(&hash[i]); 11512 11513 return hash; 11514 } 11515 11516 /* Initialize per network namespace state */ 11517 static int __net_init netdev_init(struct net *net) 11518 { 11519 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11520 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11521 11522 INIT_LIST_HEAD(&net->dev_base_head); 11523 11524 net->dev_name_head = netdev_create_hash(); 11525 if (net->dev_name_head == NULL) 11526 goto err_name; 11527 11528 net->dev_index_head = netdev_create_hash(); 11529 if (net->dev_index_head == NULL) 11530 goto err_idx; 11531 11532 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11533 11534 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11535 11536 return 0; 11537 11538 err_idx: 11539 kfree(net->dev_name_head); 11540 err_name: 11541 return -ENOMEM; 11542 } 11543 11544 /** 11545 * netdev_drivername - network driver for the device 11546 * @dev: network device 11547 * 11548 * Determine network driver for device. 11549 */ 11550 const char *netdev_drivername(const struct net_device *dev) 11551 { 11552 const struct device_driver *driver; 11553 const struct device *parent; 11554 const char *empty = ""; 11555 11556 parent = dev->dev.parent; 11557 if (!parent) 11558 return empty; 11559 11560 driver = parent->driver; 11561 if (driver && driver->name) 11562 return driver->name; 11563 return empty; 11564 } 11565 11566 static void __netdev_printk(const char *level, const struct net_device *dev, 11567 struct va_format *vaf) 11568 { 11569 if (dev && dev->dev.parent) { 11570 dev_printk_emit(level[1] - '0', 11571 dev->dev.parent, 11572 "%s %s %s%s: %pV", 11573 dev_driver_string(dev->dev.parent), 11574 dev_name(dev->dev.parent), 11575 netdev_name(dev), netdev_reg_state(dev), 11576 vaf); 11577 } else if (dev) { 11578 printk("%s%s%s: %pV", 11579 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11580 } else { 11581 printk("%s(NULL net_device): %pV", level, vaf); 11582 } 11583 } 11584 11585 void netdev_printk(const char *level, const struct net_device *dev, 11586 const char *format, ...) 11587 { 11588 struct va_format vaf; 11589 va_list args; 11590 11591 va_start(args, format); 11592 11593 vaf.fmt = format; 11594 vaf.va = &args; 11595 11596 __netdev_printk(level, dev, &vaf); 11597 11598 va_end(args); 11599 } 11600 EXPORT_SYMBOL(netdev_printk); 11601 11602 #define define_netdev_printk_level(func, level) \ 11603 void func(const struct net_device *dev, const char *fmt, ...) \ 11604 { \ 11605 struct va_format vaf; \ 11606 va_list args; \ 11607 \ 11608 va_start(args, fmt); \ 11609 \ 11610 vaf.fmt = fmt; \ 11611 vaf.va = &args; \ 11612 \ 11613 __netdev_printk(level, dev, &vaf); \ 11614 \ 11615 va_end(args); \ 11616 } \ 11617 EXPORT_SYMBOL(func); 11618 11619 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11620 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11621 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11622 define_netdev_printk_level(netdev_err, KERN_ERR); 11623 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11624 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11625 define_netdev_printk_level(netdev_info, KERN_INFO); 11626 11627 static void __net_exit netdev_exit(struct net *net) 11628 { 11629 kfree(net->dev_name_head); 11630 kfree(net->dev_index_head); 11631 xa_destroy(&net->dev_by_index); 11632 if (net != &init_net) 11633 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11634 } 11635 11636 static struct pernet_operations __net_initdata netdev_net_ops = { 11637 .init = netdev_init, 11638 .exit = netdev_exit, 11639 }; 11640 11641 static void __net_exit default_device_exit_net(struct net *net) 11642 { 11643 struct netdev_name_node *name_node, *tmp; 11644 struct net_device *dev, *aux; 11645 /* 11646 * Push all migratable network devices back to the 11647 * initial network namespace 11648 */ 11649 ASSERT_RTNL(); 11650 for_each_netdev_safe(net, dev, aux) { 11651 int err; 11652 char fb_name[IFNAMSIZ]; 11653 11654 /* Ignore unmoveable devices (i.e. loopback) */ 11655 if (dev->features & NETIF_F_NETNS_LOCAL) 11656 continue; 11657 11658 /* Leave virtual devices for the generic cleanup */ 11659 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11660 continue; 11661 11662 /* Push remaining network devices to init_net */ 11663 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11664 if (netdev_name_in_use(&init_net, fb_name)) 11665 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11666 11667 netdev_for_each_altname_safe(dev, name_node, tmp) 11668 if (netdev_name_in_use(&init_net, name_node->name)) 11669 __netdev_name_node_alt_destroy(name_node); 11670 11671 err = dev_change_net_namespace(dev, &init_net, fb_name); 11672 if (err) { 11673 pr_emerg("%s: failed to move %s to init_net: %d\n", 11674 __func__, dev->name, err); 11675 BUG(); 11676 } 11677 } 11678 } 11679 11680 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11681 { 11682 /* At exit all network devices most be removed from a network 11683 * namespace. Do this in the reverse order of registration. 11684 * Do this across as many network namespaces as possible to 11685 * improve batching efficiency. 11686 */ 11687 struct net_device *dev; 11688 struct net *net; 11689 LIST_HEAD(dev_kill_list); 11690 11691 rtnl_lock(); 11692 list_for_each_entry(net, net_list, exit_list) { 11693 default_device_exit_net(net); 11694 cond_resched(); 11695 } 11696 11697 list_for_each_entry(net, net_list, exit_list) { 11698 for_each_netdev_reverse(net, dev) { 11699 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11700 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11701 else 11702 unregister_netdevice_queue(dev, &dev_kill_list); 11703 } 11704 } 11705 unregister_netdevice_many(&dev_kill_list); 11706 rtnl_unlock(); 11707 } 11708 11709 static struct pernet_operations __net_initdata default_device_ops = { 11710 .exit_batch = default_device_exit_batch, 11711 }; 11712 11713 static void __init net_dev_struct_check(void) 11714 { 11715 /* TX read-mostly hotpath */ 11716 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags); 11717 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 11718 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 11719 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 11720 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 11721 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 11722 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 11723 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 11724 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 11725 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 11726 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 11727 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 11728 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 11729 #ifdef CONFIG_XPS 11730 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 11731 #endif 11732 #ifdef CONFIG_NETFILTER_EGRESS 11733 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 11734 #endif 11735 #ifdef CONFIG_NET_XGRESS 11736 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 11737 #endif 11738 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 11739 11740 /* TXRX read-mostly hotpath */ 11741 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); 11742 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state); 11743 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 11744 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 11745 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 11746 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 11747 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46); 11748 11749 /* RX read-mostly hotpath */ 11750 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 11751 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 11752 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 11753 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 11754 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout); 11755 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs); 11756 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 11757 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 11758 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 11759 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 11760 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 11761 #ifdef CONFIG_NETPOLL 11762 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 11763 #endif 11764 #ifdef CONFIG_NET_XGRESS 11765 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 11766 #endif 11767 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104); 11768 } 11769 11770 /* 11771 * Initialize the DEV module. At boot time this walks the device list and 11772 * unhooks any devices that fail to initialise (normally hardware not 11773 * present) and leaves us with a valid list of present and active devices. 11774 * 11775 */ 11776 11777 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ 11778 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) 11779 11780 static int net_page_pool_create(int cpuid) 11781 { 11782 #if IS_ENABLED(CONFIG_PAGE_POOL) 11783 struct page_pool_params page_pool_params = { 11784 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, 11785 .flags = PP_FLAG_SYSTEM_POOL, 11786 .nid = cpu_to_mem(cpuid), 11787 }; 11788 struct page_pool *pp_ptr; 11789 11790 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); 11791 if (IS_ERR(pp_ptr)) 11792 return -ENOMEM; 11793 11794 per_cpu(system_page_pool, cpuid) = pp_ptr; 11795 #endif 11796 return 0; 11797 } 11798 11799 static int backlog_napi_should_run(unsigned int cpu) 11800 { 11801 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 11802 struct napi_struct *napi = &sd->backlog; 11803 11804 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 11805 } 11806 11807 static void run_backlog_napi(unsigned int cpu) 11808 { 11809 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 11810 11811 napi_threaded_poll_loop(&sd->backlog); 11812 } 11813 11814 static void backlog_napi_setup(unsigned int cpu) 11815 { 11816 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu); 11817 struct napi_struct *napi = &sd->backlog; 11818 11819 napi->thread = this_cpu_read(backlog_napi); 11820 set_bit(NAPI_STATE_THREADED, &napi->state); 11821 } 11822 11823 static struct smp_hotplug_thread backlog_threads = { 11824 .store = &backlog_napi, 11825 .thread_should_run = backlog_napi_should_run, 11826 .thread_fn = run_backlog_napi, 11827 .thread_comm = "backlog_napi/%u", 11828 .setup = backlog_napi_setup, 11829 }; 11830 11831 /* 11832 * This is called single threaded during boot, so no need 11833 * to take the rtnl semaphore. 11834 */ 11835 static int __init net_dev_init(void) 11836 { 11837 int i, rc = -ENOMEM; 11838 11839 BUG_ON(!dev_boot_phase); 11840 11841 net_dev_struct_check(); 11842 11843 if (dev_proc_init()) 11844 goto out; 11845 11846 if (netdev_kobject_init()) 11847 goto out; 11848 11849 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11850 INIT_LIST_HEAD(&ptype_base[i]); 11851 11852 if (register_pernet_subsys(&netdev_net_ops)) 11853 goto out; 11854 11855 /* 11856 * Initialise the packet receive queues. 11857 */ 11858 11859 for_each_possible_cpu(i) { 11860 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11861 struct softnet_data *sd = &per_cpu(softnet_data, i); 11862 11863 INIT_WORK(flush, flush_backlog); 11864 11865 skb_queue_head_init(&sd->input_pkt_queue); 11866 skb_queue_head_init(&sd->process_queue); 11867 #ifdef CONFIG_XFRM_OFFLOAD 11868 skb_queue_head_init(&sd->xfrm_backlog); 11869 #endif 11870 INIT_LIST_HEAD(&sd->poll_list); 11871 sd->output_queue_tailp = &sd->output_queue; 11872 #ifdef CONFIG_RPS 11873 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11874 sd->cpu = i; 11875 #endif 11876 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11877 spin_lock_init(&sd->defer_lock); 11878 11879 init_gro_hash(&sd->backlog); 11880 sd->backlog.poll = process_backlog; 11881 sd->backlog.weight = weight_p; 11882 INIT_LIST_HEAD(&sd->backlog.poll_list); 11883 11884 if (net_page_pool_create(i)) 11885 goto out; 11886 } 11887 if (use_backlog_threads()) 11888 smpboot_register_percpu_thread(&backlog_threads); 11889 11890 dev_boot_phase = 0; 11891 11892 /* The loopback device is special if any other network devices 11893 * is present in a network namespace the loopback device must 11894 * be present. Since we now dynamically allocate and free the 11895 * loopback device ensure this invariant is maintained by 11896 * keeping the loopback device as the first device on the 11897 * list of network devices. Ensuring the loopback devices 11898 * is the first device that appears and the last network device 11899 * that disappears. 11900 */ 11901 if (register_pernet_device(&loopback_net_ops)) 11902 goto out; 11903 11904 if (register_pernet_device(&default_device_ops)) 11905 goto out; 11906 11907 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11908 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11909 11910 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11911 NULL, dev_cpu_dead); 11912 WARN_ON(rc < 0); 11913 rc = 0; 11914 out: 11915 if (rc < 0) { 11916 for_each_possible_cpu(i) { 11917 struct page_pool *pp_ptr; 11918 11919 pp_ptr = per_cpu(system_page_pool, i); 11920 if (!pp_ptr) 11921 continue; 11922 11923 page_pool_destroy(pp_ptr); 11924 per_cpu(system_page_pool, i) = NULL; 11925 } 11926 } 11927 11928 return rc; 11929 } 11930 11931 subsys_initcall(net_dev_init); 11932