1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.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/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 157 static DEFINE_SPINLOCK(ptype_lock); 158 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 159 struct list_head ptype_all __read_mostly; /* Taps */ 160 161 static int netif_rx_internal(struct sk_buff *skb); 162 static int call_netdevice_notifiers_info(unsigned long val, 163 struct netdev_notifier_info *info); 164 static int call_netdevice_notifiers_extack(unsigned long val, 165 struct net_device *dev, 166 struct netlink_ext_ack *extack); 167 static struct napi_struct *napi_by_id(unsigned int napi_id); 168 169 /* 170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 171 * semaphore. 172 * 173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 174 * 175 * Writers must hold the rtnl semaphore while they loop through the 176 * dev_base_head list, and hold dev_base_lock for writing when they do the 177 * actual updates. This allows pure readers to access the list even 178 * while a writer is preparing to update it. 179 * 180 * To put it another way, dev_base_lock is held for writing only to 181 * protect against pure readers; the rtnl semaphore provides the 182 * protection against other writers. 183 * 184 * See, for example usages, register_netdevice() and 185 * unregister_netdevice(), which must be called with the rtnl 186 * semaphore held. 187 */ 188 DEFINE_RWLOCK(dev_base_lock); 189 EXPORT_SYMBOL(dev_base_lock); 190 191 static DEFINE_MUTEX(ifalias_mutex); 192 193 /* protects napi_hash addition/deletion and napi_gen_id */ 194 static DEFINE_SPINLOCK(napi_hash_lock); 195 196 static unsigned int napi_gen_id = NR_CPUS; 197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 198 199 static DECLARE_RWSEM(devnet_rename_sem); 200 201 static inline void dev_base_seq_inc(struct net *net) 202 { 203 while (++net->dev_base_seq == 0) 204 ; 205 } 206 207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 208 { 209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 210 211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 212 } 213 214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 215 { 216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 217 } 218 219 static inline void rps_lock_irqsave(struct softnet_data *sd, 220 unsigned long *flags) 221 { 222 if (IS_ENABLED(CONFIG_RPS)) 223 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 224 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 225 local_irq_save(*flags); 226 } 227 228 static inline void rps_lock_irq_disable(struct softnet_data *sd) 229 { 230 if (IS_ENABLED(CONFIG_RPS)) 231 spin_lock_irq(&sd->input_pkt_queue.lock); 232 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 233 local_irq_disable(); 234 } 235 236 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 237 unsigned long *flags) 238 { 239 if (IS_ENABLED(CONFIG_RPS)) 240 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 241 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 242 local_irq_restore(*flags); 243 } 244 245 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 246 { 247 if (IS_ENABLED(CONFIG_RPS)) 248 spin_unlock_irq(&sd->input_pkt_queue.lock); 249 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 250 local_irq_enable(); 251 } 252 253 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 254 const char *name) 255 { 256 struct netdev_name_node *name_node; 257 258 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 259 if (!name_node) 260 return NULL; 261 INIT_HLIST_NODE(&name_node->hlist); 262 name_node->dev = dev; 263 name_node->name = name; 264 return name_node; 265 } 266 267 static struct netdev_name_node * 268 netdev_name_node_head_alloc(struct net_device *dev) 269 { 270 struct netdev_name_node *name_node; 271 272 name_node = netdev_name_node_alloc(dev, dev->name); 273 if (!name_node) 274 return NULL; 275 INIT_LIST_HEAD(&name_node->list); 276 return name_node; 277 } 278 279 static void netdev_name_node_free(struct netdev_name_node *name_node) 280 { 281 kfree(name_node); 282 } 283 284 static void netdev_name_node_add(struct net *net, 285 struct netdev_name_node *name_node) 286 { 287 hlist_add_head_rcu(&name_node->hlist, 288 dev_name_hash(net, name_node->name)); 289 } 290 291 static void netdev_name_node_del(struct netdev_name_node *name_node) 292 { 293 hlist_del_rcu(&name_node->hlist); 294 } 295 296 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 297 const char *name) 298 { 299 struct hlist_head *head = dev_name_hash(net, name); 300 struct netdev_name_node *name_node; 301 302 hlist_for_each_entry(name_node, head, hlist) 303 if (!strcmp(name_node->name, name)) 304 return name_node; 305 return NULL; 306 } 307 308 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 309 const char *name) 310 { 311 struct hlist_head *head = dev_name_hash(net, name); 312 struct netdev_name_node *name_node; 313 314 hlist_for_each_entry_rcu(name_node, head, hlist) 315 if (!strcmp(name_node->name, name)) 316 return name_node; 317 return NULL; 318 } 319 320 bool netdev_name_in_use(struct net *net, const char *name) 321 { 322 return netdev_name_node_lookup(net, name); 323 } 324 EXPORT_SYMBOL(netdev_name_in_use); 325 326 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 327 { 328 struct netdev_name_node *name_node; 329 struct net *net = dev_net(dev); 330 331 name_node = netdev_name_node_lookup(net, name); 332 if (name_node) 333 return -EEXIST; 334 name_node = netdev_name_node_alloc(dev, name); 335 if (!name_node) 336 return -ENOMEM; 337 netdev_name_node_add(net, name_node); 338 /* The node that holds dev->name acts as a head of per-device list. */ 339 list_add_tail(&name_node->list, &dev->name_node->list); 340 341 return 0; 342 } 343 344 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 345 { 346 list_del(&name_node->list); 347 netdev_name_node_del(name_node); 348 kfree(name_node->name); 349 netdev_name_node_free(name_node); 350 } 351 352 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 353 { 354 struct netdev_name_node *name_node; 355 struct net *net = dev_net(dev); 356 357 name_node = netdev_name_node_lookup(net, name); 358 if (!name_node) 359 return -ENOENT; 360 /* lookup might have found our primary name or a name belonging 361 * to another device. 362 */ 363 if (name_node == dev->name_node || name_node->dev != dev) 364 return -EINVAL; 365 366 __netdev_name_node_alt_destroy(name_node); 367 368 return 0; 369 } 370 371 static void netdev_name_node_alt_flush(struct net_device *dev) 372 { 373 struct netdev_name_node *name_node, *tmp; 374 375 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 376 __netdev_name_node_alt_destroy(name_node); 377 } 378 379 /* Device list insertion */ 380 static void list_netdevice(struct net_device *dev) 381 { 382 struct net *net = dev_net(dev); 383 384 ASSERT_RTNL(); 385 386 write_lock(&dev_base_lock); 387 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 388 netdev_name_node_add(net, dev->name_node); 389 hlist_add_head_rcu(&dev->index_hlist, 390 dev_index_hash(net, dev->ifindex)); 391 write_unlock(&dev_base_lock); 392 393 dev_base_seq_inc(net); 394 } 395 396 /* Device list removal 397 * caller must respect a RCU grace period before freeing/reusing dev 398 */ 399 static void unlist_netdevice(struct net_device *dev) 400 { 401 ASSERT_RTNL(); 402 403 /* Unlink dev from the device chain */ 404 write_lock(&dev_base_lock); 405 list_del_rcu(&dev->dev_list); 406 netdev_name_node_del(dev->name_node); 407 hlist_del_rcu(&dev->index_hlist); 408 write_unlock(&dev_base_lock); 409 410 dev_base_seq_inc(dev_net(dev)); 411 } 412 413 /* 414 * Our notifier list 415 */ 416 417 static RAW_NOTIFIER_HEAD(netdev_chain); 418 419 /* 420 * Device drivers call our routines to queue packets here. We empty the 421 * queue in the local softnet handler. 422 */ 423 424 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 425 EXPORT_PER_CPU_SYMBOL(softnet_data); 426 427 #ifdef CONFIG_LOCKDEP 428 /* 429 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 430 * according to dev->type 431 */ 432 static const unsigned short netdev_lock_type[] = { 433 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 434 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 435 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 436 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 437 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 438 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 439 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 440 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 441 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 442 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 443 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 444 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 445 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 446 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 447 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 448 449 static const char *const netdev_lock_name[] = { 450 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 451 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 452 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 453 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 454 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 455 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 456 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 457 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 458 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 459 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 460 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 461 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 462 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 463 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 464 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 465 466 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 467 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 468 469 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 470 { 471 int i; 472 473 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 474 if (netdev_lock_type[i] == dev_type) 475 return i; 476 /* the last key is used by default */ 477 return ARRAY_SIZE(netdev_lock_type) - 1; 478 } 479 480 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 481 unsigned short dev_type) 482 { 483 int i; 484 485 i = netdev_lock_pos(dev_type); 486 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 487 netdev_lock_name[i]); 488 } 489 490 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 491 { 492 int i; 493 494 i = netdev_lock_pos(dev->type); 495 lockdep_set_class_and_name(&dev->addr_list_lock, 496 &netdev_addr_lock_key[i], 497 netdev_lock_name[i]); 498 } 499 #else 500 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 501 unsigned short dev_type) 502 { 503 } 504 505 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 506 { 507 } 508 #endif 509 510 /******************************************************************************* 511 * 512 * Protocol management and registration routines 513 * 514 *******************************************************************************/ 515 516 517 /* 518 * Add a protocol ID to the list. Now that the input handler is 519 * smarter we can dispense with all the messy stuff that used to be 520 * here. 521 * 522 * BEWARE!!! Protocol handlers, mangling input packets, 523 * MUST BE last in hash buckets and checking protocol handlers 524 * MUST start from promiscuous ptype_all chain in net_bh. 525 * It is true now, do not change it. 526 * Explanation follows: if protocol handler, mangling packet, will 527 * be the first on list, it is not able to sense, that packet 528 * is cloned and should be copied-on-write, so that it will 529 * change it and subsequent readers will get broken packet. 530 * --ANK (980803) 531 */ 532 533 static inline struct list_head *ptype_head(const struct packet_type *pt) 534 { 535 if (pt->type == htons(ETH_P_ALL)) 536 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 537 else 538 return pt->dev ? &pt->dev->ptype_specific : 539 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 540 } 541 542 /** 543 * dev_add_pack - add packet handler 544 * @pt: packet type declaration 545 * 546 * Add a protocol handler to the networking stack. The passed &packet_type 547 * is linked into kernel lists and may not be freed until it has been 548 * removed from the kernel lists. 549 * 550 * This call does not sleep therefore it can not 551 * guarantee all CPU's that are in middle of receiving packets 552 * will see the new packet type (until the next received packet). 553 */ 554 555 void dev_add_pack(struct packet_type *pt) 556 { 557 struct list_head *head = ptype_head(pt); 558 559 spin_lock(&ptype_lock); 560 list_add_rcu(&pt->list, head); 561 spin_unlock(&ptype_lock); 562 } 563 EXPORT_SYMBOL(dev_add_pack); 564 565 /** 566 * __dev_remove_pack - remove packet handler 567 * @pt: packet type declaration 568 * 569 * Remove a protocol handler that was previously added to the kernel 570 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 571 * from the kernel lists and can be freed or reused once this function 572 * returns. 573 * 574 * The packet type might still be in use by receivers 575 * and must not be freed until after all the CPU's have gone 576 * through a quiescent state. 577 */ 578 void __dev_remove_pack(struct packet_type *pt) 579 { 580 struct list_head *head = ptype_head(pt); 581 struct packet_type *pt1; 582 583 spin_lock(&ptype_lock); 584 585 list_for_each_entry(pt1, head, list) { 586 if (pt == pt1) { 587 list_del_rcu(&pt->list); 588 goto out; 589 } 590 } 591 592 pr_warn("dev_remove_pack: %p not found\n", pt); 593 out: 594 spin_unlock(&ptype_lock); 595 } 596 EXPORT_SYMBOL(__dev_remove_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 * This call sleeps to guarantee that no CPU is looking at the packet 608 * type after return. 609 */ 610 void dev_remove_pack(struct packet_type *pt) 611 { 612 __dev_remove_pack(pt); 613 614 synchronize_net(); 615 } 616 EXPORT_SYMBOL(dev_remove_pack); 617 618 619 /******************************************************************************* 620 * 621 * Device Interface Subroutines 622 * 623 *******************************************************************************/ 624 625 /** 626 * dev_get_iflink - get 'iflink' value of a interface 627 * @dev: targeted interface 628 * 629 * Indicates the ifindex the interface is linked to. 630 * Physical interfaces have the same 'ifindex' and 'iflink' values. 631 */ 632 633 int dev_get_iflink(const struct net_device *dev) 634 { 635 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 636 return dev->netdev_ops->ndo_get_iflink(dev); 637 638 return dev->ifindex; 639 } 640 EXPORT_SYMBOL(dev_get_iflink); 641 642 /** 643 * dev_fill_metadata_dst - Retrieve tunnel egress information. 644 * @dev: targeted interface 645 * @skb: The packet. 646 * 647 * For better visibility of tunnel traffic OVS needs to retrieve 648 * egress tunnel information for a packet. Following API allows 649 * user to get this info. 650 */ 651 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 652 { 653 struct ip_tunnel_info *info; 654 655 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 656 return -EINVAL; 657 658 info = skb_tunnel_info_unclone(skb); 659 if (!info) 660 return -ENOMEM; 661 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 662 return -EINVAL; 663 664 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 665 } 666 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 667 668 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 669 { 670 int k = stack->num_paths++; 671 672 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 673 return NULL; 674 675 return &stack->path[k]; 676 } 677 678 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 679 struct net_device_path_stack *stack) 680 { 681 const struct net_device *last_dev; 682 struct net_device_path_ctx ctx = { 683 .dev = dev, 684 .daddr = daddr, 685 }; 686 struct net_device_path *path; 687 int ret = 0; 688 689 stack->num_paths = 0; 690 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 691 last_dev = ctx.dev; 692 path = dev_fwd_path(stack); 693 if (!path) 694 return -1; 695 696 memset(path, 0, sizeof(struct net_device_path)); 697 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 698 if (ret < 0) 699 return -1; 700 701 if (WARN_ON_ONCE(last_dev == ctx.dev)) 702 return -1; 703 } 704 path = dev_fwd_path(stack); 705 if (!path) 706 return -1; 707 path->type = DEV_PATH_ETHERNET; 708 path->dev = ctx.dev; 709 710 return ret; 711 } 712 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 713 714 /** 715 * __dev_get_by_name - find a device by its name 716 * @net: the applicable net namespace 717 * @name: name to find 718 * 719 * Find an interface by name. Must be called under RTNL semaphore 720 * or @dev_base_lock. If the name is found a pointer to the device 721 * is returned. If the name is not found then %NULL is returned. The 722 * reference counters are not incremented so the caller must be 723 * careful with locks. 724 */ 725 726 struct net_device *__dev_get_by_name(struct net *net, const char *name) 727 { 728 struct netdev_name_node *node_name; 729 730 node_name = netdev_name_node_lookup(net, name); 731 return node_name ? node_name->dev : NULL; 732 } 733 EXPORT_SYMBOL(__dev_get_by_name); 734 735 /** 736 * dev_get_by_name_rcu - find a device by its name 737 * @net: the applicable net namespace 738 * @name: name to find 739 * 740 * Find an interface by name. 741 * If the name is found a pointer to the device is returned. 742 * If the name is not found then %NULL is returned. 743 * The reference counters are not incremented so the caller must be 744 * careful with locks. The caller must hold RCU lock. 745 */ 746 747 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 748 { 749 struct netdev_name_node *node_name; 750 751 node_name = netdev_name_node_lookup_rcu(net, name); 752 return node_name ? node_name->dev : NULL; 753 } 754 EXPORT_SYMBOL(dev_get_by_name_rcu); 755 756 /** 757 * dev_get_by_name - find a device by its name 758 * @net: the applicable net namespace 759 * @name: name to find 760 * 761 * Find an interface by name. This can be called from any 762 * context and does its own locking. The returned handle has 763 * the usage count incremented and the caller must use dev_put() to 764 * release it when it is no longer needed. %NULL is returned if no 765 * matching device is found. 766 */ 767 768 struct net_device *dev_get_by_name(struct net *net, const char *name) 769 { 770 struct net_device *dev; 771 772 rcu_read_lock(); 773 dev = dev_get_by_name_rcu(net, name); 774 dev_hold(dev); 775 rcu_read_unlock(); 776 return dev; 777 } 778 EXPORT_SYMBOL(dev_get_by_name); 779 780 /** 781 * __dev_get_by_index - find a device by its ifindex 782 * @net: the applicable net namespace 783 * @ifindex: index of device 784 * 785 * Search for an interface by index. Returns %NULL if the device 786 * is not found or a pointer to the device. The device has not 787 * had its reference counter increased so the caller must be careful 788 * about locking. The caller must hold either the RTNL semaphore 789 * or @dev_base_lock. 790 */ 791 792 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 793 { 794 struct net_device *dev; 795 struct hlist_head *head = dev_index_hash(net, ifindex); 796 797 hlist_for_each_entry(dev, head, index_hlist) 798 if (dev->ifindex == ifindex) 799 return dev; 800 801 return NULL; 802 } 803 EXPORT_SYMBOL(__dev_get_by_index); 804 805 /** 806 * dev_get_by_index_rcu - find a device by its ifindex 807 * @net: the applicable net namespace 808 * @ifindex: index of device 809 * 810 * Search for an interface by index. Returns %NULL if the device 811 * is not found or a pointer to the device. The device has not 812 * had its reference counter increased so the caller must be careful 813 * about locking. The caller must hold RCU lock. 814 */ 815 816 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 817 { 818 struct net_device *dev; 819 struct hlist_head *head = dev_index_hash(net, ifindex); 820 821 hlist_for_each_entry_rcu(dev, head, index_hlist) 822 if (dev->ifindex == ifindex) 823 return dev; 824 825 return NULL; 826 } 827 EXPORT_SYMBOL(dev_get_by_index_rcu); 828 829 830 /** 831 * dev_get_by_index - find a device by its ifindex 832 * @net: the applicable net namespace 833 * @ifindex: index of device 834 * 835 * Search for an interface by index. Returns NULL if the device 836 * is not found or a pointer to the device. The device returned has 837 * had a reference added and the pointer is safe until the user calls 838 * dev_put to indicate they have finished with it. 839 */ 840 841 struct net_device *dev_get_by_index(struct net *net, int ifindex) 842 { 843 struct net_device *dev; 844 845 rcu_read_lock(); 846 dev = dev_get_by_index_rcu(net, ifindex); 847 dev_hold(dev); 848 rcu_read_unlock(); 849 return dev; 850 } 851 EXPORT_SYMBOL(dev_get_by_index); 852 853 /** 854 * dev_get_by_napi_id - find a device by napi_id 855 * @napi_id: ID of the NAPI struct 856 * 857 * Search for an interface by NAPI ID. Returns %NULL if the device 858 * is not found or a pointer to the device. The device has not had 859 * its reference counter increased so the caller must be careful 860 * about locking. The caller must hold RCU lock. 861 */ 862 863 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 864 { 865 struct napi_struct *napi; 866 867 WARN_ON_ONCE(!rcu_read_lock_held()); 868 869 if (napi_id < MIN_NAPI_ID) 870 return NULL; 871 872 napi = napi_by_id(napi_id); 873 874 return napi ? napi->dev : NULL; 875 } 876 EXPORT_SYMBOL(dev_get_by_napi_id); 877 878 /** 879 * netdev_get_name - get a netdevice name, knowing its ifindex. 880 * @net: network namespace 881 * @name: a pointer to the buffer where the name will be stored. 882 * @ifindex: the ifindex of the interface to get the name from. 883 */ 884 int netdev_get_name(struct net *net, char *name, int ifindex) 885 { 886 struct net_device *dev; 887 int ret; 888 889 down_read(&devnet_rename_sem); 890 rcu_read_lock(); 891 892 dev = dev_get_by_index_rcu(net, ifindex); 893 if (!dev) { 894 ret = -ENODEV; 895 goto out; 896 } 897 898 strcpy(name, dev->name); 899 900 ret = 0; 901 out: 902 rcu_read_unlock(); 903 up_read(&devnet_rename_sem); 904 return ret; 905 } 906 907 /** 908 * dev_getbyhwaddr_rcu - find a device by its hardware address 909 * @net: the applicable net namespace 910 * @type: media type of device 911 * @ha: hardware address 912 * 913 * Search for an interface by MAC address. Returns NULL if the device 914 * is not found or a pointer to the device. 915 * The caller must hold RCU or RTNL. 916 * The returned device has not had its ref count increased 917 * and the caller must therefore be careful about locking 918 * 919 */ 920 921 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 922 const char *ha) 923 { 924 struct net_device *dev; 925 926 for_each_netdev_rcu(net, dev) 927 if (dev->type == type && 928 !memcmp(dev->dev_addr, ha, dev->addr_len)) 929 return dev; 930 931 return NULL; 932 } 933 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 934 935 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 936 { 937 struct net_device *dev, *ret = NULL; 938 939 rcu_read_lock(); 940 for_each_netdev_rcu(net, dev) 941 if (dev->type == type) { 942 dev_hold(dev); 943 ret = dev; 944 break; 945 } 946 rcu_read_unlock(); 947 return ret; 948 } 949 EXPORT_SYMBOL(dev_getfirstbyhwtype); 950 951 /** 952 * __dev_get_by_flags - find any device with given flags 953 * @net: the applicable net namespace 954 * @if_flags: IFF_* values 955 * @mask: bitmask of bits in if_flags to check 956 * 957 * Search for any interface with the given flags. Returns NULL if a device 958 * is not found or a pointer to the device. Must be called inside 959 * rtnl_lock(), and result refcount is unchanged. 960 */ 961 962 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 963 unsigned short mask) 964 { 965 struct net_device *dev, *ret; 966 967 ASSERT_RTNL(); 968 969 ret = NULL; 970 for_each_netdev(net, dev) { 971 if (((dev->flags ^ if_flags) & mask) == 0) { 972 ret = dev; 973 break; 974 } 975 } 976 return ret; 977 } 978 EXPORT_SYMBOL(__dev_get_by_flags); 979 980 /** 981 * dev_valid_name - check if name is okay for network device 982 * @name: name string 983 * 984 * Network device names need to be valid file names to 985 * allow sysfs to work. We also disallow any kind of 986 * whitespace. 987 */ 988 bool dev_valid_name(const char *name) 989 { 990 if (*name == '\0') 991 return false; 992 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 993 return false; 994 if (!strcmp(name, ".") || !strcmp(name, "..")) 995 return false; 996 997 while (*name) { 998 if (*name == '/' || *name == ':' || isspace(*name)) 999 return false; 1000 name++; 1001 } 1002 return true; 1003 } 1004 EXPORT_SYMBOL(dev_valid_name); 1005 1006 /** 1007 * __dev_alloc_name - allocate a name for a device 1008 * @net: network namespace to allocate the device name in 1009 * @name: name format string 1010 * @buf: scratch buffer and result name string 1011 * 1012 * Passed a format string - eg "lt%d" it will try and find a suitable 1013 * id. It scans list of devices to build up a free map, then chooses 1014 * the first empty slot. The caller must hold the dev_base or rtnl lock 1015 * while allocating the name and adding the device in order to avoid 1016 * duplicates. 1017 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1018 * Returns the number of the unit assigned or a negative errno code. 1019 */ 1020 1021 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1022 { 1023 int i = 0; 1024 const char *p; 1025 const int max_netdevices = 8*PAGE_SIZE; 1026 unsigned long *inuse; 1027 struct net_device *d; 1028 1029 if (!dev_valid_name(name)) 1030 return -EINVAL; 1031 1032 p = strchr(name, '%'); 1033 if (p) { 1034 /* 1035 * Verify the string as this thing may have come from 1036 * the user. There must be either one "%d" and no other "%" 1037 * characters. 1038 */ 1039 if (p[1] != 'd' || strchr(p + 2, '%')) 1040 return -EINVAL; 1041 1042 /* Use one page as a bit array of possible slots */ 1043 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1044 if (!inuse) 1045 return -ENOMEM; 1046 1047 for_each_netdev(net, d) { 1048 struct netdev_name_node *name_node; 1049 list_for_each_entry(name_node, &d->name_node->list, list) { 1050 if (!sscanf(name_node->name, name, &i)) 1051 continue; 1052 if (i < 0 || i >= max_netdevices) 1053 continue; 1054 1055 /* avoid cases where sscanf is not exact inverse of printf */ 1056 snprintf(buf, IFNAMSIZ, name, i); 1057 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1058 __set_bit(i, inuse); 1059 } 1060 if (!sscanf(d->name, name, &i)) 1061 continue; 1062 if (i < 0 || i >= max_netdevices) 1063 continue; 1064 1065 /* avoid cases where sscanf is not exact inverse of printf */ 1066 snprintf(buf, IFNAMSIZ, name, i); 1067 if (!strncmp(buf, d->name, IFNAMSIZ)) 1068 __set_bit(i, inuse); 1069 } 1070 1071 i = find_first_zero_bit(inuse, max_netdevices); 1072 free_page((unsigned long) inuse); 1073 } 1074 1075 snprintf(buf, IFNAMSIZ, name, i); 1076 if (!netdev_name_in_use(net, buf)) 1077 return i; 1078 1079 /* It is possible to run out of possible slots 1080 * when the name is long and there isn't enough space left 1081 * for the digits, or if all bits are used. 1082 */ 1083 return -ENFILE; 1084 } 1085 1086 static int dev_alloc_name_ns(struct net *net, 1087 struct net_device *dev, 1088 const char *name) 1089 { 1090 char buf[IFNAMSIZ]; 1091 int ret; 1092 1093 BUG_ON(!net); 1094 ret = __dev_alloc_name(net, name, buf); 1095 if (ret >= 0) 1096 strlcpy(dev->name, buf, IFNAMSIZ); 1097 return ret; 1098 } 1099 1100 /** 1101 * dev_alloc_name - allocate a name for a device 1102 * @dev: device 1103 * @name: name format string 1104 * 1105 * Passed a format string - eg "lt%d" it will try and find a suitable 1106 * id. It scans list of devices to build up a free map, then chooses 1107 * the first empty slot. The caller must hold the dev_base or rtnl lock 1108 * while allocating the name and adding the device in order to avoid 1109 * duplicates. 1110 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1111 * Returns the number of the unit assigned or a negative errno code. 1112 */ 1113 1114 int dev_alloc_name(struct net_device *dev, const char *name) 1115 { 1116 return dev_alloc_name_ns(dev_net(dev), dev, name); 1117 } 1118 EXPORT_SYMBOL(dev_alloc_name); 1119 1120 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1121 const char *name) 1122 { 1123 BUG_ON(!net); 1124 1125 if (!dev_valid_name(name)) 1126 return -EINVAL; 1127 1128 if (strchr(name, '%')) 1129 return dev_alloc_name_ns(net, dev, name); 1130 else if (netdev_name_in_use(net, name)) 1131 return -EEXIST; 1132 else if (dev->name != name) 1133 strlcpy(dev->name, name, IFNAMSIZ); 1134 1135 return 0; 1136 } 1137 1138 /** 1139 * dev_change_name - change name of a device 1140 * @dev: device 1141 * @newname: name (or format string) must be at least IFNAMSIZ 1142 * 1143 * Change name of a device, can pass format strings "eth%d". 1144 * for wildcarding. 1145 */ 1146 int dev_change_name(struct net_device *dev, const char *newname) 1147 { 1148 unsigned char old_assign_type; 1149 char oldname[IFNAMSIZ]; 1150 int err = 0; 1151 int ret; 1152 struct net *net; 1153 1154 ASSERT_RTNL(); 1155 BUG_ON(!dev_net(dev)); 1156 1157 net = dev_net(dev); 1158 1159 /* Some auto-enslaved devices e.g. failover slaves are 1160 * special, as userspace might rename the device after 1161 * the interface had been brought up and running since 1162 * the point kernel initiated auto-enslavement. Allow 1163 * live name change even when these slave devices are 1164 * up and running. 1165 * 1166 * Typically, users of these auto-enslaving devices 1167 * don't actually care about slave name change, as 1168 * they are supposed to operate on master interface 1169 * directly. 1170 */ 1171 if (dev->flags & IFF_UP && 1172 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1173 return -EBUSY; 1174 1175 down_write(&devnet_rename_sem); 1176 1177 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1178 up_write(&devnet_rename_sem); 1179 return 0; 1180 } 1181 1182 memcpy(oldname, dev->name, IFNAMSIZ); 1183 1184 err = dev_get_valid_name(net, dev, newname); 1185 if (err < 0) { 1186 up_write(&devnet_rename_sem); 1187 return err; 1188 } 1189 1190 if (oldname[0] && !strchr(oldname, '%')) 1191 netdev_info(dev, "renamed from %s\n", oldname); 1192 1193 old_assign_type = dev->name_assign_type; 1194 dev->name_assign_type = NET_NAME_RENAMED; 1195 1196 rollback: 1197 ret = device_rename(&dev->dev, dev->name); 1198 if (ret) { 1199 memcpy(dev->name, oldname, IFNAMSIZ); 1200 dev->name_assign_type = old_assign_type; 1201 up_write(&devnet_rename_sem); 1202 return ret; 1203 } 1204 1205 up_write(&devnet_rename_sem); 1206 1207 netdev_adjacent_rename_links(dev, oldname); 1208 1209 write_lock(&dev_base_lock); 1210 netdev_name_node_del(dev->name_node); 1211 write_unlock(&dev_base_lock); 1212 1213 synchronize_rcu(); 1214 1215 write_lock(&dev_base_lock); 1216 netdev_name_node_add(net, dev->name_node); 1217 write_unlock(&dev_base_lock); 1218 1219 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1220 ret = notifier_to_errno(ret); 1221 1222 if (ret) { 1223 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1224 if (err >= 0) { 1225 err = ret; 1226 down_write(&devnet_rename_sem); 1227 memcpy(dev->name, oldname, IFNAMSIZ); 1228 memcpy(oldname, newname, IFNAMSIZ); 1229 dev->name_assign_type = old_assign_type; 1230 old_assign_type = NET_NAME_RENAMED; 1231 goto rollback; 1232 } else { 1233 netdev_err(dev, "name change rollback failed: %d\n", 1234 ret); 1235 } 1236 } 1237 1238 return err; 1239 } 1240 1241 /** 1242 * dev_set_alias - change ifalias of a device 1243 * @dev: device 1244 * @alias: name up to IFALIASZ 1245 * @len: limit of bytes to copy from info 1246 * 1247 * Set ifalias for a device, 1248 */ 1249 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1250 { 1251 struct dev_ifalias *new_alias = NULL; 1252 1253 if (len >= IFALIASZ) 1254 return -EINVAL; 1255 1256 if (len) { 1257 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1258 if (!new_alias) 1259 return -ENOMEM; 1260 1261 memcpy(new_alias->ifalias, alias, len); 1262 new_alias->ifalias[len] = 0; 1263 } 1264 1265 mutex_lock(&ifalias_mutex); 1266 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1267 mutex_is_locked(&ifalias_mutex)); 1268 mutex_unlock(&ifalias_mutex); 1269 1270 if (new_alias) 1271 kfree_rcu(new_alias, rcuhead); 1272 1273 return len; 1274 } 1275 EXPORT_SYMBOL(dev_set_alias); 1276 1277 /** 1278 * dev_get_alias - get ifalias of a device 1279 * @dev: device 1280 * @name: buffer to store name of ifalias 1281 * @len: size of buffer 1282 * 1283 * get ifalias for a device. Caller must make sure dev cannot go 1284 * away, e.g. rcu read lock or own a reference count to device. 1285 */ 1286 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1287 { 1288 const struct dev_ifalias *alias; 1289 int ret = 0; 1290 1291 rcu_read_lock(); 1292 alias = rcu_dereference(dev->ifalias); 1293 if (alias) 1294 ret = snprintf(name, len, "%s", alias->ifalias); 1295 rcu_read_unlock(); 1296 1297 return ret; 1298 } 1299 1300 /** 1301 * netdev_features_change - device changes features 1302 * @dev: device to cause notification 1303 * 1304 * Called to indicate a device has changed features. 1305 */ 1306 void netdev_features_change(struct net_device *dev) 1307 { 1308 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1309 } 1310 EXPORT_SYMBOL(netdev_features_change); 1311 1312 /** 1313 * netdev_state_change - device changes state 1314 * @dev: device to cause notification 1315 * 1316 * Called to indicate a device has changed state. This function calls 1317 * the notifier chains for netdev_chain and sends a NEWLINK message 1318 * to the routing socket. 1319 */ 1320 void netdev_state_change(struct net_device *dev) 1321 { 1322 if (dev->flags & IFF_UP) { 1323 struct netdev_notifier_change_info change_info = { 1324 .info.dev = dev, 1325 }; 1326 1327 call_netdevice_notifiers_info(NETDEV_CHANGE, 1328 &change_info.info); 1329 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1330 } 1331 } 1332 EXPORT_SYMBOL(netdev_state_change); 1333 1334 /** 1335 * __netdev_notify_peers - notify network peers about existence of @dev, 1336 * to be called when rtnl lock is already held. 1337 * @dev: network device 1338 * 1339 * Generate traffic such that interested network peers are aware of 1340 * @dev, such as by generating a gratuitous ARP. This may be used when 1341 * a device wants to inform the rest of the network about some sort of 1342 * reconfiguration such as a failover event or virtual machine 1343 * migration. 1344 */ 1345 void __netdev_notify_peers(struct net_device *dev) 1346 { 1347 ASSERT_RTNL(); 1348 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1349 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1350 } 1351 EXPORT_SYMBOL(__netdev_notify_peers); 1352 1353 /** 1354 * netdev_notify_peers - notify network peers about existence of @dev 1355 * @dev: network device 1356 * 1357 * Generate traffic such that interested network peers are aware of 1358 * @dev, such as by generating a gratuitous ARP. This may be used when 1359 * a device wants to inform the rest of the network about some sort of 1360 * reconfiguration such as a failover event or virtual machine 1361 * migration. 1362 */ 1363 void netdev_notify_peers(struct net_device *dev) 1364 { 1365 rtnl_lock(); 1366 __netdev_notify_peers(dev); 1367 rtnl_unlock(); 1368 } 1369 EXPORT_SYMBOL(netdev_notify_peers); 1370 1371 static int napi_threaded_poll(void *data); 1372 1373 static int napi_kthread_create(struct napi_struct *n) 1374 { 1375 int err = 0; 1376 1377 /* Create and wake up the kthread once to put it in 1378 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1379 * warning and work with loadavg. 1380 */ 1381 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1382 n->dev->name, n->napi_id); 1383 if (IS_ERR(n->thread)) { 1384 err = PTR_ERR(n->thread); 1385 pr_err("kthread_run failed with err %d\n", err); 1386 n->thread = NULL; 1387 } 1388 1389 return err; 1390 } 1391 1392 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1393 { 1394 const struct net_device_ops *ops = dev->netdev_ops; 1395 int ret; 1396 1397 ASSERT_RTNL(); 1398 dev_addr_check(dev); 1399 1400 if (!netif_device_present(dev)) { 1401 /* may be detached because parent is runtime-suspended */ 1402 if (dev->dev.parent) 1403 pm_runtime_resume(dev->dev.parent); 1404 if (!netif_device_present(dev)) 1405 return -ENODEV; 1406 } 1407 1408 /* Block netpoll from trying to do any rx path servicing. 1409 * If we don't do this there is a chance ndo_poll_controller 1410 * or ndo_poll may be running while we open the device 1411 */ 1412 netpoll_poll_disable(dev); 1413 1414 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1415 ret = notifier_to_errno(ret); 1416 if (ret) 1417 return ret; 1418 1419 set_bit(__LINK_STATE_START, &dev->state); 1420 1421 if (ops->ndo_validate_addr) 1422 ret = ops->ndo_validate_addr(dev); 1423 1424 if (!ret && ops->ndo_open) 1425 ret = ops->ndo_open(dev); 1426 1427 netpoll_poll_enable(dev); 1428 1429 if (ret) 1430 clear_bit(__LINK_STATE_START, &dev->state); 1431 else { 1432 dev->flags |= IFF_UP; 1433 dev_set_rx_mode(dev); 1434 dev_activate(dev); 1435 add_device_randomness(dev->dev_addr, dev->addr_len); 1436 } 1437 1438 return ret; 1439 } 1440 1441 /** 1442 * dev_open - prepare an interface for use. 1443 * @dev: device to open 1444 * @extack: netlink extended ack 1445 * 1446 * Takes a device from down to up state. The device's private open 1447 * function is invoked and then the multicast lists are loaded. Finally 1448 * the device is moved into the up state and a %NETDEV_UP message is 1449 * sent to the netdev notifier chain. 1450 * 1451 * Calling this function on an active interface is a nop. On a failure 1452 * a negative errno code is returned. 1453 */ 1454 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1455 { 1456 int ret; 1457 1458 if (dev->flags & IFF_UP) 1459 return 0; 1460 1461 ret = __dev_open(dev, extack); 1462 if (ret < 0) 1463 return ret; 1464 1465 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1466 call_netdevice_notifiers(NETDEV_UP, dev); 1467 1468 return ret; 1469 } 1470 EXPORT_SYMBOL(dev_open); 1471 1472 static void __dev_close_many(struct list_head *head) 1473 { 1474 struct net_device *dev; 1475 1476 ASSERT_RTNL(); 1477 might_sleep(); 1478 1479 list_for_each_entry(dev, head, close_list) { 1480 /* Temporarily disable netpoll until the interface is down */ 1481 netpoll_poll_disable(dev); 1482 1483 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1484 1485 clear_bit(__LINK_STATE_START, &dev->state); 1486 1487 /* Synchronize to scheduled poll. We cannot touch poll list, it 1488 * can be even on different cpu. So just clear netif_running(). 1489 * 1490 * dev->stop() will invoke napi_disable() on all of it's 1491 * napi_struct instances on this device. 1492 */ 1493 smp_mb__after_atomic(); /* Commit netif_running(). */ 1494 } 1495 1496 dev_deactivate_many(head); 1497 1498 list_for_each_entry(dev, head, close_list) { 1499 const struct net_device_ops *ops = dev->netdev_ops; 1500 1501 /* 1502 * Call the device specific close. This cannot fail. 1503 * Only if device is UP 1504 * 1505 * We allow it to be called even after a DETACH hot-plug 1506 * event. 1507 */ 1508 if (ops->ndo_stop) 1509 ops->ndo_stop(dev); 1510 1511 dev->flags &= ~IFF_UP; 1512 netpoll_poll_enable(dev); 1513 } 1514 } 1515 1516 static void __dev_close(struct net_device *dev) 1517 { 1518 LIST_HEAD(single); 1519 1520 list_add(&dev->close_list, &single); 1521 __dev_close_many(&single); 1522 list_del(&single); 1523 } 1524 1525 void dev_close_many(struct list_head *head, bool unlink) 1526 { 1527 struct net_device *dev, *tmp; 1528 1529 /* Remove the devices that don't need to be closed */ 1530 list_for_each_entry_safe(dev, tmp, head, close_list) 1531 if (!(dev->flags & IFF_UP)) 1532 list_del_init(&dev->close_list); 1533 1534 __dev_close_many(head); 1535 1536 list_for_each_entry_safe(dev, tmp, head, close_list) { 1537 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1538 call_netdevice_notifiers(NETDEV_DOWN, dev); 1539 if (unlink) 1540 list_del_init(&dev->close_list); 1541 } 1542 } 1543 EXPORT_SYMBOL(dev_close_many); 1544 1545 /** 1546 * dev_close - shutdown an interface. 1547 * @dev: device to shutdown 1548 * 1549 * This function moves an active device into down state. A 1550 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1551 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1552 * chain. 1553 */ 1554 void dev_close(struct net_device *dev) 1555 { 1556 if (dev->flags & IFF_UP) { 1557 LIST_HEAD(single); 1558 1559 list_add(&dev->close_list, &single); 1560 dev_close_many(&single, true); 1561 list_del(&single); 1562 } 1563 } 1564 EXPORT_SYMBOL(dev_close); 1565 1566 1567 /** 1568 * dev_disable_lro - disable Large Receive Offload on a device 1569 * @dev: device 1570 * 1571 * Disable Large Receive Offload (LRO) on a net device. Must be 1572 * called under RTNL. This is needed if received packets may be 1573 * forwarded to another interface. 1574 */ 1575 void dev_disable_lro(struct net_device *dev) 1576 { 1577 struct net_device *lower_dev; 1578 struct list_head *iter; 1579 1580 dev->wanted_features &= ~NETIF_F_LRO; 1581 netdev_update_features(dev); 1582 1583 if (unlikely(dev->features & NETIF_F_LRO)) 1584 netdev_WARN(dev, "failed to disable LRO!\n"); 1585 1586 netdev_for_each_lower_dev(dev, lower_dev, iter) 1587 dev_disable_lro(lower_dev); 1588 } 1589 EXPORT_SYMBOL(dev_disable_lro); 1590 1591 /** 1592 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1593 * @dev: device 1594 * 1595 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1596 * called under RTNL. This is needed if Generic XDP is installed on 1597 * the device. 1598 */ 1599 static void dev_disable_gro_hw(struct net_device *dev) 1600 { 1601 dev->wanted_features &= ~NETIF_F_GRO_HW; 1602 netdev_update_features(dev); 1603 1604 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1605 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1606 } 1607 1608 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1609 { 1610 #define N(val) \ 1611 case NETDEV_##val: \ 1612 return "NETDEV_" __stringify(val); 1613 switch (cmd) { 1614 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1615 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1616 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1617 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1618 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1619 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1620 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1621 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1622 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1623 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1624 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1625 } 1626 #undef N 1627 return "UNKNOWN_NETDEV_EVENT"; 1628 } 1629 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1630 1631 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1632 struct net_device *dev) 1633 { 1634 struct netdev_notifier_info info = { 1635 .dev = dev, 1636 }; 1637 1638 return nb->notifier_call(nb, val, &info); 1639 } 1640 1641 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1642 struct net_device *dev) 1643 { 1644 int err; 1645 1646 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1647 err = notifier_to_errno(err); 1648 if (err) 1649 return err; 1650 1651 if (!(dev->flags & IFF_UP)) 1652 return 0; 1653 1654 call_netdevice_notifier(nb, NETDEV_UP, dev); 1655 return 0; 1656 } 1657 1658 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1659 struct net_device *dev) 1660 { 1661 if (dev->flags & IFF_UP) { 1662 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1663 dev); 1664 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1665 } 1666 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1667 } 1668 1669 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1670 struct net *net) 1671 { 1672 struct net_device *dev; 1673 int err; 1674 1675 for_each_netdev(net, dev) { 1676 err = call_netdevice_register_notifiers(nb, dev); 1677 if (err) 1678 goto rollback; 1679 } 1680 return 0; 1681 1682 rollback: 1683 for_each_netdev_continue_reverse(net, dev) 1684 call_netdevice_unregister_notifiers(nb, dev); 1685 return err; 1686 } 1687 1688 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1689 struct net *net) 1690 { 1691 struct net_device *dev; 1692 1693 for_each_netdev(net, dev) 1694 call_netdevice_unregister_notifiers(nb, dev); 1695 } 1696 1697 static int dev_boot_phase = 1; 1698 1699 /** 1700 * register_netdevice_notifier - register a network notifier block 1701 * @nb: notifier 1702 * 1703 * Register a notifier to be called when network device events occur. 1704 * The notifier passed is linked into the kernel structures and must 1705 * not be reused until it has been unregistered. A negative errno code 1706 * is returned on a failure. 1707 * 1708 * When registered all registration and up events are replayed 1709 * to the new notifier to allow device to have a race free 1710 * view of the network device list. 1711 */ 1712 1713 int register_netdevice_notifier(struct notifier_block *nb) 1714 { 1715 struct net *net; 1716 int err; 1717 1718 /* Close race with setup_net() and cleanup_net() */ 1719 down_write(&pernet_ops_rwsem); 1720 rtnl_lock(); 1721 err = raw_notifier_chain_register(&netdev_chain, nb); 1722 if (err) 1723 goto unlock; 1724 if (dev_boot_phase) 1725 goto unlock; 1726 for_each_net(net) { 1727 err = call_netdevice_register_net_notifiers(nb, net); 1728 if (err) 1729 goto rollback; 1730 } 1731 1732 unlock: 1733 rtnl_unlock(); 1734 up_write(&pernet_ops_rwsem); 1735 return err; 1736 1737 rollback: 1738 for_each_net_continue_reverse(net) 1739 call_netdevice_unregister_net_notifiers(nb, net); 1740 1741 raw_notifier_chain_unregister(&netdev_chain, nb); 1742 goto unlock; 1743 } 1744 EXPORT_SYMBOL(register_netdevice_notifier); 1745 1746 /** 1747 * unregister_netdevice_notifier - unregister a network notifier block 1748 * @nb: notifier 1749 * 1750 * Unregister a notifier previously registered by 1751 * register_netdevice_notifier(). The notifier is unlinked into the 1752 * kernel structures and may then be reused. A negative errno code 1753 * is returned on a failure. 1754 * 1755 * After unregistering unregister and down device events are synthesized 1756 * for all devices on the device list to the removed notifier to remove 1757 * the need for special case cleanup code. 1758 */ 1759 1760 int unregister_netdevice_notifier(struct notifier_block *nb) 1761 { 1762 struct net *net; 1763 int err; 1764 1765 /* Close race with setup_net() and cleanup_net() */ 1766 down_write(&pernet_ops_rwsem); 1767 rtnl_lock(); 1768 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1769 if (err) 1770 goto unlock; 1771 1772 for_each_net(net) 1773 call_netdevice_unregister_net_notifiers(nb, net); 1774 1775 unlock: 1776 rtnl_unlock(); 1777 up_write(&pernet_ops_rwsem); 1778 return err; 1779 } 1780 EXPORT_SYMBOL(unregister_netdevice_notifier); 1781 1782 static int __register_netdevice_notifier_net(struct net *net, 1783 struct notifier_block *nb, 1784 bool ignore_call_fail) 1785 { 1786 int err; 1787 1788 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1789 if (err) 1790 return err; 1791 if (dev_boot_phase) 1792 return 0; 1793 1794 err = call_netdevice_register_net_notifiers(nb, net); 1795 if (err && !ignore_call_fail) 1796 goto chain_unregister; 1797 1798 return 0; 1799 1800 chain_unregister: 1801 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1802 return err; 1803 } 1804 1805 static int __unregister_netdevice_notifier_net(struct net *net, 1806 struct notifier_block *nb) 1807 { 1808 int err; 1809 1810 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1811 if (err) 1812 return err; 1813 1814 call_netdevice_unregister_net_notifiers(nb, net); 1815 return 0; 1816 } 1817 1818 /** 1819 * register_netdevice_notifier_net - register a per-netns network notifier block 1820 * @net: network namespace 1821 * @nb: notifier 1822 * 1823 * Register a notifier to be called when network device events occur. 1824 * The notifier passed is linked into the kernel structures and must 1825 * not be reused until it has been unregistered. A negative errno code 1826 * is returned on a failure. 1827 * 1828 * When registered all registration and up events are replayed 1829 * to the new notifier to allow device to have a race free 1830 * view of the network device list. 1831 */ 1832 1833 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1834 { 1835 int err; 1836 1837 rtnl_lock(); 1838 err = __register_netdevice_notifier_net(net, nb, false); 1839 rtnl_unlock(); 1840 return err; 1841 } 1842 EXPORT_SYMBOL(register_netdevice_notifier_net); 1843 1844 /** 1845 * unregister_netdevice_notifier_net - unregister a per-netns 1846 * network notifier block 1847 * @net: network namespace 1848 * @nb: notifier 1849 * 1850 * Unregister a notifier previously registered by 1851 * register_netdevice_notifier(). The notifier is unlinked into the 1852 * kernel structures and may then be reused. A negative errno code 1853 * is returned on a failure. 1854 * 1855 * After unregistering unregister and down device events are synthesized 1856 * for all devices on the device list to the removed notifier to remove 1857 * the need for special case cleanup code. 1858 */ 1859 1860 int unregister_netdevice_notifier_net(struct net *net, 1861 struct notifier_block *nb) 1862 { 1863 int err; 1864 1865 rtnl_lock(); 1866 err = __unregister_netdevice_notifier_net(net, nb); 1867 rtnl_unlock(); 1868 return err; 1869 } 1870 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1871 1872 int register_netdevice_notifier_dev_net(struct net_device *dev, 1873 struct notifier_block *nb, 1874 struct netdev_net_notifier *nn) 1875 { 1876 int err; 1877 1878 rtnl_lock(); 1879 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1880 if (!err) { 1881 nn->nb = nb; 1882 list_add(&nn->list, &dev->net_notifier_list); 1883 } 1884 rtnl_unlock(); 1885 return err; 1886 } 1887 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1888 1889 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1890 struct notifier_block *nb, 1891 struct netdev_net_notifier *nn) 1892 { 1893 int err; 1894 1895 rtnl_lock(); 1896 list_del(&nn->list); 1897 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1898 rtnl_unlock(); 1899 return err; 1900 } 1901 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1902 1903 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1904 struct net *net) 1905 { 1906 struct netdev_net_notifier *nn; 1907 1908 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1909 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1910 __register_netdevice_notifier_net(net, nn->nb, true); 1911 } 1912 } 1913 1914 /** 1915 * call_netdevice_notifiers_info - call all network notifier blocks 1916 * @val: value passed unmodified to notifier function 1917 * @info: notifier information data 1918 * 1919 * Call all network notifier blocks. Parameters and return value 1920 * are as for raw_notifier_call_chain(). 1921 */ 1922 1923 static int call_netdevice_notifiers_info(unsigned long val, 1924 struct netdev_notifier_info *info) 1925 { 1926 struct net *net = dev_net(info->dev); 1927 int ret; 1928 1929 ASSERT_RTNL(); 1930 1931 /* Run per-netns notifier block chain first, then run the global one. 1932 * Hopefully, one day, the global one is going to be removed after 1933 * all notifier block registrators get converted to be per-netns. 1934 */ 1935 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1936 if (ret & NOTIFY_STOP_MASK) 1937 return ret; 1938 return raw_notifier_call_chain(&netdev_chain, val, info); 1939 } 1940 1941 /** 1942 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1943 * for and rollback on error 1944 * @val_up: value passed unmodified to notifier function 1945 * @val_down: value passed unmodified to the notifier function when 1946 * recovering from an error on @val_up 1947 * @info: notifier information data 1948 * 1949 * Call all per-netns network notifier blocks, but not notifier blocks on 1950 * the global notifier chain. Parameters and return value are as for 1951 * raw_notifier_call_chain_robust(). 1952 */ 1953 1954 static int 1955 call_netdevice_notifiers_info_robust(unsigned long val_up, 1956 unsigned long val_down, 1957 struct netdev_notifier_info *info) 1958 { 1959 struct net *net = dev_net(info->dev); 1960 1961 ASSERT_RTNL(); 1962 1963 return raw_notifier_call_chain_robust(&net->netdev_chain, 1964 val_up, val_down, info); 1965 } 1966 1967 static int call_netdevice_notifiers_extack(unsigned long val, 1968 struct net_device *dev, 1969 struct netlink_ext_ack *extack) 1970 { 1971 struct netdev_notifier_info info = { 1972 .dev = dev, 1973 .extack = extack, 1974 }; 1975 1976 return call_netdevice_notifiers_info(val, &info); 1977 } 1978 1979 /** 1980 * call_netdevice_notifiers - call all network notifier blocks 1981 * @val: value passed unmodified to notifier function 1982 * @dev: net_device pointer passed unmodified to notifier function 1983 * 1984 * Call all network notifier blocks. Parameters and return value 1985 * are as for raw_notifier_call_chain(). 1986 */ 1987 1988 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1989 { 1990 return call_netdevice_notifiers_extack(val, dev, NULL); 1991 } 1992 EXPORT_SYMBOL(call_netdevice_notifiers); 1993 1994 /** 1995 * call_netdevice_notifiers_mtu - call all network notifier blocks 1996 * @val: value passed unmodified to notifier function 1997 * @dev: net_device pointer passed unmodified to notifier function 1998 * @arg: additional u32 argument passed to the notifier function 1999 * 2000 * Call all network notifier blocks. Parameters and return value 2001 * are as for raw_notifier_call_chain(). 2002 */ 2003 static int call_netdevice_notifiers_mtu(unsigned long val, 2004 struct net_device *dev, u32 arg) 2005 { 2006 struct netdev_notifier_info_ext info = { 2007 .info.dev = dev, 2008 .ext.mtu = arg, 2009 }; 2010 2011 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2012 2013 return call_netdevice_notifiers_info(val, &info.info); 2014 } 2015 2016 #ifdef CONFIG_NET_INGRESS 2017 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2018 2019 void net_inc_ingress_queue(void) 2020 { 2021 static_branch_inc(&ingress_needed_key); 2022 } 2023 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2024 2025 void net_dec_ingress_queue(void) 2026 { 2027 static_branch_dec(&ingress_needed_key); 2028 } 2029 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2030 #endif 2031 2032 #ifdef CONFIG_NET_EGRESS 2033 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2034 2035 void net_inc_egress_queue(void) 2036 { 2037 static_branch_inc(&egress_needed_key); 2038 } 2039 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2040 2041 void net_dec_egress_queue(void) 2042 { 2043 static_branch_dec(&egress_needed_key); 2044 } 2045 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2046 #endif 2047 2048 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2049 EXPORT_SYMBOL(netstamp_needed_key); 2050 #ifdef CONFIG_JUMP_LABEL 2051 static atomic_t netstamp_needed_deferred; 2052 static atomic_t netstamp_wanted; 2053 static void netstamp_clear(struct work_struct *work) 2054 { 2055 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2056 int wanted; 2057 2058 wanted = atomic_add_return(deferred, &netstamp_wanted); 2059 if (wanted > 0) 2060 static_branch_enable(&netstamp_needed_key); 2061 else 2062 static_branch_disable(&netstamp_needed_key); 2063 } 2064 static DECLARE_WORK(netstamp_work, netstamp_clear); 2065 #endif 2066 2067 void net_enable_timestamp(void) 2068 { 2069 #ifdef CONFIG_JUMP_LABEL 2070 int wanted; 2071 2072 while (1) { 2073 wanted = atomic_read(&netstamp_wanted); 2074 if (wanted <= 0) 2075 break; 2076 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2077 return; 2078 } 2079 atomic_inc(&netstamp_needed_deferred); 2080 schedule_work(&netstamp_work); 2081 #else 2082 static_branch_inc(&netstamp_needed_key); 2083 #endif 2084 } 2085 EXPORT_SYMBOL(net_enable_timestamp); 2086 2087 void net_disable_timestamp(void) 2088 { 2089 #ifdef CONFIG_JUMP_LABEL 2090 int wanted; 2091 2092 while (1) { 2093 wanted = atomic_read(&netstamp_wanted); 2094 if (wanted <= 1) 2095 break; 2096 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2097 return; 2098 } 2099 atomic_dec(&netstamp_needed_deferred); 2100 schedule_work(&netstamp_work); 2101 #else 2102 static_branch_dec(&netstamp_needed_key); 2103 #endif 2104 } 2105 EXPORT_SYMBOL(net_disable_timestamp); 2106 2107 static inline void net_timestamp_set(struct sk_buff *skb) 2108 { 2109 skb->tstamp = 0; 2110 skb->mono_delivery_time = 0; 2111 if (static_branch_unlikely(&netstamp_needed_key)) 2112 skb->tstamp = ktime_get_real(); 2113 } 2114 2115 #define net_timestamp_check(COND, SKB) \ 2116 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2117 if ((COND) && !(SKB)->tstamp) \ 2118 (SKB)->tstamp = ktime_get_real(); \ 2119 } \ 2120 2121 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2122 { 2123 return __is_skb_forwardable(dev, skb, true); 2124 } 2125 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2126 2127 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2128 bool check_mtu) 2129 { 2130 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2131 2132 if (likely(!ret)) { 2133 skb->protocol = eth_type_trans(skb, dev); 2134 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2135 } 2136 2137 return ret; 2138 } 2139 2140 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2141 { 2142 return __dev_forward_skb2(dev, skb, true); 2143 } 2144 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2145 2146 /** 2147 * dev_forward_skb - loopback an skb to another netif 2148 * 2149 * @dev: destination network device 2150 * @skb: buffer to forward 2151 * 2152 * return values: 2153 * NET_RX_SUCCESS (no congestion) 2154 * NET_RX_DROP (packet was dropped, but freed) 2155 * 2156 * dev_forward_skb can be used for injecting an skb from the 2157 * start_xmit function of one device into the receive queue 2158 * of another device. 2159 * 2160 * The receiving device may be in another namespace, so 2161 * we have to clear all information in the skb that could 2162 * impact namespace isolation. 2163 */ 2164 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2165 { 2166 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2167 } 2168 EXPORT_SYMBOL_GPL(dev_forward_skb); 2169 2170 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2171 { 2172 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2173 } 2174 2175 static inline int deliver_skb(struct sk_buff *skb, 2176 struct packet_type *pt_prev, 2177 struct net_device *orig_dev) 2178 { 2179 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2180 return -ENOMEM; 2181 refcount_inc(&skb->users); 2182 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2183 } 2184 2185 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2186 struct packet_type **pt, 2187 struct net_device *orig_dev, 2188 __be16 type, 2189 struct list_head *ptype_list) 2190 { 2191 struct packet_type *ptype, *pt_prev = *pt; 2192 2193 list_for_each_entry_rcu(ptype, ptype_list, list) { 2194 if (ptype->type != type) 2195 continue; 2196 if (pt_prev) 2197 deliver_skb(skb, pt_prev, orig_dev); 2198 pt_prev = ptype; 2199 } 2200 *pt = pt_prev; 2201 } 2202 2203 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2204 { 2205 if (!ptype->af_packet_priv || !skb->sk) 2206 return false; 2207 2208 if (ptype->id_match) 2209 return ptype->id_match(ptype, skb->sk); 2210 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2211 return true; 2212 2213 return false; 2214 } 2215 2216 /** 2217 * dev_nit_active - return true if any network interface taps are in use 2218 * 2219 * @dev: network device to check for the presence of taps 2220 */ 2221 bool dev_nit_active(struct net_device *dev) 2222 { 2223 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2224 } 2225 EXPORT_SYMBOL_GPL(dev_nit_active); 2226 2227 /* 2228 * Support routine. Sends outgoing frames to any network 2229 * taps currently in use. 2230 */ 2231 2232 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2233 { 2234 struct packet_type *ptype; 2235 struct sk_buff *skb2 = NULL; 2236 struct packet_type *pt_prev = NULL; 2237 struct list_head *ptype_list = &ptype_all; 2238 2239 rcu_read_lock(); 2240 again: 2241 list_for_each_entry_rcu(ptype, ptype_list, list) { 2242 if (ptype->ignore_outgoing) 2243 continue; 2244 2245 /* Never send packets back to the socket 2246 * they originated from - MvS (miquels@drinkel.ow.org) 2247 */ 2248 if (skb_loop_sk(ptype, skb)) 2249 continue; 2250 2251 if (pt_prev) { 2252 deliver_skb(skb2, pt_prev, skb->dev); 2253 pt_prev = ptype; 2254 continue; 2255 } 2256 2257 /* need to clone skb, done only once */ 2258 skb2 = skb_clone(skb, GFP_ATOMIC); 2259 if (!skb2) 2260 goto out_unlock; 2261 2262 net_timestamp_set(skb2); 2263 2264 /* skb->nh should be correctly 2265 * set by sender, so that the second statement is 2266 * just protection against buggy protocols. 2267 */ 2268 skb_reset_mac_header(skb2); 2269 2270 if (skb_network_header(skb2) < skb2->data || 2271 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2272 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2273 ntohs(skb2->protocol), 2274 dev->name); 2275 skb_reset_network_header(skb2); 2276 } 2277 2278 skb2->transport_header = skb2->network_header; 2279 skb2->pkt_type = PACKET_OUTGOING; 2280 pt_prev = ptype; 2281 } 2282 2283 if (ptype_list == &ptype_all) { 2284 ptype_list = &dev->ptype_all; 2285 goto again; 2286 } 2287 out_unlock: 2288 if (pt_prev) { 2289 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2290 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2291 else 2292 kfree_skb(skb2); 2293 } 2294 rcu_read_unlock(); 2295 } 2296 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2297 2298 /** 2299 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2300 * @dev: Network device 2301 * @txq: number of queues available 2302 * 2303 * If real_num_tx_queues is changed the tc mappings may no longer be 2304 * valid. To resolve this verify the tc mapping remains valid and if 2305 * not NULL the mapping. With no priorities mapping to this 2306 * offset/count pair it will no longer be used. In the worst case TC0 2307 * is invalid nothing can be done so disable priority mappings. If is 2308 * expected that drivers will fix this mapping if they can before 2309 * calling netif_set_real_num_tx_queues. 2310 */ 2311 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2312 { 2313 int i; 2314 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2315 2316 /* If TC0 is invalidated disable TC mapping */ 2317 if (tc->offset + tc->count > txq) { 2318 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2319 dev->num_tc = 0; 2320 return; 2321 } 2322 2323 /* Invalidated prio to tc mappings set to TC0 */ 2324 for (i = 1; i < TC_BITMASK + 1; i++) { 2325 int q = netdev_get_prio_tc_map(dev, i); 2326 2327 tc = &dev->tc_to_txq[q]; 2328 if (tc->offset + tc->count > txq) { 2329 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", 2330 i, q); 2331 netdev_set_prio_tc_map(dev, i, 0); 2332 } 2333 } 2334 } 2335 2336 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2337 { 2338 if (dev->num_tc) { 2339 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2340 int i; 2341 2342 /* walk through the TCs and see if it falls into any of them */ 2343 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2344 if ((txq - tc->offset) < tc->count) 2345 return i; 2346 } 2347 2348 /* didn't find it, just return -1 to indicate no match */ 2349 return -1; 2350 } 2351 2352 return 0; 2353 } 2354 EXPORT_SYMBOL(netdev_txq_to_tc); 2355 2356 #ifdef CONFIG_XPS 2357 static struct static_key xps_needed __read_mostly; 2358 static struct static_key xps_rxqs_needed __read_mostly; 2359 static DEFINE_MUTEX(xps_map_mutex); 2360 #define xmap_dereference(P) \ 2361 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2362 2363 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2364 struct xps_dev_maps *old_maps, int tci, u16 index) 2365 { 2366 struct xps_map *map = NULL; 2367 int pos; 2368 2369 if (dev_maps) 2370 map = xmap_dereference(dev_maps->attr_map[tci]); 2371 if (!map) 2372 return false; 2373 2374 for (pos = map->len; pos--;) { 2375 if (map->queues[pos] != index) 2376 continue; 2377 2378 if (map->len > 1) { 2379 map->queues[pos] = map->queues[--map->len]; 2380 break; 2381 } 2382 2383 if (old_maps) 2384 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2385 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2386 kfree_rcu(map, rcu); 2387 return false; 2388 } 2389 2390 return true; 2391 } 2392 2393 static bool remove_xps_queue_cpu(struct net_device *dev, 2394 struct xps_dev_maps *dev_maps, 2395 int cpu, u16 offset, u16 count) 2396 { 2397 int num_tc = dev_maps->num_tc; 2398 bool active = false; 2399 int tci; 2400 2401 for (tci = cpu * num_tc; num_tc--; tci++) { 2402 int i, j; 2403 2404 for (i = count, j = offset; i--; j++) { 2405 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2406 break; 2407 } 2408 2409 active |= i < 0; 2410 } 2411 2412 return active; 2413 } 2414 2415 static void reset_xps_maps(struct net_device *dev, 2416 struct xps_dev_maps *dev_maps, 2417 enum xps_map_type type) 2418 { 2419 static_key_slow_dec_cpuslocked(&xps_needed); 2420 if (type == XPS_RXQS) 2421 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2422 2423 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2424 2425 kfree_rcu(dev_maps, rcu); 2426 } 2427 2428 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2429 u16 offset, u16 count) 2430 { 2431 struct xps_dev_maps *dev_maps; 2432 bool active = false; 2433 int i, j; 2434 2435 dev_maps = xmap_dereference(dev->xps_maps[type]); 2436 if (!dev_maps) 2437 return; 2438 2439 for (j = 0; j < dev_maps->nr_ids; j++) 2440 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2441 if (!active) 2442 reset_xps_maps(dev, dev_maps, type); 2443 2444 if (type == XPS_CPUS) { 2445 for (i = offset + (count - 1); count--; i--) 2446 netdev_queue_numa_node_write( 2447 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2448 } 2449 } 2450 2451 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2452 u16 count) 2453 { 2454 if (!static_key_false(&xps_needed)) 2455 return; 2456 2457 cpus_read_lock(); 2458 mutex_lock(&xps_map_mutex); 2459 2460 if (static_key_false(&xps_rxqs_needed)) 2461 clean_xps_maps(dev, XPS_RXQS, offset, count); 2462 2463 clean_xps_maps(dev, XPS_CPUS, offset, count); 2464 2465 mutex_unlock(&xps_map_mutex); 2466 cpus_read_unlock(); 2467 } 2468 2469 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2470 { 2471 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2472 } 2473 2474 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2475 u16 index, bool is_rxqs_map) 2476 { 2477 struct xps_map *new_map; 2478 int alloc_len = XPS_MIN_MAP_ALLOC; 2479 int i, pos; 2480 2481 for (pos = 0; map && pos < map->len; pos++) { 2482 if (map->queues[pos] != index) 2483 continue; 2484 return map; 2485 } 2486 2487 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2488 if (map) { 2489 if (pos < map->alloc_len) 2490 return map; 2491 2492 alloc_len = map->alloc_len * 2; 2493 } 2494 2495 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2496 * map 2497 */ 2498 if (is_rxqs_map) 2499 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2500 else 2501 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2502 cpu_to_node(attr_index)); 2503 if (!new_map) 2504 return NULL; 2505 2506 for (i = 0; i < pos; i++) 2507 new_map->queues[i] = map->queues[i]; 2508 new_map->alloc_len = alloc_len; 2509 new_map->len = pos; 2510 2511 return new_map; 2512 } 2513 2514 /* Copy xps maps at a given index */ 2515 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2516 struct xps_dev_maps *new_dev_maps, int index, 2517 int tc, bool skip_tc) 2518 { 2519 int i, tci = index * dev_maps->num_tc; 2520 struct xps_map *map; 2521 2522 /* copy maps belonging to foreign traffic classes */ 2523 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2524 if (i == tc && skip_tc) 2525 continue; 2526 2527 /* fill in the new device map from the old device map */ 2528 map = xmap_dereference(dev_maps->attr_map[tci]); 2529 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2530 } 2531 } 2532 2533 /* Must be called under cpus_read_lock */ 2534 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2535 u16 index, enum xps_map_type type) 2536 { 2537 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2538 const unsigned long *online_mask = NULL; 2539 bool active = false, copy = false; 2540 int i, j, tci, numa_node_id = -2; 2541 int maps_sz, num_tc = 1, tc = 0; 2542 struct xps_map *map, *new_map; 2543 unsigned int nr_ids; 2544 2545 if (dev->num_tc) { 2546 /* Do not allow XPS on subordinate device directly */ 2547 num_tc = dev->num_tc; 2548 if (num_tc < 0) 2549 return -EINVAL; 2550 2551 /* If queue belongs to subordinate dev use its map */ 2552 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2553 2554 tc = netdev_txq_to_tc(dev, index); 2555 if (tc < 0) 2556 return -EINVAL; 2557 } 2558 2559 mutex_lock(&xps_map_mutex); 2560 2561 dev_maps = xmap_dereference(dev->xps_maps[type]); 2562 if (type == XPS_RXQS) { 2563 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2564 nr_ids = dev->num_rx_queues; 2565 } else { 2566 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2567 if (num_possible_cpus() > 1) 2568 online_mask = cpumask_bits(cpu_online_mask); 2569 nr_ids = nr_cpu_ids; 2570 } 2571 2572 if (maps_sz < L1_CACHE_BYTES) 2573 maps_sz = L1_CACHE_BYTES; 2574 2575 /* The old dev_maps could be larger or smaller than the one we're 2576 * setting up now, as dev->num_tc or nr_ids could have been updated in 2577 * between. We could try to be smart, but let's be safe instead and only 2578 * copy foreign traffic classes if the two map sizes match. 2579 */ 2580 if (dev_maps && 2581 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2582 copy = true; 2583 2584 /* allocate memory for queue storage */ 2585 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2586 j < nr_ids;) { 2587 if (!new_dev_maps) { 2588 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2589 if (!new_dev_maps) { 2590 mutex_unlock(&xps_map_mutex); 2591 return -ENOMEM; 2592 } 2593 2594 new_dev_maps->nr_ids = nr_ids; 2595 new_dev_maps->num_tc = num_tc; 2596 } 2597 2598 tci = j * num_tc + tc; 2599 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2600 2601 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2602 if (!map) 2603 goto error; 2604 2605 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2606 } 2607 2608 if (!new_dev_maps) 2609 goto out_no_new_maps; 2610 2611 if (!dev_maps) { 2612 /* Increment static keys at most once per type */ 2613 static_key_slow_inc_cpuslocked(&xps_needed); 2614 if (type == XPS_RXQS) 2615 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2616 } 2617 2618 for (j = 0; j < nr_ids; j++) { 2619 bool skip_tc = false; 2620 2621 tci = j * num_tc + tc; 2622 if (netif_attr_test_mask(j, mask, nr_ids) && 2623 netif_attr_test_online(j, online_mask, nr_ids)) { 2624 /* add tx-queue to CPU/rx-queue maps */ 2625 int pos = 0; 2626 2627 skip_tc = true; 2628 2629 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2630 while ((pos < map->len) && (map->queues[pos] != index)) 2631 pos++; 2632 2633 if (pos == map->len) 2634 map->queues[map->len++] = index; 2635 #ifdef CONFIG_NUMA 2636 if (type == XPS_CPUS) { 2637 if (numa_node_id == -2) 2638 numa_node_id = cpu_to_node(j); 2639 else if (numa_node_id != cpu_to_node(j)) 2640 numa_node_id = -1; 2641 } 2642 #endif 2643 } 2644 2645 if (copy) 2646 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2647 skip_tc); 2648 } 2649 2650 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2651 2652 /* Cleanup old maps */ 2653 if (!dev_maps) 2654 goto out_no_old_maps; 2655 2656 for (j = 0; j < dev_maps->nr_ids; j++) { 2657 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2658 map = xmap_dereference(dev_maps->attr_map[tci]); 2659 if (!map) 2660 continue; 2661 2662 if (copy) { 2663 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2664 if (map == new_map) 2665 continue; 2666 } 2667 2668 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2669 kfree_rcu(map, rcu); 2670 } 2671 } 2672 2673 old_dev_maps = dev_maps; 2674 2675 out_no_old_maps: 2676 dev_maps = new_dev_maps; 2677 active = true; 2678 2679 out_no_new_maps: 2680 if (type == XPS_CPUS) 2681 /* update Tx queue numa node */ 2682 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2683 (numa_node_id >= 0) ? 2684 numa_node_id : NUMA_NO_NODE); 2685 2686 if (!dev_maps) 2687 goto out_no_maps; 2688 2689 /* removes tx-queue from unused CPUs/rx-queues */ 2690 for (j = 0; j < dev_maps->nr_ids; j++) { 2691 tci = j * dev_maps->num_tc; 2692 2693 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2694 if (i == tc && 2695 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2696 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2697 continue; 2698 2699 active |= remove_xps_queue(dev_maps, 2700 copy ? old_dev_maps : NULL, 2701 tci, index); 2702 } 2703 } 2704 2705 if (old_dev_maps) 2706 kfree_rcu(old_dev_maps, rcu); 2707 2708 /* free map if not active */ 2709 if (!active) 2710 reset_xps_maps(dev, dev_maps, type); 2711 2712 out_no_maps: 2713 mutex_unlock(&xps_map_mutex); 2714 2715 return 0; 2716 error: 2717 /* remove any maps that we added */ 2718 for (j = 0; j < nr_ids; j++) { 2719 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2720 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2721 map = copy ? 2722 xmap_dereference(dev_maps->attr_map[tci]) : 2723 NULL; 2724 if (new_map && new_map != map) 2725 kfree(new_map); 2726 } 2727 } 2728 2729 mutex_unlock(&xps_map_mutex); 2730 2731 kfree(new_dev_maps); 2732 return -ENOMEM; 2733 } 2734 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2735 2736 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2737 u16 index) 2738 { 2739 int ret; 2740 2741 cpus_read_lock(); 2742 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2743 cpus_read_unlock(); 2744 2745 return ret; 2746 } 2747 EXPORT_SYMBOL(netif_set_xps_queue); 2748 2749 #endif 2750 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2751 { 2752 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2753 2754 /* Unbind any subordinate channels */ 2755 while (txq-- != &dev->_tx[0]) { 2756 if (txq->sb_dev) 2757 netdev_unbind_sb_channel(dev, txq->sb_dev); 2758 } 2759 } 2760 2761 void netdev_reset_tc(struct net_device *dev) 2762 { 2763 #ifdef CONFIG_XPS 2764 netif_reset_xps_queues_gt(dev, 0); 2765 #endif 2766 netdev_unbind_all_sb_channels(dev); 2767 2768 /* Reset TC configuration of device */ 2769 dev->num_tc = 0; 2770 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2771 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2772 } 2773 EXPORT_SYMBOL(netdev_reset_tc); 2774 2775 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2776 { 2777 if (tc >= dev->num_tc) 2778 return -EINVAL; 2779 2780 #ifdef CONFIG_XPS 2781 netif_reset_xps_queues(dev, offset, count); 2782 #endif 2783 dev->tc_to_txq[tc].count = count; 2784 dev->tc_to_txq[tc].offset = offset; 2785 return 0; 2786 } 2787 EXPORT_SYMBOL(netdev_set_tc_queue); 2788 2789 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2790 { 2791 if (num_tc > TC_MAX_QUEUE) 2792 return -EINVAL; 2793 2794 #ifdef CONFIG_XPS 2795 netif_reset_xps_queues_gt(dev, 0); 2796 #endif 2797 netdev_unbind_all_sb_channels(dev); 2798 2799 dev->num_tc = num_tc; 2800 return 0; 2801 } 2802 EXPORT_SYMBOL(netdev_set_num_tc); 2803 2804 void netdev_unbind_sb_channel(struct net_device *dev, 2805 struct net_device *sb_dev) 2806 { 2807 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2808 2809 #ifdef CONFIG_XPS 2810 netif_reset_xps_queues_gt(sb_dev, 0); 2811 #endif 2812 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2813 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2814 2815 while (txq-- != &dev->_tx[0]) { 2816 if (txq->sb_dev == sb_dev) 2817 txq->sb_dev = NULL; 2818 } 2819 } 2820 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2821 2822 int netdev_bind_sb_channel_queue(struct net_device *dev, 2823 struct net_device *sb_dev, 2824 u8 tc, u16 count, u16 offset) 2825 { 2826 /* Make certain the sb_dev and dev are already configured */ 2827 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2828 return -EINVAL; 2829 2830 /* We cannot hand out queues we don't have */ 2831 if ((offset + count) > dev->real_num_tx_queues) 2832 return -EINVAL; 2833 2834 /* Record the mapping */ 2835 sb_dev->tc_to_txq[tc].count = count; 2836 sb_dev->tc_to_txq[tc].offset = offset; 2837 2838 /* Provide a way for Tx queue to find the tc_to_txq map or 2839 * XPS map for itself. 2840 */ 2841 while (count--) 2842 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2843 2844 return 0; 2845 } 2846 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2847 2848 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2849 { 2850 /* Do not use a multiqueue device to represent a subordinate channel */ 2851 if (netif_is_multiqueue(dev)) 2852 return -ENODEV; 2853 2854 /* We allow channels 1 - 32767 to be used for subordinate channels. 2855 * Channel 0 is meant to be "native" mode and used only to represent 2856 * the main root device. We allow writing 0 to reset the device back 2857 * to normal mode after being used as a subordinate channel. 2858 */ 2859 if (channel > S16_MAX) 2860 return -EINVAL; 2861 2862 dev->num_tc = -channel; 2863 2864 return 0; 2865 } 2866 EXPORT_SYMBOL(netdev_set_sb_channel); 2867 2868 /* 2869 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2870 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2871 */ 2872 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2873 { 2874 bool disabling; 2875 int rc; 2876 2877 disabling = txq < dev->real_num_tx_queues; 2878 2879 if (txq < 1 || txq > dev->num_tx_queues) 2880 return -EINVAL; 2881 2882 if (dev->reg_state == NETREG_REGISTERED || 2883 dev->reg_state == NETREG_UNREGISTERING) { 2884 ASSERT_RTNL(); 2885 2886 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2887 txq); 2888 if (rc) 2889 return rc; 2890 2891 if (dev->num_tc) 2892 netif_setup_tc(dev, txq); 2893 2894 dev_qdisc_change_real_num_tx(dev, txq); 2895 2896 dev->real_num_tx_queues = txq; 2897 2898 if (disabling) { 2899 synchronize_net(); 2900 qdisc_reset_all_tx_gt(dev, txq); 2901 #ifdef CONFIG_XPS 2902 netif_reset_xps_queues_gt(dev, txq); 2903 #endif 2904 } 2905 } else { 2906 dev->real_num_tx_queues = txq; 2907 } 2908 2909 return 0; 2910 } 2911 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2912 2913 #ifdef CONFIG_SYSFS 2914 /** 2915 * netif_set_real_num_rx_queues - set actual number of RX queues used 2916 * @dev: Network device 2917 * @rxq: Actual number of RX queues 2918 * 2919 * This must be called either with the rtnl_lock held or before 2920 * registration of the net device. Returns 0 on success, or a 2921 * negative error code. If called before registration, it always 2922 * succeeds. 2923 */ 2924 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2925 { 2926 int rc; 2927 2928 if (rxq < 1 || rxq > dev->num_rx_queues) 2929 return -EINVAL; 2930 2931 if (dev->reg_state == NETREG_REGISTERED) { 2932 ASSERT_RTNL(); 2933 2934 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2935 rxq); 2936 if (rc) 2937 return rc; 2938 } 2939 2940 dev->real_num_rx_queues = rxq; 2941 return 0; 2942 } 2943 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2944 #endif 2945 2946 /** 2947 * netif_set_real_num_queues - set actual number of RX and TX queues used 2948 * @dev: Network device 2949 * @txq: Actual number of TX queues 2950 * @rxq: Actual number of RX queues 2951 * 2952 * Set the real number of both TX and RX queues. 2953 * Does nothing if the number of queues is already correct. 2954 */ 2955 int netif_set_real_num_queues(struct net_device *dev, 2956 unsigned int txq, unsigned int rxq) 2957 { 2958 unsigned int old_rxq = dev->real_num_rx_queues; 2959 int err; 2960 2961 if (txq < 1 || txq > dev->num_tx_queues || 2962 rxq < 1 || rxq > dev->num_rx_queues) 2963 return -EINVAL; 2964 2965 /* Start from increases, so the error path only does decreases - 2966 * decreases can't fail. 2967 */ 2968 if (rxq > dev->real_num_rx_queues) { 2969 err = netif_set_real_num_rx_queues(dev, rxq); 2970 if (err) 2971 return err; 2972 } 2973 if (txq > dev->real_num_tx_queues) { 2974 err = netif_set_real_num_tx_queues(dev, txq); 2975 if (err) 2976 goto undo_rx; 2977 } 2978 if (rxq < dev->real_num_rx_queues) 2979 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2980 if (txq < dev->real_num_tx_queues) 2981 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2982 2983 return 0; 2984 undo_rx: 2985 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2986 return err; 2987 } 2988 EXPORT_SYMBOL(netif_set_real_num_queues); 2989 2990 /** 2991 * netif_get_num_default_rss_queues - default number of RSS queues 2992 * 2993 * This routine should set an upper limit on the number of RSS queues 2994 * used by default by multiqueue devices. 2995 */ 2996 int netif_get_num_default_rss_queues(void) 2997 { 2998 return is_kdump_kernel() ? 2999 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3000 } 3001 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3002 3003 static void __netif_reschedule(struct Qdisc *q) 3004 { 3005 struct softnet_data *sd; 3006 unsigned long flags; 3007 3008 local_irq_save(flags); 3009 sd = this_cpu_ptr(&softnet_data); 3010 q->next_sched = NULL; 3011 *sd->output_queue_tailp = q; 3012 sd->output_queue_tailp = &q->next_sched; 3013 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3014 local_irq_restore(flags); 3015 } 3016 3017 void __netif_schedule(struct Qdisc *q) 3018 { 3019 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3020 __netif_reschedule(q); 3021 } 3022 EXPORT_SYMBOL(__netif_schedule); 3023 3024 struct dev_kfree_skb_cb { 3025 enum skb_free_reason reason; 3026 }; 3027 3028 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3029 { 3030 return (struct dev_kfree_skb_cb *)skb->cb; 3031 } 3032 3033 void netif_schedule_queue(struct netdev_queue *txq) 3034 { 3035 rcu_read_lock(); 3036 if (!netif_xmit_stopped(txq)) { 3037 struct Qdisc *q = rcu_dereference(txq->qdisc); 3038 3039 __netif_schedule(q); 3040 } 3041 rcu_read_unlock(); 3042 } 3043 EXPORT_SYMBOL(netif_schedule_queue); 3044 3045 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3046 { 3047 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3048 struct Qdisc *q; 3049 3050 rcu_read_lock(); 3051 q = rcu_dereference(dev_queue->qdisc); 3052 __netif_schedule(q); 3053 rcu_read_unlock(); 3054 } 3055 } 3056 EXPORT_SYMBOL(netif_tx_wake_queue); 3057 3058 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3059 { 3060 unsigned long flags; 3061 3062 if (unlikely(!skb)) 3063 return; 3064 3065 if (likely(refcount_read(&skb->users) == 1)) { 3066 smp_rmb(); 3067 refcount_set(&skb->users, 0); 3068 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3069 return; 3070 } 3071 get_kfree_skb_cb(skb)->reason = reason; 3072 local_irq_save(flags); 3073 skb->next = __this_cpu_read(softnet_data.completion_queue); 3074 __this_cpu_write(softnet_data.completion_queue, skb); 3075 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3076 local_irq_restore(flags); 3077 } 3078 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3079 3080 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3081 { 3082 if (in_hardirq() || irqs_disabled()) 3083 __dev_kfree_skb_irq(skb, reason); 3084 else 3085 dev_kfree_skb(skb); 3086 } 3087 EXPORT_SYMBOL(__dev_kfree_skb_any); 3088 3089 3090 /** 3091 * netif_device_detach - mark device as removed 3092 * @dev: network device 3093 * 3094 * Mark device as removed from system and therefore no longer available. 3095 */ 3096 void netif_device_detach(struct net_device *dev) 3097 { 3098 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3099 netif_running(dev)) { 3100 netif_tx_stop_all_queues(dev); 3101 } 3102 } 3103 EXPORT_SYMBOL(netif_device_detach); 3104 3105 /** 3106 * netif_device_attach - mark device as attached 3107 * @dev: network device 3108 * 3109 * Mark device as attached from system and restart if needed. 3110 */ 3111 void netif_device_attach(struct net_device *dev) 3112 { 3113 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3114 netif_running(dev)) { 3115 netif_tx_wake_all_queues(dev); 3116 __netdev_watchdog_up(dev); 3117 } 3118 } 3119 EXPORT_SYMBOL(netif_device_attach); 3120 3121 /* 3122 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3123 * to be used as a distribution range. 3124 */ 3125 static u16 skb_tx_hash(const struct net_device *dev, 3126 const struct net_device *sb_dev, 3127 struct sk_buff *skb) 3128 { 3129 u32 hash; 3130 u16 qoffset = 0; 3131 u16 qcount = dev->real_num_tx_queues; 3132 3133 if (dev->num_tc) { 3134 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3135 3136 qoffset = sb_dev->tc_to_txq[tc].offset; 3137 qcount = sb_dev->tc_to_txq[tc].count; 3138 if (unlikely(!qcount)) { 3139 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3140 sb_dev->name, qoffset, tc); 3141 qoffset = 0; 3142 qcount = dev->real_num_tx_queues; 3143 } 3144 } 3145 3146 if (skb_rx_queue_recorded(skb)) { 3147 hash = skb_get_rx_queue(skb); 3148 if (hash >= qoffset) 3149 hash -= qoffset; 3150 while (unlikely(hash >= qcount)) 3151 hash -= qcount; 3152 return hash + qoffset; 3153 } 3154 3155 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3156 } 3157 3158 static void skb_warn_bad_offload(const struct sk_buff *skb) 3159 { 3160 static const netdev_features_t null_features; 3161 struct net_device *dev = skb->dev; 3162 const char *name = ""; 3163 3164 if (!net_ratelimit()) 3165 return; 3166 3167 if (dev) { 3168 if (dev->dev.parent) 3169 name = dev_driver_string(dev->dev.parent); 3170 else 3171 name = netdev_name(dev); 3172 } 3173 skb_dump(KERN_WARNING, skb, false); 3174 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3175 name, dev ? &dev->features : &null_features, 3176 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3177 } 3178 3179 /* 3180 * Invalidate hardware checksum when packet is to be mangled, and 3181 * complete checksum manually on outgoing path. 3182 */ 3183 int skb_checksum_help(struct sk_buff *skb) 3184 { 3185 __wsum csum; 3186 int ret = 0, offset; 3187 3188 if (skb->ip_summed == CHECKSUM_COMPLETE) 3189 goto out_set_summed; 3190 3191 if (unlikely(skb_is_gso(skb))) { 3192 skb_warn_bad_offload(skb); 3193 return -EINVAL; 3194 } 3195 3196 /* Before computing a checksum, we should make sure no frag could 3197 * be modified by an external entity : checksum could be wrong. 3198 */ 3199 if (skb_has_shared_frag(skb)) { 3200 ret = __skb_linearize(skb); 3201 if (ret) 3202 goto out; 3203 } 3204 3205 offset = skb_checksum_start_offset(skb); 3206 BUG_ON(offset >= skb_headlen(skb)); 3207 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3208 3209 offset += skb->csum_offset; 3210 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3211 3212 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3213 if (ret) 3214 goto out; 3215 3216 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3217 out_set_summed: 3218 skb->ip_summed = CHECKSUM_NONE; 3219 out: 3220 return ret; 3221 } 3222 EXPORT_SYMBOL(skb_checksum_help); 3223 3224 int skb_crc32c_csum_help(struct sk_buff *skb) 3225 { 3226 __le32 crc32c_csum; 3227 int ret = 0, offset, start; 3228 3229 if (skb->ip_summed != CHECKSUM_PARTIAL) 3230 goto out; 3231 3232 if (unlikely(skb_is_gso(skb))) 3233 goto out; 3234 3235 /* Before computing a checksum, we should make sure no frag could 3236 * be modified by an external entity : checksum could be wrong. 3237 */ 3238 if (unlikely(skb_has_shared_frag(skb))) { 3239 ret = __skb_linearize(skb); 3240 if (ret) 3241 goto out; 3242 } 3243 start = skb_checksum_start_offset(skb); 3244 offset = start + offsetof(struct sctphdr, checksum); 3245 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3246 ret = -EINVAL; 3247 goto out; 3248 } 3249 3250 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3251 if (ret) 3252 goto out; 3253 3254 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3255 skb->len - start, ~(__u32)0, 3256 crc32c_csum_stub)); 3257 *(__le32 *)(skb->data + offset) = crc32c_csum; 3258 skb->ip_summed = CHECKSUM_NONE; 3259 skb->csum_not_inet = 0; 3260 out: 3261 return ret; 3262 } 3263 3264 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3265 { 3266 __be16 type = skb->protocol; 3267 3268 /* Tunnel gso handlers can set protocol to ethernet. */ 3269 if (type == htons(ETH_P_TEB)) { 3270 struct ethhdr *eth; 3271 3272 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3273 return 0; 3274 3275 eth = (struct ethhdr *)skb->data; 3276 type = eth->h_proto; 3277 } 3278 3279 return __vlan_get_protocol(skb, type, depth); 3280 } 3281 3282 /* openvswitch calls this on rx path, so we need a different check. 3283 */ 3284 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3285 { 3286 if (tx_path) 3287 return skb->ip_summed != CHECKSUM_PARTIAL && 3288 skb->ip_summed != CHECKSUM_UNNECESSARY; 3289 3290 return skb->ip_summed == CHECKSUM_NONE; 3291 } 3292 3293 /** 3294 * __skb_gso_segment - Perform segmentation on skb. 3295 * @skb: buffer to segment 3296 * @features: features for the output path (see dev->features) 3297 * @tx_path: whether it is called in TX path 3298 * 3299 * This function segments the given skb and returns a list of segments. 3300 * 3301 * It may return NULL if the skb requires no segmentation. This is 3302 * only possible when GSO is used for verifying header integrity. 3303 * 3304 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3305 */ 3306 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3307 netdev_features_t features, bool tx_path) 3308 { 3309 struct sk_buff *segs; 3310 3311 if (unlikely(skb_needs_check(skb, tx_path))) { 3312 int err; 3313 3314 /* We're going to init ->check field in TCP or UDP header */ 3315 err = skb_cow_head(skb, 0); 3316 if (err < 0) 3317 return ERR_PTR(err); 3318 } 3319 3320 /* Only report GSO partial support if it will enable us to 3321 * support segmentation on this frame without needing additional 3322 * work. 3323 */ 3324 if (features & NETIF_F_GSO_PARTIAL) { 3325 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3326 struct net_device *dev = skb->dev; 3327 3328 partial_features |= dev->features & dev->gso_partial_features; 3329 if (!skb_gso_ok(skb, features | partial_features)) 3330 features &= ~NETIF_F_GSO_PARTIAL; 3331 } 3332 3333 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3334 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3335 3336 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3337 SKB_GSO_CB(skb)->encap_level = 0; 3338 3339 skb_reset_mac_header(skb); 3340 skb_reset_mac_len(skb); 3341 3342 segs = skb_mac_gso_segment(skb, features); 3343 3344 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3345 skb_warn_bad_offload(skb); 3346 3347 return segs; 3348 } 3349 EXPORT_SYMBOL(__skb_gso_segment); 3350 3351 /* Take action when hardware reception checksum errors are detected. */ 3352 #ifdef CONFIG_BUG 3353 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3354 { 3355 netdev_err(dev, "hw csum failure\n"); 3356 skb_dump(KERN_ERR, skb, true); 3357 dump_stack(); 3358 } 3359 3360 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3361 { 3362 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3363 } 3364 EXPORT_SYMBOL(netdev_rx_csum_fault); 3365 #endif 3366 3367 /* XXX: check that highmem exists at all on the given machine. */ 3368 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3369 { 3370 #ifdef CONFIG_HIGHMEM 3371 int i; 3372 3373 if (!(dev->features & NETIF_F_HIGHDMA)) { 3374 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3375 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3376 3377 if (PageHighMem(skb_frag_page(frag))) 3378 return 1; 3379 } 3380 } 3381 #endif 3382 return 0; 3383 } 3384 3385 /* If MPLS offload request, verify we are testing hardware MPLS features 3386 * instead of standard features for the netdev. 3387 */ 3388 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3389 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3390 netdev_features_t features, 3391 __be16 type) 3392 { 3393 if (eth_p_mpls(type)) 3394 features &= skb->dev->mpls_features; 3395 3396 return features; 3397 } 3398 #else 3399 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3400 netdev_features_t features, 3401 __be16 type) 3402 { 3403 return features; 3404 } 3405 #endif 3406 3407 static netdev_features_t harmonize_features(struct sk_buff *skb, 3408 netdev_features_t features) 3409 { 3410 __be16 type; 3411 3412 type = skb_network_protocol(skb, NULL); 3413 features = net_mpls_features(skb, features, type); 3414 3415 if (skb->ip_summed != CHECKSUM_NONE && 3416 !can_checksum_protocol(features, type)) { 3417 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3418 } 3419 if (illegal_highdma(skb->dev, skb)) 3420 features &= ~NETIF_F_SG; 3421 3422 return features; 3423 } 3424 3425 netdev_features_t passthru_features_check(struct sk_buff *skb, 3426 struct net_device *dev, 3427 netdev_features_t features) 3428 { 3429 return features; 3430 } 3431 EXPORT_SYMBOL(passthru_features_check); 3432 3433 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3434 struct net_device *dev, 3435 netdev_features_t features) 3436 { 3437 return vlan_features_check(skb, features); 3438 } 3439 3440 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3441 struct net_device *dev, 3442 netdev_features_t features) 3443 { 3444 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3445 3446 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3447 return features & ~NETIF_F_GSO_MASK; 3448 3449 if (!skb_shinfo(skb)->gso_type) { 3450 skb_warn_bad_offload(skb); 3451 return features & ~NETIF_F_GSO_MASK; 3452 } 3453 3454 /* Support for GSO partial features requires software 3455 * intervention before we can actually process the packets 3456 * so we need to strip support for any partial features now 3457 * and we can pull them back in after we have partially 3458 * segmented the frame. 3459 */ 3460 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3461 features &= ~dev->gso_partial_features; 3462 3463 /* Make sure to clear the IPv4 ID mangling feature if the 3464 * IPv4 header has the potential to be fragmented. 3465 */ 3466 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3467 struct iphdr *iph = skb->encapsulation ? 3468 inner_ip_hdr(skb) : ip_hdr(skb); 3469 3470 if (!(iph->frag_off & htons(IP_DF))) 3471 features &= ~NETIF_F_TSO_MANGLEID; 3472 } 3473 3474 return features; 3475 } 3476 3477 netdev_features_t netif_skb_features(struct sk_buff *skb) 3478 { 3479 struct net_device *dev = skb->dev; 3480 netdev_features_t features = dev->features; 3481 3482 if (skb_is_gso(skb)) 3483 features = gso_features_check(skb, dev, features); 3484 3485 /* If encapsulation offload request, verify we are testing 3486 * hardware encapsulation features instead of standard 3487 * features for the netdev 3488 */ 3489 if (skb->encapsulation) 3490 features &= dev->hw_enc_features; 3491 3492 if (skb_vlan_tagged(skb)) 3493 features = netdev_intersect_features(features, 3494 dev->vlan_features | 3495 NETIF_F_HW_VLAN_CTAG_TX | 3496 NETIF_F_HW_VLAN_STAG_TX); 3497 3498 if (dev->netdev_ops->ndo_features_check) 3499 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3500 features); 3501 else 3502 features &= dflt_features_check(skb, dev, features); 3503 3504 return harmonize_features(skb, features); 3505 } 3506 EXPORT_SYMBOL(netif_skb_features); 3507 3508 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3509 struct netdev_queue *txq, bool more) 3510 { 3511 unsigned int len; 3512 int rc; 3513 3514 if (dev_nit_active(dev)) 3515 dev_queue_xmit_nit(skb, dev); 3516 3517 len = skb->len; 3518 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3519 trace_net_dev_start_xmit(skb, dev); 3520 rc = netdev_start_xmit(skb, dev, txq, more); 3521 trace_net_dev_xmit(skb, rc, dev, len); 3522 3523 return rc; 3524 } 3525 3526 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3527 struct netdev_queue *txq, int *ret) 3528 { 3529 struct sk_buff *skb = first; 3530 int rc = NETDEV_TX_OK; 3531 3532 while (skb) { 3533 struct sk_buff *next = skb->next; 3534 3535 skb_mark_not_on_list(skb); 3536 rc = xmit_one(skb, dev, txq, next != NULL); 3537 if (unlikely(!dev_xmit_complete(rc))) { 3538 skb->next = next; 3539 goto out; 3540 } 3541 3542 skb = next; 3543 if (netif_tx_queue_stopped(txq) && skb) { 3544 rc = NETDEV_TX_BUSY; 3545 break; 3546 } 3547 } 3548 3549 out: 3550 *ret = rc; 3551 return skb; 3552 } 3553 3554 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3555 netdev_features_t features) 3556 { 3557 if (skb_vlan_tag_present(skb) && 3558 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3559 skb = __vlan_hwaccel_push_inside(skb); 3560 return skb; 3561 } 3562 3563 int skb_csum_hwoffload_help(struct sk_buff *skb, 3564 const netdev_features_t features) 3565 { 3566 if (unlikely(skb_csum_is_sctp(skb))) 3567 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3568 skb_crc32c_csum_help(skb); 3569 3570 if (features & NETIF_F_HW_CSUM) 3571 return 0; 3572 3573 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3574 switch (skb->csum_offset) { 3575 case offsetof(struct tcphdr, check): 3576 case offsetof(struct udphdr, check): 3577 return 0; 3578 } 3579 } 3580 3581 return skb_checksum_help(skb); 3582 } 3583 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3584 3585 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3586 { 3587 netdev_features_t features; 3588 3589 features = netif_skb_features(skb); 3590 skb = validate_xmit_vlan(skb, features); 3591 if (unlikely(!skb)) 3592 goto out_null; 3593 3594 skb = sk_validate_xmit_skb(skb, dev); 3595 if (unlikely(!skb)) 3596 goto out_null; 3597 3598 if (netif_needs_gso(skb, features)) { 3599 struct sk_buff *segs; 3600 3601 segs = skb_gso_segment(skb, features); 3602 if (IS_ERR(segs)) { 3603 goto out_kfree_skb; 3604 } else if (segs) { 3605 consume_skb(skb); 3606 skb = segs; 3607 } 3608 } else { 3609 if (skb_needs_linearize(skb, features) && 3610 __skb_linearize(skb)) 3611 goto out_kfree_skb; 3612 3613 /* If packet is not checksummed and device does not 3614 * support checksumming for this protocol, complete 3615 * checksumming here. 3616 */ 3617 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3618 if (skb->encapsulation) 3619 skb_set_inner_transport_header(skb, 3620 skb_checksum_start_offset(skb)); 3621 else 3622 skb_set_transport_header(skb, 3623 skb_checksum_start_offset(skb)); 3624 if (skb_csum_hwoffload_help(skb, features)) 3625 goto out_kfree_skb; 3626 } 3627 } 3628 3629 skb = validate_xmit_xfrm(skb, features, again); 3630 3631 return skb; 3632 3633 out_kfree_skb: 3634 kfree_skb(skb); 3635 out_null: 3636 dev_core_stats_tx_dropped_inc(dev); 3637 return NULL; 3638 } 3639 3640 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3641 { 3642 struct sk_buff *next, *head = NULL, *tail; 3643 3644 for (; skb != NULL; skb = next) { 3645 next = skb->next; 3646 skb_mark_not_on_list(skb); 3647 3648 /* in case skb wont be segmented, point to itself */ 3649 skb->prev = skb; 3650 3651 skb = validate_xmit_skb(skb, dev, again); 3652 if (!skb) 3653 continue; 3654 3655 if (!head) 3656 head = skb; 3657 else 3658 tail->next = skb; 3659 /* If skb was segmented, skb->prev points to 3660 * the last segment. If not, it still contains skb. 3661 */ 3662 tail = skb->prev; 3663 } 3664 return head; 3665 } 3666 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3667 3668 static void qdisc_pkt_len_init(struct sk_buff *skb) 3669 { 3670 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3671 3672 qdisc_skb_cb(skb)->pkt_len = skb->len; 3673 3674 /* To get more precise estimation of bytes sent on wire, 3675 * we add to pkt_len the headers size of all segments 3676 */ 3677 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3678 unsigned int hdr_len; 3679 u16 gso_segs = shinfo->gso_segs; 3680 3681 /* mac layer + network layer */ 3682 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3683 3684 /* + transport layer */ 3685 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3686 const struct tcphdr *th; 3687 struct tcphdr _tcphdr; 3688 3689 th = skb_header_pointer(skb, skb_transport_offset(skb), 3690 sizeof(_tcphdr), &_tcphdr); 3691 if (likely(th)) 3692 hdr_len += __tcp_hdrlen(th); 3693 } else { 3694 struct udphdr _udphdr; 3695 3696 if (skb_header_pointer(skb, skb_transport_offset(skb), 3697 sizeof(_udphdr), &_udphdr)) 3698 hdr_len += sizeof(struct udphdr); 3699 } 3700 3701 if (shinfo->gso_type & SKB_GSO_DODGY) 3702 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3703 shinfo->gso_size); 3704 3705 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3706 } 3707 } 3708 3709 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3710 struct sk_buff **to_free, 3711 struct netdev_queue *txq) 3712 { 3713 int rc; 3714 3715 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3716 if (rc == NET_XMIT_SUCCESS) 3717 trace_qdisc_enqueue(q, txq, skb); 3718 return rc; 3719 } 3720 3721 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3722 struct net_device *dev, 3723 struct netdev_queue *txq) 3724 { 3725 spinlock_t *root_lock = qdisc_lock(q); 3726 struct sk_buff *to_free = NULL; 3727 bool contended; 3728 int rc; 3729 3730 qdisc_calculate_pkt_len(skb, q); 3731 3732 if (q->flags & TCQ_F_NOLOCK) { 3733 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3734 qdisc_run_begin(q)) { 3735 /* Retest nolock_qdisc_is_empty() within the protection 3736 * of q->seqlock to protect from racing with requeuing. 3737 */ 3738 if (unlikely(!nolock_qdisc_is_empty(q))) { 3739 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3740 __qdisc_run(q); 3741 qdisc_run_end(q); 3742 3743 goto no_lock_out; 3744 } 3745 3746 qdisc_bstats_cpu_update(q, skb); 3747 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3748 !nolock_qdisc_is_empty(q)) 3749 __qdisc_run(q); 3750 3751 qdisc_run_end(q); 3752 return NET_XMIT_SUCCESS; 3753 } 3754 3755 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3756 qdisc_run(q); 3757 3758 no_lock_out: 3759 if (unlikely(to_free)) 3760 kfree_skb_list_reason(to_free, 3761 SKB_DROP_REASON_QDISC_DROP); 3762 return rc; 3763 } 3764 3765 /* 3766 * Heuristic to force contended enqueues to serialize on a 3767 * separate lock before trying to get qdisc main lock. 3768 * This permits qdisc->running owner to get the lock more 3769 * often and dequeue packets faster. 3770 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3771 * and then other tasks will only enqueue packets. The packets will be 3772 * sent after the qdisc owner is scheduled again. To prevent this 3773 * scenario the task always serialize on the lock. 3774 */ 3775 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3776 if (unlikely(contended)) 3777 spin_lock(&q->busylock); 3778 3779 spin_lock(root_lock); 3780 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3781 __qdisc_drop(skb, &to_free); 3782 rc = NET_XMIT_DROP; 3783 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3784 qdisc_run_begin(q)) { 3785 /* 3786 * This is a work-conserving queue; there are no old skbs 3787 * waiting to be sent out; and the qdisc is not running - 3788 * xmit the skb directly. 3789 */ 3790 3791 qdisc_bstats_update(q, skb); 3792 3793 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3794 if (unlikely(contended)) { 3795 spin_unlock(&q->busylock); 3796 contended = false; 3797 } 3798 __qdisc_run(q); 3799 } 3800 3801 qdisc_run_end(q); 3802 rc = NET_XMIT_SUCCESS; 3803 } else { 3804 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3805 if (qdisc_run_begin(q)) { 3806 if (unlikely(contended)) { 3807 spin_unlock(&q->busylock); 3808 contended = false; 3809 } 3810 __qdisc_run(q); 3811 qdisc_run_end(q); 3812 } 3813 } 3814 spin_unlock(root_lock); 3815 if (unlikely(to_free)) 3816 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3817 if (unlikely(contended)) 3818 spin_unlock(&q->busylock); 3819 return rc; 3820 } 3821 3822 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3823 static void skb_update_prio(struct sk_buff *skb) 3824 { 3825 const struct netprio_map *map; 3826 const struct sock *sk; 3827 unsigned int prioidx; 3828 3829 if (skb->priority) 3830 return; 3831 map = rcu_dereference_bh(skb->dev->priomap); 3832 if (!map) 3833 return; 3834 sk = skb_to_full_sk(skb); 3835 if (!sk) 3836 return; 3837 3838 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3839 3840 if (prioidx < map->priomap_len) 3841 skb->priority = map->priomap[prioidx]; 3842 } 3843 #else 3844 #define skb_update_prio(skb) 3845 #endif 3846 3847 /** 3848 * dev_loopback_xmit - loop back @skb 3849 * @net: network namespace this loopback is happening in 3850 * @sk: sk needed to be a netfilter okfn 3851 * @skb: buffer to transmit 3852 */ 3853 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3854 { 3855 skb_reset_mac_header(skb); 3856 __skb_pull(skb, skb_network_offset(skb)); 3857 skb->pkt_type = PACKET_LOOPBACK; 3858 if (skb->ip_summed == CHECKSUM_NONE) 3859 skb->ip_summed = CHECKSUM_UNNECESSARY; 3860 WARN_ON(!skb_dst(skb)); 3861 skb_dst_force(skb); 3862 netif_rx(skb); 3863 return 0; 3864 } 3865 EXPORT_SYMBOL(dev_loopback_xmit); 3866 3867 #ifdef CONFIG_NET_EGRESS 3868 static struct sk_buff * 3869 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3870 { 3871 #ifdef CONFIG_NET_CLS_ACT 3872 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3873 struct tcf_result cl_res; 3874 3875 if (!miniq) 3876 return skb; 3877 3878 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3879 tc_skb_cb(skb)->mru = 0; 3880 tc_skb_cb(skb)->post_ct = false; 3881 mini_qdisc_bstats_cpu_update(miniq, skb); 3882 3883 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3884 case TC_ACT_OK: 3885 case TC_ACT_RECLASSIFY: 3886 skb->tc_index = TC_H_MIN(cl_res.classid); 3887 break; 3888 case TC_ACT_SHOT: 3889 mini_qdisc_qstats_cpu_drop(miniq); 3890 *ret = NET_XMIT_DROP; 3891 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3892 return NULL; 3893 case TC_ACT_STOLEN: 3894 case TC_ACT_QUEUED: 3895 case TC_ACT_TRAP: 3896 *ret = NET_XMIT_SUCCESS; 3897 consume_skb(skb); 3898 return NULL; 3899 case TC_ACT_REDIRECT: 3900 /* No need to push/pop skb's mac_header here on egress! */ 3901 skb_do_redirect(skb); 3902 *ret = NET_XMIT_SUCCESS; 3903 return NULL; 3904 default: 3905 break; 3906 } 3907 #endif /* CONFIG_NET_CLS_ACT */ 3908 3909 return skb; 3910 } 3911 #endif /* CONFIG_NET_EGRESS */ 3912 3913 #ifdef CONFIG_XPS 3914 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3915 struct xps_dev_maps *dev_maps, unsigned int tci) 3916 { 3917 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3918 struct xps_map *map; 3919 int queue_index = -1; 3920 3921 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3922 return queue_index; 3923 3924 tci *= dev_maps->num_tc; 3925 tci += tc; 3926 3927 map = rcu_dereference(dev_maps->attr_map[tci]); 3928 if (map) { 3929 if (map->len == 1) 3930 queue_index = map->queues[0]; 3931 else 3932 queue_index = map->queues[reciprocal_scale( 3933 skb_get_hash(skb), map->len)]; 3934 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3935 queue_index = -1; 3936 } 3937 return queue_index; 3938 } 3939 #endif 3940 3941 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3942 struct sk_buff *skb) 3943 { 3944 #ifdef CONFIG_XPS 3945 struct xps_dev_maps *dev_maps; 3946 struct sock *sk = skb->sk; 3947 int queue_index = -1; 3948 3949 if (!static_key_false(&xps_needed)) 3950 return -1; 3951 3952 rcu_read_lock(); 3953 if (!static_key_false(&xps_rxqs_needed)) 3954 goto get_cpus_map; 3955 3956 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3957 if (dev_maps) { 3958 int tci = sk_rx_queue_get(sk); 3959 3960 if (tci >= 0) 3961 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3962 tci); 3963 } 3964 3965 get_cpus_map: 3966 if (queue_index < 0) { 3967 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 3968 if (dev_maps) { 3969 unsigned int tci = skb->sender_cpu - 1; 3970 3971 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3972 tci); 3973 } 3974 } 3975 rcu_read_unlock(); 3976 3977 return queue_index; 3978 #else 3979 return -1; 3980 #endif 3981 } 3982 3983 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3984 struct net_device *sb_dev) 3985 { 3986 return 0; 3987 } 3988 EXPORT_SYMBOL(dev_pick_tx_zero); 3989 3990 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 3991 struct net_device *sb_dev) 3992 { 3993 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 3994 } 3995 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 3996 3997 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 3998 struct net_device *sb_dev) 3999 { 4000 struct sock *sk = skb->sk; 4001 int queue_index = sk_tx_queue_get(sk); 4002 4003 sb_dev = sb_dev ? : dev; 4004 4005 if (queue_index < 0 || skb->ooo_okay || 4006 queue_index >= dev->real_num_tx_queues) { 4007 int new_index = get_xps_queue(dev, sb_dev, skb); 4008 4009 if (new_index < 0) 4010 new_index = skb_tx_hash(dev, sb_dev, skb); 4011 4012 if (queue_index != new_index && sk && 4013 sk_fullsock(sk) && 4014 rcu_access_pointer(sk->sk_dst_cache)) 4015 sk_tx_queue_set(sk, new_index); 4016 4017 queue_index = new_index; 4018 } 4019 4020 return queue_index; 4021 } 4022 EXPORT_SYMBOL(netdev_pick_tx); 4023 4024 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4025 struct sk_buff *skb, 4026 struct net_device *sb_dev) 4027 { 4028 int queue_index = 0; 4029 4030 #ifdef CONFIG_XPS 4031 u32 sender_cpu = skb->sender_cpu - 1; 4032 4033 if (sender_cpu >= (u32)NR_CPUS) 4034 skb->sender_cpu = raw_smp_processor_id() + 1; 4035 #endif 4036 4037 if (dev->real_num_tx_queues != 1) { 4038 const struct net_device_ops *ops = dev->netdev_ops; 4039 4040 if (ops->ndo_select_queue) 4041 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4042 else 4043 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4044 4045 queue_index = netdev_cap_txqueue(dev, queue_index); 4046 } 4047 4048 skb_set_queue_mapping(skb, queue_index); 4049 return netdev_get_tx_queue(dev, queue_index); 4050 } 4051 4052 /** 4053 * __dev_queue_xmit - transmit a buffer 4054 * @skb: buffer to transmit 4055 * @sb_dev: suboordinate device used for L2 forwarding offload 4056 * 4057 * Queue a buffer for transmission to a network device. The caller must 4058 * have set the device and priority and built the buffer before calling 4059 * this function. The function can be called from an interrupt. 4060 * 4061 * A negative errno code is returned on a failure. A success does not 4062 * guarantee the frame will be transmitted as it may be dropped due 4063 * to congestion or traffic shaping. 4064 * 4065 * ----------------------------------------------------------------------------------- 4066 * I notice this method can also return errors from the queue disciplines, 4067 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4068 * be positive. 4069 * 4070 * Regardless of the return value, the skb is consumed, so it is currently 4071 * difficult to retry a send to this method. (You can bump the ref count 4072 * before sending to hold a reference for retry if you are careful.) 4073 * 4074 * When calling this method, interrupts MUST be enabled. This is because 4075 * the BH enable code must have IRQs enabled so that it will not deadlock. 4076 * --BLG 4077 */ 4078 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4079 { 4080 struct net_device *dev = skb->dev; 4081 struct netdev_queue *txq; 4082 struct Qdisc *q; 4083 int rc = -ENOMEM; 4084 bool again = false; 4085 4086 skb_reset_mac_header(skb); 4087 4088 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4089 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4090 4091 /* Disable soft irqs for various locks below. Also 4092 * stops preemption for RCU. 4093 */ 4094 rcu_read_lock_bh(); 4095 4096 skb_update_prio(skb); 4097 4098 qdisc_pkt_len_init(skb); 4099 #ifdef CONFIG_NET_CLS_ACT 4100 skb->tc_at_ingress = 0; 4101 #endif 4102 #ifdef CONFIG_NET_EGRESS 4103 if (static_branch_unlikely(&egress_needed_key)) { 4104 if (nf_hook_egress_active()) { 4105 skb = nf_hook_egress(skb, &rc, dev); 4106 if (!skb) 4107 goto out; 4108 } 4109 nf_skip_egress(skb, true); 4110 skb = sch_handle_egress(skb, &rc, dev); 4111 if (!skb) 4112 goto out; 4113 nf_skip_egress(skb, false); 4114 } 4115 #endif 4116 /* If device/qdisc don't need skb->dst, release it right now while 4117 * its hot in this cpu cache. 4118 */ 4119 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4120 skb_dst_drop(skb); 4121 else 4122 skb_dst_force(skb); 4123 4124 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4125 q = rcu_dereference_bh(txq->qdisc); 4126 4127 trace_net_dev_queue(skb); 4128 if (q->enqueue) { 4129 rc = __dev_xmit_skb(skb, q, dev, txq); 4130 goto out; 4131 } 4132 4133 /* The device has no queue. Common case for software devices: 4134 * loopback, all the sorts of tunnels... 4135 4136 * Really, it is unlikely that netif_tx_lock protection is necessary 4137 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4138 * counters.) 4139 * However, it is possible, that they rely on protection 4140 * made by us here. 4141 4142 * Check this and shot the lock. It is not prone from deadlocks. 4143 *Either shot noqueue qdisc, it is even simpler 8) 4144 */ 4145 if (dev->flags & IFF_UP) { 4146 int cpu = smp_processor_id(); /* ok because BHs are off */ 4147 4148 /* Other cpus might concurrently change txq->xmit_lock_owner 4149 * to -1 or to their cpu id, but not to our id. 4150 */ 4151 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4152 if (dev_xmit_recursion()) 4153 goto recursion_alert; 4154 4155 skb = validate_xmit_skb(skb, dev, &again); 4156 if (!skb) 4157 goto out; 4158 4159 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4160 HARD_TX_LOCK(dev, txq, cpu); 4161 4162 if (!netif_xmit_stopped(txq)) { 4163 dev_xmit_recursion_inc(); 4164 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4165 dev_xmit_recursion_dec(); 4166 if (dev_xmit_complete(rc)) { 4167 HARD_TX_UNLOCK(dev, txq); 4168 goto out; 4169 } 4170 } 4171 HARD_TX_UNLOCK(dev, txq); 4172 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4173 dev->name); 4174 } else { 4175 /* Recursion is detected! It is possible, 4176 * unfortunately 4177 */ 4178 recursion_alert: 4179 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4180 dev->name); 4181 } 4182 } 4183 4184 rc = -ENETDOWN; 4185 rcu_read_unlock_bh(); 4186 4187 dev_core_stats_tx_dropped_inc(dev); 4188 kfree_skb_list(skb); 4189 return rc; 4190 out: 4191 rcu_read_unlock_bh(); 4192 return rc; 4193 } 4194 4195 int dev_queue_xmit(struct sk_buff *skb) 4196 { 4197 return __dev_queue_xmit(skb, NULL); 4198 } 4199 EXPORT_SYMBOL(dev_queue_xmit); 4200 4201 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4202 { 4203 return __dev_queue_xmit(skb, sb_dev); 4204 } 4205 EXPORT_SYMBOL(dev_queue_xmit_accel); 4206 4207 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4208 { 4209 struct net_device *dev = skb->dev; 4210 struct sk_buff *orig_skb = skb; 4211 struct netdev_queue *txq; 4212 int ret = NETDEV_TX_BUSY; 4213 bool again = false; 4214 4215 if (unlikely(!netif_running(dev) || 4216 !netif_carrier_ok(dev))) 4217 goto drop; 4218 4219 skb = validate_xmit_skb_list(skb, dev, &again); 4220 if (skb != orig_skb) 4221 goto drop; 4222 4223 skb_set_queue_mapping(skb, queue_id); 4224 txq = skb_get_tx_queue(dev, skb); 4225 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4226 4227 local_bh_disable(); 4228 4229 dev_xmit_recursion_inc(); 4230 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4231 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4232 ret = netdev_start_xmit(skb, dev, txq, false); 4233 HARD_TX_UNLOCK(dev, txq); 4234 dev_xmit_recursion_dec(); 4235 4236 local_bh_enable(); 4237 return ret; 4238 drop: 4239 dev_core_stats_tx_dropped_inc(dev); 4240 kfree_skb_list(skb); 4241 return NET_XMIT_DROP; 4242 } 4243 EXPORT_SYMBOL(__dev_direct_xmit); 4244 4245 /************************************************************************* 4246 * Receiver routines 4247 *************************************************************************/ 4248 4249 int netdev_max_backlog __read_mostly = 1000; 4250 EXPORT_SYMBOL(netdev_max_backlog); 4251 4252 int netdev_tstamp_prequeue __read_mostly = 1; 4253 int netdev_budget __read_mostly = 300; 4254 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4255 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4256 int weight_p __read_mostly = 64; /* old backlog weight */ 4257 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4258 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4259 int dev_rx_weight __read_mostly = 64; 4260 int dev_tx_weight __read_mostly = 64; 4261 4262 /* Called with irq disabled */ 4263 static inline void ____napi_schedule(struct softnet_data *sd, 4264 struct napi_struct *napi) 4265 { 4266 struct task_struct *thread; 4267 4268 lockdep_assert_softirq_will_run(); 4269 lockdep_assert_irqs_disabled(); 4270 4271 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4272 /* Paired with smp_mb__before_atomic() in 4273 * napi_enable()/dev_set_threaded(). 4274 * Use READ_ONCE() to guarantee a complete 4275 * read on napi->thread. Only call 4276 * wake_up_process() when it's not NULL. 4277 */ 4278 thread = READ_ONCE(napi->thread); 4279 if (thread) { 4280 /* Avoid doing set_bit() if the thread is in 4281 * INTERRUPTIBLE state, cause napi_thread_wait() 4282 * makes sure to proceed with napi polling 4283 * if the thread is explicitly woken from here. 4284 */ 4285 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4286 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4287 wake_up_process(thread); 4288 return; 4289 } 4290 } 4291 4292 list_add_tail(&napi->poll_list, &sd->poll_list); 4293 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4294 } 4295 4296 #ifdef CONFIG_RPS 4297 4298 /* One global table that all flow-based protocols share. */ 4299 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4300 EXPORT_SYMBOL(rps_sock_flow_table); 4301 u32 rps_cpu_mask __read_mostly; 4302 EXPORT_SYMBOL(rps_cpu_mask); 4303 4304 struct static_key_false rps_needed __read_mostly; 4305 EXPORT_SYMBOL(rps_needed); 4306 struct static_key_false rfs_needed __read_mostly; 4307 EXPORT_SYMBOL(rfs_needed); 4308 4309 static struct rps_dev_flow * 4310 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4311 struct rps_dev_flow *rflow, u16 next_cpu) 4312 { 4313 if (next_cpu < nr_cpu_ids) { 4314 #ifdef CONFIG_RFS_ACCEL 4315 struct netdev_rx_queue *rxqueue; 4316 struct rps_dev_flow_table *flow_table; 4317 struct rps_dev_flow *old_rflow; 4318 u32 flow_id; 4319 u16 rxq_index; 4320 int rc; 4321 4322 /* Should we steer this flow to a different hardware queue? */ 4323 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4324 !(dev->features & NETIF_F_NTUPLE)) 4325 goto out; 4326 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4327 if (rxq_index == skb_get_rx_queue(skb)) 4328 goto out; 4329 4330 rxqueue = dev->_rx + rxq_index; 4331 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4332 if (!flow_table) 4333 goto out; 4334 flow_id = skb_get_hash(skb) & flow_table->mask; 4335 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4336 rxq_index, flow_id); 4337 if (rc < 0) 4338 goto out; 4339 old_rflow = rflow; 4340 rflow = &flow_table->flows[flow_id]; 4341 rflow->filter = rc; 4342 if (old_rflow->filter == rflow->filter) 4343 old_rflow->filter = RPS_NO_FILTER; 4344 out: 4345 #endif 4346 rflow->last_qtail = 4347 per_cpu(softnet_data, next_cpu).input_queue_head; 4348 } 4349 4350 rflow->cpu = next_cpu; 4351 return rflow; 4352 } 4353 4354 /* 4355 * get_rps_cpu is called from netif_receive_skb and returns the target 4356 * CPU from the RPS map of the receiving queue for a given skb. 4357 * rcu_read_lock must be held on entry. 4358 */ 4359 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4360 struct rps_dev_flow **rflowp) 4361 { 4362 const struct rps_sock_flow_table *sock_flow_table; 4363 struct netdev_rx_queue *rxqueue = dev->_rx; 4364 struct rps_dev_flow_table *flow_table; 4365 struct rps_map *map; 4366 int cpu = -1; 4367 u32 tcpu; 4368 u32 hash; 4369 4370 if (skb_rx_queue_recorded(skb)) { 4371 u16 index = skb_get_rx_queue(skb); 4372 4373 if (unlikely(index >= dev->real_num_rx_queues)) { 4374 WARN_ONCE(dev->real_num_rx_queues > 1, 4375 "%s received packet on queue %u, but number " 4376 "of RX queues is %u\n", 4377 dev->name, index, dev->real_num_rx_queues); 4378 goto done; 4379 } 4380 rxqueue += index; 4381 } 4382 4383 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4384 4385 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4386 map = rcu_dereference(rxqueue->rps_map); 4387 if (!flow_table && !map) 4388 goto done; 4389 4390 skb_reset_network_header(skb); 4391 hash = skb_get_hash(skb); 4392 if (!hash) 4393 goto done; 4394 4395 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4396 if (flow_table && sock_flow_table) { 4397 struct rps_dev_flow *rflow; 4398 u32 next_cpu; 4399 u32 ident; 4400 4401 /* First check into global flow table if there is a match */ 4402 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4403 if ((ident ^ hash) & ~rps_cpu_mask) 4404 goto try_rps; 4405 4406 next_cpu = ident & rps_cpu_mask; 4407 4408 /* OK, now we know there is a match, 4409 * we can look at the local (per receive queue) flow table 4410 */ 4411 rflow = &flow_table->flows[hash & flow_table->mask]; 4412 tcpu = rflow->cpu; 4413 4414 /* 4415 * If the desired CPU (where last recvmsg was done) is 4416 * different from current CPU (one in the rx-queue flow 4417 * table entry), switch if one of the following holds: 4418 * - Current CPU is unset (>= nr_cpu_ids). 4419 * - Current CPU is offline. 4420 * - The current CPU's queue tail has advanced beyond the 4421 * last packet that was enqueued using this table entry. 4422 * This guarantees that all previous packets for the flow 4423 * have been dequeued, thus preserving in order delivery. 4424 */ 4425 if (unlikely(tcpu != next_cpu) && 4426 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4427 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4428 rflow->last_qtail)) >= 0)) { 4429 tcpu = next_cpu; 4430 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4431 } 4432 4433 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4434 *rflowp = rflow; 4435 cpu = tcpu; 4436 goto done; 4437 } 4438 } 4439 4440 try_rps: 4441 4442 if (map) { 4443 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4444 if (cpu_online(tcpu)) { 4445 cpu = tcpu; 4446 goto done; 4447 } 4448 } 4449 4450 done: 4451 return cpu; 4452 } 4453 4454 #ifdef CONFIG_RFS_ACCEL 4455 4456 /** 4457 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4458 * @dev: Device on which the filter was set 4459 * @rxq_index: RX queue index 4460 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4461 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4462 * 4463 * Drivers that implement ndo_rx_flow_steer() should periodically call 4464 * this function for each installed filter and remove the filters for 4465 * which it returns %true. 4466 */ 4467 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4468 u32 flow_id, u16 filter_id) 4469 { 4470 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4471 struct rps_dev_flow_table *flow_table; 4472 struct rps_dev_flow *rflow; 4473 bool expire = true; 4474 unsigned int cpu; 4475 4476 rcu_read_lock(); 4477 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4478 if (flow_table && flow_id <= flow_table->mask) { 4479 rflow = &flow_table->flows[flow_id]; 4480 cpu = READ_ONCE(rflow->cpu); 4481 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4482 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4483 rflow->last_qtail) < 4484 (int)(10 * flow_table->mask))) 4485 expire = false; 4486 } 4487 rcu_read_unlock(); 4488 return expire; 4489 } 4490 EXPORT_SYMBOL(rps_may_expire_flow); 4491 4492 #endif /* CONFIG_RFS_ACCEL */ 4493 4494 /* Called from hardirq (IPI) context */ 4495 static void rps_trigger_softirq(void *data) 4496 { 4497 struct softnet_data *sd = data; 4498 4499 ____napi_schedule(sd, &sd->backlog); 4500 sd->received_rps++; 4501 } 4502 4503 #endif /* CONFIG_RPS */ 4504 4505 /* 4506 * Check if this softnet_data structure is another cpu one 4507 * If yes, queue it to our IPI list and return 1 4508 * If no, return 0 4509 */ 4510 static int napi_schedule_rps(struct softnet_data *sd) 4511 { 4512 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4513 4514 #ifdef CONFIG_RPS 4515 if (sd != mysd) { 4516 sd->rps_ipi_next = mysd->rps_ipi_list; 4517 mysd->rps_ipi_list = sd; 4518 4519 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4520 return 1; 4521 } 4522 #endif /* CONFIG_RPS */ 4523 __napi_schedule_irqoff(&mysd->backlog); 4524 return 0; 4525 } 4526 4527 #ifdef CONFIG_NET_FLOW_LIMIT 4528 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4529 #endif 4530 4531 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4532 { 4533 #ifdef CONFIG_NET_FLOW_LIMIT 4534 struct sd_flow_limit *fl; 4535 struct softnet_data *sd; 4536 unsigned int old_flow, new_flow; 4537 4538 if (qlen < (netdev_max_backlog >> 1)) 4539 return false; 4540 4541 sd = this_cpu_ptr(&softnet_data); 4542 4543 rcu_read_lock(); 4544 fl = rcu_dereference(sd->flow_limit); 4545 if (fl) { 4546 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4547 old_flow = fl->history[fl->history_head]; 4548 fl->history[fl->history_head] = new_flow; 4549 4550 fl->history_head++; 4551 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4552 4553 if (likely(fl->buckets[old_flow])) 4554 fl->buckets[old_flow]--; 4555 4556 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4557 fl->count++; 4558 rcu_read_unlock(); 4559 return true; 4560 } 4561 } 4562 rcu_read_unlock(); 4563 #endif 4564 return false; 4565 } 4566 4567 /* 4568 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4569 * queue (may be a remote CPU queue). 4570 */ 4571 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4572 unsigned int *qtail) 4573 { 4574 enum skb_drop_reason reason; 4575 struct softnet_data *sd; 4576 unsigned long flags; 4577 unsigned int qlen; 4578 4579 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4580 sd = &per_cpu(softnet_data, cpu); 4581 4582 rps_lock_irqsave(sd, &flags); 4583 if (!netif_running(skb->dev)) 4584 goto drop; 4585 qlen = skb_queue_len(&sd->input_pkt_queue); 4586 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4587 if (qlen) { 4588 enqueue: 4589 __skb_queue_tail(&sd->input_pkt_queue, skb); 4590 input_queue_tail_incr_save(sd, qtail); 4591 rps_unlock_irq_restore(sd, &flags); 4592 return NET_RX_SUCCESS; 4593 } 4594 4595 /* Schedule NAPI for backlog device 4596 * We can use non atomic operation since we own the queue lock 4597 */ 4598 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4599 napi_schedule_rps(sd); 4600 goto enqueue; 4601 } 4602 reason = SKB_DROP_REASON_CPU_BACKLOG; 4603 4604 drop: 4605 sd->dropped++; 4606 rps_unlock_irq_restore(sd, &flags); 4607 4608 dev_core_stats_rx_dropped_inc(skb->dev); 4609 kfree_skb_reason(skb, reason); 4610 return NET_RX_DROP; 4611 } 4612 4613 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4614 { 4615 struct net_device *dev = skb->dev; 4616 struct netdev_rx_queue *rxqueue; 4617 4618 rxqueue = dev->_rx; 4619 4620 if (skb_rx_queue_recorded(skb)) { 4621 u16 index = skb_get_rx_queue(skb); 4622 4623 if (unlikely(index >= dev->real_num_rx_queues)) { 4624 WARN_ONCE(dev->real_num_rx_queues > 1, 4625 "%s received packet on queue %u, but number " 4626 "of RX queues is %u\n", 4627 dev->name, index, dev->real_num_rx_queues); 4628 4629 return rxqueue; /* Return first rxqueue */ 4630 } 4631 rxqueue += index; 4632 } 4633 return rxqueue; 4634 } 4635 4636 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4637 struct bpf_prog *xdp_prog) 4638 { 4639 void *orig_data, *orig_data_end, *hard_start; 4640 struct netdev_rx_queue *rxqueue; 4641 bool orig_bcast, orig_host; 4642 u32 mac_len, frame_sz; 4643 __be16 orig_eth_type; 4644 struct ethhdr *eth; 4645 u32 metalen, act; 4646 int off; 4647 4648 /* The XDP program wants to see the packet starting at the MAC 4649 * header. 4650 */ 4651 mac_len = skb->data - skb_mac_header(skb); 4652 hard_start = skb->data - skb_headroom(skb); 4653 4654 /* SKB "head" area always have tailroom for skb_shared_info */ 4655 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4656 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4657 4658 rxqueue = netif_get_rxqueue(skb); 4659 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4660 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4661 skb_headlen(skb) + mac_len, true); 4662 4663 orig_data_end = xdp->data_end; 4664 orig_data = xdp->data; 4665 eth = (struct ethhdr *)xdp->data; 4666 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4667 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4668 orig_eth_type = eth->h_proto; 4669 4670 act = bpf_prog_run_xdp(xdp_prog, xdp); 4671 4672 /* check if bpf_xdp_adjust_head was used */ 4673 off = xdp->data - orig_data; 4674 if (off) { 4675 if (off > 0) 4676 __skb_pull(skb, off); 4677 else if (off < 0) 4678 __skb_push(skb, -off); 4679 4680 skb->mac_header += off; 4681 skb_reset_network_header(skb); 4682 } 4683 4684 /* check if bpf_xdp_adjust_tail was used */ 4685 off = xdp->data_end - orig_data_end; 4686 if (off != 0) { 4687 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4688 skb->len += off; /* positive on grow, negative on shrink */ 4689 } 4690 4691 /* check if XDP changed eth hdr such SKB needs update */ 4692 eth = (struct ethhdr *)xdp->data; 4693 if ((orig_eth_type != eth->h_proto) || 4694 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4695 skb->dev->dev_addr)) || 4696 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4697 __skb_push(skb, ETH_HLEN); 4698 skb->pkt_type = PACKET_HOST; 4699 skb->protocol = eth_type_trans(skb, skb->dev); 4700 } 4701 4702 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4703 * before calling us again on redirect path. We do not call do_redirect 4704 * as we leave that up to the caller. 4705 * 4706 * Caller is responsible for managing lifetime of skb (i.e. calling 4707 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4708 */ 4709 switch (act) { 4710 case XDP_REDIRECT: 4711 case XDP_TX: 4712 __skb_push(skb, mac_len); 4713 break; 4714 case XDP_PASS: 4715 metalen = xdp->data - xdp->data_meta; 4716 if (metalen) 4717 skb_metadata_set(skb, metalen); 4718 break; 4719 } 4720 4721 return act; 4722 } 4723 4724 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4725 struct xdp_buff *xdp, 4726 struct bpf_prog *xdp_prog) 4727 { 4728 u32 act = XDP_DROP; 4729 4730 /* Reinjected packets coming from act_mirred or similar should 4731 * not get XDP generic processing. 4732 */ 4733 if (skb_is_redirected(skb)) 4734 return XDP_PASS; 4735 4736 /* XDP packets must be linear and must have sufficient headroom 4737 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4738 * native XDP provides, thus we need to do it here as well. 4739 */ 4740 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4741 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4742 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4743 int troom = skb->tail + skb->data_len - skb->end; 4744 4745 /* In case we have to go down the path and also linearize, 4746 * then lets do the pskb_expand_head() work just once here. 4747 */ 4748 if (pskb_expand_head(skb, 4749 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4750 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4751 goto do_drop; 4752 if (skb_linearize(skb)) 4753 goto do_drop; 4754 } 4755 4756 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4757 switch (act) { 4758 case XDP_REDIRECT: 4759 case XDP_TX: 4760 case XDP_PASS: 4761 break; 4762 default: 4763 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4764 fallthrough; 4765 case XDP_ABORTED: 4766 trace_xdp_exception(skb->dev, xdp_prog, act); 4767 fallthrough; 4768 case XDP_DROP: 4769 do_drop: 4770 kfree_skb(skb); 4771 break; 4772 } 4773 4774 return act; 4775 } 4776 4777 /* When doing generic XDP we have to bypass the qdisc layer and the 4778 * network taps in order to match in-driver-XDP behavior. 4779 */ 4780 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4781 { 4782 struct net_device *dev = skb->dev; 4783 struct netdev_queue *txq; 4784 bool free_skb = true; 4785 int cpu, rc; 4786 4787 txq = netdev_core_pick_tx(dev, skb, NULL); 4788 cpu = smp_processor_id(); 4789 HARD_TX_LOCK(dev, txq, cpu); 4790 if (!netif_xmit_stopped(txq)) { 4791 rc = netdev_start_xmit(skb, dev, txq, 0); 4792 if (dev_xmit_complete(rc)) 4793 free_skb = false; 4794 } 4795 HARD_TX_UNLOCK(dev, txq); 4796 if (free_skb) { 4797 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4798 kfree_skb(skb); 4799 } 4800 } 4801 4802 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4803 4804 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4805 { 4806 if (xdp_prog) { 4807 struct xdp_buff xdp; 4808 u32 act; 4809 int err; 4810 4811 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4812 if (act != XDP_PASS) { 4813 switch (act) { 4814 case XDP_REDIRECT: 4815 err = xdp_do_generic_redirect(skb->dev, skb, 4816 &xdp, xdp_prog); 4817 if (err) 4818 goto out_redir; 4819 break; 4820 case XDP_TX: 4821 generic_xdp_tx(skb, xdp_prog); 4822 break; 4823 } 4824 return XDP_DROP; 4825 } 4826 } 4827 return XDP_PASS; 4828 out_redir: 4829 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4830 return XDP_DROP; 4831 } 4832 EXPORT_SYMBOL_GPL(do_xdp_generic); 4833 4834 static int netif_rx_internal(struct sk_buff *skb) 4835 { 4836 int ret; 4837 4838 net_timestamp_check(netdev_tstamp_prequeue, skb); 4839 4840 trace_netif_rx(skb); 4841 4842 #ifdef CONFIG_RPS 4843 if (static_branch_unlikely(&rps_needed)) { 4844 struct rps_dev_flow voidflow, *rflow = &voidflow; 4845 int cpu; 4846 4847 rcu_read_lock(); 4848 4849 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4850 if (cpu < 0) 4851 cpu = smp_processor_id(); 4852 4853 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4854 4855 rcu_read_unlock(); 4856 } else 4857 #endif 4858 { 4859 unsigned int qtail; 4860 4861 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4862 } 4863 return ret; 4864 } 4865 4866 /** 4867 * __netif_rx - Slightly optimized version of netif_rx 4868 * @skb: buffer to post 4869 * 4870 * This behaves as netif_rx except that it does not disable bottom halves. 4871 * As a result this function may only be invoked from the interrupt context 4872 * (either hard or soft interrupt). 4873 */ 4874 int __netif_rx(struct sk_buff *skb) 4875 { 4876 int ret; 4877 4878 lockdep_assert_softirq_will_run(); 4879 4880 trace_netif_rx_entry(skb); 4881 ret = netif_rx_internal(skb); 4882 trace_netif_rx_exit(ret); 4883 return ret; 4884 } 4885 EXPORT_SYMBOL(__netif_rx); 4886 4887 /** 4888 * netif_rx - post buffer to the network code 4889 * @skb: buffer to post 4890 * 4891 * This function receives a packet from a device driver and queues it for 4892 * the upper (protocol) levels to process via the backlog NAPI device. It 4893 * always succeeds. The buffer may be dropped during processing for 4894 * congestion control or by the protocol layers. 4895 * The network buffer is passed via the backlog NAPI device. Modern NIC 4896 * driver should use NAPI and GRO. 4897 * This function can used from interrupt and from process context. The 4898 * caller from process context must not disable interrupts before invoking 4899 * this function. 4900 * 4901 * return values: 4902 * NET_RX_SUCCESS (no congestion) 4903 * NET_RX_DROP (packet was dropped) 4904 * 4905 */ 4906 int netif_rx(struct sk_buff *skb) 4907 { 4908 bool need_bh_off = !(hardirq_count() | softirq_count()); 4909 int ret; 4910 4911 if (need_bh_off) 4912 local_bh_disable(); 4913 trace_netif_rx_entry(skb); 4914 ret = netif_rx_internal(skb); 4915 trace_netif_rx_exit(ret); 4916 if (need_bh_off) 4917 local_bh_enable(); 4918 return ret; 4919 } 4920 EXPORT_SYMBOL(netif_rx); 4921 4922 static __latent_entropy void net_tx_action(struct softirq_action *h) 4923 { 4924 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4925 4926 if (sd->completion_queue) { 4927 struct sk_buff *clist; 4928 4929 local_irq_disable(); 4930 clist = sd->completion_queue; 4931 sd->completion_queue = NULL; 4932 local_irq_enable(); 4933 4934 while (clist) { 4935 struct sk_buff *skb = clist; 4936 4937 clist = clist->next; 4938 4939 WARN_ON(refcount_read(&skb->users)); 4940 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4941 trace_consume_skb(skb); 4942 else 4943 trace_kfree_skb(skb, net_tx_action, 4944 SKB_DROP_REASON_NOT_SPECIFIED); 4945 4946 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4947 __kfree_skb(skb); 4948 else 4949 __kfree_skb_defer(skb); 4950 } 4951 } 4952 4953 if (sd->output_queue) { 4954 struct Qdisc *head; 4955 4956 local_irq_disable(); 4957 head = sd->output_queue; 4958 sd->output_queue = NULL; 4959 sd->output_queue_tailp = &sd->output_queue; 4960 local_irq_enable(); 4961 4962 rcu_read_lock(); 4963 4964 while (head) { 4965 struct Qdisc *q = head; 4966 spinlock_t *root_lock = NULL; 4967 4968 head = head->next_sched; 4969 4970 /* We need to make sure head->next_sched is read 4971 * before clearing __QDISC_STATE_SCHED 4972 */ 4973 smp_mb__before_atomic(); 4974 4975 if (!(q->flags & TCQ_F_NOLOCK)) { 4976 root_lock = qdisc_lock(q); 4977 spin_lock(root_lock); 4978 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 4979 &q->state))) { 4980 /* There is a synchronize_net() between 4981 * STATE_DEACTIVATED flag being set and 4982 * qdisc_reset()/some_qdisc_is_busy() in 4983 * dev_deactivate(), so we can safely bail out 4984 * early here to avoid data race between 4985 * qdisc_deactivate() and some_qdisc_is_busy() 4986 * for lockless qdisc. 4987 */ 4988 clear_bit(__QDISC_STATE_SCHED, &q->state); 4989 continue; 4990 } 4991 4992 clear_bit(__QDISC_STATE_SCHED, &q->state); 4993 qdisc_run(q); 4994 if (root_lock) 4995 spin_unlock(root_lock); 4996 } 4997 4998 rcu_read_unlock(); 4999 } 5000 5001 xfrm_dev_backlog(sd); 5002 } 5003 5004 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5005 /* This hook is defined here for ATM LANE */ 5006 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5007 unsigned char *addr) __read_mostly; 5008 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5009 #endif 5010 5011 static inline struct sk_buff * 5012 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5013 struct net_device *orig_dev, bool *another) 5014 { 5015 #ifdef CONFIG_NET_CLS_ACT 5016 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5017 struct tcf_result cl_res; 5018 5019 /* If there's at least one ingress present somewhere (so 5020 * we get here via enabled static key), remaining devices 5021 * that are not configured with an ingress qdisc will bail 5022 * out here. 5023 */ 5024 if (!miniq) 5025 return skb; 5026 5027 if (*pt_prev) { 5028 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5029 *pt_prev = NULL; 5030 } 5031 5032 qdisc_skb_cb(skb)->pkt_len = skb->len; 5033 tc_skb_cb(skb)->mru = 0; 5034 tc_skb_cb(skb)->post_ct = false; 5035 skb->tc_at_ingress = 1; 5036 mini_qdisc_bstats_cpu_update(miniq, skb); 5037 5038 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5039 case TC_ACT_OK: 5040 case TC_ACT_RECLASSIFY: 5041 skb->tc_index = TC_H_MIN(cl_res.classid); 5042 break; 5043 case TC_ACT_SHOT: 5044 mini_qdisc_qstats_cpu_drop(miniq); 5045 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5046 return NULL; 5047 case TC_ACT_STOLEN: 5048 case TC_ACT_QUEUED: 5049 case TC_ACT_TRAP: 5050 consume_skb(skb); 5051 return NULL; 5052 case TC_ACT_REDIRECT: 5053 /* skb_mac_header check was done by cls/act_bpf, so 5054 * we can safely push the L2 header back before 5055 * redirecting to another netdev 5056 */ 5057 __skb_push(skb, skb->mac_len); 5058 if (skb_do_redirect(skb) == -EAGAIN) { 5059 __skb_pull(skb, skb->mac_len); 5060 *another = true; 5061 break; 5062 } 5063 return NULL; 5064 case TC_ACT_CONSUMED: 5065 return NULL; 5066 default: 5067 break; 5068 } 5069 #endif /* CONFIG_NET_CLS_ACT */ 5070 return skb; 5071 } 5072 5073 /** 5074 * netdev_is_rx_handler_busy - check if receive handler is registered 5075 * @dev: device to check 5076 * 5077 * Check if a receive handler is already registered for a given device. 5078 * Return true if there one. 5079 * 5080 * The caller must hold the rtnl_mutex. 5081 */ 5082 bool netdev_is_rx_handler_busy(struct net_device *dev) 5083 { 5084 ASSERT_RTNL(); 5085 return dev && rtnl_dereference(dev->rx_handler); 5086 } 5087 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5088 5089 /** 5090 * netdev_rx_handler_register - register receive handler 5091 * @dev: device to register a handler for 5092 * @rx_handler: receive handler to register 5093 * @rx_handler_data: data pointer that is used by rx handler 5094 * 5095 * Register a receive handler for a device. This handler will then be 5096 * called from __netif_receive_skb. A negative errno code is returned 5097 * on a failure. 5098 * 5099 * The caller must hold the rtnl_mutex. 5100 * 5101 * For a general description of rx_handler, see enum rx_handler_result. 5102 */ 5103 int netdev_rx_handler_register(struct net_device *dev, 5104 rx_handler_func_t *rx_handler, 5105 void *rx_handler_data) 5106 { 5107 if (netdev_is_rx_handler_busy(dev)) 5108 return -EBUSY; 5109 5110 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5111 return -EINVAL; 5112 5113 /* Note: rx_handler_data must be set before rx_handler */ 5114 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5115 rcu_assign_pointer(dev->rx_handler, rx_handler); 5116 5117 return 0; 5118 } 5119 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5120 5121 /** 5122 * netdev_rx_handler_unregister - unregister receive handler 5123 * @dev: device to unregister a handler from 5124 * 5125 * Unregister a receive handler from a device. 5126 * 5127 * The caller must hold the rtnl_mutex. 5128 */ 5129 void netdev_rx_handler_unregister(struct net_device *dev) 5130 { 5131 5132 ASSERT_RTNL(); 5133 RCU_INIT_POINTER(dev->rx_handler, NULL); 5134 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5135 * section has a guarantee to see a non NULL rx_handler_data 5136 * as well. 5137 */ 5138 synchronize_net(); 5139 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5140 } 5141 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5142 5143 /* 5144 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5145 * the special handling of PFMEMALLOC skbs. 5146 */ 5147 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5148 { 5149 switch (skb->protocol) { 5150 case htons(ETH_P_ARP): 5151 case htons(ETH_P_IP): 5152 case htons(ETH_P_IPV6): 5153 case htons(ETH_P_8021Q): 5154 case htons(ETH_P_8021AD): 5155 return true; 5156 default: 5157 return false; 5158 } 5159 } 5160 5161 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5162 int *ret, struct net_device *orig_dev) 5163 { 5164 if (nf_hook_ingress_active(skb)) { 5165 int ingress_retval; 5166 5167 if (*pt_prev) { 5168 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5169 *pt_prev = NULL; 5170 } 5171 5172 rcu_read_lock(); 5173 ingress_retval = nf_hook_ingress(skb); 5174 rcu_read_unlock(); 5175 return ingress_retval; 5176 } 5177 return 0; 5178 } 5179 5180 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5181 struct packet_type **ppt_prev) 5182 { 5183 struct packet_type *ptype, *pt_prev; 5184 rx_handler_func_t *rx_handler; 5185 struct sk_buff *skb = *pskb; 5186 struct net_device *orig_dev; 5187 bool deliver_exact = false; 5188 int ret = NET_RX_DROP; 5189 __be16 type; 5190 5191 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5192 5193 trace_netif_receive_skb(skb); 5194 5195 orig_dev = skb->dev; 5196 5197 skb_reset_network_header(skb); 5198 if (!skb_transport_header_was_set(skb)) 5199 skb_reset_transport_header(skb); 5200 skb_reset_mac_len(skb); 5201 5202 pt_prev = NULL; 5203 5204 another_round: 5205 skb->skb_iif = skb->dev->ifindex; 5206 5207 __this_cpu_inc(softnet_data.processed); 5208 5209 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5210 int ret2; 5211 5212 migrate_disable(); 5213 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5214 migrate_enable(); 5215 5216 if (ret2 != XDP_PASS) { 5217 ret = NET_RX_DROP; 5218 goto out; 5219 } 5220 } 5221 5222 if (eth_type_vlan(skb->protocol)) { 5223 skb = skb_vlan_untag(skb); 5224 if (unlikely(!skb)) 5225 goto out; 5226 } 5227 5228 if (skb_skip_tc_classify(skb)) 5229 goto skip_classify; 5230 5231 if (pfmemalloc) 5232 goto skip_taps; 5233 5234 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5235 if (pt_prev) 5236 ret = deliver_skb(skb, pt_prev, orig_dev); 5237 pt_prev = ptype; 5238 } 5239 5240 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5241 if (pt_prev) 5242 ret = deliver_skb(skb, pt_prev, orig_dev); 5243 pt_prev = ptype; 5244 } 5245 5246 skip_taps: 5247 #ifdef CONFIG_NET_INGRESS 5248 if (static_branch_unlikely(&ingress_needed_key)) { 5249 bool another = false; 5250 5251 nf_skip_egress(skb, true); 5252 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5253 &another); 5254 if (another) 5255 goto another_round; 5256 if (!skb) 5257 goto out; 5258 5259 nf_skip_egress(skb, false); 5260 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5261 goto out; 5262 } 5263 #endif 5264 skb_reset_redirect(skb); 5265 skip_classify: 5266 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5267 goto drop; 5268 5269 if (skb_vlan_tag_present(skb)) { 5270 if (pt_prev) { 5271 ret = deliver_skb(skb, pt_prev, orig_dev); 5272 pt_prev = NULL; 5273 } 5274 if (vlan_do_receive(&skb)) 5275 goto another_round; 5276 else if (unlikely(!skb)) 5277 goto out; 5278 } 5279 5280 rx_handler = rcu_dereference(skb->dev->rx_handler); 5281 if (rx_handler) { 5282 if (pt_prev) { 5283 ret = deliver_skb(skb, pt_prev, orig_dev); 5284 pt_prev = NULL; 5285 } 5286 switch (rx_handler(&skb)) { 5287 case RX_HANDLER_CONSUMED: 5288 ret = NET_RX_SUCCESS; 5289 goto out; 5290 case RX_HANDLER_ANOTHER: 5291 goto another_round; 5292 case RX_HANDLER_EXACT: 5293 deliver_exact = true; 5294 break; 5295 case RX_HANDLER_PASS: 5296 break; 5297 default: 5298 BUG(); 5299 } 5300 } 5301 5302 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5303 check_vlan_id: 5304 if (skb_vlan_tag_get_id(skb)) { 5305 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5306 * find vlan device. 5307 */ 5308 skb->pkt_type = PACKET_OTHERHOST; 5309 } else if (eth_type_vlan(skb->protocol)) { 5310 /* Outer header is 802.1P with vlan 0, inner header is 5311 * 802.1Q or 802.1AD and vlan_do_receive() above could 5312 * not find vlan dev for vlan id 0. 5313 */ 5314 __vlan_hwaccel_clear_tag(skb); 5315 skb = skb_vlan_untag(skb); 5316 if (unlikely(!skb)) 5317 goto out; 5318 if (vlan_do_receive(&skb)) 5319 /* After stripping off 802.1P header with vlan 0 5320 * vlan dev is found for inner header. 5321 */ 5322 goto another_round; 5323 else if (unlikely(!skb)) 5324 goto out; 5325 else 5326 /* We have stripped outer 802.1P vlan 0 header. 5327 * But could not find vlan dev. 5328 * check again for vlan id to set OTHERHOST. 5329 */ 5330 goto check_vlan_id; 5331 } 5332 /* Note: we might in the future use prio bits 5333 * and set skb->priority like in vlan_do_receive() 5334 * For the time being, just ignore Priority Code Point 5335 */ 5336 __vlan_hwaccel_clear_tag(skb); 5337 } 5338 5339 type = skb->protocol; 5340 5341 /* deliver only exact match when indicated */ 5342 if (likely(!deliver_exact)) { 5343 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5344 &ptype_base[ntohs(type) & 5345 PTYPE_HASH_MASK]); 5346 } 5347 5348 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5349 &orig_dev->ptype_specific); 5350 5351 if (unlikely(skb->dev != orig_dev)) { 5352 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5353 &skb->dev->ptype_specific); 5354 } 5355 5356 if (pt_prev) { 5357 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5358 goto drop; 5359 *ppt_prev = pt_prev; 5360 } else { 5361 drop: 5362 if (!deliver_exact) { 5363 dev_core_stats_rx_dropped_inc(skb->dev); 5364 kfree_skb_reason(skb, SKB_DROP_REASON_PTYPE_ABSENT); 5365 } else { 5366 dev_core_stats_rx_nohandler_inc(skb->dev); 5367 kfree_skb(skb); 5368 } 5369 /* Jamal, now you will not able to escape explaining 5370 * me how you were going to use this. :-) 5371 */ 5372 ret = NET_RX_DROP; 5373 } 5374 5375 out: 5376 /* The invariant here is that if *ppt_prev is not NULL 5377 * then skb should also be non-NULL. 5378 * 5379 * Apparently *ppt_prev assignment above holds this invariant due to 5380 * skb dereferencing near it. 5381 */ 5382 *pskb = skb; 5383 return ret; 5384 } 5385 5386 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5387 { 5388 struct net_device *orig_dev = skb->dev; 5389 struct packet_type *pt_prev = NULL; 5390 int ret; 5391 5392 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5393 if (pt_prev) 5394 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5395 skb->dev, pt_prev, orig_dev); 5396 return ret; 5397 } 5398 5399 /** 5400 * netif_receive_skb_core - special purpose version of netif_receive_skb 5401 * @skb: buffer to process 5402 * 5403 * More direct receive version of netif_receive_skb(). It should 5404 * only be used by callers that have a need to skip RPS and Generic XDP. 5405 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5406 * 5407 * This function may only be called from softirq context and interrupts 5408 * should be enabled. 5409 * 5410 * Return values (usually ignored): 5411 * NET_RX_SUCCESS: no congestion 5412 * NET_RX_DROP: packet was dropped 5413 */ 5414 int netif_receive_skb_core(struct sk_buff *skb) 5415 { 5416 int ret; 5417 5418 rcu_read_lock(); 5419 ret = __netif_receive_skb_one_core(skb, false); 5420 rcu_read_unlock(); 5421 5422 return ret; 5423 } 5424 EXPORT_SYMBOL(netif_receive_skb_core); 5425 5426 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5427 struct packet_type *pt_prev, 5428 struct net_device *orig_dev) 5429 { 5430 struct sk_buff *skb, *next; 5431 5432 if (!pt_prev) 5433 return; 5434 if (list_empty(head)) 5435 return; 5436 if (pt_prev->list_func != NULL) 5437 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5438 ip_list_rcv, head, pt_prev, orig_dev); 5439 else 5440 list_for_each_entry_safe(skb, next, head, list) { 5441 skb_list_del_init(skb); 5442 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5443 } 5444 } 5445 5446 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5447 { 5448 /* Fast-path assumptions: 5449 * - There is no RX handler. 5450 * - Only one packet_type matches. 5451 * If either of these fails, we will end up doing some per-packet 5452 * processing in-line, then handling the 'last ptype' for the whole 5453 * sublist. This can't cause out-of-order delivery to any single ptype, 5454 * because the 'last ptype' must be constant across the sublist, and all 5455 * other ptypes are handled per-packet. 5456 */ 5457 /* Current (common) ptype of sublist */ 5458 struct packet_type *pt_curr = NULL; 5459 /* Current (common) orig_dev of sublist */ 5460 struct net_device *od_curr = NULL; 5461 struct list_head sublist; 5462 struct sk_buff *skb, *next; 5463 5464 INIT_LIST_HEAD(&sublist); 5465 list_for_each_entry_safe(skb, next, head, list) { 5466 struct net_device *orig_dev = skb->dev; 5467 struct packet_type *pt_prev = NULL; 5468 5469 skb_list_del_init(skb); 5470 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5471 if (!pt_prev) 5472 continue; 5473 if (pt_curr != pt_prev || od_curr != orig_dev) { 5474 /* dispatch old sublist */ 5475 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5476 /* start new sublist */ 5477 INIT_LIST_HEAD(&sublist); 5478 pt_curr = pt_prev; 5479 od_curr = orig_dev; 5480 } 5481 list_add_tail(&skb->list, &sublist); 5482 } 5483 5484 /* dispatch final sublist */ 5485 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5486 } 5487 5488 static int __netif_receive_skb(struct sk_buff *skb) 5489 { 5490 int ret; 5491 5492 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5493 unsigned int noreclaim_flag; 5494 5495 /* 5496 * PFMEMALLOC skbs are special, they should 5497 * - be delivered to SOCK_MEMALLOC sockets only 5498 * - stay away from userspace 5499 * - have bounded memory usage 5500 * 5501 * Use PF_MEMALLOC as this saves us from propagating the allocation 5502 * context down to all allocation sites. 5503 */ 5504 noreclaim_flag = memalloc_noreclaim_save(); 5505 ret = __netif_receive_skb_one_core(skb, true); 5506 memalloc_noreclaim_restore(noreclaim_flag); 5507 } else 5508 ret = __netif_receive_skb_one_core(skb, false); 5509 5510 return ret; 5511 } 5512 5513 static void __netif_receive_skb_list(struct list_head *head) 5514 { 5515 unsigned long noreclaim_flag = 0; 5516 struct sk_buff *skb, *next; 5517 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5518 5519 list_for_each_entry_safe(skb, next, head, list) { 5520 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5521 struct list_head sublist; 5522 5523 /* Handle the previous sublist */ 5524 list_cut_before(&sublist, head, &skb->list); 5525 if (!list_empty(&sublist)) 5526 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5527 pfmemalloc = !pfmemalloc; 5528 /* See comments in __netif_receive_skb */ 5529 if (pfmemalloc) 5530 noreclaim_flag = memalloc_noreclaim_save(); 5531 else 5532 memalloc_noreclaim_restore(noreclaim_flag); 5533 } 5534 } 5535 /* Handle the remaining sublist */ 5536 if (!list_empty(head)) 5537 __netif_receive_skb_list_core(head, pfmemalloc); 5538 /* Restore pflags */ 5539 if (pfmemalloc) 5540 memalloc_noreclaim_restore(noreclaim_flag); 5541 } 5542 5543 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5544 { 5545 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5546 struct bpf_prog *new = xdp->prog; 5547 int ret = 0; 5548 5549 switch (xdp->command) { 5550 case XDP_SETUP_PROG: 5551 rcu_assign_pointer(dev->xdp_prog, new); 5552 if (old) 5553 bpf_prog_put(old); 5554 5555 if (old && !new) { 5556 static_branch_dec(&generic_xdp_needed_key); 5557 } else if (new && !old) { 5558 static_branch_inc(&generic_xdp_needed_key); 5559 dev_disable_lro(dev); 5560 dev_disable_gro_hw(dev); 5561 } 5562 break; 5563 5564 default: 5565 ret = -EINVAL; 5566 break; 5567 } 5568 5569 return ret; 5570 } 5571 5572 static int netif_receive_skb_internal(struct sk_buff *skb) 5573 { 5574 int ret; 5575 5576 net_timestamp_check(netdev_tstamp_prequeue, skb); 5577 5578 if (skb_defer_rx_timestamp(skb)) 5579 return NET_RX_SUCCESS; 5580 5581 rcu_read_lock(); 5582 #ifdef CONFIG_RPS 5583 if (static_branch_unlikely(&rps_needed)) { 5584 struct rps_dev_flow voidflow, *rflow = &voidflow; 5585 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5586 5587 if (cpu >= 0) { 5588 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5589 rcu_read_unlock(); 5590 return ret; 5591 } 5592 } 5593 #endif 5594 ret = __netif_receive_skb(skb); 5595 rcu_read_unlock(); 5596 return ret; 5597 } 5598 5599 void netif_receive_skb_list_internal(struct list_head *head) 5600 { 5601 struct sk_buff *skb, *next; 5602 struct list_head sublist; 5603 5604 INIT_LIST_HEAD(&sublist); 5605 list_for_each_entry_safe(skb, next, head, list) { 5606 net_timestamp_check(netdev_tstamp_prequeue, skb); 5607 skb_list_del_init(skb); 5608 if (!skb_defer_rx_timestamp(skb)) 5609 list_add_tail(&skb->list, &sublist); 5610 } 5611 list_splice_init(&sublist, head); 5612 5613 rcu_read_lock(); 5614 #ifdef CONFIG_RPS 5615 if (static_branch_unlikely(&rps_needed)) { 5616 list_for_each_entry_safe(skb, next, head, list) { 5617 struct rps_dev_flow voidflow, *rflow = &voidflow; 5618 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5619 5620 if (cpu >= 0) { 5621 /* Will be handled, remove from list */ 5622 skb_list_del_init(skb); 5623 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5624 } 5625 } 5626 } 5627 #endif 5628 __netif_receive_skb_list(head); 5629 rcu_read_unlock(); 5630 } 5631 5632 /** 5633 * netif_receive_skb - process receive buffer from network 5634 * @skb: buffer to process 5635 * 5636 * netif_receive_skb() is the main receive data processing function. 5637 * It always succeeds. The buffer may be dropped during processing 5638 * for congestion control or by the protocol layers. 5639 * 5640 * This function may only be called from softirq context and interrupts 5641 * should be enabled. 5642 * 5643 * Return values (usually ignored): 5644 * NET_RX_SUCCESS: no congestion 5645 * NET_RX_DROP: packet was dropped 5646 */ 5647 int netif_receive_skb(struct sk_buff *skb) 5648 { 5649 int ret; 5650 5651 trace_netif_receive_skb_entry(skb); 5652 5653 ret = netif_receive_skb_internal(skb); 5654 trace_netif_receive_skb_exit(ret); 5655 5656 return ret; 5657 } 5658 EXPORT_SYMBOL(netif_receive_skb); 5659 5660 /** 5661 * netif_receive_skb_list - process many receive buffers from network 5662 * @head: list of skbs to process. 5663 * 5664 * Since return value of netif_receive_skb() is normally ignored, and 5665 * wouldn't be meaningful for a list, this function returns void. 5666 * 5667 * This function may only be called from softirq context and interrupts 5668 * should be enabled. 5669 */ 5670 void netif_receive_skb_list(struct list_head *head) 5671 { 5672 struct sk_buff *skb; 5673 5674 if (list_empty(head)) 5675 return; 5676 if (trace_netif_receive_skb_list_entry_enabled()) { 5677 list_for_each_entry(skb, head, list) 5678 trace_netif_receive_skb_list_entry(skb); 5679 } 5680 netif_receive_skb_list_internal(head); 5681 trace_netif_receive_skb_list_exit(0); 5682 } 5683 EXPORT_SYMBOL(netif_receive_skb_list); 5684 5685 static DEFINE_PER_CPU(struct work_struct, flush_works); 5686 5687 /* Network device is going away, flush any packets still pending */ 5688 static void flush_backlog(struct work_struct *work) 5689 { 5690 struct sk_buff *skb, *tmp; 5691 struct softnet_data *sd; 5692 5693 local_bh_disable(); 5694 sd = this_cpu_ptr(&softnet_data); 5695 5696 rps_lock_irq_disable(sd); 5697 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5698 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5699 __skb_unlink(skb, &sd->input_pkt_queue); 5700 dev_kfree_skb_irq(skb); 5701 input_queue_head_incr(sd); 5702 } 5703 } 5704 rps_unlock_irq_enable(sd); 5705 5706 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5707 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5708 __skb_unlink(skb, &sd->process_queue); 5709 kfree_skb(skb); 5710 input_queue_head_incr(sd); 5711 } 5712 } 5713 local_bh_enable(); 5714 } 5715 5716 static bool flush_required(int cpu) 5717 { 5718 #if IS_ENABLED(CONFIG_RPS) 5719 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5720 bool do_flush; 5721 5722 rps_lock_irq_disable(sd); 5723 5724 /* as insertion into process_queue happens with the rps lock held, 5725 * process_queue access may race only with dequeue 5726 */ 5727 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5728 !skb_queue_empty_lockless(&sd->process_queue); 5729 rps_unlock_irq_enable(sd); 5730 5731 return do_flush; 5732 #endif 5733 /* without RPS we can't safely check input_pkt_queue: during a 5734 * concurrent remote skb_queue_splice() we can detect as empty both 5735 * input_pkt_queue and process_queue even if the latter could end-up 5736 * containing a lot of packets. 5737 */ 5738 return true; 5739 } 5740 5741 static void flush_all_backlogs(void) 5742 { 5743 static cpumask_t flush_cpus; 5744 unsigned int cpu; 5745 5746 /* since we are under rtnl lock protection we can use static data 5747 * for the cpumask and avoid allocating on stack the possibly 5748 * large mask 5749 */ 5750 ASSERT_RTNL(); 5751 5752 cpus_read_lock(); 5753 5754 cpumask_clear(&flush_cpus); 5755 for_each_online_cpu(cpu) { 5756 if (flush_required(cpu)) { 5757 queue_work_on(cpu, system_highpri_wq, 5758 per_cpu_ptr(&flush_works, cpu)); 5759 cpumask_set_cpu(cpu, &flush_cpus); 5760 } 5761 } 5762 5763 /* we can have in flight packet[s] on the cpus we are not flushing, 5764 * synchronize_net() in unregister_netdevice_many() will take care of 5765 * them 5766 */ 5767 for_each_cpu(cpu, &flush_cpus) 5768 flush_work(per_cpu_ptr(&flush_works, cpu)); 5769 5770 cpus_read_unlock(); 5771 } 5772 5773 static void net_rps_send_ipi(struct softnet_data *remsd) 5774 { 5775 #ifdef CONFIG_RPS 5776 while (remsd) { 5777 struct softnet_data *next = remsd->rps_ipi_next; 5778 5779 if (cpu_online(remsd->cpu)) 5780 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5781 remsd = next; 5782 } 5783 #endif 5784 } 5785 5786 /* 5787 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5788 * Note: called with local irq disabled, but exits with local irq enabled. 5789 */ 5790 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5791 { 5792 #ifdef CONFIG_RPS 5793 struct softnet_data *remsd = sd->rps_ipi_list; 5794 5795 if (remsd) { 5796 sd->rps_ipi_list = NULL; 5797 5798 local_irq_enable(); 5799 5800 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5801 net_rps_send_ipi(remsd); 5802 } else 5803 #endif 5804 local_irq_enable(); 5805 } 5806 5807 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5808 { 5809 #ifdef CONFIG_RPS 5810 return sd->rps_ipi_list != NULL; 5811 #else 5812 return false; 5813 #endif 5814 } 5815 5816 static int process_backlog(struct napi_struct *napi, int quota) 5817 { 5818 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5819 bool again = true; 5820 int work = 0; 5821 5822 /* Check if we have pending ipi, its better to send them now, 5823 * not waiting net_rx_action() end. 5824 */ 5825 if (sd_has_rps_ipi_waiting(sd)) { 5826 local_irq_disable(); 5827 net_rps_action_and_irq_enable(sd); 5828 } 5829 5830 napi->weight = dev_rx_weight; 5831 while (again) { 5832 struct sk_buff *skb; 5833 5834 while ((skb = __skb_dequeue(&sd->process_queue))) { 5835 rcu_read_lock(); 5836 __netif_receive_skb(skb); 5837 rcu_read_unlock(); 5838 input_queue_head_incr(sd); 5839 if (++work >= quota) 5840 return work; 5841 5842 } 5843 5844 rps_lock_irq_disable(sd); 5845 if (skb_queue_empty(&sd->input_pkt_queue)) { 5846 /* 5847 * Inline a custom version of __napi_complete(). 5848 * only current cpu owns and manipulates this napi, 5849 * and NAPI_STATE_SCHED is the only possible flag set 5850 * on backlog. 5851 * We can use a plain write instead of clear_bit(), 5852 * and we dont need an smp_mb() memory barrier. 5853 */ 5854 napi->state = 0; 5855 again = false; 5856 } else { 5857 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5858 &sd->process_queue); 5859 } 5860 rps_unlock_irq_enable(sd); 5861 } 5862 5863 return work; 5864 } 5865 5866 /** 5867 * __napi_schedule - schedule for receive 5868 * @n: entry to schedule 5869 * 5870 * The entry's receive function will be scheduled to run. 5871 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5872 */ 5873 void __napi_schedule(struct napi_struct *n) 5874 { 5875 unsigned long flags; 5876 5877 local_irq_save(flags); 5878 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5879 local_irq_restore(flags); 5880 } 5881 EXPORT_SYMBOL(__napi_schedule); 5882 5883 /** 5884 * napi_schedule_prep - check if napi can be scheduled 5885 * @n: napi context 5886 * 5887 * Test if NAPI routine is already running, and if not mark 5888 * it as running. This is used as a condition variable to 5889 * insure only one NAPI poll instance runs. We also make 5890 * sure there is no pending NAPI disable. 5891 */ 5892 bool napi_schedule_prep(struct napi_struct *n) 5893 { 5894 unsigned long val, new; 5895 5896 do { 5897 val = READ_ONCE(n->state); 5898 if (unlikely(val & NAPIF_STATE_DISABLE)) 5899 return false; 5900 new = val | NAPIF_STATE_SCHED; 5901 5902 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5903 * This was suggested by Alexander Duyck, as compiler 5904 * emits better code than : 5905 * if (val & NAPIF_STATE_SCHED) 5906 * new |= NAPIF_STATE_MISSED; 5907 */ 5908 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5909 NAPIF_STATE_MISSED; 5910 } while (cmpxchg(&n->state, val, new) != val); 5911 5912 return !(val & NAPIF_STATE_SCHED); 5913 } 5914 EXPORT_SYMBOL(napi_schedule_prep); 5915 5916 /** 5917 * __napi_schedule_irqoff - schedule for receive 5918 * @n: entry to schedule 5919 * 5920 * Variant of __napi_schedule() assuming hard irqs are masked. 5921 * 5922 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5923 * because the interrupt disabled assumption might not be true 5924 * due to force-threaded interrupts and spinlock substitution. 5925 */ 5926 void __napi_schedule_irqoff(struct napi_struct *n) 5927 { 5928 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5929 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5930 else 5931 __napi_schedule(n); 5932 } 5933 EXPORT_SYMBOL(__napi_schedule_irqoff); 5934 5935 bool napi_complete_done(struct napi_struct *n, int work_done) 5936 { 5937 unsigned long flags, val, new, timeout = 0; 5938 bool ret = true; 5939 5940 /* 5941 * 1) Don't let napi dequeue from the cpu poll list 5942 * just in case its running on a different cpu. 5943 * 2) If we are busy polling, do nothing here, we have 5944 * the guarantee we will be called later. 5945 */ 5946 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5947 NAPIF_STATE_IN_BUSY_POLL))) 5948 return false; 5949 5950 if (work_done) { 5951 if (n->gro_bitmask) 5952 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5953 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 5954 } 5955 if (n->defer_hard_irqs_count > 0) { 5956 n->defer_hard_irqs_count--; 5957 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5958 if (timeout) 5959 ret = false; 5960 } 5961 if (n->gro_bitmask) { 5962 /* When the NAPI instance uses a timeout and keeps postponing 5963 * it, we need to bound somehow the time packets are kept in 5964 * the GRO layer 5965 */ 5966 napi_gro_flush(n, !!timeout); 5967 } 5968 5969 gro_normal_list(n); 5970 5971 if (unlikely(!list_empty(&n->poll_list))) { 5972 /* If n->poll_list is not empty, we need to mask irqs */ 5973 local_irq_save(flags); 5974 list_del_init(&n->poll_list); 5975 local_irq_restore(flags); 5976 } 5977 5978 do { 5979 val = READ_ONCE(n->state); 5980 5981 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 5982 5983 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 5984 NAPIF_STATE_SCHED_THREADED | 5985 NAPIF_STATE_PREFER_BUSY_POLL); 5986 5987 /* If STATE_MISSED was set, leave STATE_SCHED set, 5988 * because we will call napi->poll() one more time. 5989 * This C code was suggested by Alexander Duyck to help gcc. 5990 */ 5991 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 5992 NAPIF_STATE_SCHED; 5993 } while (cmpxchg(&n->state, val, new) != val); 5994 5995 if (unlikely(val & NAPIF_STATE_MISSED)) { 5996 __napi_schedule(n); 5997 return false; 5998 } 5999 6000 if (timeout) 6001 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6002 HRTIMER_MODE_REL_PINNED); 6003 return ret; 6004 } 6005 EXPORT_SYMBOL(napi_complete_done); 6006 6007 /* must be called under rcu_read_lock(), as we dont take a reference */ 6008 static struct napi_struct *napi_by_id(unsigned int napi_id) 6009 { 6010 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6011 struct napi_struct *napi; 6012 6013 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6014 if (napi->napi_id == napi_id) 6015 return napi; 6016 6017 return NULL; 6018 } 6019 6020 #if defined(CONFIG_NET_RX_BUSY_POLL) 6021 6022 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6023 { 6024 if (!skip_schedule) { 6025 gro_normal_list(napi); 6026 __napi_schedule(napi); 6027 return; 6028 } 6029 6030 if (napi->gro_bitmask) { 6031 /* flush too old packets 6032 * If HZ < 1000, flush all packets. 6033 */ 6034 napi_gro_flush(napi, HZ >= 1000); 6035 } 6036 6037 gro_normal_list(napi); 6038 clear_bit(NAPI_STATE_SCHED, &napi->state); 6039 } 6040 6041 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6042 u16 budget) 6043 { 6044 bool skip_schedule = false; 6045 unsigned long timeout; 6046 int rc; 6047 6048 /* Busy polling means there is a high chance device driver hard irq 6049 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6050 * set in napi_schedule_prep(). 6051 * Since we are about to call napi->poll() once more, we can safely 6052 * clear NAPI_STATE_MISSED. 6053 * 6054 * Note: x86 could use a single "lock and ..." instruction 6055 * to perform these two clear_bit() 6056 */ 6057 clear_bit(NAPI_STATE_MISSED, &napi->state); 6058 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6059 6060 local_bh_disable(); 6061 6062 if (prefer_busy_poll) { 6063 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6064 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6065 if (napi->defer_hard_irqs_count && timeout) { 6066 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6067 skip_schedule = true; 6068 } 6069 } 6070 6071 /* All we really want here is to re-enable device interrupts. 6072 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6073 */ 6074 rc = napi->poll(napi, budget); 6075 /* We can't gro_normal_list() here, because napi->poll() might have 6076 * rearmed the napi (napi_complete_done()) in which case it could 6077 * already be running on another CPU. 6078 */ 6079 trace_napi_poll(napi, rc, budget); 6080 netpoll_poll_unlock(have_poll_lock); 6081 if (rc == budget) 6082 __busy_poll_stop(napi, skip_schedule); 6083 local_bh_enable(); 6084 } 6085 6086 void napi_busy_loop(unsigned int napi_id, 6087 bool (*loop_end)(void *, unsigned long), 6088 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6089 { 6090 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6091 int (*napi_poll)(struct napi_struct *napi, int budget); 6092 void *have_poll_lock = NULL; 6093 struct napi_struct *napi; 6094 6095 restart: 6096 napi_poll = NULL; 6097 6098 rcu_read_lock(); 6099 6100 napi = napi_by_id(napi_id); 6101 if (!napi) 6102 goto out; 6103 6104 preempt_disable(); 6105 for (;;) { 6106 int work = 0; 6107 6108 local_bh_disable(); 6109 if (!napi_poll) { 6110 unsigned long val = READ_ONCE(napi->state); 6111 6112 /* If multiple threads are competing for this napi, 6113 * we avoid dirtying napi->state as much as we can. 6114 */ 6115 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6116 NAPIF_STATE_IN_BUSY_POLL)) { 6117 if (prefer_busy_poll) 6118 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6119 goto count; 6120 } 6121 if (cmpxchg(&napi->state, val, 6122 val | NAPIF_STATE_IN_BUSY_POLL | 6123 NAPIF_STATE_SCHED) != val) { 6124 if (prefer_busy_poll) 6125 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6126 goto count; 6127 } 6128 have_poll_lock = netpoll_poll_lock(napi); 6129 napi_poll = napi->poll; 6130 } 6131 work = napi_poll(napi, budget); 6132 trace_napi_poll(napi, work, budget); 6133 gro_normal_list(napi); 6134 count: 6135 if (work > 0) 6136 __NET_ADD_STATS(dev_net(napi->dev), 6137 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6138 local_bh_enable(); 6139 6140 if (!loop_end || loop_end(loop_end_arg, start_time)) 6141 break; 6142 6143 if (unlikely(need_resched())) { 6144 if (napi_poll) 6145 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6146 preempt_enable(); 6147 rcu_read_unlock(); 6148 cond_resched(); 6149 if (loop_end(loop_end_arg, start_time)) 6150 return; 6151 goto restart; 6152 } 6153 cpu_relax(); 6154 } 6155 if (napi_poll) 6156 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6157 preempt_enable(); 6158 out: 6159 rcu_read_unlock(); 6160 } 6161 EXPORT_SYMBOL(napi_busy_loop); 6162 6163 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6164 6165 static void napi_hash_add(struct napi_struct *napi) 6166 { 6167 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6168 return; 6169 6170 spin_lock(&napi_hash_lock); 6171 6172 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6173 do { 6174 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6175 napi_gen_id = MIN_NAPI_ID; 6176 } while (napi_by_id(napi_gen_id)); 6177 napi->napi_id = napi_gen_id; 6178 6179 hlist_add_head_rcu(&napi->napi_hash_node, 6180 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6181 6182 spin_unlock(&napi_hash_lock); 6183 } 6184 6185 /* Warning : caller is responsible to make sure rcu grace period 6186 * is respected before freeing memory containing @napi 6187 */ 6188 static void napi_hash_del(struct napi_struct *napi) 6189 { 6190 spin_lock(&napi_hash_lock); 6191 6192 hlist_del_init_rcu(&napi->napi_hash_node); 6193 6194 spin_unlock(&napi_hash_lock); 6195 } 6196 6197 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6198 { 6199 struct napi_struct *napi; 6200 6201 napi = container_of(timer, struct napi_struct, timer); 6202 6203 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6204 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6205 */ 6206 if (!napi_disable_pending(napi) && 6207 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6208 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6209 __napi_schedule_irqoff(napi); 6210 } 6211 6212 return HRTIMER_NORESTART; 6213 } 6214 6215 static void init_gro_hash(struct napi_struct *napi) 6216 { 6217 int i; 6218 6219 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6220 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6221 napi->gro_hash[i].count = 0; 6222 } 6223 napi->gro_bitmask = 0; 6224 } 6225 6226 int dev_set_threaded(struct net_device *dev, bool threaded) 6227 { 6228 struct napi_struct *napi; 6229 int err = 0; 6230 6231 if (dev->threaded == threaded) 6232 return 0; 6233 6234 if (threaded) { 6235 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6236 if (!napi->thread) { 6237 err = napi_kthread_create(napi); 6238 if (err) { 6239 threaded = false; 6240 break; 6241 } 6242 } 6243 } 6244 } 6245 6246 dev->threaded = threaded; 6247 6248 /* Make sure kthread is created before THREADED bit 6249 * is set. 6250 */ 6251 smp_mb__before_atomic(); 6252 6253 /* Setting/unsetting threaded mode on a napi might not immediately 6254 * take effect, if the current napi instance is actively being 6255 * polled. In this case, the switch between threaded mode and 6256 * softirq mode will happen in the next round of napi_schedule(). 6257 * This should not cause hiccups/stalls to the live traffic. 6258 */ 6259 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6260 if (threaded) 6261 set_bit(NAPI_STATE_THREADED, &napi->state); 6262 else 6263 clear_bit(NAPI_STATE_THREADED, &napi->state); 6264 } 6265 6266 return err; 6267 } 6268 EXPORT_SYMBOL(dev_set_threaded); 6269 6270 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6271 int (*poll)(struct napi_struct *, int), int weight) 6272 { 6273 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6274 return; 6275 6276 INIT_LIST_HEAD(&napi->poll_list); 6277 INIT_HLIST_NODE(&napi->napi_hash_node); 6278 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6279 napi->timer.function = napi_watchdog; 6280 init_gro_hash(napi); 6281 napi->skb = NULL; 6282 INIT_LIST_HEAD(&napi->rx_list); 6283 napi->rx_count = 0; 6284 napi->poll = poll; 6285 if (weight > NAPI_POLL_WEIGHT) 6286 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6287 weight); 6288 napi->weight = weight; 6289 napi->dev = dev; 6290 #ifdef CONFIG_NETPOLL 6291 napi->poll_owner = -1; 6292 #endif 6293 set_bit(NAPI_STATE_SCHED, &napi->state); 6294 set_bit(NAPI_STATE_NPSVC, &napi->state); 6295 list_add_rcu(&napi->dev_list, &dev->napi_list); 6296 napi_hash_add(napi); 6297 /* Create kthread for this napi if dev->threaded is set. 6298 * Clear dev->threaded if kthread creation failed so that 6299 * threaded mode will not be enabled in napi_enable(). 6300 */ 6301 if (dev->threaded && napi_kthread_create(napi)) 6302 dev->threaded = 0; 6303 } 6304 EXPORT_SYMBOL(netif_napi_add); 6305 6306 void napi_disable(struct napi_struct *n) 6307 { 6308 unsigned long val, new; 6309 6310 might_sleep(); 6311 set_bit(NAPI_STATE_DISABLE, &n->state); 6312 6313 for ( ; ; ) { 6314 val = READ_ONCE(n->state); 6315 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6316 usleep_range(20, 200); 6317 continue; 6318 } 6319 6320 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6321 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6322 6323 if (cmpxchg(&n->state, val, new) == val) 6324 break; 6325 } 6326 6327 hrtimer_cancel(&n->timer); 6328 6329 clear_bit(NAPI_STATE_DISABLE, &n->state); 6330 } 6331 EXPORT_SYMBOL(napi_disable); 6332 6333 /** 6334 * napi_enable - enable NAPI scheduling 6335 * @n: NAPI context 6336 * 6337 * Resume NAPI from being scheduled on this context. 6338 * Must be paired with napi_disable. 6339 */ 6340 void napi_enable(struct napi_struct *n) 6341 { 6342 unsigned long val, new; 6343 6344 do { 6345 val = READ_ONCE(n->state); 6346 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6347 6348 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6349 if (n->dev->threaded && n->thread) 6350 new |= NAPIF_STATE_THREADED; 6351 } while (cmpxchg(&n->state, val, new) != val); 6352 } 6353 EXPORT_SYMBOL(napi_enable); 6354 6355 static void flush_gro_hash(struct napi_struct *napi) 6356 { 6357 int i; 6358 6359 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6360 struct sk_buff *skb, *n; 6361 6362 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6363 kfree_skb(skb); 6364 napi->gro_hash[i].count = 0; 6365 } 6366 } 6367 6368 /* Must be called in process context */ 6369 void __netif_napi_del(struct napi_struct *napi) 6370 { 6371 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6372 return; 6373 6374 napi_hash_del(napi); 6375 list_del_rcu(&napi->dev_list); 6376 napi_free_frags(napi); 6377 6378 flush_gro_hash(napi); 6379 napi->gro_bitmask = 0; 6380 6381 if (napi->thread) { 6382 kthread_stop(napi->thread); 6383 napi->thread = NULL; 6384 } 6385 } 6386 EXPORT_SYMBOL(__netif_napi_del); 6387 6388 static int __napi_poll(struct napi_struct *n, bool *repoll) 6389 { 6390 int work, weight; 6391 6392 weight = n->weight; 6393 6394 /* This NAPI_STATE_SCHED test is for avoiding a race 6395 * with netpoll's poll_napi(). Only the entity which 6396 * obtains the lock and sees NAPI_STATE_SCHED set will 6397 * actually make the ->poll() call. Therefore we avoid 6398 * accidentally calling ->poll() when NAPI is not scheduled. 6399 */ 6400 work = 0; 6401 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6402 work = n->poll(n, weight); 6403 trace_napi_poll(n, work, weight); 6404 } 6405 6406 if (unlikely(work > weight)) 6407 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6408 n->poll, work, weight); 6409 6410 if (likely(work < weight)) 6411 return work; 6412 6413 /* Drivers must not modify the NAPI state if they 6414 * consume the entire weight. In such cases this code 6415 * still "owns" the NAPI instance and therefore can 6416 * move the instance around on the list at-will. 6417 */ 6418 if (unlikely(napi_disable_pending(n))) { 6419 napi_complete(n); 6420 return work; 6421 } 6422 6423 /* The NAPI context has more processing work, but busy-polling 6424 * is preferred. Exit early. 6425 */ 6426 if (napi_prefer_busy_poll(n)) { 6427 if (napi_complete_done(n, work)) { 6428 /* If timeout is not set, we need to make sure 6429 * that the NAPI is re-scheduled. 6430 */ 6431 napi_schedule(n); 6432 } 6433 return work; 6434 } 6435 6436 if (n->gro_bitmask) { 6437 /* flush too old packets 6438 * If HZ < 1000, flush all packets. 6439 */ 6440 napi_gro_flush(n, HZ >= 1000); 6441 } 6442 6443 gro_normal_list(n); 6444 6445 /* Some drivers may have called napi_schedule 6446 * prior to exhausting their budget. 6447 */ 6448 if (unlikely(!list_empty(&n->poll_list))) { 6449 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6450 n->dev ? n->dev->name : "backlog"); 6451 return work; 6452 } 6453 6454 *repoll = true; 6455 6456 return work; 6457 } 6458 6459 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6460 { 6461 bool do_repoll = false; 6462 void *have; 6463 int work; 6464 6465 list_del_init(&n->poll_list); 6466 6467 have = netpoll_poll_lock(n); 6468 6469 work = __napi_poll(n, &do_repoll); 6470 6471 if (do_repoll) 6472 list_add_tail(&n->poll_list, repoll); 6473 6474 netpoll_poll_unlock(have); 6475 6476 return work; 6477 } 6478 6479 static int napi_thread_wait(struct napi_struct *napi) 6480 { 6481 bool woken = false; 6482 6483 set_current_state(TASK_INTERRUPTIBLE); 6484 6485 while (!kthread_should_stop()) { 6486 /* Testing SCHED_THREADED bit here to make sure the current 6487 * kthread owns this napi and could poll on this napi. 6488 * Testing SCHED bit is not enough because SCHED bit might be 6489 * set by some other busy poll thread or by napi_disable(). 6490 */ 6491 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6492 WARN_ON(!list_empty(&napi->poll_list)); 6493 __set_current_state(TASK_RUNNING); 6494 return 0; 6495 } 6496 6497 schedule(); 6498 /* woken being true indicates this thread owns this napi. */ 6499 woken = true; 6500 set_current_state(TASK_INTERRUPTIBLE); 6501 } 6502 __set_current_state(TASK_RUNNING); 6503 6504 return -1; 6505 } 6506 6507 static int napi_threaded_poll(void *data) 6508 { 6509 struct napi_struct *napi = data; 6510 void *have; 6511 6512 while (!napi_thread_wait(napi)) { 6513 for (;;) { 6514 bool repoll = false; 6515 6516 local_bh_disable(); 6517 6518 have = netpoll_poll_lock(napi); 6519 __napi_poll(napi, &repoll); 6520 netpoll_poll_unlock(have); 6521 6522 local_bh_enable(); 6523 6524 if (!repoll) 6525 break; 6526 6527 cond_resched(); 6528 } 6529 } 6530 return 0; 6531 } 6532 6533 static __latent_entropy void net_rx_action(struct softirq_action *h) 6534 { 6535 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6536 unsigned long time_limit = jiffies + 6537 usecs_to_jiffies(netdev_budget_usecs); 6538 int budget = netdev_budget; 6539 LIST_HEAD(list); 6540 LIST_HEAD(repoll); 6541 6542 local_irq_disable(); 6543 list_splice_init(&sd->poll_list, &list); 6544 local_irq_enable(); 6545 6546 for (;;) { 6547 struct napi_struct *n; 6548 6549 if (list_empty(&list)) { 6550 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6551 return; 6552 break; 6553 } 6554 6555 n = list_first_entry(&list, struct napi_struct, poll_list); 6556 budget -= napi_poll(n, &repoll); 6557 6558 /* If softirq window is exhausted then punt. 6559 * Allow this to run for 2 jiffies since which will allow 6560 * an average latency of 1.5/HZ. 6561 */ 6562 if (unlikely(budget <= 0 || 6563 time_after_eq(jiffies, time_limit))) { 6564 sd->time_squeeze++; 6565 break; 6566 } 6567 } 6568 6569 local_irq_disable(); 6570 6571 list_splice_tail_init(&sd->poll_list, &list); 6572 list_splice_tail(&repoll, &list); 6573 list_splice(&list, &sd->poll_list); 6574 if (!list_empty(&sd->poll_list)) 6575 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6576 6577 net_rps_action_and_irq_enable(sd); 6578 } 6579 6580 struct netdev_adjacent { 6581 struct net_device *dev; 6582 netdevice_tracker dev_tracker; 6583 6584 /* upper master flag, there can only be one master device per list */ 6585 bool master; 6586 6587 /* lookup ignore flag */ 6588 bool ignore; 6589 6590 /* counter for the number of times this device was added to us */ 6591 u16 ref_nr; 6592 6593 /* private field for the users */ 6594 void *private; 6595 6596 struct list_head list; 6597 struct rcu_head rcu; 6598 }; 6599 6600 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6601 struct list_head *adj_list) 6602 { 6603 struct netdev_adjacent *adj; 6604 6605 list_for_each_entry(adj, adj_list, list) { 6606 if (adj->dev == adj_dev) 6607 return adj; 6608 } 6609 return NULL; 6610 } 6611 6612 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6613 struct netdev_nested_priv *priv) 6614 { 6615 struct net_device *dev = (struct net_device *)priv->data; 6616 6617 return upper_dev == dev; 6618 } 6619 6620 /** 6621 * netdev_has_upper_dev - Check if device is linked to an upper device 6622 * @dev: device 6623 * @upper_dev: upper device to check 6624 * 6625 * Find out if a device is linked to specified upper device and return true 6626 * in case it is. Note that this checks only immediate upper device, 6627 * not through a complete stack of devices. The caller must hold the RTNL lock. 6628 */ 6629 bool netdev_has_upper_dev(struct net_device *dev, 6630 struct net_device *upper_dev) 6631 { 6632 struct netdev_nested_priv priv = { 6633 .data = (void *)upper_dev, 6634 }; 6635 6636 ASSERT_RTNL(); 6637 6638 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6639 &priv); 6640 } 6641 EXPORT_SYMBOL(netdev_has_upper_dev); 6642 6643 /** 6644 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6645 * @dev: device 6646 * @upper_dev: upper device to check 6647 * 6648 * Find out if a device is linked to specified upper device and return true 6649 * in case it is. Note that this checks the entire upper device chain. 6650 * The caller must hold rcu lock. 6651 */ 6652 6653 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6654 struct net_device *upper_dev) 6655 { 6656 struct netdev_nested_priv priv = { 6657 .data = (void *)upper_dev, 6658 }; 6659 6660 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6661 &priv); 6662 } 6663 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6664 6665 /** 6666 * netdev_has_any_upper_dev - Check if device is linked to some device 6667 * @dev: device 6668 * 6669 * Find out if a device is linked to an upper device and return true in case 6670 * it is. The caller must hold the RTNL lock. 6671 */ 6672 bool netdev_has_any_upper_dev(struct net_device *dev) 6673 { 6674 ASSERT_RTNL(); 6675 6676 return !list_empty(&dev->adj_list.upper); 6677 } 6678 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6679 6680 /** 6681 * netdev_master_upper_dev_get - Get master upper device 6682 * @dev: device 6683 * 6684 * Find a master upper device and return pointer to it or NULL in case 6685 * it's not there. The caller must hold the RTNL lock. 6686 */ 6687 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6688 { 6689 struct netdev_adjacent *upper; 6690 6691 ASSERT_RTNL(); 6692 6693 if (list_empty(&dev->adj_list.upper)) 6694 return NULL; 6695 6696 upper = list_first_entry(&dev->adj_list.upper, 6697 struct netdev_adjacent, list); 6698 if (likely(upper->master)) 6699 return upper->dev; 6700 return NULL; 6701 } 6702 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6703 6704 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6705 { 6706 struct netdev_adjacent *upper; 6707 6708 ASSERT_RTNL(); 6709 6710 if (list_empty(&dev->adj_list.upper)) 6711 return NULL; 6712 6713 upper = list_first_entry(&dev->adj_list.upper, 6714 struct netdev_adjacent, list); 6715 if (likely(upper->master) && !upper->ignore) 6716 return upper->dev; 6717 return NULL; 6718 } 6719 6720 /** 6721 * netdev_has_any_lower_dev - Check if device is linked to some device 6722 * @dev: device 6723 * 6724 * Find out if a device is linked to a lower device and return true in case 6725 * it is. The caller must hold the RTNL lock. 6726 */ 6727 static bool netdev_has_any_lower_dev(struct net_device *dev) 6728 { 6729 ASSERT_RTNL(); 6730 6731 return !list_empty(&dev->adj_list.lower); 6732 } 6733 6734 void *netdev_adjacent_get_private(struct list_head *adj_list) 6735 { 6736 struct netdev_adjacent *adj; 6737 6738 adj = list_entry(adj_list, struct netdev_adjacent, list); 6739 6740 return adj->private; 6741 } 6742 EXPORT_SYMBOL(netdev_adjacent_get_private); 6743 6744 /** 6745 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6746 * @dev: device 6747 * @iter: list_head ** of the current position 6748 * 6749 * Gets the next device from the dev's upper list, starting from iter 6750 * position. The caller must hold RCU read lock. 6751 */ 6752 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6753 struct list_head **iter) 6754 { 6755 struct netdev_adjacent *upper; 6756 6757 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6758 6759 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6760 6761 if (&upper->list == &dev->adj_list.upper) 6762 return NULL; 6763 6764 *iter = &upper->list; 6765 6766 return upper->dev; 6767 } 6768 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6769 6770 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6771 struct list_head **iter, 6772 bool *ignore) 6773 { 6774 struct netdev_adjacent *upper; 6775 6776 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6777 6778 if (&upper->list == &dev->adj_list.upper) 6779 return NULL; 6780 6781 *iter = &upper->list; 6782 *ignore = upper->ignore; 6783 6784 return upper->dev; 6785 } 6786 6787 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6788 struct list_head **iter) 6789 { 6790 struct netdev_adjacent *upper; 6791 6792 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6793 6794 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6795 6796 if (&upper->list == &dev->adj_list.upper) 6797 return NULL; 6798 6799 *iter = &upper->list; 6800 6801 return upper->dev; 6802 } 6803 6804 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6805 int (*fn)(struct net_device *dev, 6806 struct netdev_nested_priv *priv), 6807 struct netdev_nested_priv *priv) 6808 { 6809 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6810 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6811 int ret, cur = 0; 6812 bool ignore; 6813 6814 now = dev; 6815 iter = &dev->adj_list.upper; 6816 6817 while (1) { 6818 if (now != dev) { 6819 ret = fn(now, priv); 6820 if (ret) 6821 return ret; 6822 } 6823 6824 next = NULL; 6825 while (1) { 6826 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6827 if (!udev) 6828 break; 6829 if (ignore) 6830 continue; 6831 6832 next = udev; 6833 niter = &udev->adj_list.upper; 6834 dev_stack[cur] = now; 6835 iter_stack[cur++] = iter; 6836 break; 6837 } 6838 6839 if (!next) { 6840 if (!cur) 6841 return 0; 6842 next = dev_stack[--cur]; 6843 niter = iter_stack[cur]; 6844 } 6845 6846 now = next; 6847 iter = niter; 6848 } 6849 6850 return 0; 6851 } 6852 6853 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6854 int (*fn)(struct net_device *dev, 6855 struct netdev_nested_priv *priv), 6856 struct netdev_nested_priv *priv) 6857 { 6858 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6859 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6860 int ret, cur = 0; 6861 6862 now = dev; 6863 iter = &dev->adj_list.upper; 6864 6865 while (1) { 6866 if (now != dev) { 6867 ret = fn(now, priv); 6868 if (ret) 6869 return ret; 6870 } 6871 6872 next = NULL; 6873 while (1) { 6874 udev = netdev_next_upper_dev_rcu(now, &iter); 6875 if (!udev) 6876 break; 6877 6878 next = udev; 6879 niter = &udev->adj_list.upper; 6880 dev_stack[cur] = now; 6881 iter_stack[cur++] = iter; 6882 break; 6883 } 6884 6885 if (!next) { 6886 if (!cur) 6887 return 0; 6888 next = dev_stack[--cur]; 6889 niter = iter_stack[cur]; 6890 } 6891 6892 now = next; 6893 iter = niter; 6894 } 6895 6896 return 0; 6897 } 6898 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6899 6900 static bool __netdev_has_upper_dev(struct net_device *dev, 6901 struct net_device *upper_dev) 6902 { 6903 struct netdev_nested_priv priv = { 6904 .flags = 0, 6905 .data = (void *)upper_dev, 6906 }; 6907 6908 ASSERT_RTNL(); 6909 6910 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 6911 &priv); 6912 } 6913 6914 /** 6915 * netdev_lower_get_next_private - Get the next ->private from the 6916 * lower neighbour list 6917 * @dev: device 6918 * @iter: list_head ** of the current position 6919 * 6920 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6921 * list, starting from iter position. The caller must hold either hold the 6922 * RTNL lock or its own locking that guarantees that the neighbour lower 6923 * list will remain unchanged. 6924 */ 6925 void *netdev_lower_get_next_private(struct net_device *dev, 6926 struct list_head **iter) 6927 { 6928 struct netdev_adjacent *lower; 6929 6930 lower = list_entry(*iter, struct netdev_adjacent, list); 6931 6932 if (&lower->list == &dev->adj_list.lower) 6933 return NULL; 6934 6935 *iter = lower->list.next; 6936 6937 return lower->private; 6938 } 6939 EXPORT_SYMBOL(netdev_lower_get_next_private); 6940 6941 /** 6942 * netdev_lower_get_next_private_rcu - Get the next ->private from the 6943 * lower neighbour list, RCU 6944 * variant 6945 * @dev: device 6946 * @iter: list_head ** of the current position 6947 * 6948 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6949 * list, starting from iter position. The caller must hold RCU read lock. 6950 */ 6951 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 6952 struct list_head **iter) 6953 { 6954 struct netdev_adjacent *lower; 6955 6956 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 6957 6958 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6959 6960 if (&lower->list == &dev->adj_list.lower) 6961 return NULL; 6962 6963 *iter = &lower->list; 6964 6965 return lower->private; 6966 } 6967 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 6968 6969 /** 6970 * netdev_lower_get_next - Get the next device from the lower neighbour 6971 * list 6972 * @dev: device 6973 * @iter: list_head ** of the current position 6974 * 6975 * Gets the next netdev_adjacent from the dev's lower neighbour 6976 * list, starting from iter position. The caller must hold RTNL lock or 6977 * its own locking that guarantees that the neighbour lower 6978 * list will remain unchanged. 6979 */ 6980 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 6981 { 6982 struct netdev_adjacent *lower; 6983 6984 lower = list_entry(*iter, struct netdev_adjacent, list); 6985 6986 if (&lower->list == &dev->adj_list.lower) 6987 return NULL; 6988 6989 *iter = lower->list.next; 6990 6991 return lower->dev; 6992 } 6993 EXPORT_SYMBOL(netdev_lower_get_next); 6994 6995 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 6996 struct list_head **iter) 6997 { 6998 struct netdev_adjacent *lower; 6999 7000 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7001 7002 if (&lower->list == &dev->adj_list.lower) 7003 return NULL; 7004 7005 *iter = &lower->list; 7006 7007 return lower->dev; 7008 } 7009 7010 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7011 struct list_head **iter, 7012 bool *ignore) 7013 { 7014 struct netdev_adjacent *lower; 7015 7016 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7017 7018 if (&lower->list == &dev->adj_list.lower) 7019 return NULL; 7020 7021 *iter = &lower->list; 7022 *ignore = lower->ignore; 7023 7024 return lower->dev; 7025 } 7026 7027 int netdev_walk_all_lower_dev(struct net_device *dev, 7028 int (*fn)(struct net_device *dev, 7029 struct netdev_nested_priv *priv), 7030 struct netdev_nested_priv *priv) 7031 { 7032 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7033 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7034 int ret, cur = 0; 7035 7036 now = dev; 7037 iter = &dev->adj_list.lower; 7038 7039 while (1) { 7040 if (now != dev) { 7041 ret = fn(now, priv); 7042 if (ret) 7043 return ret; 7044 } 7045 7046 next = NULL; 7047 while (1) { 7048 ldev = netdev_next_lower_dev(now, &iter); 7049 if (!ldev) 7050 break; 7051 7052 next = ldev; 7053 niter = &ldev->adj_list.lower; 7054 dev_stack[cur] = now; 7055 iter_stack[cur++] = iter; 7056 break; 7057 } 7058 7059 if (!next) { 7060 if (!cur) 7061 return 0; 7062 next = dev_stack[--cur]; 7063 niter = iter_stack[cur]; 7064 } 7065 7066 now = next; 7067 iter = niter; 7068 } 7069 7070 return 0; 7071 } 7072 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7073 7074 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7075 int (*fn)(struct net_device *dev, 7076 struct netdev_nested_priv *priv), 7077 struct netdev_nested_priv *priv) 7078 { 7079 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7080 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7081 int ret, cur = 0; 7082 bool ignore; 7083 7084 now = dev; 7085 iter = &dev->adj_list.lower; 7086 7087 while (1) { 7088 if (now != dev) { 7089 ret = fn(now, priv); 7090 if (ret) 7091 return ret; 7092 } 7093 7094 next = NULL; 7095 while (1) { 7096 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7097 if (!ldev) 7098 break; 7099 if (ignore) 7100 continue; 7101 7102 next = ldev; 7103 niter = &ldev->adj_list.lower; 7104 dev_stack[cur] = now; 7105 iter_stack[cur++] = iter; 7106 break; 7107 } 7108 7109 if (!next) { 7110 if (!cur) 7111 return 0; 7112 next = dev_stack[--cur]; 7113 niter = iter_stack[cur]; 7114 } 7115 7116 now = next; 7117 iter = niter; 7118 } 7119 7120 return 0; 7121 } 7122 7123 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7124 struct list_head **iter) 7125 { 7126 struct netdev_adjacent *lower; 7127 7128 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7129 if (&lower->list == &dev->adj_list.lower) 7130 return NULL; 7131 7132 *iter = &lower->list; 7133 7134 return lower->dev; 7135 } 7136 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7137 7138 static u8 __netdev_upper_depth(struct net_device *dev) 7139 { 7140 struct net_device *udev; 7141 struct list_head *iter; 7142 u8 max_depth = 0; 7143 bool ignore; 7144 7145 for (iter = &dev->adj_list.upper, 7146 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7147 udev; 7148 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7149 if (ignore) 7150 continue; 7151 if (max_depth < udev->upper_level) 7152 max_depth = udev->upper_level; 7153 } 7154 7155 return max_depth; 7156 } 7157 7158 static u8 __netdev_lower_depth(struct net_device *dev) 7159 { 7160 struct net_device *ldev; 7161 struct list_head *iter; 7162 u8 max_depth = 0; 7163 bool ignore; 7164 7165 for (iter = &dev->adj_list.lower, 7166 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7167 ldev; 7168 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7169 if (ignore) 7170 continue; 7171 if (max_depth < ldev->lower_level) 7172 max_depth = ldev->lower_level; 7173 } 7174 7175 return max_depth; 7176 } 7177 7178 static int __netdev_update_upper_level(struct net_device *dev, 7179 struct netdev_nested_priv *__unused) 7180 { 7181 dev->upper_level = __netdev_upper_depth(dev) + 1; 7182 return 0; 7183 } 7184 7185 static int __netdev_update_lower_level(struct net_device *dev, 7186 struct netdev_nested_priv *priv) 7187 { 7188 dev->lower_level = __netdev_lower_depth(dev) + 1; 7189 7190 #ifdef CONFIG_LOCKDEP 7191 if (!priv) 7192 return 0; 7193 7194 if (priv->flags & NESTED_SYNC_IMM) 7195 dev->nested_level = dev->lower_level - 1; 7196 if (priv->flags & NESTED_SYNC_TODO) 7197 net_unlink_todo(dev); 7198 #endif 7199 return 0; 7200 } 7201 7202 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7203 int (*fn)(struct net_device *dev, 7204 struct netdev_nested_priv *priv), 7205 struct netdev_nested_priv *priv) 7206 { 7207 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7208 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7209 int ret, cur = 0; 7210 7211 now = dev; 7212 iter = &dev->adj_list.lower; 7213 7214 while (1) { 7215 if (now != dev) { 7216 ret = fn(now, priv); 7217 if (ret) 7218 return ret; 7219 } 7220 7221 next = NULL; 7222 while (1) { 7223 ldev = netdev_next_lower_dev_rcu(now, &iter); 7224 if (!ldev) 7225 break; 7226 7227 next = ldev; 7228 niter = &ldev->adj_list.lower; 7229 dev_stack[cur] = now; 7230 iter_stack[cur++] = iter; 7231 break; 7232 } 7233 7234 if (!next) { 7235 if (!cur) 7236 return 0; 7237 next = dev_stack[--cur]; 7238 niter = iter_stack[cur]; 7239 } 7240 7241 now = next; 7242 iter = niter; 7243 } 7244 7245 return 0; 7246 } 7247 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7248 7249 /** 7250 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7251 * lower neighbour list, RCU 7252 * variant 7253 * @dev: device 7254 * 7255 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7256 * list. The caller must hold RCU read lock. 7257 */ 7258 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7259 { 7260 struct netdev_adjacent *lower; 7261 7262 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7263 struct netdev_adjacent, list); 7264 if (lower) 7265 return lower->private; 7266 return NULL; 7267 } 7268 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7269 7270 /** 7271 * netdev_master_upper_dev_get_rcu - Get master upper device 7272 * @dev: device 7273 * 7274 * Find a master upper device and return pointer to it or NULL in case 7275 * it's not there. The caller must hold the RCU read lock. 7276 */ 7277 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7278 { 7279 struct netdev_adjacent *upper; 7280 7281 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7282 struct netdev_adjacent, list); 7283 if (upper && likely(upper->master)) 7284 return upper->dev; 7285 return NULL; 7286 } 7287 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7288 7289 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7290 struct net_device *adj_dev, 7291 struct list_head *dev_list) 7292 { 7293 char linkname[IFNAMSIZ+7]; 7294 7295 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7296 "upper_%s" : "lower_%s", adj_dev->name); 7297 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7298 linkname); 7299 } 7300 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7301 char *name, 7302 struct list_head *dev_list) 7303 { 7304 char linkname[IFNAMSIZ+7]; 7305 7306 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7307 "upper_%s" : "lower_%s", name); 7308 sysfs_remove_link(&(dev->dev.kobj), linkname); 7309 } 7310 7311 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7312 struct net_device *adj_dev, 7313 struct list_head *dev_list) 7314 { 7315 return (dev_list == &dev->adj_list.upper || 7316 dev_list == &dev->adj_list.lower) && 7317 net_eq(dev_net(dev), dev_net(adj_dev)); 7318 } 7319 7320 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7321 struct net_device *adj_dev, 7322 struct list_head *dev_list, 7323 void *private, bool master) 7324 { 7325 struct netdev_adjacent *adj; 7326 int ret; 7327 7328 adj = __netdev_find_adj(adj_dev, dev_list); 7329 7330 if (adj) { 7331 adj->ref_nr += 1; 7332 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7333 dev->name, adj_dev->name, adj->ref_nr); 7334 7335 return 0; 7336 } 7337 7338 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7339 if (!adj) 7340 return -ENOMEM; 7341 7342 adj->dev = adj_dev; 7343 adj->master = master; 7344 adj->ref_nr = 1; 7345 adj->private = private; 7346 adj->ignore = false; 7347 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7348 7349 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7350 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7351 7352 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7353 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7354 if (ret) 7355 goto free_adj; 7356 } 7357 7358 /* Ensure that master link is always the first item in list. */ 7359 if (master) { 7360 ret = sysfs_create_link(&(dev->dev.kobj), 7361 &(adj_dev->dev.kobj), "master"); 7362 if (ret) 7363 goto remove_symlinks; 7364 7365 list_add_rcu(&adj->list, dev_list); 7366 } else { 7367 list_add_tail_rcu(&adj->list, dev_list); 7368 } 7369 7370 return 0; 7371 7372 remove_symlinks: 7373 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7374 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7375 free_adj: 7376 dev_put_track(adj_dev, &adj->dev_tracker); 7377 kfree(adj); 7378 7379 return ret; 7380 } 7381 7382 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7383 struct net_device *adj_dev, 7384 u16 ref_nr, 7385 struct list_head *dev_list) 7386 { 7387 struct netdev_adjacent *adj; 7388 7389 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7390 dev->name, adj_dev->name, ref_nr); 7391 7392 adj = __netdev_find_adj(adj_dev, dev_list); 7393 7394 if (!adj) { 7395 pr_err("Adjacency does not exist for device %s from %s\n", 7396 dev->name, adj_dev->name); 7397 WARN_ON(1); 7398 return; 7399 } 7400 7401 if (adj->ref_nr > ref_nr) { 7402 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7403 dev->name, adj_dev->name, ref_nr, 7404 adj->ref_nr - ref_nr); 7405 adj->ref_nr -= ref_nr; 7406 return; 7407 } 7408 7409 if (adj->master) 7410 sysfs_remove_link(&(dev->dev.kobj), "master"); 7411 7412 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7413 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7414 7415 list_del_rcu(&adj->list); 7416 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7417 adj_dev->name, dev->name, adj_dev->name); 7418 dev_put_track(adj_dev, &adj->dev_tracker); 7419 kfree_rcu(adj, rcu); 7420 } 7421 7422 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7423 struct net_device *upper_dev, 7424 struct list_head *up_list, 7425 struct list_head *down_list, 7426 void *private, bool master) 7427 { 7428 int ret; 7429 7430 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7431 private, master); 7432 if (ret) 7433 return ret; 7434 7435 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7436 private, false); 7437 if (ret) { 7438 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7439 return ret; 7440 } 7441 7442 return 0; 7443 } 7444 7445 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7446 struct net_device *upper_dev, 7447 u16 ref_nr, 7448 struct list_head *up_list, 7449 struct list_head *down_list) 7450 { 7451 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7452 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7453 } 7454 7455 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7456 struct net_device *upper_dev, 7457 void *private, bool master) 7458 { 7459 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7460 &dev->adj_list.upper, 7461 &upper_dev->adj_list.lower, 7462 private, master); 7463 } 7464 7465 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7466 struct net_device *upper_dev) 7467 { 7468 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7469 &dev->adj_list.upper, 7470 &upper_dev->adj_list.lower); 7471 } 7472 7473 static int __netdev_upper_dev_link(struct net_device *dev, 7474 struct net_device *upper_dev, bool master, 7475 void *upper_priv, void *upper_info, 7476 struct netdev_nested_priv *priv, 7477 struct netlink_ext_ack *extack) 7478 { 7479 struct netdev_notifier_changeupper_info changeupper_info = { 7480 .info = { 7481 .dev = dev, 7482 .extack = extack, 7483 }, 7484 .upper_dev = upper_dev, 7485 .master = master, 7486 .linking = true, 7487 .upper_info = upper_info, 7488 }; 7489 struct net_device *master_dev; 7490 int ret = 0; 7491 7492 ASSERT_RTNL(); 7493 7494 if (dev == upper_dev) 7495 return -EBUSY; 7496 7497 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7498 if (__netdev_has_upper_dev(upper_dev, dev)) 7499 return -EBUSY; 7500 7501 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7502 return -EMLINK; 7503 7504 if (!master) { 7505 if (__netdev_has_upper_dev(dev, upper_dev)) 7506 return -EEXIST; 7507 } else { 7508 master_dev = __netdev_master_upper_dev_get(dev); 7509 if (master_dev) 7510 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7511 } 7512 7513 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7514 &changeupper_info.info); 7515 ret = notifier_to_errno(ret); 7516 if (ret) 7517 return ret; 7518 7519 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7520 master); 7521 if (ret) 7522 return ret; 7523 7524 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7525 &changeupper_info.info); 7526 ret = notifier_to_errno(ret); 7527 if (ret) 7528 goto rollback; 7529 7530 __netdev_update_upper_level(dev, NULL); 7531 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7532 7533 __netdev_update_lower_level(upper_dev, priv); 7534 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7535 priv); 7536 7537 return 0; 7538 7539 rollback: 7540 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7541 7542 return ret; 7543 } 7544 7545 /** 7546 * netdev_upper_dev_link - Add a link to the upper device 7547 * @dev: device 7548 * @upper_dev: new upper device 7549 * @extack: netlink extended ack 7550 * 7551 * Adds a link to device which is upper to this one. The caller must hold 7552 * the RTNL lock. On a failure a negative errno code is returned. 7553 * On success the reference counts are adjusted and the function 7554 * returns zero. 7555 */ 7556 int netdev_upper_dev_link(struct net_device *dev, 7557 struct net_device *upper_dev, 7558 struct netlink_ext_ack *extack) 7559 { 7560 struct netdev_nested_priv priv = { 7561 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7562 .data = NULL, 7563 }; 7564 7565 return __netdev_upper_dev_link(dev, upper_dev, false, 7566 NULL, NULL, &priv, extack); 7567 } 7568 EXPORT_SYMBOL(netdev_upper_dev_link); 7569 7570 /** 7571 * netdev_master_upper_dev_link - Add a master link to the upper device 7572 * @dev: device 7573 * @upper_dev: new upper device 7574 * @upper_priv: upper device private 7575 * @upper_info: upper info to be passed down via notifier 7576 * @extack: netlink extended ack 7577 * 7578 * Adds a link to device which is upper to this one. In this case, only 7579 * one master upper device can be linked, although other non-master devices 7580 * might be linked as well. The caller must hold the RTNL lock. 7581 * On a failure a negative errno code is returned. On success the reference 7582 * counts are adjusted and the function returns zero. 7583 */ 7584 int netdev_master_upper_dev_link(struct net_device *dev, 7585 struct net_device *upper_dev, 7586 void *upper_priv, void *upper_info, 7587 struct netlink_ext_ack *extack) 7588 { 7589 struct netdev_nested_priv priv = { 7590 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7591 .data = NULL, 7592 }; 7593 7594 return __netdev_upper_dev_link(dev, upper_dev, true, 7595 upper_priv, upper_info, &priv, extack); 7596 } 7597 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7598 7599 static void __netdev_upper_dev_unlink(struct net_device *dev, 7600 struct net_device *upper_dev, 7601 struct netdev_nested_priv *priv) 7602 { 7603 struct netdev_notifier_changeupper_info changeupper_info = { 7604 .info = { 7605 .dev = dev, 7606 }, 7607 .upper_dev = upper_dev, 7608 .linking = false, 7609 }; 7610 7611 ASSERT_RTNL(); 7612 7613 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7614 7615 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7616 &changeupper_info.info); 7617 7618 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7619 7620 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7621 &changeupper_info.info); 7622 7623 __netdev_update_upper_level(dev, NULL); 7624 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7625 7626 __netdev_update_lower_level(upper_dev, priv); 7627 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7628 priv); 7629 } 7630 7631 /** 7632 * netdev_upper_dev_unlink - Removes a link to upper device 7633 * @dev: device 7634 * @upper_dev: new upper device 7635 * 7636 * Removes a link to device which is upper to this one. The caller must hold 7637 * the RTNL lock. 7638 */ 7639 void netdev_upper_dev_unlink(struct net_device *dev, 7640 struct net_device *upper_dev) 7641 { 7642 struct netdev_nested_priv priv = { 7643 .flags = NESTED_SYNC_TODO, 7644 .data = NULL, 7645 }; 7646 7647 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7648 } 7649 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7650 7651 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7652 struct net_device *lower_dev, 7653 bool val) 7654 { 7655 struct netdev_adjacent *adj; 7656 7657 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7658 if (adj) 7659 adj->ignore = val; 7660 7661 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7662 if (adj) 7663 adj->ignore = val; 7664 } 7665 7666 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7667 struct net_device *lower_dev) 7668 { 7669 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7670 } 7671 7672 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7673 struct net_device *lower_dev) 7674 { 7675 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7676 } 7677 7678 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7679 struct net_device *new_dev, 7680 struct net_device *dev, 7681 struct netlink_ext_ack *extack) 7682 { 7683 struct netdev_nested_priv priv = { 7684 .flags = 0, 7685 .data = NULL, 7686 }; 7687 int err; 7688 7689 if (!new_dev) 7690 return 0; 7691 7692 if (old_dev && new_dev != old_dev) 7693 netdev_adjacent_dev_disable(dev, old_dev); 7694 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7695 extack); 7696 if (err) { 7697 if (old_dev && new_dev != old_dev) 7698 netdev_adjacent_dev_enable(dev, old_dev); 7699 return err; 7700 } 7701 7702 return 0; 7703 } 7704 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7705 7706 void netdev_adjacent_change_commit(struct net_device *old_dev, 7707 struct net_device *new_dev, 7708 struct net_device *dev) 7709 { 7710 struct netdev_nested_priv priv = { 7711 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7712 .data = NULL, 7713 }; 7714 7715 if (!new_dev || !old_dev) 7716 return; 7717 7718 if (new_dev == old_dev) 7719 return; 7720 7721 netdev_adjacent_dev_enable(dev, old_dev); 7722 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7723 } 7724 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7725 7726 void netdev_adjacent_change_abort(struct net_device *old_dev, 7727 struct net_device *new_dev, 7728 struct net_device *dev) 7729 { 7730 struct netdev_nested_priv priv = { 7731 .flags = 0, 7732 .data = NULL, 7733 }; 7734 7735 if (!new_dev) 7736 return; 7737 7738 if (old_dev && new_dev != old_dev) 7739 netdev_adjacent_dev_enable(dev, old_dev); 7740 7741 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7742 } 7743 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7744 7745 /** 7746 * netdev_bonding_info_change - Dispatch event about slave change 7747 * @dev: device 7748 * @bonding_info: info to dispatch 7749 * 7750 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7751 * The caller must hold the RTNL lock. 7752 */ 7753 void netdev_bonding_info_change(struct net_device *dev, 7754 struct netdev_bonding_info *bonding_info) 7755 { 7756 struct netdev_notifier_bonding_info info = { 7757 .info.dev = dev, 7758 }; 7759 7760 memcpy(&info.bonding_info, bonding_info, 7761 sizeof(struct netdev_bonding_info)); 7762 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7763 &info.info); 7764 } 7765 EXPORT_SYMBOL(netdev_bonding_info_change); 7766 7767 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7768 struct netlink_ext_ack *extack) 7769 { 7770 struct netdev_notifier_offload_xstats_info info = { 7771 .info.dev = dev, 7772 .info.extack = extack, 7773 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7774 }; 7775 int err; 7776 int rc; 7777 7778 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7779 GFP_KERNEL); 7780 if (!dev->offload_xstats_l3) 7781 return -ENOMEM; 7782 7783 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7784 NETDEV_OFFLOAD_XSTATS_DISABLE, 7785 &info.info); 7786 err = notifier_to_errno(rc); 7787 if (err) 7788 goto free_stats; 7789 7790 return 0; 7791 7792 free_stats: 7793 kfree(dev->offload_xstats_l3); 7794 dev->offload_xstats_l3 = NULL; 7795 return err; 7796 } 7797 7798 int netdev_offload_xstats_enable(struct net_device *dev, 7799 enum netdev_offload_xstats_type type, 7800 struct netlink_ext_ack *extack) 7801 { 7802 ASSERT_RTNL(); 7803 7804 if (netdev_offload_xstats_enabled(dev, type)) 7805 return -EALREADY; 7806 7807 switch (type) { 7808 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7809 return netdev_offload_xstats_enable_l3(dev, extack); 7810 } 7811 7812 WARN_ON(1); 7813 return -EINVAL; 7814 } 7815 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7816 7817 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7818 { 7819 struct netdev_notifier_offload_xstats_info info = { 7820 .info.dev = dev, 7821 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7822 }; 7823 7824 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7825 &info.info); 7826 kfree(dev->offload_xstats_l3); 7827 dev->offload_xstats_l3 = NULL; 7828 } 7829 7830 int netdev_offload_xstats_disable(struct net_device *dev, 7831 enum netdev_offload_xstats_type type) 7832 { 7833 ASSERT_RTNL(); 7834 7835 if (!netdev_offload_xstats_enabled(dev, type)) 7836 return -EALREADY; 7837 7838 switch (type) { 7839 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7840 netdev_offload_xstats_disable_l3(dev); 7841 return 0; 7842 } 7843 7844 WARN_ON(1); 7845 return -EINVAL; 7846 } 7847 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7848 7849 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7850 { 7851 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7852 } 7853 7854 static struct rtnl_hw_stats64 * 7855 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7856 enum netdev_offload_xstats_type type) 7857 { 7858 switch (type) { 7859 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7860 return dev->offload_xstats_l3; 7861 } 7862 7863 WARN_ON(1); 7864 return NULL; 7865 } 7866 7867 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7868 enum netdev_offload_xstats_type type) 7869 { 7870 ASSERT_RTNL(); 7871 7872 return netdev_offload_xstats_get_ptr(dev, type); 7873 } 7874 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7875 7876 struct netdev_notifier_offload_xstats_ru { 7877 bool used; 7878 }; 7879 7880 struct netdev_notifier_offload_xstats_rd { 7881 struct rtnl_hw_stats64 stats; 7882 bool used; 7883 }; 7884 7885 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 7886 const struct rtnl_hw_stats64 *src) 7887 { 7888 dest->rx_packets += src->rx_packets; 7889 dest->tx_packets += src->tx_packets; 7890 dest->rx_bytes += src->rx_bytes; 7891 dest->tx_bytes += src->tx_bytes; 7892 dest->rx_errors += src->rx_errors; 7893 dest->tx_errors += src->tx_errors; 7894 dest->rx_dropped += src->rx_dropped; 7895 dest->tx_dropped += src->tx_dropped; 7896 dest->multicast += src->multicast; 7897 } 7898 7899 static int netdev_offload_xstats_get_used(struct net_device *dev, 7900 enum netdev_offload_xstats_type type, 7901 bool *p_used, 7902 struct netlink_ext_ack *extack) 7903 { 7904 struct netdev_notifier_offload_xstats_ru report_used = {}; 7905 struct netdev_notifier_offload_xstats_info info = { 7906 .info.dev = dev, 7907 .info.extack = extack, 7908 .type = type, 7909 .report_used = &report_used, 7910 }; 7911 int rc; 7912 7913 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 7914 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 7915 &info.info); 7916 *p_used = report_used.used; 7917 return notifier_to_errno(rc); 7918 } 7919 7920 static int netdev_offload_xstats_get_stats(struct net_device *dev, 7921 enum netdev_offload_xstats_type type, 7922 struct rtnl_hw_stats64 *p_stats, 7923 bool *p_used, 7924 struct netlink_ext_ack *extack) 7925 { 7926 struct netdev_notifier_offload_xstats_rd report_delta = {}; 7927 struct netdev_notifier_offload_xstats_info info = { 7928 .info.dev = dev, 7929 .info.extack = extack, 7930 .type = type, 7931 .report_delta = &report_delta, 7932 }; 7933 struct rtnl_hw_stats64 *stats; 7934 int rc; 7935 7936 stats = netdev_offload_xstats_get_ptr(dev, type); 7937 if (WARN_ON(!stats)) 7938 return -EINVAL; 7939 7940 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 7941 &info.info); 7942 7943 /* Cache whatever we got, even if there was an error, otherwise the 7944 * successful stats retrievals would get lost. 7945 */ 7946 netdev_hw_stats64_add(stats, &report_delta.stats); 7947 7948 if (p_stats) 7949 *p_stats = *stats; 7950 *p_used = report_delta.used; 7951 7952 return notifier_to_errno(rc); 7953 } 7954 7955 int netdev_offload_xstats_get(struct net_device *dev, 7956 enum netdev_offload_xstats_type type, 7957 struct rtnl_hw_stats64 *p_stats, bool *p_used, 7958 struct netlink_ext_ack *extack) 7959 { 7960 ASSERT_RTNL(); 7961 7962 if (p_stats) 7963 return netdev_offload_xstats_get_stats(dev, type, p_stats, 7964 p_used, extack); 7965 else 7966 return netdev_offload_xstats_get_used(dev, type, p_used, 7967 extack); 7968 } 7969 EXPORT_SYMBOL(netdev_offload_xstats_get); 7970 7971 void 7972 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 7973 const struct rtnl_hw_stats64 *stats) 7974 { 7975 report_delta->used = true; 7976 netdev_hw_stats64_add(&report_delta->stats, stats); 7977 } 7978 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 7979 7980 void 7981 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 7982 { 7983 report_used->used = true; 7984 } 7985 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 7986 7987 void netdev_offload_xstats_push_delta(struct net_device *dev, 7988 enum netdev_offload_xstats_type type, 7989 const struct rtnl_hw_stats64 *p_stats) 7990 { 7991 struct rtnl_hw_stats64 *stats; 7992 7993 ASSERT_RTNL(); 7994 7995 stats = netdev_offload_xstats_get_ptr(dev, type); 7996 if (WARN_ON(!stats)) 7997 return; 7998 7999 netdev_hw_stats64_add(stats, p_stats); 8000 } 8001 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8002 8003 /** 8004 * netdev_get_xmit_slave - Get the xmit slave of master device 8005 * @dev: device 8006 * @skb: The packet 8007 * @all_slaves: assume all the slaves are active 8008 * 8009 * The reference counters are not incremented so the caller must be 8010 * careful with locks. The caller must hold RCU lock. 8011 * %NULL is returned if no slave is found. 8012 */ 8013 8014 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8015 struct sk_buff *skb, 8016 bool all_slaves) 8017 { 8018 const struct net_device_ops *ops = dev->netdev_ops; 8019 8020 if (!ops->ndo_get_xmit_slave) 8021 return NULL; 8022 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8023 } 8024 EXPORT_SYMBOL(netdev_get_xmit_slave); 8025 8026 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8027 struct sock *sk) 8028 { 8029 const struct net_device_ops *ops = dev->netdev_ops; 8030 8031 if (!ops->ndo_sk_get_lower_dev) 8032 return NULL; 8033 return ops->ndo_sk_get_lower_dev(dev, sk); 8034 } 8035 8036 /** 8037 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8038 * @dev: device 8039 * @sk: the socket 8040 * 8041 * %NULL is returned if no lower device is found. 8042 */ 8043 8044 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8045 struct sock *sk) 8046 { 8047 struct net_device *lower; 8048 8049 lower = netdev_sk_get_lower_dev(dev, sk); 8050 while (lower) { 8051 dev = lower; 8052 lower = netdev_sk_get_lower_dev(dev, sk); 8053 } 8054 8055 return dev; 8056 } 8057 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8058 8059 static void netdev_adjacent_add_links(struct net_device *dev) 8060 { 8061 struct netdev_adjacent *iter; 8062 8063 struct net *net = dev_net(dev); 8064 8065 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8066 if (!net_eq(net, dev_net(iter->dev))) 8067 continue; 8068 netdev_adjacent_sysfs_add(iter->dev, dev, 8069 &iter->dev->adj_list.lower); 8070 netdev_adjacent_sysfs_add(dev, iter->dev, 8071 &dev->adj_list.upper); 8072 } 8073 8074 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8075 if (!net_eq(net, dev_net(iter->dev))) 8076 continue; 8077 netdev_adjacent_sysfs_add(iter->dev, dev, 8078 &iter->dev->adj_list.upper); 8079 netdev_adjacent_sysfs_add(dev, iter->dev, 8080 &dev->adj_list.lower); 8081 } 8082 } 8083 8084 static void netdev_adjacent_del_links(struct net_device *dev) 8085 { 8086 struct netdev_adjacent *iter; 8087 8088 struct net *net = dev_net(dev); 8089 8090 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8091 if (!net_eq(net, dev_net(iter->dev))) 8092 continue; 8093 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8094 &iter->dev->adj_list.lower); 8095 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8096 &dev->adj_list.upper); 8097 } 8098 8099 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8100 if (!net_eq(net, dev_net(iter->dev))) 8101 continue; 8102 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8103 &iter->dev->adj_list.upper); 8104 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8105 &dev->adj_list.lower); 8106 } 8107 } 8108 8109 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8110 { 8111 struct netdev_adjacent *iter; 8112 8113 struct net *net = dev_net(dev); 8114 8115 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8116 if (!net_eq(net, dev_net(iter->dev))) 8117 continue; 8118 netdev_adjacent_sysfs_del(iter->dev, oldname, 8119 &iter->dev->adj_list.lower); 8120 netdev_adjacent_sysfs_add(iter->dev, dev, 8121 &iter->dev->adj_list.lower); 8122 } 8123 8124 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8125 if (!net_eq(net, dev_net(iter->dev))) 8126 continue; 8127 netdev_adjacent_sysfs_del(iter->dev, oldname, 8128 &iter->dev->adj_list.upper); 8129 netdev_adjacent_sysfs_add(iter->dev, dev, 8130 &iter->dev->adj_list.upper); 8131 } 8132 } 8133 8134 void *netdev_lower_dev_get_private(struct net_device *dev, 8135 struct net_device *lower_dev) 8136 { 8137 struct netdev_adjacent *lower; 8138 8139 if (!lower_dev) 8140 return NULL; 8141 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8142 if (!lower) 8143 return NULL; 8144 8145 return lower->private; 8146 } 8147 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8148 8149 8150 /** 8151 * netdev_lower_state_changed - Dispatch event about lower device state change 8152 * @lower_dev: device 8153 * @lower_state_info: state to dispatch 8154 * 8155 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8156 * The caller must hold the RTNL lock. 8157 */ 8158 void netdev_lower_state_changed(struct net_device *lower_dev, 8159 void *lower_state_info) 8160 { 8161 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8162 .info.dev = lower_dev, 8163 }; 8164 8165 ASSERT_RTNL(); 8166 changelowerstate_info.lower_state_info = lower_state_info; 8167 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8168 &changelowerstate_info.info); 8169 } 8170 EXPORT_SYMBOL(netdev_lower_state_changed); 8171 8172 static void dev_change_rx_flags(struct net_device *dev, int flags) 8173 { 8174 const struct net_device_ops *ops = dev->netdev_ops; 8175 8176 if (ops->ndo_change_rx_flags) 8177 ops->ndo_change_rx_flags(dev, flags); 8178 } 8179 8180 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8181 { 8182 unsigned int old_flags = dev->flags; 8183 kuid_t uid; 8184 kgid_t gid; 8185 8186 ASSERT_RTNL(); 8187 8188 dev->flags |= IFF_PROMISC; 8189 dev->promiscuity += inc; 8190 if (dev->promiscuity == 0) { 8191 /* 8192 * Avoid overflow. 8193 * If inc causes overflow, untouch promisc and return error. 8194 */ 8195 if (inc < 0) 8196 dev->flags &= ~IFF_PROMISC; 8197 else { 8198 dev->promiscuity -= inc; 8199 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8200 return -EOVERFLOW; 8201 } 8202 } 8203 if (dev->flags != old_flags) { 8204 pr_info("device %s %s promiscuous mode\n", 8205 dev->name, 8206 dev->flags & IFF_PROMISC ? "entered" : "left"); 8207 if (audit_enabled) { 8208 current_uid_gid(&uid, &gid); 8209 audit_log(audit_context(), GFP_ATOMIC, 8210 AUDIT_ANOM_PROMISCUOUS, 8211 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8212 dev->name, (dev->flags & IFF_PROMISC), 8213 (old_flags & IFF_PROMISC), 8214 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8215 from_kuid(&init_user_ns, uid), 8216 from_kgid(&init_user_ns, gid), 8217 audit_get_sessionid(current)); 8218 } 8219 8220 dev_change_rx_flags(dev, IFF_PROMISC); 8221 } 8222 if (notify) 8223 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8224 return 0; 8225 } 8226 8227 /** 8228 * dev_set_promiscuity - update promiscuity count on a device 8229 * @dev: device 8230 * @inc: modifier 8231 * 8232 * Add or remove promiscuity from a device. While the count in the device 8233 * remains above zero the interface remains promiscuous. Once it hits zero 8234 * the device reverts back to normal filtering operation. A negative inc 8235 * value is used to drop promiscuity on the device. 8236 * Return 0 if successful or a negative errno code on error. 8237 */ 8238 int dev_set_promiscuity(struct net_device *dev, int inc) 8239 { 8240 unsigned int old_flags = dev->flags; 8241 int err; 8242 8243 err = __dev_set_promiscuity(dev, inc, true); 8244 if (err < 0) 8245 return err; 8246 if (dev->flags != old_flags) 8247 dev_set_rx_mode(dev); 8248 return err; 8249 } 8250 EXPORT_SYMBOL(dev_set_promiscuity); 8251 8252 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8253 { 8254 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8255 8256 ASSERT_RTNL(); 8257 8258 dev->flags |= IFF_ALLMULTI; 8259 dev->allmulti += inc; 8260 if (dev->allmulti == 0) { 8261 /* 8262 * Avoid overflow. 8263 * If inc causes overflow, untouch allmulti and return error. 8264 */ 8265 if (inc < 0) 8266 dev->flags &= ~IFF_ALLMULTI; 8267 else { 8268 dev->allmulti -= inc; 8269 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8270 return -EOVERFLOW; 8271 } 8272 } 8273 if (dev->flags ^ old_flags) { 8274 dev_change_rx_flags(dev, IFF_ALLMULTI); 8275 dev_set_rx_mode(dev); 8276 if (notify) 8277 __dev_notify_flags(dev, old_flags, 8278 dev->gflags ^ old_gflags); 8279 } 8280 return 0; 8281 } 8282 8283 /** 8284 * dev_set_allmulti - update allmulti count on a device 8285 * @dev: device 8286 * @inc: modifier 8287 * 8288 * Add or remove reception of all multicast frames to a device. While the 8289 * count in the device remains above zero the interface remains listening 8290 * to all interfaces. Once it hits zero the device reverts back to normal 8291 * filtering operation. A negative @inc value is used to drop the counter 8292 * when releasing a resource needing all multicasts. 8293 * Return 0 if successful or a negative errno code on error. 8294 */ 8295 8296 int dev_set_allmulti(struct net_device *dev, int inc) 8297 { 8298 return __dev_set_allmulti(dev, inc, true); 8299 } 8300 EXPORT_SYMBOL(dev_set_allmulti); 8301 8302 /* 8303 * Upload unicast and multicast address lists to device and 8304 * configure RX filtering. When the device doesn't support unicast 8305 * filtering it is put in promiscuous mode while unicast addresses 8306 * are present. 8307 */ 8308 void __dev_set_rx_mode(struct net_device *dev) 8309 { 8310 const struct net_device_ops *ops = dev->netdev_ops; 8311 8312 /* dev_open will call this function so the list will stay sane. */ 8313 if (!(dev->flags&IFF_UP)) 8314 return; 8315 8316 if (!netif_device_present(dev)) 8317 return; 8318 8319 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8320 /* Unicast addresses changes may only happen under the rtnl, 8321 * therefore calling __dev_set_promiscuity here is safe. 8322 */ 8323 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8324 __dev_set_promiscuity(dev, 1, false); 8325 dev->uc_promisc = true; 8326 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8327 __dev_set_promiscuity(dev, -1, false); 8328 dev->uc_promisc = false; 8329 } 8330 } 8331 8332 if (ops->ndo_set_rx_mode) 8333 ops->ndo_set_rx_mode(dev); 8334 } 8335 8336 void dev_set_rx_mode(struct net_device *dev) 8337 { 8338 netif_addr_lock_bh(dev); 8339 __dev_set_rx_mode(dev); 8340 netif_addr_unlock_bh(dev); 8341 } 8342 8343 /** 8344 * dev_get_flags - get flags reported to userspace 8345 * @dev: device 8346 * 8347 * Get the combination of flag bits exported through APIs to userspace. 8348 */ 8349 unsigned int dev_get_flags(const struct net_device *dev) 8350 { 8351 unsigned int flags; 8352 8353 flags = (dev->flags & ~(IFF_PROMISC | 8354 IFF_ALLMULTI | 8355 IFF_RUNNING | 8356 IFF_LOWER_UP | 8357 IFF_DORMANT)) | 8358 (dev->gflags & (IFF_PROMISC | 8359 IFF_ALLMULTI)); 8360 8361 if (netif_running(dev)) { 8362 if (netif_oper_up(dev)) 8363 flags |= IFF_RUNNING; 8364 if (netif_carrier_ok(dev)) 8365 flags |= IFF_LOWER_UP; 8366 if (netif_dormant(dev)) 8367 flags |= IFF_DORMANT; 8368 } 8369 8370 return flags; 8371 } 8372 EXPORT_SYMBOL(dev_get_flags); 8373 8374 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8375 struct netlink_ext_ack *extack) 8376 { 8377 unsigned int old_flags = dev->flags; 8378 int ret; 8379 8380 ASSERT_RTNL(); 8381 8382 /* 8383 * Set the flags on our device. 8384 */ 8385 8386 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8387 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8388 IFF_AUTOMEDIA)) | 8389 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8390 IFF_ALLMULTI)); 8391 8392 /* 8393 * Load in the correct multicast list now the flags have changed. 8394 */ 8395 8396 if ((old_flags ^ flags) & IFF_MULTICAST) 8397 dev_change_rx_flags(dev, IFF_MULTICAST); 8398 8399 dev_set_rx_mode(dev); 8400 8401 /* 8402 * Have we downed the interface. We handle IFF_UP ourselves 8403 * according to user attempts to set it, rather than blindly 8404 * setting it. 8405 */ 8406 8407 ret = 0; 8408 if ((old_flags ^ flags) & IFF_UP) { 8409 if (old_flags & IFF_UP) 8410 __dev_close(dev); 8411 else 8412 ret = __dev_open(dev, extack); 8413 } 8414 8415 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8416 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8417 unsigned int old_flags = dev->flags; 8418 8419 dev->gflags ^= IFF_PROMISC; 8420 8421 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8422 if (dev->flags != old_flags) 8423 dev_set_rx_mode(dev); 8424 } 8425 8426 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8427 * is important. Some (broken) drivers set IFF_PROMISC, when 8428 * IFF_ALLMULTI is requested not asking us and not reporting. 8429 */ 8430 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8431 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8432 8433 dev->gflags ^= IFF_ALLMULTI; 8434 __dev_set_allmulti(dev, inc, false); 8435 } 8436 8437 return ret; 8438 } 8439 8440 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8441 unsigned int gchanges) 8442 { 8443 unsigned int changes = dev->flags ^ old_flags; 8444 8445 if (gchanges) 8446 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8447 8448 if (changes & IFF_UP) { 8449 if (dev->flags & IFF_UP) 8450 call_netdevice_notifiers(NETDEV_UP, dev); 8451 else 8452 call_netdevice_notifiers(NETDEV_DOWN, dev); 8453 } 8454 8455 if (dev->flags & IFF_UP && 8456 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8457 struct netdev_notifier_change_info change_info = { 8458 .info = { 8459 .dev = dev, 8460 }, 8461 .flags_changed = changes, 8462 }; 8463 8464 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8465 } 8466 } 8467 8468 /** 8469 * dev_change_flags - change device settings 8470 * @dev: device 8471 * @flags: device state flags 8472 * @extack: netlink extended ack 8473 * 8474 * Change settings on device based state flags. The flags are 8475 * in the userspace exported format. 8476 */ 8477 int dev_change_flags(struct net_device *dev, unsigned int flags, 8478 struct netlink_ext_ack *extack) 8479 { 8480 int ret; 8481 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8482 8483 ret = __dev_change_flags(dev, flags, extack); 8484 if (ret < 0) 8485 return ret; 8486 8487 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8488 __dev_notify_flags(dev, old_flags, changes); 8489 return ret; 8490 } 8491 EXPORT_SYMBOL(dev_change_flags); 8492 8493 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8494 { 8495 const struct net_device_ops *ops = dev->netdev_ops; 8496 8497 if (ops->ndo_change_mtu) 8498 return ops->ndo_change_mtu(dev, new_mtu); 8499 8500 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8501 WRITE_ONCE(dev->mtu, new_mtu); 8502 return 0; 8503 } 8504 EXPORT_SYMBOL(__dev_set_mtu); 8505 8506 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8507 struct netlink_ext_ack *extack) 8508 { 8509 /* MTU must be positive, and in range */ 8510 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8511 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8512 return -EINVAL; 8513 } 8514 8515 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8516 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8517 return -EINVAL; 8518 } 8519 return 0; 8520 } 8521 8522 /** 8523 * dev_set_mtu_ext - Change maximum transfer unit 8524 * @dev: device 8525 * @new_mtu: new transfer unit 8526 * @extack: netlink extended ack 8527 * 8528 * Change the maximum transfer size of the network device. 8529 */ 8530 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8531 struct netlink_ext_ack *extack) 8532 { 8533 int err, orig_mtu; 8534 8535 if (new_mtu == dev->mtu) 8536 return 0; 8537 8538 err = dev_validate_mtu(dev, new_mtu, extack); 8539 if (err) 8540 return err; 8541 8542 if (!netif_device_present(dev)) 8543 return -ENODEV; 8544 8545 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8546 err = notifier_to_errno(err); 8547 if (err) 8548 return err; 8549 8550 orig_mtu = dev->mtu; 8551 err = __dev_set_mtu(dev, new_mtu); 8552 8553 if (!err) { 8554 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8555 orig_mtu); 8556 err = notifier_to_errno(err); 8557 if (err) { 8558 /* setting mtu back and notifying everyone again, 8559 * so that they have a chance to revert changes. 8560 */ 8561 __dev_set_mtu(dev, orig_mtu); 8562 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8563 new_mtu); 8564 } 8565 } 8566 return err; 8567 } 8568 8569 int dev_set_mtu(struct net_device *dev, int new_mtu) 8570 { 8571 struct netlink_ext_ack extack; 8572 int err; 8573 8574 memset(&extack, 0, sizeof(extack)); 8575 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8576 if (err && extack._msg) 8577 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8578 return err; 8579 } 8580 EXPORT_SYMBOL(dev_set_mtu); 8581 8582 /** 8583 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8584 * @dev: device 8585 * @new_len: new tx queue length 8586 */ 8587 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8588 { 8589 unsigned int orig_len = dev->tx_queue_len; 8590 int res; 8591 8592 if (new_len != (unsigned int)new_len) 8593 return -ERANGE; 8594 8595 if (new_len != orig_len) { 8596 dev->tx_queue_len = new_len; 8597 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8598 res = notifier_to_errno(res); 8599 if (res) 8600 goto err_rollback; 8601 res = dev_qdisc_change_tx_queue_len(dev); 8602 if (res) 8603 goto err_rollback; 8604 } 8605 8606 return 0; 8607 8608 err_rollback: 8609 netdev_err(dev, "refused to change device tx_queue_len\n"); 8610 dev->tx_queue_len = orig_len; 8611 return res; 8612 } 8613 8614 /** 8615 * dev_set_group - Change group this device belongs to 8616 * @dev: device 8617 * @new_group: group this device should belong to 8618 */ 8619 void dev_set_group(struct net_device *dev, int new_group) 8620 { 8621 dev->group = new_group; 8622 } 8623 EXPORT_SYMBOL(dev_set_group); 8624 8625 /** 8626 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8627 * @dev: device 8628 * @addr: new address 8629 * @extack: netlink extended ack 8630 */ 8631 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8632 struct netlink_ext_ack *extack) 8633 { 8634 struct netdev_notifier_pre_changeaddr_info info = { 8635 .info.dev = dev, 8636 .info.extack = extack, 8637 .dev_addr = addr, 8638 }; 8639 int rc; 8640 8641 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8642 return notifier_to_errno(rc); 8643 } 8644 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8645 8646 /** 8647 * dev_set_mac_address - Change Media Access Control Address 8648 * @dev: device 8649 * @sa: new address 8650 * @extack: netlink extended ack 8651 * 8652 * Change the hardware (MAC) address of the device 8653 */ 8654 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8655 struct netlink_ext_ack *extack) 8656 { 8657 const struct net_device_ops *ops = dev->netdev_ops; 8658 int err; 8659 8660 if (!ops->ndo_set_mac_address) 8661 return -EOPNOTSUPP; 8662 if (sa->sa_family != dev->type) 8663 return -EINVAL; 8664 if (!netif_device_present(dev)) 8665 return -ENODEV; 8666 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8667 if (err) 8668 return err; 8669 err = ops->ndo_set_mac_address(dev, sa); 8670 if (err) 8671 return err; 8672 dev->addr_assign_type = NET_ADDR_SET; 8673 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8674 add_device_randomness(dev->dev_addr, dev->addr_len); 8675 return 0; 8676 } 8677 EXPORT_SYMBOL(dev_set_mac_address); 8678 8679 static DECLARE_RWSEM(dev_addr_sem); 8680 8681 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8682 struct netlink_ext_ack *extack) 8683 { 8684 int ret; 8685 8686 down_write(&dev_addr_sem); 8687 ret = dev_set_mac_address(dev, sa, extack); 8688 up_write(&dev_addr_sem); 8689 return ret; 8690 } 8691 EXPORT_SYMBOL(dev_set_mac_address_user); 8692 8693 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8694 { 8695 size_t size = sizeof(sa->sa_data); 8696 struct net_device *dev; 8697 int ret = 0; 8698 8699 down_read(&dev_addr_sem); 8700 rcu_read_lock(); 8701 8702 dev = dev_get_by_name_rcu(net, dev_name); 8703 if (!dev) { 8704 ret = -ENODEV; 8705 goto unlock; 8706 } 8707 if (!dev->addr_len) 8708 memset(sa->sa_data, 0, size); 8709 else 8710 memcpy(sa->sa_data, dev->dev_addr, 8711 min_t(size_t, size, dev->addr_len)); 8712 sa->sa_family = dev->type; 8713 8714 unlock: 8715 rcu_read_unlock(); 8716 up_read(&dev_addr_sem); 8717 return ret; 8718 } 8719 EXPORT_SYMBOL(dev_get_mac_address); 8720 8721 /** 8722 * dev_change_carrier - Change device carrier 8723 * @dev: device 8724 * @new_carrier: new value 8725 * 8726 * Change device carrier 8727 */ 8728 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8729 { 8730 const struct net_device_ops *ops = dev->netdev_ops; 8731 8732 if (!ops->ndo_change_carrier) 8733 return -EOPNOTSUPP; 8734 if (!netif_device_present(dev)) 8735 return -ENODEV; 8736 return ops->ndo_change_carrier(dev, new_carrier); 8737 } 8738 EXPORT_SYMBOL(dev_change_carrier); 8739 8740 /** 8741 * dev_get_phys_port_id - Get device physical port ID 8742 * @dev: device 8743 * @ppid: port ID 8744 * 8745 * Get device physical port ID 8746 */ 8747 int dev_get_phys_port_id(struct net_device *dev, 8748 struct netdev_phys_item_id *ppid) 8749 { 8750 const struct net_device_ops *ops = dev->netdev_ops; 8751 8752 if (!ops->ndo_get_phys_port_id) 8753 return -EOPNOTSUPP; 8754 return ops->ndo_get_phys_port_id(dev, ppid); 8755 } 8756 EXPORT_SYMBOL(dev_get_phys_port_id); 8757 8758 /** 8759 * dev_get_phys_port_name - Get device physical port name 8760 * @dev: device 8761 * @name: port name 8762 * @len: limit of bytes to copy to name 8763 * 8764 * Get device physical port name 8765 */ 8766 int dev_get_phys_port_name(struct net_device *dev, 8767 char *name, size_t len) 8768 { 8769 const struct net_device_ops *ops = dev->netdev_ops; 8770 int err; 8771 8772 if (ops->ndo_get_phys_port_name) { 8773 err = ops->ndo_get_phys_port_name(dev, name, len); 8774 if (err != -EOPNOTSUPP) 8775 return err; 8776 } 8777 return devlink_compat_phys_port_name_get(dev, name, len); 8778 } 8779 EXPORT_SYMBOL(dev_get_phys_port_name); 8780 8781 /** 8782 * dev_get_port_parent_id - Get the device's port parent identifier 8783 * @dev: network device 8784 * @ppid: pointer to a storage for the port's parent identifier 8785 * @recurse: allow/disallow recursion to lower devices 8786 * 8787 * Get the devices's port parent identifier 8788 */ 8789 int dev_get_port_parent_id(struct net_device *dev, 8790 struct netdev_phys_item_id *ppid, 8791 bool recurse) 8792 { 8793 const struct net_device_ops *ops = dev->netdev_ops; 8794 struct netdev_phys_item_id first = { }; 8795 struct net_device *lower_dev; 8796 struct list_head *iter; 8797 int err; 8798 8799 if (ops->ndo_get_port_parent_id) { 8800 err = ops->ndo_get_port_parent_id(dev, ppid); 8801 if (err != -EOPNOTSUPP) 8802 return err; 8803 } 8804 8805 err = devlink_compat_switch_id_get(dev, ppid); 8806 if (!recurse || err != -EOPNOTSUPP) 8807 return err; 8808 8809 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8810 err = dev_get_port_parent_id(lower_dev, ppid, true); 8811 if (err) 8812 break; 8813 if (!first.id_len) 8814 first = *ppid; 8815 else if (memcmp(&first, ppid, sizeof(*ppid))) 8816 return -EOPNOTSUPP; 8817 } 8818 8819 return err; 8820 } 8821 EXPORT_SYMBOL(dev_get_port_parent_id); 8822 8823 /** 8824 * netdev_port_same_parent_id - Indicate if two network devices have 8825 * the same port parent identifier 8826 * @a: first network device 8827 * @b: second network device 8828 */ 8829 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8830 { 8831 struct netdev_phys_item_id a_id = { }; 8832 struct netdev_phys_item_id b_id = { }; 8833 8834 if (dev_get_port_parent_id(a, &a_id, true) || 8835 dev_get_port_parent_id(b, &b_id, true)) 8836 return false; 8837 8838 return netdev_phys_item_id_same(&a_id, &b_id); 8839 } 8840 EXPORT_SYMBOL(netdev_port_same_parent_id); 8841 8842 /** 8843 * dev_change_proto_down - set carrier according to proto_down. 8844 * 8845 * @dev: device 8846 * @proto_down: new value 8847 */ 8848 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8849 { 8850 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8851 return -EOPNOTSUPP; 8852 if (!netif_device_present(dev)) 8853 return -ENODEV; 8854 if (proto_down) 8855 netif_carrier_off(dev); 8856 else 8857 netif_carrier_on(dev); 8858 dev->proto_down = proto_down; 8859 return 0; 8860 } 8861 EXPORT_SYMBOL(dev_change_proto_down); 8862 8863 /** 8864 * dev_change_proto_down_reason - proto down reason 8865 * 8866 * @dev: device 8867 * @mask: proto down mask 8868 * @value: proto down value 8869 */ 8870 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8871 u32 value) 8872 { 8873 int b; 8874 8875 if (!mask) { 8876 dev->proto_down_reason = value; 8877 } else { 8878 for_each_set_bit(b, &mask, 32) { 8879 if (value & (1 << b)) 8880 dev->proto_down_reason |= BIT(b); 8881 else 8882 dev->proto_down_reason &= ~BIT(b); 8883 } 8884 } 8885 } 8886 EXPORT_SYMBOL(dev_change_proto_down_reason); 8887 8888 struct bpf_xdp_link { 8889 struct bpf_link link; 8890 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8891 int flags; 8892 }; 8893 8894 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8895 { 8896 if (flags & XDP_FLAGS_HW_MODE) 8897 return XDP_MODE_HW; 8898 if (flags & XDP_FLAGS_DRV_MODE) 8899 return XDP_MODE_DRV; 8900 if (flags & XDP_FLAGS_SKB_MODE) 8901 return XDP_MODE_SKB; 8902 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8903 } 8904 8905 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8906 { 8907 switch (mode) { 8908 case XDP_MODE_SKB: 8909 return generic_xdp_install; 8910 case XDP_MODE_DRV: 8911 case XDP_MODE_HW: 8912 return dev->netdev_ops->ndo_bpf; 8913 default: 8914 return NULL; 8915 } 8916 } 8917 8918 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8919 enum bpf_xdp_mode mode) 8920 { 8921 return dev->xdp_state[mode].link; 8922 } 8923 8924 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 8925 enum bpf_xdp_mode mode) 8926 { 8927 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 8928 8929 if (link) 8930 return link->link.prog; 8931 return dev->xdp_state[mode].prog; 8932 } 8933 8934 u8 dev_xdp_prog_count(struct net_device *dev) 8935 { 8936 u8 count = 0; 8937 int i; 8938 8939 for (i = 0; i < __MAX_XDP_MODE; i++) 8940 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 8941 count++; 8942 return count; 8943 } 8944 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 8945 8946 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 8947 { 8948 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 8949 8950 return prog ? prog->aux->id : 0; 8951 } 8952 8953 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 8954 struct bpf_xdp_link *link) 8955 { 8956 dev->xdp_state[mode].link = link; 8957 dev->xdp_state[mode].prog = NULL; 8958 } 8959 8960 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 8961 struct bpf_prog *prog) 8962 { 8963 dev->xdp_state[mode].link = NULL; 8964 dev->xdp_state[mode].prog = prog; 8965 } 8966 8967 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 8968 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 8969 u32 flags, struct bpf_prog *prog) 8970 { 8971 struct netdev_bpf xdp; 8972 int err; 8973 8974 memset(&xdp, 0, sizeof(xdp)); 8975 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 8976 xdp.extack = extack; 8977 xdp.flags = flags; 8978 xdp.prog = prog; 8979 8980 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 8981 * "moved" into driver), so they don't increment it on their own, but 8982 * they do decrement refcnt when program is detached or replaced. 8983 * Given net_device also owns link/prog, we need to bump refcnt here 8984 * to prevent drivers from underflowing it. 8985 */ 8986 if (prog) 8987 bpf_prog_inc(prog); 8988 err = bpf_op(dev, &xdp); 8989 if (err) { 8990 if (prog) 8991 bpf_prog_put(prog); 8992 return err; 8993 } 8994 8995 if (mode != XDP_MODE_HW) 8996 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 8997 8998 return 0; 8999 } 9000 9001 static void dev_xdp_uninstall(struct net_device *dev) 9002 { 9003 struct bpf_xdp_link *link; 9004 struct bpf_prog *prog; 9005 enum bpf_xdp_mode mode; 9006 bpf_op_t bpf_op; 9007 9008 ASSERT_RTNL(); 9009 9010 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9011 prog = dev_xdp_prog(dev, mode); 9012 if (!prog) 9013 continue; 9014 9015 bpf_op = dev_xdp_bpf_op(dev, mode); 9016 if (!bpf_op) 9017 continue; 9018 9019 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9020 9021 /* auto-detach link from net device */ 9022 link = dev_xdp_link(dev, mode); 9023 if (link) 9024 link->dev = NULL; 9025 else 9026 bpf_prog_put(prog); 9027 9028 dev_xdp_set_link(dev, mode, NULL); 9029 } 9030 } 9031 9032 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9033 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9034 struct bpf_prog *old_prog, u32 flags) 9035 { 9036 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9037 struct bpf_prog *cur_prog; 9038 struct net_device *upper; 9039 struct list_head *iter; 9040 enum bpf_xdp_mode mode; 9041 bpf_op_t bpf_op; 9042 int err; 9043 9044 ASSERT_RTNL(); 9045 9046 /* either link or prog attachment, never both */ 9047 if (link && (new_prog || old_prog)) 9048 return -EINVAL; 9049 /* link supports only XDP mode flags */ 9050 if (link && (flags & ~XDP_FLAGS_MODES)) { 9051 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9052 return -EINVAL; 9053 } 9054 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9055 if (num_modes > 1) { 9056 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9057 return -EINVAL; 9058 } 9059 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9060 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9061 NL_SET_ERR_MSG(extack, 9062 "More than one program loaded, unset mode is ambiguous"); 9063 return -EINVAL; 9064 } 9065 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9066 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9067 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9068 return -EINVAL; 9069 } 9070 9071 mode = dev_xdp_mode(dev, flags); 9072 /* can't replace attached link */ 9073 if (dev_xdp_link(dev, mode)) { 9074 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9075 return -EBUSY; 9076 } 9077 9078 /* don't allow if an upper device already has a program */ 9079 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9080 if (dev_xdp_prog_count(upper) > 0) { 9081 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9082 return -EEXIST; 9083 } 9084 } 9085 9086 cur_prog = dev_xdp_prog(dev, mode); 9087 /* can't replace attached prog with link */ 9088 if (link && cur_prog) { 9089 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9090 return -EBUSY; 9091 } 9092 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9093 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9094 return -EEXIST; 9095 } 9096 9097 /* put effective new program into new_prog */ 9098 if (link) 9099 new_prog = link->link.prog; 9100 9101 if (new_prog) { 9102 bool offload = mode == XDP_MODE_HW; 9103 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9104 ? XDP_MODE_DRV : XDP_MODE_SKB; 9105 9106 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9107 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9108 return -EBUSY; 9109 } 9110 if (!offload && dev_xdp_prog(dev, other_mode)) { 9111 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9112 return -EEXIST; 9113 } 9114 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9115 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9116 return -EINVAL; 9117 } 9118 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9119 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9120 return -EINVAL; 9121 } 9122 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9123 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9124 return -EINVAL; 9125 } 9126 } 9127 9128 /* don't call drivers if the effective program didn't change */ 9129 if (new_prog != cur_prog) { 9130 bpf_op = dev_xdp_bpf_op(dev, mode); 9131 if (!bpf_op) { 9132 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9133 return -EOPNOTSUPP; 9134 } 9135 9136 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9137 if (err) 9138 return err; 9139 } 9140 9141 if (link) 9142 dev_xdp_set_link(dev, mode, link); 9143 else 9144 dev_xdp_set_prog(dev, mode, new_prog); 9145 if (cur_prog) 9146 bpf_prog_put(cur_prog); 9147 9148 return 0; 9149 } 9150 9151 static int dev_xdp_attach_link(struct net_device *dev, 9152 struct netlink_ext_ack *extack, 9153 struct bpf_xdp_link *link) 9154 { 9155 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9156 } 9157 9158 static int dev_xdp_detach_link(struct net_device *dev, 9159 struct netlink_ext_ack *extack, 9160 struct bpf_xdp_link *link) 9161 { 9162 enum bpf_xdp_mode mode; 9163 bpf_op_t bpf_op; 9164 9165 ASSERT_RTNL(); 9166 9167 mode = dev_xdp_mode(dev, link->flags); 9168 if (dev_xdp_link(dev, mode) != link) 9169 return -EINVAL; 9170 9171 bpf_op = dev_xdp_bpf_op(dev, mode); 9172 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9173 dev_xdp_set_link(dev, mode, NULL); 9174 return 0; 9175 } 9176 9177 static void bpf_xdp_link_release(struct bpf_link *link) 9178 { 9179 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9180 9181 rtnl_lock(); 9182 9183 /* if racing with net_device's tear down, xdp_link->dev might be 9184 * already NULL, in which case link was already auto-detached 9185 */ 9186 if (xdp_link->dev) { 9187 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9188 xdp_link->dev = NULL; 9189 } 9190 9191 rtnl_unlock(); 9192 } 9193 9194 static int bpf_xdp_link_detach(struct bpf_link *link) 9195 { 9196 bpf_xdp_link_release(link); 9197 return 0; 9198 } 9199 9200 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9201 { 9202 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9203 9204 kfree(xdp_link); 9205 } 9206 9207 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9208 struct seq_file *seq) 9209 { 9210 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9211 u32 ifindex = 0; 9212 9213 rtnl_lock(); 9214 if (xdp_link->dev) 9215 ifindex = xdp_link->dev->ifindex; 9216 rtnl_unlock(); 9217 9218 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9219 } 9220 9221 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9222 struct bpf_link_info *info) 9223 { 9224 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9225 u32 ifindex = 0; 9226 9227 rtnl_lock(); 9228 if (xdp_link->dev) 9229 ifindex = xdp_link->dev->ifindex; 9230 rtnl_unlock(); 9231 9232 info->xdp.ifindex = ifindex; 9233 return 0; 9234 } 9235 9236 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9237 struct bpf_prog *old_prog) 9238 { 9239 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9240 enum bpf_xdp_mode mode; 9241 bpf_op_t bpf_op; 9242 int err = 0; 9243 9244 rtnl_lock(); 9245 9246 /* link might have been auto-released already, so fail */ 9247 if (!xdp_link->dev) { 9248 err = -ENOLINK; 9249 goto out_unlock; 9250 } 9251 9252 if (old_prog && link->prog != old_prog) { 9253 err = -EPERM; 9254 goto out_unlock; 9255 } 9256 old_prog = link->prog; 9257 if (old_prog->type != new_prog->type || 9258 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9259 err = -EINVAL; 9260 goto out_unlock; 9261 } 9262 9263 if (old_prog == new_prog) { 9264 /* no-op, don't disturb drivers */ 9265 bpf_prog_put(new_prog); 9266 goto out_unlock; 9267 } 9268 9269 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9270 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9271 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9272 xdp_link->flags, new_prog); 9273 if (err) 9274 goto out_unlock; 9275 9276 old_prog = xchg(&link->prog, new_prog); 9277 bpf_prog_put(old_prog); 9278 9279 out_unlock: 9280 rtnl_unlock(); 9281 return err; 9282 } 9283 9284 static const struct bpf_link_ops bpf_xdp_link_lops = { 9285 .release = bpf_xdp_link_release, 9286 .dealloc = bpf_xdp_link_dealloc, 9287 .detach = bpf_xdp_link_detach, 9288 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9289 .fill_link_info = bpf_xdp_link_fill_link_info, 9290 .update_prog = bpf_xdp_link_update, 9291 }; 9292 9293 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9294 { 9295 struct net *net = current->nsproxy->net_ns; 9296 struct bpf_link_primer link_primer; 9297 struct bpf_xdp_link *link; 9298 struct net_device *dev; 9299 int err, fd; 9300 9301 rtnl_lock(); 9302 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9303 if (!dev) { 9304 rtnl_unlock(); 9305 return -EINVAL; 9306 } 9307 9308 link = kzalloc(sizeof(*link), GFP_USER); 9309 if (!link) { 9310 err = -ENOMEM; 9311 goto unlock; 9312 } 9313 9314 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9315 link->dev = dev; 9316 link->flags = attr->link_create.flags; 9317 9318 err = bpf_link_prime(&link->link, &link_primer); 9319 if (err) { 9320 kfree(link); 9321 goto unlock; 9322 } 9323 9324 err = dev_xdp_attach_link(dev, NULL, link); 9325 rtnl_unlock(); 9326 9327 if (err) { 9328 link->dev = NULL; 9329 bpf_link_cleanup(&link_primer); 9330 goto out_put_dev; 9331 } 9332 9333 fd = bpf_link_settle(&link_primer); 9334 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9335 dev_put(dev); 9336 return fd; 9337 9338 unlock: 9339 rtnl_unlock(); 9340 9341 out_put_dev: 9342 dev_put(dev); 9343 return err; 9344 } 9345 9346 /** 9347 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9348 * @dev: device 9349 * @extack: netlink extended ack 9350 * @fd: new program fd or negative value to clear 9351 * @expected_fd: old program fd that userspace expects to replace or clear 9352 * @flags: xdp-related flags 9353 * 9354 * Set or clear a bpf program for a device 9355 */ 9356 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9357 int fd, int expected_fd, u32 flags) 9358 { 9359 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9360 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9361 int err; 9362 9363 ASSERT_RTNL(); 9364 9365 if (fd >= 0) { 9366 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9367 mode != XDP_MODE_SKB); 9368 if (IS_ERR(new_prog)) 9369 return PTR_ERR(new_prog); 9370 } 9371 9372 if (expected_fd >= 0) { 9373 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9374 mode != XDP_MODE_SKB); 9375 if (IS_ERR(old_prog)) { 9376 err = PTR_ERR(old_prog); 9377 old_prog = NULL; 9378 goto err_out; 9379 } 9380 } 9381 9382 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9383 9384 err_out: 9385 if (err && new_prog) 9386 bpf_prog_put(new_prog); 9387 if (old_prog) 9388 bpf_prog_put(old_prog); 9389 return err; 9390 } 9391 9392 /** 9393 * dev_new_index - allocate an ifindex 9394 * @net: the applicable net namespace 9395 * 9396 * Returns a suitable unique value for a new device interface 9397 * number. The caller must hold the rtnl semaphore or the 9398 * dev_base_lock to be sure it remains unique. 9399 */ 9400 static int dev_new_index(struct net *net) 9401 { 9402 int ifindex = net->ifindex; 9403 9404 for (;;) { 9405 if (++ifindex <= 0) 9406 ifindex = 1; 9407 if (!__dev_get_by_index(net, ifindex)) 9408 return net->ifindex = ifindex; 9409 } 9410 } 9411 9412 /* Delayed registration/unregisteration */ 9413 static LIST_HEAD(net_todo_list); 9414 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9415 9416 static void net_set_todo(struct net_device *dev) 9417 { 9418 list_add_tail(&dev->todo_list, &net_todo_list); 9419 atomic_inc(&dev_net(dev)->dev_unreg_count); 9420 } 9421 9422 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9423 struct net_device *upper, netdev_features_t features) 9424 { 9425 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9426 netdev_features_t feature; 9427 int feature_bit; 9428 9429 for_each_netdev_feature(upper_disables, feature_bit) { 9430 feature = __NETIF_F_BIT(feature_bit); 9431 if (!(upper->wanted_features & feature) 9432 && (features & feature)) { 9433 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9434 &feature, upper->name); 9435 features &= ~feature; 9436 } 9437 } 9438 9439 return features; 9440 } 9441 9442 static void netdev_sync_lower_features(struct net_device *upper, 9443 struct net_device *lower, netdev_features_t features) 9444 { 9445 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9446 netdev_features_t feature; 9447 int feature_bit; 9448 9449 for_each_netdev_feature(upper_disables, feature_bit) { 9450 feature = __NETIF_F_BIT(feature_bit); 9451 if (!(features & feature) && (lower->features & feature)) { 9452 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9453 &feature, lower->name); 9454 lower->wanted_features &= ~feature; 9455 __netdev_update_features(lower); 9456 9457 if (unlikely(lower->features & feature)) 9458 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9459 &feature, lower->name); 9460 else 9461 netdev_features_change(lower); 9462 } 9463 } 9464 } 9465 9466 static netdev_features_t netdev_fix_features(struct net_device *dev, 9467 netdev_features_t features) 9468 { 9469 /* Fix illegal checksum combinations */ 9470 if ((features & NETIF_F_HW_CSUM) && 9471 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9472 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9473 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9474 } 9475 9476 /* TSO requires that SG is present as well. */ 9477 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9478 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9479 features &= ~NETIF_F_ALL_TSO; 9480 } 9481 9482 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9483 !(features & NETIF_F_IP_CSUM)) { 9484 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9485 features &= ~NETIF_F_TSO; 9486 features &= ~NETIF_F_TSO_ECN; 9487 } 9488 9489 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9490 !(features & NETIF_F_IPV6_CSUM)) { 9491 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9492 features &= ~NETIF_F_TSO6; 9493 } 9494 9495 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9496 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9497 features &= ~NETIF_F_TSO_MANGLEID; 9498 9499 /* TSO ECN requires that TSO is present as well. */ 9500 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9501 features &= ~NETIF_F_TSO_ECN; 9502 9503 /* Software GSO depends on SG. */ 9504 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9505 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9506 features &= ~NETIF_F_GSO; 9507 } 9508 9509 /* GSO partial features require GSO partial be set */ 9510 if ((features & dev->gso_partial_features) && 9511 !(features & NETIF_F_GSO_PARTIAL)) { 9512 netdev_dbg(dev, 9513 "Dropping partially supported GSO features since no GSO partial.\n"); 9514 features &= ~dev->gso_partial_features; 9515 } 9516 9517 if (!(features & NETIF_F_RXCSUM)) { 9518 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9519 * successfully merged by hardware must also have the 9520 * checksum verified by hardware. If the user does not 9521 * want to enable RXCSUM, logically, we should disable GRO_HW. 9522 */ 9523 if (features & NETIF_F_GRO_HW) { 9524 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9525 features &= ~NETIF_F_GRO_HW; 9526 } 9527 } 9528 9529 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9530 if (features & NETIF_F_RXFCS) { 9531 if (features & NETIF_F_LRO) { 9532 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9533 features &= ~NETIF_F_LRO; 9534 } 9535 9536 if (features & NETIF_F_GRO_HW) { 9537 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9538 features &= ~NETIF_F_GRO_HW; 9539 } 9540 } 9541 9542 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9543 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9544 features &= ~NETIF_F_LRO; 9545 } 9546 9547 if (features & NETIF_F_HW_TLS_TX) { 9548 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9549 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9550 bool hw_csum = features & NETIF_F_HW_CSUM; 9551 9552 if (!ip_csum && !hw_csum) { 9553 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9554 features &= ~NETIF_F_HW_TLS_TX; 9555 } 9556 } 9557 9558 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9559 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9560 features &= ~NETIF_F_HW_TLS_RX; 9561 } 9562 9563 return features; 9564 } 9565 9566 int __netdev_update_features(struct net_device *dev) 9567 { 9568 struct net_device *upper, *lower; 9569 netdev_features_t features; 9570 struct list_head *iter; 9571 int err = -1; 9572 9573 ASSERT_RTNL(); 9574 9575 features = netdev_get_wanted_features(dev); 9576 9577 if (dev->netdev_ops->ndo_fix_features) 9578 features = dev->netdev_ops->ndo_fix_features(dev, features); 9579 9580 /* driver might be less strict about feature dependencies */ 9581 features = netdev_fix_features(dev, features); 9582 9583 /* some features can't be enabled if they're off on an upper device */ 9584 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9585 features = netdev_sync_upper_features(dev, upper, features); 9586 9587 if (dev->features == features) 9588 goto sync_lower; 9589 9590 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9591 &dev->features, &features); 9592 9593 if (dev->netdev_ops->ndo_set_features) 9594 err = dev->netdev_ops->ndo_set_features(dev, features); 9595 else 9596 err = 0; 9597 9598 if (unlikely(err < 0)) { 9599 netdev_err(dev, 9600 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9601 err, &features, &dev->features); 9602 /* return non-0 since some features might have changed and 9603 * it's better to fire a spurious notification than miss it 9604 */ 9605 return -1; 9606 } 9607 9608 sync_lower: 9609 /* some features must be disabled on lower devices when disabled 9610 * on an upper device (think: bonding master or bridge) 9611 */ 9612 netdev_for_each_lower_dev(dev, lower, iter) 9613 netdev_sync_lower_features(dev, lower, features); 9614 9615 if (!err) { 9616 netdev_features_t diff = features ^ dev->features; 9617 9618 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9619 /* udp_tunnel_{get,drop}_rx_info both need 9620 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9621 * device, or they won't do anything. 9622 * Thus we need to update dev->features 9623 * *before* calling udp_tunnel_get_rx_info, 9624 * but *after* calling udp_tunnel_drop_rx_info. 9625 */ 9626 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9627 dev->features = features; 9628 udp_tunnel_get_rx_info(dev); 9629 } else { 9630 udp_tunnel_drop_rx_info(dev); 9631 } 9632 } 9633 9634 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9635 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9636 dev->features = features; 9637 err |= vlan_get_rx_ctag_filter_info(dev); 9638 } else { 9639 vlan_drop_rx_ctag_filter_info(dev); 9640 } 9641 } 9642 9643 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9644 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9645 dev->features = features; 9646 err |= vlan_get_rx_stag_filter_info(dev); 9647 } else { 9648 vlan_drop_rx_stag_filter_info(dev); 9649 } 9650 } 9651 9652 dev->features = features; 9653 } 9654 9655 return err < 0 ? 0 : 1; 9656 } 9657 9658 /** 9659 * netdev_update_features - recalculate device features 9660 * @dev: the device to check 9661 * 9662 * Recalculate dev->features set and send notifications if it 9663 * has changed. Should be called after driver or hardware dependent 9664 * conditions might have changed that influence the features. 9665 */ 9666 void netdev_update_features(struct net_device *dev) 9667 { 9668 if (__netdev_update_features(dev)) 9669 netdev_features_change(dev); 9670 } 9671 EXPORT_SYMBOL(netdev_update_features); 9672 9673 /** 9674 * netdev_change_features - recalculate device features 9675 * @dev: the device to check 9676 * 9677 * Recalculate dev->features set and send notifications even 9678 * if they have not changed. Should be called instead of 9679 * netdev_update_features() if also dev->vlan_features might 9680 * have changed to allow the changes to be propagated to stacked 9681 * VLAN devices. 9682 */ 9683 void netdev_change_features(struct net_device *dev) 9684 { 9685 __netdev_update_features(dev); 9686 netdev_features_change(dev); 9687 } 9688 EXPORT_SYMBOL(netdev_change_features); 9689 9690 /** 9691 * netif_stacked_transfer_operstate - transfer operstate 9692 * @rootdev: the root or lower level device to transfer state from 9693 * @dev: the device to transfer operstate to 9694 * 9695 * Transfer operational state from root to device. This is normally 9696 * called when a stacking relationship exists between the root 9697 * device and the device(a leaf device). 9698 */ 9699 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9700 struct net_device *dev) 9701 { 9702 if (rootdev->operstate == IF_OPER_DORMANT) 9703 netif_dormant_on(dev); 9704 else 9705 netif_dormant_off(dev); 9706 9707 if (rootdev->operstate == IF_OPER_TESTING) 9708 netif_testing_on(dev); 9709 else 9710 netif_testing_off(dev); 9711 9712 if (netif_carrier_ok(rootdev)) 9713 netif_carrier_on(dev); 9714 else 9715 netif_carrier_off(dev); 9716 } 9717 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9718 9719 static int netif_alloc_rx_queues(struct net_device *dev) 9720 { 9721 unsigned int i, count = dev->num_rx_queues; 9722 struct netdev_rx_queue *rx; 9723 size_t sz = count * sizeof(*rx); 9724 int err = 0; 9725 9726 BUG_ON(count < 1); 9727 9728 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9729 if (!rx) 9730 return -ENOMEM; 9731 9732 dev->_rx = rx; 9733 9734 for (i = 0; i < count; i++) { 9735 rx[i].dev = dev; 9736 9737 /* XDP RX-queue setup */ 9738 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9739 if (err < 0) 9740 goto err_rxq_info; 9741 } 9742 return 0; 9743 9744 err_rxq_info: 9745 /* Rollback successful reg's and free other resources */ 9746 while (i--) 9747 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9748 kvfree(dev->_rx); 9749 dev->_rx = NULL; 9750 return err; 9751 } 9752 9753 static void netif_free_rx_queues(struct net_device *dev) 9754 { 9755 unsigned int i, count = dev->num_rx_queues; 9756 9757 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9758 if (!dev->_rx) 9759 return; 9760 9761 for (i = 0; i < count; i++) 9762 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9763 9764 kvfree(dev->_rx); 9765 } 9766 9767 static void netdev_init_one_queue(struct net_device *dev, 9768 struct netdev_queue *queue, void *_unused) 9769 { 9770 /* Initialize queue lock */ 9771 spin_lock_init(&queue->_xmit_lock); 9772 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9773 queue->xmit_lock_owner = -1; 9774 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9775 queue->dev = dev; 9776 #ifdef CONFIG_BQL 9777 dql_init(&queue->dql, HZ); 9778 #endif 9779 } 9780 9781 static void netif_free_tx_queues(struct net_device *dev) 9782 { 9783 kvfree(dev->_tx); 9784 } 9785 9786 static int netif_alloc_netdev_queues(struct net_device *dev) 9787 { 9788 unsigned int count = dev->num_tx_queues; 9789 struct netdev_queue *tx; 9790 size_t sz = count * sizeof(*tx); 9791 9792 if (count < 1 || count > 0xffff) 9793 return -EINVAL; 9794 9795 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9796 if (!tx) 9797 return -ENOMEM; 9798 9799 dev->_tx = tx; 9800 9801 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9802 spin_lock_init(&dev->tx_global_lock); 9803 9804 return 0; 9805 } 9806 9807 void netif_tx_stop_all_queues(struct net_device *dev) 9808 { 9809 unsigned int i; 9810 9811 for (i = 0; i < dev->num_tx_queues; i++) { 9812 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9813 9814 netif_tx_stop_queue(txq); 9815 } 9816 } 9817 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9818 9819 /** 9820 * register_netdevice - register a network device 9821 * @dev: device to register 9822 * 9823 * Take a completed network device structure and add it to the kernel 9824 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9825 * chain. 0 is returned on success. A negative errno code is returned 9826 * on a failure to set up the device, or if the name is a duplicate. 9827 * 9828 * Callers must hold the rtnl semaphore. You may want 9829 * register_netdev() instead of this. 9830 * 9831 * BUGS: 9832 * The locking appears insufficient to guarantee two parallel registers 9833 * will not get the same name. 9834 */ 9835 9836 int register_netdevice(struct net_device *dev) 9837 { 9838 int ret; 9839 struct net *net = dev_net(dev); 9840 9841 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9842 NETDEV_FEATURE_COUNT); 9843 BUG_ON(dev_boot_phase); 9844 ASSERT_RTNL(); 9845 9846 might_sleep(); 9847 9848 /* When net_device's are persistent, this will be fatal. */ 9849 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9850 BUG_ON(!net); 9851 9852 ret = ethtool_check_ops(dev->ethtool_ops); 9853 if (ret) 9854 return ret; 9855 9856 spin_lock_init(&dev->addr_list_lock); 9857 netdev_set_addr_lockdep_class(dev); 9858 9859 ret = dev_get_valid_name(net, dev, dev->name); 9860 if (ret < 0) 9861 goto out; 9862 9863 ret = -ENOMEM; 9864 dev->name_node = netdev_name_node_head_alloc(dev); 9865 if (!dev->name_node) 9866 goto out; 9867 9868 /* Init, if this function is available */ 9869 if (dev->netdev_ops->ndo_init) { 9870 ret = dev->netdev_ops->ndo_init(dev); 9871 if (ret) { 9872 if (ret > 0) 9873 ret = -EIO; 9874 goto err_free_name; 9875 } 9876 } 9877 9878 if (((dev->hw_features | dev->features) & 9879 NETIF_F_HW_VLAN_CTAG_FILTER) && 9880 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9881 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9882 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9883 ret = -EINVAL; 9884 goto err_uninit; 9885 } 9886 9887 ret = -EBUSY; 9888 if (!dev->ifindex) 9889 dev->ifindex = dev_new_index(net); 9890 else if (__dev_get_by_index(net, dev->ifindex)) 9891 goto err_uninit; 9892 9893 /* Transfer changeable features to wanted_features and enable 9894 * software offloads (GSO and GRO). 9895 */ 9896 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 9897 dev->features |= NETIF_F_SOFT_FEATURES; 9898 9899 if (dev->udp_tunnel_nic_info) { 9900 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9901 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9902 } 9903 9904 dev->wanted_features = dev->features & dev->hw_features; 9905 9906 if (!(dev->flags & IFF_LOOPBACK)) 9907 dev->hw_features |= NETIF_F_NOCACHE_COPY; 9908 9909 /* If IPv4 TCP segmentation offload is supported we should also 9910 * allow the device to enable segmenting the frame with the option 9911 * of ignoring a static IP ID value. This doesn't enable the 9912 * feature itself but allows the user to enable it later. 9913 */ 9914 if (dev->hw_features & NETIF_F_TSO) 9915 dev->hw_features |= NETIF_F_TSO_MANGLEID; 9916 if (dev->vlan_features & NETIF_F_TSO) 9917 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 9918 if (dev->mpls_features & NETIF_F_TSO) 9919 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 9920 if (dev->hw_enc_features & NETIF_F_TSO) 9921 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 9922 9923 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 9924 */ 9925 dev->vlan_features |= NETIF_F_HIGHDMA; 9926 9927 /* Make NETIF_F_SG inheritable to tunnel devices. 9928 */ 9929 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 9930 9931 /* Make NETIF_F_SG inheritable to MPLS. 9932 */ 9933 dev->mpls_features |= NETIF_F_SG; 9934 9935 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 9936 ret = notifier_to_errno(ret); 9937 if (ret) 9938 goto err_uninit; 9939 9940 ret = netdev_register_kobject(dev); 9941 if (ret) { 9942 dev->reg_state = NETREG_UNREGISTERED; 9943 goto err_uninit; 9944 } 9945 dev->reg_state = NETREG_REGISTERED; 9946 9947 __netdev_update_features(dev); 9948 9949 /* 9950 * Default initial state at registry is that the 9951 * device is present. 9952 */ 9953 9954 set_bit(__LINK_STATE_PRESENT, &dev->state); 9955 9956 linkwatch_init_dev(dev); 9957 9958 dev_init_scheduler(dev); 9959 9960 dev_hold_track(dev, &dev->dev_registered_tracker, GFP_KERNEL); 9961 list_netdevice(dev); 9962 9963 add_device_randomness(dev->dev_addr, dev->addr_len); 9964 9965 /* If the device has permanent device address, driver should 9966 * set dev_addr and also addr_assign_type should be set to 9967 * NET_ADDR_PERM (default value). 9968 */ 9969 if (dev->addr_assign_type == NET_ADDR_PERM) 9970 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 9971 9972 /* Notify protocols, that a new device appeared. */ 9973 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 9974 ret = notifier_to_errno(ret); 9975 if (ret) { 9976 /* Expect explicit free_netdev() on failure */ 9977 dev->needs_free_netdev = false; 9978 unregister_netdevice_queue(dev, NULL); 9979 goto out; 9980 } 9981 /* 9982 * Prevent userspace races by waiting until the network 9983 * device is fully setup before sending notifications. 9984 */ 9985 if (!dev->rtnl_link_ops || 9986 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 9987 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 9988 9989 out: 9990 return ret; 9991 9992 err_uninit: 9993 if (dev->netdev_ops->ndo_uninit) 9994 dev->netdev_ops->ndo_uninit(dev); 9995 if (dev->priv_destructor) 9996 dev->priv_destructor(dev); 9997 err_free_name: 9998 netdev_name_node_free(dev->name_node); 9999 goto out; 10000 } 10001 EXPORT_SYMBOL(register_netdevice); 10002 10003 /** 10004 * init_dummy_netdev - init a dummy network device for NAPI 10005 * @dev: device to init 10006 * 10007 * This takes a network device structure and initialize the minimum 10008 * amount of fields so it can be used to schedule NAPI polls without 10009 * registering a full blown interface. This is to be used by drivers 10010 * that need to tie several hardware interfaces to a single NAPI 10011 * poll scheduler due to HW limitations. 10012 */ 10013 int init_dummy_netdev(struct net_device *dev) 10014 { 10015 /* Clear everything. Note we don't initialize spinlocks 10016 * are they aren't supposed to be taken by any of the 10017 * NAPI code and this dummy netdev is supposed to be 10018 * only ever used for NAPI polls 10019 */ 10020 memset(dev, 0, sizeof(struct net_device)); 10021 10022 /* make sure we BUG if trying to hit standard 10023 * register/unregister code path 10024 */ 10025 dev->reg_state = NETREG_DUMMY; 10026 10027 /* NAPI wants this */ 10028 INIT_LIST_HEAD(&dev->napi_list); 10029 10030 /* a dummy interface is started by default */ 10031 set_bit(__LINK_STATE_PRESENT, &dev->state); 10032 set_bit(__LINK_STATE_START, &dev->state); 10033 10034 /* napi_busy_loop stats accounting wants this */ 10035 dev_net_set(dev, &init_net); 10036 10037 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10038 * because users of this 'device' dont need to change 10039 * its refcount. 10040 */ 10041 10042 return 0; 10043 } 10044 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10045 10046 10047 /** 10048 * register_netdev - register a network device 10049 * @dev: device to register 10050 * 10051 * Take a completed network device structure and add it to the kernel 10052 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10053 * chain. 0 is returned on success. A negative errno code is returned 10054 * on a failure to set up the device, or if the name is a duplicate. 10055 * 10056 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10057 * and expands the device name if you passed a format string to 10058 * alloc_netdev. 10059 */ 10060 int register_netdev(struct net_device *dev) 10061 { 10062 int err; 10063 10064 if (rtnl_lock_killable()) 10065 return -EINTR; 10066 err = register_netdevice(dev); 10067 rtnl_unlock(); 10068 return err; 10069 } 10070 EXPORT_SYMBOL(register_netdev); 10071 10072 int netdev_refcnt_read(const struct net_device *dev) 10073 { 10074 #ifdef CONFIG_PCPU_DEV_REFCNT 10075 int i, refcnt = 0; 10076 10077 for_each_possible_cpu(i) 10078 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10079 return refcnt; 10080 #else 10081 return refcount_read(&dev->dev_refcnt); 10082 #endif 10083 } 10084 EXPORT_SYMBOL(netdev_refcnt_read); 10085 10086 int netdev_unregister_timeout_secs __read_mostly = 10; 10087 10088 #define WAIT_REFS_MIN_MSECS 1 10089 #define WAIT_REFS_MAX_MSECS 250 10090 /** 10091 * netdev_wait_allrefs_any - wait until all references are gone. 10092 * @list: list of net_devices to wait on 10093 * 10094 * This is called when unregistering network devices. 10095 * 10096 * Any protocol or device that holds a reference should register 10097 * for netdevice notification, and cleanup and put back the 10098 * reference if they receive an UNREGISTER event. 10099 * We can get stuck here if buggy protocols don't correctly 10100 * call dev_put. 10101 */ 10102 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10103 { 10104 unsigned long rebroadcast_time, warning_time; 10105 struct net_device *dev; 10106 int wait = 0; 10107 10108 rebroadcast_time = warning_time = jiffies; 10109 10110 list_for_each_entry(dev, list, todo_list) 10111 if (netdev_refcnt_read(dev) == 1) 10112 return dev; 10113 10114 while (true) { 10115 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10116 rtnl_lock(); 10117 10118 /* Rebroadcast unregister notification */ 10119 list_for_each_entry(dev, list, todo_list) 10120 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10121 10122 __rtnl_unlock(); 10123 rcu_barrier(); 10124 rtnl_lock(); 10125 10126 list_for_each_entry(dev, list, todo_list) 10127 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10128 &dev->state)) { 10129 /* We must not have linkwatch events 10130 * pending on unregister. If this 10131 * happens, we simply run the queue 10132 * unscheduled, resulting in a noop 10133 * for this device. 10134 */ 10135 linkwatch_run_queue(); 10136 break; 10137 } 10138 10139 __rtnl_unlock(); 10140 10141 rebroadcast_time = jiffies; 10142 } 10143 10144 if (!wait) { 10145 rcu_barrier(); 10146 wait = WAIT_REFS_MIN_MSECS; 10147 } else { 10148 msleep(wait); 10149 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10150 } 10151 10152 list_for_each_entry(dev, list, todo_list) 10153 if (netdev_refcnt_read(dev) == 1) 10154 return dev; 10155 10156 if (time_after(jiffies, warning_time + 10157 netdev_unregister_timeout_secs * HZ)) { 10158 list_for_each_entry(dev, list, todo_list) { 10159 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10160 dev->name, netdev_refcnt_read(dev)); 10161 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10162 } 10163 10164 warning_time = jiffies; 10165 } 10166 } 10167 } 10168 10169 /* The sequence is: 10170 * 10171 * rtnl_lock(); 10172 * ... 10173 * register_netdevice(x1); 10174 * register_netdevice(x2); 10175 * ... 10176 * unregister_netdevice(y1); 10177 * unregister_netdevice(y2); 10178 * ... 10179 * rtnl_unlock(); 10180 * free_netdev(y1); 10181 * free_netdev(y2); 10182 * 10183 * We are invoked by rtnl_unlock(). 10184 * This allows us to deal with problems: 10185 * 1) We can delete sysfs objects which invoke hotplug 10186 * without deadlocking with linkwatch via keventd. 10187 * 2) Since we run with the RTNL semaphore not held, we can sleep 10188 * safely in order to wait for the netdev refcnt to drop to zero. 10189 * 10190 * We must not return until all unregister events added during 10191 * the interval the lock was held have been completed. 10192 */ 10193 void netdev_run_todo(void) 10194 { 10195 struct net_device *dev, *tmp; 10196 struct list_head list; 10197 #ifdef CONFIG_LOCKDEP 10198 struct list_head unlink_list; 10199 10200 list_replace_init(&net_unlink_list, &unlink_list); 10201 10202 while (!list_empty(&unlink_list)) { 10203 struct net_device *dev = list_first_entry(&unlink_list, 10204 struct net_device, 10205 unlink_list); 10206 list_del_init(&dev->unlink_list); 10207 dev->nested_level = dev->lower_level - 1; 10208 } 10209 #endif 10210 10211 /* Snapshot list, allow later requests */ 10212 list_replace_init(&net_todo_list, &list); 10213 10214 __rtnl_unlock(); 10215 10216 /* Wait for rcu callbacks to finish before next phase */ 10217 if (!list_empty(&list)) 10218 rcu_barrier(); 10219 10220 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10221 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10222 netdev_WARN(dev, "run_todo but not unregistering\n"); 10223 list_del(&dev->todo_list); 10224 continue; 10225 } 10226 10227 dev->reg_state = NETREG_UNREGISTERED; 10228 linkwatch_forget_dev(dev); 10229 } 10230 10231 while (!list_empty(&list)) { 10232 dev = netdev_wait_allrefs_any(&list); 10233 list_del(&dev->todo_list); 10234 10235 /* paranoia */ 10236 BUG_ON(netdev_refcnt_read(dev) != 1); 10237 BUG_ON(!list_empty(&dev->ptype_all)); 10238 BUG_ON(!list_empty(&dev->ptype_specific)); 10239 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10240 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10241 #if IS_ENABLED(CONFIG_DECNET) 10242 WARN_ON(dev->dn_ptr); 10243 #endif 10244 if (dev->priv_destructor) 10245 dev->priv_destructor(dev); 10246 if (dev->needs_free_netdev) 10247 free_netdev(dev); 10248 10249 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10250 wake_up(&netdev_unregistering_wq); 10251 10252 /* Free network device */ 10253 kobject_put(&dev->dev.kobj); 10254 } 10255 } 10256 10257 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10258 * all the same fields in the same order as net_device_stats, with only 10259 * the type differing, but rtnl_link_stats64 may have additional fields 10260 * at the end for newer counters. 10261 */ 10262 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10263 const struct net_device_stats *netdev_stats) 10264 { 10265 #if BITS_PER_LONG == 64 10266 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10267 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10268 /* zero out counters that only exist in rtnl_link_stats64 */ 10269 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10270 sizeof(*stats64) - sizeof(*netdev_stats)); 10271 #else 10272 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10273 const unsigned long *src = (const unsigned long *)netdev_stats; 10274 u64 *dst = (u64 *)stats64; 10275 10276 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10277 for (i = 0; i < n; i++) 10278 dst[i] = src[i]; 10279 /* zero out counters that only exist in rtnl_link_stats64 */ 10280 memset((char *)stats64 + n * sizeof(u64), 0, 10281 sizeof(*stats64) - n * sizeof(u64)); 10282 #endif 10283 } 10284 EXPORT_SYMBOL(netdev_stats_to_stats64); 10285 10286 struct net_device_core_stats *netdev_core_stats_alloc(struct net_device *dev) 10287 { 10288 struct net_device_core_stats __percpu *p; 10289 10290 p = alloc_percpu_gfp(struct net_device_core_stats, 10291 GFP_ATOMIC | __GFP_NOWARN); 10292 10293 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10294 free_percpu(p); 10295 10296 /* This READ_ONCE() pairs with the cmpxchg() above */ 10297 p = READ_ONCE(dev->core_stats); 10298 if (!p) 10299 return NULL; 10300 10301 return this_cpu_ptr(p); 10302 } 10303 EXPORT_SYMBOL(netdev_core_stats_alloc); 10304 10305 /** 10306 * dev_get_stats - get network device statistics 10307 * @dev: device to get statistics from 10308 * @storage: place to store stats 10309 * 10310 * Get network statistics from device. Return @storage. 10311 * The device driver may provide its own method by setting 10312 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10313 * otherwise the internal statistics structure is used. 10314 */ 10315 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10316 struct rtnl_link_stats64 *storage) 10317 { 10318 const struct net_device_ops *ops = dev->netdev_ops; 10319 const struct net_device_core_stats __percpu *p; 10320 10321 if (ops->ndo_get_stats64) { 10322 memset(storage, 0, sizeof(*storage)); 10323 ops->ndo_get_stats64(dev, storage); 10324 } else if (ops->ndo_get_stats) { 10325 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10326 } else { 10327 netdev_stats_to_stats64(storage, &dev->stats); 10328 } 10329 10330 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10331 p = READ_ONCE(dev->core_stats); 10332 if (p) { 10333 const struct net_device_core_stats *core_stats; 10334 int i; 10335 10336 for_each_possible_cpu(i) { 10337 core_stats = per_cpu_ptr(p, i); 10338 storage->rx_dropped += local_read(&core_stats->rx_dropped); 10339 storage->tx_dropped += local_read(&core_stats->tx_dropped); 10340 storage->rx_nohandler += local_read(&core_stats->rx_nohandler); 10341 } 10342 } 10343 return storage; 10344 } 10345 EXPORT_SYMBOL(dev_get_stats); 10346 10347 /** 10348 * dev_fetch_sw_netstats - get per-cpu network device statistics 10349 * @s: place to store stats 10350 * @netstats: per-cpu network stats to read from 10351 * 10352 * Read per-cpu network statistics and populate the related fields in @s. 10353 */ 10354 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10355 const struct pcpu_sw_netstats __percpu *netstats) 10356 { 10357 int cpu; 10358 10359 for_each_possible_cpu(cpu) { 10360 const struct pcpu_sw_netstats *stats; 10361 struct pcpu_sw_netstats tmp; 10362 unsigned int start; 10363 10364 stats = per_cpu_ptr(netstats, cpu); 10365 do { 10366 start = u64_stats_fetch_begin_irq(&stats->syncp); 10367 tmp.rx_packets = stats->rx_packets; 10368 tmp.rx_bytes = stats->rx_bytes; 10369 tmp.tx_packets = stats->tx_packets; 10370 tmp.tx_bytes = stats->tx_bytes; 10371 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10372 10373 s->rx_packets += tmp.rx_packets; 10374 s->rx_bytes += tmp.rx_bytes; 10375 s->tx_packets += tmp.tx_packets; 10376 s->tx_bytes += tmp.tx_bytes; 10377 } 10378 } 10379 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10380 10381 /** 10382 * dev_get_tstats64 - ndo_get_stats64 implementation 10383 * @dev: device to get statistics from 10384 * @s: place to store stats 10385 * 10386 * Populate @s from dev->stats and dev->tstats. Can be used as 10387 * ndo_get_stats64() callback. 10388 */ 10389 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10390 { 10391 netdev_stats_to_stats64(s, &dev->stats); 10392 dev_fetch_sw_netstats(s, dev->tstats); 10393 } 10394 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10395 10396 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10397 { 10398 struct netdev_queue *queue = dev_ingress_queue(dev); 10399 10400 #ifdef CONFIG_NET_CLS_ACT 10401 if (queue) 10402 return queue; 10403 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10404 if (!queue) 10405 return NULL; 10406 netdev_init_one_queue(dev, queue, NULL); 10407 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10408 queue->qdisc_sleeping = &noop_qdisc; 10409 rcu_assign_pointer(dev->ingress_queue, queue); 10410 #endif 10411 return queue; 10412 } 10413 10414 static const struct ethtool_ops default_ethtool_ops; 10415 10416 void netdev_set_default_ethtool_ops(struct net_device *dev, 10417 const struct ethtool_ops *ops) 10418 { 10419 if (dev->ethtool_ops == &default_ethtool_ops) 10420 dev->ethtool_ops = ops; 10421 } 10422 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10423 10424 void netdev_freemem(struct net_device *dev) 10425 { 10426 char *addr = (char *)dev - dev->padded; 10427 10428 kvfree(addr); 10429 } 10430 10431 /** 10432 * alloc_netdev_mqs - allocate network device 10433 * @sizeof_priv: size of private data to allocate space for 10434 * @name: device name format string 10435 * @name_assign_type: origin of device name 10436 * @setup: callback to initialize device 10437 * @txqs: the number of TX subqueues to allocate 10438 * @rxqs: the number of RX subqueues to allocate 10439 * 10440 * Allocates a struct net_device with private data area for driver use 10441 * and performs basic initialization. Also allocates subqueue structs 10442 * for each queue on the device. 10443 */ 10444 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10445 unsigned char name_assign_type, 10446 void (*setup)(struct net_device *), 10447 unsigned int txqs, unsigned int rxqs) 10448 { 10449 struct net_device *dev; 10450 unsigned int alloc_size; 10451 struct net_device *p; 10452 10453 BUG_ON(strlen(name) >= sizeof(dev->name)); 10454 10455 if (txqs < 1) { 10456 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10457 return NULL; 10458 } 10459 10460 if (rxqs < 1) { 10461 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10462 return NULL; 10463 } 10464 10465 alloc_size = sizeof(struct net_device); 10466 if (sizeof_priv) { 10467 /* ensure 32-byte alignment of private area */ 10468 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10469 alloc_size += sizeof_priv; 10470 } 10471 /* ensure 32-byte alignment of whole construct */ 10472 alloc_size += NETDEV_ALIGN - 1; 10473 10474 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10475 if (!p) 10476 return NULL; 10477 10478 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10479 dev->padded = (char *)dev - (char *)p; 10480 10481 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10482 #ifdef CONFIG_PCPU_DEV_REFCNT 10483 dev->pcpu_refcnt = alloc_percpu(int); 10484 if (!dev->pcpu_refcnt) 10485 goto free_dev; 10486 __dev_hold(dev); 10487 #else 10488 refcount_set(&dev->dev_refcnt, 1); 10489 #endif 10490 10491 if (dev_addr_init(dev)) 10492 goto free_pcpu; 10493 10494 dev_mc_init(dev); 10495 dev_uc_init(dev); 10496 10497 dev_net_set(dev, &init_net); 10498 10499 dev->gso_max_size = GSO_MAX_SIZE; 10500 dev->gso_max_segs = GSO_MAX_SEGS; 10501 dev->gro_max_size = GRO_MAX_SIZE; 10502 dev->upper_level = 1; 10503 dev->lower_level = 1; 10504 #ifdef CONFIG_LOCKDEP 10505 dev->nested_level = 0; 10506 INIT_LIST_HEAD(&dev->unlink_list); 10507 #endif 10508 10509 INIT_LIST_HEAD(&dev->napi_list); 10510 INIT_LIST_HEAD(&dev->unreg_list); 10511 INIT_LIST_HEAD(&dev->close_list); 10512 INIT_LIST_HEAD(&dev->link_watch_list); 10513 INIT_LIST_HEAD(&dev->adj_list.upper); 10514 INIT_LIST_HEAD(&dev->adj_list.lower); 10515 INIT_LIST_HEAD(&dev->ptype_all); 10516 INIT_LIST_HEAD(&dev->ptype_specific); 10517 INIT_LIST_HEAD(&dev->net_notifier_list); 10518 #ifdef CONFIG_NET_SCHED 10519 hash_init(dev->qdisc_hash); 10520 #endif 10521 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10522 setup(dev); 10523 10524 if (!dev->tx_queue_len) { 10525 dev->priv_flags |= IFF_NO_QUEUE; 10526 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10527 } 10528 10529 dev->num_tx_queues = txqs; 10530 dev->real_num_tx_queues = txqs; 10531 if (netif_alloc_netdev_queues(dev)) 10532 goto free_all; 10533 10534 dev->num_rx_queues = rxqs; 10535 dev->real_num_rx_queues = rxqs; 10536 if (netif_alloc_rx_queues(dev)) 10537 goto free_all; 10538 10539 strcpy(dev->name, name); 10540 dev->name_assign_type = name_assign_type; 10541 dev->group = INIT_NETDEV_GROUP; 10542 if (!dev->ethtool_ops) 10543 dev->ethtool_ops = &default_ethtool_ops; 10544 10545 nf_hook_netdev_init(dev); 10546 10547 return dev; 10548 10549 free_all: 10550 free_netdev(dev); 10551 return NULL; 10552 10553 free_pcpu: 10554 #ifdef CONFIG_PCPU_DEV_REFCNT 10555 free_percpu(dev->pcpu_refcnt); 10556 free_dev: 10557 #endif 10558 netdev_freemem(dev); 10559 return NULL; 10560 } 10561 EXPORT_SYMBOL(alloc_netdev_mqs); 10562 10563 /** 10564 * free_netdev - free network device 10565 * @dev: device 10566 * 10567 * This function does the last stage of destroying an allocated device 10568 * interface. The reference to the device object is released. If this 10569 * is the last reference then it will be freed.Must be called in process 10570 * context. 10571 */ 10572 void free_netdev(struct net_device *dev) 10573 { 10574 struct napi_struct *p, *n; 10575 10576 might_sleep(); 10577 10578 /* When called immediately after register_netdevice() failed the unwind 10579 * handling may still be dismantling the device. Handle that case by 10580 * deferring the free. 10581 */ 10582 if (dev->reg_state == NETREG_UNREGISTERING) { 10583 ASSERT_RTNL(); 10584 dev->needs_free_netdev = true; 10585 return; 10586 } 10587 10588 netif_free_tx_queues(dev); 10589 netif_free_rx_queues(dev); 10590 10591 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10592 10593 /* Flush device addresses */ 10594 dev_addr_flush(dev); 10595 10596 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10597 netif_napi_del(p); 10598 10599 ref_tracker_dir_exit(&dev->refcnt_tracker); 10600 #ifdef CONFIG_PCPU_DEV_REFCNT 10601 free_percpu(dev->pcpu_refcnt); 10602 dev->pcpu_refcnt = NULL; 10603 #endif 10604 free_percpu(dev->core_stats); 10605 dev->core_stats = NULL; 10606 free_percpu(dev->xdp_bulkq); 10607 dev->xdp_bulkq = NULL; 10608 10609 /* Compatibility with error handling in drivers */ 10610 if (dev->reg_state == NETREG_UNINITIALIZED) { 10611 netdev_freemem(dev); 10612 return; 10613 } 10614 10615 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10616 dev->reg_state = NETREG_RELEASED; 10617 10618 /* will free via device release */ 10619 put_device(&dev->dev); 10620 } 10621 EXPORT_SYMBOL(free_netdev); 10622 10623 /** 10624 * synchronize_net - Synchronize with packet receive processing 10625 * 10626 * Wait for packets currently being received to be done. 10627 * Does not block later packets from starting. 10628 */ 10629 void synchronize_net(void) 10630 { 10631 might_sleep(); 10632 if (rtnl_is_locked()) 10633 synchronize_rcu_expedited(); 10634 else 10635 synchronize_rcu(); 10636 } 10637 EXPORT_SYMBOL(synchronize_net); 10638 10639 /** 10640 * unregister_netdevice_queue - remove device from the kernel 10641 * @dev: device 10642 * @head: list 10643 * 10644 * This function shuts down a device interface and removes it 10645 * from the kernel tables. 10646 * If head not NULL, device is queued to be unregistered later. 10647 * 10648 * Callers must hold the rtnl semaphore. You may want 10649 * unregister_netdev() instead of this. 10650 */ 10651 10652 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10653 { 10654 ASSERT_RTNL(); 10655 10656 if (head) { 10657 list_move_tail(&dev->unreg_list, head); 10658 } else { 10659 LIST_HEAD(single); 10660 10661 list_add(&dev->unreg_list, &single); 10662 unregister_netdevice_many(&single); 10663 } 10664 } 10665 EXPORT_SYMBOL(unregister_netdevice_queue); 10666 10667 /** 10668 * unregister_netdevice_many - unregister many devices 10669 * @head: list of devices 10670 * 10671 * Note: As most callers use a stack allocated list_head, 10672 * we force a list_del() to make sure stack wont be corrupted later. 10673 */ 10674 void unregister_netdevice_many(struct list_head *head) 10675 { 10676 struct net_device *dev, *tmp; 10677 LIST_HEAD(close_head); 10678 10679 BUG_ON(dev_boot_phase); 10680 ASSERT_RTNL(); 10681 10682 if (list_empty(head)) 10683 return; 10684 10685 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10686 /* Some devices call without registering 10687 * for initialization unwind. Remove those 10688 * devices and proceed with the remaining. 10689 */ 10690 if (dev->reg_state == NETREG_UNINITIALIZED) { 10691 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10692 dev->name, dev); 10693 10694 WARN_ON(1); 10695 list_del(&dev->unreg_list); 10696 continue; 10697 } 10698 dev->dismantle = true; 10699 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10700 } 10701 10702 /* If device is running, close it first. */ 10703 list_for_each_entry(dev, head, unreg_list) 10704 list_add_tail(&dev->close_list, &close_head); 10705 dev_close_many(&close_head, true); 10706 10707 list_for_each_entry(dev, head, unreg_list) { 10708 /* And unlink it from device chain. */ 10709 unlist_netdevice(dev); 10710 10711 dev->reg_state = NETREG_UNREGISTERING; 10712 } 10713 flush_all_backlogs(); 10714 10715 synchronize_net(); 10716 10717 list_for_each_entry(dev, head, unreg_list) { 10718 struct sk_buff *skb = NULL; 10719 10720 /* Shutdown queueing discipline. */ 10721 dev_shutdown(dev); 10722 10723 dev_xdp_uninstall(dev); 10724 10725 netdev_offload_xstats_disable_all(dev); 10726 10727 /* Notify protocols, that we are about to destroy 10728 * this device. They should clean all the things. 10729 */ 10730 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10731 10732 if (!dev->rtnl_link_ops || 10733 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10734 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10735 GFP_KERNEL, NULL, 0); 10736 10737 /* 10738 * Flush the unicast and multicast chains 10739 */ 10740 dev_uc_flush(dev); 10741 dev_mc_flush(dev); 10742 10743 netdev_name_node_alt_flush(dev); 10744 netdev_name_node_free(dev->name_node); 10745 10746 if (dev->netdev_ops->ndo_uninit) 10747 dev->netdev_ops->ndo_uninit(dev); 10748 10749 if (skb) 10750 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10751 10752 /* Notifier chain MUST detach us all upper devices. */ 10753 WARN_ON(netdev_has_any_upper_dev(dev)); 10754 WARN_ON(netdev_has_any_lower_dev(dev)); 10755 10756 /* Remove entries from kobject tree */ 10757 netdev_unregister_kobject(dev); 10758 #ifdef CONFIG_XPS 10759 /* Remove XPS queueing entries */ 10760 netif_reset_xps_queues_gt(dev, 0); 10761 #endif 10762 } 10763 10764 synchronize_net(); 10765 10766 list_for_each_entry(dev, head, unreg_list) { 10767 dev_put_track(dev, &dev->dev_registered_tracker); 10768 net_set_todo(dev); 10769 } 10770 10771 list_del(head); 10772 } 10773 EXPORT_SYMBOL(unregister_netdevice_many); 10774 10775 /** 10776 * unregister_netdev - remove device from the kernel 10777 * @dev: device 10778 * 10779 * This function shuts down a device interface and removes it 10780 * from the kernel tables. 10781 * 10782 * This is just a wrapper for unregister_netdevice that takes 10783 * the rtnl semaphore. In general you want to use this and not 10784 * unregister_netdevice. 10785 */ 10786 void unregister_netdev(struct net_device *dev) 10787 { 10788 rtnl_lock(); 10789 unregister_netdevice(dev); 10790 rtnl_unlock(); 10791 } 10792 EXPORT_SYMBOL(unregister_netdev); 10793 10794 /** 10795 * __dev_change_net_namespace - move device to different nethost namespace 10796 * @dev: device 10797 * @net: network namespace 10798 * @pat: If not NULL name pattern to try if the current device name 10799 * is already taken in the destination network namespace. 10800 * @new_ifindex: If not zero, specifies device index in the target 10801 * namespace. 10802 * 10803 * This function shuts down a device interface and moves it 10804 * to a new network namespace. On success 0 is returned, on 10805 * a failure a netagive errno code is returned. 10806 * 10807 * Callers must hold the rtnl semaphore. 10808 */ 10809 10810 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10811 const char *pat, int new_ifindex) 10812 { 10813 struct net *net_old = dev_net(dev); 10814 int err, new_nsid; 10815 10816 ASSERT_RTNL(); 10817 10818 /* Don't allow namespace local devices to be moved. */ 10819 err = -EINVAL; 10820 if (dev->features & NETIF_F_NETNS_LOCAL) 10821 goto out; 10822 10823 /* Ensure the device has been registrered */ 10824 if (dev->reg_state != NETREG_REGISTERED) 10825 goto out; 10826 10827 /* Get out if there is nothing todo */ 10828 err = 0; 10829 if (net_eq(net_old, net)) 10830 goto out; 10831 10832 /* Pick the destination device name, and ensure 10833 * we can use it in the destination network namespace. 10834 */ 10835 err = -EEXIST; 10836 if (netdev_name_in_use(net, dev->name)) { 10837 /* We get here if we can't use the current device name */ 10838 if (!pat) 10839 goto out; 10840 err = dev_get_valid_name(net, dev, pat); 10841 if (err < 0) 10842 goto out; 10843 } 10844 10845 /* Check that new_ifindex isn't used yet. */ 10846 err = -EBUSY; 10847 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10848 goto out; 10849 10850 /* 10851 * And now a mini version of register_netdevice unregister_netdevice. 10852 */ 10853 10854 /* If device is running close it first. */ 10855 dev_close(dev); 10856 10857 /* And unlink it from device chain */ 10858 unlist_netdevice(dev); 10859 10860 synchronize_net(); 10861 10862 /* Shutdown queueing discipline. */ 10863 dev_shutdown(dev); 10864 10865 /* Notify protocols, that we are about to destroy 10866 * this device. They should clean all the things. 10867 * 10868 * Note that dev->reg_state stays at NETREG_REGISTERED. 10869 * This is wanted because this way 8021q and macvlan know 10870 * the device is just moving and can keep their slaves up. 10871 */ 10872 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10873 rcu_barrier(); 10874 10875 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10876 /* If there is an ifindex conflict assign a new one */ 10877 if (!new_ifindex) { 10878 if (__dev_get_by_index(net, dev->ifindex)) 10879 new_ifindex = dev_new_index(net); 10880 else 10881 new_ifindex = dev->ifindex; 10882 } 10883 10884 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10885 new_ifindex); 10886 10887 /* 10888 * Flush the unicast and multicast chains 10889 */ 10890 dev_uc_flush(dev); 10891 dev_mc_flush(dev); 10892 10893 /* Send a netdev-removed uevent to the old namespace */ 10894 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10895 netdev_adjacent_del_links(dev); 10896 10897 /* Move per-net netdevice notifiers that are following the netdevice */ 10898 move_netdevice_notifiers_dev_net(dev, net); 10899 10900 /* Actually switch the network namespace */ 10901 dev_net_set(dev, net); 10902 dev->ifindex = new_ifindex; 10903 10904 /* Send a netdev-add uevent to the new namespace */ 10905 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10906 netdev_adjacent_add_links(dev); 10907 10908 /* Fixup kobjects */ 10909 err = device_rename(&dev->dev, dev->name); 10910 WARN_ON(err); 10911 10912 /* Adapt owner in case owning user namespace of target network 10913 * namespace is different from the original one. 10914 */ 10915 err = netdev_change_owner(dev, net_old, net); 10916 WARN_ON(err); 10917 10918 /* Add the device back in the hashes */ 10919 list_netdevice(dev); 10920 10921 /* Notify protocols, that a new device appeared. */ 10922 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10923 10924 /* 10925 * Prevent userspace races by waiting until the network 10926 * device is fully setup before sending notifications. 10927 */ 10928 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10929 10930 synchronize_net(); 10931 err = 0; 10932 out: 10933 return err; 10934 } 10935 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 10936 10937 static int dev_cpu_dead(unsigned int oldcpu) 10938 { 10939 struct sk_buff **list_skb; 10940 struct sk_buff *skb; 10941 unsigned int cpu; 10942 struct softnet_data *sd, *oldsd, *remsd = NULL; 10943 10944 local_irq_disable(); 10945 cpu = smp_processor_id(); 10946 sd = &per_cpu(softnet_data, cpu); 10947 oldsd = &per_cpu(softnet_data, oldcpu); 10948 10949 /* Find end of our completion_queue. */ 10950 list_skb = &sd->completion_queue; 10951 while (*list_skb) 10952 list_skb = &(*list_skb)->next; 10953 /* Append completion queue from offline CPU. */ 10954 *list_skb = oldsd->completion_queue; 10955 oldsd->completion_queue = NULL; 10956 10957 /* Append output queue from offline CPU. */ 10958 if (oldsd->output_queue) { 10959 *sd->output_queue_tailp = oldsd->output_queue; 10960 sd->output_queue_tailp = oldsd->output_queue_tailp; 10961 oldsd->output_queue = NULL; 10962 oldsd->output_queue_tailp = &oldsd->output_queue; 10963 } 10964 /* Append NAPI poll list from offline CPU, with one exception : 10965 * process_backlog() must be called by cpu owning percpu backlog. 10966 * We properly handle process_queue & input_pkt_queue later. 10967 */ 10968 while (!list_empty(&oldsd->poll_list)) { 10969 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 10970 struct napi_struct, 10971 poll_list); 10972 10973 list_del_init(&napi->poll_list); 10974 if (napi->poll == process_backlog) 10975 napi->state = 0; 10976 else 10977 ____napi_schedule(sd, napi); 10978 } 10979 10980 raise_softirq_irqoff(NET_TX_SOFTIRQ); 10981 local_irq_enable(); 10982 10983 #ifdef CONFIG_RPS 10984 remsd = oldsd->rps_ipi_list; 10985 oldsd->rps_ipi_list = NULL; 10986 #endif 10987 /* send out pending IPI's on offline CPU */ 10988 net_rps_send_ipi(remsd); 10989 10990 /* Process offline CPU's input_pkt_queue */ 10991 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 10992 netif_rx(skb); 10993 input_queue_head_incr(oldsd); 10994 } 10995 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 10996 netif_rx(skb); 10997 input_queue_head_incr(oldsd); 10998 } 10999 11000 return 0; 11001 } 11002 11003 /** 11004 * netdev_increment_features - increment feature set by one 11005 * @all: current feature set 11006 * @one: new feature set 11007 * @mask: mask feature set 11008 * 11009 * Computes a new feature set after adding a device with feature set 11010 * @one to the master device with current feature set @all. Will not 11011 * enable anything that is off in @mask. Returns the new feature set. 11012 */ 11013 netdev_features_t netdev_increment_features(netdev_features_t all, 11014 netdev_features_t one, netdev_features_t mask) 11015 { 11016 if (mask & NETIF_F_HW_CSUM) 11017 mask |= NETIF_F_CSUM_MASK; 11018 mask |= NETIF_F_VLAN_CHALLENGED; 11019 11020 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11021 all &= one | ~NETIF_F_ALL_FOR_ALL; 11022 11023 /* If one device supports hw checksumming, set for all. */ 11024 if (all & NETIF_F_HW_CSUM) 11025 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11026 11027 return all; 11028 } 11029 EXPORT_SYMBOL(netdev_increment_features); 11030 11031 static struct hlist_head * __net_init netdev_create_hash(void) 11032 { 11033 int i; 11034 struct hlist_head *hash; 11035 11036 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11037 if (hash != NULL) 11038 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11039 INIT_HLIST_HEAD(&hash[i]); 11040 11041 return hash; 11042 } 11043 11044 /* Initialize per network namespace state */ 11045 static int __net_init netdev_init(struct net *net) 11046 { 11047 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11048 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11049 11050 INIT_LIST_HEAD(&net->dev_base_head); 11051 11052 net->dev_name_head = netdev_create_hash(); 11053 if (net->dev_name_head == NULL) 11054 goto err_name; 11055 11056 net->dev_index_head = netdev_create_hash(); 11057 if (net->dev_index_head == NULL) 11058 goto err_idx; 11059 11060 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11061 11062 return 0; 11063 11064 err_idx: 11065 kfree(net->dev_name_head); 11066 err_name: 11067 return -ENOMEM; 11068 } 11069 11070 /** 11071 * netdev_drivername - network driver for the device 11072 * @dev: network device 11073 * 11074 * Determine network driver for device. 11075 */ 11076 const char *netdev_drivername(const struct net_device *dev) 11077 { 11078 const struct device_driver *driver; 11079 const struct device *parent; 11080 const char *empty = ""; 11081 11082 parent = dev->dev.parent; 11083 if (!parent) 11084 return empty; 11085 11086 driver = parent->driver; 11087 if (driver && driver->name) 11088 return driver->name; 11089 return empty; 11090 } 11091 11092 static void __netdev_printk(const char *level, const struct net_device *dev, 11093 struct va_format *vaf) 11094 { 11095 if (dev && dev->dev.parent) { 11096 dev_printk_emit(level[1] - '0', 11097 dev->dev.parent, 11098 "%s %s %s%s: %pV", 11099 dev_driver_string(dev->dev.parent), 11100 dev_name(dev->dev.parent), 11101 netdev_name(dev), netdev_reg_state(dev), 11102 vaf); 11103 } else if (dev) { 11104 printk("%s%s%s: %pV", 11105 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11106 } else { 11107 printk("%s(NULL net_device): %pV", level, vaf); 11108 } 11109 } 11110 11111 void netdev_printk(const char *level, const struct net_device *dev, 11112 const char *format, ...) 11113 { 11114 struct va_format vaf; 11115 va_list args; 11116 11117 va_start(args, format); 11118 11119 vaf.fmt = format; 11120 vaf.va = &args; 11121 11122 __netdev_printk(level, dev, &vaf); 11123 11124 va_end(args); 11125 } 11126 EXPORT_SYMBOL(netdev_printk); 11127 11128 #define define_netdev_printk_level(func, level) \ 11129 void func(const struct net_device *dev, const char *fmt, ...) \ 11130 { \ 11131 struct va_format vaf; \ 11132 va_list args; \ 11133 \ 11134 va_start(args, fmt); \ 11135 \ 11136 vaf.fmt = fmt; \ 11137 vaf.va = &args; \ 11138 \ 11139 __netdev_printk(level, dev, &vaf); \ 11140 \ 11141 va_end(args); \ 11142 } \ 11143 EXPORT_SYMBOL(func); 11144 11145 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11146 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11147 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11148 define_netdev_printk_level(netdev_err, KERN_ERR); 11149 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11150 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11151 define_netdev_printk_level(netdev_info, KERN_INFO); 11152 11153 static void __net_exit netdev_exit(struct net *net) 11154 { 11155 kfree(net->dev_name_head); 11156 kfree(net->dev_index_head); 11157 if (net != &init_net) 11158 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11159 } 11160 11161 static struct pernet_operations __net_initdata netdev_net_ops = { 11162 .init = netdev_init, 11163 .exit = netdev_exit, 11164 }; 11165 11166 static void __net_exit default_device_exit_net(struct net *net) 11167 { 11168 struct net_device *dev, *aux; 11169 /* 11170 * Push all migratable network devices back to the 11171 * initial network namespace 11172 */ 11173 ASSERT_RTNL(); 11174 for_each_netdev_safe(net, dev, aux) { 11175 int err; 11176 char fb_name[IFNAMSIZ]; 11177 11178 /* Ignore unmoveable devices (i.e. loopback) */ 11179 if (dev->features & NETIF_F_NETNS_LOCAL) 11180 continue; 11181 11182 /* Leave virtual devices for the generic cleanup */ 11183 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11184 continue; 11185 11186 /* Push remaining network devices to init_net */ 11187 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11188 if (netdev_name_in_use(&init_net, fb_name)) 11189 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11190 err = dev_change_net_namespace(dev, &init_net, fb_name); 11191 if (err) { 11192 pr_emerg("%s: failed to move %s to init_net: %d\n", 11193 __func__, dev->name, err); 11194 BUG(); 11195 } 11196 } 11197 } 11198 11199 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11200 { 11201 /* At exit all network devices most be removed from a network 11202 * namespace. Do this in the reverse order of registration. 11203 * Do this across as many network namespaces as possible to 11204 * improve batching efficiency. 11205 */ 11206 struct net_device *dev; 11207 struct net *net; 11208 LIST_HEAD(dev_kill_list); 11209 11210 rtnl_lock(); 11211 list_for_each_entry(net, net_list, exit_list) { 11212 default_device_exit_net(net); 11213 cond_resched(); 11214 } 11215 11216 list_for_each_entry(net, net_list, exit_list) { 11217 for_each_netdev_reverse(net, dev) { 11218 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11219 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11220 else 11221 unregister_netdevice_queue(dev, &dev_kill_list); 11222 } 11223 } 11224 unregister_netdevice_many(&dev_kill_list); 11225 rtnl_unlock(); 11226 } 11227 11228 static struct pernet_operations __net_initdata default_device_ops = { 11229 .exit_batch = default_device_exit_batch, 11230 }; 11231 11232 /* 11233 * Initialize the DEV module. At boot time this walks the device list and 11234 * unhooks any devices that fail to initialise (normally hardware not 11235 * present) and leaves us with a valid list of present and active devices. 11236 * 11237 */ 11238 11239 /* 11240 * This is called single threaded during boot, so no need 11241 * to take the rtnl semaphore. 11242 */ 11243 static int __init net_dev_init(void) 11244 { 11245 int i, rc = -ENOMEM; 11246 11247 BUG_ON(!dev_boot_phase); 11248 11249 if (dev_proc_init()) 11250 goto out; 11251 11252 if (netdev_kobject_init()) 11253 goto out; 11254 11255 INIT_LIST_HEAD(&ptype_all); 11256 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11257 INIT_LIST_HEAD(&ptype_base[i]); 11258 11259 if (register_pernet_subsys(&netdev_net_ops)) 11260 goto out; 11261 11262 /* 11263 * Initialise the packet receive queues. 11264 */ 11265 11266 for_each_possible_cpu(i) { 11267 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11268 struct softnet_data *sd = &per_cpu(softnet_data, i); 11269 11270 INIT_WORK(flush, flush_backlog); 11271 11272 skb_queue_head_init(&sd->input_pkt_queue); 11273 skb_queue_head_init(&sd->process_queue); 11274 #ifdef CONFIG_XFRM_OFFLOAD 11275 skb_queue_head_init(&sd->xfrm_backlog); 11276 #endif 11277 INIT_LIST_HEAD(&sd->poll_list); 11278 sd->output_queue_tailp = &sd->output_queue; 11279 #ifdef CONFIG_RPS 11280 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11281 sd->cpu = i; 11282 #endif 11283 11284 init_gro_hash(&sd->backlog); 11285 sd->backlog.poll = process_backlog; 11286 sd->backlog.weight = weight_p; 11287 } 11288 11289 dev_boot_phase = 0; 11290 11291 /* The loopback device is special if any other network devices 11292 * is present in a network namespace the loopback device must 11293 * be present. Since we now dynamically allocate and free the 11294 * loopback device ensure this invariant is maintained by 11295 * keeping the loopback device as the first device on the 11296 * list of network devices. Ensuring the loopback devices 11297 * is the first device that appears and the last network device 11298 * that disappears. 11299 */ 11300 if (register_pernet_device(&loopback_net_ops)) 11301 goto out; 11302 11303 if (register_pernet_device(&default_device_ops)) 11304 goto out; 11305 11306 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11307 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11308 11309 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11310 NULL, dev_cpu_dead); 11311 WARN_ON(rc < 0); 11312 rc = 0; 11313 out: 11314 return rc; 11315 } 11316 11317 subsys_initcall(net_dev_init); 11318